Document [original]

Design of Low-Power K-Band Circuits in CMOS

Technology

Vorgelegt von

M.Sc. Mohammed Kamal Abdelrahman Ali

geb. in Fayoum, ¨

Agypten

Von der Fakult¨

at IV - Elektrotechnik und Informatik

der Technischen Universit¨

at Berlin

zur Erlangung des Akademischen Grades

Doktor der Ingenieurwissenschaften

- Dr.-Ing. -

genehmigte Dissertation

Promotionsausschuss:

Vorsitzender: Prof. Dr.-Ing. Rolf Schuhmann

Gutachter: Prof. Dr.-Ing. Georg B¨

ock

Gutachter: Prof. Dr.-Ing. Peter Weger

Gutachter: Dr.-Ing Wilhilm Keusgen

Tag der wissenschaftlichen Aussprache: 25.11.2014

Berlin, 2015

To my family

Acknowledgement

I would like to express my gratitude to my supervisor Prof. Dr.-Ing. Georg B

ock who gave me

not only the opportunity to conduct my PhD work in the Microwave Engineering Laboratory, at

the Technical University of Berlin, Germany, but he was always directing me to the right and the

shortest way to achieve the goals, i could not have accomplished my work without his support

and advices. I will always be grateful to the Deutscher Akademischer Austauschdienst (DAAD)

which financially supported me during my stay. It was also a great chance to work with and learn

from my colleagues who are belonging to different cultures, i really learned too much from them.

Special thanks to Dr.-Ing. Viswanathan Subramanian, the first team leader who gave me advices

in the beginning of my work, and helped me to learn the hand on skills. Finally, all love, and

all gratitude to my wife ”Amaal”, and my mother ”Safya” who were always encouraging and

backing me up, i will be always sincere to them. In the name of Allah, the most compassionate,

the most Merciful.

Mohammed Kamal Ali

Berlin, March 2014

Abstract

Millimeter-wave (mm-wave) CMOS transceivers have attracted interest in recent years,

especially in the

24 GHz

and

60 GHz

ISM bands. As indispensable building blocks in a

wireless transceiver, frequency generation and conversion circuits are confronted by many design

challenges. At mm-wave frequencies, the voltage controlled oscillator (VCO) suffers from a

poor phase noise and a limited tuning range, while the frequency divider is usually accompanied

by a narrow locking range and high power consumption. Direct down-conversion mixer, on

the other hand, suffers from a very poor noise figure as a result of the flicker noise. For the

generation of the LO signals with low power consumption, the design of the VCO encounters

difficulties. Moreover, the generation of IQ signals with high accuracy over a wide band is much

more challenging.

In this dissertation, new techniques and optimized topologies are proposed to improve the

performance of mm-wave frequency generation and conversion circuits. Employing a transformer

feedback topology, a

24 GHz

LC-VCO was designed using

130 nm

CMOS technology. It

achieves a wide tuning range and consumes one fourth of the power that the conventional

LC-VCO requires. The designed transformer feedback VCO has comparable phase noise to

the conventional counterpart. Two designs of the mm-wave frequency divider have also been

presented. One design is a Ka-band Miller frequency divider in 130 nm CMOS technology. A

specific optimization approach was followed to achieve a maximum locking range for low power

consumption. The loop gain is maximized over the widest possible frequency range, the Qfactor

of the load inductor, on the other hand, is chosen such that the third order harmonic is rejected

by a required predetermined value. Measurement results of the designed circuits show that their

performance compares to the state-of-the-art developments and satisfies the

24 GHz

FMCW

radar application.

The other design is a

45 GHz

static frequency divider using

90 nm

CMOS technology.

New techniques such as ”inductive peaking”, ”split resistors”, and ”asymmetric sizing of input

viii

transistors” are employed to minimize the power consumption of the divider. Moreover, direct

down-conversion mixer is designed. Overcoming the flicker noise effect was a challenge. Novel

current bleeding network was therefore proposed. Not only the noise performance is improved but

the conversion gain is increased as well. Based on the designed static frequency divider and the

direct down-conversion and up-conversion mixers, complete IQ-modulator and demodulator were

eventually implemented using

90 nm

CMOS technology. Measurement results of the designed

and tested circuits show that their performance compares to the state-of-the-art developments and

satisfies the 60 GHz high-data-rate communication application of the IEEE 802.15.3c standard.

Kurzfassung

Die Entwicklung von Millimeterwellen-Transceivern auf Basis moderner CMOS-Technologien

steht derzeit weltweit im Fokus von Wissenschaft und Forschung. Dabei sind insbesondere

Schaltungen f

ur die ISM-B

ander bei 24 GHz bzw. 60 GHz von Interesse. Beim Entwurf von

CMOS-Schaltungen zur Erzeugung und Umsetzung von Frequenzen in diesen Frequenzbereichen

ergeben sich zahlreiche unterschiedliche Problemstellungen. So ist einerseits die Performance

von spannungsgesteuerten Oszillatoren (VCOs) bei Millimeterwellen-Frequenzen durch ein

schlechtes Phasenrauschen sowie durch einen begrenzten Abstimmbereich beeintr

achtigt. Ander-

seits k

onnen Frequenzteiler mit großer Bandbreite und geringem Leistungsverbrauch schwierig

realisiert werden. Infolge des Flickerrauschens und der einhergehenden schlechten Rauschzahl

gestaltet sich des Weiteren auch das direkte Heruntermischen in das Basisband mit sog. Direct-

Down-Conversion-Mischern als schwierig. Die Erzeugung des LO-Signals bei gleichzeitig

geringer Leistungsaufnahme geht mit einigen Schwierigkeiten einher. Zu guter Letzt ist ebenso

die Erzeugung von IQ-modulierten Signalen hoher Genauigkeit

uber gr

oßere Bandbreiten eine

Herausforderung beim Entwurf solcher Transceiver.

Zur

Uberwindung dieser Problemstellungen werden in der vorliegenden Dissertation neuartige

Methoden und optimierte Schaltungen zur Erzeugung und Umsetzung von Millimeterwellenfre-

quenzen untersucht. Dies beinhaltet einen 24-GHz-LC-VCO, der auf Basis einer Transformer-

Feedback-Topologie in einer 130-nm-CMOS-Technologie realisiert wurde. Dieser VCO erreicht

einen sehr großen Abstimmbereich und zeichnet sich im Vergleich zu konventionellen LC-

VCOs mit einer deutlich reduzierten Leistungsaufnahme bei vergleichbarem Phasenrauschen

aus. Des Weiteren sind im Rahmen dieser Arbeit zwei Frequenzteiler entwickelt worden. Der

erste Teiler wurde f

ur das Ka-Band in einer 130-nm-CMOS-Technologie entworfen. Dieser sog.

Miller-Divider wurde derart optimiert, dass eine maximale Bandbreite bei gleichzeitig minimaler

Leistungsaufnahme realisiert werden konnte. Die Schleifenverst

arkung wurde

uber den gr

oßt

oglichen Frequenzbereich maximiert. Auf der anderen Seite wurde die G

ute der Lastinduktivit

derart gew

ahlt, dass die Unterdr

uckung der 3. Harmonischen auf den erforderlichen Wert gehal-

ten werden konnte. Die mit den realisierten Schaltungen erzielten Messergebnisse entsprechen

dem neusten Stand der Technik und gen

ugen den Anforderungen von 24-GHz-FMCW-Radar-

Systemen.

Der zweite Frequenzteiler, ein sog. Static-Divider, wurde in einer 90-nm-Technologie f

eine Frequenz von 45 GHz realisiert. Neuartige Techniken wie Inductive Peaking,Split Resistors

und die asymmetrische Auslegung der Eingangstransistoren wurden hierbei implementiert um

die Leistungsaufnahme der Teiler deutlich zu minimieren. Als weiterer Transceiver-Baustein

wurde außerdem ein Direct-Down-Conversion-Mischer implementiert. Dessen Flickerrauschen

konnte durch den Einsatz eines neuartigen Current-Bleeding-Netzwerkes verbessert werden.

Neben den verbesserten Rauscheigenschaften konnte auf diese Art und Weise außerdem auch ein

gesteigerter Konversionsgewinn erzielt werden. Auf Basis des entwickelten Static-Dividers sowie

der realisierten Mischer konnte schließlich ein kompletter IQ-Modulator bzw. IQ-Demodulator

in einer 90-nm-Technologie entworfen werden. Die mit den realisierten Schaltungen erzielten

Messergebnisse entsprechen dem neusten Stand von Wissenschaft und Technik und erf

ullen

die Anforderungen von Nahbereichskommunikationssystemen im 60-GHz-Band gem

aß IEEE

802.15.3c.

Contents

List of Figures xv

List of Tables xxi

1 Introduction 1

1.1 Motivation..................................... 1

1.2 ScopeofResearch................................. 1

1.2.1 FMCW Radar Transceiver . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2.2 Short Range High Data Rate WLAN Transceiver . . . . . . . . . . . . 5

1.3 DifferentialSignaling............................... 7

1.4 ThesisOrganization................................ 8

2 RFICs – Essentials and Fundamentals 9

2.1 FrequencyDividers................................ 9

2.1.1 MillerDivider .............................. 9

2.1.2 Static Frequency Divider . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Voltage Controlled Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2.1 Principles and Design Issues . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 DirectConverisonMixer ............................. 18

2.3.1 Gain of the Gilbert-Cell mixer . . . . . . . . . . . . . . . . . . . . . . 19

3 CMOS Devices for RFIC – Technology Overview 23

3.1 Introduction.................................... 23

xii CONTENTS

3.2 Active Devices in CMOS Technology . . . . . . . . . . . . . . . . . . . . . . 24

3.2.1 Gain.................................... 24

3.2.2 Noise ................................... 26

3.3 Passive Devices in CMOS Technology . . . . . . . . . . . . . . . . . . . . . . 27

3.3.1 Inductor.................................. 27

3.3.2 CMOSVaractor.............................. 29

3.3.3 MetalCapacitor.............................. 30

3.3.4 TransmissionLine ............................ 32

3.4 Conclusions.................................... 33

4 Design of RF Circuits for the 24 GHz FMCW Radar Transceiver 35

4.1 SystemOverview................................. 35

4.2 Miller Frequency Divider . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.2.1 DesignApproach............................. 36

4.2.2 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.3 Voltage Controlled Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.3.1 Conventional and Transformer Feedback VCO’s . . . . . . . . . . . . 43

4.3.2 DesignApproach............................. 45

4.3.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5 Design of the IQ Modulator for the 60 GHz WLAN Transceiver 51

5.1 SystemOverview................................. 51

5.2 Static Frequency Divider . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

5.2.1 DesignApproach............................. 51

5.2.1.1 Evaluating Amplifier Design . . . . . . . . . . . . . . . . . 53

5.2.1.2 Sizing of the Latch Stage . . . . . . . . . . . . . . . . . . . 54

5.2.1.3 Design of the Splitting Resistor Ratio . . . . . . . . . . . . . 55

5.2.1.4 OutputBuffers......................... 57

5.2.2 IQImbalance............................... 58

5.2.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

CONTENTS xiii

5.3 Direct Up-Conversion Mixer . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.3.1 DesignApproach............................. 64

5.3.2 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.4 IFAmplifier.................................... 68

5.4.1 Cascode Differential Amplifier . . . . . . . . . . . . . . . . . . . . . . 68

5.5 SystemIntegration ................................ 69

6 Design of the IQ Demodulator for 60 GHz WLAN Transceiver 75

6.1 SystemOverview................................. 75

6.2 Low-Flicker Noise Direct Down-Conversion Mixer . . . . . . . . . . . . . . . 75

6.2.1 Noise Mechanism and Current Bleeding Network . . . . . . . . . . . . 75

6.2.2 DesignApproach............................. 83

6.2.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6.3 SystemIntegration ................................ 89

7 Conclusions 93

7.1 FutureWork.................................... 94

Bibliography 95

List of Publications 103

xiv CONTENTS

List of Figures

1.1 Transmit and Receive signals from a point target . . . . . . . . . . . . . . . . . 2

1.2 Block Diagram of the 24 GHz FMCW Radar Transceiver . . . . . . . . . . . . 3

1.3 Channelization of the IEEE 802.15.3c standard . . . . . . . . . . . . . . . . . 5

1.4 Block Diagram of the WLAN Transceiver . . . . . . . . . . . . . . . . . . . . 6

2.1 Block Diagram of the Miller Divider . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Miller Divider (a) Generic Topology (b) Realized with RC filter . . . . . . . . . 11

2.3 Miller divider with wide band phase shifter (a) Waveform (b) Generic Topology 12

2.4 MillerdividerwithBPF.............................. 12

2.5 (a) Divided and third order harmonic (b) Superimposed waveform for different α12

2.6

Block Diagram of the Static Frequency Divider showing the output buffers of the

Iand Qpaths ................................... 14

2.7 Positive feedback oscillatory system . . . . . . . . . . . . . . . . . . . . . . . 15

2.8 (a) Realization of the two-port model (b) Convensional cross-coupled LC VCO 16

2.9

(a) One-port model of the LC VCO (b) One Realization of the active negative

resistivecircuitry ................................. 17

2.10 (a) Realization of the one-port model (b) Convensional cross-coupled LC VCO 18

2.11 Conventional Gilbert-Cell mixer . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.12 Typical waveform of the Taylor series coefficient p1(t)of the switching pair . . 21

3.1

Metal Stack of the (a)

130−nm

Technology [24] and the (b)

90−nm

Technology

[25] ........................................ 24

3.2 On-chip Transistor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

xvi LIST OF FIGURES

3.3

Simulated

fmax

and

NFmin

as a function of finger width and biasing current density

at a total width of 50 µm ............................. 26

3.4 On-chipInductorModel ............................. 27

3.5 On-chip Stacked Inductor for high inductance value per Area . . . . . . . . . . 28

3.6 MOS capacitance ranges from 35 f F to 310 f F with 40 total number of fingers 29

3.7

On-chip MIM capacitor, capacitor-top-metal and capacitor-bottom-metal are

inserted to increase the capacitance per unit area . . . . . . . . . . . . . . . . . 31

3.8 Lumped model of the Transmission Line . . . . . . . . . . . . . . . . . . . . . 32

3.9 Coplanar waveguide with ground plan . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Block Diagram of the 24 GHz FMCW Radar Transceiver . . . . . . . . . . . . 35

4.2 Miller divider circuit with active output balun . . . . . . . . . . . . . . . . . . 37

4.3 Simulated open loop gain and third order harmonic attenuation . . . . . . . . . 38

4.4

Simulated open loop gain versus switching transistor size for different bias current

4.5 Chip micrograph of the realized Miller divider . . . . . . . . . . . . . . . . . . 40

4.6 Input sensitivity of the Miller divider . . . . . . . . . . . . . . . . . . . . . . . 41

4.7 Output power of the Miller divider . . . . . . . . . . . . . . . . . . . . . . . . 41

4.8 Spectrum analyzer’s screen shot of the output power of the Miller divider . . . 42

4.9 (a) Conventional NMOS VCO (b) Transformer Feedback VCO . . . . . . . . . 43

4.10

Simulated phase noise, at

25 GHz

with

1MHz

offset, versus cross-coupled

transistor width for different values of bias current . . . . . . . . . . . . . . . . 46

4.11 Measured tuning range of both conventional and TF VCOs . . . . . . . . . . . 48

4.12 Chip micrograph of the all NMOS VCO . . . . . . . . . . . . . . . . . . . . . 48

4.13 Chip micrograph of the Transformer feedback VCO . . . . . . . . . . . . . . . 48

4.14 Measured phase noise at 1 MHz offset of both all NMOS and TF VCO . . . . . 49

4.15

Layout of the Transformer, L1 and L2 drawn in the MA and E1 metal layer,

respectively .................................... 49

5.1 Block diagram of the IQ-Modulator . . . . . . . . . . . . . . . . . . . . . . . 52

5.2 Circuit diagram of a single latch . . . . . . . . . . . . . . . . . . . . . . . . . 53

LIST OF FIGURES xvii

5.3

Open loop gain of the evaluating amplifier as a function of the tail current

transistorsize ................................... 54

5.4

Self oscillation frequency of the SFD and evaluating amplifier DC current are

proportional to the tail current transistor size . . . . . . . . . . . . . . . . . . . 55

5.5

Self oscillation frequency and latching transistors DC current varies in accordance

to both the tail current device size and cross coupled transistors size . . . . . . 56

5.6

Step response of the SFDs output nodes for different splitting ratios of the resistive

load ........................................ 56

5.7 Circuit diagram of the output buffer . . . . . . . . . . . . . . . . . . . . . . . 57

5.8

Single quadrature down-conversion mixer with a SFD as an IQ local oscillator

signalgenerator.................................. 58

5.9 Simulated imbalance of IQ amplitude and phase . . . . . . . . . . . . . . . . . 59

5.10 Simulated image rejection ratio due to the IQ imbalance . . . . . . . . . . . . . 60

5.11 Chip micrograph of the SFD showing the pad configuration . . . . . . . . . . . 60

5.12 Simulated and measured sensitivity curve of the static frequency divider . . . . 61

5.13 Spectrum of the free running SFD . . . . . . . . . . . . . . . . . . . . . . . . 61

5.14 Circuit schematic of the up-conversion mixer . . . . . . . . . . . . . . . . . . 63

5.15

Simulation of conversion gain of the up-conversion mixer vs. the size of the

switchingtransistors ............................... 64

5.16 Chip micrograph of the up-conversion mixer . . . . . . . . . . . . . . . . . . . 65

5.17 Optimum LO power for maximum conversion gain of the up-conversion mixer 66

5.18 Conversion gain of the up-conversion mixer . . . . . . . . . . . . . . . . . . . 66

5.19 P−1dB of the up-conversion mixer . . . . . . . . . . . . . . . . . . . . . . . 67

5.20 Circuit diagram of the cascode IF amplifier . . . . . . . . . . . . . . . . . . . 68

5.21 Power gain of the IF amplifier . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.22 Input and output matching of the IF amplifier . . . . . . . . . . . . . . . . . . 69

5.23 Frequency divider and IQ generator for the LO signal . . . . . . . . . . . . . . 70

5.24 Circuit diagram of the up-conversion mixer with the LO buffer . . . . . . . . . 71

xviii LIST OF FIGURES

5.25

The chip micrograph of the IQ-Modulator showing circuit components and

probingpads.................................... 71

5.26 Conversion gain of the IQ-Modulator . . . . . . . . . . . . . . . . . . . . . . . 72

6.1 Block diagram of the IQ-Demodulator . . . . . . . . . . . . . . . . . . . . . . 76

6.2 Gilbert-Cell mixer with conventional current bleeding network . . . . . . . . . 77

6.3

Equivalent circuit of the Gilbert-Cell mixer with conventional current bleeding

mixer ....................................... 78

6.4 Gilbert-Cell mixer with static current bleeding network and one inductor . . . . 79

6.5 Equivalent circuit of Gilbert-Cell mixer with current bleeding and one inductor 79

6.6 Gilbert-Cell mixer with dynamic current bleeding circuit and one inductor . . . 81

6.7

Gilbert-Cell mixer with the proposed dynamic current bleeding mixer and two

inductors ..................................... 82

6.8

Equivalent curcuit of the Gilbert-Cell mixer with dynamic current bleeding

network and two series inductors . . . . . . . . . . . . . . . . . . . . . . . . . 83

6.9

Simulated single side band noise figure of the Gilbert-Cell mixer with one shunt

andtwoseriesinductors.............................. 83

6.10

Simulated conversion gain of the Gilbert-Cell mixer with one shunt and two

seriesinductors .................................. 84

6.11

Simulated conversion gain of the Gilbert-Cell mixer versus the switching transis-

torsize....................................... 85

6.12

Chip micrograph of the direct down-conversion mixer with the proposed DCBN

showing the pad configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 86

6.13

Measured conversion gain of the Gilbert-Cell mixer showing the optimum LO

inputpower .................................... 86

6.14

Measured and simulated conversion gain of the Gilbert-Cell down-conversion

mixer with the proposed DCBN . . . . . . . . . . . . . . . . . . . . . . . . . 87

6.15

The

1−dB

compression point of the Gilbert-Cell mixer with degeneration inductors

6.16 Frequency divider and IQ generator for the LO signal . . . . . . . . . . . . . . 90

LIST OF FIGURES xix

6.17 Down-conversion mixer with the LO buffer . . . . . . . . . . . . . . . . . . . 91

6.18

The chip micrograph of the IQ-Demodulator showing circuit components and

probingpads.................................... 91

6.19 Conversion gain of the IQ-Demodulator . . . . . . . . . . . . . . . . . . . . . 92

xx LIST OF FIGURES

List of Tables

1.1 mm-Wave PHY chanelization . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.1 Simulation of normal and stacked inductors in the 90 nm CMOS technology . . 29

3.2 Simulation examples of the MIM capacitors in the 90 nm CMOS technology . . 32

4.1 Circuit Parameters of the Miller Divider . . . . . . . . . . . . . . . . . . . . . 37

4.2 Circuit Parameters of the Miller Divider . . . . . . . . . . . . . . . . . . . . . 37

4.3 Miller divider’s performance comparison with the state of the art . . . . . . . . 42

4.4 Circuit Parameters of the Conventional VCO . . . . . . . . . . . . . . . . . . 43

4.5 Circuit Parameters of the Transformer Feedback VCO . . . . . . . . . . . . . . 44

4.6 Performance comparison of the two VCO versions with the state of the art . . . 50

5.1 Circuit Parameters of the Latch Circuit . . . . . . . . . . . . . . . . . . . . . . 53

5.2 Circuit Parameters of the Output Buffer Circuit . . . . . . . . . . . . . . . . . 57

5.3 Performance comparison of the SFD with the state of the art . . . . . . . . . . 62

5.4 Performance comparison of the SFD with the state of the art . . . . . . . . . . 62

5.5 Circuit parameters of the up-conversion mixer . . . . . . . . . . . . . . . . . . 63

5.6 Summary and performance comparison of the direct up-conversion mixer . . . 67

5.7 Summary and performance comparison of the direct up-conversion mixer . . . 67

5.8 Circuit parameters of the IF amplifier . . . . . . . . . . . . . . . . . . . . . . 69

5.9 Summary and performance comparison of the IQ-Modulator . . . . . . . . . . 73

5.10 Summary and performance comparison of the IQ-Modulator . . . . . . . . . . 73

6.1 Circuit Parameters of the direct down-conversion mixer . . . . . . . . . . . . . 82

xxi

xxii LIST OF TABLES

6.2

Summary and performance comparison of the direct down-conversion mixer with

theproposedDCBN................................ 88

6.3

Summary and performance comparison of the direct down-conversion mixer with

theproposedDCBN................................ 88

6.4

Summary and performance comparison of the direct down-conversion mixer with

theproposedDCBN................................ 89

6.5 Summary and performance comparison of the IQ-Demodulator . . . . . . . . . 92

List of Abbreviations

3D Three Dimension

AC Alternating current

ADS Advanced Design System

CG Conversion gain

CMOS Complementary metal-oxide semiconductor

dB Decibel

dBm Decibel referred to milliwatt

DC Direct current

EM Electromagnetic

fFrequency

ftTransient frequency

fmax Maximum oscilation frequency

FM Frequency modulation

gmTransconductance of the transistor

Gmax Maximum stable gain

Gbps Giga bit per second

GHz Gigahertz

HF High frequency

HFSS High Frequency Structural Simulation

IC Integrated circuit

IF Intermediate frequency

IP3 Third-order intercept point

xxiv LIST OF TABLES

K Stability factor

LNA Low noise amplifier

mA Milliamperes

MHz Megahertz

NF Noise figure

P Power

P1dB Input referred 1dB intercept point

PA Power amplifier

Pin Input power

Pout Output power

PLL Phase locked loop

QQuality factor

RF Radio frequency

Rx Receiver

SFD Static frequency divider

Si Silicon

Tx Transmitter

TRx Transceiver

VCO Voltage controlled oscillator

W Width of the MOS transistor

Chapter 1

Introduction

1.1 Motivation

As wireless products, such as wireless networking and localization applications, become a part of

people’s everyday lives, the need for high performance at lower costs and power consumption

become even more important. Overcoming the challenges involved in the design of radio-

frequency (RF) transceivers can help satisfy this need. This thesis provides an analysis, design,

and characterization of RF circuits for portable applications. The cost of the designs is an

important issue, therefore the circuits’ layouts were carefully designed in order for the design

to occupy small chip area. Battery life, on the other hand, is an important issue for portable

applications therefore, optimization procedures are followed to minimize the consumed power.

1.2 Scope of Research

This thesis focuses on the design and characterization various K-band transceiver circuits. The

RF circuits were designed using different CMOS technologies; namely, IBM

130 nm

, and TSMC

90 nm

. Two different front-ends are to be investigated in this thesis as well; the radar transceiver

system, which is designed using the IBM

130 nm

CMOS technology, and the short range high

data rate WLAN transceiver system which is designed based on TSMC

90 nm

CMOS technology.

Scaling down the size of the transistor shifts its unity gain frequency

and its maximum

2CHAPTER 1. INTRODUCTION

Figure 1.1: Transmit and Receive signals from a point target

oscillation frequency

fmax

. The

and

fmax

of the

90 nm

CMOS technology are

120 GHz

and

150 GHz

, respectively; therefore the

90 nm

CMOS devices are adequate to impelement the

60 GHz WLAN transceiver front-end.

1.2.1 FMCW Radar Transceiver

Continuous-wave (CW) radar is a type of radar system whereby a known stable frequency CW

radio energy is transmitted and then received from any reflecting objects. The main advantage of

the CW radar is that energy is not pulsed, they are much simpler to design. Continuous-wave

radars have no minimum or maximum range, although the broadcast power level imposes a

practical limit on range. CW radar maximizes its total power on a target because the transmitter

is broadcasting continuously. Maximum distance in a CW radar is determined by the overall

bandwidth and transmitter power. There are two types of CW radars: un-modulated and modulated

CW radar.

Frequency-modulated CW radar (FMCW) is a short-range measuring radar capable of de-

termining distance [1]. This increases reliability by providing distance measurement along with

speed measurement, which is essential when there is more than one source of reflection arriving

at the radar antenna. This kind of radar is often used as ”radar altimeter” to measure exact height

during the landing procedure of aircrafts. It is also used as early-warning radar, wave radar, and

proximity sensors. Doppler shift is not always required for detection when FM is used. Sawtooth

modulation is employed in CWFM radars where range is desired for objects that lack rotating

parts.

In a CWFM radar, a chirp signal is transmitted for a certain duration. While it is transmitting,

1.2. SCOPE OF RESEARCH 3

Figure 1.2: Block Diagram of the 24 GHz FMCW Radar Transceiver

the echoes are received by the receiver and mixed with the transmitted signal, and the result is

low-pass filtered to produce a superposition of beat frequencies. Fig. 1.1 shows a transmitted

and the reflected signals from a point target. The two-way travel time is

, the bandwidth of the

transmit signal is

, and the sweep time is

. At any instance of time, the transmitted and the

received signals are multiplied by a mixer. Since multiplying two sinusoidal signals together

results in a sum and difference terms, the output signal is then low pass filtered, and only the

difference term is left as shown in fig. 1.2 [2] [3]. The frequency of this signal is given by

, the

beat frequency.

An expression for the beat frequency can easily be found. Using similar triangles and

rearranging terms, we obtain the following expression:

fb=B·τ

T(1.1)

since the beat frequency signal is time-limited to

seconds, its spectrum will be a sinc function

centered at

f=fb

, the first zero crossing will occur at

f=1

. When multiple targets exist, the

result will be a superposition of many beat frequencies. In the frequency domain, two targets beat

frequencies can be as close together as 1

T, substituting in equation 1.1 we have:

∆τ =T·∆fb

B(1.2)

substituting in

∆fb=1

, we have

∆τ =1

. Hence, the minimum resolvable separation in time

4CHAPTER 1. INTRODUCTION

between two targets is inversely proportional to the bandwidth. The two-way time to a target,

and the range to the target, R, are related by the following formula:

R=c·τ

2(1.3)

and the range resolution is given by:

∆R=c·∆τ

2(1.4)

where cis the speed of light in air, substituting for ∆τ in equation 6.6 gives:

∆R=c

2·B(1.5)

Since the beat frequency is proportional to

, and

is proportional to the range

, knowledge of

the beat frequency of any target entails knowledge of the range of that target. With many targets,

we can separate them by taking the Fourier Transform of the received signal, and determine the

range through frequency.

In contrast to the FMCW radar transceiver design presented in [4], it is desirable to achieve

a single antenna TRx. In order to fulfill the single antenna architecture, TRx must be able to

switch between the Rx and Tx. In most common architectures the Rx and Tx are connected to

the antenna through a switch which either connects the input of the Rx or the output of the Tx to

the antenna. In that topology, the switch is located in the most sensitive part of the circuit as both

the output power of the PA and the noise figure of the LNA are being affected by the insertion

loss of the switch.

In order to avoid RF signal distortion, in this design the switch has been located in the LO

path as shown in fig. 1.2. In this case, in the Rx mode, no LO power is being delivered to the Tx

and in the Tx mode, no LO power is being delivered to the Rx mixer. Although the isolation of

the switches are not infinite, DC supply switching between the Rx and Tx modes, a very high

Tx-Rx isolation can be achieved. DC supply switching, beside isolation improvement, can also

improve the power efficiency of the TRx.

In this thesis, several RF building blocks for this frond-end transceiver were designed and

1.2. SCOPE OF RESEARCH 5

Figure 1.3: Channelization of the IEEE 802.15.3c standard

tested. As shown in fig. 1.2, the marked blocks are the scope of this work.

1.2.2 Short Range High Data Rate WLAN Transceiver

The mm-Wave PHY, is defined for the frequency band of

57 – 66 GHz

, as allocated by the

regulatory agencies in Europe, Japan, Canada, and the United States as well as any other areas

where the regulatory bodies have allocated this band, and is shown in fig. 1.3. Four channels are

defined for the PHY, nevertheless regulatory requirements allow fewer in some regions.

A total of three PHYs are defined for the mm-Wave PHY. They are as follows:

•Single Carrier mode in mm-Wave PHY (SC PHY).

•High Speed Interface mode in mmWave PHY (HSI PHY).

•Audio/Visual mode in mmWave PHY (AV PHY).

The SC PHY supports a variety of modulation and coding schemes (MCSs) that support bit

rate up to

5Gb/s

. The SC PHY supports a wide range of modulations,

π/2

BPSK,

π/2

QPSK,

π/2

8-PSK,

π/2

16-QAM, pre-coded MSK, pre-coded GMSK, on-off keying (OOK), and dual

alternate mark inversion (DAMI). The coding schemes included are Reed-Solomon (RS) coding

with low-complexity implementation and low-density parity check (LDPC) coding with high

error-correcting capability. Code spreading using either linear feedback shift register (LFSR)

code or the Golay sequence is also applied to increase the robustness of the system. The HSI

PHY, is designed for NLOS operation and uses OFDM with an FEC based on LDPC.

The AV PHY, is designed for NLOS operation and the transmission of uncompressed, high

definition video and audio. It uses OFDM modulation with convolutional inner code and a Reed

6CHAPTER 1. INTRODUCTION

Figure 1.4: Block Diagram of the WLAN Transceiver

Solomon outer code. The AV mode supports omni-directional coverage via the low-rate PHY

(LRP) for the purpose of setting up high-throughput connections using the high-rate PHY (HRP).

Different PHYs are a result of demands of different market segments, which were based on

the development of the usage models for this standard. For example, one usage model is for kiosk

applications. This usage model requires a bit rate of

1.5Gb/s

at a

range. The SC-PHY can

provide such a data rate at that short range with less complexity, and thus lower cost than an

OFDM PHY. Another usage model required the streaming of uncompressed video. Due to the

nature of uncompressed video signals a special PHY, the AV PHY, was selected to provide high

throughput. A third usage model uses an ad-hoc system to connect computers and devices around

a conference table. In this usage model, all of the devices in the WPAN will have bidirectional,

NLOS high speed, low-latency communication, which is provided by the HSI PHY. Mandatory

data rates of all those PHYs are selected according to specific usage models. In addition, higher

data rates are provided to give options to the implementors so that they can best address the

different market segments.

In this thesis, the IQ- modulator and demodulator of the transceiver architecture of fig. 1.2 are

designed. The

and

signals of the LO are generated using wide locking range static frequency

divider (SFD) instead of using hybrids which have narrow band for accurate

and

signals

generation. The

40 GHz

LO signal is provided by an integer-N PLL, the LO frequencies from

the PLL and the divided frequencies fed to the

mixers are listed in table 1.1. The IF signals

1.3. DIFFERENTIAL SIGNALING 7

Channel ID Start Frequency Center Frequency Stop Frequency LO LO/2

(GHz) (GHz) (GHz) (GHz) (GHz)

1 57.24 58.32 59.40 38.88 19.44

2 59.40 60.48 61.56 40.32 20.16

3 61.56 62.64 63.72 41.76 20.88

4 63.72 64.80 65.88 43.20 21.6

Table 1.1: mm-Wave PHY chanelization

are fed from the first stage mixer of the receiver path, and fed into an up-conversion mixer for

the transmitter path. Finally, the base-band

and

signals are available from or to DSPs in the

receive and transmit paths respectively.

1.3 Differential Signaling

As single-ended and differential signals are compared, it is important to keep the system-level

performance metrics in mind for good overall receiver design. A single-ended signal, unbalanced

by definition, is measured by the difference between the signal of interest and a constant reference

point. The reference point, which is normally ground, serves as the return path for the signal. A

problem can be encountered if an error source is introduced into the signal path. Because the

ground reference will be unaffected by the injected error, the error is carried forward through

the signal. Any signal variation introduced in a single-ended configuration will be difficult to

be removed without using overly complex cancellation techniques. Single-ended signals are,

therefore, more prone to noise and electromagnetic coupled interference.

Differential signals, on the other hand, are made up of pairs of balanced signals moving at

equal but opposite amplitudes around a reference point. The difference between the positive

and negative balanced signals corresponds to the composite differential signal. If an error is

introduced to a differential system path, it will be added to each of the two balanced signals

equally. Because the return path is not a constant reference point, the error will be canceled in

the differential signal. Consequently, differential signal chains are less susceptible to noise and

interference. This inherent error cancellation also provides better common mode rejection ration

8CHAPTER 1. INTRODUCTION

(CMRR) and power supply rejection ratio (PSSR).

In the WLAN Transceiver system, differential signaling is always used for both RF signal and

LO signals to overcome the problems of common mode signals and inaccurate power supplies.

1.4 Thesis Organization

The thesis is organized as follows; Chapter 2 presents the background and analysis of various RF

circuits such as Miller and static frequency dividers, voltage controlled oscillators, up and down

direct-conversion mixers. In chapter 3, an overview of the two CMOS technologies used are

represented, metal stacks are shown, device’s gain and noise are investigated, and comments on

the characterization of the passive components are discussed. In chapter 4, the design procedures

and measured data of the

24 GHz

circuits are presented, these RF circuits are part of the radar

transceiver front-end. Chapters 5 and 6 mainly present the design of both IQ-Modulators and

IQ-Demodulators respectively; each circuit, such as, up-conversion, down-conversion mixers,

static frequency dividers, IF amplifiers, were separately measured and characterized, then all of

the blocks were integrated together to be a part of the complete front-end transceiver. Chapter 7

concludes this work, and a proposal for future work is presented as well.

Chapter 2

RFICs – Essentials and Fundamentals

The theory and analyses of various RF circuits are investigated in this chapter. Frequency dividers

are essential parts of frequency synthesizers and clock recovery circuits. Two dividers are

presented; a Miller divider designed for the

24 GHz

Radar transceiver, and a static frequency

divider designed for the

60 GHz

WLAN transceiver system. The phase noise of the voltage

controlled oscillators is a challenge; sources of phase noise were studied, an optimization

procedure for better noise performance was undertaken. Up- and down-conversion mixers are

then investigated. When the signal is converted to the vicinity of DC, problems such as flicker

noise arise and become a bottleneck. Methods to alleviate these effects are studied.

2.1 Frequency Dividers

2.1.1 Miller Divider

The general key design criteria for frequency dividers are the requirements for low input power,

low power consumption, and wide frequency range of correct division. The maximum frequency

of operation of static dividers is closely related to and much lower than the device cut-off

frequency. Moreover, the power consumption of the static divider increases progressively with the

input frequency due to the charging and discharging of capacitances at each clock cycle. On the

other hand, regenerative frequency dividers can operate at high frequencies without a proportional

10 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

Figure 2.1: Block Diagram of the Miller Divider

increase of power consumption. The high frequency and low power operation of a regenerative

frequency divider come at the expense of narrow-band operation and larger chip area due to the

utilization of on-chip inductors. In spite of the high frequency and low power operation, the

injection locked frequency dividers have narrow frequency range. The dynamic dividers, however,

have wider bandwidth than the injection-locked counterpart. Dynamic frequency dividers (DFD)

are also known as Miller dividers.

Dynamic frequency dividers or Miller dividers were firstly discovered by Miller in

1939

[5],

hence the name Miller divider. They operate on the bases of positive feedback principles;

consequently, Barkhausen criterion must be met [6] [7]. The block diagram of the DFD is shown

in fig.2.1. For regeneration, a finite signal, e.g. due to thermal noise, must be present in the loop,

and the loop gain must be greater than unity; additionally, for zero output with no input the loop

gain must be less than unity when the input signal is removed [8] [9]. The oscillation grows

until the gain is reduced to unity and amplitude sustains due to nonlinearities. As shown in fig.

2.1, the input signal at

ωin

is mixed with a feedback signal at

ωin/2

to give sidebands at

ωin/2

and

3ωin/2

. The band pass filter selects the lower sideband. The amplifier compensates for loop

losses.

While providing an intuitive understanding of the circuit operation, we normally have two

methodologies to provide enough selection of the half input frequency signal; low pass filtering,

and band pass filtering. The LPF shown in fig. 2.2a fails to stipulate the condition for proper

division. For example, the low pass filter may be realized as a first order RC network of fig.

2.2b, a reasonable model of the load seen at the output node of typical mixers [10]. Neglecting

2.1. FREQUENCY DIVIDERS 11

(a) (b)

Figure 2.2: Miller Divider (a) Generic Topology (b) Realized with RC filter

nonlinearities in the mixer, the node equation at the output node is given by:

R1C1

dt +y=βyAcos(ωint)(2.1)

where

denotes the mixer conversion factor. The differential equation is first order with constant

coefficient and its solution is given by:

y(t) = y(0)e−t

R1C1+βA

R1C1ωin sin(ωint)(2.2)

Interestingly, it is clear that

y(t)

decays to zero with a time constant of

R1C1

, and the circuit

fails to divide regardless of the value of

ωin

with respect to the LPF corner frequency,

(R1C1)−1

In other words,

ωin/2

is not regenerated even though

(R1C1)

is chosen to attenuate the third

harmonic 3ωin/2, and even if a noise current at ωin/2 is injected into the loop.

Suppose now that we have the extreme case where all time constants in the loop are negligible,

all waveforms are rectangular, and the circuit operates correctly. As illustrated in fig. 2.3b, the

mixer output resembles

y(t)

but shifted by a quarter period, suggesting that inserting a broadband

delay in the loop permits correct division. However, the RC network of fig. 2.2b does not satisfy

the conditions required in fig. 2.3a. For example, the network cannot provide a phase shift of

90o

ωin/2

and

270o

3ωin/2

. Furthermore, it attenuates the third harmonic considerably, failing

to generate the idealized waveforms shown in fig. 2.3a.

We now study another extreme case where the loop exhibits no delay at

ωin/2

but enough

selectivity to attenuate the third harmonic [11]. Fig. 2.4 exemplifies this case, with the mixer

injecting a current into the parallel tank (i.e. BPF). Solving the node equation at the output node

12 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

(a) (b)

Figure 2.3:

Miller divider with wide band phase shifter (a) Waveform (b)

Generic Topology

Figure 2.4: Miller divider with BPF

will produce the following:

y(t) = Aeatcosωin

2t+θ(2.3)

where

is a constant, depending on the loop gain, and is positive only when the loop gain is

greater than unity [11]. The positive constant

implies that an exponentially growing sinusoid

at half of the input frequency is a possible solution for the system during the transient build-up

period. When the oscillations’ amplitude is large enough, the devices become nonlinear causing

gain compression until it becomes unity and the constant

is finally zero, thus sustainable

oscillations will last, whereas the input is applied.

The impact of the third order harmonic on the correct operation is very important. Suppose

(a) (b)

Figure 2.5:

(a) Divided and third order harmonic (b) Superimposed waveform

for different α

2.1. FREQUENCY DIVIDERS 13

that the peaks of

x1(t)

(the required divided output)and

x2(t)

(the third order harmonic) have

the same amplitude. As depicted in fig. 2.5a, the product waveform displays multiple zero

crossings in each period due to the third order harmonic, revealing that such a loop fails to

divide if this harmonic is not suppressed sufficiently, i.e., if

y(t)

does not monotonically rise and

fall [10]. Fig. 2.5b illustrates the resulting waveforms for different values of the attenuation

factor,

, experienced by the third order harmonic with respect to the fundamental. To eliminate

the extraneous zero crossings, we require that the slope of

y(t)

does not change sign between a

positive peak and the next negative peak. Since the output is the superposition of the two tones,

thus we have:

y(t) = Acosωin

2t+αsin3ωin

2t (2.4)

since the third order harmonic exhibits a phase shift of about

90o

, while the required divided

output exhibits no shift, then the attenuation αshould be less than 1

2√3.

In summary, proper operation of the Miller divider requires either sufficient broadband phase

shift around the loop or enough suppression of the third harmonic. The first alternative is difficult

to realize in CMOS technology for the following reasons:

•

with the low transconductance of MOS devices, the voltage drop across the load resistors

must be large so as to provide enough loop gain

•source followers consume substantial voltage headroom and attenuate the signal

•

the limited bandwidth of the source followers prevents the divider from achieving high-

speed operation

Fortunately, these issues can be resolved by employing an LC tank as the load in the Miller

divider, provided that the small-signal loop gain at half the input frequency is greater than unity.

2.1.2 Static Frequency Divider

Frequency dividers are extensively used for various applications including frequency synthesis

and IQ generators. Static frequency dividers (SFD) are commonly used at millimeter-wave (mm-

wave) because of their wide operation range; however, they suffer from high power dissipation,

14 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

Figure 2.6:

Block Diagram of the Static Frequency Divider showing the output

buffers of the Iand Qpaths

which increases proportional to the operating frequency. Other frequency divider types such as

Miller, as well as the injection locking dividers, suffer from narrow operating frequency range.

Complex techniques should be used to improve their locking range.

In this design, the (SFD) is used not only as a frequency divider, but as an IQ signal generator

as well, as shown in fig. 2.6. Polyphase networks are commonly used as qudrature generators,

although imbalance could be minimized over the required frequency range, the polyphoase

networks are lossy and noisy circuits. Extra amplifiers should be used to supply enough LO

power to the mixer, which require more chip area and more power consumption. The (SFD) can

work as a frequency divider while quadrature signals are available at the outputs.

The main target of the (SFD) designers is to achieve the highest possible operation frequency

out of the used technology, disregarding the amount of power that the divider needs. For instance,

different (SFD) were designed in [12] and [13] with moderate power consumption by using

techniques to boost the operation frequency. On the other hand, faster response of the flip-flops is

achieved in [14], [15], and [16] with higher power dissipation. A different design technique is

used in [17] to further boost the operating frequency by scarifying input sensitivity and power

dissipation. In this thesis, techniques will be used to minimize the power consumption where

possible. Inductive peaking provides wider bandwidth and can boost the operation frequency.

Splitting resistors can speed up the charging and discharging times, thus shortening setup times

of the latches. Power consumption can also be minimized through asymmetric transistor sizing,

as will be shown below.

2.2. VOLTAGE CONTROLLED OSCILLATORS 15

2.2 Voltage Controlled Oscillators

Voltage controlled oscillators (VCOs) are critical building blocks in various communication

applications such as frequency synthesizers, timing-recovery circuits, and clock generation. One

of the most fundamental VCO realizations is LC VCOs. Despite several topologies of the LC

VCOs, conventional LC VCOs with cross coupled transistors are most common, owing to their

ease of design and simplicity in achieving the conditions of oscillation. Although these conditions

are becoming hard to satisfy with frequency increases, techniques are used to boost the

of the

cross coupled devices.

2.2.1 Principles and Design Issues

Most RF oscillators can be viewed as feedback circuits. Consider the simple linear feedback

system depicted in fig. 2.7, with the overall transfere function as follows:

Y(s)

X(s)=H(s)

1−H(s).(2.5)

Figure 2.7: Positive feedback oscillatory system

A self-sustaining mechanism arises at the frequency

H(so) = +1

and the oscillation

amplitude remains constant if

is purely imaginary, i.e.,

H(so=jωo) = +1

. Thus, for steady

oscillation, two conditions must be simultaneously met at wo:

•The loop gain, |H(jωo)|, must be equal to unity.

•The total phase shift around the loop, H(jωo), must equal to zero or 360o

16 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

(a) (b)

Figure 2.8:

(a) Realization of the two-port model (b) Convensional cross-

coupled LC VCO

The above conditions are called Barkhausen’s criteria [18] [19]. One realization of the above-

mentioned concept is shown in fig. 2.8a, a positive feedback system consists of a non-inverting

amplifier where input and output are shorted to each other. This configuration is the standard

conventional cross-coupled LC VCO as redrawn in fig. 2.8b.

The above view of oscillators is called the ”two-port” model [18]. By contrast, the ”one-port”

model treats the oscillator as two one-port networks connected to each other as shown in fig. 2.9a.

Suppose that the tank is a parallel LC tank whose resonance frequency is

ωo

and its equivalent

resistance is

at resonance. The tank by itself does not oscillate indefinitely because some of

the stored energy is dissipated in

in every cycle. The idea in the one-port model is that an

active network generates an impedance equal to

−Rp

so that the equivalent parallel resistance

seen by the intrinsic, lossless resonator is infinite. In essence, the energy lost in

is replenished

by the active circuit in every cycle, allowing steady oscillation.

One realization of the active negative resistance circuitry is shown in fig. 2.9b, the tank is

connected as shown in fig. 2.10a, and the differential version of it is again the conventional LC

VCO as shown in fig. 2.10b.

Phase noise, tuning range, and power consumption are the important issues while designing

any VCO. However, special attention should be given to the phase noise in order to have the

sharp frequency spectrum cause less interference to the nearby channels when the signal is

down-converted.

2.2. VOLTAGE CONTROLLED OSCILLATORS 17

(a) (b)

Figure 2.9:

(a) One-port model of the LC VCO (b) One Realization of the

active negative resistive circuitry

Several noise sources and complicated mechanisms contribute to the noise performance of the

VCO. Varactor capacitance variation due to the bias noise causes phase noise through an AM-FM

modulation mechanism [20]. Although techniques are used to cancel the bias noise, these are

hardly effective at high frequencies. This effect is, however, minor and can be mitigated by either

using filtering, which necessitates usage of an extra inductor, or eliminating the current source

which deteriorates VCO sensitivity to power supply and process variation.

Switching transistors, on the other hand, have the greatest percentage of contribution to the

phase noise. Their thermal and flicker noise appears directly at the output node of the VCO.

Flicker noise can, nevertheless, be minimized by maximizing the size of the switching transistors.

Despite the fact that the current source transistor is not directly connected to the oscillating

nodes of the VCO, its noise is converted to the neighborhood of the oscillation frequency by

the switching transistors. Both high-frequency and flicker noise effects are reduced using larger

devices and decoupling capacitors. Low-frequency noise, however, is converted directly to the

the VCO’s output nodes via the AM-PM mechanism.

Nonlinearity of the switching devices causes generation of

2nd

and

3rd

harmonics flowing

into the low impedance capacitance of the tank, causing in turn a phase imbalance. The oscillator

then adjusts its frequency in order to return to its stability condition. Amplitude modulation of

the bias of the level of the harmonics, results in variation of phase imbalance level and frequency

18 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

(a) (b)

Figure 2.10:

(a) Realization of the one-port model (b) Convensional cross-

coupled LC VCO

shift. This is called an AM-PM modulation mechanism.

The AM-PM effect can be minimized by increasing the linearity of the switching devices.

Linearity increases by narrowing the devices, so that the overdrive voltage is increased. Smaller

devices, however, suffer from increased levels of flicker noise. Thus, a trade off between flicker

noise and AM-PM mechanism exists, and the proper size should be selected for the minimum

possible phase noise.

2.3 Direct Converison Mixer

The mixer is an important building block of a typical front-end circuit, since the frequency

translation process will generate some unwanted spurious signals to degrade the signal-to-noise

ratio (SNR). The noise becomes an important factor of the mixer. In most transceivers, the noise

of the mixer affects the noise of RF front-end system. Therefore, a low noise mixer is very

important in the design of front-end transceivers.

For direct conversion receivers, the devices suffer from a very high intrinsic flicker noise.

Linear RF circuits are not affected by the flicker noise, such as low noise amplifiers (LNA), since

its operating frequency is much higher than the flicker noise corner frequency, but in the mixers,

the case is different, because the flicker noise lies in the output band of the mixer, especially, in

2.3. DIRECT CONVERISON MIXER 19

low-IF and Zero-IF mixers. For heterodyne receivers the problem is more relaxed, as the final

downconversion stage is preceded by a long chain of amplification, thus its noise contribution

is lower and the dynamic range is not extremely affected. Nevertheless, the reduction of their

flicker noise is required as the output frequency is in the DC vicinity.

Passive mixers have several advantages. They have zero DC power consumption, high

linearity and low noise figure. But there are some disadvantages that limit their use in high

performance and high dynamic range front ends. The large LO powers are required for driving

passive mixers which impose large power consumption for the VCO unit. The isolation between

different ports of passive mixers is small and their occupied area on the silicon is large. Beside

the above disadvantages the passive mixers impose large losses on the signal. According to the

basic Firis formula the noise of the succeeding stages would become as large as the loss of the

mixer. This is why the sensitivity of the receivers with a passive mixer is not adequate while the

noise figure of the mixer itself is very small. In this thesis, a Gilbert-Cell mixer will be adopted.

Techniques will be used to reduce the flicker noise. Power consumption is also an important

design target as well as the noise performance.

2.3.1 Gain of the Gilbert-Cell mixer

The Gilbert cell consists of a transconductance or driver stage, which is a differential pair biased

at a fixed operating point, and two switching pairs driven by the strong LO signal [21] [22].

Resistive or tuned tank loads can be connected at the output, and degeneration can be used to

linearize the transconductance stage. For a double balanced mixer, shown in fig. 2.11, since a

large AC drive is applied to the switching pair, the bias of the switch transistors is not fixed but

varies periodically with time. When a differential voltage greater than a certain value

is applied

between the gates of the two transistors, one of them switches off. When the absolute value of the

instantaneous LO voltage

VLO

is lower than

, the current of the driver stage is shared between

the two devices. In this case, it is desirable to find the drain current of each transistor for a given

LO voltage and driver-stage bias current. We will assume that the output conductance of the

devices can be neglected, and therefore

can be modeled with an ideal current source

. The

20 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

Figure 2.11: Conventional Gilbert-Cell mixer

output current of the single-balanced mixer is a function of the instantaneous LO voltage

VLO

and the current at the output of the driver stage

IB+is

, with

being the bias current and

the

small-signal current in one single output is as follows:

Io1+io1=I1−I2=F(VLO(t),IB+is)(2.6)

Since isis small, a first order Taylor’s expansion gives:

io1=dF

dIbIB·is(2.7)

io1=p1(t)·is(2.8)

p1(t)periodic waveform. In a doubly balanced structure, the total output current is as follows:

io2=p1(t)·(−is)(2.9)

io=io1−io2=2p1(t)·is(2.10)

During the time interval

∆

when the LO voltage is between

and

−Vx

and both transistors are

on,

p1(t)

depends on ,

VLO(t)

and the IV characteristics of the transistors.

p1(t)

can be written

2.3. DIRECT CONVERISON MIXER 21

LO Voltage

P1(t)

+Vx

-Vx

Figure 2.12:

Typical waveform of the Taylor series coefficient

p1(t)

of the

switching pair

as follows:

p1(t) = gm1(t)−gm2(t)

gm1(t)+gm2(t)(2.11)

Where

gm1(t)

and

gm2(t)

represent the instantaneous small-signal transconductances of the

switching transistors. A signal

is multiplied by the waveform

p1(t)

, and therefore the frequency

spectrum of the corresponding output is:

Io1(ω) =

n=∞

∑

n=−∞

p1,n·Is(ω+nωLO)(2.12)

It is worth noticing that with good device matching,

p1(t)

is an odd symmetry (i.e.

p1(t) =

−p1(t+TLO

), with

TLO

being the LO period, and hence

p1(t)

has only odd-order frequency

components (See fig. 2.12). Usually the term for

n=1

n=−1

is of interest, corresponding to

shifting up or down the input signal in the frequency domain by one multiple of the LO frequency,

and in this case

C=|p1,1|=|p1,−1|

represents the conversion gain of the switching pair alone.

Since

is(t) = gm·vin(t)

where

vin(t)

is the input voltage signal at the gate of the transconductance

transistors and

is the transconductance of it, the conversion gain of the single-balanced mixer

in transconductance form is:

gc=C·gm(2.13)

22 CHAPTER 2. RFICS – ESSENTIALS AND FUNDAMENTALS

For high LO amplitude,

p1(t)

approaches a square waveform and

approaches

. Assuming

VLO =Vx

as it should be for proper mixer operation, an estimate for

can be obtained by

approximating p1(t)with a straight line during ∆[22].

C=2

πsin(π∆ fLO)

π∆ fLO (2.14)

And Vxis given as follows [23]:

Vx=αIB

2K+sαIB

2K2

+IB

K(2.15)

Thus, the mixer gain can be improved by decreasing the bias current

, or enlarging the switching

transistor size. Limiting bias current can be achieved by using bleeding networks in order to keep

of the driver transistors, therefore, conversion gain is unaffected. Increasing the size of the

switches helps to decrease

and improves its flicker noise contribution; nevertheless, further

increasing the size of the transistor adds more parasitic capacitance at the common source nodes.

Some of the AC current from the driver transistors is shunted in the parasitic capacitances at the

common source node causing degradation for the conversion gain. Therefore, size of the switches

should be optimized in order to attain the maximum conversion gain.

Chapter 3

CMOS Devices for RFIC – Technology

Overview

3.1 Introduction

In addition to active transistors, RF circuits are composed of various passive components used

for impedance matching, resonant circuits, filters, and bias circuitry. The passive components

include inductors, capacitors, transmission lines, and resistors. Varactors with appropriate tuning

capability are also employed in the VCO. All of these devices can be implemented in standard

CMOS technologies using interconnection metals, MOS (Metal-Oxide-Silicon) transistors, and

polysilicon layers. However, geometric and process limitations such as the thickness of the metal,

the physical distance from the substrate, the maximum metal width and minimum spacing, and the

doping concentrations limit the value of the achievable impedance, therefore, the performance of

the on-chip passive devices are limited. In this chapter, the properties of the RF components that

can be implemented in CMOS technology are discussed, their impact on the circuit performance

is presented.

In the CMOS process, aluminum

(Al)

and copper

(Cu)

are usually used for the fabrication of

the metal traces. The distance between the top metal and the substrate is

16.89 µm

for the

130 nm

CMOS, and

5.265 µm

for the

90 nm

CMOS technology. Fig. 3.1(b) and fig. 3.1(a) show the metal

stack of the two processes [24] and [25]. The relative permitivity of the silicon substrate is 11.9

24 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

(a) (b)

3.4 um

Si Substrate

5.26 um

4 um

Si Substrate

16.89 um

Figure 3.1:

Metal Stack of the (a)

130 −nm

Technology [24] and the (b)

90 −nm Technology [25]

while its conductivity is

10 S/m

. The low substrate resistivity and the short distance between the

top metal and the substrate significantly limit the RF performance that can be achieved, because

of high ohmic metals and lossy dielectrics.

3.2 Active Devices in CMOS Technology

At high frequencies, the effect of the series resistive parasitics have become more significant.

Therefore, it is important to include these parasitics in the transistor model in addition to the

capacitive parasitics that are already included in the digital CMOS models. Fig. 3.2 shows the

MOSFET model.

3.2.1 Gain

The key figure-of-merits characterizing the RF performance of a transistor are the unity gain

frequency

(ft)

and the maximum oscillation frequency

(fmax)

. The unity gain frequency,

(ft)

is defined as the frequency where the short-circuited current gain of the transistor is one. This

parameter is often used to measure the speed of the device. For MOS transistors, the unity gain

3.2. ACTIVE DEVICES IN CMOS TECHNOLOGY 25

Figure 3.2: On-chip Transistor Model

frequency can be expressed as:

ft=gm

2πqC2

g−(gmriCgs −Cgd)2≈gm

2πCg

(3.1)

assuming that

|Cg|>>|gmriCgs −Cgd|

. Here

Cg=Cgs +Cgd

. An important observation is that the

current gain is independent of the gate resistance of the device. It is shown that

is dependent

on the drain-source current IDS.

The maximum oscillation frequency,

fmax

, is also important since it indicates the maximum

frequency at which useful power gain can be expected from a device.

fmax

is defined as the

frequency where the Mason’s unilateral gain becomes one. For MOS transistors, the maximum

oscillation frequency can be expressed as:

fmax =sft

8πRg,iCgd

(3.2)

It can be predicted that fmax varies with 1

Wand proportional to √gm[26].

However, the value of

fmax

is determined not only by sizing and bias conditions, but is also

highly dependent on resistive losses due to transistor and layout parasitic [27]. By using narrow

finger widths, the effect of the gate resistance can be made negligible compared to the other

parasitic resistors. Therefore, with optimal layout, fmax is not limited by the gate resistance, but

is primarily determined by the series source/ drain resistances and substrate losses. Fig. 3.3

26 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

shows

fmax

of the 130 nm CMOS transistor. It can be higher than

120 GHz

with current density

exceeding

40 µA/µm

and finger width in the range

1µm

4µm

[28]. Also the device noise

performance achieves its optimum in the same range of the finger width.

Figure 3.3:

Simulated

fmax

and

NFmin

as a function of finger width and biasing

current density at a total width of 50 µm

3.2.2 Noise

The high frequency noise of a CMOS transistor is characterized by the minimum noise figure

(NFmin)

. For CMOS transistors, the minimum noise figure can be expressed by the following

empirical equation [29]:

NFmin =1+Kf

ftqγgm(Rg+Rs)(3.3)

where it has been shown that

(NFmin)

decreases with the device scaling to smaller dimensions,

and equation 3.3 indicates that

(NFmin)

decreases as a result of the increased

(ft)

. Equation 3.3

also indicates that

is an important parameter affecting the

(NFmin)

and that transistors should

be designed to minimize the value of

in order to demonstrate lower

(NFmin)

. As shown in fig.

3.3, this can be achieved by decreasing the finger width of the device.

3.3. PASSIVE DEVICES IN CMOS TECHNOLOGY 27

Figure 3.4: On-chip Inductor Model

3.3 Passive Devices in CMOS Technology

3.3.1 Inductor

Fig. 3.4 shows a commonly used equivalent circuit of an on-chip spiral inductor [30]. In this

equivalent circuit,

represents the inductance of the spiral,

is the metal series resistance, and

is the series feed forward capacitance that accounts for the capacitance due to the overlaps

between the spiral and the underpass.

Cox

represents the capacitance due to the oxide layers

between the inductor metal and substrate, while

Rsi

and

Csi

are the resistance and capacitance

due to the presence of the substrate. Because the inductor is fabricated over a low resistivity

silicon substrate, the actual inductance and quality factor in a circuit vary considerably with

the frequency of operation. At low frequencies, the inductor behaves as an inductance with

series resistance (metal resistance due to limited conductivity), while at higher frequencies, the

inductor is shunted not only by a parasitic capacitance between the inductor turns, but also

by the capacitance between the metal layers and the substrate itself. The interplay between

the inductance, the parasitic capacitance, and resistance causes the quality factor to rise at low

frequencies, reach a peak and then fall off at high frequencies.

In CMOS technology, inductors are usually fabricated as a spiral structure by using the top

metal of the process. This is to minimize the parasitic capacitances and as a result increase the

self-resonant frequency by maximizing the distance from the inductor metal to the substrate.

Several uppermost levels of metals can be used together, as they can be connected in parallel

through vias to achieve a very low effective sheet resistance. Alternatively, the metal traces in

28 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

Figure 3.5: On-chip Stacked Inductor for high inductance value per Area

different layers can be connected in series to achieve a high inductance per unit area. An Inductor

which is formed by series connection of layers is called stacked inductor and is shown in fig. 3.5.

Higher inductance values can be achieved in a smaller area. It suffers, however, from reduced

self-resonance frequency and lower quality factors.

The inductance and quality factors of an inductor can be extracted from the admittance

(Y)

parameters which are calculated from the measured scattering

(S)

parameters by using the

equations below:

Lind =−imag(1/Y11)

ω(3.4)

Qind =−imag(Y11)

real(Y11)(3.5)

According to the equivalent circuit in fig. 3.4, Y11 can be derived by the following equation:

Y11 =1

RS+jωLS

+jωCS+jωCOX ||1

RSi

+jωCSi(3.6)

The substrate loss can be reduced by having an opening in the P-well underneath the spiral to

increase the resistivity of the substrate or by placing the patterned ground using the lowermost

metal layer. The purpose of this shield is to terminate the parasitic electric field entering the

substrate. As a result, the inductor parasitic substrate loss becomes smaller. Table 3.1 presents

examples of the characteristics of spiral inductors in 90 nm CMOS technology. The achievable

values of inductance, quality factors, and operating frequencies are limited and governed by

trade-offs. The distance between the top metal and the substrate, and the low resistivity of the

3.3. PASSIVE DEVICES IN CMOS TECHNOLOGY 29

Figure 3.6:

MOS capacitance ranges from

35 f F

310 f F

with

total

number of fingers

substrate limit the performance of any inductors.

L(pH)Type Peak (Q)fQpeak (GHz)Area (µm ×µm) Layers

285 Normal 36 32 42 ×42 M9

Stacked 10 39 24 ×24 M7 to M9

510 Normal 19.5 23 50 ×50 M9

Stacked 9 25.5 25 ×25 M6 to M9

835 Normal 15 15 62 ×62 M9

Stacked 8 17.5 26 ×26 M6 to M9

Table 3.1:

Simulation of normal and stacked inductors in the 90 nm CMOS

technology

3.3.2 CMOS Varactor

Varactors in CMOS technology can be made of transistors by connecting the source and the

drain to each others. The transistors operate in both the depletion and the accumulation regions,

hence the name accumulation-mode varactor. In this way, the capacitance

Cmos

is monotonically

changed. The tuning characteristic of the accumulation mode varactors of the

90 nm

technologies

with a total number of fingers of 40 is shown in fig. 3.6.

The capacitance and quality factor of a varactor can be found using the

parameters extracted

30 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

from the measured scattering parameters and are given by the expressions below:

Cvar =imag(Y11)

ω(3.7)

Qvar =imag(Y11)

real(Y11)(3.8)

In this example, the characteristics of the varactors in the TSMC

90 nm

CMOS process at

20 GHz

is shown. Capacitance range is from

66 f F

300 f F

with the worst case

factor of

. The achievable values of capacitance, quality factors, and operating frequencies are limited

and governed by trade-offs. While the quality factor of an inductor increases with frequency,

the quality factor of a varactor decreases with frequency. Thus, when used in the LC tank of an

oscillator, the size of a varactor should be chosen so that its quality factor does not degrade the

overall quality factor of the LC tank.

3.3.3 Metal Capacitor

In CMOS technology, capacitors can also be fabricated with two closely placed metals. A

metal-insulator-metal (MIM) capacitor is made with two parallel metal plates and an insulating

layer between them. Since the bottom plate and adjacent substrate form a parasitic capacitor,

this capacitor is non-symmetrical. To minimize the effect of this parasitic capacitance, the two

uppermost metal layers are usually used to form an MIM capacitor. The distance between these

two metals (

0.715 µm

for TSMC

90 nm

processes) is too large to achieve a desirable capacitance

per unit area. In reality, extra metal layers are inserted between the two metal layers and connected

with the upper layer through vias to decrease the gap between these two metal plates of the

capacitor and increase the capacitance value per unit area. As shown in fig. 3.7 [25], capacitor top

metal (CTM) and capacitor bottom metal (CBM) are inserted between M7 and M8 and connected

to M8 through vias. MIM capacitors of the

90 nm

CMOS are shielded by a floating metal dummy

at M6 and M5.

For decoupling and matching, a constant capacitance with respect to the operating frequency

is required. The capacitance of metal capacitors is independent of the applied input voltage, but

3.3. PASSIVE DEVICES IN CMOS TECHNOLOGY 31

Si Substrate

CTM

CBM

Figure 3.7:

On-chip MIM capacitor, capacitor-top-metal and capacitor-bottom-

metal are inserted to increase the capacitance per unit area

their capacitance per unit area is less than the MOSFET-based capacitors, and thus less area

efficient. Although MOSFET-based capacitors can have higher capacitance density over the MIM

capacitor, they suffer from capacitance variations with applied voltage.

The capacitance and quality factor of a MIM capacitor can be measured with the same method

as a varactor and are given bey the following equations:

Cvar =imag(Y11)

ω(3.9)

Qvar =imag(Y11)

real(Y11)(3.10)

Table 3.2 shows examples of the characteristics of the metal capacitors in the

90 nm

CMOS

processes. Larger area capacitors have longer metals, this means higher parasitic inductors and

thus lower resonance frequency, and higher metal resistance thus lower

. On the other hand,

shrinking the area of the capacitors limits the permissible number of vias that connects both CTM

and CBM to M8, this leads to higher series resistance and lower

. The optimum capacitor value

according to Table 3.2 is 215 f F which corresponds to an area of (10 ×10) µm2.

32 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

C(f F)Qat 20 GHz fr(GHz)Area (µm ×µm)

35 18 348 4 ×4

55 22 373 5 ×5

108 26 188 10 ×5

215 36 121 10 ×10

500 21.6 74 15 ×15

975 12.6 50 20 ×20

1800 7.2 37 25 ×25

Table 3.2:

Simulation examples of the MIM capacitors in the

90 nm

CMOS

technology

Figure 3.8: Lumped model of the Transmission Line

3.3.4 Transmission Line

A section of a transmission line can be modeled by a lumped circuit model shown in fig. 3.8,

where

, and

are the series resistance, series inductance, shunt conductance, and shunt

capacitance per unit length, respectively. When it is assumed that the loss is small, the line can be

characterized by following parameters:

Zo=sR+jωL

G+jωC≈rL

C(3.11)

λ≈2π

ωo√LC (3.12)

QL=ωoL/R(3.13)

QC=ωoC/G(3.14)

where

is the characteristic impedance of the transmission line,

is the wavelength,

is the

inductive quality factor, and QCis the capacitive quality factor.

3.4. CONCLUSIONS 33

(a) (b)

Figure 3.9: Coplanar waveguide with ground plan

On-chip transmission lines can be made using coplanar waveguide (CPW) with ground plane

underneath as shown in fig. 3.9a and fig. 3.9b. The lines are implemented in the uppermost

layer to maximize the distance from the low-resistive substrate which has the potential to lower

the capacitive quality factor

(QC)

. The ground plane helps to prevent the electric field from

penetrating into the substrate, thus, the shunt loss is mainly due to the loss tangent of the oxide,

resulting in a high capacitive quality factor value. The width of the signal line and the gap

between the signal line and ground specify the values of

, and

, as shown in the above

equations. The existence of the ground shield reduces the distance from the line to the ground

plane, leading to narrower

50 Ω

lines. As a result, the ohmic loss increase and the inductive

quality factor degrades.

3.4 Conclusions

The gain and noise performance of the transistor is significantly affected by the parasitic re-

sistances. Therefore, the layout and size of the transistors should be chosen such that these

parasitic resistances are minimized, multi-finger structures with a small width for each finger is

an optimum choice.

Passive devices such as inductors, capacitors, varactors, and transmission lines are normally

provided by the CMOS technology founder. At high frequency, the performance of these devices

is degraded because of the limited metal thickness and the short distance between the metal layer

34 CHAPTER 3. CMOS DEVICES FOR RFIC – TECHNOLOGY OVERVIEW

and the lossy substrate. The trade-off should be considered to achieve the required performance at

the required frequency. Varactors are mainly used in the LC tanks of VCOs for frequency tuning.

Unlike inductors, the quality factor of a varactor decreases with frequency. Multi-finger varactors

can be used to improve their quality factors. Metal-insulator-metal (MIM) capacitors are usually

used for decoupling and matching. Therefore, they should be designed so that their self-resonance

frequency is higher than the operating frequency of the circuit. This can be achieved by avoiding

long metal lines in the capacitors and thus large inductance.

Integrated transmission lines can be implemented with CPW with a ground patterned shield.

Shielding prevents the waves to penetrate into the lossy substrate, thus limiting the loss to only the

oxide loss tangent. Introducing ground shield increases the

of the lines. Narrowing the lines

adjust their Zoto 50 Ωat the expense of increased metal resistive losses and degraded inductive

quality factor QL.

Chapter 4

Design of RF Circuits for the 24 GHz

FMCW Radar Transceiver

4.1 System Overview

One of the most common Radar topologies is the FMCW radar [2] [3]. The designed transceiver

shown in fig. 4.1 is used as a short range automotive radar. Its principle of operation along with

its resolution calculation is discussed in the introductory chapter of this thesis. In this work, the

Miller divider, and the voltage controlled oscillator are designed and tested to be integrated into

the complete front-end transceiver.

Figure 4.1: Block Diagram of the 24 GHz FMCW Radar Transceiver

36CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

4.2 Miller Frequency Divider

4.2.1 Design Approach

The designed Miller divider is designed based on a double balanced Gilbert-Cell mixer with

inductive load as shown in fig. 4.2. The inductors

are designed to resonate with the output

nodes’ parasitic capacitances at exactly half the input frequency. These capacitances are: Gate

capacitance of the divider transistors, drain capacitances of the switching transistors, and gate

capacitances of the following stage.

The feedback capacitance

Cf b

is used only to sample the output voltage signal and feed it

back to the gates of the transconductance transistors. Their values therefore must be large enough

to behave as RF short-circuit at the desired divided frequency, and thus there is only a negligible

effect of the total capacitance at the output nodes. An output on-chip active balun [26] has been

utilized to isolate the output node of the divider from the

50 Ω

load, and to convert the differential

to single ended output.

The required

of the load inductors determines the required attenuation of the third harmonic

as well as the loop gain over the operation range. Assuming for simplicity that the switches

are ideal, the equivalent resistance at the output nodes is

, the transconductance of the driver

transistors is

, and the transfer function of the resonance

circuit is

Hf(s)

. Thus the open

loop gain which must be greater unity is given as follows:

gain =2

πgmRHf(s)(4.1)

where

Hf(s) = 2ζωns

s2+2ζωns+ω2

(4.2)

given that

ωn=1

√LC

is half the input frequency, and

2ζωn=1

. A value of the lowest

corre-

sponds to the marginal attenuation for the third order harmonic of

2√3

. This value corresponds to

the widest frequency range for correct division. Thus, the minimum selectivity of the filter is as

4.2. MILLER FREQUENCY DIVIDER 37

follows:

L2ω2

L2ω2+R2(1−LCω2)2=1

2√32

(4.3)

the

factor of the inductor is expressed as

Lωn

and equal to

1.24

from the calculation of the

previous equation and is shown in [10].

Figure 4.2: Miller divider circuit with active output balun

Ws(µm)W1,2(µm)W3,4,5,6(µm)Wcp,cn (µm)Cf b (f F)

30 60 30 30 600

Table 4.1: Circuit Parameters of the Miller Divider

Lp(pH)R(KΩ)Vb1(mV)Vb2(mV )Vb3(V)

570 10 600 900 1.2

Table 4.2: Circuit Parameters of the Miller Divider

However, much higher value is required to guarantee enough loop gain, which degrades for

the following reasons:

•

The parasitic capacitance between the transconductances’ drains creates a pole that wastes

about half of the small signal drain currents.

•

The gradual switching of the switching pairs with the nearly sinusoidal drive converts part

of the differential currents into common mode current.

38CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

•

The load inductors’ capacitances and the coupling capacitors make the resonant frequency

lower than half the input frequency.

Fig. 4.3 shows the simulation results of the magnitude of the open-loop gain and the third

order harmonic attenuation. The frequency range for correct division can be extracted from this

figure using the following conditions:

•Magnitude of the voltage gain must be at least unity

•Enough attenuation of the third harmonic

As seen in fig. 4.3, for lower frequencies the third-order attenuation factor

being out of range

specifies the lower frequency limit. In our case the rejection is not enough at nearly

38 GHz

and

beyond. On the other hand, the upper frequency range limitation is specified by the decreasing

gain which is going less than unity at about

46 GHz

. Note that the voltage gain and the attenuation

factor of the third order harmonic of fig. 4.3 are open loop simulated results, meaning that the

center frequency is shifted upward.

The load inductors

Lp=0.57 nH

with

Q−

factor of

8.5

resonate with the parasitic capacitances

at the output nodes, thus providing

300 Ω

equivalent resistance at

14 GHz

with negligible effects

on voltage headroom.

When maximum gain of the mixer at high frequency is required, the driver transistors must be

fast enough to provide the adequate gain. Referring to fig. 3.3 in the previous chapter, transistors

Figure 4.3: Simulated open loop gain and third order harmonic attenuation

4.2. MILLER FREQUENCY DIVIDER 39

should be carefully layed out. Enough current density in a narrow fingered transistors improves

their speed.

The total gain of the Gilbert-Cell is also dependent on the current gain of the switching pair.

Their function is to convey the AC current coming from the transconductance stage into the load.

It is clear that the swing of the LO signal must be large enough to be able to fully commutate the

current between both sides of the switches. The current gain is expressed as [31] [?] [32]:

C=2

πsin(π∆ flo)

π∆ flo (4.4)

where

π∆ flo

is proportional to the bias current of the switches and inversely proportional to their

size. In other words, nearly ideal switching, i.e.

C=2

, can be achieved by either biasing the

switching transistors with as small current as possible, or increasing their size. Increasing the size

of the switches over a certain limit means that the parasitic capacitances increase, which leads to

slow switching. On the other hand, although lower biasing current increases the switching speed,

the gain of the transconductance stage is reduced due to reduced

fmax

of the transconductance

devices. Thus, careful design must take care of both current and switching transistor sizes for

optimal switching along with the minimum required LO power, which is also a very important

design issue.

Fig. 4.4 shows one such optimization process of switching transistor size and bias current.

Shown here are the open loop voltage values as a function of the switching transistor size for

various bias currents. At lower values of the current it can be seen that the gain is limited by the

transconductance transistors. Although at higher values of the biasing current higher gain values

are expected, but unfortunately due to the low speed operation of the switching transistors at high

bias current the gain is lowered. The rate of the gain decrease is larger than the increase in the

transconductance introduced by

and

. Optimum switch sizes are chosen such that they are

wide enough to speed up the switching and at the same time they do not introduce large parasitic

which decreases the gain. As shown in fig. 4.4, through this technique an optimum bias current

value of 6.5mA with 30 µm switching transistors has been chosen.

40CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

Figure 4.4:

Simulated open loop gain versus switching transistor size for

different bias current

Figure 4.5: Chip micrograph of the realized Miller divider

4.2.2 Measurement Results

The circuit was fabricated using a standard

0.13 µm

CMOS technology. The chip occupies a total

area of

0.68 ×0.64 mm2

including the RF and the DC probing pads. The chip micrograph is

shown in fig. 4.5. The divider input signal which is driven in a GSG configuration is shown at the

left side of the figure, while the output signal driving a

50 Ω

load is on the right. Three biasing

voltages are required (refer also to fig. 4.2):

Vdd

Vb1

, and

Vb2

; they are provided on the top of the

chip. The single ended input is converted into differential signal through passive on-chip balun

which in turn attenuates the input signal further by about

5dB

, meaning that the actual required

input power is lower than the measured one by this amount.

4.2. MILLER FREQUENCY DIVIDER 41

Fig. 4.6 shows both the measured and simulated sensitivity and operation range of the divider.

It can be seen that a minimum input power of

2dBm

is required for a frequency range from

24 GHz

34 GHz

. It is worth mentioning here that the divider fails to correctly divide at

frequencies lower than

24 GHz

, even though the input power is made much higher. This is as a

result of the insufficient suppression of the third order harmonic. Similarly, division ceases above

34 GHz due to the insufficient loop gain, resulting in a bandwidth of 10 GHz.

Figure 4.6: Input sensitivity of the Miller divider

Figure 4.7: Output power of the Miller divider

Fig. 4.7 shows the measured and simulated output power as a function of the input frequency

42CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

range at an input power of

0dBm

. The output power values are better than

−16 dBm

for the

whole frequency range. Finally, the output spectrum for input frequency of

28 GHz

is plotted in

Fig. 4.8. The spectrum shows a measured output power of

−8dBm

at an output frequency of

GHz.

Figure 4.8:

Spectrum analyzer’s screen shot of the output power of the Miller

divider

A comparison of the realized Miller divider with the state-of-the-art results is presented in

Table 4.3. With a trade-off between the various critical divider performances including the chip

area it can be seen that the realized circuit achieves a very good performance and is comparable

to the best reported results.

Performance Parameter This work [10] [33]

CMOS Technology (µm)0.13 0.18 0.13

Frequency (GHz)28 40 56.5-72.2

Frequency Range (GHz)10.0 2.3 15.7

Division Ratio 2 4 2

Input Sensitivity (dBm)-2 3 0

Output Power (dBm)-16 -16 -20

Supply Voltage (V)1.2 2.5 1.0

Power Dissipation (mW)10.0 16.8 4.7

Chip Area (mm2)0.68×0.64 0.50×0.70 0.83×0.51

Table 4.3: Miller divider’s performance comparison with the state of the art

4.3. VOLTAGE CONTROLLED OSCILLATOR 43

(a) (b)

Figure 4.9: (a) Conventional NMOS VCO (b) Transformer Feedback VCO

4.3 Voltage Controlled Oscillator

4.3.1 Conventional and Transformer Feedback VCO’s

There are three categories of the cross-coupled transistor VCOs; All PMOS, all NMOS, and

a combination of NMOS and PMOS cross-coupled transistors. PMOS VCO topologies have

many advantages over the other two in terms of phase noise for two reasons; first, the PMOS

device has lower noise than the NMOS counterpart due to the fact that the latter has lower

critical field, and thus suffers from velocity saturation more than the PMOS device. Second,

the varactors are grounded through a low impedance at low frequency, thus the effect of the

low frequency noise coming from the bias transistor is negligible [20]. However, the PMOS

device has a much lower

fmax

than the NMOS device. This may lead to higher bias current and

increased power consumption as a consequence of larger device size. Narrow tuning range and

lower maximum oscillation frequencies are other disadvantages. In other words, with increasing

operating frequencies the advantages of the PMOS VCO topologies are more and more lost.

Thus, only NMOS transistors were considered. Fig. 4.9a shows a conventional NMOS VCO in

Ws(µm)W1(µm)W2(µm)Cvar (f F)L(pH)Vbias (V)VDD (V)

650 35 10 40 –130 350 1.1 1.2

Table 4.4: Circuit Parameters of the Conventional VCO

44CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

W1(µm)W2(µm)Cvar f F L1(pH)L2(pH)VDD (V)

30 10 60–170 430 120 0.6

Table 4.5: Circuit Parameters of the Transformer Feedback VCO

which the headroom of the bias transistor limits the peak amplitude of the oscillation that further

degrades the phase noise. A PMOS current source is used instead of the NMOS counterpart

because of its better noise performance and to limit the tuning voltage to below

Vdd

. However, by

utilizing a transformer between the source and drain terminals of the cross coupled transistor the

bias transistor can be removed while keeping the VCO noise performance unchanged or even

improved. This topology is called transformer feedback VCO (TFVCO) and is shown in fig.

4.9b. The bias noise source and its upconversion mechanism do not exist in the TFVCO. The

transformer makes the drain and the source voltages in phase, thus enabling the oscillations to

swing below zero [34]. The amount of peak drain voltage improvement is proved in [35] and is

dependent of the turn ratio of the transformer. According to Leeson’s formula shown below, the

benefit is further enhancement of the phase noise at the same supply voltage, or equivalently, the

same phase noise at much lower supply voltage.

L(fm) = 2KTReqF

Ao2fo

2Q fm21+∆f1/f3

fm(4.5)

Switching transistors are another important source of noise, they not only contributing noise but

converting it from different frequency bands to around the oscillation frequency by the AM-PM

conversion mechanism. Nonlinearity of the switch transistors is the main cause of this conversion

mechanism. It directly causes second and third harmonic to flow in the low impedance varactors.

This causes the VCO to adjust the frequency to achieve the phase balance again. On the other

hand, amplitude modulation of the bias modulates the harmonics, causing phase shift variation

and hence frequency variation [36]. This effect exists only in the conventional NMOS VCO due

to the existence of the bias transistor. The AM-PM conversion can however be minimized by

minimizing the size of the switching transistors. The reasons can be explained as follows: Firstly,

the overdrive voltage becomes higher; therefore, the device becomes more linear. Secondly, the

bandwidth of the device becomes wider, thus higher order distortion decreases. Narrowing the

4.3. VOLTAGE CONTROLLED OSCILLATOR 45

device width has an implication of increased overdrive and is limited by the voltage headroom

needed by the bias transistor and the required transconductance needed to satisfy the conditions

of oscillation.

According to Leeson’s formula, phase noise is inversely proportional to the square of the

oscillation’s peak amplitude. This indicates that the phase noise performance of the conventional

NMOS VCO is again better than that of the VCO which utilizes both NMOS and PMOS

devices. Furthermore, the cross coupled PMOS transistors modulate more and more the varactors

capacitance causing further degradation of the phase noise. Minimizing the size of the varactors

may mitigate this influence but on the expense of the tuning range [37]. On the other hand,

limited output swing of the NMOS and PMOS VCO not only impacts the phase noise but affects

the output power that the VCO can deliver as well.

4.3.2 Design Approach

Fig. 4.9a shows a conventional NMOS VCO. It consists of two NMOS cross coupled transistors

(M1 and M2) in order to generate the negative resistance needed to compensate for the losses of

the LC tank. The value of this resistance is about

−1/gm

. Its absolute value must be less than the

equivalent parallel resistance of the LC tank at resonance so that the oscillations can start. This is

particularly difficult at high frequencies because the core transistors cannot be large due to the

capacitances they add to the tank. To accommodate core transistors with a sufficient width, the

parasitic capacitances connected to the tank must be minimized. At a given operating frequency,

the reduced parasitic capacitances also allow inclusion of larger varactors for a wider tuning

range. The transistor size limitation can also be alleviated by increasing the Qfactor of the tank

to lower the losses. Therefore, low parasitic and high

factor of the resonator circuit, as well as

low parasitic and high gain transistor design is needed to increase the maximum VCO operating

frequency. The

factor of the tank not only determines the required transistor width and hence

the maximum operation frequency but directly affects the phase noise performance as well.

On the other hand, the TFVCO topology (shown in fig. 4.9b) consists of transistors

and

which are also the cross coupled transistors producing the negative resistance to compensate

46CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

Figure 4.10:

Simulated phase noise, at

25 GHz

with

1MHz

offset, versus

cross-coupled transistor width for different values of bias current

the tank losses. The required

is higher than that of the conventional NMOS counterpart for

two main reasons: First, the cross coupled transistors are source degenerated by inductors that

make the equivalent negative resistance at the same

value lower than the conventional one.

Second, the

factor of the transformer is lower than that of the normal inductor, thus higher

losses and more

are required. To grantee oscillations at much lower supply voltage, the bulk

can be biased to decrease the threshold voltage

(Vt)

of the transistor. To relax the

requirement,

the transformer has to be carefully designed to attain high Qvalues.

Fig. 4.10 shows an approach to optimize the phase noise performance of the conventional

NMOS VCO. For the switching transistor width

W1,2

below

25 µm

, the VCOs fail to oscillate,

especially at lower levels of bias current. On the other hand, increasing the transistor widths

increase the parasitics and necessitates smaller varactors to maintain oscillation on the expense of

tuning range reduction. As discussed before, to mitigate the effect of the AM-PM conversion

caused by the cross coupled transistors, the width of the transistors must be decreased to improve

their linearity, however, further decrease of the transistor width impacts their flicker noise. Below

the optimum transistor width, the degradation of the flicker noise dominates the improvement of

the AM-PM conversion. The best phase noise can be achieved at transistor a width of

35 µm

and

bias current of

15 mA

. The minimum dimensions of varactors used are with

gates fingers and

active contacts (RX) repetitions for optimum Qfactor and widest possible tuning range.

4.3. VOLTAGE CONTROLLED OSCILLATOR 47

4.3.3 Measurement Results

The two VCO versions were fabricated in a standard

130 nm

CMOS technology. Fig. 4.11

shows the tuning range of both VCOs. In the TFVCO larger varactors were used which lead to

a wider tuning range at lower frequencies of oscillation. Chip micrographs of the conventional

NMOS VCO and the TFVCO are shown in fig. 4.12 and fig. 4.13 respectively. The chip area

0.55 ×0.39 mm2

for the conventional NMOS VCO, while the chip area of the TFVCO is

0.58 ×0.49 mm2

including RF and DC probing pads. For the TFVCO, the bulk of the transistors

is connected to a positive voltage in order to allow deep supply down scaling as a result of lower

threshold. Therefore, the conventional VCO consumes

23 mW

while the TFVCO consumes only

6mW

with a supply voltage of only

0.6V

. The measured phase noise of both VCOs are shown in

fig. 4.14. Measured and simulated phase noise of the conventional VCO compares to each other

25 GHz

; the size of the transistors is chosen such that the phase noise is optimum as depicted

in fig. 4.10 and shown to be

−103 dBc/Hz

at an offset of

1MHz

which is the same as obtained

in the measurements. The phase noise of the TFVCO is however

5dBc/Hz

worse because of

the deep supply voltage down scaling. The transformer is shown in fig. 4.15, both primary and

secondary inductor traces are drawn close to each other to increase their coupling coefficient.

The patterned ground shield is inserted below the traces to minimize penetration of the field into

the lossy substrate. The differential VCO output is probed using a GSSG configuration to drive

the

50 Ω

load. A common source buffer is used to isolate the VCO output nodes from the low

impedance load and to deliver a reasonable amount of power to the output. Probes and cable

losses are then de-embedded. A comparison between both versions is presented in table 4.6.

48CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

Figure 4.11: Measured tuning range of both conventional and TF VCOs

Figure 4.12: Chip micrograph of the all NMOS VCO

Figure 4.13: Chip micrograph of the Transformer feedback VCO

4.3. VOLTAGE CONTROLLED OSCILLATOR 49

Figure 4.14:

Measured phase noise at

1MHz

offset of both all NMOS and TF

VCO

Figure 4.15:

Layout of the Transformer, L1 and L2 drawn in the MA and E1

metal layer, respectively

50CHAPTER 4. DESIGN OF RF CIRCUITS FOR THE 24 GHZ FMCW RADAR TRANSCEIVER

Performance Parameter This work This work [37] [35] [38]

Conv. TF Conv. TF TF

CMOS Technology (µm)0.13 0.13 0.13 0.18 0.13

Frequency (GHz)26 22 59 24 24

Tuning Range (GHz)1.7 3.7 5.8 0.6 2.2

Phase Noise (dBc/Hz)@1MHz -97.5 -92.5 -89 -100 -113

Tuning Voltage (V)0⇒1 0 ⇒1.2 0 ⇒1.5 0 ⇒1.2 0 ⇒1.2

Supply Voltage (V)1.2 0.6 1.5 0.65 0.6

Output Power (dBm)0.5 -3 -10 -13 -10

Power Dissipation (mW)23 6 9.8 7.8 3

Chip Area (mm2)0.39×0.62 0.58×0.49 N.A 0.7×0.6 0.6×0.6

Table 4.6:

Performance comparison of the two VCO versions with the state of

the art

Chapter 5

Design of the IQ Modulator for the 60 GHz

WLAN Transceiver

5.1 System Overview

The first stage of the short-range high data-rate transmitter according to the IEEE 802.15.3c

PHY standard is the IQ-modulator (refere to fig. 1.4). The block diagram of the IQ-modulator

is shown in fig. 5.1. The

40 GHz

LO signal generated by the PLL is fed into the input of the

static frequency divider (SFD) to provide wideband quadrature signals at

20 GHz

. The direct

up-conversion mixers translates the

2GHz

base-band signal into the IF range at around

20 GHz

The mixer outputs are combined together and fed to the IF amplifier whose input is matched to

25 Ω

. The SFD operates at the four carrier frequencies as in Table 1.1 with a required input LO

power of only 0 dBm.

5.2 Static Frequency Divider

5.2.1 Design Approach

The configuration of a SFD is shown in fig. 2.6. It is realized with two cross-coupled D flip-flops

(D-FF), each D-FF is terminated with a small buffer. The two D-FF outputs toggle at every

52CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.1: Block diagram of the IQ-Modulator

positive and negative going transitions of the clock, thereby generating two quadrature outputs

at half the clock frequency. A high speed realization of D-FF is achieved through current-mode

logic circuit (CML), using a sensitive differential amplifier or evaluating amplifier, shown in

fig. 5.2, and a latching stage responsible for holding the state of the D-FF waiting for the next

evaluation phase. During the positive half cycle of the clock, the differential sensor

Meval

should

be fast enough to correctly sample the input, and in the negative clock cycle, the latch circuit

Mlatch is switched on and latches the evaluated value.

Boosting the operation frequency of the SFD was always accomplished by using inductive

peaking [14] and split resistors [15]. In this work these techniques are used to shorten the

evaluation phase without boosting the gain of the evaluation amplifier or requiring high

the latching devices. In the conventional shunt peaking, inductor and resistor were interchanged

so that the inductor is connected directly to the latch drains. In this case the inductor parasitic

capacitance to ground cannot be distinguished from the drain capacitance. As a result, the time

constant also increases, reducing the latch speed. In this work (fig. 5.2) the inductor is inserted

close to the supply. Moreover, stacked inductors could be used to preserve area. The inductor

is not an issue as it is already in series with another resistor, so its resistive part could be absorbed

into it.

In this approach, although having effect on each other, both the evaluating amplifier and

5.2. STATIC FREQUENCY DIVIDER 53

Figure 5.2: Circuit diagram of a single latch

Weval (µm)Wlatch (µm)Wclkp (µm)Wclkn (µm)Lload (pH)R(Ω)R1 (Ω)

24 6.4 30 15 800 40 20

Table 5.1: Circuit Parameters of the Latch Circuit

the latching circuits are considered separately. The loading of both on each other is taken into

account during the design procedure. A design cycle is repeated for optimum sizing and current

at the required frequency. The value of the load inductor is chosen such that it resonates with the

latch devices’ capacitances to oscillate at the free running mode. Splitting ratio is finally decided

to provide the fastest possible charging and discharging response. Losses that are compensated

by the latch transistors should be kept low to avoid extra current into the latch.

5.2.1.1 Evaluating Amplifier Design

As it becomes selective, the gain of the evaluating amplifier determines the operation range of

the SFD; in other words, without using series resistors, gain of the evaluation amplifier is high

enough over a narrow band, and thus it correctly evaluates the SFD state over a narrow band of

frequencies. Choosing an optimum load resistor is a trade-off between the frequency band of

operation and enough gain of the evaluating amplifier. The maximum load resistor value of

60 Ω

is selected, so that the rising and falling times of the output nodes are short enough to follow the

maximum required operation frequency; at the same time, gain is enough for the lower frequency

edge. The design approach for minimum power can begin by selecting the appropriate size for

54CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.3:

Open loop gain of the evaluating amplifier as a function of the tail

current transistor size

Meval

to provide enough gain, recalling that its size does not significantly affect the speed of the

latch [12] but its optimum size can be selected to fine tune the required fosc.

Fig. 5.3 shows a family of gain curves corresponding to different sizes of

Mclkp

. Increasing

the DC current boosts the gain of the evaluating amplifier and thus shifts the unity gain frequency

upward. Removing the load resistor shifts the amplifier gain curve further up, but with a very

narrow range of not more than 2 GHz. For deeper insight into the effect of the amplifier current

on the SFD operation, fig. 5.4 shows the effect of

Mclkp

size on both current dissipation and

fosc

DC current is directly proportional to the device size

Mclkp

, it can be increased up the point at

which the transistor goes into the linear operation. Although the

fosc

also increases in proportion

to current, it increases much slower at higher current. In other words, a couple of

GHz

can

be achieved at the cost of doubling the current dissipation. A transistor

Mclkp

size of

30 µm

selected so that the current is 1.9mA at an fosc of about 20 GHz.

5.2.1.2 Sizing of the Latch Stage

During the latching phase, both (

Mlatch

and

Mclkn

) should operate in saturation while the os-

cillation condition is satisfied (i.e.

R1·gmlatch >1

) and minimum possible latch transistor size

for speed consideration. Remember that the losses of the low

stacked inductor add up to the

5.2. STATIC FREQUENCY DIVIDER 55

Figure 5.4:

Self oscillation frequency of the SFD and evaluating amplifier DC

current are proportional to the tail current transistor size

total losses of the load. A reasonable value for

is initially chosen, taking into account that

the decrease of

might lead to a requirement of larger

Mlatch

, thus pulling the frequency of

operation down.

The current need not to be the same for the evaluation and the latch phases, the current for the

latch network is actually independent of that drawn into the evaluation amplifier. Asymmetric

input device sizes design saves much current that was not needed in the cross coupled transistors.

Fig. 5.5 depicts the dependency of

fosc

and the latch DC current on sizes of the cross coupled

transistors

Mlatch

and the negative clock transistor

Mclkn

. The current of the latch can not be

decreased beyond a value that allows for full compensation of the load losses, thus oscillation

was not possible at smaller transistor size

Mclkn

. Even higher

fosc

can be achieved at lower latch

current. According to fig. 5.5,

Mclkn

transistor size is chosen to be

15 µm

at which the latching

stage draws less than

1mA

. The

Mlatch

transistor size of

6.4µm

is then selected to achieve the

required fosc of about 20 GHz.

5.2.1.3 Design of the Splitting Resistor Ratio

Starting from the fact that the two stage

networks have shorter rising and falling times than

that of single time constant networks, splitting the load resistor improves the setup time of the

56CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.5:

Self oscillation frequency and latching transistors DC current varies

in accordance to both the tail current device size and cross coupled transistors

size

Figure 5.6:

Step response of the SFDs output nodes for different splitting ratios

of the resistive load

latch, thus increasing the maximum operation frequency [15]. A total load resistance is chosen

to be

60 Ω

. Lower values necessitate higher current in the latching stage to compensate losses.

On the other hand, higher values increases the time constant of charging and discharging the

output node capacitance, this slower response translating directly to prolonged setup time and

thus decreased frequency of operation. Fig. 5.6 shows a step response of the output nodes for

different splitting ratios. The procedure begins by keeping the total value of the load resistor

5.2. STATIC FREQUENCY DIVIDER 57

Figure 5.7: Circuit diagram of the output buffer

Wtail (µm)Wdi f f (µm)Lload (pH)R(Ω)Vbias (mV )

50 10 400 45 700

Table 5.2: Circuit Parameters of the Output Buffer Circuit

unchanged and distributing its value between

and

as shown in fig. 5.2. As

increases

and

decreases by the same amount, the rise time improves. When

goes greater than

2·R1

the SFD stops oscillating, as

is not enough to maintain oscillations (i.e.

R1·gmlatch <1

)

and an excess current is needed to boost the latch transconductance. As shown in fig. 5.6, an

improvement of about 50 percent of the rise time is achieved. The

time constant circuit can

even be split into three or more stages for further response improvement - complexity is a price

that should be paid.

5.2.1.4 Output Buffers

This SFD is intended to be used as an IQ signal generator, thus both latches are made symmetrical

and both of them are terminated with identical buffers. The buffers are used to isolate the output

node of the SFD from the next stage and to provide the

50 Ω

output impedance for measurement

and integration purposes. Along the design procedure, the loading effect of the buffers on the

latches should be taken into consideration, thus they are designed at the beginning and it is taken

into account that their sizes should be kept as small as possible to save further power at the

required frequency. Fig. 5.7 shows the buffer circuit. Inductive peaking of a resistively loaded

58CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

amplifier is a good choice, as it can provide bandwidth extension and can be easily matched. The

current consumption of each is 0.5mA.

5.2.2 IQ Imbalance

Due to integration difficulty or selectivity constraints, band pass filters in the receiver front end

are not desired. Instead, image rejection mixers are adopted. In the single quadrature image

rejection mixer of fig. 5.8, only the local oscillator signal is generated in

and

paths. The

IQ signal representation of the

is, mathematically, one impulse at only positive or negative

frequency; Thus, the

signal is shifted down once to the right or to the left, respectively; thus,

no overlap of the image over the wanted signal takes place. In reality, due to tolerances and

devices mismatching, finite amplitude and phase imperfections exist, thus a residue of LO appears

at the positive frequency, as represented in the equations below:

Figure 5.8:

Single quadrature down-conversion mixer with a SFD as an IQ

local oscillator signal generator

LO =ALOe−jωLOt+ALOpejωLOt(5.1)

LO = [ALO +ALOp]cos(ωLOt)−j[ALO −ALOp]sin(ωLOt)(5.2)

Suppose now that the

and

of the

have gain and phase errors of

and

respectively. Then

the LO could be modeled as follows:

LO = (1+ε)cos(ωLOt+φ)−jsin(ωLOt)(5.3)

5.2. STATIC FREQUENCY DIVIDER 59

Figure 5.9: Simulated imbalance of IQ amplitude and phase

LO = [(1+ε)cosφ]cos(ωLOt)−j[1+j(1+ε)sinφ]sin(ωLOt)(5.4)

Equating coefficients of 5.2 and 5.4 yield:

ALO =1

2[1+(1+ε)ejφ](5.5)

ALOp =1

2[−1+(1+ε)e−jφ](5.6)

The image rejection ratio IRR in dB can be calculated as follows:

IRR =10log

ALO

ALOp 

(5.7)

IRR =10log1+2(1+ε)cosφ+(1+ε)2

1−2(1+ε)cosφ+(1+ε)2(5.8)

Fig. 5.9 shows the simulated IQ power and phase difference between the quadrature outputs.

Maximum phase and power imbalance is

2.5◦

and

0.7dB

, respectively. According to equation

(8) the IRR is about

29 dB

over the specified operation frequency range. The

IRR

is verified in

fig. 5.10.

60CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.10: Simulated image rejection ratio due to the IQ imbalance

Figure 5.11: Chip micrograph of the SFD showing the pad configuration

5.2.3 Measurement Results

The circuit was fabricated in a

90 nm

CMOS process. It occupies a chip area of

0.60 ×0.75 mm2

including probing pads. A micrograph of the frequency divider is shown in fig. 5.11. To ease the

on-chip measurement, an on chip passive balun was also designed. A

50 GHz

signal generator is

used to generate the input. Two differential outputs are available, one is the In-phase and the other

is the quadrature-phase outputs, one is terminated to

50 Ω

while the other is to be measured. The

output spectrum is measured with Agilent Spectrum Analyzer E4440A. With the

1.2V

power

supply, the divider circuit draws a current of 6.8mA including the output buffers.

The measured and simulated input sensitivity is shown in fig. 5.12. The cable losses, probe

losses, and the on-chip balun loss have been de-embedded. With a

−1dBm

input power, the

5.2. STATIC FREQUENCY DIVIDER 61

Figure 5.12:

Simulated and measured sensitivity curve of the static frequency

divider

Figure 5.13: Spectrum of the free running SFD

62CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

circuit operation range is

12 GHz

, from

33.5GHz

45.5GHz

. The divider self-oscillating

frequency is

20.5GHz

. Fig. 5.13 shows the output spectrum when no input power is applied.

Table I summarizes and compares the performance of the developed static divider with other

recently published work. It can be noticed, that the power consumption is lowest among the

others with comparable and higher operation frequencies. Due to the use of inductive peaking,

the circuit becomes more selective, explaining the fact that the locking range is narrower but still

sufficiently wideband for our application.

This work [12] [13] [14]

Technology (nm)90 130 130 90

Supply (V)1.2 1.5 1.2 1.2

Power Dissipation 6.9 9 8.4 15.7

Input Power (dBm)-1 0 0 -1

Operation Frequency (GHz)33.5 ⇒45.5 5 ⇒45 9 ⇒34 49 ⇒67

Output Power (dBm)-16 N.A N.A N.A

Chip Area (mm2)0.60×0.75 N.A 0.10×0.20 0.68×0.80

Table 5.3: Performance comparison of the SFD with the state of the art

This work [15] [16] [17]

Technology (nm)90 90 130 130

Supply (V)1.2 1.5 1.5 1.5

Power Dissipation 6.9 39.7 32 11.7

Input Power (dBm)-1 0 N.A 6.5

Operation Frequency (GHz)33.5 ⇒45.5 4 ⇒54 30 ⇒43.2 31 ⇒50

Output Power (dBm)-16 N.A N.A N.A

Chip Area (mm2)0.60×0.75 1.00×1.00 0.63×1.00 0.96×1.15

Table 5.4: Performance comparison of the SFD with the state of the art

5.3 Direct Up-Conversion Mixer

The direct up-conversion mixer translates the base-band signal up to the IF frequency at

20 GHz

Due to the low frequency base-band signal, the output up converted signal lies in the vicinity of

the LO signal frequency. Therefore, the isolation for the LO at the output should high. In other

5.3. DIRECT UP-CONVERSION MIXER 63

words, the level of the LO at the IF port should be lower than the up-converted signal. In this

sense, a double balanced mixer is used instead of single balanced structure. On the other hand,

the IQ modulator should have high conversion gain to provide enough power to the front-end

transmitter, thus a Gilbert-cell mixer is used instead of the passive counterpart.

Figure 5.14: Circuit schematic of the up-conversion mixer

Wd(µm)Ws(µm)L(pH)C(f F)Ldeg (nH)Ltail (nH)

155 44 350 105 2 2

Table 5.5: Circuit parameters of the up-conversion mixer

Fig. 5.14 shows a circuit diagram of the up-conversion mixer. The linearity of the mixer is

an important issue, especially for more complex modulation schemes where the difference in

amplitude between symbols is small. The linearity of the mixer is determined by the transconduc-

tance stage. The third order intercept point can be shifted up by using a derivative superposition

technique [39]. This method is effective only at a very low frequency, when the parasitics are

negligible. In the range of the base-band signal,

2GHz

, it does not working anymore, as a result

of the parasitic capacitances which degrade

of the driver transistors. The linearity of the up-

conversion mixer is improved instead by inserting degeneration inductors. Degeneration shares

the input power with the driver transistors

(Md)

, degrading the conversion gain but boosting the

compression point.

64CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.15:

Simulation of conversion gain of the up-conversion mixer vs. the

size of the switching transistors

5.3.1 Design Approach

Unlike the down-conversion mixers, direct up-conversion mixers do not need peaking inductors

at the output nodes of the driver transistors. The capacitance at these nodes is small and most of

the RF current flows through the switching transistors

(Ms)

. Nevertheless, increasing the size of

the

adds up extra capacitances causing more RF current to flow into the parasitic capacitances.

Increasing transistor size

(Ms)

on the other hand results in less gate-source DC voltage and better

switching behavior, and hence less loss of the switching transistor current gain. As a result, an

optimum size of the switching transistors compromises the switching speed with the RF current

loss in the parasitic capacitances. Fig. 5.15 shows an optimum switching transistor size of

44 µm

Design of the driver transistor’s size

(Ms)

determines directly the power consumption of the

mixer as well as the conversion gain. In this design,

(Md)

is biased with gate voltage of

550 mV

so that the third current derivative is zero; this leads to a maximum possible third order intercept

point. Load inductor

is chosen such that it provides maximum conversion gain at around the IF

frequency of 20 GHz. The output is then matched to the 50 Ωload.

5.3. DIRECT UP-CONVERSION MIXER 65

Figure 5.16: Chip micrograph of the up-conversion mixer

5.3.2 Measurement Results

The mixer was fabricated in a

90 nm

CMOS process. It occupies an active area of

0.66×0.42 mm2

including probing pads. A micrograph of the Gilbert-Cell mixer with the proposed dynamic

current bleeding network is shown in fig. 5.16. To ease the on-chip measurement, an on chip

passive balun was also designed and connected to the LO port of the mixer; meanwhile, the

RF input of the mixer can be driven single ended because of the tail common mode rejection

inductor

Ltail

. Nevertheless, differential base-band signal was generated using off-chip balun

and two couplers. Two

50 GHz

signal generators are used to generate LO and BB signals. The

differential output is probed using differential probe; one of single-ended output is terminated

50 Ω

while the other is connected to the input of the Agilent Spectrum Analyzer E4440A.

The cable losses, probe losses, and the on-chip balun loss have been de-embedded. LO power is

swept so that the best conversion gain is achieved, maximum gain corresponds to a LO power of

6dB

as shown in fig. 5.17, de-impeding the losses of the integrated passive balun results in an

optimum LO power of

1dBm

. Best case conversion gain is

9.5dB

at LO frequency of

24 GHz

and base band frequency of

1.5GHz

, measured and simulated conversion gain curves are shown

in fig. 5.18. The

1dB

compression point is

−8dB

as shown in fig. 5.19. With the

1.2V

power

supply; the core Gilbert-Cell mixer draws a current of

6.3mA

. Tables 5.6 and 5.7 summarize the

performance of the direct up-conversion Gilbert-Cell mixer and compare it with the state of the

art results.

66CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.17:

Optimum LO power for maximum conversion gain of the up-

conversion mixer

Figure 5.18: Conversion gain of the up-conversion mixer

5.3. DIRECT UP-CONVERSION MIXER 67

Figure 5.19: P−1dB of the up-conversion mixer

This work [40] [41] [42]

Technology 90 nm 90 nm 0.13 µm 90 nm

Power Dissipation (mW)7.56 13.2 8 11.1

Conversion Gain (dB)9.5 -11 0.7 2

Optimum LO Power (dBm)1 0 3 5

P1dB(Input Referred) (dBm)-8 1 -5.8 -14.8

LO Frequency (GHz)15 ⇒35 40 22 ⇒29 20 ⇒26

Base-Band Frequency (GHz)0.01 ⇒2 11 (IF)1.5 ⇒3.5 2.7

Chip Size (mm2)0.66 ×0.42 0.60 ×1.05 0.68 ×0.68 0.65 ×0.57

Table 5.6:

Summary and performance comparison of the direct up-conversion

mixer

This work [43] [44]

Technology 90 nm 0.13 µm 0.13 µm

Power Dissipation (mW)7.56 24 2.7

Conversion Gain (dB)9.5 4 -5.6

Optimum LO Power (dBm)1 0 N.A

P1dB(Input Referred) (dBm)-8 -7 -14

LO Frequency (GHz)15 ⇒35 59 ⇒65 58.3 ⇒62.5

Base-Band Frequency (GHz)0.01 ⇒2 0.25 0.05 ⇒3.5

Chip Size (mm2)0.66 ×0.42 0.30 ×0.70 0.88 ×0.78

Table 5.7:

Summary and performance comparison of the direct up-conversion

mixer

68CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

5.4 IF Amplifier

5.4.1 Cascode Differential Amplifier

The IF signal is amplified before or after complex mixing in the demodulator or modulator

respectively. A cascode differential amplifier is adopted in this design, as shown in fig. 5.20.

Cascode configuration is better than the common source amplifier in this design. It partially

trades-off the gain with better linearity, stability and isolation. The output of the IF amplifier is

matched to

25 Ω

in order for its output to be equally split between the two inputs of the mixers

in the IQ-demodulator circuit, while its input is matched to

25 Ω

to receive equal power from

the mixer’s outputs in the IQ-modulator circuit. Fig. 5.21 and fig. 5.22 show power gain and

input output isolation of the IF amplifier, respectively. The power gain is

7dB

while the isolation

is a bit less than

−10 dB

at the frequency of interest. The use of stacked inductors in the input

and output matching circuitry makes them highly lossy, especially when it is needed to put extra

fillers inside and near to the inductor traces. The effect of these fillers is not included in their EM

model to speed up the simulation, and therefore there is a great difference between the simulated

and measured gain, input, and output matching.

Figure 5.20: Circuit diagram of the cascode IF amplifier

5.5. SYSTEM INTEGRATION 69

Wd(µm)Wcas (µm)Lmatch (pH)Cmatch (f F)C(f F)R(KΩ)Ltail (nH)

50 50 390 85 800 1 450

Table 5.8: Circuit parameters of the IF amplifier

Figure 5.21: Power gain of the IF amplifier

Figure 5.22: Input and output matching of the IF amplifier

5.5 System Integration

Referring to the IQ-Modulator block diagram of fig. 5.1, the mixer should be driven by an enough

LO power in order to maintain the highest possible conversion gain. As shown in fig. 5.17, the

optimum LO power required by the down-conversion mixer is

1dBm

, taking into account the

70CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.23: Frequency divider and IQ generator for the LO signal

losses of the balun, while the power delivered by the static frequency divider

(SFD)

is around

−16 dBm

. In reality, the LO port of the mixer requires voltage swing rather than power level.

According to simulations, the voltage swing of the SFD’s output is

20 mV

, and the required

voltage swing that can fully switch the switching transistors of the mixer on and off is

150 mV

Therefore, an amplifier with the load inductor should resonate with the LO port’s capacitor of

the mixer at the required frequency of operation at around

20 GHz

. The block diagram of the

SFD and the circuit diagram of both the latch and the output buffer are shown in fig. 5.23; the

output buffers are mainly used to isolate the output node of the latch from the high capacitance

of the following stage, and to provide matching to the

50 Ω

load. When it is integrated in the

IQ-Demodulator, there is no need to design a matching network - its output node is directly

fed into the gate terminals of the LO buffer of the mixer. The LO buffer, along with the direct

down-conversion mixer, is shown in fig. 5.24. The load inductor of the LO buffer,

Lreso

, and the

drain capacitance of the buffer from one side, and the gate capacitances of the switches from the

other side, constitute a parallel

resonance network, whose impedance is maximum at around

20 GHz, thus a maximum voltage swing of 450 mV is produced.

As shown in fig. 5.25, the chip micrograph of the complete IQ-Demodulator, the LO buffers

are located very close to the LO port of the mixer. The

and

outputs of the SFD are routed

5.5. SYSTEM INTEGRATION 71

Figure 5.24: Circuit diagram of the up-conversion mixer with the LO buffer

Figure 5.25:

The chip micrograph of the IQ-Modulator showing circuit compo-

nents and probing pads

72CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Figure 5.26: Conversion gain of the IQ-Modulator

using transmission lines. The input of the IF amplifier is matched to

25 Ω

to receive equal power

from the two sides of the IQ-demodulator’s mixers.

The IQ-modulator has been fabricated in a TSMC

90 nm

CMOS process. It occupies an

active area of

0.82×1.17 mm2

including probing pads. A micrograph of the Modulator including

inputs, outputs, and DC pads is shown in fig. 5.25. To ease the on-chip measurement, an on-chip

passive balun was also designed and connected to the input of the SFD; meanwhile, an input

balun followed by two line couplers are used to generate two differential quadrature signals to

feed the base-band input terminals of the modulator. Two

50 GHz

signal generators are used to

generate LO and BB signals. The differential output is probed using a differential probe; one

of single ended outputs is terminated to

50 Ω

while the other is connected to the input of the

Agilent Spectrum Analyzer E4440A. The cable losses, probe losses, and the on-chip balun loss

have been de-embedded. The best case conversion gain is

6dB

at LO input frequency of

GHz

and base band frequency of

2GHz

, measured and simulated conversion gain curves are

shown in fig. 5.26. The

1−dB

compression point is

−12 dBm

. With the

1.2V

power supply,

the complete IQ-Modulator consumes a total power of

42 mW

including LO buffers and output

matching buffers. Tables 5.9 and 5.10 summarize the IQ-demodulator performance and compares

it to the state of the art.

5.5. SYSTEM INTEGRATION 73

This work [45] [46]

Technology CMOS 90 nm 0.13 µm 0.13 µm

Power Dissipation (mW)42 75.9 39

Conversion Gain (dB)6 -6 -6.5

Minimum LO Power (dBm)-1 N.A 5

P1dB(Input Referred) (dBm)-12 N.A -2.5

LO Frequency (GHz)33.5 ⇒45.5 20 ⇒32.5 50 ⇒70

Base-Band Frequency (GHz)0.01 ⇒2.17 15 1

Chip Size (mm2)0.84 ×1.15 0.98 ×0.80 1.08 ×0.65

Table 5.9: Summary and performance comparison of the IQ-Modulator

This work [47] [48]

Technology CMOS 90 nm 0.13 µm 90 nm

Power Dissipation (mW)42 21 0

Conversion Gain (dB)6 -8 -13

Minimum LO Power (dBm)-1 4 4

P1dB(Input Referred) (dBm)-12 -20

LO Frequency (GHz)33.5 ⇒45.5 30 ⇒65 50 ⇒62

Base-Band Frequency (GHz)0.01 ⇒2.17 N.A 0.1

Chip Size (mm2)0.84 ×1.15 0.89 ×0.56 0.85 ×0.6

Table 5.10: Summary and performance comparison of the IQ-Modulator

74CHAPTER 5. DESIGN OF THE IQ MODULATOR FOR THE 60 GHZ WLAN TRANSCEIVER

Chapter 6

Design of the IQ Demodulator for 60 GHz

WLAN Transceiver

6.1 System Overview

As depicted in fig. 1.4, the IQ-demodulator translates down the IF frequency of

20 GHz

, derived

from the first down-conversion stage, to the base-band frequency range of

2GHz

in a complex

form. The block diagram of the IQ-demodulator is shown in fig. 6.1. The

and

signal

imbalance determines the

EVM

of the received symbols, therefore, accurate IQ signal is highly

demanded especially for high order modulation schemes. The static frequency divider is thus

employed to grantee balanced IQ signal over the required wide band of operation. The IF output

of the amplifier is matched to

25 Ω

. The SFD can operate at the four carrier frequencies as in

Table 1.1 with a required input LO power of only 0 dBm.

6.2 Low-Flicker Noise Direct Down-Conversion Mixer

6.2.1 Noise Mechanism and Current Bleeding Network

The dominant noise source of the mixer is the flicker noise of the LO switch stage because the IF

signal is located at closed to DC frequency. The flicker noise of the mixers is generated by the

76CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.1: Block diagram of the IQ-Demodulator

transconductance stage, switching stage and load stage. Usually, the load stage does not have

much effect on flick noise, in fact, flicker noise of the transconductance stage is converted to

LO frequency. So flick noise of the mixer is mainly determined by the switching stage. The LO

switches generate noise pulse trains by the direct mechanism and the DC average of noise pulse

trains is the output flicker noise current as follows [22]:

in,out(dir)=4I×Vn

S×T(6.1)

Vn=s2×Kf

WLCox f(6.2)

S=Vn

∆t(6.3)

Where

is the bias current for the RF transconductance stage,

is the LO period,

is the

equivalent flicker noise of the switching pair and

∆t

is the slope of the LO signal.

and

are

the effective width and length,

Cox

is the oxide capacitance,

is frequency, and

is a process

parameter.

In order to decrease flicker noise in the direct mechanism, the size of the switching pairs needs

to be increased, and large switching devices increase the parasitic capacitance of the switching

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 77

Figure 6.2: Gilbert-Cell mixer with conventional current bleeding network

pairs, resulting in the flicker noise indirectly translating to the output. The second mechanism

that generates flicker-noise is the indirect mechanism, flicker-noise mainly depends on the tail

capacitance

(Cp)

at the node between the LO switches and RF transconductance stage. In order

to decrease the flicker noise in CMOS active mixers, the bias current of the local oscillator (LO)

switches should be small enough to lower the height of the noise pulses [22].

io,n(ind)=2Cp

TVn×4I×Vn

ms +(CpωLO)2(6.4)

Where

is the tail capacitance of the node between the LO switches and the RF transconductance

stage,

is the LO period,

gms

is the transconductance of the LO switches, and

is the equivalent

flicker noise of the switching pair. According to equation 6.4, the tail capacitance should be small

enough to decrease the effect of the indirect mechanism. Therefore, a trade-off between direct

and indirect mechanisms necessitates an optimum selection of the switching transistor size. In

order to not design high gain low noise mixers, current bleeding networks are commonly used for

the direct down-conversion mixers.

Usually, increasing the bias current through the switching stage can lead to a high gain and

better linearity. However, when the current through the switching stage becomes higher, the

78CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.3:

Equivalent circuit of the Gilbert-Cell mixer with conventional

current bleeding mixer

voltage headroom problems of the load will be present. The current bleeding technique is usually

adopted to reduce bias current through the switch, as shown in fig. 6.2. The equivalent circuit is

shown in fig. 6.3. By current division, as can be seen from equation 6.5, some RF current flows

into the bleeding circuit and it will decrease the conversion gain [22].

ILO =RB

RB+1

gm1(t)+gm2(t)×p1(t)×iRF (6.5)

On the other hand, the output noise current of LO switches is decreased as the bleeding current is

increased. If the amount of the bleeding current is increased, the bias current of the LO switches

is decreased. As we can see from equation 6.1, the noise current by the direct mechanism is then

decreased. However, there are a few drawbacks with the conventional current bleeding technique.

As the bias current of the LO switches is reduced, the impedance of the LO switches, as seen from

the RF stage, is increased. Therefore, more RF leakage current will flow into the bleeding circuit,

decreasing conversion gain. It also allows more RF current to be shunted by the tail capacitance

as shown in fig. 6.3. Another drawback that makes this current bleeding network ineffective,

is that the tail capacitance

increases, making the indirect noise mechanism dominate. The

conventional current bleeding technique can reduce the bias current of the LO switches, but

flicker noises are still generated by the indirect mechanism (tail capacitance). The tail capacitance

is still needed to be reduced. The best way to reduce the tail capacitance is to minimize the

size of the LO switches and RF transconductance stages. However, CMOS transistors suffer

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 79

Figure 6.4:

Gilbert-Cell mixer with static current bleeding network and one

inductor

Figure 6.5:

Equivalent circuit of Gilbert-Cell mixer with current bleeding and

one inductor

from high intrinsic flicker noise, which is inversely proportional to the device area. Therefore,

one inductor

is connected between the common source node of the LO switches, as shown in

fig. 6.4, to resonate the tail capacitance out. In this way, the conversion gain and flicker noise

performance are improved simultaneously under resonant condition. As shown in fig. 6.5, by

resonating the tail capacitance with

, the impedance at node A looking into

can be high

enough to protect some RF current from being shunted by the tail capacitance.

For the resonant circuit, the nonlinear capacitances are padded by the inductor and the

80CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

harmonics generated are reduced significantly. The admittance of the resonant circuit is:

Yin(ω) = jωRFCp+1

jωRF L=jωRFCp1−1

ω2

RF LCp(6.6)

When the resonant frequency is at RF, the impedance at node A looking into

can be high

enough to protect some RF current from being shunted by the tail capacitance. This increases the

conversion gain by 2 to 3 dB. When the resonant frequency is at 2 RF, we have:

2ωRF =1

pLCp

(6.7)

In this case the input admittance of the resonance circuit is

Yin(ω) = −jωRF ·3Cp

. Thus, the

effective variation of

is reduced by a factor of

, reducing the nonlinearity. This will improve

the linearity. Setting the resonant frequency between RF and 2RF will make a good compromise

between the conversion gain and linearity.

The static current injection technique was proposed to reduce the bias current of the LO

switches. However, the impedance of the LO switches seen from the RF stage is increased as we

reduce the bias current of the LO switches. In addition, RF leakage current flows through the

injection circuit, which decreases conversion gain and also allows more RF current to be shunted

by the tail capacitance

(Cp)

at the node between the LO switches and RF transconductance stage.

Dynamic Current Injection (DCI) technique has been used in the mixer in order to reduce direct

flicker-noise generation. As shown in fig. 6.6, two PMOS,

, were used to inject a dynamic

current equal to the bias current of each pair of switches at only the switching event. Apart from

the instant of switching, current can flow through the switches, reducing their impedance and

thus lowering loss of the RF current. A parallel tuning inductor

has been employed to tune out

the parasitic tail capacitances Cpto alleviate the indirect flicker noise source.

As shown before, by resonating the tail capacitance with

, the impedance at node A looking

into

can be high enough to protect some RF current from being shunted by the tail capacitance.

We have improved the conversion gain ranging from

3dB

in simulation by using one inductor.

However, there are still two shunted paths for RF current leakage. One is the current bleeding

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 81

Figure 6.6:

Gilbert-Cell mixer with dynamic current bleeding circuit and one

inductor

circuit, and the other one is the shunted inductor. In this work, a current bleeding technique with

two inductors connected between the common source node of the LO switches and the PMOS

is proposed, as shown in fig. 6.7. Therefore, two inductors are connected to the PMOS device

to resonate the tail capacitance out, and the conversion gain and flicker-noise performance are

improved simultaneously under resonant condition. Another important role of these two inductors

is to protect RF current flowing into the current bleeding circuit. This helps RF current flow into

the mixer output directly. Therefore, we can achieve more conversion gain than the conventional

current bleeding technique.

In the proposed design shown in fig. 6.7, parasitic capacitances of the dynamic bleeding

transistors

(Mb)

are kept separated from the switches common sources by series resonating

inductors (L). Therefore, the flicker noise due to the indirect mechanism is not increased; at the

same time the DCBN provides the required DC current of the driver stage, minimizing the effect

of the direct mechanism. The conversion gain is simultaneously improved, as there are only two

paths for the RF driver current to ground: The source impedances of the switches, which is low

during the LO period except for its zero crossings, and the resonance network, whose equivalent

resistance is 300 Ω.

82CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.7:

Gilbert-Cell mixer with the proposed dynamic current bleeding

mixer and two inductors

W1(µm)Ws(µm)Wb(µm)Lb(pH)R(Ω)Ibias (mA)Ldeg (pH)Ltail (pH)

60 35 300 420 900 3.6 450 500

Table 6.1: Circuit Parameters of the direct down-conversion mixer

As shown in fig. 6.8, the resonating inductor

(L)

divides up the total tank capacitance into

two series parasitic capacitances; drain and source capacitances of

and

, and gate and drain

parasitic capacitances of

. The series inductors

(Ls)

in this case are nearly four times larger

than the shunt inductor

(Lp)

in the conventional DCBN for two reasons; firstly because every

series inductor

(Ls)

resonating with the capacitances of only one side, secondly because the total

parasitic capacitances are now series to each other instead of being shunted. In this design,

420 pH while Lp is 125 pH.

Improvement of the noise figure is shown in fig. 6.9. The output channel bandwidth is

10 MHz

2GHz

while the LO frequency is set to

20 GHz

. Both of the designs have the same

and

sizes, so that the flicker noise by the direct mechanism is the same. Nevertheless, with the

proposed DCBN, the noise figure is improved by

8dB

more than the conventional one because

of the enhancement of the flicker noise by the indirect mechanism, which resulted from splitting

the capacitances at the common sources. Meanwhile, the conversion gain is also improved by

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 83

Figure 6.8:

Equivalent curcuit of the Gilbert-Cell mixer with dynamic current

bleeding network and two series inductors

an average of

6dB

. Driver transistors for both designs

(M1)

are also made identical, but in the

proposed design it is shown that more RF current is benefited in the switching transistors. Fig.

6.10 shows the conversion gain of both designs in the LO frequency range of

18 GHz

23 GHz

at the base band frequency of 1 GHz.

Figure 6.9:

Simulated single side band noise figure of the Gilbert-Cell mixer

with one shunt and two series inductors

6.2.2 Design Approach

Not only is flicker noise is the important issue in this design, but the linearity and power

consumption are of great concern as well. To save power, the size and bias of the driver stage did

not need to provide the maximum possible

fmax

. In this case, the current density of transistor

84CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.10:

Simulated conversion gain of the Gilbert-Cell mixer with one

shunt and two series inductors

is chosen to be

36 µA/µm

; this value provides a reasonable overall conversion gain of

4dB

when the switch transistors are optimized. The PMOS transistors of the DCBN

(Mb)

are sized so

that they provide most of the driver DC current. The switches are biased so that they work in the

deep saturation region yet have gate to source voltage that just above its threshold by 50 mV .

The total conversion gain of the mixer is proportional to the RF current gain of the switches

from the driver stage to the highly ohmic resistive load, the current gain (C)is as follows [22]:

C=2

πsin(π∆ fLO)

(π∆ fLO)(6.8)

Where

∆

is the time interval in which all the switches are on. The current gain

(C)

approaches

asymptotically the value of

2/π

when

∆

approaches zero; this situation can be achieved by biasing

the switches so that they are hardly conduct current, and by increasing their size. With these

biasing conditions, the switching transistors

(Ms)

conduct a very small amount of current of

150 µA

and the LO voltage swing of

50 mV

is enough to fully switch them off. Increasing the

size of the switches further shunts more RF current to the common sources parasitic capacitances

and causes the gain to drop. The optimum size of the switches is 35 µm as shown in fig. 6.11.

The bleeding transistors

(Mb)

should supply almost all of the DC current of the

transistors.

They are cross-coupled transistors which should be large enough; therefore, their parasitic

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 85

Figure 6.11:

Simulated conversion gain of the Gilbert-Cell mixer versus the

switching transistor size

capacitance is large compared to the Gilbert-Cell parasitic capacitances. These capacitances are

separated from the common sources of the switches by adding series inductors. The inductors

and the parasitic capacitances resonate at the RF input frequency to provide high ohmic path to

the ground in parallel to the sources of the switches. The size of

300 µm

while the inductor

value is 450 pH.

The linearity of the mixer can be improved by inserting degeneration inductors

(Ldeg)

whose

impedance is compared to

of the driver transistors

Ldeg

is also used for stability and

matching reasons. By using the degeneration inductors, the gain is reduced by

3dB

. Instead of

using tail current source, which consumes voltage headroom and limits the output voltage swing,

an inductor is used for rejecting common mode input. This makes it possible to apply single

ended input instead of differential with very small gain loss. Stacked inductors are used to save

area, nevertheless, they have lower quality factor and thus reduced gain.

6.2.3 Measurement Results

The circuit was fabricated in a

90 nm

CMOS process. It occupies an active area of

0.44×0.72 mm2

including probing pads. A micrograph of the Gilbert-Cell mixer with the proposed dynamic

current bleeding network is shown in fig. 6.12. To ease the on-chip measurement, an on chip

86CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.12:

Chip micrograph of the direct down-conversion mixer with the

proposed DCBN showing the pad configuration

Figure 6.13:

Measured conversion gain of the Gilbert-Cell mixer showing the

optimum LO input power

passive balun was also designed and connected to the LO port of the mixer; meanwhile, the RF

input of the mixer can be single ended driven because of the tail common mode rejection inductor

Ltail

. Two

50 GHz

signal generators are used to generate LO and RF signals. The differential

output is probed using a differential probe; one of single ended output is terminated to

50 Ω

while the other is connected to the input of the Agilent Spectrum Analyzer E4440A through

an off-chip coupling capacitor. The cable losses, probe losses, and the on-chip balun loss have

been de-embedded. LO power is swept so that the best conversion gain is achieved, maximum

gain corresponds to a LO power of

1dB

as shown in fig. 6.13, de-impeding the losses of the

integrated passive balun results in an optimum LO power of

−4dBm

. Best case conversion

gain is

5dB

at LO frequency of

20 GHz

and base band frequency of

500 MHz

, measured and

6.2. LOW-FLICKER NOISE DIRECT DOWN-CONVERSION MIXER 87

simulated conversion gain curves are shown in fig. 6.14. The

1dB

compression point is

−12 dB

as shown in fig. 6.15. With the 1.2Vpower supply; the core Gilbert-Cell mixer draws a current

3.6mA

. Tables 6.2, 6.3, and 6.4 summarize the performance of the direct down-conversion

Gilbert-Cell mixer with the proposed DCBN and compare it with the state of the art results.

Figure 6.14:

Measured and simulated conversion gain of the Gilbert-Cell down-

conversion mixer with the proposed DCBN

Figure 6.15:

The

1−dB

compression point of the Gilbert-Cell mixer with

degeneration inductors

88CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

This work [49] [50]

Technology 90 nm 0.18 µm 0.18 µm

Power Dissipation (mW)4.3 10.8 44.7

Conversion Gain (dB)5 15 28

Optimum LO Power (dBm)-4 N.A -5

P1dB(Input Referred) (dBm)-12 -14.2 -25

LO Frequency (GHz)18 ⇒23 5.1 5.1

Base-Band Frequency (GHz)0.01 ⇒2 0.1 N.A

Current Bleeding Network Dynamic, two Ind. Dynamic, one Ind. Dynamic, one Ind.

SSB Noise Figure (dB)12 (Simulated) 10.6 7.4

Chip Size (mm2)0.44 ×0.72 1.00 ×1.10 1.9 ×1.00

Table 6.2:

Summary and performance comparison of the direct down-

conversion mixer with the proposed DCBN

This work [22] [51]

Technology 90 nm 0.18 µm 0.18 µm

Power Dissipation (mW)4.3 7 14

Conversion Gain (dB)5 16.2 27

Optimum LO Power (dBm)-4 0 N.A

P1dB(Input Referred) (dBm)-12 -14 N.A

LO Frequency (GHz)18 ⇒23 5.2 4.8 ⇒5.7

Base-Band Frequency (GHz)0.01 ⇒2 N.A 0.01 ⇒0.2

Current Bleeding Network Dynamic, two Ind. Static, two Ind. Dynamic, one Ind.

SSB Noise Figure (dB)12 (Simulated) 9.8 7.8

Chip Size (mm2)0.44 ×0.72 0.78 ×0.98 1.11 ×1.30

Table 6.3:

Summary and performance comparison of the direct down-

conversion mixer with the proposed DCBN

6.3. SYSTEM INTEGRATION 89

This work [52] [53] [54]

Technology 90 nm 0.18 µm 90 nm 0.18 µm

Power Dissipation (mW)4.3 5.65 38 5.2

Conversion Gain (dB)5 8 10 18.7

Optimum LO Power (dBm)-4 N.A N.A N.A

P1dB(Input Referred) (dBm)-12 -8.4 -18.5 -17.6

LO Frequency (GHz)18 ⇒23 24 21.6 ⇒27 20 ⇒25

Base-Band Frequency (GHz)0.01 ⇒2 0.05 N.A N.A

Current Bleeding Network Dynamic, two Ind. Static, one Ind. – –

SSB Noise Figure (dB)12 (Simulated) 12.6 11 18

Chip Size (mm2)0.44 ×0.72 1.1 ×0.98 1.0 ×0.98 0.18

Table 6.4:

Summary and performance comparison of the direct down-

conversion mixer with the proposed DCBN

6.3 System Integration

Referring to the IQ-demodulator block diagram of fig. 6.1, the mixer should be driven by

enough LO power in order to maintain the highest possible conversion gain. As shown in fig.

6.13, the optimum LO power required by the down-conversion mixer is

−4dBm

taking into

account the losses of the balun, while the power delivered by the static frequency divider

(SFD)

is around

−16 dBm

. In reality, the LO port of the mixer requires voltage swing rather than

power level. According to simulations, the voltage swing of the SFD’s output is 20 mV, and the

required voltage swing that can fully switch the mixer’s switching transistors on and off is

50 mV

Therefore, an amplifier’s load inductor should resonate with the LO port capacitor of the mixer at

the required frequency of operation around

20 GHz

. The block diagram of the SFD and the circuit

diagram of both the latch and the output buffer are shown in fig. 6.16; the output buffers are used

mainly to isolate the output node of the latch from the high capacitance of the following stage,

and to provide matching to the 50 Ωload. When it is integrated in the IQ-Demodulator, there is

no need to design a matching network, its output node is directly fed into the gate terminals of the

LO buffer of the mixer. The LO buffer, along with the direct down-conversion mixer, is shown in

fig. 6.17. The load inductor of the LO buffer,

Lreso

, and the drain capacitance of the buffer from

one side, and the gate capacitances of the switches from the other side, constitute a parallel

90CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.16: Frequency divider and IQ generator for the LO signal

resonance network, whose impedance is maximum at around

20 GHz

, thus a maximum voltage

swing of 400 mV is produced.

As shown in fig. 6.18, the chip micrograph of the complete IQ-Demodulator, the LO buffers

are located very close to the LO port of the mixer. The

and

outputs of the SFD are routed

using transmission lines. The output of the IF amplifier is matched to

25 Ω

to equally divide its

output power between the two sides of the IQ-demodulator’s mixers.

The IQ-Demodulator has been fabricated in a TSMC

90 nm

CMOS process. It occupies

an active area of

0.84 ×1.15 mm2

including probing pads. A micrograph of the Demodulator

including inputs, outputs, and DC pads is shown in fig. 6.18. To ease the on-chip measurement,

an on chip passive balun was also designed and connected to the input of the SFD; meanwhile,

the IF input of the IF amplifier can be driven single ended because of the tail common mode

rejection inductor. Two

50 GHz

signal generators are used to generate LO and IF signals. The

differential output is probed using a differential probe; one of single ended outputs is terminated

50 Ω

while the other is connected to the input of the Agilent Spectrum Analyzer E4440A

through an off-chip coupling capacitor. While the other differential output is left unconnected.

The cable losses, probe losses, and the on-chip balun loss have been de-embedded. Best case

conversion gain is

6dB

at LO input frequency of

40 GHz

and base band frequency of

500 MHz

;

6.3. SYSTEM INTEGRATION 91

Figure 6.17: Down-conversion mixer with the LO buffer

Figure 6.18:

The chip micrograph of the IQ-Demodulator showing circuit

components and probing pads

92CHAPTER 6. DESIGN OF THE IQ DEMODULATOR FOR 60 GHZ WLAN TRANSCEIVER

Figure 6.19: Conversion gain of the IQ-Demodulator

This work [56] [55]

Technology CMOS 90 nm CMOS 90 nm CMOS 0.13 µm

Power Dissipation (mW)50 73 N.A

Conversion Gain (dB)6 20 -3

Minimum LO Power (dBm)-1 N.A 8

P1dB(Input Referred) (dBm)-12 N.A 6

LO Frequency (GHz)33.5 ⇒45.5 8.5 ⇒9.5 35 ⇒50

Base-Band Frequency (GHz)0.01 ⇒2.17 N.A 0.01

Chip Size (mm2)0.84 ×1.15 1.05 ×0.98 0.9 ×1

Table 6.5: Summary and performance comparison of the IQ-Demodulator

measured and simulated conversion gain curves are shown in fig. 6.19. There are two main

reasons causes differences between measured and simulated results: The first is the lossy stacked

inductors with relatively low

factors due to the existence of the fillers, these fillers are not

included in the EM model to save the simulation run time. The second is the insufficient power

delivered to the LO port of the mixer over the wide frequency range. The

1−dB

compression

point is

−12 dBm

with the

1.2V

power supply. The complete IQ-Demodulator consumes a total

power of

50 mW

including LO buffers and output matching buffers. The relatively low linearity

compared to [55] is due to minimum power requirements and the elevated conversion gain value.

Table 6.5 summarizes the IQ-demodulator performance and compares it to other state of the art

demodulators.

Chapter 7

Conclusions

In this dissertation, new techniques and optimized topologies are proposed to improve the

performance of mm-wave frequency generation and conversion circuits. Employing a transformer-

feedback topology, a

24 GHz

LC-VCO was designed using

130 nm

CMOS technology. It

achieves a tuning range of

3.7GHz

and consumes only

6mW

which is one fourth of the power

that the conventional LC-VCO needs. The designed transformer-feedback VCO has a phase

noise of

−92.5dBc/Hz

at an offset frequency of

1MHz

which is comparable to that of the

conventional counterpart. Two designs of frequency divider have also been presented. One design

is a Ka-band Miller frequency divider in

130 nm

CMOS technology. A specific optimization

approach was followed to achieve a maximum achievable locking range of

10 GHz

at a

10 mW

of power consumption and a maximum required input power of

−2dBm

. The other design,

on the other hand, is a

45 GHz

static frequency divider using

90 nm

CMOS technology. New

techniques such as inductive peaking, split resistors, and asymmetric sizing of input transistors

are employed to minimize the power consumption of the divider. The locking range of the static

divider is

12 GHz

, while it consumes only

6.9mW

and requires an input power of

0dBm

minimum. Moreover, direct down-conversion mixer is investigated and designed. The Flicker

noise effect is challenging. New current bleeding network was therefore proposed, not only the

noise performance is improved but the conversion gain is increased as well. The noise figure

of the mixer is

12 dB

, the conversion gain is

5dB

at a power consumption of

4.3mW

, and an

optimum LO signal power of −4dBm.

94 CHAPTER 7. CONCLUSIONS

Based on the designed static frequency divide, the direct down-conversion and up-conversion

mixers, complete IQ-modulator and demodulator were eventually demonstrated in

90 nm

CMOS

technology. It is targeted at a

60 GHz

super-heterodyne transceiver for high-data-rate commu-

nication application. Measurement results show that the IQ-demodulator operates with a LO

frequency of

33.5

45.5GHz

, it achieves a conversion gain of

6dB

at a power consumption of

only

50 mW

. This performance compares favorably with the state-of-the-art developments and

satisfies the 60 GHz high-data-rate communication application of the IEEE 802.15.3c standard.

7.1 Future Work

With the increased demand for higher data-rates for wireless LAN, and more complex modulation

schemes, front-ends need to attain more linear performance and flat gain over the whole band.

Using more advanced technologies, whose substrates have lower losses and transistor having

higher

and

fmax

, can help to design circuits with relaxed current requirements and higher

linearity.

Bibliography

[1]

T. Trinh, E. Benko, and W. Wong, “Ka-Band Microstrip Integrated Circuit FMCW

Transceiver,” in Microwave Symposium Digest, 1986 IEEE MTT-S International, June 1986,

pp. 639–642.

[2] G. S. N. Raju, Radar Engineering. I K International Publishing House, Oct. 2008.

[3]

M. Jankiraman, Design of Multi-Frequency CW Radars. Scitech Publishing Inc., Aug. 2007.

[4]

K. Yhland and C. Fager, “A FET Transceiver Suitable for FMCW Radars,” Microwave and

Guided Wave Letters, IEEE, vol. 10, no. 9, pp. 377–379, Sep 2000.

[5]

R. L. Miller, “Fractional-Frequency Generators Utilizing Regenerative Modulation,” Pro-

ceedings of the IRE, vol. 27, no. 7, pp. 446–457, 1939.

[6]

K. Koivunen and M. Jenu, “A 40-GHz MMIC Frequency Divider-by- Two,” in Microwave

Conference, 1997. 27th European, vol. 2, 1997, pp. 662–667.

[7]

A. Safarian and P. Heydari, “A Study of High-Frequency Regenerative Frequency Dividers,”

in Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on, 2005, pp.

2695–2698 Vol. 3.

[8]

R. Harrison, “Theory of Regenerative Frequency Dividers using Double-Balanced Mixers,”

in Microwave Symposium Digest, 1989., IEEE MTT-S International, 1989, pp. 459–462 vol.1.

[9]

Z. Heshmati, I. Hunter, and R. D. Pollard, “Microwave Parametric Frequency Dividers With

Conversion Gain,” Microwave Theory and Techniques, IEEE Transactions on, vol. 55, no. 10,

pp. 2059–2064, 2007.

96 BIBLIOGRAPHY

[10]

J. Lee and B. Razavi, “A 40-GHz frequency Divider in 0.18-um CMOS Technology,”

Solid-State Circuits, IEEE Journal of, vol. 39, no. 4, pp. 594–601, 2004.

[11]

K. Sengupta, T. Bhattacharyya, and H. Hashemi, “A Nonlinear Transient Analysis of Re-

generative Frequency Dividers,” Circuits and Systems I: Regular Papers, IEEE Transactions

on, vol. 54, no. 12, pp. 2646–2660, 2007.

[12]

K. Sengupta and H. Hashemi, “Maximum Frequency of Operation of CMOS Static Fre-

quency Dividers: Theory and Design techniques,” in Electronics, Circuits and Systems, 2006.

ICECS ’06. 13th IEEE International Conference on, 2006, pp. 584–587.

[13]

R. Shu, V. Subramanian, and G. Boeck, “A 8 to 1 static Frequency Divider Operating up to

34 GHz in 0.13-um CMOS technology,” in Microwave Workshop Series on Millimeter Wave

Integration Technologies (IMWS), 2011 IEEE MTT-S International, 2011, pp. 17–20.

[14]

L. Li, P. Reynaert, and M. Steyaert, “A 60GHz 15.7mW Static Frequency Divider in 90nm

CMOS,” in ESSCIRC, 2010 Proceedings of the, 2010, pp. 246–249.

[15]

J.-Y. Chang and S.-I. Liu, “A 4-54GHz Static Frequency Divider with Back-Gate Coupling,”

in VLSI Design, Automation and Test, 2007. VLSI-DAT 2007. International Symposium on,

2007, pp. 1–4.

[16]

B. Liang, D. Chen, B. Wang, D. Cheng, and T. Kwasniewski, “A 43-GHZ Static Frequency

Divider in 0.13 um Standard CMOS,” in Electrical and Computer Engineering, 2008. CCECE

2008. Canadian Conference on, 2008, pp. 000 111–000 114.

[17]

Y. Mo, E. Skafidas, R. Evans, and I. Mareels, “Analysis and Design of a 50-GHz 2:1

CMOS CML Static Frequency Divider Based on LC-tank,” in Microwave Integrated Circuit

Conference, 2008. EuMIC 2008. European, 2008, pp. 64–67.

[18]

B. Razavi, RF Microelectronics, ser. Prentice Hall Communications Engineer-

ing and Emerging Technologies. Prentice Hall PTR, 2011. [Online]. Available:

http://books.google.com/books?id= TccKQEACAAJ

BIBLIOGRAPHY 97

[19]

B. Razavi, Design of Analog CMOS Integrated Circuits, ser. McGraw-Hill higher

education. McGraw-Hill Education (India) Pvt Limited, 2002. [Online]. Available:

http://books.google.com/books?id=ZdXvVEtGrhQC

[20]

A. Jerng and C. Sodini, “The Impact of Device Type and Sizing on Phase Noise Mechanisms,”

Solid-State Circuits, IEEE Journal of, vol. 40, no. 2, pp. 360–369, 2005.

[21]

T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits. Second edition,

Press Syndicate of the University of Cambridge, pp. 419-424, 2004.

[22]

J. Park, C.-H. Lee, B.-S. Kim, and J. Laskar, “Design and Analysis of Low Flicker-Noise

CMOS Mixers for Direct-Conversion Receivers,” Microwave Theory and Techniques, IEEE

Transactions on, vol. 54, no. 12, pp. 4372–4380, 2006.

[23]

M. Terrovitis and R. Meyer, “Intermodulation Distortion in Current-Commutating CMOS

Mixers,” Solid-State Circuits, IEEE Journal of, vol. 35, no. 10, pp. 1461–1473, Oct 2000.

[24] IBM Microeletronics, “CMRF8SF Model Reference Guide,” August 20, 2009.

[25]

TSMC Taiwan Semiconductor Manufacturing Co., LTD, “TSMC 90NM/85NM CMOS

LOGIC/MS/RS and 80NM LOGIC/MS Design Rules,” July 25, 2010.

[26]

V. Subramanian, V.-H. Do, W. Keusgen, and G. Boeck, “60 GHz SiGe HBT Downconversion

Mixer,” in Microwave Integrated Circuit Conference, 2007. EuMIC 2007. European, 2007,

pp. 76–79.

[27]

T. Manku, “Microwave CMOS-Device Physics and Design,” Solid-State Circuits, IEEE

Journal of, vol. 34, no. 3, pp. 277–285, 1999.

[28]

C. Doan, S. Emami, A. Niknejad, and R. Brodersen, “Millimeter-Wave CMOS Design,”

Solid-State Circuits, IEEE Journal of, vol. 40, no. 1, pp. 144–155, 2005.

[29]

S. Chu, K. Chew, P. R. Verma, C. Ng, C. H. Cheng, N. Toledo, Y. K. Yoo, W. B. Loh, K. C.

Leong, S. Q. Zhang, B. G. Oon, Y. W. Poh, T. Zhou, K. K. Khu, and S. Lim, “Enabling Wire-

less Communications with state-of-the-art RF CMOS and SiGe BiCMOS Technologies,” in

98 BIBLIOGRAPHY

Radio-Frequency Integration Technology: Integrated Circuits for Wideband Communication

and Wireless Sensor Networks, 2005. Proceedings. 2005 IEEE International Workshop on,

2005, pp. 115–118.

[30]

M. Danesh, J. Long, R. Hadaway, and D. Harame, “A Q-Factor Enhancement Technique for

MMIC Inductors,” in Radio Frequency Integrated Circuits (RFIC) Symposium, 1998 IEEE,

1998, pp. 217–220.

[31]

F. Zhang, E. Skafidas, and W. Shieh, “A 60-GHz Double-Balanced Gilbert Cell Down-

Conversion Mixer on 130-nm CMOS,” in Radio Frequency Integrated Circuits (RFIC)

Symposium, 2007 IEEE, 2007, pp. 141–144.

[32]

A. Safarian and P. Heydari, “Design and Analysis of a Distributed Regenerative Frequency

Divider using Distributed Mixer,” in Circuits and Systems, 2004. ISCAS ’04. Proceedings of

the 2004 International Symposium on, vol. 1, 2004, pp. I–992–5 Vol.1.

[33]

G. Von Buren, C. Kromer, F. Ellinger, A. Huber, M. Schmatz, and H. Jackel, “A Combined

Dynamic and Static Frequency Divider for a 40GHz PLL in 80nm CMOS,” in Solid-State

Circuits Conference, 2006. ISSCC 2006. Digest of Technical Papers. IEEE International,

2006, pp. 2462–2471.

[34]

K.-C. Kwok and H. Luong, “Ultra-Low-Voltage High-Performance CMOS VCOs using

Transformer Feedback,” Solid-State Circuits, IEEE Journal of, vol. 40, no. 3, pp. 652–660,

2005.

[35]

J. Yang, C.-Y. Kim, D.-W. Kim, and S. Hong, “Design of a 24-GHz CMOS VCO With an

Asymmetric-Width Transformer,” Circuits and Systems II: Express Briefs, IEEE Transactions

on, vol. 57, no. 3, pp. 173–177, 2010.

[36]

E. Hegazi and A. Abidi, “Varactor Characteristics, Oscillator Tuning Curves, and AM-FM

Conversion,” Solid-State Circuits, IEEE Journal of, vol. 38, no. 6, pp. 1033–1039, 2003.

[37]

C. Cao and K. O, “Millimeter-Wave Voltage-Controlled Oscillators in 0.13-um CMOS

Technology,” Solid-State Circuits, IEEE Journal of, vol. 41, no. 6, pp. 1297–1304, 2006.

BIBLIOGRAPHY 99

[38]

C.-A. Lin, J.-L. Kuo, K.-Y. Lin, and H. Wang, “A 24 GHz Low Power VCO with Trans-

former Feedback,” in Radio Frequency Integrated Circuits Symposium, 2009. RFIC 2009.

IEEE, 2009, pp. 75–78.

[39]

S. Murad, R. Pokharel, M. Abdelghany, H. Kanaya, and K. Yoshida, “High linearity 5.2

GHz CMOS Up-Conversion Mixer using Derivative Superposition Method,” in TENCON

2010 - 2010 IEEE Region 10 Conference, Nov 2010, pp. 1509–1512.

[40]

I. Lai, Y. Kambayashi, and M. Fujishima, “50GHz Double-Balanced Up-Conversion Mixer

Using CMOS 90nm Process,” in Circuits and Systems, 2007. ISCAS 2007. IEEE International

Symposium on, May 2007, pp. 2542–2545.

[41]

A. Verma, K. O, and J. Lin, “A Low-Power Up-Conversion CMOS Mixer for 22-29-GHz

Ultra-Wideband Applications,” Microwave Theory and Techniques, IEEE Transactions on,

vol. 54, no. 8, pp. 3295–3300, Aug 2006.

[42]

I. Lai and M. Fujishima, “An Integrated 20-26 GHz CMOS Up-Conversion Mixer with Low

Power Consumption,” in Solid-State Circuits Conference, 2006. ESSCIRC 2006. Proceedings

of the 32nd European, Sept 2006, pp. 400–403.

[43]

F. Zhang, E. Skafidas, W. Shieh, B. Yang, B. Wicks, and Z. Liu, “A 60-GHz Double-

Balanced Mixer for Direct Up-Conversion Transmitter on 130-nm CMOS,” in Compound

Semiconductor Integrated Circuits Symposium, 2008. CSIC ’08. IEEE, Oct 2008, pp. 1–4.

[44]

M.-C. Chen, H.-S. Chen, T.-C. Yan, and C.-N. Kuo, “A CMOS Up-Conversion Mixer with

Wide IF Bandwidth for 60-GHz Applications,” in Silicon Monolithic Integrated Circuits in

RF Systems, 2009. SiRF ’09. IEEE Topical Meeting on, Jan 2009, pp. 1–4.

[45]

J.-H. Tsai and T.-W. Huang, “35-65 GHz CMOS Broadband Modulator and Demodulator

With Sub-Harmonic Pumping for MMW Wireless Gigabit Applications,” Microwave Theory

and Techniques, IEEE Transactions on, vol. 55, no. 10, pp. 2075–2085, Oct 2007.

100 BIBLIOGRAPHY

[46]

Y.-C. Tsai, J.-L. Kuo, J.-H. Tsai, K.-Y. Lin, and H. Wang, “A 50 - 70 GHz I/Q modulator

with Improved Sideband Suppression using HPF/LPF Based Quadrature Power Splitter,” in

Microwave Symposium Digest (MTT), 2011 IEEE MTT-S International, June 2011, pp. 1–4.

[47]

P.-H. Tsai, C.-C. Kuo, J.-L. Kuo, S. Aloui, and H. Wang, “A 30 - 65 GHz Reduced-Size

Modulator with Low LO Power using Sub-Harmonic Pumping in 90-nm CMOS Technology,”

in Radio Frequency Integrated Circuits Symposium (RFIC), 2012 IEEE, June 2012, pp.

491–494.

[48]

Y.-H. Lin, J.-L. Kuo, and H. Wang, “A 60-GHz Sub-Harmonic IQ Modulator and Demodu-

lator using Drain-Body Feedback Technique,” in Microwave Integrated Circuits Conference

(EuMIC), 2012 7th European, Oct 2012, pp. 365–368.

[49]

H. Kanaya, N. Koga, M. A. Abdelghany, R. Pokharel, and K. Yoshida, “Low Flicker-Noise

and Low Leakage Direct Conversion CMOS Mixer for 5GHz Application,” in Microwave

Conference, 2009. APMC 2009. Asia Pacific, 2009, pp. 1631–1634.

[50]

M. A. Abdelghany, A. I. A. Galal, R. Pokharel, H. Kanaya, and K. Yoshida, “A low

Flicker Noise Direct Conversion Receiver for the IEEE 802.11a Wireless LAN Standard,” in

Microwave Conference, 2009. APMC 2009. Asia Pacific, 2009, pp. 1647–1650.

[51]

A. I. A. Galal, M. A. Abdelghany, R. Pokharel, H. Kanaya, and K. Yoshida, “A Low Power

Low Flicker Noise Merged Balun LNA and Mixer for 5.2GHz Wireless LAN Receivers,” in

TENCON 2010 - 2010 IEEE Region 10 Conference, Nov 2010, pp. 1517–1520.

[52]

Y.-H. Chang, C.-Y. Huang, and Y.-C. Chiang, “A 24GHz Down-Conversion Mixer with

Low Noise and High Gain,” in Microwave Integrated Circuits Conference (EuMIC), 2012

7th European, Oct 2012, pp. 285–288.

[53]

K. Kjelgard and T. Lande, “A K-band UWB Receiver Front-End with Passive Mixer in 90

nm CMOS,” in Ultra-Wideband (ICUWB), 2013 IEEE International Conference on, Sept

2013, pp. 180–183.

BIBLIOGRAPHY 101

[54]

C.-H. Li, C.-N. Kuo, and M.-C. Kuo, “A 1.2-V 5.2-mW 20-30-GHz Wideband Receiver

Front-End in 0.18-um CMOS,” Microwave Theory and Techniques, IEEE Transactions on,

vol. 60, no. 11, pp. 3502–3512, Nov 2012.

[55]

C.-S. Lin, H.-Y. Chang, P.-S. Wu, K.-Y. Lin, and H. Wang, “A 35-50 GHz IQ-Demodulator

in 0.13-um CMOS Technology,” in Microwave Symposium, 2007. IEEE/MTT-S International,

June 2007, pp. 1397–1400.

[56]

K. Dufrene and R. Weigel, “Highly Linear IQ Downconverter for Reconfigurable Wireless

Receivers,” in Wireless Technology, 2005. The European Conference on, Oct 2005, pp. 19–22.

102 BIBLIOGRAPHY

List of Publications

[Paper A] Ali, M.K

, V. Subramanian, T. Zhang, and G. Boeck, “Design of Ka-Band Miller

Divider in 130 nm CMOS,” in Radio-Frequency Integration Technology (RFIT), 2011

IEEE International Symposium on, pp. 205–208, Nov. 2011.

[Paper B] Ali, M.K

. and Subramanian, V. and Tao Zhang and Boeck, G., “Performance Com-

parison of Differential K-Band VCOs in 130 nm CMOS Technology,” in Microwave

Conference (GeMiC), 2012 The 7th German, pp. 1–4, march 2012.

[Paper C] Ali, M.K

. and Hamidian, A. and Ran Shu and Malignaggi, A. and Boeck, G., “45

GHz Low Power Static Frequency Divider in 90 nm CMOS,” in Radio-Frequency

Integration Technology (RFIT), 2012 IEEE International Symposium on, pp. 65–67,

Nov. 2012.

[Paper D] Ali, M.K

. and Hamidian, A. and Malignaggi, A. and Boeck, G., “Utilizing Static

Frequency Divider for Quadrature Signal Generation in a 90 nm CMOS Technology,”

in Microwave Conference (GeMIC), Germany, vol., no., pp.1,4, 10-12 March 2014.

[Paper E] Ali, M.K

. and Hamidian, A. and Malignaggi, A. and Boeck, G., “IQ Demodulator for

the IEEE 802.15.3c Standard in the 90 nm CMOS Technology,” European Microwave

Conference (EuMC), 2014 44th, vol., no., pp.592,595, 6-9 Oct. 2014.

[Paper F] Ali, M.K

. and Hamidian, A. and Malignaggi, A. and Boeck, G., “Low Flicker Noise

High Linearity Direct Conversion Mixer for K-band Applications in a 90 nm CMOS

Technology,” Microwaves, Radar, and Wireless Communication (MIKON), 2014 20th

International Conference on, vol., no., pp. 1,4, June 2014.

[Paper G]

Malignaggi, A. and Hamidian, A. and Ran Shu and

Ali, M.K

and Boeck, G., “Ana-

lytical Study and Performance Comparison of mm-wave CMOS LNAs,” Microwave

Integrated Circuits Conference (EuMIC), European, pp. 260,263, Mar. 2013.

[Paper H]

Hamidian, A. and Malignaggi, A. and Shu, R. and

Ali, M.K.

and Boeck, G., “Multi-

gigabit 60 GHz OOK Front-End in 90 nm CMOS,” Silicon Monolithic Integrated

Circuits in RF Systems (SiRF), 2013 IEEE 13th Topical Meeting on, pp. 96,98, Jan.

2013.

103

104 LIST OF PUBLICATIONS

[Paper I]

Shu, R. and Subramanian, V. and Hamidian, A. and Malignaggi, A. and

Ali, M.K.

and Boeck, G., “A 36-49 GHz Injection-Locked Frequency Divider with Transformer-

Based Dual-Path Injection,” Silicon Monolithic Integrated Circuits in RF Systems

(SiRF), 2013 IEEE 13th Topical Meeting on, pp. 114,116, Jan. 2013.

[Paper J]

Hamidian, A. and Malignaggi, A. and Shu, R. and

Ali, M.K.

and Boeck, G., “A

Wideband Gilbert Cell Up-Converter in 90 nm CMOS for 60 GHz Application,” Radio-

Frequency Integration Technology (RFIT), 2012 IEEE International Symposium on, pp.

50,52, Nov. 2012.

[Paper K]

Shu, R. and Hamidian, A. and Malignaggi, A. and

Ali, M.K.

and Boeck, G., “A 40

GHz CMOS VCO with Resonated Negative-Conductance Cell,” Radio-Frequency

Integration Technology (RFIT), 2012 IEEE International Symposium on, pp. 77,79,

Nov. 2012.

[Paper L]

Malignaggi, A. and Hamidian, A. and Ran Shu and

Ali, M.K.

and Kravets, A. and

Boeck, G., “60 GHz LNAs in 90 nm CMOS Technology,” Signals, Systems, and

Electronics (ISSSE), 2012 International Symposium on, pp. 1,4, Oct. 2012.

[Paper M]

Tao Zhang and Subramanian, V. and

Ali, M.K.

and Boeck, G., “Integrated K-Band

CMOS Passive Mixers Utilizing Balun and Polyphase Filters,” Radio-Frequency

Integration Technology (RFIT), 2011 IEEE International Symposium on, pp. 89,92,

Nov. 2011.

[Paper N]

Hamidian, A. and Subramanian, V. and Doerner, R. and Shu, R. and Malignaggi, A.

and

Ali, M.K.

and Boeck, G., “60 GHz Power Amplifier Utilizing 90 nm CMOS Tech-

nology,” Radio-Frequency Integration Technology (RFIT), 2011 IEEE International

Symposium on, pp. 73,76, Nov. 2011.