Acoustic Wave and

Electromechanical Resonators
Concept to Key Applications
 For a listing of recent titles in the
Artech House Integrated Microsystems Series,
turn to the back of this book.
 Acoustic Wave and
Electromechanical Resonators
Concept to Key Applications

Humberto Campanella

artechhouse.com
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the U.S. Library of Congress.

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library.

ISBN-13: 978-1-60783-977-4

Cover design by Igor Valdman

© 2010 ARTECH HOUSE

685 Canton Street
Norwood, MA 02062

All rights reserved. Printed and bound in the United States of America. No part of this book
may be reproduced or utilized in any form or by any means, electronic or mechanical, includ-
ing photocopying, recording, or by any information storage and retrieval system, without
permission in writing from the publisher.
All terms mentioned in this book that are known to be trademarks or service marks have
been appropriately capitalized. Artech House cannot attest to the accuracy of this informa-
tion. Use of a term in this book should not be regarded as affecting the validity of any trade-
mark or service mark.

10 9 8 7 6 5 4 3 2 1
Aura matutinae: nivea et bianchissima
Alba del Mar, Aire, Río, Bosque . . .
Contents
Preface xiii
Acknowledgments xv

CHAPTER 1
MEMS and NEMS Resonator Technologies 1
1.1 What Is MEMS and What Is NEMS? 1
1.2 Physical Fundamentals of MEMS and NEMS Resonators 5
1.2.1 The Mechanical Damped Harmonic Oscillator 6
1.2.2 Quality Factor and Damping Mechanisms 8
1.2.3 Transduction in MEMS and NEMS Resonators 11
1.2.4 Resonance Frequency, Mode Shaping, and Aspect Ratio 18
1.3 Key Fabrication Technologies 22
1.3.1 The Production Cycle 22
1.3.2 Common to Integrated Circuit (MEMS) 24
1.3.3 Nanofabrication Techniques (NEMS) 31
1.4 Summary 33
References 33

CHAPTER 2
Acoustic Microresonator Technologies 37
2.1 Introduction to Acoustic Wave Resonators 37
2.1.1 Acoustic Waves 38
2.1.2 Acoustic Microresonators 40
2.2 Fundamentals of Piezoelectricity and Acoustic Wave Propagation 42
2.2.1 Theory of Piezoelectricity 42
2.2.2 Excitation and Vibration Mode Description 44
2.3 Surface Acoustic Wave (SAW) Resonators 46
2.3.1 One-Port and Two-Port Configurations 47
2.3.2 SAW Resonator Design 50
2.3.3 SAW Applications 51
2.4 Bulk Acoustic Wave Resonators 51
2.4.1 Acoustic Wave Propagation in BAW Devices 52
2.4.2 Thin-Film Bulk Acoustic Wave Resonators (FBAR) 54
2.4.3 Solidly Mounted Resonators (SMR) 57
2.4.4 FBAR and SMR Applications 60
2.5 Summary 64
References 65

vii
viii Contents

CHAPTER 3
Design and Modeling of Micro- and Nanoresonators 69
3.1 The Stages of Resonator Design and Modeling 69
3.2 The Electromechanical Transformer 75
3.2.1 MEMS and NEMS Resonators 76
3.2.2 FBAR and Other Acoustic Resonators 78
3.3 Equivalent-Circuit Models 82
3.3.1 The Resonant LC Tank 82
3.3.2 The Butterworth-Van-Dyke Model 83
3.3.3 Case Study: FBAR Process and Modeling 86
3.4 Finite Element Modeling (FEM) 88
3.4.1 Building the Model 90
3.4.2 Structural, Modal, and Harmonic Analyses 92
3.4.3 Coupled-Domain Analysis 94
3.4.4 Case Study: Modal and Harmonic Analysis of a Resonant
Mass Sensor 96
3.5 Summary 99
References 100

CHAPTER 4
Fabrication Techniques 103
4.1 Process Overview 103
4.2 FBAR Fabrication Techniques 105
4.2.1 Oxidation of Silicon 105
4.2.2 Metallization and Piezoelectric Layer Deposition 106
4.2.3 Surface-Micromachining-Based Process 108
4.2.4 Bulk-Micromachining-Based Processes 109
4.3 Instrumentation and Materials for Fabrication 110
4.4 Process Compatibility and Characterization 112
4.4.1 Thin-Film Attributes 112
4.4.2 Crystallography 114
4.4.3 Etching Performance 118
4.4.4 Structural Performance 121
4.5 Summary 123
References 125

CHAPTER 5
Characterization Techniques 127
5.1 Low- and High-Frequency Electrical Characterization 127
5.1.1 Short-Open DC and Low-Frequency Measurements 128
5.1.2 Microwave Network Theory and the Scattering-Parameter
Description 130
5.1.3 High-Frequency Measurement Setup 131
5.1.4 Quality Factor Extraction 133
5.2 Determination of Elastic, Dielectric, and Piezoelectric Constants 140
5.2.1 Elastic Constants 140
5.2.2 Dielectric Constants 143
Contents ix

5.2.3 Piezoelectric Properties 144

5.3 Equivalent-Circuit-Parameter Extraction 146
5.3.1 Parameter-Extraction Algorithm 147
5.3.2 Case Study: Equivalent-Circuit-Parameter Extraction
of an FBAR 150
5.4 AFM, Optical, and Electron-Beam-Induced Characterization 151
5.4.1 AFM-Based Characterization with Optical Detection 151
5.4.2 Optical Microscope Interferometry with Piezoelectric Actuation 154
5.4.3 Fabry-Pérot Interferometry 155
5.4.4 Electron-Beam Excitation 157
5.5 Summary 158
References 158

CHAPTER 6
Performance Optimization 163
6.1 Frequency Stability 163
6.1.1 Thin-Film Thickness Tolerance 164
6.1.2 Layout Design Effects 165
6.1.3 Time and Frequency Stability 165
6.1.4 Temperature Stability and Thermal Coefficient Factor (TCF) 168
6.2 Temperature Compensation 169
6.2.1 TCFBAR Fabrication Processes 170
6.2.2 Behavioral Description and Modeling of a TCFBAR 171
6.3 Frequency Tuning 172
6.3.1 DC Tuning 174
6.3.2 Uniform-Film Deposition 175
6.3.3 FIB-Assisted Tuning Technique 177
6.3.4 Milling of FBAR as Another FIB-Tuning Procedure 179
6.3.5 Frequency-Tuning Sensitivity and Responsivity 180
6.3.6 Quality Factor 181
6.4 Summary 183
References 184

CHAPTER 7
Integration of Resonator to CMOS Technologies 187
7.1 Integration Strategies 187
7.1.1 Hybrid Integration 188
7.1.2 Monolithic Integration 191
7.1.3 Heterogeneous Integration 194
7.2 State-of-the-Art Integrated Applications 195
7.2.1 MEMS and NEMS Resonators 196
7.2.2 SAW and FBAR 200
7.2.3 Advanced 3D Integration Technologies: Wafer Level Transfer 204
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 204
7.3.1 The Resonator Process 207
7.3.2 The CMOS Process 209
7.3.3 The Wafer-Level-Transfer Process 213
x Contents

7.3.4 Characterization and Technology Optimization 215

7.4 Summary 219
References 221

CHAPTER 8
Sensor Applications 225
8.1 Resonant Sensing Performance 225
8.1.1 The Role of the Q Factor on Resolution 226
8.1.2 Performance Features and Parameters 227
8.2 Mass Sensors 229
8.2.1 MEMS-Based Microbalances 229
8.2.2 Ultrasensitive NEMS Mass Sensors 234
8.2.3 Acoustic Resonator Distributed-Mass Sensors 238
8.2.4 FBAR-Based Localized-Mass Detection 242
8.3 Mechanical Sensors 245
8.3.1 Pressure Sensors 246
8.3.2 Accelerometers 248
8.4 Atomic Force Detection 251
8.5 Magnetic Sensors 254
8.6 Summary 259
References 260

CHAPTER 9
Radio Frequency Applications 265
9.1 Introduction 266
9.2 Passive-Circuit Applications 269
9.2.1 SAW, BAW, and FBAR-Based Band-Selection Filters 269
9.2.2 Duplexers, Triplexers, and More 271
9.2.3 Microelectromechanical Filters 277
9.2.4 RF MEMS Switches 282
9.3 Active-Circuit Applications 285
9.3.1 Oscillators 285
9.3.2 Mixers and Mixlers 289
9.3.3 Tuned Low-Noise Amplifiers 292
9.3.4 RF Front-End Systems 293
9.4 Summary 297
References 298

CHAPTER 10
Case Studies: Modeling, Design, and Fabrication of FBAR and
MEMS-Based Systems 303
10.1 Methodological Approach for MEMS-IC Integration 304
10.2 Case I: Compatibility of FBAR and Silicon Technologies 306
10.2.1 Compatibility Testing 306
10.2.2 Front-Side, Reactive-Ion-Etching-Based Process 312
10.2.3 Back-Side Wet-Etching Process 317
Contents xi

10.3 Case II: High-Level Design of a Temperature-Compensated (TC)

Oscillator 319
10.3.1 The Temperature Stability Issue in Oscillators 320
10.3.2 Low Phase Noise FBAR-Based Oscillators 322
10.3.3 Codesign of an FBAR-Based TC Oscillator 323
10.4 Case III: Read-Out Circuit Design of a 434-MHz MEMS Resonator 330
10.4.1 MEMS-CMOS Integration Technology 331
10.4.2 Read-Out Circuit Specification and Circuit Architecture 332
10.4.3 Read-Out Implementation and Characterization 333
10.5 Summary 334
References 335

About the Author 337

Index 339
Preface
Micro- and nanoresonators have become key components of virtually all modern
radio frequency (RF) systems and sensors. As fabrication and packaging technolo-
gies have enabled high-production yield and high-performance devices, miniature
resonators have left the laboratory to take up industrial applications. However, the
complexity and huge variety of acoustic and electromechanical resonator technolo-
gies sometimes make it difficult to understand the mechanisms involved in their
operation. Efficacy on decision making with regard to choosing the right technol-
ogy for our application is thus often prevented. Also required is mastery of several
disciplines, from applied physics and materials to manufacturing and system level
applications.
How can we give an effective answer to these issues so as to benefit design, pro-
cess, and application engineers? Keeping them in mind, this book provides compre-
hensive and wide coverage of thin-film bulk acoustic wave resonators (FBAR),
microelectromechanical (MEMS), and nanoelectromechanical (NEMS) resonators,
at both the technology and application levels. Hence, this work contributes to the
field with a thorough and practical resonator description that gathers, for the first
time, state-of-the-art design and modeling, fabrication and characterization tech-
niques, integration with standard integrated circuit (IC) technologies, and industrial
and laboratory applications. Moreover, practical case studies complement the theo-
retical and laboratory background.
The book assumes basic undergraduate-level preparation in physics, material
science, electronics, chemical, mechanical, or electrical engineering. The intended
audience includes the following:

1. Senior undergraduate or beginning graduate students involved in advanced

education related to MEMS, RF systems, or sensors;
2. Process and design engineers, related to clean room and micro- and
nanofabrication processes and modeling of MEMS devices;
3. MEMS, NEMS, and FBAR device researchers who are already familiar with
the fundamentals of both MEMS and circuit design.

The contents of this book stem largely from academic and clean room experi-
ence. They follow what is to the author’s view the logical sequence of FBAR,
MEMS, and NEMS resonator production flow.
Chapter 1 is an introductory chapter to MEMS and NEMS resonator technolo-
gies. It covers the context and theoretical background of electromechanical devices.
The discussion differentiates between MEMS and NEMS, their transduction mech-
anisms, and current fabrication technologies.

xiii
xiv Preface

Chapter 2, on the other hand, specializes in bulk and surface acoustic wave res-
onators, like thin-film bulk acoustic wave resonator (FBAR), solidly mounted reso-
nator (SMR), and surface acoustic wave resonator (SAW) technologies. As in
the previous chapter, physical fundamentals of acoustic wave propagation,
piezoelectricity, and applications are discussed.
Chapter 3 addresses the varied universe of resonator models by first defining a
design flow and the role of modeling in the production of resonators. Electrome-
chanical transformers, equivalent circuits, analytical models, and finite element
models are covered.
Chapter 4 explains the fabrication and technological details of resonator fabri-
cation by illustrating the case of FBAR manufacturing. Materials, processes, and
structure characterization are described.
Chapter 5 deals with characterization. The contents provide the reader with
concepts and experimental setup of electrical, atomic force microscope, and optical
interferometry techniques.
Chapter 6 discusses performance optimization, thus dealing with frequency sta-
bility, temperature compensation, and frequency tuning. Detailed coverage of a
novel focused-ion-beam-assisted technique for FBAR tuning is provided.
Chapter 7 deals with technological issues of FBAR, MEMS, NEMS resonator,
and complementary metal oxide semiconductor (CMOS) technologies integration.
As the MEMS-CMOS concept expands in current implementations, the subject is of
central concern of process engineers and IC designers.
Chapter 8 reviews state-of-the-art sensor applications of FBAR, MEMS, and
NEMS resonators. A balanced discussion of academic research and industrial prod-
ucts is preserved in the chapter.
Chapter 9 reviews classical passive and active RF applications. It revisits them
under the light of the latest FBAR and MEMS resonator technologies, enabling the
development of new architectures. Special attention is given to high-impact com-
mercial RF microdevices and microsystems.
Chapter 10 closes the book by studying three implementation cases. Practical
examples of FBAR fabrication, conceptual design of a temperature-compensated
(TC) oscillator, and 434-MHz MEMS resonator read out circuit design are studied.
Acknowledgments
Writing a book cannot be accomplished without the support and encouragement of
family, friends, and colleagues. Auto-motivation and faith also aid the author to
foresee the “light at the end of the tunnel” with no significant loss of sanity and
common sense.
First, I thank the Consejo Superior de Investigaciones Científicas (CSIC) and
particularly the management and staff of the Instituto de Microelectrónica de Bar-
celona–Centro Nacional de Microelectrónica (IMB-CNM) for allowing me to
undertake this project. I specially thank the clean room staffers for their support on
fabrication and characterization. I have no words to express my gratitude to my for-
mer advisor and mentor at CSIC, Professor Jaume Esteve, from whom I have
learned my best lessons. What I have learned from him will be transmitted through
this book. I am indebted to present or former colleagues with which I have had the
privilege to work on several MEMS and FBAR projects. I would like to thank Maria
Calle and C. Golden for reading the manuscript and providing valuable feedback. I
thank the anonymous reviewer for thorough revision of the manuscript. His or her
suggestions contributed to significantly improve the quality of the contents. I would
like to thank Dr. Zachary Davis and the intellectual property rights and permissions
departments at the IEEE, IOP, NPG-Macmillan, Elsevier, ACS, AIP, SPIE, Sony,
TriQuint, Epcos, iSupply—WTC, Avago Technologies, IBM, Agilent Technologies,
Seiko EPSON, and Discera, for granting permissions to reproduce copyrighted
figures. I gratefully acknowledge the active cooperation of the Artech House
staff—particularly, Mark Walsh, senior acquisitions manager, for having paved the
way to publish this book, and Penelope Comans, assistant editor, for her support
throughout the manuscript development. Finally, I thank you, Alba, my beloved
wife and companion, for designing and preparing high-quality artwork and, most
important, for your unconditional patience and love.

xv
 CHAPTER 1

MEMS and NEMS Resonator

Technologies
This introductory chapter addresses the fundamentals of micro- and
nanoelectromechanical (MEMS and NEMS) resonators. Throughout this book, we
employ the acronym MEMS to designate electron devices based on silicon. In the
Section 1.1, MEMS and NEMS are defined and their main attributes described so
we can identify their main differences from the beginning. Next, in the Section 1.2, a
theoretical background on the physics behind the operating principles of MEMS
and NEMS resonators is presented. The fundamental relationships of mechanical
harmonic oscillators and transduction mechanisms are reviewed, thus supporting
the discussion on the resonance modes and different resonator structures.
Second 1.2 deals with production and technological aspects of MEMS and
NEMS resonator design and fabrication. Section 1.3, the production cycle of
MEMS and NEMS shows the steps to develop and fabricate a resonator, from con-
cept to application dimensioning. The section also describes the technologies cur-
rently employed in MEMS and NEMS fabrication. Two approaches are considered:
the common to integrated circuit (IC), which is suitable for MEMS, and the
nanofabrication-based techniques, more suitable for NEMS.

1.1 What Is MEMS and What Is NEMS?

Microelectromechanical systems (MEMS) are movable structures and systems with

micrometer sizes. The concept also involves the related fabrication technology.
Nowadays, it is generally accepted that MEMS have at least one of their dimensions
in the range from hundreds of nanometers to hundreds of micrometers.
Nanoelectromechanical systems (NEMS) have dimensions below the 100-nm con-
ventional limit. Although some MEMS may also have submicron dimensions, a
more determinant aspect differentiating MEMS and NEMS is their base technol-
ogy. MEMS have been conceived under the integrated circuit (IC) paradigm, in
which silicon is the base material for substrate, integration, and device fabrication.
MEMS are fabricated according to the same fabrication techniques and processes of
ICs, especially from complementary-metal-oxide-semiconductor (CMOS) technol-
ogies. Actually, MEMS and CMOS circuits can coexist for integrated applications.
CMOS has become the predominant IC technology, whose potential is not only
exploited for ICs but also for a variety of MEMS-based applications. Recently inte-
grated microsystems featuring calibration by digital programming, self-testing, and

1
2 MEMS and NEMS Resonator Technologies

digital interfaces have been implemented on a single chip, demonstrating the

strength of CMOS-based MEMS [1].
NEMS devices, on the other hand, are not scaled-down MEMS, but a quite dif-
ferent technological approach. Although it is widely implemented, silicon is no lon-
ger the main base technology of NEMS. New materials and structures like carbon
nanotubes, nanowires, and organic composites are being investigated and imple-
mented in many NEMS demonstrators [2–5]. Scaling down has been achieved by IC
technology reduction, but mainly by introducing new fabrication tools, like electron
beam lithography (EBL) [6], atomic force microscope [7], and nano-imprint [8],
among others. The dominant physical phenomena are also different at the MEMS
and NEMS scales, which imposes new engineering challenges. Due to technological
development at both fabrication and material levels, new applications have
appeared, many of them relying on NEMS devices. This integrated approach of
NEMS technology and applications is known as the nano-bio-info-cognitive
(NBIC) convergence. The schema of Figure 1.1 illustrates the different approaches
of MEMS and NEMS technologies and the NBIC concept. Three different technolo-
gies can be differentiated: microelectronics, MEMS, and NEMS. The baseline mate-
rial for both microelectronics and MEMS is silicon, and applications can be found
in information technologies, neuroscience, biotechnology, cognitive science, and
learning, among many others [9].
Materials, base technology, and applications of micro- and nanodevices are
diverse and go beyond traditional fields of electronics after the conception of the
NBIC convergence. Figure 1.2 shows how, before the NBIC, silicon is almost exclu-
sively the material for fabricating microprocessors, digital signal processors (DSP),
microcontrollers, analog devices, silicon MEMS, and integrated circuits in general
(at left). After the NBIC, new materials and structures like polymers, organic com-
posites, carbon nanotubes, or nanowires make part of specialized systems on the
same chip. The situation has given rise to the bio-MEMS, lab-on-a-chip (LoC),
micro total analysis systems (μTAS) and system-on-a-chip concepts (SoC) [10–12].
MEMS and NEMS play a central role on the miniaturization of electronic SoC,
LoC, and μTAS. Information and communication technologies (ICT) have bene-

Microelectronic (CMOS) Information technologies

Neuroscience

Microsystems (MEMS)
Biotechnology

Nanosystems (NEMS) Cognitive Science (Learning)

Figure 1.1 The NBIC convergence concept.

1.1 What Is MEMS and What Is NEMS? 3

Figure 1.2 Impact of the NBIC on the materials and applications of technology.

fited from the MEMS and NEMS boom as well. Technological advances of micro-
and nanoelectronic engineering fields have led to a drastic reduction of size and
price in sensors and radio frequency (RF) components. New technologies have also
enabled their integration into single microelectronic chips. Ultrahigh sensitivity,
faster response times, and power and size efficiency are some of the benefits that
have leveraged the implementation of MEMS in integrated sensor systems and mod-
ern mobile communication transceivers.
A detailed, comprehensive, and rigorous discussion on MEMS and NEMS his-
tory would take a complete book. If one has to elect the first MEMS device pro-
duced in history, it is the gold resonating MOS gate transistor invented by H. C.
Nathanson at the Westinghouse Laboratories in 1967, which implemented the sur-
face micromachining technique [13]. Thus, we see how the technological break-
throughs of microelectronics and micromachining techniques have allowed the
evolution of MEMS and NEMS. Therefore, one can say that the MEMS era began
with the pressure microsensor in the 1970s. It was followed by the actuator,
microlenses and accelerometers in the 1980s; the RF devices, chemical sensors,
micromirrors, antennas, and gears in the 1990s; and the medical applications, the
LoC, μTAS, bio-MEMS, and NEMS devices developed in the following decade.
This list shows a representative group of the major advancements of electronics,
MEMS, and NEMS history [14]:

• 1947: Germanium transistor invention (Bell Labs: W. Shockley, J. Bardeen,

W. Brattain);
• 1958: Silicon-made force sensor commercialization;
• 1958: First integrated circuit (Texas Instruments: J. Kilby);
• 1961: First silicon sensor (Kulite, bare silicon strain gages);
• 1967: Surface micromachining invention (Westinghouse: J. Nathanson, reso-
nant gate transistor);
• 1967: Anisotropic silicon etching (R. M. Finne and D. L. Klein);
4 MEMS and NEMS Resonator Technologies

• 1970: First silicon accelerometer (Kulite);

• 1971: Microprocessor invention (Intel, Intel 4004);
• 1977: First capacitive pressure sensor (Stanford University: J. Angell);
• 1979: Micromachined inkjet nozzle for thermal printer (Hewlett-Packard
Labs);
• 1980: Silicon torsional scanning mirror (K. E. Petersen);
• 1982: Blood pressure transducer (Foxboro/ICT, Honeywell);
• 1982: LIGA process (Karlsruhe Nuclear Research Center);
• 1986: Atomic force microscope invention (G. Binnig, IBM);
• 1988: Rotary electrostatic side drive motors (U. California Berkeley: Fan, Tai,
and Muller);
• 1989: Lateral comb drive (U. Michigan: Tang, Nguyen, and Howe);
• 1990: BIACORE microfluidic chip;
• 1991: Polysilicon hinge (Pister, Judy, Burgett, Fearing);
• 1993: Multiuser MEMS processes-MUMPS program (Microelectronics Cen-
ter North Carolina);
• 1993: First sold accelerometer (Analog Devices: ADXL50);
• 1994: Deep reactive ion etching patented (Bosch);
• 1996: First carbon nanotubes systematically produced (R. Smalley);
• 1999: Optical micro switch for the Internet (Lucent Technologies);
• 2000s: NEMS;
• 2001: First carbon nanotube transistors (IBM);
• 2005: First MEMS-based RF front end (Martina Consortium);
• 2007: First MEMS oscillator commercialized (Discera).

The MEMS market is very dynamic, and it has grown fast, especially during the
past few years. To date, MEMS devices have broken many economical and concep-
tual barriers, enabling them to penetrate consumer segments of the market in a mas-
sive way. During the past five years, consumer electronics based on MEMS have
passed from 6% to 22% of the global $25 billion MEMS market. The pie diagram
of Figure 1.3 illustrates the situation. IT peripheral applications still dominate,
although to a lesser extent, a segmented market in which automotive, medical,
telecom, industry, aerospace and defense, household, and other applications com-
plete the share [15]. MEMS are close to the big market of street consumer—and not
only in the IT sense. This is evident in the explosion of sophisticated mobile devices
and game consoles incorporating MEMS technology [16, 17].
This book concentrates on the study of a specific kind of the many MEMS and
NEMS existing devices: resonators. A resonator can be defined as a frequency-
selective amplifier, which, after appropriate excitation, goes into vibration. The res-
onance frequency is determined by physical properties of the materials composing
the resonator, its layout, dimensions and mechanical configuration, and the excita-
tion signal. MEMS and NEMS resonators have multiple applications as sensors and
telecommunication system devices. Miniature mass, force, pressure, acceleration,
torque, and flow detectors can be implemented with resonant MEMS and NEMS
1.2 Physical Fundamentals of MEMS and NEMS Resonators 5

Figure 1.3 MEMS application fields in 2009: the global $25 billion market is dominated by IT
and consumer electronics. (© 2009 Nexus [15].)

sensors. In the telecommunications field, resonators are utilized to build filters,

oscillators, and mixers, for example. The quality factor (Q), which is studied in the
next section, is the main parameter defining the performance of a resonator.

1.2 Physical Fundamentals of MEMS and NEMS Resonators

As a physical body, the structure of MEMS and NEMS resonators has an infinite
number of eigenfrequencies, also referred to as eigenvalues. Under certain condi-
tions, virtually any of them can be excited to drive the structure into resonance.
Through the transduction mechanism exciting the MEMS or NEMS, electrome-
chanical energy conversion causes the resonator to mechanically vibrate. Various
approaches exist to implement MEMS/NEMS resonators: the micro- or
nanomechanical and the acoustical. The approaches refer to the resonance modes
excited, which lead to different vibration modes of the structure. In the micro- or
nanomechanical modes, the structure vibrates at the fundamental modes, in gen-
eral, and the resonator deflects, expands, or twists. Because the resonator deforms
when excited at the corresponding frequencies, these are also known as structural
modes. In the acoustical mode approach, deformation of the resonator is negligible,
in comparison to its structural dimensions. An acoustic wave of the corresponding
frequency propagates through the bulk, the surface, or both due to the excitation
signal and the acoustic properties of the resonator. The mechanisms and physics
behind acoustic resonators are studied in detail in Chapter 2. Let us review the
physical fundamentals of resonance and transduction mechanisms in MEMS and
NEMS resonators.
6 MEMS and NEMS Resonator Technologies

1.2.1 The Mechanical Damped Harmonic Oscillator

The resonant behavior of a MEMS or NEMS resonator can be described by the
model of the mechanical damped harmonic oscillator:
n
m eff && + Dx& + kx& = Fext ( x , t ) = F0 ⋅ sin(2πft )
x (1.1)

where x is the time-varying harmonic oscillation, m eff n

is the equivalent mass, D is
the damping constant, k is the spring constant, Fext is the external force applied to
the resonator, and F0 is the amplitude of the harmonic force signal with frequency f.
The first term on the right side of (1.1) is the inertial force of the system, the second
term is the damping and is related to the losses, and the third one is the restoring
force. The eigenfrequencies of the oscillator are calculated by setting F0 = 0 in (1.1),
by only considering the mass and spring constants:

1 k
fn = (1.2)
2π n
m eff

where fn is the resonance frequency of the nth mode. By looking at (1.2), we can
point out that the effective mass depends, of course, on the physical mass of the res-
onator, but also on its dimensions, geometry, and resonance mode, and then we can
make some interesting remarks:

• Scaling down the dimensions of the resonator leads to a reduction of the abso-
lute value of the effective mass. This increases the resonance frequency of the
current mode.
• The effective mass is inversely proportional to the order of the resonance
mode. Higher-order modes have less energy than fundamental ones.
• Scaling down from the MEMS to the NEMS regime leads to a reduction of the
effective mass normalized to the physical mass. Thus, the vibration amplitude
of NEMS resonators is smaller than it is for its MEMS counterpart.

In (1.1) the value of D determines the amount of damping in the system:

2πf n m eff
n

D= (1.3)
Q

where Q is the quality factor of the resonator. When operated in a fluid, like air or a
liquid, the Q factor mainly depends on friction, and it is an important design
parameter of the resonator applications. The energetic interpretation of the har-
monic oscillator gives us a clear definition of the Q factor:

max. energy stored in one period

Q = 2π ⋅ (1.4)
energy dissipated per period

According to the resonance frequency definition, the maximum energy is stored

when the resonator vibrates at the frequency fn. Due to losses, the oscillation energy
is spread on a bandwidth B around fn, which makes (1.4) equivalent to:
1.2 Physical Fundamentals of MEMS and NEMS Resonators 7

fn
Q= (1.5)
B

B is defined between the −3-dB frequencies around fn. The definition is illus-
trated in Figure 1.4, and it is widely used in tuned circuit and resonator character-
ization. Later in Chapter 5 we will return to practical aspects of the Q value
characterization. Depending on whether the resonator vibrates in an elastic medium
or in vacuum, the losses experimented are different. When the medium is air, for
example, the Q factor is loaded by friction, anchoring, surface damping, and
thermo-elastic damping, among other loss mechanisms. In good vacuum, the fric-
tion contribution to the Q factor can be neglected, and its value is higher than in liq-
uid or gas. Because of this situation, the resonator characterization results have to
specify the measurement conditions.
With the actual state of technology, it is not strange to reach Q factors between
103 and 105 with MEMS resonators. High Q values allow low power operation and
high force sensitivity in sensor applications. Besides, high Q is equivalent to low dis-
sipation, which translates into low insertion loss as well, which is very convenient
for radio frequency (RF) applications. Nevertheless, the Q factor diminishes in a lin-
ear way with the resonator size [18]. The idea is illustrated in Figure 1.5, where the
historical evolution of size reduction is compared against Q factor. A high Q factor
implies a reduction in bandwidth, which is useful for the selectivity of sensors and
filters. The situation relates also to the power handling of the resonator: the mini-
mum power is defined at signal levels comparable to the thermal fluctuations of the
system. The minimum power is estimated by dividing the thermal energy KBT° over
the energy-exchange time constant τ:

Figure 1.4 Quality factor of resonators: the oscillation energy is spread on a bandwidth B around
the resonance frequency fn. B is defined between the −3-dB frequencies around fn.
8 MEMS and NEMS Resonator Technologies

Figure 1.5 Quality factor and size of MEMS and NEMS resonators: the Q factor diminishes with
dimensions. (© 2000 IEEE [18].)

KBT °
PMIN = (1.6a)
τ

KBT ° f n
= (1.6b)
Q

where τ is the quotient between Q and the resonance frequency fn. For a given noise
density of the environment, (1.6) means that a high-Q resonator will be able to
reject more thermal noise power than low-Q devices, because noise integration is
performed over a smaller bandwidth. This allows enhanced discrimination of the
in-resonance signal and, equivalently, we can say that the high-Q resonator has big-
ger signal amplitudes than the low-Q resonator, for the same noise levels. Some
characterization techniques are based on the thermal limit operation to extract
some resonator parameters [19].

1.2.2 Quality Factor and Damping Mechanisms

In the previous section, we have defined the damping D and quality factor Q under
the light of the damped harmonic oscillator model. Now, we examine the physical
damping mechanisms inherent to MEMS and NEMS resonator operation.
First, we say that the overall Q factor of the resonator (QTOTAL) relays on diverse
mechanisms, namely:

1. Air damping (1/QAIR);

2. Support damping (1/QSUPPORT);
3. Surface-effect damping (1/QSURFACE);
4. Thermoelastic damping (1/QTHERMOELASTIC).
1.2 Physical Fundamentals of MEMS and NEMS Resonators 9

Depending on the resonator design and the operation environment, these mech-
anisms contribute to a lesser or large extent to the QTOTAL value. QTOTAL is calcu-
lated as the sum of the inverse Q factors attributed to each mechanism:

1 1 1 1 1
= + + + (1.7)
Q TOTAL Q AIR Q SUPPORT Q SURFACE Q THERMOELASTIC

Air damping relates to the energy released by the resonator when it collides with
the air surrounding the device at its vibrational state. At its time, three mechanisms
contribute to air damping: viscous or Stoke’s damping [20], acoustic radiation [21],
and squeezed-film damping [22], the value of QAIR being given by:

1 1 1 1
=+ + + (1.8)
Q AIR Q Stroke Q Squeezed Q Acoustic

Viscous damping, or viscous drag, removes the fluid around the resonator,
whereas acoustic radiation occurs when the resonator excites the air in the direction
perpendicular to the device motion. The energy loss parallel to the direction of
motion is described by the Stoke’s damping:

m eff 2 πf 0
Q Stroke = (1.9)
A 2 ρη atm

where meff is the effective mass, A is the area of the bottom and top surfaces, and f0 is
the resonance frequency of the resonator; ηatm is the viscosity at atmospheric pres-
sure; and ρ is the air density depending on the air’s molar mass, gas constant, and
operating temperature. For example, in parallel-plate resonators with capacitive
transduction, the fluid is pushed out of the gap: the air motion parallel to the plates
produces Couette damping, while the motion perpendicular to the plate surfaces
generates squeezed-film damping:

π 5 h0 m eff f 0
Q Squeezed = (1.10)
48ηL eff W eff3

where h0 and η are the air gap thickness and viscosity at the operating point, and Leff
and Weff are the effective length and width of the air gap [23]. Squeezed-film damp-
ing is the dominant loss mechanism of electrostatically driven MEMS and NEMS
resonators [23]. Nevertheless, the shape of the resonator determines the ultimate
environmental conditions of the dominant mechanism. Vignola et al. found out
that, at a pressure greater than 10−2 Torr, acoustic radiation dominates for their
MEMS paddle resonators, while for a diamond-shaped oscillator, 10−3 Torr is the
low-pressure condition when radiation damping is dominant [24]. More details on
the analytical description of air damping can be found in [25].
Support damping, also known as anchor losses, results when the resonator
anchors are stressed at the clamping points as a consequence of resonator displace-
ment during vibration. Thus, a fraction of the vibrational energy is lost from the res-
10 MEMS and NEMS Resonator Technologies

onator through elastic wave propagation into the substrate. All the elastic energy
transferred to the substrate is lost, although the mechanisms are only partially
understood. Anchor losses can be significant if contributions from other loss mecha-
nisms are negligible, and they degrade the Q factor and may introduce
coupled-to-clamping spurious resonances.
Analytical expressions for anchoring losses derived by Judge et al. distinguish
between two cases: supports that can be treated as plates, and supports that act as
semi-infinite elastic media, with effectively infinite thickness. The former case is
applicable to many MEMS resonators, while the latter is more appropriate for
NEMS devices [26]. However, since the losses depend on the clamping design of the
resonator, no general expression exists for calculating the QSUPPORT value. Instead, it
should be inferred from experiments, after de-embedding the contributions of air,
surface, and thermoelastic damping.
To face the support damping, the designer balances the MEMS structure to
compensate for the losses at the resonance frequency. The tuning fork is perhaps
one of the most popular and studied MEMS structures, largely implemented in
oscillator and filter applications [27]. More recently, other innovative structures
have been explored as well, like the mesa-dome resonator proposed by Pandey et al.
In this design, the resonator is surrounded by a trench, or mesa, that partially
reflects wave energy back to the resonator. Depending on the distance from the res-
onator to the mesa, the reflected wave interferes either constructively or destruc-
tively with the resonator, increasing or decreasing the Q factor. Due to the
constructive-interference anchor decoupling, Q factor improvements of up to
400% are achieved [28].
Surface-effect damping induced at the surface of the resonator increases as
the surface to volume ratio increases. This is the case of devices approaching the
NEMS scale. Roughness, contaminants, and etching residues contribute to damp-
ing and Q factor reduction [29]. Surface losses include structural dissipation
(crystallographic defects) and surface effects, which are more significant when the
dimensions of the resonator approach to those of the surface roughness and
crystallographic defects.
Thermoelastic damping is an intrinsic-material damping source due to
thermoelasticity present in most materials, and it is caused by irreversible heat flow
across the thickness of the resonator. As its name suggests, it describes the coupling
between the elastic field in the structure caused by deformation and the temperature
field. TED is the dominant loss mechanism in MEMS resonators in thicker and long
cantilevers. TED losses become less relevant when the thickness is reduced, anchor-
ing and surface losses being the ultimate loss mechanisms limiting the Q factor at
low pressure.
The earliest study of thermoelastic damping is dated back to 1937 in Zener’s
classical work [30, 31], in which he studied thermoelastic damping in beams under-
going flexural vibrations. Although the design of the resonator influences the
QTHERMOELASTIC contribution to the global factor QTOTAL, other strategies like tor-
sional-mode operation have proven to be effective on reducing the thermoelastic
losses. For a reference on MEMS resonator design with low thermoelastic damping,
refer to the work of Duwel et al. [32].
1.2 Physical Fundamentals of MEMS and NEMS Resonators 11

1.2.3 Transduction in MEMS and NEMS Resonators

Transduction in MEMS and NEMS resonators occurs through a variety of mecha-
nisms involving the physical domain conversion of the input signal. Such mecha-
nisms and the physical response of the resonator generate an output signal after the
excitation. According to this idea, the resonator is both a sensor and an actuator,
whose working principle is illustrated in Figure 1.6. The excitation signal causes the
mechanical structure of the resonator to vibrate, if this signal is of a frequency close
to one of the eigenvalues of the structure. An external component or circuit detects
the resonance in the electrical domain, through an electric current or voltage pro-
portional to the vibration amplitude. Then, a feedback electronic circuit processes
and delivers the output current to a digital signal processor (DSP) to figure out the
resonance frequency value f0.
Often, “resonant” is considered a detection-actuation mechanism because the
information of the transduction process is directly related to the frequency response
of the resonator. As we will study later, in sensor applications the resonator vibrates
at a resonance frequency f0 that is proportional to the magnitude of the physical sig-
nal sensed by the resonator. This is the operation principle of resonant mass detec-
tors, inertial force sensors, and many other applications we review in Chapter 8.
The resonant principle for sensor applications has many advantages, if we com-
pare it with other mechanisms. Perhaps, high sensitivity and resolution, low noise,
and frequency-domain output are the most attractive. It has been demonstrated that
the absolute sensitivity is directly proportional to the working frequency of the sen-
sor. Also, the low-frequency noise sources become less significant when operating
the resonator at higher frequencies. If the Q factor is high, then the resonator is a
quasi-binary device whose response to the excitation signal is highly selective. Nev-
ertheless, resonators suffer from high temperature sensitivity, which cause thermal
drifting that can degrade its frequency response. In such cases, a thermal-compensa-
tion mechanism is required to overcome the fluctuations of the resonance fre-
quency. Another aspect to be considered is that, due to its resonating operation,
resonators present high strain sensitivity and mechanical fragility. The situation
imposes the need of hermetic packaging able to absorb shocks and external
vibrations.

Magnitude

Excitation Detection

Figure 1.6 Working principle of a resonator.

12 MEMS and NEMS Resonator Technologies

There exists a variety of physical principles currently exploited in MEMS and

NEMS resonators operation. Some of them are usually combined to offer a com-
plete transduction system. The list in Table 1.1 shows just some of the possibilities,
along with their advantages and drawbacks. Of course, other combinations are also
possible, depending on the system and the size of the resonator.
Piezoelectric resonators are used as sensors, actuators, or both. An electric
potential applied to the piezoelectric material of the resonator produces a mechani-
cal strain and the resonator bends. Alternatively, a mechanical stress applied to the
piezoelectric material induces an electric charge on the resonator surface propor-
tional to the mechanical force. Piezoelectric crystals used in the fabrication of
MEMS resonators include lead-zirconium-titanate (PZT), zinc oxide (ZnO), and
aluminum nitride (AlN), among others. Most acoustic resonator technologies
involve the piezoelectric effect, which is discussed in detail in Chapter 2.
Electrostatic actuation of a MEMS cantilever with capacitive detection of the
resonance frequency is illustrated in the drawing of Figure 1.7. The cantilever is
driven into resonance when an AC electric signal of the appropriate frequency is
applied to one electrode—namely, the driver electrode. The cantilever is biased with
a DC voltage to induce the electrostatic charge of the structure at the parallel-capac-
itance system formed by the driver and the cantilever. At the other side, a read-out
electrode collects the capacitive-transduced current i0. The system is modeled as a
two-port configuration, because two independent ports are used for the driver and
read-out electrodes. In a simpler model, the same electrode is employed for driving
the cantilever and reading out the transduced current.
The variable capacitance formed between the cantilever and the electrodes can
be described by the parallel-plate capacitor model C(x):

εA
C( x , t ) = (1.11)
( 0 x (t ))
d −

where A is the parallel-plate area formed between the faced surfaces of cantilever
and electrode, d0 is the rest distance between the plates, and x(t) is the time-varying
separation induced by the actuation. The voltage V applied to the driver and cantile-
ver causes an electrostatic force FE acting on the cantilever, related to the electric
work WE:

Table 1.1 Transduction Mechanisms in MEMS and NEMS Resonators

Excitation Detection Advantage (+) Drawback (–)
Piezoelectric Piezoelectric Large dynamic range; acoustical Needs IC-compatibility development
isolation technologies available
Electrostatic Capacitive Surface micromachining Complex process if only bulk
micromachining is available
Electrothermal Piezoresistive Effective and simple; bulk Heat generation power; bandwidth
micromachining available
Optothermal Optical Electrical isolation; immunity to Hard to integrate
most electromagnetic noise
Magnetic Magnetic Effective and simple; variety of Magnetic materials or magnets required;
excitation mechanisms magnetic fields in the system
1.2 Physical Fundamentals of MEMS and NEMS Resonators 13

VDC

i0 (x, t)

vAC(f )
v 0 (x, t)

C(x, t) C(x, t)
x(t) x(t)

Figure 1.7 Electrostatic actuation of a cantilever resonator and capacitive detection of the reso-
nance: AC voltage is applied to the driver electrode, which generates the electrostatic force acting
on the cantilever. The resonance is detected by the read-out electrode in the form of a collected
AC current. The cantilever is biased to maximize the amplitude of the current.

∂W E ∂ ⎛1 2⎞ 1 εAV 2
FE ( x ) = = ⎜ CV ⎟ = (1.12)
∂x ∂x ⎝2 ⎠ 2 (d − x )2
0

V accounts for both the driving and biasing voltages vAC and VDC, and C
includes both the static capacitance C0 and C(x, t). The relationship between FE, vAC,
and VDC is defined through the change of C(x):

∂ C( x )
F E ( x ) = VDC v AC (1.13)
∂x

According to the displacement current definition of capacitors, the electric cur-

rent at the read-out electrode i0(t) is expressed as:

∂
i 0 (t ) =
∂t
(C( x , t ) ⋅ V ) (1.14)

The current i0(t) is calculated as:

∂ v AC ∂ C(t )
i 0 (t ) = (C o + C( x )) + (VDC + v AC ) (1.15)
∂t ∂t

Since C(x) and vAC are small in comparison to C0 and VDC, respectively, (1.15)
can be approximated to:
14 MEMS and NEMS Resonator Technologies

∂ v AC ∂ C(t )
i 0 (t ) ≈ C o + VDC (1.16)
∂t ∂t

In (1.16) the second term at the right side is dominant, so this equation could be
further simplified if needed. Note that the restoring force of the cantilever FX
opposes to FE. Along with the dimensions and geometry, it imposes the mechanical
conditions for the resonance frequency. The images of Figure 1.8 show a cantilever
placed between the two electrodes, when it resonates at the fundamental and second
flexural modes (upper and lower images, respectively). Considering the previous
analysis, one can conclude that the magnitude of the collected current at the
read-out electrode will be higher when the resonator and electrode gap is reduced,
their faced surfaces increased, and the biasing voltage augmented. Such parameters
need careful design to avoid sticking, collapsing, or saturation of the resonator.
Magnetic transduction of MEMS devices can be implemented through different
physical mechanisms. The main are electromagnetic induction [33], Lorentz force
[34], and ferromagnetic attraction. Specifically, resonant magnetic transducers
vibrate when they are in the presence of a time-varying magnetic field with oscilla-
tion frequency equal to one of the natural frequencies of the MEMS.
The Lorentz force explains the motional force acting on a conductor on a point
charge due to electromagnetic fields, which is done in terms of electric and magnetic
fields by:

F = q( E + v × B) (1.17)

where F is the force, E is the electric field, B is the magnetic field, q is the electric
charge of the particle, v is the instantaneous velocity of the particle, and × is the vec-
tor cross product. According to the Lorentz force mechanism, the MEMS resonator
will be accelerated and orientated in the same axis of the E field, with perpendicular

Figure 1.8 Cantilever resonator electrostatically actuated: the first and second flexural modes are
observed in the scanning electron microscope (SEM) image. (Image courtesy of Dr. Zachary J.
Davis, DTU Nanotech, Copenhagen, Denmark.)
1.2 Physical Fundamentals of MEMS and NEMS Resonators 15

and instantaneous variations following the instantaneous velocity vector v and the
B field according to the right-hand rule. The term qE is called the electric force,
while the term qv B is the magnetic force. Some definitions simplify the term
“Lorentz force,” referring specifically to the magnetic force component:

F = q( v × B) (1.18)

The magnetic force component of the Lorentz force manifests itself as the force
that acts on a current-carrying wire in a magnetic field. This is the principle used in
the design of Lorentz force–driven MEMS magnetic transducers. When
magnetomotive excitation is employed, the MEMS structure is covered by a metal-
lic layer and placed in the magnetic field, where an electric current is guided through
the metal conductor. This induces the Lorentz force used to drive the MEMS into
resonance. Alternatively, the MEMS can detect the presence of a magnetic field.
Current flowing through the conductive loop of the resonator aligns with the mag-
netic field. The MEMS is driven into resonance to maximize the sensitivity.
In ferromagnetic-based transducers, the MEMS resonator incorporates a soft
magnetic material or a hard magnet, as depicted in Figure 1.9. The resonance fre-
quency of the MEMS shifts due to the strain added to its structure when the external
magnetic field and magnetic component of the resonator interact. The magnitude of
the frequency shifting is used to determine the amplitude or the direction of the
external magnetic field.
Referring to (1.1), the external force Fext is the ferromagnetic attraction force
FM:
r
F M = ∇( m ⋅ B) (1.19)
r
where m is the magnetic moment and B is the time-varying magnetic field. If a per-
manent magnet is attached to the resonator, (1.19) can be rewritten as:
r
F M = m ⋅ ∇B (1.20)

r ∂ B( x , t )
FM = m ⋅ (1.21)
∂x

Thus, the time-varying ferromagnetic force attraction between the external field
source and the MEMS depends on the variations of the magnetic field with both dis-

x(t)

i(t)

Figure 1.9 Ferromagnetic actuation of a MEMS cantilever.

16 MEMS and NEMS Resonator Technologies

tance and time. As in the case of electrostatic transduction, the restoring force of the
MEMS FX given by Hooke’s law:

FX = − k ⋅ x (1.22)

opposes to the resonator deflection caused by FM in order to balance the system:

F X = −F M (1.23)

The spring constant k of the MEMS/NEMS resonator, along with its dimen-
sions and geometry, imposes the mechanical conditions leading the structure to res-
onate at a certain frequency. Figure 1.10 shows a quad-beam MEMS resonator with
a NdFeB permanent magnet attached to it. The structure resonates at a frequency of
270 Hz when an AC current induces the external magnetic field. A coil placed
underneath the resonator generates the current. The detection of resonance is car-
ried out by optical systems (the resonance curve is shown in the inset).
Electrothermal actuation implements a heating resistor located on a movable
MEMS/NEMS structure. The heater dissipates the power of an electric signal flow-
ing through it. The dissipated power is converted into heat and the MEMS experi-
ments thermal expansion or contraction according to its mechanical configuration,
temperature, and environmental conditions. Since the thermal signal follows the
voltage waveform, the MEMS can be driven into resonance if the electric signal is
chosen to be of a frequency corresponding to one of the MEMS eigenvalues. The
driving moment of the MEMS is proportional to the power dissipation P in the
heater:

V2
P= (1.24)
R

where V is the voltage signal and R is the resistance of the heater, and the voltage-
dependent component of P is:

Figure 1.10 Magnetic transducer implemented with MEMS resonator and permanent magnet
attached to it (resonance curve in the inset).
1.2 Physical Fundamentals of MEMS and NEMS Resonators 17

V 2 = (VDC + v AC sin 2 πft )

2
(1.25)

⎛ 2 v ⎞
= ⎜VDC + AC ⎟ + 2 ⋅ VDC ⋅ v AC ⋅ sin 2 πft − v AC
2
cos 4πft (1.26)
⎝ 2 ⎠

In (1.26) VDC and vAC are the DC and AC voltage components of V, respectively,
and f is the frequency. If this excitation signal frequency f equals one of the natural
frequencies of the MEMS f0, it resonates. The second term in (1.26) shows that the
dissipated power on the resistor depends on both the DC and AC signal amplitudes
and their frequencies. The situation will drive the MEMS with more or less vibra-
tion energy.
The detection of resonance can be performed by using implanted piezoresistors
of resistance RP, and taking advantage of their piezoresistive behavior:

1
ΔR ∝
2
∏ ⋅σ (1.27)

where ΔR is the change of the resistance value due to deformation of the

piezoresistor, Π is its piezoresistive coefficient, and σ is the stress in the MEMS
structure caused by the heating-induced deformation. In silicon MEMS, the heaters
and the piezoresistors are implanted in the substrate by ion implantation techniques
currently employed in standard IC processes. The implantation can be made of
boron, phosphorus, or a suitable material according to the available process and the
designed resistivity values. In sensor applications, the resistors are placed at a
stress-sensitive point of the MEMS, as illustrated in Figure 1.11. The piezoresistors
are usually arranged in a Wheatstone bridge configuration to balance the output
voltage signal vOUT, given that the bridge is biased with the DC voltage VBIAS. In the
example, the MEMS device is a beam, although cantilevers or membranes are also
often implemented. At resonance, vOUT is the time-varying signal of frequency f0
expressed by:

vAC (f )

Figure 1.11 Electrothermal actuation with piezoresistive detection: the heater induces time-vary-
ing power dissipation on the beam, which resonates if the driving voltage corresponds to the
beam’s natural frequency. The piezoresistors detect the resonance, converting the mechanical sig-
nal into the electrical domain by a Wheatstone bridge.
18 MEMS and NEMS Resonator Technologies

ΔR(t )
vOUT (t ) = VBIAS (1.28)
RP

Many other transduction mechanisms can induce resonance of MEMS and

NEMS devices, too, like piezoelectric excitation and detection, which will be dis-
cussed in detail in Chapter 2. Most of them are suitable for both MEMS and NEMS.
However, not all can be scaled down to the NEMS dimensions (e.g., optical detec-
tion of resonance of NEMS with dimensions less than the spot diameter of lasers in
AFM systems or other microscopy systems). On the other hand, NEMS resonators
are sensitive to electrostatic and magnetic actuations, among other mechanisms [35,
36]. When placed in the middle of electron-beam fields, the resonance can be
induced due to electrostatic mechanisms [37]. The gallium nitride (GaN) shown in
the TEM image of Figure 1.12 resonates due to the electrostatic excitation of a
probe approaching to the nanowire and carrying the oscillating signal, while the
nanowire substrate, TEM stage, and signal generator are grounded to the same volt-
age reference [38]. In the end, the suitability of transduction depends on the struc-
ture and excited mode of the resonator, as we will study next.

1.2.4 Resonance Frequency, Mode Shaping, and Aspect Ratio

There are no restrictions to design the structure of MEMS resonators. Nevertheless,
they are usually designed with certain geometries that have demonstrated their effi-
ciency: clamped-clamped beams, cantilevers, disks, rings, squares, membranes,
combs, and triple and multiple beams, for example. NEMS resonators typically
have the form of clamped-clamped beams, cantilevers, nanotubes, or nanowires,
among others. Each one of them has its own process and performance
characteristics.
Rigidity, intrinsic and design-related stress, and the layout and structural mate-
rial of the resonator make the actuation more efficient at certain resonance modes.
The mode shaping strongly links with the preferred mechanical configuration and
aspect ratio of the device, provided it is driven with a signal of the corresponding

1 μm

Figure 1.12 Gallium nitride (GaN) nanowire resonating at 2.2 MHz due to electrostatic excita-
tion mechanisms: the nanowire is excited with an oscillating signal provided by a probe approach-
ing to it, while the nanowire’s substrate is grounded to the voltage reference of the system. (©
2006 American Chemical Society [38].)
1.2 Physical Fundamentals of MEMS and NEMS Resonators 19

frequency. Thus, long beams and cantilevers resonate at flexural, torsional, or

extensional modes with good electromechanical conversion [39, 40]. By changing
the aspect ratio of dimensions, bulk acoustic modes can be promoted in beams as
well [41]. Disk resonators, on the other hand, exhibit good electromechanical per-
formance at their fundamental radial and wine-glass modes [42, 43]. The natural
modes of square and ring resonators also have preferred radial resonances [44, 45].
Long beams can resonate at higher-order radial modes, but the electromechanical
conversion efficiency will be low at those modes. The schematics and pictures of
Figure 1.13 show various examples of MEMS resonators and the mode shaping of
their resonance modes.
The generic model of the harmonic oscillator described by (1.1) and (1.2) is
adapted to each resonator design, according to its mechanical particularities and the

(a)

(b)

Ground
electrode
Input
electrode AI

Pt
Ground
electrode AIN

(c)

(e)

(d)
Figure 1.13 MEMS resonators and their resonance-mode shaping: (a) flexural (cantilever) (courtesy of:
Zachary J. Davis, DTU Nanotech, Copenhagen, Denmark); (b) bulk (beam) (© 2007 IEEE [41]); (c) radial
(ring) (© 2005 IEEE [45]); (d) radial (disk) (© 2007 IEEE [42]); and (e) wine-glass (disk) (© 2005 IEEE [43]).
20 MEMS and NEMS Resonator Technologies

Euler-Bernoulli equation. For example, the formulation of the resonance frequency

of flexural-mode beams, equivalent to (1.2), is:
2
1 ⎛γn ⎞ E⋅I ⋅L
fn = ⎜ ⎟ (1.29)
2π ⎝ L ⎠ m

where E is the Young’s module, I is the inertia moment, L is the length of the beam,
m is the mass, and γn is a dimensionless parameter. Its value is numerically calcu-
lated for the nth mode and defined by the components of the Euler-Bernoulli equa-
tion leading to (1.29). Whether the beam is clamped at both sides or only by one of
its ends—a cantilever—the values of γn are different for a given resonance mode (n).
Similarly, the cross-section area and the vibration plane of the resonance define the
value of I. If the oscillation is in the vertical plane, and the beam is a cantilever with
rectangular section of width w and thickness t, the inertia moment is:

w ⋅t 3
rect
I vert = (1.30)
12

Replacing (1.30) in (1.29), the spring constant k of (1.2) can be rewritten as

E⋅ w ⋅t 3
k= (1.31)
4L3

And the effective mass nth mode of the cantilever is

⎛ 3⎞
n
m eff = ⎜ 4 ⎟m (1.32)
⎝γn ⎠

The first three-mode values of γn calculated for a cantilever are γ1 = 1.875, γ2 =

4.694, and γ3 = 7.854. Now we ask ourselves: can this analysis be extendable to
clamped-clamped (c-c) beams or to other MEMS or NEMS resonator geometries?
The answer is yes. However, because of their different boundary conditions, c-c
beams and cantilevers have different inertia moments, spring constants, and effec-
tive masses. For the rectangular-section c-c beam, k and m effn
are, respectively:

16 ⋅ E ⋅ w ⋅ t 3
k= (1.33)
L3

⎛ 192 ⎞
n
m eff = ⎜ 4 ⎟m (1.34)
⎝ γn ⎠

Finally, the first three-mode values of γn calculated for the c-c beam are γ1 =
2.365, γ2 = 3.927, and γ3 = 5.498. These examples have shown us how the reso-
nance frequency and model constants of the resonator change by applying different
boundary conditions. According to (1.31) to (1.34), we see that, for a given reso-
nance mode, the cantilever is more flexible, has a bigger effective mass, and reso-
nates at a lower frequency, if we compare it with the c-c beam.
1.2 Physical Fundamentals of MEMS and NEMS Resonators 21

By using the previous equations, Table 1.2 shows calculated values of the flex-
ural-mode resonance frequencies of c-c beam and cantilever of resonators of differ-
ent dimensions. We assume silicon-made beams with E = 160 GPa and ρ = 2,330
kg/m3, and we corroborate that the higher rigidity of the c-c beam makes it resonate
at higher frequencies. The calculations also demonstrate that by reducing the
dimensions of the cantilevers and the beams from the MEMS to the NEMS scale, it
is possible to reach fundamental frequencies in the range of gigahertz. With these
dimensions, not only are higher resonance frequencies achieved, but also the force
constants are kept at small values. These attributes make NEMS resonators highly
sensitive in force-detection applications with ultralow power operation, as we study
in Chapter 8.
From the previous analysis and observing Table 1.2, it seems logical to be
interested in scaling down the size of MEMS to enter the NEMS regime. However,
several difficulties arise if MEMS technologies are scaled from micrometers to
nanometers. High resonance frequencies can only be achieved if the aspect ratio of
the resonator is close to unity (L/w~ 1, L/t ~ 1). This implies very high force con-
stants, thus requiring high power levels of the excitation signal to obtain an appre-
ciable mechanical response. Therefore, if the force constants are high, the
low-power feature is diluted, and then the minimum power, the dynamic range,
the tuning capability, and the Q factor performances are negatively affected.
Besides, the same fabrication technology is employed to go down from the MEMS
to the NEMS scale. As we study in Section 1.4, the planar IC-like approach for
fabricating MEMS requires that thickness is the same for big surface or small sur-
face devices. Thus, it is impossible to perform scaling down in three dimensions,
but just on the lateral (width and length) axes. The images and resonance curves of
Figure 1.14(a–d) show four resonators made of silicon carbide (SiC) within the
same process and scaled down in its lateral dimensions. The 70-nm thickness is the
same in all resonators. We can see how the aspect ratio changes, and, as we reduce
the dimensions, the resonance frequency increases. However, the vibration ampli-
tude diminishes, thus reducing the Q factor. From the figures, it is also noticeable
that the Q factor reduction causes the in-resonance signal to approach the noise
levels. Similar analysis can be done for the more complex geometries commented
earlier, in order to synthesize their spring-mass constants and design equations
(see [46]).
The previous ideas illustrate the reasons for investigating and implementing
nanofabrication techniques to fabricate NEMS devices, rather than current
IC-based processes, widely used in MEMS fabrication. This matter and the produc-
tion cycle of MEMS and NEMS resonators are studied in the next sections.

Table 1.2 Flexural-Mode Resonance Frequencies of Si Cantilever and c-c Beam

Resonators (Three Dimensions Are Compared)
Dimensions (L w t)
Boundary condition
10 mm 200 nm 1 mm 50 nm 100 nm 10 nm
100 nm 50 nm 10 nm
Clamped-clamped beam 2.13 MHz 106 MHz 2.13 GHz
Cantilever 1.34 MHz 66.9 MHz 1.34 GHz
22 MEMS and NEMS Resonator Technologies

(a) (b) (c) (d)

15 15
400

300
ΔR/R(10 )

10 10
−6

200

5 5
100

0 0 0

50 k 55 k 1.56 M 1.59 M 7.9 M 8.0 M 126 M 128 M

Frequency (MHz)

Figure 1.14 The aspect ratio of fabricated MEMS and NEMS resonators: the four resonators in
the SEM images have a thickness of 70 nm and lateral dimensions of: (a) 33 μm × 5 μm; (b) 10 μm
× 2 μm; (c) 2.5 μm × 0.8 μm; and (d) 0.6 μm × 0.4 μm. (© 2007 Nature Publishing Group [47].)

1.3 Key Fabrication Technologies

The resonator fabrication technologies are the same as those used in other MEMS
and NEMS device fabrication. We divide them in two categories: (1) common to IC
fabrication (suitable for MEMS), and (2) nanofabrication techniques (suitable for
NEMS). In both cases, MEMS and NEMS fabrication is costly in time, materials,
services, equipment, and man power. Thus, the design and production infrastruc-
tures are not affordable for most of the companies or research centers. External
foundries have adapted their production lines to MEMS manufacturing and offer
fabrication services to universities and small companies through public access pro-
grams like Europractice. The turnaround times for fabricated devices are about
three months.
For these reasons, fabrication is critical, and previous design activities have to
be rigorously completed to guarantee successful manufacturing results. From con-
cept to application, there is a production cycle involving a set of engineering activi-
ties, which follow a logical sequence within the process. The section explains this
cycle and describes some of the fabrication technologies of MEMS and NEMS
devices in a general way. Later in Chapters 4 and 10, we will provide an extensive
description of these technologies applied to acoustic microresonators fabrication.

1.3.1 The Production Cycle

Once we have defined the geometry, boundary conditions, and materials, the design
equations presented in the previous section aid the analytical modeling of the reso-
nator. Figure 1.15 depicts the flow diagram of the production cycle.
First, we define the resonator concept basing our decision on the expected appli-
cation and the possible integration with IC technologies. The concept involves
defining the generalities of the fabrication process and the interconnection technol-
ogy with the IC, compatibility issues with the IC, temperature budget, and materi-
als. At the layout level, it also requires previsions of the resonator active area,
1.3 Key Fabrication Technologies 23

Start

Resonator concept
and process design

Analytic and finite

element modeling

Prototype fabrication
and characterization

Parameter extraction Redesign

Desired No
frequency, Q?

Yes

Application
dimensioning

Figure 1.15 The production cycle of MEMS and NEMS resonators.

electrode design and contact pad area, dicing areas (if needed), or via holes for inter-
connection (if needed). If we intend to fabricate the resonator by using the layers of
a microelectronic process, we need to choose the appropriate layers of the technol-
ogy. Later in Chapter 7, we will study the integration strategies for this purpose.
Then, we perform analytical and finite element modeling (FEM) of the resona-
tor. In a first approximation, the design equations presented in the previous section
are enough to estimate the spectral range in which we expect to find the resonance
frequency. With the design dimensions, we can build a finite element model. The
FEM analysis is useful to extract the eigenvalues of the structure and to predict the
structural and harmonic response due to external signals or forces. Commercial
FEM tools offer many possibilities, so we can succeed in building a complex and
reliable resonator model, which we will use to fix the final resonator process and
dimensions.
A MEMS prototype can now be fabricated. Different test and de-embedding
structures are used at this stage to extract the resonator material constants, quality
factor, equivalent-circuit parameters, insertion losses, and parasitic impedances.
Fabricated devices are then characterized with available measurement instrumenta-
tion. The main characterization results are the resonance frequency and the quality
factor, whose values will validate our design or will force us to redesign the
resonator to reach the design values.
24 MEMS and NEMS Resonator Technologies

If the resonance frequency has acceptable design values, extraction of the

equivalent-circuit parameters is carried out. These parameters are a circuit-like elec-
trical representation of the harmonic oscillator, and they can be used in further
analyses, especially when we intend for future IC integration or read-out circuit
interconnection. Even when the resonance frequency fits with acceptable tolerance
within the design values, other design elements have to be considered, like the inser-
tion losses or the quality factor. These elements are, as we commented before, key
aspects of the application design. Since the success of the production cycle mainly
depends on the fabrication technologies, we provide a detailed analysis of them in
the next section.

1.3.2 Common to Integrated Circuit (MEMS)

Microfabrication technologies common to IC are a wide family of processes and
techniques comprising the following: carrying substrates—mainly silicon, photo-
lithography, etching, implantation, dielectric and conductive layer deposition, and,
in general, all processes currently employed in integrated circuit manufacturing.
Most of them have been adapted to produce MEMS devices, but not necessarily
modified in their essence and principle: parallel processing and planar technology.
Besides, micromachining complements the list to produce movable structures. The
process cycle involving these technologies is depicted in Figure 1.16. The following
sections discuss some of them.

1.3.2.1 Silicon Substrates

Silicon (Si) is the most employed substrate in MEMS and NEMS device fabrication.
Si crystal orientation, doping, and wafer size are chosen depending on the applica-
tion. Si wafers of types N or P with diameters of 100 to 300 mm and thicknesses
from 300 μm to 1 mm are available in the market. Crystal’s orientation (e.g., 100,
111) and doping types determine the wafer resistivity and fabrication time of
micromachined MEMS, among other parameters. Different methods exist to fabri-
cate Si wafers. Thus, the resistivity varies from the 0.002–50 Ω·cm obtained with
the Czochralski technique, to the 20–100 Ω·cm in the Floating Zone method [48,

Figure 1.16 Process of common-to-IC MEMS fabrication.

1.3 Key Fabrication Technologies 25

49]. Nevertheless, the presence of impurity materials in Si wafer fabrication is

14 3
unavoidable, with concentrations of oxygen lower than 1 × 10 atoms/cm , carbon
12 14 3
impurities between 1 × 10 − 1 × 10 atoms/cm , and heavy metals being the main
materials found in the wafers.
P-type and N-type wafers are fabricated by introducing impurities of the III and
V groups of the periodic table in the Si, respectively. III-group materials widely used
are arsenic (As) and phosphorus (P), whereas boron (B) is the V-group material
most employed in P-type doping. Ion implantation using accelerated isotopes to
50–200 keV achieves doping at rates between 1011 − 1014 atoms/cm3. However, the
carrier concentration, which depends on the wafer type (N or P), is not uniform and
may change within the wafer and between the different wafers of a lot. Impurity dif-
fusion contributes to a redistribution of charges through high temperature processes
(1,000ºC).
Silicon-on-insulator (SOI) wafers are another kind of substrate widely used in
MEMS and NEMS fabrication [50]. SOI wafers are constituted by a Si device layer
(DEV), a highly resistive buried oxide layer (BOX), and the Si substrate, as depicted
in Figure 1.17. Commercial SOI wafers are available with DEV and BOX layers
having thicknesses in the range of 1 to 100 μm and 0.5 to 5 μm, respectively [51].
The SOI technology finds good application in MEMS micromachining because it
provides high control of etching. MEMS structures can be fabricated with the DEV
or BOX layers, whose thicknesses are previously known. Also, SOI wafers aid in
reducing the high coupling losses of standard Si wafers, because the DEV and BOX
layers offer less resistance and better isolation, respectively. In RF applications, this
is a key feature for obtaining good performance.
Other substrates for MEMS and NEMS applications offer high resistivity and
flexibility so far [52]. Additionally, transferable electronics and MEMS have pro-
moted the development of flexible substrates different from silicon [53].

1.3.2.2 Oxidation of Silicon

Oxidation of silicon wafers has several applications in microdevice fabrica-
tion—substrate passivation (the formation of a chemically, electronically,
and electromechanically stable surface), diffusion, ion implantation, dielectric
thin-film making, and intersubstrate material interfacing, among others [54]. Gen-
erally speaking, two kinds of oxides are fabricated: thermal oxide and deposited
oxide.

Figure 1.17 Silicon-on-insulator (SOI) technology as a high-performance substrate for MEMS and
NEMS resonators.
26 MEMS and NEMS Resonator Technologies

Thermal silicon oxide (SiO2) grows on Si when the air-Si interface is oxidized at
room temperatures. Therefore, nanometer-thick SiO2 thin films cover the whole Si
wafer surface. Thicker SiO2 layers can be grown at elevated temperatures after dry
or wet atmosphere reactions:

• Dry reaction (with oxygen): Si + O2 → SiO2;

• Wet reaction (with water): Si + 2H2O → SiO2 + H2.

Growth rate and oxide characteristics depend on concentration and quality of

the oxidant, pressure, temperature, diffusivity, and oxide thickness, among others.
For constant-temperature oxidation, the relationship between oxide thickness and
time is parabolic. The oxygen-based process is slow but it offers high-quality oxide,
whereas the water-based oxidation is faster but at the cost of lower oxide quality. In
both cases, typical oxidation temperatures vary from 950ºC to 1,050ºC.
Deposited SiO2 is obtained through thin-film deposition techniques, and it can
be processed on Si or on other substrates as well. The obtained oxide may have
thicknesses of hundreds of nanometers and is, in general, of better quality than the
thermal oxide, thus exhibiting higher resistivity and dielectric properties. The depo-
sition techniques of SiO2 are common to other dielectric materials, as we review in
the following section.

1.3.2.3 Dielectric Layer Deposition

Dielectric layers—as SiO2—can be deposited using a variety of techniques, such as
epitaxial growing, chemical vapor deposition (CVD), and molecular-beam grow-
ing. Epitaxial is especially useful for synthesizing high-quality dielectric layers in a
monolayer basis, where low processing times and high temperature promote highly
oriented crystal growing. Highly doped buried monocrystalline silicon or columnar
piezoelectric materials like aluminum nitride (AlN) are two application examples.
High deposition temperatures—above 950ºC–1,250ºC—restrict the application of
epitaxial for previous-CMOS processing.
CVD is based on chemical reaction of the vapor-phase reactant with chemical
radicals found inside the CVD machine chamber. The product of reaction is then
deposited on the substrate, usually at a slow rate, as represented in the schematic of
Figure 1.18. Many variants of the CVD process exist, namely:

• High temperature chemical vapor deposition (HTCVD);

• Low temperature chemical vapor deposition (LTCVD);
• Plasma enhanced chemical vapor deposition (PECVD);
• Low pressure chemical vapor deposition (LPCVD).

The versatility of CVD allows the fabrication of polysilicon, numerous kinds of

oxides, BSG, PSG, BPSG, SiO2, and nitrides, Si2N3, Si3N4, among others.
1.3 Key Fabrication Technologies 27

Vapor-phase
To ,P reactant

Chemical
radicals

Deposited
species
Substrate

Figure 1.18 Chemical vapor deposition.

1.3.2.4 Conductive Layer Deposition

Metal evaporation, physical vapor deposition (PVD), and electroplating a deposi-
tion are the main techniques for conductive layer deposition. Sputtering, for exam-
ple, is a technique of the PVD family that creates a source of particles being
subsequently condensed onto the substrate, thus forming a thin film. Physical sput-
tering is the process where atoms are ejected from a surface as a consequence of the
bombardment of the latter by heavy energetic particles (ions). Under the adequate
energetic conditions and low-pressure environment, low-density plasma is formed
by electron impact ionization of the gas in a controlled electrical gas discharge. Low
ionization degrees in the range of 10−5 to 10−2 and processing pressures of 1–30
mTorr correspond to an ion density of 109 to 1012 cm3 [55]. Due to these physical
conditions, the sputtering process can be accomplished at relatively low tempera-
tures (i.e., below 400ºC, which is a favorable condition for CMOS compatibility).
Figure 1.19 illustrates the concept of physical sputtering.
A combination of AC, magnetron, and reactive sputtering reports several
advantages to single sputtering deposition. AC sputtering allows discharging of the
target (ion charging occurs in DC sputtering), magnetron increases the attracting
forces and acceleration of ions to the cathode, and reactive sputtering enables
dielectric and compound material deposition using metallic targets. At the end,
higher ion density and more efficient deposition are achieved with lower energy lev-
els (in the hundreds of electron-volts, instead of kilo-electron volts) and power
supplies.
With respect to other deposition techniques, sputtering offers several advan-
tages. Among them, we can mention film uniformity, surface smoothness and thick-
ness control, versatility (virtually any material can be introduced into a gas
discharge or sputtered from the target), good adhesion, conformal or planarized
coating, and higher deposition rates. Most important, the low process temperature
makes the sputtering technique compatible with CMOS fabrication [56].
28 MEMS and NEMS Resonator Technologies

Cathode
target

Ar

Sputtered
precursors

Substrate
Anode

Figure 1.19 Physical sputtering.

1.3.2.5 Conventional Photolithography

Photolithography is an optical technique to transfer the patterns from a photo mask
or reticule to the wafer. Thus, the system involves four elements: ultraviolet (UV)
light source, mask or reticule, photoresist, and the wafer where the patterns are to
be transferred to. Each mask or reticule contains only one layer of devices. Thus, the
photolithography exposure is repeated as far as all chip patterns are transferred to
the wafer. The process involves a consecutive number of resist spinning and coating,
baking, developing, layer patterning, and resist removal [57].
The conventional photolithography process is illustrated with one example in
Figure 1.20, which shows a simplified schema of the patterning of a SiO2 layer.
First, the photo-resist is uniformly deposited on the wafer by spin coating. Then, the
mask or reticule is charged in the exposure line and aligned with the wafer. When

Figure 1.20 Conventional optical photolithography of a SiO2 pattern: (a) SiO2 layer deposition;
(b) photo-resist coating; (c) mask alignment and UV exposure; (d) resist developing; (e) SiO2 etch-
ing; and (f) resist removal.
1.3 Key Fabrication Technologies 29

exposed to UV radiation, the molecular structure of the light-sensitive resist is bro-

ken. After resist developing, two cases are considered, depending on whether the
resist is positive or negative. With positive resist, only resist areas under the patterns
in the mask remain after developing, while the opposite occurs for negative resist.
After the wafer is cleaned, the resist remaining on the wafer serves as mask to selec-
tively pattern the SiO2, according to the resist case. After SiO2 etching, the resist is
removed and the wafer can be subsequently been processed.
The mask is made of a transparent, rigid dielectric, typically fused silica glass or
quartz covered with a chrome film defining a pattern that allows UV light to pass
through the transparencies. In standard contact photolithography, patterns are
designed at 1:1 scale, whereas stepper and scanner technologies shrink them by four
or five, thus obtaining reduced-size devices. Photo masks are manufactured by large
commercial companies, as the cost of setting up a mask shop is some hundred mil-
lion dollars [58–60]. However, major semiconductor companies fabricated their
own masks [61–64]. UV-light wavelength identifies the different types of exposure
lines: 365-nm i-line, 436-nm g-line, and deep UV (DUV) below 300 nm. DUV
examples are the 248-nm krypton fluoride, 193-nm argon fluoride, and 157 nm.
Phase shifting and immersion lithographies are other photolithographic techniques
intended for enhanced resolution below 150 nm. The immersion technique, for
example, achieves resolutions as small as 37 nm. Extreme UV (EUV, 13.5 nm),
x-ray, and electron-ion sources technologies have been developed to reduce the
pattern sizes to even smaller wavelengths of few nanometers.

1.3.2.6 Bulk Micromachining

Bulk and surface micromachining are etching techniques for MEMS, which are
based on the modified CMOS technology. In bulk micromachining, the substrate is
patterned to form the three-dimensional, movable structure of the MEMS device (it
is rarely used for NEMS).
Bulk micromachining of Si was developed to fabricate movable microstruc-
tures. It uses wet- and dry-etching techniques in conjunction with masks and
etch-stop layers to define three-dimensional microstructures [65]. Bulk
micromachining can be performed by the front or the back of the substrate through
wet- and dry-etching processes as depicted in Figure 1.21. Wet-etching processes
utilize etchants attacking the different crystallographic directions at slower or faster
rates. Examples of anisotropic etchant solutions are potassium hydroxide (KOH),
sodium hydroxide (NaOH), and ethylene-diamine-pyrocatecol (EDP).
Dry etching is carried out through reactive-ion etching (RIE) and deep reactive-
ion etching (DRIE). RIE and DRIE make use of plasma to etch straight-walled struc-
tures (e.g., cubes or rectangles). Depending on the recipe, RIE-based dry etching is
useful for both isotropic and anisotropic etching. Taking advantage of isotropy,
etching can be performed in both the vertical and lateral directions. The feature is
exploited to release movable structures from the front of the wafer. Alternatively,
the RIE recipe can be modified to provide an effective means of back-side processing
with no considerable lateral etching. In this case, the whole bulk of the wafer is
etched, whereas in most front-side etching applications, it is just partially etched
[66].
30 MEMS and NEMS Resonator Technologies

Structural layer
Stop layer

Membrane

Bulk
Si

Figure 1.21 Bulk micromachining.

RIE and DRIE are kinds of sputtering systems in which reactive species are
accelerated to the substrate. Accelerated plasma ions perform the bombardment of
substrate, which is achieved through appropriate biasing conditions and reactive
species. Figure 1.22 illustrates the process. Given a reactive species, etching is selec-
tive to certain materials. Thus, the application and target materials fix the reactive
species to avoid undesired etching of other materials on the substrate [67].

1.3.2.7 Surface Micromachining

In surface micromachining, a sacrificial layer deposited on the surface of wafers is
patterned to form the movable structure of MEMS and NEMS devices. The process
must guarantee high patterning accuracy of the sacrificial layer. After etching, it
leaves free volume for the device.
The thickness of thin films is usually limited to up to 5 μm, thus leading to pla-
nar-type microstructures [54]. In this sense, surface micromachining can be called a
thin-film technology. The advantage of surface micromachining is the use of stan-
dard CMOS fabrication processes and facilities, which makes the fabrication of
integrated sensors and actuators affordable to research institutes, small companies,
and universities. The typical surface micromachining process depicted in Figure
1.23 implements a sacrificial layer to provide mechanical support to the subsequent

To ,P
Plasma gas

Venting

Substrate

Figure 1.22 Reactive ion etching.

1.3 Key Fabrication Technologies 31

(a) (d)

(b) (e)

Sacrificial layer
Structural layer
Substrate
(c)

Figure 1.23 Surface micromachining: (a) deposition and (b) patterning (sacrificial layer);
(c) deposition and (d) patterning (structural layer); and (e) etching (sacrificial layer).

layers, and it is removed to release the microstructures. Hence, one or more

thin-film layers made of structural—the final device—and sacrificial materials are
deposited and patterned on the surface of the wafer. After completion of the struc-
tural layer deposition, the sacrificial layer is removed, and the whole device is
released [68].
Usually, the sacrificial layer is a dielectric material, although a conductive or
semiconductor material may also be employed. Currently implemented sacrificial
layers are made of SiO2, phosphorous-doped silicon dioxide (PSG), or silicon nitride
(Si3N4) [69]. After fabrication of the structures, wet etching of the sacrificial layer
forms cavities underneath the surface components, which allows releasing and
eventually the motion of the device. The wet etching can be done by using hydroflu-
oric acid (HF), buffered hydrofluoric acid (HF), KOH, EDP, TMAH, or NaOH,
among others.
The size of patterned features that can be obtained with the foregoing tech-
niques is in the order of microns, although smaller features can be obtained with
ultra-thin films—tens or hundreds of nanometers—and the help of appropriate
tools, like critical point dryers (CPDs). CPDs are useful to avoid sticking of
submicron-sized membranes, cantilevers, and other structures resulting from the
sacrificial layer etching. Nevertheless, NEMS resonators usually require advanced
fabrication technologies to guarantee accuracy and process repeatability. The next
section introduces the main nanofabrication techniques.

1.3.3 Nanofabrication Techniques (NEMS)

Nanofabrication mainly refers to lithography techniques that allow manufactur-
ing of nanometer-sized structures for nanoelectronic circuits and NEMS.
Nanofabrication is addressed by two complementary approaches: top down and
bottom up. Top down is the nanoscale-updated methodology common to IC manu-
facturing. Top down involves the fabrication of small structures by patterning the
“big” structural layer (top) to obtain the nanosized feature. On the other hand, bot-
tom up denotes structuring the matter from the “small” scale of molecules—even
atoms with the current techniques—to achieve a nanodevice. Nanosphere lithogra-
phy and molecular self-assembly are two examples of this approach [70, 71]. These
32 MEMS and NEMS Resonator Technologies

Figure 1.24 Taxonomy of nanofabrication techniques.

nanofabrication techniques and categories are depicted in the taxonomic tree of

Figure 1.24.
Top-down techniques can be categorized in two groups: nanolithography and
nanopatterning. Although the difference may be subtle, nanolithography imple-
ments additional elements to pattern the structure of interest, whereas in
nanopatterning direct interaction between the layer and the patterning tool occurs.
The nanolithographic tools are photo masks, stencils, molds, and printing buffers,
among others. Extreme UV (EUV), soft lithography implementing polymers, x-ray
lithography, nanoimprint lithography (NIL), and nanostencil lithography (nSL) are
examples of the nanolithography group.
Nanopatterning techniques are the subject of intensive research activity. Com-
mercial equipment is available to produce final devices, too. Because they do not use
masks or similar elements, they are also known as maskless lithography. The
expression, though, refers more accurately to a computer-aided serial technique
that implements a mirror and laser or another light source to transfer the patterns to
the wafer in a layer basis. Remarkable examples of the nanopatterning category are
electron beam lithography (EBL), atomic force microscopy (AFM) lithography,
focused ion beam (FIB) lithography, scanning tunneling microscope (STM), dip-pen
nanolithography (DPN), magnetolithography (ML), and a daily growing list of
novel lithography techniques.
Nanofabrication techniques offer many opportunities for NEMS production.
Their main features are lower cost, suitability for NBIC applications, increased
device complexity, biocompatibility and plastic materials implementation, and
1.4 Summary 33

Table 1.3 Engineering Attributes of MEMS and NEMS Resonators

Characteristic MEMS NEMS
Dimensions 100 nm–1 mm Below 100 nm
Resonance frequency kHz–MHz MHz–GHz
(fundamental flexural mode)
Q factor Higher Lower
Power Higher Lower
Aspect ratio >1 (high, in general) ~1 (about unity, in general)
Surface-to-volume ratio Lower Higher
Dynamic range Higher Lower
Active mass (normalized Bigger Smaller
to physical mass)
Sensitivity Good Better
Phase noise Lower Higher
Current applications Sensors, radio frequency Sensors
Technology IC-based (top-down); IC-based (top down) and bottom
parallel production up (nanostructuring); serial
production (if nanofabrication
tools are employed)

smaller sizes. Nevertheless, several challenges of nanofabrication are still to be dealt

with, like reproducibility of nanosized patterns, manufacturing throughput, and
their compatibility to current IC processes.

1.4 Summary

This chapter has introduced the main concepts regarding MEMS and NEMS reso-
nators. Physical phenomena, modeling and transduction principles of resonant
devices, and fabrication techniques have been examined (piezoelectric transduction
will be described in Chapter 2). We have learned that MEMS and NEMS resonators
are differentiated from each other by their size and by their fabrication approach
and physics scaling. Current NEMS engineering is facing challenges solved for
MEMS many years ago. Among them, we find packaging, Q factor, high sur-
face-to-volume ratio, and repeatability. Table 1.3 is a nonexhaustive list of engi-
neering items we need to consider when comparing MEMS and NEMS. The goal is
to contextualize the technologies, performance, applications, and challenges of
both devices (“lower” and “higher” are relative to MEMS and NEMS).

References

[1] Baltes, H., et al., “CMOS MEMS—Present and Future,” Proc. IEEE Intl. Conf. MEMS
2002, January 20–24, 2002, Las Vegas, NV, pp. 459–466.
[2] Bauerdick, S., et al., “Direct Wiring of Carbon Nanotubes for Integration in
Nanoelectromechanical Systems,” J. Vac. Sci. Technol., Vol. B24, 2006, p. 3144.
[3] Husain, A. et al., “Nanowire-Based Very-High-Frequency Electromechanical Resonator,”
Appl. Phys. Lett., Vol. 83, 2003, pp. 1240–1242.
34 MEMS and NEMS Resonator Technologies

[4] Ozin, G. A., A. C. Arsenault, and L. Cademartiri, Nanochemistry: A Chemical Approach to

Nanomaterials, Cambridge, U.K.: RSC Publishing, 2009.
[5] Bachand, G. D., and Carlo D. Montemagno, “Constructing Organic/Inorganic NEMS
Devices Powered by Biomolecular Motors,” Biomedical Microdev., Vol. 2, 2000,
pp. 179–184.
[6] McCord, M. A., and M. J. Rooks, Handbook of Microlithography, Micromachining and
Microfabrication, Vol. II, Bellingham, WA: SPIE, 2000.
[7] Binnig, G., C. F. Quate, and C. Gerber, “Atomic Force Microscope,” Phys. Rev. Lett., Vol.
56, No. 9, 1986, pp. 930–933.
[8] Chou, S. Y., P. R. Krauss, and P. J. Renstrom, “Imprint Lithography with 25-Nanometer
Resolution,” Science, Vol. 272, 1996, pp. 85–87.
[9] National Science Foundation (NSF), Converging Technologies for Improving Human Per-
formance (Nanotechnology, Biotechnology, Information Technology and Cognitive Sci-
ence), 2002.
[10] ST Microelectronics, “System on Chip,” http://www.st.com/stonline/products/technolo-
gies/soc/soc.htm.
[11] Geschke, O., H. Klank, and P. Telleman, (eds.), Microsystem Engineering of
Lab-on-a-Chip Devices, New York: John Wiley & Sons, 2004.
[12] Herold, K. E., and A. Rasooly, (eds.), Lab-on-a-Chip Technology: Fabrication and
Microfluidics, San Francisco, CA: Caister Academic Press, 2009.
[13] Nathanson, H. C., and R. A. Wickstrom, “The Resonant Gate Transistor,” IEEE Trans. on
Electron Dev., Vol. 14, 1967, pp. 117–133.
[14] Southwest Center for Microsystems Education and the University of New Mexico, History
of Microelectomechanical Systems (MEMS), 2008.
[15] Salomon, P., “NEXUS_MNT_Market_Report_III-2005-2009,” NEXUS Task Force (coor-
dination: WTC), 2004.
[16] Apple, http://www.apple.com/iphone/.
[17] Nintendo, http://www.nintendo.com/wii.
[18] Roukes, M. L., “Nanoelectromechanical Systems,” Tech. Dig. of the 2000 Solid-State Sen-
sor and Actuator Workshop, Hilton Head, SC, June 4–8, 2000, pp. 1–4.
[19] Gibson, C. T., D. Alastair Smith, and C. J. Roberts, “Calibration of Silicon AFM
Cantilevers,” Nanotechnology, Vol. 16, 2005, pp. 234–238.
[20] Blom, F. R., et al., “Dependence of the Quality Factor of Micromachined Silicon Beam Res-
onators on Pressure and Geometry,” J. Vac. Sci. Tech., Vol. B10, 1992, pp. 19–26.
[21] Mangiarotty, R. A., “Acoustic Radiation Damping of Vibrating Structures,” J. Acoust. Soc.
Am., Vol. 35, 1963, pp. 369–377.
[22] Hosaka, H., K. Itao, and S. Kuroda, “Damping Characteristics of Beam-Shaped
Micro-Oscillators,” Sens. Actuators A: Phys., Vol. 49, 1995, pp. 87–95.
[23] Vemuri, S., G. K. Fedder, and T. Mukherjee, “Low-Order Squeeze Film Model for Simula-
tion of MEMS Devices,” Proc. 2000 Intl. Conf. on Modeling and Simulation of Microsys-
tems Semiconductors, Sensors and Actuators, San Diego, CA, March 27–29, 2000,
pp. 205–208.
[24] Vignola, J. F., et al., “Loss Mechanisms in MEMS Oscillators,” Proc. SPIE Fifth Interna-
tional Conference on Vibration Measurements by Laser Techniques: Advances and Appli-
cations, Vol. 4827, Ancona, Italy, June 18–21, 2002, pp. 466–477.
[25] Brotz, J., “Damping in CMOS-MEMS Resonators,” M.Sc. thesis, Carnegie Mellon Univer-
sity, 2004.
[26] Quevy, E. P., and R. T. Howe, “Redundant MEMS Resonators for Precise Reference Oscil-
lators,” Dig. Tech. Papers Radio Frequency Integrated Circuits Symp. RFIC 2005, Long
Beach, CA, June 12–14, 2005, pp. 113–116.
1.4 Summary 35

[27] Judge, J. A., et al., “Attachment Loss of Micromechanical and Nanomechanical Resonators
in the Limits of Thick and Thin Support Structures,” J. Appl. Phys., Vol. 101, 2007,
013521.
[28] Pandey, M., et al., “Reducing Anchor Loss in MEMS Resonators Using Mesa Isolation,” J.
Microelectromech. Syst., Vol. 18, 2009, pp. 836–844.
[29] Ono, T., and M. Esashi, “Effect of Ion Attachment on Mechanical Dissipation of a Resona-
tor,” Appl. Phys. Lett., Vol. 87, 2005, 044105.
[30] Zener, C., “Internal Friction in Solids, I: Theory of Internal Friction in Reeds,” Phys. Rev.,
Vol. 52, 1937, pp. 230–235.
[31] Zener, C., “Internal Friction in Solids, I: General Theory of Thermoelastic Internal Fric-
tion,” Phys. Rev., Vol. 53, 1938, pp. 90–99.
[32] Duwel, A., et al., “Engineering MEMS Resonators with Low Thermoelastic Damping,” J.
Microelectromech. Syst., Vol. 15, 2006, pp. 1437–1445.
[33] Kim, Y.-S., et al., “A Class of Micromachined Magnetic Resonator for High-Frequency
Magnetic Sensor Applications,” J. Appl. Phys., Vol. 99, 2006, 08B309.
[34] Greywall, D. S., “Sensitive Magnetometer Incorporating a High-Q Nonlinear Mechanical
Resonator,” Meas. Sci. Tech., Vol. 16, 2005, pp. 2473–2482.
[35] Lassagne, B., et al., “Ultrasensitive Mass Sensing with a Nanotube Electromechanical Reso-
nator,” Nano Lett., Vol. 8, 2008, pp. 373–3738.
[36] Vasquez, D. J., and J. W. Judy, “Flexure-Based Nanomagnetic Actuators and Their Ulti-
mate Scaling Limits,” Proc. IEEE Intl. Conf. MEMS 2008, Tucson, AZ, January 13–17,
2008, pp. 737–741.
[37] Huang, Y., X. Bai, and Y. Zhang, “In Situ Mechanical Properties of Individual ZnO
Nanowires and the Mass Measurement of Nanoparticles,” J. Phys.: Cond. Matter, Vol. 18,
2006, pp. 179–184.
[38] Nam, C. Y., et al., “Diameter-Dependent Electromechanical Properties of GaN Nanowires,”
Nano Lett., Vol. 6, 2006, pp. 153–158.
[39] Holmgren, O., et al., “Analysis of Vibration Modes in a Micromechanical Square-Plate
Resonator,” J. Micromech. Microeng., Vol. 19, 2009, 015028.
[40] Pang, W., et al., “Ultrasensitive Mass Sensor Based on Lateral Extensional Mode (LEM)
Piezoelectric Resonator,” Proc. IEEE Intl. Conf. MEMS 2006, Istanbul, Turkey, Janu-
ary 22–26, 2006, pp. 78–81.
[41] Pourkamali, S., G. K. Ho, and F. Ayazi, “Low-Impedance VHF and UHF Capacitive Silicon
Bulk Acoustic-Wave Resonators—Part II: Measurement and Characterization,” IEEE
Trans. on Electron Dev., Vol. 54, 2007, pp. 2024–2030.
[42] Li, S.-S., et al., “An MSI Micromechanical Differential Disk-Array Filter,” Dig. of Tech.
Papers 14th Intl. Conf. on Solid-State Sensors, Actuators & Microsystems
TRANSDUCERS 2007, Lyon, France, June 10–14, 2007, pp. 307–311.
[43] Lin, Y.-W., et al., “Low Phase Noise Array-Composite Micromechanical Wine-Glass Disk
Oscillator,” Dig. of Tech. Papers IEEE Intl. Electron Devices Meeting IEDM 2005, Wash-
ington, D.C., December 5–7, 2005, pp. 281–284.
[44] Demirci, M. U., M. A. Abdelmoneum, and C. T.-C. Nguyen, “Mechanically Corner-Cou-
pled Square Microresonator Array for Reduced Series Motional Resistance,” Dig. of Tech.
Papers 12th Intl. Conf. on Solid-State Sensors & Actuators TRANSDUCERS 2003,
Boston, MA, June 8–12, 2003, pp. 955–958.
[45] Piazza, G., et al., “Low Motional Resistance Ring-Shaped Contour-Mode Aluminum
Nitride Piezoelectric Micromechanical Resonators for UHF Applications,” Proc. IEEE
Intl. Conf. MEMS 2005, Miami Beach, FL, January 30–February 3, 2005, pp. 20–23.
[46] Teva, J., “Integration of CMOS-MEMS Resonators for Radiofrequency Applications in the
VHF and UHF Bands,” Ph.D. thesis, U. A. Barcelona, 2004.
36 MEMS and NEMS Resonator Technologies

[47] Li, M., H. X. Tang, and M. L. Roukes, “Ultra-Sensitive NEMS-Based Cantilevers for Sens-
ing, Scanned Probe and Very High-Frequency Applications,” Nature Nanotechnology,
Vol. 114, 2007, pp. 114–120.
[48] Czochralski, J., Z. Phys. Chem., Vol. 92, 1918, pp. 219–221; Encyclopædia Britan-
nica, “Czochralski Method,” http://www.britannica.com/EBchecked/topic/149253/
Czochralski-method.
[49] Pfann, W. G., “Principles of Zonemelting,” T. AIME, Vol. 194, 1952, p. 747.
[50] IBM, “IBM Advances Chip Technology with Breakthrough for Making Faster, More Effi-
cient Semiconductors: Silicon-on-Insulator Technology,” 1998, http://www-03.ibm.com/
press/us/en/pressrelease/2521.wss.
[51] Silicon Materials, http://www.si-mat.com/.
[52] Polyakov, A., et al., “High-Resistivity Polysilicon as RF Substrate in Wafer-Level Packag-
ing,” Electron. Lett., Vol. 41, 2005, pp. 100–101.
[53] Organic Electronics Association, http://www.oe-a.org/.
[54] Lyshevski, S. E., MEMS and NEMS: Systems, Devices and Structures, Boca Raton, FL:
CRC Press, 2002.
[55] Engelmark, F., “AlN and High-K Thin Films for IC and Electroacoustic Applications,”
Ph.D. thesis, Uppsala University, Uppsala, Sweden, 2002.
[56] Parsons, R., “Sputter Deposition Processes,” in Thin Film Processes II, J. L. Vossen and W.
Kern, (eds.), San Francisco, CA: Academic Press Limited, 1991, p. 178.
[57] Jaeger, R. C., “Lithography,” in Introduction to Microelectronic Fabrication, 2nd ed.,
Upper Saddle River, NJ: Prentice-Hall, 2002.
[58] Dai Nippon Printing Co.-Ltd., http://www.dnp.co.jp/index_e.html.
[59] Photronics, Inc., http://www.photronics.com/.
[60] Toppan Photomasks, Inc., http://www.photomask.com/.
[61] Intel Mask Operations, http://www.intel.com.
[62] Advanced Micro Devices (AMD), http://www.amd.com/.
[63] Industrial Business Machines (IBM), “Semiconductor Solutions,” http://www-03.ibm.com/
technology/index.html.
[64] Nippon Electric Co. (NEC), “Pioneering Development of Immersion Lithography,”
http://www.nec.co.jp.
[65] Lyshevski, S. E., Micro- and Nano-Electromechanical Systems: Fundamentals of Micro-
and Nano-Engineering, Boca Raton, FL: CRC Press, 2000.
[66] Serre, C., et al., “Test Microstructures for Measurement of SiC Thin Film Mechanical Prop-
erties,” J. Micromech. Microeng., Vol. 9, 1999, pp. 190–193.
[67] Coburn, J. W., and H. F. Winters, “Ion- and Electron-Assisted Gas-Surface Chemistry: An
Important Effect in Plasma Etching,” J. Appl. Phys., Vol. 50, 1979, pp. 3189–3196.
[68] Benítez, M. A., et al., “A New Process for Releasing Micromechanical Structures in Surface
Micromachining,” J. Micromech. Microeng., Vol. 6, 1996, pp. 36–38.
[69] Calaza, C., et al., “A Surface Micromachining Process for the Development of a
Medium-Infrared Tuneable Fabry-Perot Interferometer,” Sens. Actuators A: Phys.,
Vol. 113, 2004, pp. 39–47.
[70] Haynes, C. L., and R. P. Van Duyne, “Nanosphere Lithography: A Versatile
Nanofabrication Tool for Studies of Size-Dependent Nanoparticle Optics,” J. Phys. Chem.,
Vol. B105, 2001, pp. 5599–5611.
[71] Whitesides, G. M., J. P. Mathias, and C. T. Seto, “Molecular Self-Assembly and
Nanochemistry: A Chemical Strategy for the Synthesis of Nanostructures,” Science,
Vol. 254, 1991, pp. 1312–1319.
 CHAPTER 2

Acoustic Microresonator Technologies

Acoustic microresonators are fueling the reduction in size and power consumption
of mobile radio equipment and sensing systems that the telecommunication and
sensors industries have been undertaking during the past few years. The kind of
acoustic resonators we study herein are microelectromechanical devices. They expe-
rience acoustic wave propagation and eventually vibrate at a resonance frequency
related to their dimensions and mechanical configuration, when driven with the
appropriate conditions. Roughly, they are classified in two categories: surface
acoustic wave (SAW) and bulk acoustic wave (BAW) resonators. Two types of the
latter are found: the solidly mounted resonator (SMR) and the thin-film bulk acous-
tic wave resonator (FBAR). In Chapter 1 we studied how silicon-MEMS resonators
can be excited at frequencies promoting acoustic wave propagation. Now, the dis-
cussion is restricted to those resonators implementing an acoustic layer made of
piezoelectric materials.
Despite the need for compatibility between CMOS and piezoelectric thin-film
processes, SAW, SMR, and FBAR devices can be fabricated within standard IC
technologies. Additionally, FBAR manufacturing entails micromachining steps, like
MEMS resonator processes. On the other hand, FBARs resonate at far-from-funda-
mental acoustic modes, instead of purely mechanical modes. Both circumstances
have thus created certain controversy regarding whether FBARs are considered
MEMS resonators.
In Section 2.1, we define the fundamental concepts of acoustic wave propaga-
tion and its differences with the electromagnetic-wave actuation in “classical”
MEMS. Moreover, emphasis is put on the working principle of acoustic wave reso-
nators. Then, Section 2.2 begins with the basics of piezoelectricity and acoustic
wave theory, and we study some of the most exploited acoustic vibration modes.
Section 2.3 goes into more detail about acoustic wave propagation, device design,
and applications of SAW resonators. Similar discussion is continued in Section 2.4,
which is devoted to BAW resonators.

2.1 Introduction to Acoustic Wave Resonators

By taking advantage of acoustic-wave propagation, acoustic wave resonators are

built. Many kinds of acoustic resonators exist. Musical instruments, for example,
amplify the vibration of a string or a shock by using resonant cavities or pipes, like
drums, guitars, pianos, and organs do. Thus, the instrument produces sound waves
of specific tones regarding the size of the acoustic cavity. Music instruments in
which the air vibrates inside the cavity with one opening are known as Helmholtz

37
38 Acoustic Microresonator Technologies

resonators [1]. Herein we focus on microelectromechanical acoustic resonators

and, more specifically, on piezoelectric-based acoustic resonators.

2.1.1 Acoustic Waves

An acoustic wave is a disturbance in an elastic medium that propagates in space and
time, thus transferring the energy supplied by an excitation source along the
medium in the form of oscillation or vibration. Acoustic wave propagation entails
elastic deformation of the medium along the propagation axis or in other axes as
well. In contrast to electromagnetic waves, acoustic waves do require a medium to
propagate, and their propagation speeds depend on the mechanical properties of the
wave-supporting material. Virtually any material is capable of supporting acoustic
wave propagation, including silicon, as we reviewed in Chapter 1. Nevertheless, the
piezoelectric properties of certain materials facilitate the wave propagation, thus
improving the electromechanical energy conversion, so piezoelectrics are usually
chosen as the acoustic layer of many acoustic-wave resonators. Also known as
sound speed, the acoustic-wave phase velocities are several times slower than those
of the electromagnetic wave traveling in the same medium [2].
In a first approach, there exist two types of acoustic waves: surface acoustic
waves (SAW) and bulk acoustic waves (BAW). Strictly speaking, a combination of
both is normally found in the form of longitudinal, shear, mixed longitudinal-shear
Rayleigh waves [3], Love waves [4], or Lamb waves [5], among others. The waves
we can see propagating on the surface of a lake after hitting the water mass with a
stone illustrate the concept of a SAW. On the other hand, the sound waves traveling
through the air until reaching our ears are of the BAW type. In these examples, the
water and the air are the propagation media of the SAW and BAW, respectively.
Figure 2.1 through Figure 2.4 let us understand the differences between sur-
face—Rayleigh and Love—and bulk—longitudinal and shear—acoustic waves.
As illustrated in Figure 2.1, the surface particles of an isotropic solid move in
ellipses in planes normal to the surface and parallel to the direction of the Rayleigh

Particle
Wave propagation
motion

Figure 2.1 Rayleigh wave propagation: the surface particles of an isotropic solid move in ellipses
in planes normal to the surface and parallel to the wave direction.
2.1 Introduction to Acoustic Wave Resonators 39

Particle
motion

Wave propagation

Figure 2.2 Love waves: the surface particles move in horizontal lines perpendicular to the wave
propagation.
Wave propagation

Particle
motion

Figure 2.3 Longitudinal-mode waves: the bulk particles oscillate or vibrate in the same axis of the
wave propagation.

wave propagation. The particle displacement is significant at a depth of about one

wavelength. This motion is retrograde at the surface and thin depths, and becomes
prograde at greater depths, as the size of the ellipses is smaller and its eccentricity
changes for particles deeper in the material. This behavior was predicted by Lord
Rayleigh in 1885, hence its name.
Love waves travel faster than Rayleigh waves. The particle motion of a Love
wave, depicted in Figure 2.2, forms a horizontal line perpendicular to the direction
of propagation, creating horizontally polarized shear waves (SH waves). Moving
deeper into the material, motion alternately increases and decreases as one exam-
ines deeper layers of particles. The amplitude, or maximum particle motion,
decreases rapidly with depth, and it decays with the squared root of the distance
traveled by the wave.
On the other hand, bulk acoustic waves are longitudinal, shear-mode, or com-
bination of both. Longitudinal waves travel through the medium parallel to the
40 Acoustic Microresonator Technologies

Wave propagation

Particle
motion

Figure 2.4 Shear or transverse-mode waves: the bulk particles oscillate in the plane perpendicular
to the wave propagation and energy transfer.

same axis of the oscillations or vibrations of the particles in the medium; that is, in
the same or opposite direction as the motion of the wave as shown in Figure 2.3.
Longitudinal mode waves are confined in a resonant cavity, thus displaying a par-
ticular standing-wave pattern. The longitudinal modes are reinforced by construc-
tive interference after many reflections from the cavity’s reflecting surfaces for
wavelengths corresponding to entire fractions of twice the length of the cavity. All
other wavelengths experience destructive interference and are suppressed.
While longitudinal modes have a pattern with their nodes located axially along
the length of the cavity, transverse modes, with nodes located perpendicular to the
axis of the cavity, may also exist. A transverse or shear-mode wave propagates and
transfers its energy in the direction perpendicular to the oscillations occurring in the
medium. If the shear wave moves in the positive x-direction, for example, particles
in the medium oscillate in the y-z plane, as represented in Figure 2.4. Shear-mode
resonance occurs at longer wavelengths than longitudinal-mode vibrations.
Another type of complex quasi-surface wave is the Lamb waves propagating in
solid plates. In this case, particle motion lies in the plane defined by the plate normal
and the direction of wave propagation. The mathematical description of Lamb
waves is quite complex, and due to its complexity Lamb waves have not been sys-
tematically explored in experimental implementations until recently. Similar but
still different than Rayleigh waves, they are also often called Rayleigh-Lamb waves.

2.1.2 Acoustic Microresonators

Acoustic microresonators are kind of microelectromechanical devices experiencing
acoustic wave propagation and eventually vibrating at a resonance frequency
related to their dimensions and mechanical properties. In some sense, the resonator
2.1 Introduction to Acoustic Wave Resonators 41

behaves as an acoustic cavity trapping the wave in the medium. To do that, trans-
mission and reflection of the wave are promoted by the appropriate means, such as
electrodes and acoustic layer functionally designed of the type and frequency of
the acoustic wave. Thus, the amplitude of the wave achieves its maximum when
the transmitted and reflected waves have λ, λ/2, or λ/4 phase shifting, according
to the separation of the electrodes, dimensions of the acoustic layer, and acoustic
mode.
To illustrate, let’s consider the case of longitudinal waves of wavelength λ prop-
agating along the bulk of a λ/2-long resonant cavity. Because of the in-phase align-
ment of the transmitted and reflected waves, the constructive interference between
them reinforces the energy inside the cavity. Otherwise, the incident and reflected
waves are out of phase, and they are suppressed after destructive interference. This
happens not only for the fundamental wavelength λ, but also for the shorter waves
of wavelength λ/n equal to an entire fraction n of the fundamental wavelength λ.
The sequence of Figure 2.5 depicts the propagation of the first five longitudinal
modes through the λ/2-long resonator.
Silicon and other materials have been used to manufacture acoustic resonators.
Nevertheless, the high-frequency requirements of modern systems, the electronic
technology evolution imposing the need for miniaturization, and the development
of thin-film piezoelectric technologies paved the way to the new generation of
thin-film acoustic resonators. In this way, new fabrication processes and materials
were developed and thin-film aluminum nitride (AlN) and zinc oxide (ZnO) became
the standard for the new kind of SAW and BAW miniature resonators. In the next
sections, we concentrate the discussion on these devices and the physics of acoustic
wave propagation based on the piezoelectricity theory.

(a) (b)

(c) (d)

Figure 2.5 (a–d) Propagation of longitudinal-mode waves inside λ/2 resonators: constructive
interference between the incident and reflected waves occurs for waves of length λ/n, where n is
an entire number.
42 Acoustic Microresonator Technologies

2.2 Fundamentals of Piezoelectricity and Acoustic Wave Propagation

Piezoelectricity is the property of some crystalline materials to deform after electric

field excitation or, alternatively, to undergo electrical displacement when an exter-
nal strain is applied to the crystallographic structure. We refer to these properties as
the piezoelectric and inverse piezoelectric effects. They promote the medium defor-
mation due to electric potentials or currents, thus easing the wave propagation and
its implementation in electronic circuits integrating sensors and/or radio frequency
components. Next, the basics of piezoelectricity theory and acoustic wave modes in
piezoelectric resonators are examined.

2.2.1 Theory of Piezoelectricity

Piezoelectric materials in commercial application like quartz, AlN, or ZnO are crys-
talline solids arranged in a polycrystalline structure presenting symmetries on cer-
tain axes. At the macroscopic level, they can exhibit orientation in a determined
direction due to a poling process. Since piezoelectric ceramics are anisotropic, their
physical constants—elasticity, permittivity, and so on—are tensor quantities. For
this reason, the study of piezoelectricity theory starts with previous definition of a
reference system of coordinates.
The constants are generally expressed with two subscript indices. The first
index refers to the axis of the excitation and the second one to that of the actuation.
These variables are in the mechanical (stress or strain) and electrical (electric dis-
placement or electric field) domains. We can define the crystallographic axes by
using the notation of a Cartesian rectangular system, where we usually choose the
direction of positive polarization to coincide with the Z-axis of a Cartesian rectan-
gular system, as Figure 2.6 depicts. Directions of X, Y, and Z are represented by 1,
2, and 3, respectively, whereas the shear about these axes is done by 4, 5, and 6,
respectively. In crystallography, the natural coordinate system is provided by the
dimensions of the unitary crystal cell, denominated by the letters a, b, and c. Thus,
we choose X, Y, and Z to coincide with the natural axes of the crystal a, b, and c,
respectively.
In longitudinal-mode resonators, the electric field is applied in the direction of
the Z-axis (“3”), thus coinciding with the preferred crystal orientation (c-axis), and,
due to the electric field, the crystal will experience strain/stress in the Z-axis (“3”),
too.
As stated before, the sound speed is some orders of magnitudes less than the
electromagnetic-wave propagation speed. This is the case of piezoelectric materials,

Z 3

Poling
axis Y 2
(c-axis)
5
4
X 1

Figure 2.6 Axes convention and directions of deformation.

2.2 Fundamentals of Piezoelectricity and Acoustic Wave Propagation 43

where the acoustic wave is about five times slower than electromagnetic waves. For
this reason, the quasi-electrostatic approximation is enough to describe the wave
propagation in piezoelectric materials, as the magnetic effects are neglected in the
analysis. Thus, the description of piezoelectricity couples the equations of linear
elasticity with the charge equation of electrostatics through the piezoelectric con-
stants of the crystal. Due to the domain coupling, the electric variables are only
quasi-static, as previously commented, and the formulation of the mutual relation-
ship between the quasi-static electric field and the applied mechanical stress is:

T6 ×1 = c 6E× 6 ⋅ S 6 ×1 − e 6 × 3 ⋅ E 3 ×1
(2.1)
D3 ×1 = e 3 × 6 ⋅ S 6 ×1 − ε S3 × 3 ⋅ E 3 ×1

This is known as the stress-charge form of the piezoelectric equations, where T

is the stress matrix, S is the strain matrix describing the deformation of the crystal, c
is the stiffness matrix, e is the piezoelectric constant matrix, E is the electric field
applied to the resonator, D is the electric density displacement matrix, and ε is the
permittivity matrix of the piezoelectric material [6]. The super-indices in the c and ε
matrices point out that they are evaluated at constant electric field and strain,
respectively.
Alternate forms of the constitutive equations describe in an equivalent fashion
the electromechanical coupling. A certain equation of the group is employed
depending on the boundary conditions that more easily simplify the description of
the system. The strain-charge form is:

S 6 ×1 = s 6E× 6 ⋅ T6 ×1 + d 6 × 3 ⋅ E 3 ×1
(2.2)
D3 ×1 = d 3 × 6 ⋅ T6 ×1 + ε 3T× 3 ⋅ E 3 ×1

The strain-voltage form is:

S 6 ×1 = s 6D× 6 ⋅ T6 ×1 + g 6 × 3 ⋅ D 3 ×1
(2.3)
E 3 ×1 = − g 3 × 6 ⋅ T6 ×1 + β 3T× 3 ⋅ D3 ×1

The stress-voltage form is:

T6 ×1 = c 6D× 6 ⋅ S 6 ×1 − h6 × 3 ⋅ D3 ×1
(2.4)
E 3 ×1 = − h3 × 6 ⋅ S 6 ×1 − β 3S × 3 ⋅ D3 ×1

where s is the compliance matrix (reciprocal of the stiffness matrix), β is the inverse
matrix of permittivity, and d, g, and h are the alternate forms of piezoelectric con-
stants. Although (2.1) to (2.4) are exact, some of the right-hand variables simplify
to zero upon certain boundary conditions, greatly simplifying the analysis and
implementation.
One can pass from one description to another one by rewriting the stiffness,
compliance, and piezoelectric constant matrices. This latter, for example, can be
restated to their corresponding alternate form by the relations:
44 Acoustic Microresonator Technologies

e = dc E d = εT g
(2.5)
g= β dT
h = gc D

In the next section, some vibration modes are analyzed in the light of these
equations. There we will see how the specific boundary conditions of the system
make some variables vanish, thus simplifying the formulation of electromechanical
coupling.

2.2.2 Excitation and Vibration Mode Description

In Section 2.1.1, acoustic wave propagation was introduced. Now, a more rigorous
analysis of different excitation and vibration modes of acoustic resonators can be
carried out, by using the constitutive equations studied in the previous section. Fol-
lowing, we revise some of the most representative vibration modes, their boundary
conditions, and simplified equations resulting from the analysis of a specific
excitation mode.
Acoustic resonances are high-order modes often denoted with the nomenclature
of the excitation-actuation axes, instead of designating them in the mode-sequence
manner of fundamental and first modes of electromechanical resonators. For exam-
ple, excitation in the vertical axis (3), leading to strain in the same axis, is referred to
as the “33” mode of the resonator (one could erroneously think that this is the
33rd-order mode). The crystal poling also gives a reference for defining the excita-
tion axis. The schemas of Figure 2.7 illustrate the excitation-wave propagation
notations of four configurations:

(a) (b)

(c) (d)

Figure 2.7 Vibration modes of acoustic resonators: (a) longitudinal (“33” mode); (b) extensional
(“31” mode); (c) thickness-transversal (“31” shear mode); and (d) lateral-shear (“15” mode).
2.2 Fundamentals of Piezoelectricity and Acoustic Wave Propagation 45

• Thickness excitation of the thickness vibration (“33” longitudinal mode);

• Thickness excitation of the extensional vibration (“31” flexural mode);
• Thickness excitation of the transversal vibration (“31” shear mode);
• Lateral excitation of the shear-mode vibration (“15” shear mode).

The poling axis, P in the figures, determines the wave propagation. The refer-
ence axes are considered according to this, as in the 15-shear-mode vibration of Fig-
ure 2.7(d), where the poling axis is along the one-axis orientation (despite the
electric field is applied along the three-axis).
The boundary conditions of these resonator configurations simplify the consti-
tutive equations of (2.1) to (2.4), easing the implementation of models and calcula-
tions. Considering the longitudinal-mode resonator of Figure 2.7(a), the electric
field E3 and mechanical stress T3 applied in the three-axis generate electrical dis-
placement D3 and mechanical strain S3. With these conditions, using the
strain-charge form of (2.2) is very convenient:

⎡ S 1 ⎤ ⎡ s11E
⋅ ⋅ ⋅ ⋅ ⋅ ⎤ ⎡ 0 ⎤ ⎡ d 11 ⋅ ⋅ ⎤
⎢ ⎥ ⎢ ⎥ ⎢
⎢S 2 ⎥ ⎢ ⋅
E
s 22 ⋅ ⋅ ⋅ ⋅ ⎥ 0⎢ ⎥ d d 22 ⋅ ⎥
⎢ ⎥ ⎢ 21 ⎥ ⎡0⎤
⎢S 3 ⎥ ⎢ ⋅ ⋅ E
s 33 ⋅ ⋅ ⋅ ⎥ ⎢T3 ⎥ ⎢d 31 d 32 d 33 ⎥ ⎢ ⎥
⎢ ⎥=⎢ ⎥⋅⎢ ⎥ + ⎢ ⋅ 0
⎢S4 ⎥ ⎢ ⋅ ⋅ ⋅ E
s 44 ⋅ ⋅ ⎥ ⎢0⎥ ⎢ ⋅ ⋅ ⋅ ⎥⎥ ⎢ ⎥
⎢E ⎥
⎢S ⎥ ⎢ ⋅ ⋅ ⋅ ⋅ E
s55 ⋅ ⎥ ⎢0⎥ ⎢ ⋅ ⋅ ⋅ ⎥ ⎣ 3⎦
⎢ 5⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢⎣ S 6 ⎥⎦ ⎣ ⋅ ⋅ ⋅ ⋅ ⋅ s 66 ⎦ ⎣ 0 ⎦ ⎣d 61
E
⋅ ⋅ ⎦
(2.6)
⎡ d 11 ⋅ ⋅ ⎤ ⎡0⎤
⎢d d 22 ⋅ ⎥ ⎢0⎥
⎡D1 ⎤ ⎢ ⎥ ⎢ ⎥ ⎡ε11 ⋅ ⎤ ⎡0⎤
21 T
⋅
⎢D ⎥ = ⎢d 31 d 32 d 33 ⎥ ⋅ ⎢T3 ⎥ + ⎢ ⋅ ε T ⎥
⋅ ⎥⋅⎢ 0 ⎥
⎢ 2⎥ ⎢ ⋅ ⋅ ⋅ ⎥ ⎢ ⎥ ⎢
0
22
⎢ ⎥
⎢⎣D3 ⎥⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⋅
⎣ ⋅ ε T33 ⎥⎦ ⎢⎣E 3 ⎥⎦
⎢ ⋅ ⋅ ⋅ ⎥ ⎢0⎥
⎢ ⎥ ⎢ ⎥
⎣d 61 ⋅ ⋅ ⎦ ⎣0⎦

In a more compact representation, (2.6) simplifies to:

S 3 = s 33
E
⋅ T3 + d 33 ⋅ E 3
(2.7)
D3 = d 33 ⋅ T3 + ε 33
T
⋅ E3

This shows that we need to know only a reduced set of elastic, dielectric,
and piezoelectric constants if some restrictions are imposed to the system. Analog
procedure is carried out to analyze and simplify the description of the
thickness-transversal “31” mode of the plate in Figure 2.7(b). The electric field E3
and mechanical stress T1 applied in the three-axis and one-axis directions, respec-
tively, generate electrical displacement D3 and mechanical strain S1, in the
three-axis and one-axis directions, respectively, thus simplifying (2.2) to:

S 1 = s11
E
⋅ T1 + d 31 ⋅ E 3
(2.8)
D3 = d 31 ⋅ T1 + ε T33 ⋅ E 3
46 Acoustic Microresonator Technologies

Transversal and lateral modes are promoted in SAW resonators, and longitudi-
nal and shear modes are usually excited in BAW resonators. Of course, due to pro-
cess and design issues, spurious generation of nondesired modes can also occur.
However, assuming that the preferred crystal orientations and excited modes are
the expected, the piezoelectric, elastic, and dielectric constants of (2.1) to (2.8) can
be used to characterize the electromechanical performance of the piezoelectric, as
we will study in Chapter 5. Actually, a SAW resonator can be used as a test structure
in order to extract the material constants of the piezoelectric layer employed in a
BAW resonator. In the next section, the principles, design, and applications of SAW
resonators are revisited.

2.3 Surface Acoustic Wave (SAW) Resonators

SAW resonators exploit SAW propagation and electromechanical transduction to

implement electronic circuits as filters, oscillators, and sensors. The transduction is
performed by two or more metallic electrodes and a piezoelectric layer, and, typi-
cally, wave propagation SAW resonators of the Rayleigh or leaky-SAW types [7]. A
SAW resonator is basically a resonant cavity in which a first transducer electrode
converts the electric signal into a lateral mechanical wave. The resulting SAW prop-
agates on the piezoelectric to reach the second electrode, where it is transduced back
into the electrical domain. When arriving at the second electrode, and typically
aided by one or more reflector electrodes, the acoustic wave bounces back in the
direction of the first electrode, and the electromechanical conversion is repeated
indefinitely, as depicted in Figure 2.8. Thus, the acoustic wave is trapped in the
cavity formed by the resonator electrodes.
The electrodes are one or more interdigitated transducers (IDTs) being fabri-
cated by state-of-the-art photolithography techniques. Many piezoelectric crystals
and ceramics have been implemented in the fabrication of SAW resonators, like
quartz, lithium niobate, lithium tantalate, lanthanum gallium silicate
(langasite-LGS), lead zirconate titanate (PZT), and, more recently, AlN and ZnO.
Because of its low temperature drifting, quartz has been preferred to other materials
and, to date, it dominates the high-accuracy SAW-resonator implementations.
The available photolithography resolution limiting the dimensions of the IDTs
and the piezoelectric layer determine the maximum operating frequency of the fun-
damental mode of the resonator. However, as well as other electromechanical reso-
nators, SAW devices can be operated at over-tone modes to bypass
near-to-fundamental bulk or other resonance modes. Typical frequencies for
SAW-resonator-based applications are in the UHF band below 1 GHz, although
high-performance commercial devices in the 1.5-GHz GPS and 1.9-GHz PCS bands
are available so far [8, 9]. The fundamental frequency of the SAW resonator mainly
depends on the pitch of the IDTs, which is chosen to be equal to the SAW wave-
length λ, and the sound speed of the piezoelectric layer v:

v
f0 = (2.9)
λ
2.3 Surface Acoustic Wave (SAW) Resonators 47

Figure 2.8 Surface acoustic wave (SAW) resonator.

Other aspects influence the final performance of the resonator, like the cavity
and reflector grating design. To date, many companies fabricate SAW resonators,
some of the major manufacturers being Epson Toyocom Corporation [10], Epcos
AG [11], Murata Corporation [12], Fujitsu Corp. [13], Hitachi Corp. [14],
Samsung [15], and NEC Corp. [16]. Commercial resonators exhibit low insertion
losses, which make them ideal for high-performance radio frequency applications.
Besides, they are relatively simple to fabricate because no micromachining tech-
niques are required, although standard photolithography techniques have imposed
a 3-GHz limit on SAW device fabrication (which will be overcome with lead-
ing-edge photolithography at sure). Through the years, design and fabrication of
SAW resonators have been improved to achieve better application performance and
device efficiency. In the following, we discuss the basic topics of SAW resonator
design.

2.3.1 One-Port and Two-Port Configurations

SAW resonators are found in two types of port configurations: one-port and
two-port resonators. One-port resonators are two-terminal devices, and they find
48 Acoustic Microresonator Technologies

application in oscillator circuits like VCOs or Colpitts oscillators. Two-port resona-

tors behave more like narrow bandpass filters. One-port SAW resonators have a
single IDT generating and receiving the SAW, and two grating reflectors, which
reflect the SAW and generate a standing wave between the two reflectors. The IDT
and reflectors are fabricated on quartz crystal substrate or another piezoelectric
material and patterned by photolithographic processes [17]. The IDT, input and
output ports, and reflectors of a one-port resonator are depicted in Figure 2.9. The
SAW escapes out from the IDT fingers and is picked back to them through the
reflectors.
On the other hand, two-port SAW resonators exhibit two IDTs, one of them
generating the SAW, and the second one picking it. As with one-port devices, two
grating reflectors aid the SAW to be reflected and confined between the IDTs. The
generic connections to the first and second IDTs, their layout, and their placement
between the reflectors are drawn in Figure 2.10.
One-port SAW resonators can be of the synchronous or the nonsynchronous
type, while many configurations of the two-port SAW resonator exist. Examples of
the two-port type are the synchronous single pole, the single-phase unidirectional
transducer (SPUDT), the proximity-coupled resonator, and the with-
drawal-weighted-reflector resonator. The different configurations attempt to bal-
ance design, cost and manufacturability, and performance issues [18]. To illustrate,
Figures 2.11(a) and 2.11(b) depict one-port resonators of both synchronous and
nonsynchronous types, respectively, and a two-port proximity-coupled resonator is
drawn in Figure 2.11(c).

Reflector IDT Reflector

(Piezoelectric substrate)
Figure 2.9 One-port SAW resonator.

Reflector IDT (in) IDT (out) Reflector

(Piezoelectric substrate)
Figure 2.10 Two-port SAW resonator.
2.3 Surface Acoustic Wave (SAW) Resonators 49

(a)

(b)

(c)
Figure 2.11 One-port and two-port configurations: (a) synchronous (one-port);
(b) nonsynchronous (one-port); and (c) proximity-coupled (two-port). (Figure based on designs of
P. V. Wright.)
50 Acoustic Microresonator Technologies

2.3.2 SAW Resonator Design

The design parameters of SAW resonators are, essentially: (1) the IDT and grating
periods; and (2) the width of the cavity and the total beam width of the IDTs and
grating. The IDT period λIDT controls the resonance frequency, while the reflector
grating period λR determines the total reflectivity (and thus the energy concentration
in the center of the IDT region). The resonant cavity width lC also plays a role in the
frequency response. We can consider the next cases:

1. The one-port synchronous, with equal λIDT and λR values, and lC equal to

(2n + 1) λ
lC = (2.10)
4

where n is an integer, and λ equals λIDT. The advantages of this configuration

are that, due to the periodicity of the structure (λIDT = λR), the resonance
frequency does not depend on the reflectivity of the electrodes and the
manufacturability yield is high. However, this is achieved at the cost of
higher insertion losses or, equivalently, having increased die area: the
number of fingers in both the IDT and grating has to be increased to make
the overall beam wider, thus diminishing the insertion losses.
2. The one-port nonsynchronous, with different λIDT and λR values. In addition,
the cavity width is no longer that of (2.10). In this case, the resonance
frequency depends on the reflectivity of the electrodes and the conductance
can be enhanced, which allows controlling the insertion losses at the cost of
manufacturability.
3. Two-port designs, in which we control the frequency stability and introduce
multiple poles in the frequency response through the cavity design, the sym-
metry between the first and second IDTs, the grating finger design, and cou-
pled-resonator designs. The transducer electrodes are apodized in both
one-port and two-port SAW resonators in order to suppress transverse
modes. Thus, the apodizing strategy is also a design parameter of the fre-
quency response. For a complete and rigorous guide on SAW resonator, the
designer should consult the work of P. V. Wright [18–21].

So far, we have considered that the resonator operates at the SAW fundamental
frequency predicted by (2.9). Nevertheless, they can also work at selected harmonic
frequencies, depending on the IDT metallization ratio η = a/b. Thus, the SAW oper-
ates at only odd or odd and even harmonic frequencies [22]. Figure 2.12 shows some
examples of modified IDTs for fundamental and harmonic frequency operation.
The role of novel quartz-cutting angles and the new materials employed in SAW
device fabrication has been increasing the performance and resonance frequency of
SAW resonators as well. For example, Epson-Toyocom developed a 2.5-GHz reso-
nator with 6-dB insertion losses and high-frequency stability, by using a novel
ST-cut quartz technique [23].
2.4 Bulk Acoustic Wave Resonators 51

(a)

(b)

(c)
Figure 2.12 SAW IDT structures for fundamental and harmonic frequency operation: (a) solid electrode
(odd-harmonic); (b) split electrode (odd-harmonic); and (c) three-electrode IDT (odd-even harmonic).

2.3.3 SAW Applications

SAW resonators have been a commercial success on radio frequency applications,
especially for filter and oscillator implementations. Their impact has made possible
considerable reductions in size and power of the chipsets of mobile devices [24, 25].
More modest, but also important, has been the impact of SAW resonators in mass
detector and pressure sensor devices, with application in bioparticle detection [26,
27]. A detailed discussion on both RF and sensor applications is carried out in
Chapters 8 and 9, respectively.

2.4 Bulk Acoustic Wave Resonators

Bulk acoustic wave (BAW) resonators are boosting the reduction in size and power
consumption of mobile radio equipment experienced by the telecommunication
industry during the past few years. Due to their commercial success, the motivation
52 Acoustic Microresonator Technologies

for developing new, integrated applications based on BAW devices has caught the
attention of many industry and academic research groups all over the world. Nowa-
days, this interest comes not only from the telecom industry, but also from sens-
ing-application companies. Keeping both features in mind—CMOS integration and
new applications—the implication is that BAW fabrication processes should be suc-
cessful in producing high-quality devices and, at the same time, integrating them
with standard CMOS technologies. This section focuses on introducing the key con-
cepts and state-of-the-art technologies related to BAWs, ideas that will be further
developed later in the book.

2.4.1 Acoustic Wave Propagation in BAW Devices

BAW devices experience acoustic wave propagation through the bulk of its active
layer structure, hence their name (Figure 2.13). This feature differentiates BAW and
surface acoustic wave (SAW) devices, the acoustic waves propagating in longitudi-
nal or shear-transversal modes. In both cases, the acoustic wave causes deformation
of the active layer, which is typically a piezoelectric material made with a thin-film
technology. As in SAW resonators, piezoelectric and the inverse piezoelectric effects
are the actuation and detection mechanisms involved in BAW device operation.
According to these principles, a voltage applied to the resonator’s electrodes induces
strain of the acoustic layer and vice versa, after a mechanical strain of the acoustic
layer a voltage can be read out the electrodes.
Another difference between BAW and SAW devices is their physical layout. In
BAW, the acoustic layer is a component element of a stacked structure in which the
acoustic wave is confined. A couple of metal layers acting as electrodes complete the
structure of BAW resonators. Therefore, the BAW device is fabricated on top of a
carrying substrate, typically silicon, and the acoustic layer and electrodes are
located on top of said substrate. In SAW, the acoustic layer may be the carrying sub-
strate by itself, and the IDT electrodes are located on top of it, both in the same
plane. Typical operation frequencies for SAW range from 30 MHz to 1 GHz, while
the central frequency of BAW can be found in the 1- to 10-GHz band (typical:
2 GHz). For both technologies, however, the value of the central frequency or the
frequency-band range might be extended depending on the operating resonant
mode—fundamental or overtone—and the available fabrication technologies [28].

Air
Top electrode

Piezoelectric λ
t= V = V0 cos 2πf 0 t
2
Bottom electrode

Air

v
λ=
f0

Figure 2.13 Acoustic-wave propagation in BAW resonators.

2.4 Bulk Acoustic Wave Resonators 53

The resonance frequency of a BAW resonator operating in fundamental, longi-

tudinal mode is mainly determined by the thickness t of the acoustic layer from [29]:

2 πf 0 t
θ= (2.11)
v

In (2.11), θ, v, and f are the phase, sound velocity, and frequency of the acoustic
wave propagating through the bulk of the acoustic layer, and t is the thickness of the
thin film. At resonance (f = f0), the acoustic phase in the film is θ = π. Under these
conditions, and solving f in (2.11), it leads to:

v
f0 = (2.12)
2t

This result is equivalent to saying that the thickness of the thin film is equal to
half the wavelength of the acoustic wave for the first longitudinal resonance mode.
However, in an electrode-piezoelectric-electrode resonator, the electrodes’ contri-
bution to this equation must also be accounted for, since their added thickness
reduces the resonance frequency [30]. Looking at Figure 2.13, t must be half a wave-
length of the acoustic wave, in order to confine the energy between the electrodes.
At resonance, this energy is magnified by the quality factor Q of the device. The
mechanism is illustrated in Figure 2.14(a). Due to the piezoelectric effect, an electric

(a)

(b)

Figure 2.14 (a) Electric charge displacement and poling in a BAW resonator due to an electric
potential applied to its electrodes, and (b) deformation of the crystal structure after electric fields
of opposite magnitudes induced in the c-axis. (© 2001 Agilent Technologies [31].)
54 Acoustic Microresonator Technologies

potential V applied to the electrodes of the resonator induces an electric field E and
an electric density displacement D in the poling axis of the crystallographic struc-
ture, also referred to as the c-axis in Figure 2.14(b). Also observed in the figure, the
crystal suffers from mechanical deformation in the axis of the electric field. Never-
theless, strain in other directions may occur, depending on the crystal orientation
and resonance mode. Extensional and shear-mode resonances are two examples of
different operating modes of the device.
The constitutive equation (2.7) previously described explains this mechanism
for longitudinal-mode BAW resonators. As it will have been noticed, design param-
eters and physical behavior of BAW devices are completely different from those of
SAW devices, in which the resonance frequency is a layout-design function of the
interdigitated transducers. At the manufacturing level, these differences translate
into more complex processes for the BAW device case, because acoustical isolation
between the resonator and the substrate should be provided by some means.
Among BAW, we find two kinds of devices, namely, thin-film bulk acoustic
wave resonators (FBAR) and solidly mounted resonators (SMR) [32]. The opera-
tion and physical principles of FBAR and SMR are the same, the only difference
being the fabrication technology providing the acoustical isolation mentioned
earlier.
Both kinds of BAW are a metal-piezoelectric-metal stack of materials. How-
ever, the FBAR exhibits a micromachined air gap to reduce the electromechanical
coupling to the carrying substrate, while the SMR device implements an array of
reflecting materials, known as a reflecting mirror or Bragg’s reflector [33] (Figure
2.15). Careful selection of the mirror materials and configuration guarantees full
impedance mismatching and improved isolation between the SMR and the sub-
strate [28]. In both FBAR and SMR, the purpose of the acoustical isolation is to
obtain a high-quality factor resonator.

2.4.2 Thin-Film Bulk Acoustic Wave Resonators (FBAR)

The first FBAR device was disclosed by Lakin and Wang in 1981 [29].
Micromachining of the FBAR takes advantage of well-established technologies and
processes largely implemented in microelectromechanical systems (MEMS) and
CMOS processes.
Concerning technological aspects, fabrication technologies and processes are
the most described matter in FBAR devices. Key points in describing the FBAR tech-
nology are the piezoelectric-layer deposition process, the FBAR-stack configura-
tion, and the micromachining implementation to obtain a high-quality factor
device.
The piezoelectric layer composing modern FBAR devices is usually made of alu-
minum nitride (AlN), zinc oxide (ZnO), or lead zirconate titanate (PZT). Other
materials like magnesium zinc oxide (MgxZn1-xO) or lanthanum gallium silicate
(langasite - LGS) are currently being explored as well [34, 35]. Its electromechanical
and piezoelectric properties, and the possibility of CMOS compatibility, have made
AlN the preferred material for FBAR implementation. According to Table 2.1 and
(2.12), the high acoustic wave velocity of AlN allows handy fabrication of a typical
device to resonate in the gigahertz range, for a thin-film thickness in the order of few
2.4 Bulk Acoustic Wave Resonators 55

Electrode

Piezoelectric
Electrode

Air gap

Substrate (Si)

(a)

Electrode

Piezoelectric
Electrode

m1
m2
m1
m2
m1

(b)
Figure 2.15 Stacked structure of BAW devices with acoustical isolation provided by: (a) a
micromachined air gap (FBAR), and (b) a Bragg reflector (SMR).

microns or less. This feature makes FBARs known as “thin-film” devices. Table 2.1
compares the main properties of AlN and some piezoelectric materials.
The deposition technique of the piezoelectric layer is another important aspect
influencing the performance of FBARs. Common fabrication techniques for achiev-

Table 2.1 Electromechanical Properties of Some Piezoelectric Materials

Acoustic Acoustic Acoustic
Density Dielectric Velocity Impedance Acoustic Loss dB/ s
3 6 2
Material (kg/m ) Constant (m/s) (10 kg/m s) Coupling at 1 GHz
AlN 3,270 8.5 10,400 34.0 0.17 ~5
ZnO 5,680 8.8 6,330 36.0 0.28 8.3
CdS 4,820 9.5 4,465 21.5 0.15 >50
LiNbO3 4,640 29 7,320 30.6 0.17 0.5–0.9
LiTaO3 7,450 43 6,160 46.4 0.19 0.8
56 Acoustic Microresonator Technologies

ing a high-quality piezoelectric layer are radio-frequency sputtering [36–38],

epitaxial growth [39], DC sputtering [40], or combination of the same, among oth-
ers. Additional postprocessing techniques such as annealing of the piezoelectric-
deposited wafer may be employed to improve the crystal’s quality [41]. Several
parameters condition the quality of the piezoelectric layer, some of the most rele-
vant being the deposition technique and the temperature of the substrate. Some
deposition techniques like sputtering are compatible with CMOS processes,
whereas others like epitaxial growth are unsuitable for CMOS integration due to
their processing at high temperatures over 400ºC [39].
Another critical topic influencing the quality of the piezoelectric layer is the sub-
strate on which it is deposited (i.e., the FBAR’s bottom electrode). It is well known
that some metals lead to better c-axis-oriented crystal structures than others, mainly
due to crystallographic compatibility between them and the AlN layer [42, 43].
Molybdenum (Mo), tungsten (W), aluminum (Al), and platinum (Pt) are some of
the materials commonly employed in AlN-based FBAR fabrication.
Taking the previously mentioned process parameters into consideration, the
crystallographic quality of the AlN is evaluated through atomic force microscopy
(AFM) and x-ray diffraction (XRD) characterization, among other techniques.
AFM aids in measuring the granularity of the crystal, whereas XRD provides infor-
mation on the preferred orientation of the crystal. Additional techniques like scan-
ning electron microscopy (SEM) may also help in the visual inspection of the
polycrystalline structure of the piezoelectric. The AFM and SEM images of Figure
2.16(a) to Figure 2.16(c) show surface and cross-sectional views of AlN crystals
deposited on different substrates. It can be noted how different the AlN’s grain qual-
ity and the crystal orientation are when deposited under different conditions on
Au/Cr [Figure 2.16(b)] and Al [Figure 2.16(c)] substrates [44].
As previously set out, micromachining is needed to release the FBAR, thus pro-
viding acoustical isolation between the device and the substrate in order to achieve
high-quality factors. The result of the process is a structure with an air gap, a cavity,
or a membrane underneath the resonator’s structure. Front-side or back-side sur-
face and bulk micromachining are the technological options for device releasing.
According to the surface micromachining technique, a sacrificial layer placed in
between the substrate and the resonator is attacked by wet etching. Processes imple-
menting germanium [45] and silicon oxide [46] as sacrificial layers for surface
micromachining, among others, have been described. On the other hand, bulk
micromachining can be implemented through wet or dry etching, either from the
front or from the back of the wafer. Different implementations of wet etching of sili-
con in KOH solutions are presented in [44, 47]. Dry etching of the silicon substrate
may also be carried out by using RIE [48] or DRIE [49]. In Figure 2.17, the profiles
of FBAR processes using some of these micromachining techniques are illustrated.
Less described is the influence of the electrodes’ layout in the FBAR’s perfor-
mance. Some electrode geometries have been demonstrated to suppress or diminish
spurious resonance modes better than others [50]. At the same time, the size of the
electrodes affects the electrical performance of FBARs [51, 52].
Tuning of FBARs is also of special interest. Due to process deviations or varia-
tion of the physical properties of the materials employed in the fabrication, the reso-
nance frequency may be slightly different from that corresponding to the process
2.4 Bulk Acoustic Wave Resonators 57

(a)

(b)

(c)
Figure 2.16 Evaluation of the thin-film quality in AlN: SEM image of AlN deposited on Au/Cr
substrate, (a) top and (b) cross-sectional views; and (c) cross-sectional SEM of AlN on Al electrode
showing amorphous layer and random crystal orientation. (© 2005 Elsevier [44].)

design. Different approaches, such as electrostatic tuning [53] or mass-loading [54,

55], for example, have been conceived in order to change the resonance frequency
of FBARs. Related to this topic, the temperature coefficient factor (TCF) of FBARs
is also an important factor conditioning the performance of the device, especially
for RF applications. Temperature-compensation strategies based on the deposition
of an opposite-TCF thin-film material on the FBAR have been studied [56, 57].

2.4.3 Solidly Mounted Resonators (SMR)

The device structure and working principle of SMRs are basically the same as those
of FBARs. Instead of the air gap in FBARs, SMR devices implement a stack of
λ/4-thick acoustically mismatched layers, which are placed underneath the resona-
tor to provide acoustic isolation from the substrate, thus ameliorating the Q factor.
The λ/4 thicknesses li are the quarter wavelength λi of the BAW at the ith layer evalu-
ated at the resonance frequency f0:

λi
li = (2.13)
4
58 Acoustic Microresonator Technologies

Electrode

Electrode Piezoelectric

Air gap

Substrate (Si)

(a)

Electrode

Piezoelectric

Electrode

Air gap

Substrate (Si)

(b)

Electrode

Piezoelectric

Electrode

Air gap

(c)
Figure 2.17 Different micromachining processes for releasing of FBAR: (a) surface, (b) front-side,
and (c) back-side bulk micromachining.
2.4 Bulk Acoustic Wave Resonators 59

vi
li = (2.14)
4f 0

The large acoustic-impedance mismatching between the alternating λ/4-thick

materials causes the BAW to be systematically reflected to the resonator, thus con-
fining the energy in the acoustic layer of the device. For this reason, the stack is
called a reflecting mirror, or Bragg reflector, after W. L. Bragg. A significant per-
centage of the BAW is effectively reflected with this method, and the energy of the
escaped acoustic wave diminishes with depth along the traveled path down to the
substrate. Figure 2.18 illustrates the operation of SMRs.
Typically, the impedance mismatching is achieved by alternating dielectric and
conductor materials, with low and high acoustic impedances, respectively. Silicon
oxide (SiO2) and tungsten (W), for example, are two materials commonly employed
in SMR implementations. SMRs exhibit good robustness, low stresses, and practi-
cally no risk of mechanical damage in dicing and assembly [58]. On the other hand,
FBARs exhibit higher Q factors and thermal isolation due to the air gap. However,
the Q factor of SMRs can be tailored to achieve design specifications as a function
of the number of reflector periods N [28]:

π s ⎡ phd ⎤
( )
2N
Q= pd phl ⎢1 + ⎥ (2.15)
4 ⎣ phl − 1⎦

where p ds is the ratio of acoustic impedances of substrate (s) and piezoelectric (d)
layer, p hl is the ratio of acoustic impedances of the low-impedance (l) and
high-impedance (h) reflector layers, and p hd is the ratio of piezoelectric and
high-impedance reflector layer. The direct relation between N and Q in (2.15) let us
calculate the loaded Q factor (due to acoustic coupling).

Electrode

Piezo Acoustic wave

(λ/2) profile
Electrode

Low-imp. (λ/4)

High-imp. (λ/4)

Low-imp. (λ/4)

High-imp. (λ/4)

Low-imp. (λ/4)

High-imp. (λ/4)

Substrate

Figure 2.18 Wave propagation through SMRs and the Bragg reflector.
60 Acoustic Microresonator Technologies

Controlling the Q factor enables the design and fabrication of highly selective
filters for radio frequency applications, a field where the SMRs are being vastly
exploited. Filter topologies as varied as ladder filters, solidly mounted stacked crys-
tal filters (SCF) [32], and coupled resonator filters (CRF) [59] have been imple-
mented. By implementing a grounded electrode in the middle of the stack, the input
and output terminals of SCFs and CRFs are shielded to attain high out-of-band
rejection. Figure 2.19 shows both SCF and CRF topologies, where the CRF exhibits
additional layers in order to decrease the mechanical coupling between the top and
bottom resonators. Because these topologies are based on resonator stacking, SCF
and CRF are smaller than ladder filters, offering them competitive advantages in
reduced-size mobile applications.

2.4.4 FBAR and SMR Applications

FBAR finds application in a variety of systems ranging from radio frequency (RF)
to sensing components. Although the telecommunication industry has been the
very first engine stimulating the development of FBAR applications, new sensing

Electrode

Piezoelectric

Piezoelectric Ground plane

Electrode
m1
m2
m1
m2
m1

(a)

Electrode

Piezoelectric
Ground plane
Coupling m1 m2
layers m1

Piezoelectric
Electrode
m1
m2
m1
m2
m1

(b)
Figure 2.19 SMR-based filters: (a) stacked crystal filter (SCF); and (b) coupled resonator filter
(CRF).
2.4 Bulk Acoustic Wave Resonators 61

applications have been demonstrated in the past few years. This can boost even
more the increasing presence and importance of FBARs in the MEMS device and
system markets, as reported by independent market forecasting, reproduced here in
Figure 2.20 [60]. Nowadays, BAW devices (FBAR+SMR) represent over 90% of the
RF MEMS share, which is roughly 10% of the RF MEMS market [61]. The
BAW-based duplexer market is expected to grow from 300 million pieces in 2005 to
a total demand of well over 900 million pieces in 2010. This represents revenues
over $1.5 billion by 2010. At this time, Avago Technologies (FBAR) and TriQuint
(SMR) are the major players, with Avago having a market share 60% over of the
total BAW device business.
The demanding requirements of emerging third generation (3G) mobile tele-
communication systems have justified the search for new RF passive technologies,
FBAR among them. All these systems operate in the 2-GHz frequency band, which
is the typical FBAR’s resonance frequency (fabricated with a 1-μm-thick AlN layer).
The first RF applications of FBARs were thus devoted to supply fully passive com-
ponents, able to compete with SAW and ceramic technologies, such as filters and
duplexers. A duplexer is an RF system comprising two filters for simultaneous
bidirectional communication, the first one being the transmission (TX) filter and
the second one the reception (RX) filter. These components succeeded in offering
lower insertion losses, higher out-of-band rejection, and a smaller size than those
made with on-the-market technologies [62–67]. Also, FBAR’s temperature coeffi-
cients are in the order of −20 to −35 ppm/ºC, which is less than the typical range of
−35 to −94 ppm/ºC of SAW devices. An FBAR-duplexer implementation from
Agilent Technologies is shown in Figure 2.21, where the six-FBAR layout [Figure
2.21(a)] and demonstration package system with two filter and assembly board
[Figure 2.21(b)] are observed [31, 68].
In FBAR-based sensor applications, one or more FBAR devices are the constitu-
ent elements of a system operating under piezoelectric actuation or detection mech-
anisms. Mass sensors and biochemical, liquid, or gas detectors are some examples.

Figure 2.20 RF MEMS market forecasting 2005–2009: nowadays, BAW devices (FBAR+SMR) rep-
resent over 90% of the RF MEMS share. (© 2005 WTC [61].)
62 Acoustic Microresonator Technologies

(a)

(b)
Figure 2.21 Agilent FBAR filter implementation: (a) layout of one of the six-FBAR filters of the
duplexer (© 2001 Agilent Technologies [31]); and (b) demo package system comprising TX and
RX FBAR-filters, assembly board and connectors (© 2002 Agilent Technologies [68].)

All these applications work under the same principle of quartz crystal microbal-
ances (QCM) [69]: mass loading of the resonator’s structure [70].
According to this principle, mass deposition on one of the FBAR elec-
trodes gives rise to down-shifting of the resonance frequency, due to changes on the
acoustic-impedance mismatching and consequent phase-shifting of the acoustic
wave between the different FBAR-layer interfaces. The performance of mass-load-
ing-based detectors is evaluated through the mass sensitivity [Hz × cm2/ng] and the
minimum detectable mass [ng/cm2]. Since the operating environment of the
FBAR-based sensor determines its sensitivity and the Q-factor loading, different
aspects of the fabrication process and resonance mode are considered. In this way,
longitudinal or shear-mode operation is desired, depending on whether the sensing
medium is air, gas, or liquid.
The 2-GHz longitudinal-mode-FBAR biosensor system presented by Gabl et al.
performs DNA and protein detection operating in a liquid environment [71]. An
improved shear-mode FBAR version of the biodetection system was also imple-
mented by the same group. In this case, the sensor performance ruled by the smallest
detectable mass attachment is already better (2.3 ng/cm2) than that of QCMs [72].
Another electro-acoustic chemical sensor based on FBAR detected low concentra-
tions of the analyte upon exposure to H2, CO, and ethanol, with a fast and repeat-
2.4 Bulk Acoustic Wave Resonators 63

able response [73]. An FBAR mass sensor at its tip, inserted into biological and
chemical environments to sense various chemical-bio species, has been shown to
detect mercury ions in water [74]. Optical images of the sensor and schematic of the
operating principle are shown in Figure 2.22.
Based on the same metal-piezoelectric-metal structure and process of the FBAR,
but operating at nonlongitudinal resonance modes, different kinds of applications
have recently appeared. For example, mechanically coupled contour-mode MEMS
filters using a thin-film AlN process have been demonstrated. The use of contour
modes whose frequencies are set by lithographically defined dimensions permits the
cofabrication of multiple filters at arbitrary frequencies on the same chip, with the
filters having center frequencies of 40 and 100 MHz [75]. Also, a resonant mass sen-
sor that is based on a lateral extensional mode (LEM) ZnO resonator and has a min-
imum detectable mass (MDM) of 10−15g in air at room temperature has been
demonstrated by Pang et al. This resonator exhibits a quality factor higher than
−15
1,400 at 60 MHz and mass detection uncertainty of only about 4.6 × 10 g [76].
Some of these applications are shown in Figure 2.23.
At this point, we still need to cover CMOS-integrated FBAR applications.
Chapter 7 presents a detailed discussion on the different integration philosophies,
requirements, and fabrication processes concerning CMOS integration.

(a) (b)

Silicone rubber ring

SiN

Au film
SiN
Al
Al
Network
analyzer

(c)
Figure 2.22 Mass sensor for biological applications: (a) top and (b) bottom views of the sensor
and membrane; and (c) schematic of the mercury-ion detecting principle, where the ions interact
with the gold coating, adding mass to the resonator and thus changing the resonance frequency.
(© 2005 IOP Publishing Ltd. [74].)
64 Acoustic Microresonator Technologies

(a)

(b)
Figure 2.23 Other FBAR-based applications (nonlongitudinal resonance modes): (a) mechanically
coupled contour-mode MEMS filters (© 2006 IEEE [75]); and (b) resonant mass sensor based on a
lateral extensional mode (LEM) ZnO resonator (© 2006 IEEE [76]).

2.5 Summary

This chapter introduced the physics fundamentals, technologies, and applications

of acoustic wave resonators. We discussed surface and bulk acoustic wave
propagations and examined some of the types of acoustic waves, including Ray-
leigh, longitudinal, and shear-mode waves. The constitutive equations of piezoelec-
tric and inverse piezoelectric effects provided us the formal background to
understand the different excitation and vibration modes of acoustic resonators.
We examined SAW and BAW devices as well, providing an overview of SAW
resonator operation, design, and typology. We paid more attention to BAW devices
due to their booming performance on the radio frequency market and recently in
new sensor applications. The mechanism of BAW propagation, the characteristics
of the two types of BAW devices—FBAR and SMR—and their fabrication technol-
ogy and applications were examined. The in-depth study of these aspects constitutes
the core of this book, and extensive attention will be dedicated to them in the next
chapters.
2.5 Summary 65

References

[1] Helmholtz, H. L. F., On the Sensations of Tone as a Physiological Basis for the Theory of
Music, 4th ed., London, U.K.: Longmans, Green, and Co., 1912.
[2] Auld, B. A., Acoustic Fields and Waves in Solids, Vols. I and II, 2nd ed., Malabar, FL:
Krieger Publishing Company, 1990.
[3] Lord Rayleigh, “On Waves Propagated Along the Plane Surface of an Elastic Solid,” Proc.
London Math. Soc., Vol. s1–17, 1885, pp. 4–11.
[4] Love, A. E. H., Some Problems of Geodynamics, Cambridge, U.K.: Cambridge University
Press, 1911, and New York: Dover, 1967.
[5] Lamb, H., “On Waves in an Elastic Plate,” Proc. Roy. Soc. London, Ser. A 93, 1917,
pp. 114–128.
[6] ANSI/IEEE Std. 176-1987, “IEEE Standard on Piezoelectricity,” Institute of Electrical and
Electronics Engineers, New York, 1988.
[7] Yamanouchi, K., and M. Takeuchi, “Applications for Piezoelectric Leaky Surface Waves,”
Proc. IEEE Ultrason. Symp. 1990, Honolulu, HI, December 4–7, 1990, pp. 11–18.
[8] Epcos AG, “1855 MHz Low-Loss Filter for Mobile Communications,” 2002,
http://www.epcos.com.
[9] TriQuint Semiconductor. “881.5/1960 MHz SAW Filter,” 2009, http://www.triquint.com.
[10] Epson Toyocom, http://epsontoyocom.co.jp/english/.
[11] Epcos AG, http://www.epcos.com.
[12] Murata Manufacturing Co., Ltd., http://www.murata.com.
[13] Fujitsu Global, http://www.fujitsu.com.
[14] Hitachi Global, http://www.hitachi.com.
[15] Samsung Electronics, http://www.hitachi.com.
[16] Nippon Electric Company, http://www.hitachi.com.
[17] Murata Manufacturing Co., Ltd., “Surface Acoustic Wave Resonators,” 2001,
http://www.murata.com.
[18] Wright, P. V., “A Review of SAW Resonator Filter Technology,” Proc. IEEE Ultrason.
Symp. 1992, Tucson, AZ, October 20–23, 1992, pp. 29–38.
[19] Wright, P. V., “Low-Cost High-Performance Resonator and Coupled-Resonator Design:
NSPUDT and Other Innovative Structures,” Proc. 43th Ann. Symp. Frequency Control,
Denver, CO, May 31–June 2, 1989, pp. 574–587.
[20] Wright, P. V., “Analysis and Design of Low-Loss SAW Devices with Internal Reflections
Using Coupling-of-Modes Theory,” Proc. IEEE Ultrason. Symp. 1989, Montreal, Quebec,
October 3–6, 1989, pp. 141–152.
[21] Thompson, D. F., and P. V. Wright, “Wide Bandwidth NSPUDT Coupled Resonator Filter
Design,” Proc. IEEE Ultrason. Symp. 1991, Orlando, FL, December 8–11, 1991,
pp. 181–184.
[22] Campbell, C. K., “Understanding Surface Acoustic Wave (SAW) Devices for Mobile and
Wireless Applications and Design Techniques,” 2008, http://www3.sympatico.ca/
colin.kydd.campbell/.
[23] Epson Toyocom, “GHz Band, High Accuracy SAW Resonators and SAW Oscillators,”
http://epsontoyocom.co.jp/english/.
[24] Fujitsu Media Devices Ltd., Cellular SAW Duplexer, FAR-D5GA-881M50-D1AA, 2006,
http://www.fujitsu.com.
[25] Epcos AG, Surface Acoustic Wave Components Division, “SAW Frontend Module GSM
850/EGSM/DCS/PCS/WCDMA 2100,” 2008, http://www.epcos.com.
[26] Martin, F., “Pulse Mode Shear Horizontal-Surface Acoustic Wave (SH-SAW) System for
Liquid Based Sensing Applications,” Biosensors and Bioelectr., Vol. 19, 2004,
pp. 627–632.
66 Acoustic Microresonator Technologies

[27] Talbi, A., et al., “ZnO/Quartz Structure Potentiality for Surface Acoustic Wave Pressure
Sensor,” Sens. Actuators A: Phys., Vol. 128, 2006, pp. 78–83.
[28] Tsarenkov, G. V., “10+ GHz BAW Resonators Based on Semiconductor Multilayer
Heterostructures,” Proc. IEEE Intl. Ultrason. Symp. 1999, Tahoe, NV, October 17–19,
1999, pp. 939–942.
[29] Lakin, K. M., and J. S. Wang, “Acoustic Bulk Wave Composite Resonators,” Appl. Phys.
Lett., Vol. 38, 1981, pp. 125–127.
[30] Chao, M.-C., et al., “Modified BVD-Equivalent Circuit of FBAR by Taking Electrodes into
Account,” Proc. IEEE Intl. Ultrason. Symp. 2002, Munich, Germany, October 8–12,
2002, pp. 973–976.
[31] Mueller, W., A Brief Overview of FBAR Technology, Agilent Technologies, technical
report AB-WCM200701a, July 20, 2001.
[32] Lakin, K. M., G. R. Kline, and K. T. McCarron, “Development of Miniature Filters for
Wireless Applications,” IEEE Trans. on Microwave Theory Tech., Vol. 43, 1995,
pp. 2933–2939.
[33] Aigner, R., “Volume Manufacturing of BAW-Filters in a CMOS Fab,” Proc. Second Inter-
national Symposium on Acoustic Wave Devices for Future Mobile Communication Sys-
tems, Chiba, Japan, March 3–5, 2004, pp. 129–134.
[34] Emanetoglu, N. W., “MgxZn1-xO: A New Piezoelectric Material,” Proc. IEEE Intl.
Ultrason. Symp. 2001, Vol. 1, Atlanta, GA, October 7–10, 2001, pp. 253–256.
[35] Fritze, H., and H. L. Tuller, “Langasite for High-Temperature Bulk Acoustic Wave Appli-
cations,” Appl. Phys. Lett., Vol. 78, 2001, pp. 976–977.
[36] Engelmark, F., “Structural and Electroacoustical Studies of AlN Thin Films During Low
Temperature Radio Frequency Sputtering Deposition,” J. Vac. Sci. Technol. A, Vol. 19,
2001, pp. 2664–2669.
[37] Shiosaki, T., “Low Temperature Growth of Piezoelectric Films by RF Reactive Planar Mag-
netron Sputtering,” Jap. J. Appl. Phys., Vol. 20, 1981, pp. 149–152.
[38] Uchiyama, S., et al., “Growth of AlN Films by Magnetron Sputtering,” J. Crystal Growth,
Vol. 189–190, 1998, pp. 448–451.
[39] Vispute, R. D., H. Wu, and J. Narayan, “High Quality Epitaxial Aluminum Nitride Layers
on Sapphire by Pulsed Laser Deposition,” Appl. Phys. Lett., Vol. 67, 1995, pp. 1549–1551.
[40] Dubois, M.-A., and P. Muralt, “Stress and Piezoelectric Properties of Aluminum Nitride
Thin Films Deposited onto Metal Electrodes by Pulsed Direct Current Reactive Sputter-
ing,” J. Appl. Phys. Vol. 89, 2001, pp. 6389–6395.
[41] Shen, L., R. K. Fu, and P. K. Chu, “Synthesis of Aluminum Nitride Films by Plasma Immer-
sion Ion Implantation–Deposition Using Hybrid Gas–Metal Cathodic Arc Gun,” Rev.
Scient. Instrum., Vol. 75, 2004, pp. 719–724.
[42] Lee, J.-B., et al., “Effects of Bottom Electrodes on the Orientation of AlN Films and the Fre-
quency Responses of Resonators in AlN-Based FBARs,” Thin Solid Films, Vol. 447–448
2004, pp. 610–614.
[43] Akiyama, M., et al., “Influence of Metal Electrodes on Crystal Orientation of Aluminum
Nitride Thin Films,” Vacuum, Vol. 74, 2004, pp. 699–703.
[44] Huang, C.-L., K.-W. Tay, and L. Wu, “Fabrication and Performance Analysis of Film Bulk
Acoustic Wave Resonators,” Mater. Lett., Vol. 59, 2005, pp. 1012–1016.
[45] Hara, M., et al., “Surface Micromachined AlN Thin Film 2 GHz Resonator for CMOS
Integration,” Sens. Actuators A: Phys., Vol. 117, 2005, pp. 211–216.
[46] Saravanan, S., et al., “A Novel Surface Micromachining Process to Fabricate AlN
Unimorph Suspensions and Its Application for RF Resonators,” Sens. Actuators A: Phys.,
Vol. 130–131, 2006, pp. 340–345.
[47] Pang, W., H. Zhang, and E. S. Kim, “Micromachined Acoustic Wave Resonator Isolated
from Substrate,” IEEE Trans. on Ultrason. Ferroelectr. Freq. Control, Vol. 52, 2005,
pp. 1239–1246.
2.5 Summary 67

[48] Piazza, G., et al., “Single-Chip Multiple-Frequency RF Microresonators Based on Alumi-

num Nitride Contour-Mode and FBAR Technologies,” Proc. IEEE Intl. Ultrason. Symp.
2005, Rotterdam, the Netherlands, September 18–21, 2005, pp. 1187–1190.
[49] Satoh, Y., et al., “Development of 5GHz FBAR Filters for Wireless Systems,” Proc. 2nd
International Symposium on Acoustic Wave Devices for Future Mobile Communication
Systems, Chiba, Japan, March 3–5, 2004, pp. 141–144.
[50] Rosén, D., J. Bjursttröm, and I. Katardjiev, “Suppression of Spurious Lateral Modes in
Thickness-Excited FBAR Resonators,” IEEE Trans. on Ultrason. Ferroelectr. Freq. Con-
trol, Vol. 52, 2005, pp. 1189–1192.
[51] Kim, Y.-D., et al., “Highly Miniaturized RF Bandpass Filter Based on Thin-Film Bulk
Acoustic-Wave Resonator for 5-GHz-Band Application,” IEEE Trans. on Microwave The-
ory Tech., Vol. 54, 2006, pp. 1218–1228.
[52] Su, Q., et al., “Thin-Film Bulk Acoustic Resonators and Filters Using ZnO and Lead-Zirco-
nium-Titanate Thin Films,” IEEE Trans. on Microwave Theory Tech., Vol. 49, 2001,
pp. 769–778.
[53] Pan, W., “A Surface Micromachined Electrostatically Tunable Film Bulk Acoustic Resona-
tor,” Sens. Actuators A: Phys., Vol. 126, 2006, pp. 436–446.
[54] Ruby, R. C., and P. P. Merchant, “Tunable Thin Film Acoustic Resonators and Method for
Making the Same,” U.S. Patent No. 5587620, December 24, 1996.
[55] Ylilammi, M. A., “Method for Performing On-Wafer Tuning of Thin Film Bulk Acoustic
Wave Resonators (FBARS),” U.S. Patent No. 6051907, April 18, 2000.
[56] Vanhelmont, F., et al., “A 2 GHz Reference Oscillator Incorporating a Temperature Com-
pensated BAW Resonator,” Proc. IEEE Intl. Ultrason. Symp. 2006, Vancouver, B.C.,
October 3–6, 2006, pp. 333–336.
[57] Larson III, J. D., “Acoustic Wave Resonator and Method of Operating the Same to Main-
tain Resonance When Subjected to Temperature Variations,” U.S. Patent No. 6,420,820,
July 16, 2002.
[58] Aigner, R., “High Performance RF-Filters Suitable for Above IC Integration: Film
Bulk-Acoustic-Resonators (FBAR) on Silicon,” Proc. IEEE 2003 Custom Integrated Cir-
cuit Conference 2003, San Jose, CA, September 21–24, 2003, pp. 141–146.
[59] Lakin, K. M., et al., “Bulk Acoustic Wave Resonators and Filters for Applications Above 2
GHz,” Dig. Tech. Papers IEEE Intl. Microwave Symposium MTT-S 2002, Seattle, WA,
June 2–7, 2002, pp. 1487–1490.
[60] Inouye, S., “Electronics.ca Research Network, 2006 MicroElectroMechanical Systems
(MEMS),” Research Report # DB2563, November 2006, pp. 1–48.
[61] Wicht Technologie Consulting, “RF MEMS Market II, 2005-2009,” 2005.
[62] Bradley, P., et al., “A Film Bulk Acoustic Resonator (FBAR) Duplexer for USPCS Handset
Applications,” Dig. of Tech. Papers IEEE International Microwave Symp. MTT-S 2001,
Phoenix, AZ, May 20–25, 2001, pp. 367–370.
[63] Ruby, R., et al, “Ultra-Miniature High-Q Filters and Duplexers Using FBAR Technology,”
Dig. of Tech. Papers IEEE Intl. Solid-State Circuits Conf., San Francisco, CA, February
5–7, 2001, pp. 120–121.
[64] Ylilammi, M., et al., “Thin Film Bulk Acoustic Wave Filter,” IEEE Trans. on Ultrason.
Ferroelectr. Freq. Control, Vol. 49, 2002, pp. 535–539.
[65] Nishihara, T., et al., “High Performance and Miniature Thin Film Bulk Acoustic Wave Fil-
ters for 5 GHz,” Proc. IEEE Intl. Ultrason. Symp. 2002, Munich, Germany, October 8–12,
2002, pp. 969–972.
[66] Morkner, H., et al., “An Integrated FBAR Filter and PHEMT Switched-Amp for Wireless
Applications,” Dig. of Tech. Papers IEEE International Microwave Symp. MTT-S 1999,
Anaheim, CA, June 13–19, 1999, pp. 1393–1396.
68 Acoustic Microresonator Technologies

[67] Feld, D., et al., “A High Performance 3.0 mm × 3.0 mm × 1.1 mm FBAR Full Band Tx Filter
for US PCS Handsets,” Proc. IEEE Intl. Ultrason. Symp. 2002, Munich, Germany, Octo-
ber 8–12, 2002, pp. 913–918.
[68] Agilent Technologies, Applications Engineering Team, Personal Systems Division, “Using
the HPMD-Series and QPMD-Series CDMA Duplexers,” Design Tip (technical report),
April 12, 2002.
[69] Lu, C. S., Applications of Piezoelectric Quartz Crystal Microbalance, London, U.K.:
Elsevier, 1984.
[70] G. Z. Sauerbrey, “Verwendung von Schwingquarzen zur Wägung dünner Schichten und
Microwägung,” Z. Phys., Vol. 155, 1959, pp. 206–222.
[71] Gabl, R., et al., “First Results on Label-Free Detection of DNA and Protein Molecules
Using a Novel Integrated Sensor Technology Based on Gravimetric Detection Principles,”
Biosens. Bioelectron., Vol. 19, 2004, pp. 615–620.
[72] Weber, J., et al., “Shear Mode FBARs as Highly Sensitive Liquid Biosensors,” Sens. Actua-
tors A: Phys., Vol. 128, 2006, pp. 84–88.
[73] Benetti, M., et al., “Microbalance Chemical Sensor Based on Thin-Film Bulk Acoustic
Wave Resonators,” Appl. Phys. Lett., Vol. 87, 2005, p. 173504.
[74] Zhang, H., et al., “A Film Bulk Acoustic Resonator in Liquid Environments,” J.
Micromech. Microeng., Vol. 15, 2005, pp. 1911–1916.
[75] Stephanou, P. J., et al., “Mechanically Coupled Contour Mode Piezoelectric Aluminum
Nitride MEMS Filters,” Proc. IEEE Intl. Conf. MEMS 2006, Istanbul, Turkey, Janu-
ary 22–26, 2006, pp. 906–909.
[76] Pang, W., et al., “Ultrasensitive Mass Sensor Based on Lateral Extensional Mode (LEM)
Piezoelectric Resonator,” Proc. IEEE Intl. Conf. MEMS 2006, Istanbul, Turkey, January
22–26, 2006, pp. 78–81.
 CHAPTER 3

Design and Modeling of Micro- and

Nanoresonators
Modeling is a fundamental task in the production cycle of FBAR, MEMS, and
NEMS resonators. Modeling allows prediction of the resonator static and dynamic
responses, and, ideally, it is performed at the early stages of the design process,
when the decisions of choosing materials, geometries, and dimensions are to be
made. Besides aiding the designer in defining the fabrication technology, modeling
allows analysis of the performance of a working device, so we can fit the actual fre-
quency response to given models in order to extract their parameters. Design of
modern system-on-a-chip blocks composed of resonators and integrated circuits is
largely based on modeling activity, where the physical interconnection of the system
creates a mutual and complex interaction that has to be quantified.
The design of resonators is addressed by different modeling approaches, each
having its own context and usefulness. Analytical models are based on mechanical,
electromechanical, and purely electrical equivalent-circuit representations of the
resonator physics. In this chapter, we study these models, getting started with the
stages of design and modeling introduced in Section 3.1. Next, Section 3.2 examines
the electromechanical transformer concept, which gives the analytical elements and
initial criteria to build more complex resonator models. Then, the circuit theory per-
spective of equivalent-circuit parameter models is presented in Section 3.3. With
this at hand, the components, structure, and analyses types of finite element models
are studied in Section 3.4.

3.1 The Stages of Resonator Design and Modeling

In Chapter 1, we studied the production cycle of MEMS and NEMS resonators. In

that cycle, the design stages previous to manufacturing are based on models and
modeling tools that represent the physics of the resonator. Each model defines an
input-output relationship with regard to its level of abstraction, assuming that the
system and manufacturing levels are the highest and lowest ones in the design
hierarchy, respectively.
Starting from the system-level specifications and descending in the direction of
manufacturing, the design goes through successive modeling stages involving less
abstraction and more physics and fabrication parameters, as depicted in Figure 3.1.
Mature, independent, and well-established modeling tools exist in the market to
accomplish the calculations and analysis. These tools specialize in the different
stages of the modeling process. If the resonator is part of a bigger system—like an IC

69
70 Design and Modeling of Micro- and Nanoresonators

Figure 3.1 Modeling hierarchy for micro and nanoresonator design.

or a network of MEMS/NEMS devices, its design can be structured or

nonstructured as long as the resonator and IC design tasks go in parallel or
nonparallel ways [1, 2].
System simulations are performed at the schematic level, and the components of
the model are connected by nets. Each component implements a behavioral model
or transfer function. Commercial tools like CADENCE (Virtuoso Schematic),
COVENTOR (Architect), and MATLAB-Simulink feature net-connection capabili-
ties for interconnecting different building blocks of a system [3–5]. COVENTOR,
for example, provides libraries for MEMS-IC interconnection. In the CADENCE
environment, Virtuoso Schematic offers powerful networking capabilities, thus
connecting Spice or Spectre models with behavioral customized models made by the
user in Verilog-A or VHDL-AMS [6–9]. Figure 3.2 shows the level of abstraction of
a system model comprising a MEMS resonator and amplifier integrated circuit for
read out of the MEMS current.
However, in the user’s eyes, the system-level model does not necessarily make
explicit reference to the inner relations of the input-output variables of the building
blocks. In this way, the user only instantiates and connects the blocks, and the simu-
lator calculates the values of the variables according to the transfer functions of the

Read-out circuit
VBIAS
VIN VDD VDD

VOUT

Amplifier Buffer
Figure 3.2 System-level model of a read-out circuit of a MEMS resonator (circuit schematic of the
resonator and the read-out circuit).
3.1 The Stages of Resonator Design and Modeling 71

blocks and the analysis type. AC analysis, for example, is performed by sweeping a
user-selected variable to obtain the response of the interest variable against time,
frequency, temperature, or any other independent variable. Figure 3.3 plots the
bode (frequency response) of the read-out circuit and the MEMS resonator [10].
The user only needs to specify the MEMS Q factor, insertion loss, and resonance
frequency, which are enough to build its second-order transfer function. Alterna-
tively, the MEMS block can be specified by its lumped-circuit elements, and the
amplifier described by its AC gain and bandwidth, in a first approximation.
The transfer function of the system can also be implemented by the user.
MATLAB is very efficient in coding the transfer function and analysis type as a
new function by following the MATLAB syntax rules. Also, the user takes advan-
tage of preexisting MATLAB functions to simplify its own function and make
the script more compact and readable. This approach gives the user more control
of the model, although it requires manual building of the blocks. Much of the
physical-level information of the components, like layout, geometries, and process
parameters, is not considered at this stage. Figure 3.4 plots the root locus of the
open-loop system composed by an amplifier and a 2-GHz resonator. The ampli-
fier is assumed to be a CMOS circuit in common-source configuration and
transimpedance gain gm, which is serially connected to the second-order resonator
so their transfer functions are multiplied in the frequency domain. With the
root-locus plot, the system stability is studied through the amplifier and resonator
parameters [11].
At this point, the physical nature of the system becomes less intuitive and begins
to be more explicitly described by the lower-level behavioral models. A behavioral
model reproduces the physical behavior of the system by integrating its mathemati-
cal model and underlying physics in an attempt to make the model as full and realis-
tic as the physical system. Behavioral modeling has become a current-day circuit

Figure 3.3 Bode plot (frequency response) of the read-out circuit and resonator.
72 Design and Modeling of Micro- and Nanoresonators

(a)

(b)
Figure 3.4 (a, b) Root locus of the open-loop FBAR and amplifier circuit.

and MEMS design technique supported by computer-aided design (CAD) tools and
description languages like VHDL-AMS and Verilog-A [12].
The behavioral model of an FBAR with linear thermal coefficient factor (TCF)
is built from its equivalent lumped-element RLC model and the TCF (the equiva-
lent-circuit model of FBARs and other resonators will be studied in Section 3.3).
The TCF, reference temperature, and RLC values at that temperature are the main
parameters of the model: the TCF is fixed, the RLC elements are extracted from
experimental measurements, and a reference temperature is fixed is chosen. This
example deals with an electrothermal model, thus integrating both electrical and
3.1 The Stages of Resonator Design and Modeling 73

thermal elements. The data plot of Figure 3.5 shows the evolution of resonance fre-
quency as a function of temperature, described by this behavioral model. Simula-
tions in the −20ºC to +80ºC temperature range were performed in 10ºC steps, with a
TCF equal to −25 ppm/ºC and a reference temperature of 40ºC (it can be noted that
the TCF equals to 0 ppm/ºC at that temperature). The resonance frequencies in the
temperature range are found in the 2.2-GHz band. This model is useful for design-
ing complete systems that integrate the resonator, like RF oscillators. The design of
a temperature-compensated FBAR-based oscillator is studied in Chapter 10, where
the behavioral model of the FBAR and the circuit are described in detail.
The system and component-level models give us a vision of the expected reso-
nance frequencies and physical parameter influence in the performance of the reso-
nator. After that, 3D finite element modeling (FEM) links the analytical
formulation, the geometry, and the materials of the resonator, so we can predict its
static and dynamic responses with high precision. Later in Section 3.4, we study the
characteristics and design flow of a reliable FEM analysis.
The ANSYS model of Figure 3.6 compares the equilibrium (edged) and
deformed (solid) configurations of a resonating gallium arsenide (GaN) nanowire.
In the FEM, the user introduces the mechanical properties of GaN—elasticity, den-
sity, Poisson modulus—as simulation parameters. The FEM is composed of a num-
ber of elements, typically hundreds or thousands, which are represented in the
figure by the small “bricks.” After modal analysis, the fundamental resonance fre-
quency of the 2-μm-long nanowire with a diameter of 50 nm equals 11.9 MHz.
A reliable FEM analysis gives the designer enough confidence to complete the
physical 2D layout, which will be used in the resonator manufacturing. The layout
is designed in layers that represent the different processes being implemented in the
fabrication. Depending on the available technology, each one of these layers is
employed in photolithography mask fabrication, laser photolithography, or other
suitable lithography or patterning means. The layout is drawn in accordance with
the design rules and process restrictions of the target technology. Commercial tools
like CADENCE offer sophisticated options to develop complex designs and to
perform design-rule checking.

Figure 3.5 Behavioral description of an FBAR: parametric frequency response of the model as a
function of temperature.
74 Design and Modeling of Micro- and Nanoresonators

Figure 3.6 3D finite element model of a GaN nanoresonator: the nanowire is 2 μm long with a
diameter equal to 50 nm, and it has a resonance frequency of 11.9 MHz.

In MEMS and NEMS manufacturing, the involved micromachining or

nanopatterning techniques add complexity to the layout design. Due to the lack of
standard processes for movable-structure releasing, the layout has to consider the
performance of the etching techniques developed for the specific case of the interest
resonator technology. Anisotropy, etching times, and selectivity are some of the fac-
tors conditioning the geometry and size of the structural and sacrificial layers of the
technology. In deep silicon etching processes like DRIE, for example, the etching
times are different for small and big-sized features. Underetching or lateral etching
can also be important, so enough margins must be kept in the lateral dimensions to
guarantee the integrity of the structure at the end of the process. Figure 3.7 depicts
the three-layer layout of an AlN-made FBAR resonator (the etching layer is not
included). As seen in the drawing, the bottom electrode is designed to have less
width than both the AlN and top electrode layers. Also, the AlN layer is wider than
the top and bottom electrodes. This is due to different reasons: First, to cover possi-
ble misalignment during photolithography (the tolerance of the photolithography
process is assumed to be known and smaller than the width margin). This guaran-
tees that the bottom electrode will always be underneath both the AlN and the top
electrode, thus fixing the duty resonator area. Second, short circuits of the top and
bottom electrodes are prevented. This may happen if the undesired lateral etching of
the AlN causes it to be narrowed beyond the acceptable limits imposed by the AlN
thickness and the etching rates. The fabrication details regarding FBARs and the
issues of lateral and underetching are covered in Chapter 4.
So far, we have given a rapid overview to the stages of the design flow. The
design and modeling processes conclude with the manufacturing of devices. Further
modeling can be carried out on fabricated resonators when the equivalent-circuit
parameters are extracted, for a given circuital model. In the next sections, we dis-
cuss in detail each level of the modeling workflow, from the higher abstraction of
equivalent circuits to the more physical FEM and layout optimization.
3.2 The Electromechanical Transformer 75

(a)

(b)
Figure 3.7 Physical layout view of an AlN-based FBAR: (a) 2D top view as seen by the designer
and (b) 3D view.

3.2 The Electromechanical Transformer

FBARs, MEMS, and NEMS resonators convert the input energy from the electrical
to the mechanical domains, and vice versa. This process entails, on the one hand,
input-to-output domain coupling and, on the other hand, an effective electrome-
chanical energy-conversion factor. The resonator is, in this sense, an electrome-
chanical transformer, and it can be modeled by lumped-circuit electrical and
mechanical components linked through the transformer, among other components
in the model. The components of such an electromechanical transformer-based
model are chosen depending on the transduction mechanism. These models help the
designer understand the operation principles and their relationship with
transduction, topology, and geometry of the resonator, thus providing useful infor-
76 Design and Modeling of Micro- and Nanoresonators

mation to the FEM design and analysis. The modeling particularities of

electromechanical and acoustic resonators are discussed next.

3.2.1 MEMS and NEMS Resonators

The mass-spring damped system depicted in Figure 3.8 illustrates the mechanical
oscillation of MEMS and NEMS resonators. This model is in close relation with the
mathematical description of the oscillatory movement and the resonance frequency
provided by (1.1) and (1.2), respectively, which we reproduce herein for the reader
convenience [13]:

&& + Dx& + kx = Fext ( x , t ) = F0 ⋅ sin(2πft )

m eff x (3.1)

1 k
f0 = (3.2)
2π m eff

where x is the time-varying harmonic oscillation, meff is the equivalent mass of the
fundamental resonance mode with frequency f0, D is the damping constant, k is the
spring constant, Fext is the external force applied to the resonator, and F0 is the
amplitude of the harmonic force signal with frequency f.
Figure 3.8 and (3.1) and (3.2) illustrate the oscillation and include the excitation
force as the system input. However, they lack a description of the physical mecha-
nism causing the vibration, because it depends on the transduction type. To exem-
plify, let’s consider a MEMS resonator electrostatically driven by both AC and DC
voltage signals, such as that previously depicted in Figure 1.7. The mechanical cir-
cuit constituted by the mass-spring-damper system is fed by the electric voltage sig-
nals vAC and VDC, as it represents the circuital model of Figure 3.9(a). Since certain
equivalence between the mechanical and electrical domains exists, this circuit can
also be depicted in terms of the electrically equivalent RLC parameter circuit of Fig-
ure 3.9(b).
A stricter equivalence between the mechanical and the RLC elements is done
by adding a couple of electromechanical transformers in the circuit, as shown in

−x1 x0 +x 1

k D
−x1

x0
meff
+x 1

Figure 3.8 Mass-spring damped mechanical model of a MEMS resonator.

3.2 The Electromechanical Transformer 77

k
VAC
D
meff
VDC

(a)

α meff Rm αk

VAC

VDC

(b)
Figure 3.9 Equivalent circuits of the electrostatically driven MEMS resonator: (a) mechanical
model and (b) electrically equivalent RLC model.

Figure 3.10. On the left side, the voltage source velec shunts the transformer and the
static capacitance C0 formed between the walls of the driver electrode and the reso-
nator. The source velec includes both the AC and DC voltages, and the electrome-
chanical transformer has efficiency ηe. Thus, the arrangement models the
transformation of input voltage into mechanical motion. Note that, while the cir-
cuit at the left side of the transformer is electrical, its right side links to the mechani-
cal-side circuit formed by the mass meff, spring k, and damper D. The electrical
analogy is the same of Figure 3.9. That is, the inductance equals meff, the capacitor
equals 1/k, and the resistance equals D. At right, the second transformer models the
mechanical impedance transformation due to the mechanical coupling between the
beam and the clamping system [14].
At the driving electrode port (shunted by C0), the equivalent RLC resonator cir-
cuit elements are transformed by the electromechanical ratio ηe, and their values are
[14]:

D
Rm = (3.3)
η e2

m eff
Lm = (3.4)
η 2e

1:ηe meff D 1/k 1:ηc

velec CO f mech

Figure 3.10 Electromechanical model of the MEMS resonator coupling the electrical and
mechanical domains through the electromechanical transformers: the circuit is bidirectional and
has both electrical (voltage velec) and mechanical (force Fmech) inputs and outputs. (After: [14].)
78 Design and Modeling of Micro- and Nanoresonators

η e2
Rm = (3.5)
k

where Rm, Lm, and Cm are the motional resistance, inductance, and capacitance of
the equivalent circuit, respectively, seen at the driving electrode port.
The electromechanical transformer model is completely bidirectional to repro-
duce the transduction physics. This means that the input voltage velec induces an
electrostatic force on the MEMS, thus causing mechanical vibration and the subse-
quent mechanical force Fmech. Alternatively, an input mechanical force Fmech creates
an electromotive displacement, and the resulting displacement current flows
through C0, where the output voltage velec can be read out.

3.2.2 FBAR and Other Acoustic Resonators

Electromechanical transformer models of FBAR, SMR, and other acoustic resona-
tors are mainly based on the Mason’s or the Krimholtz-Leedom-Matthaei’s (KLM)
models [15, 16]. The Mason’s model, for example, shows that most of the difficul-
ties in deriving the solutions for the wave propagation through the acoustic layer are
overcome by using the network theory approach. Thus, the Mason’s model presents
an exact equivalent circuit that separates the piezoelectric material into one electric
port and two acoustic ports through the use of an ideal electromechanical trans-
former. Each acoustic port represents the acoustic paths to the top and bottom elec-
trodes, while the electric port sees the electric input impedance Zin opposing to
current circulation. For longitudinal-mode FBARs, the equivalent circuit represen-
tation of the FBAR, according to the Mason’s model, is depicted in the schematic
drawings of Figure 3.11. A similar approach is followed by the KLM model.
The input impedance Zin seen from the half-phase line of the AlN layer, between
the two parallel electrodes is [17]:

1 ⎛ 2 tan φ ( Z T + Z B ) cos 2 φ + j sin 2 φ ⎞

Z in = × ⎜⎜1 − keff ⎟ (3.6)
jωC 0 ⎝ φ ( Z T + Z B ) cos 2 φ + j( Z T + Z B + 1) sin 2 φ⎟⎠

2
where φ is the half phase across the piezoelectric plate, k eff is the piezoelectric cou-
pling coefficient of the AlN film, C0 is the parallel-plate capacitance between the
two electrodes, and ZT and ZB are normalized acoustic impedances at the piezoelec-
tric layer boundaries (with respect to the acoustic impedance of the AlN layer).
Equation (3.6) can be divided into three components: the electrical one, the purely
mechanical one, and the electromechanical transformer with gain N:

tan φ
N = keff
2
(3.7)
φ

The phase across the piezoelectric layer is equal to:

φ = κ×t (3.8)
3.2 The Electromechanical Transformer 79

Top
electrode

Piezoelectric
layer

Bottom
electrode

(a)

Top electrode Bottom electrode

Acoustic Port (1) Acoustic Port (2)

Electric Port
(b)
Figure 3.11 FBAR electromechanical model: (a) cross-section view of the FBAR with equivalent
impedance values and input and output ports, and (b) equivalent circuit representation of the
FBAR, according to the Mason’s model.

where κ and t are the wave vector and the thickness of the piezoelectric layer, and κ
depends on the frequency f and acoustic velocity v in the piezoelectric film, as given
by:

2 πf
κ= (3.9)
v

The boundary impedances ZT and ZB at the interface between the AlN and the
electrodes are determined by the acoustic impedance matching between both media.
If more than one metal is used in the fabrication of electrodes, or if another material
is stacked, both the acoustic path and the input impedance seen from the electrical
side change. The values of ZT and ZB can be found by [18]:
80 Design and Modeling of Micro- and Nanoresonators

⎛ Z Load cos φ T B + jZ 0T B sin φ T ⎞

= Z 0T ×⎜ TB ⎟
B
ZT B
(3.10)
B ⎜Z ⎟
⎝ 0 cos φ T B + jZ Load sin φ T B⎠

where Z0T/B is the characteristic acoustic impedance of either the top (T) or bottom
(B) electrode’s layer, ZLoad is the input load impedance seen by either the top or bot-
tom electrode’s layer at the next interface, and φT/B is the acoustic-wave phase across
either the top or bottom electrode’s layer. In FBARs, the top and bottom electrode
surfaces have an air interface (although in real implementations, some area of the
electrodes still contacts the substrate). Since air has low acoustic impedance, the
air-to-electrode interfaces have boundary impedances equal to zero. Final calcula-
tion of the electrical impedance Zin requires recursive calculation of ZT and ZB,
beginning with the air interfaces and ending at the AlN layer. The circuit of Figure
3.12 represents the acoustic and electric impedances of the FBAR Mason’s model.
This is equivalent to the analytical formulation of (3.6) and (3.10), and it gives a
graphic understanding of the transmission line formed by the different layers of the
resonator (it may be useful to complete the recursive calculations of ZT and ZB as
well).
This model can be implemented by computing software like Mathematica [19]
or MATLAB, and it may be served to rapid prediction of the fundamental and har-
monic acoustic modes of the resonator and its sensitivity to material properties or
layer configuration. The curve of Figure 3.13 shows the ideal, lossless transmission
response of an FBAR from 1 to 20 GHz. Five modes are observed in the plot, the
first and main one being around 2 GHz.
A similar approach can be employed to model the wave propagation and the
input impedance seen from the electric port of solidly mounted resonators (SMRs).
By adding the Bragg’s mirror reflector layers to the model [20], the input impedance
Zin can be calculated, as depicted in Figure 3.14. In this case, what is modified is the

Air Pt AIN layer Pt Air

(bottom) (top)

Pt AIN Pt
jz 0 tan (φPt/2) .A jz 0 tan (φ/2) .A jz 0 tan (φPt/2) .A
jz 0 sin (φ/2) .A
jz 0 sin (φPt/2) . A

jz 0 sin (φPt/2) . A
AIN

hC 0
Pt

Pt

−C 0

C0

Z IN

Figure 3.12 Transmission-line representation of the acoustic and electric paths of the Mason’s
model of FBARs.
3.2 The Electromechanical Transformer 81

−10

−20
S21mag (dB)

−30

−40

−50

−60

−70
0 0.5 1 1.5 2
Frequency (GHz) X 10
10

Figure 3.13 Frequency response of an FBAR predicted by the Mason’s model (MATLAB simula-
tion).

Top
electrode

Piezoelectric
layer

Bottom
electrode
Low
impedance

High
impedance
Low
impedance

Figure 3.14 Equivalent acoustic impedances of solidly mounted resonator (SMR): the bottom
impedance ZB is modified.

acoustic impedance ZB seen by the bottom electrode due to the added layers under-
neath the resonator.
82 Design and Modeling of Micro- and Nanoresonators

The circuit of Figure 3.15 represents the equivalent transmission line of an SMR
that includes the acoustic impedances of the Bragg’s mirror. According to (3.6) and
(3.10), it shows how the calculation of ZB requires more steps than in the FBAR
model, because the transmission line is longer due to the reflecting layers of the mir-
ror. Electromechanical models of FBARs and SMRs are served to predict the imped-
ance mismatching and resonance frequency changes due to new layers in the stack,
which is useful for analyzing sensor and filter applications.

3.3 Equivalent-Circuit Models

Equivalent-circuit modeling is a powerful resonator analysis and design tool, which

represents the resonator as a lumped-element circuit in the electrical domain. The
parameters of the equivalent circuit characterize the physics and fabrication process
of the resonator, and they are also an electrical abstraction of the motional response
of the resonator (related to processing but without physical meaning). At the system
level, electrical equivalent-circuit models aid the fitting and parameter extraction
used in the design of circuits integrating FBARs, MEMS, and NEMS resonators.

3.3.1 The Resonant LC Tank

The basic equivalent circuit of resonators is the lossless resonant LC tank depicted
in Figure 3.16. The tank is a second-order, frequency-selective impedance that
stores electromechanical energy, whose resonance frequency f0 is given by [21]:

1
f0 = (3.11)
2 π Lt Ct

Pt AIN layer Pt Air

High Low (bottom) (top)

Hi Lo Pt AIN Pt
jz 0 tan (φHi/2) .A jz 0 tan (φLo /2) .A jz 0 tan (φPt/2) .A jz 0 tan (φ/2) .A jz 0 tan (φPt/2) .A
jz 0 sin (φHi /2) . A

jz 0 sin (φLo /2) . A

jz 0 sin (φPt/2) . A

jz 0 sin (φ/2) .A

jz 0 sin (φPt/2) . A
AIN
Lo
Hi

Pt

Pt

hC 0

−C 0

C0

Z IN

Figure 3.15 Transmission line model of an SMR.

3.3 Equivalent-Circuit Models 83

Parallel Series

L
C L
C

(a) (b)
Figure 3.16 The resonant LC tank: (a) parallel circuit (at resonance: maximum impedance, mini-
mum admittance); and (b) series circuit (at resonance: minimum impedance, maximum admit-
tance).

where Lt and Ct are the equivalent inductance and capacitance of the tank. The par-
allel and serial LC tanks behave like resonant circuits, but the frequency-dependent
impedances are opposite. At the low frequencies, far below the resonance fre-
quency, the series LC tank is essentially a capacitor of impedance 1/jωCt. Far above
resonance, the tank presents an inductive behavior with impedance approximately
equal to jωLt. At resonance, the series inductive and capacitive impedances compen-
sate among them, and the overall impedance is minimized. The parallel tank works
in the opposite way. At the resonance frequency, its impedance and admittance
reach maximum (infinite) and minimum (zero) values, respectively.
However, physical systems are lossy circuits that prevent infinite energy storage
in the tank. Thus, the basic lossless LC circuit is complemented with a resistor repre-
senting the losses in the tank that allows calculation of the quality factor [22]:

1
Q= (3.12)
2 πf 0 RC

In terms of L, this is equivalent to:

2 πf 0 L
Q= (3.13)
R

The resonant tank models to a large extent the behavior of radio frequency
components, antennas, waveguides, and transmission lines, which are actually
tuned resonant circuits. Nevertheless, the process and transduction particularities
of acoustic and electromechanical resonators require models whose parameters
describe their physics and operation. We examine this subject in the following
section.

3.3.2 The Butterworth-Van-Dyke Model

The Butterworth-Van-Dyke (BVD) model is an equivalent-circuit representation
suitable for crystal resonators, FBARs, and electrostatically actuated MEMS and
NEMS resonators [23]. Figure 3.17 shows the BVD circuit, which has two parallel
branches—namely, the motional and static capacitance arms. The motional arm
comprises the series motional inductance Lm, capacitance Cm, and resistance Rm.
84 Design and Modeling of Micro- and Nanoresonators

Lm

C0 Rm

Cm

Static capacitance Motional

arm arm
Figure 3.17 Butterworth-Van-Dyke model of micro- and nanoresonators depicting the motional
and static-capacitance branches.

Although without physical meaning, motional elements relate to process-dependent

parameters and material constants. The static capacitance branch is formed by the
parallel-plate capacitance C0 formed between the top and bottom electrodes of
FBAR and SMR resonators, or between the cantilever and the electrodes of MEMS
resonators.
As a two-branch parallel circuit, BVD models explain the series and parallel res-
onance frequencies found in FBAR and electrostatic MEMS resonators, given by:

1
fS = (3.14)
2π L m C m

1
fP = (3.15)
⎛ C C ⎞
2π Lm ⋅ ⎜ m 0 ⎟
⎝Cm + C0 ⎠

where fS and fP are the series and parallel resonance frequencies, respectively.
Because the impedance is minimized at fS, this frequency is often referred to as sim-
ply the resonance frequency, while fP is the antiresonance, as the motional imped-
ance is maximum at this frequency.
The link between the circuital BVD model and the physics of its elements
depends on the transduction mechanism. For example, the circuit elements of crys-
tal resonators and FBARs are [24]:

ε⋅A
C0 = (3.16)
t

Cm 8 ⋅ keff
2

= 2 (3.17)
C0 N ⋅ π2

v
Lm = 3 2
(3.18)
64f εAkeff
S
3.3 Equivalent-Circuit Models 85

ηε
Rm = 2
(3.19)
16f S ρAvkeff

where A is the area of the parallel-plate capacitor; , t, ρ, and v are the absolute
permittivity, thickness, density, and speed sound of the piezoelectric material,
respectively; N is the acoustic mode (N = 1, 3, 5, …), and η is the acoustic viscosity
~ ~
related to the imaginary part of the wave vector κ = k r + j ⋅ k i by:

~ ηω ⎛ ω⎞
ki = ×⎜ ⎟ (3.20)
2 ρv 2
⎝ v⎠

We carried out a similar analysis in Section 3.2.1 to find the RLC equivalent-cir-
cuit parameters of MEMS and NEMS resonators. Through (3.16) to (3.20), we can
infer the physical constants of the process from the BVD circuit elements and curve
fitting to experimentally obtained data. Process characterization based on this
approach is explained in Chapter 5.
The modified Butterworth-Van-Dyke (MBVD) model allows us to obtain a
more realistic representation of FBARs and acoustic resonators. The MBVD
accounts for dielectric and ohmic losses in the piezoelectric material and the trans-
mission line by using resistances RP and RS, respectively [25]. The MBVD circuit is
depicted in Figure 3.18, which results from adding RP and RS to the BVD circuit. In
the parallel-plate capacitance arm, RP is serially added to C0, whereas the transmis-
sion lines connecting the resonator are split in two components of resistance RS/2,
each one corresponding to the bottom and top electrode line losses.
The technological availability has made possible the application of FBAR,
MEMS, and NEMS resonators in the field of radio frequency circuits to build com-
petitive-performance blocks like filters, mixers, and oscillators. In this way, they are
designed to operate at very and ultrahigh frequencies (VHF, UHF) and microwave
bands. Typically, resonators are connected to other circuits and supported on sili-
con or SOI substrates. In the microwave regime, however, the substrate plays a
major role in the performance of the circuit. Due to the high frequency, parasitic
effects like ohmic losses and reactances become important in the working band [26].
These effects cause signal drifting from the RF path to the substrate, thus degrading
the performance of the circuit.

Bottom Top
electrode Lm Rm electrode
R S /2 Cm R S /2
(pad) (pad)

Rp C0

Line Resonator Line

Figure 3.18 The modified Butterworth-Van-Dyke model: the circuit adds ohmic losses of the
acoustic layer and the contact lines.
86 Design and Modeling of Micro- and Nanoresonators

Three main sources of signal loss in microwave resonator circuits have been
identified: conduction losses in the metal lines, dielectric losses in the substrate, and
radiation losses. Conduction and radiation losses can be minimized by proper
design of the RF conducting path, whereas high-resistivity substrates aid reducing
the dielectric losses [27]. Depending on the frequency regime, a proper model has to
be adopted to accurately describe these losses. This model is then employed to
de-embed the resonator response from the parasitic effects of the circuit, which is
useful in final-application circuit design. To exemplify this subject, we study next
the equivalent-circuit model of an FBAR embedded in a coplanar transmission line
and supported on a silicon substrate.

3.3.3 Case Study: FBAR Process and Modeling

The FBAR of our case study is fabricated on top of a SiO2 layer and a 500-μm-thick
silicon (Si) substrate. This substrate is a 1015 cm3-boron-doped low-resistivity Si
wafer. The FBAR is comprised by bottom and top titanium-platinum (Ti/Pt) elec-
trodes with thicknesses of 30 and 150 nm, a 700-nm-thick aluminum nitride (AlN)
layer. Front-side RIE-assisted micromachining releases the resonator. With this
configuration, we expect to find the resonance frequency in the 2-GHz band.
Due to characterization purposes, the resonator is embedded in a coplanar
transmission line with ground-to-signal-to-ground (GSG) connections to the FBAR.
Although the isolating SiO2 is between the signal line and the substrate, part of the
RF signal energy diverts to the substrate through the oxide, thus promoting dielec-
tric losses. In the 2-GHz regime, the SiO2-Si substrate can be modeled by resistive
and reactive elements [28], as it represents the drawing of Figure 3.19. The oxide
capacitance Cox depends on the thickness, area of the electric contact, and dielectric
constant, which is determined by the fabrication process. The resistance Rsub and
capacitance Csub values depend on the substrate thickness and resistivity, the latter
being primarily influenced by the doping characteristics.
Assuming that the FBAR is accurately modeled by the MBVD elements, the
equivalent circuit of the system is the superposition of the resonator’s MBVD, trans-
mission line, contact pads, and substrate models. Figure 3.20 shows the cross-sec-

Electric Electric
contact contact

C OX SiO 2 C OX

C sub Rsub
C sub Rsub
Substrate (Si)

Figure 3.19 Equivalent-circuit model of substrates at the microwave regime.

3.3 Equivalent-Circuit Models 87

Ti/Pt

PAD Lm
Rp PAD
AIN
Rm
L S /2 R S /2 C0 R S /2 L S /2
Cm
Ti/Pt Ti/Pt

C OX SiO 2 SiO 2 C OX

Air gap

C sub Rsub C sub Rsub

Substrate (Si)

Figure 3.20 Equivalent circuit model of FBAR, including substrate loss and reactance elements
(the FBAR is embedded in a coplanar transmission line).

tion of the process and the model elements. In addition to the MBVD elements—Rm,
Lm, Cm, C0, Rp, and Rs—and substrate elements—Rsub, Csub, and Cox—the model
includes the line inductance Ls, which describes the parasitic inductance becoming
of significance in long transmission lines [29].
Figure 3.21 shows the circuit, with the duty FBAR area in the center of the
image, the long transmission lines connecting the electrodes to the contact pads at
both the right and left side of the FBAR, and the substrate. The circuit elements are
annotated in the picture to illustrate the model. The width and length of the trans-
mission line are 70 μm and 150 μm, respectively, which gives theoretical Rs/2 and
Ls/2 values of 1Ω and 60 fH, respectively. Calculation of Cox for a 400-nm-thick
2
thermal-SiO2 layer with pad area of 70 × 70 μm leads to a theoretical value of 420
fF. Concerning the 500-μm-thick Si substrate, theoretical calculations of Rsub and
Csub result in values found in the range of 4,500Ω and 1 fF, respectively.

Figure 3.21 FBAR embedded in transmission line: the resonator, contact pads, and substrate loss
and reactance elements are signaled.
88 Design and Modeling of Micro- and Nanoresonators

These calculations give a general idea of the magnitude orders of the circuit ele-
ments, and they are useful to define the initial conditions of parameter extraction
algorithms, as we study in Chapter 5.

3.4 Finite Element Modeling (FEM)

Finite-element modeling (FEM) is a vast field that merits full coverage in a complete
book. The FEM analysis field spans from the formulation of numerical analysis and
computational efficiency to the operational aspects of FEM commercial tools [30,
31]. Instead of going in depth on these topics, this section introduces general ideas
about concepts, procedures, and working flow of FEM analysis. Computer-aided
FEM is a powerful numerical-analysis tool that allows accurate prediction of the
static and dynamic responses of a multiple-domain physical system. FEM analysis
reproduces the geometry and forces interaction of complex systems whose analyti-
cal formulation is unfeasible. Starting from the structural model of the system, FEM
analysis couples the structural physics with electrostatic, magnetostatic, piezoelec-
tric, thermal, optic, fluidic, and electromagnetic domains, among others (Figure
3.22). A number of commercial software tools are available in the market for FEM
analysis, like ANSYS (Ansys Inc., Canonsburg, Pennsylvania) [32], Coventor [4],
IDEAS [33], and COMSOL [34], to mention some of the most popular.
The FEM of FBAR, MEMS, or NEMS resonators is a system constituted of the
following components:

1. The resonator itself;

2. The environment—air, vacuum;
3. The degrees of freedom or boundary conditions;
4. The excitation sources—atmospheric or relative pressure, gravity or another
inertial force, mechanical acceleration, fluidic pressure, magnetic field, elec-
tric potential, and so on.

Electrostatic

Piezoelectric Optics

Structural

Fluidics Electromagnetic

Thermal

Figure 3.22 Physical domains in MEMS and NEMS finite element analysis.
3.4 Finite Element Modeling (FEM) 89

As its name suggests, the fundamental component of an FEM is the element: the
model geometry is structured by a group of elements, typically thousands of them,
in the same way as bricks construct the shape and structure of a building. Being con-
nected among them, such elements receive from and transmit to the rest of elements
the forces that they are the object of from external excitation sources. When model-
ing the complete system, the FEM analysis software implements a set of equations
pertinent to the physical domains involved in the interaction. These equations are
applied and solved at the element level, and the results are stored in the system mem-
ory and used as an input parameter for the next element to be analyzed. In the end,
all these results are scaled up and superposed to evaluate the global system response,
whose quality will depend on the accuracy, correction, and complexity of the built
model.
FEM analysis is a process that involves a set of necessary actions whose
sequence is observed in the flow of Figure 3.23. First, the system characteristics and
physical interactions between the system components are defined (this is not neces-
sarily done with the FEM tool). From here on, the model of the system is built, by
using a compiler or preprocessor (the /PREP primitive of ANSYS, for example).
First, we define the materials and element types; then we generate the geometry; and
next we mesh this model with a number of elements that can be specified by the
designer. Once the model is made and meshed, the degrees of freedom (DOF) or
boundary conditions of the system are applied to it. This establishes the physical ref-
erence frame, and it involves initial force definition, initial charging, and clamping
of the structure, among others. Once this task is completed, we proceed to solve the
model according to the simulation settings defined by the user. Structural, modal,
and harmonic analyses can be carried out, with or without initial stresses or loads
included in the analysis. Linear and nonlinear analysis options are also available in
commercial tools. When the analysis finishes, the postprocessing system of the FEM

Preprocessor

Solver

Post-
processor

Figure 3.23 Finite element modeling flow.

90 Design and Modeling of Micro- and Nanoresonators

tool allows evaluation of the simulation results, as they are provided in the form of
graphs, charts, tables, or 3D plots. Specific primitives of the program are invoked to
enter in postprocessing mode (e.g., /POST in ANSYS).
Generally speaking, the analyzer solves the second-order differential equations
relating the mechanical response to an external force applied to the system, which
was already defined by (3.1) and reproduced here for convenience:

&& + Dx& + kx = F
mx (3.21)

where m is the mass, k is the spring constant, D is the damping, and F is the applied
force. It should be noted that F 0 in the modal analysis because it aims to find the
eigenvalues of the system when no external forces are applied. The end products of
the FEM analysis are the material deformations, stress, induced currents or electro-
magnetic fields, current flows, resonance frequencies, frequency responses, and
many other results. The nature of applied forces will determine the implemented
models, elements, and physical domain coupling. When the system response of a
model defined in a certain physical domain to forces applied forces in another one is
studied, we speak in terms of coupled-field response. Modern tools implement the
design options to cover multiple-domain physical interaction.

3.4.1 Building the Model

Building the model starts from the concept of the desired FEM of the resonator. This
includes taking a priori decisions with regard to four topics: (1) geometry and mate-
rials of the device; (2) type of analysis; (3) type of elements; and (4) degrees of free-
dom (DOF) of the structure. Creation of the model geometry and of the material
and element database is possible due to different tools available in commercial sim-
ulators, which are useful for building complex geometries, conics, and
micromachined-like 3D structures, for example. In general, the more realistic the
geometry of the model, the more accurate the simulation results will be obtained.
However, simplified geometries and 2D models are often enough to get accu-
rate results. This mainly depends on the DOFs and symmetry, thus limiting the
model generation and meshing to the active resonator volume. Also, the chosen ele-
ment types have to be prepared to couple the physical domains involved in the anal-
ysis type. Experienced designers let a program routine select the element type in the
function of the analysis when different simulations are to be run with the same
model. Figure 3.24 shows the conceptual drawing of a MEMS cantilever that has
been prepared to define the geometry and DOF in order to build its FEM. The spa-
tial orientation, dimensions, shape, and materials of the cantilever, the clamping
area (UX, UY, UZ = 0), and electrostatic configuration (φT = 1V, φB = 0V) have been
chosen in this example.
Once the geometry is built and the element and material types are chosen, the
model is meshed with a number of elements that can be specified by the designer.
Again, many options exist to complete this task, including manual, automatic,
mapped, or free meshing. The designer has to be aware of the compatibility between
the element and analysis types, which vary with the simulation software and depend
on the shape, available DOFs, and number of nodes of the element, among other
3.4 Finite Element Modeling (FEM) 91

Top electrode
(Pt, 180 nm) φ = 1V

Acoustic layer
Ux, U y, U z = 0
(AIN, 1 μm)

x y

Bottom electrode
(Pt, 180 nm) φ = 1V

Figure 3.24 Conceptual drawing of the FEM model of a MEMS cantilever resonator depicting the
geometry and degrees of freedom.

factors. Figure 3.25 shows the 2D model of a meshed system consisting of a reso-
nant silicon-MEMS cantilever, a magnet inserted in the MEMS, the air surrounding
the device, and the space region where the magnetic field lines are parallel to the
infinite boundary. The different material types are differentiated by their tone: the
lightest tone is the infinite region, mid-gray is the air around the MEMS, the gray in
the central region represents the cantilever, and the darker tones are the elements of
the magnet.
Accuracy of the simulation is determined by the number of elements and their
aspect ratio. The aspect ratio definition depends on the element shape and model,
although, as a general rule, it is the quotient of the largest over the shortest element

Figure 3.25 FEM meshing: MEMS cantilever resonator (light gray in the center) with inserted
magnet (dark), air, and infinite boundary elements around them.
92 Design and Modeling of Micro- and Nanoresonators

dimension in the mesh. Sometimes, high-aspect ratio elements are created, which
degrades the numerical approximations assumed for regular elements. This will
alter the solution superposition and the global response of the system, which also
depends on the relative element size with respect to the structure dimensions. Thus,
denser meshing leads to better superposition and model quality. It is a good design
practice to perform a mesh testing to ensure the quality and reliability of the simula-
tions. The example of Figure 3.26 shows the influence of the aspect ratio (AR) and
element size in the resonance frequency of an FBAR. Figure 3.26(a) shows the simu-
lated results of resonance frequency against element size (the aspect ratio is constant
and equal to 11). The response of Figure 3.26(b) evidences a dependence on the
aspect ratio of elements. Therefore, we obtain different aspect ratios by changing
the element size. Although slight differences arise, they can be enough to degrade
the performance of the prediction, especially when the resonator is designed for
high-sensitivity applications.

3.4.2 Structural, Modal, and Harmonic Analyses

Once the model is meshed and the DOFs are defined, the corresponding analyses of
the MEMS are carried out. Structural, modal, and harmonic are the most-performed

(a)

(b)
Figure 3.26 Mesh study to analyze the frequency stability of the model as a function of (a) con-
stant aspect ratio (AR) equal to 11; and (b) fixed element thickness of 87 nm and variable xy
dimensions to obtain a variable AR.
3.4 Finite Element Modeling (FEM) 93

analyses for studying FBAR, MEMS, and NEMS resonators, although many other
analysis options exist, depending on the physics of the resonator transduction.
Structural analysis aims to determine stresses, strains, deformations, and deflec-
tions of the resonator due to static force loading like gravity, pressure, inertial accel-
eration, or another constant force applied to the nodes, the elements, or one of the
structure surfaces. Figure 3.27 shows the contour plot of a MEMS cantilever with
silicon mass and a magnet inserted in the mass. The picture compares the initial and
equilibrium positions of the system after considering the effect of gravity force (+1g)
applied to the structure. The calculated deflection at the maximum displacement
edge of the mass is about 168 μm.
Modal analysis extracts the natural frequencies (eigenvalues) of the MEMS
device. In this analysis, the external forces are set to zero values (F 0), as stated
before. Thus, the analysis searches for modes existing between given frequency
ranges. Otherwise, the simulator runs freely to find the first N eigenmodes of reso-
nance, which is a user option. Modal analysis does not quantify the deformation or
vibration amplitude of the structure at the so-found natural frequencies. Instead, it
solves (3.21) to calculate the node displacement, thus finding the frequencies of
maximum displacements. Commercial software can produce contour plots repre-
senting the shape of the studied modes. Contour plots of two eigenmodes of a sili-
con-FBAR resonant accelerometer are seen in Figure 3.28. While the geometry of
the silicon mass shown in Figure 3.28(a) mainly determines the 3-kHz mode, the
800-kHz mode is attributed to the fundamental frequency of the sensing FBAR res-
onator [35]. In the latter case, the mass remains essentially immobile [Figure
3.28(b), detail of the resonator shape in the inset].
Harmonic analysis studies the dynamic response of the resonator to an excita-
tion signal (F ≠ 0), between user-defined starting and ending frequencies. Thus, it
assumes that one or more modes with significant amplitude are found in the span of
interest. It gives understanding of the relative amplitude of the different modes
(something that is impossible to do with modal analysis) and allows detecting

Figure 3.27 Structural analysis results: static deflection of the cantilever due to gravity force.
94 Design and Modeling of Micro- and Nanoresonators

(a)

(b)
Figure 3.28 Modal FEM analysis of a Si-FBAR accelerometer: (a) 3-kHz mode shape of the funda-
mental value of the structure (it corresponds to the Si mass eigenfrequency); and (b) 800-kHz
mode corresponding to the sensing resonator (detailed in the inset).

small-amplitude modes that may pass unseen by the user when performing the
modal analysis. Output data of harmonic analysis include tables relating amplitude
against frequency or time, and nodal solutions used in the generation of mode-con-
tour plots.

3.4.3 Coupled-Domain Analysis

A coupled-domain or coupled-field analysis is performed when the designer wants
to evaluate the influence of external forces in one domain in the static or dynamic
response in other domains. For example, the magnetic force attracting the magnet
3.4 Finite Element Modeling (FEM) 95

(a)

(b)
Figure 3.29 Magnetic-structural coupled-field FEM analysis: (a) magnetic flux lines; and (b) mag-
netic flux density.

inserted in the MEMS shown in the previous example generates a structural deflec-
tion of the cantilever. The magnitude of the deflection can be predicted if the
amount of magnetic force is estimated in advance. This result is employed in struc-
tural design of the MEMS. In this example, the first question is, how we can deter-
mine the amount of force to apply to the structure? This case requires coupling the
magnetic and the structural domains in order to design the structure and to know
the magnetic force arising between the magnetic components.
There are two different strategies to perform a coupled-field analysis:

• Multifield solver;
• Complete transient simulation.

ANSYS Multiphysics solves the multifield strategy by defining two or more sets
of single-domain elements, grouping the element types into physical fields, and
selecting the quantities to be passed between the fields and the order in which such
96 Design and Modeling of Micro- and Nanoresonators

fields are to be solved. Additional commands control the loading of simulation vec-
tors to proceed with the system solving [36].
The second strategy proposes performing independent simulations, each one
with its own inputs and element types. Following with the example of the magnetic
MEMS-magnet system, the model is meshed with magnetic elements, and the mag-
netostatic solution is calculated. The results are then saved and stored in the simula-
tion database to be used as inputs for the second structural analysis. Next, the
model is remeshed with structural elements, and the solver executed. Several itera-
tions may be required to design a system with desired values, which is typically done
by a loop. Figure 3.29 shows the magnetic flux lines and magnetic flux density after
magnetostatic analysis of the magnetic sensor 2D model. Pursuant to this analysis,
the magnetic force arising between the magnet and the excitation source is obtained,
which is used in the structural analysis to excite and calculate the displacement of
the MEMS-magnet ensemble.

3.4.4 Case Study: Modal and Harmonic Analysis of a Resonant Mass Sensor
We now examine the model of an FBAR-based resonant mass sensor built in
ANSYS. The resonator is made of Pt electrodes and an AlN acoustic layer. Figure
3.30 depicts a section of the model meshing, with the electrodes and the AlN being
tone-differentiated by their material types. As observed in the figure, the meshing is
uniform, as the model has a total of 15,000 elements with a maximum aspect ratio
of 5.7.
The eight-node hexahedral SOLID5 and SOLID45 options were selected for the
element geometry, and the lateral dimensions (xy) of all elements were of 1 μm. The
ANSYS element types SOLID5 and SOLID45 are served to implement the piezo-
electric and electrode materials. SOLID5 has a 3D magnetic, thermal, electric,
piezoelectric, and structural field capability with limited coupling between the
fields. The element thus has eight nodes with up to six degrees of freedom at each
node and hexahedral or prism geometry options. When used in structural and

Figure 3.30 Section of the FBAR sensor meshing built-in ANSYS.

3.4 Finite Element Modeling (FEM) 97

piezoelectric analyses, SOLID5 has large deflection and stress stiffening capabilities.
SOLID45 is used for the 3D modeling of solid structures and has geometry similar
to that of SOLID5. However, the element is defined by eight nodes having only
three degrees of freedom at each node: translations in the nodal x, y, and z direc-
tions. In this model, coupled-field and translational DOFs are defined for the
SOLID45 elements, whereas only translational DOFs are defined for the SOLID5
elements. According to this meshing, the model comprises 18,200 nodes.
The material properties used in the simulations are those presented in Table 3.1.
The damping D equal to 110−3 is defined for the AlN material type. The assigned
value of D uses the ANSYS definition of D 1/(2Q), where the Q factor is chosen to
have a value similar to that of FBARs available in the market and literature (Q ~
500). The resonator is a clamped-clamped beam that has lateral dimensions of 50 ×
70 μm2, and electrode and AlN thicknesses of 180 nm and 1,000 nm, respectively.
The DOFs include displacement (Ux,y,z) and voltage (VOLT) boundary conditions
defined at all the nodes located at the clamping lateral-wall surfaces and the elec-
trodes of the FBAR. Thus, a 1-V voltage is applied to the top electrode, while the
bottom electrode is grounded to 0V, respectively. The structural DOFs translate
into mechanical displacements Ux, Uy, and Uz equal to 0 at the corresponding nodes.
With this configuration, the DOF matrix has a size of 7,344 nodes; 2,142 of them
defined for Ux,y,z and 5,202 devoted for VOLT [37].
ANSYS performs modal and harmonic analyses on this model around the
design 2-GHz frequency band. The user defines the number of extracted and
expanded modes in the frequency span. Figure 3.31 shows the contour plot of the
2.22-GHz longitudinal mode of the FBAR, which is seen at its perspective and side
views. In this “breathing-like” mode, the nodal displacements are bigger at the reso-
nator’s center, whereas they are negligible near the clamping zones.
Parametric harmonic analysis was carried out to observe the impact of the
material properties on the resonance frequency of the longitudinal mode. In Figure
3.32 the frequency response between 2.35 GHz and 2.55 GHz is plotted. There,
major peaks near 2.40 GHz and 2.48 GHz correspond to the longitudinal-mode res-
onance and antiresonance frequencies. If we look into the frequency range close to
the resonance peak, we find that frequency and mode shaping differences arise
between the three curves, due to the modified material properties. Besides the reso-

Table 3.1 Material Properties Used in the

FEM Simulations of the FBAR Sensor
AlN Pt (Electrodes)
3
ρ [kg/m ] 3,000 21,450
v 0.35 0.38
E
c 33 [GPa] 200 144
εE33 [GPa] 8.81 —
2
e13 [C/m ] −0.6 —
2
e23 [C/m ] −0.6 —
2
e33 [C/m ] 1.46 —
2
e52 [C/m ] −0.48 —
2
e61 [C/m ] −0.48 —
98 Design and Modeling of Micro- and Nanoresonators

(a)

(b)
Figure 3.31 Contour plot of the longitudinal mode of the FBAR sensor (modal analysis): (a) per-
spective view and (b) side view.

nance frequency shifting, minor modes shift to a different value, diminish or aug-
ment their relative amplitudes, superpose to other modes, or simply disappear.
Harmonic and modal FEM analysis finds application in resonator chart-mode
building. This is employed in full-mode resonator design where the evolution of lon-
gitudinal and transversal modes against dimensions of the resonator is examined
through parametric studies. As another subject of interest concerning FEM, com-
plete knowledge of the material properties is crucial for accurate modeling of the
absolute harmonic response of the FBAR. As manufactured devices have pro-
cess-dependent material properties, FEM simulations can also be served to extract
their values from experimental device characterization.
3.5 Summary 99

Dx0um
Dx10um
1E-11 Dx20um

Admitance (Y)
1E-12

1E-13

1E-14

2.35 2.40 2.45 2.50 2.55

Frequency (GHz)
(a)

Dx0um
Dx10um
Dx20um
1E-11
Admitance (Y)

1E-12
2.395 2.400 2.405 2.410 2.415
Frequency (GHz)
(b)
Figure 3.32 Parametric analysis of the FBAR frequency response after harmonic analysis: (a)
2.35-GHz to 2.55-GHz frequency span, and (b) zoomed in view around 2.4 GHz (the resonance
frequency and mode shaping differences are observed at this scale).

3.5 Summary

The design of resonators passes through different modeling approaches. Therefore,

we consider analytical models based on the physics of the resonator, electrome-
chanical models, equivalent-circuit parameter models, and finite element models
based on numerical methods and domain coupling analysis. In the end, these models
serve to perform the physical layout design and, on the other hand, as characteriza-
tion tools when fabricated devices are available.
Electromechanical transformer models offer the conceptual basis to compre-
hensive understanding of the resonator operation mechanisms. They analyze both
the electrical and mechanical domains by using a unified approach, and they can be
interconnected to model more complex resonator-based systems.
Equivalent-circuit modeling is, in general, an electrical-domain representation
of the resonator. Equivalent-circuit parameters characterize the physics and fabrica-
tion process of the resonator, and they represent an electrical abstraction of the res-
onator motional response. At the system level, electrical equivalent-circuit models
100 Design and Modeling of Micro- and Nanoresonators

aid the fitting and parameter extraction tasks performed for circuit design purposes.
The technique is currently employed for designing FBAR, MEMS, and NEMS
resonator-based integrated circuits.
Last but not least, we have gone into the physical implementation details of
FEM analysis. Besides being a powerful prediction tool, FEM simulations with
experimental characterization data are served to extract the process-dependent
material constants. The sophistication level of commercial FEM design software
makes the field of FEM analysis an enriching experience for designers and process
engineers. The available tools support the execution of multidomain analysis of
virtually any micro- or nanostructure.
While the design and modeling processes conclude with the manufacturing of
devices, one can say that the modeling cycle always continues. Existing designs and
processes need permanent refinement. Thus, modeling provides us with the valuable
information needed for iterative device improvement, from the abstract to the
physical levels.

References

[1] Fedder, G. K., “Structured Design of Integrated MEMS,” Proc. IEEE Intl. Conf. MEMS
1999, Orlando, FL, January 17–21, 1999, pp. 1–8.
[2] Fan, Z., et al., “Structured Synthesis of MEMS Using Evolutionary Approaches,” Appl.
Soft Comp., Vol. 8, 2007, pp. 579–589.
[3] Cadence Design Systems, Inc., http://www.cadence.com.
[4] Coventor, Inc., http://www.coventor.com/.
[5] The MathWorks, Inc., http://www.mathworks.com/.
[6] Pederson, D., “Simulation Program with Integrated Circuits Emphasis (SPICE),” University
of California–Berkeley, 1975.
[7] Kundert, K., and J. White, “Spectre Circuit Simulator,” Cadence Design Systems.
[8] IEEE Std. 1364-2001, “IEEE Standard Verilog Hardware Description Language,” Institute
of Electrical and Electronics Engineers, New York, 2001.
[9] IEEE Std. 1076.1-1999, “VHSIC Hardware Description Language Analog and Mixed-Sig-
nal,” Institute of Electrical and Electronics Engineers, New York, 1999.
[10] Bode, H. W., “Bode Plots,” AT & T Bell Labs, circa 1930.
[11] Evans, W. R., “The Root Locus Method,” originally published in AIEE Transactions,
1948. Republished in Control System Dynamics, New York: McGraw-Hill, 1954.
[12] Polderman, J. W., and J. C. Willems, Introduction to Mathematical Systems Theory: A
Behavioral Approach, New York: Springer, 1998.
[13] Tipler, P., Physics for Scientists and Engineers: Vol. 1, 4th ed., New York: W. H. Freeman
& Co., 1998.
[14] Bannon, F. D., J. R. Clark, and C. T.-C. Nguyen, “High-Q HF Microelectromechanical Fil-
ters,” IEEE J. Solid-State Circuits, Vol. 35, 2000, pp. 512–526.
[15] Mason, W. P., Electromechanical Transducers and Wave Filters, Princeton, NJ: Van
Nostrand, 1948.
[16] Krimholtz, R., D. A. Leedom, and G. L. Matthaei, “New Equivalent Circuit for Elementary
Piezoelectric Transducers,” Electron. Lett., Vol. 6, 1970, pp. 398–399.
[17] Lakin, K. M., G. R. Kline, and K. T. McCarron, “Development of Miniature Filters for
Wireless Applications,” IEEE Trans. on Microw. Theory Tech., Vol. 43, 1995,
pp. 2933–2939.
3.5 Summary 101

[18] Heaviside, O., “Electromagnetic Induction and Its Propagation,” The Electrician, 1885,
1886, and 1887.
[19] Wolfram Research, Inc., http://www.wolfram.com/.
[20] Aigner, R., “Volume Manufacturing of BAW-Filters in a CMOS Fab,” Proc. Second Inter-
national Symposium on Acoustic Wave Devices for Future Mobile Communication Sys-
tems, Chiba, Japan, March 3–5, 2004, pp. 129–134.
[21] Razhavi, B., Design of Analog CMOS Integrated Circuits, New York: McGraw-Hill, 2001.
[22] Davis, W. A., and K. K. Agarwal, Radio Frequency Circuit Design, New York: John Wiley
& Sons, 2001.
[23] Lin, Y. W., et al., “Series-Resonant VHF Micromechanical Resonator Reference Oscilla-
tors,” IEEE J. Solid State Circuits, Vol. 39, 2004, pp. 2477–2491.
[24] Kim, K.-W., et al., “Resonator Size Effects on the TFBAR Ladder Filter Performance,”
IEEE Microw. Wirel. Compon. Lett., Vol. 13, 2003, pp. 335–337.
[25] Larson III, J., et al., “Modified Butterworth-Van Dyke Circuit for FBAR Resonators and
Automated Measurement System,” Proc. IEEE Intl. Ultrason. Symp. 2000, San Juan,
Puerto Rico, October 22–25, 2000, pp. 863–868.
[26] Wittstruck, R. H., et al., “Properties of Transducers and Substrates for High Frequency
Resonators and Sensors,” J. Acoust. Soc. Am., Vol. 118, 2005, pp. 1414–1423.
[27] Polyakov, A., et al., “High-Resistivity Polycrystalline Silicon as RF Substrate in
Wafer-Level Packaging,” Electron. Lett., Vol. 41, 2005, pp. 100–101.
[28] Hasegawa, H., M. Furukawa, and H. Yanai, “Properties of Microstrip Line on Si-SiO2 Sys-
tem,” IEEE Trans. on Microw. Theory Tech., Vol. 19, 2001, pp. 869–881.
[29] Campanella, H., et al., “Automated On-Wafer Extraction of Equivalent-Circuit Parameters
in Thin-Film Bulk Acoustic Wave Resonators (FBAR) and Substrate,” Microwave Optical
Tech. Lett., Vol. 50, 2008, pp. 4–7.
[30] Chandrupatla, R., and A. D. Belegundu, Introduction to Finite Elements in Engineering,
2nd ed., Upper Saddle River, NJ: Prentice-Hall, 1997.
[31] Zienkiewicz, O. C., and Y. K. Cheung, The Finite Element Method in Structural and Con-
tinuum Mechanics, New York: McGraw-Hill, 1967.
[32] ANSYS, Inc., http://www.ansys.com/.
[33] IDEAS Ltd., http://www.ideas-eng.com.
[34] COMSOL, Inc., http://www.comsol.com.
[35] Campanella, H., et al., “Accelerometer Based on Thin-Film Bulk Acoustic Wave Resona-
tors,” Proc. IEEE Intl. Ultrason. Symp. 2007, New York, October 28–31, 2007,
pp. 1148–1151.
[36] ANSYS Inc. “Low Frequency Electromagnetic,” Training manual, 2004.
[37] Campanella, H., et al., “Analytical and Finite-Element Modeling of a Localized-Mass Sen-
sor Based on Thin-Film Bulk Acoustic Resonators (FBAR),” IEEE Sensors J., Vol. 9, 2009,
pp. 892–901.
 CHAPTER 4

Fabrication Techniques
There exist a great variety of MEMS, NEMS, and FBAR processes and fabrication
techniques. The fundamentals of some of those techniques were already discussed in
Chapters 1 and 2. Next, we take the particular case of FBAR devices to illustrate a
complete microfabrication process. Although FBARs are just one example of
microresonators, the techniques presented in this chapter are also applicable to the
general case of MEMS and NEMS resonators.
The main processes involved in the fabrication of FBARs are the piezoelectric
layer and electrode’s deposition and patterning, and the micromachining technol-
ogy for the resonator releasing. Nowadays, these processes are implemented by
using a variety of technologies to obtain high-quality factor devices.
The chapter discusses different realizations of the FBAR fabrication process,
putting special emphasis on the aluminum nitride (AlN) layer deposition technology
and material characterization. At the same time, it explains compatibility issues and
the actions to solve them. Pursuant to the compatibility development, fully released
devices can be fabricated and characterized. By using scanning-electron-microscope
(SEM), interferometer, and confocal inspection techniques, the FBAR’s structural
analysis is carried out, as we explain at the end of the chapter.

4.1 Process Overview

FBARs are a sandwiched piezoelectric membrane, typically made of aluminum

nitride (AlN), zinc oxide (ZnO), or lead zirconate titanate (PZT). Membranes for
current 2.4-GHz devices have a thickness of about 1 μm. Sputtering or epitaxial are
current techniques for deposition, which is performed on metallic layers of few hun-
dred nanometers.
FBARs are typically fabricated on a silicon (Si) substrate and released by surface
or bulk micromachining techniques to obtain a free-moving device with high-qual-
ity factors. Surface micromachining employs a sacrificial layer made of either a
metallic or a dielectric material. An etchant solution reacts with the sacrificial layer,
etching it. At the end of the process, the resonator exhibits its characteristic cavity
once the sacrificial layer is removed. Because the sacrificial layer is a thin film of a
few micrometers, surface micromachining produces quasi-2D devices.
On the other hand, bulk micromachining processes remove a considerable vol-
ume of the substrate, thus making the “3D” structure more fragile. The substrate is
usually Si (100), although high-resistivity substrates are implemented in RF applica-
tions to diminish the signal coupling [1]. Two main techniques accomplish this pro-
cess: dry and wet etching. Dry etching is based on RIE, DRIE, or ICP. The process is

103
104 Fabrication Techniques

carried out from either the front or the back side of the wafer. At its time, potassium
hydroxide (KOH) or similar solutions etch Si (100) in an anisotropic way, by
immersing the wafer in the solution. The schemas of Figure 4.1 depict side views of
FBAR processes using these technologies.
Before the bottom electrode deposition, the first step consists of depositing a
passivation layer on top of the Si substrate. The passivation provides a window for

Electrode

Piezoelectric
Electrode

Air gap

Substrate (Si)

(a)

Electrode

Piezoelectric
Electrode

Air gap

Substrate (Si)

(b)

Electrode

Piezoelectric
Electrode

Air gap

(c)
Figure 4.1 Micromachining technologies of FBAR processes: (a) surface, (b) front-side, and (c)
back-side bulk micromachining.
4.2 FBAR Fabrication Techniques 105

etching and reduces the electrical coupling between the FBAR and the substrate,
thus diminishing RF losses. Typical implementations of this layer involve silicon
oxide (SiO2) or silicon nitride (Si3N4) thin films with thicknesses of hundreds of
nanometers.
Top and bottom electrodes are made of metallic materials of the same or differ-
ent thicknesses, which is a design choice. Metals compatible with the crystallogra-
phy and piezoelectric layer deposition are preferred to implement the bottom
electrode. Platinum (Pt), molybdenum (Mo), tungsten (W), and chromium (Cr) are
some examples [2]. The top electrode has fewer compatibility requirements. Thus, it
can be fabricated using aluminum (Al), cupper (Cu), one of the metals previously
mentioned, or another material compatible with standard integrated-circuit pro-
cesses. The resonator contact pads may follow the current trends of standard
CMOS processes, so they can be implemented with gold (Au) or Al. Some metals,
like Pt, require an extra metal layer to allow proper adhesion to the substrate. Com-
monly used for the adhesion layer are titanium (Ti) or chromium (Cr), typically
deposited with thicknesses of about 100 nm.
The three processes can be implemented through standard microfabrication
clean room facilities and equipment. To further illustrate, Section 10.2 revises the
step-by-step fabrication of a real device. In the following sections, we study the con-
cepts, physical principles, fabrication equipment, materials, and chemical products
involved in FBAR manufacturing. The study comprises fundamental definitions on
oxidation, metallization, sputtering deposition, and micromachining techniques
regarding FBAR and MEMS fabrication.

4.2 FBAR Fabrication Techniques

This section takes the particular case of FBAR to illustrate fabrication techniques
that are common to MEMS and NEMS fabrication. Already in Chapter 1 we
defined the physical and chemical concepts explaining the techniques. Now, we dis-
cuss the application of some of those techniques to the particular case of FBAR
fabrication.

4.2.1 Oxidation of Silicon

Previously, in Section 1.3.2 the processing and applications of SiO2 were explained.
In the specific case of FBAR, oxidation will be used to passivate and interface the
substrate and the device through a dielectric SiO2 film. SiO2 grown at high tempera-
tures as previously described is also known as thermal oxide. However, SiO2 can
also be deposited on other materials (different from Si). In this case, chemical vapor
deposition (CVD) is usually employed to obtain a high-quality oxide. In the case of
FBAR implementations, either the thermal or the deposited SiO2 may have thick-
nesses of hundreds of nanometers. Another application of CVD-deposited SiO2 will
be discussed in Chapter 6. There, we will see how a SiO2 layer of appropriate thick-
ness may optimize the temperature coefficient factor (TCF) of FBARs implemented
with AlN.
106 Fabrication Techniques

4.2.2 Metallization and Piezoelectric Layer Deposition

As defined in Chapter 1, metallization is the formation of metal films for intercon-
nections, ohmic contacts, rectifying metal-semiconductor contacts, and protections.
Metallic thin films can be deposited on the surface of dielectric, conductor, or semi-
conductor materials. Vacuum evaporation (deposition of single element conduc-
tors, resistors, and dielectrics), sputtering, CVD, platting, and electroplating are
some of the employed techniques. The FBAR metallization allows the application of
the electric field through the acoustic layer, providing ohmic contacts for read-out
purposes. In FBAR processes, Pt, W, Al, Ti, Cr, Ir, and Mo may be deposited on top
of the SiO2 passivation layer by RF sputtering. Sputtering is the deposition of com-
pound materials and refractive metals through removal of the surface atoms or
molecular fragments from a solid cathode (target). By bombarding it with positive
ions from an inert gas (argon), removed atoms or molecular fragments deposit on
the substrate forming the thin film.
Now, we center the discussion on the acoustic, piezoelectric layer deposition.
The piezoelectric material is perhaps the layer that mostly determines the ultimate
quality of BAW and SAW devices. Throughout this book, we discuss the particular
case of AlN fabrication, due to its popularity, relatively simple processing, and
CMOS compatibility. As well as metallization, there are a variety of AlN deposition
methods, which are currently employed for FBAR and SAW fabrication, including
sputtering, epitaxial growth, and physical vapor deposition (PVD), among others
[3–5]. The crystallographic and piezoelectric quality of the AlN will strongly
depend on the implemented technique. In this context, sputtering succeeds on
achieving standard AlN quality with the benefits of low-temperature processes,
making it more attractive to CMOS-compatible processes.
An example setup of the sputtering system for AlN deposition is shown in Fig-
ure 4.2. Sputtering can be performed by applying DC, AC, or a combination of DC
and AC voltages. The voltage is applied between a cathode to which the target is
attached and an anode grounded to the chamber. Vacuum conditions exist when the
inert gas enters in the chamber, which is usually argon. Due to the potential differ-
ence between cathode and anode, positively charged argon ions are attracted and
accelerated toward the cathode. Thus, they collide with the target, conveniently
located near the cathode. As a consequence of the collision, different processes
occur:

• Ion reflection and returning of the same to the gas phase;

• Emission of secondary electrons;
• Deep ion penetration and implantation in the target material;
• Ion-induced mixing as well as diffusion;
• Most important: ion-energy transfer to atoms located in the target’s surface
and ejection of some of them (i.e., sputtering of target’s atoms).

If a combination of argon and another gas enter in the chamber (e.g., nitrogen),
a reaction occurs between the gas and ejected atoms, leading to reactive sputtering.
In this case, deposition of dielectric and compound materials using metallic targets
can be carried out. The AlN deposition process by reactive sputtering is illustrated
4.2 FBAR Fabrication Techniques 107

Matching
network

(a)

(b)
Figure 4.2 RF/magnetron sputtering for AlN deposition: (a) setup for AC/DC magnetron sputter-
ing; and (b) illustration of the reactive sputtering process, where the Ar+ ions impact the target
and the Al reacts with N2 to form AlN. (Source: F. Engelmark, 2002.)

in Figure 4.2(b). The control system injects the Ar/N2 gas in the chamber and, due to
the biasing and magnetron, the Ar+ ions accelerate toward the target. The ejected Al
atoms react with the reactive N2 atoms, thus forming the AlN deposited on the
substrate.
As pointed out in Section 2.3.2, process parameters have an influence on the
quality, grain size, and orientation of the AlN crystals. Gas concentration and flow,
pressure inside the sputtering chamber, DC bias, and AC power are the main vari-
ables [6]. The tilting angle and position of the wafer on the sample’s holder are pro-
cess parameters that also control the crystal orientation [7].
108 Fabrication Techniques

4.2.3 Surface-Micromachining-Based Process

Surface micromachining enables full releasing of fabricated FBARs in a fast and
clean way. In this section, we study the required steps and technologies to accom-
plish this purpose.
The simplified fabrication sequence of Figure 4.3 features a sacrificial layer,
implemented to release the FBAR. First, we deposit a passivation film of SiO2 or
Si3N4 layer on the Si substrate. When patterned, the thin film contributes to define
the region containing the sacrificial layer. This is deposited and patterned in the
next step of the process [Figure 4.3(b)]. An additional buffer layer, SiO2, for exam-
ple, may be deposited between the sacrificial layer and the substrate to reduce the
etching impact on the substrate during the micromachining process. Next, fabrica-
tion continues with a first metal deposited on the passivation layer through RF and
magnetron sputtering [Figure 4.3(c)]. The process requires high etching selectivity
between the sacrificial layer and the bottom electrode. Thus, we need to choose suit-
able metals according to this requirement and the etchant solution. Next, we sputter
the AlN and pattern it by wet etching in TMAH or a TMAH-based solution [Figure

Sacrificial
layer
Si3N4 /SiO2 Si3N4 /SiO2

Si Si
(a) (b)

Sacrificial Sacrificial
layer AIN layer
Bottom electrode Bottom electrode

Si Si
(c) (d)

Top electrode Sacrificial Top electrode

AIN layer AIN
Bottom electrode Bottom electrode
Air gap

Si Si
(e) (f)

Figure 4.3 Surface-micromachining-based FBAR processing: (a) buffer and mask layers deposition and
etching; (b) sacrificial layer deposition and patterning; (c) first electrode, (d) AlN, and (e) second electrode
deposition and patterning, respectively; and (f) etching of the sacrificial layer (in HF solution).
4.2 FBAR Fabrication Techniques 109

4.3(d)]. Then, we proceed with deposition and patterning of the top electrode,
according to the techniques already described for the bottom electrode [Figure
4.3(e)]. In the last step, immersing the wafer in the etchant removes the sacrificial
layer [Figure 4.3(f)]. The sacrificial layer can be a metallic—Ti or Al—or a dielectric
material—SiO2 or PSG. Popular etchants for sacrificial layer removal are HF
solutions. At this point, the device exhibits two air interfaces.

4.2.4 Bulk-Micromachining-Based Processes

4.2.4.1 Reactive-Ion Etching-Based Process
The first bulk-micromachining-based process implements front- or back-side RIE of
the Si substrate. In the front-side case, partial or total etching of Si is carried out,
while total etching of Si is performed from the back-side of the wafer. The simplified
fabrication sequence of Figure 4.4 depicts the cross-sectional schematic view of a
front-side RIE process. After passivation deposition and patterning, we deposit the
first metal layer by RF and magnetron sputtering [Figure 4.4(a)], defining the layout
of the bottom electrode. In the next step, the AlN layer is also deposited by RF/mag-
netron sputtering and etched according to the process described in the previous sec-
tion [Figure 4.4(b)]. After a second metallization and patterning of the resulting
metal layer, the top electrode is created and the device structure is completed [Figure
4.4(c)]. Finally, the front-side RIE completes the process by releasing the FBAR.
This achieved by using the etching window provided by the SiO2 layer [Figure
4.4(d)].
The process description and step sequence of the back-side RIE-based process is
similar to that of the previously described front-side process. However, the photo-
lithography of the etching window is done at the back side of the wafer. Although

AIN
Bottom electrode Si3N4 /SiO2 Bottom electrode

Si Si
(a) (b)

Top electrode Top electrode

AIN AIN
Bottom electrode Bottom electrode

Air gap

Si Si

(c) (d)

Figure 4.4 Front-side RIE process overview: (a) first electrode; (b) AlN; (c) second electrode deposition
and patterning, respectively; and (d) device releasing from the front side of the wafer (RIE).
110 Fabrication Techniques

the photomask may have the same features in both cases, double-sided alignment
capabilities should be provided for the back-side mask (strictly speaking, the fea-
tures may be different, since the front-side etching needs larger etching windows to
allow the etching plasma to attacking the substrate). Despite the layout design simi-
larities, different compatibility issues arise in front-side and back-side fabrication
processes.

4.2.4.2 Wet-Etching Process

Anisotropic wet etching of silicon follows, in general, the same fabrication sequence
as the processes previously described, depicted in Figure 4.5. First, the passivation
deposited on top of the Si substrate also serves as a buffer to stop the back-side etch-
ing. As in the RIE process, a resist layer and an etching-selective material are depos-
ited on the back side of the wafer, and they will serve to mask etching during the last
step of the process. The process is essentially the same as before: bottom electrode,
AlN layer, and top electrode deposition and patterning [Figure 4.5(a–c)]. In the last
step, anisotropic etching of the Si substrate succeeds in releasing the FBAR after
immersion of the wafer in KOH or in a KOH-based solution [Figure 4.5(d)]. The
remaining resist and etching mask can be removed. Since the Si etching stops at the
passivation layer on the front side of the wafer, this step can be avoided.

4.3 Instrumentation and Materials for Fabrication

There is a vast amount of technology, techniques, and processes available in the

market for FBAR and MEMS fabrication. For that reason, the list provided in Table

AIN
Bottom electrode Si3N4 /SiO2 Bottom electrode

Si Si

(a) (b)
Top electrode Top electrode
AIN AIN
Bottom electrode Bottom electrode

Membrane

Si Si

(c) (d)

Figure 4.5 Wet-etching-based bulk-micromachining process for FBAR fabrication: (a) bottom electrode;
(b) AlN layer; (c) top electrode; and (d) anisotropic etching of the Si substrate from the back side of the
wafer (in KOH solution).
4.3 Instrumentation and Materials for Fabrication 111

4.1 is only an example of the techniques, process parameters, materials, chemical

products, and processing conditions that may be involved in the device manufactur-
ing. The table lists the main processes under the assumption that (100) silicon
wafers are used as a substrate for device fabrication. This table should be used only
as an orientation to illustrate the complexity of the process, and it is suitable for
MEMS and FBAR fabrication using microfabrication techniques. The example of
the table includes the main processes involved in FBAR fabrication: photolithogra-
phy, thin-film deposition (sputtering), thermal oxidation, dielectric thin-film
deposition (CVD), and dry and wet etching for micromachining.

Table 4.1 Instruments, Chemical Products, and Materials Implemented in FBAR Fabrication
Process Technique and Process Parameter(s)
Photolithography Priming: solution and contact time (e.g., HMDS, 25 seconds for
(conventional UV) SiO2 substrates)
Coating: resist density and thickness (e.g., HiPR 6512, 1.2 μm)
Spinning: coating time and angular velocity (RPM)
Soft baking: temperature and time (e.g., 100ºC, 20 seconds)
Alignment and exposure: UV line, proximity mode, exposure time
(e.g., i-line, contact-3-μm, 10 seconds)
Developing: time, temperature, developer, and solvent
(e.g., 10–30 seconds, 22ºC, OPD4262, RER)
Hard baking: temperature and time (e.g., 115ºC, 30–60 seconds)
Cleaning-resist removal (dry etching by O2 plasma): plasma flow,
pressure, RF power, time (e.g., 10%, < 1 mbar, 500W, 30–45
minutes.)
Sputtering Target: sputtered precursor material (e.g., Ti, Pt, Al, W)
(AlN and metal deposition) Gas concentration (e.g., Ar, Ar/N2—50–50%, reactive)
Gas flow of the inert gas (e.g., 50 sccm (Ar))
RF power: between the anode and cathode (e.g., 100W)
DC power: between the anode and cathode (e.g., 500W)
DC bias: between the anode and cathode (e.g., 300 VDC)
−3 −2
Pressure (e.g., 10 –10 mbar)
Typical sputtered thickness (e.g., 10–2,000 nm)
Thin-film growth and Thermal oxidation: oxidation temperature, dry oxidation time,
deposition (oxides, nitrides) wet oxidation time; gas flow (e.g., 1,100ºC, 10–15 minutes,
90–120 minutes, 5–10 sccm—O2: dry, H2+O2: wet)
PECVD deposition: oxidation temperature (e.g., 1,100ºC)
Typical grown/deposited thickness (e.g., 10–2,000 nm)
Etching (AlN, metals, Dry etching (RIE): inert gas buffer (e.g., Ar); atmosphere
oxide, sacrificial layers) (e.g., SF6+O2 (Si), CHF3 (SiO2)); pressure (e.g., 75 mTorr); RF
power (e.g., 100W); DC bias (e.g., 80 VDC); etching rates (e.g.,
5,000 Å/min); etched quantities (e.g., 10–500 μm (thickness)).
Wet etching (isotropic, surface): etchant solution (e.g.,
OPD4262-TMAH (AlN)), HF 49% (Ti, SiO2); etching rate (e.g.,
300–500 Å/min (AlN in OPD4262)), 50,000 Å/min (Ti in HF);
drying (e.g., oven, critical point dryer (CPD, if sticking is critical));
etched quantities (typical) (e.g., 1 μm (AlN, thickness), 50 mm
(Ti, lateral)).
Wet etching (anisotropic, bulk, Si): etchant solution (e.g., KOH
40%, TMAH 25%); etching rate (e.g., 56 μm/hour (75ºC),
anisotropy 400:1 (100-Si in KOH)); etched quantities (typical)
(e.g., 500 μm (Si wafer))
112 Fabrication Techniques

4.4 Process Compatibility and Characterization

Different issues need to be considered for achieving compatibility between the dif-
ferent steps of the FBAR fabrication process, among them:

1. AlN deposition with good crystallographic quality and with no significant

alteration of the seed metal layer (bottom electrode);
2. AlN etching with no significant metal electrode etching;
3. Metal layer patterning with no significant AlN etching;
4. Si etching with no significant damage to the FBAR’s structure (for bulk
micromachining) and Si etching selectivity (against SiO 2 and AlN);
5. Sacrificial-layer etching selectivity (in the case of surface micromachining).

Achieving process compatibility thus requires the development of testing and

characterization strategies, tailored for the particular case of the materials
employed during the process. To implement these strategies, testing wafers made
with the FBAR composing materials have to be prepared. These wafers are to be
processed according to the expected process sequence and etchants. Depending on
the results of these tests, variations in the processing sequence in the materials or in
the etchants may be introduced. Usually, several iterations are required until full
compatibility is achieved.
During and after completion of the process testing, both quality and structural
characterization of the resonator will help to determine the right process or, at least,
the most convenient process combination. The characterization techniques avail-
able may include atomic force microscopy (AFM), scanning electron microscopy
(SEM), optical and confocal microscopy, and x-ray diffraction. In the following sec-
tions, we study different compatibility and quality issues.

4.4.1 Thin-Film Attributes

The thickness and profile of thin films will depend on the deposition rate, which is
determined by the type of material and the deposition conditions. Deposited thin
films are, in general, nonplanar. That means that, in the absence of planarization
processes, the thin film will display a nonuniform thickness across the wafer. Evalu-
ation of the thin-film uniformity and the thickness profile can be measured with a
system implementing standard spectroscopy and reflectometry techniques, like the
Nanospec family from Nanometrics [8]. The Nanospec system exhibits angstrom
resolution and is particularly useful for nitrides, oxides, and other transparent films.
The topographic profile of Figure 4.6 shows how the thickness of a nonplanar AlN
layer shifts as a function of the distance of the measured point to the center of a
100-mm wafer. According to this plot, the thickness standard deviation is 74 nm for
the scanned devices, which corresponds to a variation of around 10% (taken from a
mean value of 853 nm). These results show that high deviations may occur if no
external planarization means is employed (e.g., example chemical polishing).
Another topographic technique especially employed in metallic thin-film char-
acterization is mechanical profilometry. Thus, the profilometer scans the surface of
the sample with the tip of the scanner being physically in contact with the sample
4.4 Process Compatibility and Characterization 113

Figure 4.6 AlN thickness profile measured by the Nanospec AFT-200 system (distance is mea-
sured from the center of the wafer). (© 2007 IOP Publishing Ltd. [9].)

and scratching its surface. The plot of Figure 4.7 shows the profile of an FBAR with
W (750Å), AlN (5,000Å), and resist (12,000Å or 1.2 μm), measured with a Veeco
profiler. The FBAR has two W layers for bottom and top electrodes. The
photolithographic resist used for patterning of the top electrode is still on the W.
The FBAR is located on top of a passivation SiO2 layer.
The surface roughness of thin films also provides information about the success
and repeatability of the deposition process. Under certain deposition conditions, the
size of the surface’s grains is usually an indicator of some properties of the material.
A useful tool for measuring the grain size and the surface grain is the AFM.
For example, an AFM can be implemented to perform a detailed analysis of the
surface roughness of AlN film. In this application, AFM measurements relate the
size of the grain to the crystallographic quality of the AlN. The scanned samples
from Figure 4.8(a–c) present statistical analysis results of the surface roughness of
AlN films deposited onto Si, Al, and Pt bottom electrodes, respectively. In the fig-

Figure 4.7 Profilometry of an FBAR for supervision of thickness and profile of the layers: W
(750Å), AlN (5,000Å), and resist (12,000Å), measured with a Veeco profiler.
114 Fabrication Techniques

5.0 μm 5.0 μm 5.0 μm

2.5 μm 2.5 μm 2.5 μm

0.0 0.0 0.0

5.0 μm 2.5 μm 0.0 5.0 μm 2.5 μm 0.0 5.0 μm 2.5 μm 0.0
(a) (b) (c)

Figure 4.8 AFM analysis of the AlN surface roughness for different substrates: (a) Si; (b) Al; and
(c) Pt.

2
ures, the scanning area on the surface of each sample is 5 × 5 μm . For each case, the
measured roughness at root-mean-square (RMS) values is 12, 8.7, and 31.9 nm,
respectively. These results are a first indicator of the AlN crystal’s quality, taking
into account that higher grain sizes prompt higher crystallographic quality in the
preferred (002) orientation for a given deposited material [10]. In these examples,
the best grain size—and presumably best crystal quality—is obtained for the AlN
film deposited on Pt substrate.

4.4.2 Crystallography
A crystallographic study allows us to determine the material configuration and crys-
tal orientations of the evaluated sample. X-ray diffraction (XRD) equipment per-
forms this study, which is useful to analyze the presence and intensity of certain
orientations of the AlN crystals. Thus, one measures the full-width-half-maximum
(FWHM) aperture angle 2 θ/ω of the diffraction pattern and the rocking curves for
the set of materials expected to be in the tested wafer.
Diffraction occurs as waves interact with a regular crystalline structure whose
size is periodically repeated in a distance about the same as the wavelength of the
incident x-ray source. X-rays happen to have wavelengths on the order of a few ang-
stroms, the same as typical interatomic distances in crystalline solids. For this rea-
son, x-rays can be diffracted from minerals, which, by definition, are crystalline and
have periodic atomic structures. When certain geometric requirements are met,
x-rays scattered from a crystalline solid can constructively interfere, producing a
diffracted beam. In 1912, W. L. Bragg recognized a predictable relationship among
several factors [11]:

1. The distance between similar atomic planes in a mineral (the interatomic

spacing), which is called the d-spacing and is measured in angstroms.
2. The angle of diffraction, which is called the theta angle q and is measured in
degrees. For practical reasons, the diffractometer measures an angle that is
twice θ. That is why the measured angle is also called 2-theta (2θ).
3. The wavelength of the incident x-radiation, symbolized by the Greek letter
lambda (λ).
4.4 Process Compatibility and Characterization 115

These factors are combined in Bragg’s law:

n ⋅ λ = 2 d sin θ (4.1)

For the case of copper, which is the target material of the x-ray radiation in
most of the commercial XRD machines, λ = 1.54, and n is assumed to equal 1 (n =
1).
A diffractometer, a goniometer, and a scintillation counter for measuring the
x-ray intensity, among other setup elements, are used to make a diffraction pattern
of the samples. The goniometer is motorized and moves through a range of 2θ
angles. Because the scintillation counter is connected to the goniometer, we can
measure the x-ray intensity at any angle to the specimen. That is how the 2θ angles
for Braggs’s Law are determined (see Figure 4.9).
The sample patterns of Figure 4.10(a–d) show the 2θ/ω diffraction peak inten-
sity for different AlN crystal orientations, including the (002), c-axis orientation. In
this example, the AlN was deposited on a Pt seed layer. The 2θ angles for each orien-
tation are 33.24 (100), 36.12 (002), 37.94 (101), and 59.40 (110) degrees. As
observed in Figure 4.10, the AlN (002) orientation peak is several orders of magni-
tude more intense than in the AlN (101), (100), and (110) orientations, the relation-
ships being 40, 48, and 54 dB, respectively. This means that the AlN crystal in the
example exhibits a preferred orientation in the c-axis.
Assuming a hexagonal crystal structure and that each 2θ-angle peak corre-
sponds to a specific crystal orientation, the network parameters of the AlN crystal
can be calculated. The network parameters are the dimensions and distances of the
crystal-plane structure in a certain coordinate space. All the lattice planes and direc-
tions of the crystal are described by a mathematical description known as a Miller
index [12]. This allows the specification, investigation, and discussion of specific
planes and directions of a crystal. In the hexagonal lattice system, the direction [hkl]
defines a vector direction that is normal to the surface of a particular plane or facet,
where h, k, and l are coordinate axes. Referring to lattice’s d-spacing in (2.1), the
value of each network parameter is given by:
2 2 2 2
⎛ 1⎞ ⎛ h⎞ ⎛ k⎞ ⎛ 1⎞
⎜ ⎟ =⎜ ⎟ +⎜ ⎟ +⎜ ⎟ (4.2)
⎝d⎠ ⎝ a⎠ ⎝ b⎠ ⎝c⎠

Figure 4.9 X-ray diffraction measurement: simplified setup including x-ray source, target mate-
rial, sample wafer, goniometer, and counter.
116 Fabrication Techniques

5
1.2×10
1000
5
1.0×10

Intensity (counts)
800
Intensity (counts)

4
8.0×10
600
4
6.0×10
400
4
4.0×10
200
4
2.0×10
0
0.0
36 37 38 39 40
30 40 50 60 70
20/ω (deg) 20/ω (deg)
(a) (c)

1000 250

Intensity (counts)
Intensity (counts)

800 200

600 150

400 100

200 50

0 0

30 31 32 33 34 35 57 58 59 60 61 62
20/ω (deg) 20/ω (deg)
(b) (d)
Figure 4.10 X-ray diffraction peak intensity: (a) global XRD pattern (biggest peak: (002) AlN); and detail
of (b) (100) AlN (33.24°), (c) (101) AlN (37.94°), and (d) (110) AlN (59.40°) peaks.

where a, b, and c are the corresponding network parameters. In Figure 4.11(a) the
hexagonal structure of the AlN crystal is depicted. According to this geometry and
using (4.1) and (4.2), the values of a, b, c, and d are calculated to be 2.837Å,
2.941Å, 4.973Å, and 2.487Å, respectively. These results are very close to 3.084Å,
3.084Å, 4.948Å, and 2.474Å, which are reference values for hexagonal AlN crys-
tals [13]. The SEM image of Figure 4.11(b) shows the columnar structure of the
analyzed AlN sample. The columnar-crystal orientation is often associated with
good piezoelectric properties, although adequate poling is also a necessary condi-
tion [14].
Additional information concerning the AlN quality can be extracted from the
XRD analysis. The full width half maximum (FWHM) aperture angle is also an
important parameter to evaluate the quality of the crystal: the narrower the angle,
the higher the orientation in a certain axis. FWHM is measured in relation to refer-
ence materials, and its aperture angle depends on the width dispersion values for the
different planes of the crystalline structure (network parameters of crystal). As a
rule of thumb, values of less than 1° are expected for crystalline materials strongly
oriented in a specific axis.
As the fabrication technique strongly determines the crystallographic orienta-
tion, Table 4.2 compares the AlN deposition for different implementations and pro-
4.4 Process Compatibility and Characterization 117

(a)

(b)
Figure 4.11 AlN crystal orientation: (a) hexagonal lattice structure of AlN and (b) SEM image
showing the columnar structure of an AlN sample.

cesses. This table shows the relationship between deposition temperature,

full-wave-half-maximum (FWHM) width, and fabrication technique. Although the
comparative could be more exhaustive, it still contributes to estimate the quality of
films as a function of the fabrication techniques with reference information. As

Table 4.2 Different AlN-Deposition Implementations

Temperature
Source FWHM (°) (Max. ºC, Process) Thickness ( m)
[15] 1.6 Low (350, RF sputtering) 2.0
[16] 1.3 Low (<400, RF sputtering) 2.0
[9] 0.11 Low (<400, DC+RF sputtering) 1.0
[17] >0.4 Low (<500, epitaxial) N/A
[15] 0.4 High (500, epitaxial) 2.0
[3] 0.03 High (>600, epitaxial) N/A
118 Fabrication Techniques

observed, high-temperature processes offer, in general, more quality than low-tem-

perature-deposited AlN films, at the cost of CMOS-compatibility loss.
In this sense, the comparison of sputtering with epitaxial processes is not
straightforward. In spite of FWHM evaluation, further analyses must be carried out
to determine the ultimate crystal quality and piezoelectric properties of the AlN
film. Sanz-Hervás et al. found out that no direct relationship exists between low
FWHM values and the piezoelectric coefficient of sputtered AlN films, suggesting
the influence of domain inversion of the poling structure due to crystal defects [14].
On considering this background, characterization of poling domains and compara-
tives of different process conditions are required. Once the AlN is deposited on the
appropriate substrate, a device fabrication process is to be defined. Compatibility
between deposition, patterning, and micromachining at the different fabrication
stages has to be ensured in order to obtain a full and operational device.

4.4.3 Etching Performance

Appropriate releasing of the resonator is a requirement to obtain a high-quality fac-
tor. For this purpose, the micromachining technique has to provide a membrane or
cavity that allows the device to move freely with no significant damage of its struc-
ture. While insufficient etching may prevent satisfactory releasing, excessive pro-
cessing times may cause irreversible damage to the device. For this reason,
supervision and evaluation of the etching rates and selectivity is crucial to achieve a
successful fabrication. Optical microscope inspection, SEM, profile meter, and con-
focal microscopy, among others, can be used to calculate the anisotropy and etching
rates of the implemented micromachining process.
Prior to final device fabrication, we need to perform selectivity and compatibil-
ity testing of the different layers of the device. Evaluation of etching rates and integ-
rity of the device are the main goals of this testing. For this purpose, test wafers with
different layer configurations are prepared and the etching rates for the passivation,
electrodes, piezoelectric, and sacrificial layers are measured. With this information,
the etching times are adjusted to obtain a released device with minimum damage
and the shortest processing times. The selectivity of the structural layers of the reso-
nator is prioritized, especially that of the piezoelectric layer.
To illustrate the testing, let’s take the case of HF-based wet etching of a sacrifi-
cial Ti layer in a surface micromachining process. The electrodes are made of Pt
with an adhesive layer of Cr; the piezoelectric layer is AlN; and the FBAR is fabri-
cated on a Si3N4 passivation layer. AlN and Pt have proven to be highly selective to
HF, since they do not suffer significant damage during testing. Thus, the selectivity
of Cr and the etching rate of Ti are evaluated. Figure 4.12(a) shows the layout of a
test wafer with Si/Ti/Cr configuration and etching windows in the Cr layer (open-
ings in light color). The Ti lies underneath the Cr layer and is exposed to the etchant
through the etching windows. (Ti and Cr thicknesses are 1,000 and 400 nm, respec-
tively). Figure 4.12(b) shows a detailed view of the wafer with underetching dimen-
sions after 12 minutes of etching. Thus, for this specimen, the Ti etching rate is
calculated to be 40,000 Å/min. No change in the dimensions or shape of the Cr layer
is observed. However, slight differences in the Ti etching rate are observed when
measuring the size of different patterns. In considering the window-size factor and
4.4 Process Compatibility and Characterization 119

(a)

(b)
Figure 4.12 Ti/Cr etching-selectivity testing: (a) sample with Si/Ti/Cr configuration (opening
windows in light color showing underlying Ti layer); and (b) detailed view of the etch window
with underetching observed underneath the Cr layer.

after several measurements, the etching rate range is determined to be within the
range of 35,000–70,000 Å/min.
In addition, SEM is particularly useful for analyzing the anisotropy and etching
rates of bulk micromachining. By taking a cross section of the sample, the depth and
lateral underetching are measured and these parameters are calculated. The SEM
images of Figure 4.13(a) and Figure 4.13(b) show cross-sectional views of sample
profiles after RIE of the Si substrate. The RIE recipe of the example specifies an
SF6+O2 atmosphere with Ar buffer at 75 mTorr pressure, and 100W of RF power. A
patterned metallic layer highly selective to the RIE recipe covers the substrate to
allow selective etching. The metal patterning constitutes an etching window that is
opened in those areas around the FBAR where the etching is to be performed, thus
protecting the substrate and other regions of the device. Looking into Figure
4.13(a), the lateral versus vertical etching rates can be estimated to be between 1:3
120 Fabrication Techniques

(a)

(b)
Figure 4.13 RIE profile and etching rate evaluation (SEM images): (a) big etching window (> 20
μm); and (b) small etching window (< 10 μm).

and 1:4. According to the scale, the vertical etching is around 100 μm, whereas the
lateral underetching is around 30–35 μm. Thus, the lateral etching rate can be calcu-
lated to be 1,500–1,600 Å/min. However, the small-windowed structure of Figure
4.13(b) reveals a lower lateral-to-vertical aspect ratio in the order of 1:2. Also, the
etching rate is reduced for this case. Clearly, there exists a dependence between the
size of the etching window and the etching rate. Thus, the size of devices and etching
windows should be carefully designed to find the best combination: small windows
make the layout design more efficient, but they reduce the etching rate, whereas big-
ger windows increase the etching rate at the cost of bigger underetching areas.
According to this example, releasing a 50-μm-wide device would require a mini-
mum etching time of around 3 hours.
After completion of the fabrication process, the SEM images in Figure 4.14(a)
show the layout of the stacked structure of an FBAR comprising top and bottom
electrodes, the AlN, and the passivation layer. The Si substrate is found under the
4.4 Process Compatibility and Characterization 121

etching window (in dark). In Figure 4.14(b) the stacked configuration displaying
the electrode-AlN-electrode sequence can be seen. Applying a correction factor of
the vertical scale—due to the setup tilting of the electron beam—the thickness of the
composing materials can be measured (given the topographic scale, annotated in the
image). In this example, the tilting angle is 52º, as the metal-AlN-metal layers have
thicknesses of 180 nm, 1,000 nm, and 180 nm, respectively. Also, the air cavity
underneath the structure is observed. In partially released devices, some Si still
remains underneath the bottom electrode, as seen in Figure 4.14(c). In this case, fur-
ther etching is required to finish the FBAR fabrication.

4.4.4 Structural Performance

Interferometer and confocal microscopy are other useful techniques to evaluate the
structural performance of fabricated devices. Stress levels, sticking, and releasing of
the resonator are some of the aspects that can be studied.
The image of Figure 4.15(a) shows a partially released cantilever. Interferome-
ter bands are observed in the released region of the device (lower side of the image,
the area near the border of the electrode). The topographic image (up) corroborates
this observation, where bending of the device due to residual stress can be deduced
from the higher-scale levels near the border of the device (light gray). The central,
dark-gray region is still clamped to the Si substrate. At its time, the FBAR of Figure
4.15(b) exhibits relatively homogeneous scale levels (up) and interferometer bands
(down). In this case, the square-shaped, beam-type device has been fully released
from the substrate, thus vibrating at its natural mechanical frequency (MHz) all
along its structure. However, we can observe considerable stress of the beam when
contrasting the topographic levels at the center and lateral regions, where one of the
clamped electrodes is found (at left). A RIE-based bulk micromachining process was
served to fabricate both the cantilever and beam devices.
Confocal microscopy allows us to get additional details on the structure of cer-
tain processes, specifically for thin-film devices like FBARs [18, 19]. Flatness, stress,
and sticking of the structure can be analyzed. Also, the etching status of a
micromachining process can be supervised.
After micromachining, the structure or the contour regions of the resonator
may suffer damage, sticking, or deformation due to residual stress after releasing. In
Figure 4.16, two devices fabricated within bulk and surface micromachining pro-
cesses are compared. In Figure 4.16(a), the accumulated stress in the underetched
region of the passivation layer (SiO2) is observed in dark gray after Si underetching.
Such stress is evidenced by deformation and, in some parts, cracking of the mem-
brane (around the etching window in black). The situation of a surface
micromachined device is fairly different, as shown in the confocal image of Figure
4.16(b). Lower processing times and fewer fabrication steps alleviate induced stress,
thus providing relatively flat structures, in comparison to those obtained by the
bulk-based process. A quasi-flat topographic profile of the same device along the
A-A′ axis can be seen in Figure 4.16(c). In this example, the 1.5-μm-thick sacrificial
layer has already disappeared after etching. The top electrode and the overall FBAR
structure also have an almost-flat conformation.
122 Fabrication Techniques

(a)

(b)

(c)
Figure 4.14 FBAR structure after fabrication: (a) overall layout; (b) insight into the layered
metal-AlN-metal structure (tilting angle of 52°, thicknesses of Pt, AlN, and AlN of 180, 1,000, and
180 nm, respectively); and (c) partially released device (Si substrate is still observed underneath
the device). (© 2007 IOP Publishing Ltd. [9].)
4.5 Summary 123

(a)

(b)
Figure 4.15 Interferometer analysis of the FBAR structure: (a) partially released, cantilever-type
device (partial bending and nonhomogeneous interferometer bands are observed); and (b) fully
released, beam-type resonator (homogeneous bands and topographic levels all along the device).

4.5 Summary

The FBAR fabrication technology comprises many steps and has to be developed to
achieve full-process compatibility. Among these steps, piezoelectric layer deposition
and a variety of micromachining processes are the key technologies. Also, different
characterization techniques have to be implemented in order to evaluate the main
steps of the process. Structural, crystallographic, and etching analysis can be studied
by current-art instrumentation like AFM, SEM, interferometer, and confocal
microscopy. Although a variety of process parameters should be considered, Table
4.3 attempts to compare the different aspects concerning FBAR-related
microfabrication. Advantages and challenges of each technology should be sought
on considering the possibilities and limitations of the technology available to the
designer.
124 Fabrication Techniques

(a)

(b)
Z (μm)
5

1
+rms

−1
−rms
−2

−3

−4

−5

0 20 40 60 80 100 120 140 160 180 200

Y (μm)

(c)
Figure 4.16 Stress and flatness confocal supervision: (a) RIE-based FBAR (accumulated stress in
the SiO2 membrane is observed); (b) surface-micromachined FBAR (sacrificial layer cavity after
etching in dark gray); and (c) topographic profile of the device (along the A-A’ axis).
4.5 Summary 125

Table 4.3 Micromachining Implementations in the Fabrication of FBAR

Bulk Micromachining Surface Micromachining Bulk Micromachining
Fabrication Issue (Dry Etching) (Wet Etching) (Wet Etching)
Design complexity 4-mask set, including 4-mask set, including 5-mask set, including
RIE mask sacrificial layer mask additional back-etching mask
Fabrication High metal step Sticking, critical etching Front-to-back mask alignment,
complexity (top and bottom) time (to avoid metal lift-off) residual silicon on back side
Etching rate High Highest Low
Etching time Short Shortest Long
(50 μm × 50 μm,
wafer: 500 μm)
Cleanness More clean Less clean Less clean
Residual stress Medium Low High
Etching control Time-supervised Time-supervised Self-controlled
Wafer area required Underetching of Less than RIE More than RIE
non-FBAR regions

References

[1] Ylilammi, M., et al., “Thin Film Bulk Acoustic Wave Filter,” IEEE Trans. on Ultrason.
Ferroelectr. Freq. Control, Vol. 49, 2002, pp. 535–539.
[2] Lee, J. B., et al., “Effects of Bottom Electrodes on the Orientation of AlN Films and the Fre-
quency Responses of Resonators in AlN-Based FBARs,” Thin Solid Films, Vol. 447–448,
2004, pp. 610–614.
[3] Uchiyama, S., et al., “Growth of AlN Films by Magnetron Sputtering,” J. Crystal Growth,
Vol. 189–190, 1998, pp. 448–451.
[4] Vispute, R. D., H. Wu, and J. Narayan, “High Quality Epitaxial Aluminum Nitride Layers
on Sapphire by Pulsed Laser Deposition,” Appl. Physics Lett., Vol. 67, No. 11, 1995,
pp. 1549–1551.
[5] Dubois, M. A., and P. Muralt, “Stress and Piezoelectric Properties of Aluminum Nitride
Thin Films Deposited onto Metal Electrodes by Pulsed Direct Current Reactive Sputter-
ing,” J. Applied Physics, Vol. 89, No. 11, 2001, pp. 6389–6395.
[6] Oshmyansky, Y., et al., “Sputtering Processes for Bulk Acoustic Wave Filters,” Semicon-
ductor International, 2003, http://www.semiconductor.net/article/CA282270.html.
[7] Chung, C.-J., et al., “Synthesis and Bulk Acoustic Wave Properties on the Dual Mode Fre-
quency Shift of Solidly Mounted Resonators,” IEEE Trans. on Ultrason. Ferroelectr. Freq.
Control Vol. 55, No. 4, 2008, pp. 857–864.
[8] Nanometric Inc., http://www.nanometrics.com.
[9] Campanella, H., et al., “Focused-Ion-Beam-Assisted Tuning of Thin-Film Bulk Acoustic
Wave Resonators (FBAR),” J. Micromech. Microeng., Vol. 17, 2007, pp. 2380–2389.
[10] Clement, M., et al., “SAW and BAW Response of C-Axis AlN Thin Films Sputtered on Plat-
inum,” Proc. IEEE Intl. Ultrason. Symp. 2004, Montreal, Quebec, August 24–27, 2004,
pp. 1367-1370.
[11] Perutz, M. F., “How W. L. Bragg Invented X-Ray Analysis,” Acta Cryst. A, Vol. 46, 1990,
pp. 633–643.
[12] Ashcroft, N. W., and N. D. Mermin, Solid State Physics, New York: Harcourt, 1976.
[13] Wright, A. F., and J. S. Nelson, “Consistent Structural Properties for AlN, GaN, and InN,”
Phys. Rev. B: Condens. Matter, Vol. 51, 1995, pp. 7866–7869.
126 Fabrication Techniques

[14] Sanz-Hervás, A., et al., “Degradation of the Piezoelectric Response of Sputtered C-Axis
AlN Thin Films with Traces of Non-(0002) X-Ray Diffraction Peaks,” Appl. Phys. Lett.,
Vol. 88, 2006, p. 161915.
[15] Engelmark, F., et al., “Structural and Electroacoustical Studies of AlN Thin Films During
Low Temperature Radio Frequency Sputtering Deposition,” J. Vac. Sci. Technol. A.,
Vol. 19, 2001, pp. 2664–2669.
[16] Hara, M., et al., “Surface Micromachined AlN Thin Film 2 GHz Resonator for CMOS
Integration,” Sens. Actuator A-Phys., Vol. 117, 2005, pp. 211–216.
[17] Shiosaki, T., et al., “Low Temperature Growth of Piezoelectric Films by RF Reactive Planar
Magnetron Sputtering,” Jap. J. Appl. Phys., Vol. 20, 1981, pp. 149–152.
[18] Semwogerere, D., and E. R. Weeks, “Confocal Microscopy,” Encyclopedia of Biomaterials
and Biomedical Engineering, London, U.K.: Taylor & Francis, 2005.
[19] Prasad, V., D. Semwogerere, and E. R. Weeks, “Confocal Microscopy of Colloids,” J.
Phys.: Cond. Mat., Vol. 19, 2007, p. 113102.
 CHAPTER 5

Characterization Techniques
Characterization of FBAR and MEMS resonators comprises different methods and
techniques, and is performed through successive measurement stages. Material and
equivalent-circuit parameter extractions enable for a complete description of the
device, which can be used to design specific applications. Nowadays, several reso-
nator characterization techniques are available through commercial systems or spe-
cific laboratory setup. Electrical, optical, or mechanical methods, among others,
can aid the parameter-extraction process of micro- and nanodevices, although the
electrical techniques are perhaps the most powerful for MEMS resonator
measurement.
Electrical characterization involves understanding the basics of network theory
and measurement techniques necessary to extract the equivalent circuit parameters,
the quality factor, the electromechanical coupling, and the elastic, dielectric, and
piezoelectric constants of the resonator. In this chapter, we discuss two electrical
characterization approaches based on scattering parameter analysis. The setup and
applications of the low- and high-frequency techniques are explained by exemplary
applications. Other techniques are also being used to evaluate the resonant behavior
of FBAR and MEMS. In particular, the concepts and measurement setup of interfer-
ence microscopy and AFM techniques are introduced at the end of the chapter.

5.1 Low- and High-Frequency Electrical Characterization

Electrical characterization entails the feeding of an electric current or voltage to the

circuit composed by the resonator, a signal generator, and, optionally, a matching
or read-out electric circuit. In general, what is characterized is the change of the
current delivered to a loading impedance when it flows through the resonator’s
frequency-dependent impedance Zm. The schema of Figure 5.1 depicts the typical
setup of a resonator’s electrical characterization.
The definition of “low” frequency is relative to the actuation mechanism and
design of the resonator, but in general it comprises the range of frequencies between
DC and tens of megahertz. On the other hand, the “high” frequency behavior is
described in terms of energy delivered to the load, rather than in the traditional,
low-frequency circuit representation. Thus, the energetic description focuses on the
energy passing through and returning from a distributed transmission line in which
the resonator is inserted, when a lumped-circuit representation of the system is
impractical or hard to establish. In the following sections, we define both the low-
and high-frequency measurements and their usefulness in the parameter extraction
of the resonator.

127
128 Characterization Techniques

Z in Z m(f )
I L (f )
Iin (f )
VAC(f )
V L (f ) ZL

Signal generator Resonator Load

Figure 5.1 Electrical characterization of a MEMS resonator: the resonator’s frequency-dependent
current IL(f) flows through the load impedance ZL. The current Iin(f) is provided by an AC signal
generator and the loading impedance may be a passive or an active circuit matched to a signal
analyzer (like a network analyzer).

5.1.1 Short-Open DC and Low-Frequency Measurements

We define “low-frequency” as the frequency range between DC and a few mega-
hertz. This is the frequency band in which the first mechanical modes of MEMS res-
onators are found. Low-frequency measurements can monitor the success of the
fabrication process. For example, one can verify short- and open-circuit conditions
of the resonator’s electrodes. At the same time, measurements enable the extraction
of the low-frequency material constants.
First, the integrity of the resonator’s structure is verified by DC short-open mea-
surements. As FBARs and several MEMS and NEMS resonators implement two or
more electrodes to allow read out, it is crucial to assure ohmic contact between the
lines connecting the resonator and the read-out electrode system (short-circuit con-
dition). Electrical isolation between different electrodes also has to be guaranteed
(open-circuit condition).
After verification of the correctness of the structure, we evaluate DC or low-fre-
quency material constants and layout parameters. For example, in FBARs or paral-
lel plate electrostatic resonators, the pad-to-transmission-line DC resistance (Rs /2)
and the static capacitance C0 can be measured. Then, one extracts the static
permittivity ε value from the measured C0 value, for known resonator dimensions
(electrode area A and thickness t of the AlN layer, or equivalently the plate-to-plate
distance). On the other hand, characterization of the first electromechanical-mode
resonance frequencies provides useful information for the extraction of elastic con-
stants of the resonator’s materials.
Short and open-circuit can be performed using a DC probe station, a semicon-
ductor parameter analyzer, and a capacitance meter, among other instruments. A
fully functional device must accomplish both the short- and the open-circuit condi-
tions. These conditions are evaluated through current-to-voltage testing, or I/V test-
ing. In the setup of the measurement system, a saturation current is defined for the
short-circuit measurement. Thus, I(V) plots are used to evaluate the line resistance:
a high resistance prompts for the open-circuit verification, while low-resistance val-
ues indicate the short-circuit condition (line resistance in the units of ohms or less).
The short-circuit condition of a given electrode is verified by connecting two
DC probes at different points of the line, as shown in Figure 5.2(a, b). Due to the
fabrication process and to the device’s layout, it may be of special interest to check
out this condition on those electrodes with thin-film layers and high fabrication
5.1 Low- and High-Frequency Electrical Characterization 129

Short
0.10

0.05

Current (A)
Si substrate 0.00

(a) Short −0.05

Short
−0.10

−10 −5 0 5 10
Voltage (V)

(c)

(b)
Figure 5.2 Setup and connections for verification of the short-circuit condition: (a) side-view
schematic of the probe location; (b) top-view optical photograph of an FBAR indicating the probe
location; and (c) I/V plot for a typical device (line resistance Rs/2 of 12Ω).

steps. In these cases, there is a high risk of open circuit between the resonator’s elec-
trodes and the transmission line, a situation which has to be evaluated.
The short-circuit condition is affirmatively verified if the electrode shows cur-
rent-continuity between the points where the two probes are located. The experi-
mental I/V plot of a device verifying the condition is shown in Figure 5.2(c). In this
example, the current limitation is 100 mA. The line is made of Pt with thickness of
150 nm, width between 30–100 μm and length between 50–200 μm. With these val-
ues, the line resistance Rs/2 equals 12Ω. Let us note that the locations in which the
probes touch the electrode have not been systematically controlled. Hence, this
experiment is conceived only to give a rough estimation of the magnitude of Rs/2 (or
of the short-circuit condition). The value depends on the dimensions and material
(Pt) of the transmission line, which explains the high value of Rs/2, in comparison to
Mo or Al implementations (Rs/2 less than 1Ω). Since Pt is a material with higher
resistivity than Al or Mo, for example, relatively short transmission lines lead to
Rs/2 values of units of ohms. Thus, with appropriate layout and process optimiza-
tion Rs/2 may be significantly reduced.
In two-electrode resonators, the open-circuit measurements are carried out by
connecting each probe to each one of the electrodes. In Figure 5.3(a, b), the mea-
surement configuration and connections are observed. Verification of the open-cir-
cuit condition guarantees that both electrodes are isolated between them (i.e., no
electrical contact occurs), thus keeping the low-frequency static-capacitance behav-
ior of the device. Mask misalignment during the fabrication process could derive
into electrode contacting and short-circuiting of them. The open-circuit condition is
thus verified if current discontinuity is observed between the two probing points.
Figure 5.3(c) shows the I/V plot for a device verifying the open-circuit condition (the
130 Characterization Techniques

Open
−9
1.5×10 Open

−9
1.0×10
−10
5.0×10

Current (A)
Si substrate 0.0
(a) −10
−5.0×10
−9
−1.0×10
−9
−1.5×10
−10 −5 0 5 10
Voltage (V)

(c)

Open

(b)

Figure 5.3 Verification of the open-circuit condition: (a) side-view schematic of FBAR with probe
location; (b) top-view optical photograph of an FBAR indicating the probe location; and (c) I/V
plot for a device verifying the condition (minimum through-resistance of 25 GΩ (i.e., the open-cir-
cuit condition).

current was also limited to 100 mA). By calculating the mean value and standard
deviation of the current, the DC through-resistance was estimated to be between
25–90 GΩ (equivalent to an open-circuit condition).

5.1.2 Microwave Network Theory and the Scattering-Parameter Description

Transmission and reflection measurements are the best way to describe the electri-
cal response of resonators working in the gigahertz frequency regime. In particular,
the scattering-parameter representation—or S-parameters—is very useful to
explain the impedance and energy relationships of the RF system comprised by the
resonator, the transmission line connecting said device to the measurement system,
and the probing pads at which the test instrumentation is connected to the circuit.
S-parameters are used to evaluate microwave circuits and come from the two-port
circuit representation shown in Figure 5.4, where two sets of ingoing (ai) and outgo-
ing (bi) waves are generated. The indices i = 1, 2 stand for the input port and output
port, respectively.
The four quantities are related together by the scattering matrix S:

a1 a2
S
b1 b2

Figure 5.4 Two-port circuit representation with scattering parameters.

5.1 Low- and High-Frequency Electrical Characterization 131

⎡b1 ⎤ ⎡ S 11 S 12 ⎤ ⎡ a1 ⎤
⎢b ⎥ = ⎢S S 22 ⎥⎦ ⎢⎣a 2 ⎥⎦
(5.1)
⎣ 2 ⎦ ⎣ 21

Setting one of the independent variables to zero allows us to find the individual
S-parameters by:

b1
S 11 =
a1 a 2 =0

b1
S 12 =
a2 a 1 =0
(5.2)
b
S 21 = 2
a1 a 2 =0

b2
S 22 =
a2 a 1 =0

where S11 is the reflection coefficient at port 1 when port 2 is terminated with a
matched load, S12 is the reverse transmission coefficient when port 1 is terminated
with a matched load, S21 is the forward transmission coefficient when port 2 is ter-
minated with a matched load, and S22 is the reflection coefficient at port 2 when port
1 is terminated with a matched load (S11 and S22 are closely related to the input and
output impedances, respectively) [1].
Typically, the transmission line can be implemented as a coplanar waveguide
(CPW). In this way, the S-parameters can also be defined in terms of the circuit and
the CPW electrical characteristics [2]:

Z in
S 11 =
(2 Z S + Z in )(1 + λL) (5.3)
2Z S
S 21 =
(2 Z S + Z in )(1 + λL)

where Zin and Zs are the resonator’s input and source impedance, respectively, and λ
and L are the propagation constant and length of the CPW, respectively. These
equations are simplified forms of the S-parameters for the special case: Z0 = ZS,
where Z0 is the characteristic impedance of the CPW. These relationships are very
useful when S-parameter measurements are performed with a network analyzer and
a probing system with calibrated and known impedance values (typically Z0 = 50Ω).

5.1.3 High-Frequency Measurement Setup

The setup to perform high-frequency electrical characterization requires the use of
microwave network analyzer and probe stations or prototypes to connect the reso-
nator to the measurement equipment. A first setup comprises the network analyzer
and a coplanar probe station (ground-signal-ground system). The network analyzer
performs automated evaluation of the resonator’s S-parameters. The use of
132 Characterization Techniques

coplanar transmission lines provides a very convenient method to contact the reso-
nator because they offer a well-defined ground plane and avoid the introduction of
parasitic and strain inductances [1]. The typical setup for a CPW connecting an
FBAR is depicted in Figure 5.5, where the FBAR’s circuit representation and the sys-
tem-level interconnection are shown. By using appropriate calibration standards
and routines, the measurement plane is translated to the probing-pad location, thus
compensating the effects of the probing system—including cables and connectors—
and matching the network analyzer’s port impedance of 50Ω. Commercial network
analyzers may be employed for gigahertz-range measurements [3], whereas the
interconnection may be carried out with microwave probe stations [4].
Nowadays, different calibration standards are implemented in RF measure-
ments, as the coplanar thru-open-short-and-load (TOSL) structures are very popu-
lar. TOSL standards are commercially available in the form of printed-circuit
substrates and are provided by microwave-instrumentation companies [5]. The

a1
ZG bG

1 2

EG b1
V ZL

Signal generator Load

(network analyzer) ΓG Γi (network analyzer)

(a)

Network analyzer

Probe station

(b)
Figure 5.5 First setup for electrical characterization of the scattering parameters in FBAR: (a) cir-
cuit representation; and (b) physical connection of the instrumentation and probe station.
5.1 Low- and High-Frequency Electrical Characterization 133

name of the TOSL standard describes the network components employed in the cal-
ibration routine. This standard is conventionally used with an automatic network
analyzer and is suitable for calibration routines limited to coaxial transmission lines
in the frequency range from DC to 6 GHz. To perform the calibration, the standard
kit implements two components: the printed-circuit TOSL substrate and the stan-
dard software against which the measurements are compared during calibration.
For higher frequency calibrations, other standards are implemented [6].
In a second setup, a printed circuit board (PCB) will replace the coplanar probe
station. In this PCB, a CPW is designed to connect the resonator die to the measure-
ment instrumentation. The die is attached to the PCB and connected to the
PCB-made transmission lines by wire-bonding. For this purpose, the PCB’s CPW
design has to guarantee impedance matching to the 50Ω value of the port imped-
ance of the network analyzer (appropriate PCB design and careful selection and
knowledge of the dielectric constants of the PCB’s laminate are required). The cir-
cuit representation of this setup is the same as that depicted in Figure 5.5(a). A sche-
matic representation of an exemplary PCB-FBAR system is depicted in Figure
5.6(a), and the physical setup comprising the PCB and the network analyzer is
shown in Figure 5.6(b). The calibration standards used in this setup are coplanar
TOSL structures implemented as coaxial connectors fabricated by Agilent Technol-
ogies. The PCBs were fabricated using laminates made of the high-frequency dielec-
tric material Rogers 3010, which is a ceramic-filled PTFE composite with low
dielectric loss and application up to 10 GHz [7].
The measurement setup also influences the characterization results and has to
be considered when performing the evaluation of results and the parameter
extraction. As previously discussed, PCB-based and coplanar probe station char-
acterization setups can be implemented. This decision has an impact on the mea-
surement’s performance and leads to different de-embedding approximations.
While the PCB-based setup is practical for complex systems involving more than
one FBAR device, such as nonconventional devices or hybrid FBAR-CMOS cir-
cuits, its flexibility is achieved at the cost of added losses and reactance due to the
wire-bonding, and it is not very practical for single-resonator characterization.
Also, additional—and nonnegligible—design and prototyping effort has to be
dealt with.
To visualize the differences and impact of both systems, the same device can be
characterized by first measuring it directly on the wafer, with the coplanar probe
station. After data measurement and acquisition, the wafers can be diced, packaged,
and tested on chips glued and bonded to the PCB. The plots of Figure 5.7 compare
data from S-parameter measurements performed on the same resonator, where
three main resonances are observed at 1.8, 2.4, and 2.6 GHz. Although the electri-
cal response is similar in both setups, the effects of added losses and reactance
affecting the S-parameter values can be observed. Wire-bonding and PCB design
optimization would help diminish the differences between both measurements.

5.1.4 Quality Factor Extraction

The quality factor (Q) is a key parameter to evaluate the performance of both the
resonators and the resonator-based applications. Hence, its evaluation is of great
134 Characterization Techniques

(a)

Network analyzer

PCB with packaged FBAR

(b)
Figure 5.6 Second setup for electrical characterization in FBAR: (a) schematic representation of
the PCB-FBAR system; and (b) physical interconnection of the PCB and the network analyzer.

concern in order to perform accurate estimations of the device’s capabilities. There

are different methods for calculating the Q factor of a resonant device. Based on
electrical characterization, transmission and reflection S-parameters are useful tools
to extract the Q factor. Its characterization in FBARs is important because it deter-
mines the selectivity of filters and phase noise of oscillators in RF applications, and
the sensing capability of sensors [8, 9].
In a first method, which we will call the S21-S11 magnitude method, the Q value
of a resonator can be obtained from the measurements of two-port S parameters, by
using the magnitudes of both the transmission (S21) and reflection (S11) parameters.
According to this method and assuming source impedance Zs of 50Ω, the Q factor is
calculated by the following equation [2]:
5.1 Low- and High-Frequency Electrical Characterization 135

−4 320
280
−6
240
−8

S21p (deg)
200
−10
S21m (dB)

160
−12 120
−14 80
−16 40
−18 0
−20 −40
1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0
Frequency (GHz) Frequency (GHz)
(a) (c)
0

−4

−8
S11m (dB)

−12

−16

−20

1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

Frequency (GHz)
(b)
Figure 5.7 Comparison of the probe-station-based and the PCB-based characterization setups: (a) magni-
tude of S21; (b) magnitude of S11; and (c) phase of S21. The effects of added reactance are observed in
the response of PCB-packaged resonators.

⎛ ωS ⎞
⎜ ⎟
⎜ω ⎟
QS =
⎝ p⎠ (1 − S 21 Min ) (1 − S 11 Min ) (5.4)
2
⎛ω ⎞ S 21 Min S 11 Min
1 − ⎜⎜ S ⎟⎟
⎝ ωp ⎠

In (5.4) ωs and ωp are the serial and parallel resonance frequencies, and S21Min
and S11Min are the minimum values of the S21 and S11 parameters, respectively. This
method for determining the Q value has the advantage of taking account of all
acoustic and electrical loss mechanisms [10].
According to the classical BVD model presented in Chapter 3, the Q factor at
the series resonance frequency can also be defined as:

1
QS = (5.5)
ω S R mC m

From (5.3), (5.4), and (5.5), values for Rm and Cm can be extracted in a first
approximation by:
136 Characterization Techniques

S 11 Min
Rm = 2Z S (5.6a)
(1 − S 11 Min )
1
Cm = (5.6b)
ω S R mQS

Another Q factor definition considers the open-loop phase response φ(ω) of the
resonator, examined at resonance. The Q factor is then defined as [9]:

ωS ∂ φ
Q= (5.7)
2 ∂ω

If the phase slope is large, a significant change in the phase shift and Q factor
arises. This definition has an interesting interpretation for oscillator-design applica-
tions in which the resonator drives the oscillation. Thus, the open-loop Q value
measures how the closed-loop circuit will oppose the variations in the frequency of
oscillation [9]. Using this method, in practice, the Q factor at resonance can be
extracted from the S21 parameter by observing the derivative of the S21 phase.
According to this method and to the plots of Figure 5.8, the phase of the S21 parame-

40

20
S21phase (deg)

0
−20

−40

−60

−80
2.2 fS fp 2.4
Frequency (GHz)
(a)

−5
1.0×10

−6
5.0×10

0.0

2.2 fS fp 2.4
Frequency (GHz)
(b)
Figure 5.8 Phase response of a typical FBAR for Q-factor calculation: (a) phase of the S21 param-
eter indicating the minimum and maximum slope values (the same correspond to series and paral-
lel resonance frequencies, respectively); and (b) derivative of the S21 phase (the most negative
and positive peaks correspond to fs and fp).
5.1 Low- and High-Frequency Electrical Characterization 137

ter of an FBAR is used to characterize the Q factor of the resonator. The most nega-
tive derivative value coincides with the most negative phase slope, corresponding to
the series resonance frequency fs. On the other hand, the most positive derivative
value corresponds to the most positive phase slope (i.e., the parallel resonance fre-
quency fp). Characterization results of the Q factor by using this method are similar
to those found by the S21-S11-magnitude method. In this example, a Q value bigger
than 1,000 is extracted.
A third method widely employed for Q factor calculation uses the −3-dB defini-
tion of bandwidth: Q is then defined as the resonance frequency f0 divided by the
two-sided −3-dB bandwidth B (Q = f0/B). This method is very popular for
high-insertion-loss MEMS devices and bandpass filters with a several-decibel peak
response. Figure 5.9 shows the frequency response of a clamped-clamped MEMS
resonator fabricated within the UMC CMOS process [11]. The response was
obtained by mixing technique measurements (circles) and the corresponding
Lorentz fitting (continuous line). At the 25.475-MHz central frequency, the signal
power level is about −115 dBm, and the curve fits the −3-dB levels of −118 dBm at
the frequencies located at ± 15.75 kHz of the central frequency, or B 31.5 kHz,
thus leading to the Q value of 812. However, in low-loss devices such as FBARs, the
resonance peak at the series resonance frequency is poorly defined, especially—and
paradoxically—in high-Q factor resonators (insertion losses near to 0 dB are the
rule on FBAR performance). For this reason, this method is rarely used to character-
ize FBARs.
To illustrate the impact of Q factor on FBAR and MEMS resonator-based
applications, let’s consider the example of a microwave ladder filter. Ladder filters
and duplexers are nowadays the most popular and commercially successful applica-
tions of FBARs. Worldwide companies like Avago Technologies (formerly Agilent)
or Infineon are selling millions of units each year, thus replacing RF components
made with SAW and ceramic technologies [12, 13]. Such components offer lower
insertion losses, higher out-of-band rejection, and a more reduced size than those
made with previous technologies. Although FBAR filters are well developed and
mature at the design and technology levels, their performance is very sensitive to the

Figure 5.9 Frequency response around the resonance peak of a CMOS-MEMS clamped-clamped
HF resonator: the Q factor equal to 812 was extracted with the –3-dB method after mixing tech-
nique measurements performed in vacuum. (© 2009 Institute of Physics [11].)
138 Characterization Techniques

quality of their composing resonators. Since there are numerous references to FBAR
filters [14, 15], this application has also become a useful benchmark tool in order to
explore the limitations and possibilities of FBAR technologies.
A ladder filter is comprised by an nth-order interconnection of series
and shunt-located FBARs. The schematic and equivalent-circuit models are
depicted in Figure 5.10. The order of the filter refers to the number of
series-to-shunt stages, counting from the input to the output port of the filter [14].
The series and shunt FBARs have different resonance frequencies and should ideally
be of different size, in order to optimize the in-band and out-of-band performance
of the filter [16].
The performance of the filter can be predicted by circuit-level simulations and
the equivalent-circuit model of Figure 5.10(b). The equivalent-circuit model of a
third-order filter (N = 6 resonators) with 2.2-GHz central frequency and corre-
sponding simulation results are shown in Figure 5.11(a). Outstanding 2-dB inser-
tion loss and 36-dB out-of-band rejection values can thus be predicted. However,
the realization of the filter does not always lead to the expected specifications. If
process variations or design issues significantly affect the quality factor of resona-
tors, the global result is disastrous for the filter’s performance. Following with the
same example, the third-order filter performs the poor characteristics shown in Fig-
ure 5.11(b). Insertion losses higher than 14 dB, and 26-dB zeroes in the out-of-band
region, are far below the standards required for modern filters.
If we analyze these results, we can obtain a good explanation regarding the Q
factor of the FBARs. According to the insertion-loss expression for a passband filter
[17], we can see how strong the link between the Q factor and insertion losses (IL)
is:

(a)

(b)
Figure 5.10 FBAR-based ladder filter (third order, six resonators): (a) ladder filter topology; and
(b) circuit-level modeling.
5.1 Low- and High-Frequency Electrical Characterization 139

0
−20

−40

V0 (dB)
−60

−80

−100
1.8 2.0 2.2 2.4 2.6
Frequency
(a)

−14 −4
−16
−6
S21mag (dB)

S11mag (dB)
−18
−20 −8

−22 −10
−24
−12
−26
−28 −14

2.0 2.2 2.4 2.6 2.8 3.0

Frequency (GHz)
(b)
Figure 5.11 Frequency response of the ladder filter: (a) simulated response (2-dB insertion loss,
out-of-band rejection > 36 dB); and (b) experimental characterization of fabricated filter (14-dB
insertion loss, 26-dB out-of-band rejection).

⎛ ⎞
⎜ ⎟
⎜ 4 ⎟
IL( dB) = 20 log⎜ 2 ⎟
(5.8)
⎜ 4 + 3π ⎟
⎜ 2
Q ⎟⎠
⎝ keff

2
For a k eff of 3.0–3.4%, and IL values between 12–16 dB, Q factors in the range
of 60 to 80 can be predicted. If we look at the filter-composing resonator and its S21
parameter magnitude, both shown in Figure 5.12, we can verify that the extracted
Q factor value is in the same range. In summary, we see how low Q factor values of
the resonators drastically increase insertion losses of the filter. Commercial filters
meeting the stringent requirements of contemporary wireless mobile systems imple-
ment resonators with Q factors higher than 1,000. Precise understanding and care-
ful extraction of the resonator physical constants enable fine-tuning of the
fabrication process, thus promoting such high Q factors. Therefore, the following
section describes the elastic, dielectric, and piezoelectric constants of FBARs and
their extraction.
140 Characterization Techniques

−2

S21mag (dB)
−4

−6

1.8 2.1 2.4 2.7 3.0 3.3

Frequency (GHz)
Figure 5.12 Characterization of one of the filter-composing FBARs: the S21 parameter of the res-
onator in the inset allows extraction of its Q factor (60–80), which explains to a great extent the
poor performance of the filter.

5.2 Determination of Elastic, Dielectric, and Piezoelectric Constants

The electromechanical performance of FBAR, MEMS, and NEMS resonators

depends on both their design and their transduction mechanism. Physical constant
values of the materials composing the resonator ultimately determine their reso-
nance frequency, quality factor, and other performance parameters. In the next sec-
tions, we analyze the extraction of physical constants by discussing the case of
FBARs and AlN. The characterization procedures and parameter extraction meth-
odology used for the elastic and dielectric constants are also extendable to other
MEMS and NEMS devices, as we study. Moreover, the extraction of piezoelectric
constants is of use to characterize FBARs, SMRs, SAW, and, in general, piezoelec-
tric-based MEMS and NEMS resonators. The background and experimental
methods for the extraction of such constants are described next.

5.2.1 Elastic Constants

The acoustic and mechanical properties of the structural layer of FBAR and MEMS
resonators mainly determine their performance and frequency response. While the
structural layer of MEMS resonators is typically fabricated by using silicon or
polysilicon, it is a piezoelectric material in the case of BAW or SAW resonators.
Recent resonator applications implement other materials as well, like diamond or
carbon nanotubes (CNTs) [18, 19].
Among the elastic constants contributing to the resonator’s characterization,
we consider the elastic stiffness (or Young’s modulus), the mass density, and the
sound velocity. In order to extract these properties, the resonator’s low-frequency
mechanical resonance is estimated from the analytical model corresponding to its
mechanical configuration. Let’s consider the clamped-clamped beam resonator
depicted in Figure 5.13, which has the same geometrical configuration of most
FBARs.
5.2 Determination of Elastic, Dielectric, and Piezoelectric Constants 141

Electrode
Clamping (silicon)

AIN
Electrode
Figure 5.13 Clamped-clamped beam resonator with dimensions.

The resonance frequency of the beam depends on the thickness t and length l of
the beam. Assuming a structural layer of AlN with thin-metal electrodes, one can
calculate the frequency by [20]:

. 2 t
473 c
f0 = ⋅ (5.9)
2π l 2 12 ρ

The quotient of the stiffness constant c and the mass density ρ of the AlN can be
determined by measuring the fundamental resonance frequency f0. Assuming the
value of one of these constants, the other one can be extracted. Two independent
measurements performed on beams of different dimensions allow completion of the
evaluation of this quotient.
In our example, we take two beams with identical thicknesses t of 1 μm and dif-
ferent lengths l of 60 and 90 μm. Their experimental S21 parameters are plotted in
Figure 5.14 (the resonator’s layouts are observed in the insets). As observed in Fig-
ure 5.14(a), the 90-μm-long beam exhibits a fundamental frequency of 800 kHz,
whereas Figure 5.14(b) shows the 60-μm-long device resonating at a higher fre-
quency of 1.5 MHz. These results are in good agreement with the theoretical expec-
tations of the clamped beam model described by (3.2). Furthermore, if we build
finite element models (FEMs) of the beams and perform the modal analysis, we will
find out similar frequency values within small error.
In Table 5.1 we compare the values of the first resonance modes of these AlN
beams obtained by experimental, ANSYS, and analytical methods. Using these val-
ues and solving c and ρ in (3.5), the AlN stiffness and density values are estimated to
be between 180–220 GPa and 3.0–3.7 g/cm3, respectively. With these results at
hand, the sound velocity can be derived through the well-known relationship [21]:

c
υ= (5.10)
ρ

According to the extracted stiffness and density values, the sound velocity of
the AlN is calculated to be in the range of 7,000–8,500 m/s. As we reviewed in
142 Characterization Techniques

−60

S21mag (dB)
−66

−72

0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2

Frequency (MHz)
(a)
−74
−75
−76
−77
S21mag (dB)

−78
−79
−80
−81
−82
−83
1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0
Frequency (MHz)
(b)
Figure 5.14 Mechanical resonances of the beam-shaped FBARs in the insets: (a) length of 90 μm,
and (b) length of 60 μm (thickness is 1 μm for both devices).

Table 5.1 Low-Frequency Characterization of an FBAR, Comparing Experimental, Finite-

Element-Modeling (ANSYS) and Analytical Calculations for a Clamped-Clamped Beam Resonator
Device Experimental Finite-Element- Analytical (3.2)
(S-Parameter Data) Modeling (ANSYS )
Large-sized beam First mode First mode First mode
(width: 50 μm, (fundamental): 800 kHz; (fundamental): 877 kHz; (fundamental): 885 kHz
length: 90 μm, second mode: 1.3 MHz; second mode: 1.46 MHz;
thickness: 1 μm) third mode: 2.2 MHz third mode: 2.47 MHz
Small-sized beam First mode First mode First mode
(width: 50 μm, (fundamental): 1.5 MHz; (fundamental): 1.98 MHz; (fundamental): 1.9 MHz
length: 60 μm, second mode: 2.75 MHz second mode: 2.65 MHz
thickness: 1 μm)

Chapter 2, the acoustic velocity affects the high-frequency resonant behavior of

FBARs, although other electromechanical parameters must also be accounted for.
5.2 Determination of Elastic, Dielectric, and Piezoelectric Constants 143

5.2.2 Dielectric Constants

In some MEMS as well as in BAW and SAW resonators, the structural layer is a
dielectric thin film, whose properties we are interested in characterizing. Some test
structures, like parallel-plate capacitors sandwiching the dielectric layer and form-
ing a static capacitance C0, allow evaluating the permittivity ε of the dielectric. For
this purpose, two-probe measurements can be performed by using a capacitance
meter and a probe station to contact each electrode of the capacitor. The absolute
permittivity ε, also known as the dielectric constant, is defined as the dielectric dis-
placement per unit electric field. Using the same notation defined in Chapter 2, the
first subscript gives the direction of the dielectric displacement, and the second gives
the direction of the electric field. If the material is anisotropic, ε is a matrix with nine
elements of different values. In piezoelectric materials, ε11 and ε33 are particularly
interesting, and they are defined as follows:

• ε11T
[F/m] is the permittivity for the dielectric displacement and electric field in
direction 1 under conditions of constant stress.
• ε S33 [F/m] is the permittivity for the dielectric displacement and electric field in
direction 3 under conditions of constant strain.

Given that the static capacitance C0 formed by the electrodes and the dielectric
is measured, the permittivity can be evaluated by implementing the following char-
acterization procedure [22]:

1. Low frequency C0 measurement;

2. Dielectric thickness (t) measurement (by means of profilometer,
interferometer or SEM) [23];
3. Capacitor area (A): determined from the resonator’s layout;
4. Calculation of permittivity (ε):

t
ε11
T S
33
= εr ⋅ ε0 = C0 (5.11)
A

The evaluation of ε11

T
or ε S33 depends on the test structure we choose. For exam-
ple, if we have a c-axis-oriented AlN thin film we want to evaluate the ε S33 , which we
do by using parallel-plate FBARs of known electrode sizes. If the plate has an area of
2
200 × 140 μm and the thin film has a thickness of 1 μm, a theoretical C0 value of
2.21 pF is expected. From (5.11), a theoretical ε S33 value of 7.8 × 10−11 F/m should be
extracted.
Process and layout deviations may explain possible differences between experi-
mental and theoretical dielectric-constant values. First, if the thickness of the dielec-
tric is not uniform, C0 may vary from one device to another. Also, device
underetching may lead to capacitor areas bigger than expected to the design. Statis-
tical analysis and parametric FEM may complement the experimental means
described herein, which are good starting points for further refinement of the
dielectric constants.
144 Characterization Techniques

5.2.3 Piezoelectric Properties

5.2.3.1 The Piezoelectric Coupling Coefficient
The piezoelectric coupling coefficient is, together with the Q factor, the most impor-
tant parameter characterizing the electrical performance of acoustic resonators like
SMRs or FBARs. Strictly speaking, what is being characterized is the electrome-
chanical conversion capability of the acoustic layer of the resonator (e.g., AlN,
ZnO, or PZT). Furthermore, a mutual relationship exists between them, through
the figure of merit (FoM) of the resonator:

FoM = Q ⋅ kT2 (5.12)

This means that, for a given technology, the value of the process is given by the
combined Q by kT2 product, rather than by only one of them. Depending on the
application, it will be interesting to optimize one of the Q or the kT2 values (the other
one will diminish in the same proportion).
Since piezoelectric ceramics are anisotropic, their physical constants (elasticity,
permittivity, and so on) are tensor quantities. For this reason, as explained in Chap-
ter 2, the coupling coefficient kT2 of an FBAR can be measured in the longitudinal
(“33”) or transverse (“31”) wave propagation directions. Depending on the avail-
able test structures, one or both coefficients can be measured.

2
Longitudinal Coupling Coefficient k 33
2
The longitudinal coupling coefficient k 33 represents the electromechanical conver-
sion efficiency in the c-axis (“3”) when an electric field in the z-axis (“3”) is applied
to the piezoelectric. This is done by measuring the longitudinal-mode resonance fre-
quency (mode “33”) of an FBAR (or SMR), which is found in the 2-GHz band for a
typical FBAR process. The crystal orientation and thickness of the piezoelectric
layer mainly determine the coupling coefficient k233. As the thickness decreases, the
value of k233 increases, Q decreases, and the resonance frequency increases, and vice
versa. The expression to derive the value of k233 from experimental measurements is
[21]:

π 2 fs fp − fs π 2 fs ⎛ π fp − fs ⎞
2
k33 = = tan ⎜⎜ ⎟
⎟ (5.13)
4 fp fp 4 fp ⎝ 2 fp ⎠

In (5.13), fs and fp are series and parallel resonance frequencies of the FBAR or
SMR. Table 5.2 shows the k233 values for various devices implemented with AlN
thicknesses of 1,000 and 500 nm. While the resonance frequency augments for the
500-nm-thick AlN devices, k233 increases and Q decreases for similar values of the
FoM.
The theoretical limits for the magnitude of k233 in AlN are around 6.7–7.0%,
whereas typical Q factor values for commercial FBARs are between 700–2,000.
Thus, reference values for the FoM between 50 and 100 can be found. Average val-
ues of 3% for k233 and 10% for the FoM are relatively low in comparison to refer-
ence processes. The low efficiency of the electromechanical conversion is probably
due to inversion of the crystal’s poling domain [24]. Annealing or other techniques
5.2 Determination of Elastic, Dielectric, and Piezoelectric Constants 145

Table 5.2 Coupling Coefficient k 33 Characterization for

2

Different Devices and Thicknesses: The Q Factor Changes in

2
an Inverse Proportion to k 33, for Similar Values of the FoM
AlN
2
Device k33 (%) Q Factor FoM Thickness (nm)
Type I 2.42 731.5 17.72 1,000
Type II 2.53 519.0 12.34 1,000
Type III 0.85 1,517 12.90 1,000
Type IV 2.63 587 15.44 500
Type V 6.46 156.6 10.12 500
Type VI 3.45 276.3 9.54 500

intended for AlN-deposition refinement would contribute to future improvements

of the crystal’s quality.

2
Transverse Coupling Coefficient k 31
2
The k 31 coupling coefficient measures the electromechanical conversion efficiency
of a c-axis-oriented (“3”) piezoelectric when an electric field is applied along the x
axis (“1”). The measurement technique is basically the same previously described
for the k233 characterization, although a surface wave test structure, like a SAW res-
onator, is required. Again, the electromechanical coupling is a function of the sur-
face-wave series and parallel resonance frequencies f ST and f PT of the SAW device:

π 2 fS fP − fS
T T T
2
k31 = (5.14)
4 f PT f PT

Taking the AlN process of the previous example, a SAW device exhibiting series
and parallel resonance frequencies of 211.75 MHz and 212.15 MHz allows the
extraction of an effective k231 value of 0.47%. This value is between one-fourth and
one-fifth of the k233 value, which is normal for c-axis-oriented piezoelectric films.
Both k231 and k233 are utilized in the extraction of the piezoelectric constants of the
thin film, which is explained in the following item.

5.2.3.2 Piezoelectric Constants

Piezoelectric constants of a fabricated material are process-dependent physical con-
stants explaining the electromechanical performance of the crystal. Although refer-
ence values are available for commercially available material compounds, like AlN,
specific crystal characterization has to be performed for a newly developed process.
The piezoelectric constants can be expressed in voltage-to-strain (g), voltage-
to-stress (d), charge-to-stress (e), and charge-to-strain (d) forms. Linear relation-
ships exist between the different forms. Thus, the characterization of one form
allows the knowledge of the other ones. In this section, the d-form of the
piezoelectric constant is studied.
The piezoelectric constant d is defined as the electric polarization induced on a
material per unit mechanical stress applied to it. Alternatively, it is the mechanical
strain experienced by the material per unit electric field applied to it. The first sub-
146 Characterization Techniques

script refers to the direction of polarization at zero-electric field (E = 0), or to the

applied field strength. The second one refers to the direction of the applied stress, or
to the direction of the induced strain. Two relevant components of the d-constant
are as follows:

1. d33 [m/V] is the induced polarization the direction of Z-axis (“3”) per
applied unit stress in the same direction. Alternatively, it is the induced
strain per unit electric field applied in the same direction.
2. d31 [m/V] is the induced polarization in the direction of the Z-axis (“3”) per
unit stress applied in the direction of the X-axis (“1”). Alternatively, it is the
strain induced in direction 1 per unit electric field applied in direction 3.

The values of d33 and d31 can be extracted through different methods. One
method performs an experimental measurement of the mechanical displacement of
the film when an electric potential is applied to the electrodes. The measurement of
the low absolute level of the displacements in thin films necessitates the use of a pre-
cise interferometer technique. Typically, a resolution of about 10−2 Å and a complex
measurement setup—including a sensitive double-beam interferometer—are
required to determine the low-field piezoelectric coefficients [25]. For that reason,
an indirect evaluation is preferred, by calculating the values of d33 and d31 from pre-
viously extracted electromechanical constants.
In previous sections, FBAR and SAW resonators were useful to evaluate
permittivity, stiffness, and coupling-coefficient constants. These constants can be
used to obtain the piezoelectric charge constants from piezoelectric constitutive
equations:
2
d 33 ⋅ c 33
2
k33 = (5.15)
ε T33

2
d 31 ⋅ c 11
2
k31 = (5.16)
ε T33

Solving the right-sided parts of (5.15) and (5.16) and using previously extracted
parameters, the d-constants are calculated. Using the parameters of the previous
example, the values of d33 and d31 are 2.85 and 1.12 pm/V. These magnitudes are
roughly equal to half the value of previously reported epitaxial AlN films [26]. The
higher deposition temperatures of epitaxial processes partially explain the better
quality of the films obtained with this process, in comparison to sputtered-AlN
films. On the other hand, the d33/d31 ratio is 2.56, in the same order of the theoretical
relationship for clamped wurtzite structures (d33/d31 = 2) [27].

5.3 Equivalent-Circuit-Parameter Extraction

Previously, in Chapter 3, the BVD equivalent-circuit-parameter model of FBARs

and other MEMS resonators was introduced. Electrical modeling and extraction of
equivalent-circuit parameters in FBARs are well documented, the lumped-element
5.3 Equivalent-Circuit-Parameter Extraction 147

MBVD being the most accepted model for parameter extraction purposes [28]. The
extraction of the model’s parameters is carried out by evaluating and averaging the
resistive and reactive components of the S-parameters evaluated at different points,
in a frequency span where the resonance is expected to be found. Some optimization
processes can be implemented in the extraction of the parameters, like
least-mean-squares fitting. Additional dissipative elements due to substrate cou-
pling may also be accounted for in the model, or they can be omitted if on-the-wafer
calibration or de-embedding is performed. These dissipative elements are of great
interest, since they could explain insertion losses and signal drifting from the RF sig-
nal paths connecting the resonator to the substrate wafer. An accurate resonator’s
model is particularly important to design MEMS-to-CMOS integrated applications.
The MBVD model and the setup for parameter extraction are shown in the schema
and optical image of Figure 5.15.
The optimization of parameter extraction in other RF devices is well known:
extensive work has been done to extract, from measured S-parameters, the equiva-
lent-circuit parameters of microwave resonators [29], FET transistors [30, 31], and
filters [32]. In most of these cases, a least-squares strategy is adopted to optimize the
extraction of the circuit parameters. In this section, we expand this discussion by
studying a multistep procedure implementing a least-squares optimization strategy
for the extraction of equivalent-circuit elements of the resonator and of the sub-
strate carrying it. In this way, on-wafer calibration is avoided [28], allowing
model-based de-embedding of the MBVD parameters.

5.3.1 Parameter-Extraction Algorithm

Assuming a given resonator’s model, a least-square-based extraction algorithm
aims to estimate the optimal values of one or more error functions, which describe
the transfer function of the resonator by its equivalent circuit parameters. The
extraction can be done in an iterative way, processing each one of the iterations
either in a single step or in a multiple substep basis, the latter being more accurate in
complex or selective functions. In this case and for a given iteration, each substep in
the algorithm extracts a subset of one or more of the equivalent-circuit parameters.
This implies that different error functions are defined and are to be optimized. Prior
to this, the S-parameters’ sensitivity to the different circuit elements has to be ana-
lyzed, in order to determine the most adequate function to be used in the extraction
of each circuit element.
To illustrate this, let’s consider the MBVD FBAR’s model depicted in Figure
5.15(a). Thorough exploration of the model allowed us to find out that some equiv-
alent-circuit elements are more sensitive to the reflection parameter S11, whereas
others are more sensitive to the transmission parameter S21. Also, the execution
sequence of the substeps has to be studied, with the aim of achieving acceptable
numerical stability and convergence rates. Table 5.3 presents a summary of this
study, where seven different error functions involve the equivalent-circuit parame-
ters of both the FBAR and its substrate. This means that each one of the iterations
will be comprised by the execution of each one of these functions on a seven-substep
basis.
148 Characterization Techniques

Ti/Pt

PAD Lm
Rp PAD
AIN
Rm
L S /2 R S /2 C0 R S /2 L S /2
Cm
Ti/Pt Ti/Pt

C OX SiO 2 SiO 2 C OX

Air gap

C sub Rsub C sub Rsub

Substrate (Si)

(a)

(b)
Figure 5.15 MBVD model of FBAR, substrate, transmission line, and characterization pads: (a)
equivalent circuit representation and (b) characterization setup.

The number of iterations can be set in two ways: first, through an internal toler-
ance value of the error-function value, and, second, by user’s setting as a running
parameter at the time of execution of the algorithm. The execution of each substep
may be controlled by various optimization criteria. Since each substep is also an
iterative routine by itself, one of the optimization criteria may be the number of
local execution cycles (for the substep, independentof the global number of itera-
tions). Other optimization criteria may be the tolerance of the optimized variable or
the tolerance of the error-function value. Around 10 global iterations are typically
enough to achieve good convergence of the equivalent-circuit parameters in a
standard Pentium processor–based PC platform.
The flow diagram of Figure 5.16 illustrates the sequence of the parameter
extraction. As shown, a given iteration cycle may begin with the optimization of Rm.
The remaining elements in the equivalent-circuit are then set to fixed values—the
5.3 Equivalent-Circuit-Parameter Extraction 149

Table 5.3 Error Functions of the Multisubstep Parameter-

Extraction Algorithm on FBARs
Circuit-Elements S-Parameter
Substep Optimized Fitted Error Function
1 Rm S11 S11mag(Rm)
2 Lm, Cm, C0 S21 S21phase(Lm, Cm, C0)
3 Rs S11 S11mag(Rs)
4 Rp S21 S21mag(Rp)
5 Rsub S21 S21mag(Rsub)
6 Csub S21 S21phase(Csub)
7 Cox S21 S21phase(Cox)
Source: [33].

Start

Load experimental data

S11, S21

Search f s , f p , S11min , S21min

First estimate:
Rm, Lm, Cm, C0

Generate initial-condition
vectors

Return final values:

R m, L m, Cm C0 , R p , Rs,
R sub , Csub, C ox

End

Figure 5.16 Flow diagram of an FBAR’s equivalent-circuit parameter extraction algorithm.

ones obtained in the previous iteration cycle. In the second substep, the L-C-C0
ensemble is optimized by implementing the second error-function S21phase(Lm, Cm,
C0), while Rm and the other model elements remain fixed to constant values, and so
forth. After the substep is completed, the algorithm proceeds in the same way to
execute the following steps until the last substep. In that moment a new global cycle
starts, and the foregoing process is repeated until the last iteration is completed.
150 Characterization Techniques

Additional routines may also be implemented in order to complete the fully auto-
mated parameter-extraction procedure, including loading of S-parameter ASCII
data and first estimation of the BVD parameters [2], among others.

5.3.2 Case Study: Equivalent-Circuit-Parameter Extraction of an FBAR

The optical image of Figure 5.17(a) shows the top view of a rectangular-shaped
FBAR with annotated model elements. The device was fabricated according to the
RIE-based process described in Chapter 4. The transmission line is made of Ti/Pt
and has a thickness of 30 nm/150 nm, a width of 50 μm, a length of 250 μm, and
theoretical Rs/2 and Ls/2 in the range of 8–12 ohms and 90–140 fH, respectively.
The dark area around the resonator is SiO2 after Si underetching (an equivalent area
can be found underneath the FBAR). Figure 5.17(b) plots experimental and extrac-
tion results where the transmission spectrum (S21) magnitude is shown in a span of
1.5 GHz. These curves compare the BVD, MBVD, and MBVD with parasitic ele-
ment models. The extracted Rs/2 and Ls/2 values are 12.25Ω and 34 fH, respec-
tively, which are of the same order of magnitude as those expected from the
theoretical analysis.

(a)

−5
S21mag (dB)

−10

−15
1.5 2.0 2.5 3.0
Frequency (GHz)

(b)
Figure 5.17 Experimental results on the extraction of FBAR’s parameters: (a) exemplary FBAR
with annotated model elements; and (b) fitting and experimental curves (BVD and MBVD curves
are also plotted). (After: [34].)
5.4 AFM, Optical, and Electron-Beam-Induced Characterization 151

Concerning the substrate of FBAR, an 800-nm-thick thermal-SiO2 layer leads to

a theoretical value of Cox of around 430 fF, whereas the extracted value is 647 fF.
On the other hand, theoretical calculations for our low-resistivity 500-μm-width Si
wafers result in values in the range of 2,000–10,000Ω and 2 fF for resistance Rsub
and capacitance Csub, respectively. These values are of the same order as the
extracted values obtained by the algorithm, −150Ω and 1 fF, respectively. The dif-
ferences in the extracted and theoretical values of Rsub may be explained by a lower
resistivity value of the wafer, with respect to the expected doping configuration.
Further characterization of this process should allow refinement of this parameter.
Whatever the case, the extracted values are consistent with the properties of a
low-resistivity substrate.
Finally, the extracted BVD model values are also closely related to experimental
observation. In particular, for the testing device, the measured coupling coefficient
k332 of 2.32% is very close to 2.22%. The latter is calculated from the extracted val-
ues of Lm, Cm, C0; 435.5 nH, 11.1 fF, and 605.2 fF, respectively. The 10-Ω and
8.5-Ω values of Rm and Rp, respectively, are also in the range of those previously
reported for resonators [33–35].

5.4 AFM, Optical, and Electron-Beam-Induced Characterization

Nonelectrical characterization techniques have been largely employed to character-

ize physical systems. In this section, we also discuss AFM and optical interferometer
techniques.

5.4.1 AFM-Based Characterization with Optical Detection

AFM is a nanometer-scale-resolution tool able to measure the static and low-fre-
quency mechanical parameters of some MEMS resonators, especially those with a
cantilever shape and a frequency response between units of hertz to few megahertz.
More recently, AFM has been used to perform high-frequency scanning of
gigahertz-mode resonators like FBARs. Next, we discuss both characterization
approaches.

5.4.1.1 Low-Frequency Mechanical Resonance of MEMS

In standard tapping-mode AFM measurements, a cantilever with a tip scans the sur-
face of the sample after piezoelectric actuation of a device chip placed on the head of
the AFM system, as shown in Figure 5.18(a). A piezoelectric actuator excites the
cantilever substrate vertically, causing the cantilever to move up and down. In its
encounter with the sample’s surface, the tip located on the cantilever’s vertex
deflects. The reflected laser beam reveals information about the vertical height of
the sample surface and material’s characteristics. This setup, however, can be modi-
fied to perform MEMS resonator measurements. In the new system, depicted in Fig-
ure 5.18(b), the MEMS resonator replaces the AFM cantilever and is driven by the
AFM’s piezoelectric actuator and a signal generator.
152 Characterization Techniques

(a)

(b)
Figure 5.18 AFM-based MEMS resonator characterization: (a) standard SFM setup for sample
characterization; and (b) modified setup for MEMS resonator characterization.

In the resonance measurements, the AFM is configured to work in tapping

mode. Thus, the AC signal of the generator feeds the piezoelectric actuator, which is
caused to move the cantilever to the extent of its natural modes and the transduced
energy by the piezoelectric material. As the cantilever moves vertically, the reflected
laser beam, or return signal, deflects in a regular pattern over a photodiode array,
generating a sinusoidal, electronic signal. Such signal is postprocessed by fre-
quency-analysis software to extract the resonance peaks and their quality factors
within the frequency span [36].
Calibration and evaluation of the spring constants of MEMS cantilevers can
also be performed with a similar setup. There exist a number of techniques to deter-
mine the spring constant of atomic force microscope (AFM) cantilevers. These
methods can be divided into four categories: dynamic, theoretical, static, and indi-
rect. The dynamic methods rely on analyzing the resonance response of the cantile-
ver. Some of these methods require knowledge on the cantilever’s material
properties and dimensions, whereas others involve analysis of the lever’s thermal
noise spectrum. The theoretical methods involve the use of simple expressions that
require accurate knowledge of the cantilever’s Young’s modulus and dimensions. In
the static response methods, a known force is applied to an AFM cantilever to deter-
mine the spring constant. The known force can be supplied by a small added mass, a
reference lever or artifact, or an indentation device. Indirect methods have also been
employed. These techniques require that one cantilever on each chip is calibrated
accurately, and then the spring constant of the other levers on the same chip is
calculated indirectly [37].
5.4 AFM, Optical, and Electron-Beam-Induced Characterization 153

5.4.1.2 High-Frequency Dispersion Amplitude of FBARs

High-frequency modes of MEMS and FBAR can also be supervised by AFM tech-
niques. Reference [38] presents a setup based on the combination of an AFM with a
lock-in detection technique to measure displacement amplitude of an FBAR.
For example, the desired main thickness vibration mode of FBAR resonators is
observed in the gigahertz band, where lateral resonant modes are also excited
around this frequency [39]. These modes exhibit very small displacement ampli-
tudes of about units of nanometers due to the piezoelectric effect. Hence, the AFM
scans the FBAR’s surface in order to evaluate the mode shaping around the main
resonance frequency. The characterization procedure implemented by San Paulo et
al. [38] requires electrical characterization of the FBAR, prior to the AFM experi-
ments (to estimate the resonance range and also for comparison purposes). The
measurement setup of Figure 5.19 illustrates how the AFM tip is placed in contact
with the top electrode of the FBAR while a sinusoidal amplitude-modulated RF sig-
nal is used to drive the resonator. The RF oscillation amplitude is then measured by
using the modulation signal as the reference for the lock-in amplification of the can-
tilever deflection signal. The experiments require the use of AFM cantilevers with a
high resonance frequency (> 20 kHz) and a low stiffness constant (< 1 N/m). Conse-
quently, we have the selection of a particular modulation frequency, near but lower
than the cantilever’s resonance frequency. The RF driving signal has to be chosen in
the range of the FBAR resonance frequency previously obtained in the electrical
characterization. To analyze the topography, or equivalently the mode shaping
maps, or wavelength patterns, the two-dimensional Fourier transform of the ampli-
tude is calculated over a selected scanning line on the device geometry. The exis-
tence of a uniform thickness vibration mode at some driving frequency is checked as
well. Finally, the average oscillation amplitude along the selected scan line against
the line frequency is plotted.
Line-scanning the whole surface allows obtaining 2D mode shaping maps of the
FBAR electrode at the thickness-mode frequency and the near lateral-mode frequen-
cies. The method also evaluates the vibration amplitude as a function of the modu-
lation and driving frequencies.
Another measurement technique of the resonator’s displacement and mode
shaping describes how images are recorded at normal video frame rates. This is
done by using dynamic holography with photorefractive interferometric detection,

Figure 5.19 AFM and lock-in detection based experimental setup. (After: [38].)
154 Characterization Techniques

coupled with microwave electrical impedance measurements [40]. The ability of the
photorefractive technique to process optical interference over an extended area
eliminates the need for rapid serial scanning of smaller size. The experimental setup
includes a coherent solid state laser source (532 nm), a microwave generator, a
photorefractive crystal, an electro-optic modulator (EOM), a photodetector or
charge-coupled-device (CCD) camera, mirrors, lenses and optical beams.

5.4.2 Optical Microscope Interferometry with Piezoelectric Actuation

Interferometry is another characterization method employed in both static and
dynamic resonator characterization. The method is performed by an optical micro-
scope with objectives featuring light-interference capabilities at a specific wave-
length λ. Interferometry takes advantage of the interference produced when the light
passing through an object is caused to interfere with a reference beam of light that
has followed a somewhat different pathway. Under these circumstances, the opaque
surfaces of a reflected light specimen or transparent specimens are imaged when the
path difference between the two light beams is converted into intensity fluctuations.
Thus, interference bands appear on the structure’s surface when the sample is
placed on the microscope’s stage and after appropriate focusing [41].
In static measurements, the levels of flatness or curvature of the resonator’s
structure can be analyzed through this method. By studying the orientation of the
interference bands, the type of curvature—concave or convex—can be inferred. If
the bands displace to the clamping zone after moving the sample away with respect
to a reference observation point, the structure is down-tilted, or it has a concave
shape. Otherwise, the sample is up-tilted, or it is convex. Bands appearing simulta-
neously across the surface of the resonator indicate that it is highly flat. The optical
image of Figure 5.20 shows the interferometer bands of a MEMS cantilever whose
central area is still clamped to the substrate (due to insufficient etching time). Look-
ing at its interference bands, this device has a concave shape, which is explained by
the difference of residual stress between the released and the nonreleased areas.
In dynamic characterization, interferometry is very useful for characterizing the
frequency response of the resonator and its main mechanical parameters, like the
resonance frequency, its vibration amplitude, and the quality factor. Apart from the
microscope, the measurement setup comprises an AC signal generator and a com-
mercially available piezoelectric actuator [42] whose frequency response is previ-

Figure 5.20 Interferometer bands of a resonating MEMS cantilever.

5.4 AFM, Optical, and Electron-Beam-Induced Characterization 155

ously known and supposed to cover the resonator’s frequency range of interest. Due
to the piezoelectric effect, the AC signal provided by the generator induces an AC
displacement of the piezoelectric actuator, which is mechanically attached to the
resonator’s die. Due to this coupling (and if the frequency of the signal generator
approaches one of the natural modes of the resonator), it goes into resonance. When
the resonance starts up, a group of zero-contrast bands become visible. In this way,
the resonance frequency is related to the highest number of zero-contrast bands over
the resonator’s surface. Figure 5.21 shows out-of-resonance and in-resonance inter-
ferometer images of another MEMS resonator. The number of zero-contrast bands
is determined by the implemented wavelength λ of the microscope’s objective and
helps to estimate the vibration’s amplitude A, through first-order Bessel functions
relating both variables [43]. The images of Figure 5.21 show the interferometry
bands on the surface of a 1,000 × 1,000 μm2 MEMS cantilever Si resonator with
thickness of 5 μm. In the static resonator image of Figure 5.21(a), the interference
bands are seemingly observed, while the cantilever resonating at 4 kHz shows a dif-
fuse image with four zero-contrast bands. The images were taken at λ = 546 nm. The
wavelength leads to a resonating amplitude of about 500 nm [44].
Although practical and useful, it should be said that interferometry entails some
limitations. Due to the optical nature of the measurement, only rough approxima-
tions of the quality factor’s values can be obtained. Thus, calibration and evaluation
of the 3-dB value amplitudes are not straightforward tasks and may lead to consid-
erable error. Another drawback of this technique is the impossibility to state and
compare the energy values delivered to the resonator and transduced by the excita-
tion system at the different vibration modes. These limitations, however, can be
overcome with postprocessing software and automated calibration routines.

5.4.3 Fabry-Pérot Interferometry

Fabry-Pérot interferometry is an optical technique based on a structure implement-
ing reflective surfaces. They selectively transmit and reflect wavelengths of the inci-
dent light depending on the dimensions, incidence angle, and refractive index of the

(a) (b)
Figure 5.21 Interferometer-based MEMS resonator characterization: (a) static resonator interfer-
ometer image; and (b) cantilever resonating at 4 kHz (zero-contrast bands are observed). (After:
[44].)
156 Characterization Techniques

interferometer [45]. The structure can be a transparent plate with two reflecting sur-
faces—also known as an etalon—or it can be two parallel highly reflecting mir-
rors—the interferometer. The spectrum of the interferometer exhibits transmission
peaks as a function of wavelength corresponding to resonances of the etalon or
interferometer structures, which is the base of Fabry-Pérot characterization of
MEMS and NEMS resonators.
The technique can be used in two ways: in a Fabry-Pérot interferometer the dis-
tance l between the plates can be tuned in order to change the wavelengths at which
the transmission peaks occur. On the other hand, the Fabry–Pérot etalon has a fixed
size—the thickness of the MEMS or NEMS resonator—and the characterization
aims for finding out which wavelengths are transmitted (corresponding to the thick-
ness of the resonator). Due to the angle dependence of the transmission, the peaks
can also be shifted by rotating the etalon with respect to the beam. The geometry of
the Fabry-Pérot etalon is depicted in Figure 5.22.
How much of the light incident to the etalon is effectively transmitted? The
answer relies on the transmission function TF of the etalon and the interference
between the multiple light reflections between its two reflecting surfaces. Depending
on whether or not the transmitted light beams Tm are in phase, constructive or
destructive interference occurs. In the case of constructive interference a high-trans-
mission peak of the etalon is detected. Out-of-phase transmitted beams are
destructed, thus corresponding to a transmission minimum. The phase conditions
are thus dependent on the wavelength l of the light, the traveling angle through the
etalon θ, and the thickness and refractive index n of the etalon. The phase shift δ
for constructive interference is given by:

δ = 2 κl cos θ (5.17)

where κ is the wave number inside the etalon:

2 πn
κ= (5.18)
λ

Reflected
light

Transmitted
light

Incident
light beam

Figure 5.22 Fabry-Pérot interferometer (etalon): the incident light beams are selectively transmit-
ted according to their wavelength and angle of incidence, and the thickness l and refractive index
n of the etalon.
5.4 AFM, Optical, and Electron-Beam-Induced Characterization 157

As the drawing of Figure 5.22 suggests, the amplitude of the transmitted beams
AT is the sum of the individual amplitude beams Tm:
∞
AT = ∑T
m= 0
m (5.19)

Assuming no absorption of the etalon, the energy of the transmitted and

reflected light is conserved. The condition T R = 1 is met, where T and R are the
transmission and reflection magnitudes of the beams. The first transmitted beam T0
results from direct transmission through the etalon, while the second beam T1 is a
phase-delayed version of T0, and doubly attenuated by a factor R2 due to the two
reflections inside the etalon. By considering the m reflections of the (m 1)th beam,
(5.19) can also be expressed in terms of the phase shift, and the transmission and
reflection amplitudes by:
∞
A T = T ∑ R m e jmδ (5.20)
m= 0

The amplitude sum of (5.20) can be expressed in an analytical form as the solu-
tion of the series by:

T
AT = (5.21)
1 − Re jδ

Finally, the energy of the transmission function TF equals the product AT A*T as:

T2
TF = A T A*T = (5.22)
1 + R − 2R cos δ
2

The frequency spectrum of TF gives us the transmitted-peak wavelengths of the

incident light.
Fabry-Pérot interferometers or etalons are used in optical modems, spectros-
copy, lasers, and dichroic filters to measure the wavelengths of incident light
[46–48]. As a characterization tool, it is used in MEMS and NEMS resonator char-
acterization to detect the resonance frequencies of the device. Since MEMS and
NEMS resonators are relatively thin and transparent—thickness comparable with
the wavelengths of incident light—part of the light is transmitted through the struc-
ture and is reflected by the structural layer. Indeed, MEMS and NEMS resonator act
as Fabry-Pérot interferometers. This method was first used by Carr et al. to detect
the vibration of NEMS resonators [49]. A general disadvantage of the technique is
that the laser beam locally increases the temperature of the structure when
illuminating it.

5.4.4 Electron-Beam Excitation

Electron-beam-induced excitation of NEMS resonators can be used to elicit vibra-
tion of the resonator. The technique is based on quasi-electrostatic charges induced
on the surface of very thin dielectric resonators by the action of electrons in vacuum
158 Characterization Techniques

environment. The NEMS is assumed to be inside the chamber of a scanning electron

microscope (SEM), a transmission electron microscope (TEM), or similar equip-
ment where electron beam imaging and high vacuum can be available.
In a first setup, the electron beam of the SEM or TEM illuminates the sample.
Another mechanism for inducing the charges involves a conducting sharp probe
connected to an external signal generator. As the probe approaches the NEMS at
enough distance to generate charge accumulation, time-varying charges generate
electromotive forces driving the NEMS into resonance. If we know the driving fre-
quency of the signal generator, and we observe the vibration, the NEMS resonance
frequency can be extracted.
Required forces make the mechanism suitable for resonator characterization at
the NEMS scale. Under appropriate conditions, nanocantilever, nanotube and
nanowire resonators [19, 50], and nanooscillators [51] have been characterized.

5.5 Summary

In this chapter, we have learned many lessons regarding the characterization of

FBAR, MEMS, and NEMS resonators. First, we studied the basis of low- and
high-frequency characterization. Different characterization setups have been
described, and layout considerations affecting the performance of FBARs have been
commented. Based on the electrical characterization method, we have seen how the
main elastic, dielectric, and piezoelectric constants of the resonator are extracted,
and we have seen how the quality factor and the electromechanical coupling
coefficients are extracted.
The electrical characterization also supports the extraction of the equiva-
lent-circuit parameters of FBAR and MEMS. This task is particularly useful for
designing MEMS-based integrated circuit applications, as we will study in the next
chapters. Although many methods exist for extracting these parameters, we have
focused on explaining a least-mean-square-based algorithm to optimize the extrac-
tion. By this algorithm, on-wafer calibration can be avoided, thus achieving instan-
taneous de-embedding of the resonator’s equivalent-circuit parameters. Of course,
the performance of such a tool can be improved with classical test structures carry-
ing out on-the-wafer calibration and parameter de-embedding.
NEMS characterization techniques are being developing to deal with
submicron oscillation detection. AFM measurement, Fabry-Pérot interferometry,
and electron-beam-induced resonance elicitation are suitable for structural and
vibration analysis of NEMS resonators. They are replacing well-developed electri-
cal techniques that cannot be implemented at the nanometer scale or due to the
device configuration. Such techniques are increasingly leaving the laboratory to
become part of robust applications where in situ characterization is required [52],
and they are expected to deal with more application fields in the near future.

References

[1] Davis, W. A., and K. K. Agarwal, Radio Frequency Circuit Design, New York: John Wiley
& Sons, 2001.
5.5 Summary 159

[2] Su, Q., et al., “Thin-Film Bulk Acoustic Resonators and Filters Using ZnO and Lead–Zirco-
nium–Titanate Thin Films,” IEEE Trans. on Microw. Theory Tech., Vol. 49, 2001,
pp. 769–778.
[3] Rohde & Schwartz International Network Analyzers, http://www2.rohde-schwarz.com.
[4] Süss Microtec Probe Systems, http://www.suss.com.
[5] Cascade Microtech RF Microwave Probes, http://www.cmicro.com.
[6] Agilent Technologies, Mechanical and Electronic Calibration Kits (Microwave Mechanical
Calibration Kits & ECal Modules, up to 110 GHz): http://www.agilent.com.
[7] Rogers Corporation, “RO3000 Series High Frequency Circuit Materials,”
http://www.rogerscorporation.com/mwu/pdf/3000data.pdf.
[8] Ylilammi, M., et al., “Thin Film Bulk Acoustic Wave Filter,” IEEE Trans. on Ultrason.
Ferroelectr. Freq. Control, Vol. 49, 2002, pp. 535–539.
[9] Razhavi, B., RF Microelectronics, Upper Saddle River, NJ: Prentice-Hall, 1998.
[10] Kaitila, J., et al, “Spurious Resonance Free Bulk Acoustic Wave Resonators,” Proc. IEEE
Intl. Ultrason. Symp. 2003, Honolulu, HI, October 5–8, 2003, pp. 84–87.
[11] Lopez, J. L., et al., “Integration of RF-MEMS Resonators on Submicrometric Commercial
CMOS Technologies,” J. Micromech. Microeng., Vol. 19, 2009, 015002.
[12] Avago Technologies, http://www.avagotech.com/pages/en/rf_for_mobile_wlan_mmw/
fbar_filters/fbar.
[13] Infineon Technologies, http://www.infineon.com.
[14] Ruby, R., et al., “Ultra-Miniature High-Q Filters and Duplexers Using FBAR Technology,”
Digest of Tech. Papers IEEE Intl. Solid-State Circuits Conf. ISSCC 2001, San Francisco,
CA, February 5–7, 2001, pp. 120–121.
[15] Lakin, K. M., et al., “Bulk Acoustic Wave Resonators and Filters for Applications Above 2
GHz,” Digest of Tech. Papers IEEE Intl. Microwave Symp. MTT-S 2002, Vol. 3, Seattle,
WA, June 3–7, 2002, pp. 1487–1490.
[16] Kim, K. W., et al., “Resonator Size Effects on the TFBAR Ladder Filter Performance,”
IEEE Microw. Wirel. Compon. Lett., Vol. 13, 2003, pp. 335–337.
[17] Ruby, R., “Review and Comparison of Bulk Acoustic Wave FBAR and SMR Technology,”
Proc. IEEE Intl. Ultrason. Symp. 2007, New York, October 28–31, 2007, pp. 1029–1040.
[18] Gaidarzhy, A., et al., “High Quality Factor Gigahertz Frequencies in Nanomechanical Dia-
mond Resonators,” Appl. Phys. Lett., Vol. 91, 2007, 203503.
[19] Lassagne, B., et al., “Ultrasensitive Mass Sensing with a Nanotube Electromechanical Reso-
nator,” Nano Lett., Vol. 8, 2008, pp. 3735–3738.
[20] Lin, Y. W., et al., “Series-Resonant VHF Micromechanical Resonator Reference Oscilla-
tors,” IEEE J. Solid State Circuits, Vol. 39, 2004, pp. 2477–2491.
[21] ANSI/IEEE Std. 176-1987, “IEEE Standard on Piezoelectricity,” Institute of Electrical and
Electronics Engineers, New York, 1988.
[22] Alexander, T. P., et al., “Dielectric Properties of Sol-Gel Derived ZnO Thin Films,” Proc.
Tenth IEEE Intl. Symp. on Applications of Ferroelectrics Ferroelectrics ISAF’96, Vol. 2,
East Brunswick, NJ, August 18–21, 1996, pp. 585–588.
[23] Makkonen, T., et al., “Estimating Materials Parameters in Thin-Film BAW Resonators
Using Measured Dispersion Curves,” IEEE Trans. on Ultrason. Ferroelectr. Freq. Control,
Vol. 51, 2004, pp. 42–51.
[24] Sanz-Hervás, A., et al., “Degradation of the Piezoelectric Response of Sputtered C-Axis
AlN Thin Films with Traces of Non-(0002) X-Ray Diffraction Peaks,” Appl. Phys. Lett.,
Vol. 88, 2006, 161915.
[25] Kholkin, A. L., et al., “Piezoelectric Characterization of Pb(Zr,Ti)O3 Thin Films by
Interferometric Technique,” Proc. 10th IEEE Intl. Symp. on Applications of Ferroelectrics
Ferroelectrics ISAF’96, Vol. 1, East Brunswick, NJ, August 18–21, 1996, pp. 351–354.
160 Characterization Techniques

[26] Tsubouchi, K., K. Sugai, and N. Mikoshiba, “AlN Material Constants Evaluation and SAW
Properties on AlN/Al2O3 amd AlN/Si,” Proc. IEEE Intl. Ultrason. Symp. 1981, Vol. 1,
Chicago, IL, October 14–16, 1981, pp. 375–380.
[27] Berlincourt, D., H. Jaffe, and L. R. Shiozawa, “Electroelastic Properties of the Sulfides,
Selenides, and Tellurides of Zinc and Cadmium,” Phys. Rev., Vol. 129, 1963,
pp. 1009–1017.
[28] Larson III, J., et al., “Modified Butterworth-Van Dyke Circuit for FBAR Resonators and
Automated Measurement System,” Proc. IEEE Intl. Ultrason. Symp. 2000, San Juan,
Puerto Rico, October 22–25, 2000, pp. 863–868.
[29] Wheless, W., and D. Kajfez, “Microwave Resonator Circuit Model from Measured Data
Fitting,” Digest of Tech. Papers IEEE Intl. Microwave Symp. MTT-S 1986, Baltimore,
MD, June 2–4, 1986, pp. 681–684.
[30] Kondoh, H., “An Accurate FET Modeling from Measured S-Parameters,” Digest of Tech.
Papers IEEE Intl. Microwave Symp. MTT-S 1986, Baltimore, MD, June 2–4, 1986,
pp. 377–380.
[31] Kompa, G., and M. Novotny, “Highly Consistent FET Model Parameter Extraction Based
on Broadband S-Parameter Measurements,” Digest of Tech. Papers IEEE Intl. Microwave
Symp. MTT-S 1992, Albuquerque, NM, June 1–5, 1992, pp. 293–296.
[32] Seyfert, F., et al., “Extraction of Coupling Parameters for Microwave Filters: Determina-
tion of a Stable Rational Model from Scattering Data,” Digest of Tech. Papers IEEE Intl.
Microwave Symp. MTT-S 1992, Philadelphia, PA, June 8-13, 2003, pp. 25–28.
[33] Campanella, H., et al., “Automated On-Wafer Extraction of Equivalent-Circuit Parameters
in Thin-Film Bulk Acoustic Wave Resonators (FBAR) and Substrate,” Microwave Optical
Tech. Lett., Vol. 50, 2008, pp. 4–7.
[34] Campanella, H., et al., “Instantaneous De-Embedding of the On-Wafer Equivalent-Circuit
Parameters of Acoustic Resonator (FBAR) for Integrated Circuit Applications,” Proc. 20th
Symp. on Integrated Circuits and Systems Design SBCCI’07, Rio de Janeiro, Brazil, Sep-
tember 3–6, 2007, pp. 212–217.
[35] Campanella, H., et al., “Automated On-Wafer Characterization in Micro-Machined Reso-
nators: Towards an Integrated Test Vehicle for Bulk Acoustic Wave Resonators (FBAR),”
Proc. IEEE Intl. Conf. Microelectronic Test Structures ICMTS 2007, Tokyo, Japan, March
19–22, 2007, pp. 157–161.
[36] Digital Instruments, Dimension 3100 SPM AFM measurement system, Dimension 3100
Manual, Rev. B, 2009.
[37] Gibson, C. T., D. A. Smith, and C. J. Roberts, “Calibration of Silicon AFM Cantilevers,”
Nanotechnology, Vol. 16, 2005, pp. 234–238.
[38] San Paulo, A., X. Liu, and J. Bokor, “Atomic Force Microscopy Characterization of Elec-
tromechanical Properties of RF Acoustic Bulk Wave Resonators,” Proc. IEEE Intl. Conf.
MEMS 2004, Maastrich, the Netherlands, January 25–29, 2004, pp. 169–172.
[39] Telschow, K. L., V. A. Deason, and D. L. Cotte, “Full-Field Imaging of Acoustic Motion at
Nanosecond Time and Micro Length Scales,” Proc. IEEE Intl. Ultrason. Symp. 2002,
Munich, Germany, October 8–11, 2002, pp. 601–604.
[40] Telschow, K. L., et al., “Full-Field Imaging of Gigahertz Film Bulk Acoustic Resonator
Motion,” IEEE Trans. on Ultrason. Ferroelectr. Freq. Control, Vol. 50, 2003,
pp. 1279–1285.
[41] NIKON Microscopy U., http://www.microscopyu.com/articles/optics/objectivespecial.html.
[42] Physik Instrumente (PI) GmbH & Co., “PICMA ® Chip Monolithic Multilayer Piezo Actu-
ators (LVPZT),” Data Sheet, 2008.
[43] Lizorkin, P., “Bessel Functions,” in Encyclopaedia of Mathematics, M. Hazewinkel, (ed.),
Boston, MA: Kluwer Academic Publishers, 2001.
[44] Morata, M., “Resonadores micromecanizados para su aplicación en la detección de gases,”
Ph.D. Thesis, U. A. Barcelona. 2004.
[45] Fabry, C., and A. Pérot, The Interferometer, 1899.
5.5 Summary 161

[46] Langdon, R. M., and D. Dowe, “Photoacoustic Oscillator Sensors,” in Fiber Optic Sensors
II, Vol. 798, A. M. Scheggi, (ed.), The Hague, the Netherlands: SPIE, 1987, pp. 86–93.
[47] Stokes, N. A. D., R. M. A. Fatah, and S. Venkatesh, “Self-Excitation in Fibre-Optic
Microresonator Sensors,” Sens. Actuators A: Phys., Vol. 21, 1990, pp. 369–372.
[48] Kouh, T., et al., “Diffraction Effects in Optical Interferometric Displacement Detection in
Nanoelectromechanical Systems,” Appl. Phys. Lett., Vol. 86, 2005, 013106.
[49] Carr, D. W., and H. G. Craighead, “Fabrication of Nanoelectromechanical Systems in Sin-
gle Crystal Silicon Using Silicon on Insulator Substrates and Electron Beam Lithography,”
J. Vac. Sci. Technol. B, Vol. 15, 1997, pp. 2760–2763.
[50] Nam, C.-Y., et al., “Diameter-Dependent Electromechanical Properties of GaN
Nanowires,” Nano Lett., Vol. 6, 2006, pp. 153–158.
[51] Tanner, S. M., et al., “High-Q GaN Nanowire Resonators and Oscillators,” Appl. Phys.
Lett., Vol. 91, 2007, 203117.
[52] “First Atomic Force Microscope on Mars,” Scientific payload in the NASA’s Phoenix Mars
Lander mission to Mars, http://monet.physik.unibas.ch/famars/index.htm.
 CHAPTER 6

Performance Optimization
After ideal-device characterization, the main process and equivalent-circuit param-
eters are obtained. However, process deviations and stability issues introduce new
variables to the study of the resonator’s behavior. These issues may include time and
frequency stability due to differences between the designed and the implemented
fabrication process. Furthermore, frequency drifting due to the thermal conditions
of the operation has to be kept at low values, for accomplishing the competitive per-
formance of resonators.
This chapter presents the main process optimization techniques to achieve
improved performance of resonators. The optimization of an FBAR is studied in
detail as a representative case for the illustration of the concept. We begin this chap-
ter by explaining the sources of time and frequency instability of the resonance fre-
quency, mainly due to fabrication process deviations and the thermal configuration
of the resonator’s structure. In this way, the thermal coefficient factor is evaluated.
Afterward, we learn about the concept, implementation, and characterization of
different temperature-compensation techniques.
Process-based tuning of the resonance frequency in batch resonators is also
required to face the performance uniformity required in commercial, mass-pro-
duced devices. Some postfabrication tuning techniques of the FBAR’s resonance fre-
quency are studied herein, expanding the study of tuning to a novel, low-impact
technique based on focused-ion-beam.

6.1 Frequency Stability

Tolerances of the fabrication process, the layout design of the resonator, the process
design, and the physical environment of operation may affect the frequency
response of resonators. These parameters cause the resonance frequency to be
changed or to exhibit poor stability. Also, nondesired resonance modes of lower
energy may superpose in a nonconstructive way to the main resonance mode, lead-
ing to quality factor reduction and diminishing the overall resonator’s performance.
These deviations are related to inaccurate or low-controlled thin-film deposition,
oxide growing, or micromachining, among other sources, and they lead to unex-
pected material properties, dimension, or layout variations of the resonators. In this
section, it will become evident how the resonance frequency depends on these
materials and layout properties.

163
164 Performance Optimization

6.1.1 Thin-Film Thickness Tolerance

The performance of the resonator and resonator-based systems depends on how
accurately the resonance frequency’s value is predicted. In certain applications, like
FBAR-based ladder filters, the resonance frequencies of the series and the shunt res-
onators have to be fixed within very small tolerances. This allows preserving the
bandwidth-to-central frequency ratio, in-band insertion loss, and out-of-band-
rejection specifications. Since the size and frequency relationships of devices have a
big impact on the filter’s response, the design allows only small tolerances. As dis-
cussed in previous chapters, the thickness of the piezoelectric layer is the key process
parameter controlling the resonance frequency of the filter-composing FBARs. For
example, the analysis in [1] shows that a deposition process of ZnO with thickness
tolerance of ±86Å is required to obtain a competitive ladder filter with application
in the PCS band (1.9 GHz).
In Section 4.4.1, thickness and profile measurements showed how the AlN
layer may suffer variations due to fabrication in a nonplanar process. If we look at
the S-parameter characterization results, it can be seen that the resonance fre-
quency deviations are between 10% and 15%. This is in the same order of thick-
ness deviations previously found, as observed in the plot of Figure 6.1. The
variations obligate to implement a planarization step, a postprocess tuning tech-
nique, or both. The planarization process is mandatory to reduce the variation of
the AlN thickness, and it can be performed by chemical mechanical polishing
(CMP), which is a technique widely used in semiconductor fabrication. By combin-
ing the chemical and mechanical actions of a reactive and abrasive component, and
a polishing pad, the layer is flattened to guarantee a uniform thickness across the
wafer within very small tolerances. CMP is an on-process technique, because it is
performed during the fabrication process, and it can be performed on such layers
requiring planarization that are critical to the resonance frequency’s control (not
only the AlN, as we will see later). Other fabrication and postfabrication-based fre-
quency control techniques exist as well, as we will study in the next sections of this
chapter.

2.55
2.50
Frequency (GHz)

2.45
2.40
2.35
2.30
2.25
2.20
750 800 850 900 950
Thickness (nm)

Figure 6.1 Resonance frequency of FBARs with different AlN thicknesses: due to process devia-
tions and nonplanar AlN deposition, both the thickness and resonant frequency present a 10–15%
tolerance (respective to the nominal value). (© 2007 IOP Publishing Ltd. [2].)
6.1 Frequency Stability 165

6.1.2 Layout Design Effects

In addition to fundamental-mode resonances, the low frequency response of resona-
tors involves static and quasi-static mechanisms. On the other hand, the high fre-
quency behavior of resonators accounts for layout- and process-dependent
phenomena, including parasitic effects due to the packaging setup, substrate, trans-
mission line, and design-related higher-order modes.
Regarding the layout design of FBAR and other MEMS resonators, the presence
of higher-order modes in the proximity of the main resonance frequency may affect
performance at the frequency of interest. In FBARs, for example, some electrode
geometries suppress or diminish the presence of “spurious” resonance modes, while
others enhance them in a negative way [3]. These modes are lateral, Lamb-like
modes superposing to the main longitudinal-mode resonances, which may appear
due to the resonator’s layout, the crystalline characteristics of the piezoelectric
layer, or both. At the same time, the size of the electrodes has an effect in the
insertion losses of FBARs [4, 5].
For the sake of illustration, let’s take the example of two groups of FBARs fabri-
cated within a RIE-based process. Resonators of the first group have electrodes with
nonparallel walls, whereas those of the second group are designed with paral-
lel-wall electrodes. Examples of the S-parameter responses of resonators from both
groups are observed in the optical pictures of Figure 6.2. Looking at the S21 and
S11 parameter’s response of the second group in Figure 6.2(a), two resonances com-
parable in size with the main peak (2.2 GHz) can be observed around the frequen-
cies 1.6 and 2.6 GHz. However, only a main resonance mode is observed in the
resonators of the first group, according to the plots of Figure 6.2(b). The parallelism
of the FBAR’s electrode walls contributes to the appearance of important spurious
resonances, due to the systematic transmission and reflection of lateral-mode
waves.
Superposed modes of lower energies can also be notoriously close to the main
resonance frequency. While the main resonance is due to longitudinal-wave propa-
gation, as previously discussed in Chapter 2, superposed modes are caused by lat-
eral waves. The energy of these lateral modes can be diminished by the addition of
an “overlap” electrode, which has proven its effectiveness on this task [5, 6]. An
overlap electrode is an annular-shaped structure built on the top electrode of the
FBAR to modify the wave propagation conditions near the layout boundaries of the
electrode. In this way, a more uniform resonance shape is obtained, and the energy
of lateral-mode resonances attenuated. The plots of Figure 6.3 compare the fre-
quency response of two resonators of similar geometry fabricated within the same
process, with and without overlap electrode. Both the electrode design and the over-
lap-electrode addition strategies should be addressed to optimize the FBAR’s
frequency response.

6.1.3 Time and Frequency Stability

Time stability is also an important factor to be considered when studying the reso-
nator’s frequency response. To analyze the fluctuations of the resonance frequency,
we distinguish two cases: short-term and long-term stability.
166 Performance Optimization

−9

S21mag (dB)
−18

−27
1.6 2.0 2.4 2.8
Frequency (GHz)
(a)

−5

−10
S21mag (dB)

−15

−20

−25
1.6 2.0 2.4 2.8
Frequency (GHz)
(b)
Figure 6.2 Electrode geometry influence in the S-parameter response of the FBAR (f0= 2.25 GHz):
(a) parallel-side resonators (big resonances at 1.6 and 2.6 GHz); and (b) nonparallel side devices (a
single resonance peak is observed at 2.25 GHz; small lateral-mode resonances are not observed at
this scale).

Short-time stability can be explained by noise sources in the setup, and it can be
studied by two different methods:

1. Observing a wideband frequency span of the resonant signal of the FBAR

and acquiring various sets of data to calculate the frequency deviation;
2. Observing a 0-Hz frequency span near the resonance frequency to evaluate
the phase deviation.

From the conceptual point of view, results of the first method are equivalent to
those of the second one. Regarding the first method, data acquisition of the magni-
tude of the S11 or S21 parameters is performed, and the resonance frequencies of
each curve labeled and compared among them. In the second method, we acquire
the S11 or S21 parameter phase data, we identify the resonance frequency (using the
marker tool of the network analyzer), and then we reduce the frequency span to 0
Hz. Then we start a new data acquisition and store phase values in order to make
some statistical analysis, by extracting the mean value and standard deviation. The
plot of Figure 6.4 shows the phase response of the S21 parameter of a rectangu-
lar-shaped FBAR. Data acquisition is carried out along a representative number of
6.1 Frequency Stability 167

Figure 6.3 Overlap electrode effects on the high-frequency response of FBARs. (Courtesy of Seiko
EPSON Corp., 2008.)

−10.10

−10.15
S21phase (deg)

−10.20

−10.25

0 200 400 600

Sample number (freq: 2.29218 GHz)

Figure 6.4 Short-term time stability characterization: zero-span frequency-domain acquisition of

the S21 phase response.

samples to give the study statistical validity. To calculate the short-time frequency
stability, the mean value of the phase is divided by the phase slope at the resonance
frequency, which is calculated by differentiating the S21 parameter phase data of
the first acquisition (see an example of S21 phase differentiation in Figure 5.8). In
−6
this example, the phase deviation is ±0.08° and the phase slope is 2.6 × 10 deg/Hz.
Consequently, the frequency drifting has a value of 30 kHz.
The long-term stability is, at its time, observed over a longer window of time.
To study it, S-parameter acquisition is performed on different devices along days,
weeks, months, or even years depending on the quality and goals of the study. In
168 Performance Optimization

such a long window, it is hard to compare the environmental variables of the char-
acterization setup among different samples, but some conclusions regarding the
devices may be extracted. The curves of Figure 6.5 illustrate the magnitudes of the
S11 parameter of an exemplary FBAR device, taken with a 1-month difference.
Measurements were performed at room temperature and, as it can be seen in the fig-
ure, with no significant difference among the resonance frequencies. This example
can be improved and done in a more systematic way, thus exploring the minimum
frequency resolution needed to establish accurate reference-frequency values. Envi-
ronmental conditions, like temperature or vibration, can also be controlled.

6.1.4 Temperature Stability and Thermal Coefficient Factor (TCF)

The temperature stability of resonators is a third aspect worth studying to under-
stand the importance of process optimization. In reference oscillator applications of
FBAR, QCMs, or other MEMS resonators, for example, a low temperature coeffi-
cient (TCF) of the resonator is particularly important to guarantee the best
phase-noise performance [7]. The TCF of the structural layer mainly determines the
overall TCF of the resonator; thus, measuring it gives process and application
design elements to optimize the resonator and its applications.
First, the TCF characterization of the resonator is performed. The required
setup includes a network analyzer, a heater stage and temperature controller, and
an adequate probe station to connect the resonator and the network analyzer. Also,
the temperature range and values to perform the measurements between a mini-
mum and maximum value are defined, the temperature of the chamber is set up, and
the S-parameter data acquisition is performed. New data acquisitions are carried
out for each temperature value after appropriate control of the heater stage. Gather-
ing the data of all measurements in a postprocessing tool allows identification of the
changes of the resonance frequency. In the end, one obtains the curve relating the
resonance frequency against the chamber’s temperature and calculates the TCF.
Since the frequency change is usually low, it is expressed in parts per million (ppm),
referred to a reference frequency as:

−10
S211mag (dB)

−15

−20

2.25 2.30 2.35 2.40 2.45

Frequency (GHz)
Figure 6.5 Time stability characterization: two independent measurements of the S11 parameter
were performed on the same device with a 1-month difference between them.
6.2 Temperature Compensation 169

(fi − f0 ) Δf
× 10 6 [ppm] = × 10 6 [ppm] (6.1)
f0 f0

In (6.1), fi is the measured resonance frequency at temperature i, and f0 is the

resonance frequency at a known temperature of reference. The plot of Figure 6.6
shows the TCF characterization of a rectangular-shaped FBAR. The chosen temper-
ature range is in a range of 100ºC from −20ºC to +80ºC. The resonator has elec-
trodes of Ti/Pt with a thickness of 30 nm/150 nm and a layer of AlN with thickness
equal to 1,000 nm. The TCF value of this resonator is about −20 ppm/ºC, which is
in accordance to FEM simulations (−16 ppm/ºC), and is typical of FBARs fabricated
with sputtered AlN films (TCF between −20 to −40 ppm/ºC). However, this value is
rather high to accomplish the requirements of modern RF applications, in compari-
son to technologies like quartz crystals (TCFs in the units of ppm/ºC). Optimization
of the TCF toward a temperature-compensated device is discussed in the next sec-
tion, where we examine the compensation strategies and the involved processes.

6.2 Temperature Compensation

As commented in the previous section, high TCF values degrade the frequency sta-
bility and phase-noise response of the system in which the resonator is integrated.
Different authors have proposed various strategies to reduce the magnitude of the
TCF of AlN-based FBARs, and so stabilize the frequency response of FBARs. As a
mainstream approach for temperature compensation, a thin-film layer made of a
material with a TCF of opposite magnitude is added to the resonator’s material
stack.
Proper selection and dimensioning of the compensation-film material reduce
the overall TCF of the device, to values even as low as 1–5 ppm/ºC. Vanhelmont et
al. [8] proposed a temperature-compensated FBAR (TCFBAR) for RF oscillator
applications, by a SiO2 layer deposited on top of the AlN layer, which is shown in

1500
TCF: −20 ppm/°C
2
1000 R = 0.946
Δf/f0 (ppm)

500

−500

−1000
−20 0 20 40 60 80
Temperature (°C)
Figure 6.6 Thermal characterization of the FBAR process.
170 Performance Optimization

Figure 6.7(a). With this compensation, the resonator achieves thermal-coefficients

as low as −5 ppm/ºC. Alternative embodiments of the TCFBAR implement the ther-
mal-compensation layer on top of the top metal electrode of the resonator, as
shown in Figure 6.7(b) [9].

6.2.1 TCFBAR Fabrication Processes

The TCF compensation techniques rely mainly on modifying the thermal configura-
tion of the resonator, as commented earlier. The fabrication process of a TCFBAR
introduces additional steps concerning the deposition and patterning of a TCF-com-
pensating layer on top of the AlN layer or in another place of the resonator’s stack.
This layer can be made, for example, of SiO2, and its deposition can be done by
plasma-enhanced chemical vapor deposition (PECVD) or thermal growing. The
PECVD-oxide exhibits a positive TCF that compensates the negative TCF of the
AlN layer. One simplification of the design process allows choosing the photolith-
ography layer of the PECVD-oxide to be the same as for the AlN layer (thus saving a
mask).
In considering such a TCFBAR fabrication process, let’s study the TCF compen-
sation with an example in which we have three wafers with FBAR devices. Two of

Top electrode
SiO2
AIN
Bottom electrode

Bragg reflector

Substrate (Si)

(a)

Flashing cover

Compensating
Top layer
electrode

Bottom Piezoelectric
electrode

Air gap

Substrate (Si)

(b)
Figure 6.7 Temperature-compensated FBAR process alternatives: (a) a compensation layer of sili-
con oxide is deposited on top of the AlN layer (© 2006 IEEE [8]); and (b) the compensation layer
is deposited on top of the second metal electrode of the FBAR (From: © 2005 J. D. Larson [9]).
6.2 Temperature Compensation 171

them have chips with a SiO2 compensation layer and the third one, without it. The
two wafers containing the TCFBAR devices have the TE/SiO2/AlN/BE layered
structure, where TE and BE stand for the top and bottom electrodes, respectively.
Each wafer has the same configuration for the TE (180 nm), BE (180 nm), AlN (700
nm) layers, although they differ in the thickness of the SiO2 layer. The first wafer has
a 50-nm-thick SiO2 layer, whereas the second one’s layer has a thickness of 300 nm.
The third, noncompensated wafer serves as a reference of the technology’s TCF.
Thermal characterization of the three wafers is performed to extract their TCFs.
Thus, we introduce the wafers in the probe station and close the chamber to modify
the temperature between −20°C and +80ºC. Then, we measure the S-parameters of
each device and we store resulting data for postprocessing, according to the proce-
dures described in Section 6.1.4. Measurements are done in the temperature range
of −20°C to +80 ºC, with steps of 10ºC among them. Figure 6.8 shows thermal char-
acterization results.
As can be seen, the compensation-oxide layer has the effect of changing the TCF
of the resonator. According to these results, the oxide layer should have a thickness
between 50 and 300 nm in order to achieve a TCF value near to 0 ppm/°C. This
example shows how the SiO2 layer mitigates the negative TCF of the AlN layer. Dif-
ferent thicknesses of the SiO2 layer evidence a compensation trend. In this way, the
target TCF value should be designed using both experimental and modeling tools,
regarding the specific fabrication process.

6.2.2 Behavioral Description and Modeling of a TCFBAR

A behavioral model is a mathematical tool that helps predict the response of a com-
ponent as a function of a group of variables. In parallel to experimental character-
ization, the behavioral modeling of TCFBARs describes the temperature-dependent
performance of the resonator. To accomplish this task, we create a mathematical
model relating process, equivalent circuit, and temperature variables. Assuming lin-
ear temperature dependence of the resonance frequency, the frequency-shift
relationship can be described by:

1000
0 nm: −20ppm/°C
50 nm: −13ppm/°C
500 300 nm: +14ppm/°C

0
Δf/f0 (ppm)

−500

−1000

−1500

−2000
−20 0 20 40 60 80
Temperature (°C)
Figure 6.8 Thermal characterization of TCFBARs: the positive-TCF value of the SiO2 layer has the
effect of compensating the negative TCF of the AlN layer.
172 Performance Optimization

( (
f 0Ti = f 0ref 1 − TCF × Tref − Ti )) (6.2)

where f0Ti is the resonance frequency at the temperature Ti, and f0ref is the resonance
frequency at the reference temperature Tref. Previously, we studied the equivalent-
circuit equation for the resonance frequency of a crystal resonator, which is given
by:

1
f0 = (6.3)
2π LmC m

Assuming that either the motional inductance Lm or motional capacitance Cm

can be used to describe the linear TCF of the FBAR, a high-level behavioral model of
the TCFBAR can be built. For example, the Lm described by:

Lref
LTim = m
(6.4)
(1 − TCF × (T ))
2
ref − Ti

where LmTi is the Lm value at temperature Ti, and Lmref is the Lm value at the reference
temperature Tref. Replacing LmTi in (6.4) with Lm in (6.3) describes the high-level lin-
ear behavior of f0. Figure 6.9(a) depicts a schema of the equivalent-circuit model of
the FBAR with the behavioral model of Lm within the Cadence design environment.
The plot of Figure 6.9(b) shows the frequency shift-to-temperature curve of a
2.3-GHz FBAR. A TCF of −25 ppm/ºC and a reference temperature Tref = 40ºC were
set up in the model.
This model can be used for system-level and circuit design purposes. Once the
temperature dependence of the TCFBAR with a specified thermal coefficient is
described, the TCFBAR model can be integrated in a circuit-design environment
and implemented by using commercial CAD tools, like Verilog-A descriptors and
Cadence design suite.

6.3 Frequency Tuning

Due to process deviations or modification of the electromechanical properties of the

materials used for FBAR fabrication, its target resonance frequency may be slightly
different from that originally designed. For these reasons, several tuning procedures
have been incorporated in the FBAR and MEMS manufacturing processes. This
subject has become a requirement for industrial production and repeatability of the
frequency response. Some of these procedures implement postfabrication tech-
niques for changing the resonance frequency, whereas others are intended for
tuning of the device during its operation.
For example, an electrostatic-tunable FBAR is presented in [10]. Taking advan-
tage of a similar principle, it has been demonstrated that the electrical properties of
its composing materials can be changed by applying an electric field, thus modifying
the resonance frequency [11]. Indeed, some applications using this approach have
been patented for the integration of acoustically active materials or devices, as well
6.3 Frequency Tuning 173

Rm Cm
L m (T 0 )
Ls Rs

C ox
Rp C0

C sub R sub

(a)

1500

1000 TCF: −25 ppm/°C

500
Δf/f0 (ppm)

−500

−1000

−20 0 20 40 60 80
Temperature (°C)
(b)
Figure 6.9 Behavioral description of FBAR: (a) equivalent-circuit model of the FBAR with Verilog
model of Lm (in the box); and (b) frequency-to-temperature curve for a 2.3-GHz FBAR with TCF
equal to –25 ppm/ºC and reference temperature Tref = 40ºC.

[12, 13]. Frequency adjustment of microelectromechanical cantilevers by using elec-

trostatic pull down has also been explored in [14]. This approach could be espe-
cially useful in piezoelectric-based resonators (in FBAR, for example, induced strain
may change the resonance frequency). Also, an electrostatic mechanism to tune
upward and downward the resonance frequency of nanomechanical resonators was
demonstrated in [15]. Even more, the impedances of piezoelectric resonators
exposed to moderate magnetic fields have been efficiently tuned in their resonance
windows [16]. The setup and characterization results of a basic electrostatic-tuning
procedure are discussed in Section 6.3.1.
On the other hand, postfabrication techniques have had a remarkable impact
on industrial implementations for mass production. For example, FBAR tuning
based on the mass-loading principle has attracted the attention of many research
groups, mainly oriented to depositing a metal or isolating layer on top of the FBAR
stack [17, 18]. In this line of research, recent investigations have shown how the
electrical performance of FBAR boosts up when covered with carbon nanotubes,
too [19]. The mass loading is performed by growing or depositing a thin film on one
of the electrodes of the resonator. The mass loading affects the FBAR’s frequency
response, changing its resonance frequency f0 v0/2t0, where v0 and t0 are the sound
174 Performance Optimization

velocity and the thickness of the unloaded resonator, respectively. The added mass
changes the phase condition of the acoustic wave propagating through the bulk of
the acoustic layer due to impedance modification. The effect is down shifting of the
resonance frequency value [20, 21]. The mass loading in FBAR is carried out by
covering the whole surface of one of the electrodes with a uniform thin film. The
thin film is deposited by physical vapor deposition (PVD) or grown by chemical
vapor deposition (CVD) techniques. In Section 6.3.2 the uniform-film-based mass
loading of FBARs is described.
New techniques have also demonstrated their suitability for the tuning of
micro- and nanosized devices. Thus, the tools implemented in the process are able to
perform postfabrication, localized mass loading of MEMS resonators. For example,
Chiao et al. present in [22] a postpackaging tuning process for microresonators by
pulsed laser deposition (PLD). By adding materials on the surface of the structure,
the desired resonant frequency could be achieved. In the past few years,
focused-ion-beam (FIB) has become a powerful tool for the postfabrication of a
wide variety of MEMS devices. Subtractive or additive implementations of FIB tech-
niques have enabled the fabrication of three-dimensional structures on a microme-
ter scale [23–25]. In another example of an elegant FIB application, the milling of
optical waveguide MEMS is performed for both characterization and
postfabrication tool of a cantilever sensor [26]. Also, the fabrication process of
metal-oxide nanowires showing gas sensing capabilities being contacted by a
dual-beam FIB machine is revealed in [27]. In Section 6.3.3, we introduce a
FIB-based technique for tuning of FBARs.

6.3.1 DC Tuning
The resonance frequency of FBARs can be controlled by changing the electrostatic
configuration of the device. By applying a DC-bias voltage to a third electrode of the
FBAR, its resonance frequency can be tuned, because the static capacitance formed
by the FBAR is modified when the DC voltage is applied to the new electrode [10]. A
different two-electrode biasing approach has been demonstrated so far. In this case,
the DC and AC voltages are applied to the same FBAR’s electrode, thus injecting
both signals to one of the electrodes. With the aid of a bias tee or similar DC cou-
pling system, both the DC and AC voltages are superposed with no need for
additional signal lines.
According to this strategy, the measurement setup for DC tuning is depicted in
Figure 6.10 where the circuit representation, and the physical interconnection of the
FBAR to the network analyzer, the DC-power supply, the bias-tee, and the probe
station are shown. As for the AC excitation–only setup, appropriate calibration
standards and routines have to be initiated in order to compensate for the effects of
the layout-defined transmission line and probing system, keeping the matching
impedance of the network analyzer’s ports to 50Ω.
Using that setup, let’s say that we apply a DC voltage between 0 and 30 VDC to
the first electrode of an FBAR. As we observe in the curves of Figure 6.11(a), the
magnitude of the S11 parameter changes in a linear way and shifts down the reso-
nance frequency to a maximum value of 3 MHz. This leads to a DC-tuning sensitiv-
ity of around 100 kHz/VDC, which is about half the sensitivity obtained with the
6.3 Frequency Tuning 175

a1
ZG bG

1 2
DC
EG AC b1
V ZL

Signal generator Load

(network analyzer) ΓG Γi (network analyzer)

(a)

Network analyzer

Probe station

AC
Bias tee

DC

DC-power supply (b)

Figure 6.10 Measurement setup for DC tuning using a single electrode for DC+AC voltage sup-
ply: (a) circuit representation and (b) physical connection of the FBAR to the instrumentation and
the probe station.

three-electrode tunable FBAR of [10]. Those FBARs, at a resonance frequency of

6.8 GHz, need actuation voltages between 0 to 9 VDC to get a tuning range of 1.9
MHz, thus achieving a tuning sensitivity of 211 kHz/V. Nevertheless, the FBARs of
the second example exhibit an expanded tuning range with high linearity, as illus-
trated in the plot of Figure 6.11(b). Due to its simplicity, this method is demon-
strated to be suitable for fine tuning in certain RF applications.

6.3.2 Uniform-Film Deposition

Uniform-film deposition is a well-known postfabrication method to change the res-
onance frequency to lower values by taking advantage of the mass loading principle
of the electrode. The mass loading–based tuning performance of
microelectromechanical resonators may be described by one or more metrics,
though the mass responsivity is the main parameter being considered. The inverse
mass-responsivity RM is defined as the number of grams deposited on the device
needed to produce a change of 1 Hz in its resonance frequency.
176 Performance Optimization

(a)

(b)
Figure 6.11 DC tuning in FBARs: (a) magnitude of the S11 parameter for different DC-tuning
voltages; and (b) frequency shifting against DC-tuning voltage (tuning sensitivity of 100
kHz/VDC).

Uniform-film deposition can be performed by physical vapor deposition (PVD),

chemical vapor deposition (CVD), or thermal-growing techniques. In this way,
either a metallic or nonmetallic thin film is deposited all over one of the electrodes of
the resonator. This method is suitable for tuning of a wide variety of resonators,
from MEMS to FBAR, and many implementation examples are available in the
literature [11–13].
Magnesium fluoride (MgF2) is a dielectric material that can be used to demon-
strate the uniform-film–based mass-loading concept. If we use PVD to deposit MgF2
on the surface, we will experiment the down-shifting of the resonance frequency.
For instance, if we have a rectangular FBAR with electrode dimensions of 50 × 70
2
μm , and we deposit MgF2 with thicknesses between 2 and 20 nm, we obtain the
highly linear down-shift of the resonance frequency shown in the curves of Figure
−17
6.12. The responsivity achieved with this method is of 1.4 × 10 g/Hz. Since the
PVD process is not easily controllable for thin-film deposition under 1 nm, it is hard
to accomplish mass loading to a lower value, given a certain electrode area. How-
ever, if we reduce the size of the deposited mass, improved sensitivity may be
obtained, as we study in the next section.
6.3 Frequency Tuning 177

Figure 6.12 Uniform-film deposition: frequency shifting of the resonance frequency against the
amount of an MgF2 thin film deposited on top of FBARs.

6.3.3 FIB-Assisted Tuning Technique

6.3.3.1 Ion-Assisted Chemical Vapor Deposition
So far, we have discussed the uniform-film, mass-loading approach to frequency
tuning of resonators. We previously stated that the electrode size and the thin-film
deposition techniques limit the responsivity of the method. However, new local-
ized-mass-deposition techniques may improve the tuning performance of micro-
and nanoresonators. In this section, we study FIB deposition and effects on FBARs,
although it is suitable for silicon nanocantilevers, nanobridges, and other MEMS
and NEMS resonator geometries. This tuning technique has been developed and
extensively demonstrated by our group at the IMB-CNM (CSIC) [2].
First, we define the expression localized-mass deposition, which is the deposi-
tion of a material whose size is small, in comparison to the effective resonator’s or
electrode’s area (typically, the mass area is a fraction of less than the 10% of the
electrode’s area). Localized-mass deposition can be performed using machines with
nanopatterning capabilities like FIB.
To perform the deposition, a precursor material containing Pt, for example, is
injected close to the resonator in the sample’s chamber of the FIB machine and
decomposed by the ion beam. Gallium ion beams (Ga+) are accelerated at kV,
which gives them enough energy to promote the decomposition of the precursor gas
crossing their path. The technique is also known as ion-assisted chemical vapor
deposition (IACVD). If we properly adjust the dwell time of the ion beam, position-
ing, and current density, the amorphous compound resulting from the precursor
decomposition is locally deposited on the area scanned by the beam. According to
Auger Electron Spectroscopy measurements and the precursor materials, this com-
pound may contain C, Pt, and Ga [28]. Its mass density can be estimated in propor-
tion to these materials [personal communication with H. Mulders, FEI Company,
2006]. Depending on the desired size and thickness of the deposited mass, the ion
current and deposition time need to be adjusted. The schematic diagram of Figure
6.13 represents the mass deposition procedure: the Pt injector introduces the
metalorganic precursor inside the vacuum chamber, while the same is decomposed
by the Ga+ ion-beam scanning the resonator’s electrode area. The result is a mass
deposited on the top electrode of the cantilever, illustrated in this figure as a square
spot. In addition, the electron beam of the dual beam machines is used for imaging
and calibration purposes without structural damaging of the sample.
178 Performance Optimization

Figure 6.13 FIB-based tuning setup and procedure: the Pt injector introduces the metalorganic
precursor inside the vacuum chamber, whereas the same is decomposed by the Ga+ ion-beam
scanning the cantilever’s area.

The FIB-assisted tuning technique is very versatile and allows a wide variety of
geometries and types of resonators’ configurations being tuned. The SEM images of
Figure 6.14(a, b) were obtained inside the dual-beam FIB-SEM machine and show a
mass deposited on the center of the top electrode of an FBAR. The mass has a con-
tact surface of 1.5 μm × 1.5 μm, while the FBAR electrode’s area was of 50 μm × 50
μm. The tilting angle of the electron beam is 52°, so dimensioning of patterns at the
vertical scale has to be compensated for accurate calculations.
The SEM images of Figure 6.15(a–c) show examples of other FBAR geometries
in which localized tuning of the resonance frequency was performed. FBARs with a
rhomboidal shape [Figure 6.15(a)], a cantilever [Figure 6.15(b)], and a piezoelectric
bar [Figure 6.15(c)] illustrate different layout configurations with their correspond-
ing tuning load deposited on the top electrode. Since the location and size of tuning
loads give valuable information about the tuning performance, calibration and
adjustment of the setup need to be performed prior to deposition on target FBAR
devices.
Some undesired effects, like charging of the sample’s surface, may occur during
deposition, thus causing the ion beam to be shifted away from the desired location.

(a) (b)

Figure 6.14 FIB-based deposition of a C/Pt/Ga composite on top of FBAR (SEM images): (a) gen-
eral view; and (b) detailed view of the deposited spot. (© 2007 IOP Publishing Ltd. [2].)
6.3 Frequency Tuning 179

(a) (b)

(c)

Figure 6.15 Examples of other FBAR geometries in which tuning of the resonance frequency was
performed: (a) rhomboidal-shaped; (b) cantilever; and (c) piezoelectric bar. (© 2007 IOP Publish-
ing Ltd. [2].)

Although not significantly for small-sized mass deposited on big electrode resona-
tors, it could cause the tuning load to be deposited out of the surface of narrower
devices like cantilevers. In this way, a calibration procedure helps the improvement
of the deposition accuracy. Calibration is also important for analyzing the perfor-
mance of the frequency tuning. Two aspects mainly influence this performance:

1. The size of the deposited mass;

2. Collateral effects of the ion beam in the sample’s surface like residual etching
or deposition in the surroundings of the target area (this could happen due to
inappropriate selection of the ion-current or scanning conditions).

In this way, the calibration has to ensure the appropriate size and thickness of
the mass and, at the same time, minimum damage to the resonator when irradiated
by the ion beam (we will return later to this point). Also, efficiency is important in
IACVD, because a certain deposition goal can be achieved with lower or higher cur-
rent densities. This will depend on the scanning area and will affect the deposition
(or milling) time.

6.3.4 Milling of FBAR as Another FIB-Tuning Procedure

At this point, we have discussed a localized-mass-deposition–based tuning proce-
dure of resonators. However, under certain ion-beam setup conditions, milling
(etching) of the device structure can be performed, thus deloading the resonator.
This can happen for current densities of the ion beam above about pA/ìm2, but other
considerations, such as scanning speed, also influences the process. While the reso-
180 Performance Optimization

nance frequency may be down-shifted by the mass-loading procedure, with the

mass deloading this frequency is expected to be up-shifted. Since the localized etch-
ing modifies the structure of the device, we can suppose a change of the mode-shap-
ing of the resonance, too. Thus, it is important to understand the interaction
between deloading, frequency tuning, and resonance’s mode shaping. The SEM
image of Figure 6.16 shows a region of a deloaded FBAR, in which the Ti/Pt top
electrode has been locally milled by the ion beam. In this image, the underlying AlN
layer is observed in the milled area.
Based on the same principle, we can change the shape of the resonance modes of
the device by transversal milling of the structure. To illustrate this concept, let’s
look at the rectangular FBAR shown in the SEM image of Figure 6.17. Suppression
of the spurious modes was achieved by transversal cuts of the layered FBAR struc-
ture, as observed in the image. Electrical characterization of the magnitude of the
S21 parameter, before and after milling, demonstrates how rippling and some reso-
nance modes are no longer visible after the FIB processing [curves of Figure
6.17(b)]. Looking at the curves, the new resonance shape coincides with the shape
and frequency of smaller modes observed in the nonmilled device. Before the mill-
ing, these modes were superposed to the main resonance and other small spurious
modes. Now, they can be enhanced due to the ion-beam processing. Of course, sys-
tematic engineering of the mode shape requires finite element analysis, but the pre-
sented examples illustrate the potential of this tool for mode shaping.

6.3.5 Frequency-Tuning Sensitivity and Responsivity

At this point, we wonder whether or not the FIB-based tuning has a better tuning
performance in comparison to uniform-film deposition. As long as IACVD is a com-
plex process, the answer will depend on several deposition parameters and on the
configuration of the deposited mass. Nevertheless, one example of FIB-deposited
FBARs illustrates the advantages of this tuning method. Let’s say that we perform
localized-mass deposition on the geometric center of the top electrode of a group of
rectangular resonators with electrode area of 50 × 70 μm2. The squared-shaped
2
localized masses had a size of 1.5 × 1.5 μm and different thicknesses to change the
−15 −12
amount of mass; in the range of 9.0 × 10 g to 3.6 × 10 g. The frequency shift
against the amount of deposited mass is shown in the plot of Figure 6.18 (the x-axis

Figure 6.16 Illustration of the milling-based deloading concept: ion-assisted etching locally
removes the Ti/Pt top electrode of FBAR (the square-shaped etching allows observation of the AlN
layer). (© 2007 IOP Publishing Ltd. [2].)
6.3 Frequency Tuning 181

(a)

FIB
No milling

(b)
Figure 6.17 FIB-based engineering of mode shaping in FBAR: (a) SEM image of FBAR with
transversal cuttings (intended for suppression of spurious modes); and (b) magnitude of the S21
transmission parameter, before (dotted line) and after (continuous line) FIB-assisted milling (ripple
and some spurious modes disappeared for the after-FIB case). (© 2007 IOP Publishing Ltd. [2].)

is represented in log-scale, due to the wide range of deposited masses). Regarding

−19
these results, the responsivity may be calculated, its average value being 7.4 × 10
g/Hz. This value is competitive with respect to other NEMS technologies previously
reported [29–31].
The curves also show the extrapolated linear fit of the uniform-film deposition
case. Thus, we can see how the localized-mass deposition exhibits better
responsivity in comparison to the uniform-film case. In this way, the extrapolated
curve shows that the improvement is between one and three orders of magnitude for
the ion-beam-tuning procedure. However, as we have also seen in the curves, this
happens for low-mass depositions. Above the crossing value of the curves, uni-
form-film deposition would be more efficient to resonator’s tuning.

6.3.6 Quality Factor

The quality (Q) factor of the resonator mainly determines its mass sensitivity, since
it is related with a minimum change of frequency. In this sense, the Q factor is an
important issue to be considered in mass-loading-based tuning of resonators. Dur-
ing deposition, however, the ion beam might induce heat damages in the resonator’s
182 Performance Optimization

Frequency Shift (MHz)

−15
Mass (10 g)
(a)

Uniform
Localized
Linear Fit
Extrapolated Uniform
Frequency Shift (kHz)

−15
Mass (10 g)
(b)
Figure 6.18 Frequency shift of the resonance frequency against the amount of tuning load: (a)
detailed sensitivity plot; and (b) FIB-tuning performance in comparison to uniform-film deposition.
(© 2007 IOP Publishing Ltd. [2].)

structure, due to possible heating of the materials, thus reducing its Q factor. Some
practical experiments may help us know the effects of ion-beam radiation. Starting
from ion-beam–based imaging of a test resonator, the electrical response can be
evaluated before and after ion-beam irradiation.
To do that, we can take one or more resonators and extract the Q factor by
evaluating their S-parameter responses. Next, the resonators are irradiated with the
ion beam to perform imaging or very-small mass deposition (in order to not signifi-
cantly change the resonance frequency). Again, we characterize the S-parameters
and the Q factors to see if some change in the electrical response arises. The plot of
Figure 6.19 compares the magnitudes of the S21 parameter before and after
ion-beam imaging practiced on a sample FBAR. The S parameters and the Q factor
remained at a constant value of about 500 in both cases.
Some explanations can explain these results. First, not enough heating of the
structure occurs to modify the piezoelectric properties of the AlN. Also related to
this idea, the longitudinal resonance mode and the reduced size of the mass—com-
pared to the FBAR—inhibits to a great extent the mass-related damping. As a differ-
ence to flexural MEMS resonators, FBAR works in longitudinal resonance modes.
6.4 Summary 183

Figure 6.19 S21 magnitude of FBAR before and after i-beam operation (no mass deposition, only
i-beam imaging). No significant change of the response is observed. (© 2007 IOP Publishing Ltd.
[2].)

In low-frequency mechanical resonance modes, as with the case of vibrating cantile-

vers, the electromechanical damping of deposited masses strongly reduces the Q
factor [32].
Another aspect is that the mass is deposited on top of the Pt layer of the FBAR,
and no interaction occurs between the ion beam and the AlN, due to the Pt shield-
ing. In this case, the AlN quality mainly determines the Q factor of the FBAR.

6.4 Summary

In this chapter, we have learned about different resonator optimization techniques.

First, we have reviewed the main issues degrading the expected performance of reso-
nators. We have studied how process deviations change the thickness of the struc-
tural layer, which affects the resonance frequency. Also, thermal noise sources may
affect the short- and long-term frequency stability. The thermal coefficient factor
(TCF) of the structural layers of the resonator is another important issue determin-
ing the stability, which has importance in oscillator applications, for example.
Different resonator optimization techniques have been explored. First, we
described a process-based temperature compensation method. To start, we evalu-
ated the TCF of FBAR. With this information, the temperature-compensation strat-
egy was implemented by modifying the fabrication process with an additional
thermal-compensation layer. We studied different examples of TCFBARs with
reduced TCF and compared their performance with noncompensated FBARs.
On the other hand, we have made a review on various state-of-the-art fre-
quency-tuning techniques. Beginning with electrostatic tuning of the resonance fre-
quency, we studied the uniform-film deposition technique, too. However, we have
taken significant space to explain the leading-edge FIB-assisted tuning technique, its
advantages and its possibilities. Different ion-beam-tuning procedures, including
IACVD and milling of resonators, were described. Also, we have performed
benchmarking of the ion-beam and uniform-film tuning techniques, finding out that
the FIB-based procedure offers improved responsivity under small-mass-deposition
conditions, which are very useful for fine tuning of sensitive devices, like FBARs.
Improved responsivities of at least one and two orders of magnitude were achieved
with this method. The possible influence of ion-beam irradiation in the physical
184 Performance Optimization

structure and electrical response of the resonators was also analyzed. We conclude
that it depends on the type of resonator and on the resonance mode. For example,
no significant variation in the Q factor or the S parameters of FBARs was observed,
whereas low-frequency cantilevers suffer from significant damping and Q-factor
reduction.

References

[1] Andersen, B. D., and N. A. Belkerdid, “Measurement Sensitivity Analysis of One Port BAW
Resonators,” Proc. 50th IEEE International Frequency Control Symposium 1996, Hono-
lulu, HI, June 5–7, 1996, pp. 357–362.
[2] Campanella, H., et al., “Focused-Ion-Beam-Assisted Tuning of Thin-Film Bulk Acoustic
Wave Resonators (FBAR),” J. Micromech. Microeng., Vol. 17, 2007, pp. 2380–2389.
[3] Rosén, D., J. Bjursttröm, and I. Katardjiev, “Suppression of Spurious Lateral Modes in
Thickness-Excited FBAR Resonators,” IEEE Trans. on Ultrason. Ferroelectr. Freq. Con-
trol, Vol. 52, 2005, pp. 1189–1192.
[4] Su, Q., et al., “Thin-Film Bulk Acoustic Resonators and Filters Using ZnO and Lead-Zirco-
nium-Titanate Thin Films,” IEEE Trans. on Microw. Theory Tech., Vol. 49, 2001,
pp. 769–778.
[5] Kim, Y. D., et al., “Highly Miniaturized RF Bandpass Filter Based on Thin-Film Bulk
Acoustic-Wave Resonator for 5-GHz-Band Application,” IEEE Trans. on Microw. Theory
Tech., Vol. 54, 2006, pp. 1218–1228.
[6] Kaitila, J., et al., “Spurious Resonance Free Bulk Acoustic Wave Resonators,” Proc. IEEE
Intl. Ultrason. Symp. 2003, Honolulu, HI, October 5–8, 2003, pp. 84–87.
[7] Gribaldo, S., et al., “Experimental Study of Phase Noise in FBAR Resonators,” IEEE
Trans. on Ultrasonics, Ferroelectrics, Freq. Control, Vol. 53, 2006, pp. 1982–1986.
[8] Vanhelmont, F., et al., “A 2GHz Reference Oscillator Incorporating a Temperature Com-
pensated BAW Resonator,” Proc. IEEE Intl. Ultrason. Symp. 2006, Vancouver, BC,
October 3–6, 2006, pp. 333–336.
[9] Larson III, J. D., “Method of Making an Acoustic Wave Resonator,” U.S. Patent
6,874,212, April 2005.
[10] Pan, W., et al., “A Surface Micromachined Electrostatically Tunable Film Bulk Acoustic
Resonator,” Sens. Actuator A-Phys., Vol. 126, 2006, pp. 436–446.
[11] Lancaster, M. J., J. Powell, and A. Porch, “Thin-Film Ferroelectric Microwave Devices,”
Supercond. Sci. Technol., Vol. 11, 1998, pp. 1323–1334.
[12] Ella, J., “Device Incorporating a Tunable Thin Film Bulk Acoustic Resonator for Perform-
ing Amplitude and Phase Modulation,” U.S. Patent 5714917, 1998.
[13] Korden, C., T. Ostertag, and W. Ruile “Component Having an Acoustically Active Mate-
rial for Tuning During Operation,” U.S. Patent 6847271, 2005.
[14] Kafumbe, S. M. M., J. S. Burdess, and A. J. Harris, “Frequency Adjustment of
Microelectromechanical Cantilevers Using Electrostatic Pull Down,” J. Micromech.
Microeng., Vol. 15, 2005, pp. 1033–1039.
[15] Kozinsky, I., et al., “Tuning Nonlinearity, Dynamic Range, and Frequency of
Nanomechanical Resonators,” Appl. Phys. Lett., Vol. 88, 2006, 253101.
[16] Maglione, M., W. Zhu, and Z. H. Wang, “Evidence of a Strong Magnetic Effect on the
Impedance of Integrated Piezoelectric Resonators,” Appl. Phys. Lett., Vol. 87, 2005,
092904.
[17] Ruby, R. C., and P. P. Merchant, “Tunable Thin Film Acoustic Resonators and Method for
Making the Same,” U.S. Patent 5587620, December 24, 1996.
6.4 Summary 185

[18] Ylilammi, M. A., “Method for Performing On-Wafer Tuning of Thin Film Bulk Acoustic
Wave Resonators (FBARS),” U.S. Patent 6051907, April 18, 2000.
[19] Dragoman, M., et al., “High Performance Thin Film Bulk Acoustic Resonator Covered
with Carbon Nanotubes,” Appl. Phys. Lett., Vol. 89, 2006, 143122.
[20] Sauerbrey, G. Z., “Verwendung von Schwingquarzen zur Wägung dünner Schichten und
Microwägung,” Z. Phys., Vol. 155, 1959, pp. 206–222.
[21] Lostis, P., “Etude, realisation et contrôle de lames minces introduisant une difference de
marche déterminée entre deux vibrations rectangulaires, Part 2: Nouvelle methôde de
controle de l’epaisseur pendant l’évaporation,” Rev. Opt., Theor. Instrum., Vol. 38, 1959,
pp. 1–28.
[22] Chiao, M., and L. Lin “Post-Packaging Frequency Tuning of Microresonators by Pulsed
Laser Deposition,” J. Micromech. Microeng., Vol. 14, 2004, pp. 1742–1747.
[23] Fujii, T., et al., “A Nanofactory by Focused Ion Beam,” J. Micromech. Microeng., Vol. 15,
2005, pp. S286–S291.
[24] Reyntjens, S., and R. Puers, “Focused Ion Beam Induced Deposition: Fabrication of
Three-Dimensional Microstructures and Young’s Modulus of the Deposited Material,” J.
Micromech. Microeng., Vol. 10, 2000, pp. 181–188.
[25] Reyntjens, S., and R. Puers, “A Review of Focused Ion Beam Applications in Microsystem
Technology,” J. Micromech. Microeng., Vol. 11, 2001, pp. 287–300.
[26] Pruessner, M. W., et al., “End-Coupled Optical Waveguide MEMS Devices in the Indium
Phosphide Material System,” J. Micromech. Microeng., Vol. 16, 2006, pp. 832–842.
[27] Hernández-Ramírez, F., et al., “Fabrication and Electrical Characterization of Circuits
Based on Individual Tin Oxide Nanowires,” Nanotechnol., Vol. 17, 2006, pp. 5577–5583.
[28] Vilà, A., et al., “Fabrication of Metallic Contacts to Nanometer-Sized Materials Using a
Focused Ion Beam (FIB),” Mater. Sci. Eng. C, Vol. 26, 2006, pp. 1063–1066.
[29] Forsen, E., et al., “Ultrasensitive Mass Sensor Fully Integrated with Complementary
Metal-Oxide-Semiconductor Circuitry,” Appl. Phys. Lett., Vol. 87, 2005, 043507.
[30] Ekinci, K. L., X. M. H. Huang, and M. L. Roukes, “Ultrasensitive Nanoelectromechanical
Mass Detection,” Appl. Phys. Lett., Vol. 84, 2004, pp. 4469–4471.
[31] Ilic, B., et al., “Attogram Detection Using Nanoelectromechanical Oscillators,” J. of Appl.
Phys., Vol. 95, 2004, pp. 3694–3703.
[32] Enderling, S., et al., “Characterization of Frequency Tuning Using Focused Ion Beam Plati-
num Deposition,” J. Micromech. Microeng., Vol. 17, 2007, pp. 213–219.
 CHAPTER 7

Integration of Resonator to CMOS

Technologies
Complementary-metal-oxide-semiconductor (CMOS) is, nowadays, the most pop-
ular integrated-circuit (IC) technology. Although other IC technologies—like bipo-
lar, BiCMOS, or GaAs, for example—exhibit improved performance in specialized
applications such as radio frequency (RF), the presence of CMOS in the IC market
field has grown at a faster rate. Virtually all the industrial and consumer-oriented
applications are built of low-cost, low-power electronics systems implementing
CMOS circuits. Furthermore, the development of the system-on-chip (SoC) and sys-
tem-on-a-package (SoP) concepts has contributed to increase the popularity of
CMOS. More recently, the technological and application convergences between
traditional IC technologies and the emerging nano-bio-info-cogno (NBIC) devices
have boosted and motivated the integration of CMOS with other processes, materi-
als, and devices, like FBARs, MEMS and NEMS resonators, SOI/CMOS mixed sig-
nal ASICs, microscale passive components, micropower systems, and
biocompatible materials. Low cost, reduced size, and low power are the main
requirements and promises of such convergence.
This chapter deals with the case of resonator-based, IC-integrated systems.
Thus, the different approaches to integrate CMOS and resonator technologies are
studied. First, we define the current-art integration strategies and technologies,
including hybrid, monolithic, and the more recent heterogeneous integration. Then,
we review leading-edge IC applications based on integrated MEMS, NEMS, FBAR,
and SAW resonators. At this point, we pay our attention to the advanced 3D, heter-
ogeneous integration with CMOS processes. The wafer-level-transfer integration of
diverse micro- and nanotechnologies and target CMOS technologies are discussed.
The chapter closes with a case study on the development and implementation of a
wafer-level-transfer process for the heterogeneous integration of FBAR and CMOS
technologies.

7.1 Integration Strategies

Over the past decades, CMOS has become the predominant fabrication technology
for integrated circuits (IC). Research and development efforts have been made to
continuously improve process yield and reliability, while minimal feature sizes and
fabrication costs continue to decrease. Nowadays, the power of CMOS technology
is exploited for ICs and also for a variety of microsensors and MEMS benefiting
from well-established fabrication technologies and the availability of on-chip cir-

187
188 Integration of Resonator to CMOS Technologies

cuitry. Recently integrated microsystems featuring calibration by digital program-

ming, self-testing, and digital interfaces have been implemented on a single chip,
demonstrating the strength of CMOS-based MEMS [1].
SoC technologies integrate multiple function systems on a single silicon chip.
This concept involves the double integration of processes and functions, and it
promises increased capability, lower power consumption, and improved economy,
in comparison to the multiple-chip technologies [2]. Whether technology con-
straints of the SoC definition arise at the time of implementation, the more general
concept of SoP enables the integration of multiple systems fabricated within the
same or different IC processes into a single-packaged chip, through advanced pack-
aging technologies. With the contribution of MEMS technologies, the LoC and the
micro total analysis systems (μTAS) are some of the application-level concepts
developed from the SoC or SoP [3–5]. More and more, we see how MEMS and
NEMS play a central role in the miniaturization of electronic SoC, LoC, and μTAS.
In all these systems, CMOS is the transversal technology that supports the condi-
tioning and signal processing functions, and it provides, in most of the cases, the
physical substrate of the system.
Nevertheless, the engineering of integration and process compatibility of
MEMS to CMOS-based IC technologies is a tricky and challenging task at both the
technology and application levels. Several challenges, like manufacturing tempera-
ture, packaging, postprocessing of MEMS, integrity of CMOS, and reliability and
modeling of the MEMS-to-CMOS interface, are just some of the considerations that
can be mentioned [6–8]. Nowadays, hybrid and monolithic-integration are the
mainstream strategies in CMOS-to-MEMS integration. The CMOS-to-NEMS and
CMOS-to-FBAR integrations have been demonstrated as well. More recently, het-
erogeneous integration has become an important evolution of the hybrid strategy in
CMOS integration. According to the way in which the CMOS and the MEMS sub-
strates are processed and interconnected, each strategy is defined as follows.

7.1.1 Hybrid Integration

In hybrid integration, the MEMS or NEMS chip is fabricated on a different sub-
strate within various technological processes involving micromachining of the
device structure, and it is combined with a separate CMOS-processed IC chip. The
integration of both processes is performed at the chip level by wire bonding or pack-
aging techniques after fabrication of both the MEMS/NEMS and the CMOS chips.
Thus, dicing is carried out prior to process integration [1].
The CMOS integration may be performed for read-out or functional purposes.
For example, the temperature-compensated silicon I-shaped bulk acoustic resona-
tor (IBAR) reference oscillator of Figure 7.1 connects the IBAR and the CMOS cir-
cuit through wire-bonding to close the oscillation loop [9]. The IBAR is an
electrostatic micromechanical MEMS resonator whose current is amplified by the
transimpedance amplifier incorporating a mixed bandgap PTAT temperature com-
pensation stage. The IBAR is fabricated using the HARPSS-on-SOI process [10],
and the CMOS circuit is a transimpedance amplifier with a self-biased
folded-cascode operational transconductance amplifier (OTA) fabricated in a
two-poly three-metal (2P3M) 0.5-μm CMOS process.
7.1 Integration Strategies 189

(a)

(b)

(c)
Figure 7.1 Temperature-compensated silicon I-shaped bulk acoustic resonator (IBAR) reference
oscillator: (a) block diagram of the reference oscillator; (b) SEM picture of the IBAR; and (c) die pic-
ture of the interface IC. (© 2007 IEEE [9].)
190 Integration of Resonator to CMOS Technologies

Hybrid integration is the oldest and simplest way to integrate different technol-
ogies, because it does not concern on-process compatibility issues. Regarding such
compatibility, it is fast because it avoids technological developments to harmonize
the requirements of the processes to be integrated. The integration can be performed
with commercial wire bonding or multichip-module (MCM) platforms, which are
the cheapest option in most cases, instead of suffering the costs and engineering
effort of technology development. On the other hand, hybrid integration is carried
out in a one-by-one basis, which makes it slow for mass-production purposes. The
diagrams of Figure 7.2 represent the process of hybrid integration when performed
through wire bonding and stacked packaging. As long as wafers are diced to extract
the chips, a pick and place system is needed to manipulate both the resonator and IC
chips. As we can see, two technologies are available for carrying out the intercon-
nection: wire bonding and packaging. While the MEMS die is placed onside of the
IC chip in the case of the wire-bonded system, they are placed one above the other
one when using the MCM packaging technology.
The wire-bonding system saves alignment effort but is area expensive. Also, the
wires introduce parasitic reactive and resistive elements on the circuit, thus degrad-
ing the performance of the system, which is especially notorious for high-frequency
applications. These parasitics must be considered to the equivalent-circuit model of

(a)

(e)

(b)

(f)

(c)

(d) (g)

Figure 7.2 Process of hybrid integration: the MCM-based approach includes post deposition and
resonator wafer dicing (a), flip-chip of the resonator die and connection to the IC wafer (b), full
dice interconnection (c), and dicing of the IC wafer with the resonators (d). On the other hand,
the wire-bonding–based hybridization involves dicing of both the resonator and the IC wafers (e),
wire bonding of the first resonator-IC dice (f), and full dice interconnection (g).
7.1 Integration Strategies 191

the system to predict its response within reasonable margins. The packaging
approach offers increased area efficiency and reduced parasitic effects, but it is more
complex to implement, and requires careful alignment of the interconnection pads.
Hybrid integration is the most affordable option when, due to intrinsic features
of the MEMS/NEMS technology, the CMOS line may be contaminated by the
materials of the MEMS/NEMS process. The situation makes impossible the devel-
opment of a compatibility protocol. Besides, MEMS/NEMS and CMOS manufac-
turers are often specialized, have dedicated resources, or simply lack the interest to
disperse their expertise in exploring new processes or integration strategies. One
explanation can be that they are busy enough continuously updating their own tech-
nology in order to remain competitive in the market place. For these manufacturers,
hybrid integration offers a good tradeoff between functionality and cost.

7.1.2 Monolithic Integration

Monolithic integration appeared to overcome the limitations of hybrid technolo-
gies. According to this approach, the MEMS device and the IC are made and com-
bined on a single substrate, according to the same standard, CMOS-compatible
technology. The combination is performed using read-out circuitry to compensate
for MEMS deficiencies or to provide the desired functionality. Depending on the
order of fabrication, monolithic integration can be grouped into: (1) pre-CMOS
(also known as first-CMOS or before-CMOS), (2) intermediate-CMOS (or
interdigitated fabrication), and (3) post-CMOS (add-on or after-CMOS) [1].
In the pre-CMOS approach, the MEMS structures or part of them are formed
before the regular CMOS process sequence. Necessarily, the MEMS/NEMS, as a
movable device requiring micromachining, cannot be located under the IC layers so
it must be placed at one side of the IC, which increases the die area. Remarkable
examples of this approach are the embedded polysilicon microstructures
(microengine) based on the iMEMS technology of Sandia National Laboratories
shown in Figure 7.3 [11].
In the intermediate-CMOS approach, the CMOS process sequence is inter-
rupted for additional thin-film deposition or micromachining steps. This approach
is commonly exploited to implement surface micromachined polysilicon struc-
tures in CMOS technology. Either the standard gate polysilicon or an addi-
tional low-stress polysilicon layer is used as structural material. Examples of
commercially available microsensors relying on intermediate process steps are the
Infineon’s pressure-sensor integrated circuits [12]. Figure 7.4 illustrates the side
view of this process [13]. There, surface micromachining of a SiO2 sacrificial layer is
carried out to form the cavity of the polysilicon membrane. Hermetic sealing of the
membrane is achieved when depositing subsequent layers of the CMOS process on
the membrane.
In the post-CMOS approach, two general fabrication strategies can be distin-
guished. In the first strategy, the MEMS structures are completely built on top of a
finished CMOS substrate, leaving the CMOS layers untouched. An example for this
approach is the Texas Instruments’ Digital Micromirror Device (DMD) [14]). Alter-
natively, the MEMS can be obtained by machining the CMOS layers after the com-
pletion of the regular CMOS process sequence. Both strategies are implemented by
192 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.3 Pre-CMOS integration: (a) concept and (b) realization of a three-level polysilicon
structure (micro-engine) built in a trench. (© 1995 IEEE [11].)

Figure 7.4 Intermediate-CMOS process: the Infineon’s pressure-sensor IC is implemented

through surface micromachining of the polysilicon layer to form the sensor’s cavity. Sealing of the
membrane is carried out by using the subsequent CMOS layers. (After: [13].)

using a variety of bulk-micromachining and surface-micromachining CMOS-com-

patible techniques. Figure 7.5 shows a bridge-shaped MEMS resonator fabricated
in AMS035 CMOS technology. The MEMS has been released by wet-etching sur-
face micromachining and using the second strategy (micromachining of the
top-level CMOS layers after completion of the process) [15].
Monolithic integration is a complex process that requires a specific develop-
ment to control the materials, temperatures, and the possible sources of contamina-
7.1 Integration Strategies 193

(a) (b)

(c)
Figure 7.5 Post-CMOS bridge-shaped MEMS resonator fabricated in AMS035 CMOS technology:
(a) top and (b) cross-sectional views, and (c) realization of the MEMS resonator-CMOS ensemble
(SEM image of the bridge-shaped resonator at right). (© 2006 IEEE [15].)

tion. Therefore, it is highly specialized and exclusive for the modified CMOS
technology, with the MEMS/NEMS technology subordinating its possibilities to the
restrictions imposed by the CMOS process. Due to the achieved compatibility, the
process parameters of the CMOS and the MEMS/NEMS devices are the same, as
their electrical responses can be predicted with the same modeling tools. The
parasitics effects of the MEMS/NEMS-CMOS interconnection are drastically
reduced, in comparison to hybrid integration, because of the efficiency achieved by
the shorter paths provided by the CMOS via/through hole technology. However,
the development costs and the risks are high, and the CMOS manufacturer must
provide exclusive resources for the technology testing, which is not always in the
best interest of an industrial company. However, the impact on mass production is
very positive, thus guaranteeing higher volumes. Monolithic integration also
enables the fabrication of nanosized NEMS devices through advanced patterning
tools like electron-beam lithography (EBL) and FIB, among others. The impact of
these tools on the CMOS performance after postfabrication of NEMS devices has
been documented [16].
194 Integration of Resonator to CMOS Technologies

7.1.3 Heterogeneous Integration

In spite of the classical hybrid-monolithic classification of the CMOS integration
approaches, a third strategy, heterogeneous integration, is to be defined. Heteroge-
neous integration is an evolution of hybrid integration and refers to technologies
that can be integrated on one platform device. Also, it can designate materials that
are not compatible and cannot be manufactured on the same substrate—at least not
in a cost-effective manner [17]. The concept of heterogeneous integration enables
the fabrication and assembly of complete electronic subsystems from components
fabricated within a variety of processes. MEMS devices, SOI/CMOS mixed signal
ASICs, microscale passive components, and micropower systems are some of these
systems. The baseline technology incorporates bump attachment and flip-chip
between the various components and the substrates. When possible, this will entail
wafer bumping prior to dicing and attachment [18, 19]. This feature and the
three-dimensional stacking of different substrates represent a fundamental differ-
ence from traditional hybrid-integration technologies. Concept and realizations of a
Si-based system-in-package composed by passive and heterogeneous integration
can be observed in the drawings and images of Figure 7.6. First, in Figure 7.6(a), the
concept of integration of a variety of passive and active components is depicted. In

(a)

(b)
Figure 7.6 Heterogeneous integration: (a) concept of integrated high-resistivity substrate, CMOS
digital processing, RF and analog BiCMOS sections, and MEMS sensors (on top); and (b) cross-
section of a fully integrated radio module with active transceiver and passive die in a molded lead
frame. (© 2006 Elsevier [19].)
7.2 State-of-the-Art Integrated Applications 195

Figure 7.6(b) a cross-section view of a fully integrated radio module with an active
transceiver and passive die in a molded lead frame is observed.
Heterogeneous integration overcomes the technological restrictions of tradi-
tional hybrid and monolithic integrations. To complete the integration, no process
compatibility between the MEMS/NEMS and the CMOS part is required. Processes
involving diverse materials like high-temperature ceramics, piezoelectrics, and
noble metals, among others, can be integrated to a CMOS substrate avoiding the
need for CMOS line modification. Additionally, the same integration process may
be suitable for different CMOS and MEMS/NEMS technologies, if they incorporate
similar conditions. This is the case of standard CMOS processes implementing cer-
tain metals, dielectrics, and passivation layers. Thus, flexibility is an added value of
the heterogeneous integration approach.
In addition to standard CMOS processes, the clean room incorporates MCM
and postprocessing sections for stacking, interconnection, and micromachining of
the parts. As a remarkable difference to traditional hybrid processes, heterogeneous
integration can be performed either at the chip or the wafer level, which means that
dicing is no longer prioritized over integration. Instead, new procedures like
wafer-level transfer become key components of the integration technology.
Micromachining may be carried out before or after the integration, leaving the dic-
ing at the end of the process, if it is desirable. Because of this feature, heteroge-
neously integrated systems may be mass produced, thus benefiting from scale
economies. The interconnection technology, as in the monolithic integration case,
can be optimized to reduce the parasitic effects at the MEMS-to-CMOS interface.
Via or through hole, electroplating, and bump attachment are some of the available
technologies to provide the vertical connection of the stacked system, which is not
limited to only a couple of processes as illustrated in the conceptual diagram of Fig-
ure 7.6. Traditional micromachining as well as advanced laser ablation techniques
may be employed to release devices or substrates.
Table 7.1 synthesizes the requirements and features of the different integration
strategies we have reviewed on this chapter. As we may suspect, any of the strategies
can claim supremacy, and their implementation will depend more on the specific
economical and application cases. As long as FBAR, MEMS and NEMS resonators,
CMOS, and other IC processes have been integrated through these strategies, this
table may aid as a checklist for defining application requirements and economical
and technological possibilities. The design effort also has to be accounted for.

7.2 State-of-the-Art Integrated Applications

In this section, we review some of the latest integrated-resonator applications to

date. Although the technology requirements for integrating MEMS, NEMS, and
acoustic resonators may be the same, we organize the section in three main blocks.
In the first one, we discuss the case of CMOS-integrated MEMS and NEMS resona-
tors. In the second one, we explain the integration of SAW and FBAR devices. In the
last block, focus is on advanced 3D integration technologies. This section is will
describe some of the key technological aspects of the IC-to-resonator integration,
196 Integration of Resonator to CMOS Technologies

Table 7.1 Requirements and Features of Hybrid, Monolithic, and Heterogeneous Integration Strategies
Requirements Hybrid Monolithic Heterogeneous
Process specialization No Yes No
Complexity Low High High
Compatibility Low High Low
requirements
Flexibility-versatility High Low High
Development Costs Low High High
Production Costs High Low Low
Performance Low High High
Integration Wire-bonding, Via-through hole, Wafer-level-transfer,
Technologies MCM, packaging planar, surface surface micromachining,
micromachining, laser ablation, MCM,
advanced patterning packaging
tools (EBL, FIB)
MEMS-NEMS Features restricted to Only CMOS-compatible Suitable for a broad variety
Possibilities the available wire- materials, nanosized of MEMS (feature size
bonding technology features restricted to the available
MCM technology)
Geometry Onside (wire-bond- Onside, above-IC 3D-stacked (more than two
ing), MCM-stacked chips, in general)
(two chips)

thus giving emphasis to the process description rather than the application analysis.
Numerous applications of integrated resonators will be discussed in Chapter 8.

7.2.1 MEMS and NEMS Resonators

MEMS and NEMS resonators integrated with CMOS technologies are being widely
used in diverse sensor applications. Most recently, the resonators are monolithically
integrated according to the post-CMOS approach. The NEMS-resonator-based
mass sensors of Arcamone et al. implement a wafer-scale process based on post-
processing the CMOS wafer that contains read-out integrated circuits [20]. They
use nanostencil lithography (nSL) to pattern on top of the CMOS process several
types of micro- and nanoelectromechanical resonators, whose dimensions range
from 20 μm down to 200 nm. They have demonstrated the simultaneous nSL pat-
terning with a throughput of about 2,000 resonators per wafer. The technique is a
reliable method to define nanoscaled patterns and is also compatible with the
CMOS process, thus avoiding damage to the CMOS circuits.
The sequence of fabrication is depicted in the schema of Figure 7.7. Once the
CMOS process is completed, 80-nm-thick Al features are locally deposited and pat-
terned in a single step through metal evaporation and nSL. In this way, the nSL
wafer with the stencil is aligned on top of the CMOS wafer and the Al deposited by
using electron-beam evaporation [Figure 7.7(a)]. The Al features serve as a mask to
transfer its pattern to the resonator’s 600-nm-thick polysilicon structural layer. This
is carried out by RIE etching of the polysilicon layer (poly0), taking advantage of
the very high dry etching selectivity of Al with respect to Si [Figure 7.7(b)]. Before
completing the process, an UV-patterned photoresist mask is deposited and opened
7.2 State-of-the-Art Integrated Applications 197

Al material flux
NANOSTENCIL membrane

(a)

n-well
++
n implantations Resulting Al patterns Si-p bulk

Nanostencil-deposited
Al patterns

(b) n-well

Si-p bulk

UV-patterned
photoresist mask Aperture
for release

(c) n-well

Si-p bulk

Free-standing
structure

(d)
n-well

Si-p bulk
Field oxide (1 μm) poly0 (0.6 μm) poly1 (0.48 μm)

Interlevel oxide (1.3 μm) Al (1 μm) Passivation (0.8 μm)

Figure 7.7 MEMS and NEMS resonators monolithically integrated with CMOS: (a) the
post-CMOS resonator fabrication includes nanostencil lithography (nSL) to deposit an Al mask, (b)
polysilicon etching to define the structural layer of the resonators, (c) photoresist protection of the
CMOS circuits, and (d) and dry etching of the CMOS field oxide. (e) Nanocantilevers and (f) sus-
pended beams connected to the CMOS circuits are obtained. (© 2008 IOP Publishing Ltd. [20].)

in selected areas around the resonators to protect the CMOS circuits [Figure 7.7(c)].
Finally, immersing the wafer in HF etches the field oxide of the CMOS process, thus
releasing the resonators [Figure 7.7(d)]. The result of the process can be appreciated
in the pictures of Figure 7.7(e, f). The resonators are onside placed and connected to
signal interfacing and amplification CMOS circuits. Mechanical structures with line
widths down to 150 nm can be routinely achieved by the presented process
sequence.
198 Integration of Resonator to CMOS Technologies

(e)

100 μm

2 μm

(f)

100 μm

2 μm

Figure 7.7 (continued)

A second example of post-CMOS monolithic integration is the MEMS cantile-

ver resonator device fabricated through the Taiwan Semiconductor Manufacturing
Company (TSMC) 0.35-μm double polysilicon quadruple metal (2p4m) CMOS
process. According to this process, Chiou et al. fabricate a CMOS-MEMS prestress
vertical comb-drive resonator with a piezoresistive sensor to detect its static and
dynamic response [21]. The fabrication process of the comb-drive MEMS resonator
is depicted in the sequence of Figure 7.8. First, Figure 7.8(a) illustrates the cross-
section view of the CMOS process after full layer deposition and opening of the
7.2 State-of-the-Art Integrated Applications 199

Silicon Substrate Silicon Substrate

(a) (b)

Silicon Substrate Silicon Substrate

(c) (d)

Figure 7.8 Fabrication process of the comb-drive MEMS resonator: (a) completion of the CMOS
process (proving passivation apertures); (b) deposition and patterning a photoresist mask to pro-
tect the CMOS regions; (c) etching the field oxide by anisotropic RIE etching; and (d) etching the
Si substrate to release the suspended microstructure and removal of the photoresist. (© 2008 IOP
Publishing Ltd. [21].)

passivation. Prior to etching the interlayer dielectric, a thick photoresist is deposited

and patterned to protect the nonresonator areas, as illustrated in Figure 7.8(b).
Then, the dielectric is etched by anisotropic RIE with CHF3/O2 plasma, thus defin-
ing the sidewalls of the resonator and etching the field oxide as it shows Figure
7.8(c). Next, isotropic RIE of the doped region of Si is performed in a SF6/O2 atmo-
sphere, thus releasing the suspended structures of the MEMS resonators. Finally,
the photoresist is removed to complete the post-CMOS micromachining process, as
shown in Figure 7.8(d).
As we have seen, many of the CMOS-monolithically integrated MEMS and
NEMS resonators are placed onside of the CMOS process, and implemented
through the upper layers of the CMOS process. In general, the processes involve the
following:

1. The passivation layer is opened to allow etching of the structures.

2. The resonators are shaped by standard lithography. NEMS devices can be
patterned by using advanced techniques like nSL or e-beam lithography
(EBL).
3. The resonators are released by dry or wetetching of the dielectric layers of
the CMOS process, as the interlayer dielectric or the field oxide.
200 Integration of Resonator to CMOS Technologies

7.2.2 SAW and FBAR

As a difference with most CMOS-integrated NEMS and MEMS resonators, in
CMOS-based SAW and FBAR applications the process necessarily considers the
addition of piezoelectric thin films, which are nonstandard in conventional CMOS
technologies. In monolithic integration, these layers impose the need for
CMOS-compatible processes, thus implying the development of a compatibility
protocol. One of the issues is the low-temperature requirement for depositing
thin-film piezoelectric materials like AlN or ZnO. The second issue may be the
adaptation of the CMOS line to deposit nonstandard metals like Pt or Mo.

7.2.2.1 SAW Integration

Hybrid applications based on SAW and CMOS circuits exist since the miniaturiza-
tion of SAW has made feasible the size fitting between the chips of both technolo-
gies. The polylithic integration of a quartz wafer on silicon (QoS) to fabricate a
reference SAW-CMOS oscillator was demonstrated by Yunseong et al. [22]. The
SAW resonator and its connection with the ICs were fabricated after wafer-level
bonding of the quartz and silicon wafers. A few years later, Yeonwoo et al.
improved the technology to provide temperature-compensated SAW-CMOS
oscillators [23].
With the passing of time, the monolithic integration of SAW devices and CMOS
circuits has made possible the fabrication of SAW-CMOS reference oscillators on
the same chip. Furuhata et al. developed a reference oscillator that combined the
SAW device and the IC into a single unit, with the total thickness of nearly the same
as that of the IC [24]. They fabricated the IC in a standard CMOS 0.25-μm process
and flattened the wafer through chemical-mechanical polishing (CMP). After CMP,
they opened the passivation layer to access the upper Al metallization and deposited
the ZnO thin film on the wafer, as depicted in Figure 7.9(a). Next, deposition and
etching of an Al layer forms the IDTs of the SAW electrodes [Figure 7.9(b)]. A key
feature of this approach is the low-temperature deposition of the ZnO layer at
400ºC. This avoids damages to the circuits, especially the Al metallization. The
oscillator has the layout shown in Figure 7.9(c) and exhibits a central frequency of
545 MHz and a low phase noise of –120 dBc/Hz at 10 kHz.
Another example of SAW-CMOS monolithic integration is the fully integrated
two-port SAW resonator fabricated within the standard 0.6-μm American Micro-
systems Incorporated (AMI) CMOS three-metal two-poly CMOS process [25]. The
metal layers M1 and M2 of the CMOS process were used to implement the IDTs,
the reflectors, and the ground shield. The process entails three main steps:
(1) RIE-based micromachining to etch the interlayer SiO2, (2) ZnO deposition, and
(3) wet etching of ZnO. The first step allows accessing the ground shield and pat-
terning the IDTs. In the second step, the ZnO is sputtered to fill the space
between the IDT fingers, and the third step opens the contact pads for characteriza-
tion and interconnection. An improved implementation of this process employs the
third CMOS metal layer M3 to provide the resonator with an extra layer of reflec-
tors to contain the acoustic waves propagating above the transducer. Cross-
sectional cut views of both implementations are depicted in the schematic drawing
7.2 State-of-the-Art Integrated Applications 201

(a)

(b)

(c)
Figure 7.9 Monolithically integrated SAW-CMOS oscillator: the fabrication process involves: (a)
CMP, pad opening, and ZnO deposition, and (b) the upper Al layer deposition to form the SAW
electrode IDTs. (c) The layout of the SAW-CMOS oscillator. (© 2005 IEEE [24].)

of Figure 7.10. Resonators with different designs achieved parallel resonant fre-
quencies of 1.02 GHz, 941 MHz, and 605 MHz.

7.2.2.2 FBAR
The integration of FBAR has a relatively short history, in comparison to other
MEMS devices. As a high-Q factor device, FBARs have also attracted the attention
of RF IC and sensor-application designers. Recent developments in FBAR-to-
CMOS integration have stimulated the conception of integrated applications in
which FBAR is a key component, with IC integration as a requirement for proper
functionality of the system. Hybrid and monolithic integration strategies have been
investigated and implemented so far.
202 Integration of Resonator to CMOS Technologies

Figure 7.10 CMOS-based SAW resonator fabricated within the AMI standard CMOS process:
two-metal implementation that uses the first metal layer M1 to form the ground shield and the
second metal M2 to design the reflectors and the transducers; and three-metal process featuring
improved reflector design implemented through the third metal M3. (© 2007 IEEE [25].)

In the case of hybrid integration, the FBAR and IC chips are bonded to their cor-
responding circuit nodes. Examples of this integration approach are the oscillators
presented in [26–28], where the FBAR performs the crystal-like functionality in the
system. More recently, Avago Technologies [29] introduced a temperature-com-
pensated 604-MHz oscillator based on the FBAR-Colpitts topology. The quadratic
temperature function of its FBAR is compensated for by an oxide layer of the oppo-
site temperature coefficient factor (TCF) [30]. In Figure 7.11 the UC Berkeley–
Agilent’s implementation of a Pierce oscillator is shown. Double and short
wire-bonding between both chips ensures reduction of losses and parasitic induc-
tances. Area restrictions, added parasitic capacitances due to the bonding, and
batch processing of the integrated devices are some of the limitations of the hybrid
integration approach.

Figure 7.11 Hybrid integration of FBAR with low-power CMOS oscillator. (© 2003 IEEE [26].)
7.2 State-of-the-Art Integrated Applications 203

In the current art of monolithic integration, FBARs are placed above the circuit
using a post-CMOS strategy, thus saving die area. The approach disclosed in 1993
showed the concept [31] and a system integrating FBAR and radio circuitry con-
ceived in 2001 [32]. However, it was as late as 2005 when the first monolithic
FBAR-above-IC RF systems were demonstrated by the Martina consortium. Using a
0.25-μm BiCMOS process, this group implemented double-lattice filters [33], filter-
ing LNAs comprising two broadband amplifiers and one FBAR filter [34], and
a 5-GHz FBAR-based low-phase noise oscillator [35], among others, as shown in
Figure 7.12.
The interest in monolithic integration of FBAR spans from fully active ICs to
passive components as well, like CMOS inductors [36]. On the other hand,
CMOS-integrated MEMS, NEMS, and SAW resonators have already been demon-
strated as well [37, 38]. In spite of the elegance of monolithic integration, its com-

(a)

(b)
Figure 7.12 Post-CMOS monolithic integration of FBAR: (a) Martina’s concept of the FBAR-
above-IC integration; and (b) filtering LNA comprising two broadband amplifiers and
differential-lattice filter. (© 2005 IEEE [34].)
204 Integration of Resonator to CMOS Technologies

plexity, compatibility issues, costs, and technology-specific nature are the main
challenges of this approach.

7.2.3 Advanced 3D Integration Technologies: Wafer Level Transfer

Wafer level transfer (WLT) is an advanced 3D integration technology able to inte-
grate CMOS, MEMS, and other IC technologies. To accomplish this purpose, the
devices on a source wafer are transferred to a target wafer, all at the same time,
hence its name. The WLT technologies must also be able to achieve full releasing of
movable devices, as it is the case of MEMS and NEMS resonators. Nowadays, sur-
face micromachining, flip-chip, multichip module, and laser ablation, among oth-
ers, have been implemented to accomplish 3D CMOS integration of MEMS and
NEMS devices.
Two examples are representative of the 3D heterogeneous integration. One of
them was developed by IBM (Zurich Research Laboratory) in the context of the
“Millipede” project [17], and consists of three main processes. In the first process, a
cantilever array is fabricated according to conventional MEMS fabrication tech-
niques. The second process fabricates the CMOS metal and studs on which the can-
tilevers will be supported. In the third process, the MEMS-CMOS alignment and
lamination are carried out, and the cantilevers are released. Figure 7.13(a) depicts
the processes, and the final device results are observed in the SEM image of Figure
7.13(b).
A second example is the EPSON’s surface free technology by laser abla-
tion-annealing (SUFTLA) technology. The SUFTLA process enables the transfer of
thin-films or thin-film devices from their original substrate to any substrate by using
selective laser annealing. Low-temperature (below 425ºC) polycrystalline-silicon
thin film transistors (poly-Si TFTs) and TFT circuits could be successfully trans-
ferred from a glass or quartz substrate to plastic film without affecting the circuit
functionalities [39]. Using the SUFTLA technology, EPSON has demonstrated flexi-
ble TFT-LCD displays, polymer-based displays, TFT-SRAM memories, and the
industry’s first flexible 8-bit asynchronous microprocessor shown in Figure 7.14
[40].
Similar approaches have followed the 3D integrations of the monocrystalline
silicon mirrors fabricated using CMOS-compatible transfer bonding [41], which is
improved with nano-imprint resists to optimize the bonding of the CMOS and the
MEMS wafers [42]. As described in the foregoing examples, the two wafers are
placed one above the other and aligned, and their landing pads are bonded through
interconnection posts. These technologies, as we have studied, are implemented at
the wafer level, which is the main difference to previous hybridization of integrated
systems.

7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration

Taking advantage of the WLT approach, we can also fabricate FBARs integrated to
a standard CMOS process. As we have studied before, this will overcome the limita-
tions of hybrid or monolithically integrated resonators. According to this method,
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 205

I. MEMS part II. Post-CMOS part

I.1 Millipede cantilever array fabrication II.1 Metal/solder stud fabrication
1 μm thick stable lever anchor Contact pad
Tip on the heater platform
Stressed SiN layer Cantilever

SOI wafer

I.2 Wafer lamination on glass substrate III. Assembling

Glass Wafer
Polyimide III.1 MEMS-CMOS alignment and lamination

I.3 Seed wafer grinding/dry etching, BOX wet etching

III.2 Glass debonding and dry cantilever cleaning

I.4 Polyimide interlock backside structure

(a)

(b)
Figure 7.13 Heterogeneous integration technology for cantilever array fabrication: (a) process
description and (b) SEM image of the cantilever array. (© 2009 IBM [17].)

the FBARs on a first device wafer are transferred all at a time to a target the CMOS
wafer, thus obtaining 3D, floating structures. This approach is also suitable to other
MEMS and NEMS resonators.
In the WLT-based integration approach, we need to distinguish between two
main processes: (1) manufacturing of the resonator strictly speaking, and (2) its
wafer-level transfer into a CMOS substrate. On the other hand, we have a previ-
ously fabricated CMOS wafer. For example, the schematic drawings of Figure 7.15
explain the simplified sequence of a fabrication process involving this couple of pro-
cesses. First, both the CMOS and the FBAR wafers are fabricated according to inde-
pendent technological processes [Figure 7.15(a)]. Next, in either the CMOS or the
FBAR wafers, pillars providing future electrical connection and mechanical support
206 Integration of Resonator to CMOS Technologies

Figure 7.14 The industry’s first flexible 8-bit asynchronous microprocessor, fabricated with the
SUFTLA technology. (© 2005 EPSON [40].)

(a)

(e)

(b)

(f)

(c)

(g)
(d)

Figure 7.15 Heterogeneous integration process overview: (a) FBAR and CMOS wafers are fabri-
cated within independent processes; (b) pillars are fabricated on the CMOS wafer (or the FBAR
wafer, if desired); (c) the FBAR wafer is turned on and placed above the CMOS wafer (devices in
contact with pillars); (d) after soldering, FBARs are hard-connected to both the FBAR and CMOS
wafers; (e) side view of the FBAR-CMOS ensemble; (f) sacrificial layer wet etching and device
releasing; and (g) FBARs are attached to the CMOS substrate, with no presence of the former
FBAR-carrying Si substrate.
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 207

to the FBAR-CMOS ensemble are fabricated [Figure 7.15(b)]. Then, the FBAR
wafer is placed above the CMOS wafer, in order for the FBARs to contact the pillars
fabricated on the CMOS substrate [Figure 7.15(c)]. After hard interconnection at
appropriate temperature conditions, the FBAR-CMOS ensemble is integrated as
shown in Figure 7.15(d). A side view of the ensemble shows that a sacrificial layer
between the FBARs and its origin substrate provides mechanical support to them
[Figure 7.15(e)]. After the etching of the sacrificial layer, the substrate separates
from the devices, thus releasing them above the CMOS wafer [Figure 7.15(f)]. We
can appreciate that only the structural layers of the resonators remain above the
CMOS substrate, as depicted in the bird’s-eye scheme of Figure 7.15(g). This
method entails a significant step away the conventional hybrid integration strategy,
because wafer-level integration is prioritized to device-level dicing, among other
features. At this point, the wafer can be diced to extract the chips.
One can implement this method to integrated FBAR and CMOS for RF and sen-
sor applications. Its advantages are diverse: saving of die area, minimization of
power losses, reduction of the fabrication time, process compatibility, and suitabil-
ity to other MEMS devices, among others. As a difference to conventional FBARs or
the floating structure presented in [43], the whole FBAR structure is completely
transferred to the target substrate, which can also carry CMOS or another inte-
grated circuit technology. As a floating structure, the FBAR is placed above the IC
and is exclusively supported by two or more posts, which are also interconnecting
points to the CMOS substrate.

7.3.1 The Resonator Process

First, we describe the resonator manufacturing process. We can assume that the
CMOS wafer is fabricated according to a standard process and that it carries inte-
grated circuits. As we said before, the FBAR wafer is placed above and connected to
the CMOS substrate. The schematic drawings of Figure 7.16(a, b) show top and
cross-sectional views of the integrated FBAR-CMOS ensemble, with the FBAR
placed on top of the CMOS substrate. As observed, the pillars are the only support-
ing means of the FBAR, making it a three-dimensional, floating structure. At the
same time, the posts provide electrical contact between the FBARs and the CMOS
substrate. The CMOS wafer may carry passive devices, test structures, integrated
circuits, or a combination of all.
The technology can integrate other microresonator and IC technologies as well.
Since the floating configuration of the FBAR provides it with acoustical and
mechanical isolation to the substrate, the same process principles may be applied to
other micromachined movable structures, such as MEMS resonators, RF switches,
varactors, and tunable capacitors. Also, this method is suitable to integrate func-
tional blocks comprised of more than one resonator, like filters and duplexers.
Figure 7.17 depicts the main steps of the fabrication process. First, Figure
7.17(a) shows that a sacrificial layer made of a metal or dielectric material is depos-
ited on top of the FBAR wafer substrate. This is a characteristic seal of the technol-
ogy, because it enables the wafer-level releasing of the devices once transferred to
the CMOS wafer. Thus, one avoids the patterning or mask-driven processing of the
sacrificial layer. Instead, it is uniformly deposited on the wafer and etched at the
208 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.16 Schematic drawing of the floating FBAR structure, after fabrication using the hetero-
geneous integration method: (a) top view and (b) side view of the FBAR and CMOS substrate.

final step, when performing the FBAR and substrate integration. Next, Figure
7.17(b) illustrates the FBAR fabrication, where the metal of the first electrode is on
top of the sacrificial layer. Then, the piezoelectric and second electrode layers are
deposited at the steps shown in Figure 7.17(c, d), respectively. To achieve this pur-
pose, one may employ the thin-film deposition or growing technologies previously
described, like RF sputtering, epitaxial, or plasma-assisted deposition.
A realization of the previously described process may be as follows: the sacrifi-
cial layer can be made a 1-μm-thick phosphor silicate glass (PSG) film deposited on
a 500-μm-thick silicon wafer. Next, an 180-nm-thick Cr/Pt layer—30 nm of Cr for
Pt-adhering purposes—is deposited on top of the PSG, and the first electrode is
defined by liftoff. Then, the acoustic layer can be a 1μm-thick AlN layer that is
wet-etched by using OPD4262 developer to define the resonator shape. Finally, the
second electrode is also made according to the same process of the first electrode.
FBAR-device wafers fabricated with this process are observed in the optical pictures
of Figure 7.18. Broadly, this process is similar to the surface micromachining pro-
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 209

(a) (c)

(b) (d)
Figure 7.17 Schematic drawing of the FBAR fabrication process, before integration with the
CMOS substrate: (a) sacrificial layer deposition on top of the substrate; (b) deposition and pattern-
ing of the first metal electrode; (c) deposition and patterning of the acoustic layer (AlN); and (d)
deposition and patterning of the second metal electrode.

cess previously described in Chapter 4. The main differences between them are the
sacrificial layer material and the photolithography step to define its geometry. In the
surface-micromachined FBAR, the sacrificial layer is patterned only underneath the
region occupied by the resonator, according to the photolithography. On the other
hand, the 3D process described in this section performs etching on the whole wafer
surface without photolithography. Additionally, the passivation layer implemented
in the surface micromachining process disappears in this technology. Due to the
floating configuration of the resulting devices, isolating layers are avoided with this
process (the only contact points to the CMOS wafer are the pillars).
Although the previous example implements an AlN layer, compatibility issues
with the CMOS technology are irrelevant to choosing the thin-film acoustic mate-
rial, because the FBAR and the CMOS wafers have contact only after they have
been fabricated. Such a process offers more flexibility to choose the piezoelectric
material of the FBAR acoustic layer. This can be made of AlN, ZnO, lead zirconate
titanate (PZT), lead tantalum zirconate titanate (PLZT), or any other material with
good piezoelectric properties. Since an air interface between the FBAR and CMOS
wafers is provided at the end of the integration process, the FBAR manufacturing is
completed.

7.3.2 The CMOS Process

The target CMOS wafer may be fabricated following the standard process of com-
mercial CMOS technologies, and the cross-sectional view of a typical process is
210 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.18 Optical images of the floating FBAR process, before integration with the CMOS sub-
strate. The sacrificial PSG layer covers the whole surface of the wafer (no mask-driven patterning of
the PSG is required prior to etching): (a) general view of a section of the device-wafer surface; and
(b) detailed view of an FBAR device.

depicted in the schematic drawing of Figure 7.19. Alternative implementations in

bipolar, BiCMOS, SOI, or other IC technologies are also suitable for integration. As
usual, the CMOS process includes a passivation layer on top of the upper
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 211

Figure 7.19 Schematic representation of an exemplary CMOS process, suitable for integration
with the floating FBAR technology.

metallization layers. Also, opening windows are provided to allow for electrical
connection of the IC. For integration purposes, the opening windows will serve as
landing pads to the FBAR-to-CMOS interconnection. Although commercial CMOS
technologies provide a top passivation layer to protect the circuits and opening win-
dows to contact the circuit pads, rare cases without these features should be pre-
pared to complete the integration process. In those cases, additional postprocessing
steps would be required in the IC process, in order to deposit and open a passivation
layer to define the pad area on top of the integrated-circuit layers.
The passivation layer may be, for example, a 500-nm-thick PECVD SiO2 layer,
patterned through standard photolithography to open the pad connections.
Depending on the application of the CMOS technology (RF, high power, standard),
the material of the upper metal layer is made of a combination of Pt on Cr, Al on Cr,
or Al on Ti, among many possibilities. Depending on the available metal combina-
tion, the wet etching of the FBAR-CMOS ensemble has to be performed using suit-
able solutions in order to etch the sacrificial layer without damaging of the CMOS
layers. Equivalently, if we have technological options for the CMOS metallization,
we choose the upper metal configuration that meets the etching requirements of the
FBAR wafer.
Let’s study two CMOS technologies that we will consider for the heterogeneous
integration of the previously described FBAR process. The first one implements the
in-house CNM25 CMOS of the IMB-CNM (CSIC) (Barcelona), the cross-section
schematic view of which is depicted in Figure 7.20(a). In this process, an isolating
212 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.20 CNM25 2-metal technology implementation: (a) cross-sectional schematic view of
the CNM25 process; and (b) a coplanar test structure fabricated according to this process.

Si3N4 layer is deposited on the surface of the Si substrate, prior to the first Pt/Cr
metallization M1. In order to provide the process with adhesive metallization, a
Ti/Ni/Au layer is deposited on the upper metal (M2) to form landing pads to inter-
connect the CMOS with the FBAR wafer. The passivation layer can be made of a
resist or oxide. It serves as a mask for patterning of the Ti/Ni/Au, and, at the same
time, it becomes the CMOS passivation. The aspect of the CMOS surface with a
coplanar test structure designed with the M2 is observed in the optical micrograph
of Figure 7.20(b). This structure will be used to perform electrical characterization
of the FBARs once the heterogeneous integration is completed. The contact pads of
the FBAR will be connected to the CMOS pads signaled in the figure, and the reso-
nator will occupy the empty space in between them once the device transfer is
completed.
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 213

The second CMOS process that we study herein is one of the 2-poly, 4-metal
Austria Micro System’s (AMS) 0.35-μm AMS035c3b4 technologies [44]. This is a
commercial and standard CMOS process whose parameters depict Figure 7.21(a).
The top layer metallization (Metal 4) is made of Al/Ti, which is accessible by appro-
priate mask design and pad opening through etching of the protection and
passivation layers PROT1 and PROT2. The same coplanar structures of the previ-
ous example can be implemented with the Metal 4, as observed in the optical pic-
ture of Figure 7.21(b). Again, the region between the landing pads will be occupied
by the FBAR once the interconnection process of the FBAR and CMOS wafers is
completed.

7.3.3 The Wafer-Level-Transfer Process

Once the FBAR and the CMOS wafers are fabricated, they are heterogeneously inte-
grated by a wafer-level-transfer process. First, the deposition and etching of an alloy
or stack of metals suitable for soldering or electroplating is carried out on the FBAR
electrodes, thus defining two or more landing pads for the interconnection, as
shown in Figure 7.22(a). This is required if the FBAR electrodes are made of materi-
als unsuitable to adherence with soldering or electroplating processes. Thus, the
materials composing the landing pads may be selected from the group of Ni, Au,
and Ti, among others. Next, Figure 7.22(b) illustrates how the pillars are placed on
top of the landing pads. The techniques and methods required to implement them
are diverse and include mask-driven deposition of soldering paste, metal deposition,
and electroplating assisted deposition, among others. Then, the FBAR wafer is
turned over and placed above the CMOS wafer, as represented in Figure 7.22(c).
Both wafers are handled to align the FBAR pillars and the opening windows of the
CMOS wafer. Manual or automatic pick-and-place systems may be used for posi-
tioning, alignment, and soft-contacting of FBAR and CMOS wafers. Since the
FBAR wafer is flipped at this step, the second electrode becomes the closest one to
the CMOS substrate, whereas the substrate is now the upper layer of the suspended
structure. Current technologies provide pillars that are bigger than FBARs. There-
fore, an air interface is created at the gap between the FBAR and the CMOS wafers.
The mechanical interconnection can be done by soldering bumps heated and sol-
dered to the CMOS pads, at one side, and to the FBAR pads, at the other one. Alter-
natively, electroplating or deposition techniques may replace soldering. In the last
step of the process, depicted in Figure 7.22(d), the FBARs are released through wet
etching of the sacrificial layer. Since the sacrificial layer is between the FBARs and
their substrate, it is lifted off when the etching is completed, thus separating the
FBARs and completing the transfer to the CMOS wafer. Now the FBARs exhibit
two air interfaces, and they have mechanical freedom to vibrate, as no acoustical
coupling exists between them and the CMOS wafer. This feature is a necessary
condition to guarantee high-quality factors.
To perform the interconnection of the FBAR and CMOS wafers, the landing
pads can be made of Ni, Au, and Ti, for example, with thicknesses of 30 nm, 20 nm,
and 50 nm, respectively. The pillars may be soldering bumps, which are applied on
the wafer by a masking-aided appliance and an alignment system. In the stereo-
scopic picture of Figure 7.23(a), a close-up of the pillars on the resonators can be
214 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.21 AMS035 technology implementation of the CMOS-substrate for heterogeneous inte-
gration with FBAR substrate. (a) Cross-sectional schematic view of the AMS035 process. (After:
[44].) (b) Optical image of a coplanar test structure implemented within this process (the FBAR is
to be located in between the landing pads).
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 215

(a) (c)

(b) (d)

Figure 7.22 Cross-sectional view of the heterogeneous integration process of FBARs to CMOS
substrates: (a) FBAR fabrication and landing-pad deposition; (b) contacting/supporting post depo-
sition; (c) FBAR-to-CMOS interconnection (the FBAR wafer is flipped, face-down); and (d) sacrificial
layer wet etching. (© IEEE 2008 [45].)

observed. The pick-and-place system is used now to align and contact the FBAR and
CMOS wafers. In this example, the ensemble is introduced in an oven or a heater to
accomplish the hard interconnection between both FBAR and CMOS substrates.
Finally, the PSG sacrificial layer is attacked through wet etching on an HF buffered
solution, thus releasing the Si substrate of the FBAR wafer. The SEM image of Fig-
ure 7.23(b) shows the integrated system, where the FBARs are connected to the
CNM25 coplanar structures. It can be seen that the thin-film FBAR structure is
placed about 40–50 μm above the substrate.

7.3.4 Characterization and Technology Optimization

We are interested on finding out whether the heterogeneous integration process is
effective to obtain integrated devices with a high-quality factor. For this reason, we
employ electrical characterization of the resonators to evaluate the process. The
coplanar test structures will serve as this purpose. As commented in the previous
section, the heterogeneous integration process was performed on two different stan-
dard CMOS technologies. Another example of CNM25-integrated FBARs is shown
in Figure 7.24(a), whereas Figure 7.24(b) shows another resonator above an
AMS035 CMOS substrate. Both devices are located 40–50 μm above their corre-
sponding CMOS substrates (focus is laid on the resonators). In each case, the FBAR
was integrated to a coplanar GSG test structure with pitch of 150 μm, in order to
216 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.23 Heterogeneous integration implementation: (a) stereoscopic image of the FBARs and
the pillars placed on their contacting pads; and (b) SEM image of FBARs above the CNM25 CMOS
substrate.

(a) (b)
Figure 7.24 Top-view optical images of floating FBAR integrated with two different CMOS sub-
strates: (a) in-house standard CMOS, and (b) AMS 0.35-μm technology. (© IEEE 2008 [45].)
7.3 Wafer-Level-Transfer-Based FBAR-to-CMOS Integration 217

perform on-the-wafer characterization. The measurement setup comprises a

coplanar RF-probe station and a network analyzer (the signal path is through the
CMOS-FBAR-CMOS interfaces).
In Figure 7.25(a) the magnitude and phase of the S21 parameter of an exem-
plary device are plotted, with a resonance frequency of around 2.4 GHz observed.
The Q factors of heterogeneously integrated FBARS and RIE-based FBAR devices
with the same layout are compared in Figure 7.25(b). As observed, the floating
FBAR process exhibits a better performance than the RIE-based process, thus suf-
fering lower damping and achieving reduced substrate coupling. Looking at these
curves, it should be said that larger variations in both the attenuation and phase
shift are required to accomplish the in-band and out-of-band specifications of mod-
ern filter-applications. This goal can be achieved with higher Q-factor devices,
which can be obtained after appropriate layout design of the electrodes, among
other actions [46]. Due to the direct relationship between the Q factor and the
FBAR electrode size, the high-Q-factor goal can be reached with bigger-sized
devices.

(a)
Q factor

Device

(b)
Figure 7.25 Electrical characterization of FBAR integrated with CNM25 standard-CMOS sub-
strate: (a) magnitude and phase of the transmission (S21) parameter (characterized device in the
SEM image of the inset); and (b) comparative performance of the Q factors of 3D-integrated and
RIE-based FBARs (exemplary devices of each technology in the insets). (© IEEE 2008 [45].)
218 Integration of Resonator to CMOS Technologies

The thin-film structure of the FBAR makes it fragile and highly stressed. Thus,
small devices should be fabricated to diminish the break-off risk and improve the
process yield. On the other hand, the soldering technology requires deposition of
bumps with a pitch of the order of 250 μm, each bump having a diameter of 80 μm
(after heating). This leads to bigger-device layout design, thus increasing the die
area. As a consequence, the mechanical rigidity and stability of the devices are
reduced. A third issue is that postintegration cleaning of the integrated wafers is
required, due to residual composites deposited on the surface of the CMOS wafer
(and/or the FBAR wafer), as colateral products of bump melting. Additionally, the
electrical properties of the soldering bumps as interconnecting paths are still to be
studied, especially for the gigahertz-range frequency band.
Another technological possibility to pillar fabrication is electroplating. With
this option, the soldering paste is replaced by a more uniform electroplated column,
grown on top of either the CMOS or the FBAR wafers. The resulting product is a
cleaner and sharper FBAR-CMOS ensemble, with reduced pitch dimensions, and,
consequently, improved area efficiency. No additional layout design effort is
required, since the same mask for landing pad and soldering bump definition is used
for the electroplated pillar. The sequence of Figure 7.26 represents the technological
process of the optimized FBAR-CMOS integration. The main difference in respect
to Figure 7.22 is the description of the electroplating process for pillar construction.
Starting from the FBAR fabrication, the first electrode, AlN layer, and second
electrode are successively deposited on top of a sacrificial layer (e.g., PSG), as
depicted in Figure 7.26(a). Next, a passivation layer is deposited and patterned to
feature the landing pads [Figure 7.26(b)]. Typically, this layer is a thick resist and

(a) (d)

(g)

(b) (e)

(h)
(c) (f)

Figure 7.26 Electroplating-based optimized process sequence for heterogeneous integration of

FBAR and CMOS wafers: (a) FBAR fabrication; (b) resist deposition; (c) seed material; (d) oxide
deposition; (e) electroplating; (f) resist-oxide removing; (g) FBAR-CMOS wafers integration; and
(h) sacrificial layer etching (FBAR released).
7.4 Summary 219

will be used to define the electroplating area. Then, a seed material for electroplat-
ing is deposited on the wafer as depicted in Figure 7.26(c). This material is a metal
with good conductive properties, like Cu or Au. Since the process intends for
wafer-level integration, the seed material is deposited along the whole wafer sur-
face. Next, the same mask of Figure 7.26(b) is used to pattern another passivation
layer, as depicted in Figure 7.26(d). This second passivation layer will contribute to
limiting the growing area of the electroplated material, according to the supporting
post mask design [Figure 7.26(e)]. Once the electroplating deposition is finished, the
resist, seed material, and second passivation layer are removed to provide the final
shape of the FBAR wafer before the integration [Figure 7.26(f)]. Note that Figure
7.26(b–f) could also be performed on the CMOS wafer. However, in spite of the
standard nature of the CMOS process, this implementation describes a more gen-
eral solution. The last two steps are basically the same as those already described at
the beginning of this section, where the FBAR-CMOS interconnection is carried out
[Figure 7.26(g)] and the sacrificial layer is attacked to release the Si substrate, thus
completing the process [Figure 7.26(h)].
In considering this process, a new design with reduced pillar diameter and pitch
can be achieved. In the previous example, they were equal to 80 μm and 250 μm,
respectively. With the optimization introduced by this process, these values can be
reduced to 25 μm and 100 μm. Shrinking the pillar dimensions and pitch is of spe-
cial relevance to this integration approach, because it permits a more compact and
robust design of the FBAR structure. Since the AlN and electrode intersections
define the effective resonator area, the remaining electrode and AlN areas can be
reduced to provide shorter structures. This alleviates the high stress of the floating
FBAR structure. Additionally, this feature lets a higher FBAR density in the wafer,
due to reduced distance between devices. Figure 7.27 illustrates these effects. Figure
7.27(a) shows top-view representations of FBARs, the landing pads, and the effects
of optimized electroplating process in the pad size reduction. In the same way, the
pitch reduction permits a higher number of FBARs to be located on a determined
area, as depicted in Figure 7.27(b). Clearly, the new structures are expected to be
more robust and efficient.

7.4 Summary

In this chapter, we have reviewed the concepts and technological approaches for
CMOS integration of micro- and nanodevices. Hybrid, monolithic, and heteroge-
neous integration, along with implementation examples of them, was discussed.
Special emphasis was given to MEMS, NEMS, SAW, and FBAR resonators, thus
introducing state-of-the-art integrated-resonator technologies and applications. At
this point, we paid our attention to the latest 3D integration techniques implement-
ing advanced processes. With this review, we entered into Section 7.3 to study a
method of heterogeneous integration of FBAR and CMOS technologies [47]. We
have also studied two implementations of the heterogeneous integration of FBARs
and two different CMOS processes. As we have seen, the integrated FBARs exhibit
a floating, three-dimensional structure above the CMOS substrates [45]. This
method has shown us how compatibility requirements can be minimized to achieve
220 Integration of Resonator to CMOS Technologies

(a)

(b)
Figure 7.27 FBAR layout optimization due to the electroplating-based fabrication of the support-
ing posts: (a) pitch reduction; and (b) pad-size reduction, compared in both cases to the soldering
bump process.

full integration of such different technologies. Since the processes do not require
special manufacturing restrictions—like deposition temperature of the FBAR
acoustic layer—enhanced flexibility and versatility may be achieved for the
FBAR-to-CMOS interconnection technology.
Heterogeneous integration exhibits many benefits over traditional hybrid and
monolithic integration of micro- and nanoresonators. Compared to standard
flip-chip implementations [48], for example, this method presents two main advan-
tages: wafer-level integration and no carrying substrate attached to the resonator.
This saves the long processing time of batch resonator-to-CMOS integration. Het-
erogeneous integration can be seen as a kind of packaging technology serving as a
protective means [49]. Additionally, it features interconnection to the integrated
ensemble. On the other hand, it avoids the constraints of costly and technology-spe-
cific monolithic integration [50]. In summary, some of the main benefits introduced
by heterogeneous integration can be stated as follows:
7.4 Summary 221

1. Technology-independent integration process;

2. No need for integration-oriented resonator technology (e.g., process
temperature);
3. No need for integration-oriented IC technology;
4. Versatility and flexibility of the IC technology;
5. Possibility of wafer-level or die-level integration;
6. Reduced die area (compared to onside resonator integration);
7. Versatility and flexibility of the interconnection technology (e.g., solder
bumping, electroplating).

References

[1] Baltes, H., et al., “CMOS MEMS—Present and Future,” Proc. IEEE Intl. Conf. MEMS
2002, Las Vegas, NV, January 20–24, 2002, pp. 459–466.
[2] ST Microelectronics, System on Chip, http://www.st.com/stonline/products/technolo-
gies/soc/soc.htm.
[3] Geschke, O., H. Klank, and P. Telleman, (eds.), Microsystem Engineering of
Lab-on-a-Chip Devices, New York: John Wiley & Sons, 2004.
[4] Herold, K. E., and A. Rasooly, (eds.), Lab-on-a-Chip Technology: Fabrication and
Microfluidics, San Francisco, CA: Caister Academic Press, 2009.
[5] Royal Society of Chemistry, Lab on a Chip, http://www.rsc.org/Publishing/Jour-
nals/lc/index.asp.
[6] Tea, N. H., et al., “Hybrid Postprocessing Etching for CMOS-Compatible MEMS,”
IEEE/ASME J. Microelectromech. Syst., Vol. 6, 1997, pp. 363–372.
[7] Latorre, L., et al., “Characterization and Modeling of a CMOS-Compatible MEMS Tech-
nology,” Sens. Actuator A-Phys., Vol. 74, No. 1-3, 1999, pp. 143–147.
[8] Lund, J. L., et al., “A Low Temperature Bi-CMOS Compatible Process for MEMS RF Reso-
nators and Filters,” Proc. Solid-State Sensor, Actuator and Microsystems Workshop 2002,
Hilton Head Island, SC, June 2–6, 2002, pp. 38–41.
[9] Sundaresan, K., et al., “Electronically Temperature Compensated Silicon Bulk Acoustic
Resonator Reference Oscillators,” IEEE J. Solid-State Circuits, Vol. 42, 2007,
pp. 1425–1434.
[10] Pourkamali, S., Z. Hao, and F. Ayazi, “VHF Single Crystal Silicon Capacitive Elliptic
Bulk-Mode Disk Resonators—Part II: Implementation and Characterization,”
IEEE/ASME J. Microelectromech. Syst., Vol. 13, 2004, pp. 1054–1062.
[11] Smith, J. H., et al., “Embedded Micromechanical Devices for the Monolithic Integration of
MEMS with CMOS,” Proc. IEEE Intl. Electron Devices Meeting IEDM 1995, Washing-
ton, D.C., December 10–13, 1995, pp. 609–612.
[12] Infineon Technologies, Pressure Sensors, http://www.infineon.com/sensors.
[13] Hierold, C., “Intelligent CMOS Sensors,” Proc. IEEE Intl. Conf. MEMS 2000, Miyazaki,
Japan, January 23–27, 2000, pp. 1–6.
[14] Florence, J. M., and L. A. Yoder, “Display System Architectures for Digital Micromirror
Device (DMD)–Based,” Proc. SPIE Projection Displays II, San Jose, CA, January 29, 1996,
Vol. 2650, pp. 193–208.
[15] Verd, J., et al., “Integrated CMOS–MEMS with On-Chip Readout Electronics for
High-Frequency Applications,” IEEE Electron Dev. Lett., Vol. 27, 2006, pp. 495–497.
[16] Campabadal, F., et al., “CMOS Degradation Effects due to Electron Beam Lithography in
Smart NEMS Fabrication,” Proc. SPIE Symposium on Microtechnologies for the New Mil-
lennium 2005, Sevilla, Spain, May 9–11, 2005, Vol. 5836, p. 667.
222 Integration of Resonator to CMOS Technologies

[17] IBM Zurich Research Laboratory, Device Integration, http://www.zurich.ibm.com/st/

server/microdevice.html.
[18] Geske, J., et al., “Vertical and Lateral Heterogeneous Integration,” Appl. Phys. Lett.,
Vol. 79, 2001, pp. 1760–1762.
[19] Roozeboom, F., et al., “Passive and Heterogeneous Integration Towards a Si-Based Sys-
tem-in-Package Concept,” Thin Solid Films, Vol. 504, 2006, pp. 391–396.
[20] Arcamone, J., et al., “Full-Wafer Fabrication by Nanostencil Lithography of
Micro/Nanomechanical Mass Sensors Monolithically Integrated with CMOS,”
Nanotechnology, Vol. 19, 2008, 305302.
[21] Chiou, J.-C., L.-J. Shieh, and Y.-J. Lin, “CMOS-MEMS Prestress Vertical Cantilever Reso-
nator with Electrostatic Driving and Piezoresistive Sensing,” J. Phys. D: Appl. Phys.,
Vol. 41, 2008, 205102.
[22] Yunseong, E., et al., “Reference SAW Oscillator on Quartz-on-Silicon (QoS) Wafer for
Polylithic Integration of True Single Chip Radio,” IEEE Electron Dev. Lett., Vol. 21, No.
8, 2000, pp. 393–395.
[23] Yeonwoo, K., E. Yunseong, and L. Kwyro, “Polylithic Integration of a Reference SAW
Quartz-on-Siliconoscillator for Single-Chip Radio,” Digest of Technical Papers IEEE
International Solid-State Circuits Conference ISSCC 2002, Vol. 1, 2002, pp. 186–459.
[24] Furuhata, M., et al., “Development of Monolithic CMOS-SAW Oscillator,” Proc. IEEE
Intl. Ultrason. Symp., 2005, Vol. 4, Rotterdam, the Netherlands, September 18–21, 2005,
pp. 2194–2197.
[25] Nordin, A. N., and M. E. Zaghloul, “Modeling and Fabrication of CMOS Surface Acoustic
Wave Resonators,” IEEE Trans. on Microwave Theory Tech., Vol. 55, No. 5, 2007,
pp. 992–1001.
[26] Otis, B. P., and J. M. Rabaey, “A 300-μW 1.9-GHz CMOS Oscillator Utilizing
Micromachined Resonators,” IEEE J. Solid-State Circuits, Vol. 38, 2003, pp. 1271–1274.
[27] Chee, Y. H., A. M. Niknejad, and J. M. Rabaey, “An Ultra-Low-Power Injection Locked
Transmitter for Wireless Sensor Networks,” IEEE J. Solid State Circuits, Vol. 41, 2006,
pp. 1740–1748.
[28] Zhang, H., et al., “5GHz Low Phase-Noise Oscillator Based FBAR with Low TCF,” Proc.
13th International Conference on Solid-State Sensors, Actuators and Microsystems Trans-
ducers 2005, Seoul, Korea, June 5–9, 2005, pp. 1100–1101.
[29] Avago Technologies, http://www.avagotech.com.
[30] Pang, W., et al., “A Temperature-Stable Film Bulk Acoustic Wave Oscillator,” IEEE Elec-
tron Dev. Lett., Vol. 29, 2008, pp. 315–318.
[31] Dunn, W. C., et al., “Monolithic Circuit with Integrated Bulk Structure Resonator,” U.S.
Patent 5,260,596, November 1993.
[32] Lee, K., Y. S. Eo, and S. Hyun, “Single-Chip Radio Structure with Piezoelectric Crystal
Device Integrated on Monolithic Integrated Circuit and Method of Fabricating the Same,”
U.S. Patent 6,285,866, September 2001.
[33] Dubois, M.-A., et al., “Integration of High-Q BAW Resonators and Filters Above IC,”
IEEE Intl. Solid-State Circuits Conf. Dig. of Tech. Papers, 2005, San Francisco, CA, Febru-
ary 6–10, 2005, pp. 392–393.
[34] Carpentier, J. F., et al., “A SiGe:C BiCMOS WCDMA Zero-IF RF Front-End Using an
Above-IC BAW Filter,” IEEE Intl. Solid-State Circuits Conf. Dig. of Tech. Papers, 2005,
San Francisco, CA, February 6–10, 2005, pp. 394–395.
[35] Aissi, M., et al., “A 5 GHz Above-IC FBAR Low Phase Noise Balanced Oscillator,” Proc.
IEEE Radio Frequency Integrated Circuits Symposium RFIC-2006, Vol. 1, San Francisco,
CA, June 11–13, 2006, pp. 25–28.
[36] Figueredo, D., et al., “Thin Film Bulk Acoustic Resonator (FBAR) and Inductor on a
Monolithic Substrate and Method of Fabricating the Same,” U.S. Patent 6,710,681, March
2004.
7.4 Summary 223

[37] Forsen, E., et al., “Ultrasensitive Mass Sensor Fully Integrated with Complementary
Metal-Oxide-Semiconductor Circuitry,” Appl. Phys. Lett., Vol. 87, 2005, 043507-1-3.
[38] Ilic, B., et al., “Attogram Detection Using Nanoelectromechanical Oscillators,” J. Appl.
Phys., Vol. 95, 2004, pp. 3694–3703.
[39] Shimoda, T., and S. Inoue, “Surface Free Technology by Laser Annealing (SUFTLA),”
IEEE Electron Devices Meeting IEDM 1999 Digest Tech. Papers, Washington, D.C.,
December 5–8, 1999, pp. 289–292.
[40] Seiko Epson Corporation, “Epson Develops the World’s First Flexible 8-Bit Asynchronous
Microprocessor,” http://www.epson.co.jp/e/newsroom/2005/news_2005_02_09.htm
[41] Niklaus, F., S. Haasl, and G. Stemme, “Arrays of Monocrystalline Silicon Micromirrors
Fabricated Using CMOS Compatible Transfer Bonding,” IEEE/ASME J.
Microelectromech. Syst., Vol. 12, No. 4, 2003, pp. 465–469.
[42] Populin, M., et al., “Thermosetting Nano-Imprint Resists: Novel Materials for Adhesive
Wafer Bonding,” Proc. IEEE Intl. Conf. MEMS 2007, Kobe, Japan, January 21–25, 2007,
pp. 239–242.
[43] Kim, E. -K., “Thin Film Resonator and Method for Manufacturing the Same,” U.S. Patent
6,849,475 B2, February 2005.
[44] Austria Microsystems, AMS035c3b4, 2-poly 4-metal 0.35 μm CMOS technology,
http://www.austriamicrosystems.com.
[45] Campanella, H., et al., “Thin-Film Bulk Acoustic Wave Resonator Floating Above CMOS
Substrate,” IEEE Electron Device Lett., Vol. 29, 2008, pp. 28–30.
[46] Su, Q. X., et al., “Thin-Film Bulk Acoustic Resonators and Filters Using ZnO and
Leadzirconium-Titanate Thin Films,” IEEE Trans. on Microwave Theory Tech., Vol. 49,
2001, pp. 769–778.
[47] Campanella, H., et al., “Thin-Film Bulk Acoustic Wave Resonator and Method for Per-
forming Heterogeneous Integration of the Same with Complementary-Metal-Oxide-Semi-
conductor Integrated Circuit,” European Patent Office App. 07380041.9, February 2007.
[48] Vanhelmont, F., et al., “A 2GHz Reference Oscillator Incorporating a Temperature Com-
pensated BAW Resonator,” Proc. IEEE Intl. Ultrason. Symp. 2006, Vancouver, BC, Octo-
ber 3–6, 2006, pp. 333–336.
[49] Feld, D., et al., “A Wafer Level Encapsulated FBAR Chip Molded into a 2.0 mm/spl times/
1.6 mm Plastic Package for Use as a PCS Full Band Tx Filter,” Proc. IEEE Intl. Ultrason.
Symp. 2003, Vol. 2, Honolulu HI, October 5–8, 2003, pp. 1798–1801.
[50] Dubois, M. -A., et al., “Integration of High-Q BAW Resonators and Filters Above IC,”
IEEE Intl. Solid-State Circuits Conf. Dig. of Tech. Papers, 2005, San Francisco, CA, Febru-
ary 6–10, 2005, pp. 392–393.
 CHAPTER 8

Sensor Applications
A micro- or nanosensor is the part of a microsystem that inputs information into the
system comprised of the electronic circuit that conditions the sensor signal, the actu-
ator responding to the electrical signals generated within the circuit, and the sensor
itself. Thus, the sensor is the interface to the outside world. It converts the input sig-
nal from its physical domain to the electrical domain (the actuator works in the
opposite way). The technological advancements of the micro- and nanoelectronic
engineering fields have led to a drastic reduction of size and prize in sensors and
have enabled sensor integration into single microelectronic chips. In this way, many
microsensors (and most recently nanosensors) take advantage of the silicon technol-
ogies and the well-known electrical and mechanical properties of this material.
Other materials have also been studied and are important elements of modern
miniature MEMS, NEMS, and acoustic sensors.
There exist many detection mechanisms including piezoresistivity,
piezoelectricity, electrostatic, magnetic, optical, and resonant techniques. Resonant
sensors offer many benefits, like improved sensitivity and accuracy, and reduced
power consumption, among others. Due to the action of the external excitation, res-
onator-based sensors change their resonance frequencies, which is detected by a
read-out electronic circuit. This change is directly proportional to the magnitude of
the input signal, which can be given in one of the physical domains previously
mentioned.
The chapter focuses on resonant sensors, and more particularly the main
emphasis is on mass and mechanical sensors. Mass sensors detect the amount of
mass deposited on the surface of the resonator and the physical mechanism of detec-
tion is known as the mass-loading effect. Thus, the added mass of a film or body
deposited on the resonator brings about down shifting of its resonance frequency.
On the other hand, mechanical sensors perform pressure, force, acceleration,
torque, inertial, and flow sensing, and the physical mechanism involves a strain
added to the resonator structure, thus increasing the resonance frequency of the
device. In the following sections, we study the concepts, technologies, and some of
the most popular MEMS-, NEMS-, and FBAR-based resonant sensor applications.

8.1 Resonant Sensing Performance

Resonant sensing has many benefits over nonresonant techniques, like improved
sensitivity, resolution, and accuracy. The recent advances of material science and
fabrication technology have also enabled the manufacturing of low-power and

225
226 Sensor Applications

highly temperature stable resonators. Table 8.1 compares the main features of reso-
nant sensing against capacitive and piezoresistive techniques.
Nevertheless, performance benefits are achieved at the cost of added complexity
of fabrication processes, in particular those related with packaging and vacuum
operation. As we discuss next, the resonator’s Q factor, its geometry, and the char-
acterization technique determine the ultimate resolution and performance of the
sensor.

8.1.1 The Role of the Q Factor on Resolution

In Chapter 1 we reviewed the damping mechanisms affecting the Q factor of reso-
nators. Now, we revisit the Q factor from the sensor performance perspective and
more specifically with the frequency resolution, because the measurand (mass, pres-
sure, acceleration) is a function of frequency (or time). The smaller the frequency
value that can be discriminated is, the smaller the measurand magnitude that we are
able to quantify is. Thus, two aspects relate with frequency discrimination: phase
stability of the resonator, and the frequency response evaluation method.
The phase stability issue arises from the Q factor definition considering the
open-loop phase response φ(ω) of the resonator, which we discussed in Chapter 4.
We reproduce it convenience of the reader:

ωS ∂ φ
Q= (8.1)
2 ∂ω

Determining the frequency resolution Δf passes from quantifying the phase sta-
bility Δφ. The practical matter is how to proceed with experimental phase data and
(8.1). Typically, we perform statistical analysis on phase data collected at the reso-
nance frequency to extract mean values and standard deviation. The latter can be
considered the actual phase deviation Δφ around a mean phase value, using Allan’s
deviation or similar criteria [2]. On the other hand, and referring to Figure 5.8(a),
for example, we know that fast slope changes φm occur near the resonance fre-
quency. With this at hand, the frequency resolution is expressed by:

Δφ
Δf = (8.2)
φm f 0

Table 8.1 Comparative Performance of Resonant, Piezoresistive, and

Capacitive Sensing Techniques
Item Resonant Piezoresistive Capacitive
8 5 5
Resolution 1 part in 10 1 part in 10 1–10 parts in 10
Accuracy 10–1,000 ppm 500–10,000 ppm 100–10,000 ppm
Power consumption ∼1 nW ∼1 mW < 0.1 mW
Temperature factor ∼1 ppm/ºC > 1,000 ppm/ºC 1–10 ppm/ºC
Source: [1].
8.1 Resonant Sensing Performance 227

The size of resonators also affects the Q factor, as we know from the damping
models of electromechanical and acoustic resonators examined in previous chap-
ters. Therefore, we state three considerations regarding the resolution of the sensor:

1. Sensitivity augments proportionally to the Q factor, because the frequency

resolution grows with Q (regarding (8.2)).
2. Sensitivity augments proportionally to the size of the resonator [regarding
(1.9), (3.12), (3.13), and, in general, the electromechanical models of
MEMS and FBARs].
3. Sensitivity augments proportionally to the central frequency of the resonator
[3].

Characterization method is the second aspect that determines the frequency res-
olution of the sensor. Frequency-domain techniques include S-parameter character-
ization, amplitude spectrum, and power spectrum, among others. These techniques
do require specification of the intermediate frequency (IF) filter bandwidth, which
determines the amount of out-of-band noise filtering. Time-domain techniques, on
the other hand, allow the Q factor characterization and read-out circuits suitable
for electronic integration with the sensor in full system implementation. Fre-
quency-analysis integrated circuits [4], zero-crossing, and threshold-crossing fre-
quency counting techniques [5], among others, have demonstrated outstanding
resolution of 1 Hz and less. Actually, the characterization-related resolution is lim-
ited only by the available IC technology and instrumentation hardware, as the ulti-
mate resolution of the sensor relies on the resonator design and physical damping
mechanisms.

8.1.2 Performance Features and Parameters

Material properties, operation mode, and geometry also affect the sensitivity per-
formance of resonant sensors. If we look into the design equations of resonators,
and more specifically of the sensor, we realize that sensitivity is dependent on geo-
metric parameters and the mechanical configuration of the structure on which the
resonator is located.
In order to compare resonant and nonresonant techniques, we need to point out
the sensitive elements of capacitive, piezoresistive, piezoelectric, and resonant
devices. Sensitivity is measured in terms of a dimensionless quantity termed the
gauge factor (GF), which equals the relative change in the detection variable divided
by the actuated variable. For example, piezoresistive and piezoelectric devices are
strain sensitive. Capacitive transduction is sensitive to the parallel-plate distance
and overlap area. Electromechanical resonance depends on thickness and length,
and acoustical resonance on the thickness. Additionally, the sensor application
determines the GF form, as the case of mass sensors where the density (mass) is the
actuated variable. Table 8.2 summarizes the gauge factors, gauge functions, and
sensitive variables for resonant and other sensing techniques.
Sensitivity is maximized when the detection device is located at the point where
the gauge factor is maximum (i.e., where the biggest response is obtained with a
minimum effort on the actuated variable). The stress contour plot of the cantilever
228 Sensor Applications

Table 8.2 Performance Features and Parameters of Sensing Techniques

Sensing Gauge Function Gauge Factor Sensitive Variable
Piezoresistive ρL ΔR R Length L
R= ; GF = ;
A ε
R: resistance, ρ: density, ε: strain (ε ΔL/L)
L: length, A: area
Capacitive A ΔC C Area A
C = εr ε0 ; GF1 = ;
t ΔA A Thickness t
C: capacitance, ΔC C
ε: permittivity, GF1 = ;
Δt t
t: thickness
Piezoelectric Qi = d ij F j ; ΔQ Q Thickness t or length L
GF1 =
Q: charge, d: piezoelectric ε
constant, F: force ε: strain (DL/L)
2
Resonant 1 ⎛ γn ⎞ E⋅ I⋅ L ε: strain (DL/L) Length L
fn = ⎜ ⎟
(electromechanical) 2π ⎝ L ⎠ m
Resonant (acoustical) 1 E Δ f f0 Thickness t or density ρ
f0 = GF1 =
4 πt ρ Δt t

in Figure 8.1 illustrates the concept: stress is maximum near the clamping surface.
Thus, a big change in piezoresistance or frequency will be achieved with low strain
if the piezoresistor or the resonator is placed on these regions.

Figure 8.1 Stress contour plot of a cantilever: maximum sensitivity is obtained around the clamp-
ing points.
8.2 Mass Sensors 229

8.2 Mass Sensors

Miniature mass sensors have wide applications in physical, chemical, and biologic
systems, and their sensitivities have made them starring devices in new, convergent
microdevices and nanodevices. The operating principle behind mass sensors is the
mass loading effect. Therefore, the mass deposited on the resonator brings about
down shifting of its resonance frequency. Within the operation limits of the sensor
technology, the frequency shifting is directly proportional to the amount of the
deposited mass. The fabrication technology of the resonator conditions the manner
in which the mass can be deposited.
Mass loading can happen in air, gas, or liquid media, and a physical or chemical
interaction between the resonator surface and the medium is needed to fix the mass
to the resonator. In some applications, the mass is temporarily deposited, because it
is a volatile compound that evaporates or reacts with the media. Strictly speaking,
mass loading occurs when a thin-film material grows or is deposited on the resona-
tor in a localized or distributed way, thus covering part or the whole active surface
of the device. For example, thin-film devices like FBARs experience mass loading
when their composing layers are stacked to fabricate the device metal electrodes.
Resonant mass sensors are characterized by their frequency responses, which
are evaluated through frequency-domain read-out circuits or instrumentation.
Their sensitivity depends on the resonance frequency, the detection system resolu-
tion, and the Q factor of the resonator, among others. Thus, resonators with high
resonance frequencies and Q factors and low-noise detection will exhibit improved
sensitivity than that of low-frequency, low–Q factor, and poor-frequency resolu-
tion devices. An unavoidable consequence of mass loading is that it increases the
damping, thus decreasing the resonator Q factor. For that reason, the resonator is
designed with regard to the application and the sensing medium, which also con-
tributes to the Q factor reduction (damping losses are severe in liquid and dense
fluids).
Commercially available mass sensors are implemented with acoustic QCM,
SAW, and BAW resonators. Also, MEMS- and NEMS-based sensors are being
investigated to perform distributed or localized high-sensitivity mass detection.
They can be used as distributed-mass or as localized-mass sensors. A distrib-
uted-mass sensor makes use of its whole surface to detect the amount of mass on it,
which is deposited, grown, or adhered to it by physical or chemical means. On the
other hand, the localized-mass sensor detects the mass of a body with lateral dimen-
sions (in contact with the resonator surface) small enough in comparison to those of
the electrode area. The mass sensing capabilities of each technology rely on the dif-
ferent mechanisms described in the following sections.

8.2.1 MEMS-Based Microbalances

The MEMS-based resonant mass sensors perform mass detection by detecting the
frequency shifting of the fundamental mechanical or first-overtone resonance
modes. As we study later, other sensors make use of higher-order flexural modes to
detect the added mass on their surface. The sensing structure of most of the
MEMS-based sensor is made of silicon (Si). In this way, MEMS sensors exhibit high
230 Sensor Applications

mass sensitivity due to the excellent mechanical properties of Si. Another advantage
of Si-made sensors is their full compatibility with standard CMOS processes and
low fabrication costs. Nevertheless, other materials can also be implemented as the
structural layer of the resonator.
The MEMS sensors have physical dimensions with a high aspect ratio, typically
ranging from units to hundreds of micrometers. The cantilever is the preferred
structure of highly sensitive mass sensors. Optimization of its dimensions achieves
mass sensitivity maximization through a high surface area–to-mass ratio. In Chap-
ter 2 we examined the resonance frequency of the MEMS cantilever, reproduced
here for convenience:
2
1 αn EIl
fn = (8.3)
2π l 2 m eff

where fn is the n-mode resonance frequency, E is the Young’s modulus, l is the beam
length, I is the moment of inertia, and meff is the effective mass of the cantilever. The
coefficient αn depends on the n-mode number [6]. Equivalently, (8.3) may be
expressed in terms of the spring-damped-mass system:

1 k
fn = (8.4)
2π n
m eff

where k is the spring constant of the cantilever and m effn

is the effective mass of the
nth mode of resonance. The mass Δm deposited on the cantilever changes both the
stiffness and the effective mass of it. If the added mass is small enough, in compari-
son to the cantilever mass, we can assume no change of the elastic properties and
neglect the stiffness variation. Thus, the net effect of Δm contributes to decrease the
resonance frequency to a value defined by the first-order approximation:

Δf 1 Δm
=− (8.5)
fn 2 m eff
n

where Δf is the resonance frequency shifting. The optimization of the cantilever

design aims to maximize the mass sensitivity Sm of the device:

Δf
Sm = (8.6)
Δm

Looking at (8.4) and (8.5), we can see how smaller cantilevers will exhibit
higher resonance frequencies and improved mass sensitivity. This can be controlled
through different approaches. Hwang et al. optimized the design of a piezoelectric
cantilever intended for biosensing applications by the Taguchi method [7], thus
using the first resonance frequency, the separation factor between the resonance fre-
quencies, and the sensing signal of the piezoelectric cantilever as the object func-
tions [8]. According to this method, they optimize the sensitivity of the cantilever to
the binding between the antigen and the antibody by considering the combined
effect of the length, width, thickness, and width-to-length ratio of the cantilever. In
8.2 Mass Sensors 231

this way, they maximize the first resonance frequency and the separation factor
between the first and the second resonance frequencies.
Narducci et al. have studied how the sensitivity of Si T-shaped cantilever reso-
nators for mass sensing applications augments by detecting the higher-order reso-
nance modes and reducing the device dimensions [9]. The T-shaped cantilevers
shown in Figure 8.2 are released through back-side etching, and they are piezoelec-
trically actuated. The detection is carried out by four piezoresistors in a Wheatstone
bridge configuration. They demonstrated how, for these resonators, performing the
detection at the second resonance mode improves the mass sensitivity by a factor of
4.1. Also, the sensitivity gain is 16 when the length and width of the cantilever are
halved, thus achieving an Sm value equal to 1.3 Hz/pg.
Further reduction of microcantilever dimensions has been proposed by Davila
et al., with the purpose of detecting Bacillus anthracis Sterne spores in air and liquid
[10]. They implement microcantilevers whose resonance frequency is detected with
the use of a laser Doppler vibrometer (LDV) after thermal-noise source excitation.
Aside from the investigations on the viscous effects in the cantilever response,
Davila et al. performed biological experiments involving suspended spores on the
cantilevers in air and water. Figure 8.3(a) shows a dark field photograph of a
20-μm-long cantilever with spores deposited on it and then after measurements in
water, and the SEM image of Figure 8.3(b) details a small group of spores on the

Figure 8.2 T-shaped cantilever for mass detection applications (length: 400 μm, width: 300 μm,
thickness: 15 μm). (© 2009 Elsevier [9].)
232 Sensor Applications

(a) (b)
Figure 8.3 Microcantilevers for detecting Bacillus anthracis Sterne spores in air and liquid: (a) dark
field photograph of a 20-μm-long cantilever after measurements in water; and (b) SEM image
showing different spores on the cantilever. (© 2007 Elsevier [10].)

cantilever in air. For detection of spores in water, the cantilevers were

functionalized with antibodies in order to fix the spores onto the surface. With this
setup, they could detect a minimum of 2 spores (740 fg) and 50 spores (139 pg) in
air and water, respectively, with 20-μm-long, 9-μm-wide, and 200-nm-thick canti-
levers, thus obtaining measurement sensitivities of 9.23 Hz/fg for air and 0.1 Hz/fg
for water.
Aside from microcantilevers, other MEMS structures have been explored to
perform high-sensitivity mass detection. Ismail et al. designed a degenerate mode
resonant mass sensor with a cyclically symmetric structure [11]. The simplest struc-
ture with these features is a circular diaphragm, which supports pairs of independ-
ent modes of vibration sharing a common natural frequency, thus referred to as
degenerated modes. If an extra mass is added to the structure over predefined
regions, then the degeneracy can be broken, and this produces a separation of the
previously identical frequencies. The frequency split is the output of the sensor and
is proportional to the added mass. Due to its design, the sensor is frequency
self-compensated and the ambient effects equally influence both modes, thus pre-
serving the frequency split for a given mass. Figure 8.4 shows the layout of the
polysilicon diaphragm resonator.
Other interesting MEMS structures are the high-frequency silicon columnar
microresonators reported by Kehrbusch et al. [12]. Within their fabrication tech-
nique, they can control the dimensions of the microresonators on a scale of at least 1
μm, as the SEM images of Figure 8.5 show. According to the characterized mechani-
cal properties of the silicon columns, they resonate at the lowest flexural mode of
3–7 MHz with quality factors of up to 2,500 in air and ~8,800 under vacuum condi-
tions. The columnar microresonators were operated as a mass microbalance with a
sensitivity of 1 Hz/fg and an experiment-deduced mass detection limit of 25 fg.
Multiple MEMS-based mass sensors implement arrays of devices intended for
multiple particle detection. Villarroya et al. developed an on-chip array of
polysilicon cantilevers, which are monolithically integrated with the CMOS
read-out circuit [13]. They successfully demonstrated the electrostatic excitation
and detection of 4-resonator and 8-resonator cantilever arrays with submicron
dimensions of the integrated cantilevers. As long as they scaled the cantilever
dimensions, they achieved a mass sensitivity of 28 Hz/fg, the expected mass resolu-
tion in vacuum being of units of femtograms (fg). The optical images of Figure 8.6
8.2 Mass Sensors 233

Figure 8.4 Polysilicon MEMS circular diaphragm for mass detection. (© 2006 Institute of Physics
Publishing [6].)

Figure 8.5 Silicon columns of various cross sectional geometries (scale bar: 30 μm). The insets
reproduce SEM images from the top of a single column revealing the cross-sectional geometry
after RIE etching (circular [9-μm diameter], elliptical [15.1 × 7.4 μm ], square [9.1-μm width], and
2

rectangular [14.9 × 7.3 μm ]). (© 2008 American Institute of Physics [12].)

2
234 Sensor Applications

Figure 8.6 Monolithically integrated 8-cantilever mass sensor: the cantilevers are integrated with
a multiplexing and read-out CMOS circuit (a zoomed-in view of the cantilevers is on the right side
of the picture). (© 2006 Elsevier [13].)

show one of the 8-cantilever sensors monolithically integrated with a read-out and
multiplexing CMOS circuit.
So far, we have reviewed some examples of MEMS-based resonant mass sen-
sors. Although cantilevers are simple to fabricate and feature high mass-sensitivity,
diaphragm or columnar structures have also been explored. We have noticed how
the mass sensitivity is improved by reducing the dimensions of the resonator, from
the units of hertz per picogram of microsized devices to the hertz per femtogram of
nearly nanosized resonators. Hence, we would expect to improve even more the
sensitivity through further reduction of the resonator dimensions.

8.2.2 Ultrasensitive NEMS Mass Sensors

The approach of nanoelectromechanical resonators for high-sensitive mass sensing
applications has a notorious visibility in the NEMS field. Technological challenges
and the broad spectrum of convergent applications that are being envisioned for the
future explain their high impact. As long as nanomechanical resonators have been
realized and achieved fundamental resonance frequencies exceeding 1 GHz, with Q
factors in the range 103 < Q < 105, their small active masses and high Q factors have
translated into inertial mass sensitivities below 1 fg. Ekinci et al. have evaluated the
ultimate mass sensitivity limits for nanomechanical resonators operating in a vac-
uum, and they related these limits with a number of fundamental physical noise pro-
cesses. These analyses indicate that nanomechanical resonators have the potential
for mass resolution at the level of individual molecules [14].
The advancements of NEMS resonator technology have also brought new con-
cepts to mass detection and leading-edge applications, especially those in which dif-
ferent disciplines converge. Nowadays, detection of viruses is possible thanks to the
nanoresonator mass sensor technology. Ilic et al., for example, used a resonating
mechanical cantilever to detect immunospecific binding of viruses, captured from
liquid. To demonstrate the concept, they used a nonpathogenic insect baculovirus
to test the ability to specifically bind and detect small numbers of virus parti-
cles. Arrays of surface micromachined, antibody-coated polycrystalline silicon
nanomechanical cantilevers were used to detect binding of various concentra-
tions of baculoviruses in a buffer solution. The 0.5 μm × 6 μm cantilevers shown in
8.2 Mass Sensors 235

−19
Figure 8.7 exhibit mass sensitivities on the order of 10 g/Hz, thus enabling the
−15
detection of an immobilized antibody monolayer with a mass of about 3 × 10 g.
This means that the mass of single-virus particles bound to the cantilever can be
detected. Based on control experiments, they showed the nanocantilever’s capabil-
ity to adsorb very small amounts (<50 attograms) of the antibody layer [15].
NEMS resonators can also be used as self-detection instruments, and, due to
their reduced size, they can be used in high-frequency applications. In scanning
probe microscopy (SPM), the cantilever-based sensors generally use low-frequency
mechanical devices of microscale dimensions or larger. The detection systems of
these microscopes involves costly and nontransportable off-chip sensors, most of
them based on laser or photo-diode arrays. Because the dispersion of such optical
systems is greater than the actuation magnitude, they are unsuitable for detecting
the nanoscale displacement of nanocantilevers. The implementation of self-sensing
nanoresonators changes the paradigm of detection and notoriously improves the
performance of SPM. Mo-Li et al. fabricated and described the operation of
self-sensing nanocantilevers with fundamental mechanical resonances up to very
high frequencies (VHF). The devices use integrated electronic displacement trans-
ducers based on piezoresistive thin metal films to perform the nanodevice read-out
and are observed in Figure 8.8. This nonoptical transduction enables fast SPM and
VHF force sensing (the detection of 127-MHz cantilever vibrations was demon-
strated). With the smallest devices, they successfully achieved chemisorption mea-
surements in air at room temperature, with a mass resolution less than 1 attogram
[16].
Besides silicon-based resonators, carbon nanotubes (CNT) have also been dem-
onstrated to be suitable for high-resolution mass detector implementation. Jensen et
al. demonstrated that, at room temperature, carbon-nanotube-based
nanomechanical resonators can achieve atomic mass resolution. They built the
device depicted in Figure 8.9, which is essentially a mass spectrometer with a mass
sensitivity of 1.3 × 10−25 kg/Hz1/2 or, equivalently, 0.40 gold atoms /Hz1/2. Unlike
traditional mass spectrometers, nanomechanical mass spectrometers do not require

Figure 8.7 The nanocantilever used to detect immunospecific binding of viruses: the cantilever
has length l = 6 μm, width w = 0.5 μm, and thickness t = 150 nm with a 1 μm × 1 μm paddle (the
scale bar corresponds to 2 μm). (© 2004 American Institute of Physics [15].)
236 Sensor Applications

(a) (b) (c) (d)

15 15
400

300
ΔR/R(10 )

10 10
−6

200

5 5
100

0 0 0

50 k 55 k 1.56 M 1.59 M 7.9 M 8.0 M 126 M 128 M

Frequency (MHz)

Figure 8.8 Piezoresistively detected frequency response from a family of SiC nanocantilevers to a
1 nN AC drive signal at room temperature in a vacuum: fundamental-mode resonance frequencies
are: (a) 52.1 kHz, (b) 1.6 MHz, (c) 8 MHz, and (d) 127 MHz. The insets show SEM micrographs
(angled perspective) of the devices, with dimensions 33 μm × 5 μm, 10 μm × 2 μm, 2.5 μm × 0.8
μm, and 0.6 μm × 0.4 μm, respectively. All are fabricated from a single-crystal, 70-nm-thick SiC
epilayer. (© 2007 Nature Publishing Group [16].)

(a) (b)

(c)
Figure 8.9 Nanomechanical mass spectrometer device and schematics of the measurement
setup: (a) TEM images of a nanomechanical mass spectrometer device constructed from a double-
walled carbon nanotube; and (b) physical layout of the entire nanomechanical mass spectrometer
apparatus. Gold atoms are evaporated inside a UHV chamber and travel a distance dCNT before
adsorbing to the nanotube device and, consequently, lowering its resonant frequency. A shutter
may be inserted to interrupt mass loading. The QCM provides an alternative means of calibrating
the system through measurement of mass flux. (c) Schematic of the mechanical resonance detec-
tion circuit. (© 2008 Macmillan Publishers Ltd. [17].)
8.2 Mass Sensors 237

the potentially destructive ionization of the test sample. They are more sensitive to
large molecules and could eventually be incorporated on a chip [17].
High resonance frequency and low mass are the main parameters driving a high
mass sensitivity, and CNTs meet both conditions. Recent experiments have demon-
strated the ultra high mass sensitivity of CNTs. Since the mass of a nanotube is as
low as a few attograms, even a tiny amount of atoms deposited onto the nanotube
makes up a significant fraction of the total mass. Also, nanotubes are ultrarigid
mechanically, which is a key material property to push up the resonance frequency.
The 1-nm-sized carbon nanotube resonator reported by Lassagne et al. is actuated
by electrostatic interaction, as Figure 8.10(a) depicts. Using electron-beam lithogra-
phy, nanotubes are connected in transistor geometry to two Cr/Au electrodes. The
SEM image of Figure 8.10(b) shows the CNT and the electrodes. When an AC volt-
age Vg oscillating at a frequency f is applied on the back-gate of the wafer, an oscil-
lating electrostatic force is generated on the nanotube at the same frequency. The
motion of the nanotube, induced by the electrostatic force, modulates the gate
capacitance Cg, and, in turn, it modulates the charge in the nanotube. The mass
responsivity is 11 Hz/yg and the mass resolution is 25 zg at room temperature (1 yg
−24 −21
= 10 g and 1 zg = 10 g). By cooling the nanotube down to 5K in a cryostat, the

(a)

(b)
Figure 8.10 Carbon-nanotube-based mass sensor. (a) Experimental setup for mass sensing. Chro-
mium atoms are deposited onto the nanotube resonator in a Joule evaporator and the mass of the
atoms adsorbed on the nanotube is measured. (b) Scanning electron microscopy image of the
nanotube resonator connected to the electrodes. (© 2008 American Chemical Society [18].)
238 Sensor Applications

signal for the detection of mechanical vibrations is improved to a resolution of 1.4

zg [18].
As we have seen, the NEMS resonator, microfluidics, silicon, and carbon
nanotube technologies are integrated to offer ultrasensitive mass detection. The
CNT-based resonators are the most sensitive mass detectors to date. Nevertheless,
from the technological point of view, some issues should be resolved to implement
nanoelectromechanical resonators in biodetection environments. In such cases,
sample delivery is generally difficult and requires the entire device chip to be sub-
mersed into an analyte. Stiction problems are common to this scale as well. In addi-
tion, the viscous damping diminishes the mass sensitivity, so high vacuum is
required. Some innovative solutions have been presented to address these issues,
like the encapsulation of nanocantilever arrays in individually accessible, parallel
microfluidic channels, which are used for delivery of liquids and nitrogen [19]. At a
larger scale, and based on different transduction mechanisms, acoustic resonators
are also being evolved to deal with the attogram-resolution and reduced-size
technological challenges.

8.2.3 Acoustic Resonator Distributed-Mass Sensors

An acoustic resonator distributed-mass sensor features a piezoelectric layer to per-
form mass measurements, the most popular implementations being quartz crystal
microbalances (QCM) and FBAR-based sensors.
In Section 3.3.4 we studied the Mason’s model, which explains to a great extent
the frequency shifting caused by the mass loading effect: an added film or body
deposited on top of the resonator electrode changes the phase conditions of the
acoustic wave propagation. The phase shifting translates into frequency shifting of
the resonance modes of the device. Thus, the frequency change Δf is proportional to
the amount of deposited mass (Δm) and is evaluated in the Sauerbrey-Lostis equa-
tion [20] by:

⎛ ρ t ⎞
f m = f 0 ⎜1 − m m ⎟ (8.7)
⎝ ρ0t 0 ⎠

In (8.5) the term at the right of f0 corresponds to the frequency change Δf due to
added mass Δm, where ρm and tm are the density and thickness of the added-mass,
and ρ0 and t0 are the density and thickness of the unloaded resonator. In this way,
the frequency change relative to the unloaded resonance frequency can be solved as:

Δf ρ t Δm
≈− m m =− (8.8)
f0 ρ0t 0 m0

Equation (8.8) is valid if Δm is less than 2% of the initial mass of the resonator
m0 .
The sensing performance of distributed-mass sensors is evaluated by using four
parameters. First, we consider the mass segnsitivity Sm [cm2/g], which is defined as:
8.2 Mass Sensors 239

1
Sm = (8.9)
∑ ρit i
i

where ρi and ti are the density and thickness of each layer in the resonator stack. The
second parameter is the frequency responsivity Rf = Δf/f0, where Δf is the minimum
detectable frequency shift. Third, the minimum detectable mass change per unit
2
area Δm [g/cm ] can be evaluated from the mass sensitivity and the frequency
responsivity as:

Rf
Δm = (8.10)
Sm

2
Finally, the fourth parameter is the mass responsivity per area rm [g/Hz/cm ],
whose definition is related to both the resonance frequency and the mass sensitivity
as:

1
rm = (8.11)
f0 × S m

8.2.3.1 Quartz-Crystal Microbalance (QCM)

A QCM is a miniature mass sensor, or balance, that measures the amount of mass
deposited on it as the change of its resonance frequency. It is used in most applica-
tions to operate in the thickness shear resonance mode at a frequency in the 1- to
30-MHz range. When a voltage is applied to the metal electrodes, it causes oscilla-
tion of the plate at the shear-mode resonance frequency.
The QCM, as a bulk acoustic wave resonator, is built as a parallel-plate device
made of AT or SC-cut quartz crystal thin film sandwiched between two metallic
electrodes. Due to the good piezoelectric properties of quartz, the oscillation is gen-
erally very stable with the high quality of oscillation, which enables QCM to be high
sensitivity mass sensors. Its electrical behavior is described by the classical
Butterworth-Van-Dyke model. Different crystals like α-cut quartz (shear) and new
materials (langasite LGS La3Ga5SiO14, gallium ortophosphate GaPO4) are also
employed in QCM fabrication.
QCMs are operated in air, gas, or liquid, although the damping and Q factor
reduction caused by the contact of the resonator with the sensing medium imposes
the need for designing the device and its operation mode according to the applica-
tion. In air or gas phase interfaces, the damping is lower than in liquid media; for
that reason, higher-sensitivity thickness-shear resonance modes are preferred in
these cases. In liquid media the damping is very critical, thus drastically reducing the
Q factor and sensitivity of the resonator. Because of this, shear horizontal surface
acoustic wave (SH-SAW) mode [21, 22], Love-wave mode [23], and torsional mode
resonators are implemented for liquid-media applications. Although less sensitive,
lateral-wave resonances suffer less damping, and, in the end, the overall sensitivity
of the QCM is improved.
Based on the previously defined operation modes, QCMs can be categorized in
gravimetric and nongravimetric sensors. The thickness-shear mode sensors are
240 Sensor Applications

strictly gravimetric mass sensors. As stated by Mecea [24], mega-gravity inertial

forces arise in this kind of sensor, which enable the mass-sensing measurement. On
the other hand, nongravimetric sensors take advantage of viscoelasticity and viscos-
ity effects. Thus, a change in the viscoelastic properties of the material in the liquid
phase causes the resonance frequency to change because of the acoustic impedance
variation. However, the viscoelastic parameters themselves usually depend on fre-
quency, and it is often difficult to discriminate against inertial and viscoelastic
effects. The resonance frequency shift of the QCM immersed in a liquid environ-
ment is explained by models similar to the Sauerbrey equation, although they
include new terms related to the viscoelastic correction [25–28].
A typical QCM exhibits rm in the units of pg/Hz/cm2, and minimum detectable
2
mass Δm of ng/cm . With these values, QCMs find application in gravimetry; parti-
cle detection in biotechnology, using functionalized surfaces, thin-film thickness
monitoring in thermal, e-beam, sputtering, magnetron, ion, and laser deposition;
and in situ monitoring of fluid properties. Karamollaoglu et al. developed a
QCM-based DNA biosensor for detection of genetically modified organisms
(GMOs), which promises real-time, label-free, direct detection of DNA samples for
the screening of GMOs [29].

8.2.3.2 FBAR-Based Distributed-Mass Sensors

FBARs exhibit high frequency and mass sensitivities. Thus, it makes the FBAR tech-
nology a suitable candidate for biomolecular or chemical-detection applications.
Mass-sensing systems, in which quartz microbalances (QCM) have been the key
technology, are now rethought to be implemented with FBARs, taking advantage of
their higher sensitivity in comparison to conventional QCMs [30–33]. As in QCMs,
uniform loading of the active area of the electrode occurs when the FBAR is sub-
mersed in the sensing media (air or liquid).
As already discussed in Section 6.3.2, and based on the mass loading effect, the
uniform thin-film deposition is a way of tuning the FBAR resonance frequency. The
mass loading analysis of the FBAR with a magnesium fluoride (MgF2) thin film
−17
deposited on its electrode showed that a responsivity Rm value as good as 1.4 × 10
g/Hz can be reached. Based on those experiments, we now study the performance of
the thin-film-deposited FBAR as a distributed-mass sensor.
First, the theoretical and experimental values of the mass responsivity per area
rm are calculated. From the frequency shifting and deposited-mass data shown in
Figure 6.12, the theoretical value of the mass sensitivity Sm is calculated from (8.9)
by taking into account each layer in the FBAR’s stack configuration. Once obtained
Sm, the responsivity rm is obtained by the inverse product of Sm by the series reso-
nance frequency f0 (2.4 GHz), as expressed in (8.11). Table 8.3 shows the sensitivi-
ties and responsivities for each layer and for the FBAR. A thin film of 20 nm of
MgF2 is used to calculate these parameters.
Then, we find the experimental sensitivity of the FBAR-based sensor. For this
purpose, we evaluate the minimum mass change Δm and the minimum detectable
frequency Δf. The experimental, minimum detectable mass change Δm is deter-
mined by the minimum detectable frequency shift Δf. To do that, the phase noise
amplitude Δϕ is measured from zero-span S-parameter data acquired at the series
8.2 Mass Sensors 241

Table 8.3 Theoretical Mass Sensitivity and Responsivity of the FBAR-Based

Uniform-Mass Sensor
Density Layer Sensitivity Inverse Layer
3 1
FBAR’s Layer i
[g/cm ] Thickness ti [cm] Si=1/(ri ti) Sensitivity mi=Si
MgF2 3.176 2.0 × 10−6 157,430.73 6.35 × 10−6
−5
Pt (top) 21.45 1.5 × 10 3,108.00 3.22 × 10−4
Ti (top) 4.506 3.0 × 10−6 73,975.44 1.35 × 10−5
−4
AlN 3.260 1.0 × 10 3,067.48 3.26 × 10−4
Pt (bottom) 21.45 1.5 × 10−5 3,108.00 3.22 × 10−4
−6
Ti (top) 3.176 3.0 × 10 73,975.44 1.35 × 10−5
Distributed-mass 1 997.12
Sm = m0 =
∑ ρit i
2
sensitivity Sm [cm /g]
i

Mass responsivity 1 4.18 × 10

−13
rm =
per area rm f0 × S m
2
[g/Hz/cm ]

resonance frequency f0, and divided by the S21-phase slope ϕm, as given by (8.3).
Figure 8.11 plots a zero-span acquisition of the S21 parameter phase. As observed
in the figure, the maximum phase deviation from the mean value is ±0.08°, which is
−6
divided by the phase slope φm value −4.21 × 10 deg/Hz. This is calculated from
differentiation and evaluation of the S21 phase at the series resonance frequency f0
(see an example in the plots of Figure 5.8). The minimum frequency shifting is found
to be Δf = 19 kHz.
Using the foregoing results, the experimental values of Sm, Δm, and rm are calcu-
lated from (8.9) to (8.11), after the thin-film deposition. The Sm values are averaged
to obtain Δm and rm, given the minimum detectable frequency shift Δf previously
calculated. Table 8.4 shows individual and averaged Sm, mass sensitivity and
responsivity values.
According to the examples of Tables 8.3 and 8.4, the experimental uni-
form-film mass sensitivity and responsivity values are within the 80% of the theo-
2
retical values. Also, the experimental Δm equals to 9.8 ng/cm , which is pretty

−10.10
Phase noise
Mean value

−10.15 0.08
S21phase (deg)

deg

−10.20 0.08
deg

−10.25

0 200 400 600

Sample number (freq: 2.29218 GHz)

Figure 8.11 Zero-span frequency-domain acquisition of the FBAR response at resonance, for eval-
uation of the phase noise and minimum frequency shifting (with no loading).
242 Sensor Applications

Table 8.4 Experimental Mass Sensitivity and Responsivity of the

FBAR-Based Uniform-Mass Sensor
Frequency Shift Average Layer
Thickness (Average) Sensitivity
ti [cm] Δfi [Hz] Si = fi /(f0 ri ti )
MgF2 density: 3.176 g/cm3 2.0 × 10−7 9.6 × 10
5
630
f= 19 kHz 5.0 × 10−7 3.7 × 10
6
954
f0 = 2.4 GHz
1.0 × 10−6 6.2 × 10
6
817
−6 7
2.0 × 10 1.3 × 10 816
1 Δf i 804.4
Average mass
2
Sm = m0 = ∑
sensitivity Sm [cm /g] N i f0 ρit i
Minimum detectable Δf 9.84 × 10
−9
Δm =
mass m [g/cm2] f0 ⋅ S m
Mass responsivity Δm 5.18 × 10−13
rm =
Δf
2
per area rm [g/Hz/cm ]

competitive to similar mass-sensing technologies. Table 8.5 compares four realiza-

tions of the uniform-film sensor, including a quartz-based QCM sensor. Although
the Δf of QCMs is better, the highest resonance frequency of FBARs explains their
improved Sm and rm performance.

8.2.4 FBAR-Based Localized-Mass Detection

An FBAR-based localized-mass sensor performs high-sensitivity mass detection of a
material or compound deposited on the electrode of the FBAR. The SEM image of
Figure 8.12 shows an example of this concept, a C/Pt/Ga deposited by FIB on the
top electrode of an FBAR being observed [30]. Parameters different than those of
uniform-mass sensors evaluate the performance of localized-mass detectors. Mass
responsivity Rm [Hz/g] is the first one, and it is defined as the change in the fre-
quency response per unit mass change:

Δf
Rm = (8.12)
Δm

Table 8.5 Uniform-Mass-Sensing Performance for Different FBAR Realizations

(Longitudinal-Resonance Mode)
FBAR (2.2 GHz) QCM FBAR FBAR
(the Example of (10 MHz) (1.5 GHz) (2GHz)
This Section) [34] [33] [34]
Minimum detectable 19,000 1 400 15,400
frequency shift f [Hz]
Average mass sensitivity 804.4 0.54 726 937.5
2
Sm [cm /g]
Minimum detectable 9.8 × 10−9 5.2 × 10
−9
1.0 × 10
−9
21 × 10
−9

mass m [g/cm2]
−13 −12 −13 −12
Mass responsivity 5.18 × 10 1.79 × 10 9.18 × 10 1.36 × 10
2
per area rm [g/Hz/cm ]
8.2 Mass Sensors 243

Figure 8.12 Localized mass deposited on the top electrode of an FBAR. (© 2006 American Insti-
tute of Physics [30].)

Due to the experimental approach of the sensor characterization, it is often

more convenient to deal with the inverse responsivity Rm−1 [g/Hz]. Thus, the expres-
sions responsivity and inverse responsivity are used indistinctly in this context. By
−1
taking the minimum detectable frequency Δfmin and Rm , the minimum mass change
Δmmin is calculated from [35]:

Δfmin
Δmmin = (8.13)
Rm

As we can see in (8.14), the minimum detectable frequency determines the ulti-
mate sensitivity of the mass sensor. The value of Δfmin depends on the measurement
setup and requires fine characterization of the noise sources to minimize it.
Localized-mass detection using FBARs as sensing devices was demonstrated in
2006. Through FIB-assisted experiments, FBARs achieved responsivities as high as
10−19 g/Hz and minimum detectable masses of units of femtograms [30]. Further
studies on the FBAR-based sensitivity also demonstrated the sensor’s dependence
on the location and size of the deposited mass [36].
Localized-mass sensors are at least one order of magnitude more sensitive than
uniform-mass sensors, although the sensitivity decreases with the area and location
of the localized mass. It was found out that a change of the deposited mass location
causes different magnitudes of the frequency shifting. Within the 2% mass limit
imposed by the Sauerbrey-Lostis equations, the responsivity of the localized-mass
sensor changes for different-sized mass deposition, which opens a broad variety of
size-based applications. The mass-loading configuration can thus be designed
according to the purpose of the target application. The plot of Figure 8.13 shows
that the FBAR-based mass detection responsivity increases when the size or the
mass of the localized-load are reduced.
Regarding the deposited mass location, the center of the FBAR sensor is the
region of the electrode with the best responsivity, as the study of Table 8.6 details.
Previous works on atomic-force-microscopy-based scanning of FBAR electrodes
revealed that the amplitude of oscillation at resonance is related to the mode shape
at this frequency. In [37], the vertical displacement of the FBAR’s longitudinal reso-
nance modes was bigger in the central region of the resonator. This could explain
244 Sensor Applications

Figure 8.13 FBAR-based localized-mass detector responsivity [g/Hz] against normalized

mass-by-area (the reference mass is 3.6 × 10 g with a contact-area of 1.5 × 1.5 μm ). (© 2007
−12 2

Elsevier [36].)

the higher sensitivity and responsivity values for this region found in FBAR-based
localized-mass detectors. This is also coherent with experimental data obtained for
QCMs, in which localized silver spots are deposited on several positions along the
diameter of quartz resonators [38]. Similar experiments have been carried out with
crystal resonators immersed in liquid environments [39]. In these experiments, the
mass sensitivity and amplitude distribution curves were found to follow a Gaussian
function [40], the maximum value being obtained at the center of the resonator’s
electrode.
At resonance, vibration with a mode shaping induces a deformation in the verti-
cal geometry of the FBAR. Thus, an explanation for improved sensitivity of the cen-
tral region may be given in terms of different inertial fields generated on the surface
of the electrode [24]. Regarding this explanation, the central region exhibits higher
sensitivity, probably due to bigger acceleration forces—inertial field—in the center
of the FBAR.
Analytical and finite-element modeling of FBAR-based localized-mass detectors
have corroborated the location and mass size dependence on the sensor’s
responsivity, as previously referenced. The contour plots of Figure 8.14 show the
high-frequency mode shaping of an FBAR before and after localized-mass deposi-

Table 8.6 Dependence of the FBAR Responsivity

on the Axis of the Top Electrode on Which the
Localized Mass Is Deposited
Mass f Responsivity
Region (Estimated) (g) (MHz) (g/Hz)
−12
Center 3.6 × 10 1.90 1.90 × 10−19
Diagonal 3.6 × 10−12 1.13 3.19 × 10−19
−12
Lateral 3.6 × 10 0.98 3.65 × 10−19
Source: [36].
8.3 Mechanical Sensors 245

(b) (c)

(a)

(d) (e)

Figure 8.14 High-frequency mode shaping of an FBAR-based localized-mass sensor (FEM simula-
tions performed in ANSYS): (a) the FBAR with no mass; (b) mass deposited at the center of the
electrode; (c) the mass deposited 10 μm away from the center (along the y-axis); (d) mass located
at y = 20 μm; and (e) mass located at y = 30 μm. The resonance frequency shifts down with the
location of the deposited mass. (© 2009 IEEE [41].)

tion, obtained through FEM simulations in ANSYS. The FEM analysis reveals a
modification of the mode shape and the down shifting of the resonance frequency.
This is dependent on the position of the deposited mass (the center is the most sensi-
tive location exhibiting higher frequency shift).
FBAR-based localized-mass sensors are a competing technology with great
potential in highly sensitive biochemical applications. Evaluation of the active-sen-
sor area of localized-mass sensors is an interesting subject of study. Since
responsivity is location-dependent, it determines both the application and design of
the sensor. Hence, the ultimate localized-mass sensor’s sensitivity and mechanisms
are to be studied with regard to the FBAR’s size. Analytical and finite element mod-
els are suitable to accompany this topic.

8.3 Mechanical Sensors

In this section, we study the application of microresonators to mechanical sensors.

In mechanical sensing applications, a strain applied to the MEMS structure is the
most common mechanism for coupling the resonator to the measurand. In this way,
the resonator becomes a resonant strain gauge that couples to the measurand when
mounted in a suitable location on a sensing structure that deflects due to the appli-
cation of the measurand. Such a structure is specifically designed for the application
requirements and purposes.
246 Sensor Applications

The resonator output can be used to monitor the deflection of the sensing struc-
ture and thereby provide an indication of the magnitude of the measurand, which is
directly proportional to the resonance frequency of the sensor. When used as a reso-
nant strain gauge, the applied strain effectively increases the stiffness of the resona-
tor, thus increasing its resonance frequency. This is the common principle of
resonant force sensors, pressure transducers, and accelerometers [42].
The acoustic wave propagation through and reflection from a piezoelectric film
is another mechanism to sense the inertial force applied to the sensor. When the inci-
dent pressure arrives at the piezoelectric sensor, an acoustic wave propagates and
the amount of pressure can be quantified as a function of the propagated or
reflected acoustic wave. In the following sections, mechanical sensors of both the
strain gauge and acoustic wave kinds are discussed.

8.3.1 Pressure Sensors

Resonant pressure sensors are mature applications, which have been produced,
commercialized, and exploited by different companies since the 1980s. The main
sensing technology is a strain gauge integrated with a diaphragm, or it is the dia-
phragm by itself, which senses the resonance of the structure to determine the pres-
sure value as a function of frequency. Piezoelectric sensors can also be used as
resonant strain gauges to perform pressure measurement. SAW resonators have
been implemented to detect the pressure changes as the delay in the path of the
acoustic wave as well. MEMS industrial products and in-research pressure sensors
are challenging the sensitivity limits of bulk devices. They offer improved
performance in comparison to techniques.
Resonant pressure sensors have their origin at the early developments of Green-
wood [43]. The strain gauge technology developed by Greenwood was later
acquired and successfully commercialized by Druck Ltd. The sensor is a butter-
fly-shaped diaphragm attached to four pillars that deflects with pressure. Reso-
nance frequency shifting happens when the diaphragm deflects by the action of
pressure. Metal electrodes fabricated on the structure perform capacitive detection
of the strain after electrostatic actuation of the diaphragm. The boron-etch stop
technique developed by Greenwood, high vacuum, and hermetic packaging allow
the device to achieve high Q factors and sensitivities. The Druck’s Resonant Pres-
sure Transducer (RPT) series fabricated with this technology exhibits high stability
(<100 ppm) and resolution of 2–4 Hz/mbar in the 35–3,500-mbar range [44].
The Yokogawa’s silicon resonant pressure sensor is another commercial tech-
nology of high market impact. The resonant sensors are fabricated from single crys-
tal silicon using bulk micromachining techniques. Two H-shaped gauge resonators
operating at resonance are patterned on the sensor’s diaphragm. Resonance is elec-
tromagnetically elicited by placing the metallic gauges in an alternating high-fre-
quency magnetic field. As pressure is applied, the bridge gauges are simultaneously
stressed, one in compression and one in tension. The resulting change in resonance
frequency produces a differential frequency output proportional to the amount of
applied pressure. The thermal stability is higher than 10 ppm/ºC with in vacuum Q
factors above 50,000. The fabrication technology involves advanced
8.3 Mechanical Sensors 247

micromachining and packaging techniques, which are applied to Yokogawa’s

DPharp pressure transmitter and MT100/200 digital manometer [45].
The Quartzdyne’s resonant pressure sensor is a quartz resonator that changes
its frequency in response to pressure by the piezoelectric effect mechanism. The
structure is a thick-walled hollow cylinder with closed ends that encapsulates a
thickness-shear-mode disk resonator dividing the central portion of the hollow cyl-
inder, as depicted in Figure 8.15. In this device, the gauge and the diaphragm are the
same structure, namely the disk resonator. Fluid pressure on the exterior hydrostati-
cally compresses the quartz cylinder, producing internal compressive stress in the
resonator. The vibrating frequency of the sensor changes in response to this stress.
Because quartz is highly frequency stable, the sensor’s frequency output provides
high-resolution pressure measurement [46].
Recent laboratory efforts have explored the mechanisms of resonant pressure
sensing to discriminate the effects causing the resonance frequency variations. In
2003, Mertens et al. investigated the effect of pressure and temperature on the reso-
nance response of Si microcantilevers. In their studies, they identified the ranges and
mechanisms causing variations of the resonance frequency, and they demonstrated
the potential of cantilevers to distinguish between the effects of pressure, tempera-
ture, and gas properties. The resonance response as a function of pressure shows
that three different regions corresponding to molecular flow, transition, and viscous
regimes are involved. The deflection in the flow transition regime resulting from
thermal variation has a minimal effect on frequency. The frequency variation of the
cantilever is caused mainly by changes in the mean free path of gas molecules [47].
Azevedo et al. introduced a silicon carbide (SiC) MEMS resonant strain sensor
for harsh environment applications [48]. The sensor is built as a balanced-mass
double-ended tuning fork (BDETF) fabricated from 3C-SiC deposited on a Si sub-
strate. The fabricated device is shown in Figure 8.16. The resonance frequencies of
the BDETF are about 207 kHz and 210 kHz in air, with and without the on-chip
strain actuator, respectively. The resonators operate at atmospheric pressure and
from room temperature to above 300ºC, and they exhibit strain sensitivities of 66
Hz/με and strain resolution of 0.11 με in a bandwidth from 10 to 20 kHz with high
linearity [48].
The SAW pressure sensor of Talbi et al. combines the physical properties of
ZnO and quartz to produce high sensitivity and low temperature coefficient factor

Capsule

Disk resonator

Capsule

Figure 8.15 Quartzdyne’s resonant pressure sensor: a hollow cylinder with closed ends encapsu-
lates a thickness-shear-mode disk resonator (external pressure compresses the capsule and causes
resonance frequency shifting of the disk). (After: [46].)
248 Sensor Applications

Figure 8.16 SiC MEMS resonant strain sensor for harsh environment applications. (© 2007 IEEE
[48].)

(TCF). The quartz Y–X contributes to a nearly zero TCF value when it is combined
with a thin ZnO film of particular thickness. The ZnO also increases the electrome-
2
chanical coupling coefficient (k = 1%) to seven times higher than that of quartz
Y–X. The optimal structure of the 130-MHz SAW resonator yields a high electro-
mechanical coupling factor k2 of 1% and small temperature coefficient of frequency
TCF of −0.21 ppm/°C, and it achieves the pressure sensitivity to around 35
ppm/mbar [49].
The concept and operating principle of another wireless pressure microsensor
based on a SAW reflective delay line is depicted in the schematic of Figure 8.17. The
contact antenna coupled to the SAW sensor receives an RF pulse. Then, the
interdigitated transducer (IDT) transforms the received signal into an acoustic
wave. Reflectors are placed in the propagation path of the SAW, and the reflected
waves are reconverted into an electromagnetic wave by the IDT and transmitted to
the reader unit through antennas. Phase shifting evaluation of reflected impulses
determines the sensor signal. The 440-MHz SAW-based pressure sensor based on
this reflective delay line is made of 41º YX LiNbO3 with shorted circuit grating
reflectors and single-phase unidirectional transducers (SPUDT). The obtained pres-
sure sensitivity was 2.67º/kPa [50].

8.3.2 Accelerometers
Miniature accelerometers are a popular application of MEMS sensors. A resonant
accelerometer integrates a strain sensitive resonator, seismic (proof) mass, and
beam structures to support the proof mass and the resonator. The base technology is
8.3 Mechanical Sensors 249

(a)

(b)
Figure 8.17 SAW-based pressure sensor: (a) operating principle; and (b) phase response against
applied pressure. (© 2007 Elsevier [50].)

usually silicon, which is served to implement the mass and beams. The movable
proof mass aids in improving the acceleration sensitivity. Sensing resonators are
made of piezoelectric films or piezoresistive implantations. As the external force is
applied to the mass, it deflects in the acceleration axis and the added strain produces
frequency shifting of the resonator.
Ferrari et al. developed a resonant accelerometer manufactured in silicon bulk
micromachining with electrothermal excitation and piezoresistive detection. The
accelerometer’s structure is a seismic mass supported by two parallel flexure hinges
as a doubly sustained cantilever, with a resonating microbeam located between the
hinges, as shown in the schematic drawing of Figure 8.18(a). AC voltage applied to
the heaters induces a thermal excitation to the hinges, thus causing the structure to
resonate. The piezoresistors detect the resonance of the micro beam: acceleration
normal to the chip plane induces an axial stress in the microbeam and, in turn, a
proportional change in its resonance frequency. The accelerometer shown in Figure
8.18(b) resonates at a frequency around 70 kHz, and it has an acceleration sensitiv-
ity of 35 Hz/g over the range of 0 to 3 kHz [51]. Another piezoresistive accelerome-
ter with a similar performance was also demonstrated by Aikele et al. [52], thus
obtaining frequency-to-static-acceleration sensitivity of 70 Hz/g.
In piezoelectric-based resonant accelerometers, the acoustic layer of the resona-
tor is the sensitive element of the system. Current thin-film technologies employ
250 Sensor Applications

(a)

(b)
Figure 8.18 Piezoelectric accelerometer with electrothermal excitation and piezoresistive detec-
tion: (a) sensor’s concept; and (b) physical realization of the sensor chip with the accelerometer
(enlarged view). (© 2005 Elsevier [51].)

miniature FBARs fabricated on silicon accelerometers. The operating principle of

such a kind of accelerometer is depicted in Figure 8.19: acceleration applied to the
seismic mass adds stress to the device and causes strain. Since the FBAR and the Si
are mechanically coupled, the strain induced on the mass is transmitted to the beam
and FBAR. Thus, it produces charge displacement in the crystalline structure of the
piezoelectric layer, and it causes shifting of the FBAR’s resonance frequency. The
electrical charge displacement within the crystal and the resonance frequency of the
FBAR is proportional to the magnitude and orientation of the acceleration. This
principle can be extended to the case in which the FBAR couples directly with the
seismic mass. The fabrication technology of FBAR-based accelerometers requires
process compatibility between silicon and FBAR technologies. The FBAR piezoelec-
tric layer can be made of AlN, whose deposition and etching on a Si substrate is the
biggest challenge to FBAR and Si compatibility [53].
The SEM images of Figure 8.20 show front and back-side views of a dual-beam
accelerometer with embedded piezoelectric resonator and 500-μm-thick mass.
Detail of one sensing FBAR is observed in Figure 8.20(c). The thickness of the seis-
8.4 Atomic Force Detection 251

FBAR
Inertial force
Top electrode

AIN

Bottom electrode

Supporting beam
(Si)

Seismic mass Air gap

(Si)

Figure 8.19 Operating principle of a piezoelectric, FBAR-based accelerometer. (© 2008 Elsevier

[53].)

mic mass controls the accelerometer sensitivity; an example of an 80-μm thin mass
device is shown in Figure 8.20(d). The FBAR has lateral size of 20 μm × 100 μm, and
it is made of AlN and Pt electrodes, with thicknesses equal to 1,000 and 180 nm,
respectively. The supporting silicon beam has a thickness of about 80 μm. Accord-
ing to this configuration, the resonance frequency of the FBAR is 2.4 GHz.
Break-off tabs have been patterned on the Si to facilitate dicing. The device exhibits
a frequency-to-static-acceleration sensitivity of 200 kHz/g. Due to its high-fre-
quency operation mode, the embedded-FBAR accelerometer reaches a frequency
sensitivity that is more than 2,000 times higher than that of the low-frequency
devices previously discussed. However, the relative frequency shifting of the
low-frequency piezoresistive accelerometer is one order of magnitude better than
the 1 × 10−4 value of the embedded-FBAR accelerometer. Process variations can be
introduced to fabricate accelerometers with supporting beams made with the struc-
tural layers of FBARs in order to achieve improved sensitivity [54].

8.4 Atomic Force Detection

An atomic force microscope (AFM) can perform nanometer-scale and atomic-scale

force detection by implementing a MEMS cantilever as the sensing element. When
operated in tapping mode, the cantilever resonates with a frequency response
between units of hertz to few megahertz. The sharp tip located at the apex of the
cantilever scans over a surface of the sample and detects the tip-sample interactions,
while a feedback system controls the interaction force between the tip and the sam-
ple surface [55]. Using micromachining technologies, the dimensions of cantilevers
can be tuned to optimize the sensitivity, resonance frequencies, elastic constants,
and other properties of the cantilever [56]. Both the cantilever and the tip are usu-
ally made of polysilicon, although other materials, like quartz, have been imple-
mented as well [57]. Examples of AFM probes made of crystalline silicon are
observed in the SEM images of Figure 8.21. In AFM systems, the force detection is
performed by measuring the deflection of the cantilever when it is attracted by the
sample. This measurement is carried out by complex optical or interferometer tech-
252 Sensor Applications

(a)

(b)
Figure 8.20 Dual-beam FBAR-based accelerometer: (a) front side; (b) back side (thick-mass); (c)
detailed view of one of the embedded FBARs; and (d) back side (thin-mass).

niques, the setup comprising laser, optical detectors, signal generators, and lock-in
amplifiers, among others [58]. Taking advantage on the same operating principle of
FBAR-based accelerometers, a new generation of AFM-detection cantilevers can be
implemented with the FBAR technology (a detailed discussion on AFM cantilever
measurements is done in Section 5.4.1).
The concept of the FBAR-based force sensor for AFM applications is illustrated
in the schematic drawings of Figure 8.22(a, b), where two cases are differentiated.
In the configuration of Figure 8.22(a), the FBAR is embedded on the cantilever’s
structure, which can be made of silicon and has a tip close to its vertex. When the
probe and the sample approach each other, the cantilever deflects, thus bending the
FBAR and adding stress to the piezoelectric layer. The magnitude of deflection is
directly proportional to the probe-sample attraction forces and is detected as a
8.4 Atomic Force Detection 253

(c)

(d)
Figure 8.20 (continued)

change of the resonance frequency of the FBAR. The second concept is shown in
Figure 8.22(b), in which the cantilever and the tip are the same V-shaped structure.
This structure is fabricated according to the FBAR process (metal-piezoelec-
tric-metal stack configuration). As depicted in the figure, the V-FBAR cantilever
deflects when attracted by the sample, thus increasing its resonance frequency. The
fabrication technology, operation, and detection principles are the same for both
configurations. A combination of front- and back-side DRIE is performed to release
both the cantilever and the FBAR (which are the same structure in the second case).
Although there exist several AFM-probe fabrication technologies, the develop-
ment of FBAR-based force sensors is still in the early stages of analysis and design.
The pictures of Figure 8.23(a, b) show fabricated devices of the first kind with
embedded FBAR. The FBARs have a width of 40 μm, a length of 100 μm, and a
thickness of 1 μm (AlN). The second type of force sensor is observed in Figure
254 Sensor Applications

(a)

(b)
Figure 8.21 AFM probes made of crystalline silicon: (a) SEM micrographs of a typical AFM
probe-chip, with a shaft diameter ~3.8 μm, apex radius ~10 nm, and vertex angle ~10°. Cantilever
thickness is 1.5 μm, width is 25 μm, and length is 250 μm, 350 μm, and 50 μm (from left to
right). (b) Detailed view of one tip. (© 2007 Elsevier [56].)

8.23(c, d). The V-shaped tips are designed and fabricated according to the FBAR
process, and they have lengths between 5 and 50 μm, and widths of 5–20 μm. The
AlN layer and the metal electrodes are 1 μm thick and 180 nm thick, respectively.
The residual stress of the AlN layer presumably causes the bending observed on the
tips.
Both kinds of sensors could be implemented in future AFM applications if a
force tip is deposited on or grown near the FBAR [56]. In this application, the tip is
placed above the analyzed sample to perform force interaction, and the FBAR
detects the force intensity by reading out the resonance frequency’s variation.

8.5 Magnetic Sensors

Magnetic sensors of many kinds have been designed and assisted different applica-
tions throughout the past and the present century. The existing magnetic sensing
8.5 Magnetic Sensors 255

(a)

(b)

Figure 8.22 Concept of the FBAR-based AFM force detector: (a) first configuration: the FBAR is
embedded on the Si cantilever’s structure; and (b) second configuration: cantilever and tip are a
V-shaped FBAR. In both cases, the probe-sample attraction causes deflection and frequency shift-
ing of the FBAR.

technologies are capable of covering the wide range of magnetic field intensities
from over the Earth’s magnetic field down to the geomagnetic noise and below [59].
The earliest designs of magnetic sensors utilized simple magnetic attraction to fer-
rous objects to detect the magnitude or gradient of the magnetic field. In contrast,
MEMS magnetometers and gradiometers typically implement a hard magnet or
magnetic material integrated somehow to its micromachined structure. Whenever
the MEMS sensor is under the influence of the magnetic field, it happens to move
due to the ferromagnetic attraction arising between the magnetic material and the
source of the magnetic perturbation. The resulting motion is then measured through
electrical [60], optical [61], or magnetic [62] detection techniques, among others.
Current MEMS magnetometers have magnetic field sensitivities in the range
from 1 μT to 1 mT. They are also referred to as MEMS compasses because they
align in the direction of the magnetic field. Normally, MEMS compasses are
designed to operate without power consumption, although they can be connected to
an integrated read-out circuit. In the first case, optical detection is most commonly
employed to detect the movement of the MEMS. The most popular structures to
implement the compass are torsional beams with a magnetic plate [63–65] and can-
tilevers [66]. The sizes of such structures range from hundreds of micrometers to
some millimeters, and they have mainly been used as magnetic field detectors, cur-
rent meters, or optical scanners.
256 Sensor Applications

(a)

(b)
Figure 8.23 Force sensors for AFM applications: (a, b) V-shaped FBAR-made tips; and (c, d)
embedded FBARs on Si cantilevers.

Resonant magnetic field sensors of the MEMS compass and LC types are based
on magnetic resonant structures. The MEMS resonator is modified to incorporate a
soft magnetic material or to integrate a hard magnet. The resonance frequency of
the MEMS shifts due to the interaction of the external magnetic field and the mag-
netic component of the resonant system. The magnitude of the frequency shifting is
used to determine the amplitude or the direction of the external magnetic field.
According to this operation principle, Leichle et al. fabricated a CMOS-compatible
and low-temperature process sensor based on surface micromachining. The sensor,
depicted in Figure 8.24, has a resolution of 45° at 30 μT. The resonator’s low power
consumption—on the order of 20 nW—makes it useful as a magnetic compass [67].
Electromagnetic induction is a sensing mechanism that has also been imple-
mented by MEMS sensors to detect magnetic field variations. A micro-LC resonator
consisting of a solenoidal microinductor with a bundle of soft magnetic microwire
cores and a capacitor connected in parallel to the microinductor was reported by
8.5 Magnetic Sensors 257

(c)

(d)
Figure 8.23 (continued)

Kim et al. [68]. The LC resonator detects the external magnetic field by measuring
the inductance ratio as well as the magnetoimpedance ratio (MIR). The MEMS
solenoids are manufactured within lithography, electroplating, and molding (LIGA)
processes. The core magnetic material of the microinductor shown in Figure 8.25 is
a tiny glass-coated Co83.2B3.3Si5.9Mn7.6 microwire fabricated by a glass-coated
melt-spinning technique. The solenoidal microinductors fabricated by MEMS tech-
niques are 500–1,000 μm in length with 10–20 turns. The resonance frequency of
the microinductor and capacitor tank is near 105 MHz and, because the permeabil-
ity of the ultrasoft magnetic microwires changes rapidly as a function of external
magnetic field, the inductance ratio and the MIR vary very rapidly with the mag-
netic field as well [68].
The Lorentz force is another mechanism widely exploited for highly sensitive
magnetic field measurements. The resonant Lorentz-force device is typically a
mechanical resonator carrying an electrically conductive element designed for low
258 Sensor Applications

Figure 8.24 Photograph of a fabricated resonator incorporating a permanent magnet. (© 2004

Institute of Physics Publishing [67].)

Figure 8.25 Microinductors for LC-resonator-based magnetic field sensing, with and without
microwires as a core: the MEMS solenoids are made within a LIGA process, and the tiny
glass-coated Co83.2B3.3Si5.9Mn7.6 microwires are fabricated by a glass-coated melt-spinning technique.
(© 2006 American Institute of Physics [68].)

sensitivity to spurious vibration or magnetic perturbations of the environment.

Oscillation of the conductor in a magnetic field generates an electric potential that
drives a current around a closed resistive loop to damp the motion of the resonator.
Variations in the magnetic field, therefore, alter the resonator’s amplitude and so
alter the measured frequency of the overdriven resonator. According to this mecha-
nism, Greywall proposed a device that is capable of detecting perturbations in
Earth’s magnetic field smaller than one part per million (1 ppm) [69].
8.6 Summary 259

Herrera-May et al. reported another Lorentz-force-based resonant magnetic

field sensor at atmospheric pressure [70]. This sensor does not require vacuum
packaging to operate efficiently and presents a compact and simple geometry. The
MEMS structures of Figure 8.26 are built in a SOI substrate, and their geometry
3
consists of a seesaw plate (400 × 150 × 15 μm ), two torsional beams (60 × 40 × 15
3 3
μm ), four flexural beams (130 × 12 × 15 μm ), and a Wheatstone bridge with four
p-type piezoresistors. The resonant device operates at its first resonant frequency of
136.52 kHz with a Q factor over 800. A sinusoidal excitation current of 22.0 mA
with a frequency of 136.52 kHz and magnetic fields from 1G to 400G are consid-
ered. The sensor has the maximum magnetic sensitivity of 40.3 μV/G (magnetic
fields <70G), theoretical root-mean square (rms) noise voltage of 57.48 nV/Hz1/2,
theoretical resolution of 1.43 mG /Hz1/2, and power consumption less than 10.0
mW [70].
Resonant magnetic microsensors have been demonstrated at the micrometer
scale, although nanometer-sized NEMS geometries are being developed to detect
magnetic field perturbations, like the nanocantilever implementation of Vasquez et
al. [71]. Even if these NEMS are nonresonant devices, they open new paths in mag-
netic sensor miniaturization.

8.6 Summary

In this chapter, we have studied several resonant sensor applications. First, we

reviewed MEMS, NEMS, QCM, SAW, and FBAR mass detectors. The mass load-
ing effect and the sensitivity analysis of each technology were discussed. In these

Figure 8.26 Photomicrograph of the magnetic field microsensor. (© 2009 Institute of Physics
Publishing [70].)
260 Sensor Applications

analyses, we differentiated between distributed and localized mass sensing. Nowa-

days, NEMS resonators achieve the highest mass resolution and sensitivity between
the current-art technologies, the latest sensors being able to detect masses as small
as atoms. Nevertheless, the wide family of acoustic resonators is a more mature
technology and is currently being implemented in both commercial and laboratory
products. Applications of mass sensors are innumerable. Miniaturized mass sensors
have enabled the high-sensitivity, high-resolution detection of chemical particles,
biological particles like spores or viruses, DNA and proteins, gases, liquids, and a
long and growing list of convergent applications.
In the Section 8.2 we studied mechanical sensors. First, micromachined acceler-
ometers based on piezoresistive and piezoelectric actuations were introduced. The
discussion followed with the concept and fabrication technology of FBAR-based
high-frequency accelerometers. The benefits of the high-frequency detection of
FBAR accelerometers were compared to other low-frequency resonant devices. Sec-
tion 8.3 described another kind of force detector: AFM. The concept, operation
principles, and available technologies of MEMS cantilevers for AFM applications
demonstrated that a new generation of low-cost, low-size atomic-scale scanning
probes is now possible. Self-detection is one of the main features of this new genera-
tion of AFM probes, thus simplifying the detected-signal measurement setup. Even-
tually, this feature will open the path for the future implementation of portable
AFM systems.
Studied in Section 8.4, MEMS resonators also demonstrated their versatility as
resonant magnetic sensors. Most of them are based on micro cantilever or torsional
beam geometries and have made evident the advantages of miniaturization in mag-
netometry. The latest developments of NEMS resonator magnetometers promise
future resonant detection of magnetic field perturbations.
The examined applications are a fruitful and stimulating research field.
Although commercial products exist, most of them still need to solve a lot of chal-
lenges and design issues to enable the applications to pass from concept to industrial
reality. A lot of modeling, testing, and optimization work is required to consolidate
the sensor applications. Also, and related to previous chapters of this book, integra-
tion of these applications with CMOS technologies is to be considered. As has been
fully proven, CMOS integration improves the impact and versatility of the sensors,
boosting them to full systems-on-chip implementations.

References

[1] Greenwood, J. C., “Etched Silicon Vibrating Sensor,” J. Phy. E. Sci. Instrum., Vol. 17,
1984, pp. 650–652.
[2] IEEE Std. 1139, “IEEE Standard Definitions of Physical Quantities for Fundamental Fre-
quency and Time Metrology—Random Instabilities,” Institute of Electrical and Electronics
Engineers, New York, 1999.
[3] Rodriguez-Pardo, L., et al., “Resolution in Quartz Crystal Oscillator Circuits for High Sen-
sitivity Microbalance Sensors in Damping Media,” Sens. Actuators B: Chem., Vol. 103,
2004, pp. 318–324.
[4] Beeley, J. M., et al., “All-Digital Interface ASIC for a QCM-Based Electronic Nose,” Sensor
Actuators B: Chem., Vol. 103, 2004, pp. 31–36.
8.6 Summary 261

[5] Zeng, K., and C. A. Grimes, “Threshold-Crossing Counting Technique for Damping Factor
Determination of Resonator Sensors,” Rev. Sci. Instrum., Vol. 75, 2004, pp. 5257–5261.
[6] Gribaldo, S., et al., “Experimental Study of Phase Noise in FBAR Resonators,” IEEE
Trans. on Ultrasonics, Ferroelectrics, Freq. Control, Vol. 53, 2006, pp. 1982–1986.
[7] Phadke, M. S., Quality Engineering Using Robust Design, Englewood Cliffs, NJ:
Prentice-Hall, 1989.
[8] Hwang, I.-H., and J.-H. Lee, “Self-Actuating Biosensor Using a Piezoelectric Cantilever and
its Optimization,” J. Physics, Vol. 34, 2006, pp. 362–367.
[9] Narducci, M., et al., “Sensitivity Improvement of a Microcantilever Based Mass Sensor,”
Microelectronic Eng., Vol. 86, 2009, pp. 1187–1189.
[10] Davila, A. P., et al., “Microresonator Mass Sensors for Detection of Bacillus Anthracis
Sterne Spores in Air and Water,” Biosensors and Bioelectr., Vol. 22, 2007, pp. 3028–3035.
[11] Ismail, A. K., et al., “The Principle of a MEMS Circular Diaphragm Mass Sensor,” J.
Micromech. Microeng., Vol. 16, 2006, pp. 1487–1493.
[12] Kehrbusch, J., et al., “High Frequency Columnar Silicon Microresonators for Mass Detec-
tion,” Appl. Phys. Lett., Vol. 93, 2008, 023102.
[13] Villarroya, M., et al., “System on Chip Mass Sensor Based on Polysilicon Cantilevers
Arrays for Multiple Detection,” Sens. Actuators A: Phys., Vol. 132, 2006, pp. 154–164.
[14] Ekinci, K. L., Y. T. Yang, and M. L. Roukes, “Ultimate Limits to Inertial Mass Sensing
Based upon Nanoelectromechanical Systems,” J. Appl. Phys., Vol. 95, 2004,
pp. 2682–2689.
[15] Ilic, B., Y. Yang, and H. G. Craighead, “Virus Detection Using Nanoelectromechanical
Devices,” Appl. Phys. Lett., Vol. 85, 2004, pp. 2604–2606.
[16] Li, M., H. X. Tang, and M. L. Roukes, “Ultra-Sensitive NEMS-Based Cantilevers for Sens-
ing, Scanned Probe and Very High-Frequency Applications,” Nature Nanotechnology,
Vol. 114, 2007, pp. 114–120.
[17] Jensen, K., K. Kim, and A. Zettl, “An Atomic-Resolution Nanomechanical Mass Sensor,”
Nature Nanotechnology, Vol. 3, 2008, pp. 533–537.
[18] Lassagne, B., et al., “Ultrasensitive Mass Sensing with a Nanotube Electromechanical Reso-
nator,” Nano Lett., Vol. 8, 2008, pp. 3735–3738.
[19] Aubin, K. L., et al., “Microfluidic Encapsulated Nanoelectromechanical Resonators,” J.
Vac. Sci. Technol. B, Vol. 25, 2007, pp. 1171–1174.
[20] Sauerbrey, G., “Verwendung von Schwingquarzen zur Wägung dünner Schichten und zur
Mikrowägung,” Z. Phys., Vol. 155, 1959, pp. 206–222.
[21] Martin, F., et al., “Pulse Mode Shear Horizontal-Surface Acoustic Wave (SH-SAW) System
for Liquid Based Sensing Applications,” Biosensors and Bioelectr., Vol. 19, 2004,
pp. 627–632.
[22] Gulyaev, Y. V., “Review of Shear Surface Acoustic Waves in Solids,” IEEE Trans. on
Ultrason. Ferroelectr. Freq. Control, Vol. 45, 1998, pp. 935–938.
[23] Gizeli, E., et al., “A Love Plate Biosensor Utilising a Polymer Layer,” Sens. Actuators B:
Chem., Vol. 6, 1992, pp. 131–137.
[24] Mecea, V. M., “Is Quartz Crystal Microbalance Really a Mass Sensor?” Sens. Actuators A:
Phys., Vol. 128, 2006, pp. 270–277.
[25] Granstaff, V. E., and S. J. Martin, “Characterization of a Thickness-Shear Mode Quartz
Resonator with Multiple Nonpiezoelectric Layers,” J. Appl. Phys., Vol. 75, 1994,
pp. 1319–1329.
[26] Martin, S., V. Granstaff, and G. Frye, “Characterization of a Quartz Crystal Microbalance
with Simultaneous Mass and Liquid Loading,” Anal Chem., Vol. 63, 1991, pp. 2272–2281.
[27] Domack, A., et al., “Swelling of a Polymer Brush Probed with a Quartz Crystal Resonator,”
Phys. Rev. E, Vol. 56, 1997, pp. 680–689.
262 Sensor Applications

[28] Voinova, M. V., et al., “Viscoelastic Acoustic Response of Layered Polymer Films at
Fluid-Solid Interfaces: Continuum Mechanics Approach,” Physica Scripta, Vol. 59, 1999,
pp. 391–396.
[29] Karamollaoglu, I., H. A. Oktem, and M. Mutlu, “QCM-Based DNA Biosensor for Detec-
tion of Genetically Modified Organisms (GMOs),” Biochemical Engineering J., Vol. 44,
2009, pp. 142–150.
[30] Campanella, H., et al., “Localized and Distributed Mass Detectors with High Sensitivity
Based on Thin-Film Bulk Acoustic Resonators,” Appl. Phys. Lett., Vol. 89, 2006, 033507.
[31] Benetti, M., et al., “Microbalance Chemical Sensor Based on Thin-Film Bulk Acoustic
Wave Resonators,” Appl. Phys. Lett., Vol. 87, 2005, 173504.
[32] Benetti, M., et al., “Thin Film Bulk Acoustic Wave Resonator (TFBAR) Gas Sensor,” Proc.
Intl. IEEE Ultrason. Symp. 2004, Montreal, Quebec, August 24–27, 2004, pp. 1581–1584.
[33] Zhang, H., and E. S. Kim, “Micromachined Acoustic Resonant Mass Sensor,” IEEE J.
Microelectromech. Syst., Vol. 14, 2005, pp. 699–706.
[34] Weber, J., et al., “Shear Mode FBARs as Highly Sensitive Liquid Biosensors,” Sens. Actua-
tors A: Phys., Vol. 128, 2006, pp. 84–88.
[35] Ekinci, K. L., Y. T. Yang, and M. L. Roukes, “Ultimate Limits to Inertial Mass Sensing
Based upon Nanoelectromechanical Systems,” J. Appl. Phys., Vol. 95, 2004,
pp. 2682–2689.
[36] Campanella, H., et al., “Localized-Mass Detection Based on Thin-Film Bulk Acoustic
Wave Resonators (FBAR): Area and Mass Location Aspects,” Sens. Actuators A: Phys.,
Vol. 142, 2008, pp. 322–328.
[37] San Paulo, A., X. Liu, and J. Bokor, “Scanning Acoustic Force Microscopy Characteriza-
tion of Thermal Expansion Effects on the Electromechanical Properties of Film Bulk Acous-
tic Resonators,” Appl. Phys. Lett., Vol. 86, 2005, 084102.
[38] Mecea, V. M., “Loaded Vibrating Quartz Sensors,” Sens. Actuators A: Phys., Vol. 40,
1994, pp. 1–27.
[39] Kanazawa, K. K., and J. G. Gordom, “Frequency of a Quartz Microbalance in Contact
with Liquid,” Anal. Chem., Vol. 57, 1985, pp. 1770–1771.
[40] Martin, B. A., and H. E. Hager, “Velocity Profile on Quartz Crystals Oscillating in Liq-
uids,” J. Appl. Phys., Vol. 65, 1989, pp. 2630–2635.
[41] Campanella, H., et al., “Analytical and Finite-Element Modeling of a Localized-Mass Sen-
sor Based on Thin-Film Bulk Acoustic Resonators (FBAR),” IEEE Sensors J., Vol. 9, 2009,
pp. 892–901.
[42] Beeby, S., et al., MEMS Mechanical Sensors, Norwood, MA: Artech House, 2004.
[43] Greenwood, J. C., “Etched Silicon Vibrating Sensor,” J. Phy. E. Sci. Instrum., Vol. 17,
1984, pp. 650–652.
[44] Druck Ltd., Resonant Pressure Transducer (RPT) Series, http://www.druck.com.
[45] Differential Pressure High Accuracy Pressure (DPharp), Yokogawa Electric Corporation,
http://www.yokogawa.com.
[46] Quartzdyne, http://www.quartzdyne.com.
[47] Mertens, J., et al., “Effects of Temperature and Pressure on Microcantilever Resonance
Response,” Ultramicroscopy, Vol. 97, 2003, pp. 119–126.
[48] Azevedo, R. G., et al., “A SiC MEMS Resonant Strain Sensor for Harsh Environment
Applications,” IEEE Sensors J., Vol. 7, 2007, pp. 568–576.
[49] Talbi, A., et al., “ZnO/Quartz Structure Potentiality for Surface Acoustic Wave Pressure
Sensor,” Sens. Actuators A: Phys., Vol. 128, 2006, pp. 78–83.
[50] Wang, W., et al., “Optimal Design on SAW Sensor for Wireless Pressure Measurement
Based on Reflective Delay Line,” Sens. Actuators A: Phys., Vol. 139, 2007, pp. 2–6.
[51] Ferrari, V., et al., “Silicon Resonant Accelerometer with Electronic Compensation of
Input-Output Cross-Talk,” Sens. Actuators A: Phys., Vols. 123–124, 2005, pp. 258–266.
8.6 Summary 263

[52] Aikele, M., et al., “Resonant Accelerometer with Selftest,” Sens. Actuators A: Phys.,
Vol. 92, 2001, pp. 161–167.
[53] Campanella, H., et al., “High-Frequency Sensor Technologies for Inertial Force Detection
and AFM Applications Based on Thin-Film Bulk Acoustic Wave Resonators (FBAR),”
Microelectronic Eng., Vol. 86, 2009, pp. 1254–1257.
[54] Campanella, H., et al., “Accelerometer Based on Thin-Film Bulk Acoustic Wave Resona-
tors,” Proc. IEEE Intl. Ultrason. Symp. 2007, New York, October 28–31, 2007,
pp. 1148–1151.
[55] Binnig, G., C. F. Quate, and C. Gerber, “Atomic Force Microscope,” Phys. Rev. Lett.,
Vol. 56, No. 9, 1986, pp. 930–933.
[56] Villanueva, G., et al., “DRIE Based Novel Technique for AFM Probes Fabrication,” Micro-
electronic Eng., Vol. 84, 2007, pp. 1132–1135.
[57] Hida, H., et al., “Fabrication and Characterization of AFM Probe with Crystal-Quartz
Tuning Fork Structure,” Proc. IEEE Intl. Symp. Micro-NanoMechatronics and Human
Science 2005, Nagoya, Japan, November 7–9, 2005, pp. 97–101.
[58] San Paulo, A., X. Liu, and J. Bokor, “Atomic Force Microscopy Characterization of Elec-
tromechanical Properties of RF Acoustic Bulk Wave Resonators,” Proc. IEEE Intl. Conf.
MEMS 2004, Maastrich, the Netherlands, January 25–29, 2004, pp. 169–172.
[59] Lenz, J., and S. Edelstein, “Magnetic Sensors and Their Applications,” IEEE Sensors J.,
Vol. 6, 2006, pp. 631–649.
[60] Ciudad, D., et al., “Modeling and Fabrication of a MEMS Magnetostatic Magnetic Sen-
sor,” Sens. Actuators A: Phys., Vol. 115, 2004, pp. 408–416.
[61] Keplinger, F., et al., “Lorentz Force Based Magnetic Field Sensor with Optical Read Out,”
Sens. Actuators A: Phys., Vol. 110, 2004, pp. 112–118.
[62] Ripka, P., “Advances in Fluxgate Sensors,” Sens. Actuators A: Phys., Vol. 106, 2003,
pp. 8–14.
[63] Yang, H. H., et al., “Ferromagnetic Micromechanical Magnetometer,” Sens. Actuators A:
Phys., Vols. 97–98, 2002, pp. 88–97.
[64] Vasquez, D. J., and J. W. Judy, “Optically Interrogated Zero-Power MEMS Magnetome-
ter,” IEEE/ASME J. Microelectromech. Syst., Vol. 16, 2007, pp. 336–343.
[65] Cho, H. J., and Ch. H. Ahn, “Magnetically Driven Bi-Directional Optical Microscanner,”
J. Micromech. Microeng., Vol. 13, 2003, pp. 383–389.
[66] Goedeke, S. M., S. W. Allison, and P. G. Datskos, “Non-Contact Current Measurement
with Cobalt-Coated Microcantilevers,” Sens. Actuators A: Phys., Vol. 112, 2004,
pp. 32–35.
[67] Leichle, T. C., et al., “A Low-Power Resonant Micromachined Compass,” J. Micromech.
Microeng., Vol. 14, 2004, pp. 462–470.
[68] Kim, Y.-S., et al., “A Class of Micromachined Magnetic Resonator for High-Frequency
Magnetic Sensor Applications,” J. Appl. Phys., Vol. 99, 2006, 08B309.
[69] Greywall, D. S., “Sensitive Magnetometer Incorporating a High-Q Nonlinear Mechanical
Resonator,” Measurement Science and Technology, Vol. 16, 2005, pp. 2473–2482.
[70] Herrera-May, A. L., et al., “A Resonant Magnetic Field Microsensor with High Quality
Factor at Atmospheric Pressure,” J. Micromech. Microeng., Vol. 19, 2009, 015016.
[71] Vasquez, D. J., and J. W. Judy, “Flexure-Based Nanomagnetic Actuators and Their Ulti-
mate Scaling Limits,” Proc. IEEE Intl. Conf. MEMS 2008, Tucson, AZ, January 13–17,
2008, pp. 737–741.
 CHAPTER 9

Radio Frequency Applications

Radio frequency (RF) circuits are responsible for conditioning the signals transmit-
ted and received by antennas of a wireless telecommunication system. Such signals
carry the huge amount of information of modern telecommunication systems, like
public and private mobile Internet applications. Thus, RF signals need to be accom-
modated into wireless channels to meet the power, bandwidth, and carrying-fre-
quency requirements of the telecommunication standard implemented for the
transmission and reception of the information. At the transmitter side, RF systems
take digital baseband signals and convert them into analog counterparts able to
access the channel and to cross the distances specified by the application. At the
receiver, on the other hand, the incoming signal from the antenna has to be selected,
amplified, and reconverted into the digital domain to recover the transported
information.
RF systems are, in essence, analog circuits—with some exceptions—and their
miniaturization and integration with their digital counterparts have been the object
of intensive research work. Since the cost and failure rates of RF analog devices are
notoriously higher than those of digital components, diminishing of the size and
power consumption of wireless devices drives the development of new RF compo-
nents. At a certain point of history, the miniaturization road path of RF components
changed its technological paradigm and took advantage of the benefits of IC and
MEMS processes, giving rise to the RF MEMS concept. Thus, RF MEMS are the
application of MEMS technology in the fields of RF and microwave circuits.
Inductors, varactors, transmission lines, switches, and resonators are some of the
RF MEMS components fabricated with the methodologies and micromachining
techniques used in current MEMS processes. Based on these components, RF cir-
cuits and systems like oscillators, mixers, filters, and amplifiers have dramatically
reduced their size and improved their performance. Nowadays, most of the minia-
ture RF MEMS applications have matured from the early R&D stages to reach the
status of industrial and commercial products. Diverse companies are selling mil-
lions of units of RF MEMS components, which represent a significant share of the
global MEMS market.
Resonator-based RF MEMS are an important group of the RF MEMS kind, and
they are the study object of this chapter. RF MEMS resonators are useful to imple-
ment both passive- and active-circuit components and are, among the RF MEMS
components, those achieving the greatest market impact and technological matu-
rity. We have selected some of the most remarkable commercial products and labo-
ratory demonstrators to explain the competitive advantages of resonant RF MEMS
against other RF technologies. First, we contextualize the field of RF systems and
RF MEMS, and their impact on modern telecommunication systems. Then, we

265
266 Radio Frequency Applications

begin the study by describing the operation principles, functionality, specifications,

and implementations of passive-circuit applications like filters, duplexers,
triplexers, and multiplexers. These applications are implemented with either acous-
tic or electromechanical resonators. Next, focus is laid on active-circuit applica-
tions, as we study RF MEMS-resonator-based oscillators, low-noise amplifiers, and
integrated RF front ends. The chapter concludes with a discussion on the latest
state-of-the-art developments of resonant RF MEMS applications.

9.1 Introduction

The communication and information technologies revolution has carried an

unprecedented development of services and appliances, which need a physical sup-
port to be efficiently offered. The explosion of both public and private wireless
mobile Internet systems and the aggressive competition of service providers have
stimulated the market to offer innovative end-user terminals. From the past ten
years to now, several technological paradigms have been broken, passing from the
mobile phone to the convergent, ubiquitous personal device.
Nowadays, functionality and aesthetics are combined on the same apparatus
without sacrifice of portability. The requirements of modern third generation (3G)
and beyond-3G (B3G) mobile systems involve a complex set of specifications.
Among them, we find internetworking capability, multiple-band access, low power
consumption, multimedia interfaces and services, and small size. In addition, per-
sonalized interfacing, tactile screens, navigation, positioning, and personal net-
working capabilities have been incorporated into the latest mobile systems.
Provided with a variety of MEMS sensors, like silicon accelerometers and pressure
and proximity transducers, the Apple’s iPhone [1] combines a visual and
energy-efficient interface with a powerful telecommunication system able to net-
work to UMTS [2], Wi-Fi [3], HSDPA [4], GSM [5], EDGE [6], Bluetooth [7], and
GPS [8]. Other manufacturers like Nokia [9], Samsung [10], and LG [11] offer com-
peting devices as well. All are positioning as the paradigm of the convergent device.
3G specifications impose stringent requirements to the RF section of the tele-
communication system of the mobile device, such as low power consumption, low
noise amplification, and filtering low insertion losses and high selectivity. In back-
ground, the promise of the single-chip RF front end also remains a milestone of the
low-cost, low-power device. These factors have motivated the development of com-
pact technologies to replace the traditionally bulky analog components of the RF
section. The role of RF MEMS in the miniaturization of key components, like
switches, duplexers, and band-selection filters, is crucial for this purpose. RF
MEMS are fabricated with the methodologies and micromachining techniques used
in current MEMS processes. The recently demonstrated integration of RF MEMS
with CMOS processes is boosting the size reduction of the RF sections and
front ends. The now-classic diagram of Figure 9.1 illustrates the impact of RF
MEMS in the miniaturization of modern transceivers [12]. As observed in the fig-
ure, a considerable number of RF components (the shadowed ones) are potentially
replaceable by RF MEMS. Actually, commercial products have already been placed
on most of the shadowed blocks of the diagram, instead of components made with
 9.1 Introduction

Image reject IF filter

filter (ceramic) (SAW) μ mech.
resonator Receiver
Antenna 1 Antenna 2
TFR or μ mech./μ machin.
resonator IF mixer
ADC I
μ mech. IF LNA
switch RF LNA Mixer 90°
Antenna T/R
switch VCO AGC
switch ADC Q
Xstal Channel
IF PLL
tank select PLL
μ mech.
resonator
Bandpass
DAC I
filter
(ceramic) μ mech.
switch 90°
TFR
or Power amplifier DAC Q
μ mech./μ machin. Off-chip Modulator
switch passive
elements VCO On-chip
Transmitter inductor
Transmit PLL +
μ mech. Xstal μ mech.
On-chip + μ mech. or
turnable tank turnable
inductor μ mech.
resonator
capacitor resonator capacitor

Figure 9.1 Block diagram of modern telecommunication transceiver with RF MEMS replacing traditional-technology components (highlighted). (© 2000 IEEE [12].)
267
268 Radio Frequency Applications

traditional technologies like ceramics or bulky crystals, as we will review in the next
sections.
RF MEMS offer many advantages against traditional technologies: low weight,
low insertion losses, high off-state isolation, high precision, low power consump-
tion, and high reliability, among others. Due to these characteristics, the multi-
ple-band receiver integrating in-parallel filters and switches or multiplexers for
channel selection has also been envisioned. To date, duplexers, triplexers, and
quintplexers integrating band-selection filters on a single chip are available in the
market, as detailed in Section 9.2.2. The schema of Figure 9.2 depicts the concept of
the in-parallel multiple-band receiver. In this example, modern wireless standards
like UMTS or Wi-Fi are selected for baseband processing by RF MEMS switches.
The RF MEMS antenna might also be reconfigured to adapt its bandwidth to the
different frequency bands of GSM-800/900 and UMTS-2000/3000.
Not all is rose-colored concerning RF MEMS. Due to its moving-structure
nature, these devices often need hermetic packaging, and many of them are far from
being robust in hazardous environments (e.g., vibrations, temperature). Besides,
some of them need very high polarization/actuation voltages not compatible with
standard CMOS processes, as it is the case of some resonators. Large time reliability
is still to be demonstrated in switches, for example. Mass production also requires
resolving process compatibility and test issues. Nevertheless, the RF MEMS market
offers a variety of products like BAW filters, switches, inductors, and resonators.
The size of the market is still limited, in comparison with other MEMS applications
like inkjet heads or pressure and inertial sensors. However, RF MEMS components
are expected to be a major breakthrough, thus satisfying the performance, miniatur-
ization, and reconfiguration requirements of telecommunication systems [13]. The
diagram of Figure 9.3 shows the RF MEMS commercialization status, discrimi-
nated by the different products available to date. We see how BAW filters and
duplexers for microwave bands (1–10 GHz) have the largest industrial history with

Figure 9.2 Vision of the in-parallel, multiple-band receiver integrating reconfigurable RF MEMS
antenna, switches, and band-selection filters tuned at the central frequencies of current 3G and
wireless Internet standards.
9.2 Passive-Circuit Applications 269

Figure 9.3 RF MEMS commercial products road path. (© 2005 WTC [13].)

a significant number of players competing for the market. Among them, Avago
Technologies (formerly Agilent Technologies) [14] has the biggest share of the BAW
market, followed by Epcos [15], TriQuint [16], and Skyworks [17]. RF oscillators
and clocks based on micromechanical resonators (VHF and UHF bands) fabricated
by Discera (mid-2005) [18] and Si-Time (2007) [19] are also available, and Philips
and Baolab are still in the product development stage of MEMS switches [20, 21]. In
this chapter, we will focus our attention on some of these commercial BAW filters,
duplexers, resonant switches, and oscillators. Additionally, we make room for
components and subsystems still in the R&D stages.

9.2 Passive-Circuit Applications

A passive circuit application is an all-passive component circuit without signal

amplification functions. Typical examples are passive filters composed by more
than one resonator. The components are integrated at the board level with RF inte-
grated circuits (RFIC) to offer a specific RF function, mainly band selection and
channel selection filtering. Filter-based multiplexing incorporates broadband multi-
ple filtering in a single chip. In this section, we study some of the most popular reso-
nator-based filters, multiplexers, and the resonant switch.

9.2.1 SAW, BAW, and FBAR-Based Band-Selection Filters

A band-selection filter rejects the signals of adjacent channels in communication
systems. In typical wireless mobile systems, for example, the transmitted and
received signals are channelized using different frequency bands. In current trans-
ceiver architectures, the transmit filter fits between the driver and power amplifiers.
270 Radio Frequency Applications

When placed at the transmission path, the filter prevents the transmitted signal from
feedback into the receiver path, thus increasing the sensitivity. With the proper
design, the filter can also reject the out-of-band signals at the receiver path, includ-
ing images of the RF incoming signal.
The history of RF filters began with ceramic filters, which were developed using
technology similar to that of quartz crystal and electromechanical resonator filters.
Based on the type of piezoelectric materials, four stages of historical development of
ceramic filters may be identified. The first material was single-crystal quartz, the
second was single-crystal Rochelle salt, the third was barium titanate ceramics, and
the fourth was lead-zirconate-titanate (PZT) ceramics [22]. The electronic technol-
ogy evolution imposed the need for filter miniaturization. Then, new fabrication
processes and materials were developed and thin-film aluminum nitride (AlN) and
zinc oxide (ZnO) became the standard for the new kind of SAW and BAW minia-
ture resonators. Classical applications of crystal and ceramic filters provide selectiv-
ity in communications receivers at frequencies of 9 MHz or 10.7 MHz, or at higher
frequencies as roll-off filters in receivers using up-conversion. Even a lower fre-
quency (455 kHz) is used as the second intermediate frequency filters in some com-
munication receivers. Ceramic filters at 455 kHz can achieve similar bandwidths to
crystal filters at 10.7 MHz. However, the evolution of communication systems led
to new services requiring more bandwidth and higher frequency bands, where the
selectivity, insertion losses, and overall performance of traditional filter technolo-
gies were not enough to meet the requirements of new systems. By up-shifting the
frequency to the VHF and UHF passbands, SAW filters opened the way to gigahertz
band BAW filters, including both SMR- and FBAR-based filters.
BAW-based filters offer unprecedented performance, exhibiting lower insertion
losses, higher out-of-band rejection, and a more reduced size than those made with
previous technologies. Two topologies are mostly implemented in RF filter applica-
tions—namely, the stacked crystal filter (SCF) and the ladder filter. The SCF filter
stacks two or more BAW resonators with the appropriate thicknesses to build the
filtering shape, as illustrated in the schematic view of Figure 9.4 [23].
On the other hand, the ladder filter topology is comprised of an Nth-order inter-
connection of series and shunt FBARs arranged in a planar way. The schematic and

Electrode

Piezoelectric

Piezoelectric Ground plane

Electrode
m1
m2
m1
m2

Figure 9.4 SCF topology (here the filter is mounted on a Bragg’s reflector).
9.2 Passive-Circuit Applications 271

equivalent-circuit models of a filter with N 3 are depicted in Figure 9.5(a, b),

where N is the number of series-to-shunt cells, counting from the input to the output
port of the filter [24]. As suggested by the process representation of Figure 9.5(c),
series resonators are designed to be thinner than shunt FBARs. Also, shunt resona-
tors have a bigger electrode area in order to optimize the in-band and out-of-band
performances of the filter [25]. The pictures of Figure 9.5(d) show the physical lay-
out of filters fabricated within the IMB-CNM’s CNM25 technology [26].
In 2002 Avago Technologies (Agilent Technologies at that time) released the
first FBAR-based filters of the market. This launching represented a breakthrough
in miniaturization and performance and came in an opportune market moment to
challenge SAW filters. Very soon, FBAR-based filters became a commercial success,
replacing previous-technology filters in the at-the-time newest US PCS and UMTS
mobile handsets. Figure 9.6(a–d) shows the RF performance of the Avago’s
ACPF-7003 1.6 × 2.0-mm high reception and image rejection transmission (TX) fil-
ter for the US PCS band. With a passband of 1,850–1,910 MHz and typical inser-
tion losses of 2.5 dB, the filter features 35-dB minimum attenuation in the
1,930–1,990-MHz band and 25-dB image rejection at 1,830 MHz. Its main appli-
cation is in US PCS band handsets and wireless data terminals [27]. A similar high
rejection TX filter for US PCS band is presented in [28].
TriQuint Semiconductor also has a relevant market presence with their BAW
filter products. TriQuint’s filters implement solidly mounted resonators (SMRs)
with Bragg mirrors to improve the quality factor of the resonators, instead of the
micromachined air cavity of FBARs. Two examples of the TriQuint’s BAW-device
catalog are the 880404 GPS (1,575 MHz) and the 880369 Wi-Fi (802.11a, 5,775
MHz) filters, available in their 3.71 × 2.51 × 0.84 mm and 3.26 × 1.60 × 0.84 mm
packages, respectively [29, 30]. Figure 9.7(a) shows the passband response of the
GPS filter, with a central frequency of 1,575 MHz, typical insertion losses of 1.5 dB,
and 3-dB bandwidth of 35 MHz. The Wi-Fi filter, at its time, performed insertion
losses of 3.0 dB, and 3-dB bandwidth of 100 MHz at the central frequency of 5,775
MHz, as shown in the S21 plot of Figure 9.7(b).
SAW filters have also been an important technology in the RF market, because
of their good performance and low manufacturing difficulty. Companies like Epcos
and Epson Toyocom [31] are world leaders in producing components made with
this technology. For example, the Epcos B7843 SAW filters for PCS applications
exhibit maximum insertion losses of 3 dB and out-of-band attenuation bigger than
40 dB at the central PCS frequency of 1,850 MHz [32]. More recently, however,
SAW filters are gradually being replaced by the newest and technology-compatible
BAW filters. Furthermore, some products are been hybridized, thus integrating both
SAW and BAW devices in the same chip. Epcos is also commercializing BAW-based
filters for duplexer applications, which we discuss in the next section.

9.2.2 Duplexers, Triplexers, and More

A duplexer is a three-port component performing filtering of RF signals at the point
closest to the antenna in the RF chain of a wireless handset or data terminal. The
duplexer allows simultaneous transmission and reception of RF signals through a
single antenna. Thus, it reduces the complexity, size, and cost of the RF system,
272 Radio Frequency Applications

(a)

(b)

Electrode
Electrode

Piezoelectric Piezoelectric
Electrode Electrode

Air gap Air gap

Substrate (Si)

(c)

(d)

Figure 9.5 Ladder filter topology: (a) schematic view; (b) equivalent-circuit model; (c) process
description; and (d) filter implementations within the CNM25 technology.
9.2 Passive-Circuit Applications 273

(a) (b)

(c) (d)
Figure 9.6 RF performance of Avago’s ACPF-7003 high-reception and image-rejection transmission filter
for the US PCS band: (a) attenuation versus frequency (narrowband); (b) insertion loss versus frequency
(zoom in of (a) in the passband); (c) attenuation versus frequency (broadband); and (d) return loss versus
frequency. (© 2006 Avago Technologies [28].)

which is particularly important in mobile phones. In this way, the duplexer avoids
the need for dedicated antennas or switches to select the transmission (TX) and
reception (RX) channels by implementing two filters within the same circuit, as
depicted in the diagram of Figure 9.8. The first duplexer port (PORT 1) connects the
output signal of the power amplifier to the transmitter filter, while the second port
(PORT 2) is employed to connect the signal incoming from the antenna and filtered
by the passband RX filter. The third port (PORT 3) connects the antenna to both fil-
ters. Two specifications mainly characterize a duplexer:

1. The TX and RX filters have disjoint passbands to allow proper selection of

the TX and RX signals.
2. The isolation between the TX and RX ports PORT1 and PORT 2 must meet
the application requirements. Thus, both the TX and RX filters must pro-
vide enough selectivity and out-of-band rejection to reduce the signal circu-
lation occurring at the antenna port (PORT 3).

The Avago Technologies ACMD-7602 is a miniature duplexer designed for use

in UMTS band I handsets [33]. The maximum insertion losses in the TX channel are
only 1.5 dB, which minimizes current drain from the power amplifier into the RX
274 Radio Frequency Applications

(a) (b)
Figure 9.7 GPS and 802.11a RF BAW Filters from TriQuint Semiconductor: (a) attenuation versus
frequency (S21) of the 1,575-MHz GPS filter; and (b) attenuation versus frequency (S21) of the
5,775-MHz 802.11a filter. (© 2007 TriQuint Semiconductor Inc. [29, 30].)

Figure 9.8 Schematic diagram of a duplexer: the three ports connect the transmitter output
(PORT 1), the receiver input (PORT 2), and the antenna (PORT 3).

section. Insertion losses in the RX channel are achieved at a maximum of 2.0 dB,
improving receiver sensitivity. The ACMD-7602 is designed with Avago’s FBAR
technology. The duplexer enhances the sensitivity and dynamic range of WCDMA
receivers by providing more than 53-dB attenuation of the transmitted signal at the
receiver input and more than 43-dB rejection of transmit-generated noise in the
receiver band. The two duplexer filters are assembled in a molded chip-on-board
module that is less than 1.2 mm high with a maximum footprint of only 2.5 × 3.0
mm. The high power handling capability of +33 dBm is another remarkable feature
of this device. Figure 9.9(a) shows the insertion loss performances of the TX
(1,920–1,980 MHz) and RX (2,110–2,170 MHz) filters. The Epcos’ B7692 BAW
duplexer is another high-performance device for use in the WCDMA band II
(1,880–1,960 MHz) [34]. The maximum insertion losses in the usable 60-MHz
passband are 2.2 dB and 2.6 dB in the TX and RX bands, respectively. TX-RX iso-
lation levels of the transmitted signal at the receiver input and the transmit-gener-
ated noise in the receiver band are 50 dB and 48 dB, respectively. The insertion loss
9.2 Passive-Circuit Applications 275

(a)

(b)
Figure 9.9 TX rejection in RX band and RX rejection in TX band of duplexers. (a) Avago’s
ACMD-7602 UMTS band I FBAR duplexer. (© 2008 Avago Technologies [33].) (b) Epcos B7692
WCDMA band II BAW duplexer. (© 2008 Epcos [34].)

performances of the Epcos’ TX (1,850–1,910 MHz) and RX (1,930–1,990 MHz)

filters are shown in the plot of Figure 9.9(b).
So far, we have mentioned the most common use of duplexers, where the
antenna port is bidirectional, and the TX and RX are input and output ports. How-
ever, the duplexer can be designed to arbitrarily change the use of the three ports
(e.g., a single input port and two output ports). In this case, the duplexer splits the
signal incoming from the antenna and delivers filtered versions to the output ports,
which receive signals in their respective band of interest. The philosophy of the
three-port duplexer can be extended to triplexers, quintplexers, and, in general,
multiplexers. To illustrate, just consider the triplexer and quintplexer depicted in
Figure 9.10. The triplexer has four ports and the quintplexer has six ports, one of
them devoted to the antenna connection. However, the use of the remaining ports is
different in the triplexer and the quintplexer. In the Epcos B9100 SAW triplexer, the
GPS, cellular, and PCS ports are outputs to connect the RX section [35]. On the
other hand, Avago’s ACFM-7102 quintplexer features an input S-GPS port, two TX
cellular and PCS output ports, and two RX cellular and PCS input ports connecting
to the receiver [36].
The frequency response of the Epcos B9100 SAW triplexer exhibits low inser-
tion losses in its three passbands. Its features are maximum values of 0.8, 0.9, and
2.0 dB in the cellular, PCS, and GPS bands at central frequencies of 859, 1,920, and
276 Radio Frequency Applications

(a)

(b)
Figure 9.10 Schematic diagram of multiplexers. (a) Epcos B9100 SAW Cell/GPS/PCS triplexer. (©
2008 Epcos [35].) (b) Avago’s ACFM7102 PCS/Cellular/S-GPS Quintplexer. (© 2008 Avago Tech-
nologies [36].)

1,575 MHz, respectively. Typical interband attenuations are 35 dB (cellular to GPS)

and 23 dB (PCS to GPS), and power handling of +31 dBm. The transfer function of
Figure 9.11(a) shows the antenna to PCS band attenuation, where the interband
requirements specified by the standard are depicted with the rectangular masks at
850 and 1,575 MHz. Note the deep attenuations provided by the filter at the GPS
and cell bands. The Avago Technologies ACFM-7102 quintplexer combines PCS
and cellular duplexer functions with a GPS filter and is packaged in a module that is
less than 1.2 mm high with a footprint of only 5 × 8 mm and high power handling of
+33 dBm. The transfer functions of Figure 9.11(b) show the wideband insertion
losses of both the PCS and cellular duplexers of the ACFM-7102. The insertion loss
and attenuation performances of all the quintplexer ports are specified as follows
[36]:

• GPS filter (1,574.4–1,576.4 MHz): insertion loss: 2.0 dB max., isolation in

PCS/Cell Tx 45 dB min.;
• Cellular duplexer Rx (869–894 MHz): insertion loss: 3.6 dB max., noise
blocking: 40 dB min.;
• Cellular duplexer Tx (824–849 MHz): insertion loss: 2.6 dB max., interferer
blocking: 55 dB min.;
9.2 Passive-Circuit Applications 277

(a)

(b)
Figure 9.11 Transfer function of the insertion loss of multiplexers. (a) Epcos B9100 SAW
cell/GPS/PCS triplexer. (© 2008 Epcos [35].) (b) Avago’s ACFM7102 PCS/cellular/S-GPS
Quintplexer. (© 2008 Avago Technologies [36].)

• PCS duplexer Rx (1,930.5–1,989.5 MHz): insertion loss: 4.4 dB max., noise

blocking: 40 dB min.;
• PCS duplexer Tx (1,850.5–1,909.5 MHz): insertion loss: 3.9 dB max.,
interferer blocking: 53 dB min.

More than one acoustic technology can be combined to produce a duplexer. For
example, Epcos also developed the hybrid B7686 SAW-BAW duplexer for
WCDMA band II handsets. This combination allows the duplexer to have
improved TX-to-RX and RX-to-TX attenuations of 45 and 50 dB [37]. The filters
can also be combined to produce more complex RF front-end modules, as we study
in Section 9.3.4.

9.2.3 Microelectromechanical Filters

Unlike the acoustic resonator applications previously studied, the design of sili-
con-based RF MEMS circuits considers different operation principles, actuation
278 Radio Frequency Applications

mechanisms, and resonance modes. In this section, we study how movable resonant
structures made of different materials, mainly polysilicon, are combined to produce
filters. As a difference to the acoustic wave propagating through or at the surface of
the piezoelectric layer of acoustic resonators, the micromechanical resonators move
themselves in flexural, torsional, or extensional resonance modes (or combination
of them). When several of these resonators are mechanically coupled, the frequency
response is also combined, and the resulting system behaves as a filter, for example.
The concept of MEMS-resonator filters has extensively been developed at uni-
versities in the United States, Europe, and Japan. The groups of Professor C. T.-C.
Nguyen at University of Michigan and UC–Berkeley have been among those pro-
posing the first concepts and models [38–40]. One of these concepts is shown in the
SEM image of Figure 9.12(a), where the perspective view of a two-resonator filter
with electrodes is observed. In this example, the filter consists of two identical
mechanical clamped–clamped beam resonators, coupled mechanically by a flex-
ural-mode beam. Conducting strips underlie the central regions of each resonator
and serve as capacitive transducer electrodes positioned to induce resonator vibra-
tion in a direction perpendicular to the substrate. The resonator-to-electrode gaps
are 1,300Å. The MEMS filter can be modeled through an equivalent mechanical
circuit. In the circuit, a mass-spring-damper system represents each resonator, while
the coupling beam corresponds to a network of mechanical springs, as depicted in
the model of Figure 9.12(b). Such a coupled two-resonator system exhibits two
mechanical resonance modes with closely spaced frequencies that define the filter
passband [38].
As studied by Nguyen et al. and according to the previous model, the center fre-
quency of the filter is mainly determined by the frequencies of constituent resona-
tors. On the other hand, the stiffness of the coupling spring largely determines the
spacing between modes—the bandwidth. As it happens in mechanical resonators,

(a)

k r1 ks12a ks12b kr2

m r1 m r2

c r1 ks12c c r2

(b)
Figure 9.12 Two-resonator microelectromechanical filter: (a) SEM image showing the
clamped-clamped resonators, coupling beam, and electrodes; and (b) mechanical model of the fil-
ter. (© 2000 IEEE [38].)
9.2 Passive-Circuit Applications 279

the application of an electrical input signal creates an electrostatic force between the
input electrode and the first conductive resonator. The force induces vibration of
the input resonator whenever the input signals have a frequency within the
passband of the filter. In time, this vibration is transmitted to the output resonator
via the coupling spring, causing it to vibrate as well. Vibration of the output resona-
tor creates a dc-biased current and time-varying capacitor between the conductive
resonator and the output electrode, which then sources an output current flowing
through the termination, read-out impedance. If the value of the impedance is prop-
erly chosen, the filter exhibits a flattened passband as shown in Figure 9.13.
Sony Corporation proposes a bandpass filter for the VHF band using MEMS
technology [41]. Additionally, Sony has succeeded in integrating a MEMS filter
with peripheral circuits on a single chip by adding MEMS processing into the exist-
ing BiCMOS process. The bandpass filters are created by combining four resona-
tors. Thus, they are a collection of parallel-coupled clamped-clamped beam
structures, whereby the resonating beam is held fixed at its both ends. Sony’s contri-
bution is the three-port resonator design, depicted in Figure 9.14(a): the beam reso-
nance mode is changed from first order (a single wave) to second order (two waves),
and the input and output signal lines are placed under the beam independently. The
design improves the signal-to-noise ratio and reduces the influence of manufactur-
ing variations in the patterning because of the longer beam length required in the
second-order flexion mode [41]. Next, Sony increased the signal level by arranging
multiple resonators in parallel. By adjusting the resonator parallel coupling pattern,
they achieve Q factors and insertion losses appropriate for the bandpass filter. Fig-
ures 9.14(b) and 9.14(c) show schemas of the four-resonator lattice-type bandpass
filter and its frequency response, respectively. With a central frequency of 97 MHz,
the filter performs a 3.9% bandwidth (3.8 MHz), a good 3.5-dB insertion loss, and
out-of-band suppression of 33.8 dB.
Sony next created an embedded MEMS filter chip that consists of this MEMS
filter and the impedance matching circuit as the next stage. Polysilicon resonators
are formed after the front end of line (FEOL) process in their BiCMOS process.
After the multilayer interconnect process has completed, the interlayer dielectric

Figure 9.13 Frequency response of the two-resonator microelectromechanical filter: its central
frequency is 7.8 MHz with a quality factor of 435. (© 2000 IEEE [38].)
280 Radio Frequency Applications

(a)

(b)

(c)
Figure 9.14 Sony’s four-resonator MEMS filter: (a) three-port constituent MEMS resonator; (b)
lattice-type bandpass filter circuit; and (c) frequency response of the filter. (© 2009 Sony [41].)

(ILD) and the sacrificial layer around the MEMS resonators are removed with buf-
fered-HF treatment [41].
The design of MEMS filters can be done by following the same procedures and
using the same network topologies and coupled-resonator ladder-filter synthesis
techniques of electronic filters. The topology and the geometry of the filter resona-
tors can be very diverse, such as those implemented using the disk and squared array
9.2 Passive-Circuit Applications 281

filters developed by Li et al. [40] and Clark et al. [42], respectively. The MEMS filter
constituted by parallel-coupled arrays of microelectromechanical square resonators
demonstrated by Demirci et al. [43] greatly reduces the equivalent impedance of the
aggregate resonator by up to a factor of 30 in an array of 30 strongly coupled
devices while retaining high quality factors near 10,000.
Figure 9.15 shows one of these resonator array filters and its frequency response
with a central frequency of 68 MHz, stopband rejection of 25 dB, and insertion loss
less than 2.7 dB. The MEMS filter design is based on mechanical soft-coupling of
the resonators. They have been arrayed in parallel through hard mechanical cou-
pling in order to reduce the overall impedance level and improve power handling.
The hard-coupling serves to lock the frequencies of the resonators, eliminating spu-
rious responses, which might otherwise arise from simple electrical parallelism. The
single cell consists of one input resonator coupled to one output resonator, and the
resulting filters are built from this basic cell up to arrays consisting of 30 cells [42].
A more sophisticated resonator array concept involves the design of the
medium-scale integrated (MSI) vibrating microelectromechanical filter proposed by
Li et al. [40]. Through a hierarchical building block approach, one of the most sig-
nificant contributions of this work is the demonstration that mechanical circuit
design methodologies can be just as powerful as those used in the transistor world to
enhance functionality. The MSI filter circuit utilizes radial-mode disks and mechan-
ical link elements to achieve low motional resistance while suppressing unwanted
modes and feed-through signals with a 0.06% bandwidth insertion loss less than

(a)

0
−3
Transmission [dB]

−6
−9
−12
−15
−18
−21
67.60 67.80 68.00 68.20 68.40 68.60
Frequency
(b)
Figure 9.15 Square-resonator array filter: (a) parallel-coupled array of 11 microelectromechanical
square-resonator cells; and (b) frequency response of the filter (central frequency is 68 MHz). (©
2003 IEEE [43].)
282 Radio Frequency Applications

2.5 dB at the center frequency of 163 MHz. The design strategy overcomes the
impedance deficiencies by optimizing the mechanical coupling between the resona-
tors. This improves the stopband rejection of the filter response while also
suppressing unwanted modes in the same footprint.
The MSI filter circuit shown in Figure 9.16(a) is comprised of four disk-array
composites (assigned numbers from 1 to 4), each of which contains 15 con-
tour-mode disk resonators. As shown in the schematic drawing of Figure 9.16(b),
which zooms in on one of the arrays, its resonators are linked by λ/2 longitudinal
mode array-coupling beams, thus promoting in-phase resonance among the resona-
tors in each of the four arrays. This allows summing of their motional currents to
achieve a lower overall impedance and higher power handling capability. Via the
λ/2 mechanical coupling beams, each array behaves like a single composite resona-
tor with much lower impedance. The coupling strategy can be summarized as: (1)
half-wavelength λ/2 couplers accentuate the in-phase motion of disks; (2) λ couplers
force disks to mechanically vibrate out-of-phase, hence enabling differential mode
operation; and (3) λ/4 couplers spread the frequencies of the multiple-resonator sys-
tem to form the bandpass response desired for the filter [40].

9.2.4 RF MEMS Switches

Microelectromechanical switches are one of the key RF MEMS components the
market is paying more attention to. The MEMS switch can be used to select one or
more RF channels or as a part of multiple-element reconfigurable RF circuits like
phase arrays, matching networks, or reconfigurable antennas [44–46].

(a)

(b)
Figure 9.16 MSI disk-array MEMS filter: (a) 60-resonator four-array filter; and (b) zooming in on
one of the arrays detailing the coupling strategy between the resonators. (© 2007 IEEE [40].)
9.2 Passive-Circuit Applications 283

Nowadays, no commercial RF MEMS switch product is available, although big

R&D and funding efforts are being made for this purpose. The MEMS switch is a
binary component having two operation states—namely, the on (down) and the off
(up) states. Thus, the MEMS switch is placed on a microwave transmission line to
isolate or to allow the RF signal to pass through the RF circuits. When operated in
the on (or down) state, the switch is in a low impedance configuration and most of
the RF signal passes through it. In the off (or up) state, the switch exhibits a high
impedance to the transmission line and most of the RF signal is prevented from
passing through the switch.
Two placement configurations of the MEMS switch on the transmission line are
possible: the series and the shunt switch. Figure 9.17(a) depicts a conceptual draw-
ing of the series switch. In the series connection, the switch connects two segments
of the transmission line, and the RF signal flows from the first segment to the second
when the switch is in the on (down) state, because it is aligned with the segments. In
the up state (off) of the switch, the RF signal flow is interrupted because of the two
segments isolation. On the other hand, the shunt switch derives the energy to a
shunt branch of the circuit, and the operation is exactly the opposite: in the on state,
most of the RF signal is derived to the shunt and the second segment of the transmis-
sion line is isolated from the first segment. During the off state, the switch opens, the
shunt is isolated, and the RF signal energy flows between the two segments of the
line. Figure 9.17(b) depicts the shunt switch concept.
Roughly, there exist three kinds of RF MEMS switches: the ohmic-contact, the
capacitive, and the resonant switch. The performance of switches is evaluated in
terms of power consumption, insertion losses, isolation, and linearity. The switch-
ing is controlled by an actuation voltage—not always as low as desirable for IC inte-

(a)

(b)
Figure 9.17 RF MEMS switch concepts: (a) the series switch (the down state coincides with
closed-circuit and signal passing through the segments, and the up state with RF isolation between
segments 1 and 2 of the transmission line); and (b) the shunt switch isolates the segments in the
down state and allows signal passing in the up state.
284 Radio Frequency Applications

gration—that causes the movable structure of the switch to bend or to deflect. The
structure is typically a bridge or a cantilever, and it contacts the transmission line
when deflected (the on state). In this state, the switch has to be as electrically
matched as possible to the transmission line, so the insertion loss is to be at mini-
mum values, less than 1 dB. In the off state, good isolation at a wide bandwidth is
expected to be at least 25 dB or more [44]. The schematic of Figure 9.18 represents
the operation states and the mechanical configuration of a bridge switch when it is
actuated by the driving voltage.
Peroulis et al. realized that, with a proper design, the isolation properties of
capacitive switches could be improved if the impedance of the switch is controlled
by a driving AC signal. The signal makes the switch to resonate at a certain fre-
quency, thus minimizing the impedance and improving the isolation in the down
state [47–49]. Figure 9.19(a) depicts the resonant switch concept and simplified
equivalent circuit. On the other hand, Figure 9.19(b) shows a picture of the fabri-
cated MEMS switch. When the switch is in resonance, the impedance of the series
inductance L and the capacitance C is null and the switch overall impedance reduces
to R. The values of L, C, and R are designed to minimize at resonance and to
maximize the isolation.
The relevance of MEMS switches in RF circuits is especially notorious in the
state-of-the-art RF front end concepts, as we will study in Section 9.3.3. Nowadays,
RF MEMS switch products are still in early production phases, although big R&D
and funding efforts are being made to launch commercialization. Philips, Sivestra,
and Baolab are at the industrial development stage and are expected to launch
high-performance miniature MEMS switches in the next few years [20–22].

(a)

(b)
Figure 9.18 Operation states and mechanical configuration of a bridge switch: (a) off (up) state;
and (b) on (down) state.
9.3 Active-Circuit Applications 285

ZTL ZTL
L

(a)

(b)
Figure 9.19 Resonant MEMS switch: (a) concept and simplified equivalent circuit (at resonance,
the impedance equals to R); and (b) optical picture of a MEMS switch. (© 2000 IEEE [47].)

9.3 Active-Circuit Applications

An active-circuit application integrates transistor-based circuits with passive com-

ponents to perform amplification or signal processing functions. Typical examples
are oscillators, mixers, and low noise amplifiers (LNA). The integrated circuit per-
forms the amplification or signal processing, and the passive component, the reso-
nator in our case, is used as time-varying or frequency-varying impedance. The IC
and resonator integration may be hybrid, monolithic, or heterogeneous, as it has
been demonstrated by academic researchers and industrial companies. We discuss
implementations of both kinds in this section.

9.3.1 Oscillators
An oscillator is a closed-loop circuit with positive feedback. That means that the sig-
nal running around the loop is constructively amplified in such a way that the out-
put signal grows and oscillates in the absence of input signal. Instead of that, the
noise in the system is enough to start up the oscillation given that two conditions are
met:

1. The gain around the loop must be equal or higher than 1 (0 dB).
2. The total phase shift in the loop must equal to 0º.
286 Radio Frequency Applications

These conditions are known as the Barkhausen criteria [50].

The oscillator can also be viewed as a two-port negative resistance circuit,
which is somehow compensated by a positive-resistance load at the oscillation fre-
quency [51]. Both representations of the oscillator are depicted in the equivalent cir-
cuits of Figure 9.20. According to the negative resistance representation, the
oscillator is designed through large-signal analysis of the active negative resistance
circuit in order to find the circuit parameters that exactly cancel the load or any
other positive resistance in the closed loop circuit [52].
There exist many circuit topologies and kinds of oscillators, though the crystal
oscillator is the main concern of this book. Crystal oscillators are widely used to
generate standard reference frequencies in modern electronic systems. The influence
of parasitic elements on crystal oscillators and the oscillation start-up conditions
have been studied in the past [53]. In a crystal oscillator, the positive resistance com-
pensation is provided through the frequency-varying impedance of a crystal resona-
tor. The first crystal oscillators implemented quartz-crystal resonators, which have
been systematically replaced by miniaturized MEMS, NEMS, SAW, and FBAR res-
onators. The general theory that allows the linear and nonlinear analysis of any
crystal oscillator circuit was developed by Vittoz et al.; for a reference guide, see
[54]. Next, we discuss some of these implementations and the major achievements
of the miniature integrated oscillators.
The 5-GHz FBAR-based balanced oscillator developed by Aissi et al.
monolithically integrates the FBAR resonator above the AMI semiconductor
0.35-μm BiCMOS IC oscillator. The balanced, differential configuration of the
oscillator allows division by two the electrode resistance of the FBAR, which leads
to better phase noise in comparison to single-ended oscillators. The measured phase
noise is −121 dBc/Hz at 100 kHz from a 5.46-GHz carrier frequency. Another bene-
fit of the differential output configuration is that it allows direct driving of the
divider and mixer, which are generally balanced, without the need for a single-
ended to differential converter [55]. The schematic circuit of the balanced Colpitts
oscillator is shown in Figure 9.21(a). The core oscillator transistors T1 and T2 are

(a)

(b)
Figure 9.20 Equivalent circuits for oscillators: (a) positive feedback oscillator; and (b) negative
resistance oscillator.
9.3 Active-Circuit Applications 287

(a)

(b)
Figure 9.21 Above-IC FBAR low phase noise balanced oscillator: (a) schematic circuit of the bal-
anced Colpitts oscillator; and (b) chip layout micrograph. (© 2006 IEEE [55].)

equally sized, and the buffers T3 and T4 isolate the core oscillator from 50Ω imped-
ance loads also having the same size. The chip occupies 650 × 830 μm of silicon
area, with the resonator area including contacts to the BiCMOS IC last metal being
170 × 300 μm. The micrograph of Figure 9.21(b) shows the FBAR above the oscilla-
tor circuit.
A quite different approach is the MEMS oscillator developed by Lin et al. The
circuit replaces the single resonator normally used in crystal oscillators with a
mechanically coupled array to effectively raise the power-handling ability at the
working frequency. Lin and colleagues integrate a mechanically coupled array of up
to nine 60-MHz wine-glass disk resonators embedded in a positive feedback loop to
attain oscillation. The circuit achieves a phase noise of –123 dBc/Hz at 1-kHz offset
and –136 dBc/Hz at far-from-carrier offsets, beating the GSM phase noise require-
ments by 8 dB and 1 dB, respectively [56]. A detailed schematic of the circuit show-
ing the sustaining transresistance amplifier and the placement of the resonator array
is depicted in Figure 9.22(a). A SEM image of the disk array comprised by three
wine-glass resonators is observed in Figure 9.22(b).
The self-excited, nanocantilever resonator-based CMOS oscillator developed
by Verd et al. is used for mass detection applications, and it can achieve 1 ag/Hz
288 Radio Frequency Applications

(a)

(b)
Figure 9.22 Wine-glass disk-resonator array MEMS oscillator. (a) Detailed circuit schematic of the
single-stage sustaining transresistance amplifier, implemented by a fully differential amplifier in a
one-sided shunt-shunt feedback configuration. (© 2005 IEEE [56].) (b) SEM image of the
three-wine-glass disk resonator array. (© 2005 IEEE [56].)

mass sensitivity [57] (Figure 9.23). The mechanical resonator is monolithically inte-
grated above the commercial 0.35-μm AMS035 CMOS process and implemented
using the top metal layer and post-CMOS processed through simple mask-less wet
etching. The MEMS-adapted Pierce oscillator vibrates at a frequency of 6 MHz
with a 1.6-Hz frequency stability in air environment. The submicrometer-scale res-
onator is based on a cantilever structure 10 μm long, 600 nm wide, and 750 nm
thick, and it features three electrodes for DC voltage biasing, electrostatic excita-
tion, and capacitive read-out.
Discera (Ann Arbor, Michigan) has developed and sells a complete line of
silicon-based MEMS oscillators [18]. The DSC8002 is a programmable,
MEMS-based oscillator fabricated with the Discera PureSilicon technology. This
circuit can be programmed to any frequency from 1 to 150 MHz, with a nominal
operational range of 1.8V to 3.3V. The DSC8002 incorporates a robust silicon
MEMS resonator that is intended for industrial and portable applications like
mobile, consumer electronics, and CCD clocks for VTR cameras, among others.
The oscillator and resonator circuit exhibits low operating and standby current of 3
mA (at 40 MHz) and 1 μA, respectively, within a small footprint of 2.5 × 2.0 × 0.85
mm and a temperature stability of ±25 ppm to ±50 ppm. The micrograph of Figure
9.24(a) shows the footprint of the oscillator, wherein the resonator is smaller than
the ASIC. This is a remarkable difference from previous art quartz-crystal oscilla-
tors whose resonators dominated the size of the package. The block diagram of
9.3 Active-Circuit Applications 289

(a) (b)
Figure 9.23 Submicrometer resonator-based oscillator: (a) photograph of the cantilever-based
oscillator monolithically integrated in AMS’s 0.35-μm CMOS process; and (b) SEM image of a fab-
ricated 10-μm-long, 0.6-μm-wide metal cantilever. (© 2008 IEEE [57].)

Figure 9.24(b) depicts the architecture of the VCO oscillator, including an

N-divider to perform the frequency programming. The MEMS resonator and the
ASIC are integrated by wire bonding [58].
SiTime is another company that nowadays commercializes silicon MEMS oscil-
lators [19]. The SiTime SiT3701 is one of the smallest programmable voltage-con-
trolled MEMS oscillators (VCMO) on the market today. The SiT3701 all-silicon
VCMO comes in four-pin packages with the smallest footprint being 2.5 × 2.0 mm.
The circuit is programmable to obtain oscillation at any frequency from 1 MHz to
80 MHz, and at selected frequencies between 80 MHz and 110 MHz, with supply
voltage options of 1.8V, 2.5V, 2.8V, and 3.3V, and frequency stability from ±25
ppm to ±100 ppm. This oscillator is also oriented for portable consumer applica-
tions [59].
Other oscillator implementations may be mentioned as well, like the SAW oscil-
lator circuit of Schmitt et al. [60], the ultra-low-power injection locked transmitter
for wireless sensor networks based on FBAR by Chee et al. [61], or the commer-
cially available SAW and QMEMS oscillators by Epson Toyocom [31].

9.3.2 Mixers and Mixlers

A mixer is a nonlinear circuit that performs frequency conversion from the RF fre-
quency to intermediate frequencies. In essence, a mixer is a signal multiplier that
uses a reference signal of a certain frequency to downconvert the frequency spec-
trum of the input signal. Taking advantage of signal processing and well-known
trigonometric relationships, the product of the RF and the reference signals gener-
ates a couple of new signals. The first one is at the difference of the central frequen-
290 Radio Frequency Applications

(a)

VDD Output

Resonator PFD VCO

Frac-N

PLL

Standby# GNS
(pin 1)

(b)
Figure 9.24 Discera’s DSC 8002 MEMS oscillator: (a) footprint of the oscillator showing the
MEMS resonator above the ASIC; and (b) block diagram. (© 2009 Discera [58].)

cies of the RF and reference signals, and the second one is centered at the sum of
such frequencies:

f IF = f RF − f LO (9.1)

f M = f RF + f LO (9.2)

where fIF is the intermediate frequency, fM is the high-frequency image of fIF, fRF is the
RF frequency, and fLO is the local-oscillator reference frequency. In a down con-
verter, the fM signal is discarded by using a lowpass (or bandpass) filter, while the
fIF-centered signal is then delivered to the baseband stages to perform further signal
processing and information recovery. An upconverter operates in the opposite way,
by taking the added-frequency signal and filtering out the difference signal. Tradi-
tionally, the mixers have been built with conventional transistor-based circuits.
Herein, we study how new MEMS-based mixer implementations are also able to
achieve RF signal downconversion with good performance.
9.3 Active-Circuit Applications 291

MEMS mixer filters exploit the voltage-squared nonlinearity of the electrostatic

force driving the resonators, as studied by Fedder et al. [62]. Mixing and IF filtering
can be accomplished in a 0.02-mm2 area with less than 1 mW of input power. The
square geometry at the head of the 2.29-MHz cantilever of Figure 9.25 reduces
feedthrough by distancing the RF and IF electrodes. Application of RF voltage VRF
to the stationary electrode and local oscillator voltage VLO to the cantilever gener-
ates an electrostatic force proportional to (VRF − VLO)2 that acts on the moving canti-
lever. The VRF × VLO force component then has a frequency of (fRF − fLO). The
mechanical vibration of the cantilever is amplified by the mechanical Q value and
read-out capacitively through the IF electrode. Successful mixing with these resona-
tors has been observed up to 3.2-GHz input frequencies. The circuit is
monolithically integrated within a tunable RF front end fabricated within the Jazz
SiGe60 4-metal BiCMOS process [63].
The term mixler is a neologism to designate the combination of a mixer and a
filter. A mixler is a device comprised of MEMS resonators and mixer transducers
performing both frequency translation and selective filtering of the electrical input
signal [64]. Figure 9.26(a) presents the schematic representation of a mixler device.
The structure of this device is basically the same as that of a microelectromechanical
filter comprised of two clamped-clamped beam resonators, each with center fre-
quency fIF, coupled at low-velocity locations by a flexural mode beam. The SEM
image of Figure 9.26(b) shows the 37-MHz microelectromechanical mixler indicat-
ing key features and dimensions of the beams, electrodes, and so on. The device con-
verts the electrical input signal to a mechanical force. Next, it processes this signal
mechanically via its network of flexural-mode beams, and then it reconverts the
resulting signal to an electrical output signal that can be further processed by subse-
quent transceiver electronics. The key to mixing in this device is its capacitive elec-
tromechanical transduction, which converts electrical energy to mechanical energy
through a square law transfer function, as conventional mixers do [65].
A second example is the fully integrated CMOS-MEMS mixler using the com-
mercial AMS035 0.35-μm technology of Uranga et al. [66]. The MEMS structure is
a clamped-clamped beam resonator implemented with the polysilicon capacitance

− 20
2.29 MHz cantilever
− 30 Q = 1620
Frequency tuning
−40 electrode
Gain
−50
[dB]
−60
Simulation
−70
Measured
−80

−90
2.28 2.29 2.30

Frequency [MHz]
Figure 9.25 RF MEMS mixer filter fabricated within the Jazz SiGe60 4-metal BiCMOS process. (©
2005 IEEE [62].)
292 Radio Frequency Applications

(a)

(b)
Figure 9.26 MEMS mixler: (a) schematic representation comprising a filter and a mixer; and (b)
full-view of a 37-MHz microelectromechanical mixler indicating key features and dimensions of the
beams, electrodes, and ports. (© 2004 IEEE [65].)

module of the CMOS process, thus achieving a fundamental lateral resonance fre-
quency of 22.5 MHz. To operate the MEMS as a mixler, two different approaches
were proposed to generate the nonlinear signal: the quadratic relationship of the
voltage against the excitation force, and the amplitude modulation of the excitation
signal. Figure 9.27 shows a SEM image of the released MEMS structure, which is a
clamped-clamped beam 13 μm long, 350 nm wide, and having a 150-nm capacitive
transducing gap. The polysilicon capacitance module of the AMS process was used
to define both the excitation and read-out electrodes, as well as the mobile struc-
ture. Wet etching of the sacrificial silicon oxide layer was carried out to release the
MEMS. The vibration is detected through electrostatic actuation and capacitive
read-out of the generated motional current through an on-chip integrated
transimpedance amplifier [66].

9.3.3 Tuned Low-Noise Amplifiers

A low-noise amplifier (LNA) is placed on the receiver path of RF communication
systems to amplify the low-level signal incoming at the antenna port after the band
selection stage. An LNA features a low-noise figure, typically around 3–5 dB, hence
its name. LNA circuits implement moderate gain amplifiers, RF chocks and cou-
pling capacitors for frequency compensation, and input and output impedances to
match the 50-ohm transmission line impedance of most RF circuits.
9.3 Active-Circuit Applications 293

Excitation
driver

Mobile
structure

Readout
driver

Figure 9.27 Full CMOS MEMS mixler: the clamped-clamped beam resonator is fabricated with
the polysilicon capacitance module of the AMS035 CMOS process. (© 2007 IEEE [66].)

The recent monolithic integration of FBAR and CMOS technologies has

enabled the design of new circuits such as the tuned differential LNA of Figure
9.28(a) for 2.2-GHz WCDMA applications [67]. The filtering LNA comprises two
differential broadband (1 dB-BW of 5 GHz) current feedback amplifiers connected
on each side of an FBAR filter, thus ensuring the frequency selectivity of the circuit.
The curves of Figure 9.28(b) show a measured gain larger than 20 dB and a noise
figure of 3 dB in the passband. The input gain compression at 1 dB is −26 dBm. This
moderate linearity comes from the tradeoff between the amplifier’s bandwidth, the
high gain, the good noise performance, and the low power consumption (10 mA at
3V for each stage). Thus, better linearity can be achieved if one of these parameters
is sacrificed.
The design of tuned RF circuits also involves the implementation of imped-
ance-matching techniques. Although this example is the first demonstration of
FBAR-based tuned RF amplifiers, nothing impedes the realization of new concepts
like tuned power amplifiers for RF transmitters.

9.3.4 RF Front-End Systems

At this point, we have seen how individual blocks in the RF chain can be imple-
mented with available microresonator technologies. Moreover, these RF blocks are
suitable for further integration toward the single chip, thus constituting an inte-
grated RF front end. An RF front end integrates the set of components present in a
receiver from the antenna to the intermediate frequency (IF) stage. Depending on
the receiver architecture, different components can be part of the front end. How-
ever, most receivers are constituted by matching networks, channel selection
switches, band selection filters, duplexers (or any other multiplexer), LNAs, and
phase-matching arrays. The main challenge of RF front-end systems is the efficient
handling of power and bandwidth, mainly related to insertion losses of the switches
and the filters, noise figures of the LNA, or quality factors and selectivity of the
filters.
294 Radio Frequency Applications

(a)

5 30

4.5 20
Noise figure [dB]

4 10
Gain [GHz]

3.5 0

3 −10

2.5
−20

2 −30
2 2.05 2.1 2.15 2.2 2.25 2.3
Frequency [GHz]
(b)
Figure 9.28 (a, b) Tuned LNA integrated with above-IC FBAR filter: noise figure and gain. (©
2005 IEEE [67].)

Single-chip integration of RF front ends has long been a dream for RF system
designers; however, only recently, thanks to the RF MEMS contribution, has the
integration of different components and technologies been possible. New receiver
architectures, like the massively parallel switchable RF front end, enable simulta-
neous reception and transmission of narrowband signals located in different fre-
9.3 Active-Circuit Applications 295

quency bands. These architectures are suitable for multiband 3G and B3G mobile
communication systems, and they solve the problem of multiple and bulky user
handsets. The diagram of Figure 9.29 exemplifies the massively parallel front end
concept.
In the past few years, MEMS- and FBAR-based integrated front ends have
appeared in the market of RF components. The circuit proposed in [68] is a simpli-
fied implementation of a zero-IF front end, and it consists of LNA, FBAR filter, and
mixer. The RF front-end circuit architecture is depicted in Figure 9.29(a). The
design and integration of the RF front end for WCDMA applications utilize the
above-IC FBAR bandpass filter between the LNA and the mixer in order to relax
linearity constraints and, thus, power consumption for the downconversion mixers.
This first experimental chip is designed and fabricated in a 0.25-μm SiGe:C
BiCMOS process enhanced with above-IC capabilities, whose layout is observed in
the micrograph of Figure 9.29(b). The BAW filter is fabricated by CEA-LETI and
CSEM. The above-IC BAW process is plugged over the BiCMOS final passivation
layer, and the electrical contact with the rest of the circuit is realized at the last IC
metal layer [68].

Mixer

IF_I
Antenna
LNA
90 Local
0 oscillator
On-chip bandpass
BAW filter
IF_Q
Duplexer

Bandpass filter

PA From transmitter

(a)

(b)
Figure 9.29 RF front end consisting of monolithically integrated BiCMOS LNA, FBAR filter, and
downconversion mixer: (a) block diagram of the WCDMA-ZIF transceiver and (b) micrograph of
the front end (LNA, filter, and mixer). (© 2005 IEEE [68].)
296 Radio Frequency Applications

At the industrial level, Epcos offers the D5013 SAW-switch front-end module
for GSM850, EGSM, DCS, PCS, and WCDMA2100 services. The D5013 is a
low-loss SAW module for mobile telephone system integrating TX lowpass filters,
switches, and decoders. Besides, it integrates SAW filters for reception of GSM 850,
EGSM, PCN, and PCS signals, and ESD protection at the antenna port to 8 kV for
possible contact discharges [69]. The schematic drawing of Figure 9.30 depicts the
parallel architecture of the D5013 front end. Each one of the bands is selected by a
multiple-switch selector controlled by the decoder. The maximum insertion loss
value, including all the received and transmitted bands is 3.6 dB. The front end is
assembled in SMT, with an approximate weight of 100 mg, a height of 1.1 mm, and
a footprint of 3.2 × 4.5 mm.
Another integrated front-end concept is the Avago Technologies ALM-1712,
which is a 1.575-GHz GPS front-end module combining an LNA with GPS FBAR
filters. The LNA uses Avago’s proprietary GaAs enhancement-mode (pHEMT) pro-
cess to achieve a high gain with very low noise figure and high linearity. The inte-
grated filter utilizes an FBAR filter for rejection at cellular- and PCS-band
frequencies with the small footprint of 4.5 × 2.2 mm and a height of 1 mm. The
LNA has a gain of 12.8 dB, a noise figure of 1.65 dB, and cellular-band and
PCS-band rejections of 95 dBc and 90 dBc, respectively, among other specifications
[70]. The schema of Figure 9.31(a) depicts the block diagram of the ALM-1712,
where the RF input signal is filtered first by an FBAR GPS filter, then amplified by
the LNA, and postfiltered and delivered to the RF output by another FBAR GPS fil-

RX_PCS

RX_PCN

RX_EGSM

ANT RX_GMS850
ESD
WCDMA 2100

TX_PCN/PCS

TX_GSM850/
EGSM

Decoder

Figure 9.30 Epcos D5013 parallel RF front-end module for GSM850, EGSM, DCS, PCS, and
WCDMA2100: the D5013 integrated SAW filters for band selection by means of switches and the
decoder. (© 2009 Epcos [69].)
9.4 Summary 297

(a)

(b)

Figure 9.31 Avago Technologies’ ALM-1712 front-end module: the chip integrates an LNA with
GPS FBAR filters: (a) block diagram and (b) gain and return loss versus frequency. (© 2009 Avago
[70].)

ter. The 12.8-dB gain and return loss of the front-end module are observed in Figure
9.31(b). The curves are taken when the chip is powered with a voltage supply of 2.7
VDC and current of 8 mA.

9.4 Summary

In this chapter, we have reviewed the major RF applications of acoustic and electro-
mechanical microresonators. These applications and the MEMS technology behind
them are known as the RF MEMS field. As we studied, RF MEMS applications are
diverse and have high market impact. Nowadays, RF MEMS have passed from con-
cept demonstration to intense commercialization, and different acoustic and
microelectromechanical technologies are available in the market. Nevertheless,
while acoustic SAW and BAW resonator technologies—including FBAR and
SMR—have reached the required maturity for mass production and commercializa-
tion, most silicon-based RF MEMS are still in development, or they have
already been developed but are in the consolidation process. Thus, filters, duplex-
ers, and filter-based front ends have been successfully miniaturized using SAW and
298 Radio Frequency Applications

FBARs, and millions of units of them are sold every year. On the other hand,
MEMS-based RF oscillators are beginning their commercialization and have a
promising future, and the highly anticipated RF switches are still being developed to
reach industrial standards.
In this context, the RF MEMS market is very concentrated in a few companies,
in spite of the big size of the overall MEMS industry. Mobile telephony FBAR and
BAW applications dominate the market and represent nearly 40% of the total sales.
From the oscillator and switch side, they are seen as interesting because of volume
and economy scales, although reliability and CMOS integration issues are to be
solved. Besides, design efforts will be intensive, because new RF architectures are to
be developed to fully exploit the potential of integrated RF MEMS. This is a chal-
lenge and an opportunity for RF designers. We have seen in this chapter how the
integration of RF MEMS is increasing and how, to date, the single-chip front end is
a market reality that has left the laboratory and conceptual promises to occupy a
place in the RF MEMS commercial applications. Of course, further integration,
miniaturization, and process compatibility is expected for the next few years to
build even more powerful and cost-efficient RF front ends.

References

[1] Apple, iPhone, http://www.apple.com/iphone/.

[2] Universal Mobile Telecommunications System (UMTS), http://www.umts-forum.org.
[3] Wi-Fi Alliance, Wi-Fi: Local area wireless networking based on IEEE 802.11 standards,
http://www.wi-fi.org.
[4] High Speed Packet Access HSPA technology, GSM World, the GSM Association,
http://www.gsmworld.com/technology/hspa.htm.
[5] Global System for Mobile Communications, GSM World: the GSM Association,
http://www.gsmworld.com.
[6] Enhanced Data Rates for GSM Evolution EDGE Technology, GSM World: the GSM Asso-
ciation, http://www.gsmworld.com/technology/edge.htm.
[7] Bluetooth Technology, http://www.bluetooth.com.
[8] Global Positioning System, http://www.gps.gov.
[9] Nokia N97 Mobile Phone, Nokia: Connecting People, http://europe.nokia.com/n97#.
[10] Samsung Omnia mobile phone, Samsung Mobile, http://uk.samsungmobile.com/
mobile-phones/samsung-omnia.
[11] LG Arena, LG Mobile, http://www.lgmobile.com.
[12] Nguyen, C. T.-C., “Microelectromechanical Components for Miniaturized Low-Power
Communications,” Dig. of Tech. Papers IEEE MTT-S Intl. Microwave Symposium RF
MEMS Workshop 1999, Anaheim, CA, June 18, 1999, pp. 48–77.
[13] Bouchaud, J., B. Knoblich, and H. Wicht, “RF MEMS Market,” Wicht Technologie Con-
sulting, Munich, Germany, 2005.
[14] Avago Technologies, http://www.avagotech.com.
[15] Epcos AG, http://www.epcos.com.
[16] TriQuint Semiconductor, http://www.triquint.com.
[17] Skyworks, http://www.skyworksinc.com.
[18] Discera, http://www.discera.com.
[19] SiTime Corporation, http://www.sitime.com.
9.4 Summary 299

[20] Royal Philips, http://www.philips.com.

[21] Baolab Microsystems, http://www.baolab.com.
[22] Fujishima, S., “The History of Ceramic Filters,” IEEE Trans. on Ultrason. Ferroelec. Freq.
Control, Vol. 47, 2000, pp. 1–7.
[23] Lakin, K. M., G. R. Kline, and K. T. McCarron, “Development of Miniature Filters for
Wireless Applications,” IEEE Trans. on Microw. Theory Tech., Vol. 43, 1995,
pp. 2933–2939.
[24] Ruby, R., et al., “Ultra-Miniature High-Q Filters and Duplexers Using FBAR Technology,”
Dig. of Tech. Papers IEEE Intl. Solid-State Circuits Conf. ISCC 2001, San Francisco, CA,
February 5–7, 2001, pp. 120–121.
[25] Kim, K.-W., et al., “Resonator Size Effects on the TFBAR Ladder Filter Performance,”
IEEE Microw. Wirel. Compon. Lett., Vol. 13, 2003, pp. 335–337.
[26] Centro Nacional de Microelectrónica (CNM), Two-Metal CNM25 CMOS Technology,
http://www.cnm.es.
[27] Avago Technologies, ACPF-7003 “High Reception and Image Rejection Transmission Fil-
ter for the US PCS Band,” 2006, http://www.avagotech.com.
[28] Avago Technologies, “ACPF-7002 High Rejection Transmission Filter for the US PCS
Band,” 2007, http://www.avagotech.com.
[29] TriQuint Semiconductor, 880404, “1575 MHz BAW Filter,” 2007, http://www.triquint.com.
[30] TriQuint Semiconductor, 880369 “5775 MHz BAW Filter,” 2007, http://www.triquint.com.
[31] Epson Toyocom, http://epsontoyocom.co.jp/english/.
[32] Epcos AG, Surface Acoustic Wave Components Division, SAW MC WT, B7843, “1855
MHz Low-Loss Filter for Mobile Communications,” 2002, http://www.epcos.com.
[33] Avago Technologies, ACMD-7602, “Miniature UMTS Band I Duplexer,” 2008,
http://www.avagotech.com.
[34] Epcos AG, Surface Acoustic Wave Components Division, SAW MC WT, B7692, “BAW
Duplexer WCDMA Band II,” 2008, http://www.epcos.com.
[35] Epcos AG. Surface Acoustic Wave Components Division, SAW MC WT, B9100, “SAW
CELL/GPS/PCS Triplexer,” 2008, http://www.epcos.com.
[36] Avago Technologies, ACFM-7102, “PCS/Cellular/S-GPS Quintplexer,” 2007,
http://www.avagotech.com.
[37] Epcos AG, Surface Acoustic Wave Components Division, SAW MC WT, B7686,
“BAW/SAW Duplexer WCDMA Band II (PCS),” 2008, http://www.epcos.com.
[38] Bannon, F. D., J. R. Clark, and C. T. -C. Nguyen, “High-Q HF Microelectromechanical Fil-
ters,” IEEE J. Solid-State Circuits, Vol. 35, 2000, pp. 512–526.
[39] Nguyen, C. T. C., “Integrated Micromechanical Circuits for RF Front Ends,” Proc. 36th
European Solid-State Device Research Conference ESSDERC 2006, Montreux, Switzer-
land, September 18–22, 2006, pp. 7–16.
[40] Li, S.-S., et al., “An MSI Micromechanical Differential Disk-Array Filter,” Dig. of Tech.
Papers 14th International Conference on Solid-State Sensors, Actuators and Microsystems
TRANSDUCERS 2007, Lyon, France, June 10–14, 2007, pp. 307–311.
[41] Sony Corporation, Bandpass Filter Developed Using MEMS Technology Integration with
Peripheral Circuits Also Possible, http://www.sony.net/Products/SC-HP/cx_news/vol43/
pdf/sideview43.pdf.
[42] Clark, J. R., et al., “Parallel-Coupled Square-Resonator Micromechanical Filter Arrays,”
http://www.discera.com.
[43] Demirci, M. U., M. A. Abdelmoneum, and C. T.-C. Nguyen, “Mechanically Corner-Cou-
pled Square Microresonator Array for Reduced Series Motional Resistance,” Dig. of Tech.
Papers 12th Int. Conf. on Solid-State Sensors & Actuators TRANSDUCERS 2003, Boston,
MA, June 8–12, 2003, pp. 955–958.
[44] De Los Santos, H. J., RF MEMS Circuit Design for Wireless Communications, Norwood,
MA: Artech House, 2002.
300 Radio Frequency Applications

[45] De Los Santos, H. J., “MEMS for RF/Wireless Applications: The Next Wave—Part I,”
Microwave J., Vol. 44, 2001, pp. 20–41.
[46] De Los Santos, H. J., “MEMS for RF/Wireless Applications: The Next Wave—Part II,”
Microwave J., Vol. 44, 2001, pp. 142–152.
[47] Peroulis, D., et al., “RF MEMS Devices for High Isolation Switching and Tunable Filter-
ing,” Dig. of Tech. Papers IEEE MTT-S Intl. Microwave Symposium, Vol. 2, Phoenix, AZ,
June 11–16, 2000, pp. 217–1220.
[48] Peroulis, D., “RF MEMS Devices for Multifunctional Integrated Circuits and Antennas,”
Ph.D. thesis, University of Michigan, Ann Arbor, MI, 2003.
[49] Wang, X., L. P. B. Katehi, and D. Peroulis, “AC Actuation of Fixed-Fixed Beam MEMS
Switches,” Dig. of Tech. Papers Topical Meeting on Silicon Monolithic Integrated Circuits
in RF Systems, 2006, San Diego, CA, January 18–20, 2006.
[50] Jones, M. H., A Practical Introduction to Electronic Circuits, Cambridge, U.K.: Cambridge
University Press, 1982.
[51] Rhea, R. W., Oscillator Design and Computer Simulation, Atlanta, GA: Noble Publishing,
1995.
[52] Dearn, A., How to Design an RF Oscillator, London: Institution of Electrical Engineers,
2000, pp. 7/1–7/7.
[53] Unkrich, M. A., and Robert G. Meyer, “Conditions for Start-Up in Crystal Oscillators,”
IEEE J. Solid-State Circuits, Vol. SC-17, 1982, pp. 87–90.
[54] Vittoz, E. A., “High-Performance Crystal Oscillator Circuits: Theory and Application,”
IEEE J. Solid-state Circuits, Vol. SC-17, 1982, pp. 774–783.
[55] Aissi, M., et al., “A 5 GHz Above-IC FBAR Low Phase Noise Balanced Oscillator,” Proc.
IEEE Radio Frequency Integrated Circuits Symposium RFIC-2006, Vol. 1, San Francisco,
CA, June 11–13, 2006, pp. 25–28.
[56] Lin, Y.-W., et al., “Low Phase Noise Array-Composite Micromechanical Wine-Glass Disk
Oscillator,” Dig. of Tech. Papers IEEE Intl. Electron Devices Meeting IEDM 2005, Wash-
ington, D.C., December 5–7, 2005, pp. 281–284.
[57] Verd, J., et al., “Monolithic CMOS MEMS Oscillator Circuit for Sensing in the Attogram
Range,” IEEE Electron Dev. Lett., Vol. 29, 2008, pp. 146–148.
[58] Discera, DSC8002 Series 1.8 to 3.3V PureSilico Programmable Oscillator:
http://www.discera.com.
[59] SiTime Corporation, SiT3701 Smallest Programmable VCMO Voltage Controlled MEMS
Oscillator, http://www.sitime.com.
[60] Schmitt, R. F., J. W. Allen, and R. Wright, “Rapid Design of SAW Oscillator Electronics for
Sensor Applications,” Sens. Actuators B: Chem., Vol. 76, 2001, pp. 80–85.
[61] Chee, Y. H., A. M. Niknejad, and J. M. Rabaey, “An Ultra-Low-Power Injection Locked
Transmitter for Wireless Sensor Networks,” IEEE J. Solid State Circuits, Vol. 41, 2006,
pp. 1740–1748.
[62] Fedder, G. K., and T. Mukherjee, “Tunable RF and Analog Circuits Using On-Chip MEMS
Passive Components,” Dig. of Tech. Papers IEEE Intl. Solid-State Circuits Conf. ISCC
2005, San Francisco, CA, February 6–10, 2005, pp. 390–391.
[63] Jazz Semiconductor, http://www.jazzsemi.com.
[64] Wong, A.-C., H. Ding, and C.T.-C. Nguyen, “Micromechanical Mixer and Filters,” Dig. of
Tech. Papers IEEE Intl. Electron Devices Meeting IEDM 1998, San Francisco, CA, Decem-
ber 6–9, 1998, pp. 471–474.
[65] Wong, A. C., and C. T. C. Nguyen, “Micromechanical Mixer-Filters (‘Mixlers’),” J.
Microelectromech. Systems, Vol. 13, 2004, pp. 100–112.
[66] Uranga, A., et al., “Fully Integrated MIXLER Based on VHF CMOS-MEMS
Clamped-Clamped Beam Resonator,” Electron. Lett., Vol. 43, 2007, pp. 452–454.
9.4 Summary 301

[67] Dubois, M.-A., et al., “Integration of High-Q BAW Resonators and Filters Above IC,” Dig.
of Tech. Papers IEEE Intl. Solid-State Circuits Conf. ISCC 2005, San Francisco, CA, Feb-
ruary 6–10, 2005, pp. 392–393.
[68] Carpentier, J. F., et al., “A SiGe:C BiCMOS WCDMA Zero-IF RF Front-End Using an
Above-IC BAW Filter,” Dig. of Tech. Papers IEEE Intl. Solid-State Circuits Conf. ISCC
2005, San Francisco, CA, February 6–10, 2005, pp. 394–395.
[69] Epcos AG, Surface Acoustic Wave Components Division, SAW MC WT, D5013, “SAW
Frontend Module GSM 850/EGSM/DCS/PCS/WCDMA 2100,” 2008,
http://www.epcos.com.
[70] Avago Technologies, “ALM-1712 GPS Low-Noise Amplifier Front-End Module with Inte-
grated Pre and Post Filters and Variable Current/Shutdown Function,” 2009, http://www.
avagotech.com.
 CHAPTER 10

Case Studies: Modeling, Design, and

Fabrication of FBAR and MEMS-Based
Systems
So far, we have studied the concepts, modeling, design, fabrication, and integration
of FBAR and MEMS resonator-based microsystem and nanosystems. In addition,
we have reviewed some of their main sensor and RF applications. This chapter car-
ries out an in-depth study of specific implementation cases of FBAR and MEMS
systems.
To get started, we establish a methodological approach for the design of
MEMS-resonator and IC systems. Technological and conceptual aspects at both the
resonator and IC sides are considered. The purpose of this section is to define a
working reference passing through the different stages of system design. Designers
will learn the appropriate steps to should follow to build a successful resonator-IC
system, including modeling, characterization, parameter extraction, and process
design.
Next, we study the first case: the fabrication of an FBAR and its compatibility
with silicon technologies. In this way, we build a diary of the step-by-step fabrica-
tion process of an FBAR, and we go through a run by describing each of the imple-
mented processes and the required actions to supervise compatibility with the
global process. In Chapter 4 we reviewed surface and bulk-micromachined FBAR
implementations, and we defined the main issues regarding process compatibility.
In this chapter, we will learn how the compatibility requirements of FBAR and sili-
con processes can be met and discover the interactions between the different pro-
cesses. Thus, the combination of structural materials, their deposition and
patterning, and the micromachining techniques have to be compatible with the pre-
vious process steps. We propose a list of the materials currently implemented in
FBARs and their combination.
The second case we will revise herein is the high-level design of a temperature-
compensated FBAR-based oscillator. We begin the discussion by defining the topol-
ogy of FBAR oscillators and the impact of temperature in the oscillator stability.
Next, we perform the small-signal analysis of the oscillator by using the FBAR’s
MBVD model and the CMOS small-signal parameters. Then, we carry out the tem-
perature compensation analysis by high-level modeling of the temperature-compen-
sated oscillator. A Verilog-A model of the FBAR is useful to illustrate the design
approach.
In the last section of the chapter, we design a read-out circuit for a 434-MHz sil-
icon-MEMS resonator. We take the lessons learned in the first section to systemati-

303
304 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

cally design the circuit, passing from the system level to reach the circuit-level
implementation. First, we see how the circuit is specified as a function of the MEMS
characteristics and its extracted RLC model. We make a digression to describe the
integration approach of the MEMS and the read-out circuit. The study also
describes the implemented testing routines to predict and experimentally
characterize the circuit performance.

10.1 Methodological Approach for MEMS-IC Integration

In Chapter 3, the different levels of FBAR and MEMS resonator design and model-
ing were analyzed. Now, we use modeled FBAR or MEMS resonators to integrate
them with an IC. Thus, the extracted models will serve to specify the circuit require-
ments and expected performance so we can build a functional system.
The flow diagram of Figure 10.1 depicts the proposed methodology for
MEMS-IC integration, whose central axis represents all the actions involving inte-
gration of the left-side FBAR/MEMS resonator design steps and the right-side inte-
grated circuit steps. First, the integration technology has to be defined. In this step,
monolithic, hybrid, or heterogeneous integration strategy is decided. This decision

Integration
FBAR/MEMS technology Integrated
resonator circuit

Resonator concept Circuit topology and

and process design IC technology

Analytic and finite System-level

element modeling design

Prototype fabrication Circuit-level

and characterization design

Physical-level
Parameter extraction deisgn

Resonator-IC
integration/fabrication

System
characterization

Analysis and
redesign (optional)

Figure 10.1 Integrated MEMS-IC design methodology.

10.1 Methodological Approach for MEMS-IC Integration 305

will support the resonator concept and process design, involving compatibility
issues with the IC, like temperatures and materials. At the layout level, it also
requires previsions of the resonator active area, electrode design and contact pad
area, dicing design (if needed), through vias (if needed), temperature budget, and
process contamination, among other process elements. The integration approach
also affects decisions on the required IC technology and circuit topology. In the case
of monolithic integration, the considerations are the number of layers of the micro-
electronic process (typically CMOS or BiCMOS), thicknesses and materials of the
upper process layers, and other requirements concerning the postprocess of the IC
wafer. Whether the resonator is fabricated before, in the middle, or after the IC pro-
cess is also a matter of design. The IC layers to be used for this purpose have to be
decided in this step as well. On the other hand, the IC technology requirements for
hybrid and heterogeneous integrations are less demanding, and they are more
related to functional (system level) and layout (physical level) design.
The analytical and finite element modeling (FEM) of the resonator can now be
performed. After some iterations, the FEM analysis and the theoretical design will
provide information of the structural, motional, and dynamic response of the reso-
nator, which will be useful to generating the final process and layout design (pro-
vided that the resonator fabrication technology is already known and controlled). A
MEMS prototype can now be fabricated. If the monolithic approach is imple-
mented, prototyping does not necessarily require full integration with the IC, and
the resonator is usually accompanied with a dummy circuit for characterization and
parameter extraction purposes. Different test and de-embedding structures are used
at this stage to extract the resonator material constants, the quality factor, the
equivalent-circuit parameters, the insertion losses, and the parasitic impedance.
Along with the system-level design (system analysis at the transfer function level),
the extracted information is vital to complete the circuit-level design. Specifically,
input impedance matching with the resonator, amplification gain, or minimum and
maximum levels of the input current are some of the IC-design parameters depend-
ing on the resonator characterization. At the application level, the bandwidth,
oscillation frequency, output impedance and buffering, and power consumption are
among the designed specifications.
The resonator-IC interface has to be modeled at the circuit level. In this way and
depending on the integration approach, accurate quantification of the parasitic
effects of the wire-bonding, bumping, or IC via-holes will enable a reliable predic-
tion of the circuit response. Only after good-practice physical layout design and
postlayout simulations can the final response of the resonator-IC system be pre-
dicted within reasonable security margins. In general, if the resonator is
monolithically integrated and due to its nonstandard design, the design rule check-
ing (DRC) tools will notice the designer with a great amount of warning and error
messages. It is a nontrivial task of the design team to distinguish between involun-
tary mistakes and design-driven, purposely induced rule violations.
After completing the physical design, the IC is fabricated and integrated with
the resonator. Please note that the previous steps of the left-sided branch of the flow
diagram were concerned with resonator prototyping and not necessarily with the
final device. Thus, we can proceed with final-resonator fabrication, and, if the inte-
gration is monolithic, this is performed simultaneously with the IC fabrication,
306 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

given the process sequence. Otherwise, wire-bonding, flip-chip, and MCM, packag-
ing or related hybridization techniques are implemented now to complete the inte-
gration. Testing and characterization of the integrated system provide enough
feedback to determine the feasibility of our integration process and the application
by itself. Corrective design and technological actions are performed here, in case of
malfunctioning or poor system performance. The diagram of Figure 10.1 shows
only an iteration of the resonator-IC integration process, although various itera-
tions are normally required to achieve satisfactory results. This methodology is
implemented in the following sections to explain the FBAR oscillator and the
Si-MEMS resonator with integrated read-out circuit. First, we focus on the detailed
fabrication process, technology compatibility, and characterization results of an
FBAR.

10.2 Case I: Compatibility of FBAR and Silicon Technologies

In Chapter 4 we studied different FBAR fabrication processes, putting special

emphasis on the aluminum nitride (AlN) layer deposition technology and material
characterization. At the same time, the compatibility issues and the actions to solve
them were explained. We also discussed some inspection techniques, like scanning-
electron-microscope (SEM), interferometer and confocal microscopy, and their
application to structural analysis of fabricated devices. In this section, the technol-
ogy compatibility tests are explained. Then, we go through step-by-step descrip-
tions of two bulk micromachining-based FBAR processes. Although the fabrication
sequences are specific for the exemplified devices, they are suitable reference guides
of generic MEMS resonator and FBAR processes.

10.2.1 Compatibility Testing

10.2.1.1 AlN and Metal Electrode Compatibility
The FBAR studied herein is made of AlN with Pt/Ti electrodes, fabricated on silicon
substrate, and bulk micromachined by two technological options: front-side reac-
tive ion etching and back-side anisotropic wet etching. Thus, we begin the study by
reviewing the compatibility testing required to pattern the AlN in order to preserve
the electrodes, and vice versa. For this reason, the study involves different metals
and AlN etchants. Once selected Pt/Ti as the FBAR metallization, the objective is
patterning the electrodes with no damage to the AlN. The same considerations for
preserving the FBAR structure are taken into account for the micromachining
processes.
The study starts with the AlN etching techniques. AlN is patterned by wet etch-
ing, after immersion of wafers on certain etchant solutions. Different chemical solu-
tions are commercially available to accomplish this purpose. In considering the
Si/SiO2/metal-bottom/AlN/metal-top structure of the FBAR, AlN patterning has to
be completed with no damage to the bottom-electrode metal and the SiO2 layers
(i.e., the etchant has to be highly selective to AlN). The experiments required to test
the AlN selectivity consist of taking wafer samples with a different bottom-elec-
trode metal composition and photoresist configuration, and etching them with dif-
10.2 Case I: Compatibility of FBAR and Silicon Technologies 307

ferent chemical solutions. In this way, a matrix with metals in the rows and etchants
in the columns is built, like that shown in Table 10.1. The goal of these experiments
is to find the metal-resist combinations able to survive the etching process. In the
present case study, three commercial etchants are employed: KOH, TMAH, and
OPD-4262, which is a TMAH-like solution. The bottom electrode metallization is
made of Al, Ti, Pt, Au, Ni, and combinations of these materials. In all the cases, Ti is
used as adhesive metal for the structural metals (Pt, Au, or Ni) (candidate substrates
for AlN deposition), and the AlN has a thickness of 1,000 nm. Table 10.1 shows
that the OPD-4262 and Pt/Ti combination ensures reasonable etching times, with
no damage to the Pt/Ti layers. Additionally, the 1,000-nm lateral etching is negligi-
ble in comparison to the size of the FBAR electrodes (tens of micrometers). Etching
rates and etch-stop times are determined after visual inspection and profile
measurements of the sample.

Table 10.1 Compatibility Matrix for AlN Etching

1
Metal Electrode OPD-4262 KOH TMAH
(Bottom) (With/Without Resist) (With/Without Resist) (With/Without Resist)
Al Time: 30 minutes, with KOH attacks both TMAH attacks both
(2,000Å) intervals (3 minutes) resist and metal resist and metal
(mask, resist) Result: Bad (metal
attacked, resist preserved)
Ti+Pt Time: 2 hours, Time: 1 hour, with —
(300Å, 2.000Å) with intervals intervals
(mask, resist) Result: OK (no resist Result: Bad (resist is
lifting, no metal attacked, metal lifted off)
etching, no lifting)
Ti Time: 2 hours, with Time: 2 hours, —
(2.000Å) intervals with intervals
(no mask, resist) Result: OK Result: Bad (resist is
slowly attacked, metal
lifted off)
Ti+Pt+AlN Time: 5/10 minutes, Time: 1 minute Time: 1+1 minutes (AlN)
(300Å, 1,000Å, with 1-minute intervals Result: OK (metal Result: OK (metal
5,000Å) Result: OK. (AlN is preserved, AlN etched, preserved, AlN etched,
(no mask, no resist) etched, 1,000 Å/min, 5.000Å/min) 2,500 Å/min)
500 Å/min, respectively)
Ti+Au Time: 1 hour, with Time: 1 hour, with —
(300Å, 500Å) intervals intervals
(mask, resist) Result: OK Result: Bad (resist is
etched, metal “blows up”)
Ti+Au Time: 1 hour, with Time: 1 hour, with Time: 1 hour, with
(300Å, 500Å) intervals intervals intervals
(no mask, no resist) Result: OK Result: OK Result: OK
Ti+Ni Time: 1 hour, with Time: 1 hour, with —
(300Å, 500Å) intervals intervals
(mask, resist) Result: OK Result: BAD (resist is
etched, metal lifted off)
Ti+Ni Time: 1 hour, with Time: 1 hour, with Time: 1 hour, with
(300Å, 500Å) intervals intervals intervals
(no mask, no resist) Result: OK Result: OK Result: OK
308 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

Two samples are observed in the micrographs of Figure 10.2, before and after
AlN etching with OPD-4262. In the sample of Figure 10.2(a) the AlN is deposited
on Al, and we can see how the metallization has disappeared after the etching (right
side). The Al layer (the brightest in the image) remains only in those areas where the
AlN protects it (see how the connecting lines and testing pads have disappeared).
The second sample consists of AlN deposited on Ti/Pt, and Figure 10.2(b) shows it
after the same etching conditions of the previous sample. This time, the connecting
pads and the coplanar transmission line are preserved.
From the previous tests, we discarded Al as suitable bottom-electrode metal
because it was attacked by the tested etchants. Next, we examine the top electrode
compatibility with the AlN layer and the bottom electrode, the analyses being per-

(a)

(b)
Figure 10.2 AlN compatibility testing results using OPD-4262 etchant: (a) AlN deposited on Al,
before (left) and after (right) etching (the Al layer in white has disappeared after etching); and (b)
AlN deposited on Pt/Ti, which remains after AlN etching.
10.2 Case I: Compatibility of FBAR and Silicon Technologies 309

formed with Pt, Au, and Ni metallization (all of them implementing Ti as the adhe-
sive layer). In these tests, the metals are patterned through lift-off after immersion of
the samples in acetone. Additional ultrasound shaking helps increasing the efficacy
of the resist removal. Table 10.2 shows that, again, the Pt/Ti combination achieved
the best lift-off results with no damage to the previous layers.
It is worth mentioning that, in spite of the good results obtained in this study,
many other techniques and materials can also be employed to achieve similar per-
formance. Electrodes made of tungsten (W) have also been tested to be compatible
with AlN processing [1]. The compatibility of AlN with Al-, Cu-, and Mo-made
electrodes patterned through lift-off techniques have been demonstrated as well [2].
The third topic we review in this section is the selectivity of the Si etching pro-
cess. Let’s consider two bulk micromachining processes: RIE-based and
KOH-based anisotropic etchings. A short description of both processes is provided
in Section 4.2.4. Previously, it was noted that the FBAR structure is supported on a
thin SiO2 layer to isolate the metal electrodes and the substrate.

10.2.1.3 RIE-Based Si Etching Compatibility

Front-side RIE-based etching of the Si substrate requires a mask to open the SiO2
covering the wafer. The only exceptions are areas where the etching window is
opened. Also, we know that the etching selectivity between Si and SiO2 is about 1:6.
Extensive testing has demonstrated that insignificant etchings of AlN or the metal
layers occur for long exposures to etching plasma, yet the SiO2 layer may suffer
from important etching. For example, etching of 30 μm of Si leads to 5 μm of
attacked SiO2. Typically, passivation SiO2 has a thickness less than 1 μm, which
means that etching is completed in a few minutes before resonator releasing. Since
the metal electrodes of the FBAR are deposited on top of SiO2, high under-etching of
Si conducts to instability of the oxide and metal layers. Also, the SiO2 passivation
loses its functionality due to full etching along the noncovered areas. The situation
can be monitored by SEM analysis, as shown in Figure 10.3(a). The figure shows
the aspect of the test structure after RIE (the layer configuration of the structure is

Table 10.2 Compatibility Testing for

Metal-Layer Patterning
Metal Electrode Lift-Off (Only Wafers
(Bottom) with Resist, Acetone)
Ti+Pt Time: 10 minutes
(300Å, 2.000 Å) Result: OK
(mask, resist)
Ti+Au Time: 10 minutes
(300Å, 500Å) Result: Fair-poor.
(mask, resist) Resist lifted-off for
small patterns (max.
50%), and for bigger
ones (~50%)
Ti+Ni Time: 10 minutes, with
(300Å,500Å) intervals
(mask, resist) Result: Bad (metal did
not lift off)
310 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

(a)

(b)
Figure 10.3 Under-etching test of a metal-on-oxide structure attacked through front-side RIE: (a)
SEM image showing that the metal in gray bends after Si and SiO2 etching; and (b) confocal image
of the same structure (the black borders are the bent metal regions).

Si/SiO2/metal). Due to lateral under-etching of both the Si substrate and the SiO2
layer, the metal layer (in light) loses its support, thus bending and being released
from the substrate (the gray and dark-gray areas are “flying” metal with no SiO2
support). In the confocal image of Figure 10.3(b), we observe the same structure.
Here, the smooth gray regions are the metal, the rough darker-gray regions are
etched Si substrate, and the black contour is the (bent) metal with no underlying
SiO2. Longer etching times widen the bent metal contour and eventually complete
destruction of the electrode structure.
At least there are two ways for reaching compatibility of the Si RIE and the SiO2
layer. The trivial one, but not feasible, is to reduce the etching time of Si. The second
option is to protect the wafer with a material selective to the plasma recipe. Then,
the wafer is etched, and, after etching completion, the cushion material is removed.
If such a protective layer can be deposited and properly patterned to provide a win-
dow to the Si substrate, the etching time can be arbitrarily long (from the selectivity
point of view).
10.2 Case I: Compatibility of FBAR and Silicon Technologies 311

If we use the same mask of the etching window, and a protective Al/Cu layer,
for example, the previous goals can be accomplished. Figure 10.4(a) shows a sample
FBAR and its coplanar access line with the brightest Al/Cu cushion covering the
ensemble. The dark area around the device is the Si substrate seen through the etch-
ing window and the corrugated region around the window is the under-etched SiO2
area. Once the RIE finishes, the Al/Cu cushion is removed by a wet-etching process.
The micrograph of Figure 10.4(b) shows full-fabricated FBARs at the end of the
process: the darkest gray is the Si substrate at the bottom (several tens of microme-
ters), the dark gray is SiO2 with no Si underneath it, and the clear gray is SiO2 sup-
ported on nonetched Si. Some traces of the Al/Cu (bright) still remain on the
under-etched SiO2 region. The detailed, step-by-step description of this process is
carried out in Section 10.2.2.
The previous RIE process may be modified through deep RIE (DRIE) process-
ing. When implemented from the back side of wafers, DRIE of Si offers two advan-
tages: (1) under-etching of SiO2 is avoided, and (2) the Al/Cu cushion is no longer

(a)

(b)
Figure 10.4 Compatibility of the Si etching and protection of SiO2: (a) Al/Cu cushion on the SiO2
to avoid its etching (the etched region in dark black, and the laterally under-etched SiO2 in rough
gray); and (b) the FBAR and the coplanar transmission line after the Al/Cu cushion removal.
312 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

needed to protect SiO2 or structural layers of components. The drawback of this

approach is that etching of the full wafer is required to release the resonators, thus
increasing fragility.

10.2.1.4 KOH-Based Wet-Etching Compatibility

Anisotropic wet etching of Si has been used for a long time in MEMS and FBAR fab-
rication, mainly based on KOH etchants. The KOH wet etching is, in a certain way,
an auto-controlled process because the process stops when a specific angle in the
crystallographic structure of Si is reached. This advantage can be exploited if proper
mask design is carried out, to make the desired pattern size coincident with the stop-
ping angle. Selectivity of KOH to other materials also helps to control the etching.
For a given substrate thickness, KOH concentration, and appropriate mask design,
rough calculations of the etching times are enough for compatibly design the
process.
In the KOH process, the front-side Al/Cu cushion layer is replaced by a
back-side mask protecting selected regions of the Si substrate from undesired etch-
ing. The schematic drawing of Figure 10.5(a) depicts the cross-section of the pro-
cess, detailing the back-side mask, the stopping front-side SiO2 layer, and the
etching angle. The back-side mask may be a silicon nitride, Si3N4, for example. The
micrograph of Figure 10.5(b) shows an implementation of this process, the round
device being observed at the end of the wafer (500-mm depth). The SiO2 layer has a
thickness of 4,000Å and it avoids contact of the KOH with the structural layers of
the FBAR, thus preventing them from being etched. The squared shape of the
1,800-Å Si3N4 mask is also observed in the picture, where the gray area around the
etched volume is the mask. Due to the etching angle, the base area of the so-formed
truncated pyramid is bigger than the top area at the end (where the device is
located). The Si walls are the dark gray regions. In Section 10.2.3, we examine in
detail an implementation of the back-side FBAR process.

10.2.2 Front-Side, Reactive-Ion-Etching-Based Process

Our fabrication diary begins with the case of RIE-micromachined FBARs. Since the
resonators can be implemented in several ways and with an infinite variety of equip-
ment and processes, the manufacturing conditions discussed herein are exclusively a
reference guide of current-art FBAR processing. To structure the diary, we have
divided it in five main stages: (1) wafer preparation and front-side etching window
opening, (2) bottom electrode, (3) AlN deposition and patterning, (4) top electrode,
and (5) micromachining (RIE). The fabrication sequences of each stage are illus-
trated by corresponding Figures 10.6 through 10.10. The global process number is
indicated below each drawing, regardless of the stage depicted in the figure, though.

10.2.2.1 Wafer Preparation and Front-Side Etching Window Opening

We start the process by preparing the wafers for clean room processing. Commer-
cial silicon wafers of diameter 100 mm and thickness of 500 μm can be employed.
Although not necessary, it is desirable to prepare high-resistivity substrates able to
10.2 Case I: Compatibility of FBAR and Silicon Technologies 313

Electrode

AIN
Electrode

Air gap

(a)

(b)
Figure 10.5 (a, b) Compatibility of the Si wet-etching process: cut-view of the technology (the
SiO2 layer and the crystallography of the Si prevent further etching, and the Si3N4 mask protects
the Si substrate).

Figure 10.6 Front-side RIE-based process, wafer preparation, and front-side etching window
opening: (1) thermal oxidation; (2) photolithography; (3) oxide etching; and (4) resist removal.
314 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

Figure 10.7 Front-side RIE-based process, bottom electrode: (5) photolithography; (6) metal
deposition; and (7) resist removal and metal lift-off.

Figure 10.8 Front-side RIE-based process, AlN deposition, and patterning: (8) AlN deposition; (9)
photolithography; (10) AlN etching; and (11) resist removal and cleaning.

Figure 10.9 Front-side RIE-based process, top electrode: (12) photolithography; (13) metal
deposition; and (14) resist removal and metal lift-off.

Figure 10.10 Front-side RIE-based process, bulk micromachining: (15) Al/Cu cushion deposition;
(16) photolithography; (17) Al/Cu etching; (18) resist removal; (19) RIE-assisted Si
micromachining; and (20) Al/Cu cushion removal. Process completed.
10.2 Case I: Compatibility of FBAR and Silicon Technologies 315

reduce the RF losses. Once inside the clean room, standard cleaning on all the
wafers is performed with cascade and in-track rinsing. The cleaning can be carried
out in three steps: (1) 10–15 minutes of cleansing in sulfuric acid H2SO4 and H2O2
solution (2:1), (2) immersion in de-ionized water H2O for another 5 minutes, and
(3) additional 10-minute cleansing in de-ionized H20. Drying of the wafers in heater
at 110–125ºC completes the preparation.
The wafer preparation process for front-side etching is depicted in the sequence
of Figure 10.6. First, a thermal SiO2 layer of 400 nm is grown on both surfaces of
the wafer (1). This is done in an oven at 1,100ºC, alternating dry and wet environ-
ments. Typically, it starts with a short exposition to oxygen (O2) flow for about 10
minutes. Next, the flow changes to H2+O2 for the longer time of about one hour. To
finish the oxidation, O2 flows again in the chamber for a short time (another 10
minutes, for example). The flow is controlled to be between 3–6 sccm. In this pro-
cess, only the front-side oxide is useful (to isolate the electrodes and the substrate),
so the back-side oxide can be removed up to the designer decision.
Next, in step (2), a photolithographic step is performed with a mask corre-
sponding to etching window patterns. Prior to the photoresist coating, the SiO2 sur-
face is usually treated with hexa-methyl-disilazane (HMDS) vapor (20 to 30
seconds) to improve the adhesion between the wafer and the resist. With the aid of a
spinner, the resist is coated on the wafer, with typical thickness of 1–2 μm. Soft-bak-
ing is normally performed in hot plate at 100ºC and contact times of 20–30 seconds.
Then, the wafer alignment and exposure to UV are carried out. For contact-mode
masks, the resist exposition is about 10 seconds or less and the developing times
between 20 and 30 seconds. Additional hard baking in hot plate at 115ºC for about
30 seconds ameliorates the shape and consistence of the resist.
The exposed areas of SiO2 are now etched using dry-etching techniques, step
(3). For a 400-nm-thick layer, the typical etching times go from half an hour to a
couple of hours. Once the etching is completed, the remaining resist is removed by
two means: acid solution and dry etching (4). After a first acid-base resist removal,
an in-depth cleaning of the wafer is performed in oxygen plasma equipment. Typi-
cal operation ranges are 500-W power, 1-mbar pressure or less, and 10% oxygen
flow, for about 30–45 minutes.

10.2.2.2 Bottom Electrode

Patterning of the metal electrode can be done through metal etching or lift-off. In
this case study, metal lift-off for electrode patterning is implemented, with the
sequence of Figure 10.7 depicting the bottom electrode fabrication. First, the photo-
lithography defining the electrode shape is performed, with resist characteristics
and processing similar to those referred to in the previous stage (5). Thicknesses of 2
μm and soft baking at 100–115ºC are enough to obtain a resist suitable for lift-off.
Then, the metallic thin film is deposited on the wafer (6); for example, a 150-nm Pt
is sputtered by RF/magnetron sputtering (a 30-nm adhesive Ti layer is deposited
prior to Pt). By using argon (Ar) gas and a Pt target, typical deposition times are
about 20–30 minutes when the flow of the gas is 50 sccm, the RF power is 100W,
and the chamber pressure while sputtering is 110−2 mbar. The wafer is then
immersed in acetone to perform the metal lift-off (7). To aid this process, ultra-
316 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

sound shaking is combined with the acetone to increase the efficacy and the speed.
The frequency of the ultrasonic signal can be set to variable or sweeping mode.
Evaluation of this step is done through visual inspection of the wafer. Further
cleansing of the wafer may be carried out prior to the next step.

10.2.2.3 AlN Deposition and Patterning

The AlN deposition is the key technology of this process. Reactive sputtering and
epitaxial growth are the most popular techniques. In this example, reactive magne-
tron/RF sputtering is implemented in the sequence of Figure 10.8. The
1,000-nm-thick AlN layer is deposited on the wafer surface (8). Having an Ar-N2
atmosphere (50%–50%) and Al target, the deposition takes place with a rate of 6–7
−2
nm per minute when the RF power is 1 kW and the chamber pressure is 1 × 10
mbar. The same resist and setup of step (5) can also be implemented in the photo-
lithography of the AlN (9). Then, wet etching of AlN is done with the OPD-4262
etchant (10). For the thickness of 1,000 nm, the estimated etching rate is between
33–66 nm per minute. Finally, the resist and possible residues are removed through
O2-plasma-based cleaning of the wafers (11).

10.2.2.4 Top Electrode

The top electrode processing sequence is essentially the same as the bottom elec-
trode, and it is depicted in Figure 10.9: resist coating and photolithography with the
top electrode mask (12), sputtered-Pt deposition (13), and resist etching and metal
lift-off (14). Improved cleaning of the wafer may be performed by O2 plasma
etching.

10.2.2.5 FBAR Micromachining (RIE)

Once the structural layers of the FBAR are fabricated, RIE-based micromachining is
the last, and perhaps the most critical, stage of the process. In Section 10.2.1 a com-
patible process with the low Si-to-SiO2 RIE selectivity was introduced. Now, we
describe the steps required for the device releasing, the sequence illustrated in Figure
10.10. First, the wafer is covered with the Al/Cu cushion layer (15). A thickness of
500 nm is enough to protect the oxide during the RIE. The photolithography to pat-
tern the Al/Cu can thus be performed using the same mask of the etching window
opening (which coincides with the exposed SiO2 surface) (16). Etching of the Al/Cu
defines the protected SiO2 area (17), and the remaining resist on the Al/Cu can be
removed by using the previously described plasma techniques (18). The sample is
now introduced in the RIE chamber, setting up an SF6+O2 atmosphere with Ar
buffer at 75 mTorr pressure, and 100W of RF power, which etches Si with a 3:1 to
4:1 anisotropy ratio (19). FBARs are typically sized between 50 and 100 μm, so the
required lateral etching is half these values. The last step of the process is the
removal of the Al/Cu (20), which can be done in selective dry etching or wet etching,
according to the available technology. At this point, the FBAR is completely
released and the fabrication is completed.
10.2 Case I: Compatibility of FBAR and Silicon Technologies 317

10.2.3 Back-Side Wet-Etching Process

The second part of our fabrication diary is concerned with the back-side
KOH-based wet-etching process of FBARs. At the structural level of the FBAR, the
process is essentially the same as that of the RIE-processed case. However, the wafer
and etching preparation and the etching process stages substantially differ from
those presented in the previous section. Again, the diary structure is divided into five
stages: (1) wafer preparation and back-side etching window opening, (2) bottom
electrode, (3) AlN deposition and patterning, (4) top electrode, and (5)
micromachining (KOH). Since most of the processes involved in the wet-etched
FBAR fabrication are the same as described in the previous case, we will focus on
the etching preparation and the wet etching in itself.

10.2.3.1 Wafer Preparation and Back-Side Etching Window Opening

The cleaning and preparation steps are the same as described in Section 10.2.2.1. In
this case, however, it is useful to implement thinner wafers to reduce the etching
times. For this reason, 300-μm-thick wafers are usually implemented in back side
etching processes. The first step in the fabrication sequence of Figure 10.11 depicts
the growing of thermal SiO2 with a thickness of 400 nm (1), which is employed as
sacrificial material to open a region in the back side of the wafer for Si etching.
Next, a silicon nitride, Si3N4, for example, is deposited on the back-side of the
wafer (2). This nitride will be used as a mask during the Si etching due to its high
selectivity to KOH.
Next, in step (3), the first photolithographic step is performed with a mask cor-
responding to the membrane area. By considering the wafer thickness and the
KOH-driven etching angle of the wafer, one calculates the size of patterns (54.7 deg
between the {100} and {111} planes, or 35.3° respect to the normal). Thus, the
back-side membrane size (which designed in the mask) is projected through the
300-μm-thick wafer to fit the front-side membrane size. In step (4), the noncovered
nitride is attacked through plasma etching. Then, an acid solution and posterior O2
plasma etching remove the resist and remaining particles (5). The O2 plasma also
etches the thermal oxide on the front side and the noncovered areas of the wafer’s
back side (6). The wafer is now prepared for deposition of the FBAR structural lay-
ers and for the back-side etching steps performed at the end of the process.

Figure 10.11 Back-side KOH-based process, wafer preparation, and back-side etching window
opening: (1) thermal oxidation; (2) nitride deposition (back-side); (3)photolithography; (4) nitride
etching; (5) resist removal; and (6) oxide etching.
318 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

10.2.3.2 Bottom Electrode

The same techniques described in Section 10.2.2.2 are used to fabricate the bottom
electrode, whose process sequence depicts Figure 10.12. First of all, a low-pressure
plasma-enhanced chemical vapor-deposited (LPECVD) oxide layer is deposited on
the front-side surface of the wafer (7). In this way, we achieve better electrode-sub-
strate isolation and, at the same time, provide an effective barrier to the wet etching.
This means that, as a difference to the previous process, the SiO2 membrane will
support the FBAR, instead of being suspended only by its structural layers. Next,
one performs the photolithography defining the Pt electrode (8), the Pt layer deposi-
tion (9), and the lift-off of the resist and the on-resist metal (10). The Pt thickness is
again 150 nm. Further cleansing may be required prior to performing further
processing of the wafer.

10.2.3.3 AlN Deposition and Patterning

The AlN deposition and patterning sequences are identical to those presented in
Section 10.2.2.3, as depicted in Figure 10.13. The 1,000-nm-thick AlN layer is
deposited on the wafer surface (11). Then, the photolithography is performed (12),
and the AlN is etched with OPD-4262 to pattern the resonator shape (13). Finally,

Figure 10.12 Back-side KOH-based process, bottom electrode: (7) LPCV-deposited oxide; (8)
photolithography; (9) metal deposition; and (10) resist removal and metal lift-off.

Figure 10.13 Back-side KOH-based process, AlN deposition, and patterning: (11) AlN deposition;
(12) photolithography; (13) AlN etching; and (14) resist removal and cleaning.
10.3 Case II: High-Level Design of a Temperature-Compensated (TC) Oscillator 319

O2-plasma-based cleaning removes the resist and possible residues of the wafers
(14).

10.2.3.4 Top Electrode

The top electrode processing sequence is essentially the same as for the bottom elec-
trode shown in previous sections. The sequence is depicted in the schematic draw-
ings of Figure 10.14: resist coating and photolithography with the top electrode
mask (15), sputtered-Pt deposition (16), and resist etching and metal lift-off (17). In
addition, the wafer may be deeply cleaned with in O 2 plasma.

10.2.3.5 FBAR Micromachining (KOH)

Last, but not least, the KOH-assisted back-side micromachining of the wafer is car-
ried out, as depicted in step (18) of the sequence in Figure 10.15. As previously com-
mented, the angle between the back-side plane and the normal is 54.7°, and the
etching stops just at the oxide-silicon interface. Additionally, we can remove the
back-side Si3N4 and SiO2, if desired. At this point, the FBAR is released, and the pro-
cess is completed.

10.3 Case II: High-Level Design of a Temperature-Compensated (TC)

Oscillator

Resonator-based crystal oscillators (RXO) are one of the most appealing applica-
tions of FBAR and MEMS resonators. They are a subject of study and industrial
production, as we extensively reviewed in Chapter 9. In spite of the high Q factor of
such microresonators and the availability of different integration technologies with
current IC processes, RXOs (as well as virtually any oscillator) suffer frequency

Figure 10.14 Back-side KOH-based process, top electrode: (15) photolithography; (16) metal
deposition; and (17) resist removal and metal lift-off.

Figure 10.15 Back-side KOH-based process, micromachining: (18) KOH-based wet-etching of Si;
and (19) back-side nitride and oxide removal (optional).
320 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

instability from aging- and temperature-related issues. Both the IC and the resona-
tor contribute to the frequency drifting of the RXO. Already in Chapter 6, we ana-
lyzed the temperature coefficient factor of FBARs and studied a process-based
compensation strategy.
In this section, we will study the case of a temperature-compensated FBAR
oscillator. First, we introduce the impact of temperature on the frequency stabil-
ity, the phase noise, and the application requirements of RXOs. Next, we focus on
current-art FBAR-oscillator implementations and their phase noise performance.
With this at hand, both technology-level and system-level solutions for the RXO
frequency drifting are proposed. In the present study, we assume that the tempera-
ture coefficient factor (TCF) of the FBAR can be tailored at the process level. Thus,
the analysis will be oriented to reduce the TCF of the IC-plus-FBAR ensemble. For
this purpose, codesign of the RXO is carried out, by performing a Verilog-A model
of the temperature-compensated FBAR and the circuit-level modeling of the ensem-
ble. The compensation strategies and their implementations are discussed at the end
of the section.

10.3.1 The Temperature Stability Issue in Oscillators

As previously stated in Chapter 6, oscillators are closed-loop circuits with positive
feedback. If the Barkhausen criteria are met, the oscillation is sustained without the
need for input signal. An RXO is constituted by two components: the resonator and
the amplification circuit. The components are interconnected to be part of the oscil-
lator loop to attain enough gain and phase shift, imposed by the Barkhausen condi-
tions (0 dB and 360°, respectively). The resonance frequency mainly determines the
RXO frequency of the crystal, which can be a quartz plate, a MEMS resonator, a
SAW, or an FBAR. Small variations of this frequency do not normally jeopardize
the Barkhausen conditions, because the RXO is designed with adequate gain and
phase margins. However, since the performance of an oscillator depends on its fre-
quency stability, one may intuit that the frequency variations should be controlled
to guarantee operation within the application limits.
The oscillator frequency stability is quantified by defining its phase noise. The
phase noise is the amount of fluctuations in the phase of the oscillation, which leads
to frequency deviation of the ideal pure-sinusoidal oscillator behavior. The phase
noise adds to the ideal sinusoidal wave so as the oscillation power is spread to
adjacent frequencies:

a(t ) = A 0 sin(2 πf 0 + θ(t )) (10.1)

where f0 is the central frequency of the oscillation and θ(t) is the phase noise. Thus,
the phase noise modulates the ideal oscillation frequency in a random way. The
severity depends on the temperature fluctuations, among many factors. The phase
noise density is measured in dBc/Hz at various offsets referencing the central fre-
quency. It is assumed that, typically, the phase noise includes both low-frequency
flicker noise and broadband white noise, so it has a monotonically descending side-
band shape in the frequency domain, as depicted in Figure 10.16. Ultimately, the
phase noise is responsible for limiting the signal-to-noise ratio and the proper
demodulation of RF signals. Thus, the phase noise has to be kept as low as possible.
10.3 Case II: High-Level Design of a Temperature-Compensated (TC) Oscillator 321

−100

−120

Noise (dBc) v/sqrt (Hz)

−140

−160

−180

−200
1k 10k 100k 1M 10M 100M
Relative frequency (Hz)
Figure 10.16 Frequency-domain representation of phase noise.

The oscillator phase noise depends to a great extent on the resonator’s tempera-
ture coefficient factor (TCF). Thus, if we are able to reduce the TCF, the oscillator
phase noise will be improved [3]. In Chapter 6 we already defined the TCF of a reso-
nator, and some temperature compensation strategies to control the resonance fre-
quency of FBARs were discussed. At the process level, the mainstream strategy is
adding a compensation layer made of a material with a TCF of magnitude opposite
to that of the active layer. In AlN-made FBARs, a properly designed SiO2 layer can
accomplish this task to reduce the TCF to values as low as −1.5 ppm/ºC, which are
pretty competitive to the performance of quartz [4, 5].
At the system level, various circuit topologies and frequency control strategies
exist to compensate for the temperature-dependent phase noise. They include ana-
log, digital, and mixed-signal techniques. Analog techniques address temperature
compensation by building a current or voltage reference that is independent of the
power supply and fabrication process. Also, such a reference exhibits a well-known
temperature behavior. The temperature dependence may assume one of three
forms: (1) proportional to absolute temperature (PTAT), (2) constant
transconductance gm behavior, and (3) temperature independent [6]. In this way,
analog compensation circuits may implement bandgap and PTAT voltage genera-
tors [7–10], varactors or capacitor banks [11, 12], or built-in heating of the resona-
tor for constant-temperature operation [13], among many others.
Digital compensation is carried out by an EPROM memory and a resistive tem-
perature sensor. Thus, the sensor converts the resistance temperature value to a pro-
portional voltage, which is then converted to a digital signal. This signal is
subsequently used to address the EPROM containing the calibrated control voltage
necessary for compensating the measured temperature. M. A. Taslakov imple-
mented this method to temperature compensation of a 1-GHz surface transverse
wave (STW)–based oscillator with a temperature precision of ±1.25 ppm from
−30ºC to +60ºC [14]. Mixed compensation techniques attempt to combine the good
repeatability and low phase noise of analog compensation and the simpler structure
and stability of digital techniques (e.g., the double compensated TCXO presented
by Zhou et al. [15]).
322 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

10.3.2 Low Phase Noise FBAR-Based Oscillators

To date, FBAR oscillator implementations with low phase-noise performance have
been demonstrated. The high Q factor of the FBAR and the phase noise perfor-
mance of the oscillator in gigahertz bands make them attractive to 3G applications.
Although attractive, the phase noise can still be reduced by modifying the FBAR
process, the circuit design, or both. To illustrate the impact of the resonator TCF
reduction in phase noise, let’s analyze the 5-GHz RXO implemented by Zhang et al.
[16]. The noncompensated FBAR performing a −60 ppm/ºC TCF was modified to
feature a SiO2 temperature compensation layer. In this way, the compensated TCF
was reduced to −8.7 ppm/ºC, thus achieving a phase noise of −109.5 dBc/Hz at 100
kHz.
The balanced Colpitts configuration proposed by Aissi et al. succeeds in halving
the phase noise of the oscillator to a value of −121 dBc/Hz at 100 kHz from
5.46-GHz carrier frequency [17]. In this case, the FBAR does not implement any
temperature compensation layer. A third oscillator designed by Chee et al. drasti-
cally reduces the phase noise by combining a 1.9-GHz FBAR oscillator with an
injection locked power oscillator [18]. Prior to locking, the power oscillator shows a
poor phase noise of −98 dBc/Hz at 100 kHz. When the power and the FBAR oscilla-
tors are locked in, the phase noise is improved to −120 dBc/Hz at the offset fre-
quency of 100 kHz. The phase noise performances of the circuit before and after
injection locking are shown in the plot of Figure 10.17.
The previous examples are just samples of the several ways conceived to
improve the oscillator performance. Of course, many other techniques are also suit-
able for reducing the FBAR oscillator phase noise, like the varactor-controlled
low-TCF circuit. Once the FBAR is optimized to attain low-TCF, further actions
can be taken to reduce the oscillator’s TCF. Thus, codesigning the FBAR and the cir-
cuit can lead us to very low TCF values, as we analyze in the following section. We
carry out this task by modeling the temperature and frequency responses of the
FBAR, the oscillator, and the varactors.

Ref −60 dBc/Hz −98.69 dBc/Hz

10.00
dB/

100 kHz Frequency offset 10 MHz

Figure 10.17 Phase noise performance of the injection locking oscillator transmitter: a phase
noise improvement higher than 20 dBc/Hz at 100 kHz is observed after locking of the power and
FBAR oscillators. (© 2006 IEEE [18].)
10.3 Case II: High-Level Design of a Temperature-Compensated (TC) Oscillator 323

10.3.3 Codesign of an FBAR-Based TC Oscillator

Codesign of an electronic system is a working philosophy understanding that the
different components of the circuit have to be designed with regard to global system
performance. Instead of their individual optimization, codesign studies the interac-
tion among components modifying the response of the system [19]. Individual opti-
mization is thus nonuseful or, in general, it only leads to local variable optimization.
Although many methodologies for electronic system codesign exist, all of them
make use of cosimulation of the circuit components. This is done through specifi-
cally created system description languages, like Verilog-A, SystemC, or
VHDL-AMS, among others [20–23]. In recent years, the integration of MEMS and
FBARs with ICs make these methodologies necessary to optimum system design.
In our case, codesigning the FBAR and the circuit can lead us to very low TCF
value. Of course, we can optimize the circuit without considering the TCF of the
FBAR. At the process level, we can reduce the FBAR’s TCF by implementing tem-
perature compensation layers as described in Chapter 6. None of these solutions
imply that the integrated system will perform the lowest possible TCF. For this pur-
pose, we first model the temperature and frequency responses of the FBAR, the
oscillator, and the varactors. Then, we integrate them and see how the integrated
system responds to temperature, so we can determine the design values suitable for
TCF reduction.
First, it is useful to get previous knowledge of the oscillator architecture and the
circuit-level operating principles. The Figure 10.18(a) depicts the schematic circuit
of a Pierce oscillator. Transistor M1 is the oscillator’s core providing enough
transconductance gain to sustain the oscillation, IBIAS supplies the bias current of the
circuit including the M1 drain-to-source current IDS, C1 and C2 modulate the ampli-
fication gain, and the crystal is an FBAR that fixes the oscillation frequency through
the feedback network between the drain and gate of M1. C1 and C2 are external,

I BIAS

I DS C dd
C gg M1
C1 C2

(a)

(b)
Figure 10.18 Pierce oscillator: (a) schematic circuit with the FBAR between gate and drain of M1;
and (b) small-signal representation.
324 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

monolithic, or the input and output capacitances of M1. In a lesser way, they also
serve as frequency pulling parameters. The crystal resonator is implemented with an
FBAR, which we will model by the MBVD equivalent-circuit representation. IBIAS is
implemented by a CMOS current source, and the DC biasing of M1 is fixed by plac-
ing a feedback resistance R3 between gate and drain. The small-signal oscillator cir-
cuit is shown in Figure 10.18(b). A more realistic circuit should include the gate and
drain capacitances Cgg and Cdd. They may be comparable to C1 and C2 for a big-area
transistor M1 with high transconductance gain gm.
The Barkhausen conditions are examined by breaking the loop at the output of
the voltage-controlled current source (gmv1) and introducing a test current i across
R2 and C2. In this way, the open-loop gain T(s) must be higher than unity, and the
phase shift introduced by the FBAR must be equal to −180° to obtain a total phase
shift of 360°:

g m v1
T( s) = − (10.2)
i

g m R1 R 2
T( s) = (10.3)
Z c ( s)(1 + R1 C1 s)(1 + R 2 C 2 s) + R1 (1 + R 2 C 2 s) + R 2 (1 + R1 C1 s)

Equivalently, the real component of the circuit impedance (Zc) must be negative
and higher in module than the equivalent parallel impedance formed by the FBAR
and the biasing resistance R3:

1 ⎛ 1 ⎞
Z( s) = R 3 ⎜R m + L m s + ⎟ (10.4)
C0 s ⎝ C m s⎠

Z( s) = − Z c (10.5)

Note that the parasitic impedance due to FBAR-to-IC interconnection is

excluded from the analysis. Depending on the integration strategy, they can be sig-
nificant, and they should be considered for realistic results. In a first approximation,
though, we also neglect R3 and the output resistance R2 to find the values of the real
and imaginary components of the circuit impedance ZC:

g m C1 C 2
Re( Z c ) = − (10.6)
( g mC 0 ) + ω 2 (C1 C 2 + C 2 C 0 + C1 C 0 )
2 2

g m2 C 0 + 2ω 2 (C1 + C 2 )(C1 C 2 + C 2 C 3 + C1 C 3 )
Im( Z c ) = − (10.7)
[
ω ( g m C 0 ) + ω 20 (C1 C 2 + C 2 C 3 + C1 C 3 )
2
]
From the previous equations, we see how the circuit’s negative resistance can be
larger than that of the FBAR by controlling C1, C2, and gm, thus allowing the oscilla-
tion to start up. As we said before, Cgg and Cdd may be included in the analysis. Actu-
ally, they replace C1 and C2 in most circuits because of their large size. In summary,
10.3 Case II: High-Level Design of a Temperature-Compensated (TC) Oscillator 325

we have three temperature-dependent components controlling the oscillation

frequency:

1. Transistor M1, through gm, Cgg, and Cdd, all of them also dependent of IDS,
which is also affected by temperature variations;
2. External capacitors C1, C2 (monolithic or discrete components);
3. FBAR modeled through the motional impedance elements Lm, Rm, and Cm.

Now, let’s state the codesign strategies of the temperature-compensated oscilla-

tor, assuming that the circuit has been designed to meet the Barkhausen conditions
(i.e., that the gm, IDS, C1, and C2 values are enough to sustain the oscillation given an
FBAR model):

1. Controlling the equivalent-circuit parameters of the FBAR Lm, Cm, and C0:
Although not of a physical meaning, thus not controllable by circuit-design
techniques, the temperature dependence of Lm, Cm, or C0 can be modeled
through a system-level language. In this way, the FBAR behavioral model is
integrated into the circuit-level model of the oscillator to obtain the
FBAR-plus-IC temperature response.
2. Controlling the Cgg/Cdd ratio: It assumes that by controlling the IDS value, the
values of Cgg and Cdd can be changed to contribute compensating the
temperature variations. This is done by parametric analysis of the IBIAS
current source.
3. Controlling the C1/C2 ratio: The oscillation frequency depends on the total
circuit’s reactance at that frequency. The temperature variations modify the
reactance, and the oscillation frequency. The C1/C2 ratio is modified to com-
pensate the modeled Lm/Cm temperature variations.

First, we need an FBAR description suitable for integration with Spice or Spec-
tre models used in the circuit-level design environment. The parameters of such a
description are the equivalent-circuit elements and TCF of the FBAR and the tem-
perature. According to the examples of Chapter 6, where resonators with linear
TCF were studied, the mathematical model provided by (6.2) to (6.4) gives us the
basis for building the behavioral model. We reproduce here (6.2) to (6.4) for
convenience of the description:

( (
f 0Ti = f 0ref 1 − TCF × Tref − Ti )) (10.8)

1
f0 = (10.9)
2π LmC m

Lref
LTim = m
(10.10)
(1 − TCF × (T ))
2
ref − Ti

where f0 is the resonance frequency, f0Ti is the resonance frequency at temperature

Ti, f0ref is the resonance frequency at the reference temperature Tref, Lm and Cm are
326 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

the motional inductance and capacitance of the MBVD FBAR model, LmTi is the Lm
value at temperature Ti, and Lmref is the Lm value at temperature Tref. This tempera-
ture-dependent model of Lm can be implemented by different means, as the follow-
ing Verilog-A script:

‘include “constants.vams”
‘include “disciplines.vams”
module Lm_T(vp, vn) ;
inout vp, vn ;
electrical vp, vn ;
parameter real Lm = 0 ; //Lm: reference motional inductance
parameter real Temp = 40 ; //Temp: reference temperature (ºC)
parameter real TCF = −5 ; //TCF: temperature coefficient factor
analog
V(vp, vn) <+ ddt(Lm*I(vp, vn)) /((1 − TCF/1e6 *(40 − Temp))*(1-TCF/1e6 *(40 −
Temp)));
Endmodule

According to the Verilog sintaxis, Lm_T is a two-terminal module with

input-output voltage nodes vp and vn representing LmTi. Next, the initial values of
the model parameters are defined, and the voltage-to-current relationship of the
module is described. Thus, we see that the voltage across LmTi is a function of the
derivative instruction ddt of the current, the reference Lm, the reference temperature
Temp, and the TCF. This model can then be instantiated in the circuit design envi-
ronment (e.g., Cadence Virtuoso Schematic). The circuit representation of the
model and other circuit components is depicted in Figure 10.19(a), and the Cadence
dialog box for object property edition of the inductor in Figure 10.19(b). The fre-
quency response in parts per million (ppm) against the temperature is similar to that
already presented in Figure 6.9(b), although with a TCF = −5 ppm/ºC (according to
the previous script). By modifying the TCF parameter, the process-based tempera-
ture compensation of the FBAR can be simulated. The resonator model can now be
integrated with the other circuit components to analyze their combined temperature
performance.
Before the FBAR-IC integration, the circuit is simulated to check its own TCF
value. This will give us an idea of the amount of compensation due to the IC we will
need to introduce in the system. To carry out this simulation, we replace the
Verilog-A model of the FBAR by ideal RLC counterparts with no performance vari-
ations due to temperature. Also, we are interested on studying the variations of the
transconductance gm. The curves of Figures 10.20(a) and 10.20(b) show the temper-
ature-due variations of gm and the oscillation frequency, respectively. The circuit
was simulated using the Spectre models of the AMS035C3B4 CMOS technology for
the +20ºC to +80ºC range, and, according to the figures, it has a negative TCF. Ana-
lyzing the curves of gm, its TCF equals –3,125 ppm/ºC, which is very high. However,
if we look at the behavior of the oscillation frequency, its TCF has the very low
value of –0.5 ppm/ºC. This would mean that the variations of gm are absorbed or
compensated by the other circuit components.
Now, we connect the FBAR(Tº) and the IC(Tº) models. For given values of
Temp, TCF, and Lm, the available Spice or Spectre models in the CAD environment
10.3 Case II: High-Level Design of a Temperature-Compensated (TC) Oscillator 327

Rm Cm
L m (T°)

Ls Rs

C ox
Rp C0

C sub Rsub

(a)

(b)
Figure 10.19 Behavioral implementation of an FBAR in the Cadence environment: (a) schematic
Ti
circuit of the MBVD model of the FBAR with the Verilog-A model of the motional inductance Lm
(in the oval); and (b) edit dialog box for setting the initial values of the model.

perform the open-loop and close-loop circuit analysis. Let’s remember that, previ-
ously, we used TCF values equal to –5 ppm/ºC for the FBAR, and obtained an IC
TCF of –0.5 ppm/ºC. The TCF of the integrated system is carried out by sweeping
the temperature range of 20–80ºC in both the Spectre modeling setup and the
FBAR’s edit dialog box. The results are shown in Figure 10.21, where the obtained
TCF is –6.5 ppm/ºC. We conclude that the TCF of the FBAR and the IC are added,
but resulting in a higher value than the sum of their individual TCFs.
Then, we implement our compensation strategies to test their suitability. First,
we analyze the Cgg/Cdd ratio. From the previous analysis, we find out that the
drain-source current IDS is also sensitive to the temperature variations, as the points
in the plot of Figure 10.22 demonstrate. Thus, it is impossible to control the Cgg/Cdd
328 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

(a)

(b)

Figure 10.20 Circuit-level simulation of the Pierce oscillator TCF (ideal MBVD elements in the
FBAR): (a) transconductance gm; and (b) oscillation frequency (ppm).

values due to IDS via IBIAS. For this reason, we discard the implementation of this
strategy.
We consider now controlling the C1/C2 ratio. In a first stage of the analysis, we
assume that we are able to change the value of C1, C2, or both. This can be done by
implementing digitally controlled capacitor banks, or integrated MEMS or discrete
varactors. To modify C1, for example, we assume a tuning range relative to the val-
ues of Cgg. Thus, we implement a first bank tunable up to a maximum C1 value equal
to the 50% of Cgg. In the second implementation, C1 is tuned in the range of the
100% of Cgg. Results in the plots of Figure 10.23 compare these implementations
and the noncompensated circuit. A TCF reduction up to –1.8 ppm/ºC was obtained
with this approach.
 329

600

500

400

300

200

100

0
−20 −10 0 10 20 30 40 50 60 70 80
−100

−200 2
R = 0.9974

−300
Temperature (°C)

Figure 10.21 Thermal response of the integrated FBAR-IC oscillator (no temperature compensa-
tion): the system TCF is –6.5 ppm/ºC.

Figure 10.22 Thermal response of the drain-to-source current IDS of transistor M1.

In conclusion, one can say that the latter strategy will be useful to reduce the
TCF of the circuit. However, many challenges have to be faced, most of them
related to the implementation of the variable C1-C2 capacitor set. Tuning range and
quality factor of current-art varactors has to be kept in mind. Nevertheless, we have
learned that both the FBAR and the IC have to be analyzed in a joint way to obtain
more realistic results. Codesign methodologies like that introduced in these para-
graphs are useful to accomplish this purpose.
330 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

Figure 10.23 Thermal response of the integrated FBAR-IC oscillator implementing the C1/C2 ratio
compensation strategy: the system TCF is reduced up to –1.8 ppm/ºC (full range of C1 equals the
Cgg value).

10.4 Case III: Read-Out Circuit Design of a 434-MHz MEMS Resonator

Integration of MEMS resonators with microelectronic circuits often requires previ-

ous characterization and parameter extraction of the resonator, as we examined in
the previous case. These tasks provide the designer with enough information for
dimensioning the IC’s large and small signal specifications, like biasing, required
gain level, and bandwidth. However, equivalent-circuit models of small MEMS
devices may be hardly extracted due to very low output currents, high insertion
losses, and noise levels. In some cases, the engineer only has at hand rough parame-
ter estimations based on simulations to complete the circuit-level design. A second
issue is the impedance matching between the characterization instrument and the
resonator.
To overcome these issues, read-out circuits are integrated with MEMS to
accomplish two purposes: (1) signal amplification of the MEMS output current to
reach the minimum signal levels to de-embed the resonator model; and (2) imped-
ance matching with the characterization equipment. Broadly, Figure 10.24 repre-

On-chip MEMS+CMOS set

MEMS
~ A B Instrument

VAC

VBias

MEMS Amplifier Buffer

Figure 10.24 MEMS resonator and read-out circuit integration scheme: the output MEMS cur-
rent is amplified and impedance-matched to the instrument input port.
10.4 Case III: Read-Out Circuit Design of a 434-MHz MEMS Resonator 331

sents the scheme of MEMS resonator and read-out circuit integration. The drawing
assumes electrostatic actuation of the resonator with an AC voltage VAC, which is
biased through the DC voltage VBIAS. The time-varying current iAC is delivered to the
amplifier with gain A for increasing and conditioning the signal level, and the buffer
provides impedance matching to the characterization instrument. The dotted line
represents the on-chip MEMS and CMOS read-out circuit set. Typically, the char-
acterization setup for this chip includes a signal generator with sweeping capability
to excite the MEMS and a spectrum analyzer or a network analyzer to perform both
functions directly.
Several circuit architectures and measurement techniques are applicable to per-
form characterization of a MEMS resonator. In this case study, we focus on a linear-
mode circuit able to develop broadband characterization of the frequency response
of an electrostatically actuated MEMS resonator. The specification and design of
the read-out amplifier includes the amplifier architecture, values for passive compo-
nents, output impedances, dynamic range, bandwidth, and gain range, among other
parameters.

10.4.1 MEMS-CMOS Integration Technology

Better comprehension of the read-out circuit is reached if we first understand the
MEMS-CMOS integration technology and physical layout. Monolithic integration
of the MEMS and CMOS circuit reduces the parasitic effects introduced by
wire-bonding or discrete integration. In the present case, the integration is imple-
mented by designing the MEMS with the Poly1 and Poly2 layers of the
AMS035c3b4 CMOS technology [24]. The MEMS is released after a wet etching
postprocess. The cross section of the AMS device chip after CMOS process comple-
tion will look like the schema of Figure 10.25(a). As observed, the passivation layer
PROT is opened during the CMOS process to allow access to the upper IMD layer.

(a) (b)

(c) (d)
Field IMD MET1 poly2 poly1 N-well PROT
oxide
Resist

Figure 10.25 (a–d) MEMS resonator postprocess implementation using the AMS035C3B4 CMOS
process. (Courtesy of Zachary J. Davis, DTU Nanotech, Copenhagen, Denmark.)
332 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

Since the Poly1 and Poly2 layers are patterned using the AMS mask layers, we only
need to release the structures. First, a thick layer of photo-resist is deposited and a
window opened directly over the MEMS area, as shown in Figure 10.25(b). Then,
HF-based wet-etching releases the resonator, attacking the SiO2-made layers, the
field oxide, and the IMD in Figure 10.25(c). In the last step, the protection resist is
removed and the processing is finished, as depicted in Figure 10.25(d).
The driving and read-out electrodes of the MEMS current are implemented
with the first metal layer MET1, as suggested in Figures 10.25. Because of the
monolithic integration, the current flows to the IC through MET1. In this way, the
read-out circuit architecture has to be designed to match this current and make
proper signal conditioning, as we discuss in the next sections.

10.4.2 Read-Out Circuit Specification and Circuit Architecture

The main function of read-out circuits is amplification of the MEMS resonator out-
put current. We expect this current to be a resonant signal produced at a specific
mode shape in the hundreds of megahertz range and to be very weak, with AC
amplitudes in the order of nano-amperes. Regarding these matters, a set of require-
ments for the read-out amplifier is listed in Table 10.3.
By using an RLC-circuit-based electromechanical model of the resonator [25],
one can estimate the dynamic range of the read-out amplifier. As the MEMS device
is designed to resonate in a narrow frequency range, the input-current mode of the
amplifier couples to the current flowing out of the MEMS. In general, the circuit
and the MEMS present high impedance mismatching. Additionally, it is highly
desirable that the circuit has output impedance near 50Ω, given that this value cor-
responds to the input-port impedance of the network analyzer used in the MEMS
characterization. This will maximize the matching and minimize the losses. In order
to characterize the amplifier, linearity is important. However, due to the expected
high Q shape of the MEMS current, distortion would be negligible in a very narrow
bandwidth, making the linearity issue irrelevant. Other aspects concerning maxi-

Table 10.3 Read-Out Amplifier Specifications

Issue Requirement
Dynamic range Wide, running from
hundreds of nA to
units of μA
Bandwidth Cutoff frequency in
the order of 400 MHz
Power Not restricted
Size Not restricted
IC technology AMS 0.35 μm CMOS
Linearity Yes
Input mode Current mode
Feedback configuration Open/closed loop
MEMS integration Impedance matching
not required
Output matching Near 50Ω
10.4 Case III: Read-Out Circuit Design of a 434-MHz MEMS Resonator 333

mum size or power consumption can be neglected initially, only a selection of

CMOS as the target technology is required.
Different circuit architectures are applicable to perform characterization of
MEMS resonators, such as transimpedance amplifier (TIA), current comparator,
current amplifier, and low-noise amplifier. In our case, let’s choose the TIA archi-
tecture as the target configuration to perform read-out of the MEMS resonator.
Figure 10.26 depicts a TIA amplifying the MEMS current, followed by further
amplification and buffering stages. The TIA architecture takes the input current sig-
nal and amplifies it through an impedance to generate an output voltage signal.
Because of this, the TIA is also referred to as current-to-voltage converter.
Typically, the gain of the current-to-voltage converter Vout/Iin is not high enough
to meet the circuit specifications, and depending on the IC technology it assures a
higher bandwidth than a voltage-mode amplifier of the same gain. In this sense,
additional amplifying stages are usually cascaded to the TIA to achieve a desired
Vfin/Iout gain specification. Finally, buffering is required to match the output circuit
and input instrumentation impedances. If the instrument is a network analyzer with
50Ω input impedance, the output buffer impedance has to be close to this value. The
buffer gain VRF/Vfin approximates to unity, although high input-output mismatching
leads to lower gains. Additionally, external matching networks could also be inte-
grated to the setup in order to refine the impedance matching. Thus, the total AC
transimpedance gain VRF /Iin equals the cascaded gain of each stage:

VRF V Vfin VRF

= out ⋅ ⋅ (10.11)
I in I in Vout Vfin

10.4.3 Read-Out Implementation and Characterization

The implementation requirements for a typical TIA are a high bandwidth,
high transimpedance gain, adequate power consumption, low noise, low input
impedance, and small area. Improved Wilson current mirrors perform the
current-to-voltage conversion. The mirrors connect to resistive transistor loads to
attain transimpedance gain of 60 dBΩ. Its output node then connects to the DC cou-
pling and biasing stage, which is required for achieving two main objectives. First,
one isolates the voltage amplifier from the biasing deviations of the TIA. This is

Current-to- Voltage Buffer

iin voltage Vout Vfin VRF
amplifier
converter

Vout Vfin VRF

VAC ~1
iin Vout Vfin

Figure 10.26 Transimpedance amplifier: current-to-voltage converter followed by voltage ampli-

fier and buffer.
334 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

done by DC blocking at the output node of this stage. The second objective is to pro-
vide DC biasing to the subsequent voltage amplifier. Further voltage amplification
was conceived to assure sufficient voltage gain to the circuit. This block is imple-
mented as a three-stage, nondifferential, resistive load amplifier configuration add-
ing a 20-dB voltage gain.
The output buffer is implemented with transistors connected in a feedback loop
to reduce the output impedance to the range of 50Ω. When in comparison to tradi-
tional source follower topology, this configuration is capable of achieving design
values with smaller transistor sizes. Nevertheless, a reduction on the global gain
happens due to the high mismatching between the input and output impedances of
the buffer, which leads to a buffer voltage gain lower than unity. In the end, the
overall transimpedance gain of the read-out circuit is 70 dBΩ. For a detailed discus-
sion of circuit design implementation and characterization, see [26].
Traditionally, TIAs with high bandwidth are built with GaAS, InP, or SiGe
technologies. As CMOS processes were downscaled to submicron gate lengths, new
TIA designs appeared to offer higher bandwidths. On the other hand, the maximum
power supply is also reduced to prevent the field oxide breakdown, although the
threshold voltage still needs reduction [27–30]. Since the megahertz-bandwidth
requirements of the read-out circuit are within the current CMOS capabilities, we
choose this technology for implementing the circuit. Figure 10.27 shows the foot-
print of the CMOS MEMS-IC chip with the composing blocks of the system being
signaled in the picture. AC characterization of the IC is performed by a 50Ω net-
work analyzer and DC power supplies.

10.5 Summary

In this chapter, we have studied three implementation cases of FBAR and MEMS
resonators. In the first one, we reviewed the detailed fabrication processes
of bulk-micromachined FBARs. The second case addressed the design of a
temperature-compensated FBAR-based oscillator. The third one dealt with the
design of a MEMS-resonator read-out circuit.

I/O
pins
DC
coupling
Primary
current cell Buffer

Voltage
amplifier
Current-to-voltage
converter
Current
sources

I/O
pins
Figure 10.27 Die photograph of the integrated MEMS-to-read-out circuit. (© 2005 SPIE [29].)
10.5 Summary 335

From these examples, we have learned that MEMS-based system design is a

multidisciplinary area that requires understanding of different fields, from process-
ing and fabrication technologies to integrated circuit implementation. Codesign
methodologies are useful to integrate the efforts of the engineering team in the pur-
pose of building a complete MEMS-CMOS system.

References

[1] Akiyama, M., et al., “Influence of Metal Electrodes on Crystal Orientation of Aluminum
Nitride Thin Films,” Vacuum, Vol. 74, 2004, pp. 699–703.
[2] Lee, J.-B., et al., “Effects of Bottom Electrodes on the Orientation of AlN Films and the Fre-
quency Responses of Resonators in AlN-Based FBARs,” Thin Solid Films, Vol. 447–448,
2004, pp. 610–614.
[3] Gribaldo, S., et al., “Experimental Study of Phase Noise in FBAR Resonators,” IEEE
Trans. on Ultrasonics, Ferroelectrics, Freq. Control, Vol. 53, 2006, pp. 1982–1986.
[4] Vanhelmont, F., et al., “A 2 GHz Reference Oscillator Incorporating a Temperature Com-
pensated BAW Resonator,” Proc. IEEE Intl. Ultrason. Symp. 2006, Vancouver, BC, Octo-
ber 3–6, 2006, pp. 333–336.
[5] Pang, W., et al., “A Temperature-Stable Film Bulk Acoustic Wave Oscillator,” IEEE Elec-
tron Dev. Lett., Vol. 29, 2008, pp. 315–318.
[6] Razhavi, B., Design of Analog CMOS Integrated Circuits, New York: McGraw-Hill, 2001.
[7] Tanzawa, T., et al., “A 2.4-GHz Temperature-Compensated CMOS LC-VCO for Low Fre-
quency Drift Low-Power Direct-Modulation GFSK Transmitters,” IEICE Trans. Electron.,
Vol. E88-C, 2005, pp. 490–495.
[8] Stadius, K., et al., “A Monolithic Temperature Compensated 1.6 GHz VCO,” Proc. GAAS
Conference 1998, GAAS98, Amsterdam, the Netherlands, October 5–6, 1998,
pp. 626–630.
[9] Ho, G. K., et al., “Temperature Compensated IBAR Reference Oscillators,” Proc. IEEE
Intl. Conf. MEMS 2006, Istanbul, Turkey, January 22–26, 2006, pp. 737–741.
[10] Sundaresan, K., et al., “Electronically Temperature Compensated Silicon Bulk Acoustic
Resonator Reference Oscillators,” IEEE J. Solid-State Circuits, Vol. 42, 2007,
pp. 1425–1434.
[11] Schodowski, S., A New Approach to a High Stability Temperature Compensation Crystal
Oscillator, Schaumburg, IL: Motorola Communications Systems Division, 1970.
[12] Kinsman, R. G., Temperature Compensation of Crystals with Parabolic Temperature Coef-
ficients, Schaumburg, IL: Motorola Communications Systems Division, 1978.
[13] Jha, C. M., et al., “In-Chip Device-Layer Thermal Isolation of MEMS Resonator for Lower
Power Budget,” Proc. ASME International Mechanical Engineering Congress and Exposi-
tion IMECE2006, Chicago, IL, November 5–10, 2006, pp. 1–7.
[14] Taslakov, M. A., “1 GHz STW Based Oscillator with Continuous Temperature Compensa-
tion,” IEEE Trans. on Ultrason. Ferroelec. Freq. Control, Vol. 45, 1998, pp. 192–195.
[15] Zhou, W., et al., “Comparison Among Precision Temperature Compensated Crystal Oscil-
lators,” Proc. IEEE Intl. Frequency Control Symposium 2005, Vancouver, BC, August
29–21, 2005, pp. 575–579.
[16] Zhang, H., et al., “5GHz Low Phase-Noise Oscillator Based FBAR with Low TCF,” Proc.
13th International Conference on Solid-State Sensors, Actuators and Microsystems Trans-
ducers 2005, Seoul, Korea, June 5–9, 2005, pp. 1100–1101.
[17] Aissi, M., et al., “A 5 GHz Above-IC FBAR Low Phase Noise Balanced Oscillator,” Proc.
IEEE Radio Frequency Integrated Circuits Symposium RFIC-2006, Vol. 1, San Francisco,
CA, June 11–13, 2006, pp. 25–28.
336 Case Studies: Modeling, Design, and Fabrication of FBAR and MEMS-Based Systems

[18] Chee, Y. H., A. M. Niknejad, and J. M. Rabaey, “An Ultra-Low-Power Injection Locked
Transmitter for Wireless Sensor Networks,” IEEE J. Solid State Circuits, Vol. 41, 2006,
pp. 1740–1748.
[19] Churchman, C. W., The Systems Approach, New York: Delacorte Press, 1968.
[20] IEEE Std. 1364-2001, IEEE Standard Verilog Hardware Description Language, New
York: Institute of Electrical and Electronics Engineers, 2001.
[21] IEEE Std. 1666-2005, IEEE Standard SystemC Language Reference Manual, New York:
Institute of Electrical and Electronics Engineers, 2005.
[22] Bakalar, C. E., “VHDL-AMS: A Hardware Description Language for Analog and
Mixed-Signal Applications,” Proc. IEEE Trans. on Circuits Systems II: Analog Digital Sig-
nal, Vol. 46, 1999, pp. 1263–1272.
[23] IEEE Std. 1076.1-1999, VHSIC Hardware Description Language Analog And Mixed-Sig-
nal, New York: Institute of Electrical and Electronics Engineers, 1999.
[24] Austria Microsystems 2-poly 4-metal 0.35 μm CMOS technology, http://www.
austriamicrosystems.com.
[25] Nguyen, C. T.-C., and R. T. Howe, “An Integrated CMOS Micromechanical Resonator
High-Q Oscillator,” IEEE J. Solid-State Circuits, Vol. 34, 1999, pp. 440–455.
[26] Mohan, S. S., et al., “Bandwidth Extension in CMOS with Optimised On-Chip Inductors,”
IEEE J. of Solid-State Circuits, Vol. 35, 2000, pp. 346–355.
[27] Kossel, M., et al., “Wideband CMOS Transimpedance Amplifier,” Electronics Letters,
Vol. 39, 2003, pp. 587–588.
[28] Kromer, C., et al., “A Low Power 20 GHz 52 dBOhm Transimpedance Amplifier in 80 nm
CMOS,” IEEE J. Solid-State Circuits, Vol. 39, 2004, pp. 885–894.
[29] Campanella, H., et al., “Band-Pass Transimpedance Read Out Circuit for UHF MEMS
Resonator Applications,” Proc. SPIE Symposium on Microtechnologies for the New Mil-
lennium 2005, Vol. 5837, Sevilla, Spain, May 9–11, 2005, pp. 300–309.
[30] Fay, P., C. Caneau, and I. Adesia, “High-Speed MSM/HEMT and p-I-n/HEMPT Mono-
lithic Photo Receivers,” IEEE Trans. on Microwave Theory Tech., Vol. 50, 2002,
pp. 62–67.
About the Author
Humberto Campanella is a research fellow at the Consejo Superior de
Investigaciones Científicas (CSIC) assigned to the Instituto de Microelectrónica
de Barcelona IMB-CNM, Spain, and associate professor in the Department of
Telecommunications and System Engineering at the Universitat Autònoma de
Barcelona. He holds a B.Sc. in electronics engineering from the Pontificia
Universidad Javeriana, Bogotá, Colombia; an M.Sc. in telecommunication systems
from the Universidad Politécnica de Madrid, Spain; a Ph.D. in microelectronics
and automated systems from the Université de Montpellier, France; and another
Ph.D. in electronics engineering from the Universitat Autonoma de Barce-
lona, Spain. Dr. Campanella has more than 15 years of industry and academic expe-
rience in research, development, and engineering of integrated circuits,
microelectromechanical systems, telecommunications, and signal processing. He
has published several scientific papers and holds one patent on the heterogeneous
integration of FBAR and MEMS resonators with CMOS technologies. He has
served as a reviewer of indexed scientific journals and as a referee for public-funded
projects in the United States, Europe, and South America.

337
Index
1.9-GHz CMOS oscillator, 222 Acoustic sensors, 225
2-GHz longitudinal-mode-FBAR biosensor IDTs, 46, 48, 50, 200
system, 62 quartz, 29, 42, 46, 247–48
See also QCM, SAW, BAW
A Acoustic wave
Above-IC BAW cavity, 37, 40–41, 46–47, 50
FBAR, 287 definitions, 5, 38, 52–53, 246
filter, 222, 301 phase velocities, 38
process, 295 profile, 59
Acceleration propagation, 37–38, 40–45, 238, 246
axis, 249 theory, 37
mechanical, 88 Acquisition, zero-span frequency-domain, 167,
sensitivity, 249 241
Accelerometers Active-circuit applications
dual-beam, 250 definitions, 266, 285, 287, 289
micromachined, 260 See also Oscillators, mixers, mixlers, tuned
miniature, 248 LNAs, RF front ends
piezoelectric-based resonant, 249 Actuators, 11–12, 34, 66–68, 260–63
Acoustic Admittance, 99
cavity, 37 AFM (atomic force microscopy)
coupling, 55, 59, 213 analysis, 114
isolation, 54–57 applications, 252, 254, 260
layer, 37–38, 41, 52–53, 208–9 cantilevers, 151–53
loss, 55 definitions, 32, 56, 112–13, 151–53
paths, 78–79 detection cantilevers, 252
phase, 53 AFM probes
port, 78–79 chip, 254
properties, 5 fabrication technologies, 253
velocity, 55, 79, 142 tip, 153
viscosity, 85 Air gap, 55–59, 104, 108–9, 272
Acoustic impedance Al/Cu, 311, 316
characteristic, 80 cushion, 311
matching, 79 deposition, 314
normalized, 78 etching, 314
ratio of, 59 Algorithm
Acoustic resonators, 5, 37, 40, 78 equivalent-circuit parameter extraction, 149
applications, 277 least-mean-square-based, 158
manufacture, 41 AlN (aluminum nitride), 12, 26, 54, 103
microelectromechanical, 38 acoustic layer, 96
piezoelectric-based, 38 beams, 141
temperature-compensated, 188 c-axis-oriented, 143
thin-film, 41 compatibility, 308–9
vibration modes of, 44, 64 constants evaluation, 160
See also BAW, SAW, FBAR, SMR

339
340 Index

AlN (continued) reconfigurable, 282

crystal orientations, 115, 117 single, 271
crystallographic quality, 56, 112–14 Antiresonance, 84
crystals, 56, 107, 114–16 Applications
deposition, 106, 112, 314, 316–18 biological, 63
deposition and patterning, 316, 318 biosensing, 230
etchants, 306 chemical-detection, 240
etching, 112, 308, 314, 318 consumer-oriented, 187
fabrication, 106 convergent, 234, 260
FBAR fabrication, 56 force-detection, 21
layer, 78, 109–10, 169–71, 180 liquid-media, 239
metal electrode compatibility, 306 mass sensors, 260
patterning, 306 passive circuit, 269
process, 145 private mobile Internet, 265
quality, 116, 183 resonator-based, 133, 137
stiffness, 141 Ar (argon), 28, 92, 107, 111
surface roughness, 114 Architectures, transceiver, 269
AlN film Arrays
epitaxial, 146 8-resonator cantilever, 232
low-temperature-deposited, 117 coupled resonators, 287
sputtered, 118, 169 parallel-coupled resonators, 281
Aluminum, 56, 105 photodiode, 152, 235
Amplification, 285, 332–33 ASIC, 288–90
Amplifiers, 70–71, 265, 331–33 Aspect ratio, 18–19, 21–22, 33, 91–92
architecture, 331 Au (gold), 105, 213, 307, 309
broadband, 203
differential, 288 B
frequency-selective, 4 BAW (bulk acoustic wave)
LNA (low noise amplifiers), 266, 285, devices, 52, 54–55, 61, 64
292–93, 295–97, 333 resonators, 37, 46, 51–53, 55
lock-in, 252 wave propagation, 52
OTA (operational transconductance See also FBAR; SMR
amplifier), 188 Beam resonators, 17, 19–21, 141, 249–51
read-out circuit, 70–71, 303–4, 330–32, Bulk acoustic wave resonators. See BAW
334 Bulk micromachining, 12, 29–30, 56, 125
transfer functions, 70–71, 147, 276–77 See also Fabrication techniques
transimpedance amplifier (TIA), 188, (micromachining), KOH etching, RIE
333–34
C
Amplitude
modulation, 292 Cantilevers, 12–14, 17–21, 151–52, 230–32
resonator’s, 258 Capacitance
spectrum, 227 motional. See motional capacitance
vibration’s, 155 oxide (Cox), 86–87, 148–49, 151, 173
Anisotropic etching, 29, 110 substrate, parasitic (Csub), 86–87, 148
anisotropy ratio, 316 Capacitive sensing, 226–28, 283
etchant solutions, 29 Capacitive transducer electrodes, 278
RIE, 199 CCD (charge-coupled-device), 154
Annealing, 56, 144 Characterization techniques, 8, 112, 125,
selective laser, 204 127–28
ANSYS, 88–90, 96–97, 101, 141–42 AFM measurements, 113, 158
Antennas, 3, 265, 267, 273–76 standard tapping-mode, 151
port, 273, 275, 292, 296 electrical, 127–28, 132, 134, 153
electron-beam-induced, 151, 153, 155, 157
Index 341

impedance, 83–84, 130, 281–82, 284–85 DOF. See Finite element modeling (FEM)
low-frequency measurements, 128 DSP (digital signal processors), 2, 11
network analyzer, 131–34, 168, 174–75, DUV photolithography
331–34 193-nm argon fluoride, 29
network parameters, 115–16 248-nm krypton fluoride, 29
open-circuit condition, 128–30
PCB (printed circuit board), 133–34 E
probe station, 131–32, 143, 168, 174–75 Effective mass, 6, 9, 20, 230
short-circuit condition, 128–29 Electron-beam-induced characterization.
X-ray diffraction (XRD), 56, 112, 114–16, See Characterization techniques
126 Electroplating, 27, 106, 213, 218–19
See also Microwave techniques seed materials, 218–19
Clamped-clamped beam resonator, 140–42, Epitaxial growing. See Deposition techniques
291, 293 Equivalent-circuit models, 82–83, 85–87, 138,
CMOS 172–73
circuits, 71, 188, 196–97, 200 Equivalent-circuit parameters, 23–24, 74,
integration, 52, 56, 63, 187–88 146–48, 158
process, 187–88, 196–200, 209–210, 219 Etalon. See Interferometers
substrate, 195, 205–7, 209–10, 215–16 Etching. See Fabrication techniques
technologies, 187, 191–93, 196, 211 Etching rates, 74, 111, 118–21, 307
wafers, 196, 205–7, 213, 218–19 Etching windows, 109–10, 119–21, 123, 311
CMP (chemical mechanical polishing), 164, EUV (extreme UV), 29, 32
200–1 Excitation. See Transduction
CNT (carbon nanotube), 140, 235, 237
F
See also Nanotube resonators
Complementary metal oxide semiconductor, Fabrication techniques
1–2, 187–88, 193–95, 204–5 deposition, 31, 177–79, 209, 213
See also CMOS etching, 29–31, 199–200, 308–9, 311–13
CPW, 131–33 lift-off, 125, 309, 314–16, 318–19
Cr (chromium), 105–6, 118–19, 211 metallization, 105–6, 109, 200, 308
CRF (coupled resonator filters), 60 micromachining, 34, 54, 191–92, 195
crystal See also Bulk micromachining, surface
oscillators, 286–87, 300 micromachining, RIE; wet etching, dry
resonators, 83–84, 172, 244, 286 etching
CVD (chemical vapor deposition), 26–27, oxidation, 105, 315
105–6, 174, 176 passivation, 104, 110, 118, 197
See also Deposition techniques passivation layer, 121, 199–200, 209–12,
218–19
D patterning, 31, 108–10, 209–10, 314–18
Damping, 6, 8, 10, 239 photolithography, 111, 209, 313–14,
Deflection, 93, 95, 246–47, 251–52 316–19
Deformation, 42–43, 52–53, 93, 121 resist, 28–29, 307, 309, 315–19
Density, 55, 73, 227–28, 238–39 resist removal, 28, 309, 313–14, 317–19
Deposition. See Fabrication techniques See also Bulk micromachining, surface
Deposition techniques, 26–27, 55–56, 213 micromachining, RIE
epitaxial, 26, 103, 118, 208 Fabry-Pérot interferometers.
sputtering, 27, 56, 106, 111 See interferometers
See also Sputtering, CVD, PVD, IACVD FBAR (thin-film bulk acoustic wave
Detection. See Transduction resonators), 37, 54, 101, 286
Dielectric displacement (permittivity), 143 fabrication, 103, 105, 110–11, 218
Disk resonators, 18–19, 247, 280, 282 oscillators, 303, 306, 322
thickness-shear-mode, 247 resonance frequency, 57, 61, 153, 163–64
342 Index

FBAR (continued) polylithic integration, 200, 222

structure, 122–23, 207, 219, 306 strategies, 23, 187, 189, 191
See also BAW (bulk acoustic wave) technologies, 195–96, 204, 304, 319
FEM (finite element modeling), 23, 73, 88–91, wafer level transfer (WLT), 187, 204, 213
97–98 Interdigitated transducers. See Acoustic
boundary conditions, 20, 22, 43–45, 88–89 sensors, IDT
contour plots, 93, 97–98, 244 Interferometer, 103, 121, 125, 156
degrees of freedom, 88–91, 96–97 etalon, 156–57
harmonic analysis, 89, 92–94, 96–97, 99 Fabry-Pérot, 156–57
FIB (focused ion beam), 32, 174, 177, 185 Interferometry, 154–55
Filters interference bands, 121, 154–55
band-selection, 266, 268–69 zero-contrast bands, 155
ceramic, 270, 299 Isolation, 25, 273, 276, 283–84
coupled resonator, 60
design of MEMS, 280–81 L
FBAR, 137–38, 203, 293, 295–96 KOH, 110–11, 307, 312, 317
ladder filters, 60, 137–39, 270 etching, 56, 110, 312, 317–19
MEMS, 278–81 See also Bulk micromachining, TMAH
MEMS (coupled contour-mode), 63–64
L
out-of-band rejection, 61, 137, 139, 270
See also Radio frequency applications Lateral excitation. See Propagation modes
FoM (figure of merit), 144–45 Lateral extensional mode (LEM)
Frequency resolution, 226–27 LEM (lateral extensional mode), 35, 63–64, 68
Frequency response, 11, 50, 71, 278–81 See also Propagation modes
Frequency shifting, 176–77, 229, 238, 255–56 LNA (low noise amplifiers). See Amplifiers
Frequency stability, 163, 165, 288–89, 320 LoC (lab-on-a-chip), 2–3, 188
Frequency tuning, 172–73, 175, 177, 179–81 Longitudinal waves. See Propagation modes
FWHM (full-width-half-maximum), 114, Loss mechanisms
116–17 acoustic radiation, 9
air damping, 8–9
G substrate loss, 87
GF (gauge factor), 227–28 support damping, 8–10
surface-effect damping, 8, 10
H thermoelastic damping, 8, 10
Heterogeneous integration. See Integration Love waves. See Propagation modes

I M
IACVD (ion-assisted chemical vapor Magnetic field, 12, 14–15, 255, 257–59
deposition), 177, 179–80, 183 Magnetic sensors
See also Deposition techniques external field, 15–16, 256–57
IBAR, 188–89 Lorentz, 14–15, 257
IL (insertion losses), 50, 137–39, 270–71, Mass deposition
276–77 deposited mass, 176–77, 179–81, 183, 245
See also Radio frequency applications localized, 177, 180–81
Integration, 188, 194–95, 209–11, 303–6 uniform-film, 175–77, 180–82
CMOS, 187–88, 190, 192, 194 Mass loading, 62, 173–74, 176, 229
heterogeneous, 187–88, 194–95, 211–12, Mass responsivity, 175–77, 180–81, 237,
218–20 239–45
hybrid, 188, 190–91, 193–94, 202 Mass sensitivity, 62, 230–32, 234–35, 238–41
landing pads (heterogeneous), 204, 211–14, Mass sensors, 61, 229, 243, 260–61
218–19 nanomechanical mass spectrometer, 236
monolithic, 191–93, 200, 203, 220 MBVD (modified Butterworth-Van-Dyke), 85,
pillars, 205–7, 209, 213, 216 147, 150, 324
Index 343

Mechanical sensors, 225, 245–47, 249 Nanowires, 2, 18, 73–74

accelerometers, 3–4, 246, 248–51 Natural frequencies, 14, 17, 93
MEMS NBIC (nano-bio-info-cognitive), 2–3, 187
cantilever, 12, 15, 90, 93 NEMS resonators, 6, 18, 21–22, 156–58
devices, 3–4, 14, 17, 24 Network parameters. See Characterization
market, 4 techniques
processes, 265–66
sensors, 194, 229–30, 248, 255–56 O
structures, 10, 15, 191, 232 Oscillators
MEMS and NEMS, 1–4, 18, 33, 36 Barkhausen conditions, 320, 324–25
resonators, 4–5, 8–9, 33, 195–97 MEMS, 287, 290
technologies, 1–2, 4, 6, 8 oscillation frequency, 14, 136, 286, 325–26
MEMS resonators, 14–16, 18–19, 70–71, phase noise, 184, 241, 286–87, 320–22
330–31 VCMO (voltage-controlled MEMS
bridge-shaped, 192–93 oscillators), 289
comb-drive, 198–99 See also Radio frequency applications
Metallization. See Fabrication techniques OTA (operational transconductance amplifier).
Microcantilevers, 231–32 See Amplifier
Micromachining. See Fabrication techniques Out-of-band rejection. See Filters
Microwave techniques Oxidation. See Fabrication techniques
matched load, 131 Oxides, 25–26, 86, 111–12, 163
parasitic effects, 85–86, 165, 195, 305
P
S-parameters, 130–31, 147, 171, 182
transmission line, 82–83, 85–87, 129–31, Parallel-plate capacitance, 12
283–84 See also Static capacitance
See also Characterization techniques Passivation. See Fabrication techniques
Mixers. See Radio frequency applications Passive-circuit applications, 266, 269, 271,
Mode shaping, 18–19, 153, 180–81, 244 273
Modeling Patterning. See Fabrication techniques
behavioral model, 70–73, 171–72, 325 Permittivity, 42–43, 143–44, 146, 228
circuit representation, 78–79, 132–33, 148, Photolithography. See Fabrication techniques
174–75 Physical fundamentals
electromechanical transformer, 75–79, 81 MEMS and NEMS resonators, 5, 7, 9, 11
equivalent circuit, 74, 77–78, 82, 286 Piezoelectric constants, 43, 45, 139–41, 145
Mason’s model, 78–81, 238 Piezoelectric materials, 12, 42–43, 55, 85
MBVD (modified Butterworth-Van-Dyke), 85, See also AlN, ZnO, PZT
147, 150, 324 Piezoelectricity, 37, 42–43, 45, 65
parameter extraction, 23, 82, 147–48, Power consumption, 226, 255, 259, 333
303–4 ppm/ºC, 61, 169–70, 226, 326–30
stages of resonator design and, 69, 71, 73 Pressure sensors
Motional definitions, 221, 246–47
capacitance, 78, 83, 228, 284 Quartzdyne’s, 247
inductance, 77–78, 148–49, 172, 325–26 sensitivity, 248
Multiplexers. See Radio frequency applications Production cycle
design equations, 21–23, 227
N flow diagram, 22, 148–49, 304–5
Nanofabrication techniques, 22, 31–32 of MEMS and NEMS resonators, 21, 23, 69
Nanomechanical mass spectrometer. See Mass Propagation modes
sensors lateral excitation, 45
Nanomechanical resonators, 35, 173, 184, 234 lateral extensional mode, 35, 63–64, 68
Nanotube resonator, 237 longitudinal, 38–41, 44–46, 52–53, 97–98
Nanotubes, 18, 158, 237 Love wave, 38–39
344 Index

Propagation modes (continued) parallel (FBAR), 85, 87, 148–49, 151

Rayleigh waves, 38–40, 46, 64 series (FBAR), 85, 87, 128–29, 149
shear-mode waves, 38–40, 42, 54, 239 Resonance
frequency shifting, 230, 246–47
PVD (physical vapor deposition), 27, 106, parallel (antiresonance), 84, 135–36,
174, 176 144–45
See also Deposition techniques series, 135, 137, 240–41
PZT, 12, 46, 54, 103 Resonance modes, 1, 5–6, 18–20, 180
See also Piezoelectric materials Resonant LC tank, 82–83, 267
Resonator design, 9, 19, 47, 69
Q Resonator fabrication, post-CMOS, 191, 193,
QCMs (quartz crystal microbalances), 62, 168, 197, 288
238–40, 242 Resonators
See also Sensor applications 434-MHz MEMS, 330–31, 333
Quality factor concept, 22–23, 304–5
damping mechanisms, 8 RF MEMS, 265–66, 268–69, 297–98
definitions, 5–6, 8, 23–24, 53 components, 265, 268
extraction, 133 market, 61, 268, 298
reduction, 163 switches, 268, 282–83
resonators, 7, 34 RIE (reactive-ion etching), 29–30, 125,
311–12, 316
R
chamber, 316
Radio frequency (RF), 3, 7, 265–66, 289–91 process, 109, 150, 165, 313–14
circuit design, 101, 158 See also Fabrication techniques
circuits, 85 (micromachining)
components, 42, 83 RXO (resonator crystal oscillator), 319–20
high-performance applications, 47
market, 64 S
sensors, 33 Sacrificial layer, 30–31, 103, 108–9, 206–9
sputtering deposition, 66, 126 See also Surface micromachining
Radio frequency applications, 51, 60, 265–66, SAW (surface acoustic wave), 37–38, 46–47,
268 49, 52
duplexers, 61–62, 268–69, 271, 273–75 resonators, 65
filters, 61, 138–40, 270–71, 273–82 SCF (stacked crystal filters), 60, 270
mixers, 5, 285–86, 289–92, 295 Sensor applications, 7, 11, 225–28, 260
multiplexers, 266, 268–69, 275–77, 293 biological (spores), 231–32, 260
PCS, 275–76, 296, 299 minimum detectable mass (MDM), 62–63,
quintplexers, 268, 275–76 239–40, 242–43
reference oscillators, 188–89, 200, 221, 335 resolution, 11, 29, 146, 225–27
RF power, 111, 120, 315–16 See also Mass sensors; mechanical sensors;
switches, 269, 282–83, 285 biological sensors; force sensors
triplexers, 266, 268, 271, 275 Shear-mode waves. See Propagation modes
UMTS (Universal Mobile SMR (solidly mounted resonator), 54–55,
Telecommunications System), 266, 59–61, 80–82, 144
268, 298 reflectors, 48, 200, 202, 248
very high frequencies (VHF), 85, 235 Solenoids (MEMS), 257–58
See also Filters; oscillators Solidly mounted resonators. See SMR
Rayleigh waves. See Propagation modes Spectrometer. See Mass sensors
Reactive ion etching. See RIE Spring constant, 6, 16, 20, 152
Read-out Sputtering
circuit design, 330–31, 333 AC, 27
electrodes, 12–14, 292, 332 AC/DC magnetron, 107
Resistance
Index 345

physical, 27–28 U
See also Deposition techniques Universal Mobile Telecommunications System
Static capacitance, 12 (UMTS), 266, 268, 298
See also Parallel-plate capacitance UV photolithography
Strain, 42–44, 227–28, 245–46, 248–50 365-nm i-line
Structural layers, 30–31, 140–41, 183, 316–18 436-nm g-line, 29
Surface micromachining, 29–31, 103, 191–92,
196 V
See also Fabrication techniques VCMO (voltage-controlled MEMS oscillators).
(micromachining) See Oscillators
T VHF (very high frequencies), 85, 235

TC (temperature compensation), 43, 169, 171, W

319, 321, 323, 329 Wafer level transfer (WLT). See Integration
TCFBAR, 169–72, 183 Wave propagation. See Propagation modes
TCF (thermal coefficient factor), 72–73, Waves. See Propagation modes, Rayleigh,
168–69, 171–73, 325–27 Love, SAW, longitudinal waves,
definition, 169, 171, 326–27 shear-mode waves, lateral waves
Thermoelastic damping. See Loss mechanisms Wet etching, 31, 56, 110–11, 316–18
Thin-film bulk acoustic wave resonators. See See also Fabrication techniques
FBAR (thin-film bulk acoustic wave (micromachining)
resonators)
Thin films, 105–6, 112–13, 143, 239–40 X
TMAH etching, 31, 108, 111, 307 X-ray diffraction (XRD). See Characterization
See also Bulk micromachining, KOH techniques
etching
Transduction Z
detection, 18, 231–32, 234–35, 251 ZnO (zinc oxide), 41–42, 54–55, 200, 247–48
excitation, 4, 11, 44, 292 See also Piezoelectric materials
Transimpedance amplifier (TIA). See
Amplifiers
 Recent Titles in the Artech House
Integrated Microsystems Series

Acoustic Wave and Electromechanical Resonators: Concept to Key Applications,

Humberto Campanella
Adaptive Cooling of Integrated Circuits Using Digital Microfluidics, Philip Y. Paik,
Krishnendu Chakrabarty, and Vamsee K. Pamula
Fundamentals and Applications of Microfluidics, Second Edition,
Nam-Trung Nguyen and Steven T. Wereley
Integrated Interconnect Technologies for 3D Nanoelectronic Systems,
Muhannad S. Bakir and James D. Meindl, editors
Introduction to Microelectromechanical (MEM) Microwave Systems,
Héctor J. De Los Santos
An Introduction to Microelectromechanical Systems Engineering, Nadim Maluf
MEMS Mechanical Sensors, Stephen Beeby et al.
Micro and Nano Manipulations for Biomedical Applications,
Tachung C. YihIlie Talpasanu
Microfluidics for Biotechnology, Second Edition, Jean Berthier and Pascal Silberzan
Organic and Inorganic Nanostructures, Alexei Nabok
Post-Processing Techniques for Integrated MEMS, Sherif Sedky
(
Pressure-Driven Microfluidics, Václav Tesar
RFID-Enabled Sensor Design and Applicatons, Amin Rida, Li Yang, and
Manos Tentzeris
RF MEMS Circuit Design for Wireless Communications, Héctor J. De Los Santos
Wafer-Level Testing and Test During Burn-in for Integrated Circuits,
Sudarshan Bahukudumbi Krishnendu Chakrabarty
Wireless Sensor Network, Nirupama Bulusu and Sanjay Jha

For further information on these and other Artech House titles, including
previously considered out-of-print books now available through our In-Print-
Forever ® (IPF ®) program, contact:
Artech House Artech House
685 Canton Street 16 Sussex Street
Norwood, MA 02062 London SW1V 4RW UK
Phone: 781-769-9750 Phone: +44 (0)20 7596-8750
Fax: 781-769-6334 Fax: +44 (0)20 7630-0166
e-mail: [email protected] e-mail: [email protected]

Find us on the World Wide Web at: www.artechhouse.com