Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp.

1-15

1


S
S
S
e
e
e
n
n
n
s
s
s
o
o
o
r
r
r
s
s
s


&
&
&


T
T
T
r
r
r
a
a
a
n
n
n
s
s
s
d
d
d
u
u
u
c
c
c
e
e
e
r
r
r
s
s
s


ISSN 1726-5479
2010 by IFSA
http://www.sensorsportal.com







A Holistic Approach to Automated Synthesis
of Mixed-technology Digital MEMS Sensors
Part 1: Layout Synthesis of MEMS Component with Distributed
Mechanical Dynamics

Chenxu ZHAO and Tom J. KAZMIERSKI
School of Electronics and Computer Science
University of Southampton, UK
E-mail: cz05r@ecs.soton.ac.uk, tjk@ecs.soton.ac.uk


Received: 13 May 2010 /Accepted: 21 June 2010 /Published: 25 June 2010


Abstract: This contribution presents a novel, holistic methodology for automated optimal layout
synthesis of MEMS systems embedded in electronic control circuitry from user-defined high-level
performance specifications and design constraints. The proposed approach is based on simulation-based
optimization where the genetic-based synthesis of both mechanical layouts and associated electronic
control loops is coupled with calculations of optimal design parameters. The underlying MEMS models
include distributed mechanical dynamics described by partial differential equations to enable accurate
performance prediction of critical mechanical components. The proposed genetic-based synthesis
technique has been implemented in SystemC-A and named SystemC-AGNES. A practical case study of
an automated design of a capacitive MEMS accelerometer with Sigma-Delta control demonstrates the
operation of the SystemC-AGNES platform. This Part 1 of the paper focuses on the layout synthesis of
mechanical components, while the full synthesis methodology including automated and optimal
electronic control loop synthesis is outlined in Part 2. Copyright 2010 IFSA.

Keywords: MEMS accelerometer, Synthesis, Sigma-Delta control, SystemC-A



1. Introduction

This two-part paper presents an effective holistic genetic-based synthesis flow (SystemC-AGNES)
applied to automated layout synthesis of mechanical components of Micro-Electro Mechanical Systems
(MEMS) and configuration synthesis of associated electronic control system. Part 1 of this paper focuses
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

2
on the layout synthesis of mechanical sensing component and automated configuration synthesis of the
control system is demonstrated in Part 2 [1].

MEMS are currently used in a wide range of applications due to their significant advantages such as low
cost, small form factor and low power consumption. MEMS sensors, for example accelerometers and
gyroscopes, are widely used in consumer applications, mainly by the automotive industry, in
mixed-technology control designs such as safety air cushion, active suspension or anti-lock brake
system. Modern high precision inertial navigation and guidance systems are also based upon MEMS
sensors embedded in mixed-technology control loops [2].

The design of a typical MEMS system requires an integration of elements from two or more disparate
physical domains: mechanical (translational, rotational, hydraulic), electrical, magnetic, thermal etc.
Different parts of a MEMS system are traditionally designed separately using different methodologies
and different tools applied to different energy domains. Two engineering teams traditionally collaborate
to create a MEMS-based IC: one using 3-D CAD such as CoventorWare to create the MEMS
mechanical model, and the other team, meanwhile, using an EDA tool from such companies such as
Cadence to create the associated ICs. Although this approach provides accurate behaviour simulation of
MEMS devices with their associated electronics, it requires multiple tools and it is difficult to provide IC
designers with an automated synthesis and performance optimization system. This difficulty is primarily
caused by disparities between the different tools and the inconvenience of generating new MEMS
macromodels, when the MEMS layout changes, for incorporation into the IC simulations performed at
the IC design stage.

Analogue and Mixed-Signal(AMS) Hardware Description Languages(HDLs) such as VHDL-AMS
which was standardized by the IEEE in 1999 [3] and later equipped with another IEEE standard for
multiple energy domain packages [4] or Verilog-AMS [5] are able to integrate components from
different energy domains into a single model. However, automated design methodologies for the whole
integrated system supporting mixed physical domains are still lagging. This is mainly due to the fact that
state-of-the-art tools supporting AMS HDLs. such as the commonly used SystemVision from Mentor
Graphics [6] are not designed to support simulation-based synthesis and optimization where users would
be able to develop and implement complex numerical algorithms. Wang proposed a methodology to
realize a genetic optimization algorithm (GA) in a VHDL-AMS testbench [7], but the software tools
used took about 16 hours to fulfill a simple task.

Usually, the design of a MEMS system requires a significant amount of specialist human resources and
time in the iterative trial-and-error design process to determine the crucial trade-offs in meeting the
performance specifications. Therefore there is an increasing need for automated synthesis techniques
that would shorten the development cycle and facilitate the generation of optimal configurations for a
given set of performance and constraint guidelines. Some methodologies have already been proposed for
automated synthesis of mechanical parts in MEMS systems [8-13]. For example, Tamal presented a
method for rapid layout synthesis of a lateral surface-micromachined accelerometer from high-level
functional specifications and design constraints [9]. The design problem is regarded as a nonlinear
optimization problem. Standard off-the-shelf solvers (NPSOL) and a grid-based numerical optimization
algorithm are used to maximize the system's performance. In another approach, Zhun and Wang
presented a hierarchical evolutionary synthesis of MEMS device layout [12]. They divided the design
into two levels: system-level which uses behavioral macromodels and detailed physical-level based on
geometric layout models. At the system level, a combination of genetic programming and bond graphs
are used to generate and search for design candidates satisfying design specifications. At the physical
layout level, optimizations are carried out to meet more detailed design objectives.

In the above approaches the automated design of MEMS is accomplished either by simulation-based
optimization or formulating the design requirements as a numerical nonlinear constrained optimization
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

3
problem, and solved with powerful optimization techniques. However, these methodologies are
constrained to the layout synthesis of a mechanical MEMS device. The salient feature of our approach
proposed here is that it realizes an automated design of a whole MEMS system which contains not only
layout synthesis of the sensing elements but also optimal configuration synthesis for associated
electronics. Synthesis of the electronic control loop is demonstrated in Part 2 of this paper [1]. The
proposed approach integrates a MEMS component library, an electronic control loop library, an efficient
fast MEMS simulation engine implemented in SystemC-A [14] and an evolutionary computation
method (GA).

SystemC-A is a superset of SystemC developed to extend modeling capabilities of SystemC to the
analogue and mixed physical domain [14]. In addition to standard digital modelling capabilities of
SystemC, SystemC-A provides constructs to support user-defined ordinary differential and algebraic
equations (ODAEs), analogue system variables, and analogue components to enable modeling of
analogue and mixed-signal systems from very high levels of abstraction down to the circuit level.
Support for digital-analogue interfaces is also provided for smooth integration of digital and analogue
parts. The analogue simulator uses efficient linear and nonlinear solvers to assure accurate and fast
simulations of the analogue model. Most of the powerful features of VHDL-AMS and Verilog-AMS are
provided in SystemC-A in addition to a number of extra advantages such as high simulation speed,
support for hardware-software co-design and for high levels of modeling. However, the current
SystemC-A can only describe analogue systems by using ODAEs. In modern mixed-domain
applications this limits accurate modelling of system blocks, especially in the mechanical domain, which
exhibit distributed physical effects described by Partial Differential Equations (PDEs). To ensure
accurate modeling of distributed components, Finite Difference Approximation (FDA) approach is
applied to convert PDEs to a series of ODAEs which can be solved by SystemC-A. Distributed
mechanical modelling is important in MEMS designs with digital control because dynamics of
mechanical components may severely affect the system's performance. For example, it has been well
documented that sense fingers in lateral capacitive accelerometers may vibrate due to their own
dynamics, thus rendering the feedback excitation ineffective, causing an incorrect output and a failure of
the system [15, 16]. This scenario cannot be reflected by the conventional mass-damper-spring model
based on a 2nd order ordinary differential equation. In our dedicated SystemC-A model, distributed
mechanical dynamics models are implemented through FDA to enable accurate performance prediction
of critical mechanical components embedded in the mixed-technology control loop.

The synthesis technique presented here is applicable to a wide class of digital MEMS sensors with
electronic control. We demonstrate its operation using a capacitive MEMS accelerometer in a
Sigma-Delta control loop [17, 18] as a case study. The capacitive digital MEMS accelerometers are
notoriously difficult to design using traditional methods because here the mechanical element forms an
integral part of the control loop. This feature makes a separation of the two technology domains in
the design process very effortful.


2. Genetic-based Synthesis of MEMS Sensors with Electronic Control Loop

The proposed automated synthesis approach explores the design according to user defined specifications
and optimizes the structural parameters of the mechanical MEMS elements and the associated electronic
control loop parameters. The automated optimal synthesis flow is shown in Fig. 1. After specifying the
design objectives and constraints, such as the die area of the sensing element and feedback voltage in the
electronic control loop, available components in the MEMS primitive library and the electronic control
loop primitive library are combined automatically to form a valid initial design set. This set of initial
designs is loaded into the synthesis module after parameter initialization and encoding phase. The
synthesis module uses a genetic algorithm to create new MEMS structures and optimizes their
parameters for best performance. Our approach integrates mixed-technology models into a single
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

4
simulation engine which could be easily invoked from various optimization loops. Unlike traditional
MEMS design tool sets, this approach avoids a generation of macromodels in order to realize co-design
and co-simulation.




Fig. 1. Automated synthesis flow in SystemC-AGNES.


2.1. Synthesis Initialization

The two libraries, MEMS primitive library and electronic control loop primitive library, contain typical
components that are widely used in practical MEMS designs. Every member in the libraries is a data
structure record which includes its type code, geometrical parameters for MEMS primitives,
system-level design parameters for electronic primitives and constraints. This Part 1 of the paper focuses
on layout synthesis of the mechanical component while the configuration of the associated electronic
control loop is fixed.


2.1.1. MEMS Primitive Library

The mechanical part of a lateral capacitive MEMS accelerometer is composed of a proof mass, springs
and comb fingers. In the lateral capacitive structure, the proof mass is suspended by springs and it is
equipped with sense and force comb fingers which are placed between fixed fingers to form a capacitive
bridge. The sense fingers moves with the proof mass resulting in a differential imbalance in capacitance
which is measured. The electrostatic force acting on the force fingers is used as the feedback signal to
pull the proof mass in the desired direction. The available mechanical components in the MEMS
primitive library (Fig. 2) are discussed below.

1. Springs: 4 typical springs are available in the MEMS components library for this case study: classic
serpentine spring, rotated serpentine spring, folded spring and spring beam. The layout and geometrical
parameters with constraints are shown in Fig. 2.

2. Proof mass: The proof mass contains unit squires with etch holes for release. The number of these
holes is determined by the size of proof mass and size of holes. There are 4 connecting nodes and
2 connecting sides on the proof mass, 4 connecting nodes is used to connect springs and 2 connecting
sides are used for comb sense and force fingers connection.

Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

5
3. Comb fingers: The sensing element dynamics in the sense-direction is normally modeled to reflect
only one resonant mode by a lumped mass, spring, and damper, which is represented by a simple 2
nd

order ordinary differential equation:


2
2
( ) ( )
( ) ( )
in
d z t dz t
M D Kz t Ma t
dt dt
+ + = , (1)

where M is the total mass of the structure, D and K are damping and spring coefficients correspondingly.
( ) z t is the deflecttrion of the proof mass and ( )
in
a t is the input acceleration.


m m
m m
m m
m m
m
m
m
m
m
m
m m
m m
m
m m
m m
m m
m m
m m
m
m
m
m m
m
m m
m m
m m
m m
m
m m
m m
m m
m m
m m
m m
m m
m


Fig. 2. MEMS primitive component library.
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

6
In reality, the sensing fingers in a lateral structure are distributed elements with many resonant modes.
As their dynamics affect the performance of a control system, the motion of the sense beam should
be distributed, for example using the following partial differential equation [16]:


2 5 4
2 4 4
( , ) ( , ) ( , )
( , ) ( )
D in
y x t y x t y x t
A C I EI Fe x t Aa t
t x t x

c c c
+ + = +
c c c c
, (2)

where y(x, t), a function of time and position, represents the beam deflection, E, I,
D
C , , A are physical
properties of the beam: is the material density, A is the cross sectional area (Wf * T), where Wf and T
are width and thickness of the beam, E is Young's modulus and I is the second moment of area and
D
C is
the internal damping modulus. The product EI is usually regarded as the stiffness. Fe(x, t) is the
distributed electrostatic force along the finger.

The boundary conditions at the clamped end and the free end are shown in the following equations. At
the clamped end (x=0):

(0, ) ( ) y t z t =
(3)

(0, )
0
y t
x
u
c
= =
c
(4)

and at the free end (x=l):


2
2
( , )
0
y l t
M
x
c
= =
c
(5)


3
3
( , )
0
y l t
Q
x
c
= =
c
, (6)

where u , M and Q denote the slope angle, the bending moment and the shear force respectively and l is
the finger length.

The total distributed sense capacitances between the sense fingers and electrodes are:


1
0
0
1
( )
( , )
l
s
C t N T dx
d y x t
c =

}

(7)


2
0
0
1
( )
( , )
l
s
C t N T dx
d y x t
c =
+
}
,
(8)

where Ns is the number of sense fingers. The output voltage can be calculated as:


1 2
1 2
( ) ( )
out m
C C
V t V t
C C

=
+
,
(9)

where Vm(t) is high frequency carrier voltage applied on the fixed electrode in comb fingers unit.

Here the Finite Difference Approximation (FDA) is applied to convert PDEs to a series of ODAEs. If the
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

7
beam is divided into N segments and the deflection of the beam is discretized as:

( ) ( , )
n
y t y n x t = A n=1,2,3... N
(10)

The first order spatial derivatives can be approximated by finite differences:


1
( ) ( ) ( )
n n n
y t y t y t
x x

c
=
c A
(11)

Similar approximation can be applied to higher order spatial derivatives. Eq. (2) is hence transformed to
a system of the following ODAEs:


2
2 1 1 2
2 4
( ) ( ) ( ) ( ) ( ) ( )
( 4 6 4 )
( )
n n n n n n D
d y t dy t dy t dy t dy t dy t C I
A
t x t t t t t

+ +
+ + +
c A c c c c c



2 1 1 2 4
( )
( ( ) 4 ( ) 6 ( ) 4 ( ) ( )) ( )
( )
n
n n n n n in
Fe t EI
y t y t y t y t y t Aa t
x x

+ +
+ + + = +
A A
; n=3,4,5...N-2
(12)


1
( ) ( ) y t z t = ; n=1
(13)


2
3 2 4 2 1
2 4
( ) ( ) ( ) ( ) ( )
( 4 6 3 )
( )
D
dy t d y t C I dy t dy t dy t
A
t x t t t t
+ +
c A c c c c



2
4 3 2 1 4
( )
( ( ) 4 ( ) 6 ( ) 3 ( )) ( )
( )
in
Fe t EI
y t y t y t y t Aa t
x x
+ + = +
A A
; n=2
(14)


2
1 1 2 3
2 4
( ) ( ) ( ) ( ) ( )
( 2 5 4 )
( )
N N N N N D
d y t dy t dy t dy t dy t C I
A
t x t t t t

+ + +
c A c c c c



1
1 2 3 4
( )
( 2 ( ) 5 ( ) 4 ( ) ( )) ( )
( )
N
N N N N in
Fe t EI
y t y t y t y t Aa t
x x

+ + + = +
A A
; n=N-1
(15)



2
1 2
2 4
( ) ( ) ( ) ( )
( 2 )
( )
N N N N D
d y t dy t dy t dy t C I
A
t x t t t

+ +
c A c c c



1 2 4
( )
( ( ) 2 ( ) ( )) ( )
( )
N
N N N in
Fe t EI
y t y t y t Aa t
x x


+ + = +
A A
; n=N
(16)

Boundary conditions provide additional equations. The slope angle at the fixed end is approximated as:


1 0 1
( ) ( ) ( )
0
y t y t y t
x x
u
c
= = =
c A
(17)

and the bending moment M and shear force Q at the free end as:

Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

8

2
1 1
2 2
( ) ( ) 2 ( ) ( )
0
N N N N
y t y t y t y t
M
x x
+
c +
= = =
c A
(18)


3
2 1 1
3 3
( ) ( ) 3 ( ) 3 ( ) ( )
0
N N N N N
y t y t y t y t y t
Q
x x
+ +
c +
= = =
c A
(19)

Eq. (13) represents the motion of the clamped end of the sense fingers (
1
( ) y t ) which moves with the
lumped proof mass whose deflection z(t) is obtained from the solution of Eq.(1).


2.1.2. Electronic Control Loop

High-performance MEMS sensors exploit the advantages of closed-loop control strategy to increase the
dynamic range, linearity, and bandwidth of sensor. In particular, digital modulators for closed-loop
feedback control schemes, whose output is digital in the form of pulse-density-modulated bitstream,
have become very attractive in a number of MEMS applications [17-19]. A conventional 2
nd
order
electromechanical control systems is shown in Fig. 3. In this configuration, mechanical sensing
element is used as a loop filter to form the 2
nd
order electromechanical modulator. Vf1 and Vf2 are the
feedback voltages obtained from the DAC and Vm(t) is a high frequency modulation carrier voltage. The
gain Kcv represents the signal pick-off from differential change in capacitance to voltage and K is the
gain of the voltage booster amplifier following the pick-off stage. The lead compensator is used to
ensure the stability of the control loop. It is an optional component in electronic control primitive library
depending on whether the sensing element is over damped or under damped. A clocked 1-bit quantizer is
used for oversampling and generating a pulse-density modulated digital output signal. However, the
equivalent DC gain of the mechanical integrator in the 2
nd
-order electromechanical modulator is
relatively low and this leads to a poor signal-to-noise ratio (SNR). To improve the SNR, the mechanical
element can be cascaded with additional electronic integrators to form high order topologies [17]. The
example of automated synthesis discussed in this section focuses on the synthesis of MEMS layout and
the electromechanical accelerometer is fixed and of 2
nd
order. Full synthesis which includes both
MEMS layout and electronic control loop is presented in Part 2 [1].




Fig. 3. 2
nd
order electromechanical modulator.


2.1.3. Parameter Initialization and Encoding

The automated design process starts with a specification of the design objectives and constraints.
Drawing from the MEMS primitive component library and electronic control loop library, a set of
configurations is automatically selected (parents of first generation in GA) and loaded into the synthesis
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

9
module. These feasible configurations not only contain MEMS mechanical layouts but also associated
electronic control system topologies. Fig. 4 and Table 1 show an example of a feasible configuration to
illustrate the parameter initialization and encoding phase. This MEMS accelerometer here contains 4
spring beams, 14 force fingers, 20 sense fingers and a proof mass with associated 2
nd
order control
loop. The component code of each component is shown in the Fig. 2. Then the geometrical layout
parameters of mechanical part and the associated system-level design parameters of electrical control
systems are generated to describe the feasible layouts combining with the component code (Fig. 4).




Fig. 4. Example of Parameter Initialization and Encoding.


Table 1. Representation of a population member in GA for the MEMS accelerometer example.

MEMS component library Code Description
Spring 2 Beam spring
Proof mass 5 Proof mass with etching holes
Comb drive 6 Sense and force fingers
Electronic Control loop library Code Description
Sigma-Delta Control system 1(fixed) 2
nd
order Sigma-Delta Control


2.2. Genetic Approach to Synthesis

Genetic Algorithm (GA) has been selected for our case studies as it is a very popular and well tested
optimization algorithm which has demonstrated good performance in a wide variety of complex global
optimization problems where modelling difficulties arise and there is no obvious way to find optimal
solutions [20]. It has already been used for mechanical layout optimization [21].

The optimization problem is considered as a constrained optimization as both of the design and
performance parameters are bound by inequality constraints that must be met:

Maximize: ( ) F x
(20)

Subject to:


_ _
[ , ]
n n low n high
x V V e , n=1,2,3,
(21)

where F(x) is the fitness function to be optimized with design parameter vector x,
n
x represents the nth
design parameter, Vn_low and Vn_high are the lower and upper constraints of the nth design parameter.

Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

10
Performance figures of the candidate designs are evaluated by a fitness function that rates the solutions
according to their performance parameters. Fitness function is usually constructed in a weighted scalar
error form:


( )
R
F x w
R
=
'
,
(22)

where w is the weight coefficient, R is the system performance measure obtained from each simulation
while R
'
is the designer specified objective value. In this case study, w is equal to 1 if all user defined
performance constraints are met, otherwise w is set to 0.0001. If minimization of a fitness parameter is
required, e.g. the sensing element area, w is set to -1 if performance constraints are met or -10 otherwise.
In the case study discussed below, a performance evaluation engine is added to the simulator to enable
measurements of the power spectrum density (PSD) and signal-to-noise ratio (SNR), as the design
objectives, through FFT of the output bitstream. The die area of and static sensitivity of the mechanical
sensing element are also used as system performance objectives or constraints.

After the synthesis initialization, the classical genetic operations of selection, crossover and mutation are
applied to the current generation parents in order to create a new generation. In the selection operation,
designs with better performance (higher fitness) are retained. After the selection, if the crossover
operation is triggered, i.e. crossover probability is higher than a fixed threshold, new MEMS layouts are
composed from primitives and associated control systems by exchanging elements of randomly selected
parents, such as mechanical springs and electronic control blocks. Details of an example of crossover
operation in mechanical sensing element synthesis are illustrated in Fig. 5. As shown in the figure, in this
example only the crossover probability of the spring component is higher than the trigger probability of
70%, so the spring components of parents A and B exchange leaving the other components unchanged in
the creation of new offspring.




Fig. 5. An example of crossover operation in mechanical layout synthesis.


For each individual in the new generation, the genes in their chromosomes have a fixed probability to
mutate at random positions. The mutation operation for the mechanical sensing element is illustrated by
the example shown in Fig. 6 and Table 2. The mutation operation contains two phases: component
mutation and components parameter mutation. In the first phase, if the mutation probability for the
components is higher than the fixed trigger (50 % in this example) such as the beam spring and comb
fingers, new components are automatically composed using the MEMS primitive library and each
parameter of the mutated components gets a random value within its specified range. If there is no
mutation in the first phase, the mutation probability of each component parameter will be compared with
the trigger (60 % in the example) to decide whether this parameter should mutate. As shown in Fig. 6,
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

11
after the mutation the beam spring mutated to a folded spring and comb fingers mutated to themselves
but with different parameters such as a shorter length and a higher number of force fingers. For the proof
mass, only the number of holes and width were changed at the second mutation phase.

This evolution process finishes when the generation number exceeds the specified maximum number.
The optimal solution within a given generation is that with the highest fitness.




Fig. 6. An example of mutation operation in mechanical layout synthesis.


Table 2. An example of mutation operation in mechanical layout synthesis.

MEMS
component
MEMS component
Mutation
probability
(trigger 50 %)
Component
parameters
Component parameters
Mutation probability
(trigger 60 %)
Beam Spring 56 % Spring mutation
Each parameter of the
mutated spring get random
value within range
Lm: length of proof mass 55 % No mutation for length
Wm: Width of proof
mass
70 % Mutation
Wh: Size of holes 23 % No Mutation
Ns: Number of holes 92 % Mutation
Proof mass with
etching holes
30 % No mutation for
proof mass
T: Thickness of proof
mass
10% No mutation
Comb fingers
73 % Comb fingers
mutation

Each parameter of the
mutated comb sense and force
fingers such as Lf, d0 and Ns
get random value within
range


3. Synthesis Verification to Provide Appropriate Performance Metrics for the
Synthesized MEMS Geometries

The practical operation of the proposed synthesis flow for the accelerometer embedded in a conventional
control loop is demonstrated below by three experiments listed in Table 3. In the first experiment, the
system is optimized for maximum SNR with performance constraints, and in the second and third
experiments - for maximum static sensitivity and minimum area respectively.

As can be seen from the results presented in Table 3 and Fig. 8 for each experiment, the genetic synthesis
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

12
algorithm composed different layout structures and produced different performance parameters. Kx, Kz
are the stiffnesses in the axes X and Z correspondingly, and Ky is the stiffness in the sensing axis Y in
this case study. The larger the stiffness ratios Kx/Ky, and Kz/Ky the larger the relative movement of the
accelerometer along the sensing axis.

The synthesis process was carried out using the following design parameters:
1) Oversampling ratio: OSR=128
2) Bandwidth: 512 Hz
3) Oversampling frequency: fs=
17
2 Hz
4) Input force: 100 Hz acceleration with 1g amplitude


Table 3. Summary of synthesis experiments.

Design
objective
Performance constrants Synthesized
layout
SNR
(dB)
Static
sensitivity
Kzy Kzy

Area
(
2
m )
1 Maximum
SNR
SNR>30dB
Area<2.0e-7
2
m
Static sensitivity>1fF/G
Fig. 8. (a) 39.8 1.8fF/G 202 1.875 1.82e-7
2 Maximum
Static
sensitivity
SNR>30dB
Static sensitivity>2fF/G
Fig. 8. (b) 32.9 4.77fF/G 4.91 10.72 3.78e-7
3 Minimum
area of
mechanical
sensing
element
SNR>30dB
Area<1.5e-7
2
m
Fig. 8. (c) 31.5 0.27fF/G 10346 1.95 1.07e-7


Design of MEMS accelerometer in a force feedback control loop contains many crucial trade-offs.
For example, in lateral structure, static sensitivity is dependent on the length and number of the sense
fingers. However, the performance of modulation may be severely affected by the length of sense
fingers to the extent that a complete failure of the control may occur when the fingers are too long.
The maximum number of fingers is also limited by the length of proof mass. To maintain the same
resonant frequency, the finger width should be reduced if the length of the proof mass increases. This
results in a sensitivity decrease. The presented genetic-based synthesis approach deals with these
trade-offs effectively for a given choice of the design objectives.

The fitness improvement during the synthesis flow is shown in Fig. 7. The synthesized mechanical
layouts and parameters of its associated electronic control system are shown in Fig. 8 and Table 4. The
synthesized accelerometer in experiment 1 does not need compensator to assure the stability of the
control loop as the mechanical sensing element is over damped system. As expected, the structure
optimized for maximum sensitivity has more and longer sense fingers. Area optimized accelerometer in
experiment 3 shows a great area improvement over other experiments. The control loop is fixed in this
case study to form a conventional 2nd order electromechanical accelerometer. However, the noise
floor in higher order accelerometer can be reduced drastically leading to great improvement of the
SNR comparing with 2nd order accelerometer. It is discussed in Part 2 of the paper [1] where the
higher order control system is automated optimal synthesized with layout synthesis of mechanical
sensing element simultaneously.


Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

13
0 20 40 60 80 100
0
0.5
1
1.5
2
2.5
Generation
F
i
t
n
e
s
s
(b) Experiment 2: Maximum static sensitivity
Optimized result:
SNR=32.9dB
Static Sensitivity=4.77fF/G
Area=3.78e-7m
2
0 20 40 60 80 100
-14
-12
-10
-8
-6
-4
-2
0
Generation
F
i
t
n
e
s
s
(c) Experiment 3: Minimum area
Optimized result:
SNR=31.5dB
Static Sensitivity=0.27fF/G
Area=1.07e-7m
2
0 20 40 60 80 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Generation
F
i
t
n
e
s
s
(a) Experiment 1: Maximum SNR


Optimized result:
SNR=39.8dB
Static Sensitivity=1.8fF/G
Area=1.82e-7m
2

Fig. 7. Fitness improvement between generations.


Table 4. Summary of synthesized results for Experiments 1, 2 and 3.

MEMS components Experiment 1 Experiment 2 Experiment 3
Proof mass Ml = 341 m
Mw = 73 m
T = 2.9 m
Wh = 4.9 m
Nh = 28
Ml = 695 m
Mw = 136 m
T = 2.85 m
Wh = 4.2 m
Nh = 496
Ml = 205 m
Mw = 125 m
T = 2.5 m
Wh = 5.7 m
Nh = 40
Comb fingers Lf = 122 m
Tf = 2.2 m
d0 = 1.0 m
Ns = 42
Nf = 4
Wanchor=4 m
Lf = 183 m
Tf = 2.1 m
d0 = 1.5 m
Ns = 50
Nf = 10
Wanchor=4 m
Lf = 84.6 m
Tf = 2 m
d0 = 1.36 m
Ns = 24
Nf = 8
Wanchor=4 m
Spring (Folded spring)
Lo1 = 218 m
Lo2=255 m
Wo = 2 m
Lp = 11.5 m
Wp = 2.1 m
(Classic serpentine spring)
N = 2
Lo = 182 m Wo = 2.0 m
Lp =4.5 m Wp = 2.6 m
Lroot=45 m
W=5 m
(Beam spring)
Lo =200 m
Wo = 2.0 m

Control system
(2
nd
order Sigma-Delta)
Vf = 0.6V
Vm = 1.2V
K=23.8
Zero=0.1
Pole=12
Vf = 0.94V
Vm = 1.0V
K=4.8
Zero = 0.2
Pole= 13
Vf = 0.72V
Vm = 1.5V
K=9
ZERO = 0.1
POLE= 10


4. Conclusions

Part 1 of this two-part paper presents an effective simulation-based synthesis flow (SystemC-AGNES)
for automated layout synthesis of a MEMS component in a mixed-domain electrical-mechanical design.
Due to the complex nature of the synthesis process, the synthesis algorithm has been implemented in
SystemC-A. This platform is extremely well suited for complex modeling, implementation of
post-processing of simulation results and optimization algorithms [14]. A distributed model of the
mechanical sensing element is developed to ensure accurate behaviour of the MEMS accelerometer
model when embedded in a force feedback control loop. The proposed approach is fully automated
and it effectively deals with the trade-offs in complex digital MEMS sensor design to generate the layout
of the mechanical sensing element according to user defined performance constraints. Synthesis of a full
mixed-technology system which combines the layout synthesis methodology outlined above with the
synthesis of the associated electronic control is discussed in Part 2 [1].
Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

14

100 m
a)
10
1
10
2
10
3
10
4
-140
-120
-100
-80
-60
-40
-20
0
PSD of output bitstream
Frequency [Hz]
P
S
D

[
d
B
]
System Performance
SNR=39.8dB Sensitivity= 1.8e-015F/g area=1.82e-007 Kyx=202 Kzx=1.875 Resonant frequency= 8.2KHz
2
m
100 m
2
m
System Performance
SNR=32.9dB Sensitivity= 4.77e-015F/g area=3.78e-007 Kyx=4.91 Kzx=10.72 Resonant frequency= 4.6KHz
b)
10
1
10
2
10
3
10
4
-140
-120
-100
-80
-60
-40
-20
0
PSD of output bitstream
Frequency [Hz]
P
S
D

[
d
B
]
2
m
System Performance
SNR=31.5dB Sensitivity= 2.7e-016F/g area=1.07e-7 Kyx=10346 Kzx=1.95 Resonant frequency= 7.9KHz
c)
100 m
10
1
10
2
10
3
10
4
-140
-120
-100
-80
-60
-40
-20
0
PSD of output bitstream
Frequency [Hz]
P
S
D

[
d
B
]


Fig. 8. Synthesized results a) and (b: Experiment 1 and 2 (Maximum SNR); c): Experiment 3 (Maximum Static
Sensitivity); d): Experiment 4 (Minimum area of mechanical sensing element).



Sensors & Transducers Journal, Vol. 117, Issue 6, June 2010, pp. 1-15

15
References

[1]. Zhao C. and Kazmierski T. J., Holistic approach to automated synthesis and optimization of
mixed-technology digital MEMS sensors-Part 2: Synthesis of MEMS system with associated control loop,
Sensor & Transducer, Vol. 117, Issue 6, pp. 16-28.
[2]. Kraft M. Micromachined inertial sensors: The state of the art and a look into the future, IMC Measurement
and Control, 33, 6, 2000, pp. 164-168.
[3]. IEEE standard VHDL analog and mixed-signal extensions, Design Automation Standards Committee of the
IEEE Computer Society, 1999.
[4]. IEEE Standard VHDL Analog and Mixed-Signal Extensions-Packages for Multiple Energy Domain Support,
Design Automation Standards Committee of the IEEE Computer Society, 2005.
[5]. Pecheux F., Lallement C., and Vachoux A., VHDL-AMS and Verilog-AMS as alternative hardware
description languages for efficient modeling of multidiscipline systems, IEEE Trans. Computer-Aided
Design of Integrated Circuits and Systems, 24, 2, 2005, pp. 204-225.
[6]. Datasheet. SystemVision for Mechatronic System Modeling, Mentor Graphics Corporation, 2006.
(http://www.mentor.com/products/sm/system_integration_simulation_analysis/systemvision/upload/Syste
mVision datasheet.pdf)
[7]. Wang L. and Kazmierski T. J., VHDL-AMS based genetic optimization of a fuzzy logic controller for
automotive active suspension systems, in Proc. of the Int. Conf. Behacioral Modeling and Simulation
Workshop, BMAS, Sept 2005, pp. 124-127.
[8]. Tran Due Tan, Roy S., Nguyen Phu Thuy, and Huu Tue Huynh, Streamlining the design of MEMS devices:
an acceleration sensor, Circuits and System Magazine, IEEE, 8, 1, 2008, pp. 18-27.
[9]. Mukherjee T., Zhou Y., and Fedder G. K., Automated optimal synthesis of microaccelerometers, Proc. Int.
Conf. Micro Electro Mechanical Systems, MEMS'99, Jan 1999, pp. 326-331.
[10]. Mukherjee T., Zhou Y., and Fedder G. K., Automated optimal synthesis of microresonator, in Proc. of the
Int. Conf. Solid State Sensors and Actuators, 1997., 2, 1997, pp. 1109-1112.
[11]. Mukherjee T. and Fedder G. K., Structured Design of Microelectromechanical Systems, in Proc. of the
Design Automation Conference, 2, Jun 1997, pp. 680-685.
[12]. Fan Z., Goodman E. D., Wang J., Rosenberg R., Kisung Seo, and Hu J., Hierarchical evolutionary synthesis
of MEMS, Congress on Evolutionary Computation, 2004. CEC2004, 2, Jun 2004, pp. 2320-2327.
[13]. Gupta V. and Mukherjee T., Layout synthesis of CMOS MEMS accelerometers, MSM 2000, 2000,
pp. 150-153.
[14]. H. Al-Junaid and T. Kazmierski, An Analogue and Mixed-Signal Extension to SystemC, IEE Proc. Circuits,
Devices and Systems, 152, 6, December 2005, pp. 682-690.
[15]. Seeger J. I., Xuesong J., Kraft M., and Boser B. E., Sense finger dynamics in a force feedback gyroscope,
Tech. Digest of Solid State Sensor and Actuator Workshop, Hilton Head Island, USA., Jun 2000, pp. 296-299.
[16]. Zhao C., Wang L., and Kazmierski T. J., An efficient and accurate MEMS accelerometer model with sense
finger dynamics for applications in mixed-technology control loops, in Proc. of the Int. Conf. Behavioral
Modeling and Simulation Workshop, 2007. BMAS, 2007, Sept. 2007, pp. 143-147.
[17]. Dong Y., Kraft M., Gollasch C., and Redman-White W., A high-performance accelerometer with a
fifth-order Sigma-Delta modulator, Journal of Micromechanics and Microengineering, 2, 2005,
pp. S22-S29.
[18]. Kraft M., Redman-White W., and Mokhtari M. E., Closed loop micromachined sensors with higher order
modulators, in Proc. of the 4
rd
Conf. on Modeling and Simulation of Microsystems, 2001, pp. 104-107.
[19]. Dong Y., Kraft M., Hedenstierna N., and Redman-White W., Microgyroscope control system using a
high-order band-pass continuous-time Sigma-Delta modulator, in Proc. of Transducers 2007, 2, June 2007,
pp. 2533-2536.
[20]. J. Carnahan and R. Sinha, Nature's algorithm [genetic algorithms], IEEE Journal of Potential, 20, 2,
Apr-May 2001, pp. 21-24.
[21]. Y. Zhang, R. Kamalian, A. M. Agogino, and H. S. Carlo, Hierarchical MEMS synthesis and optimization.
SPIE Conference on Smart Structures and Materials, 5763, 12, Mar. 2005.

___________________

2010 Copyright , International Frequency Sensor Association (IFSA). All rights reserved.
(http://www.sensorsportal.com)