High Performance Piezoelectric MEMS

Microphones

by

Robert John Littrell

A dissertation submitted in partial fulfillment

of the requirements for the degree of
Doctor of Philosophy
(Mechanical Engineering)
in The University of Michigan
2010

Doctoral Committee:
Professor Karl Grosh, Chair
Professor David R. Dowling
Professor Khalil Najafi
Assistant Professor Kenn Richard Oldham
 TABLE OF CONTENTS

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

LIST OF ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . viii

LIST OF SYMBOLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

CHAPTER

I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Capacitive and Piezoelectric MEMS Microphones . . . . . . . 1

1.2 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Piezoelectric Materials . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Capacitive Microphone Embodiments . . . . . . . . . . . . . 11
1.7 Piezoelectric Microphone Embodiments . . . . . . . . . . . . 12
1.8 Gaps in the Literature . . . . . . . . . . . . . . . . . . . . . . 13

II. Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.1 Microphone Transducer Noise . . . . . . . . . . . . 16
2.1.2 Microphone Circuitry Noise . . . . . . . . . . . . . . 22
2.2 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.1 Cantilever Transducer Sensitivity . . . . . . . . . . 25
2.2.2 Diaphragm Transducer Sensitivity . . . . . . . . . . 31
2.3 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.1 THD of a Common Source Amplifier . . . . . . . . 36
2.3.2 THD Measurement Resonator . . . . . . . . . . . . 37
2.4 Combining Piezoelectric Elements . . . . . . . . . . . . . . . 38
2.5 Microphone Optimization . . . . . . . . . . . . . . . . . . . . 41

ii
 2.5.1 Cantilever Transducer . . . . . . . . . . . . . . . . . 42
2.5.2 Diaphragm Transducer . . . . . . . . . . . . . . . . 52
2.6 Packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7 Vibration Sensitivity . . . . . . . . . . . . . . . . . . . . . . . 57

III. First Generation Device . . . . . . . . . . . . . . . . . . . . . . . 59

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

IV. Second Generation Device . . . . . . . . . . . . . . . . . . . . . . 68

4.1 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.1 Permittivity and Loss Angle . . . . . . . . . . . . . 75
4.2.2 Piezoelectric Coupling Coefficient (d31 ) . . . . . . . 76
4.2.3 Sensitivity and Noise Floor . . . . . . . . . . . . . . 77
4.2.4 Linearity . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

V. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5.1 Demonstrated Performance . . . . . . . . . . . . . . . . . . . 83

5.2 Further Performance Improvements . . . . . . . . . . . . . . . 85
5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

iii
 LIST OF FIGURES

Figure

1.1 Capacitive and piezoelectric microphone noise levels as of December

2009. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.1 Electrical equivalent circuit of piezoelectric sensing element. . . . . . 18

2.2 Diagram of electrode/dielectric/electrode stack modeled as a contin-

uous system and associated boundary conditions. . . . . . . . . . . 19

2.3 The voltage in the material stack varies from the applied voltage (v0 )
at the base of the electrode to half of that value at the center of the
stack. Only the top half of the stack is shown due to symmetry. . . 21

2.4 Common source amplifier circuit schematic. . . . . . . . . . . . . . . 23

2.5 Small signal model of the common source amplifier with noise sources. 23

2.6 Output noise of common source amplifier. Plot includes noise from
the bias resistor, shot noise, channel noise, and noise from the load
resistor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.7 Piezoelectric laminate beam layer numbering and orientation. . . . . 25

2.8 Tensioned Diaphragm with assumed boundary conditions. . . . . . . 32

2.9 System of resonant tubes used to measure total harmonic distortion. 38

2.10 Modeled frequency response of two-tube system. Harmonics of 500

Hz are attenuated. . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.11 Effect of piezoelectric noise when sensing elements are combined in

series and parallel. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

iv
2.12 As the beams bend due to residual film stress, the gap around the
beams increases, reducing the acoustic resistance of the microphone
and decreasing sensitivity. . . . . . . . . . . . . . . . . . . . . . . . 43

2.13 Theoretical output energy per unit sensor area. Thinner electrodes
and thinner piezoelectric layers increase the output energy per sensor
area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.14 Material properties of Mo and AlN as a function of layer thickness. 47

2.15 Optimization parameter, Ψ, as a function of electrode and piezoelec-

tric layer thicknesses. The model gives optimal thicknesses around
1.5 µm AlN and 10 nm Mo. . . . . . . . . . . . . . . . . . . . . . . 48

2.16 Theoretical effect of electrode length on normalized piezoelectric out-

put energy. Decreasing the electrode length to around 45% increases
the output energy by 54%. . . . . . . . . . . . . . . . . . . . . . . . 49

2.17 Models indicate that even small amounts of diaphragm stress sig-
nificantly degrade the normalized output energy of a diaphragm of
constant bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.18 Two common embodiments of MEMS microphone packages. . . . . 54

2.19 Package and device equivalent circuit. Includes any front cavity res-
onances, device response, and back cavity compliance. . . . . . . . . 54

3.1 Four step beam fabrication process. . . . . . . . . . . . . . . . . . . 61

3.2 Microscope photograph showing the top view of twenty piezoelec-

tric beams. These beams have five material layers on them and are
released from the Si substrate. . . . . . . . . . . . . . . . . . . . . . 61

3.3 Packaged microphone wire bonded to JFET in TO can. . . . . . . . 62

3.4 Beam actuation displacement measurements and quadratic fit used

to calculate d31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.5 Test set-up used for measuring beam deflection and output voltage
resulting from an applied acoustic pressure. . . . . . . . . . . . . . . 64

3.6 Beam deflection profile resulting from a uniform pressure load. The
laser measurements towards the base of the beam were obstructed by
TO can. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

v
3.7 Amplitude and phase of the frequency response of the beams resulting
from a uniform pressure load. . . . . . . . . . . . . . . . . . . . . . 65

3.8 Measured and modeled output referred noise of common source am-
plifier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

4.1 Piezoelectric MEMS microphone fabrication process. This is a 7 mask

process without boron implantation and a 9 mask process with boron
implantation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.2 After fabrication is complete, the released beams appear as shown.

The large compressive stress in the first electrode layer caused the
beams to bend upward and are slightly out of focus. . . . . . . . . . 73

4.3 Several different PCB designs were fabricated for use with a variety of
buffering techniques. Some designs have a hole behind the transducer
in order to provide a larger back cavity volume. . . . . . . . . . . . 73

4.4 Packaged microphone transducer and JFET for signal buffering. There
is no cap on this package to avoid any package resonance in the fre-
quency response measurement. . . . . . . . . . . . . . . . . . . . . . 74

4.5 Four test capacitors. The capacitors on the left side of the figure are
used to measure the bottom AlN film while those on the right side
are used to measure the top AlN film . . . . . . . . . . . . . . . . . 75

4.6 Sensitivity of a piezoelectric MEMS microphone with a 18.4 kHz res-

onant frequency. The increase in sensitivity at high frequencies is a
characteristic of the plane wave tube. . . . . . . . . . . . . . . . . . 78

4.7 Noise floor of a piezoelectric MEMS microphone with a 18.4 kHz

resonant frequency at room temperature. The modeled noise floor is
roughly 2 dB lower than that measured but within the range expected
due to the measured cross-wafer variation . . . . . . . . . . . . . . . 79

4.8 Measured frequency response of the resonant speaker/tube system. . 81

5.1 Piezoelectric MEMS microphone of present work compared to pub-

lished works. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

vi
 LIST OF TABLES

Table

1.1 Range of properties of common MEMS piezoelectric materials. . . . 11

2.1 Beam results with and without small piezoelectric coupling . . . . . 32

2.2 Range of properties of common MEMS piezoelectric materials. . . . 50

4.1 AlN capacitance measured for capacitors with different distances to

etched AlN. The capacitance and loss angle was independent of the
distance to etched AlN. . . . . . . . . . . . . . . . . . . . . . . . . . 76

4.2 Measured d31 . The top layer exhibited a slightly larger d31 . . . . . . 76

vii
 LIST OF ABBREVIATIONS

AlN aluminum nitride

AFM atomic force microscope

Ar argon

ASIC application specific integrated circuit

dBA A-weighted decibel scale

BOE buffered oxide etch

CMOS complimentary metal oxide semiconductor

DRIE deep reactive ion etching

FET field effect transistor

FWHM full width half maximum

H2 O water

HCl hyrdochloric acid

HNO3 nitric acid

IEEE Institute of Electrical and Electronics Engineers

JFET junction field effect transistor

LCR inductance-capacitance-resistance

LNF Lurie Nanofabrication Facility

Mo molybdenum

MEMS microelectromechanical systems

N2 nitrogen

viii
PCB printed circuit board

PECVD plasma enhanced chemical vapor deposition

P(VDF-TrFE) piezoelectric copolymer of polyvinylidene flouride

PZT lead zirconate titanate

p-JFET P-type junction field effect transistor

RMS root mean square

SPC small piezoelectric coupling

SPL sound pressure level

THD total harmonic distortion

TMAH tetramethylammonium hydroxide

TO transistor outline

XRD X-ray diffraction

ZnO zinc oxide

ix
 LIST OF SYMBOLS

a acceleration

Ac , Ae , As cross-sectional, electrode, sensor area

An , Am , A0 Fourier Series coefficient

b beam width

c speed of sound in air

c 6 x 6 stiffness matrix

C electrical capacitance

Cfront , Ct , Cback acoustic compliance

C1 , C2 constants of integration

d 3 x 6 piezoelectric coupling coefficient matrix

d31 element of the piezoelectric coupling coefficient matrix

D electric displacement

D diaphragm flexural rigidity

D 3 x 1 electric displacement vector

e 3 x 6 piezoelectric coupling coefficient matrix

E electric field

E 3 x 1 electric field vector

f frequency

F force

g gap width

x
gfs transistor transconductance

h layer height

i layer index

ig , id small signal current noise

I area moment of inertia

IG , IQ large signal current

I0 , I1 modified bessel function

j square root of -1

J0 , J1 bessel function

k wave number

kB Boltzmann constant

l gap length

L beam length

Lfront , Lt acoustic inductance

M moment

MV piezoelectric induced moment

n layer index

N total number of layers

N diaphragm tension per unit length

pz pressure in the z-direction

P peak amplitude of a harmonic pressure

q electron charge

r radial coordinate

R acoustic, electrical, or mechanical resistance

R diaphragm radius

Rr radiation resistance

Rf dielectric film equivalent series resistance

xi
RL load resistance

Rfront , Rt , Rback acoustic resistance

s, s11 element of the compliance matrix

s 6 x 6 compliance matrix

T diaphragm tension per unit area

Tk temperature

v voltage

V voltage

∀ volume

VTO threshold voltage

w diaphragm displacement; beam displacement

x cartesian coordinate

y cartesian coordinate

Y Young’s modulus

YI beam flexural rigidity

YI beam flexural rigidity without assuming SPC

z cartesian coordinate

zmi distance to the middle of layer i

zn distance to the neutral axis

ZQi SPC constant

ZCi SPC constant

α, α1 , α2 SPC constant, diaphragm constant

β beam constant

γ diaphragm constant

δ material loss angle

ǫ electric permittivity

ǫ 3 x 3 electric permittivity matrix

xii
ε strain

ε 6 x 1 strain vector

θ azimuthal coordinate

κ tension parameter

λ SPC constant

Λ SPC constant

ν Poisson’s ratio

ξ loss angle calculation constant

ρ density

σ stress

σ 6 x 1 stress matrix

ς conductivity

φ electric potential

Ψ optimization parameter

ω radian frequency

xiii
 CHAPTER I

Introduction

Microphones are used in a variety of everyday devices such as telephones and

computers but also have many specialized applications such as studio recording and
laboratory testing. Every year, more than two billion microphones are built for a
range of applications. Most of these microphones convert sound into an electrical sig-
nal by utilizing capacitive transduction. Piezoelectric transduction, however, offers
unique advantages over capacitive transduction such as simplicity of fabrication and

linearity. These advantages have led to a further investigation of the typically cited
disadvantages of piezoelectric technology such as noise floor and sensitivity. Micro-
phones utilizing piezoelectric transduction have been designed, fabricated, and tested.
This work has led to new models of piezoelectric cantilevers and more accurate meth-

ods for determining appropriate model assumptions. New, more accurate, methods
of determining piezoelectric coupling coefficients have been developed. Optimization
techniques which apply to a broad range of piezoelectric sensors have been identified.
Piezoelectric microphone advantages and limitations will be demonstrated.

1.1 Capacitive and Piezoelectric MEMS Microphones

Microelectromechanical systems (MEMS) is a term used to describe a variety of

electro-mechanical devices built by utilizing equipment and techniques originally de-

1
 70 15kHz, Franz

10kHz, Royer

60 16kHz, Schellin

Noise Level (dBA)

10kHz, Ried
17kHz, Kressman

50

40
9kHz, Zou 10kHz, Bourouina
8.5kHz, Leoppart

17kHz, Bergqvist
30 14kHz, Scheeper
Capacitive Mics
Piezoelectric Mics 20kHz, Scheeper

20 −7 −6 −5 −4
10 10 10 10
Diaphragm Area (m2)

Figure 1.1: Capacitive and piezoelectric microphone noise levels as of December 2009.

veloped for integrated circuit manufacturing. These techniques, therefore, allow for
mechanical devices with very small, well defined features to be built. Since the early
1980’s, researchers have been developing MEMS microphones that utilize both capaci-
tive and piezoelectric transduction [1, 2]. Typically, MEMS piezoelectric microphones

have had a much higher noise floor (>10×) than capacitive microphones as seen in
Figure 1.1. This large disparity between capacitive and piezoelectric performance has
contributed to the adoption of capacitive transduction as the dominant technique for
MEMS microphones. Today, millions of capacitive MEMS microphones are built each

year.
The basic parameters of concern for any microphone, regardless of sensing tech-
nique, are input referred noise (also measured as minimum detectable signal, signal-
to-noise ratio, or noise floor), total harmonic distortion (THD, also measured as

maximum input level or dynamic range), and bandwidth (also measured as reso-
nant frequency). Also of interest are factors such as sensitivity, power consumption,

2
cost, and output impedance but these will not be stressed as heavily in the following
work because they are not inherent limitations of the device itself. For example, the
sensitivity of any microphone can be increased (without significant change in input
referred noise, bandwidth, or distortion) by adding an appropriate amplifier to an ex-

isting microphone. Likewise, the power consumption of capacitive and piezoelectric

microphones is more a function of their amplifying electronics than of the specific
devices. It is, however, important to evaluate the total microphone/amplifier (or
preamplifier) system because the system output must be such that it can be sensed,
amplified, or buffered without affecting any of the basic parameters of concern.

1.2 Noise

Noise is referred to as the inherent system fluctuations in system inputs or outputs

such as voltage, current, pressure, and displacement. This is separate from interfer-
ence from external signals because this interference is not an inherent limitation as it

can be removed. This work will primarily be concerned with thermal noise. Thermal
noise (also referred to as Johnson noise) was first documented by Johnson in 1928
[3]. Johnson noticed an electromotive force in conductors that is related to their tem-
perature. Nyquist then explained these results as a consequence of Brownian motion
[4, 5]. The theory of Nyquist was proved by Callen and Welton in 1951 [6]. Callen

and Welton also gave examples of mechanical systems that exhibit noise resulting
from mechanical dissipation [6]. Simply, any mechanism that converts mechanical
or electrical energy to thermal energy, such as a resistor or damper, also converts
thermal energy to mechanical or electrical energy. Therefore, any dissipative system

at a temperature above absolute zero will have noise associated with random thermal
agitation.
In a microphone/amplifier system, each significant noise source must be taken into
account. To do this, the effect of each noise source can be traced through the system

3
to the output. The noise on the output is referred to as output referred noise. The
sensitivity of the device (for a microphone this is given in V/Pa) can then be used
to determine the equivalent noise at the input called the input referred noise. This
input referred noise is ultimately the noise of interest in any sensing system. Input

referred microphone noise is typically quoted as a sound pressure level (SPL) given
on an A-weighted scale (dBA). The A-weighted scale weights specific frequencies to
mimic the sensitivity of human hearing. The noise level is then converted to SPL, a
decibel scale referenced to 20 µPa, the nominal lower limit of human hearing.
The noise floor of capacitive microphones is typically limited by noise in the micro-

phone itself, the microphone preamplifier, or both. The dominant noise source in the
microphone is thermal noise resulting from damping seen by air entering and leaving
the small capacitive gap between the diaphragm and backplate [7]. The microphone
preamplifier noise is determined by the circuitry and can be affected by the device

capacitance, depending on the amplification scheme used [8, 9].

As a simplified example of capacitive microphone optimization, consider a MEMS
microphone consisting of a diaphragm and backplate. The backplate has holes to
reduce the resistance to air escaping the gap, thereby reducing the noise in the micro-

phone. If the microphone noise is dominant over the preamplifier noise, the number
of holes in the backplate can be increased or the distance between the backplate
and the diaphragm can be increased, both of which reduce air flow resistance and,
therefore, microphone noise [8]. Both of these changes also reduce the microphone

capacitance and sensitivity, thereby increasing the input referred preamplifier noise
[8]. It is beneficial to reduce the microphone noise at the expense of preamplifier noise
until the preamplifier noise becomes larger than the microphone noise. At the point
where both contribute equally to the overall noise of interest, the total microphone
noise has been minimized. This general microphone optimization technique can be

used for any capacitive microphone, MEMS or otherwise.

4
 Although several groups have developed piezoelectric MEMS microphones, the
fundamental limiting factor for noise floor remains unclear. Typical piezoelectric
microphones consist of multiple sensing electrodes due to the varying stress in piezo-
electric material on a diaphragm [1, 10, 11, 12, 13]. Researchers have exploited the

fact that these electrodes can be combined to trade off sensitivity for device capaci-
tance [10, 11]. When combined optimally, the combination of electrodes will preserve
the total output energy of the piezoelectric device but the capacitance can be adjusted
to minimize the effect of circuit noise on the microphone [10]. This method of min-
imizing noise, however, neglects all noise coming from the piezoelectric microphone

and only minimizes the circuit noise.

The piezoelectric microphone will have thermal noise that stems from radiation
resistance [14] as well as structural damping [8]. The device will also have thermal
noise caused by the real part of the electrical impedance of the piezoelectric material

as described by Levinzon in a paper addressing piezoelectric accelerometers [15]. This

noise is determined by the loss angle (or dissipation factor) of the material and will
be filtered by the capacitance of the device [15]. Depending on the piezoelectric
material and amplification scheme used, this noise can be the dominant noise source

and cannot be ignored. This work will present complete microphone/amplifier system
noise models. This work will also present optimization techniques for reducing the
noise floor of piezoelectric devices. The noise models will be validated experimentally.

1.3 Sensitivity

Microphone sensitivity is also of importance because this sensitivity is used to

map electrical noise back to the input. Piezoelectric microphone sensitivities rang-
ing from 85 µV/Pa [13] to 920 µV/Pa [10] have been reported with typical values
around 200 µV/Pa [1, 11, 12]. Microphone sensitivity can be modeled by using a
variety of piezoelectric laminate models that have been under development since the

5
discovery of piezoelectric materials more than 100 years ago by the brothers Curie
[5]. A piezoelectric material is one that becomes electrically polarized when strained
(called the direct piezoelectric effect) and conversely, becomes strained when placed
in an electric field (called the converse piezoelectric effect) [16]. A sensor such as a

microphone utilizes the direct piezoelectric effect but it is not always appropriate to
ignore the strain resulting from the converse piezoelectric effect. The converse piezo-
electric effect can also be used to build actuators or measure piezoelectric coupling
coefficients [17].
The Institute of Electrical and Electronics Engineers (IEEE) has released a stan-

dard on piezoelectricity [18]. This standard lists the piezoelectric constitutive equa-
tions as

σ = cE ε − et E (1.1)

D = eε − ǫε E (1.2)

where σ is the 6×1 stress vector, cE is the 6 × 6 stiffness matrix measured under zero

electric field, ε is the 6 × 1 strain vector, e is the 3 × 6 piezoelectric coupling coefficient

matrix, E is the 3 × 1 electric field vector, D is the 3 × 1 electric displacement
vector, and ǫε is the 3 × 3 electric permittivity matrix measured under zero strain.
Equation 1.1 is similar to Hooke’s law but includes an additional term which couples
the mechanical and electrical domains. Equation 1.2 is similar to the definition of

electric displacement but includes an additional term coupling the mechanical and
electrical domains. This coupling between mechanical and electrical domains means
that a complete model of a piezoelectric material must include Equations 1.1 and 1.2 in
some form. Often, however, if the piezoelectric material is used for actuation (converse

piezoelectric effect), Equation 1.2 is ignored [19]. If the piezoelectric material is used
as a sensor (direct piezoelectric effect), the electric field term is often ignored in

6
Equation 1.1 and the electric field and electric displacement can be computed using
Equation 1.2 and the electrical boundary conditions [20]. Small piezoelectric coupling
(SPC) is the term given to these two assumptions [21].
Several piezoelectric materials of interest such as aluminum nitride (AlN), zinc

oxide (ZnO), and lead zirconate titanate (PZT) belong to the dihexagonal polar class
of crystals (6 mm). Many models assume the piezoelectric crystal is of this class
because these materials are common and this assumption simplifies Equations 1 and
2. Microphones of this class are characterized by a piezoelectric coupling coefficient
matrix of the form  
 0 0 0 0 d15 0 
 
d=
 0 0 0 d15 0 0 
 
d31 d31 d33 0 0 0

and a compliance matrix of the form

 
 s11 s12 s13 0 0 0 
 

 s12 s11 s13 0 0 0 

 
s13 s13 s33 0 0 0
 
 
s=




 0 0 0 s14 0 0 

 

 0 0 0 0 s14 0 

 
0 0 0 0 0 2(s11 − s12 )

These forms are most convenient for the strain-charge form of the piezoelectric con-
stitutive equations given in Chapter II. The form of the stiffness matrix, c, and
the piezoelectric coupling matrix, e, are the same as s and d, respectively, except

c66 = 1/2(c11 − c12 ).

Microphones are most easily realized using a diaphragm or cantilever beam [1, 22].
Cantilevers will be the focus of this work because they are free of residual stress while
diaphragm stiffness is often dominated by residual stress. This residual stress can

7
be difficult to predict and control and can lead to degraded device performance [23].
Most models of piezoelectric multilayer beams assume that plane sections remain
plane (referred to as the Euler-Bernoulli assumption) [24, 25, 26] while others make
the Timoshenko assumption that allows for shear and rotary inertia [27]. This work

will focus on beam models utilizing the Euler-Bernoulli assumption because the added
accuracy of the Timoshenko assumption is not necessary for the designs and tests of
interest.
The assumption of small piezoelectric coupling (SPC) will be investigated. Some
have said that SPC can be assumed whenever the piezoelectric coupling coefficient is

small [24] while others claim that the SPC assumption can lead to significant errors
for structures in which the substrate thickness is similar to or smaller than that of the
piezoelectric layer [28]. Criterion for determining the validity of the SPC assumption
will be presented in the following work. This work will also present analytical models

of multilayer piezoelectric beams both with and without SPC. These models will
be used to measure piezoelectric coupling coefficients as well as device sensitivity
experimentally in order to validate these models.

1.4 Linearity

The linearity of the microphone/amplifier system can be limited by the micro-

phone, amplifier, or both. Capacitive devices are inherently nonlinear due to the
nonlinear relationship between capacitance and gap size. As diaphragm deflection in-
creases, distortion increases so THD at a specific SPL can be decreased by increasing
the diaphragm stiffness. In this manner, microphone distortion and bandwidth can

be improved at the expense of noise and sensitivity. With the application of appropri-
ate electronics, the capacitive nonlinearity will limit the SPL at which low harmonic
distortion is achieved. This tradeoff can be observed by comparing two capacitive
microphones, one designed for low noise and one for high dynamic range. The low

8
noise microphone has a diaphragm area of 2.8 × 10−7 m2 , a noise floor of 37 dBA, and
a 1% THD at 100 dB SPL [29]. This A-weighted noise floor would be equivalent to 13
√
µPa/ Hz. The high dynamic range microphone has a diaphragm area of 1.7 × 10−7
√
m2 , a resonant frequency of 178 kHz, input referred noise of roughly 2.2 mPa/ Hz

at 1 kHz, and has linear operation up to 164 dB SPL. The first microphone’s input
referred noise is roughly 45 dB lower than the second but the second is linear up to
a SPL that is roughly 64 dB higher than the first.
Piezoelectric transduction can be assumed to be linear in the presence of small
electric fields but becomes nonlinear as the electric fields become large [30]. The

nonlinearity of PZT is primarily attributed to its ferroelectric hysteresis [31]. The

nonlinearity of non-ferroelectric piezoelectric materials such as ZnO and AlN has also
been investigated [32, 33] but well established, experimentally validated nonlinear
models of piezoelectric AlN or ZnO have not been found. Experiments performed in

the completion of the present work have indicated that AlN remains linear for electric
fields in excess of 1 × 106 V/m. The structure can also provide a source of sensor
nonlinearity. The primary source of cantilever beam nonlinearity is in the relationship
between curvature and displacement. At large deflections, the beam curvature cannot

be approximated as the second derivative of the displacement and the deflection is,
therefore, a nonlinear function of applied pressure. Although the deflection is not
linearly related to applied pressure when deflections are large, the layer stress and
therefore the output voltage remains a linear function of applied acoustic pressure.

The primary source of diaphragm nonlinearity is due to the stretching of the neutral
axis and does result in nonlinear stress in the piezoelectric layers [34].

1.5 Piezoelectric Materials

Piezoelectric material selection can have significant impact on device performance.

Different piezoelectric materials exhibit vastly different qualities and it is extremely

9
important to use the best material to suit a particular application. The most common
piezoelectric materials for MEMS devices are ZnO, PZT, and AlN.
ZnO and AlN are quite similar in that they are both dihexagonal polar crystals
(6mm) and are non-ferroelectric [35, 36]. AlN has the advantage of being compatible

with silicon semiconductor technology while ZnO can be problematic. AlN also has
a higher resistivity than ZnO. Because it can be difficult to deposit ZnO with a high
resistivity, sensors and actuators operating below 10 kHz often require an insulating
layer, usually SiO2, to reduce charge leakage [36, 37]. Despite these advantages,
ZnO has been used more commonly due to the better availability of ZnO films, less

demanding vacuum conditions for ZnO, and some early negative experiences with
residual stresses in AlN films. Over the past few years, AlN deposition has become
more prevalent and the residual stress has been shown to be more controllable [35, 38].
Both AlN and ZnO are commonly sputter deposited and as such, AlN and ZnO

thin films are almost always polycrystalline materials. The piezoelectric coupling
coefficients result from the average effect of all crystals. These materials must be
deposited in such a manner as to align the crystals or they will exhibit degraded
piezoelectric properties. X-ray diffraction is typically used to evaluate the degree of

orientation of these materials [36]. X-ray diffraction rocking curves with full width
half maximum (FWHM) values below 2◦ are typically considered well oriented for AlN
films [35]. PZT differs from ZnO and AlN in that it is a ferroelectric material. This
means that, unlike that of AlN and ZnO, the polar axis of PZT can be reoriented

after deposition [36]. PZT has been primarily designed to have high piezoelectric
coupling coefficients [36] while relatively little effort has been put into improving its
other properties. Some of the most relavent properties of ZnO, PZT, and AlN for
this study are shown in Table 1.1 for comparison.

10
 ZnO PZT AlN
d31 (C/N) −5.74 × 10−12 [39] −40 × 10 −12
- −70 × 10−12 [36] −2.65 × 10−12 [40]
d33 (C/N) 10.4 × 10−12 [39] 92 × 10−12 -203 × 10−12 [36] 5.53 × 10−12 [40]
s11 (P a−1 ) 9.63 × 10−12 [39] 9.52 × 10−12 -14.9 × 10−12 [41] 3.53 × 10−12 [40]
tan(δ) 0.01-0.1 [36] 0.03-0.05 [42] 0.002 [42]
ǫ33 /ǫo 10.9[36] 1100-1300 [42] 10.4 [42]
ρ (kg/m3 ) 5670 7600 3300

Table 1.1: Range of properties of common MEMS piezoelectric materials.

1.6 Capacitive Microphone Embodiments

Figure 1.1 shows, briefly, the performance of several capacitive MEMS micro-
phones. The smallest of these microphones was built by researchers at Knowles

Electronics [29]. This microphone has a 600 µm polysilicon diaphragm positioned

4 µm from the backplate. As noted in the paper, film stress significantly impacts
the sensitivity of the microphone. If not controlled, variations in film stress will
lead to unacceptable variations in microphone sensitivity. For this reason, the di-

aphragm is left floating with only a single attachment between the silicon wafer and
the diaphragm for electrical connection. When in use, an 11 V bias voltage holds
the diaphragm 4 µm from the backplate, giving a device capacitance of 0.5 pF. The
output is buffered by an application specific integrated circuit (ASIC) resulting in a
packaged microphone with a 37 dBA noise floor. As this microphone became a com-

mercial device, it probably represents the most highly optimized capacitive MEMS
microphone of those in Figure 1.1.
The microphone built by Bergqvist and Rudolf is another representation of a high
performance capacitive microphone [43]. This microphone has a 5 µm thick 2 mm

× 2 mm silicon diaphragm and a 4.5 pF working capacitance. When a 6 V bias was

applied, the microphone exhibited a 30 dBA noise floor. This device has a lower
noise floor than that described above but a larger area. As the area of a microphone
increases, the noise floor would be expected to drop. Because noise combines as the

11
square root of the sum of the squares and the sensitivity sums, doubling the sensor
area should result in a 3 dB improvement in noise floor. This device has roughly 14
times the area of a microphone with a 600 µm diameter and should therefore have a
noise floor 11.5 dB lower. This analysis would indicate that this microphone is not

as well optimized as the smaller microphone detailed above.

1.7 Piezoelectric Microphone Embodiments

The performance of several piezoelectric microphones has also been illustrated in

Figure 1.1. The lowest noise piezoelectric MEMS microphone detailed above is that
built by Kressman et. al [11]. This microphone uses a diaphragm design consisting of

a 230 nm oxide/nitride diaphragm, aluminum electrodes, and a 4.3 µm thick P(VDF-

TrFE) piezoelectric layer. The piezoelectric material is quite thick because it is very
compliant relative to the other materials comprising the diaphragm. The stress of
the oxide/nitride diaphragm is reported to be greater than 107 MPa. The diaphragm

is 1 mm × 1 mm and has separate middle and outer electrodes due to differences in

stress. The signal is buffered with a p-JFET with a gain of 0.85. The microphone has
a capacitance of 4 - 8 pF and a sensitivity of 0.21 mV/Pa. A bandwidth of 17 kHz
is given but the authors expect the resonant frequency to be around 100 kHz. These
microphones have a noise floor of 54.6 dBA. The authors attribute the increased

performance of this microphone to the thickness of the piezoelectric material layer

but note that this thickness is limited by an eventual reduction in sensitivity due to a
reduction in membrane displacement and by the movement of the neutral axis toward
the center of the piezoelectric material.

The microphone designed by Ried et al. is also noteworthy [44]. This microphone
is built on the same chip as its CMOS circuitry and some of the CMOS layers are
used to form the microphone. The microphone has a 2.4 mm × 2.4 mm diaphragm
that consists of a 1.7 µm silicon nitride layer, a 0.4 µm polysilicon bottom electrode,

12
a 0.2 µm SiO2 layer, a 0.5 µm ZnO piezoelectric layer, a 0.1 µm SiO2 layer, and a 0.6
µm Al top electrode. The diaphragm is nominally 3.5 µm thick. The stress of the
silicon nitride layer is 140 MPa tensile and that of the ZnO is 1 GPa compressive.
These layer stresses are designed to cancel and give a relatively stress-free device. The

stress in the diaphragm, however, resulted in a resonant frequency of 18.3 kHz instead
of the designed 6.6 kHz and a sensitivity of 0.92 mV/Pa instead of the designed 8.9
mV/Pa. The sensitivity of this device is so high because this device has 8 elements
wired in series, providing a high sensitivity but a capacitance of only 2.49 pF. This
microphone had a 57 dBA noise floor.

1.8 Gaps in the Literature

Although several piezoelectric MEMS microphones have been fabricated and tested,
the fundamental tradeoffs and performance limitations have not yet been fully inves-
tigated. While only one of the piezoelectric microphones in Figure 1.1 used AlN, the

material properties of AlN indicate that it will lead to better microphones than the
more commonly used ZnO and PZT when noise floor is of primary concern. Further,
as AlN becomes more prevalent and deposition techniques improve, the properties
and applications of AlN will also likely improve.
Additionally, while many piezoelectric MEMS microphone publications cited high

residual stress as a factor that reduced microphone sensitivity and increased band-
width [44, 11, 13], none examined the effect of residual stress when considering a
design with a fixed bandwidth. Similarly, many designs focused on maximizing the
device sensitivity either by making the piezoelectric layer thick [11] or by wiring sev-

eral electrodes in series [44]. Only the work of Ried [44], however, attempted to
minimize the noise by examining the full transducer/amplifier system. Further, none
of these works considered noise from the piezoelectric material loss.
Chapter II will present models of both the MEMS transducer and a low noise

13
amplifier. Noise resulting from material loss will be included. Together, these models
will form a complete system model that will be used to identify the important figures
of merit for the MEMS transducer design. The MEMS transducer design will then
be optimized to minimize the noise floor of a piezoelectric microphone of fixed area

and bandwidth. Cantilever and diaphragm designs will be compared.

Chapter III will then present the design, fabrication, and performance of a piezo-
electric MEMS microphone built with these models in mind. While the performance
of this first microphone did not meet expectations, it validated the models and illu-
minated certain fabrication difficulties. Chapter IV will then present a new design

aimed at overcoming fabrication issues experienced with the first microphone design.
This microphone design is unique in many ways. It is the one of only a few piezoelec-
tric MEMS microphones to utilize AlN as the piezoelectric material, uses a cantilever
design in order to avoid the detrimental effects of residual stress, has optimized elec-

trode shapes, and has relatively thin layers. This second iteration will be shown to
significantly outperform all previously reported piezoelectric MEMS microphones.

14
 CHAPTER II

Modeling

This chapter will cover several models used to predict microphone performance.

First, the dominant noise sources in the microphone will be modeled in Section 2.1.
These noise sources include those from the transducer as well as those from a common
source amplifier used to buffer the microphone output. The following section, Section
2.2 will present models for the sensitivity of cantilever and diaphragm based micro-
phones. Section 2.3 will provide a discussion of microphone linearity. The linearity

of the common source amplifier used to buffer the output will be examined as this is
expected to be the much more non-linear than the piezoelectric transducer. A model
of a high sound pressure level (SPL) test apparatus used to measure microphone dis-
tortion will also be presented in this section. These three sections should provide

a fundamental understanding of piezoelectric transduction and the associated noise

and linearity concerns.
Sections 2.4 and 2.5 give insight into the optimization of the entire microphone.
First, Section 2.4 provides a method of separating the transducer optimization from

the amplifier optimization so they can be addressed separately. Section 2.5 then
shows how to optimize the transducer alone. As noise has been the most significant
hindrance to piezoelectric microphone utilization, this is the focus of the optimization.
Section 2.6 contains a discussion of the two most common microphone packages.

15
This was not incorporated into the optimization because the packaging is designed to
prevent environmental factors from disrupting microphone performance but should
not inhibit performance in any way. The microphone cross-sensitivity to vibration is
then briefly discussed in Section 2.7.

2.1 Noise

The noise floor of a piezoelectric sensing system can be limited by the sensor itself,
the amplifying electronics, or both. As such, an electromechanical system model is
necessary to determine the noise floor of any piezoelectric device. An example of such
a model will be provided for a microphone amplified by a common source amplifier

utilizing a junction field effect transistor (JFET). A common source amplifier with
a JFET has been selected because JFETs are the leading devices for low-noise ap-
plications where the signal source impedance is large [45]. The noise floor of such
a common source amplifier is similar to that of other, more complex amplification

schemes but its simplicity allows for easier identification of noise sources and design
tradeoffs.

2.1.1 Microphone Transducer Noise

The microphone element itself has thermal noise associated with radiation resis-
tance, structural dissipation, electrical loss in the piezoelectric material, and electrode
resistance. The radiation resistance is a mechanical resistance and, therefore, has a

mechanical thermal noise associated with it. The Nyquist relation gives the thermal
noise associated with any resistive element as

F 2 = 4kB RTk (2.1)

16
where F is the root mean square (rms) value of the force spectral density, kB is
Boltzmann’s constant, Tk is the temperature in Kelvin, and R is the resistance. When
R is electrical resistance, mechanical resistance, or acoustic resistance, F has units of
√ √ √
V/ Hz, N/ Hz, or Pa/ Hz, respectively.

2.1.1.1 Radiation Resistance and Damping

The acoustical radiation resistance of a non-baffled circular radiating piston is

ρo c(kAs )2
Rr = (2.2)
4π

where ρo is the density of the fluid, c is the speed of sound in the fluid, k is the wave
number defined as the radian frequency (ω) divided by the speed of sound, and As
is the area of the radiating surface [14]. Equation 2.2 applies to circular radiating
pistons but can be applied to radiating surfaces of any shape whenever as is the case

for most MEMS microphones and pressure sensors operating in the acoustic range in
air [46]. Combining equations 2.1 and 2.2 gives the resulting acoustic noise pressure
as a function of frequency. For example, the rms spectral density of pressure noise at
1kHz using a 500 µm × 500 µm square radiating surface in air at room temperature
√
(300 K) is 1.35 × 10−8 Pa/ Hz or an SPL of -63.4 dB re 20 µPa using a 1 Hz bin.
This will be shown to be negligible compared to other noise sources.
The thermal noise resulting from structural dissipation is also quite low. It has
been shown that as beams are miniaturized, air damping is the dominant source

of damping at atmospheric pressure [47]. Since a microphone cannot be vacuum

encapsulated, thermal noise resulting from structural damping will be ignored.

17
 Figure 2.1: Electrical equivalent circuit of piezoelectric sensing element.

2.1.1.2 Material Loss

Electrical noise from the resistance of the piezoelectric material can be calculated
using equation 2.1 [15]. The piezoelectric material with electrodes on either side forms
a capacitor which can be modeled as capacitance and resistance in parallel as shown

in Figure 2.1. The resistance of a film as a function of frequency is

1
Rf = (2.3)
ωCf tan δ

where Cf is the film capacitance and tan δ is the loss angle. The loss angle is the
tangent of the angle by which the current in a lossy capacitor lags the current in an
ideal capacitor [48]. The loss angle is also referred to as the dissipation factor or the
dielectric loss. This loss angle can be used to determine the sensor resistance which

can be plugged into equation 2.1 to get a voltage noise spectral density. This voltage
noise source is modeled in series with the resistor as shown in Figure 2.1.

2.1.1.3 Electrode Loss

As will be shown, thin electrodes are desirable for high performance. As the
electrodes become thin, however, they contribute to the device noise floor so it is

important to approximate the impact of thinning electrodes. In order to simulate the

effect of thinning electrodes, it is important to capture two separate effects. First, as

18
 (a) Material stack and variables. (b) Material stack and boundary conditions

Figure 2.2: Diagram of electrode/dielectric/electrode stack modeled as a continuous

system and associated boundary conditions.

the electrode becomes thin, the electrode material resistivity will increase. This has
been described and modeled by Namba [49]. Secondly, as the electrode becomes thin,
its resistance will increase because the cross-section of the electrode has decreased.

This can be investigated by modeling a single layer of the piezoelectric material sand-
wiched by two electrode layers as a continuous system. A diagram of the continuous
system can be seen in Figure 2.2(a). The electrode material, material 1, will be mod-
eled as a conductor with finite, real conductivity, ς1 and the piezoelectric material will

be modeled as a dielectric with complex conductivity given as ς2 = ǫo ω(tan δ + j). Be-

cause the electrode is much longer than it is thick and has a much larger conductivity
than the dielectric, the electrode potential is assumed to be constant in the thickness
direction and vary only along the length of the beam. Laplace’s equation is assumed
to hold in the dielectric because the dielectric has very few free charges. These as-

sumptions result in a 1-D model of the conductor coupled to a 2-D model of the
dielectric. Figure 2.2(b) illustrates the geometry and assumed boundary conditions.
Solving this problem for voltage gives

∞  nπx   nπy 
X vo
v2 = Ao y + An cos sinh + (2.4)
n=1
L L 2

19
with RL
0
v1 (x)dx
Ao = (2.5)
Lh2

and RL  
0
v1 (x) cos nπx
L
dx
An =   (2.6)
L sinh nπh2L
2

where v1 and v2 are the voltages in the conductor and dielectric respectively, v0 is the
applied voltage, h2 is the height of the dielectric, and L is the length of the beam.

The solution for the voltage in the electrode is

ς2 dv2 (x, y = h2 /2)

Z Z
v1 = − dxdx + C1 (x = L)x + C2 (x = 0) (2.7)
ς1 h1 dy

where
ς2 dv2 (x, y = h2 /2)
Z
C1 = dx (2.8)
ς1 h1 dy

and
ς2 dv2 (x, y = h2 /2)
Z Z
C2 = vo + dxdx (2.9)
ς1 h1 dy

As can be seen, the equations are coupled together and the problem can be solved

numerically by guessing a value of v1 , solving for v2 , calculating the new v1 , and

iterating until the solution converges. A sample solution with ς1 = 1 × 107 Ω−1 m−1 ,
ς2 = 1 Ω−1 m−1 , L = 400 µm, h1 = 10 nm, and h2 = 1 µm is shown in Figure 2.3.
Only the voltage in the top half of the electrode/material stack is shown due to
symmetry. While the values used to create Figure 2.3 show significant voltage loss

across the length of the top electrode, realistic values for conductors and dielectrics
of interest would show almost no change in voltage across this electrode. Therefore,
the problem can be simplified by recognizing that the slope of the voltage in the x-
direction will be much less than that in the y-direction and can be ignored. When the

slope of the voltage in the x-direction is ignored, instead of using Laplace’s equation

20
Figure 2.3: The voltage in the material stack varies from the applied voltage (v0 ) at
the base of the electrode to half of that value at the center of the stack.
Only the top half of the stack is shown due to symmetry.

in the dielectric, one can use the equation

d2 v2
=0 (2.10)
dy 2

The voltage in the dielectric then becomes

y 1
v2 (x, y) = v1 (x) + (2.11)
h2 2

and that in the conductor is

vo h r 2ς  r 2ς  r 2ς i v
2 2 2 o
v1 (x) = cosh x − tanh L sinh x + . (2.12)
2 ς1 h1 h2 ς1 h1 h2 ς1 h1 h2 2

In order to determine the effect on the noise, the effective increase in loss angle must

be determined. First, the impedance is determined by dividing the applied voltage

by the current into the conductor (as the current into the conductor is equal to the
total current through the device),

2
z= (2.13)
ς1 h1 bξ tanh(ξL)

21
where z is the impedance, b is the beam width, and

r
2ς2
ξ= (2.14)
ς1 h1 h2

In equation 2.43, the conductivity of material two, ς2 , is complex which makes the
impedance complex. The loss angle is then found by dividing the real part of the
impedance by the imaginary part of the impedance. However, this quotient has no

closed form solution because the real and imaginary parts of the impedance cannot
be separated analytically. The impedance is therefore approximated using a Laurent
series for the 1/ tanh(ξL) term leading to a loss angle of

2ǫo ωL2
tan(δ)total = tan(δ)piezo + (2.15)
3ς1 h1 h2

This expression for the total loss angle (that which includes losses in the electrodes)
can be substituted into equation 2.3 when calculating overall noise.

2.1.2 Microphone Circuitry Noise

The circuitry, in this case a common source amplifier, will be modeled in a manner
consistent with the literature [45]. This model will include thermal noise from the
bias resistor, thermal noise in the channel of the JFET, thermal noise from the load
resistor, and shot noise in the gate. A model of this circuit can be seen in Figure

2.4. A small signal model with all these noise sources can be seen in Figure 2.5. The
thermal noise from resistors Rb and RL can be computed using equation 2.1. The
rms spectral density of the current noise, ig , can be computed as

i2g = 2qIG (2.16)

22
where q is the electron charge (1.6 × 10−19 C) and IG is the DC gate current. This
noise is similar to quantization error and is caused by the fact that current is carried
in discrete quantities. The channel noise, i2d , is thermal noise and can be computed
as
2
i2d = 4kB Tk gf s (2.17)
3

where gf s is the JFET transconductance. This noise is caused by the resistance of

the channel where the 2/3 factor is typically used for long channel devices [50]. In

Figure 2.4: Common source amplifier circuit schematic.

Figure 2.5: Small signal model of the common source amplifier with noise sources.

order to compute the effect of each of these noise sources, all other noise sources are

23
removed (current sources are open and voltage sources are shorted) and its effect on
the output voltage is computed. Each noise source is then combined as the square
root of the sum of the squares. The output noise using an n-channel JFET (2sk3719)
with a 2.2 kΩ load resistor, a sensor capacitance of 10 pF, and a 3 V VDD can be seen

in Figure 2.6. Each individual noise source as well as the total summation of noise
can be seen in the figure. The sensor resistance has been ignored in this drawing.
The thermal noise from the bias resistor, Rb , is filtered by the combined capaci-
tance of the sensor and the gate source capacitance as seen in Figure 2.6. Therefore,
as the sensor capacitance increases, the output noise from the bias resistor is reduced.

This will be an important relationship as the sensor/amplifier system is optimized.

Figure 2.6: Output noise of common source amplifier. Plot includes noise from the
bias resistor, shot noise, channel noise, and noise from the load resistor.

24
 hN l=N
zN-1

...
l=3
l=2
z
y l=1
x
b

Figure 2.7: Piezoelectric laminate beam layer numbering and orientation.

2.2 Sensitivity

Many of the noise calculations in the last chapter provide an output noise in volts.

This noise is only relevant when related to the input via the microphone sensitivity.
This section will focus on multi-layer cantilevers of uniform width and electrodes
along their entire length as shown in Figure 2.7. These cantilevers will be used to
determine piezoelectric coupling coefficients of films and to sense acoustic pressure.
Most piezoelectric microphones use either a diaphragm or cantilever design. A can-

tilever is a good choice because it is free from residual stress which has been shown
to significantly reduce the sensitivity of diaphragm-based microphones. Following
an in depth cantilever sensitivity derivation, the sensitivity of a diaphragm based
piezoelectric sensor will also be presented.

2.2.1 Cantilever Transducer Sensitivity

For this analysis, it is more convenient to use the piezoelectric coupling equations
in the strain charge form written as

ε = sE σ + dt E (2.18)

D = dσ + ǫσ E (2.19)

25
instead of the stress charge form given in equations 1.1 and 1.2. Here ε is the strain
vector, sE is the compliance matrix measured under zero electric field, σ is the stress
vector, d is the piezoelectric coupling coefficient matrix, E is the electric field vector,
D is the electric displacement vector, and ǫσ is the electric permittivity matrix mea-

sured under zero stress [18]. Equations 2.18 and 2.19 can be derived from equations
1.1 and 1.2 using
cE = s−1
E (2.20)

e = d · s−1
E (2.21)

ǫε = ǫσ − d · s−1
E ·d
t
(2.22)

The analysis of a multilayer piezoelectric cantilever will make some additional as-
sumptions but will not make the assumption of small piezoelectric coupling (SPC) as
most authors have done in the past. The stress is assumed to be uniaxial (σ11 6= 0
while σ22 = σ33 = 0), and the electric field is assumed to be negligible in the x and y

directions (E3 6= 0 while E1 = E2 = 0). It is also assumed that the piezoelectric ma-
terial under consideration has the symmetry properties of a dihexagonal polar crystal
of class 6mm. Due to the simplifications inherent in these assumptions, s will denote
s11 , E will denote E3 , ǫ will denote ǫ33 .

The equilibrium equations for the multilayer beam can be integrated through the
thickness to get equations for axial force and bending moment given [26]. If axial
inertia and traction are neglected, the beam curvature can be expressed as

M = −w,xx Y I − MV (2.23)

where
N
X 1 h1 1 d231n 3
i
YI = b ZCk + h (2.24)
s 3
n=1 n
12 ǫn sn − d231n n

26
and
N
1 X 1 d31n Vn
MV = b ZQk . (2.25)
2 n=1 sn hn

In equations 2.23-2.25, a comma followed by a subscript denotes the derivative, the

subscript k refers to the layer, s is compliance, d31 the piezoelectric coupling coefficient
 
in the strain-charge form, ǫ is the electric permittivity, Vn = φn  −φn  where
 
z=zk−1 z=zn
3 3 2
φ is the electric potential, ZCk = (zn − zn) − (zk−1 − zn) , ZQk = (zn − zn) − (zk−1 −
zn)2 , hn = zn − zk−1 , and zn is the neutral axis in the absence of piezoelectricity and

is defined as PN 1 2 2
1 n=1 sn (zn − zk−1 )
zn = P N hn . (2.26)
2 n=1 sn

With the expression for bending moment given in equation 2.23, the beam equation

governing the bending motion of a slender beam can be written

−(w,xx Y I + MV ),xx + pz = ρAc ẅ0 . (2.27)

Here, pz is some uniform pressure in the z direction, Ac is the beam cross-sectional

area, and ρ is the thickness averaged beam density. The beam has boundary condi-
tions


w =0 (2.28)

x=0

w,x  =0 (2.29)

x=0

M =0 (2.30)

x=L

M,x  = 0. (2.31)

x=L

In sensing applications, the voltage, Vn , in equation 2.25 is unknown while under

voltage actuation, the voltage is a known inhomogeneous forcing term. Because

of this difference, the case of applied voltage and voltage sensing will be handled

27
separately.
In the case of beam actuation, some voltage Vn is prescribed across each piezo-
electric layer. If the applied force, pz , is assumed to be zero and assuming harmonic
excitation, the displacement profile for the beam is determined to be

1 MV h sinh(βL) − sin(βL)
w= (sinh(βx) − sin(βx))
2 Y Iβ 2 1 + cos(βL) cosh(βL)
(2.32)
cosh(βL) + cos(βL) i
− (cosh(βx) − cos(βx))
1 + cos(βL) cosh(βL)

where
PN
4 ρω 2 b n=1 hn
β = (2.33)
YI

and ω is the radian frequency. The short circuited resonant frequency of the beam

(neglecting damping) occurs when cos(βL) cosh(βL) = −1 and the first resonance
occurs when βL = 1.875. For cases where βL << 1.875, the actuation frequency is
much less than the natural frequency of the beam and inertia can be neglected. In
this case, the curvature is approximately constant with rexpect to x and given by

MV
w,xx = − . (2.34)
YI

This equation for beam curvature can be used to determine the d31 coefficient of a
cantilever of known geometry by applying a voltage and measuring the beam curva-
ture. Previous authors have suggested extracting d31 by measuring the tip deflection.
A more robust measurement would result from using multiple measurements along
the length of the beam to determine beam curvature and extract the d31 coefficient

from equation 2.34.

The assumption of SPC, when applied to a piezoelectric actuator, is the assump-
tion that the electric field is constant in the thickness direction of the piezoelectric

28
material. This assumption is valid if and only if

N N
1X d231n h3n X ZCk
<< . (2.35)
4 n=1 ǫn sn − d231n sn n=1
sn

This condition is satisfied whenever d231 /(ǫs) << 1 for all materials in the laminate.
This criterion is sometimes given as that necessary to assume SPC [24] but, as pointed
out by others [28], the SPC assumption can be dependent upon the ratio of piezoelec-
tric to structure thickness as well. When SPC can be assumed, Y I given in equation

2.24 can be replaced with the simpler form

N
bX 1
YI = ZCk . (2.36)
3 n=1 sn

If the SPC assumption is valid and if only one piezoelectric layer is being actuated,
the d31 coupling coefficient of layer k can be computed as

hn sn Y I
d31n = −2 w,xx (2.37)
ZQk Vn b

when a voltage Vn is prescribed across the layer and a curvature w,xx results. If SPC
can be assumed, other piezoelectric layers can be either shorted or open-circuited.
In the sensing configuration, we consider some form of external mechanical or
acoustical stimulation of the cantilever. When sensing voltage, it is assumed that the

electrodes are connected to sensing circuitry with large input impedance such that
electrons cannot flow from one side of the piezoelectric material to the other. The
stress in the piezoelectric material, therefore, gives rise to an electric field.
Without assuming SPC, the deflection profile of a clamped-free multi-layer piezo-

electric beam is

Pb
w= (a1 sin(βx) + a2 cos(βx) + a3 sinh(βx) + a4 cosh(βx) − 2) (2.38)
2Y Iβ 4

29
where

βL[sinh(βL) cos(βL) + sin(βL) cosh(βL)] + 2α sinh(βL) sin(βL)

a1 = , (2.39)
βL[1 + cosh(βL) cos(βL)] + α[cosh(βL) sin(βL) + cos(βL) sinh(βL)]
βL[sinh(βL) sin(βL) − cos(βL) cosh(βL) − 1] + 2α sinh(βL) cos(βL)
a2 = − , (2.40)
βL[1 + cosh(βL) cos(βL)] + α[cosh(βL) sin(βL) + cos(βL) sinh(βL)]
βL[sinh(βL) cos(βL) + sin(βL) cosh(βL)] + 2α sinh(βL) sin(βL)
a3 = − , (2.41)
βL[1 + cosh(βL) cos(βL)] + α[cosh(βL) sin(βL) + cos(βL) sinh(βL)]
βL[sinh(βL) sin(βL) + cos(βL) cosh(βL) + 1] + 2α cosh(βL) sin(βL)
a4 = . (2.42)
βL[1 + cosh(βL) cos(βL)] + α[cosh(βL) sin(βL) + cos(βL) sinh(βL)]

In equations 2.39-2.42, PN d Λn
1 b n=1 s31n n hn
ZQk
α=− , (2.43)
2 YI
PN hi P d31i
P d31i
En h ZQk i=1 si − i6=k si Ei hi − hn i6=k si
Ei ZQi i
Λn = PN hi d31i , (2.44)
2


i=1 si − si Ei hi

d31n
and En = ǫn sn −d231n
. The voltage developed across layer k of this beam is

1 Pb
Vn = − Λ n [a1 cos(βL) − a2 sin(βL) + a3 cosh(βL) + a4 sinh(βL)]. (2.45)
L 2Y Iβ 3

In the case of free vibrations, the characteristic equation can be found by setting the
denominator of equation 2.39 to zero giving

α
cosh(βn L) cos(βn L) + [cosh(βn L) sin(βn L) + cos(βn L) sinh(βn L)] = −1. (2.46)
βn L

This equation can be used to find the natural frequencies of the beam. If the beam

is operating significantly below its first natural frequency, inertia can be neglected
and the equations for displacement and output voltage simplify. The low-frequency
displacement profile is

Pb 1 4 1 3 3 − 2α 2 2 
w= x − Lx + Lx (2.47)
Y I 24 6 12(1 − α)

30
and the voltage across a layer is

P bL2 1
Vn = − Λn . (2.48)
6Y I 1 − α

When SPC can be assumed, the sensing equations simplify considerably. The most
robust method of determining the validity of the SPC assumption is to determine if
α << 1 and if equation 2.36 is true. Any time both of these equations are true, the

SPC assumption is valid. The SPC assumption is also valid whenever d231 /(ǫs) << 1
for all materials. This check of SPC validity depends only on the piezoelectric material
parameters. This shows that when using AlN, the SPC assumption is always valid
but when using PZT (depending on the material parameters), the more extensive
check might be necessary.

When SPC can be assumed, the beam deflection profile is equal to that of a
non-piezoelectric multi-layer beam. The output voltage simplifies to

P bL2
Vn = − λn (2.49)
6Y I

where
1 d31n
λn = ZQk . (2.50)
2 ǫn sn

Table 2.1 summarizes the findings of this section.

2.2.2 Diaphragm Transducer Sensitivity

The use of a diaphragm instead of a cantilever can have advantages. If the SPC as-
sumption can be made, as is the case for most piezoelectric materials used for MEMS,

the sensitivity of a diaphragm transducer can be calculated by simply calculating the

stress in the diaphragm piezoelectric layers and integrating over the electroded area.
A low-frequency model of a circular tensioned plate subject to a uniform pressure load

31
  Without SPC   With SPC 
Pb 1 4 1 3−2α Pb 1 4
Deflection Profile w *
YI 24
x − 6
Lx3 + 12(1−α)
L2 x2 YI 24
x − 16 Lx3 + 14 L2 x2
 
d31n bVn ZQk
Curvature w,xx † − MY VI 2hn sn Y I
 2q q
βn YI 1.8752 YI
Natural Frequency f ‡ 2π ρAc 2πL2 ρAc

* P bL2 Λn 1 P bL2 d31n ZQk
Output Voltage Vn − 6Y I 1−α
− 12Y Iǫ n sn

*
Assuming voltage sensing, low frequency
†
Assuming voltage actuation, low frequency, one piezoelectric layer n
‡
Assuming voltage sensing

Table 2.1: Beam results with and without small piezoelectric coupling

Figure 2.8: Tensioned Diaphragm with assumed boundary conditions.

will be presented. The dynamics will be ignored in the following model because it will
be assumed that the A-weighted noise floor will be dominated by the low-frequency
sensitivity of the device. Figure 2.8 shows a diagram of the diaphragm and assumed
boundary conditions. A circular tensioned plate under a uniform pressure load is

governed by the equation

D∇4 w − N ∇2 w = P (2.51)

for small displacements where w is the lateral displacement of the diaphragm, R is

the diaphragm radius, P is the applied pressure, T is the diaphragm tension per unit
area given as
PN
n=1 Tk hk
T = P N
(2.52)
n=1 hk

32
 PN
N = T· n=1 hk , the diaphragm tension per unit length, and the flexural rigidity of
the plate is
N
1 X Yk
D= ZCk (2.53)
3 n=1 1 − νk2

where
1
Yk = (2.54)
s11k

and
s12k
νk = − . (2.55)
s11k

Following the derivation of Timoshenko [51], the displacement is

h
P R4 I0 ( κr
R
) − I0 (κ)] P R2 (R2 − r 2 )
w= + (2.56)
2κ3 I1 (κ)D 4κ2 D

where r
N
κ=R (2.57)
D

and I is the modified Bessel function. In order to calculate the sensitivity, the stress
must first be calculated. The stresses in the radial and azimuthal directions are

Y
σr = [(ν − 1)ǫr − νǫθ ] (2.58)
(ν + 1)(2ν − 1)
Y
σθ = [(ν − 1)ǫθ − νǫr ] (2.59)
(ν + 1)(2ν − 1)

where

d2 w
εr = (zn − zm), (2.60)
dr 2
1 dw
εθ = (zn − zm), (2.61)
r dr

33
Y is the modulus of elasticity of the layer, ν is the poisson’s ratio of the layer, zn is the
distance to the neutral axis, and zm is the distance to the center of the piezoelectric
layer. After the stress is calculated, the output voltage can be found using equation
2.19, and setting the average electric displacement over the electroded area to zero

(assuming the output of the device has high impedance). This gives

d31 hp
Z
Vout = (σr + σθ )dAe (2.62)
Ae ǫp
Ae

where Ae is the electroded area. After solving for stresses and plugging these values
into equation 2.62, the output voltage is given as

1 P d31 hp (zn − zm) s11 I0 ( κr ) − 2I1 (κ)

Z
R
Vout = dAe . (2.63)
2 Ae ǫp κ2 D (s11 − s12 )(s11 + 2s12 ) I1 (κ)
Ae

The natural frequency of a tensioned circular plate can be found by solving

−γ2 I0 (γ1 R)J1 (γ2R) − γ1 J0 (γ2 R)I1 (γ1 R) = 0 (2.64)

where J and I are the Bessel function and modified Bessel function,

p
N+ N 2 + 4ρhω 2 D
γ12 = (2.65)
2D

and p
−N + N 2 + 4ρhω 2 D
γ22 = [52]. (2.66)
2D

In some cases, the diaphragm rigidity due to bending stiffness is much greater than
that due to tension. If the value of the tension parameter, κ, is less than 2, then

the tension in the plate can be ignored. This simplifies the expression for natural

34
frequency to s
 3.196 2 D
ωn = (2.67)
R ρh

and that for displacement to

P
w= (R2 − r 2 ) [51]. (2.68)
64D

The output voltage then becomes

1 P d31 hp (zn − zm) s11

Z
Vout = R2 − 2r 2dAe . (2.69)
8 Ae ǫp D (s11 − s12 )(s11 + 2s12 )
Ae

A uniformly pressurized diaphragm with clamped boundary conditions will have

stresses of the opposite polarity on the inside (toward the center) versus the outside

(toward the rim) of the diaphragm. If the electrodes covered the entire diaphragm
uniformly, there would be no output voltage because the compressive stress would
cancel with the tensile stress in equation 2.62. If the diaphragm has piezoelectric ele-
ments split where the stress transitions from positive to negative, the output voltage

of either element would be

P d31 hp (zn − zm)s11 R2

Vout = ± . (2.70)
64ǫp D(s11 − s12 )(s11 + 2s12 )

Note that in all equations, 2(zn − zm)hp = ZQk . This substitution makes the form
of equation 2.70 appear similar to equation 2.49.

2.3 Linearity

Linearity is typically measured as a percent total harmonic distortion (THD) at

a specific amplitude. THD is commonly calculated as either a ratio of harmonic to

fundamental power or a ratio of harmonic to fundamental amplitude. The latter is

35
commonly used in audio so THD will be defined as

P
harmonic amplitudes
THD = (2.71)
fundamental frequency amplitude

The piezoelectric element is expected to be extremely linear and so the THD will be
dominated by the non-linearity of the common source amplifier. The total harmonic
distortion of a common source amplifier with a JFET will be presented. To test the

non-linearity of the microphone, high SPLs must be generated without generating

harmonics of the fundamental frequency. This can be done by using a system of
tubes designed to resonate at the frequency of interest but not at harmonics of this
frequency. The design of this system will be presented as well.

2.3.1 THD of a Common Source Amplifier

To calculate the THD, a non-linear FET model must be used. This can be done
by starting with the equation for a FET in the saturation region

ID = β0 (VGS − VT O )2 (2.72)

where βo is the FET transconductance coefficient or Beta factor, VGS is the gate-
source voltage, VT O is the FET threshold voltage, ID is the total drain current, and
channel length modulation has been ignored. Here, VGS is a combination of the DC

bias voltage and the signal giving VGS = VB + vin . This gives the total drain current
as
ID = IQ + 2β0 (VB − VT O )vi + β0 vi2 (2.73)

where IQ is the quiescent current. The output voltage of a common source amplifier
is this current multiplied by the load resistor

vo = 2β0 (VB − VT O )RL vi + RL β0 vi2 (2.74)

36
where the quiescent current has been dropped because it is a constant. A sinusoidal
input voltage of vi = V0 cos(ωt) would result in an output of

1 1
vo = 2β0 (VB − VT O )RL V0 cos(ωt) + RL β0 V02 + RL β0 V02 cos(2ωt). (2.75)
2 2

Because only the second harmonic exists in this model, the THD will be the amplitude
of the second harmonic divided by that of the fundamental and is therefore

V0
T HD = (2.76)
4(VB − VT O )

Since the 2sk3426 JFET has a threshold voltage of -0.312 V and the bias voltage will
be held at 0 V due to the bias resistor between the gate and source, the overdrive
will be 312 mV. This means that a THD of 3% will occur when the input voltage is
at 26.5 mV and a THD of 10% will occur when the input voltage is at 88.2 mV.

2.3.2 THD Measurement Resonator

In order to measure the total harmonic distortion of the microphone, it is necessary

to create a tone at a high SPL while keeping the harmonics of this tone quite low. Most
loudspeakers have significant distortion at high SPLs so a system of resonant tubes
has been designed. This system of tubes would ideally have a resonant frequency in
the range of microphone sensitivity (preferably between 500 Hz-1 kHz) and nulls at

harmonics of this frequency. These nulls will ensure that very little pressure reaches
the microphones at harmonics of the resonant frequency of interest. A two-tube
system has been devised to meet these requirements. A diagram of this system can
be seen in Figure 2.9. In each tube, a right and left traveling plane wave is shown.

The amplitudes of these plane waves will be determined by the boundary conditions.
At x = −L1 , the volume velocity will be determined by that of the speaker. Pressure
and volume velocity continuity apply at x = 0, and the velocity at x = L2 is zero.

37
 Figure 2.9: System of resonant tubes used to measure total harmonic distortion.

These four boundary conditions provide enough information to solve for the pressure
at x = L2 for a given speaker velocity. The frequency response of this system for L1 =
305 mm and L2 = 152.5 mm can be seen in Figure 2.10. The first resonant frequency

of this system occurs at roughly 500 Hz while the amplitudes at 1 kHz, 1.5 kHz, 2
kHz, 2.5 kHz, etc. are relatively low. This means that when the speaker in the front
of the first tube is driven at 500 Hz, the harmonic levels are very low compared to
this fundamental frequency. Therefore, any microphone output at harmonics of 500

Hz is due to microphone distortion.

2.4 Combining Piezoelectric Elements

As mentioned in the introduction, most piezoelectric microphones consist of at

least two separate sensing elements due to the opposing stresses on the outside and
inside of a deflected diaphragm. These two elements can be combined in series or

parallel and, when done optimally, will conserve the output energy of the sensor. In
other words, the square of the output voltage multiplied by the sensor capacitance
will remain constant. A parallel combination of two identical elements will give twice
the sensor capacitance of a single element and the same voltage. A series combination

of two identical elements will give one half the capacitance of a single element but

38
 140
Tube Response (dB re 1 Pa.s/m)

120

100

80

60

40
0 500 1000 1500 2000 2500 3000
Frequency (Hz)

Figure 2.10: Modeled frequency response of two-tube system. Harmonics of 500 Hz

are attenuated.

39
twice the output voltage. In both of these cases, the output energy is the same but
the sensor capacitance is different.
Instead of limiting a design to two sensing elements, several elements can be
used to provide a wide range of sensor capacitances but maintaining constant output

energy. This idea can be used to determine the sensor capacitance which minimizes
the input referred noise. For the example of a common source amplifier given above,
an extremely large sensor capacitance can filter the thermal noise from Rb and the
shot noise to the point that they are negligible and only channel noise dominates,
however, this also reduces the output voltage coming from the sensor significantly.

An extremely small sensor capacitance provides a large output voltage from the sensor
but does not filter the thermal noise from Rb and can form a voltage divider between
the sensor capacitance Cf , and the amplifier input capacitance Cgs so that the output
voltage is attenuated significantly. The optimal capacitance value is one between these

two extremes. For the example of the JFET used in a common source amplifier given
above, the optimal sensor capacitance is 77 pF.
If the sensor capacitance is then fixed at this optimal value, the output noise of the
circuit (ignoring the noise caused by the piezoelectric film resistance) will be fixed.

If the output noise is fixed, the input referred noise is minimized by increasing the
sensitivity of the sensor which is done by increasing the output energy of the sensor in
response to an input. This shows that if piezoelectric film resistance is large enough
to be ignored, a better sensor is one with greater output energy.

Now, it is important to determine how the piezoelectric film resistance affects the
overall circuit noise. This can be done by comparing two sensing systems with two
elements each, one wired in series and one in parallel as seen in Figure 2.11. Here, only
the noise from the piezoelectric film resistance is being considered. Although each
sensing element is identical, the two systems have different sensitivity, capacitance,

resistance, and output referred noise as noted in Figure 2.11 using Laplace notation.

40
The input referred noise (vin ) however, is the same for both systems. This finding is
true for any system of sensing elements regardless of the number of elements as long
as they are combined in such a way as to preserve the output energy.

Figure 2.11: Effect of piezoelectric noise when sensing elements are combined in series
and parallel.

2.5 Microphone Optimization

In the following section, piezoelectric microphones will be optimized using two

different techniques. The first, simpler technique involves computing the output en-
ergy due to an applied acoustic pressure. While this technique is a simple way of
comparing various beam shapes, electrode shapes, layer thicknesses and transducer

geometries, it cannot be used to compare different materials and does not take into
account electrode losses or material loss angle variations with material thickness. The

41
second optimization technique is similar to computing the total device signal to noise
ratio and does include electrode losses and piezoelectric material loss angle. While the
derivation of the optimization parameter is not as straight forward as output energy,
it can be used to compare various materials, different transducer geometries, and can

include all relationships between material properties and thickness.

2.5.1 Cantilever Transducer

2.5.1.1 Output Energy

The ideas presented in the noise section in combination with the sensitivity anal-

ysis shown above can be combined to design a piezoelectric cantilever microphone

with a low noise level. For this design, it is reasonable to assume SPC because
d231 /(ǫs) << 1 for both ZnO, AlN, and most forms of thin film PZT. Any errors
caused by the SPC assumption can be checked after the design has been optimized.

If a cantilever beam is used and SPC is assumed, the output energy of one layer of
a single beam can be computed. It is obvious that the addition of an identical beam
will increase the output energy by two and so it makes sense to normalize the output
energy to sensing area giving

2
Vout Cf b2 L4 d231n ZQk
2
= . (2.77)
2p2 bL 288(Y I)2 ǫn s2n hn

Although this equation does not seem to provide much insight, a few important design

aspects can be seen. First, since Y I includes the width, b, the output energy per area
is independent of beam width as would be expected with the given assumptions. Some
of the parameters in equation 2.77 such as d31 , sn , and ǫn , are material properties
and are, therefore, fixed for a specific piezoelectric material. The design parameters

are basically the length, L, and parameters that are related to layer thicknesses and
relative layer compliances. This equation also shows that the output energy can be

42
increased by increasing the length of the beam but this, according to Table 2.1, will
decrease the natural frequency, decreasing the bandwidth. Therefore, comparisons of
output energy should be done while adjusting the length of the beam such that the
natural frequency is constant.

Gap enlarges as beams bend

Figure 2.12: As the beams bend due to residual film stress, the gap around the beams
increases, reducing the acoustic resistance of the microphone and de-
creasing sensitivity.

The general cantilever layer stack used for the optimization will be a five layer

stack consisting of electrode/piezoelectric/electrode/piezoelectric/electrode layers. It

is important to make the piezoelectric stack symmetrical when building a cantilever
microphone because asymmetry in the stack can cause beam curvature. Beam curva-
ture will increase the gap around the cantilevers, thus providing a path for pressure
equalization and desensitizing the microphone as seen in Figure 2.12. Only the elec-

trode material is used between the two piezoelectric layers because the piezoelectric
material orientation depends upon the surface roughness and orientation [53]. The
addition of a different material may cause the piezoelectric material quality to de-

43
grade and leave the top piezoelectric material useless. This would decrease the output
energy per area by half.
With the cantilever layer stack determined, the output energy per area can be
optimized by adjusting layer thicknesses and beam length while keeping the beam

natural frequency constant. As a modeling example, a beam with Molybdenum (Mo)

electrodes, AlN, and a natural frequency of 20 kHz has been analyzed. The Mo
thickness has been adjusted from 1 nm to 1 µm and the AlN thickness has been
adjusted from 500 nm to 2.3 µm. In this example, all three electrode layers were of
equal thickness as were both piezoelectric layers. The results of output energy per

area can be seen in Figure 2.13. Figure 2.13 shows that the output energy of a five

Figure 2.13: Theoretical output energy per unit sensor area. Thinner electrodes and
thinner piezoelectric layers increase the output energy per sensor area.

layer cantilever beam under a uniform pressure load is increased as the piezoelectric
layer thicknesses and electrode layer thicknesses are reduced. This reduction in layer

44
thickness, however, cannot go on infinitely. As the piezoelectric layer thickness is
reduced, the average grain orientation is less consistent leading to reduced piezoelec-
tric coupling coefficients. This reduction in piezoelectric coupling coefficient was not
included in the calculation of normalized output energy. Additionally, the thinner

layers have a much larger loss angle leading to additional noise in the sensor. The
electrode layers, also, cannot be too thin. As the electrodes become thin, their resis-
tance increases and they start to contribute to the overall noise as discussed in the
noise section. Both thicknesses are also limited by manufacturability.

2.5.1.2 Optimization Parameter

To further investigate the limit to which these material thicknesses can be reduced,

the total signal to noise ratio can be examined. For the same reasons given above,
this signal-to-noise ratio will be normalized by the device area and natural frequency
giving the following optimization parameter

2
Vout C
Ψ= f2 (2.78)
P As tan(δ) res
2

where Vout is the voltage out of the device, C is the capacitance of the device, fres

is the resonant frequency of the device, P is the applied pressure, As is the device
area and tan(δ) is the device loss angle. The only difference between this parameter
and optimizing the normalized output energy is that it includes the loss angle. The
square of the output voltage multiplied by the capacitance can be taken from equation

2.51. The resonant frequency can be taken from Table 2.1. The loss angle is given in
equation 2.15. Putting these equations together, assuming the frequency in the loss
angle calculation is equal to the resonant frequency and removing constants gives,

1.8754 bd231p ZQ2 p

Ψ= h i. (2.79)
576π 2 Y Iǫ s2 h PN ρ h tan(δ) +
q
2∗1.8752 ǫp
p p p PNY I
i=1 i i p 3σe hp he b ρi hi
i=1

45
While this model captures all the critical design variations, the material properties
in the equation are constants while in reality, they will vary with material thickness.
Most notably, both d31 and tan(δ) tend to degrade as the film becomes extremely
thin. These material property changes can be included by experimentally measuring

their value as a function of material thickness, fitting a function to the data, and
inserting this value in the optimization parameter calculation.
The Mo and AlN properties as a function of thickness are plotted in Figures
2.14(a)-2.14(c). Here, the AlN values have been taken from a study measuring AlN
properties as a function of material thickness on platinum electrodes [54], and the

Mo resistivity has been calculated using a model from the literature [49]. The values
from these plots along with the other, less variable material properties can be used
to calculate the optimization parameter as a function of layer thickness. When using
the five layer device consisting of three Mo electrodes and two AlN layers used to

generate Figure 2.15, the optimal sizing is seen in Figure 2.15.

Instead of the continual increase in performance with reduction in layer thickness
shown in Figures 2.14(a)-2.14(c), Figure 2.15 shows that an optimal device will have
10 nm thick Mo electrodes and 1.5 µm thick AlN layers. These values, however, will

vary because AlN loss angle and piezoelectric coefficients will depend on the tool
used, the configuration of the tool, the electrode material used, surface roughness,
and many other factors. For example, while the plot in Figure 2.14(b) shows that
AlN of 0.5 µm thickness will have a loss angle of roughly 0.0034, the same author

has deposited 0.5 µm thick AlN with a loss angle of 0.001 on Mo electrodes [55].
Additionally, manufacturing devices with 10 nm electrodes will be difficult without
the use of advanced etch stop techniques or highly selective etches so thicker electrodes
may be more practical.

46
 −3
200 x 10
8
Mo Resistivity (µΩ.cm)

150
6

Dissipation Factor
100 4

50 2

0 0
0 20 40 60 80 100 0 0.5 1 1.5 2
Mo Thickness (nm) AlN Thickness (µm)

(a) Modeled Mo resistivity as a function of (b) AlN loss angle as a function of AlN
Mo film thickness. film thickness extrapolated from measure-
ments.
2.8

2.7

2.6

2.5
d31 (pC/N)

2.4

2.3

2.2

2.1

2
0 0.5 1 1.5 2
AlN Thickness (µm)

(c) AlN piezoelectric coupling coefficient

as a function of AlN film thickness extrap-
olated from measurements.

Figure 2.14: Material properties of Mo and AlN as a function of layer thickness.

47
Figure 2.15: Optimization parameter, Ψ, as a function of electrode and piezoelectric
layer thicknesses. The model gives optimal thicknesses around 1.5 µm
AlN and 10 nm Mo.

2.5.1.3 Electrode Sizing

Not only can the output energy and optimization parameter presented in the

previous sections be used to optimize layer thicknesses but they can also be used to
optimize electrode coverage. Because changes in the electrode coverage cause only
small variations in the noise power (shorter electrodes would have slightly less total
loss), it makes sense to use the output energy to determine optimal electrode coverage

for a cantilever microphone.

An applied acoustic pressure will induce the greatest amount of stress at the base
of the cantilever, where the curvature is the largest. At the tip of the cantilever,
the curvature, and therefore the stress, is small. If the electrode covers the beam

48
 2

Output Energy per Sensor Area (J/Pa m )

2
2.
1.5

.5

0
0 20 40 60 80 100
Percent Length Covered by Electrode (%)

Figure 2.16: Theoretical effect of electrode length on normalized piezoelectric output

energy. Decreasing the electrode length to around 45% increases the
output energy by 54%.

from the base to the tip, the electroded area towards the tip of the beam acts as stray
capacitance, reducing the sensitivity and therefore the output energy. If the electrode
only coveres the first few microns towards the base of the beam, the output voltage

would be very high as this is the area of greatest stress however, the output energy
would suffer due to the small area and therefore low device capacitance.
Figure 2.16 illustrates this trade-off by adjusting the electrode coverage for a
cantilever-based microphone with 1 µm thick AlN layers and 100 nm thick Mo elec-

trodes. As shown in the Figure, if the electrode covers about 54% of the beam length,
meaning that the electrode extends from the base of the beam to 54% down the length
of the beam, then the output energy is maximized. By reducing the electrode cov-
erage from the full length of the beam to the optimal value, the noise floor would

be reduced by about 2 dBA. Similar results can be obtained for a diaphragm-based

microphone.

49
 ZnO PZT AlN
d31 (C/N) −5.74 × 10−12 [39] −40 × 10 −12
- −70 × 10−12 [36] −2.65 × 10−12 [40]
d33 (C/N) 10.4 × 10−12 [39] 92 × 10−12 -203 × 10−12 [36] 5.53 × 10−12 [40]
s11 (P a−1 ) 9.63 × 10−12 [39] 9.52 × 10−12 -14.9 × 10−12 [41] 3.53 × 10−12 [40]
tan(δ) 0.01-0.1 [36] 0.03-0.05 [42] 0.002 [42]
ǫ33 /ǫo 10.9[36] 1100-1300 [42] 10.4 [42]
ρ (kg/m3 ) 5670 7600 3300

Table 2.2: Range of properties of common MEMS piezoelectric materials.

2.5.1.4 Piezoelectric Materials

Although output energy can be used to make several design decisions, it cannot
be used to compare piezoelectric materials because it does not consider loss angle.
If the noise of the amplifying circuitry is low, piezoelectric material with a large
loss angle will be the dominant source of noise. Additionally, piezoelectric material

properties depend on the electrode material upon which they are deposited so com-
parisons cannot be made by simply changing the piezoelectric material parameters.
The optimization parameter is therefore necessary to determine which piezoelectric
and electrode material combination is best. The material parameters taken from var-

ious sources in the literature can be seen in Table 1.1 which has been reproduced in
this chapter for convenience.
Because ZnO and AlN have similar piezoelectric coupling coefficients but the loss
angle of ZnO can be 100 times larger than that of AlN, this comparison will focus

primarily on AlN and PZT. If PZT is used in place AlN and Platinum (Pt, commonly
used as an electrode material for PZT) is used in place of Mo, the normalized output
energy of a beam with two 1 µm thick PZT layers, three 100 nm thick Pt electrodes,
and a resonant frequency of 20 kHz is 1.17 × 10−11 J/Pa2 m2 when using properties in

the middle of the ranges given in Table 2.2. This normalized output energy is slightly
lower than that of a beam with equal thickness AlN and Mo layers with the same
resonant frequency which has a normalized output energy of 1.25 × 10−11 J/Pa2 m2

50
as can be seen in Figure 2.13. If comparing only output energy, one would assume
that PZT would perform about the same as AlN but the lower loss angle of AlN gives
it a much lower noise floor.
This lower loss angle is taken into account by using the optimization parameter for

comparison. For the same Pt/PZT beam given above, the optimization parameter,
Ψ = 0.11 m2 /kg while that for the AlN/Mo beam is Ψ = 2.63 m2 /kg. This analysis
indicates that the beam utilizing AlN and Mo will be much better. When the model
for sensitivity is combined with a circuit model of a JFET (2sk3426) in a common
source amplifier circuit, assuming a sensor area of 1 mm2 and a capacitance of 62 pF,

the input referred noise is predicted to be 36.5 dBA for the AlN/Mo beam and 45.8
dBA for the PZT/Pt beam.

2.5.1.5 Other Considerations

Both Figures 2.13 and 2.15 show the motivation behind using MEMS technology.
As the layer thicknesses are increased, keeping the ratio between layers the same,

the output energy drops off significantly. This indicates that a high performance
piezoelectric cantilever microphone cannot be built without the use of extremely thin
material layers such as those possible with MEMS techniques.
The analyses used to create Figures 2.13 and 2.15 ignore stray capacitance. Stray

capacitance will, undoubtedly exist in the traces on the chip, bond pads, and beam
anchors but should be small for well designed systems. Fluid loading was, however,
considered in the generation of these plots. The fluid loading is a function of sensor
area and so, a sensor area of 1 mm2 was assumed. Fluid loading can be included by
adding some height of material with the density of air above and below the beam.
√
This height can be approximated as hair = 8 AS /3π where As is the sensor area
[14, 46]. Fluid loading may be significant and will have a greater impact as beam
thickness is decreased.

51
 Despite not including stray capacitance, the noise floor can be estimated by calcu-
lating the output energy of the optimal device from Figure 2.15. Take, for example, a
microphone that has a Mo thickness of 10 nm, an AlN thickness of 1.5 µm, generates
output energy of 1.5 × 10−11 J/Pa2 m2 and has an area of 1 mm2 . The loss angle of

AlN given in Table 1.1, in combination with the small signal model given in Figure
2.3 and an assumed sensor capacitance of 62 pF can be used to determine that this
microphone would have a noise floor of 33 dBA. Judging from Figure 1.1, this noise
floor would be extremely low for a piezoelectric microphone with 20 kHz resonant
frequency.

2.5.2 Diaphragm Transducer

The use of a diaphragm instead of a cantilever is quite common for piezoelectric

MEMS microphones [1, 44, 11, 12, 13]. For materials for which SPC is a valid assump-
tion (criteria are slightly different for beam and diaphragm models), it is fairly simple
to compare the output energy of a low stress diaphragm to that of a beam using

equation 2.69. If the diaphragm is assumed to have low residual stress and act as a
plate, thinner layers increase the output energy and, when optimized, the diaphragm
provides more normalized output energy than a cantilever. However, when the layers
become thin, residual stress becomes more likely to dominate the structural stiffness.

For a Mo/AlN diaphragm with 100 nm electrodes and 1 µm AlN, the diaphragm
residual stress must be less than 1.1 MPa to consider the diaphragm to be a plate
[52]. The device normalized output energy is plotted against diaphragm stress for a
five layer diaphragm with two 1 µm AlN piezoelectric layers and three 100 nm Mo
electrodes in Figure 2.17. The diaphragm radius was adjusted to keep the natural

frequency at 20 kHz. Residual stress below 1MPa is very difficult to achieve and is
roughly equivalent to the resolution of common stress measurement tools [56]. As can
be seen from Figure 2.17, increasing residual stress decreases the normalized output

52
 −11
x 10

Normalized Output Energy (J/Pa2.m2)

4

0 4 6 8
10 10 10
Diaphragm Stress (Pa)

Figure 2.17: Models indicate that even small amounts of diaphragm stress signifi-
cantly degrade the normalized output energy of a diaphragm of constant
bandwidth

energy which will lead to an increased noise floor. Inconsistency in stress will also
lead to inconsistency in sensitivity and bandwidth. For these reasons, any diaphragm
based device must have low residual stress. If low residual stress cannot be achieved
in the deposition tool, residual stress can be decreased after deposition by partially

removing the diaphragm from the substrate and then reattaching it.

2.6 Packaging

The packaging requirements of a piezoelectric MEMS microphone are similar to

those of capacitive MEMS microphones. The basic packages are shown in Figure 2.18.

Both of these packages are similar to those used by commercial MEMS microphone
manufacturers. The key aspects of these packages are an acoustic input, a back cavity
volume behind the transducer, and a chip for buffering or amplification. There are
many other methods of providing a package with these key aspects. These packages
will be modeled along with the transducer to show the influence that packaging has

53
 (a) (b)

Figure 2.18: Two common embodiments of MEMS microphone packages.

Figure 2.19: Package and device equivalent circuit. Includes any front cavity reso-
nances, device response, and back cavity compliance.

on the overall microphone performance.

An equivalent circuit for the microphone and package can be seen in Figure 2.19.
Because this is a lumped element model, only the first resonance of the microphone

is included. In Figure 2.19, the front cavity may or may not be present depending on
the packaging design. If using a design with a front cavity, the air mass moving in the
acoustic input combines with the front cavity compliance to form a Helmholtz res-
onator [14]. The response of this Helmholtz resonator can be modeled using standard
Helmholtz resonator equations [14] where lumped acoustic impedance parameters are

54
given as

ρ0
Lf ront = (L + 1.7R) (2.80)
πR2
∀f ront
Cf ront = (2.81)
ρ0 c2
ρ0 ck 2
Rf ront = (2.82)
4π

In these equations, ρ0 is the density of air, R is the radius of the acoustic input, L is
the thickness of the acoustic input, Vf ront is the front cavity volume, c is the speed of

sound (343 m/s in air at room temperature), and k is the wave number (k = ω/c).
The cantilever or diaphragm will also have a dynamic response as modeled by Rt ,
Ct , and Lt . Because the transducer is now approximated as a second order system, the
transducer model will only have a single resonant frequency. The acoustic impedance

parameters for the transducer can be found as a function of diaphragm or cantilever

displacement

R
As
w(x, y)dAs
Ct = (2.83)
P
w(x, y)2dAs
R
As
Lt = R
2 (2.84)
As
w(x, y)dAs

Here, the mass loading can be included by adding an equivalent height of material
with the density of air above and below the transducer. This equivalent height is

8R
hair = (2.85)
3π

for a circular diaphragm [14]. Similarly, the acoustic resistance of a small diaphragm

in air at acoustic frequencies can be modeled as

1
Rr = ρ0 ck 2 (2.86)
4π

55
The acoustic resistance to a volume of air traveling through the transducer will be
determined by the size of hole in the diaphragm or gaps around the cantilever beams.
Assuming the cantilever beams are flat, this acoustic resistance is given by

12µh
Rt = (2.87)
g3l

where µ is the coefficient of shear viscosity (1.85 × 10−5 Pa. s for air at room temper-

ature), h is the total thickness of the cantilever, g is the width of the gap around the
beams, and l is the total length of the gap, and l >> g.
Like the front cavity, the back cavity also has an acoustic compliance.

∀back
Cback = (2.88)
ρ0 c2

If the back cavity has some channel connecting it to atmospheric pressure, it will also
have an acoustic resistance. If this channel is rectangular, the resistance will be

12µI
Rback = (2.89)
h3 w

where l is the length of the channel, h is the height of the channel, w is the width of
the channel, and w >> h.
Typically, it is desirable to make the packaging as transparent as possible. This
means that, if the package has a front cavity acting as a Helmholtz resonator, the res-

onant frequency of the resonator is set above the frequencies of interest. Commercial
capacitive microphones typically set this resonance around 15 kHz. In some cases,
the back cavity compliance can be in the same range as that of the transducer. This
is more common for cases in which the back cavity is formed by the back side of the

MEMS chip. If the back cavity is not compliant enough, the equivalent circuit will,
at intermediate frequencies, appear as a capacitive divider. This becomes a concern

56
as the transducer compliance becomes large, as is the case with the cantilever design.
For example, a cantilever microphone with three 100 nm Mo electrodes and two 1
µm AlN piezoelectric layers requires a length of 384 µm to have a 20 kHz resonant
frequency. A design consisting of two 384 µm long, 768 µm wide beams will have an

acoustic compliance of 2.13 × 10−15 m3 /Pa. The back cavity compliance of a 768 µm
× 768 µm × 500 µm volume is 2.07 × 10−15 m3 /Pa. This low back cavity compliance
would reduce the signal by roughly 50%, therefore adding 6 dB to the noise floor and
reducing the sensitivity by the same amount. It is therefore recommended that the
cantilever design be used in a package where the back cavity volume is roughly that

of the package. If the package is 3.75 × 4.75 × 1 mm and 50% is filled with air, the
back cavity compliance would be 62.6 × 10−15 m3 /Pa. This compliance is roughly
30 times larger than that of the transducer and would have very little impact on the
noise or sensitivity.

The diaphragm design is less compliant than the cantilever design making the
above described performance degradation less severe. The same Mo/AlN stack given
above would require a diaphragm with a 641 µm radius to achieve a 20 kHz resonant
frequency (assuming the stress in the diaphragm is low). This diaphragm would have

a compliance of 2.22 × 10−15 m3 /Pa. The back cavity, consisting of the volume behind
the diaphragm, would have a compliance of 4.54 × 10−15 m3 /Pa if a 500 µm thick
wafer was used and 5.90 × 10−15 m3 /Pa if a 650 µm thick wafer were used. Therefore,
using a 650 µm thick wafer would result in only a 3 dB reduction in sensitivity and

3 dB increase in noise floor.

2.7 Vibration Sensitivity

An ideal microphone would be completely insensitive to vibration and sense only

acoustic pressure. This, however, is not the case whenever a moving element with
non-zero mass is used for transduction. The sensitivity to vibration, however, is fairly

57
simple to calculate. In order to calculate the sensitivity of a diaphragm or cantilever
beam to vibration, it is only necessary to compute the acceleration that would result
in the same force as an applied acoustic pressure. This acceleration depends only on
the mass of the diaphragm or cantilever.

When modeling vibration sensitivity instead of pressure sensitivity, a uniform

pressure load, P , can be replaced by an acceleration, a, using the equation

N
X
Pequiv = a ρi hi (2.90)
i=1

where the device has i different layers of density, ρ, and height, h. This equation

indicates that the vibration of a diaphragm or cantilever with a high mass per area
would be sensed as a large pressure while one with a low mass per area would expe-
rience very little cross-sensitivity to vibration. The mass of the air above and below
the diaphragm can be added into the summation for improved accuracy but typically

contributes little to the overall sensitivity.

As an example of vibration sensitivity, a five layer microphone consisting of two 100
µm AlN layers and three 100 µm Mo electrodes with an area of 1 mm × 1 mm would
have a vibration sensitivity of 0.011 Pa·s2 /m or 0.11 Pa/g. This means that it would
require an acceleration of 9.4 g to equal 1 Pa. When compared to other microphones,

this vibration sensitivity is better than that of large capacitive microphones and
electret microphones but slightly worse than capacitive MEMS microphones with
lower mass diaphragms.

58
 CHAPTER III

First Generation Device

3.1 Introduction

This section will cover the design, fabrication, and testing of the first genera-

tion piezoelectric MEMS microphone built for the present study. The design was
based on the models developed in Chapter II. Many of the dimensions were rela-
tively conservative due to fabrication uncertainties while others aspects of the design
were overly optimistic. The fabrication was a four mask process which kept the fab-
rication relatively simple and quick but sacrificed some performance. The methods

used to measure the piezoelectric material properties and microphone performance

will be summarized here and used in Chapter IV as well. After measuring the piezo-
electric material properties, the microphone performance measurements matched the
performance anticipated by the model.

3.2 Design

The models and optimization in Chapter II indicate that high performance piezo-
electric MEMS microphones are obtained for thin piezoelectric and electrode layers.
Because high quality AlN has been deposited on Mo by others [57, 55], Mo was se-
lected as the electrode material. Other materials such as Ti and Pt were investigated

59
but Ti did not work with the rest of the processing and Pt was denser than Mo,
leading to reduced microphone performance. A Mo thickness of 200 nm was selected
because the AlN etch had not been developed and the ability to etch through the AlN
and stop before removing all Mo was crucial to device operation. An AlN thickness of

0.5 µm was selected because thinner AlN films led to better microphone performance
and high quality AlN films at 0.5 µm had been demonstrated. The beam lengths
were designed to provide microphone bandwidths spanning the audible range and the
electrode length was equal to the beam length because the same mask was used to
pattern both the AlN and Mo. In some cases, 24 beams extended over a large hole in

the silicon while in other cases, only four wider beams covered the same area. Fewer
beams is better because this creates fewer gaps leading to a larger acoustic resistance
but it was feared that extremely wide beams may bend and twist due to film stresses.

3.3 Fabrication

Piezoelectric beams of lengths ranging from 200 µm to 450 µm and widths of 30

µm to 300 µm were constructed. These beams had a five layer stack of materials
consisting of 0.2 µm Mo/0.5 µm AlN/0.2 µm Mo/0.5 µm AlN/0.2 µm Mo. The AlN
layers were deposited in an AMS AlN sputtering tool at the University of California
Berkeley facilities by Harmonic Devices. All other fabrication was performed in the

University of Michigan Lurie Nanofabrication Facility (LNF).

The fabrication followed the steps shown in Figure 3.1. First, a 200 nm thermal
oxide was grown on a bare Si wafer and 200 nm of Mo was sputtered and patterned
using the first mask. All Mo layers were etched with a dilute mixture of Aqua Regia

(9 H2 O : 3 HCl : 1 HNO3 ). Next, AlN and Mo were deposited and patterned with the
second mask. All AlN layers were etched with hot (85◦ C) phosphoric acid. This AlN
etch caused significant undercutting of the top Mo electrodes. These same AlN and
Mo depositions were repeated and then patterned with the third mask. The fourth

60
 Figure 3.1: Four step beam fabrication process.

Figure 3.2: Microscope photograph showing the top view of twenty piezoelectric
beams. These beams have five material layers on them and are released
from the Si substrate.

mask was then used to pattern the back side hole which was etched through the wafer
with deep reactive ion etching (DRIE). When completed, the beams appeared as seen

in Figure 3.2.
When finished, the wafer was diced into 1.5 mm × 1.5 mm chips. Very few beams
broke during processing. One drawback of this process is that the cross-wafer non-
uniformity of DRIE means that the beam length varies from chip to chip. As the

length varies, so does the sensitivity and bandwidth of the microphone. Another
drawback of this process is that, because only four masks were used, the Mo cannot
be patterned independently of the AlN. This increases stray capacitance and, if the

61
 Figure 3.3: Packaged microphone wire bonded to JFET in TO can.

AlN etch undercuts too much, the two Mo layers can short together.
After dicing the wafer into chips, the chips were glued onto a transistor outline
(TO) header with a hole drilled behind the chip. A JFET (2sk3426) was also glued
onto the TO header and the chip was wire bonded to the JFET. Finally, a hole

was drilled into a TO cap which was placed on the header. When completed, the
microphone appeared as pictured in Figure 3.3.

3.4 Testing

After packaging, the microphones underwent a series of tests to determine mate-

rial properties and microphone performance. First, the loss angle of the films was
measured on both cantilevers and test capacitors using both an Agilent LCR meter
and custom circuitry at 1 kHz. Unfortunately, the loss angle of the films tended
to vary from about 0.01 to 0.1. It is believed that the loss angle of these films is
significantly above that quoted in the literature for AlN because they are only 0.5

µm thick. Conversations with AlN tool manufacturers and with Harmonic Devices
have indicated that X-ray diffraction rocking curves typically improve until the films

62
 −8
x 10
8
Measurements
Quadratic Fit
6

Deflection (m/V)
4

0
0 1 2 3 4
Distance (m) −4
x 10

Figure 3.4: Beam actuation displacement measurements and quadratic fit used to
calculate d31 .

reach 1-1.5 µm. Therefore, it is expected that these 0.5 µm films have degraded

performance.
The d31 coefficient of the films was then measured by actuating the beams and
measuring the beam deflection at several points along the length of the beam using
a Polytec laser vibrometer as seen in Figure 3.4. These measurements were used to

extract a resulting beam curvature and Equation 2.34 was used to calculate d31 . The
d31 coefficient varied from -1 to -1.8 pC/N. This is slightly less than the d31 coefficient
typically quoted in the literature and is consistent with what would be expected given
the 0.5 µm films.

After measuring the d31 coefficient, the beams were exposed to a uniform pressure
and the displacement and output voltage were measured. This was accomplished by
using a plane wave tube as shown in Figure 3.5. A small speaker was placed at one
end of the tube and the microphone and a calibrated Larson Davis 2520 reference

microphone were placed equidistant from the speaker through holes in the side of the
tube. Foam was placed in the end of the tube to provide an anechoic end condition.
A Polytec laser vibrometer was focused on the back of the beams through the hole in

63
Figure 3.5: Test set-up used for measuring beam deflection and output voltage re-
sulting from an applied acoustic pressure.

the TO header. This way, the beam displacement profile can be measured, as shown
in Figure 3.6. A circuit with a 2.2 kΩ resistor and a 3 V supply was connected to
the TO can to complete the common source amplifier. The output voltage of the

microphone as a function of frequency was then measured as shown in Figure 3.7.

Figures 3.6 and 3.7 show that the beam displacement and microphone sensitivity
match the models quite well when the measured d31 coefficient is used. The noise
floor can then be measured by placing the microphone in a quiet, anechoic environ-

ment and measuring the signal out of the amplifier. The noise out of the amplifier,
when connected to capacitors can be seen in Figure 3.8. The noise models given in
Chapter II have been validated by placing low loss capacitors in place of the sensor
and measuring noise. The piezoelectric noise dominates this sensor because the loss

angle of the film is high.

3.5 Results

When the microphone measurements were completed, microphones with noise

floors around 58 dBA were measured. Although these microphones were small, 2 ×
10−7 m2 , the noise floor is only slightly better than average for piezoelectric micro-

64
 −8
x 10
3
Measured
2.5
Model
2

Deflection (m/Pa)
1.5

0.5

−0.5
0 1 2 3 4
Distance (m) −4
x 10

Figure 3.6: Beam deflection profile resulting from a uniform pressure load. The laser
measurements towards the base of the beam were obstructed by TO can.
Sensitivity (dB re 1 V/Pa)

−60

−80
Measured
−100
Model
2 3 4
10 10 10
Phase (degrees)

100

−100

2 3 4
10 10 10
Frequency (Hz)

Figure 3.7: Amplitude and phase of the frequency response of the beams resulting
from a uniform pressure load.

65
 −110
Measurement
−115 Model

Noise Level (dB re 1V/√Hz)

−120

−125

−130

−135

−140

−145

−150 2 3 4
10 10 10
Frequency (Hz)

Figure 3.8: Measured and modeled output referred noise of common source amplifier.

phones of this size. This noise floor is somewhat limited by the piezoelectric coupling
coefficient but mostly limited by the large loss angle. The inconsistency of films also
prevented the use of both the top and bottom piezoelectric layers for sensing. The use

of both layers will double the output energy and decrease the noise floor by roughly
3 dB.
The testing of this device provides validation that the sensitivity of the cantilever
beam and the gain of the JFET were both modeled accurately. This testing also

demonstrates, for the first time, that the piezoelectric noise of a piezoelectric MEMS
microphone can be a dominant source of overall transducer noise. This first generation
device also highlighted the difficulty of obtaining high quality AlN even when using
a tool designed for AlN deposition. The etching of AlN also proved to be more
difficult than anticipated. While the hot phosphoric acid etch provided a good way of

etching AlN without etching Mo, the undercutting caused the Mo electrodes and leads
between electrodes to be significantly undercut. This undercutting likely caused the
electrodes to be shorted in some areas and may have been related to high loss angle
measurements in others. Despite these difficulties, this first generation device proved

66
that several AlN based cantilevers could be fabricated with small enough deflection
for use as a microphone. This device also showed that wide beams could be built
without concern for beam curling or twisting. These lessons learned are applied to
the second generation device discussed in Chapter IV.

67
 CHAPTER IV

Second Generation Device

Many lessons were learned from the first AlN microphone design. First, it became

clear that obtaining high quality AlN was not as simple as depositing the material
with the right tool. It was hypothesized that the AlN loss angle was so high in the
first design for three reasons. First, the AlN was only 0.5 µm thick and while some
studies shown that high quality AlN can be deposited at 0.5 µm thickness [55], others
show significant degredation of material properties at this thickness [54]. Second, the

wet etching of AlN caused significant undercutting of the electrode material, often
causing the electrode to crack or become discontinuous in narrow areas. As the
electrode connections become narrow, the resistance increases, thereby increasing the
loss angle. Third, it was thought that the AlN near the etched edge would be of poor

quality and therefore have reduced performance. If this material was extremely lossy,
it could increase the loss angle of the entire device.
The changes to the second design were primarily aimed at improving the loss angle.
Two additional masks were added to the process so the Mo could be patterned sepa-

rately from the AlN. This allowed the AlN to extend beyond the Mo electrodes so any
AlN near an etch boundary would have a diminished impact on device performance.
These additional mask layers also allowed for the electroded area to differ from the
beam outline. By extending the electroded area over the first 50% of the beam length,

68
performance can be improved as shown in Figure 2.16 in Chapter II. The ability to
pattern Mo separately from AlN also allowed for the design of a top interconnection
layer to be used to connect various layers prior to wire bonding. Additionally, AlN
thicknesses of 1.0 µm was deposited to increase the piezoelectric coupling coefficient

and, more importantly, decrease the loss angle from the devices of the first gener-
ation. Improvements in material properties were also sought by improving the Mo
properties. Studies have shown that the properties of the electrode material affect
the AlN quality and models in Chapter II show that Mo of higher conductivity can
improve the device loss angle. For these reasons, the Mo deposition parameters were

adjusted to improve the Mo orientation and conductivity. Because the Mo of design

two was of higher conductivity and because stopping the AlN etch on 200 nm of Mo
was not a problem, the Mo thickness of design two was reduced to 100 nm.
The second microphone design also used two large flaps rather than several beams

for the cantilevers. This change was made to reduce the total gap around the beams,
increasing the acoustic resistance through the device. This change was aimed at
reducing the necessary back cavity volume.

4.1 Fabrication

Using the design considerations of Chapter II along with lessons learned from

the first device presented in Chapter III, cantilever microphones were built with Mo
electrodes and AlN. These devices consisted of two wide piezoelectric beams with
three, 100 nm Mo electrodes and two 1.0 µm AlN layers. As in design 1, the AMS
AlN sputtering tool at the University of California Berkeley was used to deposit all

AlN. This was again done by Harmonic Devices Inc. A 10-15 nm AlN seed layer
was also used below the first Mo electrode because Tegal, the manufacturer of the
AMS sputter tool used by Berkeley, had suggested it would improve Mo and AlN
orientation. All processing other than the AlN deposition was performed at the

69
Figure 4.1: Piezoelectric MEMS microphone fabrication process. This is a 7 mask
process without boron implantation and a 9 mask process with boron
implantation.

University of Michigan LNF.

The fabrication followed the steps shown in Figure 4.1. In some cases, a deep
boron implantation was used to pattern the front and back sides of the wafer. This

boron implantation would later act as an etch stop, setting the beam length. The
boron doping was used because cross-wafer variations in deep reactive ion etching
(DRIE) led to variations in beam length in the previous generation of devices. Some
devices did not use this implantation and, in these cases, a slower, high aspect ratio

DRIE etch recipe was used to reduce the cross-wafer variation experienced in the
first generation of devices. After boron implantation, 400 nm of thermal oxide was
grown on the front and back sides of the wafer and the 10-15 nm AlN seed layer was
deposited.

70
 The first 100 nm Mo layer was sputter deposited at a pressure of 3 mTorr and 750
W power on the AlN. This layer was measured as 98 nm thick using a VEECO atomic
force microscope (AFM) with an RMS resolution of 0.05 nm. This nominally 100 nm
thick Mo layer had an average resistivity of 28 µΩcm measured by an FPP5000 four

point probe and an average roughness of 0.592 nm RMS measured by an AFM. This
Mo has a residual stress that varied from wafer to wafer but was approximately -
700 MPa, where negative indicates compressive. Next, this Mo layer was patterned
with dilute Aqua Regia (9:3:1 H2 O:HCl:HNO3 ) using a photoresist mask. The first
AlN layer was then deposited on the patterned Mo. This AlN layer had an x-ray

diffraction (XRD) full width half maximum (FWHM) of 1.8◦ for 1.0 µm thick layers
and 2.2◦ - 2.6◦ for 0.5 µm layers. The first AlN layer had a roughness of 2.01 nm
RMS. The residual stresses in these layers were controlled by adjusting Ar and N2
flow rates and the target stress was around 220 MPa for the 1.0 µm layer. The stress

of the deposited material tended to vary due to inconsistencies in boron doping but
was typically between 100 and 300 MPa.
After the first AlN deposition, the second Mo layer was deposited and patterned in
an identical manner to the first. This second Mo layer had a slightly larger resistivity

of roughly 35 µΩcm and a roughness of 2.02 nm RMS but the stress was almost zero.
This low stress is significantly different than the extremely compressive stress of the
first layer. The second AlN layer was then deposited in a manner identical to that of
the first AlN layer. This AlN layer had an XRD FWHM of 2.0◦ for 1.0 µm thick layers

and 2.3◦ to 2.6◦ for 0.5 µm thick layers. The top Mo layer was then deposited and
patterned in the same manner as the first two and exhibited properties similar to the
second. The AlN was then etched to form vias for metal interconnects and to form
the beams. The first AlN etch went through only the top AlN layer and stopped on
the middle Mo surface. This etch was performed in a LAM 9400 inductively coupled

plasma etcher. The second AlN etch was performed in a similar manner but etched

71
through both AlN layers and stopped on the bottom electrode. In some cases, this
etch was also used to form the beams while in other cases, a third AlN etch was used
to form the beams. After all AlN etching, a top metal layer was deposited for metal
interconnects. This layer was typically a 50 nm/500 nm Cr/Au layer. The Cr/Au

was selected for its high conductivity.

Following the Cr/Au deposition, 12 µm of PECVD oxide was deposited on the top
of the wafer and 3 µm PECVD oxide were deposited on the back side. The top oxide
was deposited to protect the beams while dicing. The bottom oxide is then patterned
and etched. This back side pattern aligns either the DRIE or TMAH etch to the

front side of the wafer. If the wafer does not have boron doping, a DRIE pattern was
always used. DRIE or TMAH were then used to etch from the back side of the wafer
to the 400 nm thermal oxide layer on the front side of the wafer.
At this point, the beams are surrounded by oxide but free from the Si substrate.

The wafer is then diced on a dicing saw. The dicing saw blade depth is slightly less
than that of the wafer in order to prevent the saw from completely separating the
chips. The wafer is then placed in a 7:1 buffered oxide etch (BOE) with surfactant to
remove all oxide. The surfactant helps the BOE get to the bottom thermal oxide. The

thin AlN seed layer is not specifically removed but is thin enough that it either breaks
or is etched away in the BOE. After all oxide is removed, the chips are separated along
the dicing lines and packaged. When finished, the devices appear as shown in Figure
4.2.

The packaging was also redesigned after the first iteration. The packaging consists
of a 3.76mm × 4.72mm patterned PCB, several of which are shown in Figure 4.3.
The top side of the PCB is patterned for solder reflow of either a JFET, an amplifier
similar to the National Semiconductor LMV1012, or allows the output of the MEMS
transducer to be connected directly to the output pads. In some cases, a hole is

drilled behind the MEMS transducer to allow the back cavity to be arbitrarily large

72
Figure 4.2: After fabrication is complete, the released beams appear as shown. The
large compressive stress in the first electrode layer caused the beams to
bend upward and are slightly out of focus.

Figure 4.3: Several different PCB designs were fabricated for use with a variety of
buffering techniques. Some designs have a hole behind the transducer in
order to provide a larger back cavity volume.

73
Figure 4.4: Packaged microphone transducer and JFET for signal buffering. There is
no cap on this package to avoid any package resonance in the frequency
response measurement.

or to allow the acoustic input to be placed behind the sensor. After separating a
transducer chip from the wafer, the chip is glued to the PCB with Epo-tek T7139

electrically insulating epoxy. The chip is then wire bonded down to the PCB surface.
Following wire bonding, the readout chip, usually a JFET, is attached to the PCB
via solder reflow. After solder reflow, the microphone appears as shown in Figure 4.4.
The back side of the PCB has pads for the signal output. In the case of the PCB

with the JFET, there are pads for ground and output. The transducer and JFET are
connected by wiring two 1.5 V batteries through a 2.2 kΩ resistor to the output pad
and grounding the ground pad as shown in Figure 2.4. The output of the JFET can
then go directly into a high impedance device for sound recording or can go through a
preamplifier for additional signal buffering and electromagnetic interference rejection.

74
Figure 4.5: Four test capacitors. The capacitors on the left side of the figure are used
to measure the bottom AlN film while those on the right side are used to
measure the top AlN film

4.2 Testing

After packaging, the microphone and other test structures underwent tests to
determine material properties and microphone performance parameters of interest.
Most of the test procedures were covered in Chapter III, all other test procedures and

results are detailed in this testing section.

4.2.1 Permittivity and Loss Angle

In order to test the hypothesis that low quality AlN near etch borders caused
increases in loss angles, the loss angles of devices shown in Figure 4.5 were measured.
These devices consisted of approximately 300 µm × 300 µm capacitors with AlN
etches varying from 0 to 15 µm from the electrode edge. The figure shows two top

(right side of the figure) and two bottom (left side of the figure) test capacitors with
a square AlN etch 6 µm (top of the figure) and 10 µm (bottom of the figure) away
from the capacitor edge. Four test capacitors of each type were measured at various
points across the wafer and the results of these tests can be seen in Table 4.1.

75
 AlN etch-Mo distance Bottom C (pF) Bottom tan(δ)(%) Top C (pF) Top tan(δ)(%)
0µm 8.04 ± 0.02 0.12 ± 0.04 7.67 ± 0.04 0.17 ± 0.05
2µm 8.07 ± 0.03 0.12 ± 0.04 7.68 ± 0.03 0.16 ± 0.05
6µm 8.06 ± 0.03 0.14 ± 0.04 7.69 ± 0.03 0.16 ± 0.05
10µm 8.07 ± 0.03 0.16 ± 0.04 7.69 ± 0.03 0.15 ± 0.05
15µm 8.06 ± 0.03 0.16 ± 0.04 7.67 ± 0.03 0.16 ± 0.06

Table 4.1: AlN capacitance measured for capacitors with different distances to etched
AlN. The capacitance and loss angle was independent of the distance to
etched AlN.

Bottom d31 (pm/V ) Top d31 (pm/V )

2.27 ± 0.2 2.48 ± 0.2

Table 4.2: Measured d31 . The top layer exhibited a slightly larger d31 .

These measurements were taken with an Agilent 4284A LCR meter using a shielded
two terminal connection and a correction for the open circuit stray capacitance. This

table indicates that neither the capacitance nor the loss angle varied significantly with
the distance between the AlN etch and the Mo electrode. The top layer of AlN was
975 nm thick and the test capacitors were 295.5 µm × 295.5 µm. The bottom layer
of AlN was 967 nm thick and the test capacitors were 297.3 µm × 297.3 µm. This
gives an average relative dielectric constant of 9.7 for the top layer and 10.0 for the

bottom layer. This relative dielectric constant is lower than that measured by most
authors.

4.2.2 Piezoelectric Coupling Coefficient (d31 )

After measuring the test capacitors to determine relative dielectric constant and
loss angle, beams were actuated to determine the d31 piezoelectric coupling coefficient

using equation 2.37. These measurements were performed on six different test can-
tilevers. On each test cantilever, both the top and bottom piezoelectric layers were
measured separately. The results can be seen in Table 4.2.

76
 This data indicates that the d31 piezoelectric coupling coefficient is slightly better
in the top layer than in the bottom. Although the loss angles of the two layers
would suggest that the they have approximately the same quality, other authors have
shown that AlN with equivalent loss angles can have different piezoelectric coupling

coefficients [54, 55].

4.2.3 Sensitivity and Noise Floor

With the given loss angles and piezoelectric coupling coefficients, the sensitivity
and noise floor of the designed microphones can be predicted. The sensitivity can
be estimated using equation 2.49. The noise floor can be estimated by modeling
the output voltage noise of the common source amplifier, including the piezoelectric

noise at the input of the common source amplifier, and then referring the noise back
to the input using the sensitivity. Several different beam lengths were designed and
fabricated. Here, a microphone consisting of two 395 µm long, 790 µm wide beams
will be investigated. These beams have a 20 µm long, 790 µm wide anchor at the base,

adding stray capacitance. These microphones are expected to have a low frequency
sensitivity of 5.69e-4 V/Pa out of the piezoelectric transducer, 1.76e-3 V/Pa out of
the common source amplifier, a natural frequency of 18.4 kHz, and an A-weighted
noise floor of 35 dBA SPL. Several microphones of this size demonstrated similar

performance with natural frequencies just above 18 kHz and sensitivity of roughly
1.8 mV/Pa at the output of the common source amplifier. One of these was selected
at random for a more detailed performance characterization. All measurements were
made in the same manner as those described in Chapter III.
The selected microphone had a sensitivity of 1.82e-3 V/Pa at 1 kHz out of the

common source amplifier, and a natural frequency of 18.4 kHz. This microphone
had an A-weighted noise floor of 37 dBA. A frequency response of this microphone
can be seen in Figure 4.6. The plane wave tube used to measure the frequency

77
 −45
Measured
Model
Sensitivity (dB re 1 V/Pa) −50

−55

−60

−65 2 3 4
10 10 10
Frequency (Hz)

Figure 4.6: Sensitivity of a piezoelectric MEMS microphone with a 18.4 kHz resonant
frequency. The increase in sensitivity at high frequencies is a character-
istic of the plane wave tube.

response as described in Chapter III could not be used to measure the response of
this microphone at frequencies as high as the microphone resonance. This is because
at high frequencies, propagating modes other than the plane wave mode exist while
at low frequencies, these higher modes are evanescent. This tube cut-off frequency is

a function of its diameter.

The sensitivity of the microphone at 10 kHz increases beyond that of the model.
This is caused by the plane wave tube and is not a true characteristic of the mi-
crophone. The noise floor measurement can be seen in Figure 4.7. The measured
A-weighted noise floor (37 dBA) is roughly 2 dB higher than the modeled noise floor

(35 dBA). Figure 4.7 shows that the difference between the modeled and measured
noise floor occurred uniformly across all frequencies. The main cause for this dis-
crepancy is most likely the material loss angle. A material loss angle of 0.3% would
cause the observed increase in noise floor and increases the noise floor almost uni-

78
 25
Measured
Noise Floor (dB re 1 Pa/√(Hz))

20 Model

15

10

−5

−10 2 3 4
10 10 10
Frequency (Hz)

Figure 4.7: Noise floor of a piezoelectric MEMS microphone with a 18.4 kHz resonant
frequency at room temperature. The modeled noise floor is roughly 2
dB lower than that measured but within the range expected due to the
measured cross-wafer variation

79
formly through the frequency range shown above. Although this loss tangent would
be higher than those measured, it would be less than two standard deviations away
from the mean. Another cause could be a lower value of d31 in combination with
a slightly higher common source amplifier gain. Although the sensitivity of the mi-

crophone matches the model quite well, the transconductance of the JFETs used in
the common source amplifier can vary significantly from part-to-part. If the JFET
transconductance was slightly higher than modeled and the d31 was slightly lower than
modeled, the modeled sensitivity may still match the measurement but the input re-
ferred noise would be lower than expected. The 2 dB difference between measurement

and model is likely a combination of these effects. A microphone with a 12.4 kHz
resonant frequency was also measured and exhibited a 34 dBA noise floor.
These measurements were made with the back of the microphone open to the room.
This makes the back cavity volume very large and so it has very little influence on

the frequency response. The front cavity was also left off the microphone in order to
avoid any package resonance. After releasing the beams, they became bent roughly
30 µm upward due to the large compressive stress of the bottom Mo layer. Because
the beams were bent roughly 30 µm upward, the acoustic resistance, Rt , around the

beams is lower than expected. This requires the back cavity volume to be large in
order to have a low cut-off frequency set by Rt and Cback given in the packaging
section of Chapter II.

4.2.4 Linearity

The linearity of the microphone was also measured. This measurement was made
in the two tube resonator described in Section 2.3.2. The frequency response of this

system was measured by applying a signal to the speaker and measuring the resulting
output with a calibrated Larson Davis 2520 reference microphone. This system fre-
quency response will, therefore, include the frequency response of the speaker as well

80
 0

−10
Response (dB)

−20

−30

−40

−50
0 500 1000 1500 2000 2500 3000
Frequency (Hz)

Figure 4.8: Measured frequency response of the resonant speaker/tube system.

as that of the tube system. The measured frequency response is shown in Figure 4.8.

This response is similar to the modeled response of the system. When a tone is played
at 500Hz, the harmonics will be attenuated providing a low distortion pressure signal
to the microphone.
After calibrating the microphone sensitivity, it was placed in the tube and the

speaker was excited at 500 Hz. The total harmonic distortion of the signal was
measured. The model predicts that this particular microphone amplified by a common
source amplifier using a JFET (model 2sk3426) will have 3% THD at an output
voltage of 82.0 mV RMS, 5% THD at 137 mV RMS, and 10% THD at 254 mV
RMS. The measured output voltage at 3% THD was 85.5 mV RMS, at 5% THD was

149 mV RMS, and at 10% THD was 360 mV RMS. As the distortion increases, the
measurements tend to stray further from the model but the predicted 3% THD level
is quite accurate. In the other cases, the device outperforms the model. Using the
calibrated sensitivity of this microphone, these distortion levels correspond to sound

81
pressure levels of 128 dB SPL at 3% THD, 133 dB SPL at 5% THD, and 140 dB SPL
at 10% THD. With the use of more linear electronics, the distortion could be much
lower at higher sound pressure levels.

4.3 Results

Microphones with two different resonant frequencies, 18.4 kHz and 12.4 kHz, were

characterized and these microphones had noise floors of 37 dBA and 34 dBA. These
measured devices did not have boron doping because the high aspect ratio DRIE
etch recipe improved cross-wafer variation to an acceptable level. The boron doping
tended to add roughness to the SiO2 leading to a slight reduction in piezoelectric

film quality. Unlike the first generation devices, these devices had a center electrode
that spanned only about 50% of the beam which increased sensitivity and reduced
the noise floor as expected. The properties of the 1 µm thick AlN were much better
than those of the 0.5 µm micron thick AlN used in the first generation devices. This

improvement is due primarily to the increase in thickness but also due to the use of
an RIE etch instead of a wet etch used in the first generation devices. Unfortunately,
the high stress in the bottom Mo layer led to a relatively large beam curvature which
prevented these microphones from being placed in a package with a small volume.
Future improvements should be made to reduce the necessary package volume.

82
 CHAPTER V

Conclusions

Given the performance obtained in the previous chapter and that which is possi-

ble with further improvements, piezoelectric MEMS microphones show the potential
for providing small microphones with much higher performance than that which is
currently available commercially.

5.1 Demonstrated Performance

Figure 5.1 shows the noise floor of the microphones discussed in Chapter IV com-

pared to other published works. These microphones have a noise floor 8 - 11 times
lower than the best previously reported piezoelectric MEMS microphones using a
sensing area only 62% - 92% that of previous devices.
Another way to compare microphones is to look at the output energy given as
2
the 1/2·Vout Cmic product or the 1/2·Q2out/Cmic quotient at the output of these trans-
ducers caused by 1 Pa input pressure. With a sensitivity of 0.588 mV/Pa out of the
piezoelectric transducer and a capacitance of 58 pf, the tested microphone with an
18.4 kHz resonant frequency develops a potential energy of 1.00e-17 Joules at the
output. While the output signal prior to amplification of the capacitive microphone

fabricated by Loeppert and Zinserling [29] is not given, it can be estimated through
modeling. If the diaphragm has clamped boundary conditions, a 600 µm diameter

83
 70 15kHz, Franz
10kHz, Royer

60
Noise Level (dBA)
16kHz, Schellin
10kHz, Ried
17kHz, Kressman

50

40
10kHz, PresentWork
8kHz, PresentWork
30

20 −7 −6 −5 −4
10 10 10 10
Diaphragm Area (m2)

Figure 5.1: Piezoelectric MEMS microphone of present work compared to published

works.

diaphragm would have a resonant frequency of 40 kHz and a center displacement of

2.7 nm/Pa. This displacement, in combination with an 11 V bias would result in an

output energy of 3.13e-18 Joules. If the boundary conditions were not clamped but
instead some combination of clamped and pinned, the output energy would increase
and the resonant frequency would decrease. This means that if the tested 18.4 kHz
resonance piezoelectric microphone had the same device capacitance as the capacitive

MEMS microphone, it would be expected to have 1.8 times the sensitivity. Even if
these devices had the same area, the piezoelectric MEMS microphone would be ex-
pected to have a greater sensitivity. Therefore, using the same circuitry, the input
referred circuit noise would be lower for the piezoelectric MEMS microphone than this

capacitive MEMS microphone. While the circuitry is not the only source of noise,
this analysis shows that the circuitry used for these measurements does not have
exceptionally low noise and further improvements to circuitry could be anticipated.

84
 While the tested piezoelectric microphone discussed above has a greater output
energy than the capacitive microphone, it also has a much lower resonant frequency.
It would seem that a capacitive microphone with a lower natural frequency would have
a greater output energy and a lower noise floor. While it is true that a more compliant

diaphragm would result in a lower noise floor, it would also result in more distortion.
A patent by Wang et al. illustrates this tradeoff [58]. This patent demonstrates
capacitive MEMS microphones with a resonant frequency between 2 kHz and 3 kHz
and a noise floor of 9.4 dBA but their models indicate that microphones like this will
deflect 2/3 the gap height at a sound pressure level of 128 dB at which point the

diaphragm will be pulled to the back plate. If a microphone stopped working when
exposed to spikes in sound pressure levels exceeding 128 dB SPL, this would pose
a problem as spikes in sound pressure levels above 128 dB SPL could be seen in a
variety of situations. Of course, mechanical stops could be built into the microphone

to prevent it from exceeding these levels but these would add fabrication complexity
and more severe nonlinearity.

5.2 Further Performance Improvements

Although these microphones exhibit significantly improved performance over pre-

viously built piezoelectric MEMS microphones, there is still room for improvements.

While the tested microphones used 100 nm thick electrodes and 1 µm thick AlN lay-
ers, the use of 25 nm thick electrodes and high quality, AlN at 0.6 µm thick would
result in a 28 dBA noise floor for a device with a 15 kHz resonant frequency. A reduc-
tion in Mo thickness should be easily achievable with the use of an optical emission

spectrometer to stop the AlN etch on the Mo electrode [59]. Further, the use of a
low stress diaphragm should reduce the noise floor by an additional 4 dB over that
of a cantilever based device of equal area. Additionally, the use of lower noise elec-
tronics could reduce the noise floor of these microphones by another 3 dB. All these

85
improvements will roughly add together resulting in extremely low noise microphones.
Over time, it is not unreasonable to expect that high quality AlN will be deposited
at 0.5 µm thick because this has already been done in a laboratory environment [55].
It is also reasonable to expect that, by optimizing the circuitry for a piezoelectric

MEMS microphone, the noise floor of the circuitry could be significantly below that
of the piezoelectric material. If this is the case, a diaphragm with a 15 kHz natural
frequency would have a radius of 509 µm and a capacitance (when sizing electrodes
optimally) of 102 pf. This microphone would have a sensitivity of 1.45 mV/Pa and
a noise floor of 20.5 dBA with noise free circuitry and a loss angle of 0.1%. If the

natural frequency is increased to 40 kHz, the radius would be 312 µm, and the noise
floor would be 33.2 dBA.
In addition to the potential for a lower noise floor, piezoelectric transduction
offers greater linearity versus capacitive sensing. While the linearity of the fabricated

devices has only been tested up to the limit of the common source amplifier, the use of
a different amplification scheme should enable these microphones to sense extremely
high sound pressure levels. The second generation devices resonate at 18.4 kHz, have
an area of 6.27 × 10−7 , a noise floor of 37 dBA and experience 3% THD at 128 dB SPL.

Although the linearity of these devices is limited by the amplification scheme, they are
much more linear than any other microphone of similar size and noise floor. If a more
linear amplification scheme was used, it is expected that these microphones would
experience less than 3% THD at 165 dB SPL. If this is the case, these microphones

would have a noise floor and dynamic range similar to 1/8” capacitive microphones
which are roughly ten times the size of these piezoelectric microphones. For any
application where dynamic range and size are the primary specifications of interest,
well optimized piezoelectric microphones vastly outperform capacitive microphones.

86
5.3 Applications

The simplicity, linearity, small size, and low noise of the fabricated piezoelectric
transducers make them ideal for a wide range of applications. The low noise, linearity,
and simplicity make them ideal for aeroacoustic testing. Many aeroacoustic tests
require the ability to sense loud sounds (165 dB SPL) without distortion. Most
commercial capacitive microphones are used for consumer electronics devices and so

most experience 3% THD between 105 dB SPL - 135 dB SPL. Some instrumentation
quality capacitive microphones experience 3% THD between 160 dB SPL - 170 dB
SPL but these are much more expensive than the microphones built for consumer
electronics applications partly because they are larger and require more complex

electronics. A piezoelectric MEMS microphone could offer a lower cost alternative to

the capacitive microphones currently used for aeroacoustic tests.
The simplicity, small size, and low noise of these microphones make them ideal for
consumer electronics applications. From the 1970s to the present, most microphones

used in consumer electronics applications have been electret microphones utilizing an

amplification scheme nearly identical to that used for these piezoelectric microphones.
These electret microphones are very simple and relatively low noise. In 2003, com-
mercial MEMS microphones were introduced. These microphones had slightly higher
noise levels than many electret microphones but were smaller and could withstand

higher temperatures, reducing the overall manufacturing cost. Piezoelectric MEMS

microphones could offer the advantages of both microphone technologies by providing
a low noise microphone with a simple amplification scheme while maintaining a small
size and withstanding high temperatures.

87
5.4 Contributions

The contributions of this thesis are primarily aimed at reducing the noise floor of
piezoelectric MEMS microphones. A complete electromechanical model of the micro-
phone was used to investigate the limits of piezoelectric microphone technology. After
developing the complete electromechanical model, it became clear that the MEMS
transducer design and optimization could be separated from the circuit design. The

ability to separate these two designs made design trade-offs and limitations more
clear and understandable. Further, new models were developed to more accurately
predict microphone performance and extract piezoelectric film properties from de-
posited films. Once the design trade-offs were fully understood, piezoelectric MEMS

microphones were fabricated and tested. The models were experimentally validated.
The contributions of this thesis have been summarized below.

• Developed a closed form solution for the sensitivity of a multi-layer cantilever

beam without making the small piezoelectric coupling assumption

• Derived an expression used to determine the validity of the small piezoelectric

coupling assumption or a multi-layer beam

• Developed and tested a method for measuring piezoelectric coupling coefficients

using beam curvature

• Introduced the use of output energy as a metric for designing piezoelectric

sensors, detailed the limitations of this metric, and provided design examples.

– Used output energy to determine optimal electrode coverage for minimizing

noise in a piezoelectric MEMS microphone

– Used output energy to determine the influence of residual stress on the

noise floor of a diaphragm-based piezoelectric MEMS microphone

88
 – Used output energy to compare cantilever-based piezoelectric microphones
to diaphragm-based piezoelectric microphones

• Introduced an optimization parameter that can be used for designing piezoelec-

tric sensors and provided design examples

– Used the optimization parameter to determine the optimal layer thick-

nesses for minimizing noise in a piezoelectric MEMS microphone

– Used the optimization parameter to compare piezoelectric materials for

the fabrication of a piezoelectric MEMS microphone

• Built and tested piezoelectric MEMS microphones having a noise floor roughly
ten times lower than the best previously published piezoelectric MEMS micro-

phones.

5.5 Future Work

Future work should be focused on improving the microphone packaging and elec-
tronics. For some applications, the circuitry used during testing may be adequate
but many applications will demand higher sensitivity, lower output impedance, bet-

ter power supply rejection ratio, or lower power consumption. The use of an ASIC
would improve all these metrics and could also reduce the noise floor. Because this
is not a capacitive sensor, a bias is not necessary so the ASIC can be slightly simpler
than those used for capacitive MEMS microphones.
The packaging also needs to be improved. For simplicity, packaging similar to

that used for common commercial MEMS microphones was used for testing but a
large back cavity volume was necessary because the beams were bent upwards. The
beams need bend less than 2.5 µm at the tip in order to package the microphones in
a reasonably small back cavity. To do this, AlN and Mo stress must be well matched

89
from one layer to the next. The ability to control residual stress of both of these
materials has been demonstrated but a low deposition to deposition variation of this
stress is crucial to ensuring flat beams. If beams with low tip deflection cannot be
repeatably fabricated, a design with less sensitivity to stress must be examined. If

this is the case, the fabrication of a stress relieved diaphragm or clamped-clamped

beam will be necessary.

90
BIBLIOGRAPHY

91
 BIBLIOGRAPHY

[1] M. Royer, J. O. Holmen, M. A. Wurm, and O. S. Aadland. Zno on si integrated

acoustic sensor. Sensors Acuators, 4:357–362, 1983.

[2] D. Hohm and G. Hess. A subminiature condenser microphone with silicon nitride
membrane and silicon back plate. Journal of the Acoustical Society of America,
85:467–480, 1989.

[3] J. B. Johnson. Thermal agitation of electricity in conductors. Physical Review,

32:97–109, July 1928.

[4] H. Nyquist. Thermal agitation of electric charge in conductors. Physical Review,

32:110–113, 1928.

[5] T. H. Lee. The Design of CMOS Radio-Frequency Integrated Circuits. Cam-

bridge University Press, 1998.

[6] H. B. Callen and T. Welton. Irreversibility and generalized noise. Physical

Review, 83:34–40, July 1951.

[7] V. Tarnow. The lower limit of detectable sound pressures. Journal of the Acous-
tical Society of America, 82:379–381, 1987.

[8] T. B. Gabrielson. Mechanical-thermal noise in micromachined acoustic and vi-

bration sensors. IEEE Transactions on Electron Devices, 40(5):903–909, MAY
1993.

92
 [9] P. R. Scheeper, B. Nordstrand, J. O. Gullov, B. Liu, T. Clausen, L. Midjord,
and T. Storgaard-Larsen. A new measurement microphone based on mems tech-
nology. Journal of Microelectromechanical Systems, 12(6):880–891, DEC 2003.

[10] R. P. Ried, E. S. Kim, D. M. Hong, and R. S. Muller. Piezoelectric microphone

with on-chip cmos circuits. Journal Microelectromechanical Systems, 2:111–120,

September 1993.

[11] R. Kressman, G. Hess, and R. Schellin. New results of micromachined silicon

subminiature microphones using piezoelectric polymer layers. 9th International

Symposium on Electrets, pages 1044–1049, 1996.

[12] R. Schellin, G. Hess, W. Kuehnel, G. M. Sessler, and E. Fukada. Silicon sub-

miniature microphones with organic piezoelectric layers - fabrication and acous-
tical behavior. Ieee Transactions on Electrical Insulation, 27(4):867–871, AUG
1992.

[13] J. Franz. Funktionsweise und technishe realisierung eines piezoelektrischen siliz-

iumsensors fur akustiche grossen. VDI-Berichte, 677:299–302, 1988.

[14] Lawrence E. Kinsler and Austin R. Frey. Fundamentals of acoustics. Wiley, New
York, 1982.

[15] F. A. Levinzon. Fundamental noise limit of piezoelectric accelerometer. Ieee

Sensors Journal, 4(1):108–111, FEB 2004.

[16] B. A. Auld. Acoustic Fields and Waves in Solids. Krieger, 1973.

[17] R. Herdier, D. Jenkins, D. Remiens, M. Dupont, and D. Osmont. A silicon

cantilever beam structure for the evaluation of d31 , d33 and e31 piezoelectric co-
efficients of pzt thin films. 16th IEEE International Symposium on Applications
of Ferroelectrics, pages 725–727, May 2007.

93
[18] Ieee standard on piezoelectricity. Technical report, 1987.

[19] Q. M. Wang and L. E. Cross. Performance analysis of piezoelectric cantilever

bending actuators. Ferroelectrics, 215:187–213, March 1998.

[20] C. K. Lee. Theory of laminated piezoelectric plates for the design of distributed
sensors/ actuators. part i: Governing equations and reciprocal relationships.
Journal of the Acoustical Society of America, 87:1144–1158, March 1990.

[21] H. F. Tiersten. Linear Piezoelectric Plate Vibration. Plenum, 1969.

[22] T. L. Ren, L. T. Zhang, H. S. Liu, L. T. Liu, and Z. H. Li. A novel ferroelectric

based microphone. Microelectronic Engineering, 66(1-4):683–687, APR 2003.

[23] R. S. Fazzio, T. Lamers, O. Buccafusca, A. Goel, and W. Dauksher. Design and

performance of aluminum nitride piezoelectric microphones. 14th International
Conference on Solid-State Sensors, Actuators and Microsystems, pages U632–
U633, June 2007.

[24] M. S. Weinberg. Working equations for piezoelectric actuators and sensors. IEEE

Journal of Microelectromechanical Systems, 8:529–533, December 1999.

[25] D. Devoe and A. P. Pisano. Modeling and optimal design of piezoelectric can-
tilever microactuators. IEEE Journal of Microelectromechanical Systems, 6:266–
270, September 1997.

[26] M. Krommer. On the correction of the bernoulli-euler beam theory for smart
piezoelectric beams. Journal of Smart Materials and Structures, 10:668–680,

2001.

[27] M. Krommer and H Irschik. On the influence of the electric field on free trans-
verse vibrations of smart beams. Journal Smart Materials and Structures, 8:401–
410, 1999.

94
[28] E. Elka, D. Elata, and H. Abramovich. The electromechanical response of mul-
tilayered piezoelectric structures. IEEE Journal of Microelectromechanical Sys-
tems, 13:332–341, April 2004.

[29] P. V. Loeppert and B. Zinserling. Silicon microphone for high volume consumer
electronic applications. Fortschritte Der Akustik, 31:393–394, 2005.

[30] H. F. Tiersten. Electroelastic equations for electroded thin plates subject to large

driving voltages. Journal of Applied Physics, 74:3389–3393, September 1993.

[31] Q. M. Wang, Q. Zhang, B. Xu, R. Liu, and L. E. Cross. Nonlinear piezoelectric

behavior of ceramic bending mode actuators under strong electric fields. Journal
of Applied Physics, 86:3352–3360, 1999.

[32] K. Shimada, T. Sota, K. Suzuki, and H. Okumura. First-principles study of

piezoelectric constants in strained bn, aln, and gan. Japneese Journal of Applied

Physics, 37:1421–1423, 1998.

[33] S. N. Mahmoodi, M. F. Daqaq, and N. Jalili. On the nonlinear-flexural response

of piezoelectrically driven microcantilever sensors. Sensors and Actuators A,
153:171–179, 2009.

[34] S. Braun, D. J. Ewins, and S. S. Rao. Eneyclopedia of vibration. Academic

Press, San Diego, 2002.

[35] G. Piazza. Piezoelectric Aluminum Nitride Vibrating RF MEMS for Radio Front-

End Technology. PhD thesis, University of California Berkeley, 2005.

[36] S. Trolier-McKinstry and P. Muralt. Thin film piezoelectrics for mems. Journal
of Electroceramics, 12(1-2):7–17, JAN-MAR 2004.

[37] P. L. Chen, R. S. Muller, R. M. White, and R. Jolly. Thin film zno-mos trans-
ducer with virtually dc response. Ultrasonics Symosium, pages 945–948, 1980.

95
[38] P. Soussan, K. O’Donnell, J. D’Haen, G. Vanhoyland, E. Beyne, and H. A. C.
Tilmans. Pulsed dc sputtered aluminum nitride: A novel approach to con-
trol stress and c-axis orientation. Matr.Res.Soc.Symp.Proc., 833:G2.2.1–G2.2.6,
2005.

[39] G. Carlotti, G. Socino, A. Petri, and E. Verona. Acoustic investigation of the

elastic properties of zno films. Applied Physics Letters, 51:1889–1891, December
1987.

[40] K. Tsubouchi and N. Mikoshiba. Zero-temperature-coefficient saw devices on

aln epitaxial films. IEEE Transactions on Sonics and Ultrasonics, 32:634–644,

September 1985.

[41] P. Delobelle, E. Fribourg-Blanc, and D. Remiens. Mechanical properties de-

termined by nanoindentation tests of [pb(zr, ti)o3 ] and [pb(mg1/3 nb2/3 )1−x tix o3 ]
sputtered thin films. Thin Solid Films, 515:1385–1393, 2006.

[42] M. A. Dubois and P. Muralt. Measurement of the effective transverse piezo-

electric coefficient e(31,f) of aln and pb(zr-x,ti1-x)o-3 thin films. Sensors and
Actuators A-Physical, 77(2):106–112, OCT 12 1999.

[43] J. Bergqvist and F. Rudolf. A silicon condenser microphone using bond and

etch-back technology. Sensors and Actuators A, 45:115–124, 1994.

[44] R. P. Ried, E. S. Kim, D. M. Hong, and R. S. Muller. Piezoelectric microphone

with on-chip cmos circuits. IEEE Journal of Microelectromechanical Systems,
2(3):111–120, SEPT 1993.

[45] F. A. Levinzon. Noise of the jfet amplifier. IEEE Transactions on Circuits and
Systems, 47:981–985, 2000.

96
[46] F. P. Mechel. Notes on the radiation impedance, especially of piston-like radia-
tors. Journal of Sound and Vibration, 123:537–572, October 1987.

[47] W. E. Newell. Miniaturization of tuning forks. Science, 161:1320–1326, Septem-

ber 1968.

[48] M. Ahmad. A simple scheme for loss angle measurement of a capacitor. IEEE
Transactions on Energy Conversion, 19:228–229, March 2004.

[49] Namba. Resistivity and temperature coefficient of thin metal films with rough

surface. Japanese Journal of Applied Physics, 9(11):1326, 1970.

[50] P. R. Gray and R. G. Meyer. Analysis and Design of Analog Integrated Circuits.
Wiley, 3 edition, 1993.

[51] Stephen Timoshenko and S. Woinowsky-Krieger. Theory of plates and shells.

McGraw-Hill, New York, 1959.

[52] M. Yu and B. Balachandran. Sensor diaphragm under initial tension: Linear

analysis. Experimental Mechanics, 45(2):123–129, APR 2005.

[53] G. F. Iriarte, J. Bjurstrom, J. Westlinder, F. Engelmark, and I. V. Katardjiev.

Synthesis of c-axis-oriented aln thin films on high-conducting layers: Al, mo, ti,
tin, and ni. IEEE Transactions On Ultrasonics, Ferroelectrics and Frequency
Control, 52:1170–1174, July 2005.

[54] F. Martin, P. Muralt, M.A. Dubois, and A. Pezous. Thickness dependence of the
properties of highly c-axis textured aln thin films. Journal of Vacuum Science

& Technology A, 22(2):361–365, March 2004.

[55] F. Martin, P. Muralt, and M.A. Dubois. Process optimization for the sputter
deposition of molybdenum thin films as electrode for aln thin films. Journal of
Vacuum Science & Technology A, 24(4):946–952, July 2006.

97
[56] J. Gupta. A comparative study of stress, resistivity and microstructure in copper
and copper-alloy thin films. Master’s thesis, Massachusetts Institute of Technol-
ogy, 1992.

[57] S. Mishin, D. R. Marx, B. Sylvia, V. Lughi, K. L. Turner, and D. R. Clark.

Sputtered aln thin films on si and electrodes for mems resonators: Relationship
between surface quality microstructure and film properties. IEEE Ultrasonics
Symposium, pages 2028 – 2032, 2003.

[58] Z. Wang, Q. Zhang, and H. Feng. High performance silicon condenser micro-

phone with perforated single crystal silicon backplate. United States Patent
7,132,307, November 2006.

[59] Scott Alan Smith. Inductively Coupled Plasma Etching of III-N Semiconductors.
PhD thesis, North Carolina State University, 1999.

98