CAPACITIVE CMOS READOUT CIRCUITS FOR HIGH PERFORMANCE

MEMS ACCELEROMETERS

A THESIS SUBMITTED TO
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES
OF
MIDDLE EAST TECHNICAL UNIVERSITY

BY

REHA KEPENEK

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR
THE DEGREE OF MASTER OF SCIENCE
IN
ELECTRICAL AND ELECTRONICS ENGINEERING

FEBRUARY 2008
 Approval of the thesis:

CAPACITIVE CMOS READOUT CIRCUITS FOR HIGH PERFORMANCE

MEMS ACCELEROMETERS

submitted by REHA KEPENEK in partial fulfillment of the requirements for the

degree of Master of Science in Electrical and Electronics Engineering
Department, Middle East Technical University by,

Prof. Dr. Canan Özgen

Dean, Graduate School of Natural and Applied Sciences

Prof. Dr. İsmet Erkmen

Head of Department, Electrical and Electronics Engineering

Asst. Prof. Dr. Haluk Külah

Supervisor, Electrical and Electronics Eng. Dept., METU

Prof. Dr. Tayfun Akın

Co-Supervisor, Electrical and Electronics Eng. Dept., METU

Examining Committee Members:

Prof. Dr. Murat Aşkar

Electrical and Electronics Engineering Dept., METU

Asst. Prof. Dr. Haluk Külah

Electrical and Electronics Engineering Dept., METU

Prof. Dr. Tayfun Akın

Electrical and Electronics Engineering Dept., METU

Prof. Dr. Cengiz Beşikçi

Electrical and Electronics Engineering Dept., METU

Dr. Said Emre Alper

Technical Vocational School of Higher Education, METU

Date: 08.02.2008

ii
I hereby declare that all information in this document has been obtained and
presented in accordance with academic rules and ethical conduct. I also
declare that, as required by these rules and conduct, I have fully cited and
referenced all material and results that are not original to this work.

Kepenek, Reha

iii
 ABSTRACT

CAPACITIVE CMOS READOUT CIRCUITS FOR HIGH

PERFORMANCE MEMS ACCELEROMETERS

Kepenek, Reha
M.Sc., Department of Electrical and Electronics Engineering
Supervisor: Asst. Prof. Haluk Kulah
Co-Supervisor: Prof. Dr. Tayfun Akın

February 2008, 144 pages

This thesis presents the development of high resolution, wide dynamic range sigma-
delta type readout circuits for capacitive MEMS accelerometers. Designed readout
circuit employs fully differential closed loop structure with digital output, achieving
high oversampling ratio and high resolution. The simulations of the readout circuit
together with the accelerometer sensor are performed using the models constructed
in Cadence and Matlab Simulink environments. The simulations verified the
stability and proper operation of the accelerometer system. The sigma-delta readout
circuit is implemented using XFab 0.6 µm CMOS process. Readout circuit is
combined with Silicon-On-Glass (SOG) and Dissolved Wafer Process (DWP)
accelerometers. Both open loop and closed loop tests of the accelerometer system
are performed. Open loop test results showed high sensitivity up to 8.1 V/g and low
noise level of 4.8 µg/Hz. Closed loop circuit is implemented on a PCB together

iv
with the external filtering and decimation electronics, providing 16-bit digital output
at 800 Hz sampling rate. High acceleration tests showed ±18.5 g of linear
acceleration range with high linearity, using DWP accelerometers. The noise tests
in closed loop mode are performed using Allan variance technique, by acquiring the
digital data. Allan variance tests provided 86 µg/Hz of noise level and 74 µg of
bias drift. Temperature sensitivity tests of the readout circuit in closed loop mode is
also performed, which resulted in 44 mg/ºC of temperature dependency.

Two different types of new adaptive sigma-delta readout circuits are designed in
order to improve the resolution of the systems by higher frequency operation. The
two circuits both change the acceleration range of operation of the system,
according to the level of acceleration. One of the adaptive circuits uses variation of
feedback time, while the other circuit uses multi-bit feedback method. The
simulation results showed micro-g level noise in closed loop mode without the
addition of the mechanical noise of the sensor.

Key words: Sensor interface electronics, Capacitive readout circuit, Sigma-Delta

Modulator, Closed Loop Systems, MEMS Accelerometers, Inertial Sensors.

v
 ÖZ

YÜKSEK PERFORMANS MEMS İVMEÖLÇERLER

İÇİN SIĞASAL CMOS OKUMA DEVRELERİ

Kepenek, Reha
Yüksek Lisans, Elektrik ve Elektronik Mühendisliği Bölümü
Tez Yöneticisi: Yard. Doç. Dr. Haluk Külah
Ortak Tez Yöneticisi: Prof. Dr. Tayfun Akın

Şubat 2008, 144 sayfa

Bu tezde yüksek çözünürlüklü, geniş ölçüm aralığı olan, sığasal MEMS

ivmeölçerler için sigma-delta tipi okuma devrelerinin geliştirilmesi anlatılmaktadır.
Tasarlanan okuma devresinin, sayısal çıkışlı, tam farksal, kapalı devre yapısıyla,
yüksek örnekleme ve yüksek çözünürlük elde edilir. Okuma devresi ve ivmeölçer
duyargasının modelleri Cadence ve Matlab Simulink ortamlaında hazırlanıp,
benzetimleri yapılmıştır. Benzetimler, ivmeölçer sisteminin kararlı ve doğru
çalışmasını doğrulamıştır. Sigma-delta okuma devresi XFab 0.6 µm CMOS süreci
kullanılarak üretilmiştir. Okuma devresi Silicon-On-Glass (SOG) ve Dissolved
Wafer Process (DWP) ivmeölçer yapılarıyla birleştirilmiştir. İvmeölçer sisteminin,
açık ve kapalı döngü testleri yapılmıştır. Açık döngü test sonuçları 8.1 V/g‟ye kadar
yüksek hassasiyet ve 4.8 µg/Hz‟lik düşük gürültü seviyesi göstermiştir. Kapalı
döngü devre, dış filtreleme ve seyreltme elektroniği ile birlikte baskı devre kartına

vi
yerleştirilmiş ve 800 Hz örnekleme frekansında 16-bit sayısal çıktı elde edilmiştir.
DWP ivmeölçerler kullanılarak yapılan yüksek ivme testleri, ±18.5 g ivme ölçüm
aralığında yüksek doğrusallık göstermiştir. Kapalı devre gürültü testleri veriler
toplanıp, Allan Variance tekniği kullanılarak gerçekleştirilmiştir. Allan Variance
testleri 86 µg/Hz gürültü seviyesi ve 74 µg sabit kayma değeri göstermiştir.
İvmeölçer sisteminin sıcaklık testleri de yapılmış ve 44mg/ºC sıcaklık hassasiyeti
gözlemlenmiştir.

Çözünürlüğü artırmak için, yüksek örnekleme frekansı kullanılarak, iki farklı, yeni,
uyarlamalı sigma-delta okuma devresi tasarlanmıştır. İki devre de, uygulana ivme
değerine göre ivme ölçüm aralığını değiştirmektedir. Uyarlamalı devrelerin biri,
geri besleme süresi değişimi kullanırken, diğer devre çok-bitli geri besleme
metodunu kullanır. Benzetim sonuçları kapalı döngüde, duyarga gürültüsü
katılmadan mikro-g seviyesinde gürültü elde edildiğini göstermiştir.

Anahtar kelimeler: Duyarga Arayüz Elektroniği, Sığasal Okuma Devresi, Sigma-

Delta Modülator, Kapalı Devre Sistemler, MEMS İvmeölçerler, Eylemsizlik
Duyargası.

vii
DEDICATION

To My Sister, Damla

viii
 ACKNOWLEDGEMENTS

First of all, I would like to express my appreciation and thanks to my supervisors

Asst. Prof. Dr. Haluk Külah and Prof. Dr. Tayfun Akın, for their support, guidance,
and friendly attitude during my graduate study and the development of this thesis.

I would like to present my special thanks to İlker Ender Ocak, for his support, great
friendship, sharing his knowledge, and his helps in packaging and testing of readout
circuits.

I thank to Dr. Said Emre Alper for his guidance and support during my study. I also
thank to Orhan Ş. Akar for his help in packaging the readout chips. Thanks to
Murat Tepegöz for his support in software and network issues. I would like to
express my gratitude to Yüksel Temiz and Batuhan Dayanık for the discussions
about the readout electronics, and their friendship. Thanks to Halil İbrahim Atasoy
for his help and friendship. Thanks to all METU MEMS research group members
for creating a friendly research environment.

I would like to thank to TÜBİTAK for their support of my study with their
scholarship.

I would like to thank to my parents and my sister, for their patience, support, and
encouragement.

Lastly, I would like to express my special thanks to Ishtar Dawn Haas, for her help,
encouragement and lovely friendship.

ix
 TABLE OF CONTENTS

ABSTRACT ............................................................................................................... iv
ÖZ .............................................................................................................................. vi
DEDICATION .........................................................................................................viii
ACKNOWLEDGEMENTS ....................................................................................... ix
TABLE OF CONTENTS ............................................................................................ x
LIST OF TABLES ...................................................................................................xiii
LIST OF FIGURES ................................................................................................. xiv
1. INTRODUCTION ............................................................................................... 1
1.1. Capacitive Accelerometers ........................................................................... 2
1.2. Capacitive Interfaces .................................................................................... 3
1.3. Previous Work .............................................................................................. 7
1.4. Accelerometers Fabricated at METU ......................................................... 12
1.5. Objectives and Organization of the Thesis ................................................. 15
2. THEORY OF SIGMA DELTA MODULATORS ............................................ 17
2.1. Sigma – Delta Modulators .......................................................................... 17
2.2. Oversampling and Quantization ................................................................. 19
2.3. Noise Shaping............................................................................................. 21
2.4. Filtering and Decimation ............................................................................ 23
2.5. Electromechanical Sigma – Delta Modulator............................................. 25
2.6. Conclusion .................................................................................................. 26
3. MODELING & SIMULATIONS ...................................................................... 27
3.1. Modeling the Accelerometer ...................................................................... 27
3.1.1. Accelerometer Transfer Function ....................................................... 30
3.1.2. Cadence Modeling of Accelerometer Sensor ...................................... 35
3.2. Complete System Modeling in Matlab Simulink ....................................... 38
3.3. Closed Loop Response and Stability Analysis ........................................... 42

x
 3.4. Conclusion .................................................................................................. 52
4. READOUT CIRCUIT DESIGN ........................................................................ 53
4.1. Structure of the Readout Circuit ................................................................. 54
4.2. Description of the Main Blocks of the Readout Circuit ............................. 56
4.2.1. Switched – Capacitor Network ........................................................... 56
4.2.2. Charge Integrator Operation ............................................................... 61
4.2.3. Lead Compensator .............................................................................. 71
4.2.4. Comparator .......................................................................................... 73
4.2.5. Start-up Circuit .................................................................................... 75
4.2.6. Multiphase Clock Generator ............................................................... 76
4.2.7. Test Structures and Control Signals .................................................... 78
4.3. Overall Circuit Simulation Results............................................................. 80
4.4. Layout Considerations ................................................................................ 83
4.5. Performance Limitations ............................................................................ 84
4.5.1. Mechanical Noise ................................................................................ 84
4.5.2. Readout Circuit Noise ......................................................................... 85
4.5.3. Open Loop Mode Non-Linearity and Operation Range ..................... 86
4.5.4. Closed Loop Operational Input Acceleration Range .......................... 87
4.5.5. Quantization Noise .............................................................................. 89
4.5.6. Mass Residual Motion ........................................................................ 89
4.6. Conclusion .................................................................................................. 91
5. ADAPTIVE SIGMA DELTA READOUT CIRCUITS .................................... 92
5.1. Operation Principle of the Readout Circuits .............................................. 93
5.1.1. Sigma-Delta Readout Circuit with Adaptive Feedback Duration ....... 94
5.1.2. Sigma-Delta Readout Circuit with Multi-bit Feedback ...................... 95
5.2. Description of the Main Blocks of the Readout Circuit ............................. 98
5.2.1. Operational Trans-conductance Amplifier (OTA) .............................. 98
5.2.2. Lead Compensator ............................................................................ 101
5.2.3. Digital Blocks ................................................................................... 104
5.3. Simulation Results Using Matlab Simulink ............................................. 107
5.4. Layouts ..................................................................................................... 110

xi
 5.5. Comparison of Readout Circuits and Conclusion .................................... 111
6. IMPLEMENTATION AND TESTS ............................................................... 113
6.1. Implementation of the Readout Circuit .................................................... 113
6.2. Open Loop Test Results ........................................................................... 117
6.2.1. Sensitivity of the Readout Circuit ..................................................... 117
6.2.2. Sensitivity of the System Using Accelerometer................................ 119
6.2.3. Noise Tests ........................................................................................ 120
6.3. Closed Loop Tests .................................................................................... 123
6.3.1. Closed Loop Linearity and Sensitivity.............................................. 123
6.3.2. Closed Loop Dynamic Response ...................................................... 125
6.3.3. High G Acceleration Tests ................................................................ 127
6.3.4. Noise Tests ........................................................................................ 130
6.3.5. Temperature Tests ............................................................................. 133
6.4. Conclusions .............................................................................................. 134
7. CONCLUSION AND FUTURE WORK ........................................................ 136
7.1. Future Directions ...................................................................................... 138
REFERENCES........................................................................................................ 140

xii
 LIST OF TABLES

Table 1.1: A comparison of previous studies on capacitive accelerometer systems. 11

Table 1.2: Parameters of capacitive accelerometers fabricated at METU. ............... 14
Table 4.1: Size of the transistors used in the OTA circuit. ....................................... 66
Table 4.2: Summary of the simulation results of the folded cascode OTA. ............. 69
Table 4.3: Size of the transistors used in the bias generator circuit. ......................... 70
Table 4.4: Size of the transistors used in the comparator circuit. ............................. 74
Table 4.5: Readout circuit parameters and design achievements.............................. 91
Table 5.1: Size of the transistors used in the OTA circuit. ..................................... 100
Table 5.2. Summary of the simulation results for the folded cascode amplifier. .... 101
Table 5.3: Feedback duration ratios and the compared values of the readout circuit.106
Table 5.4: Comparison of Sigma-Delta readout circuit in terms of quantization
noise and mass residual motion................................................................................ 112
Table 5.5: Comparison of Sigma-Delta readout circuit in terms of noise sources.. 112
Table 6.1. Regions in Allan variance graph and their meanings. ............................ 132
Table 6.2. Summary of system level test results with SOG and DWP
accelerometers. ......................................................................................................... 135

xiii
 LIST OF FIGURES

Figure 1.1: The application areas of accelerometers according to the operation

range and bandwidth. ................................................................................................... 2
Figure 1.2: Illustration of basic capacitive accelerometer. ......................................... 3
Figure 1.3: Capacitive topologies. (a) Half bridge structure. (b) Full bridge
structure. (c) Pseudo-differential half bridge structure. ............................................... 4
Figure 1.4: AC bridge amplifier structure [12]. .......................................................... 5
Figure 1.5: Trans-impedance amplifier structure [12]. ............................................... 6
Figure 1.6: Switched-capacitor charge integrator structure [12]. ............................... 6
Figure 1.7: Surface micro machined 3-axis monolithic capacitive accelerometer.
[15]. .............................................................................................................................. 7
Figure 1.8: Switched capacitor sigma-delta readout circuit designed by University
of California, Berkeley [15]. ........................................................................................ 8
Figure 1.9: The photograph of the surface micro-machined accelerometer chip
implemented by Carnegie Mellon University [18, 19]. ............................................... 8
Figure 1.10: Fully differential readout circuit designed by Carnegie Mellon
University [18, 19]. ...................................................................................................... 9
Figure 1.11: Z-axis capacitive accelerometer designed and fabricated at University
of Michigan [20]. ......................................................................................................... 9
Figure 1.12: Fully differential switched capacitor readout circuit designed in
University of Michigan [21]....................................................................................... 10
Figure 1.13: Illustration of SOI accelerometer designed and fabricated in Georgia
Institute of Technology [23]....................................................................................... 10
Figure 1.14: Switched capacitor readout circuit with a new switching scheme,
designed in Georgia Institute of Technology [25]. ................................................... 11
Figure 1.15: Silicon-On-Glass (SOG) accelerometer structure designed and
fabricated in METU [29]. ........................................................................................... 12
Figure 1.16: Fabrication process of the SOG accelerometer [29]............................. 13

xiv
Figure 1.17: Dissolved Wafer Process (DWP) accelerometer structure designed
and fabricated in METU [29]. .................................................................................... 13
Figure 1.18: Fabrication process of the DWP accelerometer [29]. ......................... 14
Figure 2.1: Block diagram of the delta modulator structure. .................................... 18
Figure 2.2: Block diagram of Sigma – Delta modulator (a) with two integrators, (b)
with the integrator blocks combined into one. ........................................................... 19
Figure 2.3: Block diagram of the sigma – delta modulator in s-domain, with the
quantization error. ...................................................................................................... 21
Figure 2.4: Noise response of various order sigma – delta modulators .................... 22
Figure 2.5: In band quantization noise of sigma – delta modulators, depending on
the oversampling ratio and modulator order. ............................................................. 23
Figure 2.6: (a) Input signal, (b) output of the modulator with the quantization noise,
(c) low pass filtration, (d) low pass filtered & decimated output............................... 24
Figure 2.7: The block diagram of the closed loop electromechanical system. ......... 25
Figure 3.1: (a) The structure of the SOG accelerometer designed in METU MEMS
VLSI research group. (b) Illustration of the accelerometer dimensions. ................... 28
Figure 3.2: Mass-spring-damper system. .................................................................. 30
Figure 3.3: Magnitude and phase response of an accelerometer with different
quality factors. ............................................................................................................ 32
Figure 3.4: Folded beam spring structure. ................................................................ 33
Figure 3.5: Illustration of (a) squeeze film damping, (b) Coutte flow damping. ...... 34
Figure 3.6: The dependence of coefficient c on the dimensions of the plates. ......... 35
Figure 3.7: Block diagram of the model constructed in Cadence environment. ....... 36
Figure 3.8: RLC circuit constructed for modeling the transfer function of the
accelerometer. ............................................................................................................ 37
Figure 3.9: Simulink model of the complete electromechanical closed loop system.39
Figure 3.10: Block diagram of the feedback block. .................................................. 39
Figure 3.11: Front-end readout circuit modeling. ..................................................... 40
Figure 3.12: Block diagram of the compensator. ...................................................... 41
Figure 3.13: Filtering and decimation blocks. .......................................................... 42
Figure 3.14: Block diagram of the sigma – delta loop. ............................................. 43
Figure 3.15: Closed loop frequency response of the electromechanical system. ..... 45

xv
Figure 3.16: Block diagram of a basic feedback system........................................... 46
Figure 3.17: Basic illustration of displacement, velocity and feedback acceleration
for a sigma delta loop. ................................................................................................ 48
Figure 3.18: The phase responses of the blocks in the closed loop system. ............. 50
Figure 3.19: Total phase of the of the system (a) without the lead compensator (b)
with the lead compensator. ......................................................................................... 51
Figure 4.1: Block diagram of the fully differential sigma-delta readout circuit. ...... 54
Figure 4.2. Complementary CMOS switch. .............................................................. 57
Figure 4.3. Switch resistance versus W/L ratios of the transistors. .......................... 57
Figure 4.4. Complementary CMOS switch with dummy transistors. ....................... 58
Figure 4.5: Schematic view of the switched capacitor network and the charge
integrator stages. ........................................................................................................ 58
Figure 4.6. Timing diagram of the signals used in the readout circuit. ..................... 59
Figure 4.7. Complementary CMOS switch with dummy transistors. ....................... 60
Figure 4.8. Basic schematic of the charge integrator circuit. .................................... 61
Figure 4.9. Schematic view of the folded cascode OTA. .......................................... 65
Figure 4.10. AC simulation results of the folded cascode OTA. .............................. 68
Figure 4.11. Noise simulation results of the folded cascode OTA. .......................... 68
Figure 4.12. Schematic view of the bias generator circuit. ....................................... 70
Figure 4.13. Temperature simulation results of the bias generator circuit. ............... 71
Figure 4.14. Schematic of the lead compensator circuitry. ....................................... 72
Figure 4.15. Timing diagram of the lead compensator, type 1. ................................ 73
Figure 4.16. Schematic view of the dynamic comparator. ........................................ 74
Figure 4.17. Input offset voltage of the comparator versus temperature. ................. 75
Figure 4.18. Schematic view of the start-up circuit. ................................................. 76
Figure 4.19. Schematic view of the multiphase clock generator circuit. .................. 77
Figure 4.20. Timing diagram of the signals generated by the multiphase clock
generator..................................................................................................................... 77
Figure 4.21. Schematic view of multiplexer structures, used for applying external
timing signals. ............................................................................................................ 78
Figure 4.22. Schematic view of the test structure for observing the internal signals.79
Figure 4.23. Schematic view of the sample and hold circuitry. ................................ 79

xvi
Figure 4.24: Open loop simulation results of the sigma delta readout circuit
together with the model of the SOG accelerometer. .................................................. 80
Figure 4.25: Closed loop simulation results of the sigma delta readout circuit
together with the model of the SOG accelerometer ................................................... 81
Figure 4.26: Simulink model of the complete electromechanical closed loop system.82
Figure 4.27: Simulation results of the accelerometer system using the model created
in Matlab. ................................................................................................................... 82
Figure 4.28: Layout view and floor plan of the fully differential readout circuit
chip. ............................................................................................................................ 83
Figure 4.29. Schematic view of the basic switched capacitor charge integrator
structure. ..................................................................................................................... 85
Figure 4.30. Differential capacitance versus displacement graph for a capacitive
accelerometer. ............................................................................................................ 87
Figure 4.31. Illustration of the sensor fingers and parameters. ................................. 88
Figure 5.1. Illustration of the effect of decreasing feedback time. (a) with 60%
feedback duration. (b) with 20% feedback duration. ................................................. 93
Figure 5.2. Illustration of the sense and feedback timings for different acceleration
ranges, for the varying feedback-time readout circuit. .............................................. 94
Figure 5.3. The structure of the accelerometer sensor, to be used with the multi-bit
adaptive sigma-delta circuit structure. ....................................................................... 96
Figure 5.4. Block Diagram of the readout circuit. .................................................... 97
Figure 5.5. Schematic view of the folded cascode OTA. .......................................... 99
Figure 5.6. AC magnitude and phase response of the OTA. ................................... 100
Figure 5.7. Simulation results for the input referred noise of OTA. ....................... 101
Figure 5.8: Schematic view of the lead compensator circuit. ................................. 102
Figure 5.9: Compensator switching timing diagram. .............................................. 102
Figure 5.10: Block diagram of the digital part of the readout circuit. .................... 104
Figure 5.11: Timing diagram of generated major timing signals. .......................... 105
Figure 5.12: Illustration of the input acceleration coverage according to different
operation ranges. ...................................................................................................... 106
Figure 5.13. Simulink model of the accelerometer system. .................................... 108
Figure 5.14. Simulink simulation results for observing mass-residual motion....... 108

xvii
Figure 5.15. Simulink simulation results of the readout circuit. ............................. 109
Figure 5.16: Layout of the adaptive readout circuits with (a) varying feedback time,
(b) multi-bit feedback. .............................................................................................. 110
Figure 6.1: Photo of fully differential sigma-delta capacitive accelerometer readout
circuit........................................................................................................................ 114
Figure 6.2: Printed Circuit Board used for the tests of the readout circuit and the
accelerometer. .......................................................................................................... 114
Figure 6.3: Photograph of the second readout chip. ............................................... 115
Figure 6.4: The accelerometer and the readout circuit wire-bonded in the same
package, using an alumina substrate PCB................................................................ 116
Figure 6.5: Accelerometer system together with the external filtering and
decimation circuitry, placed on a PCB. .................................................................... 116
Figure 6.6: Sensitivity test results of the readout circuit. ....................................... 117
Figure 6.7: Differential output voltage versus inverse of the integration capacitance.118
Figure 6.8: Open loop sensitivity tests of the readout circuit together with the
accelerometer. .......................................................................................................... 119
Figure 6.9: Test setup for noise measurements of the readout circuit. ................... 120
Figure 6.10: (a) Gain of the external circuitry, (b)Noise of the external circuitry.. 121
Figure 6.11: Open loop noise test results of the readout circuit, with and without the
accelerometer. .......................................................................................................... 122
Figure 6.12: Closed loop linearity and sensitivity test results of the accelerometer
system, with SOG accelerometer. ............................................................................ 123
Figure 6.13: Output data versus time graph of the linearity test results of the closed
loop accelerometer system, with 12 angular positions. ........................................... 124
Figure 6.14: Averaged output data versus acceleration. ........................................ 125
Figure 6.15: Test setup for measuring the dynamic response of the accelerometer,
using the rate table. .................................................................................................. 126
Figure 6.16: Differential analog outputs of the readout circuit, with 70mg, 140 mg
and 210 mg amplitude sinusoidal inputs. ................................................................. 127
Figure 6.17: Illustration of high acceleration tests performed on centrifuge table. 128
Figure 6.18: Fixture constructed for mounting the PCB on the centrifuge table. ... 128
Figure 6.19: High acceleration test results of the closed loop accelerometer system.129

xviii
Figure 6.20: Linearity graph obtained from high acceleration tests. ...................... 129
Figure 6.21: Illustration of the test setup constructed for closed loop noise tests. . 131
Figure 6.22: Differential analog outputs of the readout circuit............................... 131
Figure 6.23: Sample Allan Variance graph. ............................................................ 132
Figure 6.24: Allan variance graph obtained from the tests. .................................... 133
Figure 6.25: Linearity graph obtained from high acceleration tests. ...................... 134

xix
 CHAPTER 1

1. INTRODUCTION

Silicon-based inertial sensors, including accelerometer and gyroscopes, are one of

the most crucial types of Micro-Electro-Mechanical-Systems (MEMS) devices. Due
to their low cost, low power, small size, and high reliability, MEMS accelerometers
have extensive use, such as in airbag safety systems in automotive area, in digital
cameras for picture stabilization, for vibration monitoring in industry, microgravity
measurements in space, tilt control and platform stabilization, seismometry, and
inertial navigation and guidance [1]. Measurement bandwidth and acceleration
needs of these application areas are illustrated in Figure 1.1. From all of these areas,
MEMS accelerometers have an increasing vital role in navigational systems. The
navigational area needs high resolution, particularly having sub-micro-g resolution
accelerometers. In order to achieve sub micro-g resolution, both of the two main
components, the sensor and the readout electronics, of an accelerometer system
should be designed with extreme attention. Recently, capacitive accelerometers
have become very attractive for high precision g applications due to their high
sensitivity, low temperature sensitivity, low power, wide dynamic range and simple
structure [2-7]. There is still a necessity of high performance readout circuits to
read the extremely small movements of capacitive accelerometers. Therefore, the
research presented in this thesis focuses on the navigational purpose readout circuits
for capacitive MEMS accelerometer sensors, achieving both high resolution and
high range.

A basic explanation of capacitive type accelerometers is given in Section 1.1. In

Section 1.2, a general discussion is made on the interface electronics used in
capacitive sensing accelerometers. Section 1.3 gives a brief explanation and

1
comparison of previously designed and implemented accelerometer systems in the
literature. Then the accelerometer structures designed and fabricated in METU are
discussed in Section 1.4. Finally, Section 1.5 provides the objectives of this
research and the organization of thesis.

Smart Ammunition
104
Shock
103 Airbag Measurement
Medical
Bandwidth (Hz)

102 Active suspension

Pointing device
10 Shipping
Space Head Mounted Display
1
Navigation
10-1

10-6 10-4 10-2 1 102 104 106

Acceleration (g)

Figure 1.1: The application areas of accelerometers according to the operation range and
bandwidth.

1.1. Capacitive Accelerometers

Accelerometers need to measure the movement of an inertial mass in order to detect

acceleration. In micro-g level acceleration measurements, displacements of the
mass are generally in the order of angstroms. To detect such small displacements,
there is a variety of acceleration sensing methods presented in the literature, such as
magnetic, tunneling [8], optical, [9] and capacitive sensing. The advantage of
capacitive sensing is the ability to obtain highly sensitive and reliable
accelerometers, and ease of interfacing with the electronic circuitry.

Basic structure of a capacitive accelerometer is illustrated in Figure 1.2. The

accelerometer uses a movable mass, which is connected to base through spring
structures. There are also stationary electrode structures on the two sides of the

2
mass, where the capacitance in between changes according to the displacement of
the mass. Commonly, finger-like structures are used to increase the capacitive area
between the mass and the electrodes. In order to achieve a highly sensitive
capacitive accelerometer, capacitive area should be increased while decreasing the
gaps in between the plates. In addition, the mass should be designed large, to
accomplish low mechanical noise and high sensitivity. The construction of
capacitances on the mechanical structure enables the conversion of a mechanical
signal into electrical domain. The next section gives a discussion on basic electronic
interface structures, and the capacitive configurations.

Electrode

Mass
Springs

Electrode
Figure 1.2: Illustration of basic capacitive accelerometer.

1.2. Capacitive Interfaces

There are mainly two capacitive topologies, which are half bridge and full bridge
configurations, as shown in Figure 1.3 (a) and (b), respectively. The half bridge
configuration is used with single ended readout circuits, either with two sense
capacitors or with a sense and a reference capacitor. Square or sinusoidal signals,
frequencies of which are much higher compared to the resonance frequency of the
accelerometer, are applied at two ends of the half bridge. This creates a charge flow
or a voltage at the common node, proportional to the difference of two capacitors.

3
With this configuration, some non-idealities, such as offset, power supply noises,
and other common mode noises are observed, due to the single ended operation.

Full bridge configuration is composed of four capacitors, at least two of which are
sense capacitors. In this case, a differential charge flow, or voltage is generated at
the sensing nodes of the full bridge. This configuration is used with a differential
readout circuitry, and therefore, the common mode noises are not observed, and the
configuration has higher sensitivity.

Another capacitor configuration is a pseudo-differential structure as shown in Figure

1.3 (c). In this configuration, the square wave is applied from the common node,
and the sensing is performed from the ends of a half bridge structure, which allows
the usage of a differential readout circuit, and hence cancelling the common mode
non-idealities at the readout circuit.

(a) (b) (c)

Figure 1.3: Capacitive topologies. (a) Half bridge structure. (b) Full bridge structure.
(c) Pseudo-differential half bridge structure.

For reading out the capacitance differences, there are a few main types of readout
circuit structures. The readout structures are given with half bridge structures, for
the ease of understanding. A basic readout structure, given in Figure 1.4 uses
capacitive voltage division between two capacitances [10, 11]. This structure is also

4
named as ac-bridge amplifier [12, 13, 14], where the voltage created at the sense
node of the half bridge is amplified, demodulated and then low-pass filtered to
obtain meaningful analog data. The output voltage of the readout circuit is given in
Equation (1.1). This circuitry has the disadvantage of having lower performance
with high parasitic capacitances; hence, it is not compatible with the non-monolithic
structures.

∆𝐶
𝑉𝑜𝑢𝑡 = 𝑉𝑝 𝐴 (1.1)
2𝐶𝑠0 + 𝐶𝑝 𝑣

Figure 1.4: AC bridge amplifier structure [12].

Figure 1.5 shows trans-impedance amplifier readout structure, where the capacitors
are driven by two sinusoidal signals with 180 phase difference. Sensing node of
the half bridge is held at virtual ground by the resistive feedback. The current
created due to the capacitance difference flows through the feedback resistance, and
a voltage proportional to the capacitance difference is formed, as shown in
Equation (1.2). The output is then demodulated and low-pass filtered. This
circuitry is most suitable for the resonance mode sensors, because of the need for the
sinusoidal drive signal. Moreover, the noise of the circuit is usually dominated by
the feedback resistor.

𝑉𝑜𝑢𝑡 = 2𝜋𝑓𝑑𝑟𝑖𝑣𝑒 𝑉𝑚 𝑅𝑓 ∆𝐶 (1.2)

5
Figure 1.5: Trans-impedance amplifier structure [12].

The readout circuit structure shown in Figure 1.6 is named as a charge integrator,
which is the circuit structure selected in this study. Switched capacitor technique is
used in the circuitry. In this circuit, the excess charge created due to the capacitance
difference, is transferred on to the integration capacitor, Cint, and the charge is reset
at each readout cycle. The output of the charge integrator should only be low-pass
filtered without the need of a separate demodulator. The output of this circuit is
independent of the parasitic capacitances, as given in equation (2.1), which makes it
suitable for non-monolithic applications. With the use of a switched capacitor
circuitry, correlated double sampling (CDS) technique can also be added to the
circuitry in order to cancel the low frequency noise sources. Due to these
advantages, the switched capacitor readout circuit is the optimum choice for non-
monolithic capacitive accelerometers to achieve a high performance accelerometer
system.

∆𝐶
𝑉𝑜𝑢𝑡 = 𝑉𝑝 (1.3)
𝐶𝑖𝑛𝑡

Figure 1.6: Switched-capacitor charge integrator structure [12].

6
The next section gives some main examples on these readout circuit structures and
capacitive accelerometers existing in the literature, and their system level
achievements.

1.3. Previous Work

There are several studies of capacitive accelerometers in the literature. The Figure
1.7 shows the chip photograph of a 3-axis surface micro-machined accelerometer,
including the readout circuit on the same chip, which was designed and
implemented in University of California, Berkeley [15-17]. There are two sense
capacitors, connected in a pseudo differential half-bridge configuration. The
readout circuit uses switched capacitor charge integrator to sense the capacitive
difference, as shown in Figure 1.8. Since, the sensing nodes of the half bridge
structure is floating, an input common mode feedback circuitry is used to set the
input common mode voltage. The overall structure is composed of a sigma-delta
loop in order to hold the proof mass stationary by applying electrostatic feedback
force, which also enables digital output generation. This study achieved 110
µg/Hz noise with 84 dB dynamic range.

Figure 1.7: Surface micro machined 3-axis monolithic capacitive accelerometer. [15].

7
Figure 1.8: Switched capacitor sigma-delta readout circuit designed by University of
California, Berkeley [15].

Another surface micro-machined, monolithic accelerometer was designed and

implemented in Carnegie Mellon University [18, 19]. The chip photograph of the
accelerometer system is given in Figure 1.9. There are 4 separate sense capacitors
on the mechanical structure, which allows fully differential configuration. For the
readout circuit, ac bridge structure is used, as shown in Figure 1.10. The circuit
accomplishes 50 µg/Hz noise at 400 Hz with ±6g linear range, which results in
106 dB of dynamic range.

Figure 1.9: The photograph of the surface micro-machined accelerometer chip

implemented by Carnegie Mellon University [18, 19].

8
Figure 1.10: Fully differential readout circuit designed by Carnegie Mellon University [18,
19].

In the research carried out in University of Michigan, a z-axis bulk micro-machined

accelerometer was designed and implemented, with a large proof mass, as shown in
Figure 1.11 [20]. The large proof mass enables high sensitivity and low mechanical
noise of the sensor. A fully differential switch capacitor readout circuit is designed
and implemented for position sensing with closed loop sigma-delta configuration as
shown in Figure 1.12 [21, 22]. Hence, the system achieves noises of 3.5 µg/Hz in
open loop, and 25 µg/Hz in closed loop force feedback operation, with ±1.35 g
acceleration range.

Figure 1.11: Z-axis capacitive accelerometer designed and fabricated at University of

Michigan [20].

9
Figure 1.12: Fully differential switched capacitor readout circuit designed in University of
Michigan [21].

The research on MEMS accelerometers is also being performed in Georgia Institute

of Technology, where an SOI accelerometer structure is fabricated, with large mass
and high sensitivity, which is shown in Figure 1.13 [23-25]. The process allows 4
different sense capacitors to be constructed, however all the capacitors are
connected at the common node, proof mass of the sensor. This connection of the
sense capacitors does not allow construction of full bridge configuration. Therefore,
a switching scheme is proposed and implemented to perform the position sensing
with 4 sense capacitors, as shown in Figure 1.14. The readout circuit uses switched
capacitor charge integration technique, and the system achieves 4.4 µg/Hz noise at
150 Hz with ±2 g operation range.

Figure 1.13: Illustration of SOI accelerometer designed and fabricated in Georgia Institute
of Technology [23].

10
Figure 1.14: Switched capacitor readout circuit with a new switching scheme, designed in
Georgia Institute of Technology [25].

Table 1.1 gives a comparison of the performance of some previous capacitive

accelerometer studies, existing in the literature, and the proposed achievements of
this work. The main performance criteria for an accelerometer are the operation
range, resolution and bandwidth. In these studies, both the sensor structure and the
readout circuit play an important role, in achieving high performance accelerometer
systems. Hence, the purpose of this thesis is to obtain high resolution and wide
operational range in closed loop, i.e. improving the dynamic range, by improving
the readout circuit structure.

Table 1.1: A comparison of previous studies on capacitive accelerometer systems.

Linear
Source Accelerometer type Bandwidth Resolution
Range
M. Lemkin et.al., 1999 Surf. micro-machined 110, 160, 990
100 ~ ±1 g
[16] Monolithic, 3-axis, µg/Hz
J. Chae, et.al., 2000
SOG, Lateral ~ 2 kHz > ±2 g <100 µg/Hz
[26]
G. Fedder et.al., 2004 Surf. Micro-machined
2 kHz ±6 g 50 µg/Hz
[18] Monolithic
M. Lemkin, B. Boser, 1996 Surf. Micro-machined
~8 kHz ±3.5 g 500 µg/Hz
[27] Monolithic
X. Jiang et.al., 2002 Surf. Micro-machined
~1.4 kHz 0.125 g 2 µg/Hz
[28] Lateral
H. Kulah, 2003 Surf. – Bulk
100 Hz ±1.35 g 3.5 µg/Hz
[22] micromachining
B. V. Amini et.al., 2004 Lateral, SOI Bulk
150 Hz ±2 g 4.4 µg/Hz
[23] micromachining
Proposed study Lateral, Bulk
~ 1 kHz > ±20 g <100 µg/Hz
of this thesis (closed loop) micromachining

11
1.4. Accelerometers Fabricated at METU

Accelerometer sensors used in this research were fabricated by MEMS-VLSI

research group at Middle East Technical University. There are two main types of
sensor structures, one of which is named as Silicon-On-Glass (SOG) accelerometers,
as shown in Figure 1.15. The structure is placed on a glass substrate, and the silicon
is shaped using DRIE processing [29]. Accelerometers have comb finger structures
on each side of the proof mass, which constitutes two differential capacitances. The
thickness of the structural layer is 100 µm, which allows large proof mass, hence,
high sensitivity and low mechanical noise. However due to the large mass of the
sensor, the operational range of the accelerometer is limited, in both open loop and
closed loop operation.

Figure 1.15: Silicon-On-Glass (SOG) accelerometer structure designed and fabricated in

METU [29].

The fabrication process of the SOG accelerometer requires only 4 masks. First, a
glass substrate is etched to form anchor regions (Figure 1.16 (a)). Then, a shielding
metal layer is patterned on a 100m-thick silicon wafer (Figure 1.16 (b)). This layer
prevents DRIE notching and acts as a heat sink during DRIE. Next, the silicon and
glass wafers are anodically bonded (Figure 1.16 (c)). Metal contacts are evaporated
and patterned, and finally the wafer is etched with DRIE to define the proof mass
and sensing electrodes (Figure 1.16 (d, e, f)). Then the shielding layer is removed.

12
Figure 1.16: Fabrication process of the SOG accelerometer [29].

The second structure fabricated by the research group uses dissolved wafer process
(DWP) technique [29], as illustrated in Figure 1.17, which allows smaller gaps
between the finger structures. Smaller gaps results in higher sensitivity, due to the
increased capacitance. However, the structural thicknesses of these devices are also
decreased down to 15 µm, which decreases the sensitivity. The proof masses of the
DWP accelerometers are smaller due to the decreased structural thickness.
Although this results in higher mechanical noise, the operational linear range of the
system is increased up to 20 to 30 g‟s depending on the designed accelerometer.

Fingers Proof Mass

182 Side Fingers

m
m 0u
0u 55
40

15 um Spring

124
Central Fingers

Figure 1.17: Dissolved Wafer Process (DWP) accelerometer structure designed and
fabricated in METU [29].

13
Fabrication process of the DWP accelerometer requires 3 masks. First, a glass
substrate is etched to form anchor regions of the accelerometer (Figure 1.18 (a)).
After the formation of the anchor regions chromium and gold is sputtered on the
glass wafer and patterned to form the electrical connections (Figure 1.18 (b)). Then,
a 100µm thick silicon wafer is doped with boron (Figure 1.18 (c))and etched
reactively to form the structural layer (Figure 1.18 (d)). Next, silicon and glass
wafers are anodically bonded (Figure 1.18 (e)) and the undoped silicon is
completely etched (Figure 1.18 (f)). The parameters of two sample accelerometers
of the two kinds designed and fabricated in METU are given in Table 1.2.

(a)

(e)
(b)

(f)

(c)
GLASS GOLD
BORON DOPED SILICON CHROMIUM
SILICON

(d)

Figure 1.18: Fabrication process of the DWP accelerometer [29].

Table 1.2: Parameters of capacitive accelerometers fabricated at METU.

Parameter Value (SOG) Value (DWP)

Mass of proof mass 0.72 milli-g 0.18 milli-g
Resonant frequency 1.53 kHz 1.72 kHz
Proof mass thickness 100 µm 15 µm
Sensing gap 4.48 µm 1.1 µm
Sense capacitance 24.8 pF 16.2 pF
Sense fingers length 500 µm 400 / 350 µm
Number of fingers 290 182 / 124
Sensitivity 1.88 pF/g 0.96 pF/g
Brownian noise 2.81 µg/√Hz 3.62 µg/√Hz

14
1.5. Objectives and Organization of the Thesis

The main objective of this thesis is to design and implement a readout circuit to
achieve high resolution and high dynamic range together with the accelerometer
sensors. The following is a summary of the objectives.

 Design and implementation of a complete accelerometer readout circuit.

A sigma-delta closed loop capacitive accelerometer readout circuit is to be

designed, and implemented with a resolution in the order of micro-g level,
with high dynamic range. The aim is to obtain a low power, low cost,
closed loop accelerometer system with a linear range of ±20 g, and
resolution below 100 µg.

 Modeling and simulation of the accelerometer system and verification of

operation.

Both the accelerometer and readout circuit are to be modeled in MATLAB

Simulink and Cadence environments, and the verification of operation and
stability of the system is to be done.

 Development of new readout techniques to improve the performance of the

accelerometer system.

Adaptive readout structures are to be designed and implemented in order to

achieve higher resolution and higher dynamic range.

 System level testing of the accelerometer with the readout electronics.

Full testing of the accelerometer system is to be performed, including
sensitivity, linearity, range and noise tests.

Chapter 2 of this thesis gives a brief explanation of the theory of sigma-delta

modulators, where the main properties and limitations are discussed.

Chapter 3 presents the modeling of the accelerometer system and obtained

simulation results using the constructed models. First, it explains the accelerometer

15
sensor structure, by defining the sensor parameters. Then, the modeling and
simulations of the sensor and readout circuit are examined. Finally, the model
constructed in Cadence environment is discussed and the simulation results are
given.

Chapter 4 describes the design stage of sigma-delta fully differential readout circuit.
After explaining the operation of each block of the readout circuit and supplying the
simulation results, the layout considerations and performance limitations are
discussed.

Chapter 5 describes the design of adaptive readout circuits. After presenting the
operation of two different adaptive readout circuits, the simulation results are given.
Finally, a comparison of the readout circuits is given.

Chapter 6 of the thesis provides the test results obtained throughout the research of
this subject. First, the implementation of the readout circuit together with the sensor
and the external electronics for closed loop operation is explained. Then, the open
loop and closed loop test results are given separately.

Lastly, Chapter 7 puts a conclusion to the thesis, and gives a direction for the future
work of the research.

16
 CHAPTER 2

2. THEORY OF SIGMA DELTA MODULATORS

This chapter presents the theory of operation of sigma-delta modulators. Sigma-

delta modulators provide high resolution, especially at low bandwidth analog to
digital converter applications, with the aid of oversampling and noise shaping
concepts. Since the acceleration signals to be measured has low bandwidth, sigma-
delta modulation gives great advantage in providing high resolution, with low cost.
For applying the modulation technique with a mechanical structure, sigma-delta
modulation itself should be understood, primarily. In this purpose, the concepts for
comprehending the theory of sigma-delta modulation are explained in this chapter.
Section 2.1 gives a brief description of the structure of a first order sigma-delta
modulator. The oversampling concept is explained in Section 2.2. Section 2.3
explains the noise-shaping concept in sigma-delta modulators. Section 2.4 gives the
application of sigma-delta modulator as an electromechanical system, combined
with the accelerometer sensor.

2.1. Sigma – Delta Modulators

Sigma-delta modulation was developed from delta modulation [30]. Delta

modulation is an A/D conversion technique, where the output is quantized according
to how fast the input signal amplitude varies. Hence, if the output is 1-bit, the bit
stream at the output indicates only the sign of the variations of the input signal.
Figure 2.1 shows the basic block diagram of a delta modulator. Integrator in the
feedback loop is trying to predict the input signal and an error signal is generated
after taking the difference between the prediction and the input signal. This error
signal is then quantized using a comparator. Depending on the sign of the error

17
signal, another prediction is made by increasing or decreasing the value at the output
of the integrator. On the demodulation side, the 1-bit output stream should be
integrated to obtain the quantized signal. Then, with the use of a low-pass filter, the
analog input signal can be regenerated. [30, 31]

A/D (1-bit) x(t)

x(t) y(t) x’(t)
+
∑
-

y(t)
x’(t)
∫
Modulation

y(t)
∫ Low-pass
Filter

Demodulation

Figure 2.1: Block diagram of the delta modulator structure.

The operations performed in this system are linear, so the integrator stage at the
demodulator can be carried to the input stage, as shown in Figure 2.2(a). Moreover,
in the block diagram in Figure 2.2(b), the two integrators are combined into one
integrator. This structure forms the first order sigma-delta modulator. In this
structure, the output is directly dependent on the input signal; hence, the
demodulator side only needs a low-pass filter. The operation is also performed using
a single integrator on the modulator side; hence, it is much simpler than the delta
modulator structure. The output of a sigma-delta modulator is commonly single bit;
the resolution in amplitude is carried to the resolution in time, which is achieved by
oversampling.

18
 A/D (1-bit)
y(t)
∫ + Low-pass
∑ Filter
-

x’(t)
∫
(a)

A/D (1-bit)
y(t)
∫
+ Low-pass
∑ Filter
-

(b)
Figure 2.2: Block diagram of Sigma – Delta modulator (a) with two integrators, (b) with
the integrator blocks combined into one.

The sigma-delta A/D converters are widely used for high resolution and low
bandwidth applications due to the noise shaping and oversampling techniques.
Oversampling sigma-delta modulators are extensively used for low frequency
analog-to-digital converters especially in audio applications where the over-
sampling ratio can be considerably high and the noise rejection is very efficient [32,
33]. In micromechanical accelerometers, since the mechanical bandwidth is usually
quite small (<2kHz), sigma-delta conversion can effectively reduce noise and
improve overall performance [34-36]. The following sections describe the
properties of sigma-delta modulators.

2.2. Oversampling and Quantization

Oversampling is a crucial concept in sigma–delta modulators in order to increase the

resolution of the system by increasing the sampling frequency of the system.
Oversampling increases the resolution in the time domain, and decreases the in-band

19
noise. The Nyquist rate A/D converters have a sampling rate twice the bandwidth of
the signal frequency; but the oversampling converters use higher sampling rates.
The oversampling ratio is defined as in Equation (2.1).

𝑓𝑠
𝑀= (2.1)
2𝑓𝐵𝑊

where fs and fBW are the sampling frequency and the signal frequency bandwidth,
respectively.

Quantization and the error caused by quantization is a significant point, which

should be considered primarily in an A/D system. In Nyquist sampling quantizers,
the rms value of the error is given as in Equation (2.2) [31].

∆
2
1 2 ∆2
𝑒𝑟𝑚𝑠 = 𝑒 2 𝑑𝑒 = (2.2)
∆ −∆ 12
2

where,  is the quantization level spacing. Hence the quantization error is bounded
between /2 and -/2, and have equal probability of taking any value in between.

If there is a dither signal, with sufficiently large in amplitude, the quantization error
can be assumed to be a white noise [31]. Using this assumption, for an
oversampling quantizer, the noise power inside the signal bandwidth is given as in
Equation (2.3).

2
2
𝑒𝑟𝑚𝑠 ∆2
𝑣𝑞𝑛 ,𝑟𝑚𝑠 = = (2.3)
𝑀 12. 𝑀

Conceptually, oversampling provides resolution in time, instead of resolution in

amplitude. By decimation process, the high-resolution result can be obtained.
However, as can be observed from the quantization noise expression given in

20
Equation (2.3), doubling the sampling frequency results only a 3 dB enhancement in
the quantization noise. Therefore, oversampling by itself does not improve the
resolution of the system as desired. The sigma-delta modulation, not only does
oversampling but also the noise shaping, which decreases the in-band quantization
error considerably.

2.3. Noise Shaping

Noise shaping concept is the major purpose of usage of sigma-delta modulation.

For a first order sigma-delta modulator, the quantization error is added in the last
stage, where analog data is converted to digital, as shown in Figure 2.3.

N(s)
integrator

X(s) + Y(s)
∑ 1/s ∑
- quantizer

Figure 2.3: Block diagram of the sigma – delta modulator in s-domain, with the
quantization error.

The transfer function of the system can be calculated as given in Equation (2.4),
which results in a low pass filter characteristic. For calculating the transfer function
from input to output, the quantization noise is taken to be zero.

𝑌 𝑠 1
= (2.4)
𝑋 𝑠 𝑠+1

For calculating the noise transfer function, input signal is assumed zero. So, the
noise of the system becomes as given in Equation (2.5).

𝑌 𝑠 𝑠
= (2.5)
𝑁 𝑠 𝑠+1

21
This result has a high pass filter characteristic. In this way, the quantization noise is
shaped and carried to high frequencies. Hence, in-band noise power of the system is
decreased by using a sigma-delta structure, compared to an oversampling quantizer.
The in-band quantization noise of a sigma-delta modulator is expressed as given in
Equation (2.6) [31].

𝜋𝐿
𝑉𝑞𝑛 ,𝑟𝑚𝑠 = 𝑒𝑟𝑚𝑠 (2.6)
𝑀𝐿+0.5 2𝐿 + 1

where, L is the order of the modulator, M is the oversampling ratio and erms is the
rms value of the quantization noise calculated by Equation (2.2). The above
equation shows that, as the order of the sigma-delta modulator increases, the more
of the quantization noise is carried to high frequencies and the less in-band
quantization noise is observed. Figure 2.4 gives the noise transfer function of multi
order sigma-delta modulators and makes a comparison between noise shaping of the
different order of sigma-delta modulators. As can be observed from this figure,
increasing the order of the modulator decreases the in-band noise contribution.
Figure 2.5 illustrates the dependence of the in-band quantization of the modulator to
the oversampling ratio and the modulator order.

3rd order

2nd order
magnitude

1st order

0 frequency fs/2
Figure 2.4: Noise response of various order sigma – delta modulators

22
 0
10

-2
10
In-band quantization noise

-4
10
L=1

-6
10

L=2
-8
10

L=3
-10
10
0 200 400 600 800 1000 1200
Oversampling ratio, M
Figure 2.5: In band quantization noise of sigma – delta modulators, depending on the
oversampling ratio and modulator order.

2.4. Filtering and Decimation

The instantaneous output of a sigma-delta modulator is generally not meaningful by

itself, because of the high quantization noise included at high frequencies, especially
in the case of using high sampling rate and low resolution in amplitude quantization.
The signals at different stages of a sigma-delta modulator are illustrated in Figure
2.6. The output stream includes the input signal at the low frequency band with the
quantization noise. As explained in the previous section, the quantization noise is
shaped and mostly carried to the high frequency band. Hence, to extract the signal,
from the output data stream, a low pass filtration is needed. The low pass filtration
is preferred to be a digital stage, where high quality and low cost filters can be
implemented using digital signal processing.

23
 (a)

-fB fB

(b)

-fS/2 -fB fB fS/2

(c)

-fS/2 -fB fB fS/2

(d)

-fS/N -fB fB fS/N

Figure 2.6: (a) Input signal, (b) output of the modulator with the quantization noise, (c) low
pass filtration, (d) low pass filtered & decimated output.

The low pass filtered data has a low bandwidth; however, the sampling rate does not
change with the filtration process. A sampling rate at the Nyquist frequency is
sufficient at the output of the low pass filter. Hence, following the digital filtration
stage, a decimation process is generally needed, for removing the unnecessary data,
and ease of data processing. In sigma-delta modulator systems, generally the
decimation and filtering are carried out in the same stage, which decreases the
computation time during digital filtering.

24
2.5. Electromechanical Sigma – Delta Modulator

Sigma-delta modulators are well suited for low frequency applications, because of
the fact that high oversampling ratio is needed for achieving high resolution. Hence,
for a mechanical structure, which has a bandwidth in the range below a few kHz, a
sigma-delta modulation would result in high resolution. Moreover, a mechanical
structure, such as an accelerometer, is a second order integrator by itself, and hence
directly can be used as the integrator stage in the sigma-delta modulator structure.
As shown in Figure 2.7, the input to the system is acceleration and the feedback is
applied in the form of acceleration, with the application of an electrostatic feedback
force. The measured parameter from the mechanical sensor is the position. Hence,
the usage of a second order mechanical system together with the readout circuit in
closed loop, results in a second order sigma-delta loop to be formed. For reading
the sensor mass position, generally capacitive techniques are used, where a
capacitive change occurs depending on the position of the mass of the mechanical
sensor. For reading out the capacitance, an amplifier is needed. For operating at
high oversampling ratios, the slew rate and gain bandwidth product of this amplifier
carries high importance.

Accelerometer

A x Y
∑ ∫∫

Af Force
Feedback

Figure 2.7: The block diagram of the closed loop electromechanical system.

25
In sigma-delta modulators, multi-bit structures are not preferred commonly due to
more complex readout circuit, and the non-linearity of the quantized levels of analog
to digital converter. Thus, a single bit output is commonly used, in which the
circuitry becomes much simpler and the non-linearity problem caused by the A/D
conversion does not exist because of the two-level force feedback. Using a single
bit output results in using a single comparator at the output stage following the
amplifier. The feedback is applied to the mechanical sensor according to the output
generated at the comparator stage. Electrostatic feedback force is applied in either
direction of the mechanical sensor, which corresponds to an acceleration value
applied to the mechanical mass. In the sigma-delta loop, the displacement of the
mechanical mass is held around zero, with feedback pulses at high sampling rate,
which results in a motion of the mechanical mass around zero displacement. This
motion is called mass residual motion, which is one of the noise sources of the
electromechanical sigma-delta modulators. Mass residual motion will be discussed
in detail in Chapter 3 and 4.

2.6. Conclusion

This chapter gave a brief explanation of sigma-delta modulators and their significant
properties, such as oversampling and noise shaping. Sigma-delta modulation is well
suited for the accelerometer application, because of the low bandwidth of
acceleration signals, and the integration of the sensor into the loop, forming an
electro-mechanical sigma-delta structure. With the use of high oversampling ratios,
it is possible to carry the quantization noise to high frequencies, and obtain a high
resolution system. The next chapter gives the modeling of the sigma-delta system,
together with the readout circuit and the accelerometer, and discusses the stability of
the sigma-delta loop.

26
 CHAPTER 3

3. MODELING & SIMULATIONS

This chapter gives the modeling of the mechanical sensor and the readout circuit in
different environments and the simulations performed for understanding the
operation of electromechanical sigma-delta modulator system. In section 3.1, a
detailed explanation of the accelerometer mechanical system is presented, including
the clarification of the parameters included in the transfer function of the system,
such as spring constant and damping. Then, the accelerometer models constructed
in Cadence environment is explained. Section 3.2 gives the complete system model
generated in Matlab Simulink environment. Lastly, in Section 3.3, the closed loop
analysis and stability concerns are discussed.

3.1. Modeling the Accelerometer

Both of the sensor structures used in this research has similar lateral comb finger
structures to obtain capacitive sensing. Figure 3.1 illustrates the structure of Silicon-
On-Glass (SOG) accelerometers designed in the research group, together with the
accelerometer parameter dimensions.

There are two capacitances formed in this structure, between the two electrode
fingers and the proof mass fingers. The capacitance value between one of the
electrodes and the proof mass is given as,

𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
𝐶1,2 = + (3.1)
𝑑1 𝑑2

27
where, N is the number of fingers on one side of the accelerometer, 0 is the
permittivity of the air, A is the area of the overlapping area of one of the proof mass
and electrode finger pair, d1 and d2 are the smaller and larger gaps in between the
proof mass and electrode fingers.

(a)

Finger overlap length (L)

Electrode 1

Electrode 2
Finger antigap (d2)
F1 F2
Finger gap (d1) Proof Mass
(M) Finger gap (d1)
Finger antigap (d2)
Finger overlap length (L)

(b)
Figure 3.1: (a) The structure of the SOG accelerometer designed in METU MEMS VLSI
research group. (b) Illustration of the accelerometer dimensions.

The smaller distance between the fingers, d1 is called the gap and the larger distance
d2 is called the anti-gap, because of the existence of the anti-gap decreases the
sensitivity of the sensor. When a displacement in the position of the proof mass
occurs, the gap and anti-gap values change, hence the capacitance values can be
given as,

28
 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
𝐶1,2 = + (3.2)
𝑑1 ∓ 𝑥 𝑑2 ± 𝑥

where, x is the displacement of the proof mass from the rest position. For small
values of displacements, the capacitance variation can be assumed linearly
proportional to the displacement x. The linearity under small displacement can be
observed by taking the first two terms of Taylor expansion as follows,

𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
𝐶1,2(𝑥→0) ≅ + + 2 − 𝑥 (3.3)
𝑑1 𝑑2 𝑑1 𝑑2 2

If we take the derivative of the capacitance with respect to x, the capacitive

sensitivity of the sensor is obtained, dependent on the displacement as in
Equation (3.4). In this equation, it is observed that the sensitivity is highly
dependent on the finger gap spacings. Hence, smaller the finger gap spacings,
higher the sensitivity of the accelerometer.

𝜕𝐶1,2 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
= 2
− (3.4)
𝜕𝑥 𝑑1 ∓ 𝑥 𝑑2 ± 𝑥 2

Another important parameter of the accelerometer sensor is the electrostatic force

generated when potential is applied between the fingers. The significance of this
parameter comes from the closed loop operation and applied feedback. The
electrostatic pulling force generated between two conducting objects is given by,

1 𝜕𝐶 2
𝐹= 𝑉 (3.5)
2 𝜕𝑥

29
Hence, the force generated between the proof mass and the electrodes becomes as in
Equation (3.6). The corresponding acceleration caused by the electrostatic force is
given in Equation (3.7), in terms of g‟s.

1 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴 2
𝐹= 2
+ 𝑉𝑒𝑙𝑐 1,2 − 𝑉𝑝𝑓𝑚 (3.6)
2 𝑑1 ∓ 𝑥 𝑑2 ± 𝑥 2

𝐹 1 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴 2
𝐺= = 2
+ 𝑉𝑒𝑙𝑐 1,2 − 𝑉𝑝𝑓𝑚 (3.7)
𝑚𝑔 2𝑚𝑔 𝑑1 ∓ 𝑥 𝑑2 ± 𝑥 2

3.1.1. Accelerometer Transfer Function

The accelerometer can be modeled as a basic mass spring damper system as shown
in Figure 3.2.

K B

x m

Figure 3.2: Mass-spring-damper system.

For an external acceleration of a applied to the sensor, the equations of motion can
be shown as in equation (3.8). In this equation x is the displacement of the proof
mass according to a reference frame placed on the body of the accelerometer, where
the electrodes and the anchor points are fixed.

30
 𝑑2 𝑥 𝑑𝑥
𝐹 = 𝑚𝑎 = 𝑚 2
+𝐵 + 𝐾𝑥 (3.8)
𝑑𝑡 𝑑𝑡

The Laplace transform of the equation gives the Equation (3.9). Hence, the transfer
function of the accelerometer is obtained as in Equation (3.10).

𝑚𝐴 𝑠 = 𝑚𝑠 2 𝑋 𝑠 + 𝐵𝑠𝑋 𝑠 + 𝐾𝑋 𝑠 (3.9)

𝑋 𝑠 1
𝐻 𝑠 = = 2 (3.10)
𝐴 𝑠 𝑠 +𝑠 𝐵 𝑚 + 𝐾 𝑚

The equations (3.11) and (3.12) give the resonance frequency and the quality factor
of the accelerometer, respectively.

𝐾
𝑤0 = (3.11)
𝑚

𝑤0 𝑚
𝑄= (3.12)
𝐵

The magnitude of the accelerometer response is almost constant in the frequency

band below the resonance frequency, as shown in Figure 3.3. Hence, the
accelerometer is operated in this linear region. Figure 3.3 also shows the
dependence of the quality factor on the transfer function. As the quality factor is
increased, the peak at the resonance frequency increases, which is the case generally
obtained if the mechanical sensor is operated in vacuum environment.

31
Magnitude (dB)

Q = 26

Q = 2.6

Q = 0.26 (@ atmospheric pressure)

Phase (degree)

101 102 103 104 105

Frequency (Hz)

Figure 3.3: Magnitude and phase response of an accelerometer with different quality
factors.

In the linear region of the transfer function, at low frequency band, the magnitude of
the response is given by Equation (3.13). Hence, the capacitive sensitivity of the
accelerometer in open loop is calculated as given in Equation (3.14).

𝜕𝑥 𝑚𝑔
𝐻 0 = = (3.13)
𝜕𝑔 𝐾

𝜕𝐶 𝜕𝑥 𝜕𝐶 𝑚𝑔 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
= = 2
+ (𝑝𝐹/𝑔) (3.14)
𝜕𝑔 𝜕𝑔 𝜕𝑥 𝐾 𝑑1 ∓ 𝑥 𝑑2 ± 𝑥 2

32
3.1.1.1. Spring Constant

The accelerometer sensor uses beam type springs, which differ in size and structure.
The beam types used in the accelerometers in this research have the folded beam
type structure, as shown in Figure 3.4.

Figure 3.4: Folded beam spring structure.

The spring constant of this beam structure is given as,.

1 𝐸𝑕𝑤 3 𝐸𝑕𝑤 3
𝐾= = (3.15)
2 4𝑙 3 8𝑙 3

where, E is the Young‟s Modulus, h is the thickness, w is the width, and l is the
length of the cantilever beam.

3.1.1.2. Damping

In MEMS devices the two main sources of damping is “Squeeze Film Damping”
and “Couette-Flow Damping”. Squeeze film damping occurs when two parallel
plates move towards each other and compress the fluid molecules in between.
Couette-Flow damping results from the movement of two parallel plates in a sliding
action. Figure 3.5 illustrates the squeeze film damping and the couette flow
damping.

33
 (a)

(b)

Figure 3.5: Illustration of (a) squeeze film damping, (b) Coutte flow damping.

In the accelerometer structure used in this research, squeeze film damping results
from the fingers, where a varying gap structure is used. The movement of fingers
does not cause Couette flow damping, however, sliding of the proof mass over the
substrate results in a Couette flow damping. Nevertheless, the Coutte flow damping
has an ignorable effect when compared with the squeeze film damping, in this
structure. Therefore, the total damping constant of the accelerometer can be
expressed by only squeeze film damping as given in equation (3.16) [37]

𝑊3𝐿 3
2𝑁 2 𝑁 − 1
𝐵𝑠𝑞𝑢𝑒𝑒𝑧𝑒 = 𝑐𝜇𝑒𝑓𝑓 = 𝑐𝜇𝑒𝑓𝑓 𝑊 𝐿 + (3.16)
𝑑3 𝑑13 𝑑23

where, µeff is the effective viscosity of the environment, W is the width, L is the
length of the rectangular plates, d is the distance between two rectangular plates, d1
and d2 are the gap and the anti-gap spacings of the fingers, and c is a factor, which
depends on the W/L ratio of the rectangular plates. The dependence of the factor c
on the W/L ratio is given in the graph shown in Figure 3.6 [37].

34
 1.0

0.9

0.8

0.7

c
0.6

0.5

0.4
0 0.2 0.4 0.6 0.8 1.0
W/L

Figure 3.6: The dependence of coefficient c on the dimensions of the plates.

3.1.2. Cadence Modeling of Accelerometer Sensor

In order to observe and simulate the full accelerometer system, both the mechanical
sensor and the interface electronics should be modeled in the same environment.
One of the ways to implement the model is using the Cadence environment, where
the electronic circuitry already exists. In this case, the mechanical structure should
be modeled by electrical parameters. Another way is modeling both the interface
electronics and the mechanical structure in a seperate environment, such as Matlab.

To obtain an accurate model and simulation results, the accelerometer sensor is

modeled in Cadence environment. In this fashion, the overall system could be
simulated in the same environment.

The block diagram of the accelerometer model is given in Figure 3.7. The blocks
are generated using VerilogA language, except the transfer function block of the

35
accelerometer. The transfer function is constructed using a second order RLC
circuit.

The model is mainly composed of two variable capacitors, the transfer function of
the mechanical sensor and the blocks to convert the variables, using the defined
equations.

Figure 3.7: Block diagram of the model constructed in Cadence environment.

There are two variable capacitors having values dependent on the accelerometer
dimensions and structure, and the instantaneous value of the displacement of the
proof mass. The variable capacitors are the interface between the mechanical
structure and the electrical circuitry. Variable capacitors are generated using
VerilogA language, in which the capacitance value is set by an input value. The
capacitance value is calculated in “X_to_C” block by using the formula given in
Equation (3.2).

The displacement x is obtained as an output from the transfer function of the

accelerometer. The transfer function block takes total acceleration value as an input,

36
which is the sum of external acceleration and the acceleration caused by the
electrostatic feedback. For obtaining the second order transfer function of the
accelerometer, an RLC circuit is constructed. The RLC circuitry for obtaining the
transfer function is given in Figure 3.8.

Figure 3.8: RLC circuit constructed for modeling the transfer function of the accelerometer.

For both the input and output of the transfer function block, voltage values are used
to represent the acceleration and the displacement. There is also a gain block to set
the magnitude of the transfer function properly. For finding the component values,
the transfer function of the RLC circuit should be equated to the transfer function of
the accelerometer, which is given in Equation (3.17), where acceleration is in terms
of g‟s.

𝑋 𝑠 𝑔 𝐴
= 2 = 2 (3.17)
𝐺 𝑠 𝑠 +𝑠 𝐵 𝑚 + 𝐾 𝑚 𝑠 (𝐿𝐶) + 𝑠 𝑅𝐶 + 1

From this equation, solving for the R, L and C values gives infinite number of
solutions. Hence, to obtain a solution, value of R is selected to be 1 Ω. Then, the
other parameters can be found as in the Equations (3.18), (3.19) and (3.20), given
below.

𝑚∙𝑔
𝐴= (3.18)
𝐾

37
 𝐵
𝐶= (3.19)
𝐾

𝑚
𝐿= (3.20)
𝐵

The acceleration value, which is applied to the transfer function block, has two
sources. One of the sources is the externally applied acceleration, and the other
acceleration source results from the electrostatic force generated from the applied
voltages to the accelerometer electrodes and the proof mass. These two sources of
acceleration is directly added and given to the transfer function as an input. The
external acceleration value is directly an input, where a voltage signal, having the
value of acceleration in terms of g‟s, is applied while performing the simulations.
The acceleration created because of the electrostatic force is calculated by the
“V_to_G” block, using the electrostatic force formula given in Equation (3.7). The
net acceleration generated is calculated by taking the difference of accelerations
generated on each side of the accelerometer, i.e. G1 – G2.

The accelerometer model generated in Cadence environment is used with the

designed readout circuit, and the simulations are performed. The simulations
performed in Cadence environment give accurate results, since the readout circuit is
already designed in the same environment. However, the simulation times are so
long that complete results cannot be observed. Therefore, for examining the
operation and stability of the system, Matlab Simulink models are used, which is
discussed in the following section.

3.2. Complete System Modeling in Matlab Simulink

The model of the accelerometer combined with the model of the readout circuit is
constructed in Matlab Simulink environment. The block diagram of the model is
given in Figure 3.9.

38
Figure 3.9: Simulink model of the complete electromechanical closed loop system.

In this model, 3 blocks, which are accelerometer transfer function, displacement to

capacitance converter “X to C”, and the feedback block, represents the
accelerometer. The accelerometer transfer function is modeled with a transfer
function block in s-domain, as given in Equation (3.10). The acceleration entered
into this block is composed of external acceleration and the feedback acceleration.
Feedback acceleration is generated by the applied feedback voltage to the electrodes
and the proof mass. The internal blocks of the feedback block is shown in Figure
3.10.

The feedback acceleration is generated only when the readout circuit is in feedback
phase, timings of which are set by the pulse generator. The direction of the
feedback acceleration is defined by the output of the comparator. The magnitude of
the force is calculated according to the Equation (3.6), and the net force is then
converted to acceleration.

Figure 3.10: Block diagram of the feedback block.

39
The output of the transfer function gives the displacement as the output, and the
displacement should be converted to capacitance value in order to be useful for the
readout circuit. The “X to C” block calculates the capacitance on each side of the
accelerometer using the displacement value, and gives the capacitance difference at
the output. The capacitive difference value is then used in the front-end readout
block, to be converted into voltage values in the readout circuit. The model of the
front-end readout circuit is given in Figure 3.11. The front-end readout circuit
consists of a charge integrator, which is modeled by a gain block. The gain block
converts the capacitive difference into the voltage value as given in Equation (3.21).

𝑉𝐷𝐷 𝐶𝑆1 − 𝐶𝑆2

𝑉𝑂𝑈𝑇 = (3.21)
𝐶𝑖𝑛𝑡

where Cint is the integration capacitance value used in the charge integrator block
and Cs1, Cs2 are the values of accelerometer capacitances. The saturation of the
readout circuit at the supply voltages is also included in the model using a saturation
block. The charge integrator block samples the capacitance difference, by
transferring the excess charge generated on the sensor capacitances, to the
integration capacitances. This operation is performed once in a sampling period.
This is modeled in Simulink using a sample and hold block, with proper timings.

Figure 3.11: Front-end readout circuit modeling.

40
A compensator block is constructed, the block diagram of which is given in Figure
3.12. The compensator block maintains the stability of the sigma-delta system, by
adding a zero to the loop, and carrying the oscillation of the mass to higher
frequencies. The block holds the previous output value, and subtracts a ratio of
previous data from the current output data. The performed operation is expressed in
z-domain as given in Equation (3.22). A detailed discussion on the compensator
and stability is given in the following section.

𝐻(𝑧)𝑐𝑜𝑚𝑝 = 𝐴 − 𝐵𝑧 −1 (3.22)

Figure 3.12: Block diagram of the compensator.

After the lead compensator stage, the output value is compared with zero, with the
usage of a comparator. The output of the comparator has two levels, either “+1” or
“–1”. Hence, an output bit-stream is generated at the output of the readout circuit.
This bit-stream is supplied back to the feedback block, and the loop is closed.

The output bit-stream carries the acceleration information including a great amount
of quantization noise. The quantization noise is at the high frequency, while the
acceleration information is in the low frequency band. Therefore, for observing a
meaningful output, the bit-stream should be low-pass filtered, using the digital low
pass filtration blocks of Matlab Simulink. The low-pass filter should be a sharp
filter to eliminate the quantization noise; hence, it will have a complex structure

41
with very high order. However, using a complex filter structure increases the
simulation time incredibly. For decreasing the simulation time, three stages of
filtering stages are cascaded, while decimating the signal at each stage, as shown in

Figure 3.13. In this way, the number of data is reduced at each stage, and much
simpler low-pass filter blocks are used.

Decimation Decimation Decimation

Digital LP Digital LP Digital LP

filter filter filter

Figure 3.13: Filtering and decimation blocks.

3.3. Closed Loop Response and Stability Analysis

When an accelerometer system is operated in open loop, the bandwidth of the

system is defined by the bandwidth of the accelerometer and the measurement range
is limited by the finger spacings. According to the Equation (3.14), for increasing
the open-loop sensitivity of the accelerometer, it should be designed to have large
mass, low spring constant and high capacitance values. All these parameters have
effects on decreasing the bandwidth and the range of the system. Moreover, the
change of capacitance with respect to the displacement can be assumed to be linear
if and only if the displacement value is much smaller than the finger gap spacings.
Hence, usage of an open-loop accelerometer results in high amounts of non-linearity
of capacitance with respect to the external acceleration.

By using the accelerometer with a closed loop circuitry, the position of the
accelerometer is held around the rest position. The output of the system becomes
the value of applied feedback force to hold the proof mass position constant. Hence,
the output of the system is directly proportional to the applied acceleration, and non-
linearity is not observed. In closed loop mode, the range of the accelerometer is not

42
anymore limited with the finger gap spacings, but the maximum applicable feedback
acceleration on the proof mass. For an accelerometer having a low spring constant,
closed loop mode results in higher measurement range. Furthermore, in the closed
loop system, the bandwidth is not anymore dependent on the accelerometer transfer
function. The closed loop bandwidth depends on the closed loop parameters and
hence the bandwidth of the system increases in a considerable amount.

The theoretical calculation of the bandwidth of a sigma-delta closed system is not an

easy process, and it can vary according to the applied external acceleration. For
considering the bandwidth of the system, Figure 3.14 should be considered where a
basic block diagram of the sigma-delta loop is given.

Accelerometer

A 1 x Y
∑ 2
s +Bs/m+K/m

Af Force Feedback
(KFF)

Figure 3.14: Block diagram of the sigma – delta loop.

In this figure, the comparator block is a non-linear element, which gives an output
with two levels. The average of the comparator output pulses will converge to the
external acceleration value, however the displacement of the proof mass is held in a
small value. Hence, this results in a high gain value at the comparator stage. The
total gain  from displacement x to feedback acceleration Af, can be several orders
of magnitude larger than (K/m). Hence according to the Equation (3.23), the
resonance frequency of the closed response is carried to a high frequency value with

43
the addition of . Then, the total closed loop transfer function can have a bandwidth
several orders larger than the open loop bandwidth of the accelerometer.

𝑌(𝑠) 1 1
= 2
(3.23)
𝐴(𝑠) 𝐾𝐹𝐹 𝑠 + 𝐵 𝑚 𝑠 + 𝐾 𝑚 + 𝛽

where KFF is the factor of conversion of electrostatic voltage to feedback

acceleration, and  is a nonlinear function resulting from the comparator and the
feedback stages. In the above analysis of the bandwidth of the closed loop system,
the delay resulted from the readout circuit is assumed to be zero, which is not the
actual case. For calculating the delay caused by the readout circuit, the time
between the capacitance readout to the midpoint of feedback time should be
considered. Using this delay time, td, the phase response of the readout can be given
as in Equation (3.24) .

𝜃𝑟𝑒𝑎𝑑𝑜𝑢𝑡 = −180 ∙ 𝑡𝑑 ∙ 𝑓 (3.24)

The accurate way of determining the bandwidth of the closed loop system is
observing the output magnitude of the system by sweeping the applied acceleration
frequency. A sample result is obtained with a readout circuit of 2s sampling period
and 1.2s readout delay. For obtaining the frequency response of the closed loop
system, accelerations at 1g amplitude at various frequencies are applied to the
system and the results are as shown in Figure 3.15. The magnitude of the output at
the input frequency is observed with proper filtering, and hence the frequency
response is obtained. The result shows that the magnitude of the frequency response
is kept constant up to a frequency much higher than the resonance frequency of the
accelerometer. Increasing the input frequency results in increasing the contributed

44
quantization noise to the signal, hence the response becomes distorted at higher
frequencies.

2,5
Normalized magnitude

1,5

0,5

0
10 100 1000 10000 100000
Frequency (Hz)

Figure 3.15: Closed loop frequency response of the electromechanical system.

The closed loop system aims to hold the proof mass by applying a pulse stream of
feedback force, which creates an oscillation on the proof mass position. The
generated oscillation is randomized by the changing value of the input, and hence,
creates an undeterminable motion of the proof mass. This motion is called Mass
Residual Motion.

For analyzing the closed loop operation and mass residual motion, one should
consider the oscillation conditions. There are two conditions for oscillation in a
closed loop feedback system. The Figure 3.16 shows the block diagram of the basic
feedback circuitry. The first condition of oscillation is the magnitude condition,
where the magnitude of the loop gain H(jw).K should be set to unity. The second
condition is the phase condition, where the total phase of the closed loop system
must be set to 360 degrees.

45
 x e y
H(jw)
z

Figure 3.16: Block diagram of a basic feedback system.

For the analysis of oscillation, the input acceleration is assumed zero, and proof
mass oscillation is considered. In the sigma-delta closed loop system, a single bit
feedback is applied as feedback to the accelerometer. The applied feedback has two
levels, without regarding the amplitude of the displacement of the proof mass. This
creates an infinite gain element, which has non-linearity due to the saturation.
Independent of the amplitude of the displacement of the proof mass, the feedback
force to be applied will remain the same. Hence, in this case, the amplitude
condition settles to the desired value of unity automatically.

Since the amplitude condition is already satisfied by the system, the main concern
for oscillation is the phase of the system. 180 degrees of phase difference results
from the negative feedback. Hence, the condition that should be satisfied can be
given as in Equation (3.25). There are three sources of phase difference, which are
the accelerometer itself, the phase of the readout circuit and the lead compensator
circuitry. The lead compensator is used for adding a zero to the system and hence
carrying the proof mass oscillation frequency to a higher value. As the mass
residual motion is carried to higher frequencies, the amplitude of the oscillation
decreases and the contribution of this motion into the signal band are reduced.

46
 𝜃𝑡𝑜𝑡𝑎𝑙 = −180° + 𝜃𝑠𝑒𝑛𝑠𝑜𝑟 + 𝜃𝑟𝑒𝑎𝑑𝑜𝑢𝑡 + 𝜃𝑐𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑜𝑟 = −360° (3.25)

From the transfer function of the accelerometer, the phase of the mechanical sensor
can be extracted as given in Equation (3.26). The most part of the phase is caused
by the accelerometer, which converges to 180 degrees at high frequencies.

𝐵
1 360 𝑚 𝑤
∠𝐻 𝑗𝑤 = ∠ =− 𝑡𝑎𝑛−1 (3.26)
𝐵 𝐾 2𝜋 𝐾
𝑗𝑤 2 + 𝑚 𝑗𝑤 + 𝑚 𝑚−𝑤
2

As discussed previously, the readout phase delay is given by the time between the
sampling of the displacement value through the measurement of capacitances, and
the midpoint of the feedback time. Hence, the readout phase was given as in
Equation (3.24). The phase of the readout circuit starts to become considerable at
much higher frequencies.

The lead compensator is an important block for the stability of the system and
decreasing the mass residual motion. It has a structure, which can be shown in
z-domain as given in Equation (3.22). The phase of the lead compensator is
expressed as shown in Equation (3.27). The parameters A and B should be defined
according to the needs of the closed loop system. If the oscillation frequency is
assumed much larger than the accelerometer resonance frequency, the phase lag of
the accelerometer can be assumed 180 degrees. Hence, to obtain 180 degrees in
total, the sum of the phases of readout circuit and lead compensator should be equal
to zero, i.e. the phase lag of the readout circuit should be equal to the phase lead
value of the compensator.

47
 2𝜋𝑓
360 𝐵 ∙ 𝑠𝑖𝑛
𝑓𝑠
𝜃𝑐𝑜𝑚𝑝 = ∙ 𝑡𝑎𝑛−1 (3.27)
2𝜋 2𝜋𝑓
𝐴 − 𝐵 ∙ 𝑐𝑜𝑠
𝑓𝑠

There is an achievable limit of oscillation frequency, which is the quarter of the

sampling frequency. This maximum oscillation frequency results from the second
order nature of the accelerometer. For a basic second order system, having zero
damping and spring constant, used in the sigma delta structure, the signals are
illustrated in Figure 3.17.

-10
x 10 Accelerometer Model
5
displacement

-5
0 0.5 1 1.5 2 2.5
time -5
x 10
-4
x 10
5
velocity

-5
0 0.5 1 1.5 2 2.5
time -5
x 10
feedback acceleration

50

-50
0 0.5 1 1.5 2 2.5
time -5
x 10

Figure 3.17: Basic illustration of displacement, velocity and feedback acceleration for a
sigma delta loop.

48
By analyzing the signal waveforms, it can be said that the minimum period of the
oscillation is four times the sampling period. This fact can be explained by
examining the waveforms. Initially, assume that the displacement of the proof mass
is slightly above zero position. A feedback force will be applied on the opposite
direction, and give acceleration to the proof mass. The applied acceleration will
cause the velocity of the proof mass to increase, and stay constant after the
acceleration. The next feedback pulse should be in the opposite direction with the
previous one, because of the proof mass is moved to one side with the effect of the
previous feedback pulse. The second feedback pulse will only make the velocity of
the proof mass zero, but cannot make it start to move on the opposite direction.
Hence, third feedback pulse will be in the same direction with the second one, and
will move the proof mass to zero position. The next two feedback pulses will be on
the opposite direction. Hence, the cycle repeats in four feedback pulses. This
results in a maximum oscillation frequency of quarter of the sampling frequency.

The phase responses of the accelerometer and readout blocks are given in Figure
3.18. In the given graphs, the readout circuit is taken to have 500 kHz sampling
frequency and 1.2 µsec readout to feedback delay and some compensator responses
are shown with different coefficients.

The total phase of the system without the compensator is given in Figure 3.19. In
this configuration, the oscillation settles at the frequency around 32 kHz, which
results to have high amplitude of displacement. With the usage of a lead
compensator with the proper coefficients, the phase response of the closed loop
system becomes as shown in Figure 3.19. Hence, in this case, the frequency of
oscillation increases up to fs/4.

The main concern about the mass residual motion is its contribution on the noise in
the signal band. The calculation of the noise due to mass residual motion is given in
the next chapter.

49
 0

Readout Phase (deg)

-50

-100

-150
0 1 2 3 4 5
10 10 10 10 10 10
Frequency (Hz)

0
Sensor Phase (deg)

-50

-100

-150

0 1 2 3 4 5
10 10 10 10 10 10
Frequency (Hz)

60 9-8z-1
5-4z-1
3-2z-1
Compensator Phase (deg)

40
2-z-1

20

-20

-40
0 1 2 3 4 5
10 10 10 10 10 10
Frequency (Hz)

Figure 3.18: The phase responses of the blocks in the closed loop system.

50
 (a)

0
Total phase with compensator (deg)

-50

-100

-150

-200 X: 1.264e+005
Y: -180
-250

-300
0 1 2 3 4 5
10 10 10 10 10 10
Frequency (Hz)

(b)

Figure 3.19: Total phase of the of the system (a) without the lead compensator (b) with the
lead compensator.

51
3.4. Conclusion

This chapter gave the model of the accelerometer and the readout circuit constructed
in Cadence and Matlab environments. Then, closed loop stability analysis of the
system is given, and the necessity of the lead compensator structure is explained.
Using these models and analysis a readout circuit is designed, which is described in
next chapter. The next chapter also gives the performance limitations of the system
according to the designed readout circuit.

52
 CHAPTER 4

4. READOUT CIRCUIT DESIGN

The displacements of an accelerometer sensor are commonly in the order of an

angstrom, and the capacitive differences are below atto-farads. In order to detect
such small capacitive differences, a high-resolution readout circuit is required. The
most effective solution for non-monolithic capacitive accelerometers is the use of a
switched capacitor sigma-delta type readout circuit, to achieve high resolution and
high linearity with digital output. The aim is to design a closed loop, low noise
readout circuit, with low power consumption. Since the accelerometers used in this
study have capacitive sensitivities around 1 pF/g, to achieve micro-g resolution, the
readout circuit should be capable of detecting below 1 atto-farad capacitive
differences. Thus, an electro-mechanical sigma-delta readout circuit is designed and
implemented. In later study, another regular sigma-delta structure is designed and
implemented with minor improvements and addition of a compensator circuit in the
first design. This second design achieved significant results in terms of noise
performance, together with the sensor, both in open loop and closed loop mode.
With the gathered knowledge and experience on sigma-delta readout circuits, two
new readout circuits employing adaptive sigma-delta techniques are designed, to
improve the performance of the accelerometer systems, further. The design of the
adaptive readout circuits will be discussed in the next chapter. This section
describes the second regular sigma-delta readout circuit design.

In Section 4.1, the general structure and operation of the readout circuit is explained.
Section 4.2 describes internal blocks of the readout circuit, supplying the circuit
schematics and simulation results. Section 4.3 gives the overall simulation results of
the readout circuit with the accelerometer models in Cadence and Matlab
environments. Section 4.4 gives information on layout considerations, and in

53
Section 4.5 performance limitations of the readout circuit and the accelerometer
system is described. Finally a conclusion is made in Section 4.6.

4.1. Structure of the Readout Circuit

The complete block diagram of the sigma-delta readout circuit is given in Figure
3.2. The mechanical sensor is included in the sigma-delta loop and hence the
overall system constitutes an electromechanical structure. Besides using readout
circuit in the closed loop, it can also be operated in open loop mode. In open loop
mode, in one sampling period, the readout circuit measures the capacitance
difference of the accelerometer and gives a differential analog output voltage and no
feedback force is applied to the sensor. In closed loop mode, there are two phases in
a sampling period, which are sense and feedback phases. The capacitive difference
is sensed in the sense phase, and an analog voltage is formed. Then, the differential
analog voltage is quantized in two levels using a comparator, and according to the
result, an electrostatic feedback force is applied to the proof mass, in the feedback
phase of the sampling period.

VDD

VSS ΦC
Charge Integrator

CRef2 CRef1
Comparator

Accelerometer

ΦC VDD FEEDBACK

VSS

Multiphase Bias
CLK START-UP
Clock Generator Generator

Figure 4.1: Block diagram of the fully differential sigma-delta readout circuit.

54
Oversampling ratio of a sigma-delta system should be high to achieve high
resolution. Therefore, the components of the readout circuit should be designed to
operate at the highest possible frequency. Designed readout circuit operates at
500 kHz sampling frequency, using its on-chip clock generator circuitry, which can
be increased up to 700 kHz by applying an external clock source. Hence, the speeds
of the components are designed to be able to operate with this frequency. Using this
clock source, multiphase clocks are generated for operating various parts of the
circuit and the switch timings.

The readout circuit has a fully differential structure for reducing the effects of
common mode noise sources, such as the power supply noise. The initial stage of
the readout circuit is a charge integrator circuitry, including a switch capacitor
network. The switch capacitor network, together with the charge integrator,
provides the measurement of the sensor capacitor values by charge transferring.
The capacitive difference on the accelerometer is converted to differential analog
voltage at the charge integrator stage, which uses an Operational Trans-conductance
Amplifier (OTA) for performing the operation. In this stage, correlated double
sampling (CDS) technique is also employed, for reducing the low frequency noise
and offset at the input stage of the charge integrator.

The differential voltage at the output of the charge integrator is then supplied to the
compensator stage, which is needed for the stable operation in closed loop mode.
After the compensator, a latching comparator is used for deciding on the sign of the
differential signal. The output of the comparator is used as the closed loop output of
the readout circuit. It is also used for applying feedback voltages between the
accelerometer electrodes and proof mass.

The readout circuit also includes a start-up circuitry, which is used to bring the proof
mass of the sensor to the rest position in the case of large deflections at start-up.

55
4.2. Description of the Main Blocks of the Readout Circuit

This section describes the readout circuit blocks including the simulation results.
There are mainly four blocks of the readout circuit, first of which is the charge
integrator block. The OTA used in the charge integrator block is one of the most
critical components, which enormously affects the performance parameters of the
readout circuit. The lead compensator used following the charge integrator block
satisfies the stability of sigma delta loop, and decreases the mass residual motion by
adding a zero to the system and carrying mass residual motion frequency to a higher
value. Another critical component is the comparator block, which must operate
precisely, and at a high rate.

4.2.1. Switched – Capacitor Network

The switched capacitor circuitry shown in Figure 4.5 is used in order to perform
both readout and feedback, in different time intervals. This circuitry also employs
correlated double sampling (CDS) to decrease the low frequency noises of the
circuit.

There are two kinds of switch structures used in the switched capacitor network.
The first kind of switch is the complementary CMOS switch as shown in Figure 4.2.
This kind of switch is used in stationary switches, which are not switched during the
operation of the readout circuit. Because of these switches are not operational
during the circuit operation, there is no charge injection problem. However, the
problem to be considered is the resistances and parasitic capacitances of the
switches. At high current passing nodes of the readout circuit, the W/L ratios of the
transistors are designed large, so that the resistance of the switch is kept low.
However increasing the gate area of the transistors increases the parasitic
capacitances. Hence, the switch transistors have the minimum length, with large
widths. The W/L ratios of switches change depending on the connected node and
the current passing through the switches. The resistance of a switch depending on
its W/L ratio is given in Figure 4.3.

56
 clk

PMOS

VDD

Gnd

NMOS

clk

Figure 4.2. Complementary CMOS switch.

1400

1200
Switch Resistance (Ohm)

1000

800

600

400

200

0
0 50 100 150 200
W/L
Figure 4.3. Switch resistance versus W/L ratios of the transistors.

The second type of switch is the complementary switch with dummy switches, as
shown in Figure 4.4. This switch is used for the switch structures, which are
functional during the readout circuit operation. To overcome the charge injection
and charge feed through problems, dummy switches are used. The dummy switches
M3 – M6 have the half size of the switch transistors M1 and M2. By applying the
inverse of the control signals to the dummy switches, the injected charges are
collected on the dummy transistors.

57
 clk clk clk

PMOS PMOS PMOS

VDD VDD VDD

Gnd Gnd Gnd

NMOS NMOS NMOS

clk clk clk

Figure 4.4. Complementary CMOS switch with dummy transistors.

4.2.1.1. Readout and Feedback Switching Timings

The switch capacitor network is shown in Figure 4.5. For understanding the
operation of switching and charge transfers, first assume that the Correlated Double
Sampling (CDS) technique is not applied, which means “pcds” and “pcdsn”
switches are short-circuited and “phcd” switch is kept off. The timings of the
signals applied to the switches are shown in Figure 4.6.
VDD

VSS

Rs
phf1
pcdsn
Cint1
VDD

VSS

CR1 CR2
M2 M1 Gd phcd pcds
CCDS
Out+
pfdn
pfdn
Out-
CCDS
İvmeölçer

M1 M2 Gd phcd pcds
CS1 CS2
VDD

VSS

Cint2
pcdsn
phf2
Rs
VDD

VSS

Figure 4.5: Schematic view of the switched capacitor network and the charge integrator
stages.

58
 pfd

pfdn

phf1

phf2

Rs

Figure 4.6. Timing diagram of the signals used in the readout circuit.

The readout operation has two main phases of operation, readout and feedback
phases. In the readout phase, the charge due to the sensor capacitance difference is
transferred to the charge integrator and a proportional output voltage is generated.
Then, after the comparator block settles to a valid output, the circuit enters to
feedback phase. In the feedback phase, the sensor is disconnected from the charge
integrator and an electrostatic voltage is applied to the sensor electrodes.

At the beginning of the readout phase, “Rs”, “pfdn” and “Gd” switches are on.
Hence, the integration capacitors, Cint, are reset and the input and output nodes of
the OTA are set to mid-point voltage. After switching “Rs” and “Gd” off, the
voltages at the top and bottom nodes of the full bridge capacitor structure switches
from high to low and low to high, respectively. At this instant, charge transfer to the
integration capacitors takes place. Hence, at the charge integrator stage, the
differential output voltage related with the full bridge capacitance difference is
generated. After this generated voltage being stable and read by the comparator, the
circuit enters to feedback phase. When the circuit is in feedback phase, the
connection between the readout blocks and the sensor is cutoff by switching “pfdn”
switch off. Before starting to apply the feedback force, the top and bottom nodes of
the full bridge structure switch their states, and stay constant during the feedback
phase. Then, feedback voltage is applied at the electrodes of the sensor. During the

59
feedback, the proof mass voltage is held at 0 V. According to the last generated
output of the comparator, 5 V is applied to one of the electrodes of the sensor, which
creates a pulling force from one of the electrodes of the accelerometer. At the end
of the feedback phase, “Gd” switch is turned on, so the electrode nodes are pulled
to mid-point voltage. Then, the readout phase starts again by turning “pfdn” switch
on.

4.2.1.2. Correlated Double Sampling

CDS technique is a useful technique for decreasing the low frequency noises and the
offset created at the input node of the OTA. The switching timings for CDS
operation are given in Figure 4.7. While the circuit is in feedback phase and the
integration capacitors are reset, the output voltage of OTA is sampled on the CDS
capacitors. The sampled voltage includes any noise and offset created by the
amplifier. When the readout is performed, the CDS capacitors are connected series
to the input of the OTA, hence the sampled offset and noise is subtracted from the
signal in the readout phase. Hence, the sampled low frequency noise and the offset
are reduced. Although the addition of CDS switches and capacitors contributes to a
little amount of noise on the readout circuit, the CDS technique improves overall
noise performance of the readout circuit.

pfd

phcd

phcdn

Figure 4.7. Complementary CMOS switch with dummy transistors.

60
4.2.2. Charge Integrator Operation

The charge integrator circuit constitutes the first stage of the readout circuit. In this
part, only the operation of the charge integrator is discussed with the basic
configuration as given in Figure 4.8. The switched capacitor network, placed before
the charge integrator stage, will be discussed in the following section.

Figure 4.8 shows the full bridge capacitor structure with the fully differential charge
integrator. The full bridge capacitors are composed of two sensing capacitors of the
accelerometer and two reference capacitors which are chosen to have the same value
with the sensor rest capacitances.

The charge integrator block includes a fully differential OTA, which drives the
integration capacitors, Cint. It also includes a bias generator for biasing the
amplifier. The function of the charge integrator block is to convert the capacitance
difference of the accelerometer to a differential analog voltage at the output nodes of
the OTA.

Charge Integrator

CR1 CR2
Cint

VS1 VS2
VOUT

CS1
CS2 Cint

Figure 4.8. Basic schematic of the charge integrator circuit.

61
The operation principle of the charge integrator block is based on charge transfer
from the bridge capacitors to the integration capacitors. Square wave is applied to
the top and bottom electrodes of the bridge configuration. The square wave applied
to the bottom node is in the opposite phase with the square wave at the top node.
The readout is performed at the instant of the applied square waves change states.
Let us now consider the time when the signal at the top node switches from high to
low; hence, the bottom node signal switches from low to high. Before the switching
time, assume that all the capacitor voltages are set to VDD/2, which is 2.5 V. After
the switching, the charge must be conserved. The Equations (4.1) to (4.4) gives the
initial and final charges of each capacitor.

𝑄𝐶𝑅1,𝑖 = 𝐶𝑅1 × 𝑉𝐷𝐷 2  𝑄𝐶𝑅1,𝑓 = 𝐶𝑅1 × −𝑉𝑆1 (4.1)

𝑄𝐶𝑅2,𝑖 = 𝐶𝑅2 × 𝑉𝐷𝐷 2  𝑄𝐶𝑅2,𝑓 = 𝐶𝑅2 × −𝑉𝑆2 (4.2)

𝑄𝐶𝑆1,𝑖 = 𝐶𝑆1 × 𝑉𝐷𝐷 2  𝑄𝐶𝑆1,𝑓 = 𝐶𝑆1 × −𝑉𝑆1 (4.3)

𝑄𝐶𝑆2,𝑖 = 𝐶𝑆2 × 𝑉𝐷𝐷 2  𝑄𝐶𝑆2,𝑓 = 𝐶𝑆2 × −𝑉𝑆2 (4.4)

Here, because of the high amplifier gain, the input nodes of the amplifier will be
held at the same voltage. After equating the voltages at these nodes, the charge
transferred QT to the integration at each node is calculated as in Equations (4.5) to
(4.7).

𝑉𝑆 = 𝑉𝑆1 = 𝑉𝑆2 (4.5)

𝑄𝑇1 = 𝑄𝐶𝑆1,𝑖 − 𝑄𝐶𝑆1,𝑓 − 𝑄𝐶𝑅1,𝑖 − 𝑄𝐶𝑅1,𝑓

(4.6)
= 𝐶𝑆1 𝑉𝐷𝐷 2 − 𝑉𝑆1 + 𝑉𝐷𝐷 − 𝐶𝑅1 𝑉𝐷𝐷 2 + 𝑉𝑆1

𝑄𝑇2 = 𝑄𝐶𝑆2,𝑖 − 𝑄𝐶𝑆2,𝑓 − 𝑄𝐶𝑅2,𝑖 − 𝑄𝐶𝑅2,𝑓

(4.7)
= 𝐶𝑆2 𝑉𝐷𝐷 2 − 𝑉𝑆2 + 𝑉𝐷𝐷 − 𝐶𝑅2 𝑉𝐷𝐷 2 + 𝑉𝑆2

62
The charge transferred to the integration capacitors, generates a voltage at the output
nodes of the charge integrator. Calculation of the differential output voltage VOUT is
given in equations (4.8) and (4.9).

𝑄𝑇1 − 𝑄𝑇2
𝑉𝑂𝑈𝑇 = 𝑉𝐶𝑖𝑛𝑡 1 − 𝑉𝐶𝑖𝑛𝑡 2 = (4.8)
𝐶𝑖𝑛𝑡

3𝑉𝐷𝐷 2 − 𝑉𝑆 𝐶𝑆1 − 𝐶𝑆2 − 𝑉𝐷𝐷 2 + 𝑉𝑆 𝐶𝑅1 − 𝐶𝑅2

𝑉𝑂𝑈𝑇 = (4.9)
𝐶𝑖𝑛𝑡

If the voltage VS is assumed to be VDD/2, and the reference capacitors are selected to
be equal, the output equation simplifies to Equation (4.10).

𝑉𝐷𝐷 𝐶𝑆1 − 𝐶𝑆2

𝑉𝑂𝑈𝑇 = (4.10)
𝐶𝑖𝑛𝑡

For the VS voltage assumption to be true, the reference capacitors should be selected
equal to the sensor rest capacitances. The relation between the value of reference
capacitors and VS voltage can be given as in the Equation (4.11). This equation can
be derived by equating the total charge transferred to the integration capacitors to
zero, i.e. QT1 + QT2 = 0.

𝑉𝐷𝐷 3𝐶𝑆 − 𝐶𝑅
𝑉𝑆 = (4.11)
2 𝐶𝑆 + 𝐶𝑅

Hence, the reference capacitors being not equal to the value of the accelerometer
rest capacitors, results in the change of scale factor. Moreover, by having the Vs
voltage different from VDD/2, it can get out of the input common mode range of the
amplifier, which results in improper operation of the amplifier.

63
4.2.2.1. Operational Transconductance Amplifier (OTA)

The operational transconductance amplifier (OTA) used in the charge integrator is a

critical component for the sampling rate and noise considerations. In order to
operate the circuit at high sampling frequency, the slew rate of the OTA is the most
significant parameter. Moreover, OTA is the component, which dissipates most of
the power in the readout circuit. Thus, a low noise, low power and high slew rate
OTA is needed. For this purpose, a folded cascode, fully differential OTA is
designed. This structure utilizes high power supply rejection ratio (PSRR), high
gain-bandwidth product and low noise.

The cascode structure has the advantages of having high open-loop gain, high
supply rejection ratio and lower noise compared to other amplifier structures. The
reason for choosing folded cascode structure instead of telescopic structure is to
obtain higher output voltage swing and wider input common mode range. However,
folded structure has relatively higher noise and higher power consumption than the
telescopic structure.

There is a trade of between noise and the speed of the amplifier for setting the
widths of the input transistors. Increasing width of the input transistors increases the
gate capacitances and hence, decreases the slew rate and the unity gain frequency of
the amplifier. Besides, decreasing noise results in higher currents in the branches;
therefore increases the power dissipation of the circuit. A high slew rate, low noise,
low power dissipation OTA was designed considering all these requirements and
tradeoffs.

The schematic of the folded cascode OTA is given in Figure 4.9. M1 and M2
transistors are input differential pair, which provides the gain on the folded branch.
M3 and M4 transistors source the current to the folded branch with high output
resistance. The open loop gain expression of the folded cascode OTA is given in
Equation (4.12).

𝐴𝑣 ≈ 𝑔𝑚 1 𝑔𝑚 7 + 𝑔𝑚𝑏 7 𝑟𝑂7 𝑟𝑂1 //𝑟𝑂5 // 𝑔𝑚9 + 𝑔𝑚𝑏 9 𝑟𝑂9 𝑟𝑂11 (4.12)

64
 CMFB
VDD

Vb1 M4 M11 M12 M17 M16

Vb2 M3 M9 M10

+
+Vin -Vin Vout
M1 M2
-

M7 M8

Vb3

M5 M6 M18 Vb4 M15

M13 M14

VEE

Figure 4.9. Schematic view of the folded cascode OTA.

The input transistors M1 and M2 should have higher trans-conductance for

obtaining higher gain. Therefore, the input transistors M1 and M2 should carry
higher currents. For the gain of the amplifier to be high, the output resistances of
the transistors M9 to M12 should be high, and since output resistance is inversely
proportional to the current, currents of the transistors M5 to M14 should be low.
Hence, in order to obtain high gain, the W/L ratios of the input transistors are
designed to be high, and W/L ratios of M13 and M14 transistors are designed low.
Table 4.1 shows the W/L ratios of the transistors.

The noise of the OPAMP is an important factor in order to achieve a high-resolution

readout circuit. The main noise sources of the amplifier are the noises from the
input transistors M1, M2, as shown in Equation (4.13). Wide transistors are used as
the input transistors for decreasing the noise. The trans-conductance of transistors
M13 and M14 are kept low by having a lower bias current. For the transistors M5 to
M8, through which a high current flows, the trans-conductance is decreased by
keeping the W/L ratios low and increasing the biasing VGS voltages.

65
Table 4.1: Size of the transistors used in the OTA circuit.

Transistor W/L
M1, M2 (PMOS) 800 / 4 m
M3 (PMOS) 1200 / 4 m
M4 (PMOS) 800 / 4 m
M5, M6 (NMOS) 93 / 4 m
M7, M8 (NMOS) 152 / 4 m
M9, M10 (NMOS) 177 / 4 m
M11, M12 (PMOS) 80 / 4 m
M13, M14 (PMOS) 40 / 4 m
M15 (NMOS) 8 / 4 m
M16 (PMOS) 8 / 4 m
M17 (PMOS) 32 / 4 m
M18 (NMOS) 29 / 4 m

2 2 𝑔𝑚 5,6 + 𝑔𝑚 7,8 2𝑔𝑚 13,14

𝑉𝑛 2 = 8𝑘𝑇 + 2 + 2 (4.13)
3𝑔𝑚 1,2 3 𝑔𝑚 1,2 3 𝑔𝑚 1,2

Because of having high output resistance, the output common mode voltage of the
OTA is not stable at the mid-point voltage and highly dependent on the circuit and
process parameters and bias voltages. Hence, with a little disturbance of circuit
parameters, the output common mode settles to the supply voltages. In order to
keep the common mode voltage at the mid-point voltage, a common mode feedback
(CMFB) circuitry is needed. The CMFB circuitry measures the common mode
output of the amplifier and gives a feedback to the amplifier to settle it to the desired
value. For this purpose, a parameter proportional to the common mode voltage
should be obtained. This is achieved by using two transistors M13 and M14
operating in deep triode region, gates of which are connected to the outputs of the
OTA, and sink a total current proportional to the common mode output voltage. The
current expressions of the transistors in deep triode region are given in Equations
(4.14) and (4.15).

𝑣𝐷𝑆
𝑖𝐷𝑆,14 = 𝐾𝑛 𝑣𝐺𝑆,14 − 𝑉𝑇𝑁 − 𝑣𝐷𝑆 ≈ 𝐾𝑛 𝑣𝑂𝑈𝑇+ − 𝑉𝑇𝑁 𝑣𝐷𝑆 (4.14)
2

66
 𝑣𝐷𝑆
𝑖𝐷𝑆,13 = 𝐾𝑛 𝑣𝐺𝑆,13 − 𝑉𝑇𝑁 − 𝑣𝐷𝑆 ≈ 𝐾𝑛 𝑣𝑂𝑈𝑇− − 𝑉𝑇𝑁 𝑣𝐷𝑆 (4.15)
2

where, M13 and M14 transistors have the same vDS voltage. When the drain source
voltages of the transistors are assumed sufficiently low, then the sum of the drain
source currents of the transistors become linearly dependent on the sum of the
voltages applied at the gates of the transistors as given in Equation (4.16).

𝑖𝐷𝑆,15 = 𝑖𝐷𝑆,14 + 𝑖𝐷𝑆,13 = 𝐾𝑛 𝑣𝑂𝑈𝑇 + + 𝑉𝑂𝑈𝑇− − 2𝑉𝑇𝑁 𝑣𝐷𝑆 (4.16)

The deep triode operation of these transistors is ensured by the transistor M15,
which operates in saturation region. When the gates of the deep triode transistors
are connected to the differential output of the OTA, the total current passing through
is proportional to the common mode voltage. After obtaining this current value, it
should be fed back to the amplifier; the current is mirrored with a factor to the M5
and M6 transistors of the amplifier. In this way, if the common mode voltage
increases, M5, M6 transistors sink more current and pull the common mode voltage
at the output to a lower value. Therefore, it keeps the common mode voltage at a
value defined by the CMFB circuitry.

AC simulation results of the amplifier are given in Figure 4.10. As seen in the
simulation results, the bandwidth of the amplifier is narrow, but since the amplifier
is operated in closed loop, the decrease of open loop gain at higher frequencies is
not significant. Instead, gain bandwidth product is an important parameter, in the
design of the OTA.

Noise of the amplifier is another critical parameter of the OTA. The input referred
noise simulation result of the amplifier is given in Figure 4.11. The design of the
circuit is optimized in order to minimize the noise, and hence, 4.4 nV/Hz thermal
noise is achieved. The noise levels could be further decreased by increasing the

67
widths of the input transistors, however this would cause high power dissipation,
and a decreased speed due to the high input capacitances. Table 4.2 summarizes the
simulation results of the OTA.

A = 96.6 dB f3dB = 986 Hz

GBW = 56.6 MHz

PM = 56

Figure 4.10. AC simulation results of the folded cascode OTA.

@ 1 Hz  580 nV/Hz

4.4 nV/Hz

Figure 4.11. Noise simulation results of the folded cascode OTA.

68
Table 4.2: Summary of the simulation results of the folded cascode OTA.

Parameter Value
Gain 96.6 dB
Bandwidth 986 Hz
GBW 56.6 MHz
Phase margin 56°
Power Dissipation 5.6 mW
Input referred noise (@ 10MHz) 4.4 nV/Hz
Input referred noise (@ 1 Hz) 580 nV/Hz

4.2.2.2. Bias Generator

The folded cascode OTA needs supply voltage and temperature independent bias
voltages for proper operation. For generating bias voltages of the folded cascode
amplifier, a bias-voltage generator circuit, given in Figure 4.12 was designed. The
transistors M1 to M8 generate the two bias voltages Vb1 and Vb4. The transistors
from M9 to M14 generate the other bias voltages Vb2 and Vb3 and the transistors M3
to M6 are biased by these voltages. The currents IDS1 and IDS2 are defined by the
W/L ratios of M1 and M2 and the value of R. In addition, these currents are
guaranteed to be equal by the current mirroring action of M7 and M8. Table 4.3
gives the W/L ratios of the transistors used in the bias generator.

The bias generator circuit itself has another stable state where all the transistors are
in cutoff; hence, for operating in normal condition, it needs a start-up circuit. The
start-up circuit is constituted by the transistors M15 to M18. After the supply is
turned on, M15 pulls the Vb1 voltage low, then the currents start to flow through the
transistors, and when the Vb3 becomes higher, M18 makes M15 to enter into cutoff,
so the start-up circuit does not consume any power while the bias generator starts to
operate.

Figure 4.13 gives the simulation results of the bias generator between -40C/+75C.
The most critical bias voltage, which affects the operation of OTA mostly, is Vb1.
Therefore, the temperature response of the circuit is optimized for Vb1 voltage,
where a parabolic behavior of temperature dependency is obtained. The temperature

69
dependencies at room temperature are given on the figure, which shows a low
temperature dependency of Vb1 bias voltage.

Figure 4.12. Schematic view of the bias generator circuit.

Table 4.3: Size of the transistors used in the bias generator circuit.

Transistor W/L
M1 (PMOS) 160 / 4 m
M2 (PMOS) 80 / 4 m
M3, M4 (PMOS) 150 / 4 m
M5, M6 (NMOS) 150 / 4 m
M7, M8 (NMOS) 16 / 4 m
M9 (PMOS) 80 / 4 m
M10 (PMOS) 200 / 4 m
M11 (NMOS) 7.6 / 4 m
M12 (PMOS) 21 / 4 m
M13 (NMOS) 200 / 4 m
M14 (NMOS) 16 / 4 m
M15 (NMOS) 3 / 2 m
M16 (NMOS) 1 / 4 m
M17 (PMOS) 1 / 30 m
M18 (NMOS) 0.8 / 0.8 m

70
 1.41 mV/°C

2.82 mV/°C

1.62 mV/°C

0.084 mV/°C

Figure 4.13. Temperature simulation results of the bias generator circuit.

4.2.3. Lead Compensator

Sigma-delta modulators operate at high sampling frequencies to accomplish high

resolution. However, high frequency operation brings some stability problems. In
electro-mechanical sigma-delta systems, the mass residual motion should be settled
at high frequency for the system to be stable and having low amplitude of residual
motion. However, the transfer function of the closed loop system can result the
system to settle at a lower frequency because of the poles arising from the
mechanical structure and the readout circuit. Therefore, generally it is required to
add a zero into the sigma-delta loop. In this study, adding a zero to the system was
realized using a switched capacitor lead compensator circuit. The schematic view of
the lead compensator circuit is shown in Figure 4.14, which uses a differential
structure.

71
 2
C2

3
C1

2
+ +
1

Vin Vout

1
_ _
2

C1
3

C2
2

Figure 4.14. Schematic of the lead compensator circuitry.

The switching timings of the lead compensator are given in Figure 4.15. At the time
between t0 and t1, where 1 and 2 signals are both high and 3 is low, the input
signal is sampled on the C1 capacitors. Then, 2 switches are turned off,
disconnecting the input from the compensator capacitors. At the time t2, 3 switches
are turned on, therefore the charge on the C1 and C2 capacitors are shared with the
C3 and C4 capacitors.

At the t3 instant, 3 and 1 switches are turned off, then 2 is turned on, hence the
input is connected to the C1 and C2 capacitors, which are keeping a factor of the
previous input voltage value. Thus, the output of the compensator becomes the
input value subtracted with a factor of the previous input value, as given in equation
(4.17).

72
 𝐶1
𝑉𝑂𝑈𝑇 = 𝑉𝑛 − 𝑉𝑛−1 (4.17)
𝐶1 + 𝐶2

Sampling Vn-1

Charge
sharing

Output
valid
1

2

3
t0 t1 t2 t3 t4 t5
Figure 4.15. Timing diagram of the lead compensator, type 1.

4.2.4. Comparator

In the sigma-delta loop, a high precision, high speed comparator is needed, to obtain
a high sampling rate and high resolution in overall readout circuit. Since, sample
based operations are performed in the readout circuit, the comparator should
compare the output voltages of the charge integrator stage at a single instant in the
sampling period, and keep the result until the next sample. Consequently, a
dynamic latching comparator is used, which compares its inputs at the rising edge of
a clock signal, and keeps the output value until the falling edge of the same clock
signal.

The schematic of the dynamic latching comparator circuit is given in Figure 1.1, and
the W/L ratios of the transistors are given in Table 4.4. The operation of the circuit
can be described as follows. When the Clk is low, the comparator is in pre-charge

73
phase and both outputs of the comparator are low, and there is no current flowing
through the branches. As soon as the Clk signal is switched to high, it compares the
input voltages and gives an output accordingly. Cross-coupled M3 and M4
transistors result in faster decision of the comparator. After the decision is made,
M5 and M6 cut the currents on the branches, and the output is stored until the Clk is
made low.

VDD

Clk M11 M9 M10 M12

M14 M16

M7 M8

Vout+ Vout-

M5 M6

M13 M15
Vin+ M1 M3 M4 M2 Vin-

VEE

Figure 4.16. Schematic view of the dynamic comparator.

Table 4.4: Size of the transistors used in the comparator circuit.

Transistor W/L
M1, M2 (NMOS) 4 / 4 m
M3, M4 (NMOS) 40 / 4 m
M5, M6 (NMOS) 8 / 4 m
M7, M8 (NMOS) 10 / 4 m
M9, M10 (PMOS) 20 / 4 m
M11, M12 (PMOS) 8 / 4 m
M13, M15 (NMOS) 4 / 4 m
M14, M16 (PMOS) 10 / 4 m

74
The offset voltage of the comparator is simulated in a temperature range of -40C to
60C. A 3 V/C dependency on temperature is observed, as shown in Figure 4.17.
The temperature dependency of the offset voltage results in a dependency of output
bit-stream to the temperature. However, the effect of the offset voltage on the
output temperature dependency is in the order of a few micro-g‟s per degree.

-1,00

-1,05
Input Offset Voltage (mV)

-1,10
3 V/C
-1,15

-1,20

-1,25

-1,30
-50 -30 -10 10 30 50 70
Temperature (C)

Figure 4.17. Input offset voltage of the comparator versus temperature.

4.2.5. Start-up Circuit

If there is a high deflection on the proof-mass position, with the effect of a large
input acceleration or because of a non-ideality, the force feedback applied during the
normal closed loop operation of the circuit may not be enough for holding the proof
mass at the midway between the electrodes. For holding the proof mass in the case
of large deflections, a start-up circuit is used. The schematic of the start-up circuit is
given in Figure 4.18. The output of the charge integrator stage is used as the input
of the start-up circuit. At the input of the start-up circuit, there are two comparators,
which have internal offset values of 1 V at the inputs. Hence, if differential output
of the charge integrator exceeds the range +1V, -1V, the OR gate gives a high value,

75
and the capacitor is discharged. In this way, the clock of the readout circuit is
turned off, until the capacitor is charged again through the PMOS transistor. The
PMOS transistor has a low W/L ratio, so that the charging of the capacitor takes
longer time. In this period, constant feedback force is applied to the accelerometer.
The charging time of the capacitor is designed to be around 40sec.

VDD

STUP_EN
Vb

CLKout

Ch_int CLKin
output C

Figure 4.18. Schematic view of the start-up circuit.

4.2.6. Multiphase Clock Generator

The multiphase clock generator creates the switching signals and phases of the
readout circuit, as shown in Figure 4.19. The circuit uses a single-phase clock
signal for generating the various phases of clock signals. The clock source is
supplied from the on-chip 500kHz clock generator circuit. The clock input can also
be supplied from an external clock source, which can be increased up to 700kHz.
Increasing the frequency higher than 700 kHz results in incorrect generation of
clock timings, hence the readout circuit does not operate properly. For generating
the multiphase clock signals, first a pulse is generated, where the width of the pulse
is independent of the input clock period. Therefore, changing the period of the input
clock signal does not change the pulse with, but only the frequency of the pulse.
Then this pulse is shifted many times by using the delay elements, hence multiphase
pulses are generated. The delay elements used are composed of two inverters and a
capacitor in between. Then, these pulses are used to generate various signals needed
for the readout circuit, by the use of basic logic operations. All the pulses are

76
independent of the clock period, so when the period of the input clock signal is
increased, only the feedback duration is increased but the readout timings are kept
constant. The simulation result of the output waveforms of the multiphase clock
generator is given in Figure 4.20.

p25 phf1
p22
Clk
DELAY phf2

p20 Rs
p23

p17 phc1
p20
p11 p13

p19 phcd
p22

p27
delay delay delay delay p21 phcdn

p18 pfdn
p24

p12 p26
p15 pfd
p25

Figure 4.19. Schematic view of the multiphase clock generator circuit.

pfd
pfdn
phf1

phf2
Rs
phc1

pcds
pcdsn
Figure 4.20. Timing diagram of the signals generated by the multiphase clock generator.

77
4.2.7. Test Structures and Control Signals

There are various testing structures for making different parts of the readout circuit
available to be tested. First test structure included in the circuit is for controlling
timing signals generated by the multiphase clock generator, as shown in Figure 4.21.
This circuit consists of 8 multiplexers, which are controlled by a single signal
“Ext_tim”. When the control signal is low, the readout circuit uses the internally
generated timing signals and when “Ext_tim” is high, the 8 pins are used as inputs to
the circuit and readout circuit uses externally applied timing signals.

8 x MUX2

8
i_phf1, i_phf2, ...., i_Rs
8
phf1, phf2, ...., Rs
8
xt_phf1, xt_phf2, ...., xt_Rs

Ext_tim

Figure 4.21. Schematic view of multiplexer structures, used for applying external timing
signals.

As a second test structure, for observing the internally generated timing signals, one
pad is used, and 8 timing signals are decoded using the circuit shown in Figure 4.22.
The selection signals are the same signals, which are used for selecting the internal
reference capacitors; hence, no additional pads are used for selecting the timing
signal to be observed.

Another test structure, the sample and hold circuitry, is implemented to observe and
test the open loop operation of the readout circuit. Buffer op-amps with sample and
hold capacitances are used as shown in Figure 4.23. This circuit enables observing
the sampled analog outputs at a longer duration of the sampling period, provides
driving of high external capacitances.

78
 phf1 Test_timing

phf2

pfdn

pfd

phcdn

phcd

phc1

Rs

D0
D1
D2
D3
D4
D5
D6
D7
Decoder

S0

S1

S2
Figure 4.22. Schematic view of the test structure for observing the internal signals.

S&H
An_out+
Charge Integrator Ouptut

An_out-
S&H
Figure 4.23. Schematic view of the sample and hold circuitry.

For increasing the testability of the circuit, there are also various selection signals to
set the capacitor values and to enable and disable various parts of the circuitry. The
reference capacitors, which are used in the full bridge structure with the
accelerometer capacitances, are implemented in the chip, and are selectable, using 5
control signals, enabling 5 different sized capacitor. The sizes of the capacitors are
as binary coded, with minimum sized capacitor having value of 0.875 pF and the

79
largest capacitor having 14 pF value. Therefore, a maximum of 27.125 pF of
reference capacitor value can be selected. Similarly, the integration capacitors used
in the charge integrator stage is selectable by 4 control signals up to 15 pF, with 1
pF increments.

4.3. Overall Circuit Simulation Results

After the design of individual block diagram of the readout circuit, the simulations
of the overall system is performed, to observe the operation of the readout circuitry.
The simulations are performed in Cadence environment with the model of the
mechanical sensor, both in open loop and closed loop modes. The open loop
simulation results are given in Figure 4.24. In this simulation, 1 kHz of sinusoidal
input acceleration is applied to the system, and the charge integrator differential
output is observed. Capacitive difference on the sensor follows the input
acceleration signal with a phase delay. The analog output samples are observed to
be proportional to the capacitive difference.

Acceleration
pfd

Acceleration
phcd

Input
Input
Sense
Sense Cap.Difference
Capacitance Diff.
output
Analog output
Differential

Figure 4.24: Open loop simulation results of the sigma delta readout circuit together with
the model of the SOG accelerometer.

80
Figure 4.25 gives the closed loop simulation results of the readout circuit combined
with the mechanical sensor model. In this simulation, a 1 kHz, 1 g amplitude of
input acceleration is applied to the system. The displacement of the proof mass and
digital output signals are observed. The average value of output pulses increase for
increasing input accelerations. The residual motion of the proof mass is observed to
be around 1 nm in amplitude.

Mass displacement
Proof Mass
Input
Input displacement
Acceleration
Acceleration
output
Comparator
Digital output
Figure 4.25: Closed loop simulation results of the sigma delta readout circuit together with
the model of the SOG accelerometer

The simulation times in Cadence environment cannot exceed a few milliseconds due
to the long computation time. Simulations performed in Matlab environment gives
much fast results accurately. The model of the readout circuit with the
accelerometer is given in Figure 3.9. With this model a simulation is performed
using 1 kHz, 1 g amplitude of input acceleration. The simulation results are given in
Figure 4.27. The density of the output bit-stream is observed to be changing in
parallel with the input acceleration. The simulation results of Cadence and Matlab
are consistent.

81
Figure 4.26: Simulink model of the complete electromechanical closed loop system.

Accelerometer Model
input acceleration (g)

0.5

-0.5

-1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time -3
x 10
proof mass displacement

-9
x 10
3

-1

-2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time -3
x 10

1
digital output

0.5

-0.5

-1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time -3
x 10

Figure 4.27: Simulation results of the accelerometer system using the model created in
Matlab.

82
4.4. Layout Considerations

The layout of the readout chip is drawn and the chip is implemented with 0.6µm
XFAB CMOS process. The layout of the sigma-delta readout chip is given in
Figure 4.28, including the floor plan of the chip.

Since the readout circuit has a fully differential structure, the layout should be drawn
fully symmetrically, in order to reduce any differential offset, parasitic impedance
and other non-idealities in the circuit. The analog blocks and the digital blocks of
the circuit are separated as much as possible, by placement and the usage of guard
rings around the blocks. The guard rings are used for isolating the part of the
circuit, and reducing the substrate noise, due to the other components of the readout
circuit.

25 24 23 22 21 20 19 18 17 16
Csen_m D_VDD Scds Cint0 Cint1 Cint2 Cint3 SH_en Cmpsdis A_VDD

26 15
Vt Aoutp
Multiphase Clock Generator

Switch Capacitor Network

Sample
& 14
27 Hold Doutp
1179 m

Compensator

Comparator

Clkout
BIAS OTA 13
GND
28
Analog
Clkin output 12
Switch Capacitor Network buffers
Doutn

29 11
Vb Aoutn

1 2 3 4 5 6 7 8 9 10
Csen_p D_VSS Openloop Sclk Cref0 Cref1 Cref2 Cref3 Cref4 A_VSS
(0,0) 1717.75 m

Figure 4.28: Layout view and floor plan of the fully differential readout circuit chip.

83
4.5. Performance Limitations

There are various parameters, limiting the performance of the readout circuit
together with the accelerometer. These limitations include the noise sources, which
are the readout circuit noise due to the OTA, switches and other circuit components;
sigma – delta loop noises, such as the quantization noise and mass residual motion;
and the mechanical noise due to the accelerometer. These noise sources limit the
minimum measurable acceleration value. On the other end, maximum operation
range is limited by the sensor dimensions in open loop mode and by maximum
applicable feedback force in closed loop mode. In addition, offsets are created due
to the non-linearity in both readout circuit and accelerometer. In the following
subsections, these non-idealities limiting the performance of the system will be
discussed.

4.5.1. Mechanical Noise

Mechanical noise exists on any mechanical structure, as the force applied on the
mass of the mechanical structure. This mechanical noise, which is also called
Brownian noise, is due to the thermal motion of the molecules. Brownian noise has
a white noise characteristic with the noise amplitude in terms of acceleration given
in equation (4.18).

4𝑘𝐵 𝑇𝐵
𝑎𝑛,𝑚𝑒𝑐 𝑕 = (𝑚 𝑠 2 ) (4.18)
𝑚

where, kB is Boltzmann‟s constant, T is the temperature, B is the mechanical

damping constant and m is the mass of the mechanical structure. Hence, in order to
achieve low noise mechanical devices, the mass of the device should be selected to
be large or the structure should have low damping constant. Decreasing the
damping constant can be achieved by decreasing the pressure of the environment.
However, the vacuum condition can result in stability problems.

84
4.5.2. Readout Circuit Noise

The noises resulting from the readout circuit are noise due to the amplifier, and
kT/C noise created due to the switching of the capacitors. The thermal noise of the
OTA was already calculated in section 4.2.2.1, as given in Equation (4.13). When
this OTA is used in a switched capacitor structure, the thermal noise is filtered
according to the capacitor values. The basic switched capacitor charge-integrator
structure is shown in Figure 4.29. The thermal noise of this circuitry is given as in
Equation (4.19) [14].

Cint

CS CP
COUT

Figure 4.29. Schematic view of the basic switched capacitor charge integrator structure.

16 𝐶𝑆 + 𝐶𝑃 𝑘𝐵 𝑇
𝑉𝑛,𝑟𝑒𝑎𝑑𝑜𝑢𝑡 = (𝑉 𝐻𝑧) (4.19)
3 𝐶𝑖𝑛𝑡 𝐶𝑂𝑈𝑇 𝑓𝑆

where, kB is the Boltzmann‟s constant, T is the temperature, fS is the sampling

frequency of the readout circuit, and the capacitors are as shown in Figure 4.29. In
the figure, CS is the sensor capacitance; CP is the parasitic capacitance at the input of
the amplifier, Cint is the integration capacitor and Cout is the load capacitance at the
output of the amplifier. According to the equation (4.19), increasing the value of

85
integration capacitor Cint, decreases the noise of the readout circuit, which is
inversely proportional with the square root of Cint. However, capacitance to voltage
sensitivity of the readout is inversely proportional with the value of Cint. Hence, if
SNR of the circuit is considered, decreasing the value of integration capacitor,
increases the SNR of the circuit.

Another source of noise at the readout circuit is the kT/C noise, which mainly
results from switching of the integration capacitors. Hence the kT/C noise
expression is given as in equation (4.20).

4𝑘𝐵 𝑇
𝑉𝑛,𝑘𝑇/𝐶 = (𝑉 𝐻𝑧) (4.20)
𝐶𝑖𝑛𝑡 𝑓𝑆

4.5.3. Open Loop Mode Non-Linearity and Operation Range

In open loop operation of the readout circuit, differential capacitance is measured by

the readout circuit. However, the relation between the capacitance and proof mass
displacement is linear only if the displacements are very small in amplitude, i.e. are
ignorable when compared with the finger gap spacings. The Equation (3.2) gives
the relation between capacitance difference and displacement of proof mass, which
was already discussed in the previous chapter. Figure 4.30 gives the graph of
differential capacitance versus displacement, where the highly non-linear
dependence can be observed. The finger gap spacings of the accelerometer used to
obtain the graph, is 4.5µm. Hence, the linearity for small displacements is observed
in this graph. The capacitance non-linearity increases enormously, as the
displacement becomes comparable with finger gap spacings.

𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴 𝑁𝜀0 𝐴 𝑁 − 1 𝜀0 𝐴
∆𝐶 = 𝐶1 − 𝐶2 = + − − (4.21)
𝑑1 − 𝑥 𝑑2 + 𝑥 𝑑1 + 𝑥 𝑑2 − 𝑥

86
Figure 4.30. Differential capacitance versus displacement graph for a capacitive
accelerometer.

The operation range is limited by the finger gap spacings in open loop mode. From
the scale factor of the accelerometer at low frequencies, if the value of finger gap
spacing is substituted to displacement x, maximum operational acceleration range is
calculated, as shown in Equation (4.22).

𝐾𝑥𝑚𝑎𝑥 𝐾𝑑1
𝑎𝑚𝑎𝑥 ,𝑜𝑝𝑒𝑛𝑙𝑜𝑜𝑝 = = (𝑔) (4.22)
𝑚𝑔 𝑚𝑔

4.5.4. Closed Loop Operational Input Acceleration Range

The input acceleration range of the closed loop system is mostly dependent on the
accelerometer parameters. The maximum input acceleration is determined by the
maximum feedback force that can be applied to the accelerometer sensor. Equation
(4.23) gives the feedback acceleration applied to the sensor proof-mass related to the
electrostatic voltage applied to one of the electrodes. In this equation, m is the

87
proof-mass, g is the gravitational acceleration, and x is the amount of displacement
of the proof-mass from the rest position.

1 𝜕𝐶 2
𝑎𝑓 = 𝑉 (g) (4.23)
2𝑚𝑔 𝜕𝑥

Since, feedback is not applied full time, but only in the feedback phase, for the sense
electrodes, the maximum feedback acceleration is obtained by multiplying af by
feedback time to total readout time ratio. Hence, one can obtain equation (4.24),
which is fully dependent on accelerometer parameters except the feedback ratio and
feedback voltage. Since the feedback voltage is supplied from the power source, it
cannot be easily changed. The remaining parameter to set the operation range using
the readout circuit is the feedback time ratio. As seen from the equations, the
acceleration range is linearly proportional with the feedback duration ratio.

𝑇𝑓 𝑇𝑓 1 𝜕𝐶 2 𝑇𝑓 1 𝑁 𝑁−1
𝑎𝑚𝑎𝑥 = 𝑎𝑓 = 𝑉 = 𝐴𝜀 2
− 2
𝑉 2 (4.24)
𝑇𝑡𝑜 𝑡 𝑇𝑡𝑜𝑡 2𝑚𝑔 𝜕𝑥 𝑇𝑡𝑜𝑡 2𝑚𝑔 𝑑1 − 𝑥 𝑑2 + 𝑥

Where, A is the overlap area of one finger of the accelerometer, and  is the
permittivity of air. Other accelerometer parameters used in equation (4.24) are
illustrated in Figure 4.31.

Finger overlap length (L)

Finger antigap (d2)
Electrode 1

Electrode 2

F1 F2
Finger gap (d1)
Proof Mass
(M) Finger gap (d1)
Finger antigap (d2)
Finger overlap length (L)

Figure 4.31. Illustration of the sensor fingers and parameters.

88
4.5.5. Quantization Noise

The quantization noise of a second order sigma-delta system is given as in equation

(4.25).

𝜋2
𝑉𝑞𝑛 ,𝑟𝑚𝑠 = 𝑒𝑟𝑚𝑠 (4.25)
𝑀2.5 5

In this equation, L is the order of the sigma delta system; M is the oversampling
ratio, erms is the rms value of unshaped quantization noise. For single bit
quantization of /2 and -/2, erms becomes /12. For the case of accelerometer
readout circuit,  is 2amax. Hence, for a second order sigma-delta readout circuit, the
quantization noise can be shown as in equation (4.26).

𝑎𝑚𝑎𝑥 𝜋 2
𝑉𝑞𝑛 ,𝑟𝑚𝑠 = (4.26)
15 𝑀2.5

Equation (4.26) shows that quantization is highly dependent on oversampling ratio.

Therefore, for decreasing the quantization noise, increasing the frequency has a
significant effect.

4.5.6. Mass Residual Motion

The proof-mass of the sensor is being held in a constant position using one-bit
feedback pulses. This creates a motion around the rest position of the proof-mass.
The motion becomes random when a varying input acceleration is applied to the
sensor. However, for calculation of the amplitude of the mass residual motion, zero
input acceleration is assumed. In this case, with proper lead compensator usage, the
mass residual motion settles to the frequency fs/4, as discussed in the previous
chapter, where fs is the sampling frequency of the readout circuit.

89
The critical parameter is the magnitude of the mass residual motion and its
contribution to the acceleration noise in the bandwidth of interest. The displacement
magnitude of the residual motion can be found from the feedback acceleration
multiplied with the transfer function of the sensor at the oscillation frequency, as in
the equation (4.27) given. For calculating the contribution of the mass residual
motion to the noise in the bandwidth of interest in terms of acceleration, the
equation (4.28) is given. Here, an assumption is made such that the mass residual
motion is uniformly distributed in the frequency between 0 and the mass residual
oscillation frequency. Since various inputs will be given to the system, the residual
motion changes its frequency according to the input magnitude, and hence a white
noise assumption can be made to the mass residual motion.

1
𝑥𝑟𝑚 = 𝑎𝑚𝑎𝑥 2
(4.27)
𝑗2𝜋𝑓 + 𝐵 𝑚 𝑗2𝜋𝑓 + 𝐾 𝑚

𝑓𝐵𝑊 𝐾
𝑁𝑟𝑚 = 𝑥 (g) (4.28)
𝑓 𝑚 𝑟𝑚

If the mass residual motion has a frequency of fs/4, the equations can be
approximated as in the equations (4.29) and (4.30).

𝑎𝑚𝑎𝑥
𝑥𝑟𝑚 = 2
2𝜋𝑓𝑠 (4.29)
4

4𝑓𝐵𝑊 𝐾 𝑎𝑚𝑎𝑥
𝑁𝑟𝑚 = 2
(4.30)
𝑓𝑠 𝑚 2𝜋𝑓𝑠 4

Hence, both decreasing maximum feedback acceleration and increasing the

frequency of the readout circuit decreases the noise caused by the mass residual
motion.

90
4.6. Conclusion

In this chapter, the design and simulations of the sigma-delta readout circuit and the
performance limitations of the system are given. Table 4.5 gives the main
parameters and the noise sources of the designed circuit with both the SOG and
DWP accelerometers. The results show that mass residual motion and quantization
noise dominating the noise sources. Hence, to decrease these noises, a higher
sampling rate operation should be obtained, in order to increase the resolution of the
system. Considering the case, new adaptive readout circuits are designed to increase
sampling rates and achieving higher performance, which are explained in the
following chapter.

Table 4.5: Readout circuit parameters and design achievements.

Parameter Value (with SOG) Value (with DWP)

Closed loop acceleration range ±4 g ± 19 g
Noise due to mass residual 54 µg (1 kHz BW) 324 µg (1 kHz BW)
motion (1 kHz BW) 1.7 µg/Hz 10.5 µg/Hz
Quantization noise 10.3 µg/Hz 48.9 µg/Hz
Capacitance Sensitivity 0.3 – 5 V/pF
Power consumption 16 mW
Sampling rate 500 kHz
Readout circuit noise 0.94 µV/Hz

91
 CHAPTER 5

5. ADAPTIVE SIGMA DELTA READOUT CIRCUITS

To achieve higher performance in terms of resolution, the sigma-delta readout

circuits need higher frequency operation, since most of the noise sources are
dependent on the sampling frequency and decreases with the increasing frequency.
However, increasing the sampling frequency of the readout circuit is not an easy
process. To increase the sampling frequency, the speed of the circuit components
should be increased, which is mostly limited by the used process. Another way of
increasing the sampling frequency is to decrease the feedback duration of the
sampling period, without decreasing the sense duration. However, decreasing the
feedback time decreases maximum applicable feedback force, hence, decreases the
operation range of the closed loop system.

To increase the sampling frequency, without decreasing the operation range of the
readout circuit, two novel techniques are proposed and the readout circuits are
designed and simulated. These readout circuits will be fabricated after this thesis
study and the test results will be obtained. Therefore, only the design and the
simulation results are discussed in this thesis. The readout circuits use adaptive
techniques, where they change their operation state according to the rough
magnitude of the input acceleration. This chapter presents the design and simulation
results of these adaptive readout circuit designs. Section 5.1 presents the operation
principle of the two readout circuits. Section 5.2 provides the explanation of some
significant blocks of the readout circuits and gives the simulation results of these
circuit blocks. Section 5.3 gives simulation results of the readout circuits. The
complete simulations are performed in Matlab Simulink environment, and the
operations of the readout circuits are verified. Section 5.4 gives the layouts of the

92
designed circuits. Finally, Section 5.5 gives a comparison of the designed circuits in
terms of some noise components and the operation range, and concludes the chapter.

5.1. Operation Principle of the Readout Circuits

The readout circuits employ the sigma-delta structure, including a feedback

mechanism, which improves the resolution and dynamic range of the system. The
improvements are achieved by decreasing the mass residual motion and quantization
noise. The main issue for decreasing the mass residual motion and quantization
noise is increasing the readout sampling frequency. The limiting cases for achieving
higher sampling frequency are the OTA speed, and the feedback duration needed to
attain desired operation range. The OTA is designed to have best performance in
order to have high speed, high current drive capability, low noise and low power.
Hence, it has a feasible limit of speed, which prevents decreasing the readout time
further. Another way of increasing the sampling rate is to decrease feedback
duration. However, decreasing the feedback duration decreases the maximum
average feedback force applied to the sensor. Therefore, the operation range is
reduced. Figure 5.1 illustrates the increasing sampling frequency and decreasing
feedback acceleration, with the change of feedback duration, and without changing
the sense duration.

(a) sense feedback fs amax

(b) sense f sense f 2fs amax/6

b b

Figure 5.1. Illustration of the effect of decreasing feedback time. (a) with 60% feedback
duration. (b) with 20% feedback duration.

93
5.1.1. Sigma-Delta Readout Circuit with Adaptive Feedback Duration

It is required to increase the sampling frequency without reducing the operation

range. To achieve this, one method is to decrease the feedback duration for small
input accelerations, and gradually increase the feedback duration according to the
acceleration. In this way, the accelerometer system has high resolution for small
accelerations and lower resolutions for higher acceleration ranges. Hence, both high
resolution and high range is achieved, not at the same time but with the same
system.

The circuit has almost the same operation with the second order sigma-delta system.
However, it has variable feedback time, which is digitally controlled according to
the input acceleration range. The circuit includes a digital control mechanism for
applying feedback related with the magnitude of the input acceleration. The system
has 15 different operation ranges, which is divided linearly. Range value is hold in a
4-bit register. If the range is 0, the circuit reads the position of the proof mass in the
sense phase in 7 clock cycles and generates a feedback pulse and applies a positive
or negative feedback accordingly to the sensor in 1 clock cycle duration. This
generates a 12.5% of feedback duration, with a readout frequency of 2.5 MHz, if a
20MHz clock is used. At the maximum range, sense phase is again 7 clock cycles,
and the feedback duration increases to 31 clock cycles. So, 82 % of the readout
duration is spent for feedback and the readout frequency decreases to 513 kHz, with
a 20 MHz clock frequency. Figure 5.2 illustrates the sense feedback timings for
different operation ranges of the readout circuit.

SN FB SN FB SN FB SN FB SN FB SN FB SN FB SN FB
Rng = 0 Rng = 1 Rng = 2 Rng =6

Figure 5.2. Illustration of the sense and feedback timings for different acceleration ranges,
for the varying feedback-time readout circuit.

94
The feedback duration is increased if the range is positive, which means the input
acceleration is positive. At the beginning, the proof mass makes a positive
displacement, and a feedback force is applied in negative direction. However, when
the proof mass passes over to the negative position with the feedback, the circuit
does not enter to feedback phase after sense phase; but it enters to another sense
phase without giving any feedback. The reason of not applying a feedback force on
the other direction is to decrease the mass residual motion, since it is known that
there is already acceleration, pulling the mass to the positive position. Hence, the
proof mass is pulled back by the already existing external acceleration.

For determining the operation range, the digital output history of the circuit is
examined. The feedback pulses for 128 clock periods are counted and an average
value of acceleration is extracted from the data. If this acceleration value is not in
the range of operation, the circuit changes state by changing the feedback time
duration, and increases its acceleration range.

For counting the feedback force, a digital counter is used, and it counts at each clock
cycle according to the feedback pulse. The counter counts for 128 clock cycles. If
the clock frequency is applied as 20 MHz, this corresponds to 6.4 sec. Since the
maximum frequency of the applied acceleration is 500 Hz, in 6.4 sec, the range of
the system cannot change abruptly; so this duration does not create any problem.

5.1.2. Sigma-Delta Readout Circuit with Multi-bit Feedback

Another way of decreasing the feedback time, without decreasing the operation
range is using separate electrodes in the sensor, to apply electrostatic feedback force.
From the sense electrode, regular sigma-delta technique is applied, with short
feedback time, and hence with high resolution, and the high acceleration range is
guaranteed by applying multi-bit feedback from the feedback electrodes. Hence, the
designed readout circuit uses a different kind of accelerometer sensor, which has 3
sets of electrodes, with a single proof mass, a basic illustration is given in Figure
5.3. One of these 3 sets of electrodes is used for reading the position of the proof

95
mass, and also applying feedback with short duration. With short feedback
duration, high sampling frequency is achieved up to 2.5 MHz. This set of electrodes
can be named as “sense electrodes”, even though they are both used for sensing and
applying feedback.

The remaining 2 electrodes are used just for applying feedback force. The feedback
force that will be applied from one set of electrode can take 3 values, positive,
negative or no feedback force. When the electrodes of the accelerometer is arranged
such that one set of feedback electrodes is twice the size of the other feedback
electrode, 7 evenly distributed values of feedback force can be generated using only
the feedback electrodes. If the feedback acceleration that can be generated by the
smaller electrode is a, then the total feedback acceleration that can be applied from
feedback electrodes ranges from +3a to -3a.

30% *2a =
a 2a 0.6a
Fe1 Fe2 Sense

mass

Fe1' Fe2' Sense

-a -2a -0.6a

Figure 5.3. The structure of the accelerometer sensor, to be used with the multi-bit adaptive
sigma-delta circuit structure.

For obtaining a suitable size for the sense electrodes, the average maximum
feedback force that can be applied from the sense electrodes should be considered.
To cover all values of acceleration, the maximum feedback acceleration from sense
electrodes must be at least a/2. So, if the acceleration applied from the feedback
electrodes is zero, the sigma-delta system constituted with the sense electrodes will

96
cover the range between +a/2 and -a/2. For instance, if a feedback acceleration of a
is applied from feedback electrodes, then the operation range is changed to between
+3a/2 and +a/2. Assuming the feedback duration is the 25% of the total sampling
period, the feedback acceleration that can be applied is 25% of the full time
feedback acceleration. Hence, to obtain a feedback acceleration of a/2, twice the
size of smaller feedback electrode should be used, for the sense electrodes.
However, for safe operation, the size of the sense electrodes should be selected
larger than twice of the size of smaller feedback electrodes, or the feedback time
should be a bit more than 25%. For this purpose, the feedback duration is chosen to
be 30%. Hence, the accelerometer sensor can have electrode size ratios of 1:2:2.

After obtaining the sensor structure that can be used for the readout circuit, the basic
principle of operation of the readout circuit should be explained. The block diagram
of the readout circuit is given in Figure 5.4. The readout circuit uses the same
sigma-delta structure, which is composed of a charge integrator, a compensator and
a comparator. The charge integrator senses the capacitance difference of the sense
electrodes, by transferring the charge difference created on the sensor capacitances
to the integration capacitors. After passing from a lead compensator, the read value
is entered to a comparator. Hence, according to the result of the comparator, either a
positive or negative feedback is applied to the sense electrodes in the feedback time
of the sampling period.

Charge Integrator

CRef2 CRef1
Comparator
Digital
Outputs

Compensator Digital Blocks

Accelerometer

Figure 5.4. Block Diagram of the readout circuit.

97
For determining the feedback force to be applied from the feedback electrodes, a
digital control block is used. This block counts the applied feedback to the sense
electrodes in a period of 128 clock cycles, and if the applied feedback is exceeding a
previously defined value, the system increases the applied feedback to the feedback
electrodes.

5.2. Description of the Main Blocks of the Readout Circuit

This section describes the readout circuit blocks, which differentiate from the circuit
blocks used in sigma-delta readout circuit designs explained in Chapter 4. There are
mainly 4 blocks of the readout circuit, most of which are the same with the previous
readout circuits. The OTA is improved in these readout circuits, having a better
performance in terms of speed and noise, and using a different structure of Common
Mode Feedback. The other component changed is the lead compensator circuit,
which uses a different kind of capacitor switching mechanism. The main changes of
the circuits are made structurally by timings or feedback control, which are
controlled by the digital blocks of the readout circuit.

5.2.1. Operational Trans-conductance Amplifier (OTA)

The folded cascode fully differential OTA structure is used also in the adaptive
readout circuits. A better design in terms of noise and speed is achieved. Also, the
structure of the CMFB circuitry is changed, as shown in Figure 5.5. The sizes of the
transistors used in the OTA circuit is given in Table 5.1.

For decreasing the flicker noise of the amplifier, the area of the transistors M5 and
M6 are increased, while keeping the W/L ratio low, in order to keep the thermal
noise of the circuit low.

The CMFB circuitry used in the previous designs, highly depend on the process
parameters, and does not include a reference for setting the common mode voltage,
which results in setting the common mode of the circuit to a value not exactly at the

98
mid-point voltage. The output nodes of the OTA are connected to the gates of the
two transistors (M15, M16) of the CMFB circuit. These transistors are operated in
deep triode region; hence, the sum of the drain currents of these transistors is
proportional to the sum of the voltages at the output nodes of the OTA. With the
diode connected PMOS transistor, the first stage of the CMFB circuit generates a
voltage depending on the sum of the voltages at the output nodes of the OTA. This
stage is replicated, with the inputs of the transistors are grounded. The replicated
stage generates a reference voltage. Then, in the second stage, a diode loaded
differential amplifier is used. Diode loaded structure is used to obtain wide
bandwidth, low gain amplifier, and hence maintain the stability of the CMFB
circuit. The output of the CMFB circuit biases the bottom NMOS transistors of the
OTA, and hence keeps the output common mode very close to the reference voltage.

Figure 5.5. Schematic view of the folded cascode OTA.

AC simulation results of the amplifier are given in Figure 5.6. As seen in the
simulation results, the bandwidth of the amplifier is narrow, but since the amplifier
is operated in closed loop, the decrease of open-loop gain at higher frequencies is
not significant. Instead, gain bandwidth product is an important parameter, in the
design of the OTA.

99
Table 5.1: Size of the transistors used in the OTA circuit.

Transistor W/L
M1, M2 (PMOS) 600 / 3 m
M3 (PMOS) 400 / 3 m
M4 (PMOS) 400 / 3 m
M5, M6 (NMOS) 116 / 8 m
M7, M8 (NMOS) 8 / 8 m
M9, M10 (NMOS) 98 / 3 m
M11, M12 (PMOS) 120 / 3 m
M13, M14 (PMOS) 50 / 3 m
M15, M16, M19, M20 (NMOS) 3 / 5 m
M17, M21 (NMOS) 8 / 4 m
M18, M22 (PMOS) 3 / 10 m
M23, M24 (PMOS) 125 / 3 m
M25, M26 (NMOS) 3 / 5 m
M27 13 / 3 m

A = 97 dB f3dB = 860 Hz

GBW = 54.7 MHz

PM = 64

Figure 5.6. AC magnitude and phase response of the OTA.

Noise of the amplifier is another critical parameter of the OTA. The noise
simulation result of the amplifier is given in Figure 5.7. The low frequency noise
performance of the OTA is considerably improved in this design. The summary of
the simulation results of the OTA is given in Table 5.2.

100
 @ 1 Hz  367 nV/Hz

4.2 nV/Hz

Figure 5.7. Simulation results for the input referred noise of OTA.

Table 5.2. Summary of the simulation results for the folded cascode amplifier.

Parameter Value
Gain 97 dB
Bandwidth 860 Hz
GBW 54.7 MHz
Phase margin 64°
Power Dissipation 3.5 mW
Input referred noise (1 MHz BW) 4.2 nV/Hz
Input referred noise (@ 1 Hz) 367 nV/Hz

5.2.2. Lead Compensator

The schematic of the lead compensator structure is given in Figure 5.8. The
compensator operates in 4 non-overlapping clock phases as shown in the timing
diagram given in Figure 5.9.

101
Figure 5.8: Schematic view of the lead compensator circuit.

P1

P2

P3

Clr

Figure 5.9: Compensator switching timing diagram.

Assume that „P1‟ signal is at high state, and all other signals are at low state. Also,
assume that C2 capacitors are charged with the previous value of the voltage read
out by the charge integrator stage, named as Vn-1. Another assumption is that the C3
capacitors are discharged. In that case, with P1 switch closed, C2 and C3 capacitors

102
are paralleled inversely. After the charge sharing, the resultant voltage on C3 will
be as given in Equation (5.1).

𝐶2
𝑉𝐶3 = −𝑉𝑛−1 (5.1)
𝐶2 + 𝐶3

Next, in the „P2‟ phase, C1 and C2 capacitors are charged to the output voltage of
the charge integrator, Vn. Following this, in „P3‟ phase, C1 and C3 capacitors are
directly connected to the output and the compensator output is provided. After the
output is provided, the C2 still stores the voltage. In „Clr‟ phase, C3 capacitors are
discharged. The cycle again starts from the „P1‟ phase, C2 shares the charge with C3
and the cycle repeats.

The voltage resulting from the charge sharing of C1 and C3 in the „P3‟ phase
becomes as in Equation (5.2).

𝐶
𝐶3 𝑉𝑛 𝐶1 − 𝑉𝑛−1 𝐶 +2 𝐶
𝑉𝑜 = 2 3 (5.2)
𝐶1 + 𝐶3

If all the capacitors are selected to be 1 pF, which is the case used in the readout
circuit, the output becomes as in Equation (5.3). The equation (5.4) is the z-domain
representation of this expression.

2𝑉𝑛 − 𝑉𝑛−1
𝑉𝑜 = (5.3)
4

2 − 𝑧 −1
𝑉𝑜 𝑧 = (5.4)
4

103
5.2.3. Digital Blocks

Digital blocks for the readout circuits mainly consist of a ring oscillator, a digital
control and a multiphase clock generator circuits as shown in Figure 5.10. The ring
oscillator generates a 20 MHz clock signal needed for the operation of circuit. The
clock signal can also be supplied from an external source. The digital control circuit
controls the feedback and sense timings, the operation range of the readout circuit,
and counts the feedback pulses and gives to the output. Multiphase clock generator
block provides clock signals for all the switches used in the readout circuit.

Range Tot_pls

Multiphase
Ring Digital
Clock
Oscillator Control
Generator

Figure 5.10: Block diagram of the digital part of the readout circuit.

The digital blocks of the readout circuit consists of two parts. One of those parts is
for generating the multiphase timing signals, which are needed in various parts of
the circuit. The other part is for controlling the feedback mechanism, and
generating the necessary feedback signals.

5.2.3.1. Timing Signal Generation

The sense phase of the both readout circuits are completed in seven clock cycles.
For the readout circuit with adaptive feedback duration, the feedback time depends
on the operating range value, hence on the input acceleration value. For this circuit,
the feedback duration can vary from 1 clock cycle to 31 clock cycles. For the multi-

104
bit readout structure, the duration of the feedback cycle is selected externally from
the digital rng<0:1> inputs. The feedback duration can take values of 1, 3, 7 and 17
clock cycles, which corresponds to 12, 30, 50 and 71 percent of feedback durations.
These timings are controlled by using a counter, and other sequential and
combinational logic circuitry.

There are various timing signals used in different parts of the readout circuit. For
generating these signals, initially, three major timing signals, 1, 2, 3 are
generated. These timing signals are high for one clock cycle period, during the
sense phase, as shown in the timing diagram in Figure 5.11. After generating these
three signals, they are given to inverters as delay elements; hence, the delayed
versions of the signals are generated. After obtaining these, for each timing signal
needed in the circuit, RS latch circuit is used, where the set and reset times of the
latches are defined by the three major signals and their delayed versions.

1

2

3
Sense phase
Figure 5.11: Timing diagram of generated major timing signals.

5.2.3.2. Digital Control Circuit

For the multi-bit readout structure, the digital control circuit is used for both
defining the sense and feedback phases, by a counter, and averaging the output bit-
stream in a defined period. The averaging is used for determining the feedback to
be applied to the sensor. For averaging, a signed magnitude counter is used, which

105
is reset in each 128 clock cycles. If the feedback is in negative direction, the counter
counts -1, and for positive feedback, it counts +1. When the readout circuit is in
sense phase, the counter does not count. Hence, at the end of this 128 clock cycles,
an average output value is generated. This value is compared with a previously
defined value, and if it is higher than that value, an additional feedback is applied
from the additional electrodes of the sensor. The feedback duration of the circuit
can be set to 4 different values, and for each feedback ratio the compared value of
counted feedback pulse changes accordingly, as shown Table 5.3. As seen, when
the output value reaches around 85% of the maximum feedback time value,
additional feedback is applied to the sensor.

Table 5.3: Feedback duration ratios and the compared values of the readout circuit.

Compared value of Compared value Compared value /

Feedback ratio Feedback ratio (%)
counted pulse (%) feedback ratio
1/8 12.5 % 14/128 10.9 % 87.2 %
3/10 30 % 32/128 25 % 83.3 %
7/14 50 % 51/128 39.8 % 79.6 %
17/24 70.8 % 78/128 60.9 % 86.0 %

When the electrode ratio of the accelerometer sensor is 1:2:2, where the last 2 stands
for the sense electrodes, the input acceleration coverage can be illustrated as given
in Figure 5.12.

0.12 af 0.12 af

-0.6 af -0.4 af -0.2 af 0 0.2 af 0.4 af 0.6 af

Figure 5.12: Illustration of the input acceleration coverage according to different operation
ranges.

106
5.3. Simulation Results Using Matlab Simulink

Matlab simulations for the circuit with adaptive feedback duration are performed
using the same models of the previous sigma-delta readout circuits, which are given
in Chapter 3 of the thesis. The feedback duration and timings of the model is
changed for different input accelerations and the simulations of the circuit are
performed.

The simulation results of the circuit are performed using Matlab Simulink model of
the circuit for observing the system operation. The Simulink model of the circuit is
given in Figure 5.13. The model uses accelerometer as a second order transfer
function. The input is applied in the form of acceleration and at the output of the
accelerometer proof-mass displacement value is obtained and converted to
capacitance. Then, the readout circuit block converts the capacitance to voltage,
which includes the readout gain, saturation voltages and the timings of the readout
circuit. After passing from the compensator stage, the signal is entered to the
comparator stage. At the feedback part of the model, there are three different
blocks, for applying feedback from three different electrodes. First feedback block
applies feedback in a time-multiplexed fashion according to the comparator output
of the circuit. The other blocks apply constant voltages to the electrodes, which are
used when high value of acceleration is applied to input. Basically, additional
feedback blocks generate offset feedback acceleration.

Two different simulations are performed with the simulink model. First, zero
acceleration is applied and the mass residual motion of the system is observed, as
shown in Figure 5.14. With an applied input acceleration of 1 micro-g the rms value
of mass residual motion becomes 7 picometers.

107
Figure 5.13. Simulink model of the accelerometer system.

Input Acceleration

Proof-mass Displacement

Figure 5.14. Simulink simulation results for observing mass-residual motion.

108
Another simulation is performed with 0.4 g peak input acceleration. The results are
shown in Figure 5.15. It is observed that the average value of the output bit stream
changes according to the input acceleration value.

Input Acceleration

Feedback acceleration

Proof-mass displacement

Output bit-stream

Figure 5.15. Simulink simulation results of the readout circuit.

109
5.4. Layouts

Figure 5.16 (a) and Figure 5.16 (b) give the layouts of the adaptive readout circuits
with varying feedback time and with multi-bit feedback, respectively. Both of the
readout circuits are designed with XFab 0.6m 2 metal – 2 poly CMOS process.
The total area of each chip is 2440 x 1180 m2 including the pads. The pin
descriptions of the readout circuits are given in Appendix B.

(a) (b)
Figure 5.16: Layout of the adaptive readout circuits with (a) varying feedback time, (b)
multi-bit feedback.

110
5.5. Comparison of Readout Circuits and Conclusion

There are 3 different versions of sigma-delta readout circuits designed. First design
has a classical sigma-delta readout circuit structure, where feedback duration ratio is
0.6, and a sampling frequency of 500 kHz.

The second design is an adaptive sigma-delta structure, which changes the feedback
duration according to the input acceleration value. If the input acceleration is
around zero, the feedback duration is kept shortest. Because of the sense phase is
constant, decreasing the feedback duration results in increasing sampling frequency.
Hence, maximum acceleration is decreased and sampling frequency is increased,
which both results in decreasing the mass residual motion and quantization noise.
For also obtaining wide operation range, when the input acceleration value is
increased, the feedback duration is increased. This results in having higher noise as
the input acceleration is increased. For zero input acceleration, the sampling
frequency is increased to 2.5 MHz, while the feedback duration ratio is 0.125. For
the maximum acceleration value, feedback duration ratio increases to 0.82, and the
sampling frequency is decreased down to 513 kHz.

The last design also employs low feedback duration ratio, but achieves the desired
input acceleration range by applying feedback from separate electrodes. The
feedback duration ratio is selectable in this circuit, and designed to operate best for
the feedback duration ratio of 0.3. In this case, since feedback is applied from the
40% of the electrode, amax is decreased down to 0.12af . However, the maximum
acceleration that can be applied to the sensor as feedback can be found by adding
the feedback acceleration supplied from feedback electrodes and the amax, which is
the maximum acceleration value at the sense electrodes. Hence, 0.72af is obtained
as the input acceleration range. In the case of this circuit, there is no increase in the
noise while increasing the range of the system. The sampling frequency of this
circuit is 2 MHz when operated with 0.3 feedback duration ratio.

111
A comparison of the noises of these three circuits is given in Table 5.4 and Table
5.5. Here, a sample accelerometer sensor is selected, and all the calculations are
made according to that sensor.

Table 5.4: Comparison of Sigma-Delta readout circuit in terms of quantization noise and
mass residual motion.

Calculated according Quantization Nrm

Sampling xrm
(1kHz BW)
to SOG Noise 𝒂𝒎𝒂𝒙 Linear
Frequency 𝟐
Accelerometers 𝒂𝒎𝒂𝒙 𝝅𝟐 𝟐𝝅𝒇𝒔 𝟒𝒇𝑩𝑾 𝑲 Range
fs 𝟒 𝒙
M = 0.719e-6 kg, K = 67 𝟏𝟓 𝑴𝟐.𝟓 𝒇𝒔 𝒎 𝒓𝒎

S-D Readout Circuit 500 kHz 10.3 µg 0.636 Å 54 µg ± 4g

S-D with
0g 2.5 MHz 0.185 µg 0.0045Å 0.171 µg
feedback pulse
± 5.3g
width High
513 kHz 9.67 µg 0.805 Å 67.5 µg
modulation g
S-D with multi-bit
2.0 MHz 0.11 µg 0.0051 Å 0.194 µg ± 4.7g
feedback

Table 5.5: Comparison of Sigma-Delta readout circuit in terms of noise sources.

Calculated according to SOG Quantization &

Mechanical Readout Circuit Expected Total
Accelerometers Mass Residual
Noise Noise Noise
M = 0.719e-6 kg, K = 67 Motion Noise
S-D Readout Circuit 55 µg/Hz 2.81 µg/Hz 1.22 µg/Hz 55 µg/Hz
S-D with feedback
0g 0.252 µg/Hz 2.81 µg/Hz 0.55 µg/Hz 2.87 µg/Hz
pulse width
modulation High g 68.2 µg/Hz 2.81 µg/Hz 1.22 µg/Hz 68 µg/Hz
S-D with multi-bit feedback 0.223 µg/Hz 2.81 µg/Hz 0.61 µg/Hz 2.87 µg/Hz

The next chapter gives the implementation and test results of the regular sigma-delta
readout circuit, which were explained in Chapter 4.

112
 CHAPTER 6

6. IMPLEMENTATION AND TESTS

This chapter presents the implementation of the readout circuits and the test results
of the system. There are 2 chips implemented and tested, both of which are sigma-
delta type fully differential readout circuits. With the second readout chip, the
performance of the circuit is improved, and a compensator circuit is included.
Section 6.1 of this chapter shows the implementation of the readout circuits together
with the accelerometer sensors, and describes the implemented external electronics,
used for filtering and decimating the output signal. Section 6.2 gives the open loop
test results of the readout circuit, which includes linearity, sensitivity and noise tests
with and without the accelerometer. Section 6.3 gives the closed loop test results.
Closed loop tests include linearity and sensitivity of the accelerometer system,
dynamic response, high acceleration tests, noise and temperature tests.

6.1. Implementation of the Readout Circuit

The first readout circuit is designed and implemented using XFAB 0.6µm, two
metal, two poly CMOS process. The photo of the chip is given in Figure 3.2. The
readout chip has 38 pads, with total dimensions of 2440x1180µm2. A test structure
of some analog blocks of the readout circuit is also implemented separately, which
is placed inside the area of the readout circuit, as seen in Figure 3.2. The chip is
bonded to a 44 pin smd package. A PCB is designed and manufactured, as shown in
Figure 6.2 and the package is soldered on the PCB. The accelerometer is also
bonded on the PCB, so that it is placed near the readout circuit as much as possible.
In this way, parasitic capacitances at the sense nodes are kept low.

113
Figure 6.1: Photo of fully differential sigma-delta capacitive accelerometer readout circuit.

Figure 6.2: Printed Circuit Board used for the tests of the readout circuit and the
accelerometer.

114
A better way is to wire-bond the accelerometer inside the same package with the
readout circuit. Hence, in the later times of this research, the second readout chip
and accelerometer is placed in the same package. The photograph of the second
readout chip is shown in Figure 6.3, which has 29 pads and has dimensions of
1180x1120µm2. An alumina substrate PCB is implemented and placed in the
package, as shown in Figure 6.4. Using this package, which isolates the system
from the electromagnetic noise sources with the metal covering, minimizes the
parasitic capacitances and impedance in between readout circuit and accelerometer.
This package is mounted on a PCB as shown in Figure 6.5. This PCB includes the
filtering and decimation circuitry. Hence, the closed loop tests are mostly
performed using this PCB. The open loop and closed loop tests performed are given
in the next sections separately.

Figure 6.3: Photograph of the second readout chip.

115
Figure 6.4: The accelerometer and the readout circuit wire-bonded in the same package,
using an alumina substrate PCB.

Figure 6.5: Accelerometer system together with the external filtering and decimation
circuitry, placed on a PCB.

116
6.2. Open Loop Test Results

For checking the functionality of the readout circuit, open loop tests are performed
initially. The open loop tests include the tests with and without the accelerometer
sensor.

6.2.1. Sensitivity of the Readout Circuit

First test is performed without the accelerometer, where input capacitive full-bridge
structure is constructed using 4 external capacitors. Then the differential analog
output of the circuit is observed. In this test, since the capacitor values have some
tolerances around 10%, first, value of each capacitor is measured using an
impedance analyzer. Then, by using various capacitor values, several data points
are collected. Test are performed with 4pF and 8pF integration capacitors. The
results of the test are shown in Figure 6.6, which give the capacitance to voltage
sensitivity of the readout circuit. The theoretical sensitivities are 0.625V/pF and
1.25V/pF for 8pF and 4pF integration capacitors, respectively. The experimental
results show 1.16 V/pF and 0.54 V/pF sensitivities as seen in Figure 6.6.

2,5
2,0
1,5 Cint = 4pF
Differential Output Voltage (V)

1,0 y = 1,157x - 0,891

0,5
Cint = 8pF
0,0 y = 0,544x - 0,971
-0,5
-1,0
-1,5
-2,0
-2,5
-2 0 2 4 6
Differential Capacitance (pF)
Figure 6.6: Sensitivity test results of the readout circuit.

117
As it is observed from the graph, there is an offset and non-linearity on the data.
These errors are mostly caused by the parasitic capacitances of the test environment.
Another reason of this non-linearity is the changing scale factor, because of the
changing total capacitance, while trying to change the differential value of the
capacitors. The dependence of the scale factor on the full-bridge capacitor values
was given in Chapter 3.

Another test is performed for observing the dependence of differential output

voltage to the integration capacitors, while there is a constant input capacitance
difference. The results of the test are shown in Figure 6.7. The test is performed for
two different input capacitance differences, in opposite signs. In the graph, the x-
axis show the inverse of the integration capacitor value. Theoretically, the output
voltage is linearly dependent on the inverse of integration capacitors. The results of
this test show the linear dependence of the output differential voltage. However, it
is observed that the output voltage starts saturating at the differential output voltage
of around ±1.2 V.

1,5

y = 3,445x + 0,066
1,0
Differential Output Voltage (V)

0,5

0,0

-0,5

-1,0

y = -5,170x - 0,055
-1,5
0,0 0,1 0,1 0,2 0,2 0,3 0,3
1/Cint (1/pF)

Figure 6.7: Differential output voltage versus inverse of the integration capacitance.

118
6.2.2. Sensitivity of the System Using Accelerometer

For observing the operation and sensitivity of the readout circuit with the
accelerometer, an SOG type accelerometer is bonded to the PCB. The parameters of
the accelerometer used were given in Table 1.2 in Chapter 1. The system is placed
on a rotating head, and accelerations in the range of -1g and +1g is applied by
changing the angle of the accelerometer according to the gravitational acceleration.
The differential output of the circuit is measured using a multimeter. The test is
repeated using different integration capacitor values. By collecting the data at
various angles, with different integration capacitor values, the test results shown in
Figure 6.8 is obtained. The test result show high linearity on acceleration in the
differential output range of -0.8V and +0.8V. The sensitivity of the system reaches
up to 4.37 V/g.

Figure 6.8: Open loop sensitivity tests of the readout circuit together with the
accelerometer.

119
6.2.3. Noise Tests

The noise tests of the circuit alone by itself and with the accelerometer are
performed separately. The open loop noise tests with the accelerometer are
performed using the same SOG accelerometer used in open loop sensitivity tests.
The noise test setup is shown in Figure 6.9. For measuring the noise of the circuit,
the readout circuit and the external circuitry for measuring noise is placed in a metal
box, to prevent any electromagnetic interference on the circuits. The noise
measurement is performed using “Agilent 35670A Dynamic Signal Analyzer”

The block diagram of the external circuitry used for noise measurements are also
shown in Figure 6.9. With this circuit, the differential output is converted into
single-ended output using a difference amplifier. Then, the signal is low pass
filtered using a second order low pass filter, and the noise signal is amplified using a
low noise amplifier structure. The output of the last stage amplifier is connected to
the Dynamic Signal Analyzer through a coaxial cable. Batteries are used as the
power supply of the circuitry in order to eliminate the supply noise.

Agilent 35670A
Dynamic Signal Analyzer

Faraday Cage

Difference Amplifier Low Noise Amplifier

Accelerometer Low-Pass
& D(jw)=1 - Filter G(jw) -
Readout Circuit LP(jw)

H(jw)

Figure 6.9: Test setup for noise measurements of the readout circuit.

120
To obtain the noise characteristic of the readout circuit, the total gain H(jw), and
noise response of the external circuitry is primarily measured. The measured gain
and noise characteristics of the external noise circuitry are shown in Figure 6.10.
Then, total noise of the readout circuit with the external noise circuitry is measured.
For obtaining the noise response of the readout circuit, squared noise of the external
circuitry is subtracted from squared value of total noise and divided by the gain of
the external circuitry at each frequency, as in Equation (6.1).

350

300
External Circuit Gain

250

200

150

100

50

0
0 200 400 600 800 1000 1200 1400 1600

Frequency

(a)
1E-04
External Circuit Noise (V/rtHz)

13.4 µV/Hz

1E-05
0 200 400 600 800 1000 1200 1400 1600

Frequency

(b)
Figure 6.10: (a) Gain of the external circuitry, (b)Noise of the external circuitry.

121
 2 2
𝑉𝑛,𝑡𝑜𝑡 𝑗𝑤 − 𝑉𝑛,𝑒𝑥𝑡 𝑗𝑤
𝑉𝑛,𝑎𝑐𝑐𝑒 (𝑗𝑤) = 2
(6.1)
𝐻𝑒𝑥𝑡 (𝑗𝑤)

With this noise measurement setup, the noise of the readout circuit and the open
loop noise with accelerometer are measured. The noise test results of the readout
circuit are shown in Figure 6.11, including the tests with and without the
accelerometer. The noise of a resistance is also measured and shown in Figure 6.11,
for being a reference in the noise measurements. The test results show that the
readout circuit has 0.81 µV/Hz of noise. When this noise value is multiplied by
the scale factor, it corresponds to a noise value of 1.22 µg/Hz. When the noise of
the readout circuit with the accelerometer is measured, 4.8 µg/Hz noise level is
obtained.

10,0

Readout Circuit &

Accelerometer
Noise (V/Hz)

1,0

Readout Circuit Noise

0,1 560 KW Resistor noise

0,01
0 50 100 150 200 250 300 350 400
Frequency (Hz)

Figure 6.11: Open loop noise test results of the readout circuit, with and without the
accelerometer.

122
6.3. Closed Loop Tests

Closed loop readout tests include linearity and sensitivity tests, dynamic response
test, high acceleration test, noise and bias drift tests.

6.3.1. Closed Loop Linearity and Sensitivity

First test for measuring the sensitivity of the closed loop system is performed using
the readout circuit with SOG accelerometer. The single bit output of the comparator
is acquired with an oscilloscope. With this way, a maximum of 32768 single-bit
data could be acquired at once. The system is placed on a rotating head, and 32768
data point is acquired and saved at various angle positions, applying accelerations
from -1g to +1g. For each position, the average value of the acquired data is
calculated. After calculating the accelerations from the measured angles, the graph
in Figure 6.12 is obtained. The graph shows the ratio of high pulses to the total, in
percentage, which can be named as pulse density.

Figure 6.12: Closed loop linearity and sensitivity test results of the accelerometer system,
with SOG accelerometer.

123
Later, using DWP process accelerometers, with the PCB given in Figure 6.5, the
linearity tests are repeated. The circuitry on the PCB used for this test includes two
PIC microcontrollers for filtering and decimating output bit-stream. At the output of
the microcontrollers, 16-bit data is obtained at 800 Hz sampling frequency. Both
the readout circuit and the accelerometer are bonded in the same package, and the
package is placed on the PCB. For performing the linearity tests, PCB is mounted
on a rotating head. 16-bit output data is acquired using a data acquisition system
with a computer. The rotating head is rotated 30 degrees for each measurement.
The data is acquired at 12 positions, completing the 360 degrees, which corresponds
to 7 different acceleration values. Data is acquired for 1 minute at each position.
The result obtained by 12 position measurements is shown in Figure 6.13. The
decimal equivalent of 16-bit output data, which is obtained by averaging the 1
minute data for each position, versus acceleration is shown in Figure 6.13, where the
acceleration is calculated according to the position angle. The scale factor
according to the decimal value of output data is found as 1660 per g, where the
output can range from 0 to 64000.

4
x 10
3.3

3.25

3.2
Raw Output Data

3.15

3.1

3.05

2.95
0 0.5 1 1.5 2 2.5 3
5
x 10

Figure 6.13: Output data versus time graph of the linearity test results of the closed loop
accelerometer system, with 12 angular positions.

124
As seen on the Figure 6.14, R2 non-linearity is found as 0.0035%. If the system is
assumed linear also for high acceleration values, the measurement range can be
calculated from the scale factor obtained, by dividing the total output value range to
the scale factor. This calculation results in an operating range between -19g and
+19g. If the closed loop operation range of the SOG accelerometer is calculated
similarly from the previous closed loop linearity test, it is calculated as ±4g.

33000

32500

32000
Output Data

31500

31000

30500

30000

29500

29000
-1,5 -1 -0,5 0 0,5 1 1,5
Input Acceleration (g)

Figure 6.14: Averaged output data versus acceleration.

6.3.2. Closed Loop Dynamic Response

With this test, the response of the accelerometer system to a sinusoidal input is
observed. For applying a sinusoidal acceleration, the system is placed in a rate
table. The PCB is placed near the border of the plate of the rate table, so that the
distance between the accelerometer and the center of the rate table is maximized.
Then, a rotational vibration is applied on the accelerometer. Since the amplitude of

125
motion is small, the acceleration can be assumed linear, in the direction parallel to
the boundary of the circle of the rate table plate, which is illustrated in

Figure 6.15. The rate table sets the amplitude of sinusoidal vibration displacement.
Knowing the positional amplitude of the vibration, the acceleration can be
calculated by taking the twice derivative of the position, as given in Equation (6.2).

𝑑2 𝑑2 ∆𝑥 ∆𝑥
𝑎 𝑡 = 𝑥 𝑡 = sin 2𝜋𝑓𝑡 = 2𝜋𝑓 2 sin 2𝜋𝑓𝑡 (6.2)
𝑑𝑡 2 𝑑𝑡 2 2 2

The acceleration is applied at 10 Hz, with 3 different amplitudes of 70 mg, 140 mg

and 210 mg. The data is collected for 5 seconds, using the data acquisition board of
“National Instruments”. The acquired data is processed in Matlab Simulink using
the digital low pass filtration blocks as explained in Chapter 3. Hence, the filtered
output waveform is obtained as shown in Figure 6.16.

Figure 6.15: Test setup for measuring the dynamic response of the accelerometer, using the
rate table.

126
Figure 6.16: Differential analog outputs of the readout circuit, with 70mg, 140 mg and 210
mg amplitude sinusoidal inputs.

6.3.3. High G Acceleration Tests

Using a rotating head, maximum acceleration that can be applied on the

accelerometer cannot exceed the gravitational acceleration. Hence, to apply higher
values of acceleration, centrifugal force is used. For this purpose, the accelerometer
system is mounted on a centrifuge table. The applied acceleration to the system can
be expressed as in equation (6.3).

𝑎𝑐𝑒𝑛𝑡 = 𝑤 2 ∙ 𝑟 (6.3)

where, r is the length of the arm of centrifugal table and w is the angular speed of
the table.

The test setup prepared for the centrifuge test is illustrated in Figure 6.17. For
mounting the PCB on the centrifuge table, a durable fixture is prepared as shown in
Figure 6.18. With this fixture, it is able to change the direction of the applied

127
acceleration, just by rotating the top part of the fixture. After the PCB is mounted,
16-bit parallel data cables are connected to the data acquisition system through the
slip rings of the centrifuge table. The data is acquired and observed real time on the
computer.

Figure 6.17: Illustration of high acceleration tests performed on centrifuge table.

Figure 6.18: Fixture constructed for mounting the PCB on the centrifuge table.

The centrifuge table is rotated to achieve accelerations of ±4g, ±6g, ±8g, ±10g.
Hence, the output of the accelerometer is observed as given in Figure 6.19. In this
figure, the starting and ending of the rotation is shown. The output of the circuit is
given in terms of applied acceleration in g‟s, which is normalized by multiplying
with the scale factor.

128
 10 g
8g
6g
4g

-4 g
-6 g
-8 g
-10 g

Figure 6.19: High acceleration test results of the closed loop accelerometer system.

Then, the average of each data is calculated at different accelerations, so the

linearity plot is obtained as given in Figure 6.20. R2 non-linearity of the closed loop
system is 0.0004% as shown in the plot.

Figure 6.20: Linearity graph obtained from high acceleration tests.

129
6.3.4. Noise Tests

The closed loop noise tests are performed with two different methods. The first
method is directly measuring the output bit-stream, after passing through a low pass
filter. With this method, test setup is prepared as illustrated in Figure 6.21. For
measuring the noise, “Agilent 35670A Dynamic Signal Analyzer” is used. The
results of the noise test are shown in Figure 6.22. As it is expected, the noise of the
closed loop system increases with the increasing frequency, because of the shaped
quantization noise. Figure 6.22 also includes the open loop noise test results and
makes a comparison in between. The open loop test results are higher than the
results given earlier, since the used accelerometers in this test is DWP
accelerometers and both the scale factor and mechanical noise of the accelerometer
is different.

130
Figure 6.21: Illustration of the test setup constructed for closed loop noise tests.

Closed Loop
Noise µg/√Hz

Open Loop

Frequency (Hz)

Figure 6.22: Differential analog outputs of the readout circuit.

With the second method of measuring the closed loop noise, 16-bit filtered and
decimated output, using the PIC microcontrollers, are acquired for an hour, using a
data acquisition board.

Then, “Allan Variance” method is used to extract the noise and bias drift values of
the accelerometer system. The collected data is processed and Allen-variance graph
is plotted. Figure 6.23 shows a sample plot of Allan variance analysis graph. By
using the 6 different parts of this plot, various sources of non-idealities can be found
as shown in Figure 6.23 and Table 6.1. By using  and  values at any point in the
specified region, one can find the parameters given in Table 6.1 using the given
equations. Allan Variance is a commonly used technique for determining the noise
and bias drift of various systems. Hence, there are commercial software programs
for processing the raw data according to Allan Variance method. By using one of
these software programs, Alavar 5.2, Allan Variance plot of the acquired data is
obtained, as shown in Figure 6.24. From the Allan Variance plot, the bias drift is
calculated as 74µg and the noise of the system is calculated as 86µg/Hz.

131
 Variance

Time (s)

Figure 6.23: Sample Allan Variance graph.

Table 6.1. Regions in Allan variance graph and their meanings.

Region Name Symbol Equation Region Slope

Quantization Noise Q  a  3.Q /  -1

Angle Random Walk N  WN  N /  -1/2

Sinusoidal W0  S  wO . sin 2 (f 0 ) (f 0 ) sinusoidal

Bias Instability B  B  0.6648 B 0

Random Walk K  RW  K  1/2

3
Ramp R  R  R / 2 1

132
 Angle
Random
Walk

Bias Instability

Figure 6.24: Allan variance graph obtained from the tests.

6.3.5. Temperature Tests

For measuring the temperature dependency of the system, a temperature sensor chip,
designed in the research group is used. The temperature sensor has 0.05C of
maximum inaccuracy, and -165C to 200C measurement range. The closed loop
output of the system is acquired with the data acquisition board while changing the
temperature of the system. The temperature data is given as analog voltage, and it is
collected using an HPVEE program, with a multimeter. The results are shown in
Figure 6.25. It is observed that the output of the system is highly dependent on the
temperature, where one degree change in the temperature results in 44 mg
corresponding change at the closed loop output of the system. The temperature
dependency may result from both the temperature dependency of the accelerometer
and the readout circuit components.

133
 1.8

1.6

1.4

Acceleration Output (g)

1.2

1.0

0.8

0.6

0.4

0.2
0 1 2 3 4 5 6 7
815 Hz ~15 min x105
45

40
Temperature Output (Co)

35

30

25

20

15
0 2000 4000 6000 8000 10000 12000 14000
15 Hz ~15 min

Figure 6.25: Linearity graph obtained from high acceleration tests.

6.4. Conclusions

The test results of the readout circuits showed high sensitivity, high resolution in
open loop mode with SOG accelerometers. The closed loop tests were performed
using DWP accelerometers and high operation range with low noise of below 100
µg is obtained. However, there are some temperature dependency problems of the
readout circuit, and can be solved using a temperature sensor together with the
accelerometer and performing temperature compensation at the digital output data.

134
Table 6.2. Summary of system level test results with SOG and DWP accelerometers.

Test Test Result (SOG) Test Result (DWP)

Open loop sensitivity (Cint = 4 pF) 2.36 V/g 0.53 V/g
Open loop noise 4.8 µg/Hz 8.8 µg/Hz
Closed loop sensitivity 12.5%/g pulse density 2.6%/g pulse density
Closed loop range ±4 g ±18.5 g
Closed loop noise ~50 µg/Hz 86 µg/Hz

135
 CHAPTER 7

7. CONCLUSION AND FUTURE WORK

This thesis presented the design and implementation of a high performance readout
circuit for capacitive MEMS accelerometers. The following is a summary of the
accomplishment of this research.

 The design of a complete accelerometer readout circuit.

A sigma-delta closed loop capacitive accelerometer readout circuit was

designed, with 0.6µm XFAB CMOS process. The first readout circuit
implemented was fully functional, and open-loop noise results of
4.81 µg/rtHz were obtained using SOG type accelerometers. To improve
the first readout circuit with minor changes, and with the addition of a lead
compensator to guarantee the system stability, a second version of the
readout circuit was implemented using the same CMOS process.

 The modeling and simulation of the accelerometer system and the

verification of the operation.

The models of the accelerometer were constructed in both Cadence and

Matlab Simulink environment. The model constructed in Simulink
included the modeling of sigma-delta readout circuit together with the
accelerometer sensor. This model and the performed simulations gave a
better understanding of electromechanical sigma-delta modulating concept,
and they were used for observing the system stability. The Accelerometer
model constructed in Cadence environment gives accurate simulations
results. However, due to long simulation times, the use of this model is not
practical for simulations. For future research, if the readout circuit is

136
 designed with mixed signal circuitry in Cadence, which gives simpler
models for the digital structures in the readout circuit, the simulation times
will be considerably reduced. Hence, more accurate closed loop complete
system simulation results can be obtained.

 The development of new readout techniques to improve the performance of

the accelerometer system.

Two adaptive readout circuit structures were designed, which were verified
to improve the performance of the closed loop system. The first adaptive
circuit design decreased the quantization noise and mass residual motion by
making the circuit available to operate at higher sampling frequencies.
Operation at higher frequency was achieved by decreasing the feedback
time of the readout. The feedback time of the circuit was decreased only
for low input accelerations. For the circuit to be capable of applying
adequate feedback force at high input accelerations, the feedback time was
lengthened. Hence, the resolution was decreased at high input
accelerations. This resulted in a high resolution at low input acceleration
values and lower resolution at high input accelerations.

The second adaptive readout circuit design uses multi-bit feedback to the
accelerometer, and hence increases the resolution of the system, by
decreasing the feedback time and increasing sampling frequency.
However, this readout circuitry needs a multi-electrode accelerometer
structure, to be able to apply the multi-bit feedback to the accelerometer.

 The implementation and tests of the sigma-delta readout circuit together

with the sensors.

Using the sigma-delta readout circuit, low noise open loop and closed loop
test results were obtained. With the circuitry combined with DWP
accelerometers, an operational range of around ±18.5 g was obtained and
verified with the tests. The closed loop test results showed high linearity in

137
 this operational range. The noise of the system was measured as 86
µg/Hz, and a bias drift of 74 µg was obtained.

There was a bias drift at the output in the first 5-10 minutes of the closed
loop tests. The reason of this drift was understood after the temperature
tests of the accelerometer system were performed. It was observed that the
output of sigma-delta loop was highly dependent on the temperature of the
environment.

7.1. Future Directions

A micro-g resolution closed loop readout circuitry, together with the accelerometer,
is still a need for navigational purposes. In order to achieve higher resolution, the
first thing is to increase the sampling frequency of the readout circuit by employing
different structures, or increasing the speed of the readout circuitry by speeding up
the components used in the readout. While increasing the sampling frequency, the
stability of the system should be taken into consideration. The stability can be
verified using the models of accelerometer system.

The noise sources of the mechanical sensor and readout circuit should be added in
the MAtlab Simulink model of the accelerometer, in order to achieve more accurate
modeling, and the observation of non-idealities.

High temperature dependency of the system is a problem, and hence the temperature
independence of the. Readout circuit components and the accelerometer should be
improved. Another way of solving this problem is to use temperature compensation
for the output data of the readout circuit.

The new readout techniques can be used to increase the performance of

accelerometer systems. One of these techniques can be a multi-bit structure, which
has already been designed. After the implementation of this readout circuit, it needs
to use a multiple electrode accelerometer structure. Hence a properly sized multiple
electrode accelerometer is needed.

138
Another readout circuit using multi-bit feedback can be designed, where, to prevent
non-linearity, time-based quantized feedback can be applied. Hence, instead of
using multi-bit quantization in magnitude domain, it would be used in time domain.
However, with this configuration, the linearity of the analog to digital conversion
stage should be taken into consideration.

139
 REFERENCES

[1] Navid Yazdi, Farrokh Ayazi, Khalil Najafi, “Micromachined Inertial

Sensors” Proceedings of the IEEE, Vol. 86, No. 8, pp. 1640-1659, August
1998.

[2] B. V. Amini, R. Abdolvand, and F. Ayazi, “Sub-Micro-Gravity Capacitive

SOI Microaccelerometers,” Transducers 2005, pp. 515-518.

[3] J. Chae, H. Kulah and K. Najafi, “Hybrid Silicon-On-Glass (SOG) Lateral

Micro-Accelerometer with CMOS Readout Circuitry,” MEMS 2002,
pp. 623-626.

[4] B. Amini, S. Pourkamali and F. Ayazi, “A High Resolution, Stictionless

CMOS Compatible SOI Accelerometer with a Low Noise, Low Power
0.25μm CMOS Interface,” MEMS 2004, pp. 572-575.

[5] H. Kulah, J. Chae and K. Najafi, “Noise Analysis and Characterization of A

Sigma-Delta Capacitive Silicon Microaccelerometer,” Transducers 2003, pp.
95-98.

[6] X. Jiang, F. Wang, M. Kraft and B. E. Boser “An Integrated Surface

Micromachined Capacitive Lateral Accelerometer with 2 g/Hz
Resolution,” Solid State Sensor and Actuator Workshop, 2002, pp. 202-205.

[7] M. .A. Lemkin, M. A. Ortiz, N. Wongkomet, B. E. Boser and J. H. Smith “A

3-Axis Surface Micromachined g Accelerometer,” ISSCC 1997, pp. 202-
203.

[8] P. M. Zavracky, F. Hertley, N. Sherman, T. Hansen, an K. Warner, “A new

force balanced accelerometer using tunneling tip position sensing,” in Digest
International Conference onSolid-State Sensors and Actuators, pp. 50-51,
1993.

140
[9] A. Garcia-Valenzuela and M. Tabib-Azar, “Comparative study of
piezoelectric, piezoresistive, electrostatic, magnetic and optical sensors,”
Integrated Optics and Microstructures II, vol. SPIE-2291, pp. 125-142, 1995.

[10] Bernhard E. Boser, “Capacitive Interfaces for Monolithic Integrated

Sensors,” RF Analog-to-Digital Converters; Sensor and Actuator Interfaces;
Low-Noise Oscillators, PLLs and Synthesizers, R. J. van de Plassche, J. H.
Huijsing, W. M.C. Sansen (eds.), Kluwer Academic Publishers, November
1997.

[11] Weijie Yun, R.T. Howe, P.R. Gray, “Surface micromachined, digitally force-
balanced accelerometer with integrated CMOS detection circuitry,” 5 th
IEEE Solid-State Sensor and Actuator Workshop, Hilton Head Island,
S.C., pp. 126-131. June 21-25, 1992.

[12] S. J. Sherman, et.al, “Low cost monolithic accelerometer,” Dig. VLSI

Circuits Symp., pp. 34-35, June 1992.

[13] W. Yun, R. T. Howe, and P. R. Gray, “Surface micromachined digitally

force balanced accelerometer with integrated CMOS detection circuitry,”
Solid-State Sensor and Actuator Workshop, Hilton Head Island, SC, USA,
pp. 126-131, June 1992.

[14] N. Yazdi, H. Kulah, K. Najafi, “Precision Readout Circuits for Capacitive

Microaccelerometers,” Proceedings of IEEE Sensors 2004, Vol.1, pp. 28-31,
October 2004.

[15] Mark A. Lemkin, Bernhard E. Boser, David Auslander, Jim H. Smith, “A 3-

Axis Force Balanced Accelerometer Using a Single Proof-Mass,”
Transducers Dig. of Tech. Papers, pp. 1185-1188, June 1997.

[16] Mark Lemkin, Bernhard E. Boser, “A Three-Axis Micromachined

Accelerometer with a CMOS Position-Sense Interface and Digital Offset-
Trim Electronics,” IEEE Journal of Solid-State Circuits, Vol. 34, No. 4,
April 1999.

141
[17] Crist Lu, Mark Lemkin, Bernhard E. Boser, “A Monolithic Surface
Micromachined Accelerometer with Digital Output,” ISSCC Dig. Tech.
Papers, pp. 160–161, Feb. 1995.

[18] Jiangfeng Wu, Gary K. Fedder, L. Richard Carley, “A Low-Noise Low-

Offset Capacitive Sensing Amplifier for a 50 g/Hz Monolithic CMOS
MEMS Accelerometer,” IEEE Journal of Solid-State Circuits, vol. 39, pp.
722 - 730, May 2004.

[19] G. Zhang, H. Xie, L.E. de Rosset, and G. Fedder, “A Lateral Capacitive

CMOS Accelerometer with Structural Curl Compensation,” Proceedings of
The 12th IEEE International Conference on Micro Electro Mechanical
Systems (MEMS '99), pp. 606 – 611, January, 1999.

[20] Navid Yazdi, Khalil Najafi, “An All-Silicon Single-Wafer Micro-g

Accelerometer with a Combined Surface and Bulk Micromachining
Process,” IEEE/ASME Journal of Microelectromechanical Systems, vol. 9,
no. 4, pp. 1-8, December 2000.

[21] H. Kulah, N. Yazdi, K. Najafi, “A CMOS Switched-Capacitor Interface

Circuit for an Integrated Accelerometer,” MIDwest Symposium on Circuits
and Systems, Lansing MI, pp. 252-255, August 8-11, 2000.

[22] Haluk Kulah, “Closed-Loop Electromechanical Sigma-Delta Microgravity

Accelerometers,” Ph.D. Thesis, Department of Electrical Engineering and
Computer Sciences, University of Michigan, February 2003.

[23] Babak Vakili Amini, Farrokh Ayazi, “A 2.5V 14-bit  CMOS SOI
Capacitive Accelerometer,” Tech. Dig. IEEE International Solid-State
Circuits Conference (ISSCC 2004), pp. 314-315, San Francisco, CA,
Feb. 2004.

[24] B. Vakili Amini, R. Abdolvand, and F. Ayazi, “Sub-Micro-Gravity

Capacitive SOI Microaccelerometers,” Proc. Transducers '05, pp. 515-518
Seoul, Korea, June 2005.

142
[25] B.V. Amini, S. Pourkamali, M. Zaman, F. Ayazi, “A New Input switching
Scheme for a Capacitive Micro-g Accelerometer,” Tech. Dig. 2004
Symposium on VLSI Circuits, pp. 310-313, Honolulu, HI, June 2004.

[26] Junseok Chae, Haluk Kulah, Khalil Najafi, “A Hybrid Silicon-On-Glass

(SOG) Lateral Micro-Accelerometer with CMOS Readout Circuitry,” The
Proceedings of the 15th IEEE International Conference on Micro Electro
Mechanical Systems (MEMS 2002), Las Vegas, NV, January 2002.

[27] Mark Lemkin, Bernhard E. Boser, “A Micromachined Fully Differential

Lateral Accelerometer,” CICC Dig. Tech. Papers, pp. 315-318, May 1996.

[28] Xuesong Jiang, Feiyue Wang, Michael Kraft, Bernhard E. Boser, “An
Integrated Surface Micromachined Capacitive Lateral Accelerometer with 2
g/Hz Resolution,” Tech. Digest of Solid State Sensor and Actuator
Workshop, Hilton Head Island, pp. 202-205, USA, June 2002.

[29] S. E. Alper, “MEMS Gyroscopes for Tactical-Grade Inertial

Measurement Applications,” Ph.D. Dissertation, Middle East Technical
University, September 2005.

[30] S. Park, “Principles of Sigma-Delta Modulation for Analog-to-Digital

Converters,” Motorola Inc., USA, 1990.

[31] S. R. Norsworty, R. Schreier, G. C. Temes, “Delta-Sigma Data Converters,

Theory, Design and Simulation,” IEEE Press, 1996.

[32] B. P. Brandt and B. A. Wooley, “A 50 MHz Multibit Sigma-Delta

Modulator for 12-b 2 MHz A/D Conversion,” IEEE Journal of Solid
State Circuits 1991, pp. 1746-1756.

[33] B. P. Brandt, D. E. Wingard and B. A. Wooley, “Second-Order Sigma-Delta

Modulation for Digital-Audio Signal Acquisition,” IEEE Journal of Solid
State Circuits 1991, pp. 618-627.

143
[34] M. A. Lemkin, M. A. Ortiz, N. Wongkomet, B. E. Boser and J. H.
Smith “A 3-Axis Surface Micromachined Σ∆ Accelerometer,” ISSCC
1997, pp. 202-203.

[35] R. Gallorini and N. Abouchi, “A Capacitance Meter based on an

Oversampling Sigma-Delta Modulator and Its Application to Capacitive
Sensor Interface,” ICECS 2001, pp. 1537-1540.

[36] T. Kajita, U. Moon and G. C. Temes, “A Noise Shaping Accelerometer

Interface Circuit for Two Chip Implementation,” ISCAS 2000, pp. 337-
340.

[37] J. B. Starr, “Squeeze-Film Damping in Solid-State Accelerometers,”

Techical Digest Solid-State Sensor and Actuator Workshop, pp. 44-47,
Hilton Head, S.C.,June 1990.

144