6777

6.777J/2.
372J: The MEMSclass

Introduction to MEMS and MEMS Design
Joel Voldman
(with ideas from SDS)
Massachusetts Institute of Technology
Cite as: Joel Voldman, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechanical Devices, Spring 2007. MIT
OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
JV: 2.372J/6.777J Spring 2007, Lecture 1 - 1
Outline
> Class odds and ends

> Intro to MEMS
> The challenge of MEMS Design
> Course outline
> Design projects
and then, Microfab Part I
Handouts
> General Information Handout
Lecturers: Carol Livermore, Joel Voldman

Text: Senturias Microsystem Design
Beware: Errata on website
> Schedule
> Student Information Sheet
VERY IMPORTANT
Fill out and hand back at end of class
Handouts
> Lecture notes
Handed out at beginning of class

Extra copies available at Carols office
> Library Orientation NEXT FRIDAY FEBRUARY 16

Learn how to use online databases, journals, etc.
This will be VERY useful for Problem Set 2
and for life!
Course overview
> Course is broken into two halves
> First half
MEMS design and modeling

Seven problem sets
Differing lengths and complexity
Due on due date IN CLASS
> Second half
Case studies
Design projects
> Grading
15% Problem sets

Regrades on psets must be requested promptly
35% Take-home design problem
50% Final project
Course conduct and ethics

> See policy on cooperation in General Info handout
> We encourage teamwork during the psets
Literature solutions are OK
Students must follow ethical guidelines
All students must write up their own pset
List those you work with on problem set
Cite any literature solutions used
Some behavior is patently unacceptable
Use of prior years homework solutions
>
>
>
>
>
Cooperation is essential in final design project

No cooperation is allowed on take-home design problem
Any breaches will be dealt severely, with no warnings
Please consult us before doing anything questionable
web.mit.edu/academicintegrity/handbook/handbook.pdf
Course overview
> What makes this course challenging?

> Relevant physics in lots of fields must be grasped
quickly
We teach a great deal of material in ~2/3 semester
> Every student will learn new concepts

> Design projects
Complex open-ended design problems

Team dynamics
> All of you can learn MEMS design, and we will try to
make it easier and fun!
Outline

> Intro to MEMS
> Course outline
> Design projects
What are MEMS?
> Micro-Electro-Mechanical Systems

> Microsystems
> Microfabrication
Image removed due to copyright restrictions.
> Microtechnology
> Nanotechnology
> Etc.
Lucas Novasensor
What are MEMS?
> Microfabrication is a manufacturing technology

A way to make stuff
Adapted from semiconductor industry
With changes
Therefore, MANY standard design principles hold
> But has unique elements
New materials: SU-8, PDMS

New ways to shape them: DRIE
New material properties
Bulk vs. thin film
Different physics regimes
Si at small scales
Sandia
JV: 2.372J/6.777J Spring 2007, Lecture 1 -
What are MEMS?
> Definitions vary
Usually made via semiconductor batch fabrication

Usually small
Some important dimension is <1 mm
Ideally, useful
Used to be actual electro-mechanical systems
Sensors: Something moves and is sensed electrically
OR
Actuators: An electrical signal moves something
What are MEMS?
> Now, many MEMS have no

E or M
Static microfluidic structures
But often are multi-domain
Electroother domain is very
Aluminum
heater
Silicon
e.g., ElectroThermalFluidic
pumps
Pumping
chamber
Nozzle valve
Dual bubble
Duty = 5%
Duty = 10%
Duty = 15%
5
4
3
2
1
0
Image by MIT OpenCourseWare.
Diffuser valve
Electric connection
through hole
6
Volume flow rate (l/min)
Microbubble
Liquid inlet
Liquid outlet
popular
actuation
Pyrex glass
(B)
Images by MIT OpenCourseWare.
0
100
200
300
Pulse frequency (Hz)
400
500
Liwei Lin (UCB)
MEMS: Starting points
> Some starting points:
1961 first silicon pressure sensor (Kulite)

Diffused Si piezoresistors mounted onto package
to form diaphragm
Dr. Kurtz (founder) is MIT graduate, of course
Mid 60s: Westinghouse Resonant Gate Transistor

H.C. Nathanson, et al.,
The Resonant Gate Transistor,

IEEE Trans. Electron Devices,
March 1967, 14(3), 117-133.
Figure 1 on page 119 in: Nathanson, H. C., W. E. Newell, R. A. Wickstrom,

and J. R. Davis, Jr. "The Resonant Gate Transistor." IEEE Transacationson
Electron Devices 14, no. 3 (1967): 117-133. 1967 IEEE.
MEMS: Important early work
> Stanford Gas

Chromatograph (1975)
SC Terry, JH Jerman and JB
Angell, IEEE Trans Electron
Devices ED-26 (1979) 1880
WAY ahead of its time
> 70s to today: Ken Wise
Figure 3 on page 1882 in: Terry, S. C., J. H. Jerman, and J. B.

Angell. "A Gas Chromatographic Air Analyzer Fabricated on a
Silicon Wafer." IEEE Transactions on Electron Devices 26,
no. 12 (1979): 1880-1886. 1979 IEEE.
(Michigan) neural probes
> 70s Inkjet printheads

> 70s Start of TI DMD
project
Courtesy of Kensall D. Wise. Used with permission.
> MEMS blossomed in the 80s

> 1982 Kurt Petersen Silicon as a
mechanical material
Proc. IEEE, 70(5), 420-457, 1982.
> Mid-80s BSAC folks (Howe, Muller,

etc.) polysilicon surface
micromachining
T. Lober, MIT
R. T. Howe and R. S. Muller,

Polycrystalline silicon micromechanical
beams, J. of the Electrochemical
Society, 130, 1420-1423, (1983).
> Electrostatic Micromotors

Introduced in 1988-1990
MIT and Berkeley
> Microchip capillary

electrophoresis and lab-on-achip
Introduced ~1990-1994
A. Manz, D.J. Harrison,
Fan et al., IEDM 88, p 666.
others

Harrison et al., Science 261:895, 1993

MEMS: Some current hot topics
> Optical MEMS
Switching of optical signals

Big boom in the late 90s
Big bust in the early 00s
Muller et al., Proc. IEEE, 8:1705, 1998.

Fig. 1 on page 1706 in: Muller, R. S., and K. Y. Lau. "SurfaceMicromachined Microoptical Elements and Systems." Proceedings
of the IEEE 86, no. 8 (August 1998): 1705-1720. 1998 IEEE.
Lucent micromirror
> RF MEMS
Smaller, cheaper, better

way to manipulate RF
signals

Figure 9 on p. 17 in: Nguyen, C. T.-C. "Vibrating RF MEMS
Overview: Applications to Wireless Communications."
Proceedings of SPIE Int Soc Opt Eng 5715 (Jan. 2005): 11-25.
Reliability is issue, but

getting there
Figure 15 on p. 64 in: Nguyen, C. T.-C."Micromechanical Filters for Miniaturized

Low-power Communications." Proceedings of SPIE Int Soc Opt Eng 3673 (July 1999): 55-66.
Cite as: Joel Voldman, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechanical Devices, Spring 20
07. MIT
> BioMEMS
Shows promise for

diagnostics
Next era of quantitative

biology
No commercial winners
yet
Wise (UMich)
Courtesy of Kensall D. Wise. Used with permission.
Voldman (MIT)
Courtesy of Joel Voldman. Used with permission.
Chen (UPenn)
Mathies (UCB)
Courtesy of Richard A. Mathies.
Used with permission.
MEMS: Commercial success
> This isnt just academic curiosity
> There are products you can actually buy
Pressure sensors in your car & in your body

Accelerometers EVERYWHERE
Gyroscopes
HP
Ink-jet print heads

Texas Instruments micro-mirror array
Motorola Razr
Nintendo Wii
MEMS: Commercial success
> The major successes have been

pressure and inertial sensors
Why?
Most mature: 40+ years
Huge initial market: automotive
Recent access to huger
commercial market
Easy access to physical signal
Smaller than alternatives
Cheaper than alternatives
In medical market, that
means disposable
Can be integrated with
electronics
Honeywell microswitch

Analog Devices pressure sensor.
Moderately precise & accurate

Outline

> Intro to MEMS
> Course outline
> Design projects
JV: 2.372/6.777 Spring 2007, Lecture 1 - 22
MEMS Design
> For our purposes, design means
Create a device or system

With quantitative performance parameters (e.g., sensitivity)
Subject to constraints
Size, price, materials, physics
Some clearly defined some not
> This is hard no matter what the device is
MEMS Design
> MEMS design is hard because
The manufacturing technology is actually quite imprecise

10% tolerance on in-plane dimensions is typical
Out-of-plane tolerances may be much better
or much worse
Fabrication success is NOT a given AND is tied to the design
The material properties are unknown or poorly known
The physics are often different
Not the traditional size scales
The system must be partitioned
Which parts to integrate on-chip?
Packaging is non-trivial
NOT like ICs
All these questions should be answered early on

Some solutions to this challenge

> Approach #1
Make something easy or not useful, etc.
> Approach #2
Do incorrect back-of-the-envelope design and then proceed
> Approach #3 (grad student favorite)
Create a large range of structures One of them will work, hopefully
> Approach #4 (the MEMS class way)
Predictive design of all you know to enable chance of 1st round
success
Determine necessary modeling strategies for a given problem
From analytical to numerical
In THIS class we concentrate on analytical and tell you where it
fails
Be aware of what you dont know, cant control, and what your
assumptions are
MEMS Design
> Different levels of design
Analytical design
Abstracted physics
ODEs, Scaling, Lumped-element models
Numerical design
Intermediate approach between physical and analytical
design
Physical level:
3-D simulation of fundamental physics
PDEs, finite-element modeling, etc.
> Tradeoff between accuracy and effort/time

> Always limited by fundamental knowledge of
properties or specifications
Outline

> Intro to MEMS
> Course outline
> Design projects
Course goal
> Course goal: Learn how to design any

microfabricated device/system
> Learn how to
Understand the design process

Partition the system
Determine and model relevant physics
Evaluate different designs & fabrication technologies
Understand the linkage between fabrication and design
Course outline
> First up: fabrication and material properties (4.5 lectures)

> MEMS fabrication is intimately coupled with design
Not true of many other worlds
Example: diaphragm pressure sensor
Would like to use Si because of
piezoresistors
Material choice sets fabrication
technology: KOH
Fabrication technology determines
shapes and physical limits:
diaphragm thickness
This in turn affects performance Photograph of Motorola MPX200 x-ducer.
deflection ~ (thickness)-3
> Material properties also matter greatly
MEMS material properties are often poorly characterized
Course outline
> Fabrication lectures will focus on MEMS process

development
Unit processes
Order-of-operations
Front-end and back-end processing
> These themes will be broadcast throughout the term
Course outline
> Next we introduce the electrical and mechanical domains (2
lectures)
> This gives us something concrete to design

> Split class into two groups for Lectures 6 & 7
> Group 1: Basic Elasticity and Structures
> Group 2: Basic Electronics (Circuits, Devices, Opamps)
> Goal is to teach fundamentals at a slower pace without boring
experts
> Rejoin at Lecture 8

Course outline
> Split sessions

> Which session should I attend?
> Go for material you are less familiar with
MEs go to Electronics
EEs go to Elasticity/Structures
> Notes for both lectures will be available to all

> What if you dont know either subject?
We will hold makeup lectures of Elasticity/Structures

Please let us know ASAP if you need a makeup
Course outline
> Next we present an approach to design (3 lectures)

Lumped-element modeling
Different energy domains all use common language
Electrical, Magnetic, Structural, Fluidic, Thermal
Therefore, when you encounter a new domain, you can
quickly attach it to existing knowledge
Enables quick design

But has limits
Course outline
> Then we explore additional energy domains (6

lectures)
Structures, Thermal, Fluids & Transport
What physics are relevant?
Not all of fluids, just low-Reynolds-number flows
How do we extract lumped-element or analytical models?
What is the resistance of a microfluidic channel?
Course outline
> Systems issues (4 lectures)
Noise, feedback, packaging, design tradeoffs
> Partitioning
A major theme of the course

Cant design device with process
Also cant design device without package
Should you put any electronics on-chip?
Can you design MEMS to make read-out easier?
What are the trade-offs between different choices
Course outline
> Finally, case studies
Integrate everything we have up to now to learn about design

process of actual devices
Analog Devices accelerometer

TI micro-mirror
BioMEMS such as integrated PCR devices
Outline

> Intro to MEMS
> Course outline
> Take-home and team design projects
Design projects
> Design is the heart of this course

> We will have short design problems on the psets
> In March, we will have a take-home graded design
problem
Multifaceted: fabrication, electromechanical analytical design
Students will prove their design to staff
> In April, team design projects start
Design projects
> Projects, projects, projects
Teams of 4-6 people

Chosen by US, with input from you
Only students taking class for CREDIT can participate
All teams have a mentor
> Paper design of a MEMS-based device
Quantitative system-level specifications

Analytical design, finite-element modeling, fabrication,
packaging, electronics, calibration, etc.

Final project grade 50% due to team, 50% due to individual
> Lectures will focus on case studies

> Almost no more homeworks
Design projects
> Project timeline
Short description March 9

Your preferences March 23
Teams assigned April 4
Preliminary report
Is team functioning and has it started?
Intermediate report
Is team functioning and is it going to finish?
Final presentations and report
~30-min presentation in front of judges
20-pg manuscript-quality report
Significant prize to winning team
Design projects
> Use to illustrate course approach

> A piezoresistive sensor for biomolecular recognition
(2003)
The goal of this project is to create cantilever-based device
that detects stress induced by molecular binding.
Two cantilevers (operated differentially) will be created out of

Si with integrated poly-Si piezoresistors.
The packaged device will be used in a hand-held point-ofcare diagnostic monitor and so must be robust, small, and
connected to a circuit that gives an output proportional to the
logarithm of the concentration ratio.
> Show slides from presentation to illustrate design

process
Images removed due copyright restrictions.

Student final presentation: Gerhardt, Antimony L., Saif A. Kahn, Adam D. Rosenthal,
Nicaulas A. Sabourin, and Keng-Hoong Wee. "A Piezoresistive Molecular Binding Detector."
Microfabrication for MEMS: Part I

Carol Livermore
* With thanks to Steve Senturia, from whose lecture notes

some of these materials are adapted.
Cite as: Carol Livermore, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechanical Devices, Spring 2007. MIT
CL: 6.777J/2.372J Spring 2007, Lecture 1 - 1
Outline
> A way to think about process design
> Surface and bulk micromachining concepts
> Next class: substrates, lithography and patterning, and thin
films
> Class #3: etching, wafer bonding (including bulk

micromachining), surface micromachining and process
integration
> Class #4: in class process design exercise

> Class #5: fabrication for the life sciences and material
properties
Learning to speak fab

> There are huge lists of things that you can and cant do in
microfab (vocabulary)
> There are limits on how you can combine steps together to form
a process (grammar)
> An experienced microfabricator knows most of these

capabilities and limits from long experience (fluency)
> A beginning microfabricator must struggle to master details and

rules
> Our goal: teach enough vocabulary and grammar to get you
started (today starts with grammar)
> Only practice will make you fluent!

The promise and the pain

> Microfabrication can produce a near infinite array of structures
> But making your particular structure can be very difficult!
> Youd think it would be easy
Functional requirements dictate materials and geometry
> but its not! Fabrication processes are not ideal.

Materials property variability
Geometric variability
> Early phase design must seek a self-consistent solution to the

question of what will I make and how will I make it?
> But for that, we must be familiar with our tools and how they
work
Crayon Engineering
> At the early stages of MEMS design, conceptual process design
rather than highly detailed process design is required
> It is necessary to obey

The laws of physics
Constraints on temperatures
Constraints on chemical compatibility
Design rules on geometry
> But we can draw our processes with crayons
Hence the phrase crayon engineering
> As we approach the final design, we need CAD tools
Making integrated circuits

Wafers in
1. The Front End (creating

transistors, etc.):
Hot processes: growing
defect-free oxides, diffusing
dopants to create
transistors
Clean processes: Junk will
diffuse in, too, if you let it
2. The Back End (wiring up

the system):
Cooler processes:
depositing metal interconnects, separating layers
with less perfect insulators
Not-so-clean processes:
Dirt is still bad, but you can
wash it off
Integrated circuits out

Things to remember about IC fab

> A LOT of effort has gone into designing IC processes so that
they achieve the desired result and dont violate the laws of
physics
> Why these processes are hard to design and improve:

What if the circuit would be better if it included feature A, but
making feature A requires a back end step followed by a front end
step?
Typical solutions: omit feature A or create a back end version of

the front end step
An unacceptable solution: just go from the back end process into

the front end tool. This will contaminate the fab line and may melt
your back end features in the process.
> Bottom line: IC processing is always done the same way, with a
few variants
Things to remember about MEMS fab

> You are subject to many of the same restrictions as the IC
designers
Laws of physics
Restrictions imposed by vendors: you cant do x because it would
contaminate my tools from the point of view of my other customers
> but you have additional challenges and advantages

Often there is no baseline process to work from (though the
literature can often give you a good start)
There is no guarantee that conventional processes and tools can

produce your part
The MEMS tool kit is significantly broader than the IC tool kit (so
theres more available, but its up to you to figure out how to turn
the raw material into a functional process)
The MEMS tool kit
The Front End

(with some front
end-compatible
MEMS additions)
The Back End

(with some back
end-compatible
MEMS additions)
Everything else:
Soft lithography (polymers, stamping,
microcontact printing)
Electroplating
Electron beam lithography
Spin-cast materials (e.g. spin-on glass)
Piezoelectrics
Magnetic materials
Anodic bonding of glass to silicon
An important caveat
> MEMS processes also basically flow from the front end to the
back end
> The lines between front end and back end processes are not set
in stone they depend on your tolerance for risk/contamination.
How clean is clean?

How hot is hot?
> We will present unit processes according to whether they are

roughly front end, back end, or everything else processes, but
it is up to you, the process designer, to consider the actual
temperatures, contamination, and materials compatibilities that
are involved. What is clean for my application may not be clean
enough for yours.
Implementation of rules is lab-specific (MTL example)
Chart removed due to copyright restrictions.
Complete chart is at http://www-mtl.mit.edu/services/fabrication/ptc_matrix.html

Outline
> A way to think about process design
> Surface and bulk micromachining concepts
> Next class: substrates, lithography and patterning, and thin
films
> Class #3: etching, wafer bonding (including bulk

micromachining), surface micromachining and process
integration
> Class #4: in class process design exercise

> Class #5: fabrication for the life sciences and material
properties
Fundamentals of Surface Micromachining

Surface Micromachining
> A technique for fabricating micromechanical devices by

selectively depositing and etching thin films on the surface of a
substrate
> Structural layers are the thin films that form the final structure
> Sacrificial layers are thin films that support a MEMS structure
during fabrication, until it is released (that is, until the sacrificial
layer is etched away)
Important techniques for surface micromachining
> Lithography and patterning

> Wet and dry etching
> Deposit thin films by CVD, PECVD, thermal oxidation,
evaporation, etc.
Sandia: surface micromachined gear chain
Courtesy of Sandia National Laborator

i
es, SUMMiT Technologies, www.mems.sandia.gov
Texas Instruments Digital Mirror Display

> One MEMS example:
inside the projector
Images removed due to copyright restrictions.

Figures 48 and 50 in Hornbeck, Larry J. "From Cathode Rays to Digital Micromirrors: A History of Electronic Projection Display
Technology." Texas Instruments Technical Journal 15, no. 3 (July-September 1998): 7-46.
http://www.dlp.com/
Berkeley: surface micromachined microphotonics
Figure 5 on page 316 in: Friedberger, A., and R. S. Muller. "Improved Surface-micromachined Hinges
for Fold-out Structures." Journal of Microelectromechanical Systems 7, no. 3 (1998): 315-319. 1998 IEEE.
Fundamentals of Bulk Micromachining

Bulk Micromachining
> Selectively etch bulk substrate to create microelectronic or

micromechanical structures
> Remaining material forms the device structure

> Moving structures, trenches, and membranes possible
Important Techniques for Bulk Micromachining

> Wet and dry etching
Etch rate and profiles are important
Selectivity and masking materials, too!
> Wafer bonding
A process for permanently attaching two wafers together
MEMS energy harvester
Figure 5 on page 67 in: Meninger, S., J. O. Mur-Miranda, R. Amirtharajah,

A. Chandrakasan, and J. H. Lang. "Vibration-to-electric Energy Conversion."
IEEE Transactions on Very Large Scale Integration (VLSI) Systems 9, no. 1
(2001): 64-76. 2001 IEEE.
Designed for microWatts.
Mur-Miranda, Jose Oscar. "Electrostatic Vibration-to-Electric

Energy Conversion." Doctoral diss., Massachusetts Institute
of Technology, 2004, p. 88.
MIT: bulk micromachined microengine

3-D cutaway of a micromachined microengine. Photograph by Jonathan Protz.
NIST: bulk micromachined micro-heater
Polysilicon microheater suspended

above silicon substrate
Figure 1 on page 57 in: Parameswaran, M., A. M. Robinson, D. L. Blackburn, M.

Gaitan, and J. Geist. "Micromachined Thermal Radiation Emitter from a Commercial
CMOS Process." IEEE Electron Device Letters 12, no. 2 (Feb. 1991): 57-59. 1991 IEEE.
Preview: Unit Processes and Integration

> Next class we will start to fill in the details of the front end, back
end, and everything else with a set of unit process steps out of
which processes can be built (vocabulary)
Oxidation
Implantation and Diffusion
Lithography
Selective Etching
etc.
> But always remember to consider compatibility of the steps in

the overall, integrated process sequence (grammar) this is
critical to the success of the MEMS device
> Also remember that theoretical compatibility isnt enough! The

process sequence must be available at your facility or your
vendors facility in order for you to use it!
Microfabrication for MEMS: Part II
Carol Livermore
* With thanks to Steve Senturia, from whose lecture notes some of

these materials are adapted.
Outline
> Substrates
> Doping
> Thin films
Silicon Wafers
> Silicon is a diamond-structure cubic crystal
> Comes with different amounts of either n-type or p-type doping
z
Simple cubic crystal

z
(100) Plane
z
(110) Plane
(111) Plane
Adapted from Figure 3.1 in Senturia, Stephen D. Microsystem Design. Boston,

MA: Kluwer Academic Publishers, 2001, p. 31. ISBN: 9780792372462.
Notation
> A direction in crystal coordinates is denoted by square brackets, e.g.
[100]
> The set of symmetrically equivalent directions is written with braces,

e.g. <100>
> The plane perpendicular to a direction is denoted with parentheses,

e.g. (100)
> The set of symmetrically equivalent planes is written with curly

brackets, e.g. {100}
Diamond Structure
> The diamond structure is two face-centered cubic lattices shifted by
of the body diagonal. There are four silicon atoms per cubic unit
cell.

Adapted from Figure 3.2 in: Senturia, Stephen D. Microsystem Design. Boston,
Wafer orientation
(110) plane
(100) plane
45
(100) planes
(110) primary flat
(100) type wafer
Adapted from: Maluf, Nadim. An Introduction to M icroelectromechanical Systems Enginee ring.
Boston, MA: Artech House, 2000. ISBN: 9780890065815.
Picture from N. Maluf, An Introduction to Microelectromechanical

Systems Engineering
Other substrates
> Silicon wafers with embedded layers, such as silicon-oninsulator (SOI) wafers
device layer
buried oxide layer
substrate
Initial purpose: build ICs on device layer, and buried oxide

minimizes stray capacitance to substrate
Common MEMS purpose: bulk micromachine top layer into

moveable structures with well-controlled thickness
$$$$
> Quartz wafers
Single crystal
Fused quartz amorphous quartz wafers
Other substrates
> Glass (cheap, high impurity content)
Inexpensive base for soft lithography
Transparent for optical access
Can be very strongly attached to silicon wafers via anodic bonding
> Compound semiconductors (III-Vs, II-VIs)

Optical applications
> Sapphire
Strong, wear resistant, transparent, insulating substrate
Compatible with CMOS (so transparent CMOS MEMS)
Expensive, hard to etch
Substrate summary
Substrate
Front end
compatible
Back end
compatible
Everything else
compatible
Silicon
yes +
yes
yes, but only use

if needed
Silicon on
insulator (SOI)
yes +
yes
yes, but only use

if needed
Quartz
yes
yes
yes, but only use

if needed
Glass (pyrex)
no
yes, sometimes
yes +
Compound
semiconductor
no
yes
yes, but only use

if needed
Sapphire
yes, but only use yes, but only use yes, but only use
if needed
if needed
if needed
Outline
> Substrates
> Doping
> Thin films
Optical Lithography
Spin-cast a photosensitive
resist layer; bake out solvent
Collimated UV exposure
through a mask; resist either
cross-links or becomes
soluble
positive
negative
Develop by dissolving the

exposed/unexposed
(positive/negative) resist; can
now transfer pattern to
substrate
Alignment fiducials permit alignment of subsequent masks.

Methods of optical lithography I

> Contact
Mask touches wafer

Inexpensive
Contact degrades mask
No die size limit
Resolution: down to 1 micron nervously; down to several
microns comfortably
> Proximity
Mask of order 10 microns from wafer

Inexpensive
Less mask damage
Diffraction means lower resolution
No die size limit
Resolution: down to several microns nervously, somewhat
larger comfortably
Projection Lithography
> Projection lithography, especially
when combined with an optical
imaging system that reduces the
image size, is used for high-resolution
patterning (submicron to very
submicron)
> Larger mask features, no contact with

mask
UV Light
Condensing
Lens
Mask
Projection
Optics
Wafer
> Wafer steppers expose one die at a

time, assuring good focus and
registration
> Something to consider: if your device

Adapted from Figure 3.15 in: Senturia, Stephen D.
Microsystem Design. Boston, MA: Kluwer
Academic Publishers, 2001, p. 53.
ISBN: 9780792372462.
needs fine features, a stepper may be

required. But steppers have limits on
die size of about 1 cm.
Mask making
> Highest quality chromium on fused quartz written with an electron
beam exposing an electron-beam resist (PMMA)
Also very high quality: laser-writing

> Photographic emulsion on fused quartz exposed with UV light flashes
through a programmable aperture
> Patterns printed from an AutoCAD file on transparencies with a veryhigh-resolution printer low resolution, but cheap and fast
Positive thin resist

> Spin cast
> Thickness of order 1 micron
> Developer removes exposed resist
> Creates sloped profile at resist edge
> Some applications
Wet etching
Shallow dry etching
> Front end standard
Negative/image reversal photoresists

> Spin cast
> Thickness of order 1 micron
> Developer removes unexposed resist
> Creates a re-entrant profile
> Typical application: liftoff processes
(in acetone), often seen in back end
processing
> Rule of thumb: resist thickness should

be 3x thickness of layer to be lifted off
> Not a standard front end material, but

not inherently incompatible with it
Positive thick photoresist

> Spin cast
> Thicknesses of order 10 microns
> Sloped profiles
Slope somewhat controllable
through process conditions
> Some planarizing capability

> Typical applications:
Prolonged or low selectivity dry
etch
Deep reactive ion etch

Masking any etch over topography
> Not a standard front end material,

but not inherently incompatible
Courtesy of Reza Ghodssi. Used with permission.
with it
5 m thick
Double-sided aligned lithography

> Goal: align features on the back of the wafer to features on the
front
Common requirement in bulk micromachining

Not a standard IC capability
Functionality more common as market grows
> What you need:
Double side polished wafer
Double sided alignment tool
IR alignment, registration to global fiducials in the tool,
through holes, etc.
Special purpose lithographic techniques I

> X-ray lithography
One application: making molds for LIGA
Requires an x-ray source and x-ray mask
> Electron beam lithography
High resolution (tens of nanometers)
NEMS (NanoElectroMechanical Systems)
A slow, serial process
> Lithographic techniques that are rarely seen in front end
processing
Special purpose lithographic techniques II

> Shadow masking
Direct evaporation or sputtering through physical holes in a
shadow mask (think stencils)
Back end/everything else process

Last ditch technique for patterning surfaces that cannot be
coated with resist (large topography, fragile features)
> Soft lithography (Whitesides, Harvard)

Using polydimethylsiloxane (PDMS, a rubber material) as a
physical mold to replicate structures

Advantages:
Patterning curved surfaces
Rapid, inexpensive fabrication
A set of everything else processes
More on this later
Very thick photoresists

> SU-8 epoxy
Spin cast
Negative resist, optical
exposure
SU-8 epoxy photograph.

Can planarize extreme

topographies
Can be structural, not easily

dissolved
> Polyimide
Spin cast
Can planarize topographies
Humidity sensitive
500 micrometer SU-8 epoxy within deep silicon trench.

> Classified as everything else

Details that matter for lithographic patterning

> Existing topography: if your existing feature heights are comparable
to or greater than the thickness of the resist that you are putting
down, you will not have good coverage
Incompletely covered sidewalls, holes full of resist, resist that never
enters a hole at all

Solutions: eliminate the topography, thicker resist, alternate coating
technology (spray on, electrophoretic photoresist?), use of a previously
patterned hard mask instead of a resist mask
> Patterned resist does not have a square profile can affect the
topography of whatever you pattern with the resist
> Resist adhesion

If the surface of the wafer is hydrophilic (like SiO2), the resist might peel
during subsequent wet processing steps
Surface preparation is key (e.g. dehydration bake and HMDS coating to

render surface hydrophobic)
Cleaning!
> When we say (for example) that positive thin resist is compatible with
front end processing, we do not mean that you can have resist on
your wafer during most front end processes!
> Must remove resist and clean wafer thoroughly before high
temperature processes
> Always include cleaning in process flows, starting at the crayon

engineering level
> Resist removal techniques:

O2 plasma ash
Chemical removal of organics: piranha clean (3:1 H2SO4:H202) or
Nanostrip (a weaker version of piranha)
Solvents (acetone)
More cleaning!
> Additional cleans typically needed at specified points in the flow
> Example: RCA clean before very high temperature processing (as in furnace
for front end processing)
Step 1: Organic clean, 5:1:1 H2O:H2O2:NH4OH at 75 - 80C

Step 2: Thin oxide removal, 50:1 H2 O:HF
Step 3: Metal/ionic contamination removal, 6:1:1 H2O:H2O2: HCl at 75
80C
> Example: remove organics before moderately high temperature, fairly clean
processing (upper part of back end processing)
Piranha clean (3:1 H2SO4:H202)
> Materials compatibility (what cleans your structures can tolerate) often
determine what processes you can and cant use
> If you wait until the last minute to put cleans into your process flow, you will
likely be redesigning your device and process at the last minute
Another important detail: process bias

> The feature drawn on the mask is not the same size as the
feature produced on the wafer
> Exposed area usually extends beyond clear area on mask

> Resist selection impacts process bias
Resist thickness
Resist tone
Design Rules
After Mask 1
> Alignment of one

pattern to the next is
critical to device
fabrication
> Design rules are

created to assure that
fabrication tolerances
do not destroy devices
Feature on
Mask 1
Mandatory
Overlap
Design Rule
Feature on
Mask 2
After Mask 2:
Desired alignment
Barely acceptable misalignment
Unacceptable misalignment
Adapted from Figure 3.16 in: Senturia, Stephen D. Microsystem Design.
Norwell, MA: Kluwer Academic Publishers, 2001, p. 54. ISBN: 9780792372462.
Outline
> Substrates
> Doping
> Thin films
Doping
> Doping is the introduction of a controlled amount of

impurities to change the conductivity type and degree of a
semiconductor
> In silicon, boron is a p-type dopant (creating holes), while

phosphorus, arsenic, and antimony are n-type dopants
(creating conduction electrons)
> Some doping incorporated in initial silicon melt

> All modern thin film doping is done with ion implanation
> Doping doesnt add a new thin film, but it modifies the
properties of a thin film at the surface of existing material
Ion implantation
> A high-voltage accelerator is used to shoot ions at the wafer.
> The beam must be rastered and the wafer must be rotated to achieve
uniform dose
> Usually a thin protective layer, such as oxide, is used to prevent

sputtering of the surface and to reduce channeling
> The depth of the implant dose depends on energy

> Activation anneal after implantation allows dopants to reach proper
positions in crystal
Effective range
> The effective range measures the location of the

peak concentration of an implanted species.
10
Projected range (m)
rus
ho
osp
Ph
0.1
ron
Bo
enic
Ars
0.01
0.001
10
100
Energy (keV)
1000

This image is for illustration purposes only; it should not be used for design calculations.
Adapted from Figure 3.6 in Senturia, Stephen D. Microsystem Design. Boston, MA:
Kluwer Academic Publishers, 2001, p. 39. ISBN: 9780792372462.
Masking of implants
> Control of which regions of a wafer receive the

implant is achieved with masking layers
Ions
Photoresist
Oxide
Silicon
Location of implanted ions
Boston, MA: Kluwer Academic Publishers, 2001, p. 39. ISBN: 9780792372462.
Diffusion
> After implantation, ions are driven deeper into the substrate by
6
5
Implant dose
x j = (4 Dt ) ln
N D Dt
Junction depth (microns)
>
diffusion, a high-temperature process

The junction depth is the point at which the implanted ion
concentration is equal (but of opposite type) to the substrate doping
4
3
2
1
Background doping
concentration
0
103
104
105
Time (seconds)

This image should not be used for design calculations. Adapted from Example 2.2 in:
Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers,
2001, p. 43. ISBN: 9780792372462.
Outline
> Substrates
> Doping
> Thin films
Creating thin (and thick) films

> Many techniques to choose from
> Differences:
Front or back end processes
Quality of resulting films (electrical properties, etch
selectivity, defects, residual stress)

Conformality
Deposition rate, cost
> Physical techniques

Material is removed from a source, carried to the substrate,
and dropped there
> Chemical techniques

Reactants are transported to the substrate, a chemical
reaction occurs, and the products deposit on the substrate to
form the desired film
Taxonomy of deposition techniques

> Chemical
Thermal Oxidation
Chemical Vapor Deposition (CVD)
Low Pressure (LPCVD), Atmospheric Pressure
(APCVD), Plasma Enhanced (PECVD)
Epitaxy
Electrodeposition (Electroplating)
> Physical
Physical Vapor Deposition (PVD)
Evaporation
Sputtering
Spin-casting
Oxidation I
> Silicon forms a high quality, stable oxide
> How it works:
Oxygen diffuses through oxide to Si/oxide interface
Si + O2 + high temperature (~1100 C) furnace SiO2
Some Si is consumed
tox
0.46 tox
> Rate determined by diffusion of oxygen through oxide
> Diffusion limits practical oxide thickness to < 2 m
> A key front end process
Oxidation II
> Dry oxidation (O2)
High quality, slow oxidation rate, smaller maximum thickness
(i.e. gate oxide)
> Wet oxidation (steam)

H2 to speed the diffusion
Lower quality, faster oxidation rate
> The Deal-Grove model describes the kinetics of oxidation quite
well for oxides greater in thickness than about 30 nm.
Deal-Grove Model
For oxides greater than about 30 nm thick:
4 B DG
x final = 0.5 ADG 1 + 2 (t + DG ) 1
ADG
where
DG =
x
B
2
i
DG
x
B
DG
/ ADG
(Constants are given in the text; beware units

of BDG, m2/hour)
Growth goes approximately as t for short times, and
approximately as t for long times.
Local Oxidation
> Oxidation can be masked locally by an oxidation barrier, such as
silicon nitride
> Oxide undercuts edge of mask layer to form a birds beak

> Oxidation followed by oxide etch can also be used to sharpen silicon
features.
Oxidation barrier
Silicon wafer
Birds beaks
After oxidation
Senturia, Microsystem Design.

Norwell, MA: Kluwer Academic Publishers, 2001, p. 37. ISBN: 9780792372462.
Chemical Vapor Deposition (CVD)

> How CVD works:
Gaseous reactants, often at low pressure
Long mean free path; reactants reach substrate
Reactants react and deposit products on the substrate
Unlike oxidation, does not consume substrate material
> Energy sources facilitate CVD reactions:
High temperature, plasma, laser
> Processing temperatures vary widely
> Commonly deposited films: Oxide, silicon nitride, polysilicon
> CVD results depend on pressures, gas flows, temperature
Film composition, uniformity, deposition rate, and electrical and
mechanical characteristics can vary
> Near the boundary between front and back end, depending
Some reasons to use CVD

> Oxide formation:
To get a thicker layer than thermal oxidation can provide
To create oxide on a wafer that cant withstand high
temperatures (for example because of metal features)

To create oxide on top of a material that is not silicon
> For film formation in general:

To tailor the film properties (like film stress) by adjusting
pressures, flow rates, external energy supply, ratios of different

precursor gases (to adjust proportions of different materials in
the final product)
Conformality: (more or less) even coating on all surfaces
> Drawbacks
Films deposited at low temperature are often lower quality than
high temperature versions, and have less predictable properties

Flammable, toxic, or corrosive source gases
Thick Film Formation

> Chemical vapor deposition is a common MEMS tool for creating thick
films on the wafer surface
In practice, film stress limits thickness (film delamination or cracking, or
curvature of underlying structures)

Can deposit thick oxides; nitrides are still typically submicron
Must anneal deposited oxides for some applications lose low stress
property on anneal
> Example: PECVD deposition (350-400C) of 10 to 20 m oxides
25 m thick cracked oxide, X.

Zhang et al., Hilton Head 2000
18 m thick oxide insulation for a

microgenerator w/ Pt features
CVD enables conformal coating
MEMS-based fuel processor.
Figure 4 on p. 605 in: Arana, L.R., S. B. Schaevitz, A. J. Franz,
Figure 5 on p. 606 in: Arana, L.R., S. B. Schaevitz, A. J. Franz,
M. A. Schmidt, K. F. Jensen, "A Microfabricated Suspended-tube
M. A. Schmidt, K. F. Jensen, "A Microfabricated Suspended-tube
Chemical Reactor for Thermally Efficient Fuel Processing." Journal

of Microelectromechanical Systems 12, no. 5 (2003): 600-612.
Chemical Reactor for Thermally Efficient Fuel Processing." Journal

of Microelectromechanical Systems 12, no. 5 (2003): 600-612.
2003 IEEE.
2003 IEEE.
LPCVD Polysilicon
> Amorphous at lower deposition temperatures and higher
deposition rates
Typical temperature: ~ 590 C

> Polycrystalline at higher deposition temperatures and lower
deposition rates
Typical temperature: ~ 625 C

> Grain size and structure depend on detailed deposition
conditions
e.g. thicker films larger grains

> Structure, electrical properties, and mechanical properties also
vary with post-deposition thermal processing
Grain growth
Dopant activation or diffusion
Polysilicon stress depends on deposition details

Kurt Petersen, Trans Sensory Devices
Lober, Theresa Ann. "A Microfabricated Electrostatic

Motor Design and Process." Masters thesis, Massachusetts
Institute of Technology, 1988. 132 pages.
Epitaxy
> CVD deposition process in which atoms move to lattice sites,
continuing the substrates crystal structure
Homoepitaxy: same material, i.e. Si on Si

Heteroepitaxy: different materials, i.e. AlGaAs on GaAs
> How it happens
Slow deposition rate (enough time to find a lattice site)
High temperature (enough energy to move to a lattice site)
> Selective epitaxy is possible through masking
> Can grow a doped Si layer of known thickness
Electroplating: basics
> Pass a current through an aqueous metal solution
Anode is made of the metal that you want to deposit
Cathode is the conductive seed material on your wafer
Positive metal ions travel to the negatively charged cathode on
your wafer and deposit there
> Preparing your wafer

If you want to plate metal in some places and not in others, you will
need a patterned metal seed layer (and typically a sticky metal
adhesion layer under that)
For very short features, just plate onto the seed layer
For taller features, need to plate into a mold
Molds can be photoresist, silicon, SU-8, etc., depending on the
needs of your device
Electroplating
Electroplating for LIGA
40 m thick films of
nickel fabricated by
electroplating into a mold
This picture belongs to Professor Reza Ghodssi (University of Maryland) and was published in the following reference:
Ghodssi, R., D. J. Beebe, V. White, and D. D. Denton. "Development of a Tangential Tactor Using a LIGA/MEMS Linear
Microactuator Technology." Proceedings of the 1996 ASME Winter Annual Mtg., Symposium on Micro-Mechanical Systems,
Atlanta, Georgia, pp. 379-386, November 17-22, 1996.
Electroplating realities
Test run w/ bump

plating - perfect
Real device forms
keyholes different
loading pattern
Courtesy of Dariusz Golda. Used with permission.
Solution: Cu damascene
fill, with additives/agitation
to promote fill at bottom
Conformality and keyholes

> To lowest order, conformal films coat sidewalls and horizontal
surfaces at the same rate.
> But high aspect ratio trenches are prone to keyholes (CVD,
electroplating, etc.)
What you want:
What you get:
Physical Vapor Deposition (PVD)

> Remove material from a solid source
> Transport material to substrate
> Deposit material on substrate
> Differences among PVD techniques
How material is removed from source
Directionality when it arrives at substrate
Cleanliness of deposition
> A family of quick, low temperature processes
> All back end processes or worse
Thermal Evaporation
Rsource
> Source is resistively heated in high vacuum

Typical source: metal
> Hot source atoms are emitted in all directions and stick where
they land
> Substrate receives a directional flux of source material

Good for liftoff processes, otherwise poor conformality
> Possible contamination from generalized heating
E-beam Evaporation
electron
gun
V
esource
> Electron beam heats source in high vacuum

Typical source: metal
> Hot source atoms are emitted in all directions and stick where
they land
> Substrate receives a directional flux of source material

Good for liftoff processes, otherwise poor conformality
> Heating is less generalized less contamination
Sputtering
= Ar ion
target
> Unreactive ions (i.e. Ar) knock material off a target by momentum
transfer
> Targets: metals, dielectrics, piezoelectrics, etc.

> Different methods of obtaining energetic ions
Magnetron, plasma
> Low pressure, but not high vacuum
> Less directional and faster than evaporation
Etching, liftoff, and adhesion layers

> Films are patterned differently depending on whether the material in
question tends to react with other materials
> Materials that react (for example, aluminum):

Deposit a blanket film (sputtering good for better conformality), do
photolithography, and etch it into the desired shape
> Materials that dont react readily (for example, noble metals):
Hard to etch: typically use liftoff instead
Pattern resist, then deposit metal on top with a directional deposition tool
Not very sticky: typically need an adhesion layer to stick the noble metal
to what lies beneath
Example: use a few hundred A thick layer of Cr or Ti to adhere Au to an

underlying oxide (deposited without breaking vacuum between layers)
Is this all you can do with deposited films?
> No!
> Spin-casting: put the stuff that you want to deposit in a
liquid, spin it onto the surface like resist, and bake out the
solvent (spin on glass, PZT piezoelectrics)
> Other forms of vapor deposition designed for a particular

purpose (depositing the inert polymer parylene by vapor
deposition followed by polymerization)
> Lamination of free-standing resist films onto surfaces

> Self assembled monolayers
>
Microfabrication for MEMS: Part III

Carol Livermore

Outline
> Etching
> Wafer bonding
> Surface micromachining
> Process integration
Etching
> Wet etching

Isotropic
Anisotropic (for crystals only)
> Dry etching using plasma reactors

Isotropic plasma etching at relatively high gas pressures
Anisotropic reaction-ion etching at relatively lower gas
pressures
> Sputter etching or ion-beam milling

Not very selective
> A useful reference (what etches what and how fast):

Williams, Gupta, and Wasilik, Etch Rates for Micromachining
Processing Part II, JMEMS 12, 761-778 (2003).
Considerations for etching

> Isotropic
Etch rate the same in all crystal directions
> Anisotropic
For wet etches, rate depends on crystal plane
For dry etches, directionality determined by process
> Selectivity
Etch rate of substrate vs. etch rate of mask
> Mask adhesion (for wet etching)
Increased etching along mask/substrate interface
> Temperature
Reaction rate limited?
> Stirring
Mass transfer limited?
Isotropic etching
> Etch rate is independent of orientation
> Isotropic etch profile
Assume a well-adhered mask with infinite selectivity
Mask undercut, rounded etch profile
> Applications:
Flow channels
Removal of sacrificial layers in surface micromachining
Isotropic etching
> Some wet etches:
Si
mixture of nitric, acetic, and hydrofluoric acid
SiO2
buffered HF (BOE), also HF vapor
SiN
hot phosphoric acid
PolySi
KOH
Al
PAN etch (phosphoric, acetic, nitric acids)
> Some dry etches:

Si
XeF2 vapor
Organics O2 plasma
> Mostly clean enough for front end, with the exception of
KOH, which is a contamination risk for very high T
processes. XeF2 vapor is often used as a final release etch.
Anisotropic wet etching

> Depends on having a single-crystal substrate
> The effect depends on the different etch rates of different
exposed crystal planes
> Silicon etchants for which <111> planes etch slowly

Strong bases (KOH, NaOH, NH4OH)
TMAH
Ethylene diamine pyrochatechol
Hydrazine
Making a Trench with KOH

Before Etching
[110]
Masking Layer
Top
View
> A rectangular
(100) Silicon Substrate
Cross Section
After Etching
[110]
Top
View
[111]
[100]
pattern is aligned
to a [110]
direction on a
<100> silicon
wafer
54.70
Cross Section
Making a V-groove
> The previous etch is allowed to go to termination,

i.e. the slowing of etch rate when only {111} planes
are exposed
[110]
Top
View
54.7 0
Cross Section
54.7 0
Cross Section
> Can also make a square, pyramidal hole

Convex corners
[110]
> Convex corners

become
undercut, as
there is no single
slow-etching
(111) plane to
stop on
Masking Layer
[100]
[110] Silicon
Cross Section
[110]
Convex corners are

rapidly undercut
[110]
[100]
Cross Section

Adapted from Figure 3.23 in: Senturia, Stephen D. Microsystem Design
Corner Compensation

Figure 3 on p. 143 in: Enoksson, Peter. "New Structure for Corner Compensation in Anisotropic KOH Etching." Journal
of Micromechanics and Microengineering 7, no. 3 (September 1997): 141-144.
A common approach to corner compensation as shown in Enoksson,

J. Micromech. Microeng. 7 (1997), 141-144.
> To etch a convex corner with KOH, add extra material at corner
> Amount of material is chosen so that it will etch away just when
the overall etch reaches the desired depth
> Extra material protects convex corner from attack

Arbitrary shapes
> Any mask feature, if etched long enough, will result in

a V-groove tangent to the mask along <110>
directions
[110]
Boundary of
rectangular pit
Masking Layer
[100]
Undercut
Regions
Cross Section
Figure 3.24 in: Senturia, Stephen D. Microsystem Design.
Misalignment
> Misalignment of the mask relative to the [110]

direction always results in a larger etched region
[110]
Boundary of
Rectangular Pit
50 misalignment

Figure 3.25 in: Senturia, Stephen D. Microsystem Design.
Selectivity and etch masks for KOH etches

> Deep etches are long selectivity matters
> Mask must last long enough to bring the etch to completion
> Sidewall erosion must be at an acceptably slow rate
> Etch rate of {111} planes is finite but small
Condition-dependent, of order 400:1 for {100} rate/{111} rate
> Etch rate of mask
Si:SiO2 selectivity about 100:1
Si:LPCVD SiN selectivity at least 1000:1
PECVD SiN not effective (low quality)
Dont use photoresist!
Etch stops
> When etching into a wafer to leave a specific
thickness of material, it is necessary to have some
kind of etch stop.
> Example: diaphragm pressure sensor

Termination on {111} planes? (V-grooves only)
Prayer and a stop watch?
Usually gives poor thickness control
Chemical etch stop
An unetched material, e.g. oxide or nitride
Heavily boron doped silicon, p+, as etch
Motorola
stop for strong bases

(etches several orders of magnitude
more slowly than lightly doped if
concentration > 5 x 1019 cm3)
Electrochemical etch stop

Dry (Plasma) Etching

> At reduced pressure, a glow discharge is set up in a reactive gas
environment
> This produces

Ions that can be accelerated by the electric fields at the
bounding edges of the plasma so that they strike the surface
these can be quite directional in their impact
Free radicals (uncharged) that can diffuse to the surface and

undergo reaction
> Etching depends on reaction followed by creation of a gaseous

byproduct which is pumped away
Applicability
> Most materials can be plasma etched
Oxide
Nitride
Silicon
Most metals (not the noble metals)
Polymers
> The art is in achieving suitable selectivity both for masking
layers and to layers that lie beneath the layer being etched
Known recipes (gas mixtures, plasma conditions) with

desired selectivity
End-point detection is an important part of best practice

when using plasma etching
Shape
> The higher the pressure, the more isotropic the etch
because reactants are scattered many times before
reaching the surface (this is called plasma etching)
> To achieve directional anisotropy, one must go to low

pressure to achieve long mean-free paths for the ions
(this is called reactive-ion etching or RIE)
> Deep reactive ion etching is another thing altogether
Deep Reactive Ion Etching (The Bosch Process)

> Photoresist mask: selectivity about 50:1
> Oxide mask: selectivity > 100:1
1. Pattern photoresist
2. Reactive ion etch in SF6
3. Deposit passivation (C4F8)

(produces a teflon-like
polymer)
4. Etch and repeat cycle

(directional ions clear passivation
from bottom only)
Figure 1 on p. 265 in: Chen, K.-S., A. A. Ayon, X. Zhang, and S. M. Spearing. "Effect of process parameters on the
surface morphology and mechanical performance of silicon structures after deep reactive ion etching (DRIE)." Journal
of Microelectromechanical Systems 11, no. 3 (2002): 264-275. 2002 IEEE.
Depth depends on features and layout

> Features of different width etch at different rates (recipe
dependent)

Figure 3.28 in: Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers,
2001, p. 70. ISBN: 9780792372462.
Multi-level Etching
> Making multi level etches can be challenging
> For through etches with two different depths, simply etch from
both sides of the wafer, with double-sided alignment
Pattern side 1
Etch side 1
Flip wafer and

pattern side 2
Etch side 2
Multi-level Etching with Nested Masks

> Etching two sets of deep (> about 10 m) features on the same
side of the wafer requires a nested mask
1. Grow oxide mask
5. Etch to first depth
2. Define resist mask
6. Strip resist mask
3. Etch oxide to form mask
7. Using oxide as a mask,

etch to second depth
4. Strip resist; pattern

with new resist mask
8. Strip oxide mask
DRIE with Etch Stop
> SOI substrate

> Buried oxide acts as an etch
stop
> Charging can lead to

footing
+ + +
Outline
> Etching
> Wafer bonding
Fundamentals of Wafer Level Bonding

>
Two separate and distinct steps
The wafers are aligned to each other in a bond aligner with

a possible alignment accuracy of one micron or less
The bond fixture is loaded into a vacuum bond chamber

where the wafers are contacted together
>
Three most prevalent types
Direct or fusion wafer bonding (high temperature, ~ 1000 C)

Anodic or field-assisted bonding, ~ 500 C
Bonding with an intermediate glue layer
Gold (thermocompression), ~ 300 C

Polymer or epoxy layer
Direct Wafer Bonding

Si
Hydrate
surface
Contact and
Anneal
Optional: Thin
top wafer
Si
Spontaneous bonding reduces

surface energy; compensates
some strain energy cost.
Si to Si, Si to oxide, oxide to
oxide.
A high quality Si to Si bond

can have bulk strength.
DRIE and wafer bonding, London et al.
IR visualization of bond formation
Kevin Turner, 2003.

Courtesy of Kevin Turner. Used with permission.
Silicon-on-insulator
> Bonding to oxidized wafers is also possible, leading

to silicon-on-insulator wafers
Si
Hydrate
surface
Si
Contact and
Anneal
device layer
Thin top wafer
buried oxide layer

substrate
Wafer Geometry Impacts Bonding

> Spontaneous wafer bonding reduces surface energy
Two smooth, clean, perfectly flat wafers will bond
spontaneously
> When wafers are not perfectly flat, bonding requires them to
bend
Strain energy increases
> How far will two wafers bond?

Wafers bond until the surface energy reduction equals the
strain energy cost
> Important factors

Wafer thickness
Radius of curvature
Wafer bow innate or from stressed films
Waviness locally greater curvature
Wafer Geometry Impacts Bonding

> Bonding order and strain energy
For given total stack thickness, the strain energy accumulates
fastest for wafers of equal thickness (goes as thickness cubed)
K.T. Turner and S.M. Spearing, J. App. Phys., 92 (12) 2002,

7658-66.
To bond n wafers, add them one at a time

+
+
BAD
GOOD
> Etched features

Shallow etch hinders bonding (less interaction area)
Deep etch aids bonding (less stiffness)
Wafer bonding and yield

> Yield in MEMS can require a whole-wafer outlook, unlike IC
processing
> A micron-scale defect can create a mm- to cm-scale defect

Amplification by wafer stiffness
> Can have a die yield of 100% on individual wafers and not get
any devices if defects outside the die area prevent wafer
bonding
> Cleanliness (particulates, organics) is critical to prevent defects;

organics can outgas on anneal.
> Adjust process to minimize stiffness in bonding

At least one of the wafers should be thin (and therefore relatively
pliable) when going into the bonding process
Anodic bonding
> The mobility of sodium ions in the glass drives anodic bonding
> The wafers are heated to temperatures of about 500C; a positive
voltage (300 V 700 V) applied to the Si repels sodium ions from
the glass surface
> Suspectible to particulates, but less so than direct bonding

> Commonly used as a packaging step
+
_
Silicon
Heater
Heater
Glass
Designing process flows for cleanliness

> If you are planning to do a fusion bond, design your process
flow to prevent exposure of bonding surfaces to junk
Cleanliness is a good idea for anodic bonding, too, but
anodic bonding is less picky
> Some junk washes off easily, but some doesnt

> Example: deep reactive ion etchings passivation layer is
reluctant to come off (ashing helps somewhat but isnt perfect)
> Work around: if possible, start your process by coating your

wafer with a protective layer, like oxide. When you remove it
right before bonding, it carries the junk away with it.
Outline
> Etching
> Wafer bonding
> Surface micromachining refers to the selective removal of
sacrificial layers beneath structural layers to create suspended
structures
> Many materials choices possible

> Structural polysilicon and sacrificial PSG oxide is welldeveloped, fully characterized, and available as a foundry
service
> Why use surface micromachining?

Complex multi-layer structures are possible without the need
for wafer bonding
Structure thickness is controlled by thickness of deposited

film, not by etch
Illustrating surface micromachining
> Example
Structural layer:
polysilicon
Sacrifical layer:
Oxide
Etchant
HF
Top View
Cross Section
Substrate
Deposit
sacrificial layer
design
Deposit
structural layer
design
Remove
sacrificial layer

In-plane processing; potentially out of plane structures.

Courtesy of Sandia National Laboratories, SUMMiT Technologies, www.mems.sandia.gov
Solder assembly of surface micromachined parts

> Build surface micromachined parts
> Place solder over the joint
> Melt solder; surface tension bends part up until it hits limiter
Courtesy of Elsevier, Inc., http://www.sciencedirect.com.

Figure 10 on p. 242 in: Harsh, K. F., V. M. Bright, and
Y. C. Lee. "Solder Self-assembly for Three-dimensional
Microelectromechanical Systems." Sensors and Actuators
A: Physical 77, no. 3 (Nov. 1999): 237-244.

Figure 11 on p. 243 in: Harsh, K. F., V. M. Bright, and
Y. C. Lee. "Solder Self-assembly for Three-dimensional
Microelectromechanical Systems." Sensors and Actuators
A: Physical 77, no. 3 (Nov. 1999): 237-244.
Introduction to MUMPs
Courtesy of MEMSCAP, Inc. Used with permission.
PolyMUMPs (Multi-User MEMS Process) is a three-layer polysilicon

surface micromachining commercial process established in BSAC and
now available from MEMSCAP.
MUMPs Process
Cross section before sacrificial etch

Structural layers: Poly 0, Poly 1, Poly 2
Sacrificial layers: Oxide 1, Oxide 2 (phosphosilicate glass)
LPCVD nitride acts as passivation, electrical isolation layer
MUMPS Process: Step 1
4 100 silicon wafers, 1-2 resistivity

600 nm LPCVD Si3N4
500 nm LPCVD polysilicon (Poly-0)
Lithography poly-0 (Hole 0) and RIE poly-0
Step 2
2.0 m LPCVD PSG (oxide-1) and 1050 C anneal

Lithography Dimples and RIE PSG (750 nm)
Step 3
Lithography Anchor1 and RIE PSG
Step 4
2.0 m LPCVD polysilicon (poly-1)

Lithography Poly-1 (Hole 1) and RIE poly-1
Step 5
0.75 m LPCVD PSG(oxide 2 ) and 1050 C anneal
Step 6
Lithography poly 2-poly1-via and RIE PSG (oxide 2)
Step 7
Lithography Anchor 2 and RIE PSG (oxide-2 and oxide-1)
Step 8
1.5 m LPCVD polysilicon (poly-2)
Step 9
Lithography poly-2 (Hole 2) and RIE poly-2
Step 10
0.5 m Cr/Au evaporation (Metal)

Lithography Metal (HoleM) and lift off
Step 11
Releasing
1.5-2 min 49% HF sacrificial oxide etch at room temp.
CO2 critical point drying
Limitations of Surface MM
Stiction
> Permanent adhesion between movable structures or structure

and substrate
> Caused mainly by van der Waals forces due to hydrogen content
or moisture on surface and close proximity of movable
structures (due to thin films used)
Dealing with stiction

> Thin structures are susceptible
Deposited films
Narrow, deep-etched structures
> Prevention is key
> Options
Low surface tension liquid rinse after sacrificial etch
Surface roughening or surface coating (hydrophobic)
Critical point CO2 or sublimation drying to prevent meniscus
formation
One release recipe

Acetone soak to remove photoresist (30 min)
Isopropanol soak to remove acetone (30 sec)
Rinse in water to remove isopropanol (1 min)
Soak in 49% HF (3:30 min)
Soak in 4:1 methanol/water (9 min)
CO2 supercritical release. Liquid CO2 is used to flush the
methanol. The CO2 is then heated and will subliminate around
35 C.
HF only removes the oxide that it can reach
Holes in plates
are important!
Figure 11 on p. 243 in Harsh, K. F., V. M. Bright, and Y. C. Lee.
"Solder Self-assembly for Three-dimensional Microelectromechanical
Systems." Sensors and Actuators A: Physical 77, no. 3
(November 1999): 237-244.
Outline
> Etching
> Wafer bonding
Concerns in designing a process and mask set

> We know that we have to obey the laws of physics and the
vendors constraints when designing a process flow
> Its easier to design a robust, effective process if youve

designed your device well
Selecting your device architecture wisely if you dont

design an unbuildable structure, you wont have to build it
Designing the package and packaging process during the

device design
One concern: accumulated topography

> Successive steps of lithography and etching or deposition
create non-planar surface topography
> This can interfere with further fabrication:

Getting good coverage with photoresist
Depth of focus of lithographic tool
Wafer bonding
Stringers left over from etching
> Chemical mechanical polishing (CMP) can be used to remove or
reduce unwanted topographic features
> Other techniques available for special cases
Chemical Mechanical Polishing (CMP)

> Often used to planarize interlayer dielectric insulators
> Typical surface roughness less than of 1 nm (but waviness can
be much bigger)
> Combination of mechanical polishing and chemical etching

> Using an abrasive slurry dispersed in an alkaline solution
> High, narrow features polish faster than low, uniform features
Stringers
> Stringers form when a conformal film that covers topographic
features is etched directionally, e.g. with RIE or dilute plasma
Conformal coating
over steps
Stringers remain
after directional
plasma etch.

A Bad Stringer Location

> The material in the cusp is almost impossible to remove by
etching
> Using a thicker white layer and polishing it back with CMP cures
the stringer problem (so part of the cure is dont let it happen in
the first place a good lesson to remember)
Stringer location

System partitioning
> You know that the overall system requires a set of functionalities.
How many of them should you put on the MEMS chip? This will
govern your fab process!
> Prime example: electronics

If you have a tiny signal that you cant detect without amplifying it as
soon as its produced, then you need at least first stage amplifying
electronics
If on-chip electronics are not functionally required, you must choose

whether it will save you money (fewer chips to make and package
together) or cost you money (more ways to ruin your MEMS chip in
the fab, and fewer process options) to include electronics on chip
> Some commercially successful devices are made with on-chip

electronics
> More commonly, they are not

Die separation
> Usually you use single MEMS chips, rather than whole wafers
> When and how to cut the chips apart?
> If youre going to slice them apart with a (very ungentle) die saw,
you must identify where in the process you will do it without
breaking your structures
> One alternate approach: include etch features on your mask

that will separate the dies most of the way so they snap apart at
the end
> Either way, must think about this when creating your process
flow
Design MEMS device and package together!

> You cant make the microfabricated part without photomasks
> The photomasks include interfaces to package:
Electrical bond pads, access required for MEMS function
> Until you make the system partitioning decision, you dont know
the bond-pad requirements.
> Until you design the package, you dont know what the
constraints on the physical access will be
> Therefore, until you make the system partitioning decision and
design the package, you cant make the masks!
> Second order package-device interactions:

High temperature packaging step can affect device: thermal
stresses, outgassing, etc.
Process design philosophy

> People publish their fab accomplishments
> Students read the papers and take home the wrong message
They published X. X must be straightforward, and my process can
probably count on accomplishing even a little more than X.
W, X, Y, and Z have all been demonstrated. I cant count on doing

any better than W, X, Y, and Z, but Im sure I can accomplish all of
them simultaneously.
> Some advice:

Dont design your processes on the hairy edge of impossibility.
Including a very difficult process may be unavoidable, but a) dont
include a lot of them and b) be prepared to put a lot of work into
making that process robust.
On the design projects, we will know if your process is too

ambitious. In your thesis or in your job, Mother Nature will know if
your process is too ambitious.
Process Flow Design Challenge: Thermal Pneumatic Actuator

A thermal pneumatic actuator consists of a sealed cavity bounded on one side by a
deformable, corrugated (wavy) membrane and on the other side by a surface with an
ohmic heater. When the heater is turned on, the air inside the cavity heats up, expands,
and makes the deformable membrane bulge out. Your teams job is to design a
fabrication process flow to create such a device. The cross sections are shown below.
The membrane is 2 m thick, 4 mm on a side, and is patterned out of a 450 m thick
silicon wafer. The cavity depth is 5 m. You may choose the detailed cross-sectional
profile of the corrugations, but they should be concentric circles with a wavelength of
250 m and a depth of 10 m. The heater should be made of gold and patterned on a
Pyrex wafer.
With your team, first sketch out a process flow and mask set; then fill in the details as far
as possible. After each team has a chance to work on the process flow, a team or teams
will be chosen to present their solution(s) to the class. If your team is not chosen to
present its design, your job will be to critique the process that is presented.
Top view of silicon wafer containing

corrugated membrane
Cross section of complete device
Courtesy of Elsevier, Inc., http://www.sciencedirect.com. Used with permission.

Figure 1a) on p. 63 in: Jeong, O. C., and S. S. Yang. "Fabrication of a Thermopneumatic
Microactuator with a Corrugated p+ Silicon Diaphragm." Sensors and Actuators A:
Physical 80, no. 1 (2000): 62-67.
Close up cross-section of silicon wafer containing

corrugated membrane
Figure 2 on p. 63 in: Jeong, O. C., and S. S. Yang. "Fabrication of a Thermopneumatic
Microactuator with a Corrugated p+ Silicon Diaphragm." Sensors and Actuators A: Physical 80,
no. 1 (2000): 62-67.
This problem is based on a real device written up in the following paper; the cross
sections are from the paper as well.
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechanical
Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on
[DD Month YYYY].
Soft Lithography and

Materials Properties in MEMS
Carol Livermore
* With thanks to Steve Senturia and Joel Voldman, from

whose lecture notes some of these materials are
adapted.
Outline
> Soft Lithography

Materials and processes
Patterning biomaterials
> Material Properties in MEMS

Role of material properties in MEMS
Some examples
Determining material properties
SU-8 Epoxy
> Near-UV photosensitive epoxy that acts as a negative resist
> High aspect ratio structures possible (20:1), up to mm
> BUT it is not readily removed like photoresist would be
> Can be a structural material or a mask/mold for other materials
> Multilayer processing for thicker structures or two layer
structures, including enclosed flow paths
> Hydrophobic (adheres best to hydrophobic materials)

> Challenges:
Mechanical stress, cracking
When used as a mask, difficult to remove
Sample process: multilayer SU-8 microfluidics
> Spin coat, prebake, expose, and

postbake first layer
> Spin coat, prebake, expose, and

postbake second layer
> Develop both layers

> Cap with SU-8 coated transparent plate
> Expose to crosslink SU-8 glue, final
bake
Described in Jackman, J. Micromech. & Microeng. 11, 2001, 263.

Multilayer SU-8 options

> Challenge: suspended structures
Unintended exposure of first layer can destroy undercut
> Some demonstrated options:

Deposit metal on top of undeveloped first layer to protect it
from second exposure
Develop first layer channels and fill with a sacrificial polymer

to be removed later
Expose only the upper layer of SU-8 by controlling dose

and/or focal depth
Cracking in SU-8
> SU-8 shrinks in developer, causing cracks and loss of
adhesion
cracks
100 m
SU-8 Removal
> When using SU-8 as a resist and not a structural material, it must
be removed!
Enduring challenge best option is not to strip the SU-8
> Postbaked material extremely chemically resistant

Wet
Piranha, also removes other materials
Nanostrip, but much slower
Organic SU-8 removers
Work by swelling and peeling, not dissolution
Dry
O2 plasma for cleanup, too slow for bulk removal
Sacrificial layer: Omnicoat
Organic layer, O2 plasma patternable
Spin-coat sacrificial layer, process SU-8 on top
Sacrificial layer permits SU-8 liftoff
PDMS
> Polydimethylsiloxane
> Flexible elastomer
> Used to replicate topography from a master (Si, SU-8, etc)
> Used as a conformable stamp for patterning onto other surfaces
> Good for sealing microfluidic devices; can be sealed to many
materials
> Can be spin-coated

> Possible to dry etch
> Low cost pattern replication
PDMS
CH3
CH3
CH3
O
Si
CH3
Si
CH3
CH3
Si
> Upon treatment in oxygen

CH3
n CH3
plasma, PDMS seals to itself,

glass, silicon, silicon nitride,
and some plastic materials.
Plasma oxidation
Air (~ 10 min)
contact PDMS
surfaces
irreversible seal:
formation of
covalent bonds
Courtesy of Hang Lu and Rebecca Jackman. Used with permission.

Pattern replication by molding

> Master can be silicon, SU-8,
Master
apply
photoresist, or elastomer
> Apply prepolymer liquid to master

> Cure (by baking) and peel off
Prepolymer
Master
cure and peel
> When molding with PDMS, can

exploit PDMS sealing to form
enclosed microchannels
Master
Sample PDMS Process

Unexposed
SU-8 (50 m)
Si
Surface treatment &

casting PDMS
photolithography
PDMS
UV light
Si
mask
removing elastomer from

master
Si
PDMS
development
seal against glass after plasma

treatment and insert tubing
Si
master
tubing
Courtesy of Hang Lu. Used with permission.
Microcontact printing
Stamp
> Apply ink to an elastomer stamp

> Bring stamp in contact with surface
(need not be flat because stamp can
deform)
Stamp
> Ink transfers to surface of substrate

> Can be used to mask etches,
depositions, etc
> Inks:
Stamp
SAMs (alkanethiols for coating on

noble metals), organics, proteins,
etc.
> Size scale: submicron to

millimeters
Microcontact printing on non-planar surfaces
Figure 2 on p. 186 in Rogers, J. A., R. J. Jackman, and G. M. Whitesides. "Constructing Single- and
Multiple-helical Microcoils and Characterizing Their Performance as Components of Microinductors
and Microelectromagnets." Journal of Microelectromechanical Systems 6, no. 3 (Sept. 1997): 184-192.
1997 IEEE.
Pattern replication by imprinting

Master
deformable
> Imprint in a thermoplastic material

by heating and applying pressure
> Typical material: PMMA

(polymethyl-methacrylate)
Master
> Or imprint in UV-curable fluid, like

polyurethane
> Process usually leaves trace

Master
material in clear areas, which may

be removed by dry etch
> Can replicate nanoscale features

(nanoimprinting, S. Chou)
Parylene
> A vapor-deposited polymer that provides very conformal
coatings
> Thickness range: submicron to about 75 microns

> Chemically resistant, relatively inpermeable
Component encapsulation
> Low friction film can act as a dry lubricant

> Low-defect dielectric insulating layer
> Relatively biologically inert
> Thin films nearly transparent
> Low stress minimizes interaction with components
> Can be dry etched
Some other useful materials

> Polyurethane
UV-curable is particularly useful, can be used for molding
process
> Special purpose polymers (not all terribly mainstream, but an

active and useful area of development):
Photodefinable polycarbonate that decomposes at 250C (ie

Unity Sacrificial Polymer)
Avatrel, a polymer material through which decomposed Unity

can diffuse
Outline
> Soft Lithography


Some examples
Biomaterials processing
> Biomolecules DNA, proteins, cells

Needed for biosensors, cell arrays, etc.
Challenge is integrating fragile molecules with
semiconductor processing
Multiple methods
Micro-contact printing
Microfluidic patterning
Lift-off
Stencils
Biomaterials processing by microcontact printing

> Microcontact printing
Developed by Whitesides at
Harvard Univ.
Xia & Whitesides, Annu.

Rev. Mater. Sci. 28:153,
1998.
No harsh solvents needed

Figure 4 on p. 601 in: Blawas, A. S., and W. M. Reichert. "Protein Patterning.
Biomaterials 19, nos. 7-9 (April 1998): 595-609.
Biomaterials processing by microfluidic patterning
> Microfluidic patterning
Attach PDMS stamp to substrate

Add biomolecule to solution
Flow solution through channels
Attachment via adsorption
Courtesy of Annual Reviews. Used with permission.

Figure 1c) on p. 230: in Folch, A., and M. Toner. "Microengineering
of Cellular Interactions." Annual Review of Biomedical Engineering
2 (August 2000): 227-256.
Biomaterials processing by liftoff
> Lift-off
Use standard photoresist lift-off
Molecules must withstand acetone
or other solvent

Figure 1 on p. 597 in: Blawas, A. S., and W. M. Reichert. "Protein Patterning.
Biomaterials 19, nos. 7-9 (April 1998): 595-609.
Biomaterials processing by stencils
> Stencils
Use PDMS stamps as dry resists

Physically pattern biomaterials
Can use with most any substrate
Potential damage to cells
on feature periphery
PDMS stencil
A Folch, Univ. Washington
a)
b)
c)
d)

Figure 3 on p. 351 in: Folch, A., B. -H. Jo, O. Hurtado, D. J.
Beebe, M. Toner. "Microfabricated Elastomeric Stencils for
Micropatterning Cell Cultures." Journal of Biomedical Materials
Research 52, no. 2 (2000): 346-353.
Attach stencil to substrate

Add cells
Let cells attach to substrate
Remove stencil
Courtesy of Albert Folch. Used with permission.
Outline
> Soft Lithography


Some examples
Material properties and coupled domains
> The basic functionality of many MEMS devices is in

coupling one domain to another, and this coupling is
typically described by material properties
Mechanical to electrical
Electrical to thermal
Thermal to fluids
> The failure modes of many MEMS devices are in

coupling one domain to another
For example, package stress interacting with piezoresistor
> Some properties of MEMS materials are exceptional

Pronounced piezoresistivity of silicon
Off-label use of materials
> MEMS devices often depend on material properties

that are less important for other uses of the material
> Fracture strength of Si

> Biocompatibility of Si-based materials
Process dependence
> Some microelectronic materials, like single crystal silicon, have
highly predictable and repeatable constitutive properties
> Most microelectronic materials, however, exhibit some degree of

process dependence in their material properties, especially
deposited or thermally formed thin films
> Some properties, like thin-film residual stress, can be wildly

dependent on deposition conditions, even changing sign from
compressive to tensile
Planning around material properties

> Well-controlled material properties can pose a design challenge
Many factors, may point design in opposite directions
> Poorly-controlled material properties are worse
Every device has specifications, which must be met by either
getting it right the first time or employing some combination
of trim (fixing the hardware later) and calibration
(compensating through software)
Getting it right includes both materials and geometry

The greater the variation in properties, the greater the
headache and the greater the cost
Constitutive Properties
> Constitutive properties are normally expressed as relationships
between applied loads (causes) and resulting responses
(effects).
> It is not always clear which is which. If the functional relation

can be inverted, it doesnt actually matter.
Examples of constitutive properties

Load
Response
Property
Distributed mechanical
force
Acceleration
Mass density
Temperature rise
Increase in internal energy

per unit volume
Specific heat per unit

volume
E-field
D-field
Dielectric permittivity
H-field
B-field
Magnetic permeability
Stress
Strain
Elastic compliance
E-field
Current density
Electrical conductivity
Temperature gradient
Heat flux
Thermal conductivity
Shear stress
Shear rate
Inverse of viscosity
Scalar properties
> Scalar properties are those which involve no

orientation-dependent or direction-dependent effects
> Examples:
Mass density
Specific heat
Viscosity of a gas or unoriented liquid (i. e. not a liquid
crystal)
And, for isotropic materials: permittivity, permeability,

electrical and thermal conductivity
Tensor properties
> Properties that involve directions, either the relative directions of
applied vector loads and vector responses, or the orientation of
loads and/or responses relative to internal (crystalline) axes,
require tensors for their specification
> Examples:
Permittivity, permeability, index of refraction and conductivity
of non-cubic solids are second-rank tensors

Piezoelectric responses which couple stress and strain to
electric fields are described with third-rank tensors
Elastic constants and piezoresistive responses require fourthrank tensors
> Because constitutive properties have lots of symmetry, we can

usually boil these higher-order tensors down to something
manageable.
Outline
> Soft Lithography


Some examples
Which properties to consider?

> One of the challenges for the MEMS designer is knowing what
you should care about.
> One course goal is acquiring domain knowledge, which gives

you some insight into which material properties are important in
a given situation.
> Today: a sneak preview of what you might worry about

> A useful resource:
A previous years assignment involved looking up material
properties for many MEMS materials.
Results are posted on the web site.

Buyer beware! This is the work of 20 + students and surely
contains some errors.
Electrical Conductivity
> Relates electric field to current density; Ohms law minus the
geometry.
> Importance: circuits, sensors

> Elemental metals: observed values can vary from tabulated bulk
values, depending on deposition method.
Example: a factor of two variation in metals printed as

nanoparticles in a solvent and then annealed.*
> Other materials: variations much more pronounced

Polysilicon conductivity depends on grain size, doping:
orders of magnitude with relatively small process variations.
> Characterize by four-point measurement of test structures with

known geometry.
* Fuller et al, JMEMS, vol. 11, p. 54 (2002).
Breakdown Strength
> The maximum electric field that an insulating material in the gap
between two flat electrodes can withstand without suffering
dielectric breakdown
> Depends on the size of the interelectrode gap

> Importance: high voltage actuators, maximum performance
> Can vary with film composition, defect density
> Tabulated values are sufficient if the design does not push the
limits; otherwise characterization is necessary.
> One test approach: fabricate a capacitor and fill gap with
material in question. Match gap thickness to actual device.
Detect onset of breakdown both optically and by current flow.
Elastic constants
> Relate stress (how hard you push or pull on something in a
given direction, per unit area) to strain (the resulting fractional
change in the objects length or shape)
> Importance: how far will your cantilever or membrane deflect?

> Single crystal materials: good reproducibility
> Deposited materials: must be characterized
> Characterizing elastic constants in MEMS is a significant
challenge. Stay tuned for some sample approaches at the end.
Residual film stress

> Stress in a film deposited on a Si wafer, in the absence of
external loading.
> Two flavors:

Intrinsic stress: related to structure
Thermal stress: accumulated from a change in temperature
> Residual stress is a VERY VARIABLE PROPERTY, and must be

measured.
> Can play games, such as adjusting deposition conditions to

ensure that intrinsic and thermal stresses cancel out at given T
> One basic characterization: wafer curvature in the presence of a

blanket film
> More advanced approaches later in lecture

Known problems from residual stress

> Deformation of free, moveable structures
Compressive film on a wafer
Wafer bows
Compressive membrane or beam
Buckling
Operational deflection can be enhanced
Tensile film on a wafer
Wafer bows
Tensile membrane or beam
Operational deflection can be reduced
> Film cracking and delamination
Coefficient of Thermal Expansion (CTE)

> Fractional change in length per unit temperature change
> Importance: CTE mismatch plus high T processing or operation
creates stress (and deformation and/or destruction)
> Examples:
Bonding glass (quartz or Pyrex) to Si
Thermal stress in a film that is deposited at high T
> CTE is tabulated, and one of the less variable material

properties.
Piezoresistance
> Straining the silicon lattice shifts the band-edge energies of the
conduction band and the valence band
The conduction band, in particular, consists of multiple
minima and, depending on the direction of the strain, some
go up more than others
This shifts the relative electron population in some minima
compared to others and modifies scattering rates
> Result: an orientation-dependent shift in the conductivity

> For historical reasons, the piezoresistive coefficients relate
resistivity to stress.
> Piezoresistivity is good for strain sensors, and sometimes

changes the resistance of otherwise perfectly good resistors.
> Coefficients are material, orientation, temperature, and doping

dependent
Analytic Formulation of Piezoresistance
> The electric field E is a vector (first-rank tensor)

> The current density J is a vector (first-rank tensor)
> Therefore, the resistivity e (and/or conductivity) is a
second-rank tensor, as is the stress
> The piezoresistive effect is described by a fourth-rank

tensor
E = [e + ] J
The Details
> Good news: in cubic materials, there are just three independent
non-zero piezoresistance coefficients, 11, 12, 44.
Effect of stress along
direction of E
Effect of stress
perpendicular to E
Ohms law
Effect of
shear stress
E1
= [1 + 11 1 + 12 ( 2 + 3 )]J1 + 44 ( 12 J 2 + 13 J 3 )
E2
= [1 + 11 2 + 12 ( 1 + 3 )]J 2 + 44 ( 12 J1 + 23 J 3 )
E3
= [1 + 11 3 + 12 ( 2 + 1 )]J 3 + 44 ( 13 J1 + 23 J 2 )
where
e 11 = 1111 , e 12 = 1122 , e 44 = 2 2323

Piezoresistivity in Silicon
> Coefficients depend on doping, and decrease rapidly above
about 1019 cm-3
> Coefficients are functions of temperature

> Typical values
Type
Units
n-type
p-type
Resistivity
-cm
11.7
7.8
11
10-11 Pa-1
-102.2
6.6
12
10-11 Pa-1
53.4
-1.1
44
10-11 Pa-1
-13.6
138.1
Piezoelectricity
> Ionic crystals that lack a center of inversion symmetry can
exhibit a net polarization within each unit cell. Materials which
have this at zero strain are called ferroelectrics. Materials in
which the dipole results from strain are called piezoelectrics
> Examples
Quartz
Zinc oxide
Lithium niobate
Lead zirconate-titanate (PZT)
Aluminum nitride
poly (vinylidene fluoride)
Piezoelectricity
> Piezoelectrics can have a net electric polarization which
interacts with mechanical strain
Applying a voltage across a piezoelectric creates strains

both parallel and perpendicular to the applied electric field
(actuation)
Straining a piezoelectric creates an electric field both parallel

and perpendicular to the imposed strain (sensing)
> Interaction between stored mechanical energy and stored

electrostatic energy
Permits both sensing and actuation

Unlike piezoresistivity, which is purely dissipative (no
actuation)
Piezoelectric coefficients
> The piezoelectric tensor links the stress tensor to the electric
polarization tensor
> There are four variables: stress, strain, electric field, and
electric displacement
> Simplification by symmetry: the behavior can be captured by

either the d coefficients (when stress and electric field are the
independent variables) or by the e coefficients (when strain
and electric field are the independent variables)
> The d coefficients (units C/N) relate strain to electric field

Consider an electric field in the z direction
Strain in the x or y direction = d31*Electric field in the z direction
Strain in the z direction = d33*Electric field in the z direction
Piezoelectric materials
> Some piezoelectrics have no net polarization at zero strain
Quartz
Zinc oxide
Quartz and ZnO are stiff materials, with low strain
> Some materials have a net polarization at zero strain
Common example: PZT (Lead zirconate titanate)

Higher strain ceramic material for larger deflections
Common choice for MEMS devices
Fabrication: deposited in layers by a sol-gel (spin on) process,
thermally cured, then poled in an electric field
> Piezoelectric polymers

Example: poly (vinylidene fluoride)
Very high strains
Examples of piezoelectric coefficients

> Zinc oxide
d31 = - 5.4 pC/N
d33 = 11.7 pC/N
Note: expansion along the field corresponds to contraction
in the transverse direction
> PZT
Coefficients depend on exact PZT material, on underlying
material, on frequency, and on electric field
- d31 is in the ballpark of 100 pC/N to several hundred pC/N

d33 is in the ballpark of several hundred pC/N to 1000 pC/N
Again, expansion along the field corresponds to contraction
in the transverse direction
Variation over temperature

> Electrical conductivity
> Thermal conductivity
> Specific heat
> Coefficient of thermal expansion (CTE)
Variations in the coefficient itself
> Piezoresistive coefficients
> Piezoelectric response (pyroelectricity)
> Youngs modulus
Outline
> Soft Lithography


Some examples
The bottom line

> You need to use test structures to characterize materials with
variable properties
> Measuring electrical properties requires electrical test devices

> Measuring mechanical properties requires a mechanical test
device
Basically a MEMS device designed to help you measure a

particular property value or values
Fabrication of test device must accurately reflect fabrication

processes that you will use to create your real device
> Measuring material properties requires as much careful design

and fabrication as is required to create a device
Whats Hard?
> Knowing the geometry of the test device well enough to be able
to make good measurements of the constitutive properties and
their repeatability
Lateral and vertical dimensions not too difficult

Sidewall angles, mask undercut are harder
> Modeling the test device accurately enough so that the accuracy
of the extracted properties are limited by geometric errors
A good procedure is to make a family of test structures with

a systematic variation in geometry, and extract constitutive
properties from all of the data.
> Preventing systematic errors in the measurement
Membrane Load-Deflection Example

> The problem: residual stress and stiffness of membranes affect
their deflection under load
> Example: pressure sensors

d
2a
Pressure (p)
> Approach:
Apply different pressures and measure resulting deflections
Fit to an energy-based model for large membrane deflections
Ideal rigid boundary conditions are a benefit
> Weakness:
Deflection is very sensitive to variations in membrane
thickness and edge length, so metrology errors appear
How to get a family of test geometries?

Pull-in of beams: M-Test Example

> The problem: elastic constants and residual stress affect the
actuation of mechanical MEMS structures
> Approach:
Fabricate an array of microbeams of different lengths
Measure the voltage at which they pull in, and fit to models
Excellent agreement with known values when boundary
Gap Between Plates
conditions are ideal

g
2
g
3
Stable Region
Unstable
Region

Figure 10 on page 116 in: Osterberg, P. M., and S. D. Senturia. "M-TEST:
A Test Chip for MEMS Material Property Measurement Using
Electrostatically Actuated Test Structures." Journal of
Microelectromechanical Systems 6, no. 2 (1997): 107-118.
Pull-In
Instability
Applied Voltage
VPI
> Weakness: surface micromachined beams often have some

support compliance
Electronics A
Joel Voldman
JV: 2.372J/6.777J Spring 2007, Lecture 6E - 1
Outline
> Elements of circuit analysis
TODAY
> Elements of semiconductor physics

Semiconductor diodes and resistors
The MOSFET and the MOSFET inverter/amplifier
> Op-amps
Elements of circuit analysis
> There are many ways to analyze circuits

> Here well go over a few of them
Elements laws, connection laws and KVL/KCL

Nodal analysis
Intuitive approaches
Superposition
Lumped elements in circuits
> Circuit elements (R, L,
closed
surface
C) are lumped
approximations of
complex devices
EA = Q
> The electrical capacitor

between Q and V?
E(r, t ) = V (r, t )
b
V (b) V (a) = V = Eg
A
+
V
-
E = 0
V (b) V (a ) = E dl
What is the relation
E da = Q
Q = AV
g
A
C=
g
E =V
V = CV
Lumped elements in circuits
> The electrical capacitor

We can replace all of field theory with terminal relations
And introduce an element with an element law
As long as capacitor size << wavelength of electrical signal
In general, MEMS are small
e.g., =50 m 600 GHz
Q = CV
dQ d
I=
= (CV )
dt dt
dV
I =C
dt
Elements and element laws

> Do this with all three basic
elements
> Resistor
> Capacitor
> Inductor
V
V = RI
V
dV
I =C
dt
dI
V =L
dt
Source elements
> We need elements to provide energy into the circuit

> Two common ones are voltage source and current
source
I
+
+
V V0
V = V0
I
+
V I0
I = I0
KVL and KCL
> These are continuity laws that allow us to solve

circuits
> Kirchhoffs voltage law

The oriented sum of voltages around a loop is zero
> Kirchhoffs voltage law

The sum of currents entering a node is zero
Complex impedances
> For LTI systems, use

complex impedances
instead
Implicitly working in
1
V = IZ C = I
Cs
-
frequency domain
> Much easier circuit
+
I
analysis
the same, like

resistors
> All elements treated
V
Z
V = IZ L = ILs
V = IZ R = IR
V (s) = I ( s) Z ( s)
Lets analyze a circuit
1. Figure out what are

you trying to determine
+
V0
-
2. Replace elements with

3. Assign across and
4. Use KVL
5. Substitute in element
laws
6. Solve
L
R
complex impedances
through variables
iV
i C + VC -
+
VV V0+
-
ZC
ZL
ZR
- VR +
VL
+
iL
iR


complex impedances
iV
i C + VC -
+
VV V0+
-
ZL
ZR
- VR +

through variables
ZC
VL
+
iL
iR
VV VC + VL VR = 0
4. Use KVL
V0 iC Z C + iL Z L iR Z R = 0
iC = iL = iR
laws
6. Solve
iV
+
VV V0+
-

complex impedances

through variables
4. Use KVL
laws
i C + VC ZC
ZL
ZR
- VR +
VL
+
iL
iR
V0 iR Z C iR Z L iR Z R = 0
iR =
V0
=
ZC + Z L + Z R 1
V0
Cs
+ Ls + R
Cs
iR =
V0
2
LCs + RCs + 1
6. Solve
Example #1
> Solve for VL

+
V0
-
C
L
R
VL
-
Nodal analysis
> Element law approach becomes tedious for circuits

with multiple loops
> Nodal analysis is a KCL-based approach
Nodal analysis
1. Figure out what are you
VC
trying to determine

complex impedances
V0
+
-
3. Assign node voltages &

ground node
4. Write KCL at each node

5. Solve for node voltages
6. Use node voltages to
v2
v1
V0
+
-
C
R
find what you care about

Nodal analysis

trying to determine

complex impedances
i1
v1
V0
i2
i3
+
-
v2
L

ground node

Node 1:
Node 2:
v1 = V0
i1 + i2 + i3 = 0
v1 v2 0 v2 0 v2
+
+
=0
ZC
ZL
ZR

Nodal analysis

trying to determine

complex impedances
i1
v1
V0
+
-
i2
v2
i3
C
R
1
1
1 V0
v2
+
+
=
ZC Z L Z R ZC
v2 ( Z L Z R + Z C Z R + Z L Z C ) = V0 Z L Z R
ground node

ZLZR
v2 = V0
Z L Z R + ZC Z R + Z L ZC

Nodal analysis

trying to determine

complex impedances
i1
v1
V0
+
-
i2
v2
i3

ground node
6.
LRs + 1
R + Ls 1
Cs
LRCs 2
Solve for node voltages v2 = V0 LRCs 2 + Ls + R

5.
v2 = V0
LRs
Cs
2
LRCs
Use node voltages to
VC = v1 v2 = V0 V0
2
LRCs
+ Ls + R
Example #2
R1
+
-
V0
R2
L
VL
Intuitive methods
> Instead of solving the circuit using
equations, use series/parallel tricks to
analyze the circuit by inspection
> Current divider & impedances in

parallel
Both elements have SAME voltage

Terminals connected together
+
V
-
i
i1
i2
Z1
Z2
i
Z
V
-
Z2
i1 = i
Z1 + Z 2
Z1
i2 = i
Z1 + Z 2
Z1Z 2
V = i1Z1 = i
Z1 + Z 2
Z1Z 2
= Z1 // Z 2
Z=
Z1 + Z 2
1 1
1
= +
Z Z1 Z 2
Intuitive methods
> Voltage divider & impedances in
series
Both elements have SAME current
i
+
V1
+
V2
-
Z2
V2 = V
Z1 + Z 2
Z1
Z2
V1
V
=
=i
i1 =
Z1 Z1 + Z 2
Z = Z1 + Z 2
i
V
Z1
V1 = V
Z1 + Z 2
Intuitive methods
> Examples of elements NOT in series OR parallel
Z1
Z2
Z1
Z3
Z4
Z3
Z1 and Z3 in series
Z2 and Z4 in series
Z1 and Z2 NOT in parallel
Z3 and Z4 NOT in parallel
Z4
Z3 and Z4 in parallel
Z1 and Z3 NOT in series
Intuitive methods
iL
> Lets use this approach to solve

a circuit
V0
+
-
C
L

trying to determine
+

complex impedances
V0
+
-
C
Za
4. Re-expand to find signal of

interest
Z a = Z R // Z L
3. Collapse circuit in terms of

series/parallel relations till
circuit is trivial
Va
Za
Va = V0
Z a + ZC
Va
iL =
ZL
Intuitive methods
> Lets use this approach
to solve a circuit
Za
Z R // Z L
1
1
iL = V0
= V0
Z a + ZC Z L
Z R // Z L + Z C Z L

trying to determine

complex impedances
3. Collapse circuit in terms of

series/parallel relations till
circuit is trivial
4. Re-expand to find signal of

interest
ZRZL
ZR + ZL
1
= V0
ZRZL
+ ZC Z L
ZR + ZL
= V0
= V0
ZRZL
1
Z R Z L + ( Z R + Z L ) ZC Z L
RLs
RLs + ( R + Ls ) 1
Cs
1
Ls
RCs
iL = V0
RLCs 2 + Ls + R
Example #3
R1
+
-
V0
R2
VR
Superposition
> These equivalent circuits are linear
and obey the principles of
superposition
This can be useful
> For circuits with multiple sources,

Turn off all independent sources
except one
Solve circuit
Repeat for all sources, then add
responses
V0
+
-
short
> Turning off a voltage source gives

a short circuit
> Turning off a current source gives
I0
open
an open circuit
Superposition
v2
v1
> For circuits with multiple

sources,
Turn off all independent
I0
sources except one
Solve circuit
v2 = I 0 Z R // Z C
responses
v2
v1
V0
+
-
V0
C
R
Find v2
I0
v2
v1
+
-
C
R
v2 = V0
ZR
Z R + ZC
Superposition
> For circuits with multiple

sources,
Turn off all independent
sources except one
ZR
v2 = I 0 Z R // Z C + V0
Z R + ZC
Z Z
ZR
Solve circuit
v2 = I 0 R C + V0
Z R + ZC
Z R + ZC
responses
v2
v1
V0
+
-
C
R
I0
v2 =
I0 R 1
Cs
R+ 1
+ V0 R
Cs
I 0 R + V0 RCs
v2 =
RCs + 1
Find v2
Conclusions
> There are many ways to analyze equivalent circuits

> Use the simplest method at hand
> Element laws & connection laws are OK for simple
ckts
> Nodal analysis works for most any circuit, but will be
tedious for complicated circuits
> Try to use intuitive approaches whenever possible
Elasticity
(and other useful things to know)
Carol Livermore
* With thanks to Steve Senturia, from whose lecture notes some

of these materials are adapted.
C. Livermore: 6.777J/2.372J Spring 2007, Lecture 6 - 1
Outline
> Overview
> Some definitions
Stress
Strain
> Isotropic materials
Constitutive equations of linear elasticity
Plane stress
Thin films: residual and thermal stress
> A few important things
Storing elastic energy
Linear elasticity in anisotropic materials
Behavior at large strains
> Using this to find the stiffness of structures
Why we care about mechanics

> Mechanics makes up half of the Ms in MEMS!
DLP projection display
www.dlp.com
Pressure (p)
Pressure sensors
Images removed due to copyright restrictions. Figure 11 on p.
342 in: Zavracky, P. M., N. E. McGruer, R. H. Morrison, and D.
Potter. "Microswitches and Microrelays with a View Toward Microwave
Applications." International Journal of RF and Microwave Comput-Aided
Engineering 9, no. 4 (1999): 338-347.
1 m
Silicon
0.5 m
Cantilever
Pull-down
electrode
Veeco.com
Anchor

Adapted from Rebeiz, Gabriel M. RF MEMS: Theory, Design, and Technology.
Hoboken, NJ: John Wiley, 2003. I SBN: 9780471201694.
AFM cantilevers
Courtesy of Veeco Instruments, Inc. Used
with permission.
et al., Int. J. RF Microwave

Switches Zavracky
CAE, 9:338, 1999, via Rebeiz RF MEMS
What do we need to calculate?

> Eager beaver suggestion: everything
When I apply forces to this structure, it bends.
Heres the function that describes its deformed shape at
every point on the structure when the deformations are

small.
Here are numerical calculations of the shape at every point
on the structure when the deformations are large.
The structure is stressed, and the stress at every point in the
structure is
> Shortcut suggestion: just what we really need to know

When I apply a force F to the structure, how far does the point of
interest (the end, the middle, etc) move?

This boils down to a stiffness, as in F = kx
What is the stress at a particular point of interest (like where my
sensors are, or at the point of maximum stress)?
How much load can I apply without breaking the structure?
Why things have stiffness I

Unloaded beam is undeformed:
Axially loaded beam is stretched:
Stretching costs energy, which is stored as elastic energy.

Exactly how much energy is determined by material and
geometry.
Why things have stiffness II

Unloaded beam is undeformed:
Loaded beam is bent:

Stretched
Compressed
Stretching and compressing cost energy, which is stored in

elastic energy. Exactly how much energy is determined by
material and geometry.
Example: relating load to displacement in bending

> What are the loads, and where on the structure are they applied?
1 m
Silicon
0.5 m
Cantilever
Pull-down
electrode
Anchor

Adapted from Rebeiz, Gabriel M. RF MEMS: Theory, Design, and Technology.
Hoboken, NJ: John Wiley, 2003. ISBN: 9780471201694.
> Given the loads, what is going on at point (x,y,z)?

M
> How much curvature does that bending moment create in the
structure at a given point?
What is the geometry of the structure?

What is it made of, and how does the material respond to the
kind of load in question?
Elasticity
> Elasticity: the ability of a body to deform in response to applied
forces, and to recover its original shape when the forces are
removed
> Contrast with plasticity, which describes permanent deformation

under load
> Elasticity is described in terms of differential volume elements to

which distributed forces are applied
> Of course, all real structural elements have finite dimensions

> We will ultimately use partial differential equations to relate
applied loads and deformations
Outline
> Overview
> Some definitions
Stress
Strain
Plane stress
Stress
> Stress is force per unit area

> Normal stress
x, y, or z
> Compressive: < 0

> Tensile: > 0
x
z
x
xy
> Shear stress

xy, xz, or yz
x
z
xy
Stress
> Can have all components

at a given point in space
> SI Units: the Pascal

1 Pascal = 1 N/m2
z
zy
zx
> Other units:
yz
xz
1 atm = 14 psi = 100 kPa

1 dyne/cm2 = 0.1 Pa
z
xy
yx
y
x
> Notation: face,direction
x
Adapted from Senturia, Stephen D. Microsystem Design. Boston, MA:
Kluwer Academic Publishers, 2001. ISBN: 9780792372462.
Deformation
> Illustrating a combination of translation, rotation, and

deformation
x3, y3
x2, y2
x3, y3
x2, y2
x4, y4
u(x1, y1)
x1, y1
x4, y4
x1, y1
Adapted from Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers,
2001. ISBN: 9780792372462.
Strain
> Strain is a continuum description of deformation.
> Center of mass translation and rigid rotation are NOT strains
> Strain is expressed in terms of the displacements of each point
in a differential volume, u(x) where u is the displacement and x is
the original coordinate
> Deformation is present only when certain derivatives of these

displacements u are nonzero
Normal Strains (x, y, z)

> Something changes length
xf
> Normal strain is fractional

change in length (dimensionless)
> > 0: gets longer, < 0: gets

shorter
yi
yf
xi
x x+x
Initial length : ( x + x) x = x
Final length : ( x + x + u x ( x + x) ) ( x + u x ( x) ) =
= x + u x ( x + x) u x ( x)
x =
x+ux(x)
u x ( x + x) u x ( x) u x
=
x
x
x+x + ux(x+x)
Shear Strains (xy, xz , yz)

> Angles change
ux
> Comes from shear stresses

> Quantified as change in angle
in radians
2
y
x
xy
uy
u x u y u x u y
=
+
=
+
x
x y
y
Different regimes
> How are stress and strain related? It depends on the regime in
which youre operating.
> Linear vs nonlinear

Linear: strain is proportional to stress
Most things start out linear
> Elastic vs. plastic
Elastic: deformation is recovered when the load is removed
Plastic: some deformation remains when unloaded
> Isotropic vs. anisotropic
Life is simpler when properties are the same in all directions;
however, anisotropic silicon is a part of life
Outline
> Overview
> Some definitions
Stress
Strain
Plane stress
Linear Elasticity in Isotropic Materials
> Youngs modulus, E

The ratio of axial stress to axial strain, under uniaxial loading
Typical units in solids: GPa = 109 Pa
Typical values 100 GPa in solids, less in polymers
x
L + L
x = E x
x = L L
(for uniaxial loading)
Linear Elasticity in Isotropic Materials
> Poisson ratio,

Some things get narrower in the transverse direction when
you extend them axially.
Some things get wider in the transverse direction when you

compress them axially.
This is described by the Poisson ratio: the negative ratio of

transverse strain to axial strain
Poisson ratio is in the range 0.1 0.5 (material dependent)
y = x
Poissons ratio relates to volume change

> Volume change is
z
proportional to (1-2)
> As Poisson ratio
approaches , volume
change goes to zero
We call such materials
z (1 x )
incompressible
> Example of incompressible

material:
Rubber
y (1 x )
x(1 + x )
V = xyz (1 + x )(1 x ) xyz

2
V = xyz (1 2 ) x
Isotropic Linear Elasticity

> For a general case of loading, the constitutive relationships
between stress and elastic strain are as follows
> 6 equations, one for each normal stress and shear stress
xy
1
x = x ( y + z )
E
1
y = y ( z + x )
E
1
z = z ( x + y )
E
Shear modulus G is given by
yz
zx
G=
1
= xy
G
1
= yz
G
1
= zx
G
E
2(1 + )
Other Elastic Constants
> Other elastic constants in

isotropic materials can
always be expressed in
terms of the Youngs
modulus and Poisson
ratio
Shear modulus G
Bulk modulus (inverse of
compressibility)
E
K=
3(1 2 )
Plane stress
> Special case: when all stresses are confined to a single plane
Often seen in thin films on substrates (will discuss origin of
these stresses shortly)
> Zero normal stress in z direction (z = 0)

> No constraint on normal strain in z, z
1
1
x = ( x ( y + z )) = ( x y )
E
E
1
1
(
)
y = y ( x + z ) = ( y x )
E
E
1
( x + y )
z = ( z ( x + y )) =
E
E
often get insight

about these from
boundary conditions
Plane stress: directional dependence

> Principal axes: those
x
directions in which the

load appears to be
entirely normal stresses
(no shear)
Here, principal axes are in x and y.
> In general, there are shear

stresses in other
directions
Stresses on Inclined Sections
> Can resolve axial forces into normal and shear forces
on a tilted plane
FN = F cos
FV = F sin
FN
FV
FV
FN

Adapted from Figure 9.3 in: Senturia, StephenD. Microsystem Design. Boston, MA:
A
Area =
cos
F
= cos2
A
F
= cos sin
A
Resultant stresses vary with angle
Normalized stress
1.2
0.8
0.4
0
-0.4
-0.5
-0.25
0.25
0.5
/
Failure in shear occurs at an angle of 45 degrees

Special case: biaxial stress

> A special case of plane stress
Stresses x and y along principal axes are equal
Strains x and y along principal axes are equal
> Leads to definition of biaxial modulus
1
( x y )
E
1
y = ( y x )
E
x =
1
(1 )
E
E
=
(1 )
Biaxial modulus =
E
(1 )
Thin Film Stress

> A thin film on a substrate can have residual stress
Intrinsic stress
Thermal stress
> Mostly well-described as a plane stress
Thin film
Plane stress region
Edge
region
Substrate

Adapted from Figure 8.5 in: Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer
Academic Publishers, 2001, p. 190. ISBN: 9780792372462.
Types of strain
> What we have just talked about is elastic strain

Strains caused by loading; returns to undeformed
configuration when load is removed
Described by the isotropic equations of linear elasticity
> There are other kinds of strain as well

Thermal strain, which is related to thermal expansion
Plastic strain: if you stretch something too far, it doesnt
return to its undeformed configuration when the load is
removed (permanent component)
Total strain: the sum of all strains
Thermal expansion
> Thermal expansion: if you
change an objects temperature,
its length changes
> This is a thermally-induced strain

> An unopposed thermal
expansion produces a strain, but
not a stress
> If you oppose the thermal

expansion, there will be a stress
> Coefficient of thermal expansion,
xthermal (T ) = T T
x (T ) = x (T0 ) + T (T T0 )
and
V
= 3 T (T T0 )
V
T
Thermally Induced Residual Stress

> If a thin film is adhered to a substrate, mismatch of thermal
expansion coefficient between film and substrate can lead to
stresses in the film (and, to a lesser degree, stresses in the
substrate)
> The stresses also set up bending moments

You care about this if you dont want your wafer to curl up
like a saucer or potato chip
> And the vertical expansion of the film is also modified
Thermally Induced Residual Stress

Substrate:
s = T , s T
where
T = Td Tr
Film:
f , free = T , f T
f ,attached = T , s T
Some of the final strain is

accounted for by the strain that the
film would have if it were free. The
remainder, or mismatch strain, will
be associated with a stress through
constitutive relationships.
Mismatch:
f ,mismatch = ( T , f T , s )T
Biaxial stress:
f ,mismatch
E
=
f ,mismatch
(1 )
Assuming that the film is much thinner than

the substrate, the films actual strain is
whatever the substrate imposes.
Intrinsic residual stress

> Any thin film residual stress that cannot be explained by thermal
expansion mismatch is called an intrinsic stress
> Sources of intrinsic stress

Deposition far from equilibrium
Secondary grain growth can modify stresses
Ion implantation can produce compressive stress
Substitutional impurities can modify stress
etc.
Edge effects
> If a bonded thin film is in a state of plane stress due to residual
stress created when the film is formed, there are extra stresses
at the edges of these films
Shear stresses
F=0
F=0
Extra peel force

Adapted from Figure 8.6 in: Senturia, Stephen D. Microsystem Design. Boston, MA:
Outline
> Overview
> Some definitions
Stress
Strain
Plane stress

> Remember calculating potential energy in physics
xf
U = Fx dx (for example, U = mgh)

xi
> Deforming a material stores elastic energy

> Stress = F/A, strain = L/L
(x,y,z)
()d = ???
> Together, they contribute 1/length3: strain energy density at a

point in space
Elastic Energy
> Elastic stored energy density is the integral of stress with
respect to strain
Elastic energy density :

When ( ) = E :
(x,y,z)
~
W(x,y,z) =
()d
0
1
~
2
W(x,y,z) = E [(x,y,z)]
2
> The total elastic stored energy is the volume integral of the
elastic energy density
Total stored elastic energy :
W=
~
W
(x,y,z)dxdydz
Volume
> You must know the distribution of stress and strain through a
structure in order to find the elastic energy stored in it (next
time).
Including Shear Strains

> More generally, the energy density in a linear elastic medium is
related to the product of stress and strain (both normal and
shear)
~ 1
For axial strains : W =
2
~ 1
For shear strains : W =
2
This leads to a total elastic strain energy :
W=
1
( x x + y y + z z + xy xy + xz xz + yz yz ) dxdydz
2 Volume

> General case:
Stress is a second
rank tensor
Strain is a second
rank tensor
Elastic constants
form a fourth rank
tensor
> There is lots of

symmetry in all the
tensors
> Can represent stress

as a 1 x 6 array and
strain as a 1 x 6 array
x C11

y C12
C
z = 13
yz C14

zx C15
C16
xy
C12
C13
C14
C15
C22
C23 C24
C25
C23 C33 C34

C24 C34 C44
C35
C45
C25
C26
C55
C56
C35
C36
C45
C46
C16 x

C26 y
C36 z

C46 yz

C56 zx
C66 xy
> The elastic constants

form a 6 x 6 array,
also with symmetry
Stiffness and Compliance

> The matrix of stiffness coefficients,
analogous to Youngs modulus, are
denoted by Cij
> The matrix of compliance

coefficients, which is the inverse of
Cij, is denoted by Sij
> Yes, the notation is cruel

> Some texts use different symbols,
I = CIJ J
J
and
I = S IJ J
J
but these are quite widely used in

the literature
Cubic materials
> Only three

independent elastic
constants
C11 = C22 = C33
C12 = C23 = C31 = C21
= C32 = C13
C44 = C55 = C66

All others zero
> Values for silicon

C11 = 166 GPa
C12 = 64 GPa
C44 = 80 GPa
C11
C12
C
12
0
0
C12
C11
C12
C12
0
0
0
0
C12
C11
0
C44
0
0
0
0
0
0
0
0
C44
0
0
0
0
0
C44
Materials with Lower Symmetry
> Examples:
Zinc oxide 5 elastic constants
Quartz 6 elastic constants
> These materials come up in piezoelectricity

> Otherwise, we can enjoy the fact that most materials
we deal with are either isotropic or cubic
What lies beyond linear elasticity?
> So far, we have assumed linear elasticity.
> Linear elasticity fails at large strains

Some of the deformation becomes permanent (plastic strain)
Things get stiffer
Things break
Plastic deformation
> Beyond the yield
> This is exploited

in the bending
and stamping of
metals
5
4
Stress (arbitrary units)
point, a plastic
material develops
a permanent set
Loading curve
Unloading curve
2
1
Strain if unloaded to zero stress
Stress if unloaded to zero strain
-1
-2
4
2
3
Strain (arbitrary units)

Material behavior at large strain

> Brittle and ductile materials are very different
6
Brittle Fracture
Stress (arbitrary units)
Yield
Ductile Fracture
3
2
Elastomeric or flow region
1
0
2
3
4
Strain (arbitrary units)

Any thoughts on this device?
Figures 2, 3, and 4 on pp. 236-237 in: Kinoshita, H., K. Hoshino,

K., K. Matsumoto, and I. Shimoyama. "Thin Compound eye Camera with a Zooming
Function by Reflective Optics." In MEMS 2005 Miami: 18th IEEE International
Conference on Micro Electro Mechanical Systems: technical digest, Miami Beach,
Florida, USA, Jan. 30-Feb. 3, 2005. Piscataway, NJ: IEEE, 2005, pp. 235-238.
ISBN: 9780780387324. 2005 IEEE.
Outline
> Overview
> Some definitions
Stress
Strain
Plane stress
A simple example: axially loaded beams

> In equilibrium, force is uniform; hence stress is inversely
proportional to area (as long as area changes slowly with
position)
Geometry :
F
F
L
= =
and =
A WH
L
Uniaxial stress :
= E
F
L
=E
WH
L
EWH
F=
L
L
F = kL k =
Plug in for L=100 m,

W=5 m, H=1 m,
E=160 GPa:
EWH
L
k=8000 N/m
Another example: bending of beams and plates

> Stress and strain underlie bending, too
> Unlike uniaxial tension, where stress and strain are uniform,
bending of beams and plates is all about how the spatially
varying stress and strain contribute to an overall deformation.
Stretched: tensile stress

Compressive stress
> Next time!
Electronics B
Joel Voldman
Outline

> Op-amps
TODAY
Elements of semiconductor physics

> Discrete molecules (e.g., hydrogen
E (eV)
atom) have discrete energy levels that

the electrons can occupy
E (0 eV)
E5 (-0.54 eV)
E4 (-0.85 eV)
E3 (-1.51 eV)
E2 (-3.40 eV)
Determined via quantum mechanics
Paschen
series
Balmer
series
Hydrogen
> Adding a discrete amount of energy to

the electrons (via light, thermal energy,
etc.) can excite an electron from its
ground state to an excited state
> Different atoms have different number
E1 (-13.6 eV)
Lyman
series
(n = 3, 1 = 1)
p orbital 6
allowed levels
Silicon
(n = 2)
8 electrons
of filled states
All the action typically happens at the

highest unoccupied state
> Two distinct molecules have identical

and independent energy-level
structures
(n = 1)
2 electrons
(n = 3, 1 = 0)
s orbital 2
allowed levels

Adapted from Figures 2.1 and 2.6 in Razeghi, M. Fundamentals
of Solid State Engineering. 2nd ed. New York, NY: Springer,
2006, pp. 48 and 59. ISBN: 9780387281520.
Razeghi, Fundamentals of Solid State Engineering


>
>
>
>
>
Atoms packed into a lattice behave differently than discrete atoms

Their discrete energy levels coalesce into continuous energy bands
There may be many energy bands for the molecule
We dont care about the ones that are filled and inaccessible
We care about the highest one with electrons
6
4
Electron energy
2
0
p
s
4N empty
states
2N+2N
filled states
Isolated
Si atoms
2'
L1
Si
2'
15
15
X4
-4
X1
L1
-8
L2'
1
-12
L

Adapted from Figure 2.5 in: Pierret, Robert F. Semiconductor DeviceFundamentals.
Reading, MA: Addison-Wesley, 1996, p. 28. ISBN: 9780131784598.
Gap
L3,
-6
Si lattice
spacing
Energy
25'
X1
25'
-2
-10
Decreasing
atom spacing
L3
U,K
Wave Vector k

Figure 1 on p. 559 in: Chelikowsky, J. R., and M. L. Cohen. "Nonlocal
Pseudopotential Calculations for the Electronic Structure of Eleven
Diamond and Zinc-blende Semiconductors." Physical Review B 14, no.
2 (July 1976): 556-582.


>
>
>
>
>
The highest normally filled set of electronic states is the valence band
The lowest normally empty set of electronic states is the conduction band
An energy gap separates these states
At T=0 K, all the valence band states are filled
A filled band cannot conduct electricity This is an insulator
6
4
2
0
L3
2'
L1
15
Si
2'
15
Energy
25'
X1
25'
-2
conduction
band
Gap
L3,
X4
-4
-6
-10
energy gap
X1
L1
-8
L2'
1
-12
L
U,K
valence
band
Wave Vector k
T=0 K
Figure by MIT OpenCourseWare.
Figure 1 on p. 559 in: Chelikowsky, J. R., and M. L. Cohen. "Nonlocal
Pseudopotential Calculations for the Electronic Structure of Eleven
Diamond and Zinc-blende Semiconductors." Physical Review B 14, no.
2 (July 1976): 556-582.


> T>0 K, electrons have thermal energy (~kT)
> At equilibrium, the only way for an electron to cross the energy barrier
is to be thermally excited
> The number of electrons that can do this is

increases with thermal energy (and thus T)
decreases with larger bandgap
> The thermally excited electrons give rise to electrons

> The resulting empty states in the valence band give rise to holes
Behave similar to electrons but with opposite charge and different mass
electrons
holes
T=0 K
T=300 K
Key terminology
> Energy gap EG (1.1 eV in silicon at RT)
> Si atom concentration = 5x1022 /cm3
> The number of carriers in intrinsic Si is
related to
Probability distribution function of energies

This obeys Fermi-Dirac statistics
Allowable states
> ~10-9 (prob of being filled) x ~1019 (density of

states) = ~1010 (carriers)
> ni is the intrinsic carrier concentration

the equilibrium concentration of holes and
electrons in the absence of dopants
~1.5x1010 cm-3 at RT
This is a material property
> n is the concentration [#/cm3] of electrons

> p is the concentration [#/cm3] of holes
ni e
EG
2 k BT
ni = pi
Every thermally
generated electron
leaves behind a hole
Key terminology
> Intrinsic carrier concentration of
1014
Si, Ge, and GaAs

1013
Ge
GaAs
Si
Intrinsic carrier concentration (cm-3)
1012
1011
1010
109
108
107
Adapted from Figure 2.20 in: Pierret, Robert F. Semiconductor Device Fundamentals.
106
105
200
300
400 500
T (K)
600
700

Doped semiconductors
> Dopants: substitutional impurity
atoms introduced having one
different valence than the
semiconductor
> Acceptor dopant concentrations are

NA
[#/cm3]
p-type material
Si
Si
Si
Si
Si
Si
Si
Si
Boron (3 valence electrons)
> Donor dopant concentrations are ND

[#/cm3] n-type material
Phosphorous (5 valence electrons)
> These introduce new energy levels

Adapted from Figure 7.5 in: Razeghi, M. Fundamentals of Solid
State Engineering. 2nd ed. New York, NY: Springer, 2006, p. 238.
ISBN: 9780387281520.
close to valence or conduction

bands (~0.05 eV)
> Dopant concentrations >> ni

T=0 K
Doped semiconductors
> Energy to ionize << kT (at RT)

> We assume all dopants are ionized at RT
Ec
ED
n-type dopant
Ev
T=0
T>0K
T = 300 K
Ec
p-type dopant
EA
Ev
Adapted from Figure 2.13 in: Pierret, Robert F. Semiconductor Device Fundamentals. Reading,
MA: Addison-Wesley, 1996, p. 38. ISBN: 9780131784598.
Main results
> Donors or acceptors are fully ionized
> Define n0 and p0 as the electron and hole
N D N D+ + e
N A N A + h +
concentrations at equilibrium
> n0 and p0 follow a mass-action law
N D N D+
N A N A
n0 p0 = ni2
+
+
N A = p N D+ = n
n= p
Main results
> Overall, silicon is neutral
Can use to determine n0 and p0
> Carrier concentrations typically

vary over many orders of
magnitude
Charge neutrality requires:

N A + n0 = N D+ + p0
N A + n0 N D + p0
Can use this to simplify
> Ex:
ni ~ 1010 cm-3
NA, ND ~ 1016-1019 cm-3
> In a given material at equilibrium

NA > 100ND (p-type)
ND > 100NA (n-type)
For p-type material:

p0 N A
ni2 ni2
n0 =
=
p0 N A
Main results
> Therefore, the equilibrium
majority carrier concentration is
determined by the net doping
and the minority carrier
concentration is determined by
the n0p0 product.
> For typical numbers, minority

carrier concentration is much
less majority carrier
concentration
p-type
n-type
p0 = N A
n0 = N D
ni2
n0 =
NA
ni2
p0 =
ND
N A ~ 1017 cm 3
p0 = N A = 1017 cm 3
n0 =
2
i
(1010 cm3 )
n
3
3
10
cm
=
=
NA
1017 cm 3
Excess carriers
> We can do various things to
create excess carriers
Shine light
Apply electric fields
> Excess carriers (n, p)
n = n n0
p = p p0
Generally, n = p
represent a departure from

equilibrium
> Excess holes and electrons

are created in pairs
> Excess carriers recombine
Recombination
rate
Electron-hole
pair (EHP)
dn
n
dt
m
n(t ) ~ n(0)e
exponentially in pairs,
governed by lifetime
The minority carrier lifetime
Excess carriers
> GaAs w/ NA=p0=1015 cm-3

> GaAs ni = 106 cm-3
> Create 1014 EHP/cm3 and
calculate carrier
concentrations over time
p
Carrier concentrations (cm-3)
> Therefore, n0= 10-3 cm-3
p(t)
1015
1014
n(t)
1013
1012
Adapted from Figure 4.7 in: Streetman, Ben G., and Sanjay Kumar Banerjee.
Solid State Electronic Devices. 6th ed. Upper Saddle River, NJ: Pearson
Prentice Hall, 2006, p. 127. ISBN: 9780131497269.
10
20
30
40
50
time (ns)
Drift and diffusion of carriers

> Carriers in semiconductors
obey a drift/diffusion flux
relation
> Drift: carriers move in an

electric field
> Diffusion: carriers move in a

concentration gradient
Electric field [V/cm]
drift
diffusion
J n = qe ( n nE + Dnn )
Diffusivity [cm2/s]
Mobility [cm2/V-s]
Carrier concentration
[cm-3]
> For p-type material
n is small drift current is
small
n can be big diffusion
current dominates
Drift
Figure 3.1b) in: Pierret, Robert F.Semiconductor Device Fundamentals.
Outline

> Semiconductor diodes and resistors
> The MOSFET and the MOSFET
inverter/amplifier
> Op-amps
The semiconductor diode

> A p-n junction has very different
concentrations of carriers in the two
regions, creating a diffusive driving
force
> At equilibrium diffusive driving force =
electric field in the vicinity of the
junction
ionized donors and acceptors are

relatively depleted of mobile carriers
near the junction the space-charge
layer (SCL) or depletion region
+ + + + + +
+ + + + + +
+ + + + + +
- + + + + + +
- + + + + + +
- + + + + + +
E
-
Oxide
n-type substrate
> In order to set up this electric field, the
p-type region
n-type
p-type
- + + + + + +
- + + + + + +
- + + + + + +
xp0
xn0
XJ0

,
Stephen D. Microsystem Design.
Adapted from Figure 14.1 in: Senturia
The exponential diode

> An external voltage modifies the net
potential drop across the space charge
layer
p
Ec
> Forward bias reduces the barrier to
diffusion, leading to an increase in

current
Ev
Equilibrium
> Reverse bias increases the barrier, so

only current is due to minority carriers
generated in or near the space-charge
layer
IN
Ec
Oxide
Metal
+
p-type region
n-type substrate
ID
Ev
VD
_
IP
Forward bias
Adapted from Figu
re 6.1 in: Pierret, Robert F. Semiconductor
Device Fundamentals. Reading, MA: Addison-Wesley, 1996,
p. 236. ISBN: 9780131784598.
Ideal Exponential Diode Analysis

> Linear changes in
voltage lead to
exponential changes in
carrier concentrations
I D = I0 e
qeVD
k BT
Forward bias
0.005
Current (A)
The total current is
0.010
-VB
0.000
Reverse blocking
-0.005
Reverse breakdown
-0.010
-6
-5
-4
-3
-2
-1
Voltage (V)
The Junction-Isolated Diffused Resistor

> The structure is a diode, but with two
Oxide
Metal
contacts
> The diode action prevents currents

into the substrate provided the diode
is always reverse biased
p-type region
n-type substrate
> The conductivity is controlled by

resistor
doping
> The resistor value is determined by

geometry
desired
+
-
undesired
+
-
distributed
diode

Adapted from Figure 14.7 in: Senturia, Stephen D. Microsystem
Design. Boston, MA: Kluwer Academic Publishers, 2001, p.
363. ISBN: 9780792372462.
Outline

> Semiconductor diodes and resistors
> The MOSFET and the MOSFET
inverter/amplifier
> Op-amps
MOSFET Structure
> The MOSFET exploits the concept of a field-induced junction
> The electric field between the gate and the channel region of the
substrate can either increase the surface concentration
(accumulation) or deplete the surface and eventually invert the
surface
Source
Gate
Drain
Oxide
D
Channels
p-type substrate
n+ regions
S
Body
MOSFET qualitative operation

> With D & S grounded,
back-to-back diodes
prevent current flow
between D & S
> To reduce barrier,

apply positive voltage
to G (w.r.t. D & S)
> At some threshold

voltage, this will form
an n-channel inversion
layer that will connect
D&S
MOSFET Characteristics
1000
> Need VGS>VT for device to turn on

between D & S
> As VDS increases, voltage between
triode
800
ID (A)
> As VDS increases, current will flow
VGS = 4 V
ID,sat
600
VGS = 3 V
400
G and D decreases
> When VDS gets too big, one side of
saturation
200
channel pinches off, preventing

further increases in current
VGS = 2 V
0

S
e-
e-
VDS
S
e-
e-
S
e-
e-
n-channel
VDS=0
VDS>0
VDS>VGS-VT
Different kinds of MOSFETs
S
G
S
B
p-channel
Enhancement
Depletion
n-channel

Large-signal and small-signal MOSFET models

D
> Electrical engineers use

circuit models to analyze
circuits involving MOSFETS
K/2(VGS-VTn)
> Can use either full nonlinear
characteristics
Simple large-signal
model, in saturation
> Or linearized small-signal

model
Different models include

different components
gm vgs
CGS
r0
+
-
S
Simple small-signal
model, in saturation
Outline

> Op-amps
Operational Amplifiers op-amps

> Let someone else design a high-performance amplifier
> Basic structure and transfer characteristic
v0
Vsat+
Signal path
vv+
Slope A0
v0
v+ - v-
+
Differential
amplifier
High-gain
amplifier
Output
amplifier
VsatDIP/SO Package
VS+
v-
Output A 1
v0
v+
Inverting input A 2
Non-inverting
input A
GND
VS-
_A
+ +
B_
v+
Output B
Inverting input B
Non-inverting
Input B
Top view

Adapted from Figures 14.23 and 14.24 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers,
2001, p. 382. ISBN: 9780792372462.
LM158
Ideal and non-ideal behavior

> We do first analysis using ideal
linear model
> Op-amp data sheets have pages

and pages of limitations
> A sampling
Input offset voltage voff: zero
v0 ( s ) = A( s ) ( v+ ( s ) v ( s ) )
where
A( s ) =
A0 s0
s + s0
volts at input gives non-zero

output
Frequency limitations: op-amps

can only amplify up to a
maximum frequency
The Inverting Amplifier

Assume op-amp draws no
current
R2
R1
Vs
v-
+
-
Use Nodal analysis:
Vo
v+
R1
+
-
vv+
+
-
Vs v v V0
=
R1
R2
V0 = A(0 v )
R2
Vs
KCL:
Vo
A(v+-v-)
Vo
R2
1
=
Vs
R1 1 R2
1 + 1 +
R1
A
A
Vo
R
= 2
Vs
R1
Short method for analyzing op-amps
> Assume linear region operation

This implies that the two inputs are essentially at the same
voltage (but never exactly equal)
> Assume zero currents at both inputs

> Analyze the external circuit with these constraints
> Check to verify that the output is not at either
saturation limit (VS)
More op-amp configurations
Vs
v = v+ = VS
+
V0
R1
R2
Vs = V0
R2 + R1
V0
R1
= 1+
VS
R2
R2
Non-inverting
Adapted from Figure 14.28 in Senturia, Stephen D. Microsystem Design.
More op-amp configurations

R1
Is
_
V0
+

Transimpedance amplifier
V0 = R1 I S
Integrator
C
R1
Vs
vC _
+
_
V0
+

1
V0 =
Vs (t )dt
R1C
Differentiator
R1
C
Vs
+
_
V0
+

The differentiator is less "ideal"

Vs
0 V0
=
1
R1
sC
V0
= sR1C
Vs
Structures
Carol Livermore

Outline
> Regroup
> Beam bending
Loading and supports

Bending moments and shear forces
Curvature and the beam equation
Examples: cantilevers and doubly supported beams
> A quick look at torsion and plates
Recall: Isotropic Elasticity

> For a general case of loading, the constitutive relationships
between stress and elastic strain are as follows
> 6 equations, one for each normal stress and shear stress
xy
1
x = x ( y + z )
E
1
y = y ( z + x )
E
1
z = z ( x + y )
E
yz
zx
Shear modulus G is given by G =
1
= xy
G
1
= yz
G
1
= zx
G
E
2(1 + )
What we are considering today

> Bending in the limit of small deflections
> For axial loading, deflections are small until something bad
happens
Nonlinearity, plastic deformation, cracking, buckling

Strains typically of order 0.1% to 1%
> For bending, small deflections are typically less than the
thickness of the element (i.e. beam, plate) in question
What we are NOT considering today

> Basically, anything that makes todays theory not apply (not as
well, or not at all)
> Large deflections

Axial stretching becomes a noticeable effect
> Residual stresses
Can increase or decrease the ease of bending
Our trajectory
> What are the loads and the supports?
1 m
Silicon
0.5 m
Cantilever
Pull-down
electrode
Anchor

Adapted from Rebeiz, Gabriel M. RF MEMS: Theory, Design, and
Technology. Hoboken, NJ: John Wiley, 2003. ISBN: 9780471201694.
> What is the bending moment at point x along the beam?

M
> How much curvature does that bending moment create in the
structure at x? (Now you have the beam equation.)
> Integrate to find deformed shape

Outline
> Regroup
> Bending

Types of Loads
> Three basic types of loads:
Point force (an old friend, with its own specific point of
application)
Distributed loads (pressure)

Concentrated moment (what you get from a screwdriver, with
a specific point of application)
The forces and moments work together to make internal

bending moments more on this shortly
F=
point load
q = distributed load
M=
concentrated moment
Types of supports
> Four basic boundary conditions:

Fixed: cant translate at all, cant rotate
Pinned: cant translate at all, but free to rotate (like a hinge)
Pinned on rollers: can translate along the surface but not off
the surface, free to rotate

Free: unconstrained boundary condition
Fixed
Pinned
Free
Pinned on Rollers

Reaction Forces and Moments

> Equilibrium requires that the total force on an object be zero
and that the total moment about any axis be zero
> This gives rise to reaction forces and moments

> Cant translate means support can have reaction forces
> Cant rotate means support can have reaction moments
Point Load
F
x
MR
O
Total moment about support :

M T = M R FL
Moment must be zero in equilibrium :
M R = FL
Net force must be zero in equilibrium :
FR = F
FR
Adapted from Figure 9.7 in: Senturia, Stephen D. Microsystem Design .
Internal forces and moments

> Each segment of beam must also be in equilibrium
> This leads to internal shear forces V(x) and bending moments
M(x)
For this case,
M (x ) = F ( L x)
V (x ) = F

Adapted from Figure 9.7 in Senturia, Stephen D. Microsystem Design
Some conventions
Moments
Positive
Negative
Moments:
Image by MIT OpenCou rseWare.

Microsystem Design. Bo ston, MA: Kluwer Academic
Publishers, 2001, p. 210 . ISBN: 9780792372462.
Shears
Positive
Negative
Shears:

Microsystem Design. Boston, MA: Kluwer Academic
Publishers, 2001, p. 210. ISBN: 9780792372462.
Combining all loads

> A differential beam element, subjected to point loads,
distributed loads and moments in equilibrium, must obey
governing differential equations
All Loads:
q
V
M+dM
V+dV
dx

Adapted from Figure 9.8 in: Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic
FT = qdx + (V + dV ) V
dV
= q
dx
= ( M + dM ) M (V + dV ) dx
qdx
dx
2
dM
=V
dx
Outline
> Regroup
> Bending

Pure bending
> Important concept: THE NEUTRAL AXIS
> Axial stress varies with transverse position relative to the
neutral axis
dL = ( z )d
dx = d
Neutral Axis
Tension
z
MO
MO
d?
?
Compression
H/2
Tension
?x
Compression
H/2
Adapted from Figures 9.9 and 9.10 in: Senturia, Stephen D.
Microsystem Design . Boston, MA: Kluwer Academic
Publishers, 2001, pp. 211, 213. ISBN: 9780792372462.
z
dL = dx dx
z
x =
zE
x =
Locating the neutral axis

> In pure bending, locate the neutral axis by imposing equilibrium
of axial forces
N = ( z )dA = 0
A
thickness
E ( z )W ( z )z
One material, rectangular beam

dz = 0
EW
z dz = 0
thickness
> The neutral axis is in the middle for a one material beam of
symmetric cross-section.
> Composite beams: if the beam just has a very thin film on it, can
approximate neutral axis unchanged
> Composite beams: with films of comparable thickness, change

in E biases the location of the neutral axis
Curvature in pure bending

Curvature is related to the internal bending moment M.
Internal Moment :
M = z x dA
A
x =
M =
zE
For a one - material beam,

M =
2
z
dA
A
Moment of inertia I :
I = z 2 dA
A
2
E
(
z
)
z
dA
M =
EI
In pure bending, the internal moment M equals the externally

M0
1
=
applied moment M0. Then

for one material; for two
EI
or more materials, calculate an effective EI.
Useful case: rectangular beam

W
M0
M0
For a uniform rectangular beam,

H /2
1
I = Wz dz = WH 3
12
H / 2
2
1
3 E
M = WH
12
Differential equation of beam bending
> Relation between curvature and the applied load

x
dx
x
?(x)
ds
w(x)
d 2w
M
=
2
dx
EI
and, by successive differentiation
d?

Boston, MA: Kluwer Academic Publishers, 2001, pp. 214. ISBN: 9780792372462.
ds =
dx
cos
dx
dw
= tan
dx
ds = d
d 1 d 2 w
=
dx dx 2
d 3w
V
=
dx 3
EI
d 4w q
=
dx 4 EI
For large - angle bending

1
w
=
1 + (w)2 3 / 2
Large-angle bending is
rare in MEMS structures
Anticlastic curvature
> If a beam is bent, then the Poisson effect causes

opposite bending in the transverse direction
z
x =
y = x
/v
Original
cross-section
which creates a y - directed internal moment

y =

Boston, MA: Kluwer Academic Publishers, 2001, pp. 219. ISBN: 9780792372462.
Example: Cantilever with point load

d 2w F
=
( L x)
2
dx
EI
dw
F 2 FL
=
x +
x+ A
dx
2 EI
EI

d 2w
M (x )
=
2
dx
EI
and
M (x ) = F ( L x)
V (x ) = F
F 3 FL 2
w=
x +
x + Ax + B
6 EI
2 EI
BC : w(0) = 0,
dw
=0
dx x =0
A= B=0
w=
FL 2
x
x 1
2 EI 3L
Spring Constant for Cantilever
> Since force is applied at tip, if we find maximum tip

displacement, the ratio of displacement to force is the
spring constant.
wmax
kcantilever
L3
F
=
3EI
3EI EWH 3
= 3 =
L
4 L3
For the same dimensions as the uniaxially loaded beam,

kcantilever = 0.2 N/m
Stress in the Bent Cantilever
> To find bending stress, we find the radius of

curvature, then use the pure-bending case to find
stress
1 d 2w F
Radius of curvature : = 2 =
( L x)
EI
dx
Maximum value is at support ( x = 0)
FL
1
=
max EI
Maximum axial strain is at surface (z = H/2 )

LH
6L
F= 2
max =
F
H WE
2 EI
6L
max = 2 F
HW
Tabulated solutions
> Solutions to simple situations available in introductory
mechanics books
Point loads, distributed loads, applied moments

Handout from Crandall, Dahl, and Lardner, An Introduction to
the Mechanics of Solids, 1999, p. 531.
> Linearity: you can superpose the solutions

> Can save a bit of time
> Solutions use nomenclature of singularity functions
Singularity functions
xa
xa
xa
= 0 if x a < 0
= ( x a ) if x a > 0
n
= what is variably called an impulse or a delta function
Integrate as if it were just functions of (x-a); evaluate at the end.

Value: a single expression describes whats going on in different
regions of the beam
F
x
a
M (x ) = F a x
b
L
Overconstraint
> A cantilevers single support provides the necessary support reactions
and no more
> A fixed-fixed beam has an additional support, so it is overconstrained
> Static indeterminacy: must consider deformations and reactions to

determine state of the structure
> Many MEMS structures are statically indeterminate: flexures, optical

MEMS, switches,
> What this means for us

Failure modes and important operational effects: stress stiffening,
buckling
Your choice of how to calculate deflections
Example: center-loaded fixed-fixed beam

> Option 1: (general)
Start with beam equation in terms of q

Express load as a delta function
Integrate four times
d 4w q
=
4
dx
EI
Four B.C. give four constants
> Option 2: (not general)

Invoke symmetry
> Option 3: (general)

Pretend beam is a cantilever with as yet unknown moment and force
applied at end such that w(L) = slope(L) = 0
Using superposition, solve for deflection and slope everywhere

Impose B.C. to determine moment and force at end
Plug newly-determined moment and force into solution, and youre
done
Integration using singularity functions

F
q
F
1
=
xL 2
EI EI
F
0
x L 2 + C1
EI
F
1
x L 2 + C1 x + C2
EI
dw F x L 2
=
2
dx EI
L/2
L/2
d 4w
=
4
dx
d 3w
=
dx 3
d 2w
=
2
dx
q=F xL 2
Use boundary conditions to find

constants: no displacement at
supports, slope = 0 at supports
C1 x 2
+
+ C2 x + C3
2
3
C1 x 3 C2 x 2
F xL 2
w( x ) =
+
+
+ C3 x + C4
EI
6
6
2
Comparing spring constants

> Center-loaded fixed-fixed beam (same dimensions as previous)
F
16 EWH 3
k=
= 12.8 N/m
3
L
> Tip-loaded cantilever beam, same dimensions

F
EWH 3
k=
= 0.2 N/m
3
4L
> Uniaxially loaded beam, same dimensions

F
EWH
k=
= 8000 N/m
L
Outline
> Regroup
> Bending

Torsion
The treatment of torsion mirrors that of bending.
Figures 48 and 50 in: Hornbeck, Larry J. "From Cathode Rays to Digital Micromirrors: A History of
Electronic Projection Display Technology." Texas Instruments Technical Journal 15, no. 3 (July-Sept ember 1998): 7-46.

Figure 51 on p. 39 in: Hornbeck, Larry J. "From Cathode Rays to Digital Micromirrors: A History of
Electronic Projection Display Technology." Texas Instruments Technical Journal 15, no. 3 (July-Sept ember 1998): 7-46.
Digital Light Processing Technology: Texas Instruments

Digital Light Processing is a trademark of Texas Instruments.
Bending of plates
> A plate is a beam that is so
wide that the transverse
strains are inhibited, both
the Poisson contraction
and its associated
anticlastic curvature
> This leads to additional

stiffness when trying to
bend a plate
x =
x y
E
But y is constrained to be zero
0 = y =
y x
E
x =
2 x
1
Plate Modulus
Plate in pure bending

> Analogous to beam bending, with the limit on transverse strains
> Two radii of curvature along principal axes
1
2w
= 2
x x
x =
2w
= 2
y y
y =
> Stresses along principal axes

Ez
x =
1 2
Ez
y =
1 2
1
x = ( x y )
E
x
z
y =
1
+
x
y
y x
1
( y x )
E

> Relate moment per unit width of plate to curvature
> Treat x and y equivalently
Mx
= z x ( z )dz
W thickness
1 H 2 2
z dz
y H 2
x
1
Mx
= D
+
W
x y
1 EH 3
where D =
2
12 1
Mx
E
=
1 2
W
flexural
rigidity
> Note that stiffness comes from flexural rigidity as for a beam

> Recall that M is two derivatives away from a distributed load,
and that
2w
= 2
x x
1
2w
= 2
y y
1
> This leads to the equation for small amplitude bending of a plate
4w
4w
4w
D 4 + 2 2 2 + 4 = P( x, y )
x y
y
x
distributed load
> Often solve with polynomial solutions (simple cases) or

eigenfunction expansions
Where are we now?
> We can handle small deflections of beams and plates

> Physics intervenes for large deflections and residual
stress, and our solutions are no longer correct
> Now what do we do?

Residual stress: include it as an effective load
Large deflections: use Energy Methods
Lumped-element Modeling
with Equivalent Circuits
Joel Voldman
Outline
> Context and motivation

> Lumped-element modeling
> Equivalent circuits and circuit elements
> Connection laws
Context
> Where are we?

We have just learned how to make structures
About the properties of the constituent materials
And about elements in two domains
structures and electronics
> Now we are going to learn about modeling

Modeling for arbitrary energy domains
How to exchange energy between domains
Especially electrical and mechanical
How to model dynamics
> After, we start to learn about the rest of the domains

Inertial MEMS
> Analog Devices Accelerometer

ADXL150
Acceleration
capacitance
Changes gap
electrical output
Photograph of a circuit board.
Acceleration
Beam
Plate capacitances
Fixed plate
Anchor
Unit cell
Motio
1.3 Micron gap

125 Micron
overlap

Micrograph of machined microchannels.
2 Microns thick
RF MEMS
> Use electrical signal to create mechanical motion

> Series RF Switch (Northeastern & ADI)
Cantilever closes circuit when actuated

Figure 11 on p. 342 in: Zavracky, P. M., N. E. McGruer, R. H. Morrison,
and D. Potter. "Microswitches and Microrelays with a View Toward
Microwave Applications." International Journal of RF and Microwave
Computer-Aided Engineering 9, no. 4 (1999): 338-347.
1 m
Silicon
0.5 m
relay
Cantilever
Pull-down
electrode
Anchor
Imageby MIT OpenCourseWare.

Adapted from Rebeiz, Gabriel M.RF MEMS: Theory, Design, and Technology.
Zavracky et al., Int. J. RF Microwave CAE, 9:338, 1999, via Rebeiz RF MEMS
What wed like to do
> These systems are complicated 3D geometries

> Transform electrical energy
mechanical energy
> How do we design such structures?

Multiphysics FEM
Solve constitutive equations
at each node
Tedious but potentially
most accurate
> Is there an easier way?
Distorted switch (Coventor)
That will capture dimensional dependencies?

Allow for quick iterative design?
Maybe get us within 10-20%?
RF MEMS Switch
> What wed really like to know
What voltage will close the switch?

What voltage will open the switch (when closed)?
How fast will this happen?
What are the tradeoffs between these variables?
Actuation voltage vs. maximum switching frequency
> So lets restrict ourselves to relations between

voltage and tip deflection
Hah! we have lumped our system
Outline

> Connection laws
Lumped-element modeling
> What is a lumped element?

A discrete object that can exchange energy with other
objects
An object whose internal physics can be combined into

terminal relations
Whose size is smaller than wavelength of the appropriate

signal
Signals do not take time to propagate
Lumped elements
> Electrical capacitor
> Spring
+
V
-
> Rigid mass
I =C
Push on it and it moves

Relation between force
dV
dt
and displacement
> Fluidic channel

Apply pressure and fluid
flows instantaneously
Relation between
pressure and volumetric
flow rate
Point Load
F
x
MR
L
FR
wmax =
1
kcantilever

Pros/cons of lumped elements
> Pros
Simplified representations that carry dimensional
dependencies
Can do equivalent circuits

Static and dynamic analyses
> Cons
Lose information
Deflection along length of cantilever
Will not get things completely right
Capacitance due to fringing fields
So how do we go about lumping?
> First, we need input/output relations

This requires solving physics
This is what we do in the individual domains
We have already done this in electrical and mechanical
domains
> For cantilever RF switch

What is relation between force and tip deflection?
Not voltage and deflection
Different energy domains
RF Switch mechanical model

> We have seen that there is a linear relation
between force and tip deflection
Cantilever behaves as linear spring k

CAVEAT: k is specific for this problem
Different ks for same cantilever but
Distributed force applied over whole
cantilever
Point force applied at end

Deflection of cantilever middle is needed
Etc.
F = kx
> Lesson: Dont just use equation out of a book
RF Switch mechanical model
> What else is needed for

model?
> Inertia of cantilever

Lumped mass
> Energy loss
Lumped
dv
d 2x
F = ma = m = m 2
dt
dt
dashpot
Due to air damping
dx
F = bv = b
dt
How do we connect these together?
> Intuition and physics

> Example: cantilever switch
Tip movement (x) stretches spring

And causes damping
Tip has mass associated with it
All elements have same displacement
F
m
b
x
Outline

> Connection laws
Why use equivalent circuits?
> One modeling approach

Use circuits for electrical domain
Solve via KCL, KVL
Use mechanical lumped elements in mechanical domain
Solve via Newtons laws
Connect two using ODEs or matrices or other representation
> Our approach

Lumped elements have electrical equivalents
Can hook them together such that solving circuit intrinsically
solves Newtons laws (or continuity relationships)
Now we have ONE representation for many different domains

VERY POWERFUL
Onward to equivalent circuits

> Each lumped element has one or
more ports
through
> Each port is associated with

two variables
A through variable
An across variable
across
> Power into the port is defined by

the product of these two variables
Onward to equivalent circuits

> In electrical circuits, voltage is physically across and current
is physically through
voltage
across
current
through
> What happens when we translate mechanics into equivalent

circuits?
force
across (V)
velocity
through (I)
OR
force
through (I)
velocity
across (V)
> Why does this matter?
What circuit element is the spring?
> It stores elastic energy

> Is it a capacitor or an inductor?
F = k xdt
force
across (V)
velocity
through (I)
F = kx
V
dV
I =C
dt
C=1
k
x= 1 F
k
dF
1
x =
k dt
I
L= 1
k
dI
V =L
dt
1
I = Vdt
L
force
through (I)
velocity
across (V)
Which is correct?
> Both are correct
> And both are used
> Velocity
beware!
voltage
Indirect or mobility analogy

Cleaner match between physical system and circuit
Velocity is naturally across (e.g., relative) variable
But stores mechanical PE in inductors, KE in capacitors
Springs Inductors
> Force
voltage
Direct analogy
This is
Always store PE in capacitors what we
Springs Capacitors
will use
> Circuit topologies are dual of each other

Generalized variables
> We want a consistent modeling approach across

different domains
> Can we generalize what we just did?

YES
Generalized variables
> Formalize terminal
relations
> Displacement q(t)

> Flow f(t): the derivative of
displacement
> Effort e(t)

> Momentum p(t): the
integral of effort
> Net power into device is

effort times flow
General
Mechanical
dq
f =
dt
dx
v=
dt
q = qo + fdt
0
x = xo + vdt
0
dp
F=
dt
dp
e=
dt
t
p = po + edt
0
p = po + Fdt
0
Pnet = e f
Examples
> Effort-flow relations occur in MANY different energy domains
General
Electrical
Mechanical
Fluidic
Thermal
Effort (e)
Voltage, V
Force, F
Pressure, P
Temp. diff., T
Flow (f)
Current, I
Velocity, v
Vol. flow rate, Q
Heat flow
Displacement (q)
Charge, Q
Displacement, x
Volume, V
Heat, Q
Momentum (p)
Momentum, p
Pressure
Momentum,
Resistance
Resistor, R
Damper, b
Fluidic
resistance, R
Thermal
resistance, R
Capacitance
Capacitor, C
Spring, k
Fluid
capacitance, C
Heat capacity,
mcp
Inertance
Inductor, L
Mass, m
Inertance, M
Node law
KCL
Continuity of space
Mass
conservation
Heat energy
conservation
Mesh law
KVL
Newtons 2nd law
Pressure is
relative
Temperature is
relative
Other conventions
> Thermal convention: T becomes the across variable

(voltage) and heat-flow becomes the through variable
(current)
Conserved quantity is heat energy
Building equivalent circuits
> Need power sources

> Passive elements
> Topology and connection rules
Figure out how to put things together
> What do we get?

An intuitive representation of the relevant physics
Ability to model many domains in one representation
Access to extremely mature circuit analysis techniques and
software
One-port source elements
> Effort source and flow source

> Effort source establishes a
time-dependent effort
independent of flow
Electrical voltage source
Pressure source
> Flow source establishes a

time-dependent flow
independent of effort
Electrical current source
Syringe pump
f
+
e
-
+ e (t)
- 0
effort source
+
e
f0(t)
flow source
f,I,v
Power IN
Power OUT
f0,I0,v0
e0
V0
F0
Power IN
e
V
F
Power OUT
One-port circuit elements
> Three general passive elements

> Represent different functional relationships
Energy storage, dissipation
f
+
+
e
R
-
Relates e & f
Directly relates e & f
+
C
L
-
Relates e & q
Differentiates e
Integrates f
Relates f & p
Integrates e
Differentiates f
Analogies between mechanics and electronics

> Electrical Domain
> Mechanical Domain
A resistor
A damper (dashpot)
dQ
V = RI = R
dt
R=b
dx
F = bv = b
dt
> There is again a correspondence between
V and F
I and v
> Electrical Power = VI
Q and x
> Mechanical Power = Fv
R and b
Generalized resistor
> For the resistor,

e is an algebraic function of f
(or vice versa)
Can be a nonlinear function

f,I,v
Nonlinear resistor
f = f (e)
Linear resistor
I=1 V
R
v= 1 F
b
e,V,F

> Electrical Domain
> Mechanical Domain
A capacitor
A spring
Q = CV
dV
I =C
dt
1
C=
k
x= 1 F
k
dx
dF
1
= x =
k dt
dt
> There is again a correspondence between
V and F
I and v
Q and x
R and b
Generalized capacitance
> For a generalized capacitance, the effort e is a

function of the generalized displacement q.
Linear capacitor
Nonlinear capacitor
e = (q )
e,V,F
V = 1 Q = (Q)
C
(Q) = Q
C
e1,V1,F1
q1
Q1
x1
q
Q
x
Generalized capacitance
> Capacitors store potential energy
How much?
> Leads to concept of energy and co-energy

Co-energy
Energy
e1
W (e1 ) = qde
*
q1
q1
W (q1 ) = edq = (q )dq
e,V,F
e1
= 1 (e)de
0
e1,V1,F1
W ( q1 ) + W * ( e1 ) = e1q1
W * ( e1 ) = e1q1 W ( q1 )
q1
Q1
x1
q
Q
x
Parallel-plate capacitor
> A linear parallel-plate capacitor

> Its energy and co-energy are numerically equal
A
C=
+
V
-
V = (Q) = Q
Q1
Q1
W (Q) = (Q)dQ = Q dQ
C
W (Q) =
Q
2C
I
Q = 1 (V ) = CV
V1
V1
W * (V ) = 1 (V )dV = CVdV
2
CV
W * (V ) =
2

> Electrical Domain
> Mechanical Domain
An inductor
A mass
dI
d 2Q
V =L =L 2
dt
dt
dv
d 2x
L = m F = ma = m = m 2
dt
dt
> There is a correspondence between
V and F
I and v
Q and x
L and m
Generalized Inertance
> For a generalized inertance, flow f is a function of

momentum p.
> This once again leads to concepts of energy and coenergy

p1
W ( p1 ) = 0 fdp
f1
f,v
f1,v1
W ( f1 ) = 0 pdf
*
W ( p1 ) + W * ( f1 ) = f1 p1
p1
W * ( f1 ) = f1 p1 W ( p1 )
Outline

> Connection laws
Circuits in the e
V convention
> Elements that share flow (e.g., current) and

displacement (e.g., charge) are placed in series in an
electric circuit
> Elements that share a common effort (e.g., Voltage)

are placed in parallel in an electric circuit
.
x
k
1/k
F
m
F
b
x
Spring-mass-dashpot system
+
-
+ ek - +
em
- eb + -
Equivalent circuit
Solving circuit solves the physics
> Apply force balance to spring-mass-damper system

k
F = ma
F kx bx = mx
b
x
> Solving KVL gives same result as Newtons laws!

F Fk Fm Fb = 0
Fk = kx, Fb = bx , Fm = mx
F = kx + bx + mx
.
x
+
-
1/k
+ Fk - +
Fm
- Fb + -
> Can also do this with complex impedances

Generating equivalent circuits
> Possible to go directly

But hard with e V analogy
See slide at end and text for details
> Easier to do via circuit duals

> Use convenience of f V convention, then switch to
e V
Force is current source
Each displacement variable is a node
Masses connected between nodes and ground
Other elements connected as shown in diagram
Example
F
k2
k1
m1
m2
b1
k1
a
k2
x1
x2
.
x2 1/k2
a
F
+
-
m2
.
x1
c
b1
m1
1/k1
e
Where does this leave us?
> A 2nd-order system is a 2nd-order system

> Analogies between RLC and SMD system
n =
k
1
1
=
=
m
LC
m1
k
> Use what you already know to understand the

intricacies of what you dont know
Energy coupling
> Where is coupling between domains?

> How does voltage
deflection?
> We need transducers
two-port elements that store
energy
> We will do this next time
Conclusion
> Can model complicated systems with lumped

elements
> Lumped elements from different domains have

equivalent-circuit representations
> These representations are not unique

We use the e
V convention in assigning voltage to the
effort variable
> Once we have circuits, we have access to

POWERFUL analysis tools
For more info
> Course text chapter 5

> H.A.C. Tilmans. Equivalent circuit representation of
electromechanical transducers
Part I: lumped elements: J. Micromech. Microeng. 6:157, 1996.
Part II: distributed systems: J. Micromech. Microeng. 7:285, 1997.
Errata: J. Micromech. Microeng. 6:359, 1996.
> R. A. Johnson. Mechanical filters in electronics

> Woodson and Melcher. Electromechanical Dynamics
> M. Rossi. Acoustics and electroacoustics
> Lots and lots of papers
Finding equivalent circuit: direct approach
> Find e V equivalent

circuit of following
k2
k1
m1
m2
> Note:
b1
k2 and m2 share same
x1
x2
displacement, caused
by F
.
x2
b1, and k1 share same
1/k2
.
x1
m2
displacement, x2 x1
If k1 , m2 and m1
share same
displacement
+
-
b1
m1
1/k1
Energy-conserving Transducers
Joel Voldman*
(*with thanks to SDS)
Outline
> Last time

> The two-port capacitor as a model for energyconserving transducers
> The transverse electrostatic actuator

> A look at pull-in
> Formulating state equations
Last time: equivalent circuits
> Learned how to describe systems as lumped

elements and equivalent circuits
k
F
m
Images removed due to copyright restrictions. Figure 11 on p.

342 in: Zavracky, P. M., N. E. McGruer, R. H. Morrison, and D.
Potter. "Microswitches and Microrelays with a View Toward Microwave
Engineering 9, no. 4 (1999): 338-347.
1 m
Silicon
0.5 m
b
x
.
x
Cantilever
Pull-down
electrode
Anchor
F
Adapted from Rebeiz, Gabriel M. RF MEMS: Theory, Design, and Technology
.
+
-
1/k
+ ek - +
em
- eb + -
b
Last time: equivalent circuits
> Saw that lumped elements in different domains all

had equivalent circuits
> Introduced generalized notation to describe many

different domains
dq
f =
dt
dp
e=
dt
t
p = po + edt
0
q = qo + fdt
0
Equivalent circuit elements

General
Electrical
Mechanical
Fluidic
Thermal
Effort (e)
Voltage, V
Force, F
Pressure, P
Temp. diff., T
Flow (f)
Current, I
Velocity, v
Vol. flow rate, Q
Heat flow,
Q
Displacement (q)
Charge, Q
Displacement, x
Volume, V
Heat, Q
Momentum (p)
Momentum, p
Pressure
Momentum,
Resistance
Resistor, R
Damper, b
Fluidic
resistance, R
Thermal
resistance, R
Capacitance
Capacitor, C
Spring, k
Fluid
capacitance, C
Heat capacity,
mcp
Inertance
Inductor, L
Mass, m
Inertance, M
Node law
KCL
Continuity of space
Mass
conservation
Heat energy
conservation
Mesh law
KVL
Newtons 2nd law
Pressure is
relative
Temperature is
relative
Todays goal
> How do we model an electrical force applied to the

cantilever?
> How can we describe converting energy between

domains?
> This leads to energy-conserving transducers
Outline
> Last time


> A look at pull-in
General Considerations
> In MEMS, we are often interested in sensors and actuators
> We can classify sensors and actuators by the way they handle
energy:
Energy-conserving transducers
Examples: electrostatic, magnetostatic, and piezoelectric
actuators
Transducers that use a dissipative effect
Examples: resistive or piezoresistive sensors
> There are fundamental reasons why these two classes must be
treated differently.
Energy-conserving transducers depend only on the state variables
that control energy storage. Therefore, quasi-static analysis is OK.

Dissipative transducers depend, in addition, on state variables that
determine the rate of energy dissipation, and are more complex as a
result.
An Energy-Conserving Transducer
> By definition, it dissipates no energy, hence contains

no resistive elements in its representation
> Instead, it can store energy from different domains

this creates the transducer action
> Because the stored energy is potential energy, we

use a capacitor to represent the element, but because
there are both mechanical and electrical inputs, this
must be a new element: a two-port capacitor
Capacitor with moveable plate
> A charged capacitor has a force of attraction between

its two plates
> If one of the plates is moveable, one can make an

electrostatic actuator.
I
+
Moveable plate
V
g
-
z
Fixed plate
Adapted from Figure 6.1 in: Senturia, Stephen D. Microsyst em Design . Boston, MA: Kluwer
Various ways of charging
> Charging at fixed gap

An external force is required to
prevent plate motion
io
work
lifting
No electrical energy at zero gap
Must do mechanical work to lift
the plate
> Either method results in

stored energy
No movement No mechanical
> Charging at zero gap, then
Force
Charge at fixed gap

+
I
+
io
+Q
-Q
Charge at zero gap, then....
Force
+
V
-
I
+Q
g
-Q
Pull up

Adapted from Figure 6.2 in Senturia, Stephen D. Microsy stem
Design.. Boston, MA: Kluwer Academic Publishers, 2001, p. 127.
ISBN: 9780792372462.
Charging at Fixed Gap
> The stored energy is

obtained directly from the
definition for a linear
capacitor
W=
> Anticipating that the gap

might vary, we now
explicitly include the gap
as a variable that
determines the stored
energy
e V
qQ
Q
V=
C
Q
0 edq =0 VdQ = 0 C dQ
Q2 Q2 g
W (Q, g ) =
=
2C 2A
q
C=
A
g
Pulling Up at Fixed Charge

> Putting charge at zero gap stores no electrical energy
Q2
0
C W =
g 0
2C
> Once charge is applied, determining stored energy is a

mechanics problem.
> In determining the force, we must avoid double-counting of

charge
E=
E-field of bottom plate
Q
2A
Q on top plate
+++++++
Q2
F = QE =
2A
g
W (Q , g ) = 0
Q2g
Fdg =
2 A
-------
The final stored energy is same as before!

ONLY depends on Q and g, not the path!
Lossless transducers
> The energy in the system ONLY depends on the

STATE variables (e.g., Q, g) and NOT how we put the
energy in
The system is lossless/conservative
dW
= Pelectrical + Pmechanical
dt
= VI + Fg
dW
dQ
dg
=V
+F
dt
dt
dt
dW = VdQ + Fdg
I
+
V
-
C
W(Q,g)
+
F
-

Adapted from Figure 6.3 in Senturia, Stephen D. Microsystem Design
.
A Differential Version
> Since we can modify the stored energy either by changing the
charge or moving the plate, we can think of the stored energy as
defined differentially
dW = VdQ + Fdg
This leads to a pair of differential relations for the force and
voltage
W (Q , g )
F=
g
Q
W (Q , g )
V=
Q
g
Revisit charging the capacitor
> The energy only depends on Q, g

These are thus the STATE variables for this transducer
Q
2
Q g
W (Q, g ) =
2A
Charge, then move plates
Q1
Move plates, then charge
g1
The two-port capacitor
> This transducer is what will couple our electrical

domain to our mechanical domain
g
I
+
V
-
C
W(Q,g)
+
F
-

Adapted from Figure 6.3 in Senturia, Stephen D. Microsystem De
. sign
Q2 g
W (Q, g ) =
2A
W (Q, g )
Qg
=
V=
Q
A
g
W (Q, g )
Q2
=
F=
g
2A
Q
A different example
> What if the material in the gap could move?

x
+Q
0
-Q
x0
C=
Q2
W (Q, x) =
2C
l
( 0 x + ( x0 x) )
g
Q2 g
W (Q, x)
1
F=
=
x
2l x ( 0 x + ( x0 x) )
Q
0
Q2 g
F=
2l ( 0 x + ( x0 x) )2
Outline
> Last time


> A look at pull-in
The Electrostatic Actuator
> If we now add a spring to the upper plate to supply

the external mechanical force, a practical actuator
results
> We are getting closer to our RF switch

.
Fixed support
+
V
-
Spring k
V
g
z
Fixed plate
C
W(Q,g)
+
F
1/k

Adapted from Figure 6.4 in: Senturia,Stephen D. Microsystem Design..

Two methods of electrical control
> Charge control

Capacitor is charged from a current source, specifically
controlling the charge regardless of the motion of the plate
This method is analyzed with the stored energy
> Voltage control

Capacitor is charged from a voltage source, specifically
controlling the voltage regardless of the motion of the plate
This method is analyzed with the stored co-energy
Charge control
>
Following the causal path

1. Current source
determines the charge
2. Charge determines the
force (at any gap!)
3. Force determines the
extension of the spring
4. Extension of the spring
determines the gap
5. Charge and gap together
determine the voltage
in(t)
+
V
-
W(Q,g)
1/k

t
1)
2)
3)
z
Fixed plate
Adapted from Figure 6.4 in: Senturia,Stephen D. Microsystem Design.

.
Q = iin (t )dt
0
Spring k
Fixed support
I
4)
W
F=
g
Q2
=
2A
F
initial displacement
k
g = g0 z
z=
Q2
g = g0
2Ak
Charge control
> Lets get voltage, normalize and plot

W
V=
Q
Q2
Q g 0
2Ak
Qg
=
=
A
A
> Normalize variables to make easier to plot

First normalize V and Q to some nominal values
Introduce (normalized displacement) that goes from 0 (g=g0)
to 1 (g=0)
v =V
V0
q=Q
Q0
g g)
= zg =( 0
g0
0
Define Q0 and V0 using expression above

V0 =
Q0 g 0
A
Q02 = 2Akg 0
Charge control
> Now, plug in to non-dimensionalize
Q2
Q g 0
2Ak
V=
A
2
(
qQ0 )
(qQ0 ) g 0
2Ak (qQ0 ) g 0 q 2 g 0
V=
=
A
A
Qg
V = 0 0 q (1 q 2 ) v = q (1 q 2 )
A
= 1 g g = 1 (1 q 2 ) = q 2
0
> Now we get expressions relating voltage and

displacement to charge
Charge control
> Actuator is stable at all
multivalued the
charge uniquely
determines the state
and thus the energy
W
Q
Q2
Q g 0
2Ak
Qg
=
=
A
A
0.4
normalized
voltage
> The voltage is
V=
normalized
displacement ( )
gaps the voltage

goes to zero at zero
gap
0.2
0
0
1
0.5
0.5
0
0
0.5
1
normalized charge (q)
Co-Energy
> For voltage control, we cannot use W(Q,g) directly, because we
cannot maintain constant charge. Instead we use the co-energy
So we change variables
Recall: W * (e1 ) = q1e1 W ( q1 )
W * (V , g ) = QV W (Q, g )
dW * (V , g ) = d (QV ) dW (Q, g )
dW * (V , g ) = [QdV + VdQ ] [VdQ + Fdg ]
dW * (V , g ) = QdV Fdg
W * (V , g )
Q=
V
g
W * (V , g )
F =
g
V
Voltage control
> Following the causal path
1. Voltage and gap (implicitly)
determines the force
1
A 2
W * (Vin , g ) = CVin2 =
Vin
2
2g
2. Force determines the

spring extension
3. And thus the gap

4. Voltage and gap together
AVin2
W *
1) F =
=
2g 2
g V
determine the charge

.
I
+
Vin(t)
+
-
V
-
g = g0 z
2)
z=
C
W*(V,g)
F
k
1/k

3)
4)
g = g0
Q=
A
g
AVin2
2kg 2
Vin = CVin
Outline
> Last time


> A look at pull-in
Forces and stability

> Lets examine the net force
FNet
on the actuator
= k ( g0 g )
AV 2
2g
=0
> Nondimensionalize again
= (g 0 g ) / g 0
v =V
2
PI
V =
VPI
8kg 03
27A
4v 2 g 02
=0
2
27 g
positive
force
increases
gap
normalized force
FNet = Fmech Felec = 0
Av 2 8kg 03
= kg 0
=0
2
2 g 27A
4v 2
=0
2
27(1 )
Spring force
Electrical force
0.8
0.6
0.4
Increasing v
0.2
0
0 0.2 0.4 0.6 0.8 1
normalized displacement ( )
Stability criterion
> At low voltage, there are

two intersections
Which is stable?
The position of the actuator

is stable only when there is a
net restoring force when the
system is disturbed from
equilibrium
> At higher voltages,

there are none
What is happening?
stable
unstable
Fnet
Fnet
g
Stability criterion
> We can plot the normalized NET force versus

normalized gap and check
FNet = Fmech Felec
= k ( g0 g )
normalized force
unstable
stable
0.5
1 =
g
g0
-1
0
f net
Increasing v
0.5
1
normalized gap (g/g )
2g 2
4v 2
=
27(1 ) 2
0
-0.5
AV 2
g0
= + 1
g
4v 2
g0
27
g
Stability criterion
> So what we want is a

negative slope
FNet = k ( g 0 g )
> In this example, this

means that the spring
constant must exceed a
critical value that varies
with voltage
Stability:
AV 2
2g 2
FNet
AV 2
<0
= k +
3
g
g
AV 2
<k
3
g
Stability criterion
> If the voltage is too

large, the system
becomes unstable, and
we encounter pull-in
> Right at pull-in, the

spring constant is AT
the critical value AND
static equilibrium is
maintained
k=
At pull-in:
AVPI2
3
g PI
k ( g 0 g PI ) =
AVPI2
2
2 g PI
kg PI
k ( g 0 g PI ) =
2
8kg 03
VPI =
27 A
2
g PI = g 0
3
Stability analysis of pull-in

> Plot normalized gap versus
normalized voltage
> Solve cubic equation
g = g0
AVin2
2kg 2
normalized gap
stable
In Matlab:
1
0.5
0
0
unstable
0.5
1
normalized voltage
g = fzero(@(g)(g - g0 + eps*A*V^2/(2*k*g^2)),g0);
Release voltage after pull-in

> After pull-in less voltage is
needed to keep beam down
> Find force when pulled down

> Equate to mechanical force to
get hold-down voltage
> Is usually much less than pullin voltage

Normalize
to VPI
8kg 03
V =
27A
2
PI
g0
Felec
Fmech
g =
g =
AVin2
=
2 2
= k (g 0 ) kg 0
2
AVHD
= kg 0
2
2
2 2 kg 0
2
VHD =
A
VHD
27
< 1
=
4 g0
VPI
Macro pull-in?
> Can we do a macroscopic

pull-in demo?
> Use soft spring k = 1 N/m

> Use
A = 8.5 x 11 plates
g0 = 1 cm
8kg 03
VPI =
27 A
=
8(1)(0.01)3
27 ( 8.85 1012 ) 8.5 11 ( 0.0254 )
750 V
> Not easy this is why pullin is a MEMS-specific

phenomenon
Outline
> Last time


> A look at pull-in
Adding dynamics
> Add components to complete
Fixed support
Resistor
R
the system:
Source resistor for the voltage
Dashpot b
Mass m
Spring k
source
Inertial mass, dashpot
Vin
Fixed plate
> This is now our RF switch!

> System is nonlinear, so we
cant use Laplace to get
transfer functions
+
Vin
+
V
-
1/k
W(Q,g)
> Instead, model with state

equations

Electrical domain
Mechanical domain
The System is Now General
> The addition of the source resistor breaks up the

distinction between voltage-controlled and chargecontrolled actuation:
For small R, the system behaves like a voltage-controlled
actuator
For large R, the system behaves like a charge-controlled

actuator at short times because the impedance of the rest
of the circuit is negligible the voltage source delivers a
constant current V/R*
*See, for example, Castaner and Senturia, JMEMS, 8, 290 (1999)

State Equations
> Dynamic equations for general
system (linear or nonlinear) can be
formulated by solving equivalent
circuit
> In general, there is one state

variable for each independent
energy-storage element (port)
Goal:
Q
> Good choices for state variables:
d functions of
g =
the charge on a capacitor
Q,g,g or constants
dt
(displacement) and the current in an

g
inductor (momentum)
> For electrostatic transducer, need

three state variables
Two for transducer (Q,g)

One for mass (dg/dt)
Formulating state equations
> Start with Q
+ eR - I
> We know that dQ/dt=I

> Find relation between I and
state variables and constants
KVL : Vin eR V = 0
Vin IR V = 0
dQ
1
= I = (Vin V )
dt
R
dQ 1
Qg
= Vin
dt R
A
Vin
+
-
W(Q,g)
eR = IR
V=
Qg
A
> Now well do
g
> We know that dg
.
g
dt
KVL :
F ek em eb = 0
F kz mz bz = 0
= g
ek = kz
em = mz
eb = bz
z = g 0 g z = g , z = g
.
z
1/k
+ ek - +
W(Q,g)
em
- eb + b
F k ( g 0 g ) + mg + bg = 0
1
[F k ( g 0 g ) + bg ]
m
dg
1 Q2
=
k ( g 0 g ) + bg
dt
m 2A
g =
> State equation for g is easy:
dg
= g
dt
> Collect all three nonlinear state equations
1
Qg
Vin
R
A
Q

=
g
g
dt
g 1 Q 2

m 2 A k ( g 0 g ) + bg
> Now we are ready to simulate dynamics (WED)
What have we wrought?
> We have modeled a complex multi-domain 3D

structure using
Equivalent circuits
A two-port nonlinear capacitor
> What can we now get

Actuation voltage: VPI
Tip dynamics
> What have we lost
Figure 9 on p. 17 in: Nguyen, C. T.-C.

"Vibrating RF MEMS Overview: Applic ations to W i reless Communications."
Proceedings of SPI E Int Soc Opt Eng 5715 (January 2005): 11-25.
I m ages rem oved due to copyright restrictions. Figure 11 on p.

342 in: Za vrack y, P. M ., N. E. Mc Grue r, R. H. Mo rris on, an d D.
Potter. "Mi crosw itches and Microrelays w i th a Vie w To w ard Micro w ave
Applications." Internati onal Journal of RF and Microw ave Com put-Aided
Engineering 9, no. 4 (1999): 338-347.
Capacitor plates are not really parallel during actuation

Neglected fringing fields
Neglected stiction forces when beam is pulled in
Conclusions
> We can successfully model nonlinear transducers

with a new element: the two-port capacitor
> Know when to use energy or co-energy for forces

At best a sign error
At worst just wrong
> Under charge control, transverse electrostatic

actuator is well-behaved
> Under voltage control, exhibits pull-in
Lumped-Element System Dynamics
Joel Voldman*
*(with thanks to SDS)
Outline
> Our progress so far

> Quasistatic analysis
> Large-signal analysis
> Small-signal analysis
> Addendum: Review of 2nd-order system dynamics
Our progress so far
> Our goal has been to model multi-domain systems

> We first learned to create lumped models for each
domain
> Then we figured out how to move energy between

domains
> Now we want to see how the multi-domain system

behaves over time (or frequency)
Our progress so far
> The Northeastern/ADI RF Switch

> We first lumped the mechanical domain
Fixed support
Figure 11 on p. 342 in: Zavracky, P. M., N. E. McGruer, R. H. Morrison,
and D. Potter. "Microswitches and Microrelays with a View Toward Microwave
Engineering 9, no. 4 (1999): 338-347.
Dashpot b
+
V
1 m
Silicon
0.5 m
Cantilever
Pull-down
electrode
Anchor
Spring k
Mass m
z
Fixed plate

Adapted from Figure 6.9 in Senturia, Stephen D. Microsystem
Design. Boston, MA: Kluwer Academic Publishers, 2001, p.
138. ISBN: 9780792372462.

Adapted from Rebeiz, Gabriel M. RF MEMS: Theory, Design, and
Technology. Hoboken, NJ: John Wiley, 2003. ISBN: 9780471201694.
Our progress so far
> Then we introduced a two-port capacitor to convert

energy between domains
Capacitor because it stores potential energy
Two ports because there are two ways to store energy
Mechanical: Move plates (with charge on plates)
Electrical: Add charge (with plates apart)
The system is conservative: system energy only depends on
state variables
.
g
I
2
Q g
W (Q, g ) =
2A
+
V
1/k
+
C
Our progress so far

W (Q, g )
Q2
=
F=
g
2A
Q
> We first analyzed system

quasistatically
> Saw that there is VERY different
V=
W (Q, g )
Qg
=
Q
A
g
behavior depending on whether
Charge is controlled
stable behavior at all gaps
Voltage is controlled
dW = VdQ + Fdg
dW * = QdV Fdg
pull-in at g=2/3g0
> Use of energy or co-energy

depends on what is controlled
Simplifies math
W *
Q=
V
=
g
A
g
Vin
AVin2
W *
=
F=
g V
2g 2
Todays goal
> How to move from quasi-static to dynamic analysis

> Specific questions:
How fast will RF switch close?
> General questions:

How do we model the dynamics of non-linear systems?
How are mechanical dynamics affected by electrical domain?
> What are the different ways to get from model to

answer?
Outline

Adding dynamics
Fixed support
> Add components to complete
Resistor
R
the system:
Dashpot b
Mass m
Spring k
Source resistor for the voltage
Vin
source
Inertial mass, dashpot
Fixed plate
> This is now our RF switch!

> System is nonlinear, so we
cant use Laplace to get
transfer functions
> Instead, model with state
I
+
Vin
+
V
-
1/k
+
m
W(Q,g)
equations
Electrical domain
Mechanical domain
State Equations
> Dynamic equations for general
system (linear or nonlinear) can be
formulated by solving equivalent
circuit
> In general, there is one state variable

for each independent energy-storage
element (port)
> Good choices for state variables:

the charge on a capacitor
(displacement) and the current in an
inductor (momentum)
Goal:
Q
d functions of
g =
Q,g,g or constants
dt
g
> For electrostatic transducer, need

three state variables
Two for transducer (Q,g)

One for mass (dg/dt)
> Start with Q
+ eR - I
> We know that dQ/dt=I

> Find relation between I and
state variables and constants
KVL : Vin eR V = 0
Vin IR V = 0
dQ
1
= I = (Vin V )
dt
R
dQ 1
Qg
= Vin
dt R
A
Vin
+
-
W(Q,g)
eR = IR
Qg
V=
A
> Now well do
g
> We know that dg
.
g
dt
KVL :
F ek em eb = 0
F kz mz bz = 0
= g
ek = kz
em = mz
eb = bz
z = g 0 g z = g , z = g
.
z
1/k
+ ek - +
W(Q,g)
em
- eb + b
F k ( g 0 g ) + mg + bg = 0
1
[F k ( g 0 g ) + bg ]
m
dg
1 Q2
=
k ( g 0 g ) + bg
dt
m 2A
g =
> State equation for g is easy:
dg
= g
dt
> Collect all three nonlinear state equations
1
Qg
Vin
R
A
Q

g
g
=
dt
g 1 Q 2

m 2 A k ( g 0 g ) + bg
> Now we are ready to simulate dynamics
Outline

Quasistatic analysis
1
Qg
Vin
A
R
d
g = g
dt g 1 Q 2

k
g
g
b
g
(
)
+
0
m 2A
Fixed-point
analysis
Given static Vin, etc.
State eqns
What is deflection,
charge, etc.?
(Just now)
.
Vin
+
-
W(Q,g)
1/k
+
m
Follow
causal path
(Wed)
Equivalent circuit
State equations
For transverse
electrostatic actuator:
State variables
Inputs
x = f (x, u)
y = g (x, u)
Outputs
State variables:
Q
x = g
g
Inputs:
u = [Vin ]
Qg
Vin
A
R
f (x,u) = g
1Q
m 2 A k ( g 0 g ) + bg
Outputs: y = [g ] = g (x, u)

Fixed points
> Definition of a fixed point

Solution of f(x,u) = 0
time derivatives 0
> Global fixed point

A fixed point when u = 0
Systems can have multiple global fixed points
Some might be stable, others unstable (consider a
pendulum)
> Operating point

Fixed point when u is a non-zero constant
Fixed points of the electrostatic actuator

> This analysis is analogous to what we did last time
Qg
1
0 = Vin
R
A
0 = g
1 Q2

0=
k ( g 0 g ) + bg
m 2A
Qg
A
Q 2 Vin 2 A
=
= k ( g0 g )
2
2 A
2g
Vin =
normalized gap
Operating point
stable
1
0.5
0
0
0.5
1
normalized voltage
Last time
g = g0
AVin2
2kg 2
Outline

Large-signal analysis
1
Qg
Vin
A
R
d
g = g
dt g 1 Q 2

k
g
g
b
g
(
)
+
0
m 2A
Integrate
state eqns
Given a step input
Vin(t)u(t)
State eqns
(Earlier)
.
Vin
+
-
W(Q,g)
What is g(t), Q(t), etc.?
1/k
SPICE
+
m
Equivalent circuit
Direct Integration
> This is a brute force approach: integrate the state

equations
Via MATLAB (ODExx)
Via Simulink
> We show the SIMULINK version here

Matlab version later
Electrostatic actuator in Simulink

Q
x = g , u = [Vin ]
g
1
Qg
V
in
A
R
f(x, u) = g
1 Q
m 2A k ( g 0 g ) + bg
(u[1]^2)/(2*e*A)
+
+
+
Electrostatic force
Spring
Sum2
-1/m
Inertia
Damping
1
s
Velocity
gdot
1
s
Position
At-rest gap g _ 0
1/(e*A)
*
Qg
_
1
V_in
1/R
1
s
1/R
Charge

Adapted from Figure 7.8 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer
Electrostatic actuator with contact

<
Accel > 0 ?
+ +
g_min
>
g > g_min?
+
+
+
(u[1]^2)/(2*e*A)
Electrostatic force
-1/m
1
s
Inertia
Switch
Spring
Sum2
Velocity
Damping
gdot
Zero
1
s
Position
At-rest gap g _ 0
1/(e*A)
*
Qg
_
1
V_in
1/R
1/R
1
s
Charge
1
Q

Adapted from Figure 7.9 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001,
p. 175. ISBN: 9780792372462.
Behavior through pull-in
7 10 8
1.6
Drive (scaled)
Charge
6 10 8
1.4
1.2
5 10 8
Position
4 10 8
3 10 8
2 10 8
1.0
Pull-in
0.8
0.6
0.4
1 10 8
0 10 0
Release
0.2
0
50
100
150
Time
200
250
300
0.0
50
100
150
200
250
300
Time

Adapted from Figure 7.10 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001,
p. 176. ISBN: 9780792372462.
Behavior through pull-in
1.0
1.4
1.2
Release
0.0
Drive
1.0
Position
Velocity
0.5
Pull-in
-0.5
Release
0.8
0.6
Discharge
0.4
-1.0
0.2
-1.5
50
100
150
Time
200
250
300
0.0
Pull-in
50
100
150
200
250
300
Time

Adapted from Figure 7.11 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001, p. 177. ISBN: 9780792372462.
Outline

Small-signal analysis
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Time
response
Integrate
Take LT
n
ct
nf s
g e si
Ei aly
An
form TFs
Equivalent Linearize Linearized

circuit
circuit
What is g(t)
due to small
changes in
Vin(t)?
Form TFs
Transfer SSS Frequency

functions
response
poles &
zeros
Natural
system
dynamics
Given O.P.
How fast
can I wiggle
tip?
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Integrate
Time
response
What is g(t)
due to small
changes in
Vin(t)?
Given O.P.
Equivalent
circuit
How fast
can I wiggle
tip?
Linearization about a fixed point
> This is EXTREMELY common in MEMS literature

> This is also done in many other fields, with different
names
Incremental analysis
Etc.
Linearization About an Operating Point

> Using Taylors theorem, a system can
x = f (x, u)
x (t ) = X 0 + x (t )
u(t ) = U0 + u(t )
be linearized about any fixed point
> We can do this in one dimension or
Operating point
many
f(x)
df/dx
~ f(X0+ x)
f( X0)=Y0
f
(
)
d
x
= f (X 0 , U0 ) + i
X 0 +
x
dt
j
X0
f ( X 0 + x ) f ( X 0 ) +
df
dx
d (x )
X 0 +
= f (X 0 + x, U0 + u)
dt
Multi-dimensional Taylor
x
X0
x (t ) + f i
u
X 0 ,U 0
j
u(t )
X 0 ,U 0
Cancel
d (xi (t ) ) f i
=
x
dt
j
x (t ) + f i
i
u
X 0 ,U 0
j
J1
u (t )
i
X 0 ,U 0
J2
Linearization About an Operating Point

x (t ) = J1x + J2u(t)
> The resulting set of

equations are linear, and
have dynamics described
by the Jacobians of f(x,u)
evaluted at the fixed point.
> These describe how much

a small change in one state
variable affects itself or
another state variable
>
>
f1
Q O.P.
f
J1 = 2
Q O.P.
f 3
Q O.P.
g 0
RA
Q
The O.P. must be evaluated d g = 0
dt
to use the Jacobian
g Q0
Example linearization of
mA
the voltage-controlled
electrostatic actuator
Q0
RA
0
k
m
J1
f1
g
f 2
g
f 3
g
O. P.
O. P.
O.P.
f1
g
f 2
g
f 3
g
O. P.
O. P.
O. P.
1 R
(V )
g +
0 in

b g 0

m
J2
State Equations for Linear Systems
> Normally expressed with:
x: a vector of state variables

u: a vector of inputs
y: a vector of outputs
Four matrices, A,B,C, D
x = Ax + Bu
y = Cx + Du
> For us, Jacobian matrices take the place of A and B

> C and D depend on what outputs are desired
Often C is identity and D is zero
> Can use to simulate time responses to arbitrary

SMALL inputs
Remember, this is only valid for small deviations from O.P.
Direct Integration in Time

Charge (Q)
Charge
> Can integrate via Simulink

model (as before) or
MATLAB
500
> First define system in
0
2
> Can use MATLAB

commands step, initial,
impulse etc.
> Response of electrostatic

actuator to impulse of voltage
Parameters from text (pg

167)
Velocity
alternate method
Displacement
MATLAB
using ss(J1,J2,C,D) or
Impulse Response
1000
0
-2
-4
4
2
0
-2
-4
-6
0
10
15
20
25
Time (sec)
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Integrate
Time
response
Take LT
What is g(t)
due to small
changes in
Vin(t)?
Form TFs

functions
response
poles &
zeros
Equivalent
circuit
Given O.P.
How fast
can I wiggle
tip?
Solve via Laplace transform
> Use Laplace Transforms

to solve in frequency
domain
Transform DE to algebraic
equations
Use unilateral Laplace to

allow for non-zero ICs
x = Ax + Bu
Unilateral Laplace
sX ( s ) x (0) = AX( s ) + BU( s )
sIX ( s ) x (0) = AX( s ) + BU( s )
(sI A )X( s) = x(0) + BU( s)
Transfer Functions
> Transfer functions H(s) are useful for obtaining
compact expression of input-output relation
What is the tip displacement as a function of voltage
> Most easily obtained from equivalent circuit

> But can also be obtained from linearized state eqns
Depends on A, B, C (or J1, J2, C) matrices
Can do this for fun analytically (see attachment at
end)
Matlab can automatically convert from s.s to t.f.

formulations
> For our actuator, we would get three transfer

functions
Q( s )
V
(
s
)
in
g( s)
H(s) =
(
s
)
V
in
g ( s )
(
s
)
V
in
Sinusoidal Steady State

> When a LTI system is
driven with a sinusoid, the
steady-state response is a
sinusoid at the same
frequency
> The amplitude of the

response is |H(j)|
u (t ) = U 0 cos(t )
Y ( j ) = H ( j )U ( j )
y sss (t ) = Y0 cos(t + )
> The phase of the response

relative to the drive is the
angle of H(j)
> A plot of log magnitude vs

log frequency and angle
vs log frequency is called
a Bode plot
Y0 = H ( j ) U 0
Im{H ( j )}
tan =
Re{H ( j )}
Bode plot of electrostatic actuator

command
bode with previously
defined system sys
> Evaluate only one of TFs
40
20
0
Magnitude (dB)
> Use
Bode Diagram
Matlab
-40
-60
-80
-100
g( s)
H(s) =
Vin ( s )
-120
-140
180
> This tells us how quickly
135
Phase (deg)
we can wiggle tip!

At a certain OP!
-20
90
45
0
-45
-90
-2
10
-1
10
10
10
Frequency (rad/sec)
10
10
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Time
response
Integrate
Take LT
What is g(t)
due to small
changes in
Vin(t)?
Form TFs

functions
response
poles &
zeros
Equivalent
circuit
Natural
system
dynamics
Given O.P.
How fast
can I wiggle
tip?
Poles and Zeros

> For our models, system function is a ratio of polynomials in s
> Roots of denominator are called poles
They describe the natural (unforced) response of the system
> Roots of the numerator are called zeros

They describe particular frequencies that fail to excite any output
> System functions with the same poles and zeros have the
same dynamics
Q0
g( s)
ARm
H(s) =
=
Vin ( s )
1
Q02 1
b 2 1 b k 1 k
3
+ s +
+ s+
2 2
s +
RC
m
RC
m
m
RC
m
A
Rm
0
0
0
where C0 =
A
g 0
> MATLAB solution for poles is VERY long

Pole-zero diagram
> Displays information
tip velocity
Imaginary Axis
about dynamics of
system function
Matlab command
Pole-Zero Map
1
pole
zero
0.5
0
-0.5
pzmap
-1
-9
-8
-7
-6
> Useful for examining
-4
-3
-2
-1
-3
-2
-1
R l Aaxis
i
Real
Pole-Zero Map
1
gap
Imaginary Axis
dynamics, stability,
etc.
-5
0.5
0
-0.5
-1
-9
-8
-7
-6
-5
-4
Real Axis
Real axis
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Time
response
Integrate
Take LT
n
ct
nf s
g e si
Ei aly
An
Form TFs

functions
response
poles &
zeros
Equivalent
circuit
What is g(t)
due to small
changes in
Vin(t)?
Natural
system
dynamics
Given O.P.
How fast
can I wiggle
tip?
Eigenfunction Analysis
> For an LTI system, we can find the eigenvalues and
eigenvectors of the J1 (or A) matrix describing the internal
dynamics
For scalar 1st-order system:
dx
= x
dt
x(t ) = K 0 e t + K1
If we try solution:
x (t ) = Ke t
Plug into DE:
Our linear (or linearized)

homogeneous systems look like:
x (t ) = J1x + J2u(t)
dx
= Ax
dt
d (x )
= J1x
dt
x = Ax
This is an eigenvalue
equation
If we find we can find
natural frequencies of
system
> These are the same as the poles si of the system
> Can solve analytically
Find from det(A-I)=0
> Or numerically eig(sys)

-8.9904
-0.2627 + 0.8455i
-0.2627 - 0.8455i
Linearized system poles

> We can use either i or si to
determine natural frequencies
of system
> As we increase applied

voltage
Stable damped resonant
Increasing voltage
> Plotting poles as system
frequency decreases
changes is a root-locus plot
Spring softening
> Plot damped resonant frequency versus applied voltage
> Resonant frequency is changing because net spring constant
k changes with frequency
k'=k
AV 2
g
This is called
spring softening
Damped resonant frequency
> This is an electrically tuned mechanical resonator

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.01 0.02 0.03 0.04 0.05 0.06
Voltage
Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Time
response
Integrate
Take LT
n
ct
nf s
g e si
Ei aly
An
form TFs

circuit
circuit
What is g(t)
due to small
changes in
Vin(t)?
Form TFs

functions
response
poles &
zeros
Natural
system
dynamics
Given O.P.
How fast
can I wiggle
tip?
Linearized Transducers
> Can we directly linearize
our equivalent circuit?
YES!
> This is perhaps the most
Vin(t) +
-
common analysis in the

literature
> First, choose what is load
1/k
Fout
Source
Transducer
Load
Find OP
and what is transducer
Linearize
Here we include spring

with transducer
Vin(t)
+
-
+
V
Linearized
Transducer
+
Fout
Source
Load
Linearized Transducer Model

> First, find O.P.
V0 , g 0 , Q0
> Next, generate matrix to

relate incremental port
variables to each other
Start from energy and force

relations
Qg
A
Q2
=
k ( g0 g )
2 A
V=
Fout
Linearize (take partials)

g 0
V A
F = Q
out 0
A
Q0
A Q

g
k
> Recast in terms of port variables

I
Q
s
g = U

> Define intermediate variables

Q0
A
C0 =
, V0 =
g 0
C0
> Final expression
1
V sC0
F = V
out
0
sg
0
V0
sg 0 I

k U
s
> Now we want to convert this relation into a circuit
> Many circuit topologies are consistent with this matrix relation
> THIS IS NOT UNIQUE!
i'
-1/k'
1/k
u'
1:
'
Co
F'
2/
Co
Co
1/k
1:
1/k*
1:/2
i
+
Co
1_
1
k_/C
Co
+
o
F
_

Adapted from Figure 5 on p. 163 in Tilmans, Harrie A. C. "Equivalent Circuit Representations of Electromechanical
Transducers: I. Lumped-parameter Systems." Journal of Micromechanics and Microengineering 6, no. 1 (1996): 157-176.
> This is the one used in the text
V Z EB Z EB I
F = Z
U
Z
MO
out EB
Z MS
2 Z EB
= Z MO 1
Z MO
> Uses a transformer

Transforms port variables
Doesnt store energy
> What we want to do now is

identify ZEB, ZMS and , and
figure out what they mean
e2
=
f 2 0
0 e1
1 f
1
V Z EB Z EB I
F = Z
U
Z
MO
out EB
2 Z EB
Z MS = Z MO 1
Z MO
1
V sC0
F = V
0
sg 0
Z MS
V0
sg 0 I
k U
s
2 1
2
k Q0 sC0 k Q0 1
= 1
= 1
k
s g 0 C0 k
s g 0
2
Q
k
Q02
0
= 1
k
k
=
'
s Akg 0
Ag
C0V0
g 0
> C0 represents the

capacitance of the
structure seen from the
electrical port
> It is simply the capacitance

at the gap given by the
operating point
C0 =
A
g 0
> As Vin increases, C0 will

increase until the structure
pulls in
> This is a tunable capacitor

> k represents the effective

spring
> A combination of the

mechanical spring k and
the electrical spring
> This is an electrically

tunable spring!
Spring softening shows up
Q02
k'= k
Ag 0
in k
> As Vin increases, k will

decrease from k (at Vin=0)
to 0 (at Vin=Vpi)
> represents the

electromechanical
coupling
> Represents how much the

capacitance changes with
gap
> A measure of sensitivity
= V0
= V0
=
C
g
C0V0 Q0
=
g 0
g 0
= V0
O. P.
A
g g O. P.
A
g 02
C0V0
g 0
Transfer Functions
> Can use linearized
>
circuit to construct H(s)

using complex
impedances
Usually helpful to
eliminate transformer
> Transformer changes

impedances
Z2 =
Z1
1:
Z1
Z2
1:
Vin
U
m
+
-
C0
2/k
Vin
1/k
+
-
U
m/2
C0
b/
Can now get any transfer

function using standard
circuit analysis
Linearized Transducer Models

> Now we can understand Nguyens filter!
Figure 9 on p. 17 in Nguyen, C. T.-C. "Vibrating RF MEMS
Figure 12 on p. 62 in: Nguyen, C. T.-C. "Micromechanical
Overview: Applications to Wireless Communications."
Filters for Miniaturized Low-power Communications."
Proceedings of SPIE Int Soc Opt Eng 5715 (January 2005): 11-25.
Proceedings of SPIE Int Soc Opt Eng 3673 (July 1999): 55-66.
Small-signal analysis summary

Given O.P.
State eqns
Linearize
Linearized
state eqns
(Jacobians)
Time
response
Integrate
Take LT
n
ct
nf s
g e si
Ei aly
An
form TFs

circuit
circuit
What is g(t)
due to small
changes in
Vin(t)?
Form TFs

functions
response
poles &
zeros
Natural
system
dynamics
Given O.P.
How fast
can I wiggle
tip?
Conclusions
> We can now analyze and design both quasistatic and

dynamic behavior of our multi-domain MEMS
> We have much more powerful tools to analyze linear

systems than nonlinear systems
> But most systems we encounter are nonlinear

> Linearization permits the study of small-signal inputs
> Next up: special topics in structures, heat transfer,
fluids
Review: analysis of a 2nd-order linear system

> Spring-mass-dashpot
.
x
+
-
1/k
+ ek - +
em
- eb + -
d x x
= 1
dt x m (F kx bx )
State
eqns
x = Ax + Bu
y = Cx + Du
Direct Integration in Time

> Example: Spring-mass-dashpot step response
k=m=1;b=0.5;
Step Response
Amplitude
>> A=[0 1;-1 -0.5]; B=[0;1];

>> C=[1 0;0 1]; D=[0;0];
>> sys=ss(A,B,C,D);
>> step(sys)
Position
1.5
1
0.5
Velocity
0
1
0.5
0
-0.5
0
10
15
Tim e (sec)
20
25
Transfer Functions
> Can get TFs from A,B,C matrices
Y ( s ) = CX ( s ) + DU( s )
Y ( s ) = C (sI A ) x (0) + (sI A ) BU( s ) + DU( s )

1
Assume transient has died out (XZIR=0)

No feed-through (D=0)
1
Y( s ) = C ( sI A ) B U( s )
Y( s ) = H ( s )U( s )
1
H ( s ) = C ( sI A ) B
Transfer Functions
> Lets do analytically & via MATLAB
1
s 0 0
k
sI A =
s
0
m
m
1
s
= k
b
+
s
m
m
1 s + b m 1
1
(sI A ) = k
s
m
= s ( s + b m) + k m = s 2 + s b m + k m
1 0 1 s + b m 1 0
1

C(sI A ) B =
1
k
0 1 m s m
1
1 m
=
s
m
X( s )
1
F( s) 2
ms + sb + k
H( s) =
=

X( s )
s
ms 2 + sb + k
F( s)
s 2 + 0.5s + 1
H( s) =
2
s + 0.5s + 1
>> [n,d]=ss2tf(A,B,C,D)
n=
s1
s0
s2
0 -0.0000 1.0000
0 1.0000 -0.0000
d=
1.0000
0.5000
1.0000
Transfer Functions
> Can also construct H(s) directly
using complex impedances and
circuit model
.
1/k
x
+
-
+ ek - +
em
- eb + b
F ek em eb = 0
k
ek = kx = x
s
eb = bx
em = mx = msx
(s)
X
1
= H2 ( s ) =
F( s )
Z( s )
1
=
b + ms + k s
s
=
ms 2 + bs + k
( s ) sX( s )
X
=
F( s ) F( s )
1
X( s )
= H1 ( s ) =
ms 2 + bs + k
F( s )
Poles and Zeros

> For 2nd-order system, easy to get
poles and zeros from TFs
1
2
1 s + s b m + k m
H(s) =
s
m
s2 + s b m + k m
1 ( s s1 )( s s2 )
=
s
m
( s s )( s s )
1
2
where
2
b
k
b

s1,2 =
2m
2
m
m
these are the poles
Spring-mass-dashpot system
> It is a second order system, with
two poles
> We conventionally define
Undamped resonant frequency

Damping constant
Damped resonant frequency
Quality factor
s2 + b
s+k
= s 2 + 2s + 02
k
m
b
=
2m
0 =
s1, 2 = 2 02
For underdamped systems ( < 0 )
s1, 2 = j d
where
d = 02 2
Quality factor :
Q=
0 m 0
=
2
b
Pole-zero diagram
> Displays information about dynamics of system

function
s
H2 ( s ) = 2
s + 0.5s + 1
s1, 2 = 0.25 j 1 1 = 0.25 j 0.97
16
Pole-Zero Map
2
1.5
pole
Imaginary Axis
1
0.5
zero
0
-0.5
-1
-1.5
-2
-2
-1
0
Real Axis
SMD-position frequency response
1
1
H( j ) =
m ( 02 2 ) 2 + 4 2 2
Bode Diagram
0
-10
-20
-30
-40
360
Phase (deg)
H( j ) = atan 2
2
0
Magnitude (dB)
1
1
H( j ) =
m 2 + 2j + 02
10
315
270
225
180
-1
10
10
Frequency (rad/sec)
10
> Find eigenvalues numerically using MATLAB and A matrix
1
0
A=
0
0
.
5
[ V , ] = eig( A)
0
1 0 0.25 + 0.97 j
=
=

0
0
.
25
0
.
97
j
0.707
0.707
V = [v1 v2 ] =
0.18 0.68 j 0.18 0.68 j
Special Topics in Structures: Residual

Stress and Energy Methods
Carol Livermore

Outline
> Effects of residual stresses on structures
> Energy methods
Elastic energy
Principle of virtual work: variational methods
Examples
> Rayleigh-Ritz methods for resonant frequencies and
extracting lumped-element masses for structures
Reminder: Thin Film Stress

> If a thin film is adhered to a substrate, mismatch of thermal
expansion coefficient between film and substrate can lead to
stresses in the film (and, to a lesser degree, stresses in the
substrate)
> Residual stress can also come from film structure: intrinsic
stress
> Stresses set up bending moments that can bend the substrate
> When we release a residually stressed MEMS structure,
interesting effects can ensue
Reminder: Differential equation of beam bending

> Small angle bending:

Adapted from Figure 9.11 in: Senturia, Stephen D. Microsystem
Design. Boston, MA: Kluwer Academic Publishers, 2001, pp. 214.
ISBN: 9780792372462.
> Beam equation:

q = distributed load
w = vertical displacement
x = axial position along beam
d 4w q
=
4
dx
EI
Example: Fixed-fixed beam

> Fixed-fixed beams are common in MEMS: switches, diffraction
gratings, flexures
> Example: Silicon Light Machines Grating Light Valve display

deflects a beam in order to diffract light

Please see: Figure 1.4 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001, p. 7. ISBN: 9780792372462.
> Residual stress in beams can enhance or reduce response to an

applied load, and impact flatness of actuated beam
> Residual stress can be included in the basic beam bending

equation by the addition of an extra term
Residual Axial Stress in Beams

> Residual axial stress in a beam
>
contributes to its bending stiffness

Leads to the Euler beam equation
Thin beam
s0
s0

0 H
P0 =
which is equivalent to a distributed load

d 2w
q0 = P0W = 0WH 2
dx
Insert as added load into beam equation :
P0
s0WH
2WP0 = 20WH
s0WH
d 4w
EI 4 = q + q0
dx
d 4w
d 2w
EI 4 0WH 2 = q
dx
dx

Example: Effect of tensile stress on stiffness
-0.2
Displacement
0
s = 250 MPa
0.2
0.4
100
0.6
0.8
50
25
1
1.2
20
40
60
Position
80
100

100 m long, 2 m wide, 2 m high fixed-fixed silicon beam

Stress impacts flatness: leveraged bending

> Pull-in is modified if the actuating electrodes are away from the
point of closest approach
1)
2)
with stressstiffening
3)
Figure 3 on p. 499 in: Hung, E. S., and S. D. Senturia. "Extending the Travel Range of Analog-tuned Electrostatic Actuators."
Journal of Microelectromechanical Systems 8, no. 4 (December 1999): 497-505. 1999 IEEE.
Buckling of Axially Loaded Beams

> If the compressive stress is too large, a beam will spontaneously
bend this is called buckling
> The basic theory of buckling is in Sec. 9.6.3

> The Euler buckling criterion:
Euler
2 EH 2
=
3 L2
Plates with in-plane stress and membranes

> As with the Euler beam equation, in plane stress can be included
4w
4w
4w N x 2w N y 2w
= P ( x, y )
D 4 + 2 2 2 + 4
+
2
2
x y
y W x
W y
x
Axial stresses in x
and y directions
> When tensile stress dominates over flexural rigidity (thin,

tensioned plate), the plate may be considered a membrane
N x 2w N y 2w
= P
+
2
2
W y
W x
How about cantilevers?

> Example: residually stressed cantilever, where stress is
constant throughout structure
> Before release: stressed cantilever is attached to surface

> After release: cantilever relieves stress by expanding or
contracting to its desired length
> No bending of released structure
How about nonuniform axial stress?

> Nonuniform axial stress through the thickness of a beam creates
a bending moment
> It can arise from two sources

Intrinsic stress gradients, created during formation of the
cantilever material (e.g. polysilicon)
Residual stress in thin films deposited onto the cantilever

> The bending moment curls the cantilever
Example: Cantilever with stress gradient

> Think about it in three steps:
Relax the average stress to zero after release
Compute the moment when the beam is flat
Compute the curvature that results from the moment
Before release
After release
but before bending
-H/2
x
o
Compression
H/2
z
Stress before release
-H/2
-1
Tension
1
x
Compression
H/2
z
Stress after release
but before bending
After bending
-H/2
Tension
x
Compression
H/2
z
After bending

Adapted from Figure 9.13 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic
EI
1
1 EH
2
x =
M x = z x dA = WH 1 and x =
A
Mx
6
2 1
Example: Thin Film on Cantilever
> In this case, the curling does not relieve all the stress
After release
Before release
t
-H/2
-H/2
o
H/2
2
-3
H/2
y
y
Stress before release
Stress after release

and after bending

> See text for math
Barbastathis group, MIT

Courtesy of George Barbastathis. Used with permission.
Outline
> Energy methods
Elastic energy
Examples
Elastic Energy
> Elastic stored energy density is the integral of stress with
respect to strain
Elastic energy density :

When ( ) = E :
(x,y,z)
~
W(x,y,z) =
()d
0
1
~
2
W(x,y,z) = E [(x,y,z)]
2
> The total elastic stored energy is the volume integral of the
elastic energy density
Total stored elastic energy :
~
W = W(x,y,z)dxdydz
Volume
Including Shear Strains

> More generally, the energy density in a linear elastic medium is
related to the product of stress and strain
> A similar approach can be used for electrostatic stored energy

density (1/2)D*E and magnetostatic stored energy density
(1/2)B*H.
~ 1
For axial strains : W =
2
~ 1
For shear strains : W =
2
This leads to a total elastic strain energy :
W=
1
( x x + y y + z z + xy xy + xz xz + yz yz ) dxdydz
2 Volume
Concept: Principle of Virtual Work

> The question: how to determine the deformation that results
from an applied load
F
(F) = ??
> Known: the work done on an energy-conserving system by

external forces must result in an equal amount of stored
potential energy
> Imposing this condition can provide an exact solution to many

problems
For example, if functional dependence between quantities is

known, and you just need to find what the actual values are
Concept: Principle of Virtual Work

> Can approach this from a guessing point of view
Guess values for ; whichever one best equates stored
energy and work done is the right answer
x
Stored energy work done
are guesses
is best guess
x
xx
Deformation
> What if you dont know the functional form of your

deformations/displacements does this still work?
> Yes! You can choose a plausible shape function for the
displacement with a few adjustable parameters and iteratively
guess the constants to best equate stored energy and work
done
Principle of Virtual Work

> Goal: a variational method for solving energy-conserving
problems (a mathematical way of approaching the guessing)
> Define total potential U, including work and stored energy

U = Stored energy - Work done
> A system in equilibrium has a total potential U that is a minimum

with respect to any virtual displacement
No matter what you change, you wont get any closer to
matching work and stored energy
> Requirement: the virtual displacement must obey B.C.

> Nomenclature for small virtual displacements
In the x direction: u
In the y direction: v
In the z direction: w
Math: Principle of Virtual Work

> Consider all possible virtual displacements; evaluate change in
strains
u and xy = v + u
y
x
x
This implies changes in strain energy density
x =
>
~
W = x x + + xy xy +
> The principle of virtual work states that in equilibrium, for any
virtual displacement that is compatible with the B.C.,
~
Wdxdydz
Volume
(F
(F
s,x
u + Fs , yv + Fs , zw)dS
Surface
u + Fb , yv + Fb , zw)dxdydz = 0
b, x
Volume
Differential Version
> The previous equation is equivalent to the following:
~
Wdxdydz (Fs , x u + Fs , y v + Fs , z w )dS (Fb, x u + Fb, y v + Fb, z w)dxdydz = 0
Surface
Volume
Volume
This can be restated in the following form :
U = 0
where
~
U = Wdxdydz (Fs , x u + Fs , y v + Fs , z w )dS (Fb, x u + Fb, y v + Fb, z w)dxdydz
Volume
Surface
Volume
Variational methods
> Select a trial solution with parameters that can be varied
(x, y, z; c1, c2,cn) = trial displacement in x
v(x, y, z; c1, c2,cn) = trial displacement in y
(x,
y, z; c1, c2,cn) = trial displacement in z
> Formulate the total potential U of the system as functions of
these parameters
> Find the potential minimum with respect to the values of the
parameters
U
U
U
= 0,
= 0,....
=0
c1
c2
cn
> The result is the best solution possible with the assumed trial
function
Why Bother?
> Nonlinear partial differential equations are basically very nasty.
> Approximate analytical solutions can always be found with
variational methods
> The analytical solutions have the correct dependence on

geometry and material properties, hence, serve as the basis for
good macro-models
> Accurate numerical answers may require finite-element

modeling
Analytic vs. Numerical

> Analytic variational methods and numerical finite-element
methods both depend on the Principal of Virtual Work
> Both methods minimize total potential energy

> FEM methods use local trial functions (one per element).
Variational parameters are the nodal displacements
> Analytic methods use global trial functions
Outline
> Energy methods
Elastic energy
Examples
Example: fixed-fixed beam, small deflections

> Doubly-fixed beam with a point load at some position along the
beam, in the small deflection limit
> Our present choice: use a fourth degree polynomial trial

solution
w ( x) = c0 + c1 x + c2 x 2 + c3 x 3 + c4 x 4
Boundary conditions : w = 0 and w = 0 at x = 0, L
> Apply boundary conditions:

c0 = c1 = 0 from BC at x = 0
BC at x = L eliminate two more constants
Result is a shape function with one undetermined amplitude
parameter
w ( x) = c4 L2 x 2 2 Lx 3 + x 4

> Formulate total potential energy and find the minimum
> Calculate strain energy from bending
width of beam
total strain energy
d 2 w
= = z 2 = zc4 2 L2 12 Lx + 12 x 2
dx
EW
W=
2
1
3 5 2
=
dxdz
EWH
L c4
H / 2
30
H /2
> Calculate work done by external force applied at x0
(
> This yields total potential energy
Work = Fw ( x0 ) = Fc4 L2 x02 2 Lx03 + x04
1
U=
EWH 3 L5c42 L2 x02 2 Lx03 + x04 Fc4
30

> Minimize total potential energy with respect to c4, determine c4,
and plug in to find variational solution for deflection w(x)
U
=0
c4
L2 x02 2 Lx03 + x04
c4 = 15
F
3 5
EWH L
L2 x02 2 Lx03 + x04 L2 x 2 2 Lx 3 + x 4
(
w = 15
)(
)F
EWH 3 L5
> Compare stiffness for the case of a center-applied load
( )
3
15
L
wL =
F
2 256 EWH 3
256 EWH 3
EWH 3
k=
17
3
15
L
L3
Recall solution of beam equation

EWH 3
k = 16
L3
Properties of the Variational Solution
> Does it solve the beam equation? NO

> Is the point of maximum deflection near where the
load is applied? NOT IN GENERAL
> How can we determine how accurate the solution is?

TRY A BETTER FUNCTION
> Was this a good trial function? NO
A Better Trial Function

> Fifth-order polynomial allows both the amplitude and shape of
0.000
0.000
0.010
0.010
0.020
0.020
Vertical position
Vertical position
the deformation to be varied
0.030
0.040
0.050
0.060
Exact when xo = L/2
0.070
0.080
0.2
0.4
0.6
0.030
0.040
0.050
0.060
Exact when xo = L/2
0.070
0.8
Axial position
0.080
0.2
0.4
0.6
Axial position
0.8
Adapted from Figure 10.1 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001, p. 248.
ISBN: 9780792372462. The artist's representation of the fourth and fifth degree polynomials is approximate.
Fourth degree polynomial
Fifth degree polynomial
What about large deflections?

> For small deflections, pure bending is a good approximation
The geometrically constructed neutral axis really does have
about zero strain
> For large deflections, the beam gets longer

Tensile side gets even more tensile
Compressed side gets less compressed
Neutral axis becomes tensile
> We can treat this as a superposition of two events
First, the beam bends in pure bending, which draws the end
of the beam away from the second support

Then, the beam is stretched to reconnect with the second
support
Quantify the stretching by the strain at the originally neutral
axis
Analysis of Large Deflections

> When deflections are large, on the order of the beam
thickness, stretching becomes important
As a result of stretching, the arc length increases
ds =
-L/2
L/2
0
c
Deformed shape
of original neutral
axis
Segment of original neutral axis
x
x+dx
w(x)
[dx + u( x + dx ) u( x )]2 + [w( x + dx ) w( x )]2

Using the result that 1 + = 1 +
du 1 dw 2
+
ds = dx 1 +

2
dx
dx
The axial strain is given by

ds dx du 1 dw
x =
=
+
dx
dx 2 dx
The change in length is
L/2
w(x+dx)
u(x+dx)
Same segment after deformation
u(x)
L/2
ds dx
L =
dx = x dx
dx
L / 2
L / 2

Example: Center-Loaded Beam

> Potential energy has three terms:
Bending strain energy
Stretching strain energy
External work
-L/2
L/2
Adapted from Figure 10.2 in Senturia, Stephen

D. Microsystem Design. Boston, MA: Kluwer
> Bending and external work already calculated for one trial
function
> Pick another trial function (same weakness as last attempt, but
easy to use) and include large deflections
c
2x
w = 1 + cos
L
2
Why not a ?

> First, calculate the strain due to stretching (aggregate axial
strain)
L
L2
1
=
a =
x dx
L L L 2
L/2
du 1 dw 2
1
a =
dx
+
L L / 2 dx 2 dx
L/2
1 dw 2
1 L L 1
a = u u + dx
L 2 2 L L / 2 2 dx
> Total strain = bending strain + aggregate axial strain

T = bending + stretching
d 2 w
T = z 2 + a
dx

> Calculate total stored elastic energy from total strain
EW
W=
2
H /2 L/2
EWH 4 8 H 2 + 3c 2 c 2
dxdz =
3
96
L
H / 2 L / 2
2
T
> Finally, potential energy
EWH 4 8 H 2 + 3c 2 c 2
U = W Fc =
Fc
3
96 L
> which we minimize with respect to c
U
4 EWH 3 4 EWH 3
=0
c + 3 c
F =
3
c
6 L 8 L
> Compare linear term with solution to beam equation: prefactor
16.2 instead of 16
Results from example

> Force-displacement relationship: an amplitude-stiffened Duffing
spring
4 EWH 3 4 EWH 3
F =
c + 3 c
3
6
L

8 L
> Solution shows geometry dependence; constants may or may

not be correct
EWH 3
EWH 3
+
F = Cb
c
C
c
s
3
3
L
L
> Once youve found the elastic strain energy, finding results for
another load is easy
Work = Fc
L2
Work =
q( x)w (x )dx
L 2
Example: uniform pressure load P

> Adopt the elastic strain energy
> Calculate the work for a uniform pressure load
Work = WP
L/2
2x
c
WLPc
1
cos
dx
+
=
2
2
L
L / 2
> Minimize U to find relationship between load and deflection

4 EH 3 4 EH 3
P = 4 c + 4 c
3 L 4 L
> The geometry dependence appears!
Combining Variational and FEM Methods

> Use the analytic variational method to find a good functional
form for the result
> Establish non-dimensional numerical parameters within the

solution
> Perform well-meshed FEM simulations over the design space

> Fit the analytic solution to the FEM results
Residual Stress In Clamped Structures

> Must add a new term to the elastic energy to capture the effects
of the residual stress
a
~
W = d
~
W=
+ E )d
0
0
> Now there is a residual
stress term in the stored elastic energy
Wr = 0W
H /2
H / 2
L/ 2
dz
dx
a
L / 2
> For the fixed-fixed beam example, the residual stress term is:
>
2 2
Wr = 0WLH 2 c
4L
This leads to a general form of the load-deflection relationship
for beams, which can be extended to plates and membranes
Results for Doubly-Clamped Beam

For the case of a central point load :
2 0WH 4 EWH 3 4 EWH 3
+
F =
c + 8 L3 c
3
L
L
2
6

and for the pressure loaded case :
2 0 H 4 EH 3 4 EH 3
P = 2 + 4 c + 4 c
L 3 L 4 L
The general form for pressure loading, useful for fitting to FEM results, is :
0 H
EH 3
EH
P = Cr 2 + Cb 4 c + Cs 4 c 3
L
L
L
Finally, we note that the stress term dominates over bending when
EH 2
0 2
L
Outline
> Energy methods
Elastic energy
Examples
Estimating Resonance Frequencies

> We have achieved part of our goal of converting structures into
lumped elements
We can calculate elastic stiffness of almost any structure, for

small and large deflections
But we still dont know how to find the mass term associated
with structures
> We can get the mass term from the resonance frequency and the
stiffness
> The resonance frequency comes from Rayleigh-Ritz analysis

In simple harmonic motion at resonance, the maximum kinetic
energy equals the maximum potential energy
Determine kinetic energy; equate its maximum value to the

maximum potential energy; find 0.
Estimating Resonance Frequencies

> Guess a time dependent trial function from (x)
w ( x, t ) = w ( x) cos(t )
> Find maximum kinetic energy from maximum velocity

w ( x, t )
Maximum velocity :
t =
= w ( x )
2
1 2
mvmax
2
1
2
= m ( x ) 2 w (x )
2
Max kinetic energy, lumped : Wk ,max =

~
Max kinetic energy density : Wk ,max
Max kinetic energy : Wk ,max =
2
2
(
)
(
)
x
w
x
dxdydz
m
volume beam
> Calculate maximum potential energy from (x) as before

Rayleigh-Ritz
> The resonance frequency is obtained from the ratio of

potential energy to kinetic energy, using a variational
trial function
> The result is remarkably insensitive to the specific

trial function
=
2
0
Welastic
1
2
m ( x)w ( x)dx
2 Volume
Example: Tensioned Beam

> Compare two trial solutions:
Tensioned wire the exact solution (1/2 of a cosine)
Bent beam a very poor solution
0.00
Deflection
0.02
0.04
Bending solution
0.06
0.08
0.10
0.12
0.14
-0.5
Tensioned wire
-0.25
0
x/L
0.25
0.5

> Resonant frequencies differ by only 15%

> Worse trial functions yield higher stiffness, higher resonant
frequencies
Extracting Lumped Masses

> Use variational methods to calculate the stiffness
> Use Rayleigh-Ritz with the same trial function to calculate
the resonant frequency 2
> Extract the mass from the relation between mass, stiffness,
and resonant frequency.
2 = k/m
Dissipation and
The Thermal Energy Domain Part I
Joel Voldman*
J. Voldman: 2.372J/6.777J Spring 2007, Lecture 12 - 1
Outline
> Thermal energy domain: why we love it and why we

hate it
> Dissipative processes

Example: Charging a capacitor through a resistor
> The Thermal Energy Domain

Governing equations
> Equivalent-circuit elements & the electrothermal

transducer
> Modeling the bolometer
Thermal MEMS
> Everything is affected by temperature

> Therefore, anything can be detected or measured or
actuated via the thermal domain
> Sometimes this is good
MEMS Imagers
> A bolometer heats up due to
Poor residual stress control
incoming radiation
> This results in a temperature

change that changes the
resistance across the pixel
Figure 1 on p. 55 in: Leonov, V. N., N. A. Perova, P. De Moor, B. Du Bois,

C. Goessens, B. Grietens, A. Verbist, C. A. Van Hoof, and
J. P. Vermeiren. "Micromachined Poly-SiGe Bolometer
Arrays for Infrared Imaging and Spectroscopy."
Proceedings of SPIE Int Soc Opt Eng 4945 (2003): 54-63.
Silicon nitride and

vanadium oxide
50 m
0.5 m
IR
Radiation
Y-metal
Better design
2.5 m
B
E
X-metal
Monolithic bipolar
transistor
Figure 2 on p. 56 in: Leonov, V. N., N. A. Perova, P. De Moor, B. Du Bois,

C. Goessens, B. Grietens, A. Verbist, C. A. Van Hoof, and
J. P. Vermeiren. "Micromachined Poly-SiGe Bolometer
Arrays for Infrared Imaging and Spectroscopy.
Proceedings of SPIE Int Soc Opt Eng 4945 (2003): 54-63.
Honeywell
MEMS Imagers
> This application illustrates key features of thermal

MEMS
Excellent thermal isolation creates excellent sensitivity
Response is proportional to thermal resistance
Low thermal mass creates fast response time
Response time is proportional to thermal capacitance
Easy integration with sense electronics
Thermal flow sensing

> A time-of-flight flowrate sensor
> One resistor creates a heat pulse
> A downstream resistor acts as a
temperature sensor
> Time for heat pulse to drift

downstream is inversely related to
Figure 1 on p. 686 in: Meng, E., and Y.-C. Tai. "A Parylene MEMS Flow Sensing Array.
flowrate
Technical Digest of Transducers'03: The 12th International Conference on Solid-State
> This example illustrates the

important benefit
materials
Sensors, Actuators, and Microsystems, Boston, June 9-12, 2003. Vol. 1. Piscataway, NJ:
IEEE Electron Devices Society, 2003, pp. 686-689. ISBN: 9780780377318. 2003 IEEE.
of MEMS
Large range in thermal
conductivities
From vacuum (~0) to metal (~100s
W/m-K)
Figure 3 on p. 687 in: Meng, E., and Y.-C. Tai. "A Parylene MEMS Flow Sensing Array.
Technical Digest of Transducers'03: The 12th International Conference on Solid-State
Sensors, Actuators, and Microsystems, Boston, June 9-12, 2003. Vol. 1. Piscataway, NJ:
IEEE Electron Devices Society, 2003, pp. 686-689. ISBN: 9780780377318. 2003 IEEE.
MEMS flow sensing: commercial

> OMRON flow sensor uses measures
temperature distribution around a
heat source
> Convection alters temperature
Image removed due to copyright restrictions. OMRON flow sensor.
profile in a predictable fashion
Flow
Calm
Thermopile B
Heater
Temperature distribution
The Thermal Domain

Academic
> Sometimes this is bad

> Reason #1
Everything is a temperature sensor
Evaluation of MEMS devices over
temperature is often critical to

success
Figure 12 on p. 115 in: Takao, H., Y. Matsumoto, and M. Ishida. "A Monolithically
> Reason #2
Integrated Three-axis Accelerometer Using CMOS Compatible Stress-sensitive

Differential Amplifiers." IEEE Transactions on Electron Devices 46, no. 1 (1999):
109-116. 1999 IEEE.
Commercial
As we will see, we can never
0.179
Sensitivity (V/g)
0.180
transfer energy between domains

perfectly
The extra energy goes into the
thermal domain (e.g., heat)
We can never totally recover that
heat energy
0.178
0.177
0.176
0.175
0.174
0.173
0.172
0.171
0.170
-30 -20 -10 0 10 20 30 40 50 60 70 80
ADI ADXL320
Temperature (oC)
Outline

hate it


Governing equations

transducer
Charging a capacitor
> We will transfer energy from a power supply to a

capacitor
> Ideally, all energy delivered from supply goes to

capacitor
> In actuality, there is ALWAYS dissipation

And this is true for ALL domains
> Thus, we will lose some energy
Example: Charging a Capacitor

> Use step input
VS
> Voltage source must supply
Vc
twice the amount of energy as

goes into the capacitor
> One half the energy is

dissipated in the resistor,
independent of the value of R!
I
R
+
VS
+
C
-
VC
-
VC = VS (1 e t / RC )
I =C
dVC VS t / RC
= e
dt
R
VR
time
1
Energy stored in capacitor: WC = CVS2
2
Energy delivered by power supply:
VS2 t / RC
PS (t ) = IVs =
e
R
Ws = Ps dt =CVS2
0
Power Considerations: Joule Heating

> The extra energy is lost
to Joule heating in the
resistor
VR2
PR = IVR = I R =
R
For normal "positive" resistors
PR 0
2
> Globally, the power

entering a resistor is
given by the IV
product.
> Locally, there is power

dissipation given by
the product of the
charge flux and the
electric field.
This means
[ ]/m3
2
P%
R = J eE = eE
For normal positive conductivity

P% 0
R
and, one is the integral of the other

PR =
3
P%
d
r
R
Volume
Outline

hate it


Governing equations

transducer
Thermal Energy Domain

> It is easier to put ENERGY
INTO than get WORK OUT of
thermal domain
> All domains are linked to

thermal domain via
dissipation
> Thermal domain is linked to

all domains because
temperature affects
constitutive properties
Elastic
Magnetic
Electric
Chemical
Fluids
Thermal energy domain

Adapted from Figure 11.2 in Senturia, S tephen D. Microsystem Design.
Thermal Energy Domain

> Heat engines convert heat into
mechanical work, but not
perfectly efficiently, just like the
charging of a capacitor cannot be
done perfectly efficiently
> This is a statement of

irreversibility: the 2nd Law of
Thermodynamics
Courtesy of Zyvex Corporation. Used with permission.
Electrical energy
> Zyvex heatuator

One skinny leg and one fat leg
Run a current and skinny leg will
heat up
Structure will bend in response
Thermal energy
Lost to thermal
reservoir
Strain energy and
deflection
Governing Equations
> Some introductory
notation
> Be careful with units

and normalizations
X ~
=X
V
Q
Thermal energy [J]
~
Q
Thermal energy/volume [J/m3]
Q&= I
Q
Heat flow [W]
J Q Heat flux [W/m2]

CV =
CP =
Q
Heat capacity at constant volume (J/K)
T Volume
Heat capacity at constant pressure (J/K)
Q
T
Pressure
CV = C P = C
~ C
C=
V
~ C~
Cm =
Are the same for incompressible materials

Heat capacity/unit volume (J/K-m3)
Heat capacity/unit mass (J/K-kg) AKA specific heat
Governing Equations
> Like many domains,

conservation of energy
leads to a continuity
equation for thermal
energy
n
S
generation

+
of
stuff
in

volume

d
bdV = F ndS + gdV
dt
Get point relation
Divergence theorem and

bring derivative inside
net stuff
d stuff in
= entering
dt volume
volume
b
t dV = FdV + gdV
b
= F + g
For heat
t
~
transfer
dQ
~
+ JQ = P
sources
dt
Types of Heat Flow

> Flow proportional to a temperature
gradient
Heat conduction
J Q = T
> Convective heat transfer

A subject coupling heat transfer to
fluid mechanics
Often not important for MEMS, but
J Q = hc (T2 T1 )
sometimes is
Talk more about this later
> Radiative heat transfer

Between two bodies (at T1 and T2)
Stefan-Boltzmann Law
Can NEVER turn off
J Q = SB F12 T24 T14
The Heat-Flow Equation

> If we assume linear heat conduction, we are led to the heat-flow
equation
J Q = T
J Q = (T )
~
Q
~
= (T ) + Psources
t
dQ% %
=C
For homogeneous materials, with
dT
T 2
1 ~
= ~ T + ~ P
t C
C sources
Thermodynamic Realities
> The First law of thermodynamics implies that entropy is a
generalized displacement, and temperature is a generalized
effort; their product is energy
> The Second Law states that entropy production is 0 for any
process
In practice it always increases
> Entropy is not a conserved quantity.

> Thus, it does not make for a good generalized variable
> Therefore, we use a new convention, the thermal modeling
convention, with temperature as effort and heat flow (power)
as the flow. Note that the product of effort and flow is no
longer power!!!
But heat energy (thermal displacement) is conserved

Just like charge (electrical displacement) is conserved
Outline

hate it


Governing equations

transducer
Thermal sources
> Heat current source IQ is

represented with a flow
source
IQ
> Temperature difference

source T is represented
with an effort source
+
-
Thermal equivalent-circuit elements
> Use direct analogy again

Thermal
conductivity
J Q = T
Electrical
conductivity
Relation between
effort and flow
T 2
~
= ~ T +P
sources
t C
2T = 0
T2 = T1
No heat
storage
(J
n )1 = (J Q n )2
Laplaces Eqn.
Continuity of effort
Continuity of flow
J E = e E = eV
2V = 0
V2 = V1
No
charge
2 storage
(J E n )1 = (J E n )
> Therefore we can derive a thermal resistance

IQ
RT
1 L
RT =
[K/W]
A
For bar of uniform

cross-section
> Plus, heat conduction and current flow obey the same
differential equation
> Thus, we can use exactly the same solutions for

thermal resistors as for electrical resistors
Just change to
> What about other

T1
types of heat flow?
> Convection
IQ
> Use linear resistor

T2
J Q = hc (T2 T1 )
I Q = hc A (T2 T1 ) = hc AT
RT ,conv
1
=
hc A
IQ
RT,conv
> Radiation
Nonlinear with temperature
Large-signal model
IQ
- T1
T2 +
J Q = SB F12 T24 T14
I Q = SB F12 A T24 T14
IQ
T2
T1
> Radiation
Many ways to linearize
We show two approaches
T2>>T1
I Q = SB F12 A (T24 T14 )

T2T1
I Q SB F12 AT24
I Q = ( 4 SB F12 AT23 ) T
IQ
RT,rad
I Q SB F12 A (T1 + T ) T14

4
I Q = ( 4 SB F12 AT13 ) T
> What about energy storage?

> Just like electrical capacitors
Q
CE =
V
store charge (Q),
> We can store thermal energy (Q)
Q
CT =
T
~
CT = Cm mV
> No thermal inductor /
IQ
CT
Electro-thermal transducer
> In electromechanical energy transduction, we introduced the twoport capacitor
Energy-storing element coupling the two domains

Capacitor because it stores potential energy
> What will we use to couple electrical energy into thermal energy?
In the electrical domain, this is due to Joule dissipation, a loss
mechanism
Therefore it looks like an electrical resistor
In the thermal domain it looks like a heat source
Therefore it looks like a thermal current source
Electro-thermal transducer
> Our transducer is a resistor and a
dependent current source
The thermal current source depends

on R and I in the electrical domain
> We reverse convention on direction of

port variables in thermal domain
OK, because IQT does not track
power
Reflects fact that heat current will
always be positive out of transducer
IQ
+
V
IR
R
+
T
> This is not energy-conserving

Dissipation is intrinsic to transducer
> This is not reciprocal

Heat current does not cause a voltage
> Thermal domain can couple back to

the electrical domain
See next time
Outline

hate it


Governing equations

transducer
Different heat transfer components

> How does one know which heat transfer processes are
important for thermal modeling?
Silicon nitride and
vanadium oxide
50 m
0.5 m
IR
Radiation
Y-metal
2.5 m
B
E
X-metal
IQ
Monolithic bipolar
transistor
CT
RT
> Heat input is a current

> Heat loss is a resistor
RT I Q
1
=
T = I Q RT //
CT s 1 + RT CT s
> Heat storage in mass is a capacitor
Tss = RT I Q
= RT CT
Implications
> Increasing RT increases response

This means better thermal isolation
There is always some limiting value determined by radiation
> Given a fixed RT, decreasing CT improves response

time
This means reducing the mass or volume of the system
Tss = RT I Q
= RT CT
Thermal resistance
> What goes into RT?

> RT is the parallel combination of all loss terms
Conduction through the air and legs
Convection
Radiation
> We can determine when different terms dominate
Conduction resistance
> Conduction
Si is too thermally
conductive
SiO2 is compressively
stressed
Try SiN
11 L
2 A
(50 m )
2(20 W/m - K )(0.5 m )(5 m )
RT ,legs = 5 105 K/W
Material
(W/m-K)
Silicon
148
Silicon Nitride
RT ,legs =
20
Thermal Oxide 1.5

Air (1 atm)
0.03
Air (1 mtorr)
10-5
RT ,air =
1 L
A
2.5 m )
(
=
(105 W/m-K ) ( 50 m )( 50 m )
RT ,air = 108 K/W
Conduction through legs dominates
Other resistances
> Convection
RT ,rad =
At low pressure, there

will be no air movement
Convection will not exist
> Radiation
There is transfer
between plate and body
RT ,rad =
1
4 SB F12 AT13
1
3
4 5.67 10 8 W/m 2 K 4 (0.5)(50 m )(50 m )(300 K )
RT ,legs = 1.3 108 K/W
Leg conduction dominates loss
Use case where bodies

are close in temp
Radiation negligible
> Very fast time constant

~ms is typical for
thermal MEMS
CT = C%m mV
= ( 700 J/kg-K ) ( 3000 kg/m3 ) ( 50 m )( 50 m )( 0.5 m )

= 3 109 J/K
RT CT = 1.5 ms
Improving the design
> How can we make
Silicon nitride and

vanadium oxide
responsivity higher?
> Change materials
50 m
0.5 m
IR
Radiation
Y-metal
2.5 m
B
> Decrease thickness or

width of legs, or increase
length
This reduces mechanical
X-metal
Monolithic bipolar
transistor
rigidity
Does convection ever matter?
> Usually not

> But it can come into play in
microfluidics
> The key is whether energy is

transported faster by fluid flow or
heat conduction
> Well analyze this a bit better after

we do fluids
The Next Step
> When dealing with conservative systems, we found

general modeling methods based on energy
conservation
> With dissipative systems, we must always be coupled

to the thermal energy domain, and must address
time-dependence
> This is the topic for the next Lecture
The Thermal Domain II
Joel Voldman*
Outline
> Review
> Lumped-element modeling: self-heating of resistor
> Analyzing problems in space and 1/space
The DC Steady State the Poisson equation
Finite-difference methods
Eigenfunction methods
Transient Response
> Thermoelectricity
The generalized heat-flow equation
> Last time we generated
a general conservation
equation
> Include a flux that

depends on a force
gradient
> And a capacity
b
= F + G
t
~
dQ
~
+ JQ = P
sources
dt
n
F
S
relation
J Q = T
Q
=C
T
The generalized heat-flow equation
> We get a generalized

conduction equation
Assume homogeneous
region
> Applies to
Heat flow
Mass transport (diffusion)
Squeezed-film damping
> Provides a rich set of

solution methods
T
2T = P
sources
t
1
T
D 2T = P
t
C sources
[m2/s]
D= ~
C
[W/m-K]
[J/K-m3]
Thermal diffusivity
Thermal domain lumped elements
> Thermal resistor

Resistance to heat flow
Three types
Conduction
Convection
Radiation
IQ
RT
IQ
CT
CT = Cm mV
> Thermal capacitor

Store thermal energy
Specific heat volume
> Electrothermal transducer

Converts electrical dissipation
IQ
density
into heat current
1 L
RT =
A
+
V
IR
R
+
T
Outline
> Review
Transient Response
> Thermoelectricity
Measuring temperature with the bolometer
> So far, we know how to

convert an input heat flux
into a temp change
Silicon nitride and

vanadium oxide
50 m
0.5 m
IR
Radiation
Y-metal
2.5 m
> How do we convert that

temp change back into
the electrical domain?
CT
X-metal
Monolithic bipolar
transistor
RT I Q
1
=
T = I Q RT //
CT s 1 + RT CT s
+
IQ
RT
Tss = RT I Q
= RT CT
TCR
> Resistance changes with
temperature (TCR)
Beware, TCR is not constant!
> We can use resistor as a
R (T ) = R0 [1 + R (T T0 ) ]
R R0
R =
= R T
R0
hotplate or a temperature
sensor
V2O5
-200
Coupling back into the electrical domain
> We can define a

transducer that uses
TCR to convert back
into electrical domain IQ
> In order to measure

electrical R, we need
to introduce a voltage
& current
+
T CT
RT
R( T) V
R(T ) = R0 (1 + R T )
> This current will

couple back and
induce its own T
Thermo electrical coupling

> This is our prior
electrothermal transducer
IQ
+
V
IR
R
+
IQ
+
T CT
RT
R( T) V
2
I
> We can add in the current R
source due to Joule heating
> The current source is
IQ
T CT
RT
R( T) V
dependent on R, which is
dependent on T, and so on
> What we will want is for IQI2R

Example: Self-heating of a resistor

> First, assume input IQ=0
+
> Two ways to drive the resistor,
V R( T)
current source or voltage

source it sometimes matters
RT
T
=
I Q 1 + RT CT s
T (1 + RT CT s ) = I Q RT
dT
T +
RT CT = I Q RT
dt
I Q = I 2 R = I 2 R0 (1 + R T )
RT
CT
temp change is output
IQ=I2R
> Now, electrical port is input,
Current-source drive
Expand out
into D.E.
T +
d T
RT CT = I 2 R0 (1 + R T ) RT
dt
Plug in for
IQ

> First-order system with feedback results
d T
RT CT = T (1 I 2 R0 R RT ) + I 2 R0 RT
dt
2
I
R0 R RT ) I 2 R0
1
(
d T
+ T
=
dt
RT CT
CT
dy
+ ay = b
dt
I =
RT CT
1 I 2 R0 R RT
TSS , I
)
)
Recognize
D.E. form
Pick out quantities

of interest
I 2 R0
R0 RT I 2
CT
=
=
1 I 2 R0 R RT
1 R R0 RT I 2
RT CT
Collect terms
and rearrange
This blows up when

I2 =
1
R R0 RT
For R>0

> What changes for voltage-drive?
V R( T)
RT
dT
T +
RT CT = I Q RT
dt
+
-
T (1 + RT CT s ) = I Q RT
Same
T.F.
IQ=I2R
RT
T
=
I Q 1 + RT CT s
IQ is now
different
V2
V2
IQ =
=
R R0 (1 + R T )
IQ
Voltage-source drive
V
(1 R T )
R0
This leads to
negative feedback
for R>0
TSS ,V
RT V 2 / R0
=
1 + R RT V 2 / R0
CT
Results of modeling
> A positive TCR resistor driven from a current source can go
unstable fuse effect
> When dealing with the electrostatic actuator, we observed that

very different behavior was found depending on whether the
system was voltage-driven or current-driven
> Here we see that, depending on the way the electrical domain
couples to the thermal energy domain, it is also important to
look at the drive conditions of a system.
Back to the bolometer

> Assume we want to measure
IQ=1 nW with 1% accuracy
> This limits current one can use

for measurement
TSS , I
R0 RT I 2
2
=
R
R
I
0 T
1 R R0 RT I 2
TSS ,V
RT V 2 / R0
2
=
R
V
/ R0
T
2
1 + R RT V / R0
> For Honeywell bolometer,

Rmeas
R0~50 k, R~-2%/K, RT~107

K/W
Rsignal = R RT I Q
> Input signal will create T=10

mK
> This produces Rsignal=2x10-4,
Rmeas
V2
= R RT
0.01 R RT I Q
R0
I Q R0
or a 10 resistance change
> Voltage must be < 0.7 mV
V2
= R RT
R0
100
Outline
> Review
Transient Response
> Thermoelectricity
Space and reciprocal space
> We have thus far focused on big lumped-element

modeling to design and analyze systems
> This isnt the only way to proceed

> We can chop up the model into many small lumped
elements discretize in space
> Or we can approximate the answer using series

methods discretize in reciprocal space
DC Steady State
> The Poisson Equation
> Boundary conditions
T
1
2
D T = P
t
C sources
Dirichlet sets value on
At steady state
boundary
Fixes T (r ) boundary
Neumann sets slope on

boundary Flux
Fixes
dT
dn
1 ~
D T = ~ P
C sources
2
boundary
Mixed sets some function of

value and slope
> The Poisson Equation is linear

Can use superposition methods
Finite-Difference Solution
1 ~
D 2T = ~ P
C sources
d 2T
1 ~
D 2 = ~P
dx
C sources
> We can generate an equivalent

circuit by discretizing the
equation in space
> A numerical algorithm with a

circuit equivalent
> In 1-D, divide bar into N

segments and N+1 nodes
Ie
0
A
T=0
L
x
d 2T
dx 2
xn
T ( xn + h) + T ( xn h) 2T ( xn )
h2
x1
x2
xn-1
xn
xn+1
h
xN+1
x
Equivalent Circuit
> Can create an equivalent
circuit for this equation
T ( xn + h) + T ( xn h) 2T ( xn )
P ( xn )
=
2
h
2 P ( xn )
(Tn+1 Tn ) + (Tn1 Tn ) = h
I2
n-1
~
I S ,n = (hA)P ( xn )
I1
Rn - 1
IS,n
Rn
IS,n + 1
n+1
Rn + 1
Define local
current source
(Tn+1 Tn ) + (Tn1 Tn ) = h
IS,n + 2
(Tn+1 Tn ) + (Tn1 Tn ) + I
Rn
Rn 1
I S ,n
A
S ,n
=0
This is KCL at a node
Define local
resistance
Rn = Rn 1 =
1 h
A
Equivalent Circuit
> Lets apply this to 1-D

self-heated resistor
x1
x2
xn-1
xn
Gn=1/Rn
xn+1
h
xN+1
x
Gn=Gn+1
At node N:
(Tn+1 Tn ) + (Tn1 Tn ) = I
Rn
Rn 1
S ,n
Gn (Tn +1 Tn ) + Gn 1 (Tn 1 Tn ) = I S ,n
GTn +1 + 2GTn GTn 1 = I S ,n
Example: Self-heated resistor

> Set up conductance
matrix
> Solve
> Very appropriate for
MATLAB
> Can even generate

the conductance
matrix with MATLAB
scripts
> At edges, impose

B.Cs
0 ...
1 0
- G 2G - G ...
0 - G 2G
0
2G - G 0
- G 2G - G
0
0
0 - G 2G
0 T1
0

0 T2
1
1
0 T3
1
0 Tn 1

~
0 Tn = hAPo 1

0 Tn +1
1

2G - G 0 TN 1
1
1
- G 2G - G TN
0 0 1 TN +1
0
G = P
Eigenfunction Solution
> This is a standard method for solving linear partial differential
equations
> It leads to what amount to series expansion solutions, discretized in

reciprocal space
> Typically problems converge with only a few terms THIS IS WHY IT IS
USEFUL
2T =
P ( x)
d 2 i
Eigenfunctions of :
=
i i
2
dx
Can use any linear combination of e jkx , including s in( kx) and cos( kx)
Values of k are determined by the boundary conditions
2
Eigenfunctions can be made orthonormal

*
j i dx = ij
Eigenfunction Expansion
Ie
n ( x) = cn sin ( kn x )
sin( k n L) = 0 k n =
Apply BC at x=0, L:
T=0
n
for n = 1,2,3,...
L
1 = ( x)dx = cn2 sin 2 (k n x)dx

0
Plug into DE:
n =1
Eigenfunctions
for this problem:
Normalize:
T ( x) = An n ( x)
Assume:
2
n
L
x
2
sin(k n x)
L
d 2T
P( x)
=
2
dx
P( x)
2
kn An n ( x) =
1
2
k
n An n ( x) =
1
Multiply by
orthogonal
eigenfunction and
integrate:
~
P ( x)
2
k
n An n ( x) m ( x)dx =
0
k Am =
2
m
Am =
1
k m2
~
P ( x)
m ( x)dx
2
mx ~
sin
P ( x)dx
L L
2
mx ~
sin
P ( x)dx
L0 L
L
For uniform
power density:
~
L
P0 2
mx
Am = 2
sin
dx
k m L 0 L
~
P0 2
n
An = 3
1 ( 1)
k n L
T ( x) = An n ( x)
n =1
~
P
2
2
nx
n
T ( x) = 30
1 ( 1)
sin
L
L L
n =1 k n
~
4 L2 P0
1
nx
T ( x) = 3 3 sin
n odd n
L
The Details
Final answer: T ( x) =
Power density:
At x=L/2:
Tmax
4 Po L2
n odd
sin ( n x / L )
n3
I e2 R
I e2
=
Po =
volume e A2
2 2
4 Ie L
= 3
2
e A
1 1
1 33 + 53 +
Even if we consider only the first term in the expansion, we find

Tmax
2 2
1 Ie L
=
2
7.75 e A
2 2
1 Ie L
compared to the exact solution of
2
8 e A
Outline
> Review
Transient Response
> Thermoelectricity
Transient Modeling
> Finite-difference method

Simply add a thermal capacitance to ground at each node of
the finite-difference network. These circuits can be analyzed
with SPICE or other circuit simulators.
n-2
IS,n - 2
Rn - 2
IS,n - 1
n-1
Rn - 1
Rn
IS,n
IS,n + 1
n+1
Rn + 1
n+2
IS,n + 2
C = hA mCm

Boston, MA: Kluwer Academic Publishers, 2001, p. 302.
ISBN: 9780792372462.
Node volume
Transient Modeling
> Finite-difference method

What does the matrix representation look like now?
Tn-1
Tn
Rn-1
IS,n
Rn
Tn+1
Cn
I S ,n + Gn 1 (Tn 1 Tn ) + Gn (Tn +1 Tn ) CnTn = 0

Gn 1 (Tn 1 Tn ) Gn (Tn +1 Tn ) + CnTn = I S ,n
G + CT = P
T = AT + BP
Transient Modeling (continued)
> Eigenfunction method:
T ( x, t ) = T ( x)Y (t )
Spatial response same

as before
dT
1 ~
2
D T = ~ P
dt
C sources
Use impulse response in

time to eventually get
Laplace transfer function
Use separation of
variables to separate
space and time
~
~
P ( x, t ) = Q0 ( x) (t )
Separate
variables
dT
D 2T = 0
dt
[J/m3]
t>0+
2
dY
(
t
)
d
T ( x)
T ( x)
DY (t )
=0
2
dt
dx
1 d 2T ( x)
1 dY (t )
=
=
D
2
Y (t ) dt
T ( x) dx

Time response is a sum
of decaying
exponentials
Time and space are

linked via eigenvalues
1 d 2T ( x)
1 dY (t )
= =
D
2
Y (t ) dt
T ( x) dx
Y (t ) = e t
T ( x, t ) = T ( x)e t
d 2T
D 2 = T
dx
2
T ( x) = An
sin (k n x )
L
n
n
2
2
kn D = kn = =
D L

Match I.C. at t=0 to get
T ( x, t ) = An
n
series coefficients
T(x,0) is related to
instantaneous heat input
and heat capacity
T ( x,0) = An
n
~
Q0
An = ~
C
An ,odd
2
sin (k n x )e nt
L
~
Q0
2
sin (k n x ) = ~
L
C
2
sin (k n x )dx
L0
~
~
Q0 2 2 Q0 2 2 L
= ~
= ~
C L k n C L n
~
4 Q0
n t
(
)
sin
k
x
e
T ( x, t ) =
~
n
n
C
n , odd
Example: Impulse Response
> Uniformly heated bar, an impulse in time

> Result is a series of decaying exponentials in time
z
Uniform internal heat generation

x
y
L
Heat flow from ends

W

~
Q
T ( x, t ) = ~0
C
where
n odd n
nx nt
sin
e
L
n 2 2 D
n =
L2
lower spatial frequencies

decay slower
Using the Eigenfunction Solution

> We can go from solution to
equivalent circuit
> First, we will lump

Heat current conducted out
as output
Choose heat current

source as input P = Q0 (t )
> Then take Laplace
W H T
I Q =
0 0 x
T
dzdy
x
x =0
0 0
~
Q
8
I Q (t ) = WH e nt ~0
L n odd
C
WH
8WH
1
I Q ,n ( s) =
nL 1+ s
Y (s) =
> Then identify equivalent
1
1+ s
This is NOT unique
~
Q0
~
C
X (s)
circuit for 1st order system

X(s)
dzdy
x=L
Y(s)

Using the Eigenfunction Solution

> Each term in the
eigenfunction solution has a
simple circuit representation
> This means that if the

eigenfunction solution
converges with a few terms,
the lumped circuit is very
simple
+
Q0,n(s)
Tn
Cn
Rn
IQ,n(s)

~
Cn = (mode shape )(volume)C
2
~
~
nx
Cn = C sin
LHWC
dxdydz =
n
L
0 0 0
HW L
n 2 2 D
1
= n =
RnCn
L2
1
Rn =
n
Q0,n ( s ) =
L/2
WH
8WH Q0 8WHL Q0
= 2 2
n L C
n D C
8
Q0,n ( s ) = 2 2 (WHL ) Q0
n
8
Q0,n (t ) = 2 2 (WHL ) Q0 (t )
n
A Three-Mode Equivalent Circuit
> For the first three terms in the eigenfunction

expansion, we combine the three single-term circuits
appropriately
100
C3
IS,2(t)
R3
C2
IS,1(t)
10-1
R2
R1
IQ
C1
Magnitude
IS,3(t)
10-2
10-3
10-4
10-3

1 Term
3 Terms
100 Terms
10-2
10-1
100
101
102
103
R1C1
Outline
> Review
> The DC Steady State the Poisson equation
> Transient Response

> Thermoelectricity
Microscale temperature measurement/control
> We have seen that a resistor can be used as

a temperature sensor and hotplate
> There are other techniques to measure or

control temperature at microscale
Couple temperature to material properties
> Sensors
TCR: temperature resistance change
Thermal bimorph: temperature deflection
Thermoelectrics: temperature induced voltage
Coupled Flows
> In an ideal world, one driving
force creates one flux
> In our world, multiple forces

create multiple fluxes
J Q = T
J e = e
Drift-diffusion in
semiconductors or
electrolytes
> In general, all the different

fluxes are coupled
> If you set it up right, the Lij
J e = z n qe Dnn qe n n
n
J i = Lij F j
j =1
matrix is reciprocal
The Onsager Relations
Quantities in the Onsager Relations

> To explain thermoelectrics,
J e = L11 L12T
we must look at coupling

between heat flow and
electric field
JQ
T
= L21 L22T
> This is written in a standard

form
Resistivity
Seebeck coefficient
= e J e + S T
J Q = J e T
Peltier coefficient
Thermal conductivity
Thermocouples
> Analyze the potential gradient around
a closed loop under the assumption
of zero current (Je=0)
> Thermocouple voltage depends on

the difference in Seebeck Coefficient
between the two materials, integrated
from one temperature to the other
> It is a BULK EFFECT, not a junction

effect
Isothermal voltmeter
Hot
S,1
S,2
a
Cold

Adapted from Figure 11.10 in Senturia, Stephen D.
= S T
Tb
Vab = S (T )dT
Ta
> It is possible to make thermocouples

by accident when using different
materials in MEMS devices in regions
that might have temperature
gradients!
Go around the loop

VTC =
TH
S ,2
S ,1 ) dT
TC
VTC = ( S ,2 S ,1 ) T
For small temp rises
MEMS Thermocouples
> Many thermocouples in series
create higher sensitivity (V/K)
> These are known as thermopiles

> In MEMS thermopiles, often use
Al/Si or Al/polySi
> Able to get good thermal

isolation of sensing element
Courtesy of Thermometrics Corporation. Used with permission.
Thermometrics commercial
Si thermopile
> Number of thermocouples is

limited by leg width
Increasing leg width decreases

thermal resistance and thus
temperature response
Conclusions
> The thermal domain is a great way to transfer energy

around
Except that you have to pay the tax
> We can model thermal problems using

Equivalent circuits via lumped element models in space
Big and small
Equivalent circuits via lumped element models in reciprocal
space
For Further Information

> Introduction to Heat Transfer, Incropera and DeWitt
> Analysis of Transport Phenomena, William Deen
> Solid-State Physics, Ashcroft and Mermim
Fluids A
Joel Voldman*
Outline
> Intro to microfluidics

> Basic concepts: viscosity and surface tension
> Governing equations
> Incompressible laminar flow
Microfluidics
> The manipulation and use of fluids at the microscale

> Most fluid domains are in use at the microscale
Explosive thermofluidic flows
Inkjet printheads
We will not cover this regime
High-speed gas flows
Micro-turbomachinery
We will not cover this regime
Low-speed gas flows
Squeeze-film damping
Well do a bit of this to get b for SMD
Liquid-based slow flow
This will be the focus

3-D cutaway of a micromachined microengine.
Photograph by Jonathan Protz.
Microfluidics
> This has been one of the most important domains of

MEMS
Even though most microfluidics is not MEMS
And there are few commercial products
> The overall driver has been the life sciences

Though the only major commercial success is inkjets
> The initial driver was analytical chemistry

Separation of organic molecules
> More recently, this has shifted to biology

Manipulation of DNA, proteins, cells, tissues, etc.
Microfluidics examples
> H-filter
Developed by Yager and colleagues
at UWash in mid-90s
Being commercialized by Micronics
> An intrinsically microscale device

Uses diffusion in laminar flow to
separate molecules
Yager et al., Nature 2006

Courtesy of Paul Yager, Thayne Edwards, Elain Fu, Kristen Helton,
KjellNelson, Milton R. Tam, and Bernhard H. Weigl.
Courtesy of Paul Yager, Thayne Edwards, Elain Fu, Kristen

Helton, KjellNelson, Milton R. Tam, and Bernhard H. Weigl.
Yager, P., T. Edwards, E. Fu, K. Helton, K. Nelson,
M. R. Tam, and B. H. Weigl. "Micro fluidic Diagnostic
Technologies for Global Public Health." Nature 442
(July 27, 2006): 412-418.
Microfluidics
> Multi-layer elastomeric microfluidics
(Quake, etc.)
Use low modulus of silicone

elastomers to create hydraulic valves
Move liquids around

Use diffusivity of gas in elastomer to
enable dead-end filling

Please see: Figure 3 on p. 582 in: Thorsen, T.,
S. J. Maerkl, and S. R. Quake. "Microfluidic Large-Scale
Integration." Science, New Series 298, no. 5593 (October 18,
2002): 580-584.
Fluidigm
Goals
> To design microfluidics we need to understand

What pressures are needed for given flows
How do I size my channels?
What can fluids do at these scales
What are the relevant physics?
What things get better as we scale down
Mixing times, reagent volumes
What things get worse, and how can we manage them
Surface tension
Outline

Viscosity
> When a solid experiences shear stress, it
U
h
and, in the differential limit
deforms (e.g., strains)
Shear Modulus relates the two
> When a fluid experiences shear stress, it

Viscosity relates the two
> Constitutive property describing relationship

between shear stress [Pa] and shear rate [s-1]
> Units: Pa-s
Kinematic Viscosity
* =
Water: 0.001 Pa-s

Air: 10-5 Pa-s
U x
y
A related quantity :
deforms continuously
=*
( mU x )
y

[m2/s]
This is a diffusivity
for momentum
Surface Tension
> A liquid drop minimizes its free energy by minimizing its
surface area. The effective force responsible for this is
called surface tension () [J/m2 = N/m]
> The surface tension creates a differential pressure on the

two sides of a curved liquid surface
2r
(2r ) = P (r 2 )
solving for P
2
P =
r

Capillary Effects
> Surface forces can
2r
actually transport
liquids
> Contact angles

determine what
happens, and these
depend on the wetting
properties of the liquid
and the solid surface.

m gh (r 2 ) = 2r cos
2 cos
h=
m gr
Capillary Effects
> A hydrophobic
valve
Hydrophobic barrier
Fill
Stop
Burst
Surface tension
> The scaling as 1/r is what makes

dealing with surface tension
HARD at the microscale
> Solutions
Prime with low-surface tension
liquids
Methanol (=22.6 mN/m) vs.

water (=72.8 mN/m)
Or use surfactants
Use CO2 instead of air
Dissolves more readily in water
Zengerle et al., IEEE MEMS 1995, p340
Use diffusivity of gas in PDMS

Outline

Fluid Mechanics Governing Equations
> Mass is conserved Continuity equation

> Momentum is conserved Navier-Stokes equation
> Energy is conserved Euler equation
> We consider only the first two in this lecture
Continuity equation
> Conservation of mass
> In this case, for a control or
fixed fluid volume
both S and V are constant in time
> Point conservation relation is

valid for fixed or moving point
m=
m dV
volume
d
m dV = surface
m U n dS
dt volume
m
dV + m U n dS = 0
t
volume
surface
Apply the divergence theorem:
d
bdV = F ndS + gdV
dt
n
F
b
= F + g
t
U
(
)
m
t
dV = 0
volume
which implies
m
+ ( m U ) = 0
t
m
+ U m + m U = 0
t
Material Derivative
> The density can change due
to three effects:
An explicit time dependence

m
(e.g. local heating)
t
Flow carrying fluid through
time
changing density regions
Divergence of the fluid

velocity
> The first two of these are
U m
grouped into the material

derivative
Rate of change for an

observer moving with the
fluid
m ( U )
Material Derivative
> The first two of these

are grouped into the
material derivative
Incompressible
m
+ U m + m U = 0
t
If we define
D m m
=
+ U m
t
Dt
or. more generally, define the operator
D
= + U
Dt t
we can write the continuity equation as
D m
+ m U = 0
Dt
If the density is uniform, then in steady state
U = 0
Momentum Conservation
> We want to write
Newtons 2nd Law for
fluids
> We start with a volume

that travels with the fluid
contains a constant
amount of mass
Material volume
dp d (mU)
=
dt
dt
d
F=
m UdVm
velocity of
dt Vm (t )
F=
surface
> Then go to an arbitrary F = d

m UdV + m U ( U U s ) n dS
dt V ( t )
control volume
S (t )
Must account for flux of
momentum through
surface
flux of mU thru surface
> Pull time derivative into integral
> Cancel terms
> Apply divergence theorem
d
F=
m UdV + m U ( U U s ) n dS
dt V (t )
S (t )
=
m UdV + m U [ U s n ] dS + m U ( U U s ) n dS
t
V (t )
S (t )
S (t )
Leibnizs rule
m UdV + m UU ndS
=
t
V (t )
S (t )
a ndS = adV
S (t )
V (t )
> Final result does not

depend on control volume
=
m UdV + m UUdV
t
V (t )
V (t )
m
= U
+ m U + m
+ U U dV
t
V (t )
DU
F = m
dV
Dt
V
> Add in force terms

> Include body forces
(e.g., gravity)
Fb = m g dV
V
Fs =
s(n)dS
S (t )
> And surface forces

(i.e., stresses)
s(n) = n
Fs =
n dS
S (t )
s(n)
Fs =
dV
V (t )
DU
= mg +
m
Dt
Navier-Stokes Equations
> Substitute in for
= Pn +
= P +
stress tensor
> Compressible
Newtonian fluid
constitutive relation
> Compressible
Navier-Stokes
equations
P* = P m g r
U x
y
2 U + ( U )
3
DU
2
= P + U + ( U ) + m g
m
Dt
3
DU
= P * + 2 U + ( U )
m
Dt
3
Navier-Stokes Equations
> Terms in
compressible N-S
equations
time-dependence
pressure
compressibility
2
+ U U = P + U + ( U ) + m g
m
3
t
inertial
viscous stresses
gravity
Dimensionless Numbers
> Fluid mechanics is full of nondimensional numbers that help
classify the types of flow
> Reynolds number is most

important
> Reynolds number:

The ratio of inertial to viscous
~ ~~
~ ~ ~2~
Re U U = ( A)P + U
Non-dimensionalized
steady incompressible flow
m L0U 0 L0U 0
U0
=
=
Re =
* *
L0
effects
Ratio of convective to diffusive
momentum transport
Small Reynolds number means
neglect of inertia
Flow at low Reynolds number is
laminar
See Deen, Analysis of Transport Phenomena

Outline

Incompressible Laminar Flow

> The Navier-Stokes
equation becomes very
heat-flow-equation-like,
although the presence of
DU/Dt instead of U/t
makes the equation
nonlinear, hence HARD
Navier-Stokes becomes
DU
m
= P * + 2 U
Dt
to obtain a "diffusion-like" equation:
DU
m
= 2 U P *
Dt
As aside on heat convection

> Our heat-flow equation looked like
> Compare to incompressible N-S eqn
> If we allow fluid to moveto
convectwe can include
convection in our heat conservation
> At steady state, we get a relation

that allows us to compare
convective heat transport to
conduction
> This is the Peclet number

> For microscale water flows, L~100
m, U~0.1 mm/s, D~150x10-6 m2/s
T
1 %
2
= D T + P
t
C% sources
DU
m
= 2 U P *
Dt
DT
1 %
2
= D T + P
Dt
C% sources
1 %
2
U T = D T + P
C% sources
4
4
LU (10 m )(10 m / s )
Pe =
~ 0.1
=
2
6
D
0.15 10 m
s
Couette or Shear Flow

> Pure shear flow with a
>
The flow is one-dimensional

U = U x ( y )n x
linear velocity profile

No pressure gradient
dU
+ U U = 2 U P *
dt
> Relative velocity goes to
zero at the walls (the socalled no-slip boundary

condition)
N-S Eqns collapse to the Laplace eqn

U
W
W
2U x
=0
2
y
U x = c1 y + c2
B.C.'s: U x (0) = 0,U x (h) = U
Ux
y
Ux = U
h

Poiseuille Flow
> Pressure-driven flow through a pipe

In our case, two parallel plates
> Velocity profile is parabolic

> This is the most common flow in microfluidics
Assumes that h<<W
y
h
high P
low P
W
Umax
Ux

Solution for Poiseuille Flow

> Assume a uniform pressure
gradient along the pipe
> Assume x-velocity only

depends on y
> Enforce zero-flow boundary

conditions at walls
> Maximum velocity is at center

> Volumetric flowrate is
dP
= K
dx
2U x
K
=
2
y
1
Ux =
y ( h y ) K
2
U max
h2
=
K
8
Wh3
K
Q = W U x dy =
12
0
h
Lumped Model for Poiseuille Flow

> Can get lumped resistor
using the fluidic convention
> Note STRONG dependence

on h
> This relation is more

complicated when the aspect
ratio is not very high
P = effort = KL
12L
Q
P =
3
Wh
12L
RPois =
Wh 3
Development Length
> It takes a certain characteristic length, called the
development length, to establish the Poiseuille velocity
profile
> This development length corresponds to a development

time for viscous stresses to diffuse from wall
> Development length is proportional to the characteristic

length scale and to the Reynolds number, both of which
tend to be small in microfluidic devices
L
* Re
U
LD (time)U Re L
2
L
time
A note on vorticity
> A common statement is to say
that laminar flow has no vorticity
> What is meant is that laminar flow

has no turbulence
> Vorticity and turbulence are

different
> Can the pinwheel spin?

Then there is vorticity
> Demonstrate for Poiseuille flow
1
[ y(h y )]K
Ux =
2
= U
U x
U x
= ny
nz
z
y
K
= n z
(h 2 y )
2
Fluids - B
Joel Voldman
Outline
> Review of last time

> Poiseuille flow
> Stokes drag on a sphere
> Squeezed-film damping
> Electrolytes & Electrokinetic separations
Last time
> Surface Tension
Force at a liquid-fluid interface
U x
y
> Viscosity
Constitutive property relation
D
= + U
Dt t
shear stress to shear rate
> Material Derivative

Time derivative taking into effect
convection
> Mass continuity

> Navier-Stokes Equation
Fundamental relation for
D m
+ m U = 0
Dt
DU
2
m
= P + U + ( U ) + m g
Dt
3
Newtonian fluids
> Reynolds Number

The MOST IMPORTANT
dimensionless number
m LU
Re =
Outline

> Poiseuille flow
Poiseuille Flow
> Pressure-driven flow through a pipe

In our case, two parallel plates
> Velocity profile is parabolic

> This is the most common flow in microfluidics
Assumes that h<<W
y
h
high P
low P
W
Umax
Ux


> Assume
Incompressible
Steady
2
m
+ U U = P + U + ( U ) + m g
3
t
Ux only depends on y
Ignore gravity
> Assume a uniform pressure

gradient along the pipe
> Result is Poissons eqn

> Boundary conditions:
dP
= K
dx
2U x
K
=
2
Relative velocity goes to zero

at the walls
no-slip boundary
condition

> Integrate twice to get solution
> Maximum velocity is at center
> Can get linear flowrate [m/s] and
>
volumetric flowrate [m3/s]

Can get lumped resistor using the fluidic
convention
> Note STRONG dependence on h

> This relation is more complicated when
1
Ux =
y ( h y ) K
2
U max
h2
K
=
8
Wh3
Q = W U x dy =
K
12
0
h
the aspect ratio is not very high

y
h
Umax
Ux

P = effort = KL
12L
Q
P =
3
Wh
12L
RPois =
Wh 3
Development Length
> It takes a certain characteristic length, called the
development length, to establish the Poiseuille velocity
profile
> This development length corresponds to a development

time for viscous stresses to diffuse from wall
> Development length is proportional to the characteristic

length scale and to the Reynolds number, both of which
tend to be small in microfluidic devices
L
* Re
U
LD (time)U Re L
2
L
time
A note on vorticity
> A common statement is to say
that laminar flow has no vorticity
> What is meant is that laminar flow

has no turbulence
> Vorticity and turbulence are

different
> Can the pinwheel spin?

Then there is vorticity
> Demonstrate for Poiseuille flow
= U
Vorticity
1
[ y(h y )]K
Ux =
2
U x
U x
nz
= ny
z
y
K
(h 2 y )
= n z
2
Outline

Stokes Flow
> Steady-state flow in which
inertial effects can be neglected,
Re0
> The result is a vector Poisson

equation
DU
When m
can be neglected
Dt
> Also called creeping flow

> Action is instantaneous
No mass in system
Incompressible: no springs
2 U = P *
> This is a typical approximation

made in microfluidics
Stokes drag on a sphere

> In creeping flow, one can solve for
the flow field around a sphere

placed in an initially uniform flow
field
> This can be used to find the

stresses on the sphere and sum
them to find the total drag
> This is called the Stokes drag

> This is often the predominant
particle force in microfluidic
systems
> See Deens text for derivation
Vz = U
Fd = 6RU
Stokes drag on a sphere

> This is strictly valid ONLY in a uniform
flow
> These are hard to make

> Instead, we can use Ux(y) of the
parabolic flow profile to calculate a
height-dependent drag force
> This approach fails when the particle is

too big
> Instead, take advantage of published

solutions
Shear flow: Goldman et al., Chem. Eng.

Sci. 22, 653 (1967).
U ( y)
Poiseuille flow: Ganatos et al. J. Fluid

Mech. 99, 755 (1980).
Outline

Squeezed-Film Damping
> This is how we will get our b (or R) for the parallel-plate actuator
> The result of motion against a fluid boundary
If the fluid is incompressible, there can be a large pressure rise, so
large back forces result

If the fluid is compressible, it takes finite motion to create a
pressure rise
> In either case, the dissipation due to viscous flow provides a

damping mechanism for the motion
> This is related to lubrication theory

F
Moveable
h(t)
Fixed
The Reynolds Equation

> Assumptions for compressible
isothermal squeezed-film
damping
One-dimensional pressure
gradient: P(r,t) = P(y,t) only

No pressure gradient in z or
along plate (x)
Stokes flow
Poiseuille flow profile in the plane
Ideal gas law
Isothermal (temperature rise due
to compression is small, and heat
flow to the walls is rapid)
No-slip BCs
Rigid plate: h(r,t) = h(t) only
> The result is a version of the
( Ph) h3 1 2 2
=
P
t
12 2
z
y
x

Reynolds equation
Example: Rigid Plate Damping

> Now, assume small motions
Linearize
> The result is (guess what!) the

heat-flow equation
( Ph) h3 1 2 2
=
P
t
12 2
If we linearize:
h = h0 + h P = Po + P
z
y
x
Normalize: =

y
W
p =
P
Po
h02 Po 2 p h
p
=
2
2
ho
t 12W
Suddenly Applied Motion

> We already solved a very
similar problem
Impulse of heat into 1-D

resistor
96 LW 3
F ( s) =
4 3
ho
n =
> Now we have a velocity

impulse: sudden change of
height
1 1
sz ( s )
4
s
n odd n
1+
n
2 2
n ho Po
= n
2
12W
n=1
96 LW 3
b=
4 ho3
> We get a series of 1st-order

terms, as before
2 ho2 Po
c =
12W 2
> Only need 1st term, which

is an RC circuit
> R for viscous damping

> C for gas compressibility
> Details are in the book
R=b
C=
1
bc

Outline

Electrokinetic Phenomena
> It is a coupled-domain problem in which electrostatic forces
result in fluid flow (and vice versa)
> Start with electrolytes, move into double layer, and finally show
how to manipulate the double layer
> It is the driver behind ALOT of early micro-TAS work
Electrolytes
> Electrolytes: liquids with
mobile ions
Ion mobility (cm2/V-s)
zi
Ni =
ui Ci E Di Ci
zi
Flux (cm-2-s-1)
Examples: water, PBS
> Ions can move via

concentration gradients
(diffusion) or electric fields
(drift)
> Macroscopically, the liquid is

approximately charge-neutral
(called quasineutrality)
Valence
Diffusivity (cm2/s)
Concentration (cm-3)
e = zi qeCi 0 in the bulk

i
In neutral regions: 2 =
=0
N + = u+ C+ E D+ C+
For a binary electrolyte
(e.g., NaCl in water)
N = uC E DC
e = qe (C+ C ) 0
Electrolytes
> Surfaces with fixed
charge can lead to net
space charge in the
liquid
> Diffusion competes

with drift, and at
equilibrium, Boltzmann
distribution follows
> This leads to the

Poisson-Boltzmann
equation
Glass
OOOOOO-
Glass
OOOOOO-
+ -+ -+
+ + + + + + -+ -+
+ - + +
+ + +
C+ C-
+ +
+ + - + +
+ +
+ +
+ + + + - + + + +
C+
C-
Ni =
zi
ui Ci E Di Ci = 0 at equilibrium
zi
Ci ( x ) = Ci ,o e
Near the wall: 2 = e
zi qe ( ( x ) o )
k BT
2 =
zqC
i e
i ,o
zi qe ( o )
k BT
Diffuse Double Layer

Express in terms of = 0 :
2 =
zqC
i e
i ,o
zi qe
k BT
Expand for small potential variations
2
q
2
e
zi qeCi ,o +
z
C
i
i ,o
i
k BT i
Assuming the reference region is charge neutral

1
2 = 2
For binary
LD
monovalent
/
z
L
electrolyte
= e D
2
LD
Debye
Length
(e.g., NaCl)
2qe2Ci ,o
qe2
1
2
=
z C = k T
k BT i i i ,o
LD
B
LD=1 nm for
LD ~ 1
Ci ,o 0.1 M NaCl
Double-layer charge
> We can calculate the total charge in the double layer

> It must balance the charge at the wall
Near the wall:
= w e
z
LD
T he charge density is
=
e = 2
LD
2
Total charge per unit area

LD
in diffuse layer
d = e dz =
0
DL charge/area
[C/cm2]
w
LD
w LD L
e
=
= w
wall charge/area
[C/cm2]
Actual Double Layers

> The actual situation is MUCH more complicated
> Some ions are tightly or specifically adsorbed, forming the Stern layer
and screening the wall charge
> The rest distribute in a diffuse double-layer: the Gouy-Chapman layer

> This is an active area of research
Stern layer
(up to ~0.2 nm)
Gouy-Chapman layer
(~1-20 nm)
Potential
Shear boundary
Zeta potential
0
Phase
boundary
(Nernst)
potential
GouyChapman (up to several
potential hundred mV)
Distance from interface
Electroosmotic Flow
> An axial electric field exerts a

force on the charge in the
diffuse double layer, which
drags the fluid down the pipe
Insulating solid
Electric field
Electrolyte
Net charge in diffuse layer
Insulating solid
Analysis of Electroosmotic Flow

> Assume
Creeping flow
One dimensional flow Ux(z)
No pressure drop
Electrical body force
2U x = e Ex
> Express charge density in

terms of wall surface charge
density
e =
> Set up differential equation

=
z
d Ux
w LD
e
=
2
dz
LD
2
LD
w
LD
z
LD
Ex

> This leads to a PLUG-FLOW
profile
~3LD
d 2U x w LzD
=
e Ex
2
dz
LD
Integrate twice, and use boundary conditions:

Ux ( z = ) = 0
Ux(z)
dU x
dz
( ) =
w LD L
L
Ux =
e e

z
z
w LD L
( z ) =
e
=0
h/2
Ex
Plug flow
Ux
~3LD
U0
Ux =
Ex


> This leads to a PLUG-FLOW
profile
> Flow depends on zeta

potential
The potential at the slip

plane
Which is in a different place

than the wall, the Stern layer,
or LD
It is what is measured
Ux =
Ex
w LD L
U0 =
e Ex =
E
x
D
for z > 3LD
experimentally
For h>>LD, one typically

assumes
=0
w =
Electrophoresis
> This is just electroosmosis
around a solid surface
> Each ionic species has its own

mobility

U ep =
E x = epE x
> Therefore, in an electrolyte in

which there is a net electric field,
ions will drift at various rates
> This is the basis of a separation

technology called electrophoretic
separation
For a
large
particle
More
generally
Electrokinetic separation
> This was THE original driver for micro-TAS
TAS = total analysis systems
> Create fully integrated microsystems that would

go from sample to answer
> The KEY enabler was the integration of nonmechanical valves with the separation column
This creates extremely narrow sample plugs
> Also important is the ability to multiplex
> The actual sample prep was (and is) usually

ignored
> This is/was the raison dtre of Caliper & Aclara

> Aclara died, unclear if Caliper will succeed
> We use a channel-crossing
structure to select a sample
plug
> Then we switch voltages to

drag the plug down a
separation column
> In one approach, EO and EP co-
Ls
EO
EP,1
EP,2
exist, but EO dominates
> Different species travel at

different rates separation
Ls = t sep EP ,1 EP , 2 E x
Separation time
Schematic Illustration
> The key discovery was that
liquid samples could be
controlled with voltages
V1 (+)
VJ
Injection
V3 > VJ
> This allows one to valve
V4 > VJ
V2 = 0
and pump liquids
V1 < VJ
Create small sample plugs
Injected sample plug
V3 (+)
V4 = 0
U0
V2 < VJ
V1 < VJ
V3 (+)
ORNL movie
Slower component
U0
V2 < VJ
U0
V4 = 0
Faster component
> The macro technology is
conventional capillary
electrophoresis
Most notably used in the

human genome project
Ls
> Sample loading is the problem

> To separate two species
(ignoring diffusion), we need
> Smaller starting W means we
Ls W
can use a shorter channel
> And get a faster separation
> Conventional capillaries
Richard Mathies (UCB)
have larger W because

injection is not integrated
Though this is always
getting better
> This thus requires a

longer channel
> Microfab also allows for

integration
Courtesy of Richard A. Mathies. Used with permission.
An aside on electrodes
> Current in the electrolyte is carried by ions

> Current in the wire is carried by electrons
> At the surface, something must happen to transfer
this current
> This is electrochemistry and the typical byproduct are

gases bubbles
Electrokinetics
> That is just the beginning

> Dielectrophoresis
Force on dipoles in non-uniform electric fields
Can use AC fields
Can hold things in place
> Other phenomena

Electrohydrodynamics
Electrowetting
Induced-charge electrophoresis/electroosmosis
Comparing EOF & pressure-driven flow
> If you are designing a chip that needs to move liquids

around, which method is best?
> Issues to consider
Flowrate scaling
Liquid composition
System partitioning
Materials
Species Transport
> Flowrate
Water in rectangular SiO2 channel
h varies, W=1000 m, L=2 cm, =800, =50 mV
Drive with Ex=100 V/cm, P=5 psi
Ex
hW
Q = U 0 hW =
Ex
U0 =
Wh3
P
Q=
12 L
Q
h2
P
U0 = =
A 12 L
> Scaling
Pressure-driven flow
2
larger in large channels
10
Due to cubic
10
dependence of flow
resistance
-2
channels
Q (l/s)
EO flow larger in small
10
EO
-4
10
> Both scale equivalently

with channel length
Larger L requires more
voltage and higher
pressure to get same flow
-6
10
Poiseuille
-8
10 -6
10
-5
-4
10
10
-3
10
h (m)

> Other issues typically matter more
> Valving
EOF has built-in valving using E-fields
Poiseuille needs mechanical valves
> Liquid limitations

Poiseuille flow can pump any liquid
EOF ionic strength limits
Debye length depends on 1/C0
Increasing ionic strength decreases LD and thus
EOF
Typically use ~10-100 mM salt buffer
EOF pH limits
pH affects wall charge affects EOF
Typically use pH ~7 buffers

> Materials issues
Poiseuille flow can use any material
EOF requires defined (and stable) surface charge
Best is silica (or at least glass)
Polymers are more difficult
Surface charge can change as molecules
adsorb, etc.
> System partitioning

Both approaches involve off-chip components
Electrodes for EOF
Pumps for Poiseuille flow
Integrating complete lab-on-a-chip usually BAD idea
EOF: Easier to use external electronics
Poiseuille: Pumps hard to make on-chip
> Transport limitations

EOF
Will separate molecules as they are convected
downstream
Species in flow must tolerate E-fields (DNA/proteins

OK, cells not so good)
Plugs remain plugs

Poiseuille flow
Will not separate molecules
Objects in flow must tolerate shear/pressure
(DNA/proteins OK, cells OK depending on
shear/pressure)
Will distort plugs
Whats next
> We will discuss the behavior of the stuff in the liquids

> How to manipulate that stuff at the microscale
> And then we head into system-level issues
Feedback, Noise, etc.
Mass Transport in liquids
Joel Voldman
Outline
> Chemical potential

> Species conservation including convection
> H-filter design & eigenfunction expansion
> Taylor dispersion, the microfluidicists enemy
> Mixing
Chemical potential
> It comes from

thermodynamics

gradients are the driving
force for the movement of
molecules
> It is the electron Fermi level
T ,V
For an ideal solution:
i ( x) = i0 + k BT ln
ci ( x)
ci0
in semiconductors
> At equilibrium, there are no

gradients in
Chemical potential
> We can derive Ficks first law
from the chemical potential
> First, note that there are two

concentration units
> Relate flux to velocity

> Then relate the velocity to a
force f, using a mobility M
> Then the force to a potential

(P ) gradient
[m2/V-s]
U = n E =
[m/s]
[V/m]
ci = N ACi
# # mol
m 3 = mol m 3
J i = ciU i = N ACiU i
U i = Mf = M
P
x
[s/kg]
n
qe
( qe E ) =
n
qe
( qe )
[N]
Chemical potential
> Finally, find flux due to a

chemical potential
gradient
> Can relate diffusivity to

mobility
D
k BT =
M
Einstein Relation
i ( x) = i0 + k BT ln
ci ( x)
ci0
i
c ( x)
= ci Mk BT ln i 0
x
x
ci
J i = ci Mk BT ( ln ci ( x ) ln ci0 )
x
J i = Mk BTci ( ln ci ( x ) )
x
1 ci
J i = Mk BTci
ci x
J i = ci M
ci
ci
= D
J i = Mk BT
x
x
Outline

> Mixing
Species conservation equation

> One more conservation
equation
V
d
bdV = F ndS + gdV
dt
> Flux now includes

convection and diffusion
> Incompressible flow
b
= F + g
t
n
F
ci
= J i + RVi
t
convection
J i = Di ci + ci U i
diffusion
ci
= ( Di ci + ci U i ) + RVi
t
ci
= Di 2 ci ci U i U i ci + RVi
t
ci
+ U i ci = Di 2 ci + RVi
t
Dci
= Di 2 ci + RVi
Dt
Convection-Diffusion Equation
Convective term
> We have seen this equation
before
> We can compare the convective

to diffusive flux terms, and get a
Peclet number again
Now for diffusive vs. convective

mass transport
> For BSA (66 kDA) in microscale

flows, L~100 m, U~1 mm/s,
D~7x10-11 m2/s
> Convection is important because

times
molecular diffusivity is
slower than heat diffusivity and
105 times slower than momentum
diffusivity
107
ci
t
c
convection U i ci U L LU
~
~
~
2
c
diffusion
Di ci D 2
D
L
)(
LU 10 4 m 10 3 m / s
3
Pe =
=
~
10
11 m 2
D
7 10
s
Dheat
~10-4 m2/s for water
Dmomentum~10-6 m2/s for water
Diffusivities
> How can we get diffusivities for
different objects?
> Use mobility due to Stokes drag

> Result is Stokes-Einstein
relation
> Larger particles have smaller

diffusivity
> Often used to get an effective
Ui
k BT
D = Mk BT =
f
Ui
1
=
f = 6RU i
f
6R
D = Mk BT =
k BT
6R
R=45 nm
Rh=44.8 nm
radius (Rh) for a species

Outline

> Mixing
H-filter
> What are the minimum

diffusivity differences that
we can separate?
> How to choose channel

width, length, flowrate
Courtesy of Paul Yager, Thayne Edwards, Elain Fu, Kristen Helton,

Kjell Nelson, Milton R. Tam, and Bernhard H. Weigl.
Used with permission. Please see:
Yager, P., T. Edwards, E. Fu, K. Helton, K. Nelson, M. R. Tam,
and B. H. Weigl. "Microfluidic Diagnostic Technologies for Global
Public Health." Nature 442 (July 27, 2006): 412-418.
H-filter
> First, lets try a quick and

dirty diffusion calculation
> Assume 1-D diffusion

across width of channel
> Ignore convection effects

along length of channel
> No generation terms

> Result suggests that
separation will go as D
ci
t
ci
2 ci
= Di 2
t
x
ci
ci
~ Di 2
~ Di
~ Di
L
U
H-filter
> Can we do better?

> Yes, using eigenfunction
analysis
> Assumptions
Ignore convection
No generation
No concentration gradients
along channel height or
length
ci
t
ci
= Di 2 ci
t
c
2c
=D
t
x 2
1-D diffusion
L
One dilute component in

solvent
h
W
y
x
H-filter
> Initial condition

Solute initially fills part of
channel
> Boundary condition

No solute flux through walls
Initial condition:
c0
x
0
c0 for 0 < x < d

c( x,0) =
0 for d < x < W
Boundary condition:
c
x
= 0 for all t
x = 0 ,W
H-filter
> First, separate variables
> Time response is exponential
> Spatial eigenfunctions are
Y
= Y Y = e t
t
sinusoids
d 2C
D 2 = C
dx
> Must include DC term in series
2
2
C ( x) = a0 + An
sin kn x + Bn
cos kn x
W
W
n =1
c
2c
=D 2
t
x
c( x, t ) = C ( x)Y (t )
kn2 =
H-filter
> Sine does not meet BCs
c
x
> Cosine does
x =0
dC
=
dx
=0
x =0
2
2
0 = An kn
cos kn 0 Bn kn
sin kn 0
W
W
n =1
An = 0
2
2
sin kn x + Bn
cos kn x
C ( x) = a0 + An
W
W
n =1
c
x
x =W
dC
=
dx
0 = Bn kn
n =1
n
kn =
W
=0
x =W
2
sin knW
W
for n = 1, 2,3,
H-filter
> Finally, use eigenfunction expansion to
meet initial concentration profile
c( x, t ) = a0 + Bn
n =1
2
cos kn x e nt
W
t=0
c( x, 0) = a0 + Bn
n =1
c0 for 0 < x < d

2
cos kn x =
W
0 for d < x < W
multiply both sides by eigenfctn & integrate
2
2
2
2
=
+
c
x
k
xdx
a
k
xdx
B
k
x
(
,
0)
cos
cos
cos
cos km xdx
0
m
m
n
n
0
W
W
W
W
n =1 0
0
extract coefficient
W
Bn = c( x, 0)
0
2
cos ( kn x ) dx
W
H-filter
> Get coefficients and
Bn = c( x, 0)
DC term
2
cos ( kn x ) dx
W
d
W
2
Bn =
c0 cos ( kn x ) dx + 0 cos ( kn x ) dx
W 0
d
d
2 c0
Bn =
sin ( kn x ) 0
W kn
Bn =
2 c0
sin ( kn d )
W kn
for n = 1, 2,3,...
cd
1
a0 = c( x, 0)dx = 0
W 0
W
W
H-filter
> Can plot time evolution
for d=W/2
> Lowest-order mode

(n=1) is dominant
2c0 nd nx nt c0 d
sin
cos
e +
W
W W
n =1 n
c ( x, t ) =
n
n = D
W
H-filter
> What wed like to

know is how
separation scales
with D, t, etc.
cout
> We can determine the

concentration of
solute in output
channel
> Solve for case of

d=W/2
1
c( x, t )dx
=
W d d
cout
2
=
W
W /2
2c0
n
sin
2
n =1 n
c ( x, t ) =
c( x, t )dx
n x nt c0
cos
e +
2
n +1
2c0
n x nt c0
2
1
cos
c ( x, t ) =
( )
e +
2
W
n odd n
H-filter
> Only focus on 1st mode
Simplifies math
Is dominant mode
c ( x, t )
> First mode has error at
x
cos e
2c0

Dt
W
c0
2

2c
Dt
x
c
0
0

+
cout =
e
dx
cos
t=0
W W /2
2
W
Need other terms to meet

2
I.C.
W

Dt
c
W
x
2 2c0W
cout = 2 sin e W + 0
W
4
W W
2
cout

2 2c0W W
=
e
2
W
cout
Dt

c0
8 W
= 1 2 e
2
c0W
+
4
Dt
H-filter
> Can also look at all modes
at short time
> Result is that increases as

Dt for short times
2c0
n
sin
2
n =1 n
c ( x, t ) =
n x nt c0
cos
e +
2
Take average over
output channel
cout =
c0 4c0
2
2
1 n t
e
2
n odd n
?
cout
Dt
1.1
W2
H-filter design implications
> Since cout scales with both D and t, cout,1/cout,2 will be

independent of time at short times
> If D1>>D2, then increasing time and decreasing W
cout
0.3
helps
Minimum W is set by
0.25
Pressure drop increases 0.2
Clogging and bubbles
1st mode
0.15
Dt approx
0.1
0.05
0
0
Complete solution
0.5
1
time
1.5
Outline

> Mixing
Taylor dispersion
> Was ignoring convection OK?
Not really
> One can solve the 1-D

convection-diffusion problem
> This is called Taylor Dispersion

Axial convection + transverse
diffusion
> The result is that the plug

spreads out faster than from
simple diffusion
> The apparent diffusivity is K

> EOF does NOT suffer from
Taylor dispersion
Uniform flow field
Pe 2
U 2h2
= Di 1 +
K i = Di +
210 Di
210
Parallel-plate flow channel
U 2h2 h
K i = Di +
f
210 Di W
Rectangular flow channel
Taylor dispersion
> Can determine Ki for
8.5hW
h W
f
2
2
W h h + 2.4hW + W
rectangular channels
> As h/W0, f(h/W) ~7.95

NOT 1
Because of 2-D profile at
10
wall
> But this means a smaller

channel cross-section
and higher U, therefore
possibly more dispersion
Chatwin solution
Approximate model
Two-dimensional result
Isotropic etch profile
Double etch profile
9
8
> This implies that for a
f (h/W)
given h, bigger h/W is

better area small
Approximation
6
5
4
Exact solution
3
2
Parallel plate sol'n
1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
h/W
Convection, diffusion, and mixing

> We can use convection for good as well as evil
> At steady state, fluid mixing time turns into distance
> Short distances from inlet, two fluids appear not to mix
Outline

> Mixing
Mixing
> Mixing is driven by diffusion
> Macroscale mixing uses
turbulence (e.g., stirring) to
reduce length for diffusive mixing
> In liquid microfluidics, there is no

turbulence to decrease mixing
lengths

Cover of Science 285, no. 5425 (July 2, 1999): 1-156.
THEREFORE,
> Microfluidic mixing is EASY

> Microfluidic mixing is HARD
> Mixing length scales with Pe
W2
L ~U
~ Pe W
D
2.5 s for a 50 m channel ( D = 10 -5 cm 2 /s)

40 s for a 200 m channel ( D = 10-5 cm 2 /s)
Mixing
> How does one define mixing?
> No universal definition
> One definition:
When concentration profile is
cmax = c(0, t ) c(W , t )
uniform to within 1% (or 5%)
> For our rectangular channel,

concentration difference is
biggest between x=0 and x=W
2c
nd nx nt c0 d
c( x, t ) = 0 sin
cos
e +
n
W
W
W
n =1
2c0 nd
n
t
=
sin
1 ( 1) e n
W
n =1 n
4c0 nd nt
sin
e
W
n odd n
d
sin e
W
cmax
= 0.01
d
c0
W
W 2 1 400 W
d
2 ln
=
sin
W
d
D
cmax
Tmix

Dt
W
4c0
n
n = D
W
Mixing
2
Tmix
relative
to W
mixing
time relative
to h2/D
/D
> Mixing time scales as expected for semi-infinite diffusion

Tmix
W 2 1 400 W
d
2 ln
=
sin
W
d
D
Tmix
W 2
0.5
D
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.2
0.4
0.6
0.8
d/W
Mixing
> At the microscale various
approaches exist for
reducing diffusion lengths
Depends on how fast you

need to mix
> Approaches trade off

fabrication complexity,
generality, mixing time, etc.
> All find ways to laminate

two fluids
Figure 2 on p. 249 in Miyake, R., T. S. J. Lammerink, M. Elwenspoek, and J. H. J.

Fluitman. "Micro-Mixer with Fast Diffusion." In Micro Electro Mechanical Systems,
1993, MEMS '93: An Investigation of Micro Structures, Sensors, Actuators,
Machines and Systems, February 7-10, 1993. New York, NY: Institute of Electrical
and Electronics Engineers, 1993, pp. 248-253. ISBN: 9780780309579. 1993 IEEE.
Mixing
> 3-D split and recombine
lamination
> Complicated fab

> Typical of early designs
that focused on Si
Figure 1 on p. 442 in Branebjerg, J., P. Gravesen, J. P. Krog, and C. R. Nielsen, C.R. "Fast Mixing by Lamination."
In Micro Electro Mechanical Systems, 1996, MEMS '96: An Investigation of Micro Structures, Sensors, Actuators,
Machines and Systems, February 11-15, 1996. New York, NY: Institute of Electrical and Electronics Engineers,
1996, pp. 441-446. ISBN: 9780780329850. 1996 IEEE.
Mixing
> Laminate in one level of
channels by moving complexity
from fab to packaging

Figures 2 and 4 on pp. 266-267 in Jackman, R. J., T. M. Floyd, R. Ghodssi, M. A. Schmidt, and K. F. Jensen.
"Microfluidic Systems with On-line UV Detection Fabricated in Photodefinable Epoxy."
Journal of Micromechanics and Microengineering 11, no. 3 (May 2001): 263-269.
Passive chaotic micromixer

> Fairly simple to make
pw
> Uses simple pressure-
driven flow
pw
m
x
d
c u
}
}
u c
> Anisotropic boundary

induces anisotropic
flow
> Stroock et al., Science

295(2002):647
w
h
2/q
B
z
0 Cycles:
1/2 Cycle:
x
1 Cycle:
x
100 m
Image by MIT OpenCourseWare. Adapted from Figure 2 on p. 648 in Stroock, A. D., S. K. W. Dertinger,
A. Ajdari, I. Mezic, H. A. Stone, and G. M. Whitesides. "Chaotic Mixer for Microchannels." Science
New Series, 295, no. 5555 (January 25, 2002): 647-651.
'
Passive chaotic micromixer
> Mixing length
A
y = 3 cm
scales with ln(Pe)

Rather than Pe in
pure diffusive
mixing
Cycles 1-5:
0.2 cm
1.9
y = 3 cm
y90 (cm)
100 m
1.7
1.5
1.3
1.1
0.9
0.7
0.5
200 m
10
11
12
13
Cycle 15:
14
In(Pe)
Images by MIT OpenCourseWare. Adapted from Figure 3 on p. 649 in Stroock, A. D., S. K. W. Dertinger, A. Ajdari, I. Mezic, H. A. Stone, and
G. M. Whitesides. "Chaotic Mixer for Microchannels." Science New Series, 295, no. 5555 (January 25, 2002): 647-651.
More info
> Microfluidic flow
Viscous fluid flow, F. White
Low Reynolds Number Hydrodynamic, Happel & Brenner
Gravesen et al., Microfluidics, A Review, JMME 3(1993) 168
Includes lumped resistances for turns, constrictions, etc.
Life at Low Reynolds Number, E.M. Purcell
http://brodylab.eng.uci.edu/%7Ejpbrody/reynolds/lowpurcell.html
Stone et al., Ann. Rev. Fluid Mech., 36(2004) 381.
> Transport
Analysis of Transport Phenomena, W. Deen
Transport Phenomena, Bird, Stewart & Lightfoot
> Taylor Dispersion

Dutta & Leighton, Anal. Chem., 73(2001), 504
Chatwin & Sullivan, J. Fluid Mech., 120(1982), 347
Feedback
Joel Voldman*
Cite as: Joel Voldman, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechanical Devices, Spring 2007.
MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Outline
> Motivation for using feedback

> The uses of (linear) feedback
> Feedback on Nonlinear Systems
Quasi-static systems
Oscillators
Why use feedback?
> For actuators, how do you

know when you have
actuated?
You can calibrate/calculate/etc.,
but what about drifts?
> Adding a sensor can tell you

where you are
> Combining the sensor +

actuator with feedback can
keep you where you are
An optical attenuator
that uses
MEMS actuator
Senses optical output
Uses feedback to
control attenuation
Why use feedback?
> For sensors, feedback can

be used to enhance sensor
response
E.g., keep sensitivity constant
> Must add an actuator to do

this
An accelerometer that
uses
MEMS tunneling sensor
Electrostatic actuation
Uses feedback to control
tunneling current (and
thus gap)
Figure 1 on p. 426 in: Liu, C.-H., and T. W. Kenny. "A High-precision,

Wide-bandwidth, Micromachined Tunneling Accelerometer." Journal of
Microelectromechanical Systems 10, no. 3 (September 2001): 425-433.
2001 IEEE.
Figure 3a) on p. 426 in: Liu, C.-H., and T. W. Kenny. "A High-precision,
Wide-bandwidth, Micromachined Tunneling Accelerometer." Journal of
2001 IEEE.
Feedback in MEMS
> Since MEMS is often concerned with making sensors

for measurements or actuators to do something,
feedback is integral to the subject
> Here we will examine some of the basic uses of

feedback
Limit sensitivity to variations
Speed up system
Stabilize unstable systems
> At the end, we will look at feedback in nonlinear

systems, which is useful for making oscillators
Outline

> Feedback on Nonlinear Systems
Oscillators
Example: A MEMS hotplate

> Used for gas sensor
Membrane
low-stress SiNx, 2x 500 nm
> Heat up active material, which

reacts with gas and changes
resistance
Heater coil, bond pad

TiN/Ti, 200/10 nm
Wet SiO2
1 mm
Si wafer
> Thermal MEMS is used because

Low power
Fast
Arrayable
0.33 mm
330 m
Image removed due to

copyright restrictions.
The Canonical Feedback System

> In controls, terminology refers to the plant, the controller, the
state sensor, and the comparison point
1
Set point
+
_
Error
Sum
K(s)
Control
Actual Output
H(s)
Plant
Controller
Sensed output
1
Output
M(s)
Sensor

Example: micro hotplate

1
Set point
temp
+
_
Sum
Heater resistor
External amp
Error
K(s)
Voltage
Controller
Measured temperature
Temperature
H(s)
1
Output
Plant
M(s)
Sense resistor

Adding Noise and Disturbances

> Noise corrupts the sensor output
> Disturbances modify the control input to the plant
> In some cases, what we want to measure is the disturbance (a
feedback-controlled accelerometer)
Convection
Disturbance 2
+
_
Error
Set point
Sensed output
temp
(noisy)
K(s)
Controller
Voltage
M(s)
Sense resistor
+
+
Hotplate
Disturbed resistor
control
H(s)
Actual
output
Plant
+
_
1
Output
Temperature
3
Johnson noise

Linear Feedback: Blacks Formula

> For a LTI system, we can use Laplace transforms to create an
algebraic closed-loop transfer function
Assume sensor has (perfect) unity transfer function

X in ( s ) X out ( s )
X in (s ) 1
Set point
+
_
Error
K ( s ) [ X in ( s ) X out ( s )]
K(s)
1
H(s)
1
Control
X out ( s ) = H ( s ) Feedback
K ( s ) [XPath
in ( s ) X out ( s ) ]
X out ( s ) =
H ( s) K ( s)
X in ( s )
1 + H (s) K (s)
Blacks formula
1
Output
X out (s )

Adapted from Figure 15.3 in Senturia, Stephen D.
X out ( s ) forward gain

=
X in ( s ) 1 loop gain
Open-Loop Operation
> Control hot plate via calibration
Remember
Assume hotplate has 1st-order response
with s0~400 rad/s (f0~65 Hz)
IQ
Assume controller has no dynamics

As
H ( s) = 0 0
K (s) = K0
s + s0
or drifts in system
Any deviations cause steady-state error
Troom
Tset
Tcal
RT
T
=
=
I Q 1 + RT CT s 1
RT CT
RT CT
+s
+
-
RT
RT 1
> Works great if there are no disturbances

>
CT
K(s)
H(s)
Thotplate
+
Thotplate = Troom + H ( s ) K ( s )(Tset Tcal )

Open-Loop Operation
> Response is sensitive to variations in controller,

plant, and disturbances
Troom=23 C, K0=1.2
40
A0=1
Tcal=23 C
Troom=25 C, K0=1
38
Troom=23 C, K0=1
36
Temp (C)
34
32
30
28
Tset=37 C
26
24
22
0
0.2
0.4
0.6
0.8
tim e (s)
Feedback use #1: limit sensitivity to variations

Troom
> Add in term

proportional T
set
to error
> This is called

proportional
control
+
K(s)
H(s)
Thotplate
+
HK
1
Tset +
Troom
1 + HK
1 + HK
A0 s0
K0
s + s0
1
=
Tset +
Troom
A0 s0
A0 s0
K0
K0
1+
1+
s + s0
s + s0
Thotplate =
Closed-loop TF Thotplate =
A0 K 0 s0
s + s0
Tset +
Troom
s + s0 (1 + A0 K 0 )
s + s0 (1 + A0 K 0 )
Close the loop
> Error 0 as K0 increases,

despite
Variations in device (A0)
Variations in plant (K0)
Disturbances (Troom)
Open-loop
40
Set-point
38
Temp (C)
36
> In limit of large K0, system

responds perfectly
Though DC error never goes
exactly to zero
42
K0=10
34
32
K0=1
30
28
Heater A0=1.2
Troom=25 C
26
24
0
0.05
0.1
tim e (s)
0.15
Tset=37 C
Steady-state (DC) Error
( s ) s =0 = Tset Thotplate =
0.2
1
1
Tset
Troom
1 + A0 K 0
1 + A0 K 0
Feedback use #2: increase system bandwidth

42
> Settling time goes down
Thotplate
Tset
A0 K 0 s0
= H cl ( s )
s + s0 (1 + A0 K 0 )
36
Temp (C)
A0 s0
s + s0
K0=10
34
32
K0=1
30
28
Heater A0=1.2
Troom=25
26
24
0
s0 s0 (1 + A0 K 0 )
Set-point
38
> Bandwidth goes up

H ( s) =
Open-loop
40
0.05
0.1
tim e (s)
0.15
0.2
Controlling a 2nd-order system

> Vibration sensor
Really just an z-axis accelerometer
> Use feedback to keep gap constant

> In this case, control signal
measures vibration
> Mechanical plant is a SMD
Figure 4 on p. 435 in: Bernste in, J., R. Miller, W. Kelley, and P. Ward.
"Low-noise MEMS Vibration Sensor for Geophysical Applications." Journal of
Microelectromechanical Systems 8, no. 4 (December 1999): 433-438.
1999 IEEE.
X out ( s ) 1
1
=
F (s)
k s 2 + 1 s + 1
Q
m0
where: s = s , 0 = k , Q =
m
b
0
H ( s) =
Set Q=1/2 (critically damped)

Set k=1 for convenience
Figure 6 on p. 435 in: Bernste in, J., R. Miller, W. Kelley, and P. Ward.
"Low-noise MEMS Vibration Sensor for Geophysical Applications." Journal of
Microelectromechanical Systems 8, no. 4 (December 1999): 433-438.
1999 IEEE.
Proportional control of 2nd-order system

> Use ideal controller K(s)=K0
X out ( s )
H ( s) K ( s)
=
=
> This gives us two overall poles:
Two from SMD H(s)
None from controller K(s)
X in ( s )
1 + H (s) K (s)
order system
Q of closed-loop response
increases with increasing DC

gain
1.5
K0=100
> This means that our critically

damped system is now
underdamped
1
s + ( K 0 + 1)
Q
Qcl = Q K 0 + 1
Decreasing DC error as K0
> Some differences:
s 2 +
0,cl = K 0 + 1
> Some results are same as 1stincreases

System speeds up
K0
K0=10
0.5
This can be bad or fatal for our

system
0
0
10
Control of complex systems
> Dynamics of closed-loop system are determined by

H(s)K(s)
> Thus, behavior seen with 2nd-order SMD system will

also occur with 1st-order thermal system coupled to
1st-order controller
> What happens when we add an additional pole?
Single-Pole Controller (Real amp)

> Take SMD and control
with
1st-order
controller
> The system now has

three poles
> When going to large K0,

the system goes
unstable
This happens if one of

the roots has real
positive part
> Routh test can be used
K0
K ( s) =
1 + s
= 0
controller
time constant
X out ( s)
K0
= 3
X in ( s ) s + (1 + 2 )s 2 + (2 + )s + K 0 + 1
Routh test for third - order system :
a3 s 3 + a2 s 2 + a1s + a0
All coefficients ( an ) must have same sign
AND a2 a1 > a0 a3
to find maximum gain
K0 <
11
1
+ +
QQ

2nd-order vs. 3rd-order systems

> Stable system at all loop
> Unstable system at
gains
sufficiently high loop

gain
5
15
^ = 1
Imaginary
Imaginary
10
0
-5
-10
-15
-3
-2
-1
Real
-5
-6
-5 -4 -3 -2 -1
Real


Effect of controller bandwidth

> Controller bandwidth < Plant
bandwidth causes controller to
dominate overall response
=1
Phase (deg)
Magnitude (dB)
Bode Diagram
=100
Bode Diagram
50
-50
-50
-100
-100
0
-45
-90
-135
-180
-225
-270
10-2
Q=1/2
0=1
K0=6
10-1
100
Frequency (rad/sec)
101
Controller
-150
0
-45
-90
-135
-180
-225
-270
2
10
10-4
Plant
10-2
100
Frequency (rad/sec)
102
Overall

.
Cite as: Joel Voldman, course materials for 6.777J / 2.372J Design and Fabrication of Microelectromechan
ical Devices, Spring 2007.
Effect of controller bandwidth

> Controller bandwidth > Plant
bandwidth causes plant to
dominate overall response
=1
Q=1/2
0=1
K0=6
Phase (deg)
Magnitude (dB)
Bode Diagram
Bode Diagram
50
0
-50
-50
-100
-150
-100
0
-45
-90
-135
-180
-225
-270
10-2
=0.01
10-1
100
Frequency (rad/sec)
101
Controller
102
-200
0
-45
-90
-135
-180
-225
-270
10-2
Plant
100
102
Frequency (rad/sec)
104
Overall

PI control
Proportional - Integral (PI) Control :
> Add pole at s=0

K ( s ) = K 0 1 +
s
> This gives K(0)

And thus no DC error
> Benefits
As long as 0, will get
perfect DC tracking, but
it may take awhile
1.8
1.6
K0=100, =0
1.4
1.2
K0=1, =2
Completely insensitive
0.8
to changes in plant at
DC
0.6
K0=1, =0.1
K0=1, =0
0.4
> Drawbacks
Additional pole means
possibility of ringing
and instability
0.2
0
0
20
40
60
80
100
PID control
> Final generic term is to
Proportional - Integral - Differential (PID) Control :

K ( s ) = K 0 1 + + s
s
add in differential
feedback
Anticipate future
10
8
instability due to
integral and
proportional control
> Methods exist to tune

PID controllers
Imaginary Axis
> Tame ringing and
= 0.03
= 2.2
=0
= 2.2
0
-2
-4
-6
-8
-10
-1.5
-1
-0.5
Real Axis
0.5
Imag e by MIT OpenCourseWare.
Stabilization of unstable systems

Piyabongkarn (2005), IEEE Trans. Control Systems Tech.
> Use of feedback

#3: Stabilize an
unstable system
> Stabilize
Figure 2 on p. 139 in: Piyabongkarn, D., Y. Sun, R. Rajamani, A. Sezen, and B. J. Nelson. "Travel Range
Extension of a MEMS Electrostatic Microactuator." IEEE Transactions on Control Systems Technology 13, no. 1
(January 2005): 138-145 2005 IEEE.
.
electrostatic
actuator beyond
pull-in
> Most approaches

use feedback to
approximate
charge control
Figure 6 on p. 140 in: Piyabongkarn, D., Y. Sun, R. Rajamani, A.

Sezen, and B. J. Nelson. "Travel Range Extension of a MEMS
Electrostatic Microactuator." IEEE Transactions on Control Systems
Technology 13, no. 1 (January 2005): 138-145. 2005 IEEE.
no feedback
Figure 11 on p. 144 in: Piyabongkarn, D., Y. Sun, R. Rajamani, A.

Sezen, and B. J. Nelson. "Travel Range Extension of a MEMS
Electrostatic Microactuator." IEEE Transactions on Control Systems
Technology 13, no. 1 (January 2005): 138-145. 2005 IEEE.
with feedback
Stabilization of unstable systems
> However:
All potentially unstable modes must be both observable and
controllable
Observable means that the sensor provides state information
about the mode
Controllable means that the control inputs can modify the
mode
If a mode has both attributes, it can be stabilized (at least in
theory) with feedback
> Adding sensors to a system improves observability of

modes
> Adding actuators improves controllability

> This can be generalized from unstable to unwanted
Control for MEMS

> Electrostatic traps for cells
Exploded Pixel
> The goal is to trap single cells at
IMD hole
Metal 2
Assembled Pixel
each site
PECVD SiO2
Metal 1
> System is currently run open loop
Thermal SiO2
Silicon
> Could we do better if we ran closedloop?
> Need to sense: optical or electrical

> Need to actuate
Bead
This is hard
Cell
100 m
Control for MEMS
> Many MEMS devices/systems are run open loop

why?
> Open loop

Does not need additional sensors or actuators
These increase fab complexity, chip size, cost, etc.
But is sensitive to perturbations
> Closed loop

Requires extra complexity
More stable performance
> If you dont need closed-loop control, dont use it

Outline

> Feedback of Nonlinear Systems
Oscillators
Feedback in Nonlinear Systems

> Can no longer use nice algebraic forms
> However, the same idea still holds:
The controller pre-distorts the control signal so as to compensate
for nonlinearities in the plant
X/X0
F
X out = X 0 tan 1
F0
/2
/2
-20 -15 -10 -5
1
Error
Error
C ontrol
F=K0(Xin-Xout)
C ontrol
Adapted from Figure 15.11 in Senturia, Stephen

D. Microsystem Design . Boston, MA: Kluwer
Academic Publishers, 2001, p. 412.
ISBN: 9780792372462.
K0
Xin
In1
0 5 10 15 20
F/F0
f(u)
Plant
1
Out1
Xout
Controller
Feedback Path
Feedback in Nonlinear Systems
> Controller linearizes in nonlinear system

> This also occurs in op-amps
1.0
K ( X X out )
X out = X 0 tan 1 0 in
F
0
0.04
0.03
0.02
0.5
X in X out
0.0
Error
X
F0
tan out
K0
X0
0 for K 0 >> F0
X in X out =
Xout
0.01
-0.01
-0.02
-0.5
-0.03
-1.0
-1
-0.5
0.5
-0.04
Xin
Imag e by MIT OpenCourseWare
Adapted from Figure 15.12 in: Senturia, Stephen D. Microsystem Design. Boston,MA:
Resonators, Oscillators and Limit Cycles

> Resonator: a passive element that exhibits underdamped
oscillatory behavior
> Oscillator: a resonator plus an active gain element that

compensates the resonator losses and results in steady
oscillatory behavior
> Limiting: a required nonlinearity in either the resonator or gain

element
> Limit Cycle: stable closed path in state space
Example: Resonant RLC Circuit

> While a linear amplifier can theoretically produce an undamped
linear system, it cannot create an oscillator
+ vC -
vs
+
-
iL
vo
L
R
+ vC C
iL
+
L
R
dvC 1
= iL
dt
C
dvC 1
= iL
dt C
diL 1
= (vS vC iL R )
dt L
v
diL 1
= ( v0 vC v0 ) = C
dt L
L
vo
Example: Resonant RLC Circuit

> No stable limit cycle
> Vary gain A of op-amp circuit
2000
Vc
20
A>1
1000
10
-1000
-10
-2000
-2000
20
-20
-20
2000
20
A=1, Vc0=11.5
10
10
-10
-10
-20
-20
-20
-20
20
IL
A<1
20
A=1, Vc0=5
20
Adding a limiter creates an oscillator

> Add in arctangent
state
eqns.
limiter
v
vs = V1 tan 0
V2
1
dvC 1
= iL
dt C
Ai R
diL 1
= V1 tan 1 L vC iL R
dt L
V2
For
+ Vc+
iL
Amplifier
Limiter
V5
V0

AV1
>1
V2
iL R
Ai R
V1 tan 1 L
V2
Net gain
Net loss
iL R
Marginal Oscillator
15.0
> Gradual limiting leads to

nearly sinusoidal
waveforms for weakly
damped resonators
iL
5.0
-10.0
-15.0
90
4
10
Damped
92
94
96
98
100

Adapted from Figure 15.17 in: Senturia, Stephen D. Microsystem Design
.
103
102
15.0
101
10.0
100
10-1
10-2
0
0.5
1.5
2
Frequency (Hz)
2.5
Capacitor voltage
Power spectral density
0.0
-5.0
105
10-3
Damped
10.0
5.0
0.0
-5.0
-10.0
-15.0
-15
-10

Boston,MA: Kluwer Academic Publishers, 2001, p. 419. ISBN: 9780792372462.
-5
10
15
Inductor current
Oscillator Parting Comments
> Do not confuse a resonator with an oscillator

> The oscillator is the combined result of a resonator
with a suitably designed circuit
> The oscillator is intrinsically nonlinear

> The limit cycle obeys its own dynamics, which can be
discovered by analyzing the perturbation of a limit
cycle and the time required to recover
Conclusions
> When properly designed, feedback can
Reduce sensitivity to variations

Decrease response time of system
Control output with zero DC error
Stabilize unstable systems
> But it may be too complicated or unnecessary for

your MEMS part a systems issue
> All elements in the feedback path have poles, and

these can cause instabilities
> Numerous methods exist to analyze control systems

in frequency, time, root-locus, and state-space
domains
Noise
Carol Livermore
Outline
> Where does noise come from?
> Interference and how to deal with it
> Noise definitions and characterization
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
Electronics (diodes, amplifiers)
Resistance thermometer
> Modulation
What limits measurements?

> Fundamental physics of the sensing mechanism, or interactions
among parts of your sensor
> Example: using a resistor to sense temperature

Apply a voltage across a resistor and measure the current,
which tells you resistance

Relate change in resistance to change in temperature
But measuring resistance dissipates some heat in the
resistor, thereby changing the temperature
> Solutions:
Make sure that the temperature change that you impose is
less than the temperature change that you need to measure

Or, use a different sensing technique
> But how low can you lower the voltage and still get enough
signal?
An Example: Tunneling Accelerometer
Figure 1 on p. 236 in Liu, C.-H.,

A. M. Barzilai, J. K. Reynolds, A.
Partridge, T. W. Kenny, J. D. Grade,
and H. K. Rockstad. "Characterization
of a High-sensitivity Micromachined
Tunneling Accelerometer with
Micro-g Resolution." Journal of
Microelectromechanical Systems 7,
no. 2 (June 1998): 235-244.
1998 IEEE.
Tunneling accelerometer
Figure 2 on p. 236 in Liu, C.-H., A. M. Barzilai, J. K. Reynolds,

A. Partridge, T. W. Kenny, J. D. Grade, and H. K. Rockstad.
"Characterization of a High-sensitivity Micromachined
Tunneling Accelerometer with Micro-g Resolution." Journal of
Microelectromechanical Systems 7, no. 2 (June 1998): 235-244.
1998 IEEE.

"Characterization of a High-sensitivity Micromachined Tunneling
Accelerometer with Micro-g Resolution." Journal of
1998 IEEE.
Tunneling current is exponentially sensitive to separation of

tip and substrate. WITH FEEDBACK, this offers very
sensitive measurements of acceleration.
How sensitive? Must consider sources of NOISE.
Types of Noise Sources
> Device noise: statistical fluctuations in devices

Johnson noise, shot noise, 1/f or flicker noise
These noise sources seem immensely unlikely in the face of
our usual assumptions, but they are real and limit results.
> Interference: corruption of the signal of interest by

interference from another signal
Power line, radio stations, TV stations, ground loop inductive
effects
> Drifts: variations in device characteristics
Sources of noise in the tunneling accelerometer

> Amplifier noise
> Electrons are quantized, so
tunneling current is inherently
digitized
An apparently constant
current really isnt gives
noise
> The air molecules in the gap

arent perfectly evenly
distributed (Brownian motion)
Can build up a force from
fluctuations

"Characterization of a High-sensitivity Micromachined
Tunneling Accelerometer with Micro-g Resolution." Journal of
> Drifts? Interference? Air

currents? Package creep?
Outline
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
> Modulation
Interference and noise are different

> Interfering signals:
When an external signal couples into your circuit in an
unintended way
Capacitive pickup of power line signal, input feeding directly

through to output, etc.
Often reflects a circuit design problem that you can prevent

or remedy
Math looks like superposition and mixing of signals

> Random noise:
Random noise is random!
Statistical fluctuations in voltages, currents, forces,
pressures, etc.
Math is simpler with random, uncorrelated signals

The Interference Problem
> Power-line interference can result from grounding

problems or capacitive coupling
Capacitive coupling through
power-line transformer
60 Hz power line
DC Power
Supply
Power-line
ground
Stray capacitance
V0
+
_
Signal ground
R0
Amp
Load
Rc
Amplifier ground Load ground

(same as power line)

Shields
> A shield is a grounded conductor that surrounds the space
being shielded
> But where to connect the ground?

Amplifier shield
60 Hz power line
DC power
supply
Source shield
V0
+
_
Signal ground
Rc
R0
Amp
Load
Rs
Potential difference
Amplifier ground
(power line)
Load ground

Ground Loops
> Multiple grounds

create ground loops
> Large inductive
V0
+
_
Rc
R0
Amp
Rs
EMFs can be
created
Ground loop
V0
+
_
Rc
R0
Amp
Rs
Ground strap
(low resistance and low inductance)

Boston, MA:Kluwer Academic Publishers, 2001, p. 428. ISBN: 9780792372462.
Guards
> A guard is a shield driven at the common-mode

potential of the signal of interest
> Guards can be used to intercept surface currents or

to nullify the effect of wiring capacitance
Signal line
+
_
Oxide
p-Type substrate
n Diffusion

Interference in the tunneling accelerometer?

no. 2 (June 1998): 235-244.
1998 IEEE.
Outline
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
> Modulation
Noise characterization
> Consider a timevarying, random noise

voltage vn(t)
> Also have a signal vs(t)

> Random noise has zero
average, but a non-zero
mean-square average
> The mean square noise

from uncorrelated
noise sources adds
vs (t ) = signal
vn (t ) = noise
t / 2
1
vn = lim
vn (t )dt = 0
t t
t / 2
t / 2
1
2
2
[
]
(
)
vn = lim
v
t
dt 0
t t n
t / 2
v = v s + vn
v = vs
v 2 = vs2 + vn2
Signal-to-Noise Ratio
> Signal-to-noise ratio: a

power ratio
> Normally expressed in

logarithmic coordinates
> Decibels is usual unit

> Can use power ratio, or
rms amplitude ratio
S/N =
vs2
v
2
n
(linear scale)
decibel scale :
vs2
S/N = 10 log 2
v
n
vs ,rms
S/N = 20 log
v
n
,
rms
Spectral Density Function

> We imagine an ideal narrow-band filter that allows all frequency
content within a small band around a selected frequency to pass
> The mean square output of this filter is assumed to be

proportional to the bandwidth, provided the bandwidth is small
enough. This is equivalent to saying that the spectral content of
the signal of interest has no impulses in it (hence, is not a perfect
sine wave)
> The spectral density function expresses that proportionality
vo2 ( f o , f ) = Sn ( f o )f
vn2 = Sn ( f ) df
0
Noise in Linear Systems
> The spectral density function at the output is the

spectral density at the input times the squared
magnitude of the transfer function
2
S o ( f ) = H ( j 2 f ) S n ( f )
Vn
+
_
H (s)
+
V0
_
v02 = H ( j 2f ) Sn ( f ) df

Outline
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
> Modulation
Thermal Noise
> Imagine a damped harmonic oscillator that is not driven
mx + bx + kx = 0
> Any initial motion will be damped out
> BUT this cannot be an accurate description of reality
> Equipartition: in thermal equilibrium, each energy storage
mode has an average energy kBT/2
> Energies mv2/2 and kx2/2 cannot be zero!

> Correct equation has a noise force
mx + bx + kx = f n (b, t )

Thermal Noise (Johnson Noise)

> Any dissipative process coupled to a thermal reservoir results
in fluctuations, even in equilibrium
> Example: thermomechanical noise force

> Example: electrical resistors have a fluctuating zero-average
noise voltage
> The spectrum of Johnson noise is white (at least at all

frequencies of interest)
R (noiseless)
R (noisy)
+
_
Vn
Sn ( f ) = 4k BTR

Noise Bandwidth
> The noise bandwidth, when noise sources are white,

measures the magnitude of the mean-square noise.
vn2 = S n ( f ) df
0
Consider an ideal filter of bandwidth f

The mean - square noise from a resistor that can pass this filter is
vn2 = 4k BTRf
Thermal Noise: Other Domains

> Our lumped element view of the world makes this very easy to
apply to other domains
> Electrical (R = electrical resistance):
Sn ( f ) = 4k BTR
vn2 = 4k BTRf
> Mechanical (b = damping):
S n ( f ) = 4k BTb
f n2 = 4k BTbf
> Fluidic (R = fluidic resistance):
S n ( f ) = 4k BTR
pn2 = 4k BTRf
A good reference: T. Gabrielson, IEEE Trans. Elec. Devices, 40 (1993).

Noise on Capacitors
> When a capacitor is connected
to a resistor that is connected

to a thermal reservoir, the
mean-square noise on the
capacitor is kBT/C
vC2 = SC ( f ) df
0
> Applies to capacitance
VC (s )
1
H (s ) =
=
Vn (s ) 1 + sRC
equivalents: k(kBT) noise in a

spring
R(noiseless)
R
(noisy)
Vn
+
_
vC2 = H ( j 2f ) S n ( f )df
v =
2
C
Vc
_
1
4k BTRdf
2
1 + (2fRC )
1 k BT
v = 4k BTR
=
4 RC C
2
C

Noise bandwidth
> Can define a noise bandwidth
H ( s) =
for the capacitor noise
1
VC ( s )
=
Vn ( s ) 1 + sRC
> Noise source white
1
f =
df
2
0 1 + (2 fRC )
> Transfer function not flat
R(noiseless)
R
(noisy)
Vn
+
_
Vc
_

1
f =
4 RC
k T
vC2 = B
C
Thermomechanical noise in the tunneling accelerometer

> Wide plates (7mm x 7mm), narrowly spaced (50 m): consider
squeeze-film damping
> Squeeze-film damping holes: estimate about 30 holes, covering

between 1% and 10% of the plate area
Solid plates, at atmosphere :
bsolid
96LW 3
=
0.2 Ns
4 3
m
h0
Disk - shaped plates, one perforated, at 1 atm :

Aholes
0.01 Ns
b perf =
F
3
m
N holesh0 Aplate
Approximate net damping :
2
12Aplate

no. 2 (June 1998): 235-244. 1998 IEEE.
bnet bsolid b perf =
b perf bsolid
bsolid + b perf
0.01 Ns
Thermomechanical noise in the tunneling accelerometer

> Estimate noise at room temperature
S n ( f ) = 4k BTb 1.7 10
So noise is 1.3 10 11 N
22
N2
Hz
Hz
> Given a proof mass of 30 mg, relate noise force to noise

acceleration
Thermal noise in acceleration : 4.4 10 7 g' s
Hz
With a 1 kHz bandwidth, this is 1.4 g' s sensitivity
Thermal noise from the electronics?

> Electronics always contribute noise, and frequently are the
dominant noise source
> Amplifiers have some resistive elements in them; hence they

have thermal noise
> Amplifiers have other kinds of noise, too, such as 1/f noise: stay
tuned until we have the tools
> Electronics noise turns out not to be dominant for the tunneling
accelerometer
Tunneling is exponentially sensitive: high change in current for

small change in distance (x3 for 0.1 nm motion near operating point)
High change in current for small acceleration

Compare with small change in capacitance for 0.1 nm motion
Here, electronics introduce noise when signal is already huge
Shot noise
> The granularity of the electronic charge leads to a noise
associated with the flow of electric current
t
Discrete charge,
no noise
t
Constant
current
Charge
vs. time
t
Discrete charge,
with noise
Shot noise
> The spectral density function of shot noise is white

(at least at all frequencies of interest)
i (t ) = I DC + in (t )
The shot - noise spectral density function is :
Si ( f ) = 2qe I DC
in2 = Si ( f ) df
0
Shot noise in the tunneling accelerometer

> Linearize about operating point: output ~ 5 x 10-5 A/g
Not grams acceleration of gravity
Current depends exponentially on tunneling distance
Acceleration changes linearly with position
> Tunneling current ~ 1 nA
Si ( f ) = 2qe I DC 3 10
28
A2
Hz
Current noise is about 2 10-14 A
Hz
Acceleration noise is about 4 10-10 g' s
Hz
> Thermomechanical noise is three orders of magnitude larger

than shot noise!
Flicker Noise (1/f noise)

> Noise with a 1/f spectral density shows up in many systems
and is caused by a wide variety of phenomena
Diodes, earthquakes, biological systems, material creep/relaxation,

sand piles,
> Although the exponent is not always exactly 1, it is generally

called simply 1/f noise or flicker noise
> Important point: 1/f noise shows up a low frequencies

> At higher frequencies, white noise dominates
> The main problem with 1/f noise is that it is often hard to model,
except empirically. You may not know that you have a noise
problem until you measure it.
1/f noise in the tunneling accelerometer

> The earlier tunneling accelerometer papers classify noise
according to type: thermomechanical noise, shot noise, and
excess noise
> 1/f tunneling noise

Migration of atoms on the surface of the tunneling tip
Mobility of adsorbed contaminant molecules
> 1/f mechanical noise
Package relaxation
Thermal creep
Thermal bimorph (metal over the silicon hinge of the proof
mass)
> How to fix it? Through luck or design, find and control the
factors that correlate with the noise.
> In this case, control fluctuations in environmental temperature

Outline
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
> Modulation
Noise in a Diode
> Charge carriers are trapped in
and released from trap states in
the semiconductor
> Trap and release times depends

on energy of trap state
> A uniform-ish distribution of

states leads to a 1/f-ish spectral
density function
rs
rd
contacts and diode neutral regions.

rd is incremental resistance of
exponential diode.
1 dI D qe I D
=
=
rd dVD k BT
There is normal Johnson noise
associated with rs
and a combined shot - noise flicker - noise
spectrum associated with rd
+_ vn
ID
rs is incremental resistance of
Si ( f ) = 2qe I D
in

a
(
ID )
+K
f
The exponent a is in the range 0.5 - 2
Amplifier Noise
> MOSFETs have relatively high 1/f-noise corner frequencies
> One way to model the noise in a MOSFET is to refer all noise
sources back to one equivalent input noise source
> Noise at first stage is most important, because it is amplified in

successive stages
gate
Vn
_
+
Vin
id
Cgd
Cgb
_
_ source
Vbs
Cgs
+
Vgs
drain
gmbVbs
gmVgs
ro
_
Csb
+ body
Cdb
Kf
2
(1 + Fn ) +
Sn ( f ) = 4k BT
WLC ox f
3g m

Example: Resistance thermometer

> Toy resistance thermometer system
Use thermometer as one of the resistors for an inverting
amplifier
Assume a metal film resistor: ~ 3 x 10-3 K-1

> We create a noise model for a resistance thermometer by adding
Johnson noise source for the resistor itself
Two input-referred amplifier noise sources for the op-amp
Johnson noise source for the feedback resistor
> We calculate the transfer function for each noise source, and
using its square, find the contribution to the total mean-square
noise
Transfer function of system

> Contains four terms, one for each noise source.
> We can calculate the minimum detectable temperature change by
computing the root-mean-square noise
RF
Vs
RT
VnT
_
Vn1
_
Cin
Vn3
_
+
+
_
Signal portion of output :
V0
V0, s =
RF
Vs
RT
Incremental signal in response
Vn2
to a temperature change T
V0 =
RF
(VS + vnT + vn1 ) + 1 + RF vn 2 + vn 3
RT
RT
vo , s =
RFVS R
T
RT
Need to compare this to mean - square noise

and, for that, we need the spectral density function
Noise Spectral Density Function

> We assume uncorrelated noise sources; hence, we can add
spectral density functions
R
S n.o ( f ) = F
RT
R
[S n ,T ( f ) + S n ,1 ( f )] + 1 + F
RT
S n , 2 ( f ) + S n ,3 ( f )
Assume RF = 100 RT and RT = 1 k

Input transistor noise is as follows :
S n ,1 = S n , 2
Kf
2
= 4kbT
+ WLC f
g
3
m
ox
Resistor noise is as follows :

S n ,T = 4kbTRT
S n , F = 4kbTRF = 400kbTRT
Noise Spectral Density Function

> Noticing that RF/RT >> 1 simplies things
R
S n.o ( f ) = F
RT
R
[S n ,T ( f ) + S n ,1 ( f )] + 1 + F
RT
S n , 2 ( f ) + S n ,3 ( f )
For the FET, we assume L = 2 microns, W = 30 microns, tox = 15 nm

We assume an operating point for the input transistors of 100 microamps,
which sets g m to 300 microSiemens (about 3 k).
2K f
R0
4
+
S n.o ( f ) = 10 4k BT R0 +
+
g
3
100
WL
C
f
m
ox
S n.o ( f ) = 8.9 x 10
-13
7.2 x 10 8
+
f
Outline
> Types of noise
Thermal noise
Shot noise
Flicker noise
> Examples
> Modulation
Two methods of measuring temperature
> Use a DC source:

Result is that the minimum detectable temperature change is
inversely proportional to the source voltage.
But self-heating increases with source voltage

There clearly will be an optimum value of source voltage to
use
> Modulation (use an AC source followed by

synchronous detection)
Shifts the measurement to the white noise portion of the
spectrum
S/N ratio improves by a factor of 100 or so

DC Source
> Need to assume some finite bandwidth to filter the signal after
amplification. We assume a single-pole filter, with noise
bandwidth 1/4b (recall the noise bandwidth for the RC filter.)
v
2
n ,o
8.9 10-13
7.2 108
=
+
df
2
4 b
1 /tm f 1 + (2 f b )
where the lower limit is set by the measurement time
tm
1
1 tm
=
+
df
ln
ln
1

2 2 b
f 1 + (2f b )
2 b
1 / m
For this example assume b = 50 ms, then vo ,rms = 290 V

Assuming R = 3 103 , the final sensitivity for unity S/N is
Tmin
9.7 104
=
VS
AC Source
> If we use an AC source, the signal appears within a bandwidth
Log noise
about the source frequency, and limits of integration of the

spectral density function to twice the noise bandwidth about that
frequency
Same noise bandwidth, but

lower spectral density at the
higher carrier frequency
Log frequency
vo2,n =
Sn ( fo )
= 7.2 10 10 V 2 which corresponds to 27 V rms
2 b
Modulation Doesnt Always Work

> If the resistor itself has a
fluctuation in value due, for
example, to temperature
fluctuations, these
fluctuations always show up
directly in the same band as
the signal
> If these fluctuations have 1/f

character, then even the
modulated and then
demodulated signal will
show the 1/f spectrum
RT = R0 [1 + (t ) + R T ]
R V
1
V0 (t ) = F S
R0 [1 + (t ) + R T ]
RFVS
[1 (t ) R T ]
V0 (t )
R0
Conclusions
> Many noise calculations are easily integrated into our existing
approach to modeling, so you can (and should) consider noise
constraints from the earliest stages of design
> The dominant noise sources vary widely from system to system,
so dont make any assumptions
> 1/f noise is less easily dealt with in advance, but careful
debugging and circuit design can help
> Working at a higher frequency (modulation) can sometimes buy

you a lot a device sensitivity
Packaging
Carol Livermore*
*With thanks to Steve Senturia, from whose lecture notes some of
these materials are adapted
Outline
> Evolution of the packaging dilemma

> How to approach the challenge of getting a working
system, including:
Packaging
Partitioning
Test
Calibration
> Some common tools and considerations

> Examples
Package requirements
> Man the gates: let the right things in and out and prevent
other things from entering or leaving
> Protect the die

> Make it easy to interface the die with the rest of the world (outer
layer of packaging must conform to industry expectations)
> Be as inexpensive as possible

> The details can vary greatly
> Some elements that are often seen:
Fabrication/packaging cross-over, with first level packaging
occurring at the wafer scale, in the fab

Die attach into a package
Die encapsulation
Making connections
The need to plan for calibration and test!
Evolution of the MEMS package challenge

> MEMS micromachining and packaging began as answers to a
practical question: If I want to take advantage of silicon
piezoresistivity to measure pressure, what does the rest of the
product look like?
> Subsequent products adapt what they can from the earlier
approach and develop specific solutions where they must
Problem: no universal solutions

Must solve the details for every device design, and the solutions are
almost always different
> Enthusiasm for MEMS grows: packaging often neglected with

painful results
High costs (package cost 10x device cost!), devices that must be
redesigned
Devices cant make the jump to market!
> Present drive to create more widely-applicable packaging solutions

A diversity of devices to package

Motorola accelerometer.

Hymite/MEMSCAP packaged in-vivo magnetic switch.

Ultrasonic transducer array made by Siemens Corporation.

Motorola manifold absolute pressure sensor.
Outline

system, including:
Packaging
Partitioning
Test
Calibration

> Examples
Key Ideas
> Concurrent design
Design the device and the package at the same time
Often companies have different teams for the two parts
> Partition carefully
Which functionalities go in the chips, and which
functionalities go in the package?
How many chips will it have?

Which functionality goes on each chip?
How will the chips be connected together, and to the
package?
How will the way we partition it affect the way that we have to
test it?
The question of electronics

> Partitioning electronics and MEMS
Ask yourself honestly if integrating electronic functionality
monolithically with the MEMS device is worth it
It is difficult to optimize two different things simultaneously

Separating fab processes into pre-electronics and postelectronics can limit your options severely
Two chips separately are often cheaper than one integrated

chip (remember the Nmasks*Adevice rule?)
A cute, monolithically integrated device that doesnt work

because of unintended interactions between the MEMS
function and the electronics is useless
Some successful commercial products are monolithically

integrated
Designing for testability

> If the device is expensive, the package is expensive, and the
process of attaching the two is expensive, how much testing
should you do before committing a device to a package?
> There is no universal answer to this question, but it must be

considered for every device. Do a cost analysis with your best
estimates of costs and yields to balance risk and cost.
> The package/test procedure will have a significant impact on the

ultimate cost of building the part, and on its economic viability
> Many systems cant be tested at all without some form of

packaging
Planning for variations

> Given the many variations possible from fabrication and
packaging (both random and systematic), how do you ensure a
working device at the end?
> Aim for a perfect as-fabricated device and no shifts from the
package?
> Aim for a device that is perfect once it has interacted with the
package?
> Admit that you cant control all the factors and simply trim the
hardware once its packaged?
> Admit that you cant control all the factors and simply calibrate
whatever you get in the software?
> All of these are expensive do you want to put a lot of

resources into getting the fab just right, or into fixing each and
every part after its made?
Checklist before detailed design

> Youre ready to do the detailed design of the system and all of
its components when you have the following:
Complete specifications for both the chip(s) and the package

Specifications for all interconnections
A list of all of the parasitics that you can think of, and some
assessment of their effect on the system operation
Provision for calibration and test
To do list: the detailed design

> For the MEMS device
A process flow
A mask set
The corresponding device geometry and materials
A model supporting predicted performance
Specification of the test and calibration method
> For the package

Artwork
Specification of components and how they are made (or where
purchased)
Acceptance procedures for packages
> Full assembly

The packaging procedure
Test, acceptance, and calibration
Outline

system, including:
Packaging
Partitioning
Test
Calibration

> Examples
Die separation
> Multiple dies fabricated on the same wafer must be separated in
order to be packaged and sold
> Die saw common

> Die saw blades are of order 30 m to 250 m thick
> High speed blade rotation, cooled by a flow of water
> Lots of debris, vibrations, and general dirt
> How to protect the device?
Release etch after die saw
Wafer bonding as first-level packaging, before the die saw step
Die saw the device upside down
Dont die saw: etch mostly through and crack chips apart
> Question: do you need to do any testing before you separate the
dies?
Die attach
> Whether ICs or MEMS devices, dies must be attached to a
package
> Package: application specific

> Solder as adhesive: commonly soft solder (lower melting
temperature, about 100 C 400 C)
> Epoxy as adhesive: cross-link when heated (about 50 175 C)

> Polyimide or silicone as adhesive
> Adhesive application: dispense through a nozzle, screen print
> Can use pick and place tools to position dies on top of
adhesive in the package
Plastic packaging
> The integrated circuit standard
> Very inexpensive, pennies per electrical connection pin
> A thermosetting plastic is melted (ballpark 175 C) and injected
into a mold
> The plastic cools and hardens

> The least expensive approach:
Attach the die to a metal lead frame with an adhesive
Injection mold the plastic around it
Question: will your device and electrical connections survive
this?
Question: will encapsulation in plastic impair its

functionality?
Plastic packaging
> Second approach: a more flexible, more expensive, gentler
approach to plastic packaging
> Injection mold the package around a lead frame before the die is
attached
> Attach the die with an adhesive

> Cap the package
> Applications:
Fragile devices or electrical connections
Achieving connections that are not standard in IC packages,
such as fluid connections or optical transparency
Special requirements are integrated into the package cap

Ceramic packaging
> Collections of particles are sintered at temperatures ranging
from 800 C to 1600 C to form the package (often alumina, Al2O3)
> Ceramic packages are often processed as laminates

Individual layers can have screen printed electric
interconnects, metal film resistors, and even interlayer via
connections
Anneal the package; braze pins to package

Make electrical connection between die and package
Cap and seal the package (application-specific)
Resistors on the outer layer can be trimmed after the die is
packaged
> Durable, potentially well-sealed (hermetic)

> Higher cost (ballpark a few tens of dollars vs. less than a dollar)
Metal packaging
> A solution for harsh environments or relatively quick
prototyping
> Can be well-sealed (hermetic)

> Common materials: Kovar or stainless steel
> When features are more important than cost, can simply
machine the package with the needed features
> Example: a pressure sensor for environments that silicon

cannot tolerate
Package a Si pressure sensor in a stainless steel package

Cap the package in part with a thin stainless steel diaphragm
Fill the gap between Si and steel with oil to transfer the
pressure
Making electrical connections

> Techniques adopted from IC packaging
> Wire bonding connects electrical contact pads on die to
electrical contact pads in package
An ultrasonic sewing machine that stitches with wire

Heat + pressure + ultrasonic energy joins wire to contact pad
Frequency considerations: ultrasound may excite a
resonance
Thickness of bond pads

> Flip-chip bonding
Chip turns upside down; solder bumps attach it directly to
package
Heat to flow solder (think about temperature)

Smaller connections/lower inductance than with wirebonding
Making fluid connections

> Not standardized
> Techniques at the laboratory scale
O-ring seals and conventional tubing
Glue a tube in a hole (includes more sophisticated
approaches)
Stick a needle through a polymer structure, and inject

through it
> For lab-on-a-chip applications: various proposals for the world

to chip interface
> For a pressure sensor: protect the membrane with a cap layer
and route a hole in the package to the connector of your choice
(ie screw thread)
> For your application?

Packaging in the fab why do it?

> Your structure will become so fragile when you do the release etch that
you will no longer be able to package it
> Your device wont survive die saw unless the fragile parts are
encapsulated
> You want to minimize the package size

> You cant afford to have any particulates on your device, so it needs to
be encapsulated before it leaves the cleanroom
> Its less expensive in your case

> You expect the quality of the seal to be better in a microscale process
such as anodic bonding than in a macroscale process such as gluing
> You want to create a sealed cavity with vacuum inside

> You want to create a sealed cavity with a controlled atmosphere inside
Sealing
> Vacuum operation
Device needs to operate at vacuum, or some fixed pressure,
without ever being pumped out again
Examples: measuring absolute pressure, high Q resonators

Typically accomplished in the fabrication process rather than
in packaging
Approaches: wafer bonding, deposited films as sealants

Concern: outgassing
> Isolation from environment
> Hermetic packaging
Device doesnt need to operate at vacuum, but if water gets
inside, the device will eventually be destroyed
Vacuum sealing by anodic bonding

> Anodic bonding to seal a pressure sensor
> Fabricate on a Si wafer; dissolve it away at the end
Figure 4 on p. 582 in: Chavan, A. V., and K. D. Wise. "Batch-Processed, Vacuum-Sealed Capacitive Pressure Sensors.
Journal of Microelectromechanical Systems 10, no. 4 (December 2001): 580-588. 2001 IEEE.
Vacuum sealing by fusion bonding

> An ultrasonic transducer (a microphone, essentially) sealed by
silicon fusion bonding
> Contact in vacuum; risk of outgassing on anneal
Figure 1 on p. 130 in Huang, Y.,

A. S. Ergun, E. Haeggstrom, M. H. Badi, and
B. T. Khuri-Yakub. "Fabricating Capacitive
Micromachined Ultrasonic Transducers with
Wafer-bonding Technology." Journal of
Microelectromechanical Systems 12, no. 2
(April 2003): 128-137. 2003 IEEE.
Vacuum sealing by film deposition

> An ultrasonic transducer sealed by a deposited film
> Using the stringer effect for a good cause
> Could worry about outgassing
Figure 16 on p. 106 in Jin, X., I. Ladabaum, F. L. Degertekin, S. Calmes, and B. T.

Khuri-Yakub. "Fabrication and Characterization of Surface Micromachined
Capacitive Ultrasonic Immersion Transducers. Journal of Microelectromechanical
Systems 8, no. 1 (March 1999): 100-114. 1999 IEEE.
Figure 2 on p. 101 in Jin, X., I. Ladabaum, F. L. Degertekin, S. Calmes,

and B. T. Khuri-Yakub. "Fabrication and Characterization of Surface
Micromachined Capacitive Ultrasonic Immersion Transducers." Journal
of Microelectromechanical Systems 8, no. 1 (March 1999): 100-114.
1999 IEEE.
Resonator with thin film vacuum packaging
Figure 10 on p. 290 in Lin, L., R. T. Howe, and A. P. Pisano. "Microelectromechanical Filters for Signal Processing."
Journal of Microelectromechanical Systems 7, no. 3 (September 1998): 286-294. 1998 IEEE.
Resonator with thin film vacuum packaging
Figure 11 on p. 291 in Lin, L., R. T. Howe, and A. P. Pisano. "Microelectromechanical Filters for Signal Processing."
Journal of Microelectromechanical Systems 7, no. 3 (September 1998): 286-294. 1998 IEEE.
Isolation from the environment

> Example: an automotive pressure sensor
The membrane must be able to feel the pressure
But, the environment is not inert
Solution: coat the business part to protect it while not
compromising the functionality
> Example: bioMEMS, especially in the body

Biocompatible but functional
> Parylene is useful here
Conformal material, deposition near room temperature
Minimal structural impact
Resistant to water, many organic solvents, weak acids/bases,
salt, fuel but it is not invulnerable.
Hermetic packaging
> Moisture is bad and can lead to corrosion in the most benign of
circumstances
> Electronics and MEMS often last much longer and are more
reliable if moisture is kept out
> Definition of hermetic: prevents the diffusion of helium, with a

definition for the maximum allowable helium leak rate
Perfect hermeticity does not exist in practice
> Working definition of hermetic: keeps the moisture out

> Good material choices: silicon, metal, ceramic, thick glass (mm
thickness or above)
> Bad material choices: plastics, organic materials

> All connectors must also be hermetically sealed
Permeability chart
Considerations when selecting packaging

> First, everything that you need to think about in designing the
device itself
> Electrical parasitics

> Stress: will your package squeeze your device and change its
characteristics?
> Will you be able to calibrate and test your device? When and
how? All significant package-induced variations must occur
before that point, or else be accounted for in advance.
> Other concerns: fragility, range of temperature operation and

CTE, etc.
Outline

system, including:
Packaging
Partitioning
Test
Calibration

> Examples
Lab-scale packaging: micro rocket

> If Im just trying to get a degree, and I dont plan on selling this
device, do I still have to design the package with the device?
> YES!!! (You want the degree, right?)
Courtesy of Adam London. Used with permission.
Courtesy of Adam London. Used with permission.
Source: Adam London, MIT.

Lab-scale packaging: micro turbine

> Micro turbine device, with fluidic and electrical connections
> Fluid connections: seal acrylic plate to die via o-rings
> Questions to think about BEFORE sending out the masks: will
the pressure distort the die? Is there room for the o-rings?
Pressure Sensor Case Study

> A Motorola manifold-absolute-pressure (MAP) sensor
> Silicon micromachined diaphragm with piezoresistive sensing

Motorola manifold absolute pressure sensor.
Device and Package Concept

> Monolithic pressure sensor with circuitry in a custom bipolar
process mounted on silicon ($$$) support mounted in plastic
package
Silicone gel die coat
Pressure
Stainless steel cap
Wire bond
Sensor on
backplate
Lead frame
Pre-molded plastic case
Silicone die bond

Adapted from Motorola. Sensor Device Data/Handbook. 4th ed. Phoenix, AZ: Motorola, Inc., 1998.
The Arguments for Monolithic

> Smaller overall system
> More reliable interconnect
> Solves a customers problem:
Does away with a circuit board
Therefore, may be a cheaper total solution for the customer
even if the integrated sensor by itself is more expensive than
the hybrid sensor
> Its a business decision.
Define Interfaces
> Electrical interface: how many pins?
The sensor needs only three pins
> Mechanical interface: a stainless steel lid on the package
> When to calibrate? Before or after packaging?
Things to consider
Package-induced stress
Gel-seal induced shift in calibration
> The decision: Calibrate after bonding into the package, but
before the gel-seal.
Cost: EXTRA PINS ON THE PACKAGE
Wafer-level packaging
> Cant use fusion bonding
Silicon
(circuitry cant stand it)
> What about anodic bonding?
Glass frit
> Bonding to silicon backplane

Silicon
using a glass-frit bond (more

forgiving of deviations from
wafer flatness)
WD7
> Firing temperature between 450

and 500 C.
> Must be void free and more
Die attach
pure than standard glass frits

Details and headaches

> Package material and molding process: avoiding leaks
> Die attach
Originally used RTV (room-temperature vulcanized) silicone
adhesives
Now use a gel-like silicone or fluorosilicone that requires 150 C cure

(rubbery material transfers low stress to the chip)
Biggest problem: batch to batch variation of the die-attach material
> Wire bonding
With the chip sitting on a gel base, how can you wire bond to it?
Heat transfer through the gel is poor
Mechanical support is soft
Motorola developed custom tooling to apply heat directly to the chip
during wire bonding
More details and headaches

> Final gel encapsulant
Must wire bond before trimming, but
cannot put final gel coat on until after

trimming
Gel does produce a small shift in
response
Calibration targets must be predistorted to anticipate the effect of the
gel coat
Batch to variations in gel materials can
then create problems
> Next level assembly: keeping

customers happy
Market fragmentation
Different customers want different
next-level assemblies (Ford doesnt
want a Chrysler package)
Example: a newer capping technology

> Disclaimer:
This is not a sales pitch. This particular technology is described
here simply because it is well-described in the literature. Flexible
packaging techniques are of wide interest.
> Concept: bulk-micromachined silicon caps with integrated

interconnects, assembled onto dies at the wafer-scale
> Electrical connections and hermetic package seal made by flipchip solder bonding and through-cap interconnects
> Small overall package sizes possible

> Clean (no particles from ceramic components), CTE matched to
minimize damage and package-device interaction
> Reference: Elger et al., presented at IMAPS 04, cap technology

by Hymite
Application for this example

> MEMS switch for in vivo use, to be switched by an external
magnetic field
Details
> Caps made by standard KOH/electroplating bulk
micromachining process
SOI substrate, KOH etched from both sides
> Including interconnects in device would lower yield and affect

device process flow
> Through-cap interconnects permit flip-chip bonding to device

and to outside world with minimal disturbance to device process
> Silicon caps can be cleaned with chips before assembly to

minimize unwanted material in sealed-cavity
> Moderately low melting solder with no flux (no organic

contamination)
> Pick and place assembly, or wafer to wafer transfer; heat in the
pick and place tool to prevent caps from sliding
Bottom line
> People are actively trying to create and sell techniques like this
that can package a greater (though still not infinite) set of MEMS
devices with a single process and vendor
> This may make the life of the MEMS designer easier, permit
smaller packages, and (hopefully some day) smooth the trip to
market for some MEMS devices
> But there is still no universal solution: what if you need optical
access, fluidic access?
> And you still have to design the device and the packaging
process together, even as the number and convenience of
available tools increases
Where to learn more

> Nadim Maluf, An Introduction to Microelectromechanical
Systems Engineering, Artech House, 2000.
> Handbooks of microelectronics packaging, such as:

R.R. Tummala and E.J. Rymaszewski (eds.), Microelectronics
Packaging Handbook, Reinhold, 1989.
6.777J/2.372J Design and Fabrication of Microelectromechanical Devices

Spring Term 2007
In-Class Design Exercise
April 25, 2007
Accelerometers are starting to be deployed for consumer applications such as cell phones. Accelerometers in
this market do not need high precision or accuracy, but they must occupy a small-volume, low-profile
package. Below is a simplified schematic for the accelerometer that we will consider in this exercise. Youll
later find out that they are fancier than this, but for now, this is quite challenging enough.

Specifications
Area of capacitive fingers: 100,000 m2
Mass of proof mass: 3 g (Si=2300 kg/m3)
Max area of system: 15 mm2
Max height: 2.0 mm
Number of bond pads: 4
On-chip bond pad size: 50 m x 50 m (unless using CSP, in which case use those dimensions)
Spacing between on-chip bond pads: 50 m (unless using CSP)
Packaged device must withstand reflow temperatures for flip chip mounting of packaged device.
Packaged in vacuum.
Electrical connection from proof mass fingers to 4 bond pads.
Your task is to design a fabrication process flow and a package process to create an accelerometer that meets
the above specs. You should determine the important dimensions of your device and package. Youre free
to use bulk micromachining or surface micromachining, to start your packaging at the wafer scale or not, etc.
You need to consider everything: when the wafer is diced, when the proof mass is released, attachment into
the package, etc. Be sure to think about unintended interactions, such as thermal mismatch. Make sure that
you meet the specs above; within those constraints, your goal is to come up with the smallest packaged device
that can be mounted on a circuit board.
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J / 2.372J Design and
Fabrication of Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare
(http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Packaging options
Wafer-level chip-scale packaging (CSP)
Hermetic: Dictated by chip.

Package thickness: Equal to chip thickness + 100 m for solder bumps.
Maximum chip thickness: none
Note: solder bumps are applied at the wafer level. The wafer must be able to withstand
electroplating of a 5 m Cu seed layer onto the pads, screen printing a solder paste, and
then reflowing the paste at 225 C to form bumps.
Dimensions:
e
e
a
c
a
c,d
e
(mm)
0.18
0.5
0.32
a
b
c,d
e
f
g
(mm)
0.64
1.27
1.27
1.27
1.0
1.8
Leadless ceramic chip carrier (LCCC)

Hermetic: Yes.
Package thickness: 1.8 mm
Maximum chip thickness: 800 m
Dimensions:
d
a
c
Die
Dual flat no-lead package (DFN)

Hermetic: No.
Package thickness: 0.9 mm
Maximum chip thickness: 350 m
Dimensions:
d
a,b
c
d
e
f
g
a
c
(mm)
0.5
1.0
1.5
0.5
0.8
0.9
Die
Design choices: MEMS actuators
Carol Livermore
With thanks to Prof. S. Mark Spearing, from whose work some of

this material is obtained.
Outline
> Sorting through design options

> Actuator characteristics, examples of actuators, and
theoretical and practical limits
Electrostatic
Thermal
Piezoelectric
Magnetic
Motivation
> We now have lots of domain knowledge that we can apply to
decide whether a particular design is good enough for a
particular application
> How do you develop your intuition about which approaches are
likely to meet the needs of a given application?
Experience?
Literature search?
> One approach: borrow the concept of design charts, a tool
commonly used at the macroscale for basic design choices
(which material to choose?)
> D.J. Bell, T.J. Lu, N.A. Fleck, and S.M. Spearing, MEMS Sensor
and Actuator Selection: Database and Case Study, J.
Micromechanics and Microengineering, v 15, p. S153, 2005.
Sample Materials Comparison Chart

1000
Youngs modulus E
(G = 3E/8 ; K = E.)
Be
MFA:88-91
Youngs modulus, E (GPa)
12
10
Engineering
composites
104
Parallel
to grain
Balsa
Wood
Products
Woods
1.0
3x10
ASH
OAK
PINE
FIR
Lower E limit
for true solids
Perpendicular
to grain
Spruce
3x102
0.01
0.1
Mg
Alloys
Polymers foams
0.3
ZrO2
BeO
Porous
ceramics
Engineering
alloys
Engineering
polymers
Polyesters
PP
HDPE
Hard
BUTYL PU
Silicone
Soft
BUTYL
1.0
Guide lines for

minimum weight
design
PTFE
E
=C
Plasticised D
PVC
103
Cork
Sialons
MEL
PC
Epoxies
PS
PMMA
PVC
Nylon
LDPE
Balsa
0.1
KFRP
ASH
OAK
PINE
FIR
Aluminas
Mo W-Alloys
Alloys
Si
Ni Alloys
CFRP Glasses
Steels
UNI-PLY
3e
Cu Alloys
Pottery Ti Alloys
Zn Alloys
KFRP
Al Alloys
GFRP
CFRP
Tin Alloys
Rock, stone
Cement, concrete
Laminates
GFRP
Lead alloys
100
) E ) (m/s)
WC-Co
Engineering
Ceramics
Diamond
B SIC Sl2N4
1. Modulus-Density
Elastomers
1
E 2
=C
1
E 3
=C
Density, (Mg /m3)
10
30
Material Selection in
Mechanical Design,
M.F Ashby, Pergamon
Press, Oxford, 1995.

Adapted from Ashby, M. F. Material Selection in Mechanical Design. Boston, MA: Butterworth-Heinemann,
1995. ISBN: 9780750627276.
Different criteria for different actuator applications

> Nanopositioners
Speed, resolution, range, repeatability, robustness against
disturbances, ability to sense as well as actuate
> Micro mechanical testing apparatus

Range of motion, force, ability to sense as well as actuate,
speed
> RF MEMS
Speed, range of motion, force, power dissipation, voltage
required, ability to sense as well as actuate
RF MEMS Approach
> Active RF circuit components
Implement through CMOS compatible circuits
> Reconfigurable system that allows multiple functionalities
MEMS ohmic contact switches with low losses and good
isolation and linearity

MEMS variable capacitors (instead of CMOS capacitors)
> Integrate passive components in place, saving space in the

overall system
> Some issues:

Reliability can be an issue for MEMS switches (accumulated
damage to contacting surface)

Limited capacitance range for MEMS variable capacitors,
though losses are low, and linearity and high power
performance are good
RF MEMS Components
> Switches and relays open circuits, close circuits, or connect
signals to ground
Includes RF signal electrodes and actuating electrodes

If the actuating signal and the RF signal exist on the same
pair of electrodes, its a switch
If the actuating signal and the RF signal exist on different

sets of electrodes, its a relay
> Variable capacitors can be tuned continually over their range

> Switchable capacitors can switch between two discrete
capacitance values
> Resonators vibrate at a particular (often tunable) frequency and

>
are useful for filters

Components use a common set of actuation mechanisms
The Wish List for RF MEMS

> Low power operation
> Fast switching
> High force (especially for making electrical contact)
> Large analog-controllable actuator travel
> Simple fabrication
Few masks
Standard, CMOS-compatible processes and materials
Ability to be fabricated on the same chip as circuits
> Low voltage
> Robust operation, not prone to failures
> Ability to sense as well as actuate
Outline
> Sorting through design options

> Actuator characteristics, examples of actuators, and
theoretical and practical limits
Electrostatic
Thermal
Piezoelectric
Magnetic
Major Classes of MEMS Actuators

> Electrostatic
Attraction between oppositely charged
conductors
Figure 1 on p. 222 in Agrawal, V. "A Latching MEMS Relay for DC and RF

Applications." Electrical Contacts-2004: Proceedings of the 50th IEEE Holm
Conference on Electrical Contacts; The 22nd International Conference on
Electrical Contacts, Seattle, WA. Piscataway, NJ: IEEE, 2004, pp. 222-225.
ISBN: 9780780384606. 2004 IEEE.
> Thermal
Displacement due to thermal
expansion
> Piezoelectric
Displacement due to strain induced by

Figure 11 on p. 342 in Zavracky, P. M., N. E.
McGruer, R. H. Morrison, and D. Potter.
"Microswitches and Microrelays with a View
Toward Microwave Applications." International
Journal of RF and Microwave Computer-Aided
Engineering 9, no. 4 (1999): 338-347.

4 actuators displayed side-by-side to show
how small they are.
an electric field
> Magnetic
Displacement due to interaction among
various magnetic elements:
permanent magnets, external magnetic
Figure 8 on p. 43 in De Los Santos, H. J., G. Fischer, H. A. C. Tilmans, and J. T. M.
fields, magnetizable material, andvan
Beek. "RF MEMS for Ubiquitous Wireless Connectivity. Part I. Fabrication." IEEE
Mirowave Magazine 5, no. 4 (December 2004) 36-49. 2004 IEEE.
current-carrying conduct
Recessed Actuation Electrodes
Spring
Upper capacitor plate
Lower capacitor plate
Gaining Additional Travel with Leveraged Bending
> Pull-in is modified if the actuating electrodes are

away from the point of closest approach
1)
2)
Figure 3 on p. 499 in Hung, E. S., and S. D. Senturia.

"Extending the Travel Range of Analog-tuned Electrostatic Actuators."
Journal of Microelectromechanical Systems 8, no. 4 (December 1999):
497-505. 1999 IEEE.
with stressstiffening
3)
Zipper electrostatic actuators
Figure 1 on p. 29 in Ionis, G. V., A. Dec, and K. Suyama. "A Zipper-action

Differential Micro-mechanical Tunable Capacitor." MEMS 2001: 2001
Microelectromechanical Systems Conference, Berkeley, CA, August, 24-26,
2001. New York, NY: Institute of Electrical and Electronics Engineers, 2002.
ISBN: 9780780372245. 2002 IEEE.
Curved electrode actuators

Variable capacitor for RF
Figure 10 on p. 263 in Legtenberg, R., J. Gilbert, S. D. Senturia, and M.

Elwenspoek. "Electrostatic Curved Electrode Actuators." Journal of
1997 IEEE.
Figure 2 on p. 30 in Ionis, G. V., A. Dec, and K. Suyama. "A Zipper-action

Differential Micro-mechanical Tunable Capacitor." MEMS 2001: 2001
Microelectromechanical Systems Conference, Berkeley, CA, August, 24-26,
2001. New York, NY: Institute of Electrical and Electronics Engineers, 2002.
ISBN: 9780780372245. 2002 IEEE.
Comb Drive Application: Variable Capacitor

Figure 2 on p. 126 in Yao, J. J., S. Park, and J. DeNatale. "High Tuning-ratio MEMS-based Tunable Capacitors for RF
Communications Applications." 1998 Hilton Head Workshop on Solid-State Sensors and Actuators, Technical Digest 98TRF-0001.
Area-tuning capacitor made by deep reactive ion etching.

Tuning with area rather than gap promotes stability and tuning
range, but fabrication is more challenging and expensive.
Electrostatic inchworm actuatorss
Figure 4d on p. 332 in Yeh, R., S. Hollar,

and K. S. J. Pister. "Single Mask, Large
Force, and Large Displacement
Electrostatic Linear Inchworm Motors."
Journal of Microelectromechanical Systems
11, no. 4 (August 2002): 330-336. 2002 IEEE.
Figure 5 on p. 333 in Yeh, R., S. Hollar, and K. S. J. Pister. "Single Mask, Large Force, and
Large Displacement Electrostatic Linear Inchworm Motors." Journal of Microelectromechanical
ystems 11, no. 4 (August 2002): 330-336. 2002 IEEE.
Inchworm (more travel by

repeated motion)
Figure 4c on p. 332 in Yeh, R., S. Hollar, and K. S. J. Pister. "Single Mask,
Large Force, and Large Displacement Electrostatic Linear Inchworm Motors."
Journal of Microelectromechanical Systems 11, no. 4 (August 2002): 330-336.
2002 IEEE.
Electrostatic distributed actuators
Figure 1 on p. 122 in Minami, K., S. Kawamura, and M. Esashi. "Fabrication of Distributed Electrostatic Micro
Actuator (DEMA)." Journal of Microelectromechanical Systems 2, no. 3 (September 1993): 121-127. 1993 IEEE.
Distributed actuators (more force and distance by series and parallel

combination)
Electrostatic scratch drive actuators

Figure 1 on p. 737 in Li, L., J. G. Brown, and
D. Uttamchandri. "Study of Scratch Drive
Actuator Force Characteristics." Journal of
Micromechanics and Microengineering 12,
no. 6 (November 2002): 736-741.
no. 6 (November 2002): 736-741.

no. 6 (November 2002): 736-741.
10000
10000
1000
1000
Breakdown voltage
Breakdown field
100
100
10
10
10
100
1000
Breakdown field (V/m)
Breakdown voltage (V)
Limits on electrostatic actuators: Paschens curve
1
10000
Separation-Pressure (m-atm)
J. Judy, masters thesis, Berkeley.

Original reference: F. Paschen, Ann. Physik, 273 (1889) pp. 69-96.
Electrostatic actuators: force and distance
1 mN
1 m
1 mm
Courtesy of S. Mark Spearing. Used with permission.
Electrostatic actuators: frequency and force
1 MHz
1 kHz
1 Hz
1 mN
Max frequency reflects reported values and resonant frequencies.

Electrostatic Actuators and the Wish List

> Fast switching (electrical RC times, maybe resonance)
> High force (obtain N to maybe mN if heroic)
> Large analog-controllable actuator travel
Few masks
> Low voltage
> Robust operation, not prone to failures (stiction, dust)
so-so
Thermal Actuation
> Heat up a moveable structure; when it expands, you get force
and displacement
> A variety of different structures can take advantage of this type

of actuation (example: thermal bimorph in a circuit breaker)
> Good features:

High force, moderate displacement
Lower actuation voltages
Relatively simple fabrication
> Not so good features:
Slower switching speeds (hundreds of microseconds)
Higher power dissipation: continuous current needed to
maintain displacement unless structure latches
Thermal bimorph actuator

Figure 1 on p. 411 and Figure 4a) on p. 412 in Sehr, H., I. S. Tomlin, B. Huang, S. P. Beeby, A. G. R. Evans, A.
Brunnschweiler, G. J. Ensell, C. G. J. Schabmueller, and T. E. G. Niblock. "Time Constant and Lateral Resonances
of Thermal Vertical Bimorph Actuators." Journal of Micromechanics and Microengineering 12, no. 4 (July 2002): 410-413.
Buckle Beam Thermal Actuator

> Concept:
Run a current through a
V-shaped beam
Power dissipation heats

beam and makes it
expand
Pre-buckled shape
ensures that point of V
moves outward
> Uniform heat dissipation in

beam
Figure 1 on p. 652 in Girbau, D., A. Lazaro, and L. Pradell. "RF MEMS

Switches Based on the Buckle-beam Thermal Actuator." 33rd
European Microwave Conference, 2003, Munich, Germany,
October 7-9, 2003: Conference Proceedings 2 (2003): 651-654. 2003 IEEE.
> Center heats up more

because its more isolated
from the supports
Buckle Beam Thermal Actuator

> Force increases linearly when
actuators are connected in
parallel
> Displacement is unchanged

> Used here to close RF
switches in series and
parallel configurations
> One consideration: it can be

harder to get low contact
resistance for contact
between vertical sidewalls
than for contact between
horizontal surfaces
Figures 6 and 7 on p. 653 in Girbau, D., A. Lazaro, and L. Pradell.

RF MEMS Switches Based on the Buckle-beam Thermal Actuator.
33rd European Microwave Conference, 2003, Munich, Germany,
October 7-9, 2003: Conference Proceedings 2 (2003): 651-654. 2003 IEEE.
Electrothermal Actuator: Heatuator

Basic heatuator with one wide
and one narrow beam
Figure 1 on p. 222 in Agrawal, V. "A Latching MEMS Relay

for DC and RF Applications." Electrical Contacts-2004:
Proceedings of the 50th IEEE Holm Conference on Electrical
Contacts; The 22nd International Conference on Electrical
Contacts, Seattle, WA. Piscataway, NJ: IEEE, 2004, pp. 222-225.
ISBN: 9780780384606. 2004 IEEE.
Modified heatuator: two narrow

beams and one unheated signal beam
Figure 2 on p. 222 in Agrawal, V. "A Latching MEMS Relay for DC and RF

Applications." Electrical Contacts-2004: Proceedings of the 50th IEEE Holm Conference on
Electrical Contacts; The 22nd International Conference on Electrical Contacts, Seattle, WA,
Piscataway, NJ: IEEE, 2004, pp. 222-225. ISBN: 9780780384606. 2004 IEEE.
> Analogous to a thermal bimorph, except bimaterial

sandwich is replaced by a wide signal beam (cold)
and narrow beams (hot).
Electrothermal Relay with Mechanical Latch
Mechanical latch reduces

power consumption after
switching and closes the
relay.
Figure 3 on p. 223 in Agrawal, V. "A Latching MEMS Relay for DC and RF Applications." Electrical
contacts-2004: proceedings of the 50th IEEE Holm Conference on Electrical Contacts; the 22nd
International Conference on Electrical Contacts, Seattle, WA. Piscataway, NJ: IEEE, 2004, pp. 222-225.
ISBN: 9780780384606. 2004 IEEE.
Thermal actuators: force and distance
1N
1 mN
1 m
Thermal actuators: frequency and force
1 kHz
1 mN
1N
Max frequency reflects reported values and thermal time constants.

Thermal Actuators and the Wish List

> Fast switching
so-so
> High force (N to N, mN typical)
> Large analog-controllable actuator travel (1-100 um)
> Simple fabrication (is possible, at least)

3
Few masks
> Low voltage
Piezoelectric Actuation
> Apply an electric field across a piezoelectric material;

deformation (strain) results, along with actuator
deflection and force
> Different piezoelectric geometries are possible

> Good features:
High force
High switching speeds
Low power dissipation

Small strains must be engineered into useful displacements
Complex fabrication and nontrivial materials
Piezoelectric Actuators
___________________ Used with permission. Figure 1a on p. 92 in

Zhang, Q. Q., S. J. Gross, S. Tadigadapa, T. N. Jackson, F. T. Djuth, and S. Trolier-McKinstry. "Lead
Zirconate Titanate Films for d33 Mode Cantilever Actuators." Sensors and Actuators A, Physical 105,
no. 1 (June 2003): 91-97.
> PZT layer is fabricated on top of cantilever, sandwiched between

electrodes, and poled in the vertical direction
> Electric field is applied between top and bottom electrodes,

parallel to polarization
> d31: PZT develops a negative strain in the transverse direction;

>
rest of cantilever does not

Cantilever bends up
___________________ Used with permission. Figure 1a on p. 92 in

no. 1 (June 2003): 91-97.
> If electric field is opposite to the polarization, the cantilever will

deflect in the opposite direction, up to a point
> Consequence: stronger restoring forces and faster responses

possible than with an unactuated return
> If antiparallel electric field is too large, PZT will repole in the
other direction
Courtesy of Elsevier, Inc., ____________________

http://www.sciencedirect.com. Used with permission. Figure 1b on p. 92 in
no. 1 (June 2003): 91-97.
> PZT layer is fabricated on top of cantilever, under interdigitated

electrodes
> Electric field is in the plane; PZT is poled in the plane

> d33: With E parallel to poling, PZT develops a positive strain in
the direction of its length; cantilever bends down
> Again, antiparallel E has opposite effect within limits

Piezoelectric RF MEMS Switch

> Piezoelectric cantilever driven by
interdigitated electrodes, d33
> Cantilever is multilayer stack,

both for film stress compensation
(so the cantilever starts out flat)
and for maximum actuation force
and deflection
> Cantilever layers, from the

bottom: 50 nm oxide barrier, 500
nm SiN structural layer, 50 nm
oxide adhesion layer, 300 nm
zirconia, 230 nm PZT, Cr/Au
electrodes and contacts
Figures 1a and 1b on p. 175 in Gross, S. J., S. Tadigadapa,

T. N. Jackson, S. Trolier-McKinstry, and Q. Q. Zhang.
"Lead-zirconate-titanate-based Piezoelectric Micromachined
Switch." Applied Physics Letters 83, no. 1 (July 2003): 174-176.
> Few microsecond switching

Thin Film Bulk Acoustic Resonator (FBAR)

> Suspended structure: a sandwich
of piezoelectric material between
electrodes
> Apply an AC signal to electrodes

at the proper frequency, and the
piezoelectric resonates
mechanically
Figure 1 on p. 1687 in Wang, K., P. Bradley, and M. Gat.

"Micromachined Bulk Acoustical-wave RF Filters." In 2004
7th International Conference on Solid-State and Integrated
Circuits Technology proceedings, ICSICT 2004: October
18-21, 2004, Beijing, China, edited by Ru Huang. Piscataway,
NJ: IEEE Press, 2004, pp. 1687-1690. ISBN: 9780780385115.
2004 IEEE.
> Air interfaces minimize acoustic

losses into surrounding medium
> In commercial use (sold by Agilent

for cell phones)
Thin Film Bulk Acoustic Resonator (FBAR)

> Materials challenges:
Repeatability of piezoelectrics
properties (choose AlN and work
the process until it is repeatable)
Low acoustic losses and low

electrical losses (choose
Figure 2 on p. 814 in Ruby, R. C., P. Bradley, Y. Oshmyansky, A. Chien, and J. D. Larson, III. "Thin
molybdenum)
Film Bulk Wave Acoustic Resonators (FBAR) for Wireless Applications." In 2001 IEEE Ultrasonics
proceedings, an international symposium: October 7-10, 2001, Omni Hotel, Atlanta,
> Fabrication challenges: Symposium
Georgia, edited by Donald E. Yuhas and Susan C. Schneider. Piscataway, NJ: IEEE, 2001,
pp. 813-821. ISBN: 9780780371774. 2001 IEEE.
Precise control of layer thickness
Process compatibility (with IC and
piezoelectric)
Structure built over an oxide-filled

cavity in the substrate; oxide
removed at end to release

4 actuators displayed side-by-side to show how small they are.
Packaging in the fab, by wafer

bonding
Piezoelectric Actuators and the Wish List

> Fast switching
> High force (10 N to mN)
> Large analog-controllable actuator travel (<0.1 m to mm) so-so

2
Few masks
> Low voltage
so-so
Magnetic Actuation
> Many variations on the magnetic actuators theme
Permanent magnets interacting with an external field

Permanent magnets interacting with current-carrying coils
Current carrying conductors interacting with an external field
Variable reluctance devices
> An important good feature: high force

Current drive means high power dissipation
Fabrication complexity and severe materials challenges
Electromagnetic actuators
Figure 3 on p. 79 in Cho, H. J., and C. H. Ahn. "A Bidirectional

Magnetic Microactuator Using Electroplated Permanent Magnet
Arrays." Journal of Microelectromechanical Systems 11, no. 1
(February 2002): 78-84. 2002 IEEE.
Figure 1 on p. 79 in Cho, H. J., and C. H. Ahn. "A Bidirectional

Magnetic Microactuator Using Electroplated Permanent Magnet
Arrays." Journal of Microelectromechanical Systems 11, no. 1
(February 2002): 78-84. 2002 IEEE.
External magnetic field actuators

http://www.sciencedirect.com. Used with permission.
Figure 1 on p. 260 in Khoo, M., and C. Liu. "Micro Magnetic Silicone Elastome
Membrane Actuator. Sensors and Actuators A, Physical 89, no. 3 (April 2001):
259-266.

http://www.sciencedirect.com. Used with permission.
Figure 6 on p. 262 in Khoo, M., and C. Liu. "Micro Magnetic Silicone Elastome
http://www.sciencedirect.com. Used with permission. Membrane Actuator. Sensors and Actuators A, Physical 89, no. 3 (April 2001):
Figure 2 on p. 260 in Khoo, M., and C. Liu. "Micro Magnetic Silicone Elastome 259-266.
Membrane Actuator. Sensors and Actuators A, Physical 89, no. 3 (April 2001):
259-266.
External magnetic field actuator
Variable Reluctance Actuator Concept

> High permeability magnetic
deformable and fixed
structures form a magnetic
circuit that flux penetrates
> Increasing current through coil

increases magnetic flux
> Increasing magnetic flux

through the circuit narrows
actuator gap and draws
structures together
Figure 8 on p. 43 in De Los Santos, H. J., G. Fischer,

H. A. C. Tilmans, and J. T. M. van Beek. "RF MEMS for
Ubiquitous Wireless Connectivity. Part I. Fabrication."
IEEE Microwave Magazine 5, no. 4 (December 2004): 36-49.
2004 IEEE.
Variable Reluctance Relay
Figure 2 on p. 26 in Tilmans, H.A.C., E. Fullin, H. Ziad, M. D. J. Van de Peer, J. Kesters, E. Van Geffen,
J. Bergqvist, M. Pantus, E. Beyne, K. Baert, and F. Naso. "A Fully-packaged Electromagnetic Microrelay."
MEMS '99: Twelfth IEEE International Conference on Micro Electro Mechanical Systems technical digest:
Orlando, Florida, USA, January 17-21, 1999. Piscataway, NJ: IEEE, 1999, pp. 25-30. ISBN: 9780780351943.
1999 IEEE.
> Two-wafer, flip-chip bonded implementation of variable

>
reluctance concept; thick electroplated structures

Switching speed > 500 Hz, operated at 2 V and 8 mA, 2 mN force
Variable Reluctance Relay
Figure 5 on p. 27 in Tilmans, H. A. C., E. Fullin, H. Ziad, M. D. J. Van de

Peer, J. Kesters, E. Van Geffen, J. Bergqvist, M. Pantus, E. Beyne, K.
Baert, and F. Naso. "A Fully-packaged Electromagnetic Microrelay.
MEMS '99: Twelfth IEEE International Conference on Micro Electro
Mechanical Systems technical digest: Orlando, Florida, USA, January
17-21, 1999. Piscataway, NJ: IEEE, 1999, pp. 25-30. ISBN: 9780780351943.
1999 IEEE.
Figure 6 on p. 27 in Tilmans, H. A. C., E. Fullin, H. Ziad, M. D. J.

Van de Peer, J. Kesters, E. Van Geffen, J. Bergqvist, M. Pantus,
E. Beyne, K. Baert, and F. Naso. "A Fully-packaged Electromagnetic
Microrelay. MEMS '99: Twelfth IEEE International Conference on Micro
Electro Mechanical Systems technical digest: Orlando, Florida, USA,
January 17-21, 1999. Piscataway, NJ: IEEE, 1999, pp. 25-30. ISBN:
9780780351943. 1999 IEEE.
> Wafer 1: a magnetic substrate with thick plated coils, poles, and
>
infrastructure for flip-chip packaging

Wafer 2: a silicon substrate with a plated armature and
provision for electrical contact and packaging
Issues with Magnetic Actuation
> Fabrication challenges:

Thick conductors and magnetic features
Standard approach: plating into a mold
Plating often produces surfaces that are not flat
planarization required
> Material limitations:

Not all magnetic materials can be deposited with MEMS
fabrication techniques
> Process limitations:

Not CMOS compatible
Minimize temperature during processing to avoid Curie
temperature
Eliminates some process steps from consideration and
modifies many others
Magnetic Actuators and the Wish List

> Low power operation (but latching can help)
> Fast switching
so-so
> High force (approaching mN)
> Large analog-controllable actuator travel (large gaps)

2
Few masks
> Low voltage
Force and distance

108
106
MACRO
Magnetostrictive
Electroactive polymer
Maximum force (N)
104
Thermal bimorph
Piezoelectric
Pneumatic
102
Hydraulic
SMA
State change
Topology optimized
State change
Fluid expansion
1
Magnetostrictive
10-2
Electromagnetic
Magnetic relay
Solid expansion
Piezoelectric expansion
10-4
SMA
External field
Scratch drive
Comb drive
Inchworm
Curved electrode
Thermal bimorph
Parallel-Plate
Impact actuator
10-6
10-8
Electrostatic relay
Thermal relay
Piezoelectric bimorph
Distributed actuator
Repulsive force
Electromagnetic
10-7
10-6
10-5
10-4
10-3
10-2
10-1
Maximum displacement (m)

Adapted from Figure 1 on p. S155 in Bell, D. J., T. J. Lu, N. A. Fleck, and S. M. Spearing. "MEMS Actuators and
Sensors: Observations on Their Performance and Selection for Purpose." Journal of Micromechanics and Microengineering
15 (June 2005): S153-S164.
Not all macro actuators have micro versions, and

not all micro actuators have macro versions!
Frequency and displacement

109
108
Magnetostrictive
MACRO
Parallel-Plate
Maximum frequency (Hz)
106
Piezoelectric
Piezoelectric bimorph
Electrostatic relay
Magnetic relay
Curved electrode
Repulsive force
Piezoelectric expansion
Electromagnetic
104
External field
Solid expansion
Electroactive polymer
Comb drive
102
Magnetostrictive
Hydraulic
Thermal bimorph
Thermal relay
Electromagnetic
Thermal bimorph
Fluid expansion
Topology optimized
Pneumatic
State change
State change
SMA
S-Shaped actuator
SMA
Inchworm
Distributed actuator
10-2
Scratch drive
Impact actuator
10-4
10-7
10-6
10-5
10-4
10-3
10-2
10-1
Maximum displacement (m)

Adapted from Figure 3 on p. S157 in Bell, D. J., T. J. Lu, N. A. Fleck, and S. M. Spearing.
"MEMS Actuators and Sensors: Observations on Their Performance and Selection for Purpose."
Journal of Micromechanics and Microengineering 15 (June 2005): S153-S164.
Different Options Have Different Advantages

Electrostatic
Thermal
Piezo
Magnetic
Low power
Fast switch
High force
Large travel
Simple fab
Low voltage
Robustness
Sense/actuate
> What is most important for your application?

> What suboptimal parameters can your application tolerate?
Conclusions
> An awareness of the options can give you a head start in
designing a device or process, and improve your device
intuition
Literature awareness
Design charts
> Other factors come into play as well:
Ease of fabrication
Materials compatibilities
Resolution
Calibration
Robustness
Power consumption
Case Study: Power MEMS
Carol Livermore
Thanks to the MIT microengine team, past and present, for many of
these materials. Thanks also to A. Forte, J. Yoon, and T. Lyszczarz of
Lincoln Laboratory.
Outline
> Brief overview of power MEMS
> MIT microengine
What to make it from, and how?
High speed rotation
Combustion
Motors and generators
Putting it all together
Motivation for power MEMS

> MEMS are well established for many sensing and actuation
applications (accelerometers, pressure sensors, fine positioning
of small components)
> Many useful macroscale systems have high power requirements

and typically use macro-scale, non-MEMS solutions
Electric power generation

Propulsion
Cooling
Lasers
> When the performance of high power macro-scale systems

limits the system performance, look for solutions wherever they
may be!
Sometimes scaling favors MEMS

and sometimes it doesnt.
What to expect today

> No commercial case study today
Consider portable power your laptop still uses batteries!
But, the market is there for a truly improved system-level
solution
> What we hope to get out of this

Contrast with the common temptation to focus on some parts
of a design and assume that a solution exists for the

remaining parts
This is a failure often seen in packaging
But, power MEMS tend to push against all limits at the same
time: thermal limits, strength of materials, breakdown
voltages, thermal
See examples in which these were (or werent) successfully
considered
Working within fabrication limits
Options for electric power generation

> Macroscale heat engines
A clear win when you can find a wall to plug into, because you have
dense energy storage in combustible fuels and efficient generation
in large scale systems
> Batteries
The usual portable power solution, but they can be cumbersome for
high power/long usage applications
> Fuel cells
Hydrogen + oxygen = water + electricity

One concern: size of total system required to make it work
> Power MEMS
Fuel burning systems: miniature heat engines of various types,

micro fuel cells, thermally driven systems (i.e. thermoelectric)
Energy harvesters: vibrations and other motions, waste heat can

provide low levels of power
Extractable energy
Portable power metrics
batteries
fueled generator
System weight
hardware weight
Other metric: power level vs. system weight
Piezoelectric energy harvester
Piezoelectric cantilever converts strain

energy to electric energy.
___________________ Used with permission. Figure 3
on p. 18 in Jeon, Y. B., R. Sood, J.-h. Jeong, and S.-G. Kim. "MEMS Power Generator with
Transverse Mode Thin Film PZT." Sensors and Actuators A, Physical 122, no. 1 (July 2005): 16-22.
1 W power output
Fabrication and stress control are not

trivial!
___________________ Used
with permission. Figure 11 on p. 21 in Jeon, Y. B., R. Sood,
J.-h. Jeong, and S.-G. Kim. "MEMS Power Generator with
Transverse Mode Thin Film PZT." Sensors and Actuators A,
Physical 122, no. 1 (July 2005): 16-22.
___________________ Used with permission. Figure 5 on p. 19 in

Jeon, Y. B., R. Sood, J.-h. Jeong, and S.-G. Kim. "MEMS Power Generator with Transverse Mode
Thin Film PZT." Sensors and Actuators A, Physical 122, no. 1 (July 2005): 16-22.
MEMS fuel processor

Figure 1 on p. 601 in Arana, L.R., S. B. Schaevitz, A. J. Franz, M. A. Schmidt, and
K. F. Jensen, "A Microfabricated Suspended-tube Chemical Reactor for Thermally Efficient
Fuel Processing." Journal of Microelectromechanical Systems 12, no. 5 (2003): 600-612.
2003 IEEE.
Figure 2 on p. 602 in Arana, L. R., S. B.

Schaevitz, A. J. Franz, M. A. Schmidt, and K.
F. Jensen, "A Microfabricated Suspended-tube
Chemical Reactor for Thermally Efficient
Fuel Processing." Journal of
Microelectromechanical Systems 12, no.
5 (2003): 600-612. 2003 IEEE.
MEMS fuel processor
Figure 4 on p. 605 in Arana, L. R., S. B. Schaevitz, A. J. Franz,

M. A. Schmidt, and K. F. Jensen. "A Microfabricated Suspended-tube
Chemical Reactor for Thermally Efficient Fuel Processing."
Journal of Microelectromechanical Systems 12, no. 5
(2003): 600-612. 2003 IEEE.
Figure 5 on p. 606 in Arana, L. R., S. B. Schaevitz, A. J. Franz, M. A. Schmidt,

and K. F. Jensen, "A Microfabricated Suspended-tube Chemical Reactor for
Thermally Efficient Fuel Processing." Journal of Microelectromechanical
Systems 12, no. 5 (2003): 600-612. 2003 IEEE.
Outline
> Overview of power MEMS
> MIT microengine
What to make it from, and how?
High speed rotation
Combustion
Motors and generators
Detailed case study: MIT microengine project

> Goal: an electric generator driven by a miniature gas turbine
engine, with overall performance exceeding that of the best
batteries
> Unlikely to compete favorably with macroscale gas turbines

lower efficiency
> But, hydrocarbon fuels have high energy density (of order
13,000 W-hr/kg), so even a lower efficiency may outperform
batteries (up to ~200 W-hr/kg for rechargeables) on overall
energy density
> Other consideration: hardware size

> Similar considerations for all fuel-burning power generators
Micro gas turbine engine

Starting air in
Compressor
P3
Inlet
3.7 mm
Exhaust
21 mm
Combustor
Turbine
Thrust = 11 g
Turbine inlet temp = 1600K (2421F)
Fuel burn = 16 g/hr
Rotor speed = 1.2 x 106 RPM
Engine weight = 2 grams
Exhaust gas temp = 970C
Portable compact power sources
> Approach
Simple cycle gas turbine

Direct drive generator (1.2M RPM)
MEMS fabrication
> Near-term performance goals
5% efficiency (chemical to electrical)

1-10 watts output
What to make the microengine from?

> Considerations:
Combustion and high temperature
Stresses from high rotational speed
Need for highly-controlled feature geometries
> Candidate materials:

Nothing from the polymer family
Conventionally-machined but tiny metal parts (known to be okay
from high T/strength point of view, not limited to 2D patterns,
possibly easier to machine, likely harder to assemble)
Single crystal silicon (high T OK, readily micromachined, strength?)

Other microfabrication-compatible materials, such as silicon
carbide (better high T performance, not so easy to micromachine)
Design choice: micromachined structure

> Silicon has deep reactive ion etching (DRIE)
> DRIE exists for silicon carbide as well, but it is slower and does
not give as good a profile
> Conclusion: to start out, we are stuck with silicon or nothing.

So can silicon do the job?
Good news: single crystal silicon has close to zero defects, so it

will have few points of inherent weakness
More good news: silicon is lightweight, so it will have less

tendency to tear itself apart than a heavier material would
Bad news: silicon is brittle, so if something bad happens, it will

likely involve catastrophic failure
> Parallel approach chosen: demonstrate engine in silicon,

measure silicons properties, and look into silicon carbide
technology
Microfabricated vs. conventional materials for high

temperature structures
Static Structure
Rotating Structure
Courtesy of H. Moon, L. Anand, and S. Mark Spearing. Used with permission.

Other structural concerns

> Factors affecting the strength of
micromachined silicon devices:
Surface finish of etched surfaces (etch

defects can initiate fracture)
Fillet radius
roughness control
Strength of wafer bonds

Courtesy of S. Mark Spearing andcompany. Used with permission.
Deep etch profiles: does base have a
fillet radius or a notched undercut?
> Manufacturability and performance of silicon

carbide structures
Rotor FEM
Courtesy of S. Mark Spearing and company.
CVD SiC Moulding
TEM of CVD SiC/Si
Courtesy of S. Mark Spearing and company. Used with permission.

Courtesy of S. Mark Spearing and company. Used with permission.
Fabrication approach: DRIE and wafer bonding

1. Pattern front side of Si wafer using
photolithography
2. Etch deep, straight-walled etch with an
anisotropic deep reactive ion etcher
3. Pattern back side of wafer, aligning to
front side features
4. Deep reactive ion etch the back side of
wafer
5. Align and fusion bond first patterned
wafer to a second patterned wafer
Patterning of six-wafer engine stack

Pros and cons of fabrication approach

> To first order, in plane complexity is free
Just use a more complicated mask
> DRIE can produce high aspect ratio features (30:1) and etch
through wafers
BUT
> Deep reactive ion etching produces extruded 2D rather than true
3D
Limits design, for example turbomachinery design
> Vertical thickness and vertical complexity mean adding more

wafers to the stack
More processing, more wafer bonding
> Pushing technologies to limits: attrition

Microengines
Cross-sectioned microengine

3-D cutaway of a micromachined microengine. Photograph by Jonathan Protz.
MIT microturbine
x20
2mm
10kv
12mm
C.C. Lin
Development approach
> System is too complex to develop as a whole

> Split off subsystems and technologies
Bearings, turbomachinery, high speed rotation
Combustion
Packaging for high temperature and pressure
Materials properties
Electromechanics: motors and generators
> Approach each through a combination of modeling

and creation of stand-alone subsystem test devices
Micromachined gas bearings

> Engine rotor must spin > 1 Mrpm
> Rotation supported on micromachined gas bearings
Pressurized air is injected into narrow gaps between rotating
and stationary components
As rotor approaches wall, pressure increases and returns

rotor to center
> Thrust bearings

Provide axial support to prevent up-and-down motion of rotor
disk
> Journal bearing

Provides in-plane support to prevent rotor from crashing
sideways into housing
Micro-bearing test device
Figure 8 on p. 162 in Wong, Chee Wei, Xin Zhang, S. A. Jacobson, and A. H. Epstein. "A Self-acting Gas Thrust
Bearing for High-speed Microrotors." Journal of Microelectromechanical Systems 13, no. 2 (April 2004): 158-164.
2004 IEEE.
Micro vs. conventional gas bearings

> Gas bearings are used in macro devices
Same concept as microdevice, very effective
> Aspect ratio of rotor is different
Macro: long, cylindrical rotor
Micro: flat disk with very little outer edge area on which to
carry load
> Balancing scheme

Macro rotors are tested and balanced iteratively
How to balance a micromachined rotor? Difficult to build it
evenly, difficult to adjust balance later.
> Bottom line for micro: precision, precision
300 m long journal bearing

300 micrometer long journal bearing.
Journal Bearing Quality Control: Wafer to Wafer

Wafer #1
Wafer #2
Courtesy of Hanqing Li, et al. Used with permission.

Rotor balance drives fabrication requirements

-Compressor Etch Uniformity Measurements-
Etch Depth (um)
Etch Depth Variation on a Single Die

238
236
234
232
230
228
226
224
222
220
218
Data A
Best Fit SIN
Data B
Best Fit SIN
Data C
Best Fit SIN
Data D
Best Fit SIN
0
200
400
Angular Location (deg)
Courtesy of Chiang Juay Teo. Used with permission.
Micro air bearing testing
Figure 1 on p. 142 in Frechette, L. G., S. A.

Jacobson, K. S. Breuer, F. F. Ehrich, R.
Ghodssi, R. Khanna, Chee Wei Wong, Xin
Zhang, M. A. Schmidt, and A. H. Epstein. "High
-speed Microfabricated Silicon Turbomachinery
and Fluid Film Bearings" Journal of Microelectro
mechanical Systems 14, no. 1 (February 2005):
141-152. 2005 IEEE.
Courtesy of The MIT MicroEngine Project.

Figure 8 on p. 162 in Wong, Chee Wei, Xin Zhang, S. A. Jacobson,

and A. H. Epstein. "A Self-acting Gas Thrust Bearing for High-speed Microrotors."
Journal of Microelectromechanical Systems 13, no. 2 (April 2004): 158-164.
2004 IEEE.
Figure 10 on p. 148 in Frechette, L. G., S. A. Jacobson, K. S. Breuer, F.

F. Ehrich, R. Ghodssi, R. Khanna, Chee Wei Wong, Xin Zhang, M. A.
Schmidt, and A. H. Epstein. "High-speed Microfabricated Silicon Turbomachinery
and Fluid Film Bearings" Journal of Microelectromechanical Systems 14, no.
1 (February 2005): 141-152. 2005 IEEE.
Combustion in microengine
> Requirements
Sustain a flame in a microcombustor at the design mass flow rates

Make sure that the reactants have enough time to burn in
combustor
Make sure that the structure can survive the high temperatures and
(for rotating components) the high stress
> Reactant mixing times can scale with size and geometry
> Necessary reaction times are independent of scale
Tradeoffs among high flow rate (short residence time), reaction

times for different fuels, and complexity (design of combustion
chamber, use of catalysts)
> Burn hydrogen first, then more difficult hydrocarbons
200 Watt microcombustor
1.5 mm
Numerical Simulation Results

Courtesy of Amit Mehra and Ian Waitz, et al. Used with permission.
Operation at 1600K
Courtesy of Amit Mehra and Ian Waitz, et al.
Catalytic microcombustor
Courtesy of C. Spadaccini, et al. Used with permission.

Courtesy of C. Spadaccini, et al. Used with permission.
Combustion with catalyst permits high mass flow rate

hydrocarbon combustion but complicates fabrication.
Electric motors and generators

> Thoroughly used and understood at the macro scale
Optimized for macro performance, cost, etc.
> Some different concerns at the microscale guide type of
machine chosen
Materials and process compatibility: for example,

reconciling metals or magnetic materials with high T in wafer
bonding
The most convenient structural material (Si) conducts

The less control required the better, since instrumenting
MEMS is difficult
Change of geometry: for microsystems, easier to put active

area on planar surface than on outside of a cylinder
> Potential for Watt-scale outputs

Magnetic induction machine
Courtesy of Hur Koser, and Jeffrey Lang, et al. Used with permission.
Conceptually, its like the motor in your refrigerator.

Uses thick, electroplated structures (copper and magnetic materials).
Pros and cons of the magnetic machine
> Pros
Well understood at the macro scale

High power (multiple Watts) possible
Large gaps between moving parts minimizes viscous losses
> Cons
Challenges of microfabrication with magnetic materials

Magnetic materials heavier and less uniform than silicon:
problems with high speed rotation

Many magnetic materials dont like high temperatures:
engine operation? wafer bond anneals?
> Status
MEMS magnetic generators demonstrated at low speeds,

with macroscale bearings: Watts of power produced
Electric induction motor schematic

insulator
stator
E
-----
+++++
++++ ------
-----
+++++
electrodes
-----
+++++ ------ +++++
poor conductor
rotor
insulator
speed of stator excitation

speed of rotor film conduction
rotor speed
How to implement the electric induction machine?

> Requirements
Connect electrodes into phases, and connect phases to external
leads
Low resistance electrodes and interconnections to minimize power
losses
Low stray capacitance to minimize power losses (insulator beneath
electrodes and leads)
Bearings that can support high speed rotation (tight tolerances, no
leaks!)
Tolerate high temperatures in fabrication and, ultimately, in engine
operation
Final structure must be wafer-bondable
> Contradictions
Metals have low resistivity but dont like high temperatures

Need silicon micromachining to make good bearings, but silicon
conducts and contributes to stray capacitance problem
A more realistic picture of the motor/generator

Si substrate
20 m STATOR
oxide
1 m
3 m
10 m
ROTOR
oxide
Thick oxides minimize stray

capacitance to substrate
Low resistivity platinum
electrodes and interconnects
polySi rotor film
(moderately doped)
Si substrate
Device diameter = 4.0 mm

786 electrodes grouped in 6 phases
Electrode voltage = 300 V peak
Electrical frequency = 1.5 MHz
Mechanical frequency = 1 Mrpm
Major challenge: wafer bonding

> Thick oxides minimize capacitance to substrate, but they create
wafer bow that complicates or prevents wafer bonding
> Wafer bonding works best with pristine silicon surfaces
Preferred: coat surfaces to protect during processing, then uncover
at the very end

This case: need to wafer bond surfaces that are covered with thin
films (film roughness, build up of dirt?)
Approach: polishing
> Metal electrodes and leads
Typical bond anneal temperature > 1000 C

Thin films of platinum agglomerate by about 800 C
Compromise: 700 C for 21 hours
Also, wafers need to be clean in order to bond, but metals (Ti) etch
in typical cleaning chemistries
Compromise between cleanliness and preservation of features:
nanostrip to remove organics instead of RCA or piranha
Motor/generator schematic cross-section
ROTOR
PLATE
STATOR
PLATE
TiPt
SiO2
PolySi
Si
Rotating Si
300 V, 1.5 MHz excitation

Stator
Parts the stator:

Electrodes and interconnections are 0.3 um thick platinum, deposited
over an adhesion layer of titanium. Thick oxide (20 um thick)
minimizes capacitance between electrodes and substrate.
Stator
1 m aluminum sacrificial layer
protective backside oxide
Stator
Remove aluminum where pits will be.

Leave resist mask.
Stator
Etch 20 m deep pits in substrate.
Stator
Fill pits with 20 m PECVD TEOS oxide.
Stator
Lift off field oxide by etching Al (HCl for a week).
Stator
Remove spikes: deposit thin oxide and CMP.
Stator
Deposit interconnect rings (Pt on Ti) with an image reversal liftoff

process.
Rings will group stator electrodes into six phases.
Stator
Deposit interlayer dielectric (PECVD TEOS oxide).

Etch via contact holes.
Stator
Deposit electrodes (Pt on Ti) with an image reversal liftoff process.
Stator
Deposit PECVD TEOS bonding oxide on both sides.

CMP oxide until surface is bondable.
Anneal to outgas hydrogen from PECVD oxide! (Otherwise, gas
bubbles will split wafers apart during bonding.)
Stator
Etch plena: 575 um deep.
Stator
Etch nozzles: 100 um deep.
Electric stator
Stator withstands electrode-to-electrode voltages of 300 Vac across 4 m

gaps defect free lithography and smooth features.
Wafer levels: motor/generator
Bonding in the presence of local bow

oxide
Si
4 mm
On some bubbles it bonds; in others, it doesnt.

Generator Fabrication: Stack assembly

After contact: ~ 5/12 die yield
After long press: ~7/12 die yield
After thermal press, 4h, 500C,
3kN, uniform steel field
electrode: 12/12 die bonded
Contact
Courtesy
Of Lodewyk
Steyn, et al.
Used with
permission.
Press
Courtesy
Of Lodewyk
Steyn, et al.
Used with
permission.
Thermal
Press
Courtesy of Lodewyk Steyn, et al. Used with permission.
Bonded generator die
Courtesy of Lodewyk Steyn, et al. Used with permission.
The payoff for all of this care

> The motors and generators work
> Generators spin at up to 850,000 rpm and source net power to a
load (Lodewyk Steyns Ph.D. thesis)
> Why this was possible:
Careful design, including all strays both inside device and out and
taking into account what is fabricatable
Careful fabrication: meeting multiple hard-to-achieve specs at once

Careful models and test

> Microengine successes: careful design and detail-oriented,
consistent fabrication process development can meet exacting
specifications, and work.
> Challenge: attrition! Maybe you can achieve ten miracles, but
what is your yield at achieving them all at the same time?
> Which approaches will be commercially successful?
Alternatives to batteries are desperately needed

Solutions must meet the requirements adequately and offer an
advantage: size, convenience, cost?
Consider everything: all of the system components (balance of

plant, external electronics, any necessary fuels, etc.) when making
comparisons
CASE STUDY:
MEMS-Based Projection Displays
Carol Livermore*
Outline
> Reflection vs. diffraction

Texas Instruments DMD reflective display
Silicon Light Machines diffractive display
> DMD-based display: the basics

What it is
How its made
How it works
> DMD-based display: the details

Reliability: why might this fail, and why doesnt it usually
fail?
Packaging
Test procedures
The Texas Instruments DMD
D LP
TM
Digital light processing
A texas instruments technology
5 KO
1.0 mm
X1, 800
9 mirror pixels
1,310,720 mirror pixels
(1280 x 1024)
Projecting with the DMD
Image removed due to copyright restrictions. Figure 20.2 in Senturia, Stephen D. Microsystem
Design. Boston, MA: Kluwer Academic Publishers, 2001, p. 533. ISBN: 9780792372462.
The Silicon Light Machines Approach

Figure 1.4 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers, 2001, p. 7. ISBN: 9780792372462.
With no beam deflection,

light is reflected
With alternate beam deflection,

light is strongly diffracted
Pixel Operation
Incident light
Screen
GLV Pixel
(ribbon axis
into page)
Diffracted light
Direction of ribbon motion
Lens

Projecting an Image
GLV array
(array axis
into page)
Horizontal scan mirror
Ribbon axis
Direction of
ribbon motion
Screen
Diffracted light path
Horizontal image direction

(vertical image direction into page)
Device Wafer

Figure 20.5 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic
Outline


What it is
How its made
How it works

Reliability: why might this fail, and why doesnt it usually
fail?
Packaging
Test procedures
Timeline of the DMD at TI

> 1977: Initial explorations (DARPA contract)
> 1987: Demonstration of the DMD
> 1992: Is this commercially viable?
> 1994: Public demonstration of prototype
> 1996: First units shipped
> More than ten million units shipped
> Initial focus limited to projectors to establish base market
> Jump to TVs, theater projection
> Now branching out into other markets: lithography, medical
imaging, scientific imaging
The pixels
> One mechanical mirror per optical
Mirror
pixel
> 16 m aluminum mirrors, 17 m on

center
Mirror support post

Landing tips
> Address electronics under each
Torsion hinge
pixel
Address
electrode
Yoke

Figure 51 on p. 39 in Hornbeck, Larry J. "From
Cathode Rays to Digital Micromirrors: A History of Electronic
Projection Display Technology." Texas Instruments
Technical Journal 15, no. 3 (July-September 1998): 7-46.
Hinge support post
Electrode support post
Metal 3 address pads

Landing sites
Bias/Reset bus
To SRAM
Pixel operation
> Pixels rotate 10 degrees in
either direction
> Mirrors pull in

> Motion is limited by
mechanical stops
> On: +10 degrees

Figures 48 and 50 in Hornbeck, Larry J. "From
Cathode Rays to Digital Micromirrors: A History of Electronic
Projection Display Technology." Texas Instruments Technical
Journal 15, no. 3 (July-September 1998): 7-46.
> Off: -10 degrees
System operation
> Grayscale obtained by alternating
each mirror between on and off
positions in time
Multiple switch events per
frame update
> Color obtained by rotating color

wheel
Mirror switching events are
synchronized with wheel

See________________________
http://www.dlp.com/tech/what.aspx for more information.
> Color alternative: use three chips

> Other system elements: light
source, drive electronics,
switching algorithm, projection
optics
The product
> MEMS are fun, but products sell
> The core of the product is the digital display engine, or DDE
f/3 Projection lens
45 mm reflector f/1.5
Condenser lens
6 mm spot
size
Fold
mirrors
Condenser lens
Relay optics
Condenser f/1
Lamp 270W MH
DMD
Magnification = 3x
DMD diagonal = 17mm
Rotation axis
RGB color wheel
Fabrication considerations
> MEMS parts must be fabricated over SRAM memory cells
> MEMS processing must not damage circuits, including
aluminum interconnects
> Polysilicon? High temperature oxides?

> Alternate approach: aluminum as a structural material, with
photoresist as a sacrificial layer
> Dry release by plasma strip is a benefit
Fabrication process
Spacer-1
Metal-3
CMP oxide
(via 2 not shown)
Hinge
support post
Yoke
Hinge
Spacervia-1
Substrate with
CMDS address circuitry
After spacer-1 patterning
Oxide hinge mask
After yoke/Hinge etch and oxide strip

Mirror
Oxide mirror mask
Mirror
support post
Hinge metal
Spacervia-2
After oxide hinge mask patterning
Oxide yoke mask
Yoke (beam)
metal
After mirror oxide patterning

Mirror support post
Mirror
Hinge
Metal-3
After yoke oxide patterning
Spacer-2
Yoke Hinge support post
CMOS
Substrate
CMP Oxide
Completed device
Electromechanics: DMD Structure

Figure 51 on p. 39 in Hornbeck, Larry J. "From Cathode Rays to Digital Micromirrors: A History of Electronic
Projection Display Technology." Texas Instruments Technical Journal 15, no. 3 (July-September 1998): 7-46.
Torsional Pull-in Model

Torsional spring
Section view along diagonal
mirror support
Mirror
support
g
x1
L
x
Electrode
x
Actuation electrode
beneath mirror support

s
o
g
Electric field line

r
x1
x
P
Electrode
Capacitance Modeling
> Calculate capacitance
vs. tilt angle
C=
> Fit to cubic polynomial

> Perform conventional
C ( 0 )
1 + a10 + a330
C ( 0)
1
W * ( 0 ) = C ( 0 )V 2
2
W * ( 0 )
=
0
pull-in analysis
Normalized capacitance
2.4
2.2
2.0
1.8
1.6
1.4
0 =
1.2
1.0
0WL tan 0
g
0
0.02
0.04
0.06
Angle
k
k
a1
3a C V 2 3a
3a3C0V 2
3
3 0
1/ 4
k2
VPI =
2
3a1a3C0
Outline


What it is
How its made
How it works

Reliability: why might this fail, and why doesnt it happen
most of the time?
Packaging
Test procedures
Brainstorm: why might this fail?

> Breakage due to handling/shock
> Stiction (from surface contamination, moisture, or van der Waals
forces)
> Light exposure

> Thermal cycling
> Particle effects (electrical short, stuck mirrors, etc.)
> Metal fatigue in hinges
> Hinge memory (permanent deformation)
> Other mechanisms can impact yield right out of the fab: CMOS
defects, particles
The ratings
> Stiction (from surface contamination, moisture, or van der Waals
forces)
> Light exposure

> Thermal cycling
> Particle effects (electrical short, stuck mirrors, etc.)
> Hinge memory (permanent deformation)
> Green: no problem, Yellow: use preventive measures, Red: use
preventive measures and cross your fingers
Things not to worry about

Resonant frequencies range from about 100 kHz to the MHz range
Macroscopic shocks and vibrations cannot couple to those modes
Might worry about the package, though
Initially expected to be a problem

Test didnt show fatigue
Subsequent modeling shows that small size has a protective effect
Bulk materials: Dislocations accumulate at grain boundaries,
causing cracks
Thin film material: Structures are one grain thick, so stresses are
immediately relieved on the surface
Big picture: some solutions

> Stiction from surface contamination
Monitor voltage required to lift mirrors out of pull in
Too much voltage indicates a possible increase in surface
contamination and a need to check the process

Include spring tips at the contact point; stored energy provides a
mechanical assist
> Stiction from moisture

Package design (hermeticity, getters)
> Stiction from van der Waals forces

Anti-stiction passivation layers
> Light exposure

Figure 51 on p. 39 in Hornbeck, Larry J. "From Cathode
Rays to Digital Micromirrors: A History of Electronic
Projection Display Technology." Texas Instruments
Technical Journal 15, no. 3 (July-September 1998): 7-46.
No fundamental degradation observed after light exposure

However, UV exposure slightly increases the rate of stuck pixels
Solution: include a UV filter to limit exposure below 400 nm
Particles
> Particles limit yield AND reliability, since loose particles are a
failure waiting to happen
> Not many failures, but most are traceable to particles

Detailed analysis of each and every returned unit: what went
wrong, where did this particle come from, and how can I prevent it?
> Particle sources

Die attach adhesive can interact with antistiction coating
Debris from die separation
Generic handling
> Some elements of the ongoing anti-particle battle
Be careful!
Particle monitoring
Change die attach adhesive
Adjust die separation process
Hinge memory and thermal cycling

> The problem: if you leave a mirror actuated in one direction for
too long, the metal can creep
> Mirror develops a permanent tilt in that direction and ultimately

cannot be switched
> High temperatures are an aggravating factor

> Some solutions:
Choose a hinge material that is less prone to creep
Tailor the actuating voltage pulses to be able to transition mirrors
from a wider range of starting positions (this also offers higher
transition speed)
Reset pulse jiggles mirror out of position, even if its just going to
switch back to that position after the reset
Design projector system to control temperature

Packaging process I
> Preliminary die separation steps
Before release, spin coat a protective layer
Die saw partway through the wafer to form cleave lines
Clean, removing debris and protective layer
> Test for functionality at the wafer scale
Plasma ash to remove the sacrificial photoresist spacer
layers
Deposit an anti-adhesion passivation layer to prevent stiction

of landing tips during testing
Test for electrical and optical functionality on a test station

> Break to separate into dies
Packaging process II
> Final preparation for die attach
Plasma clean
Repassivate to prevent stiction in operation
> Attach die to a ceramic package with an unspecified adhesive
> Wirebond to make electrical connections
> Cap package with a welded-on metal lid containing an optical
window to form a hermetic seal
> Include an unspecified getter to control moisture, along the lines

of a zeolite
> Moisture control not only limits stiction, but impacts hinge
memory as well
The package
> Ceramic package
> Heat sink for temperature
3. Gold wire bonds

5. Seam weld
2. DMD
4. Hermetic optical Window
control
> Dust control critical to

prevent future failures
> Package validation:
7. Getter strips
1. Ceramic header
6. Heat sink stud

accelerated lifetime tests

(humidity and up to 100C)
on a selection of devices
Testing
> If one mirror on a chip doesnt work, the projector is broken
> For good reliability, the failure rate of projectors, EVER, should
be well below 1%
> Question: how do you ensure that youre not sending out a
batch of projectors that are just waiting to fail?
> Testing with more than just binary information

> Custom tool: the MirrorMaster
Drive DMD with electronics, inspect with a CCD camera on a
microscope
> Careful protocols
Bias Adhesion Mapping

> Gradually increase voltage to actuate mirrors, capturing an
image of mirrors at each step
> Distribution of switching and release voltages is an early

warning system for structural variations, surface contamination,
process problems
Conclusions
> Intuition can be deceiving. Who would have thought that you
could get reliability at such an immense scale?
> If you want people to get excited about your MEMS technology,
show them the product.
> If the MEMS part alone doesnt meet the spec, ask yourself if the
overall system can be designed to meet the spec.
Hinge memory was partly cured by materials and partly by design of

the control system
For more information, and credits

> Some of these images are from the Texas Instruments web site
http://www.dlp.com/
> Some are from these articles:

P.F. Van Kessel et al, A MEMS-Based Projection Display, Proc. of
the IEEE, vol. 86, p. 1687-1703 (1998).
M.R. Douglass, Lifetime Estimates and Unique Failure Mechanisms

of the Digital Micromirror Device (DMD), IEEE 36th Annual
International Reliability Physics Symposium, Reno, Nevada, 1998.
S. Jacobs et al, Hermeticity and Stiction in MEMS Packaging, IEEE

40th Annual International Reliability Physics Symposium, Dallas, TX,
2002.
CASE STUDY:
A Capacitive Accelerometer
Joel Voldman
Thanks to SDS and Tim Dennison (ADI)
Outline
> Accelerometer fundamentals

> Analog Devices accelerometer
History
Structure
Design and modeling
Fabrication and packaging
Noise and accuracy
Measurement choices
> Two approaches to measuring acceleration

Open loop: Measure change due to acceleration
Closed loop: A disturbance in a position control system
Accelerometer types
> Open vs. closed loop sensing
Open loop: Measure change due to acceleration
Closed loop: A disturbance in a position control system
> Quasi-static vs. resonant sensing

Quasi-static sensing
Mechanical resonant frequency > Frequency of
acceleration
Measure displacement due to acceleration

Optical, Capacitive, Piezoresistive, Tunneling
Resonant sensing
Measure change in resonant frequency
Due to position-dependent nonlinear spring
> Todays example involves a quasi-static accelerometer

Accelerometer fundamentals
> Displacement and acceleration
are coupled together by a
fundamental scaling law
A higher resonant frequency

implies less displacement
high f & low sensitivity
Measuring small accelerations

requires floppier structures
F ma
x= =
k
k
a
x= 2
high sensitivity and low f
> Johnson noise in the damping

mechanism gives rise to a
fundamental noise floor for
acceleration measurement
an ,rms
4 k B T 0
=
mQ
Accelerometer specifications
> Initial application arena was
automotive crash sensor
> Navigation sensors have

tighter specs
Table 1 on p. 1642 in Yazdi, N., F. Ayazi, and K. Najafi. "Micromachined Inertial

Sensors." Proceedings of the IEEE 86, no. 8 (August 1998): 1640-1659. 1998 IEEE.
Piezoresistive accelerometers
> Use piezoresistors to convert stress
in suspension beam change in
resistance change in voltage
> First MEMS accelerometer used

piezoresistors
Roylance and Angell, IEEE-TED
ED26:1911, 1979
Figure 1 on p. 1911 in Roylance, L. M., and J. B. Angell. "A Batch-fabricated Silicon

Accelerometer." IEEE Transactions on Electron Devices 26, no. 12 (December 1979):
1911-1917. 1979 IEEE.
Bulk micromachined
Glass capping wafers to damp and stop
motion
> Simple electronics

> Piezoresistors generally less
sensitive than capacitive detection
Figure 2 on p. 1911 in Roylance, L. M., and J. B. Angell. "A Batch-fabricated

Silicon Accelerometer." IEEE Transactions on Electron Devices 26, no. 12
(December 1979): 1911-1917. 1979 IEEE.
Capacitors for position measurement
> Single capacitors

Capacitance is
function of gap or area
Can be nonlinear

Parallel plate
> Differential
capacitors
One capacitor
increases while the
other decreases
Fixed
Interdigital
Fixed
Floating electrode
C1
C2
Differential overlap
Circuit model

Using a differential capacitor

> Differential drive creates sense signal proportional to
capacitance difference
Gives zero output for zero change
>
> Output linear with gap
+Vs
0
C1
C2
V0
C1
C1 C2
V0 = VS +
VS
( 2VS ) =
C1 + C2
C1 + C2
for parallel-plate capacitors where
0
-Vs
only g changes, this becomes

g 2 g1
V0 =
VS
g1 + g 2
Bulk-micromachined capacitive accelerometer

> Fabrication not reported, but
likely uses nested-mask
process
First mask
Second Mask
Silicon
Third Mask
A. First silicon etching
Second Mask
Silicon
Third Mask
B. Second silicon etching
Third Mask
Silicon
C. Third silicon etching
Adapted from Figure 3 on p. 688 in Sasayama, T., S. Suzuki, S. Tsuch itaii, A. Koide,
M. Suzuki, T. Nakazawa, and N. Ichikawa. "Highly Reliable Silicon Micro-machined Physical
Sensors In Mass Production." The 8th International Conference on Solid-State Sensors and
Actuators and Eurosensors IX: digest of technical papers. June 25-29, 1995, Stockholm, Sweden.
Stockholm, Sweden: Foundation for Sensor and Actuator Technology, 1995, pp. 687-690. ISBN: 9789163034732.
Figure originally from Koide, A., K. Sato, S. Suzuki, and M. Miki. Technical Digest of the 11th Sensor Symposium:
June 4-5, 1992, Arcadia Ichigaya, Tokyo. Tokyo, Japan: Institute of Electrical Engineers of Japan, 1992, pp. 23-26.
Thermal accelerometer
> MEMSIC thermal

convection
accelerometer
> Gas is proof mass
> Movement of gas

under acceleration
changes thermal
profile
Transimpedance circuits
> The simplest type of circuit measures the displacement current
in a capacitor using a transimpedance amplifier
Transimpedance converts current to voltage

Nulls out parasitic capacitance
> If source is DC, measure velocity of motion

RF
iC
> Velocity is not really what we want

Q = C ( x, t )Vc
v v+ = 0 Vc = Vs
dVS
dQ
C dx
iC =
= C ( x, t )
+ VS
dt
dt
x VS dt
V
C dx
iC = VS
= 0
x VS dt
RF
Vs
+
-
C(x)
CP
V0

V0 = RFVS
C dx
x VS dt
constant
Transimpedance circuits
CF
> If source is AC, can measure position

> First, must use frequency high
enough such that velocity term is
negligible
iC
Vs
> Second, operate above corner
+
-
RF
C(x)
CP
frequency of LP filter
iC =
V0
dV
dQ
C dx
= C ( x, t ) c + Vc
dt
dt
x Vc dt
VS = VS 0 cos(t ) = Re VS 0 e
j t
C dx
iC = C ( x, t ) j +
VS
x Vc dt
iC C ( x, t ) jVs

V0 = ic (C F // RF ) = ic
RF
1 + j C F RF
ic RF
ic
C ( x, t ) jVs
=
V0
jC F RF
j C F
j C F
V0
C ( x, t )
VS
CF
Non-inverting op-amp circuits
> Now there is no virtual ground and parasitic

capacitance appears in output
+Vs
C1
Vx
C2
+
CP
-Vs
V0
C1 C2
V0 = Vx =
VS
C1 + C2 + CP

AC Methods Require Demodulation

> For AC methods, output signal is a HF sinusoid
(carrier) multiplied by a LF signal
V0
C ( x, t )
VS (t )
CF
Voltage
> This is an amplitude-modulated (AM) signal

> We want to retrieve the low-frequency component
0
V0
Time
Synchronous Demodulation
> Use a nonlinear circuit to multiply V0 by an in-phase sinusoid
> This demodulates to baseband
> Relative phase is important
Differential capacitor
Vs0 cos(t)
A
Signal
source
V0
Vd
LPF
Output
Vout
-1
Inverter
C ( x, t )
V0
VS
CF
Low-pass filter
Amplifier
Synchronous
demodulator

Adapted from Figure 19.13 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic
C ( x, t )
V0
VS 0 cos(t )
CF
Vd = V0 Vs = V0 Vs 0 cos(t + )
Allow phase shift
Vd =
Vout
C ( x, t ) 2
Vs 0 [cos( ) + cos(2t + )]
2C F
C ( x, t ) 2
=
Vs 0 cos( )
2C F
Signal-to-noise Issues
> To get a big signal, use a big voltage

BUT
V0
C ( x, t )
VS (t )
CF
Voltage creates a force that can modify the state of

the mechanical system (analogous to the self-heating
problem in resistance measurement)
> Noise floor minimum often set by LPF bandwidth

> But amplifier noise will often dominate
> Genesis: An ADI engineer heard about forming

mechanical sensors on silicon
> Market pull was airbag accelerometers (50g)

Current product was $50
Auto manufacturers wanted $5 price point
> Team was formed in 1986, first product in 1993

Fabrication process was under development since early 80s
at Berkeley
> Initially partitioned system to integrate electronics onchip

This ensured that they could achieve good SNR
> BUT
Entailed large infrastructure costs that essentially hemmed
future opportunities
> This is an example where up-front system partitioning

has multi-decade consequences
ADI system partitioning
> How to integrate MEMS + circuits?

> Circuits typically are run on continually changing
(and improving) fabrication lines
> MEMS typically cannot economically support such

high throughputs
> Foundries will not accept non-pristine wafers

> Thus, must combine both in-house in a dedicated
foundry
This usually sets circuit technology
For ADI, foundry is in Cambridge and Limerick
ADI system partitioning
> How to integrate MEMS + circuits?

> Several different approaches
MEMS first
Circuits first
MEMS in the middle
> ADI chose MEMS-in-the-middle

Mostly developed at Berkeley
6 fab line
~1 million sensors/week (as of 2005)
Analog Devices ADXL50
Courtesy of Analog Devices, Inc. Used with permission.
Differential capacitor structure

Anchor
Fixed electrodes
Shuttle motion (proof mass)
Folded spring
Position sense region

(42 cells)
Self-test region
(12 cells)
Anchor

Closer views
Even closer
Fabrication sequence
BPSG
LPCVD
Nitride
Ox
p
p- substrate
Circuit area
BPSG
n+
n+ runner
Sacrifical oxide
LPCVD
Nitride
Gate oxide
Ground plane
Sensor poly
Ox
p
n+ runner
Circuit area
Sensor area
p- substrate
n+
Sensor area
PECVD Nitride
Metallization
PECVD Oxide
BPSG
Ox
p
n+ runner
p- substrate
Circuit area
n+
Sensor area

Adapted from Figures 19.23, 19.24, and 19.26 in Senturia, Stephen D. Microsystem Design. Boston, MA: Kluwer Academic Publishers,
2001, pp. 521 and 523. ISBN: 9780792372462.
Packaging
> When to do die saw, before or after release?

> ADI decided to do die separation after release and
invented a wafer-handling method to protect the
released region during sawing
One tape layer with holes corresponding to mechanical
region
A second tape layer covering the entire chip

Saw from the back (must have pre-positioned alignment
marks on wafer back to do this)
Packaging
> Processing issues

Stiction at- and post-release
Solved at-release stiction with bumps under poly
structures
Post-release stiction avoided with proprietary coating

Thermally evaporated silicone coating
Has to withstand packaging temps
> Laser trimming

Set offsets, slopes, etc.
At wafer scale
Before packaging
System diagram
> Oscillator provides AC waveform for sensing

> Waveforms:
V+
Vsense
V-

Stiffness of springs
> Parallel-plate approximation to
C sense
sense capacitance is off by

about 50%
y
42
60fF1
g0 y
g0
0 HL0
L1
> Beam bending model gives
L2
good estimate of stiffness
Folded
Motion
Unfolded
k2
5.6 N/m
3
3
6 ( 2 L1 ) + ( 2 L2 )
4
EWH
0 = 24.7 kHz
L1
L2
L2
L1
Almost rigid
Motion
Design and modeling

> Q of mechanical resonance is 5
Extremely hard to model
accurately
Squeezed film damping between
fingers
Couette drag beneath proof mass
Complex actual geometry
Rough model gives Q =34, a poor
estimate
stationary
finger
proof mass
fluid
ADXL50
> First accelerometer used feedback control to keep

plates fixed
> Lets use PD control to see what it affects
K ( s ) = K 0 (1 + s )
> Input is disturbance D(s) acceleration

> Output is force from controller F(s)
> H(s) is accelerometer: SMD
K0
(1 + s )
F (s)
HK
m
=
=
K
k + K0
D( s)
1 + HK
b
s2 + + 0 s +
m
m
m
ADXL50
> Use feedback to get both
Critical damping (when ON)
Insensitivity to material
properties
> Choose K0>>k

ADI chose 10x
> Critical damping is when

b2=4ac (Qclosed-loop=1/2)
Can pick K0 and to meet

both requirements
> Sensor response will be

insensitive to changes in k
K0
(1 + s )
F (s)
m
=
K k + K0
D( s)
b
s 2 + + 0 s +
m
m
m
K0
b
k + K0
4
+
=
m
m
m
K0
K0
b
2
+
m
m
m
0
K0
K0
+
Q
m
m
o.l .
K0
K0
1
2
m
m Qo.l .
K0
k
2
m
m
m
K0
The next generations
> In the next generation, ADI abandoned feedback

> Why?
After years of testing, ADI found that polySi was structurally
stable for intended markets
Feedback required extra electronics bigger chip $$

Needed external capacitor to set LPF
Extra cost, extra complexity
Closed-loop design was not ratiometric to power supply
Customers needed to measure supply voltage
DC bias at fingers for force feedback caused charges to
move and thus devices to drift
> Therefore, they removed feedback

Analog Devices Dies shots
Next generation specifications

> Text study is of ADXL150 (XL76)
> Text lists 22 specifications, covering sensitivity, range,
temperature range, supply voltage, nonlinearity, cross-axis
response, bandwidth, clock noise, drop test, shock survival, etc
etc.
> Also, response is ratiometric, proportional to the supply voltage

> Sensitivity is Vs = 38 mV/g
Vout
VS
=
+ aVS
2
offset
Noise and accuracy

> Noise is specified as 1 mg/Hz in a bandwidth from 10 Hz to 1000
Hz
> Corresponding Brownian noise estimate is half that value,

corresponding to a rms position noise of 0.013 nm
> Offset errors

If device is not perfectly balanced at zero g, turning on voltage
aggravates the offset
Accurate etching required special dummy features to ensure that

all cuts had the same profile (we have seen similar effects when we
looked at DRIE)
> Cross-axis sensitivity is low because of squeeze-film damping

and differential capacitor measurement
ADXL202 2-axis accelerometer
> Then moved from two 1-axis sensors to one 2-axis
Newer designs
> ADXL203 two-axis accelerometer

> Supports are in center of die to cancel 1st-order
stresses due to packaging
The latest design: ADXL40

> The newest designs use an SOI-MEMS process
Also developed at Berkeley
> Enables several circuit features

0.6 m CMOS allows 10x more transistors in same size
Allows poly fuse trims to be set on-chip
Can trim AFTER packaging
Figure 1 on p. 637 in Brosnihan, T. J., J. M. Bustillo, A. P. Pisano, and R. T. Howe. "Embedded Interconnect and Electrical Isolation for High-aspect-ratio,
SOI Inertial Instruments. In Transducers '97 Chicago: 1997 International Conference on Solid State Sensors and Actuators: digest of technical papers, June
16-19, 1997. Vol. 1. Piscataway, NJ: IEEE, 1997, pp. 637-640. ISBN: 9780780338296. 1997 IEEE.
The latest design: ADXL40
> MEMS
Higher-aspect ratio structures lead to more squeezed-film
damping Q=1
Trench isolation allows self-test to be electrically isolated

from sensing fingers
Allows 2x voltage applied 4x force

ADXL40
> Die shots
New accelerometer markets







Summary
> Accelerometers are a MEMS success story

> Early system partitioning decisions have had
profound downstream effects
Eases sensor design and sensing
Requires large internal infrastructure
And can never go to TSMC
BioMEMS Case Study: Microdevices for

PCR
Joel Voldman*
*with thanks to SDS
Outline
> What is hard about BioMEMS

> BioMEMS success stories
> DNA amplification and PCR
> Two designs
A static PCR thermocycler
A flow-thru design
Comparison
> Design evolution of static approach

> Conclusions
BioMEMS
> Applications of microsystems
Berkeley
to bioscience
Neural probes
Capillary electrophoresis
Drug delivery
Cellular engineering
Tissue engineering
Univ. Michigan
Courtesy of Kensall D. Wise.
Courtesy of Richard A. Mathies.


What is hard about BioMEMS
> The biological system is poorly defined

We fundamentally understand physics
We DONT fundamentally understand biology
Thus, only part of system can be truly predictively designed
> Intrinsic biological limitations can dictate system

performance
Protein-protein interaction kinetics
Polymerase error rates
> The materials (and thus processes) are often NOT

silicon (and thus harder)
We must move away from the most established fabrication
technologies
BioMEMS success stories
> Depending on the definition,

there are very few
> Commercial successes

PCB-mountable pressure sensor for medical applications.
Blood pressure sensors

Low-cost widget allows
devices to be disposable
Affymetrix DNA microarrays

Vastly decreases time and
cost for analyzing nucleic
acids
But these are not really

bioMEMS
> Depending on the definition,

there are very few
> Commercial successes

Blood pressure sensors
Low-cost widget allows
devices to be disposable
Courtesy of Affymetrix, Inc. Used with permission.
Affymetrix DNA microarrays

Vastly decreases time and
cost for analyzing nucleic
acids
But these are not really

bioMEMS
Courtesy of Affymetrix,
Inc. Used with permission.
Courtesy of Affymetrix, Inc. Used with permission.
> In the commercial sector,

there has been lots of hype
Success is uncertain

Lab-on-a-chip is about the size of a U. S. dime.
> Caliper/Aclara
Lab-on-a-chip
> I-stat
Portable blood analyzer
Uses ion-selective electrodes,
conductivity, etc. to measure

salts, glucose, etc.
Introduced ~1997
Purchased by Abbott
> Away from commercial sector and into basic science,

more successes arise
> Success can be defined as having impact on the

target audience
> Ken Wises neural probes
Univ. Michigan

CMOS + bulk micromachining
Puts op-amp right near neural recording sites amplifies
Courtesy of Kensall D. Wise.
and buffers weak (~V) signals

These are being used by neuroscientists in actual
experiments
Outline

> Two designs
A flow-thru design
Comparison

> Conclusions
DNA (deoxyribonucleic acid)

-O
information (genotype) that

determines phenotype (i.e., you)
T 3N
N1 A 4
3
8
9
5 6
N
4 G 1
3 2
3' 2'
4' 1'
-O
O
O
3' 2'
1'
4'
O
P
3' 2'
4' 1'
CH2
of amino acids in proteins
4'
2' 3'
-O
O
1'
4'
2' 3'
P
H
N1 G
4
2
5'
-O
C 3N
1'
6
N3 T1
N
A 1
-O
-O
5 6
N
H
5
4
C 6
3
N
1
O
P
4'
1'
3'
2'
5' CH2
-O
7 8
9
5'
3' 2'
4' 1'
contain genetic code
> This code determines the sequence
helical strands
Read 5 to 3
A sugar-phosphate backbone
Specific bases (A, C, G, T) that
CH3
> It consists of a two antiparallel
3'
> DNA contains the genetic
(B)
O
1'
4'
2' 3'
O
H
O
-O
3'
Lodish, MCB
DNA Amplification
> The bases pair specifically
A with T
C with G
> Specific enzymes (DNA

polymerases) can add
complementary nucleotides to an
existing template + primer
This is done in vivo in DNA
Repaired
DNA helix
Incoming
deoxyribonucleoside
triphosphate
Template strand
3'
3'
5'
Gap in helix
replication
> Was capitalized in vitro in

polymerase chain reaction (PCR)
Invented in 1985, Nobel Prize in
Alberts, MBC
1993
Polymerase chain reaction (PCR)
> Specifically amplify DNA starting from 1 double-
Cycle 2
Cycle 1
stranded copy
Region to be amplified
3'
5'
5'
3'
Primer 1
Target DNA
Add excess primers 1 and 2, dNTPs,

and Taq polymerase
o
Heat to 95 C to melt strands
o
Cool to 60 C to anneal primers
3'
Primer 2
5'
Primers extended by Taq
o
polymerase at 72 C
5'
3'
3'
5'
3'
5'
o

o
3'
5'
5'
3'
3'
o
polymerase at 72 C
5'
5'
3'
o
o
Lodish, MCB
Cycle 4
Cycle 3
Polymerase chain reaction (PCR)

polymerase at 72oC
Heat to 95oC to melt strands

Cool to 60oC to anneal primers

polymerase at 72oC
And so on
Lodish, MCB
PCR
> Key technological improvement
was use of polymerase that could
withstand high temperatures
Isolated from Thermus aquaticus
(Taq)
Dont have to add new polymerase
at each step
> The device is a simple

thermocycler

PCR machine by MJ Research.
> Allows amplification and detection

of small quantities of DNA
PCR cycles
> Taq extension rate ~60 nt/sec
Step 1: Denature template

100
> PCR products are typically a
> Typical protocols

~25-35 cycles at 1-3 min/cycle
~30 cycles 75 minutes
Temperature (oC)
few hundred bases
Need ~5 sec for extension

Plus time for diffusion
Step 3:
Primer
extension
90
80
70
60
Step 2:
Anneal primer
50
40
30
One " Cycle"

0
8
10
Minutes
12
14
16
PCR
> Cycle time is dominated by
ramp times due to thermal
inertia
Usually much longer than

kinetically needed
Property
Temp range
5-105 C
Set-point accuracy
0.25 C
Temperature uniformity
0.4 C
within 30
sec
Heating/cooling rate
~ 3 C/sec
Sample volume
50 l
Number of samples
96
Power required
850 W
> Transient and steady-state

temperature uniformity limits
cycle time & specificity
Spec
BioRad DNA Engine
Outline

> Two designs
A flow-thru design
Comparison

> Conclusions
Two approaches to miniaturization
> Decrease size of chamber

Vary temperature in time
> Use a flow-thru approach

Vary temperature in space (and therefore time)
> In both cases, the device is a thermal MEMS device

and the key is to reduce thermal response times
Batch PCR
> First reported by Northrup et al.
in 1993
Essentially a miniaturized
thermal cycler
Cover glass
Polyethylene tubing
10 mm
Silicone rubber
Silicon
0.5 mm
Low-stress
silicon nitride
Aluminum bond pad
Polysilicon heater

Adapted from Northrup, M. A., M. T. Ching, R. M. White, and R. T. Watson. "DNA Amplification with a
Microfabricated Reaction Chamber." In Transducers '93: the 7th International Conference on Solid-State
Sensors and Actuators: digest of technical papers: June 7-10, 1993, PACIFICO, Yokohama, Japan. Japan:
Hiroyuki Fujita, 1993, p. 924-926. ISBN: 9784990024727.
Northrup et al., Transducers 93, p924

Batch PCR
> Daniel et al. improved
thermal isolation

___________________ Used with permission.
Figure 5 on p. 84 in: Daniel, J. H., S. Iqbal, R. B. Millington, D. F. Moore, C. R. Lowe,
D. L. Leslie, M. A. Lee, and M. J. Pearce. "Silicon Microchambers for DNA Amplification.
Sensors and Actuators A, Physical 71, nos. 1-2 (November 1998): 81-88.
chamber
on
mesh isolati
beams
___________________ Used with permission.

Figure 3 on p. 83 in: Daniel, J. H., S. Iqbal, R. B. Millington, D. F. Moore, C. R. Lowe,
D. L. Leslie, M. A. Lee, and M. J. Pearce. "Silicon Microchambers for DNA Amplification.
Sensors and Actuators A, Physical 71, nos. 1-2 (November 1998): 81-88.
Continuous-flow PCR
> Developed by Kopp et al. in 1998

Buffer in
Sample in
Product out
60oC
Output capillary
Input capillaries
MM
77oC
Annealing
zone
Extension zone
95oC
Denaturing zone
Two-layer glass with etched flow channels
Adapted from Figure 1 on p. 1046 in Kopp, M. U., A. J. de Mello, and
A. Manz. "Chemical Amplification: Continuous-flow PCR on a Chip."
Science, New Series 280, no. 5366 (May 15, 1998): 1046-1048.

Adapted from Kopp, M. U., A. J. de Mello, and A. Manz. "Chemical Amplification:
Continuous-flow PCR on a Chip." Science, New Series 280, no. 5366 (May 15,
1998): 1046-1048.
Batch PCR
> Daniel reactor
SiN mesh structure, undercut with KOH
Made hydrophobic to keep water in chamber during
loading
Platinum heater resistors heat up beams

Two temperature sensing resistors
One on beams for feedback control
One on membrane to sense liquid temp
Use oil drop on top of liquid to prevent evaporation
___________________
Used with permission. Figure 3 on p. 83 in: Daniel, J. H.,
S. Iqbal, R. B. Millington, D. F. Moore, C. R. Lowe, D. L. Leslie,
M. A. Lee, and M. J. Pearce. "Silicon Microchambers for DNA
Amplification." Sensors and Actuators A, Physical 71,
nos. 1-2 (November 1998): 81-88.
Thermal modeling batch system

> Three steps
> Model chamber
> Model beams
> Combine the two, with
heating at beams
substrate
chamber substrate
beams

> Chamber model
> Assume rectangular cross-
y
x
section
> Assume dominant heat loss

through beams
2-D heat flow problem
Neglect conduction along top

and bottom
Temperature does not vary in

z
Eigenfunction expansion
solution to heat-flow eqn.
n x
m y m ,nt
T ( x, y, t ) = An ,m cos
cos
e
L
L
n m
x
y 1t
T1 ( x, y, t ) = A1 cos
cos
e
L
L
Lowest mode
> Interested in dominant time

constant
~
L2Cm m
f = = 2 =
1 2 D
2 2
1
L2

> Obtain lumped heat
capacity by weighing over
mode volume
> Extract thermal resistance

from time constant
L/2 L/2 H
x y
cos
cos dxdydz
L L
L / 2 L / 2 0
~ 2L 2L
C f = mCm
H
~
C f = mCm
Cf =
> Thermal resistance same

as zeroth-order model
suggests
~
4 mCm L2 H
f = Rf C f Rf =
1
8H
1
L 1
1 length 4
1
= 2 4=
Rf =
area
LH
8H
> For L=2 mm, =1.4 s

> H=400 m, volume = 2 l

1 1 length
Rb =
4 area
~
Cb = mCm volume
> Lumped elements for beams

Include beam capacitance
> Two circuits to model beams

Capacitor in center
Capacitor at edge
wall
> Both circuits contain same

energy in capacitor at
steady state
Capacitor at edge is simpler
substrate
Rb /2
beam
Rb
Rb /2
Substrate
Wall
Cb
Substrate
Wall
Cb /2


> Lumped circuit model of reactor
wall
> First-order lag between wall and

fluid temperature
1
Tf =
Tw
1+ f s
f =
fluid
substrate
L2
Rf
Rb
2 2 D
+
P
> Making L smaller reduces
Tw
-
lag
Substrate
Wall
+
Cb /2
Tf
Cf
-
Fluid in chamber

> Simulink proportional control circuit

> Saturation needed for maximum +/- voltage swing
Set to 0 to 15 V
Signal 1
+
-
Signal builder
As0
s + As0/K
Op-amp gain K
Fluid
u[1]^2/Rh
Saturation
Joule heat P
Wall
Switch
1
Rf*Cf.s + 1
Fluid temp
0
Rb*Rf*Cf*Cb/2.s2
Rb*Rf*Cf.s + Rb
+ (Rb *Cb/2 + (Rf + Rb) *Cf)s + 1
Transfer Fcn
Select feedback

Simulation response
100
90
> Wall temperature can be
80
> Wall to fluid heat transfer limits
Temp
controlled very quickly

performance
60
50
40
30
> Sensing fluid temperature
20
marginally reduces response

times
10
20
T (s)
30
40
10
20
30
40
120
Control
Fluid
Wall
100
Temp
But creates high-temp regions

at chamber wall
70
80
60
40
20
T (s)
Continuous-flow device
> System partitioning: put heaters off-chip

> Etch channels in glass, bond glass cover
Buffer in
Sample in
Product out
60oC
77oC
95oC
Two-layer glass with etched flow channels

Adapted from Figure 1 on p. 1046 in Kopp, M. U., A. J. de Mello, and A. Manz.
"Chemical Amplification: Continuous-flow PCR on a Chip." Science, New Series
280, no. 5366 (May 15, 1998): 1046-1048.
Thermal model of continuous-flow device

> Wall-to-fluid time constant is
same as in batch device
> L1=40 m, L2=90 m

> D=1.4x10-7 m2/s
> f= 1.2 ms
> 1000 faster than batch device!
L1 L2
f = 2
2 D
> Entrance length for thermal

equilibration
Le 3v f f
> Average flow velocity
vf =
Qf
area
= 20 mm/s
Le 60 m
This is much smaller
than zone lengths
L1
L2
> Thermal Pe number is ~10-3
Continuous-flow device
> What about Taylor dispersion?
> Pressure-driven may cause
k BT
D=
6R
multiple samples to coalesce
> Hydrodynamic radius of 1 kb

DNA ~ 50 nm
> Dispersivity is dominated by

convection
Samples will spread out a lot,

limiting usefulness for
multiple samples
LU (40 m)(0.02m / s )
5
~
2
10
Pe =
x
=
4.4 x10 8 cm 2 / s
D
Pe 2 L1
f
K = D1 +
210 L2
L1
f = 4
L2
K ~ 8 x108 D ~ 35 cm 2 / s
L1
~ 0.4
L2
Thermal lessons
> Wall temperature in most microsystems can be

quickly controlled
> Limiting step is wall-to-fluid heat transfer

> Solution is to minimize fluid characteristic length for
heat diffusion
Detecting PCR products

> Speeding up amplification is only half the battle
> DNA is not normally visible
> In conventional PCR, detect products by separating stained DNA
using electrophoresis in gel sieving matrix
This can take 0.5-2 hrs

Loading gel in electrophoresis apparatus.

Photograph of an illuminated gel after electrophoresis.

> Newer techniques allow real-time detection
Real-time PCR
> Integrate illumination/detection optics, thermal cycler, and

chemistry
Polymerization
5'
3'
5'
Forward primer
Cleavage
Reverse primer
5'
3'
5'
3'
5'
Strand displacement
5'
3'
5'
Polymerization completed
Q
5'
3'
5'
5'
3'
5'
5'
3'
5'
3'
5'
3'
5'
3'
ABI Taqman assay

Roche LightCycler
> Real-time thermocycler

> 1.5 mm-wide capillaries
> ~2-3 sec up and down

Schematic of the thermal chamber for the LightCycler(R) 2.0.
Comparison of two microfluidic PCR approaches
Continuous flow
Batch
Faster thermal response
Slower
No temp overshoot
Depends on control system
Static protocol
Can change protocol easily
Taylor dispersion effects,

and sample carryover
Sample carryover only
Optical detection more

complicated
Simpler optical detection
Materials issues
> Reactor surface must be compatible with PCR reagents
DNA, nucleoside triphosphates, polymerase, buffers
> Decreasing length scale and increasing SA/V hurts here

More molecules start to interact with surface
> Bare Si or SiN inhibits PCR

Probably due to denaturing of polymerase at surface
Silanizing or depositing/growing SiO2 helps
Add carrier protein (e.g., BSA) to block surface
> Kopp uses glass, silanization, surfactant, and buffer!

> Northrup use deposited SiO2 plus BSA
Outline

> Two designs
A flow-thru design
Comparison

> Conclusions
Evolution of chamber device

> Initial device introduced in
Cover glass
Polyethylene tubing
1993
> 1995-6
Silicone rubber
Silicon
0.5 mm
Two-heater chambers
Improved surface coating:
10 mm
Low-stress
silicon nitride
Aluminum bond pad
Polysilicon heater
silanize+BSA+polypro
Adapted from Northrup, M. A., M. T. Ching, R. M. White, and R. T. Watson. "DNA Amplification with a
insert
Microfabricated Reaction Chamber." In Transducers '93: the 7th International Conference on Solid-State
Fan for cooling
Sensors and Actuators: digest of technical papers: June 7-10, 1993, PACIFICO, Yokohama, Japan. Japan:
Hiroyuki Fujita, 1993, p. 924-926. ISBN: 9784990024727.
Chamber volume 20 l
30 sec cycle time
~10 C/s up, 2.5 C/s down
Real time-PCR via
coupled electrophoresis


Cepheid formed
> 1998
> Same two-heater chambers
> Portable application
> Up to 30 C/s up, 4 C/s down
> 17 sec minimum cycle time

> Usually use 35+ sec

Heater - rapid, precise
temperature control
> 2001 to present

> Abandoned Silicon Move to
plastic and NOT
microfabricated
Thin-film
wall
> Tubes are disposable thin-wall
Optical
excitation
window
(50 m) plastic that expands

upon introduction
> Tube is flat to decrease thermal
Circuitry - passes optical

information to computer
for analysis and display
response
> ~30 sec cycle time

> Ceramic chamber with thin film
tio
ta
i
c
Ex
D
et
ec
heater
> Thermistor temp sensor
Optical detection
window
Optics blocks - optical analysis, detect

and quantify up to four different DNA
targets simultaneously
tio

Conclusions
> BioMEMS commerical successes are still not here

> Designing the engineered part is often routine
> Interfacing with biology is where it gets hard
> Sometimes the right solution is to NOT microfab
MEMS Device Treasure Hunt assignments
Device Type
Accelerometer using capacitive sensing
Accelerometer using tunneling-based sensing
Silicon pressure sensor
Silicon pressure sensor
Microfabricated capillary electrophoresis
device
Microfluidic flow cytometer
Room-temperature microfabricated bolometer
Room-temperature microfabricated bolometer
MEMS thermal actuator
MEMS nanopositioner
MEMS tunable capacitor
Microfabricated optical switch
Surface micromachined tunable capacitor
Microfabricated electrophoresis chip for
proteins
Microfabricated RF switch
Active gaseous valve
Active liquid valve
Gyroscope (any type)
Micro mechanical resonator
Micro mechanical resonator
Specification for Best Performance

Sensitivity in gs
Sensitivity in gs
Operating temperature range
Pressure-measuring range (e.g., span)
DNA separation speed
Sorting rate
Response time
Sensitivity
Fastest response time
Largest force
Largest displacement
Best resolution
Actuation voltage
Power and insertion loss
Tuning ratio or range
Number of theoretical plates
Switching speed
Highest back pressure
Highest flow rate
Noise floor
Resonant frequency
Quality factor (Q)
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J Design and Fabrication
of Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/),
Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

Spring Term 2007
PROBLEM SET 1 (12 pts)
Issued: Lecture 2
Due: Lecture 4
MIT Server/MATLAB orientation

If you are not familiar with Matlab, work through the Matlab Primer so that you can enter variables, set
up matrices, make simple graphs, and create an M-file script to define a function. Postscript and PDF
versions of the Matlab Primer are available on the 6.777 web site.
If you are having trouble with MIT Server access or with this Matlab introduction, let a TA know
immediately.
Problem 2.6 (4pts): Dynamics with Matlab and Simulink
Dynamic modeling of MEMS devices will require us to perform time integrations of vector functions. In
this problem we will develop some of the nitty-gritty details needed to do these integrations successfully.
It will be necessary to become familiar with functions defined by M-files, and with the integration
functions (e.g., ode45). There are also some syntactical issues that must be handled correctly. In addition
to the plots, please include your m-file or Matlab diary showing your work.
(a) Consider a linear time-invariant dynamical system described by the following transfer function
H(s) relating an input X(s) to an output Y1(s):
H ( s) =
Y1 ( s )
1
= 2
X ( s) s + 2s + 4
This transfer function models some of the kinds of 2nd-order systems that we will encounter in this course
(e.g., accelerometer). Determine the differential equation relating y1(t) to x(t). Write the system as a pair
of coupled first-order differential equations with state variables y1(t) and y2 (t ) = y1 (t ) . Use the Matlab
command ode45 to integrate these equations forward in time from an initial state y1(0) = y2(0) = 0
subject to a step input x(t)=u(t). Plot y1(t) to show that this corresponds to an underdamped 2nd-order
system.
(b) An alternate approach is to define Matlab LTI models using the control systems toolbox. Define
your transfer function in Matlab using the tf command and plot its step response and Bode plot
using step and bode.
(c) Simulink can also be used to model dynamical systems. Create a Simulink model of this
dynamical system, using either a transfer function or differential equation formulation, and plot
y1(t) in response to a unit step as above.
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J Design and Fabrication of
Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/),
Problem 3.8 (2 pts): KOH etched diaphragm

We intend to use KOH etching to form a diaphragm on a (100) silicon wafer. It will be a square
membrane with a thickness well-defined by an etch stop (e.g., electrochemical etch stop or SOI wafer),
such that there are no first order thickness variations in the diaphragm itself. What feature size is required
to produce a square diaphragm with 400 m side length and 20 m thickness on a silicon wafer that is
500 m thick (the wafer thickness includes the membrane thickness)? What is the edge-length variation
of the diaphragm if the etch mask is misaligned 1 to the <110> direction? Assuming that the sensitivity
of a pressure sensor varies as the inverse fourth power of the diaphragm edge length, what percentage of
variation can be attributed to mask misalignment?
Problem 4.15 (2 pts): Crayon engineering: Debug and recreate a process and mask set for a
pressure-sensing silicon diaphragm
You are a young junior faculty member who has just hired your first graduate student, Wayford Roppar.
You have developed an idea for using a sealed-cavity pressure-sensing silicon diaphragm (1 mm across
and 15 m thick) that youre sure will make you famous and assure your tenure. You ask Wayford to
design a process flow for creating this simple structure, and Wayford returns with the process flow
detailed in Figure 1.
Being a seasoned MEMS designer, you immediately notice several critical errors with Wayfords
process (things that wont work or wont produce the result that Wayford shows in the cross
sections). Please find the critical errors in this process flow and, where possible, suggest
alternate approaches. Do not worry about the accumulation of errors, but rather treat each step
assuming that the structure up to that step could be created.
Then recreate a correct process flow along with the device cross sections at each step and the
associated mask set (with dimensions).
1
Figure 1: Process flow for a pressure-sensing silicon diaphragm.

Process steps:
1. Start with a double-side-polished n-type silicon wafer.
2. Perform photolithography using 1-m-thick positive photoresist to define the
diaphragm area.
3. Deep-reactive-ion etch the silicon to form the diaphragm; ash resist.
4. Anodically bond the silicon wafer to a pyrex wafer.
Problem 4.13 (4 pts): Crayon engineering: Create process and mask set for a DEP trap
Figure 2 shows an electrical trap that uses dielectrophoresis (DEP) to trap cells. Using the principles of
crayon engineering, design a process and mask set that will produce this structure (not to scale). Both
metal layers must be 0.5-m-thick gold (though other metals are acceptable beneath the gold). The
substrate B and layer A both must be electrical insulators. The gold linewidth is 10 m, and all other
critical parameters are specified in the figure.
You are asked to create a table of process steps as shown in Table 4.1 in the book, along with process
flow cross-sectional diagrams and masks as shown in Figure 4.1. Specify materials and the proposed
etch methods, and be sure to include as steps in your process the required wafer cleans, application of
photoresist, and stripping of photoresist. You do not need to include dimensions in your mask set in this
problem (but do draw the geometries correspondingly).
10 m
20 m
10 m
A
B
Figure 2. Schematic of a DEP trap (not to scale)

Spring Term 2007
Solution to Problem Set 1 (12 pts)
By Feras Eid (Solutions adapted from Xueen Yang )
Problem 2.6 (4pts): Dynamics with Matlab and Simulink

(a) From the transfer function,
H (s) =
we have,
Y1 ( s)
1
= 2
X ( s) s + 2s + 4
s 2 Y1 ( s ) + 2sY1 ( s ) + 4Y1 (s ) = X ( s )
Taking the inverse Laplace transform, the time domain expression becomes:
y1 (t ) + 2 y1 (t ) + 4 y1 (t ) = x(t ) ,

with initial values y1 (0) = y1 (0) = 0 .
Defining the two state variables to be y1(t) and y2 (t ) = y1 (t ) , we can write
y1 (t ) = y2
y 2 (t ) = 4 y1 (t ) 2 y2 (t ) + x(t )
The system is subjected to a step input x(t) = u(t), which can be expressed as:
x0
0
u (t ) =
x>0
1
We can create an ode function called fnc as follows:
function dy = fnc(t,y)
dy = zeros (2,1)
% create a column vector of all zeros

% to define the [y1; y2] array
% then define the state derivatives
dy(1) = y(2) ;
dy(2) = -4*y(1) - 2*y(2) + 1;
Matlab command ode45 can be used to integrate these equations for a time interval of 10 seconds:
[T,Y] = ode45(@fnc,[0 10],[0 0]);
Figure 1 shows the system response of y1 (t), which corresponds to an underdamped 2nd-order system (overshoot,
oscillation, etc..). Alternatively, we can integrate using ode23. The difference between ode23 and ode45 is the
algorithm: ode23 uses a lower order algorithm and is more efficient for a coarse estimate. In this simple case, there
is not much difference using either algorithm. But whenever you use numerical methods, it is wise to choose the
integration parameters (tolerance, for example) to ensure an accurate representation of the system and a practically
acceptable computation time.
The full MATLAB script for the problem is posted at the end of the solution. Note that the program fnc.m (which
defines the function fnc) and the program p2_6a.m must be in the same directory when p2_6a.m is called.
(b) Using tf object

H=tf(1,[1 2 4])
1
, by including the coefficients of the powers of s, as
s 2 + 2s + 4
arrays in descending order (of the powers), for the numerator and denominator respectively.
we create the transfer function corresponding to
Then we use the MATLAB commands

step(H)
bode(H)
to plot the step response of the above transfer function and its Bode plot. The step response is the same as Figure 1;
the bode plot is shown in Figure 2.
(c) Using SIMULINK
The system can also be represented using a Simulink block diagram as shown in Figure 3 (here using the transfer
function formulation). The resulting plot is again similar to that in Figure 1.
Problem 3.8 (2 pts): KOH etched diaphragm

Figure 4 shows the desired diaphragm features on a (100) silicon wafer. The mask feature size a can be expressed
as a function of wafer thickness tw, diaphragm thickness td, diaphragm dimension d and the intersection angle
between {111} and {100} planes:
a = d + 2(t w t d ) cot
(1)
Substituting the numbers, we get a = 1079 .7 m .

If the pattern is misaligned by = 1 , the actual size of the KOH pit will be a (cos + sin ) and hence the edge
length variation will be
a = a (cos + sin 1) = 18.7 m
(2)
This in turn translates to the edge-length variation for the diaphragm
d .
20 m
d = 400 m
500 m
54.7
Figure 4. Cross section of a (100) silicon wafer placed in an anisotropic etchant.
If the sensitivity S of a pressure sensor varies as the inverse fourth power of the diaphragm edge length d, i.e.
S d 4
then the percentage variation attributed to variations in wafer thickness is:
S
%
4d 1 d 100%
S
= 18.7%
(3)
pressure-sensing silicon diaphragm
Proposed process step

1. Start with a double-side-polished n-type silicon
wafer.
2. Perform photolithography using 1-m-thick
positive photoresist to define the diaphragm area
3. Deep-reactive-ion etch the silicon to form the
diaphragm; ash resist.
4. Anodically bond the silicon wafer to a pyrex wafer
Error
Must precede with a clean (RCA or piranha)

1 m of PR is very thin when used as a mask in DRIE. In
other words, since the selectivity of DRIE to silicon over
PR is ~ 50:1, etching though ~ 500m of Si would
require more than 10m of PR. Also, DRIE would lead
to a non-uniform membrane thickness with variations on
the order of the required thickness (15 m). This will
make the device function improperly if fabricated at all.
Must precede with wafer cleaning
Corrected process:
1. Start with double-side polished SOI wafer, device layer 15 m thick, oxide layer 1m thick,
substrate 500 m thick. RCA clean.
2. Deposit LPCVD nitride, 0.5 m thick (will coat both sides).
3. Spin 1-m-thick positive photoresist on bottom side and perform photolithography using Mask 1
to the bottom side.
4. Dry etch the nitride on the bottom side using CF4 / H2 plasma for example. Ash resist .
5. KOH etch the silicon from the bottom side using the embedded oxide layer as an etch stop. If the
dimensions of Mask 1 were calculated correctly, the resultant profile on the top side must be 1
mm across.
6. Piranha clean to remove all resist residue.
7. Etch the remaining nitride in 85% phosphoric acid.
8. Etch the exposed oxide using BOE for ~10 minutes. RCA clean.
9. Anodically bond the patterned SOI wafer to a Pyrex wafer.
The device cross sections at different steps and the mask needed are shown in Figure 5 below.
The side of the square on the mask is calculated from equation 1 in Problem 3.8 above.
Figure 5: Corrected process flow and mask for a pressure-sensing silicon diaphragm
Cross sections
Mask 1
1.9 mm
2
3
Materials
Silicon substrate
Embedded oxide layer
Si device layer
Nitride
Photoresist
Pyrex wafer
Problem 4.13 (4 pts): Crayon engineering: Create process and mask set for a DEP trap
Step
Description
Starting Material: Pyrex wafer
4 or 6; will be used as insulating layer B
1
2
Clean
Photolithography
Piranha
Using image reversal resist and Mask 1. Thickness of resist at least 1.5 m
(3 times that of the layer to be lifted off). Image reversal (negative) resist
necessary for lift off process later on
Deposit Au-Ti bilayer
E-beam evaporation (good for lift-off to be performed in next step).

thickness of gold ~ 0.5 m, thickness of titanium ~ 100 . Ti used as
adhesion layer.
Lift-off Au-Ti bilayer
Acetone. Follow by water rinse.
Clean
Using Nanostrip. It might be possible to use Piranha however it will eat up

some of the Ti layer and might lead to delamination of gold layer.
Deposit silicon oxide
PECVD, about 10 m thickness. Will be used as insulating layer A.
Photolithography
Using image reversal resist and Mask 2. Thickness of resist at least 1.5 m.
Deposit Au-Ti bilayer
E-beam evaporation. Thickness of gold ~ 0.5 m, thickness of titanium ~

100 . Ti used as adhesion layer.
Lift-off Au-Ti bilayer
Acetone. Follow by water rinse.
10
Photolithography
Spin cast positive thick photoresist, prebake; expose MASK 3, develop,

postbake
11
Etch oxide
Dry etch using CF4/H2 plasma. Anisotropic and selective over Si.
12
13
Strip resist
Clean
By ashing for example.

Using Nanostrip.
Process Steps
Masks
3
Mask 1
4
Mask 2
7
Mask 3
9
Materials
Pyrex wafer
10
Image reversal photoresist

Gold-titanium bilayer
13
Silicon oxide
Positive photoresist
Figure 6: Process flow and mask set for a DEP trap

MATLAB CODE FOR PROBLEM 2.6

%-----------------------------fnc.m----------------------------------------%Create a function fnc to define the relations between the state
%derivatives and the state variables. fnc is then called using ode45.
function dy = fnc(t,y)
dy = zeros (2,1)
% create a column vector of all zeros

% to define the [y1;y2] array
dy(1) = y(2) ;
% then define the state derivatives
dy(2) = -4*y(1) - 2*y(2) + 1;
%----------------------------END OF PROGRAM---------------------------------%-----------------------------p2_6a.m---------------------------------------%MATLAB code for Problem 2.6a
%Use the code below to call the defined function fnc and plot the response
%over 10 sec
[T,Y] = ode45(@fnc,[0 10],[0 0]);
plot(T,Y(:,1))
title('Figure 1. Problem 2.6 Step Response Plot');
xlabel('Time t');
ylabel('Response y_1(t)');
%----------------------------END OF PROGRAM---------------------------------%-----------------------------p2_6b.m---------------------------------------%MATLAB code for Problem 2.6b
H=tf(1,[1 2 4]);
% Use tf to create the
step(H)
% Use step to plot the
title('Figure 1. Problem 2.6 Step Response
figure
% Open a new figure
bode(H)
% Use bode to generate
title('Figure 2. Problem 2.6b Bode Plot');
transfer function
step response
Plot');
the bode plot
%----------------------------END OF PROGRAM----------------------------------

Spring Term 2007
Issued: Lecture 4
Due: Lecture 6
polysilicon surface micromachined cantilever
You are a young junior faculty member who has just hired your first graduate student, Terry Ibelfabber.
You have developed an idea for using a polySi surface-micromachined cantilever that youre sure will
make you famous and assure your tenure. You ask Terry to design a process flow for creating this simple
structure, and Terry returns with the process flow detailed in Figure 1.
Being a seasoned MEMS designer, you immediately notice several critical errors with Terrys process
(things that wont work or wont produce the result that Terry shows in his cross sections). Please find
the critical errors in this process flow and, where possible, suggest alternate approaches. Do not worry
about the accumulation of errors, but rather treat each step assuming that the structure up to that step
could be created.
This structure is actually quite simple to make. Develop a simpler process flow and associated masks to
create the final structure. Be sure to show cross-sectional and planar views of all key steps in the process.
Process steps:
1. Start with a silicon wafer.
2. Deposit 1 m of polysilicon.
3. Perform photolithography using positive photoresist (not shown) and wet-etch the
polySi using KOH.
4. Thermally grow 1 m of thermal oxide.
5. Perform photolithography using positive photoresist (not shown) and wet etch the oxide
in 49% HF.
7. Perform photolithography using positive photoresist (not shown) amd dry etch the
polysilicon using SF6 plasma.
8. Release the cantilever by etching the oxide with 49% HF.
1/5
Figure 1. Process flow for a surface-micromachined cantilever.

Problem 4.13 (7 pts): Crayon engineering: Create process and mask set for a thermal bimorph
cantilever
A thermal bimorph can be used as an actuator. In this problem, you will use the principles of crayon
engineering to design a process and mask set that will produce a silicon-based cantilever thermal
bimorph with an integrated heater and an underlying hole structure as shown below. (Silicon-based
means that the final structure is made of silicon, plus oxide, nitride, and metal as needed. You dont have
to use a plain silicon wafer, but you cant make the whole thing out of a completely different material like
metal or SU8.) A description of the structure follows; a top view is shown in Figure 2. Where a
dimension is not specified (like the lateral extent of the hole), you are free to choose a process that you
think makes sense. This may turn out to be an economic trade-off (for example, cost of processes vs.
wasted space on the wafer).
Cantilever composition: The cantilever includes a silicon structure, a metal layer on top of that (you can
choose either Al or Au), an integrated heater to actuate the bimorph, and either nitride or oxide layer(s) to
keep the heater from shorting out to the bimorph. The order of the layers is not specified up front; you
can pick any order that is buildable.
Silicon cantilever: 100 microns long, 20 microns wide, and 1.5 micron thick
2/5
Metal layer: 0.5 microns thick and covers the whole top surface of the cantilever, to within process
biases.
Underlying hole: at least 10 microns deep.
Integrated heater: 0.5 microns thick, made of doped polysilicon. Make sure that the heater has an
accessible contact pad.
Si cantilever
metal layer
polySi heater
Figure 2: Top view of the thermal bimorph cantilever structure showing

the cantilever, metal layer, and polySi heater layers in no particular order.
(a)
(b)
(c)
(1 pt) When we do crayon engineering, it is useful to identify the challenges of the

process flow (those points where we must be particularly careful to obey the laws of
physics) early on.
Examples could include thermal compatibility, chemical
compatibility, and the ability to pattern the device geometry. Identify what you see as the
major challenges for this process (a few words each). Pick three, and explain why they
are an issue.
(2 pts) Brainstorm three different ways of approaching the process, and explain them
briefly. You dont have to have all of the details ironed out on these approaches.
(4 pts) Choose one approach and flesh it out. You need to sketch the mask set with key
dimensional relations and write out the steps of the process flow. Specify materials and
the proposed deposition and etch methods, and be sure to include as steps in your process
the required wafer cleans, application of photoresist, and stripping of photoresist. If a
dimension on the mask affects the success of the process, make sure you specify it. Be
sure to show cross-sectional and planar views of all key steps in the process.
3/5
Problem 1.3 (5pts): Literature Treasure Hunt

This goal of this problem is to encourage you to learn to use the library to find real information. You will
be given a MEMS device with certain specifications. Your job is to go to the literature and find three
distinct examples of that device, including the one that you believe has the best value for the desired
specification.
You will evaluate the devices along the following metrics:
Reference(s): Give at least one complete citation to the work that you are citing, including authors,
title, source, year, etc.
Unpublished material found on the web is not acceptable.
A combination of patent literature and product specifications is acceptable for

commercial devices.
Affiliation(s): List the affiliation(s) of the authors.
Device description: Describe the design and function of the device. Use schematics when possible.
Fabrication techniques: Give a brief description of the overall fabrication approach, use diagrams
when possible. Highlight the major fabrication techniques.
Pros of approach: When compared to the other two devices, what advantages does this device have
(e.g., simpler fab, lower noise, greater reliability, etc.).
Cons of approach: When compared to the other two devices, what disadvantages does this device
have (e.g., harder fab, higher noise, worse reliability, etc.).
Value of reported metric: For the specification that you need to find, what is the value reported in
the citation, including measurement errors or any caveats. You can also report other performance
factors that you find important.
Measurement methods: How did they perform their measurement?
Confidence in results (1-5), 5 is best, 1 is dont believe and Why?: Critically evaluate the results.
Do you believe them? Were they ever repeated? Did they do it right?
Comments: Comment on the device on the overall performance by considering other performance
factors as well. Which one would you choose to buy among the three?
Three distinct examples typically means devices by different research groups. Three devices by the same
group may be acceptable if they are significantly different. Use your judgment.
You can download a word document worksheet from the assignments section named treasurehunt.pdf.
Fill in the formatted table from the worksheet and submit it on paper with your homework. Your
assigned device and specification is attached in the device treasure hunt assignments sheet. An example
of the device hunt with two devices is also attached (you do need to fill all three devices out). This
merely serves as an example; the details you put in the table depend on your knowledge on the subject as
well as technical reading skills.
4/5
It is possible to spend a lot of time on this, and still not find the best. Shrewd searching, using forward
searches or review articles, can save lots of time. However, please use judgment in allocating a reasonable
amount of effort for this problem.
5/5

Spring Term 2007
Solution to Problem Set 2 (16 pts)
By Brian Taff (Adapted from work by Feras Eid)
Issued: Lecture 4
Due: Lecture 6
polysilicon surface micromachined cantilever
Proposed process step
1. Start with a silicon wafer
Error
2. Deposit 1 m of polysilicon
Must precede with RCA clean. Also, must have some

layer (such as oxide or nitride) separating poly from
substrate to serve as a stop layer during etching.
3. Perform photolithography using positive

photoresist (not shown) and wet-etch the
polySi using KOH.
Can't perform KOH etch when PR mask is used; PR will

lift off immediately. Also, this step must be followed by
PR removal .
4. Thermally grow 1 m of thermal oxide
Can't grow thermal oxide on polysilion; the oxide will

eat much of the polysilicon and the surface will not be
planar since the oxide will grow much slower in-between
the poly plates. Deposit oxide instead. Also, must
precede with RCA clean of wafer.

photoresist (not shown) and wet etch the oxide
in 49% HF.
Can't use 49% HF with PR as a mask; the PR will lift off.

Use BOE instead. Also, this step must be followed by
PR removal. Also, we wont get perfect alignment
between the oxide opening and right edge of poly 1.
A wafer clean such as Piranha should precede this

deposition to ensure a quality deposition.

photoresist (not shown) and dry etch the polysilicon
using SF6 plasma.
SF6 leads to isotropic etching and undercutting, which

might or might not be an issue depending on the
dimensions of the cantilever. Also, we wont get perfect
alignment between the left edge of poly level 2 and the
left edge of oxide cutout. Also, this step must be
followed by PR removal.
8. Release the cantilever by etching the oxide with

49% HF
Must be followed by some drying process, as wet etching

alone will lead to stiction problems.
Simpler Process:
1.
2.
3.
4.
5.
Start with a silicon wafer, perform RCA clean with HF dip.

PECVD 1 m of sacrificial oxide.
Perform photolithography using positive photoresist (not shown) and Mask 1; then wet-etch the
oxide using BOE. Then ash resist and perform RCA clean (without HF dip).
LPCVD 1 m of polysilicon.
Perform photolithography using positive photoresist (not shown) and Mask 2; then dry-etch the
polysilicon using reactive-ion etching. Then ash resist and perform RCA clean (without HF dip).
1/7
6.
Release the cantilever by etching the oxide with BOE followed by super-critical freeze drying.
Figure 1: Simplified process flow & mask set for a surface micromachined cantilever.
Note: Darker shades of the layer colors shown in the planar views indicate geometric curvature.
2/7
Problem 4.13 (7 pts): Crayon engineering: Create a process and mask set for a thermal bimorph
cantilever
a)
Some challenges:
1. Order of depositions: the polysilicon must go down before the metal because of thermal compatibility (poly
deposition is carried at a high temperature)
2. Because of topography concerns, you have to pattern the films on the cantilever before you etch the hole
beneath it. This requirement makes the organization of the process flow a bit more challenging and involved.
Additionally, the cantilever, once released, won't be robust enough to tolerate lithography.
4. A wet etch like KOH would undercut the cantilever effectively, but you cant mask it with resist. You must
know what materials will effectively mask the wet etch. A lack of consideration here can cause inadvertent
damage to structural/functional parts of the device.
5. How do you define a cantilever that is precisely 1.5 microns thick? You will need to incorporate a
thickness-determining layer into the structure of the wafer.
b) Three approaches:
1. Use an SOI wafer to set the thickness of the cantilever. Deposit LPCVD nitride to provide electrical isolation
between the cantilever and the polySi heater. Deposit polySi for the heater, dope it, then perform
photolithography and dry etch the poly, stopping on nitride, defining heater pattern. Strip the resist, clean the
wafer, and deposit LPCVD nitride everywhere to isolate the heater from metal layer. Deposit aluminum,
perform photolithography, and etch it into the correct shape. Use thin resist photolithography to mask around
the etch hole. Dry etch the nitride to expose the silicon and open contact pads, then define the contour of the
cantilever by dry etching to the buried oxide. (*) Use BOE to remove the exposed oxide. Flip the wafer over,
and dry etch several times to strip blanket films from the back (nitride, then poly, then nitride). Pattern the back
with thick photoresist, then DRIE through to define cantilever.
2. This option is very similar to the preceding sequence, but it omits the final DRIE step. Follow the process to
point (*). Repattern so that the resist sticks out a little beyond the cantilever edges, and dry etch the exposed
buried oxide. Strip resist and recoat, patterning so that the resist extends beyond the cantilevers Si, covering it
entirely (very important.) Use a XeF2 isotropic dry etch to undercut the cantilever; the cantilever itself is
protected by resist and oxide, while the Si underneath gets removed.
3. This listing is the approach we will use in part c. Briefly: Grow doped, KOH resistant Si by epitaxy, deposit
nitride, followed by polysilicon. Dope the poly and then pattern it. Deposit another layer of nitride and then
pattern it to open contact pads. Deposit a Ti-Au bilayer and use liftoff to define the metal layer. Dry etch the
nitride and the epitaxial silicon to define the cantilever. KOH etch the substrate to release it. See part c.
c) Using the third approach from above 1 , we detail the process flow here. Key device cross sections, planar views,
and the mask set are shown in Figures 2-4. Note that the cross section taken (A-A) is not straight but rather bent
around a corner to show the different critical parts of the device during the fabrication process.
adapted from: Riethmuller and Benecke, "Thermally Excited Silicon Microactuators," IEEE Transactions on
Electron Devices, Vol 35, No. 6. June 1988.
3/7
Step
Starting Material: Silicon wafer
1
Clean
2
Deposit p+ Si
Description
4 or 6; (100) orientation; 500 m or 650 m thick
Standard RCA.
Using epitaxy, thickness = 1.5 m. Heavily doped Si (example: Boron
concentration ~ 1.3 x 1020 cm-3 ) is KOH resistant.
Deposit nitride
LPCVD, thickness = 0.1m. Used to electrically insulate the heater from the
p+ Si. Will cover both sides of wafer.
Deposit polysilicon
LPCVD, thickness = 0.5 m. Will act as the heater. Also covers both sides.
Deposit oxide
LPCVD, thickness = 0.2 m. Will act as a protective shield to poly during

the ion implantation step. Also covers both sides
Dope polysilicon and anneal
Using ion implantation of Boron. Follow with annealing (at ~1000oC) to

activate the implant.
7
8
Etch oxide
Photolithography
BOE for ~ 2 minutes.

Spin cast 1 m thick positive photoresist, prebake; expose MASK 1,
develop, postbake. Mask 1 defines the shape of the polysilicon heater.
Dry etch polysilicon
10
Strip PR
Using HBr for example (used by AME5000 in MTL-ICL). HBr etches poly
anisotropically. The heater is now patterned.
By ashing for example.
11
12
Clean
Deposit nitride
13
Photolithography
14
Dry etch nitride
15
Strip resist
16
Photolithography
17
Deposit Ti-Au bilayer
18
Lift-off Ti-Au bilayer
Acetone. Follow by water rinse. Now the metal layer is patterned.
19
Clean
Using Nanostrip
20
Photolithography

develop, postbake. Mask 4 defines the shape of the cantilever.
21
Dry etch nitride
Using CHF3/O2 plasma for example.
22
Dry etch epitaxial Si
Using HBr, for example.
23
24
Strip resist
Etch silicon (from top)
By ashing, for example.

Using KOH and the top (patterned) and bottom nitride layers and metal layer
above the contact pads as masks. Etch 30 m deep (etch rate and hence
timing for this step depend on the KOH temperature and concentration).
Since convex corners become undercut, the cantilever will be released. We
are making the hole deeper then required to make sure that the cantilever
doesnt stick to its bottom because of the wet etch.
Standard RCA.
LPCVD, thickness = 0.1 m. Used to electrically insulate the heater from the
metal layer to be deposited next. Will cover both sides of wafer.
develop, postbake. Mask 2 is used to open contact pads to the poly heater.
Using CHF3/O2 plasma for example . This patterns the contact pads into the
heater.
By ashing, for example
Using image reversal resist and MASK 3. Thickness of resist about 2 m (3
times that of the layer to be lifted off, plus an extra 0.5 m to counter the
step coverage problem due to the presence of the 0.6 m poly-nitride step ).
Image reversal (negative) resist necessary for lift off process later on.
E-beam evaporation (good for lift-off). Thickness of titanium ~ 100 ,
thickness of gold ~ 0.5 m. Ti used as adhesion layer.
4/7
Figure 2: Important cross-sectional views

Note: Some films are deposited on the back side of the wafer. Those films are not explicitly shown in these
cross sections simply because the process flow does not mandate any backside processing. Not including
such films aims to minimize the clutter in this figure.
5/7
Figure 3: Important planar views

Note: We highlight the two nitride layers using different shading in this planar view. (They are simply
shown in black in the cross-sectional views.) Additionally, the Ti-Au bilayer is drawn with some degree
of transparency to avoid obscuring underlying features. In reality the Ti-Au bilayer is not transparent.
6/7
Figure 4: The mask set used for this process along with some key dimensional constraints
Note: Generally, bond pads are at least 100m 100m in size. Anything smaller is difficult to bond
using a wirebonder. There also has to be adequate spacing between the bondpads. Generally, anything
less than 100 m is considered risky. The bondpads are not drawn to scale in this process flow. If
they were drawn accurately, it would be difficult to see the features on the cantilever.
7/7

Spring Term 2007
Issued: Lecture 6
Due: Lecture 9
Problem 9.14 (2 pts): Bending of an AFM Cantilever

A silicon nitride (LPCVD, stoichiometric) AFM cantilever has a thickness of 0.5 m and the triangular
shape (top view) shown below. It is subjected to a point load F at the tip, pointing into the paper.
Assuming small deformations, find the tip deflection in terms of F, and calculate an effective spring
constant for this beam. For a load which produces a tip deflection of 2 m, calculate the maximum stress
at the support.
a = 50 m
L = 250 m
Problem 9.10 (4pts): Thermal Stress Induced Wafer Curvature

Wafer-scale fusion bonding is commonly used to create complex silicon structures. However, two wafers
can be bonded together only if their bonding surfaces are about the same shape. If the shapes are too
different, the energy required to deform the wafers exceeds the energy gained by bringing them into
contact, and the wafers will not bond. Because stressed films deform substrates, film thicknesses are
limited on wafers that will be wafer-bonded.
Consider a PECVD silicon dioxide film that is deposited on a 650 m thick silicon wafer. The wafer is
then annealed at 1100 C and cooled to room temperature. Assume that the oxide is stress-free when it is
above the oxide softening temperature, which we will take to be 950 C. Also assume biaxial plane stress
in the film, and approximate the substrate as isotropic.
(a) Using constitutive relationships, calculate the thermal stress r in the film for the case where the
film thickness is much less than the substrate thickness.
Experience shows that the wafers radius of curvature must be greater than about 100 m to ensure that the
wafer can be bonded to a second, flat wafer. In the next two parts, we will determine the maximum
thickness of the oxide film to ensure bondability.
(b) The wafer curvature is determined by moment balance over the thickness of the wafer. Find an
expression for the net moment per unit width as a function of the geometry, materials properties,
and thermal stress in the film.
(c) Write an expression relating the film thickness to the radius of curvature. What is the maximum
film thickness to ensure bondability for the case described above?
Problem 14.12 (2 pts): Circuit loading

In this problem we will explore the differences among different op-amp topologies. Assume we are
trying to measure the signal Vs from a linear sensor and would like to amplify that before we measure it
with a computer-based data acquisition card. Thevenins theorem states that any linear electrical sensor
can be represented by a voltage source and a series impedance (in this case, a resistance RS). Your data
acquisition board, like all circuits, has some input resistance denoted by RL.
Sensor
Computer
Rs
+
R1
Sensor
R2
Vs
+
-
(a)
(b)
(c)
R2
VL
R1
Rs
Computer
+
RL
Vs
+
-
VL
RL
One way to amplify the signal VS is to use a non-inverting amplifier (Figure, left). If we
use this circuit, what is VL as a function of Vs and the four resistors, assuming that the opamp is ideal?
A second way to amplify the signal is to use the inverting amplifier (Figure, right).
Repeat part (a) for this circuit.
As RS changes, the gain of each circuit may change, leading to unpredictable behavior,
which wed like to avoid. What is the gain of each circuit when RS = 0? This is the ideal
gain. Determine analytically the (actual gain)/(ideal gain) for each circuit. Give the
constraints on each circuit such that the actual gain approaches the ideal gain. Which
circuit is less sensitive to changes in RS?
Problem 9.13 (3pts): Leveraged Bending with Electrostatic Actuation

As we will learn in lecture 9, pull-in limits the useful analog travel of a voltage-controlled moveable
capacitor plate to one third of the initial gap between the plates. One trick that MEMS designers
sometimes use to get greater controllable travel at the center of the beam is leveraged bending (illustrated
in the left hand figure below). In this scheme, the bottom capacitor plate is limited to the ends of the
beam instead of extending to the center. Because no voltage difference is applied between the center of
the beam and the substrate, the center can deflect more than one third of the initial gap before pull-in
occurs.
b
L
underlying electrodes
Consider a fixed-fixed silicon beam of width w, length L, and height h that is actuated by leveraged
bending. If the electrodes are narrow enough, we may approximate the load as two point forces applied at
distances a and b from the ends of the beam, as shown in the right hand figure above. (We would
typically design the system so that a = b, but fabrication nonuniformities may lead to some asymmetry.)
Assuming small deformations, calculate the deflection of the beam at the center. You are welcome to use
any analytic method you like (explicit integration of the beam equation, use of tabulated beam bending
solutions, etc.)
Problem 5.9 (2 pts): Circuit representation of a lumped mechanical system
A mechanically coupled micromechanical resonator can be lumped modeled as shown in the figure.
Derive an equivalent circuit in the eV convention and find the transfer function x1(s)/F(s).
k2
k1
k3
m1
m2
b1
x2
x1

Spring Term 2007
By Brian Taff (with acknowledgement to Feras Eid and Xue'en Yang)
Due: Lecture 9
Problem 9.14 (2 pts): Bending of an AFM Cantilever

For LPCVD stoichiometric silicon nitride, E = 270 GPa.
From the lecture on structures, for a cantilever with the following configuration and loading
F
d w
the differential equation of beam bending is:
dx
EI
z
where M is the internal moment at distance x:
M = -F (L-x)
and I is the moment of inertia of the cross section at the distance x. For a rectangular cross section:
I = bh3/12
For our case, h = 0.5 m and b is a function of x due to the triangular shape of the cantilever.
For our geometry:
Lx
a
L
a = 50m
a ( L x ) h3
I=
12 L
Therefore:
and
d w
dx
Integrating twice:
12 F ( L x ) L
Ea ( L x ) h
w=
L= 250m
12 FL
Eah3
6FL 2
x + Ax + B
Eah3
Applying the boundary conditions: w(x=0) = dw/dx (x=0) =0, we get: A = B = 0.
Therefore:
w( x ) =
6 FL 2
x
Eah3
6 FL3
= 55.55 F [w in m, F in N ]
Eah3
F
Eah3
=
=
= 0.018 N / m
6 L3
w( L )
w( L ) =
At the tip (x = L), the deflection is:

And the equivalent spring constant is:
keq
For a tip deflection of 2 m:
F = keq w( L ) = 0.018 2 x10 = 3.6x10 N

Mh F ( L x )h12 L 6 FL
max =
=
= 2 = 4.32 MPa
2I
2ah3 ( L x )
ah
And the maximum stress at the support is:
-8
Problem 14.12 (2 pts): Circuit loading

Rs
a) Assuming we have an ideal op-amp
i+ = 0 V+ = Vs
i- = 0 V- = V1, i1 = i2
R2
and V = V Vs = V1
+
i2
Vs
+
-
V2
Also V1 =i1 R1 , VL = i1 ( R1 + R2 )
i1
R1
VL
RL
V1
Therefore VL / Vs = 1+ R2 / R1
b) Assuming again we have an ideal op-amp

i- = 0
R1
is = iL
is
and V+ = V- = 0
Rs
R2
iL
Also V- - VL = iL R1 , Vs - V-= is ( Rs + R2 )
Vs
Therefore VL / Vs = - R1 / (Rs + R2 )
+
-
VL
-
The ratio is negative, hence the name "inverting".
c) The gain G is defined as the ratio of the output voltage to the input voltage, or VL / Vs.
For the non-inverting amplifier: Gideal = 1+ R2 / R1 = Gactual for all values of Rs
Gactual / Gideal = 1
For the inverting amplifier:
Gideal = - R1 / R2
Gactual / Gideal = R2 / (Rs + R2 )
Thus for the non-inverting amplifier, the ideal and actual gains are always equal, while for the inverting
amplifier, the actual gain approaches the ideal gain only when the sensor impedance approaches zero
(assuming both amplifiers are ideal). Clearly, the non-inverting amplifier is less sensitive to changes in Rs than
the inverting one.
RL
Problem 9.13 (3pts): Leveraged Bending with Electrostatic Actuation
F
a) The governing equation for the deflection of a doubly clamped beam
under a point load as shown in the figure is,
d 4w
EI 4 = F < x a > 1 ,
a
dx
1
where < x a > denotes a concentrated load at a. Integrating, we
have,
d 3w
EI 3 = F < x a > 0 +C1
dx
d 2w
EI 2 = F < x a >1 +C1 x + C2
dx
dw
F
1
EI
= < x a > 2 + C1 x 2 + C2 x + C3
dx
2
2
F
1
1
EIw = < x a > 3 + C1 x 3 + C2 x 2 + C3 x + C4
6
6
2
Apply the boundary conditions
w(0) = w( L ) = w '(0) = w '( L ) = 0 ,
we get,
F ( L a ) 2 ( L + 2a )
C1 =
L3
F ( L a )2 a
C2 =
L2
C3 = C4 = 0
Hence, the center deflection is,
2
L Fa (4a 3L)
w =
48EI
2
If we next examine the effect of the point load acting at the right hand side of the beam (acting alone as if
the load at a had been removed), we get a similar setup.
d 4w
= F < x ( L b ) > 1
dx 4
d 3w
EI 3 = F < x ( L b ) > 0 +C1
dx
d 2w
EI 2 = F < x ( L b ) >1 +C1 x + C2
dx
dw
F
1
EI
= < x ( L b ) > 2 + C1 x 2 + C2 x + C3
dx
2
2
1
1
F
EIw = < x ( L b ) >3 + C1 x3 + C2 x 2 + C3 x + C4
6
6
2
EI
Applying the boundary conditions
w(0) = w( L ) = w '(0) = w '( L ) = 0 ,
we get,
C1 =
Fb 2 ( 3L 2b )
L3
Fb 2 ( b L )
C2 =
L2
C3 = C4 = 0
In this case, the center deflection is:

2
L Fb (4b 3L)
w =
48EI
2
Applying superposition to the beam, we can obtain the center deflection due to both loads to be,
F a 2 (4a 3L) + b2 (4b 3L)

48EI
Problem 5.9 (2 pts): Circuit representation of a lumped mechanical system
k2
k1
k3
m1
m2
b1
x2
x1
In the e V convention, the flows are the time derivatives of the displacements xi of the different bodies, and
effort is the external force. Since k1 and b1 share the same displacement, and hence, the same flow, they are in
series. The same holds for k2 and m2 and also for k3 and m1. By observation, the net flow through b1 is ( x 2 x1 ) .
Hence, the equivalent circuit can be represented as shown in Figure.
m2
1/k2
m1
1/k1
x 2
m2s
1/k3
x1
k3/s
k1/s
x2s
s domain
b1
m1s
k2/s
x1s
b1
Apply KVL for the left loop, we have

2
m 2 s x 2 + k 2 x 2 + ( k1 / s + b1 )( x 2 s x1 s ) = F
(b1 s + k1 ) x1 + ( m 2 s 2 + b1 s + k1 + k 2 ) x 2 = F
(1)
Apply KVL for the right loop, we have

2
m1s x1 + k 3 x1 + ( k1 / s + b1 )( x1s x2 s ) = 0
( m1s 2 + b1s + k1 + k 3 ) x1 (b1s + k1 ) x2 = 0

x2 =
m1s 2 + b1s + k1 + k 3
x1
b1s + k1
Substitute into (1), we have,

x1 ( s )
F (s)
b1s + k1
(b1s + k1 ) + ( m1s + b1s + k1 + k 3 )( m2 s 2 + b1s + k1 + k 2 )
2

Spring Term 2007
Issued: Lecture 9
Due: Lecture 12
Problem 6.5 (5 pts): The in-plane interdigitated electrostatic (or comb drive) transducer
Many MEMS components use interdigitated electrodes as transducers. In this and subsequent problems
we will explore this structure. Below on the left is a moveable silicon mass with many interdigitated
(comb) fingers that is connected to the fixed substrate by a flexible spring. Using DRIE, it is possible to
make such a structure with very thick fingers (>50 m out of the plane of the paper). On the right is a
schematic of one finger of this transducer, with the rest of the system specified by lumped parameters b
and k.
i
fixed
+
v
-
spring
interdigitated electrodes
k
h
g
b
cross-section
a
mass
t
t
l
For our one-finger

model, assume that the
device is constrained to
only move in the z
direction.
The
thickness of the plate in
the plane of the paper
(t) is 200 m and its
width w is 5 m. The
spacing h is 5 m on
either side of the finger,
the initial gap g0=10
m, while the length
l=100 m. The spring
constant k=1 kN/m, the
damping
coefficient
b=0.05 mN-s/m, and the
device operates in air.
(a)
(b)
(c)
(d)
(e)
Assume a current drive. Determine analytically the force, the position variable g (the
gap), and the voltage on the plates as a function of the charge Q.
State any assumptions that you are making in calculating the force in (a).
Assume a voltage drive. Determine analytically the force, the gap, and the charge on the
plates as a function of the voltage V.
Show that the effective spring constant under voltage drive does not change with position
(i.e., Fnet g is constant with V and g). What does this imply about spring
softening/hardening and instabilities for this system?
Using a similar approach for current drive, show that spring softening occurs. Will this
transducer experience pull-in? Why?
Cite as: Carol Livermore and Joel Voldman, course 1materials for 6.777J Design and Fabrication of
Problem 6.9 (2 pts): Design the spring

The comb-drive transducer in Problem 6.5 is designed to have a spring constant of k=1 kN/m. Given the
plate thickness of 200 m and a spring width a of 10 m, how long must the spring be to produce this
spring constant? You may assume small-deflections. Describe in words and pictures how you might go
about fitting such a spring into a minimal wafer area.
Problem 6.10 (7 pts): Design a simple switch

Your task is to design an inexpensive surface-micromachined MEMS electrical switch. The general form
of the switch is shown in the figure below. It is composed of a polySi cantilever electrode above a polySi
actuating electrode, and is separated from the substrate by an insulator. The cantilever has length l,
thickness t, width w (not shown), and is separated from the actuating electrode by a gap g. You are given
the following set of constraints:
Voltage to close the device is Vc=10 V.
When closed, the cantilever is used to conduct current, and therefore must have no more than 10
of resistance along its length.
Max cantilever thickness t of 2 m.
o You can increase t, but the extra fabrication cost is $2/m for each micron over 2 m.
Minimize cost by minimizing cantilever beam area, which is priced at $200/mm2.
Cantilever length l must be at least 5 the width to ensure beam-like behavior (l5w).
Minimum initial gap g is 0.5 m.
The overlap length lo between the actuating electrode and the cantilever is 10 m.
The target cost of the part is $1.
The width w must be at least 10 the thickness t of the cantilever.
Assume E=150 GPa, mass density =2300 kg/m3, and electrical conductivity =105 S/m for
polysilicon.
You can ignore any deflection of the cantilever due to its own weight, and assume an ideal
cantilever support.
l
lo
t
(a)
(b)
Determine analytically the set of equations governing the design constraints. You will need
to make engineering approximations to solve this problem expeditiously. For instance, you
can ignore the effects of the non-ideal support and the extra stiffness induced by the
cantilever deforming at the contact. You can also model the electrical force as being applied
as a point to the tip of the cantilever (if lo<<l)..
Using hand calculations, Matlab, or any other suitable approach, determine l, t, w, and g that
will meet these specifications and result in the lowest cost. State your Vc, cost, and
resistance. If you believe that you cannot meet the cost objective, then give the dimensions
that result in the lowest cost.
(c)
Using the principles of crayon engineering, design a process flow and mask set to create your
device.
Problem 6.11 (7 pts): A simple MEMS resonator

One of the things that make MEMS resonators attractive to circuit designers is that one can recreate large
electrical inductors and capacitors using vibrating mechanical components, and further that one can tune
the value of these components by changing an applied DC bias. In this problem well examine some of
the design issues with a simple clamped-clamped MEMS resonator shown below, which has length l,
thickness t, width w, gap go, and electrode width le. We will first determine the equations describing its
static (i.e., fixed point) behavior, and then use a linearized model to examine the tenability of the
resonator.
t
g0
lelec
(a)
(b)
Determine analytically the set of equations governing the static behavior of the beam.
Specifically, provide an expression for the mechanical spring constant k and the lumped
effective mass, which we will approximate as meff = 0.4mactual. Also find the pull-in voltage
VPI for this structure. You may assume the lelec << l and thus that the electrical force acts a
point force in the mechanical domain.
Using the linearized model from the figure below, determine C0, k, and , in terms of each
other, physical dimensions, material constants, the bias voltage V0, and the static actuated gap
g 0 .
1/k
1:
+
V C0
(c)
Next, remove the transformer to obtain the following equivalent circuit, and determine
expressions for C1 and L.
C1
+
V C0
(d)
Determine the input impedance Zin of this circuit as a function of Co , C1, and L. This is a 3rd
order system that has one pole at =0. Find the frequencies () of the other poles and zeros.
(e)
The way that we tune this device is to set the fixed point of the nonlinear system by applying
a bias voltage V0 that will deflect the beam and set k (or C1), C0, and , and then superpose a
small sinusoidal voltage as our signal. As we actuate from V0=0 to V0=VPI we will tune the
location of the zero. Determine analytically 1, the zero of Zin when V0=0, and 2, the zero of
Zin when V0=VPI, in terms of physical dimensions, material constants, the bias voltage V0,
and the static actuated gap g 0 .
(f)
As approaches (but does not quite reach) 1, you can simplify 2 into a particularly simple
form. Find this form (in terms of 1 and ) and use it to estimate the maximum change in 2
as V0 is varied from zero to 95% of VPI. The actual change will be less. Why?
Now that we know how to pick where the zero occurs and how it tunes as we change V0, we
are ready to design. Pick l, w, g0, and le subject to the following constraints:
t = 1 m
VPI = 25 V
1 = 2106 rad/s
le 0.1l
g0 0.25 m
E= 150 GPa, =2300 kg/m3 for polySi, and the gap is vacuum
(g)
You may use hand calculations or Matlab to determine suitable parameters. You may want to
familiarize yourself with the Matlab function fzero to determine the gap. Plot the zero of Zin from
V0=0.05VPI to V0=0.95VPI. Also generate a Bode plot at those two voltages to show how the zero (and
pole) move as we change VPI. For the Bode plot you will need to use a very-fine frequency vector and
concentrate on the frequency range from 106 to 107 Hz to easily see the behavior.
4

Spring Term 2007
PROBLEM SET 4 SOLUTIONS (25 pts)
Issued 3/7/07
Due 3/16/07
Problem 6.5 (5 pts): The in-plane interdigitated electrostatic (or comb drive) transducer
a). Current drive case
The gap g between the finger tip and the electrode can be expressed in terms of the displacement z as follow,
g = g0 z
The effective capacitor length is hence, l g = l g 0 + z . The capacitance of the upper part of comb is (by
neglecting the electric fields around the finger tip),
Cupper =
0 (l g )t
h
Since the one-finger model is equivalent to two capacitors in parallel (upper and lower parts), the total capacitance is
multiplied by two,
C=
2 0 (l g )t
h
The energy stored in the system is

Q
W = vdQ =
0
Q2
C dQ = 2C
0
Substitute C into the equation above, we have

W=
hQ 2
4 0 (l g )t
The force becomes

W
Q2
h
F =
=
g
4
t
(
l
g) 2
Q
0
To find the gap, we have

F = kz = k(g g 0 )
g = g0
F
Q 2h
= g0
k
4 0tk (l g) 2
The voltage can be expressed as,

W
hQ
v=
=
Q g 2 0t (l g)
1
Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare

(http://ocw.mit.edu/),
b). Assumptions made in calculating the force in a).

a.
The upper and lower gaps of the comb fingers are exactly the same as fabricated
b. Fringing effects are neglected.
c.
The energy stored between the finger tip and the electrode (gap g) is neglected.
d. The electrical permittivity of air is assumed to be the one of free space.
e.
There is no space charge in the space between the plates.
f.
The actuation of the comb drive is quasi-static.
g. Charge distributed uniformly on the plates
c). Voltage drive case
We can use co-energy to find force and charge.
The co-energy in the capacitor is
V
W * ( v, g ) = Qdv = Cvdv =
0
Cv 2 0 (l g )tv 2
=
2
h
The force
W *
F =
g V
F=
0 tv 2
h
And the charge

W *
Q =
v g
Q=
2 0 (l g)tv
h
To find the gap, we again have

F = kz = k(g g 0 )
0tv 2
h
= k ( g g 0 )
g = g0
0tv 2
kh
d). The net force for the voltage drive case:

Fnet =
0tv 2
h
+ K (g 0 g)
Fnet
= k < 0
g
The effective spring constant is a negative constant, which means the increase in gap will cause the decrease in
force, therefore, the system is always stable and there is no spring softening/hardening.
e). The net force for the current drive case:
Fnet =
Q2
h
+ k(g 0 g)
4 0t(l g) 2
Fnet
Q2h
= keff =
k < 0
g
2 0t (l g)3
The effective spring constant is also negative, however, its not a constant. Since g always decreases from g0 to
0 when actuation starts, keff decreases as the comb drive is actuated, creating the spring softening effect. However,
since keff is always negative regardless of the value of g, the system is always stable, and hence, no pull-in will
occur. This conclusion is based on the assumption that the electric fields around the fingertips are negligible. Also,
it is assumed that the upper gap and lower gap of the comb fingers are the same, while in reality, they might vary
due to nonuniform etching or other fabrication effects. Pull-in, hence, can occur due to these secondary effects.
2
Problem 6.9 (2 pts): Design the spring

Since we have 2 springs in parallel (as shown in the figure), the effective spring
constant of each is keq = k parallel/2 = 0.5 kN/m
Each spring acts as a beam with a fixed support at one end (the anchor) and subjected at the other end to a zero slope
boundary condition (because of its connection to the mass), despite its ability to translate. This BC imposes a
moment reaction at the other end. Hence the resultant deflection is the superposition of 2 tabulated deflections: that
of a cantilever beam subjected to a point force at its end and that of a cantilever beam subjected to a moment at its
end. The moment is unknown yet, and will be found by substituting the zero-slope BC into the resultant deflection
profile.
2F
(x 3 + 3x 2 L)
3
Eta
6M 2
2 ( x) =
For a cantilever (length L, width a, thickness t) with a moment M at the end: w
x
Eta 3
2F
6M 2
( x ) = w 1 ( x ) + w 2 ( x ) =
The resulting deflection is thus: w
(x 3 + 3x 2 L ) +
x
3
Eta
Eta 3
1 ( x) =
For a cantilever (length L, width a, thickness t) with point force F at the end: w
Using the boundary condition,
dw ( x)
12M
2F
=
(3x 2 + 6xL) +
3
dx x = L Eta
Eta 3
x = 0
x= L
we can solve for M and find:
M =
FL
2
We then plug back into our deflection equation to find:
w ( x) =
2F
(x3 +1.5 x 2 L)
Eta 3
Springs
Evaluating at x=L provides:
Mass
FL3
w ( L) =
Eta 3
We can then solve for the equivalent spring constant.
keq =
F
Eta 3
= 3
w ( L)
L
Subsituting E 150 GPa, a = 10 m, and t = 200 m, we get: L = 310.7 m.

We can fit such a long spring into a minimal wafer area by folding it, as shown:
3
Massachusetts Institute of Technology. Downloaded
on [DD Month YYYY].
Problem 6.10 (7 pts): Design a simple switch

(a) U
sing t, l, w , and g in [m] to avoid confusion, and all other parameters in standard SI units:

V PI =
8kg 3
= 10V -------(1)
27Acap
(assuming electrical force applied as point force at tip)

Ewt 3
------------(2)
k=
4l
3
(assuming ideal cantilever support)
Acap = l o w ------------(3)
(assuming capacitor area does not vary much as switch is deflected)
1 l
R=
10 ----------(4)
wt
(assuming changes in length and width of cantilever between the closed and open switch modes are
negligible)
l 5w ------------(5)
w 10t -----------(6)
Cost constraint: lw 200 10 6 $ / m 2 + (t 2 10 6 ) u(t 2 10 6 m) 2 10 6 $ / m 1$ -------(7)
(u is the unit step function, since we start paying for the thickness once it exceeds 2 m, according to
the problem statement).
(b) Substituting E = 150 GPa, = 105 S/m, =2300 kg/m 3, l0 = 10 m, and = 0 = 8.854 x 10
assuming the gap is vacuum), we get:
gt
= 9.2712 10 9 m
l
l
10 6 m 1
wt
5
w
-12
F/m (i.e.
--------- (A) (from (1), (2), and (3))

----------(B)
----------(C)
Combining (B) and (C) yields the following relation:
l
t 106 m 1
w
5 t 106 m 1
5
5 10 6 m t
Therefore, from physical constraints alone tmin = 5 10 6 m = 5 m
Because we are forced to use a beam thickness greater than 2 m, in the cost function the term that involves the step
function is in the active state. We therefore find ourselves in a linear regime where any increases in the beam area
and/or the beam thickness beyond prescribed physical minima will add to the cost of the part.
From (A), l = gt / 9.2712 10 9 . The l corresponding with the tmin of 5 m and the minimum gap (0.5 m) is given
by: l min = 0.5 10 6 5 10 6 /(9.2712 10 9 ) = 2.696 10 4 m 270 m .
4
on [DD Month YYYY].

The w follows from (B):
l
mw
t 106
270106
m = 54106 m w
( 5 106 ) 106
The smallest w we can have and still satisfy all of the physical constraints associated with the problem is therefore:
wmin = 54 10 6 m = 54 m
We also know from (6) that the condition w 10t must be satisfied. Since tmin = 5 10 6 m = 5 m and
wmin = 54106 m = 54 m , this last condition is met, and we have therefore found our design values.
Thus g = 0.5m, t =5m, w = 54m, l = 270m. Substituting into equations (1)-(4) and (7), we get:
Vc= 9.98 V, R = 10, and the minimum cost is: 54 270 200 10 6 + (5 2) 2 = 8.92$
(c) Process: (Note that the actual dimensions of the device are slightly larger than those of the cantilever. Hence
the cost will be slightly higher than that calculated above.
1. Start with a silicon wafer, perform RCA clean with HF dip.
2. LPCVD 0.2 m of silicon nitride to act as insulator between the electrode and switch along the
substrate path.
3. LPCVD 0.2 m of polysilicon.
4. Perform photolithography using positive photoresist (not shown) and Mask 1 to define the
electrode.
5. Dry-etch the polysilicon using reactive-ion etching. Then ash resist and perform RCA clean
(without HF dip).
6. PECVD 0.5 m of sacrificial oxide
7. Perform photolithography using positive photoresist (not shown) and Mask 2 to pattern the
sacrificial layer.
8. Wet-etch the oxide using BOE. Then ash resist and perform RCA clean (without HF dip).
9. LPCVD 5 m of polysilicon.
10. Perform photolithography using positive photoresist (not shown) and Mask 3 to define the
cantilever.
11. Dry-etch the polysilicon using reactive-ion etching. Then ash resist and perform RCA clean
(without HF dip).
12. Release the cantilever by etching the oxide with BOE followed by super-critical freeze drying.
5
on [DD Month YYYY].
Cross sections
Masks
1
58 m
20 m
5
Mask 1
282 m
8
62 m
11
Mask 2
292 m
12
54 m
Mask 3
Silicon wafer
Nitride
Polysilicon
Oxide
Figure 1. Process flow and mask set for micromachined cantilever switch
Problem 6.11 (7 pts): A simple MEMS resonator

(a) For a doubly clamped cantilever beam, the effective spring constant is:
k=
16Ewt 3
l 3
(See p. 690 in Gere & Timoshenko, Mechanics of Materials, 4th Ed.)
mactual = lwt
The mass is :
meff = 0.4mactual
The pull in voltage is V PI =
8kg o 3
=
27Acap
8 16Ewt 3 g o 3
27 l elec wl 3
128Et 3 g o 3
27 l elec l 3
-------- (1)
(b) Based on the linearized model in the text:

C0 =
Acap
g o
l elec w
g o
k' = k
Vo2 C o2
V 2 l w
= k o 3elec ,
Acap g o
g o
and =
C oVo l elec wVo

=
g o
g o 2
6
Microelectromechanical Devices, Spring 2007. MIT
OpenCourseWare (http://ocw.mit.edu/),
(c) When we refer the impedances to the electrical side, we multiply each impedance on the mechanical side by
(1/)2, thus:
ZC 1 =
C1 =
Z1/ k ' =
2
k'
4
o
( l elec wVo )2
( l elec wVo )2
k V o l elec w
3
g o
1
=
s
C1 s
kg V l elec wg o
4
o
Zm
2
o
L=
g o is found by solving : g o = g o
(d) Z in = Z Co
ZL =
lwtg o4 s
( l elec wVo )2
= Ls
lwtg o4
( l elec wVo
)2
Vo2 l elec w
2kg o2
2
1
s +
LC1
1+ LC1 s
1 1
+ sL =
//(Z C1 + Z L ) =
//
=
2
sCo sC1
(s)(Co + C1 + LC o C1 s ) s 2 + 1 1 + 1
L C1 C0
2
1
sC0
-------- (2)
This is a third order system with 3 poles and 2 zeros.
Substituting s = j and setting the denominator and numerator to zero to find the poles and zeros respectively, we
find that:
The poles occur at s=0 ( = 0) and s= j where =

1
LC1
The zeros occur at s= j where =
(e) When V0 = 0, 1
k
4t
= 2
lwt l
1
=
LC1
2 =
=
When V0=VPI,
g o = g o
2V PI2 l elec w
2kg o2
C o + C1
LC o C1
---------(3)
k Vo 2 l elec w / g o 3
16Ewt 3 / l 3 Vo 2 l elec w / g o 3
=
lwt
lwt
16Ewt 3 / l 3 2 128Et 3 g o 3 l elec w /(27 l elec l 3 g o 3 )

lwt
16Et 2 2 128(g o / g o ) 3 Et 2 / 27
l 4
4t E
l2
(1
8 2 (g o / g o ) 3
)
27
4 2
= g o 1
27
go
g
o
(4)
--------------------(5)
7
(f) As 1 in equation (4) above, VVPI, g o
2
g o , and thus:
3
4t
l2
(1 2 ) = 1 1 2
At V0=0, 2 = 1.
At V0=0.95VPI , using the above approximation: 2 = 1 1 0.95 2 = 0.311
And thus the maximum change in 2 according to this approximation is:
0.311
100 = 69%
1
The actual change in 2 will be less than that calculated above because the approximate form for 2 is really only
accurate as 1.
In fact, if we solve (5) numerically , we get g o = 0.7805 g o
Substituting in (4) gives: 2 =
4t
l
E 8 0.95 2 (1 / 0.7805) 3
1
27

Thus the actual maximum change in 2 will only be :
(g) From (3), l =
4t
4 110 6
2 10
= 2.646t
l2
--------------(6)
2.646 4
100 = 33.85%
4
150 10 9
= 72 m
2300
Hence lelec must be 7.2 m. Use lelec = 7 m.

1
27 l elecV PI 2 l 3 3
= 0.27 m
Now from (1): g o =

3
128Et
w must be l/5 for the structure to act as a beam (rather than a plate), hence we will pick w = 10 m.
Figure 2 below plots f2 = 2 / (2) versus Vo as Vo varies from 0.05VPI to 0.95VPI . Figure 3 is a Bode plot of Zin at
Vo = 0.05VPI and Vo = 0.95VPI .The MATLAB code is attached at the end of the solution.
8
9
on [DD Month YYYY].
% MATLAB code for prob 6.11

function out=prob6_11()
g0 = 0.27e-6;
lelec = 7e-6;
w = 10e-6;
l = 71e-6;
t = 1e-6;
e = 8.85e-12;
rho = 2300;
E = 150e9;
A = lelec * w;
k = 16*(E*w*t^3/l^3);
m = t*l*w*rho;
VPI = sqrt(8*k*g0^3/(27*e*A));
V = linspace(0.05*VPI,0.99*VPI,100);
for in=1:length(V)
V0 = V(in);
G = fzero(@(G) gap(G,V0,k,e,A,g0),g0);
C0 = e*A/G;
phi = C0*V0/G;
kp(in) = k*(1-C0^2*V0^2/(e*A*k*G));
C1 = phi^2/kp(in);
L = m/phi^2;
w_r(in) = 1/(2*pi)*sqrt(1/(L*C1));
if in==1
Z1 = tf([1/C0 0 1/(L*C0*C1)],[1 0 1/L*(1/C0+1/C1) 0]);
end
if in==100
Z2 = tf([1/C0 0 1/(L*C0*C1)],[1 0 1/L*(1/C0+1/C1) 0]);
end
end
w = logspace(6,7,5000);
bode(Z1,'r',Z2,'b',w)
figure
semilogy(V,w_r)
function y = gap(G,V0,k,e,A,g0)
y = G - g0 + e*A*V0^2/(2*k*G^2);
10

Spring Term 2007
Issued: Lecure 13
Due: Lecture 14
Problem 12.6 (4 pts): Heating a thermal source to a uniform temperature

In class we examined a bolometer that detects incoming radiation by means of the temperature rise that it causes
in a suspended structure; an array of such structures acts as a MEMS thermal imager. In this problem we will
examine an array of similar structures, which could be used as a thermal source for calibrating an imager. In
this case, resistive heaters would be used to heat up an array of suspended structures. For this application,
separate pixels can be heated up to different temperatures, but the temperature at different points on a given
pixel should be as uniform as possible.
The pixel body is a square of silicon-rich silicon nitride (assume = 25 W/m-K) with a 50 m edge length, a 1
m thickness, and a 3 m gap between it and the substrate surface. The pixel is anchored to the substrate by 1
m thick silicon nitride tethers that are 5 m in width and 50 m long. Heating resistor(s) are patterned on the
surface of the pixel body. The details of the resistor layout may be arranged to minimize thermal nonuniformities across the pixel. The pixels are packaged in vacuum.
Schematic top view of pixel supported on

tethers, with a simple meandering resistor
covering the pixel body
Equivalent circuit overlaid on the pixel

diagram.
a) Write expressions for and calculate the thermal resistances that govern heat loss from the pixel,
assuming that it is in air and the substrate is maintained at 300 K. What is the dominant heat removal
mechanism (i.e. radiation, conduction, etc.)? We can also define a characteristic thermal resistance for
heat flow through the pixel body. As an approximation, we will define this as the thermal resistance
between centers of the individual quadrants of the pixel body. Write an expression for and then
calculate this characteristic thermal resistance (ie. Find an expression for R2 as shown the thermal
circuit.) . In what follows, the term relative temperature will refer to the difference between the absolute
temperature and that of the substrate (300K).
Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts
Institute of Technology. Downloaded on [DD Month YYYY].
We can capture some of the physics that governs temperature distribution across the pixel by means of a simple
equivalent thermal circuit, which is shown overlaid on the pixel above. In this circuit, each quadrant of the pixel
has its own thermal current source (provided by the heating resistor in that quadrant) and thermal resistances
between it and the neighboring quadrants (R2, which may be approximated from the characteristic resistance
calculated above). Some quadrants are also connected to the thermal resistance of the support tethers (R1).
b) Lets start by assuming that the heaters are laid out as a single, symmetric resistor as in the figure above,
so that the thermal currents to the four quadrants are identical. Create a finite difference matrix
G G
representation of the thermal circuit in the form: GT = I where T and I are column vectors representing
the relative temperature of and the thermal current into each quadrant, and G is a matrix whose elements
are functions of R1 and R2. Using MATLAB, find the value of the thermal current that will drive the
hottest node to 500K What is the absolute temperature of each quadrant in this case?
c) Now, by assuming that the thermal currents can have different values, find those values which will give
a uniform absolute temperature distribution of 500K as measured at each of the 4 nodes on the pixel.
What resistor layout would produce these thermal currents? What do you think is the best way to
implement variable thermal currents in each quadrant: by varying the layout of the heating resistor
between quadrants, or by having the quadrants supplied by separate circuits with their own electrical
currents? Why?
Problem 12.7 (4 pts): Transient analysis of the thermal source pixel
In this problem we will build upon our results from problem 12.6 and add in the transient response to determine
the rate at which the display can be refreshed. To do this, we need to add thermal capacitances to our circuit as
shown in the figure below to capture the heat capacity of the pixel plate.
R2
R2
R1
R2
I1
R1
I2
R2
I4
I3
a) Write an expression for and calculate the thermal capacitance C of each quadrant of the pixel. Assume
3
a specific heat for SiN of about 170 J/kg-K and a density of 3300 kg/m .
b) Using MATLAB, create a finite-difference matrix representation of the thermal circuit shown above in
the form
G
G
G
T = AT + B I
where T is a vector composed of the node temperatures of the thermal circuit relative to ground, I is
a vector of current source inputs, and A and B are matrices derived by discretizing the heat flow
equation.
c) Now we want to solve this equation to get the transient response. Noticing that the relation is in statespace form, we can use MATLABs control system toolbox to define a state-space system using ss,
where the C matrix is the identity matrix and D is all zeros. From this point, use step to determine the
step response of the system and thus the transient evolution of the system when the thermal current
sources are set to their optimum values, with all nodes starting at an absolute temperature T=300 K.
Plot the absolute temperature of each quadrant vs. time. What is the transient time to get to 95% of the
relative steady state temperature distribution that you calculated in problem 12.6?
d) Now we will simulate cooling down. The easiest way to do this is to use the Matlab command initial
using as initial conditions the node temperatures set to their steady state values as found in problem
12.6. Find the transient time to cool down the pixel to a relative temperature within 5% of the endpoint
value (when all nodes are at the substrate temperature). Plot the absolute temperature of each quadrant
vs. time. Is the cooling-down transient time the same as the heating up transient time? Why?
e) What is the maximum refresh rate for the thermal calibration source as described in this problem?
Describe in words how you could modify the design to increase the refresh rate of the pixel.
Problem 12.8 (4 pts): Design and analysis of a thermocouple microcalorimeter
Microcalorimeters are macroscale instruments that measure the heat produced by samples (such as heats of
reaction, phase change, etc.). In this problem you will design a microscale version of a microcalorimeter (micro
in microcalorimeter comes from the amount of heat (J), not the size of the instrument). The microcalorimeter
consists of a silicon nitride cantilever (thickness t=2 m) connected to a silicon substrate that is thermally
grounded at T=300 K. The samplewhich acts as the thermal sourceis placed on the end of the cantilever
over length a=50 m, and delivers 10 W of heat. The temperature difference is converted to a voltage
difference using an Al/polySi thermopile with a net Seebeck coefficient (between the two materials) of s=248
V/K. Each segment of the thermopile requires b=15 m of cantilever width. You can assume that
the silicon nitride has a thermal conductivity of =20 W/m-K, specific heat of 170 J/kg-K, and density
of 3300 kg/m3.
T( x,t)
0
SiN
b
w
l
+
V
Si
a) We could model this device by using a traditional lumped-element model. However, given that the heat
source area is not easily distinguished from the conduction area, we will instead lump the device in
terms of modes using an eigenfunction expansion. Assuming that heat conduction is the primary
mechanism for heat loss, determine the relevant boundary conditions on T(x). Next, determine a set of
eigenfunctions that will both solve the Poisson equation and meet these boundary conditions, along with
the separation coefficients kn. Finally, determine the coefficients of the eigenfunctions, incorporating
the fact that heat is generated over only part of the cantilever (from x=l-a to x=l).
b) Now extend your solution for T(x) to time-dependent temperature distribution T(x,t) by finding the
natural response of the system subject to the initial condition that T(x,0) is the temperature found in part
a. This models suddenly turning off the heat source (after reaching steady state) and letting the
temperature decay. Lump your system by assuming that we are interested in the temperature at
x=l. Take the Laplace transform of this solution to get T(s) and generate an equivalent circuit,
identifying Rn, Cn, and heat current source IQ,n(s). A good way to determine Cn is to look at the
mode volume, as we did in class.
st
c) Using the 1 -term of the solutions, determine R1, C1, and IQ,1(s) for our equivalent circuit.
d) We would like our microcalorimeter to have a 1 ms response time. Determine the dimensions l
and w that will result in this response time. Given a maximal thermoucouple packing of m=w/b
thermocouples, what is the output voltage? Is it possible to increase this without affecting the
time constant? How?
Problem 10.7 (4 pts): Effective mass of a cantilever beam
When we create lumped element models of electromechanical systems, we often need to include the lumped
mass of the element. The lumped mass for a given structure depends on how it is supported, and in general will
be different for different types of structures. In this problem, you will find the lumped mass of a cantilever
beam of Youngs modulus E, width b, thickness h, and length L. In particular, you will be looking at the case of
a tip-loaded cantilever beam. Since we have solved this problem previously (in class), you are welcome to use
the results of those calculations as a starting point.
a) First, find the elastic energy that is stored in the cantilever beam when it is deflected by a tip load F.
b) Next assume that the cantilever beam is undergoing simple harmonic motion at resonance, and find an
expression for the maximum kinetic energy of the beam.
c) Based on those results, find the resonant frequency of the cantilever beam.
d) Finally, what is the lumped mass of this cantilever beam?

Spring Term 2006
Issued: Lecture 14
Due: Lecture 16
Problem 13.9 (3 pts): Calculating fluidic resistances

Here we will examine the fluidic resistances of square versus rectangular channels. Assume a channel with crosssection of 50 m x 50 m and length L=1 cm, and that the working fluid is water.
a) Using a parallel-plate Poiseuille flow approximation, calculate the fluidic resistance of the channel in
Pa/(l/min).
b) The parallel-plate flow approximation is not strictly valid when the width of the channel approaches
the height. A more general relation between flow and pressure for a rectangular channel of width W
and height h (where Wh), is
tanh (2 n + 1)
W h dP
192 h
2 h
1 5
Q=
12 dx W n =0
(2 n + 1)5
Use this relation to calculate the resistance of the channel. What is the error of the parallel-plate
approximation?
c)
Determine analytically or numerically the minimum W/h where the Poiseuille approximation has a
10% deviation from the exact solution.
Problem 13.10 (4 pts): Timescales in microfluidic flows

One commonly assumes that in the creeping flow regime encountered in microfluidics, flows start and stop
instantaneously. However, what happens in the real world, where Re is not exactly zero? In this problem we will
find out.
a)
First, we will derive the relevant dimensionless quantities. Starting from the incompressible Navier-Stokes
equation in the text (Eq. 13.25), assume incompressible flow, neglect gravity, and expand out the total
derivative. What is the N-S equation now?
b) Next, non-dimensionalize the N-S equation using the following relations:

~
~
~2
~
2 2
u~ = U
,t = t ,P = P
U
U / L , = L , = L
0
where U0, , and L are a characteristic velocity, timescale, and length scale for the flow. Place the nondimensional N-S equation in the form of
1 u
P +
2 u
+ u u =
Re
Sr t
What are Re and Sr? One now sees how the L in the Reynolds number comes from the characteristic
~
dimension over which spatial change occurs (e.g., from = L ).
c)
In creeping flow we can neglect the convective term leading to

Re u
P +
2 u
=
Sr t
If Re<<Sr we return to our quasistatic creeping flow equation where time doesnt matter. Lets examine
this coefficient. The Re and Sr numbers can be represented as ratios of timescales. First, express Sr and Re
as
Sr =
Re =
v
c
What are c and v? What do they represent? The ratio is then

Re v
=
Sr
and thus the flow is quasistatic when v is much less than the timescale imposed by the external force.
d)
Suppose you have a square 100 m 100 m microchannel filled with water ( =10-3 Pa-s, m =1000
kg/m3). You apply a step input of pressure. Approximately how long will it take for the flow to reach
steady state? Which timescale dominates the flow development in this case?
Problem 13.11 (3 pts): Microfluidic networks and fabrication variations

We commonly create microfluidic networks that take an input flow and distribute that flow into a number of
streams. By changing the geometries of the fluidic channels we can create defined fluidic resistors and thus defined
flowrate variations across an array. In this problem we will examine how to do this, and, importantly, how robust
those predetermined variations are to process non-uniformities.
Given an input volumetric flowrate of water ( =0.001 Pa-s at 100 l/min) to the network shown in the diagram, we
would like to establish a 1:3:9:27 ratio in volumetric flowrate across the array (Q1:Q4). The height of the channels
is fixed at 50 m, while the width must be 150 m. The pressure at the channel exits is fixed and the maximum
pressure drop across the channels must be 1 psi. Finally, we want to minimize the total chip area. Approximating
the area of each section n as 2lnwn (to incorporate spaces between adjacent channels), we must keep the total area of
the resistive channels to <25 mm2. You can assume parallel-plate Poiseuille flow.
Q=100 l/min
Q1
Q2
Q3
Q4
l1
w1
a)
Determine a set of geometries (l1, w1, l2, w2, etc.) of the channels that meets the above requirements, and
then calculate the flow in each channel and the pressure drop across the channels.
b) Now assume that there is a fabrication variation across the chip, such that the channel height varies 10%
across the chip. Assume first that the channel height increases as one goes from channels 1-4, and then that
the channel height decreases as one goes from channels 1-4. How does each of these variations change the
flowrate ratios across the array? You may assume that the channel height is constant for a given section,
and increases stepwise from section to section.

Spring Term 2007
SOLUTIONS TO PROBLEM SET 6 (10 pts)
By Xueen Yang (Solutions adapted from Yong Shi) minor edits Brian Taff
Posted on 4/11/07
Problem 13.9 (3pts): Calculating fluidic resistances

a). For a parallel-plate Poiseuille flow approximation, the flow rate and the pressure drop is related by
12L
P =
Q
Wh 3
Using the eV convention, the fluidic resistance of the channel is
R pois =
12 L
Wh 3
1
Pa min) 0.1dm
60
R pois =
5
50 10 dm (50 105 dm)3
Pa min
Pa min
= 320 106
= 320
L
L
12 (0.001
b). We can express the general relation between flow and pressure to be,
Q = (1 A)
Wh 3 dP
12 dx
where we have used A to denote,

W
tanh (2 n + 1)
2h
5
W n=0
(2 n + 1)
A= 192 h
For Poiseuille flow, pressure drop is linear with length, so we have, dP = P , and hence,
dx
P =
12L
Q
(1 A)Wh 3
The fluidic resistance of the channel is
12L
(1 A)Wh 3
Using Matlab, we can find that A converges at 0.5783, substitute in, we have
R poisFull =
1
Pa min) 0.1dm
60
R poisFull =
5
(1 0.5783) 50 10 dm (50 105 dm)3
Pa min
Pa min
320
=
= 758.83
1 0.5783 L
L
12 (0.001
The error is
Error =
R poisFull R pois
R poisFull
12L
12L
(1 A)Wh 3 Wh 3
=
= A = 57.83%
12L
(1 A)Wh 3
c). As we derived from b), the error can be expressed by A, where
tanh (2 n + 1)
2 h
A=
192 h

5 W n=0
(2 n + 1)5
Using Matlab, it can be determined that the min W/h ratio for error to be less than 10% is 6.305.
The Matlab script is posted here:

clear all;
close all;
format long eng
L=1E-2;
visc=0.001*(1/60);
W=50E-6;
h=linspace(50E-6,5E-6);
h=h';
Wperh=zeros(100,1);
Error=zeros(100,1);
Rpoisfull=zeros(100,1);
Rpois=zeros(100,1);
for m=1:100
h_now=h(m,1);
Wperh(m,1)=W/h_now;
for i=0:24
Aterm(i+1,1)=((192*h_now)/(pi^5*W))*(tanh((2*i+1)*(pi/2 ...
)*(W/h_now))/(2*i+1)^5);
end
A=sum(Aterm);
Rpoisfull(m,1)=(12*visc*L*10*1E-6)/((1-A)*(W*10*(h_now*10)^3));
Rpois(m,1)=(12*visc*L*10*1E-6)/(W*10*(h_now*10)^3);
Error(m,1)=(Rpoisfull(m,1)-Rpois(m,1))/Rpoisfull(m,1);
end
plot(Wperh,Error)
ylabel('Error')
xlabel('Channel width and height ratio W/h')
Problem 13.10 (4 pts): Timescales in microfluidic flows

a). The Navier Stokes equation is
m
DU
= P + m g + 2U + (iU )
Dt
3
Assuming incompressible flow and neglecting gravity,

iU = 0 and
The equation becomes

m (
m g = 0
U
+ U iU ) = P + 2U
t
b). Substitute the following expressions into the equation:

U = U 0 u~
t = ~
t
~
P = ( U 0 / L)P
~
=
L
~
2
2 = 2
L
We have
U2
U
U
u U 0
+ u iu 0 ) = P 2 0 + 2 u 20
t
L
L
L
m LU 0 1 u
+ u iu = P + 2 u
U
0 t
m (
Re =
Sr =
m LU 0
U 0
L
c).
Sr =
c =
Re =
L
U0
v
c
v = Re c =
=
c is the time scale of convective flow
m LU 0 L
U 0
m L2
and v is the time scale of viscous flow.
d). Applying a step input of pressure, the flow near the wall can not violate the no-slip condition. The time scale to
establish steady flow is determined by the visous flow time scale. Using a length scale on the order of the hydraulic
diameter, we have
2Wh
2 100 100
= 100m
L=
=
100 + 100
W +h
3
v =
(100 10 6 ) 2 m 2
= 0.01sec
10 6 (m 2 / s)
Problem 13.11 (3pts): Microfluidic networks and fabrication variations
Q=100 l/min
Q1
Q2
Q3
Q4
l1
w1
a). First let us summarize the constraints of the design of the fluidic channels:
1. Total volumetric flow rate is 100 L/min;
2. Flow ratio across the four channels are Q1 : Q2 : Q3 : Q4 = 1: 3 : 9 : 27 ;
3. Channel height h is fixed at 50 m; width w 150 um and total area A 25 mm 2 ;
4. The maximum pressure at the inlet must be 1 psi.
Assuming parallel-plate Poiseuille flow, the flow resistance is,
P 12 l
=
R=
Q
wh 3
The volumetric flow rate across each channel can be found to be,
Q1 = 2.5, Q2 = 7.5, Q3 = 22.5, Q4 = 67.5 L / min
Since the pressure drop across the channels is the same (both the inlets and outlets are connected), we can
derive that,
R1 : R2 : R3 : R4 = 27 : 9 : 3 :1
l1 l2 l3 l4
:
:
:
= 27 : 9 : 3 :1
w1 w2 w3 w4
From total area constraint, we can write,
2(l1w1 + l2 w2 + l3 w3 + l4 w4 ) < Amax
To minimize the area, we can set the width at its minimum, w = 150 um , and we can reduce the above
expression in terms of l1,
1 1 1
2w 1+ + + l1 < Amax
3 9 27
27 Amax
27 25 mm 2
l1 <
=
= 56.25 mm
80w
80 150 103 mm
We can choose a set of channel dimensions, such that, l1 = 54 mm, l2 = 18 mm, l3 = 6 mm, l4 = 2 mm . The
pressure drop across the channel is,
12 l1
12 0.001 Pa s 54 mm
2.5 106 103 m 3 / min
P =
Q1 =
= 1.44 103 Pa = 0.21 psi < 1 psi

3
3
3
6
wh
60
s
/
min
150 10 mm 50 10 m
So the constraint on the inlet pressure is met.

4
b). Now we assume that the channel height varies 10% across the chip in stepwise fashion and we will examine two
cases: case 1, the height increases from 50 m for channel 1, to 55 m for channel 4, and case 2, the height
decreases from 50 m to 45 m.
Since all the channels are designed to have the same width and same pressure drop, the flow rate is only a
function of length and height,
h3
Q
l
Therefore, the flow resistance ratios for both cases are,
503 51.673 53.333 553
case 1 Q1 : Q2 : Q3 : Q4 =
= 1: 3.3 :10.9 : 35.9
:
:
:
27
9
3
1
503 48.333 46.673 453
case 2 Q1 : Q2 : Q3 : Q4 =
= 1: 2.7 : 7.3 :19.7
:
:
:
27
9
3
1
As expected, since Q is proportional to the cube of the height, the flow rate is very sensitive to any height
variation. And a 10% height variation across the chip can cause the flow rate to vary from 26% to 33% in this case
for the longest channel.

Spring Term 2007
Issued 4/11/07
Due 4/20/07
Problem 15.7 (4 pts): Design a PID controller

We are trying to control an electrostatic electrode actuator with a lumped mass of 1 g, a resonant frequency of 50
kHz, and a Q of 10. We will model the linearized system as a spring-mass-damper 2nd-order system. The plant
transfer function H(s) will be X(s)/F(s).
a.
First, recast the transfer function in terms of the non-dimensional complex frequency s = s / 0 .
b.
Now, assume that you are using a single-pole controller of the form in equation (15.14), with non
dimensionalized time constant = 0.1 . Using a root-locus plot for the overall control system transfer
function Xout(s)/Xin(s), determine the maximum controller gain (K0) at which the system is stable.
c.
Instead of using a single-pole controller, design a PID controller that achieves overall critically
damped system response with no DC error. Demonstrate results with SIMULINK or MATLAB
simulations.
Problem 15.8 (4 pts): Stability and capacitive loads

We are often faced with the task of driving capacitive loads including electrostatic transducers such as parallel-plate
actuators. Such loads, when driven at high frequencies, can create circuit instabilities without proper care. In this
example, assume we are driving a capacitive load of C=200 pF with a buffer op-amp controller K(s) (figure, A).
+
+
-
Ro
Vo
v+
K(s)
V +
+ K(s)(v+-v )
-
v-
(A)
Vo
C
(B)
a. Assume that the amplifier (like ALL amplifiers) has a finite output resistance R0=40 , using the
circuit model in (B), with the load capacitor C, determine analytically the loop transmission function
H(s)K(s) for the circuit, where the loop transmission function is defined as V0 = H (s)K (s)(V V0 ) .
b. Now assume that K(s) is a 2nd-order transfer function of the form
103.5
K (s) =
1+ s
1+ s
2 105
2 108
which approximates the National Semiconductor LM359 high-speed op-amp. Determine the overall loop
transmission function H(s)K(s). Make a Bode plot for the loop transmission function and determine the
phase margin with and without the capacitor. What is the maximum capacitive load that this amplifier can
drive and be stable?
)(
1
Problem 16.6 (10 pts): Noise in a (vastly simplified) capacitive accelerometer

In this problem, we will compare noise sources for a differential capacitive accelerometer. The accelerometer
employs interdigitated capacitive fingers, which are read differentially. When the proof mass moves, one
capacitance increases and the other decreases.
Vs
-Vs
The circuit assumed here is vastly simplified; we will learn more about accelerometer readouts in the case studies.
In the meantime, we assume that voltages are applied to the fingers in pulses with an amplitude Vs = 2.5 V. We will
look at the noise from various sources while the pulse is applied (in other words, we dont look very long, so it looks
like DC.) However, we will assume a finite bandwidth due to the remainder of the circuit, which is not shown here.
a. First we will examine thermomechanical noise. Assume that the microaccelerometer has a proof mass of
300 ng, a resonant frequency of 25 kHz, and the bandwidth of the measurement is 1 kHz. Also assume that
the accelerometer is critically damped and is in thermal equilibrium at room temperature. What is the total
thermomechanical noise in the bandwidth of interest?
b. Assume that at zero displacement the inter-finger gaps are 1.3m and the inter-finger capacitances C1 = C2
= 100 fF (purely parallel plate capacitances). Also assume that the parasitic capacitance Cp = 100 fF.
What is the effective variation in the capacitances C1 and C2 due to the fluctuation of the position of the
proof mass? Find the relationship between Vx (the voltage between the capacitors) and Vs, and use it to
calculate an effective voltage noise source from the thermomechanical noise at the point labeled Vx.
2
c. We saw in class that a capacitor connected to a thermal reservoir has a voltage noise associated with it. In
this case, the result is modified because the bandwidth of interest is limited to 1 kHz. Estimate the total
voltage noise introduced by the capacitors in the bandwidth of interest. You may assume that the dominant
resistor to which the capacitors are connected is the resistance of the long interdigitated fingers (assume 25
fingers in parallel). Each finger is made of heavily-doped polysilicon and has a resistivity of about 400
-cm. Each finger is 100 m long and has a cross-sectional area of 25 m2.
d. We will now consider the noise sources for the amplifier. Write an equation for the amplifier noise at the
inputs, and calculate the spectral density as a function of frequency. Assume a transconductance of 300
Siemens, a channel length L = 2 m, a channel width W = 30 m, and a gate oxide thickness of 15 nm;
also take Kf = 10-24 V2F. Calculate the total voltage noise at the amplifier inputs over the bandwidth of
interest, assuming a low frequency cut-off of at 10 Hz.
e. Redraw the accelerometer circuit diagram to include all of the noise sources. Analyze the circuit to find V0
in terms of Vs, the effective noise voltages, and the zero-noise component values. Assume that the noise
sources are uncorrelated, and calculate the total voltage noise at the output V0. Which term is dominant?
Does the form of the transfer function for this circuit (with the unity gain buffer) have a significant effect
on which terms are dominant?

Spring Term 2007
By Xueen Yang (Solutions adapted from Yong Shi) Simplifications and edits by Brian Taff
Posted 4/20/07
Problem 15.7 (4pts): Design of a PID controller
a). From Eq. (15.5), the transfer function after normalizing the frequency to the undamped resonant frequency is,
X ( s)
1/ k
,
= H ( s) =
1
F ( s)
s2 + s + 1
Q
where we have,
s
, where o = 2 50kHz = 105 rad / s
s =
k = o2 m = 2 1010 ( rad / s ) 109 ( kg ) = 10 2

Q = 10
And hence,
H ( s) =
10
1
s 2 + 0.1s +1
b). We will use a single pole controller, which has the form
K ( s) =
Ko
1 + 0.1s
The loop transfer function is hence,

H ( s) K ( s) =
Ko /10 2
( 0.1s +1) s2 + 0.1s +1
Using Matlab command rlocus(sys), the root locus plot of the closed loop transfer function is shown in
Figure 1 below. A zoomed-in view of the root near the imaginary axis as shown Figure 2 reveals that the maximum
gain Ko at which the two poles will coincide with the imaginary axis and hence the system will become unstable is
98.9. The Matlab code used to generate these two plots is provided.
Figure 1. Root locus plot of the closed loop transfer function.
Figure 2. Zoomed-in view near the imaginary axis.
% root locus plot

clear all;
close all;
Hden=[1 0.1 1];
Hnum=[1/10/pi^2];
H=tf(Hnum,Hden)
Kden=[0.1 1];
Knum=[1];
K=tf(Knum,Kden);
sys=H*K;
rlocus(sys);
c). Instead of using a single-pole controller, we will design a PID controller that achieves overall critically damped
system. A normalized PID controller has the form,

K ( s) = K o 1+ + s
The closed loop transfer function now becomes,

Ko
s 2 + s +
X out
k
=
X in
1 K
K
K
s3 + + o s 2 + 1 + o s + o
k
k
k
This is a third order system with 3 poles and 2 zeros. We know that for a second order system, critically damped
response means that the system transfer function has two real poles that are equal, and as a result, the system
achieves steady state with the fastest response without overshoot. In order for the system to exhibit a second order
behavior, we ideally would want to have 2 equal real poles and a third pole that would cancel out a zero. It turns out
that for this problem, it is not possible to achieve the critical damping with a PI controller.
What we can do,
however, is to have a third pole that is much larger than the 2 real poles such that its fast response does not have
much noticeable effect on the second order system behavior. Therefore, we can express the denominator as,
2
( s + a )( s + b ) , where a b
And by comparison, we have,
1 Ko
Q + k = a + 2b
Ko
= b2 + 2ab
1+
k
Ko
2
k = ab
There could be many choices for a and b that could satisfy the criteria of critical damping, and without
further specification on the rise time, for example, we will choose one pair that works. From part a), we can derive
that the plant H(s) has two complex poles 0.05 j . In order to have faster response, we want to have two real poles
move to the left of the S-plane. Lets choose b=1, and a=1000, say. We can derive that,
Ko = 2000k = 2 104 2
= 0.5
= 0.501
Hence the PID controller that is chosen is,

2
0.5

+ 0.501s
K ( s) = 2 104 2 1+
s
The root locus plot of the new loop transfer function
0.5
+ 0.501s
2 103 1+
s
H ( s) K ( s) =
s 2 + 0.1s +1
is shown in Figure 3. The two zeros are very close to the real poles. Although it is advantageous to have the zeros
far away from the dominant poles, since the zeros of a system affect pretty much the amplitude, rather than the
oscillation nature of the system (as long as they are in the negative half of the s-plane), we do not have to worry
about them too much. The step response of the overall closed loop system is shown in Figure 4, demonstrating the
critical damped response, with a rise time on the order of 3 ms.
Figure 3. Root locus plot of the system with PID controller.
Figure 4. Step response of the system wit PID controller.
The MATLAB coding used for this portion of the problem is provided below:
% root locus plot

clear all;
close all;
Hden=[1 0.1 1];
Hnum=[1/10/pi^2];
H=tf(Hnum,Hden)
Kden=[1 0];
ko=20000*pi^2;
Knum=ko*[0.501 1 0.5];
K=tf(Knum,Kden);
sys=H*K
rlocus(sys);
figure
step(sys/(1+sys),0.01)
Problem 15.8 (4pts): Stability and capacitive loads

a). Determine the loop transmission function H(s)K(s)
+
-
Ro
Vo
Vo
v+
K(s)
V +
v-
+ K(s)(v+-v-)
-
(A)
(B)
From the circuit model on the left, we have

V = V0
V += V
Applying KCL at the node connecting the output resistor and the capacitor:
V0 K s (V V0 ) V0
+
=0
1
R0
sC
Rearrange terms, we get,
K (s)
Vo =
(V V0 )
1 + Ro s C
The model is equivalent to a linear feedback system with an overall transfer function as,
V0 = H (s)K (s)(V V0 )
which is referred to as Blacks formula.
The Loop transmission function is hence,
H (s)K (s) =
K (s)
1 + Ro s C
b). Assuming
K ( s) =
103.5
(1+ s 2 10 )(1+ s 2 10 )
5
The transmission function then becomes

H (s)K ( s) =
103.5
(1+ s 2 10 )(1+ s 2 10 )
5
2 + s(0.5005 10 5 + Ro C 2 )
1
1+ Ro s C
3162.28 2
+ s 2 0.25 10 13 + 0.5005 10 5 Ro C + s 3 0.25 10 13 Ro C
For the case C = 0, the loop transmission function becomes

H (s)K (s) =
3162.28 2
+ s 0.5005 10 5 + s 2 0.25 10 13
2
For the case C = 200x10-9 F, the loop transmission function becomes

4
H (s)K (s) =
2 + s (1.58 10 5 )
3162.28 2
+ s 2 1.5079 10 13 + s 3 2 10 22
The phase margin and bode plots of the two cases are shown below.
The phase margin angle is defined as
= 180 o +
where is the phase angle where the amplitude of the output signal is equal to the amplitude of the input signal. A
system is stable implies a positive phase margin value.
For the capacitance value of 0 F, the phase margin is 31.3510.
For the case with C = 200 pF = 200x10-12 F the phase margin is -19.2640. The system is unstable. We can
see that a larger capacitance implies a smaller phase margin.
5
The phase margin is zero when the capacitance is decreased to about 18.44 pF, which is the maximum
capacitance to be driven stably and the corresponding frequency is 147 MHz.
Problem 16.6 (10 pts): Noise in a (vastly simplified) capacitive accelerometer
a). In general, the force-displacement characteristic for a spring-mass-damper system is described as follows:
Knowing that the mass of the accelerometer is m=300 ng, and a resonant frequency o = 2 25000 rad/s, and they
system is critically damped, we can expressed the system transfer function for position x as,
x( s)
1
= 2
F ( s ) ms + bs + k
1
m
=
2
b
s +
s+k
m
m
H ( s) =
( )
In our system, we know that we are in a critically damped situation. We can there for insert the following
expression in the generalized transfer function expression.
m0 1
= = Q for the critically damped case
b
2
k
m
where 0 =
This substitution yields the following simplified relation:
1
x( s)
=
F ( s ) ms 2 + bs + k
1
m
= 2
s + 20 s + 02
H (s) =
( s + 0 )
The intrinsic noise due to damping is similar to the noise generated by a resistor, and the spectral density function of
this noise has the form,
S n ( f ) = 4k BTb
where b=2m o for the case of critical damping.
Though we could solve this problem using the following integral:
f
x =
2
n
H ( j 2 f ) S n ( f ) df
2
.
we can instead approach to the problem by using the fact that the system bandwidth is much smaller than the
resonant frequency of the accelerometer. Hence, we can approximate the quasi-static response, and use the
bandwidth as the noise bandwidth. The mean square force acting on the mass is,
f n2 = 4k BTbf
So the displacement is,
4k BTbf
fn
=
= 5.337 pm
k
mo2
which is nearly the same as obtained by performing the integral
xn =
b). The effective capacitance C1 and C2 can be expressed as

0 HL0
Csense =
=
G0 xn , rms
0 HL0
G0
= C0
1
1
xn, rms
G0
1
xn, rms
G0
7
where C0 is the capacitance at zero displacement, and G0 is the initial gap. We can further simplify this relation by
linearizing around the initial gap condition using Taylor series expansion where we retain only the first term. We
outline such a procedure here:
f = C0
1
1
x n , rms
f
=
y
and
G 0
C 0
f
=
y
G
C0
x
G0 1 n, rms
G0
C0
G0
0
G0 1
G
0
The linearized form when retaining only the 1st Taylor series term:
0
Csense f
G0
f
+
y
G0
f
xn, rms = C0 +
G0
xn, rms = C0 1 n, rms
G
0
The effective capacitance due to the fluctuation of position is therefore:

5.337 1012
)
1.3106
The effective variation in C1 and C2 due to the fluctuation is 4.105 1019 F .
Csense 100 1015 (1
To find the relation between Vs and Vx, apply KCL at the node connecting C1 and C2,
Vs V x V x (Vs ) V x
=0
1
1
1
sC1
sC 2
sCp
Simplify the equation, we have

Vx =
C1 C2
Vs
C1 + C2 + C p
We reasonably assume that with 2.5 V actuation, the nominal capacitances are relatively unchanged. Hence, the
effective voltage noise source at the point of Vx is
x
x
C0 1+ n, rms C0 1 n , rms
G0
G0
vT =
3C0
x
2 n, rms
G0
V =
Vs
s
3
5.337 1012
1.3 106 2.5
=
3
= 6.842106 V
2
c). The mean square voltage noise on the capacitors in our system is described using equation 16.23 from the text.
We repeat it here in a slightly altered form.
vC , rms = vC1 = vC 2 = vCp = 4k BTRf
In this expression, R represents the resistor that connects to the capacitors and also the thermal reservoir. f is the
band width of interest. For the architecture outlined in this problem, we calculate the resistance using the following
routine:
R=
L
nA
= 400 106 102 m
100 106 m
25 (25 1012 m 2 )
= 0.64
8
Pluging in, the root mean square voltage noise on the capacitors is:
vC , rms = vC1 = vC 2 = vCp = 4k BTRf
= 4 1.38 1023 J / K 300 K 0.64 1000 Hz
= 3.256 109 V
d). The mean square noise at the two inputs of the operational amplifier can be written using equation 16.19.
v A2 = v B2 =
1000
S n ( f )df
f cut off
where, from equation 16.32

S n ( f ) = 4k B T (
K f
2
)(1+ Fn ) +
3g m
WLC 0 x f
In this expression:
g m is the transistor transconductance
Fn is the noise factor (2~5), assuming 5
K f is a scale factor for the 1/f noise, which is 10
-24
V2 F
WLC 0 x is the gate-to-channel capacitance and
C 0 x = 0 x
t ox
ox is the permittivity of oxide which is 3.9 0
t ox is the gate oxide thickness
Substituting in gives:
S n ( f ) = 4 1.3810 23 300
= 2.2080 10 16 +
2
(1+ 5) +
3 300 10 6
10 24
3.9 8.8510 12
f
30 2 10 12
1510 9
7.2432 10 12 2
V / Hz
f
The mean-square voltage noise

1000
v A2 = v B2 =
1000
10
f cut off
S n ( f )df
(2.2080 10 16 +
7.2432 10 12
)df
f
= 3.3649 10 11V 2
v A,rms =v B,rms = 5.008 10 6 V
e). The equivalent circuit is shown below.
I1 =
(Vs vc1 ) (V x vT )
1
sC1
I2 =
(V x vT ) ( Vs + vc 2 )
1
sC 2
Ip =
Vx v p
1
sC p
Apply KCL:
I1 = I 2 + I p
Substitute in
(Vs vc1 ) (V x vT ) (V x vT ) ( Vs + vc 2 ) V x v p
+
=
1
1
1
sC1
sC 2
sC p
Simplify the equation, we have
(Vs + vT vc1 )C1 + v p C p (Vs vC 2 vT )C 2
Vx =
C1 + C 2 + C p
where Vs is the signal. Since the noise sources are not correlated, the equivalent mean square noise at Vx is
2
n,x
C1
=
C +C +C
2
p
1
C2
(vT + vc12 ) +
C +C +C
2
p
Cp
2
vT + vc2
+
C +C +C
2
p
2
vp
in accordance with the remark at the bottom of p. 434 in the text.

Using the short method for the op-amp circuit, we have
V0 v B = V x + v A
V0 = v B + v A + V x
The three noise sources are again uncorrelated, the mean square noise at the output is
Vn2,0 = Vn2,x + vB2 + v A2

Substitute in Vn ,x and C1 = C 2 = C p , we have
2
2 2 1 2 1 2 1 2
vT + vc1 + vc2 + vcp + vB2 + v A2
9
9
9
9
2
1
= (6.842 106 )2 + (3.256 109 ) 2 + 2 (5.008 106 ) 2
9
3
= 6.0563 1011V 2
2
Vn,0
=
The root mean square of the noise at the output is

10
vn ,o , rms = 7.7822 10 6 V
The total noise output is seen to be dominant by the thermal noise and the amplifier noise. These two noise
sources have same order of magnitude and are three orders of magnitude larger than the capacitive noises. The
transfer function of this circuit does not change the proportion of the contribution of the noise sources and hence we
conclude that the form of the transfer function has no significant affect on the dominant terms of the noise.
11
Reference(s)
Affiliation(s)
Example of MEMS Device Treasure Hunt: Tunable capacitor/Tuning ratio

By Xueen Yang
Device #1
Device #2
Goldsmith, C.L., Malczewski, A., Yao, Z.J., Chen, S.,
Borwick, R.L., III, Stupar, P.A.,

Ehmke, J. and Hinzel, D.H., RF MEMS variable
DeNatale, J., Anderson, R., Chialun
capacitors for tunable filters, International Journal of
Tsai, Garrett, K. and Erlandson, R., A
RF and Microwave Computer-Aided Engineering, 1999.
high Q, large tuning range MEMS
capacitor for RF filter systems, Sensors 9: p. 362-374.
and Actuators A (Physical), v A103, n 12, 15 Jan. 2003, p 33-41
Borwick, R.L., III, Stupar, P.A.,
DeNatale, J., Anderson, R., Chialun
Tsai, and Garrett, K., A high Q, large
tuning range, tunable capacitor for RF
applications, MEMS 2002, Las Vegas,
NV, USA, pp 669-672
Rockwell Scientific Company
Raytheon Systems Corporation
Electrostatic comb drive with double

beam suspension as shown in this SEM
micro graph:
Bistable metallic membrane suspending over an

insulated electrode
A DC voltage applied between the membrane
and the electrode snaps down the membrane and
the change of the gap results in change of the
capacitance between the membrane and the
electrode
The schematic below shows the device before
and after actuation:
Device #3
Device
description
The comb fingers form the many

capacitors in parallel
Applied DC voltage changes the
overlapped areas of the fingers and
hence the value of the total capacitance
Below is a SEM micro graph showing
the etched comb fingers:
A device top view is shown here:
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J Design and Fabrication of Microelectromechanical Devices, Spring 2007. MIT
1/3
Bulk micromachining is used

Use SOI (silicon on insulator) wafers
Epoxy bond of silicon to glass
Comb fingers are etched with DRIE
(deep reactive ion etch) technique
Fabrication process is shown below:
Surface micromachining is used

Use high resistivity silicon as substrate
Thermally grown oxide is used as the buffer
layer
Nitride is used as dielectric
Aluminum is used as metal posts and membrane
Fabrication sequence is shown below:
Fabrication
technology
2/3
Pros of
approach
Cons of
approach
Use double beam suspension to increase

resistance and hence reduce sensitivity
to vibration
Use double-sided metallization to reduce
stress caused by deposition of metal
Use epoxy to bond silicon to glass to
reduce cost and simplify process;
oxygen plasma release of structure does
not cause post-release stiction
Process involves front to back alignment
Device might subject to low break-down
because of large areas and small gap
Process is not IC compatible
Value of
reported
metric
Measurement
method
Confidence in
results (1- 5)
and why
Comments
Analog tuned
Tuning ratio is 8.4:1 for 8V actuation
voltage
Q is above 100 for frequencies below
700 MHz
A HP network analyzer is used to measure the
capacitance and the S parameters
4 device yield is not reported; low break-down
voltage might limit operation range
The device has high analog tuning ratio and
relatively low actuation voltage compared to
other literature found. Although device with
higher tuning ratio exists, in terms of overall
performance, this is the best device so far.
Simple surface micromachining processes and

simple operation
Fast and robust switching of capacitance
Refractory metal is used as bottom electrode to
provide good conductivity and smooth surface to
minimize the air gap and hence maximize the
close capacitance
Process is IC compatible
Bistable nature of the device allows only digital

tuning
Residual stress in aluminum during deposition
and post release will cause membrane to warp
and hence initial capacitance and pull-in voltage
can vary
Use of high resistivity silicon substrate is more
expensive while the substrate loss is higher
compared to glass
Digital tuned
Tuning ratio of single device is 70-100:1 for 3050 V actuation voltage
Q is 20 at 1 GHz
S-parameter measurements were performed using a

Wiltron 37279 vector network analyzer and a Cascade
Summit 10000 RF probing system.
4 device yield is not reported; stiction could be a
problem during operation
A tunable capacitor with desirable tuning configurations

can be achieved by using a network of this switching
capacitor device. But the overall Q is lowered as a result
of parasitic effects. Also, the actuation voltage of this
device is too high to be useful in portable communication
applications.
3/3
Device/Specification
Name (Email)
Device #2
Device #1
Device #3
Reference(s)
Affiliation(s)
Device
description
Fabrication
technology
Pros of
approach
Cons of
approach
Value of
reported
metric
Measurement
method
Confidence in
results (1- 5)
and why
Comments
1/1
Design Projects
Some sample design projects used in previous terms are described briefly below, and
are representative of the types of projects we will use. Descriptions, specifications
and design goals for this term's projects will be provided later in the term. The scope
of each project will include a microfabricated device, the drive/detection electronics,
and a packaging concept. Each project will have a team of four or five students.
Depending on enrollment, there may be more than one team on a given topic.
1. A fast dielectrophoretic cell sorter
Cell sorting is a commonly used biological technique that can isolate target cells from
a bulk population based on differences in size, shape, or molecular specificity. In this
project you will explore the fundamental limits for micro-flow sorters by designing a
truly fast (>1,000 cells/sec) micro-cell sorter. The sorter will be created out of PDMS
on glass and will use dielectrophoresis to push incoming cells into one of two outlet
microchannels. In the process you will couple it to an optically based detection
system, actuation electronics, and identify the fundamental limits on sorting.
2. A piezoresistive sensor for biomolecular recognition
The goal of this project is to create cantilever-based device that detects stress
induced by molecular binding. Two cantilevers (operated differentially) will be
created out of SU-8 with integrated poly-Si piezoresistors. The packaged device will
be used in a hand-held point-of-care diagnostic monitor and so must be robust,
small, and connected to a circuit that gives an output proportional to the logarithm of
the concentration ratio.
3. RF-IF filter
The superheterodyne radio receiver, one of the truly brilliant inventions of the 20th
century, uses a nonlinear frequency converter in combination with a tunable local
oscillator to convert the incoming radio signal to a lower fixed frequency (the socalled intermediate frequency or if). The goal of this project is to design a
surface-micromachined silicon resonator or combination of resonators that are used
to create an if filter for incorporation into a one-chip radio.
4. A micro flex-tester for measuring the compliance of microstructures
The goal of this project is to build a microfabricated force-displacement sensor device
to characterize the compliance of microstructures. The micro-flextester is a
metrology tool used to measure the actual force-displacement characteristic of
microfabricated compliant structures. You will select a sensing scheme (i.e.,
piezoresistive strain gauges or capacitive displacement sensors) and use it in a
device that must be small, have tunable force resolution, and integrated
displacement and force sensing.
Cite as: Carol Livermore and Joel Voldman, course materials for 6.777J Design and Fabrication
of Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/),
2.372J/6.777J Design and Fabrication and Microeletromechanical Devices

Spring Term 2007
Design Problem Reports
Preliminary Report (Lecture 17)
This is a 1-2 pg brief outline of
the major challenges in the project, as seen by the team
the fundamental approach (e.g., a piezoresistive bulk-micromachined accelerometer)
which tasks are assigned to which team members, along with a rough timeline
This report only needs to be shared with the project mentor, not the entire teaching staff. Email of MS WORD
or PDF documents is fine.
Interim Report (Lecture 22)
This is a 2-5 pg more in-depth analysis summarizing the progress to-date and what the endgame is for the team
to finish the design by mid May. Identify outstanding issues and how the team will solve them. This report only
needs to be shared with the project mentor, not the entire teaching staff. Email of MS WORD or PDF
documents is fine.
Final Presentations (5 days after Lecture 25)
During the day (times are TBA) each team will give a talk and answer questions from the audience and a panel of
independent judges. Total time including questions will be 30-min/team, and this will be strictly maintained.
After all the talks, an award is given to the best presentation. Each team makes ONE presentationeach
member can give part of the presentation, or one/two member(s) can give the whole presentation.
Final Report (6 days after Lecture 25, 5 PM SHARP)
The final report should summarize the final device, including major design issues, modeling approaches,
performance specs, etc. The report should aim for 25-30 pages. Shorter is acceptable. Longer is not. Fonts
should be at least 10 pt and margins at least 0.75". It is important that the report be coherent and concise
these will affect the grade.
Each team must submit:
One Hardcopy of the REPORT to either Joel or Carol.
One Electronic version of the REPORT in PDF format and the PRESENTATION in PDF or
PPT format, either via email, burned to a cdrom, or EASILY downloadable from the web.
The electronic version is in addition to the hardcopy. Each team MUST submit a PHYSICAL
HARDCOPY upon which we will base the team's grade.
Both the hardcopy and electronic versions must be handed in by 5 PM on the due date, no
exceptions.
Grading
Each persons project grade is based on the teams project grade (50%) and that persons individual project
grade (50%). The team grade will be based on the physical hardcopy of the final report. It reflects the work
of the team as a whole and is the same for each team member. The individual project grade will be assigned
based on a peer grading process. At the conclusion of the project (after the final report is complete but before
noon 8 days after Lecture 25), each person must send an email to Joel and Carol assigning a letter grade to
each of their fellow team members contributions. An A is a solid team member, and an F is a no show. The
instructors will use the results of the peer grading to assign an individual grade to each person. Peer grading
Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute
of Technology. Downloaded on [DD Month YYYY].
pacts (for example, lets all give each other As whether we contributed or not) are not allowed and will
result in each member of the team being assigned a B for their individual team grade.
Microelectromechanical Devices, Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute
of Technology. Downloaded on [DD Month YYYY].

6777

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

6777

Încărcat de

Drepturi de autor:

Formate disponibile

6.777J/2.

372J: The MEMSclass

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 1

> Class odds and ends

and then, Microfab Part I

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 2

> General Information Handout

Lecturers: Carol Livermore, Joel Voldman

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 3

> Lecture notes

Handed out at beginning of class

> Library Orientation NEXT FRIDAY FEBRUARY 16

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 4

MEMS design and modeling

> Second half

15% Problem sets

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 5

Course conduct and ethics

Cooperation is essential in final design project

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 6

> What makes this course challenging?

> Every student will learn new concepts

Complex open-ended design problems

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 7

> Class odds and ends

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 8

What are MEMS?

> Micro-Electro-Mechanical Systems

Image removed due to copyright restrictions.

Image removed due to copyright restrictions.

JV: 2.372J/6.777J Spring 2007, Lecture 1 - 9

What are MEMS?

> Microfabrication is a manufacturing technology

> But has unique elements

New materials: SU-8, PDMS

Image removed due to copyright restrictions.

JV: 2.372J/6.777J Spring 2007, Lecture 1 -

What are MEMS?

> Definitions vary

Usually made via semiconductor batch fabrication

Actuators: An electrical signal moves something

JV: 2.372J/6.777J Spring 2007, Lecture 1 -

What are MEMS?

> Now, many MEMS have no

Image by MIT OpenCourseWare.

Pulse frequency (Hz)

Liwei Lin (UCB)

JV: 2.372J/6.777J Spring 2007, Lecture 1 -

MEMS: Starting points

> Some starting points:

1961 first silicon pressure sensor (Kulite)

Dr. Kurtz (founder) is MIT graduate, of course

Mid 60s: Westinghouse Resonant Gate Transistor

The Resonant Gate Transistor,

Figure 1 on page 119 in: Nathanson, H. C., W. E. Newell, R. A. Wickstrom,

JV: 2.372J/6.777J Spring 2007, Lecture 1 -

MEMS: Important early work

> Stanford Gas

WAY ahead of its time

> 70s to today: Ken Wise

Figure 3 on page 1882 in: Terry, S. C., J. H. Jerman, and J. B.

(Michigan) neural probes

> 70s Inkjet printheads