Sunteți pe pagina 1din 19

Chapter 1

Introduction

1.1 Thesis Aims

The purpose of this thesis is to characterize a new resonator being developed at The
Charles Stark Draper Laboratory. First, expected theoretical behavior of the resonator
will be summarized, along with a model based on these calculations suitable for use in
filter design. Next, radio frequency (RF) measurement and parameter extraction
procedures will be described, applied to prototype resonator devices, and the results
analyzed. Finally, the possibilities of implementing this resonator in high-performance
bandpass filters with be discussed using several design examples.

1.2 Project Origins and Goals

The Draper resonator is the chief element of a project which seeks to develop an array of
RF channel-select filters made from MEMS resonators and integrated with an RF low
noise amplifier, funded by the Defense Advanced Research Projects Agency (DARPA).
To date, the available integrated filters have performance limits in very narrowband, low
insertion loss applications; the large size of current non-integrated RF resonators and
filters makes a compact, portable filter bank infeasible. Integrated MEMS resonators
offer a promising solution for this difficulty, because of their high Q, mechanical tuning
techniques (e.g. laser trimming), compatibility with conventional silicon active device
processes, and small size at GHz frequencies (on the order of 10 m).

11
The goal of this project was to create resonators with center frequencies ranging from 200
MHz ~ 1.5 GHz, selectable by device geometry, with high enough Q to allow bandwidths
on the order of 100 kHz to 10 MHz. Approximately twenty of these resonators would be
fabricated to form a filter bank onto a 1-10 mm2 chip with a low noise amplifier.

1.3 Chapter Summaries

The remainder of this thesis is organized into five chapters. Chapter 2 expounds upon the
problem of integration between RF filters and the other circuitry required for
communications devices. Chapter 3 describes three major miniature resonator
technologies and their capabilities, and then presents the theoretical characteristics of the
Draper resonator. Chapter 4 summarizes in detail the measurement of prototype Draper
devices and the characterization procedures, which were first validated on known TFR
resonators. Chapter 5 begins with an overview of bandpass filter design, and discusses in
detail three filter topologies: simple ladder, dual-resonator ladder, and full lattice, with
numerical examples using the theoretical Draper resonator parameters given. Next the
manufacturing limits and tolerances of the resonator are discussed, and their effect on
filter characteristics. Given the achieveable limits of the Draper resonator, the filter
design process is described, focusing on the suitability of using the Draper resonator for a
given set of requirements. Finally, several topics concerning filter bank design are
discussed. Chapter 6 provides final conclusions for the thesis work and suggestions for
future efforts.

12
Chapter 2
Background

2.1 Current Applications of Miniature Resonators in RF


Communication Technology

As the number of wireless applications have increased, allocations of the frequency


spectrum have grown increasingly crowded. Newer technologies have therefore pushed
their carrier frequencies higher and higher, which complements the demand for smaller
and more portable devices. Greater portability also gives rise to a demand for lower
power consumption. In general, reducing the total size of a piece of circuitry enables
higher frequency operation while reducing power consumption. In current RF devices, all
the major active components can be integrated monolithically, resulting in a small size
and the best performance due to minimal parasitics. However, the passive resonators and
filters necessary for frequency selection and duplex function have no viable integrated
option. The filters used in commercial wireless products today are on the order of
millimeters in dimension [1, 2], which is still by far the largest single component of
modern RF circuitry.

2.2 Typical Communications Front-End

A typical RF receiving circuit begins with an antenna to receive wideband transmissions,


producing an analog electronic signal. The next major step is to bandpass filter this signal
to extract the appropriate frequency band, with the possibility of an intermediate gain

13
stage depending on the expected signal strength of the antenna signal and the insertion
loss of the filter. After filtering, the signal is amplified and sent to a downmixing element,

to down-
Gain Gain
converter

BPF
to an-
tenna

from
Gain modulator

BPF
Figure 2.1: Block diagram of typical RF circuitry for a wireless communications device.

and from there it is commonly sampled and processed digitally. In a duplex


communications device, there will also be a transmit circuit which takes a signal
modulated to the carrier frequency, amplifies it, bandpass filters it to prevent interference
from being generated in other frequency bands, then broadcasts it via the antenna (Figure
2.1). The various amplifiers can be integrated with standard silicon fabrication
technologies, but the filters to date have not been integrated in commercial devices. They
are frequently built with ceramic transmission-line resonators, or surface acoustic wave
(SAW) resonators, neither of which is compatible with silicon IC technologies [1, 3]. An
integrated filter solution is highly desirable for its reduced size and consequently smaller
parasitics, and to eliminate the packaging parasitics and longer interconnects due to
switching between on-chip and off-chip circuitry, particularly in the cases where an
additional amplifier stage is required between the antenna and filter.

2.3 Requirements for an Integrated Resonator

14
In order for a resonator to be integrated with silicon active devices, its production must be
compatible with silicon IC fabrication processes. However, once this requirement is
satisfied, the resonator must also be able to compete with the best non-integrated
technologies in performance, reliability, and cost before it can be considered a
worthwhile alternative to them. In addition to process compatibility, the resonator
fabrication must either maintain a certain degree of consistency or a suitable tuning
method must be available for it. Since integrated devices will be much smaller than
discrete elements, fabricating them to the same relative tolerances becomes that much
more difficult.

The other major improvement required of integrated resonators is higher device Q. First,
as frequencies grow higher, the transition from passband to stopband of the bandpass
filters must grow sharper to allow the same channel spacing, and the sharpness of this
transition is proportional to Q. For example, if a filter allows a 50 kHz channel spacing at
1 MHz center frequency, then with the same Q at 1 GHz the minimum channel spacing
would be 50 MHz. Most protocols of communication do not require more than a couple
MHz of bandwidth to contain all the information they need, so such large channels are a
very inefficient use of an increasingly valuable commodity. Second, as resonator
dimensions shrink and operational frequencies rise, in general the impedance level of the
resonator increases. At GHz+ frequencies, impedance matching becomes very important
when connecting different circuit elements. In order for high-frequency filters to
effectively couple to other active elements without losing too much signal strength to
parasitics, the resonators should have high Q, since the impedance level at resonance is
inversely proportional to Q, which in turn affects the matching impedances required.
Unfortunately as resonator dimensions decrease, relative loss mechanisms tend to
worsen, making higher Q more difficult to achieve.

15
16
Chapter 3
Miniature Resonators

3.1 Current Miniature Resonator Technologies

Several different types of piezoelectric resonators are under investigation to produce a


viable integratable solution. This section presents a brief overview of the various
competing technologies.

3.1.1 SAW

Surface Acoustic Wave (SAW) resonators are currently one of those most common filter
elements used in commercial RF telecommunications products today, along with ceramic
resonators. They generally are produced by depositing a pair of transducers onto a
smooth piezoelectric substrate. Each transducer consists of thin metal multi-fingered
electrodes interdigitated in a pair; thus there is an input port and an output port. When an
oscillation is applied across the input terminal, the input transducer converts the electrical
signal into acoustic vibrations. A surface wave vibration is generated which propagates in
a direction perpendicular to the electrode fingers, strongly favoring the frequency whose
half-wavelength equals the interdigital spacing. The output transducer converts this
vibration back to an electrical signal which is detected at the output.

The frequency response is determined exclusively by the structure of the electrodes. They
are suitable for use in many applications, but they have one major drawback: high
insertion loss. Each transducer emits energy in both directions, resulting in 6 dB of total

17
loss. The electrodes are very thin and thus have high resistance. Finally, intentional RF
mismatches must be introduced to avoid a rippling phenomenon known as triple-transit
echo. Total insertion losses of SAW filters range from 7 to 30 dB [4].

Conventional SAW resonators are the inline type, where the wave propagates in a straight
line from input to output. These have a typical packaged area of about 15 mm x 6.5 mm
[4]. A smaller type, the Z-path SAW, uses reflectors to bounce the acoustic wave twice to
make a Z-shaped pattern, allowing package dimensions on the order of 5 mm x 5 mm [5].

Author Type Q Size Frequency Year


King, inline SAW - 15.3 x 6.45 210 MHz 1999
Gopani4 mm²*
Franz5 Z-path SAW - 5 x 5 mm²* 210 MHz 1999

Table 3.1: Example SAW resonators listed with important device characteristics.

3.1.2 TFBAR

Thin Film Bulk Acoustic Resonators (TFBARs) are under heavy development by a
number of research groups due to their low-temperature processing methods, which make
integration with a silicon IC process much more likely, though other obstacles have
barred successful integration to date. TFBARs require two basic ingredients to function: a
transduction method to convert between electrical oscillations and acoustic vibrations,
and an acoustic cavity to trap these vibrations. The former is provided by a piezoelectric
membrane, and the latter by a large acoustic impedance mismatch at each interface of the
membrane, so that most acoustic energy hitting the interfaces is reflected. The method of
generating this impedance mismatch defines the type of TFBAR. The more traditional
method creates an air/crystal interface by either etching away the substrate from the
bottom, or using a sacrificial substrate layer right beneath the resonator which is etched
away after the membrane is deposited, leaving a small gap. Alternatively, the resonator
membrane may be deposited onto a stack of quarter-wavelength thick layers of acoustic
materials forming a Bragg reflector [6]. These solidly mounted resonators (SMRs) enjoy

18
greater structural stability than the air/crystal type since the resonating membrane and
electrodes are fully supported from below.

The acoustic cavity contains the piezoelectric membrane sandwiched between two
electrodes. An electrical signal of the right frequency across the electrodes excites a
standing longitudinal wave in the membrane. The major loss mechanism, and limiting
factor for Q, is the coupling at the cavity interfaces with the adjacent air and supporting
substrate. The film is usually about 1-5 m thick, with lateral dimensions anywhere from
100 to 1000 times the thickness [7].

In general, TFBARs surpass SAW resonators in every way, except for process simplicity
and cost to manufacture [8]. Thus, due to this fact and the desire to invent a process fully
compatible with silicon active devices, experimentation on these devices centers on novel
processing methods.

Author Type Q Size Frequency Year


Lakin9 SM-TFBAR 717 - 644 MHz 1999
Lakin10 TFBAR 1090 - 1.6 GHz 2001
Plessky6 SM-TFBAR 641 0.033 mm² 2 GHz 1998
Ruby11 TFBAR 500-1300 ~0.001-0.1 mm² 2-4GHz 1994

Table 3.2: Example TFBAR resonators listed with important device characteristics.

3.1.3 Mechanical Resonators

Mechanically resonant structures such as deflecting cantilevers and flexural beams show
promise as filter elements due to their high Q and possible silicon IC process
compatibility. Their principle of operation is very simple: a simple symmetric structure
with mechanically resonant modes is built out of a piezoelectric material. If an oscillating
force at some frequency is applied to the structure, actuated either mechanically or
piezoelectrically, the structure will vibrate and produce a charge separation proportional
to the amplitude of its vibration. Frequencies near resonance will produce larger
amplitude responses, which can be viewed as electrical signals if electrodes are attached

19
at appropriate points on the structure. When used as a filter element, these types of
resonators are usually one-port devices acting as variable impedances: a voltage signal is
applied to the port and a current waveform is drawn from the source with amplitude
varying with the frequency of the input signal.

A major limiting factor to date with this type of resonator is the loss associated with the
anchor between the resonating element and the substrate. In addition, interfacing
electrically to the resonating element also introduces loss in general. Thus a mechanical
element which could achieve a Q of perhaps 100,000 in isolation may only get one-tenth
that when modified as an electrical resonator. Another drawback is poor linearity with
amplitude of vibration in many cases.

Author Type Q Size Frequency Year


Cleland12 Flexural 21000 @ 3.3 x 2.4 m² 82 MHz 2001
beam 4.2 K
Nguyen 13
Flexural 7450 ~30 x 10 m² 90 MHz 2000
beam

Table 3.3: Example mechanical resonators listed with important device characteristics.

3.2 Draper Resonator Bar

The Draper resonator bar was designed to allow production of arrays of RF


communications filters integrated with silicon active circuitry on a single die. It is a
mechanical resonator designed to operate at the fundamental longitudinal mode. Its range
of dimensions should allow center frequencies from 200 MHz to 1.5 GHz while being
compact enough to fit an entire filter bank into an area on the order of 10 mm².
Preliminary loss analyses predict a device Q of 104 should be attainable [27].

3.2.1 Device Overview

20
Figure 3.1 provides a schematic of the resonator bar with the major geometrical
parameters labeled. The bar is suspended over a 1 m-deep well in the substrate by two
tethers attached to the midpoints of its sides. The bar and the tethers are a single
continuous film of aluminum nitride (AlN). Electrodes cover the top and bottom of the
bar, producing an off-resonance characteristic of a parallel-plate capacitor.

2a

Figure 3.1: Resonator bar schematic. The bar is suspended over an empty well while the
tethers rest on the lower electrode and the substrate.

Figure 3.2: Fundamental longitudinal vibration. One cycle is pictured. The amplitude of
motion is exaggerated to illustrate the motion.

Longitudinal Mode Resonance

21
The primary engineered resonance is a longitudinal vibration where the bar expands and
constricts lengthwise (Figure 3.2). The midpoint of the bar, where the tethers attach, is a
node, while the two ends of the bar experience the greatest amplitude of displacement.
The magnitude of vibration will be on the order of nanometers.

The bar has many different natural modes of resonance with several at frequencies below
that of the intended longitudinal mode of operation. However, due to symmetry and the
placement of the electrodes, the charge contribution of these lower order modes at the
device’s port should cancel to nearly zero. This calculation was confirmed by a finite
elements simulation of the device’s I-V transfer characteristic (Figure 3.3).

Figure 3.3: Finite elements model of the transfer characteristic of a resonator bar with
0.78 GHz longitudinal resonance [14].

3.2.2 Analytic Model of Longitudinal Resonance

By applying the piezoelectric equations of state and standard electrostatics relations to the
resonator bar, an analytic solution for the primary resonance may be obtained. This
solution only models a single resonance, but is immensely helpful during filter design.

A Brief Introduction to Piezoelectric Materials Properties

Piezoelectric properties arise in certain crystals with asymmetric structure. Mechanical


stressing causes a polarization of charge in these materials, with the converse occurring
as well: an applied electrical field induces physical deformation. A right-handed

22
Cartesian set of axes is traditionally introduced to facilitate mathematical analysis of the
various piezoelectric interactions, in accordance with the IEEE Standard on
Piezoelectricity [15]. An electric field in one direction will elicit, in general, a mechanical
response along each of the three axes, and a mechanical stress in one direction will
induce, in general, an electric polarization in each axis as well.

Linear Theory of Piezoelectricity

The piezoelectric equations of state relate the major mechanical variables of interest (e.g.
stress, strain, displacement) to the electric field and charge polarization. These equations
may be linearized by assuming all displacements and vibrational amplitudes are very
small. The various partial derivatives appearing in these equations may now be
considered constants, and define the various material parameters such as stiffness,
permittivity, and the piezoelectric stress and strain constants [16]. In addition, because the
phase velocities of acoustic waves are several orders of magnitude less than the velocities
of electromagnetic waves, quasi-electrostatic conditions are assumed [15]. A more
detailed treatment of the formulation of these equations and the definition of the
mechanical and electrical field variables may be found in [15],[16], and [17].

Longitudinal Mode Analytic Transfer Function

The following analysis was originally performed by Dr. Amy Duwel, and is summarized
here with her permission. Applying the piezoelectric equations of state to AlN results in
the following constitutive equations:

T1   c11 c12 c13 0 0 0   S1  0 0 e31 


T  c    0 e32 
 2  21 c 22 c 23 0 0 0  S 2 
 0
 E1 
T3  c31 c32 c33 0 0 0  S3  0 0 e33   
(3.1)         E2
T4  0 0 0 c 44 0 0  S 4  0 e24 0  
 E3 
T5  0 0 0 0 c55 0  S 5  e15 0 0  
      
T6   0 0 0 0 0 c66   S 6   0 0 0 

23
c= stiffness matrix: N/m2 eT= piezoelectric coupling: C/m2

 S1 
S 
 D1  0 0 0 0 e15 0  2   11 0 0   E1 
D   0 S3 
(3.2)  2  0 0 e24 0 0     0  22 0   E 2 
 
S
 D3  e31 e32 e33 0 0 0  4   0 0  33   E3 
S5 
 
 S 6 

= dielectric constants: F/m

where T is the stress tensor, S is the strain tensor, E is the electric field vector, and D is
the polarization charge vector. The axis along the bar’s length is subscript 1 ( x̂1 = x),
across its width is subscript 2 ( x̂ 2 = y), and vertically through its thickness is subscript 3 (
x̂3 = z). Due to symmetry, T and S only have 6 independent elements, with the reduced

subscript mapping as follows:


(3.3) Tij  Ti for i  j , Tij  T9 i  j for i  j

(3.4) S ij  S i for i  j , 2 S ij  S 9i  j for i  j

The next step is to apply force balance to Eqn. (3.1), substituting in displacement u and
electric potential :

T1   c11 c12 c13 0 0 0   u1,1  0 0 e31 


T  c   u  
 2  21 c 22 c 23 0 0 0  2, 2  0 0 e32 
 ,1 
T3  c31 c32 c33 0 0 0   u 3,3  0 0 e33   
(3.5)           , 2 
T
 4  0 0 0 c 44 0 0  u 2 , 3  u 3, 2   0 e 24 0   
T5  0 0 c55 0   u1,3  u 3,1  e15 0 0  
,3
0 0
      
T6   0 0 0 0 0 c66   u1, 2  u 2,1   0 0 0 
In this notation, any subscripts after the comma refer to derivatives with respect to that
variable. A number of approximations are now made. First, for a longitudinal vibration in
the x-direction, no y-dependence will be assumed, so u2 and all its derivatives are set to
zero. Second, for the lowest-order longitudinal mode, set stress in the z-direction and x-z
shear to be zero for all z, meaning T3 = T5 = 0. Third, the inertia in the z direction is
considered negligible. This gives the following acoustic wave equation for the
fundamental longitudinal mode in the x direction, coupled to :

24
 c2   c e 
(3.6) u1  u1,11 c11  13    ,31 e31  13 33 
 c33   c33 

Solution of the electrostatics equations begins by assuming AlN is non-conducting and


thus its free charge density is 0:
D1 D2 D3
(3.7) D  0   
x1 x 2 x3

Making the same approximations as above and some algebra results in this equation:
 e2   e2   e e c 
(3.8)  ,11  11  15    ,33  33  33   u1,31 e15  e31  15  33 31 
 c55   c33   c55 c33 

which has the form of Laplace’s equation for an anisotropic material. The system is
described by the coupled equations (3.6) and (3.8), which are difficult to solve due to the
x-z coupling. Eqn. (3.8) may be split into two parts, since the system is linear. If we let
   BC  u ,  BC is the solution to (3.8) with the RHS set to zero and the voltage

boundary conditions applied, and  u is the solution to (3.8) with zero voltage boundary
conditions. For  BC , the solution can be approximated as:
z
(3.9)  ( x, z )  f ( x )
2a
under the constraint that the bar is much longer than it is thick. The boundary conditions
are set as follows: at z = 0 (bottom electrode),  = 0, at z = 2a (top electrode),  = f(x). As
a first order approximation, Eqn. (3.6) is solved using only  BC on the RHS [14].
Plugging this solution into Eqn. (3.6) allows calculation of the total equation of motion of
the fundamental longitudinal resonance due to a driving voltage function V(s):
2e  x  V ( s)
(3.10) u1 ( x, s )  cos  2
aL  L  s  s n   n
2

where the following intermediate variables have been defined:


c132
(3.11) c  c11 
c33

c13
(3.12) e  e31  e33
c33

25
n c
(3.13)  n 
L 

(3.14)  n   n / Q
e332
(3.15)  z   33 
c33

Finally, with the complete solution we can integrate D3 over the electrode area, and then
differentiate with respect to time to express the current as a function of voltage:
4we 2 sV ( s )  z wL
(3.16) I   sV ( s )
aL s  s n   n2
2
2a

3.3 Butterworth Van-Dyke Model

R C L

C0

Figure 3.4: The Butterworth Van-Dyke model for a crystal resonator.

The Butterworth Van-Dyke (BVD) model is a common lumped element circuit model
used by crystal filter designers to simplify the transcendental functions that completely
characterize the resonators used as filter elements [19]. It generally provides an accurate
fit for a single resonance plus the other regions of a resonator’s transfer function not near
another resonance, modeling those parts as a capacitance. Qualitatively, the R-L-C branch
determines the “series” resonance, where the impedance drops sharply to a minimum
value of R at the frequency where the series inductance and capacitance cancel each other
out. At some higher frequency, the loop reactance hits zero and causes a “parallel”
resonance where most current will travel around the loop instead of past it.

3.3.1 BVD Impedance

26
The exact transfer function representing the BVD impedance is:
1 s 2 LC  sRC  1
Z BVD 
(3.17) sC 0 C
s 2 LC  sRC  1 
C0

This function can also be expressed directly in terms of the major resonator figures of
merit, series resonance ws, parallel resonance wp, and Q:
1
(3.18) ws 
LC

C
(3.19) w p  ws 1 
C0

1 wL
(3.20) Q   s
ws RC R

ws
s2  s
 ws2
1 Q 1 s 2  s  ws2
(3.21) Z BVD  
sC 0 2 w sC 0 s 2  s  wp2
s  s s  wp2
Q
The first term in this impedance is the impedance of the static capacitance C0. If C << C0,
as is generally true of high-Q resonators, the second fraction is very close to unity except
near the resonant frequencies. At those frequencies, there is a complex zero pair followed
closely by a complex pole pair, which creates a very low impedance peak followed by a
very high impedance peak, as expected.

The only thing preventing the series and parallel resonant impedances from going to zero
and infinity, respectively, is the resistance R. R is a direct result of finite device Q and
represents all the losses in the device. R is related to the reactance of the inductor and
capacitor at resonance by a factor of Q. For high-Q resonators, R is frequently ignored
during filter design, and only causes a slightly larger insertion loss in the passband than
calculated otherwise.

27
160

140
Magnitude (dB)

120

100

80

60

40
780 785 790 795 800 805 810 815 820 825 830 835

100

50
Phase (degrees)

-50

-100
780 785 790 795 800 805 810 815 820 825 830 835
Frequency (MHz)

Figure 3.5: Typical Butterworth Van-dyke impedance function.

3.3.2 Draper Resonator Equivalent Circuit Parameters

The form of Eqn. (3.16) matches exactly with the BVD impedance. By comparing the
two expressions, the equivalent circuit parameters for the Draper Resonator are found to
be:
 z wl
(3.22) C 0 
2a
al
(3.23) L 
4we 2

28
4 we 2 l
(3.24) C 
ac 2
a c
(3.25) R 
Q 4we 2

where the bar dimensions l, w, and 2a are as defined in Figure 3.1, and the other
parameters are material constants for AlN which may be found in the appendix. There are
a couple noteworthy aspects to these relations. First, the resonant frequency depends only
on the length of the bar:
1  c
(3.26) ws  
LC l 

Second, the ratio of the motional to the static capacitance (termed r) is a constant with
respect to geometry:
C 8e 2 1
(3.27) r   2   3.22%
C 0 c  z 31.04

29

S-ar putea să vă placă și