TUNNEL DIODE/TRANSISTOR INTEGRATED CIRCUITS

Abstract

by

Qingmin Liu

As the minimum feature sizes in transistor technology are reached, circuit per-

formance may also saturate. For this reason, it is important to consider new and

extraordinary ways to extend the performance of circuits. Integrated tunnel diodes

enable a variety of design alternatives for signal processing, analog-to-digital con-

version, communications, and memory. It is the goal of this work to analyze and

explore the potential of tunnel diode/transistor (TDT) technology for increasing

speed and reducing power dissipation beyond what can be achieved with transistors

alone.

Circuit design requires accurate device models. In this work, a physics-based

small-signal equivalent circuit model for the resonant tunneling diode (RTD) has

been developed, which unifies previous models by Brown et al. for quantum in-

ductance and by Lake and Yang for quantum capacitance, and provides analytic

expressions for both the quantum inductance and quantum capacitance. Further,

two new TDT circuits: a TDT differential comparator and a TDT frequency trans-

lator have been invented.

The TDT differential comparator is of special interest for use in direct digital syn-

thesis applications. Circuit simulation shows a power dissipation of 3.5 mW/latch

at 100-GHz clock frequency with 60-dBc spur-free dynamic range (SFDR) can be
 Qingmin Liu

obtained in the TDT comparator. In comparison with the conventional transistor

approach, power is reduced by approximately 1.6x at the same speed and SFDR.

The TDT frequency translator is of special interest for use in communication

systems for upconverting digital signals. The circuit consists of a transistor, a

tunnel diode, and an inductor. The transistor provides input-output isolation and

power gain relative to prior art at the expense of the immunity to the input voltage

variation.

A scalable self-aligned contact process for fabrication of the TDT circuits has

been developed using InP-based RTD and double heterojunction bipolar transistor

(DHBT). This novel approach uses silicon nitride sidewalls and a benzocyclobutene

(BCB) etchback to form self-aligned emitter-base contacts. InP/InGaAs DHBTs

have been fabricated and the test results demonstrate the feasibility of this sidewall

and etchback process. AlAs/InGaAs/InAs RTDs were also fabricated and demon-

strated a peak current density of 1.8 mA/µm2 and a peak-to-valley current ratio of

1.8.
 To my dear parents

Debao Liu and Shumin Guo

To my dear wife

Ying Shang

ii
 CONTENTS

FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

CHAPTER 1: INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Resonant Tunneling Diode . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Prior Art - Tunnel Diode/Transistor Circuits . . . . . . . . . . . . . . 5
1.3 Accomplishments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

CHAPTER 2: DEVICE DESIGN AND MODELING . . . . . . . . . . . . . . 9

2.1 Resonant Tunneling Diode . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 RTD Heterostructure Design . . . . . . . . . . . . . . . . . . . 9
2.1.2 RTD DC Model . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.3 RTD AC Model . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.4 RTD Maximum Frequency of Oscillation . . . . . . . . . . . . 29
2.2 Heterojunction Bipolar Transistor . . . . . . . . . . . . . . . . . . . . 33
2.2.1 HBT Heterostructure Design . . . . . . . . . . . . . . . . . . . 33
2.2.2 HBT SPICE Model . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.3 HBT Cut-Off Frequency and Maximum Oscillation Frequency 41

CHAPTER 3: CIRCUIT DESIGN AND SIMULATION . . . . . . . . . . . . 44

3.1 Circuit Design Using Tunnel Diodes . . . . . . . . . . . . . . . . . . . 44
3.2 Differential Comparator . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 Frequency Translator . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

CHAPTER 4: SELF-ALIGNED NITRIDE SIDEWALL PROCESS . . . . . . 65

4.1 Process Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Nitride Sidewall Formation and BCB Etchback Process . . . . . . . . 71
4.3 DHBT Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4 RTD Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

iii
CHAPTER 5: SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . . . 84
5.1 Summary of Achievements . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2 Recommendations for Future Research . . . . . . . . . . . . . . . . . 87

APPENDIX A: FABRICATION PROCESS FLOW . . . . . . . . . . . . . . 91

APPENDIX B: ESTIMATION OF THE HETEROJUNCTION BIPOLAR

TRANSISTOR CUT-OFF FREQUENCY AND MAXIMUM OSCILLA-
TION FREQUENCY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

APPENDIX C: DESIGN APPROACH USING TUNNEL DIODES FOR

LOWERING POWER IN DIFFERENTIAL COMPARATORS . . . . . . 105

APPENDIX D: TUNNEL DIODE/TRANSISTOR DIFFERENTIAL COM-

PARATORS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

APPENDIX E: UNIFIED AC MODEL FOR THE RESONANT TUNNEL-

ING DIODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

APPENDIX F: LOW POWER, HIGH SPEED, AND MIXED-SIGNAL TUN-

NELING DEVICE TECHNOLOGY . . . . . . . . . . . . . . . . . . . . . 123

APPENDIX G: VERTICAL SILICON TUNNEL DIODE ON HIGH RESIS-

TIVITY SILICON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

APPENDIX H: UNIFIED PHYSICS-BASED AC MODEL FOR THE RES-

ONANT TUNNELING DIODE . . . . . . . . . . . . . . . . . . . . . . . . 129

APPENDIX I: PERFORMANCE-AUGMENTED CMOS USING BACK-

END UNIAXIAL STRAIN . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

APPENDIX J: SILICON-BASED TUNNEL DIODES AND INTEGRATED

CIRCUITS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

iv
 FIGURES

1.1 The measured current-voltage characteristic of an InP-based resonant

tunneling diode. The circuit symbol is defined in the inset. . . . . . . 2
1.2 Conduction band profiles of a resonant tunneling diode under differ-
ent biases: (a) equilibrium, (b) resonance, and (c) off-resonance. . . . 4

2.1 Comparison of measured (circles) and simulated (line) current-voltage

characteristic of an AlAs/InGaAs/AlAs RTD. The simulation is based
on Broekaert’s model [19]. . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Resonant tunneling diode small-signal equivalent circuit models: se-
ries inductance model of (a) Gering et al. [20], and parallel inductance
model of (b) Brown et al. [21]. Model (c) of Broekaert et al. [19]
and Lake and Yang [23] adds the quantum capacitance to the series
inductance model. Unified small signal model proposed in this work
by (d) Liu and Seabaugh [22]. . . . . . . . . . . . . . . . . . . . . . . 16
2.3 InP-based resonant tunneling diode computed energy band diagram
(computed using the Schrödinger-Poisson solver, BandProf, of W. R.
Frensley). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 The Fermi surface of the degenerate emitter electron gas. Tunneling
current is directed along the kZ direction. kF is the Fermi wave num-
ber in the emitter and kZ is the electron wave vector projected onto
the z-direction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Comparison of measured (circles) and simulated (line) S-parameter
at two selected biases. (a) 0.15 V (pre-peak region); (b) 0.45 V (near
the center of the NDR region). To avoid being indistinguishable be-
tween the measured and simulated curves, only 20% of the total 801
measured data points in each curve are shown. . . . . . . . . . . . . . 23
2.6 Comparison of the differential conductance extracted from the dc I-V
measurement (line) and S-parameter measurements (circles). . . . . . 24
2.7 Schematic band diagram of the RTD at (a) equilibrium, and (b) an
applied bias voltage of VA . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.8 The reciprocal of the quantum inductance, LQ , vs. bias showing good
agreement between extracted inductance from S-parameter measure-
ments (circles) and the unified ac model (line). . . . . . . . . . . . . . 26

v
2.9 The total capacitance, CP , vs. bias showing close agreement between
extracted capacitance from S-parameter measurements (circles) and
the unified ac model (line). . . . . . . . . . . . . . . . . . . . . . . . . 28
2.10 The calculated RTD maximum oscillation frequency, fmax , vs. the
negative differential resistance, RD , where RS , CP , and τ are assumed
to be constant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.11 The calculated RTD maximum oscillation frequency, fmax , vs. the
series resistance, RS , where RD , CP , and τ are assumed to be constant. 32
2.12 Calculated energy band diagram of InP/InGaAs double heterojunc-
tion bipolar transistor (computed using the Schrödinger-Poisson solver,
BandProf, of W. R. Frensley). . . . . . . . . . . . . . . . . . . . . . . 34
2.13 Calculated energy band diagram of InP/GaAsSb double heterojunc-
tion bipolar transistor (computed using the Schrödinger-Poisson solver,
BandProf, of W. R. Frensley). . . . . . . . . . . . . . . . . . . . . . . 35
2.14 The small-signal Gummel-Poon model for an InP HBT. . . . . . . . . 36
2.15 SPICE simulated Gummel plot and comparison with measured data.
The parameters used in the SPICE simulation are shown in Table 2.3. 38
2.16 SPICE simulated common-emitter characteristics. The parameters
used in the SPICE simulation are shown in Table 2.3. . . . . . . . . . 39
2.17 SPICE simulated small signal current gain and power gain frequency
dependence for a 1 µm2 device. The parameters used in the SPICE
simulation are shown in Table 2.3. . . . . . . . . . . . . . . . . . . . . 40
2.18 Schematic cross section of a heterojunction bipolar transistor. . . . . 43

3.1 Current-controlled clocked tunnel diode pair. . . . . . . . . . . . . . . 45

3.2 Load lines of the tunnel diode pair at two conditions: (a) ICT L is off;
and (b) ICT L is on. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Schematic of a conventional transistor-only differential comparator. . 48
3.4 Schematic of a tunnel diode/transistor differential comparator. . . . . 49
3.5 Simulated output waveforms for the two comparator circuits of (a)
tunnel diode/transistor comparator, and (b) conventional bipolar tran-
sistor comparator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Simulated output spectrum of the synthesized passband signal for (a)
tunnel diode/transistor and (b) transistor-only differential compara-
tor showing 60 dBc SFDR around 37.3 GHz. This simulation uses
high speed InP-based HBT and RTD models. . . . . . . . . . . . . . 54

vi
3.7 Layout for the RTD/HBT differential comparator. The circuit schematic
is shown in Fig. 3.4. Layout area is 1.8×1.1 mm2 . . . . . . . . . . . . 55
3.8 Schematic architectures of two types of receivers: (a) conventional
superheterodyne receiver [78] and (b) Cellonics receiver [41]. . . . . . 57
3.9 Cellonics tunnel diode frequency translation circuit diagram [79]. . . . 59
3.10 Notre Dame tunnel diode/transistor frequency translation circuit.
The circuit component and source values in the simulation are: L = 1
nH, VCC = 1 V, and the HBT and RTD areas are both 2 × 2 µm2 . . . 60
3.11 Simulated output waveform for the tunnel diode/transistor frequency
translation circuit. Circuit component values are given in the caption
of Fig. 3.10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.12 Simulated frequency and magnitude of the output signal vs. the input
voltage for (a) the Cellonics frequency translation circuit of Fig. 3.9,
and (b) the tunnel diode/transistor frequency translation circuit of
Fig. 3.10. The circuit component values in the simulation are: L = 1
nH, VCC = 1 V, and the RTD area is 2 × 2 µm2 . . . . . . . . . . . . . 62
3.13 Layout for the RTD/HBT frequency translator. The circuit schematic
is shown in Fig. 3.10. Layout area is 0.8×0.5 mm2 . . . . . . . . . . . 64

4.1 Scale drawings of self-aligned emitter-base contact formation using a

nitride sidewall spacer. These drawings describe: (a) the emitter met-
allization, (b) the emitter mesa etch, (c) the silicon nitride sidewall
formation, and (d) the base metal deposition. . . . . . . . . . . . . . 67
4.2 Scale drawings of self-aligned emitter-base contact formation using
a nitride sidewall spacer. These drawings describe: (a) the BCB
deposition, (b) the BCB etchback, (c) the tungsten RIE etch, and
(d) the removal of the BCB. . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Optical micrographs of a 4x4 µm2 emitter HBT. The graphs show
the step-by-step HBT emitter, base and collector contact formation. . 70
4.4 RIE etch characteristic of silicon nitride in CHF3 . . . . . . . . . . . . 72
4.5 SEM micrograph of the silicon nitride sidewall on a GaAs substrate.
Silicon nitride is RIE etched in CHF3 plasma. . . . . . . . . . . . . . 73
4.6 RIE etch characteristic of silicon nitride in SF6 and Ar. . . . . . . . . 74
4.7 SEM micrograph of the silicon nitride sidewall on a GaAs substrate.
Silicon nitride is RIE etched in SF6 /Ar plasma. . . . . . . . . . . . . 74
4.8 RIE etch characteristic of BCB in SF6 and O2 . . . . . . . . . . . . . . 75
4.9 SEM micrograph of the BCB etchback process on an AlGaAs/GaAs
HBT wafer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

vii
4.10 Measured Gummel plot of an InP/InGaAs double heterojunction
bipolar transistor with nitride sidewall process. . . . . . . . . . . . . . 78
4.11 Measured common-emitter IC −VCE characteristic of an InP/InGaAs
double heterojunction bipolar transistor with nitride sidewall process. 80
4.12 Resonant tunneling diode DC test structure, (a) top view, (b) side
view. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.13 Measured current-voltage characteristics of the resonant tunneling
diodes on wafers: (a) 060729A2BA02, (b) 060745A3AA01. . . . . . . 83

5.1 Scale drawings of a fully self-aligned HBT process. These drawings

describe the emitter metallization, the emitter mesa etch, the silicon
nitride sidewall formation, and the base metal deposition and etch. . 89
5.2 Scale drawings of a fully self-aligned HBT process. These drawings
describe the base and collector mesa etch, and the collector metal
deposition and etch. . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

viii
 ACKNOWLEDGMENTS

I would like to thank many people who have supported and helped me to finish

this thesis.

First, I wish to express my sincere gratitude to my advisor, Dr. Alan C.

Seabaugh, for his valuable suggestions and directions concerning my work and life

during these years. He gave me a lot of encouragement and support when I felt

discouraged in the research work. I am very grateful to him for giving me this

opportunity to be his student.

Second, I would like to thank Dr. Fay, Dr. Porod, Dr. Csurgay and Dr. Jena for

agreeing to be my thesis readers and be members of my thesis defense examination

board. I am also very grateful to them for giving me a lot of valuable advice on my

research during these years.

I gratefully acknowledge the partial financial support of the Office of Naval Re-

search and Raytheon Systems. Especially, I wish to thank Prem Chahal and Frank

Morris at Raytheon for providing RTDs. I would also like to thank Yung-Chung

Kao, Paul Pinsukanjana, Haijun Zhu, and Kevin Vargason, IntelliEPI Inc. for their

supporting material growth.

Finally, I am very grateful to my parents and my wife for their encouragement

and support. I really appreciate my friends and colleagues, Surajit Sutar, Yan Yan,

Jialin Zhao, Sajid Kabeer, Bin Wu, Shishir Rai, and Notre Dame colleagues, for

their help and for the unforgettable days we had.

ix
 CHAPTER 1

INTRODUCTION

Since 1957, the tunnel diode, also called the Esaki diode, has been widely in-

vestigated for high-speed circuit applications, due to its intrinsic high switching

speed, and its multi-valued current-voltage (I−V) characteristic featuring negative

differential resistance (NDR) [1, 2, 3]. The resonant tunneling diode (RTD), first

demonstrated in 1974 [4], also features a multi-valued I−V characteristic, shown

in Fig. 1.1, and exhibits a high switching speed due to its small capacitance and

high current density. However, the tunnel diode (TD), denoting both the Esaki

diode and RTD, has not found wide use in commercial electronics. Therefore, it is

important to consider the impediments for use of tunnel diodes and how these can

be overcome.

First, the tunnel diode is a two-terminal device. Circuits constructed using

tunnel diodes lack isolation between input and output, i.e. the signals can propagate

from the output back to the input. The TD-only circuits also suffer from low

gain and fan-out [5, 6]. These deficiencies can be corrected by adding transistors

to the circuit, thereby providing input-output isolation, gain and fan-out. This

solution, however, requires development of an integrated tunnel diode technology,

i.e. a process for integration of tunnel diodes with transistors.

In the 1960s and 70s, lack of integration was a big problem for tunnel diodes,

as integrated circuits replaced discrete circuits becoming the mainstream technol-

1
 0.15
I
P

Current (mA)
0.1 Circuit
Symbol
0.05
1x1 µm2
0
0 0.2 0.4 0.6 0.8 1 1.2
Voltage (V) 4a0301-R7556

Figure 1.1. The measured current-voltage characteristic of an InP-based resonant

tunneling diode. The circuit symbol is defined in the inset.

ogy in modern electronics [5]. Today, however, methods for integrating tunnel

diodes with transistors in compound semiconductor technologies are well known,

e.g. [7, 8, 9, 10, 11]. Silicon and SiGe tunnel diodes have also recently been devel-

oped using molecular beam epitaxy [12, 13] or rapid thermal and chemical vapor

deposition processes [14, 15], and integration methods for incorporation of tunnel

diodes with CMOS have been shown [16]. These integration methods provide a basis

for incorporating tunnel diodes with transistors to augment circuits performance.

Development of tunnel diode circuits has also been impeded by the lack of a

representative SPICE model in commercial computer-aided design (CAD) softwares.

A piecewise-linear approximation for the “N-shaped” I−V characteristic has been

used for a long time, e.g. [17]. Recently, Schulman [18] and Broekaert [19] have

derived a physics-based I−V model to describe the “N-shaped” characteristic of the

RTD. For the ac model of the RTD, several different equivalent circuits have been

2
proposed, e.g. [20, 21]. In the course of this work, an analytic model has been found

which yields the small-signal equivalent circuit model directly for the RTD [22], and

which unifies previous models by Brown [21] for quantum inductance and by Lake

and Yang [23] for quantum capacitance.

Nowadays commercial technologies such as CMOS depend largely on device scal-

ing for improving circuit performance. As physical and technological limits deter-

mining minimum feature sizes are achieved, system improvement will depend more

and more on innovations in circuit design. It is important to consider augmenta-

tion technologies to enable new circuit explorations, e.g. in signal processing [24],

analog-to-digital conversion (ADC) [19], memory [25], and logic [11].

The remainder of this introductory chapter is organized as follows. Section 1.1

reviews the fundamentals of the resonant tunneling diode. Section 1.2 reviews the

prior art in tunnel diode/transistor (TDT) circuits. Finally, the accomplishments

of this research are summarized in Section 1.3.

1.1 Resonant Tunneling Diode

The RTD structure consists of a thin quantum well sandwiched between two

thin barriers and degenerately-doped emitter and collector regions as shown in Fig.

1.2, where E0 represents the quantized energy state in the quantum well. The

emitter and collector are degenerately doped so that the Fermi levels lie above the

conduction band edge in these two regions.

At equilibrium, the quantized energy state lies above the Fermi level in the emit-

ter, Fig. 1.2(a), the net tunneling current is zero. As the applied positive bias at

the collector lowers the quantized state in the quantum well until it aligns with the

occupied electron states in the emitter, resonant tunneling occurs, Fig. 1.2(b). Elec-

3
 Emitter Quantum Collector
Well

EF E0 EF

(a)

EF
E
0
EF qV

(b)
E
F

E
0
qV

E
F

(c)

Figure 1.2. Conduction band profiles of a resonant tunneling diode under different
biases: (a) equilibrium, (b) resonance, and (c) off-resonance.

trons in the emitter, whose longitudinal energy aligns with the quantized state, can

tunnel through this quantized state into the collector with a tunneling probability

approaching 1. The selection conditions for resonant tunneling, which depend in

detail on the energy-momentum relations between the emitter and quantum well,

have been worked out by Schulman [26]. The description given here is sufficient for

first-order understanding. As the number of these electrons available for tunneling

reaches a maximum, the tunneling current will reach a peak, and the corresponding

bias voltage is referred to as the peak voltage. As the bias exceeds the peak voltage,

4
the number of electrons which can tunnel into the quantum well decreases, resulting

in a decreased tunneling current which gives a negative differential resistance region

in the I−V characteristic. As the applied bias increases further, the quantized state

drops below the conduction band edge in the emitter and the resonant tunneling

path is broken, Fig. 1.2(c). Consequently, the current reaches a minimum value,

IV , which is determined by phonon-assisted tunneling and thermionic processes, and

the corresponding bias voltage is called the valley voltage. As the bias exceeds the

valley voltage, the current increases again. This current is comprised of at least

two components. One is the thermionic emission current since the bias lowers the

barriers as a result of potential drop in the emitter accumulation region. The other

excess current component is due to resonant tunneling through the second quantized

state in the quantum well.

The above discussion provides a physical picture of the tunneling process giving

rise to the multi-valued I−V characteristic and negative differential resistance of

the RTD. A detailed discussion of the RTD structures, dc and ac models, and high

frequency performance is given in Chapter 2.

1.2 Prior Art - Tunnel Diode/Transistor Circuits

Tunnel diode/transistor integrated circuits have been widely investigated due

to their simple architectures, high speed, and low power dissipation. TDT circuits

have been demonstrated for analog-to-digital conversion [19, 27], memory [7, 25],

logic (gates and flip-flops) [9, 11, 28], oscillators [29, 30], and chaos circuits [31, 32].

For analog-to-digital conversion, Broekaert et al. [19] first demonstrated a 4-bit

2-Gsps monolithic flash analog-to-digital converter, where RTDs were integrated

with heterostructure field-effect transistors (HFETs) and used in a flash circuit

5
architecture. More recently, a 3-bit 5-Gsps analog-to-digital converter was reported

also using RTDs and HFETs [27]. These converters show promise for extending the

ADC to X-band applications, but their use will depend on development of foundry

processes for TDT circuits.

In memory, an RTD/HFET low-standby-power static random-access memory

(SRAM) cell based on InP technology was demonstrated by van der Wagt et al.,

and it saved the power by 200x over the best previous GaAs-based memory cell

[25, 33]. This cell consisted of one HFET and two RTDs. In silicon technology, the

CMOS/TD SRAM, consisting of one transistor, two tunnel diodes and one stacked-

capacitor, can reduce the chip area by 2x and standby power by 7x over conventional

6T SRAM [24].

In logic, a monostable-bistable transition circuit topology [34] has been explored

for high-speed logic gates and flip-flops. This circuit topology consists of two series-

connected tunnel diodes monolithically-integrated with a parallel field-effect tran-

sistor [9, 16] or a bipolar transistor [11, 28]. Both 35 GHz single-ended [9] and 20

GHz differential [11] flip-flops have been demonstrated.

When the tunnel diode is biased in the NDR region, the circuit will oscillate. This

oscillation can be utilized to implement oscillators. A 5.8 GHz oscillator using a RTD

integrated with a heterojunction bipolar transistor (HBT) and a bond wire inductor

has been demonstrated with an output power of 3.3 dBm and a power efficiency of

8.84% [29]. Recently, a varactor-tuned voltage-controlled oscillator (VCO) using InP

RTD/HBT has been demonstrated with a center oscillation frequency of 5 GHz, a

tuning range of 710 MHz, a phase noise of -115 dBc/Hz with 1-MHz offset, and power

dissipation of 7.6 mW [30]. Compared to the conventional CMOS VCOs [35, 36],

this RTD/HBT VCO achieves comparable specifications with less components in

the circuit.

6
 The strong nonlinearity of the tunnel diode has been investigated for chaos gen-

eration, a phenomenon which is being explored for communication and information

processing [37]. A simple chaos generation circuit, which consists of a tunnel diode,

an inductor, and a capacitor, has been proposed [37]. The output of this circuit

can be periodical or chaotic, which is determined by the circuit parameters. As the

output signal is designed to be periodical, the circuit can be used as a frequency

divider. Recently, Kawano et al. [31] has demonstrated an 88-GHz 2:1 RTD/HEMT

(high electron mobility transistor) frequency divider with a single sideband phase

noise less than -100 dBc/Hz at 1-MHz offset frequency, which is comparable to con-

ventional frequency dividers [38, 39].

1.3 Accomplishments

The primary purpose of this thesis is to explore, through design, analysis, and

fabrication, InP-based RTD and HBT high speed and low power integrated circuits.

At the initial phase of this research, the equivalent circuit model of the resonant

tunneling diode was examined. InP based RTDs, provided by Raytheon Systems,

were measured using an Agilent 8510XF network analyzer. Previously published

models [19, 20, 21, 23] could not fit the measured S-parameters through the entire

frequency range (from 45-MHz to 30-GHz) and bias range (from 0 to post valley).

To resolve deficiencies in the prior models, a new unified small-signal equivalent

circuit model for the RTD was derived, and good agreement between this model

and measurement was obtained and published [22]. A detailed discussion of this

model is given in Chapter 2 and Appendix E.

In this research two new circuits which exploit the high speed and low power po-

tential of resonant tunneling diodes have been conceived, designed and simulated: a

7
TDT differential comparator and a TDT frequency translator. The TDT differential

comparator circuit allows an increase in circuit speed while simultaneously lowering

power compared to a conventional transistor-only circuit [40]. The TDT frequency

translation circuit also lowers power relative to conventional circuit approach [41].

The design and simulation of these two new circuits are given in Chapter 3.

To fabricate the TDT circuits, a new self-aligned contact process using silicon

nitride sidewalls and a benzocyclobutene (BCB) etchback has been developed. The

emitter-base separation is set by the nitride sidewall thickness allowing scaling to

reduce access resistance. The nitride sidewall is formed by depositing a plasma-

enhanced chemical vapor deposition (PECVD) Si3 N4 film, followed by an anisotropic

reactive ion etching (RIE). The nitride RIE etching conditions have been investi-

gated and the formation of nitride sidewalls has been verified by scanning electron

microscopy (SEM). Device electric test results have demonstrated the feasibility of

this fabrication process. The detailed discussion of the fabrication process and the

device test results are given in Chapter 4 and Appendix A.

The remainder of this thesis is organized as follows. Chapter 2 discusses the RTD

and HBT structures design and SPICE models. Chapter 3 discusses circuit design

using tunnel diodes and two new TDT circuits: a TDT differential comparator

and a frequency translator. Chapter 4 discusses the fabrication process used to

construct the circuit. Device results for fabrication of InP RTDs and HBTs are

described. Chapter 5 summarizes the accomplishments and proposes future work.

The RTD/HBT process traveler is given in Appendix A. The equations for the HBT

performance estimates are attached as Appendix B. Published work is included as

Appendices C to J.

8
 CHAPTER 2

DEVICE DESIGN AND MODELING

InP-based RTDs have been selected for this work because of their demonstrated

high peak current densities and high peak-to-valley current ratios [43, 44, 45]. Two

types of InP-based HBTs: InP/GaAsSb HBTs and InP/InGaAs HBTs are also uti-

lized as they have demonstrated excellent high frequency performance with measured

cut-off frequency and maximum oscillation frequency over 300 GHz [46, 47, 48, 49].

This chapter summarizes the RTD and HBT heterostructure design and models

used for circuit simulation. In addition, the RTD maximum frequency of oscillation

is revisited in light of the new understanding of the RTD equivalent circuit obtained

in this work, and its dependence on the RTD negative differential resistance and

series resistance is also shown.

2.1 Resonant Tunneling Diode

2.1.1 RTD Heterostructure Design

The slew rate of a tunnel diode can be evaluated by [50]

dV IP 1
= (1 − ), (2.1)
dt C PV R

where IP is the peak current, C is the junction capacitance, IP /C is the speed

index, and P V R is the peak-to-valley current ratio. To obtain high switching speed,

9
it is necessary to maximize the peak current and peak-to-valley current ratio and

minimize the capacitance.

Since the first InP-based InAlAs/InGaAs/InAlAs RTD was reported by Inata

et al. [51], much research has been done to improve the peak current density and

peak-to-valley current ratio. By replacing the InAlAs barrier with a strained AlAs

barrier, Inata et al. [52] later demonstrated a peak-to-valley current ratio as high

as 14 with a peak current density of 0.23 mA/µm2 at room temperature. The

AlAs/InGaAs RTD has demonstrated the highest peak current density of any RTD,

6.8 mA/µm2 with the peak-to-valley current ratio of 2.5 [43]. Broekaert et al. [54]

incorporated InAs in the quantum well structure to lower the quantized energy state

in the well with respect to the emitter conduction band edge, thereby lowering the

RTD turn-on voltage. The AlAs/InGaAs/InAs RTDs have also demonstrated high

peak-to-valley current ratio of 50 [45] and peak current density of 4.5 mA/µm2 [44].

Another fully lattice-matched heterostructure is the AlAsSb/InGaAs barrier-well

structure. The AlAsSb/InGaAs RTD demonstrated a peak-to-valley current ratio

of 11 with a peak current density of 0.15 mA/µm2 and showed potential for further

development [53]. For this research, the AlAs/InGaAs/InAs RTD has been selected

because of its record current density and low turn-on voltage. The layer diagram

for the heterostructure is given in Table 2.1.

10
 TABLE 2.1

STRUCTURAL PARAMETERS FOR AN InP-BASED RESONANT

TUNNELING DIODE. TYPICAL UNINTENTIONAL DOPING (UID) IN
THESE MATERIAL SYSTEMS IS N-TYPE WITH A DENSITY ≤ 1016 cm−3 .
THE In MOLE FRACTION IS 53% WHICH IS LATTICE-MATCHED TO InP

Material Thickness (nm) Dopant Density (cm−3 ) Function

InGaAs 40 Si 3.0E19 emitter
InGaAs 5 Si 1.0E18 emitter
InGaAs 2 UID spacer
AlAs 1.6 UID barrier
InGaAs 1.5 UID well
InAs 2.1 UID well/notch
InGaAs 1.5 UID well
AlAs 1.6 UID barrier
InGaAs 2 UID spacer
InGaAs 5 Si 1.0E18 collector
InGaAs 60 Si 3.0E19 collector

11
2.1.2 RTD DC Model

The tunneling current of the RTD is given by the Tsu-Esaki formula [56],
Z ∞ µ ¶
qm∗ kT 1 + e(EF −EZ )/kT
J= dEZ Θ(EZ , V ) ln , (2.2)
2π 2 ~3 0 1 + e(EF −EZ −eV )/kT

where q is the fundamental charge, m∗ is the effective mass, k is Boltzmann’s con-

stant, T is the temperature, ~ is Plank’s constant over 2π, EZ is the longitudinal

energy in the direction of current flow, V is the bias voltage, Θ(EZ , V ) is the trans-

mission coefficient, and EF is the Fermi energy of the emitter. The transmission

coefficient can be approximated by a Lorentzian function [18],

(Γ/2)2
Θ(EZ , V ) = , (2.3)
[EZ − (E0 − eV /2)]2 + (Γ/2)2

where Γ is the resonance width. The transmission coefficient is negligible except

when EZ is close to E0 − eV /2. Therefore, we can replace EZ by E0 − eV /2 in the

logarithmic term in Eq. 2.2, leading to the tunneling current given by the Schulman

formula [18],
· ¸· µ ¶¸
qm∗ kT Γ 1 + e(EF −E0 +eV /2)/kT π E0 − eV /2
J= ln + arctan . (2.4)
4π 2 ~3 1 + e(EF −E0 −ev/2)/kT 2 Γ/2

Equation 2.4 gives the peak current and the NDR, but does not account for the post-

valley current, which is due to tunneling through higher energy states in the well and

the barrier-controlled diode current. Broekaert [19] has provided additional analytic

expressions to describe the physics of this post-resonance I−V characteristic. In the

Broekaert model, the total current is expressed by,

IRT D (V ) = Ires1 (V ) + Ires2 (V ) + Ilkg (V ) (2.5)

where Ires1 and Ires2 are the tunneling current through the first and second quantized

states in the quantum well, Ilkg is the thermionic leakage current. The resonant

12
tunneling current for a symmetric RTD is expressed by,

Ires (V ) = IE (V ) − IE (−V ) (2.6)

with
· µ ¶¸
AJP 2 VN − V
IE (V ) = 1 + arctan
2f π Γ
· µ ¶¸
nkT V − VT
× ln 1 + exp (2.7)
VN − VT nkT

and s
2Γ
f =1− , (2.8)
π(VN − VT )
where A is the device area, JP is the approximate peak current density, VN is the

voltage at which the maximum negative differential resistance occurs, VT is the

resonance turn-on voltage, Γ is the full-width at half maximum of the resonance, n

is a “lever” factor accounting for the voltage drop across the quantum well relative

to the applied bias, and kT is the thermal voltage. The thermal leakage current is

given by, ³ ´
V
sinh nV kT
Ilkg (V ) = AJV ³ ´. (2.9)
VV
sinh nV kT

Based on Broekaert’s model, the RTD dc model parameters can be extracted by

fitting the measured I-V data to Eq. 2.5, as shown in Fig. 2.1. A close agreement

is typically obtained between the measured data and the fits based on Broekaert’s

model. Extracted parameters are listed in Table. 2.2 for the fit of Fig. 2.1; this is

the RTD model used for the TDT circuit simulations, described in Chapter 3. This

model is for an RTD fabricated by Raytheon Systems with a peak current density

of 1.1 mA/µm2 and peak-to-valley current ratio of 2.1.

13
 6
Measured InP-based RTD
5 Simulated (1.6) µm
2 2

4
Current (mA)

0
0 0.2 0.4 0.6 0.8 1
Voltage (V)

Figure 2.1. Comparison of measured (circles) and simulated (line) current-

voltage characteristic of an AlAs/InGaAs/AlAs RTD. The simulation is based on
Broekaert’s model [19].

14
 TABLE 2.2

InP RESONANT TUNNELING DIODE SPICE DC MODEL PARAMETERS

PARAM. VALUE UNIT DESCRIPTION

JP 1 1.255 mA/µm2 Peak current density (Resonance 1)
JP 2 4.170 mA/µm2 Peak current density (Resonance 2)
VN 1 0.3653 V Voltage of maximum NDR (Resonance 1)
VN 2 1.505 V Voltage of maximum NDR (Resonance 2)
VT 1 -0.03286 V Resonant turn-on voltage (Resonance 1)
VT 2 0.6562 V Resonant turn-on voltage (Resonance 2)
Γ1 0.09220 V Full-width at half maximum
of the resonance (Resonance 1)
Γ2 0.1871 V Full-width at half maximum
of the resonance (Resonance 2)
n1 1.278 Resonance lever factor (Resonance 1)
n2 1.472 Resonance lever factor (Resonance 2)
JV 0.4510 mA/µm2 Thermionic leakage current at VV
nV 1.289 Thermionic leakage current ideality factor
VV 1.000 V Valley voltage

2.1.3 RTD AC Model

The prior art in small-signal equivalent circuits for the RTD can be classified

into two categories: a series-inductance model [20] and a parallel-inductance model

[21], see Fig. 2.2, where RS represents the contact resistance plus the semiconduc-

tor sheet resistance, GD represents the differential conductance (first derivative of

the dc I-V curve), L represents the series inductance, LQ represents the quantum

inductance, and CQ and C0 represent the quantum capacitance and the geometrical

depletion capacitance, respectively.

15
 GD LQ GD

RS L RS
C0 C0

Gering Brown
(a) (b)

GD LQ GD

RS L RS
C0+CQ C0+CQ

Broekaert, Lake Liu and Seabaugh

(c) (d)

Figure 2.2. Resonant tunneling diode small-signal equivalent circuit models: series
inductance model of (a) Gering et al. [20], and parallel inductance model of (b)
Brown et al. [21]. Model (c) of Broekaert et al. [19] and Lake and Yang [23] adds
the quantum capacitance to the series inductance model. Unified small signal model
proposed in this work by (d) Liu and Seabaugh [22].

In the series-inductance model, Fig. 2.2(a) and (c), a bias-independent induc-

tance, which can be attributed to the wiring [57] appears in series with the para-

sitic resistance and the parallel combination of the tunneling conductance and the

junction capacitance. In the parallel-inductance model, the inductance, which is

attributed to the charging and discharging of the quantum well, appears in series

with the tunneling conductance which, in total, is in parallel with the junction

capacitance.

The capacitance in the equivalent circuit models should include both the geo-

metrical capacitance of the device and the quantum capacitance, which is due to

the density of electrons stored in the quantum well changing as a function of bias.

16
However, in prior work, the quantum capacitance is either neglected [21, 59], or com-

puted numerically [58]. Broekaert [19] gives a general expression for the geometrical

capacitance of the depletion layer and adds the quantum capacitance. However, in

his expression for the quantum capacitance, the sign is wrong. Recently, Lake and

Young [23] derived a simple analytic expression for the quantum capacitance and

corrected the sign. In their expression, the quantum capacitance exists only in the

NDR region and the quantum inductance is not considered.

The unified ac model for the RTD determined in the course of this work was

published [22]. This model provides the form of the small-signal equivalent cir-

cuit of the RTD, shown in Fig. 2.2(d), and leads to analytic expressions for both

the quantum inductance and quantum capacitance. Measurements of both dc I-V

characteristic and microwave frequency S-parameters of AlAs/InGaAs/InAs RTDs

support the validity of this equivalent circuit.

The unified ac RTD model is readily described using the band diagram of an

AlAs/InGaAs/InAs RTD under the bias voltage of V in Fig. 2.3. In this figure, JE

represents the emitter-to-well tunneling current density and JC represents the well-

to-collector tunneling current density. The change in collector tunneling current in

response to a change in the quantum well charge can be expressed as

∆JC = ∆(νC QW ) ≈ νC ∆QW (2.10)

where ∆QW is the change in stored charge in the quantum well and νC is the

electron escape rate (s−1 ) from the quantum well to the collector. Assuming νC is

weakly dependent on bias, ∆(νC QW ) is approximately equal to νC ∆QW for small

variations in applied bias. The tunneling current due to electrons tunneling back

from the collector to the quantum well is considered to be negligible.

17
 2
Emitter Quantum Collector
1.5 Well
Energy (eV) 1

0.5 J J
E C
E
F
0 E
0 E − qV
F
-0.5
InGaAs

InGaAs
InGaAs

InGaAs
AlAs

AlAs
InAs
-1

-1.5
0 5 10 15 20
Position (nm)

Figure 2.3. InP-based resonant tunneling diode computed energy band diagram
(computed using the Schrödinger-Poisson solver, BandProf, of W. R. Frensley).

Similarly, the emitter current change resulting from a small bias variation is

given by

∆JE = ∆(ν0 QE ) − ∆(νE QW ) ≈ ν0 ∆QE − νE ∆QW (2.11)

where QE is the change in the available tunneling charges in the emitter, ν0 is the

electron escape rate (s−1 ) from the emitter to the quantum well, and νE is the

electron escape rate (s−1 ) from the quantum well to the emitter. As in Eq. 2.10,

ν0 and νE are assumed to be weakly dependent on the bias, therefore ∆ν0 and ∆νE

are negligible.

18
 The difference between JE and JC then describe the change in charge stored in

the quantum well,

d(∆QW )
= ∆JE − ∆JC . (2.12)
dt
Solving Eq. 2.10−2.12 yields

∆QW = ν0 ∆QE τ (1 − e−t/τ ) (2.13)

where at t = 0 we assume ∆QW = 0, and τ is the electron lifetime in the quantum

well, which is defined as (νE + νC )−1 .

The electrons in the emitter that can tunnel into the quantum well through the

resonance state are those electrons whose longitudinal energy, EZ = ~2 kZ2 /2m∗ , are

resonant with the quantum-well ground-state energy. This corresponds with the

electrons indicated by the shaded disk in the E-k diagram of Fig. 2.4. The number

of these electrons is proportional to the shaded area of the disks, and to first order,

the area change is proportional to the bias voltage change. Thus, ∆QE can be

expressed as

∆QE = α∆V, (2.14)

where α is a proportionality factor. Substituting Eq. 2.14 into Eq. 2.13 yields

∆QW = αν0 τ (1 − e−t/τ )∆V. (2.15)

The change in the tunneling current density can then be written as

∆JC = αν0 νC τ (1 − e−t/τ )∆V (2.16)

from which follows the differential conductance,

∆JC
GD = A (t → ∞) = −Aqαν0 νC τ (2.17)
∆V

where A is the device area. Thus, Eq. 2.16 can be rewritten as

GD
∆JC = (1 − e−t/τ )∆V. (2.18)
A

19
 2
kX ∆kZ
∆(k Z ) = q∆V
2
*
2m
∆QE ∝ π ∆(k Z ) ∝ ∆V
2
kF

kZ

kY

Figure 2.4. The Fermi surface of the degenerate emitter electron gas. Tunneling
current is directed along the kZ direction. kF is the Fermi wave number in the
emitter and kZ is the electron wave vector projected onto the z-direction.

Eq. 2.18 can be transformed into the frequency domain to reveal the impedance

of the RTD. The Laplace transform of Eq. 2.18 is given by

GD ∆V
∆JC (s) = , (2.19)
A s(1 + sτ )

where ∆V (s) = ∆V /s. Thus, the impedance of the conduction current path is given

by
∆V (s) 1 τ
ZC (s) = = +s . (2.20)
A∆J(s) GD GD
From Eq. 2.20 the form of the equivalent circuit is apparently a differential conduc-

tance, GD , in series with a quantum inductance,

τ
LQ = . (2.21)
GD

Since the differential conductance is a function of the bias voltage, the quantum

inductance should be bias dependent.

20
 The electrons stored in the quantum well image a positive charge in the collec-

tor. This mirrored charge changes with bias, resulting in a quantum capacitance.

Combining Eq. 2.15 and 2.17, the change in the image charge in the collector, ∆QC ,

is given by
GD
∆QC = −∆QW = − ∆V. (2.22)
AνC
Therefore the corresponding quantum capacitance, CQ , is given by

∆QC GD
CQ = A =− . (2.23)
∆V νC

The total capacitance, thus, can be expressed as

²A GD
CP = C0 + CQ = − (2.24)
LW + 2LB + LD νC

where C0 represents the geometrical capacitance, ² is the dielectric constant, and

LW , LB , and LD are the widths of the quantum well, barrier and depletion region,

respectively.

To verify this proposed equivalent circuit model, both dc I-V and microwave

frequency S-parameter measurements were made on AlAs/InGaAs/InAs RTDs with

dimensions as given in Fig. 2.3. The S-parameter measurements were made on

a coplanar microwave structure using 100 µm pitch ground-signal-ground probes

connected to a Raytheon 1.6 µm × 1.6 µm RTD in parallel with an integrated 25

Ω thin film resistor. The resistor was incorporated to suppress oscillations in the

RTD for biases in the NDR region. This RTD is the same as that shown in Fig.

2.1, which has the current through the parallel resistor removed by subtraction. In

this configuration, oscillations are circumvented, resulting in an accurate equivalent

circuit extraction through the NDR region and over the entire bias range.

S-parameters were measured from 45 MHz to 30 GHz, using an Agilent 8510XF

vector network analyzer with a port power of -33 dBm (corresponding to a maximum

21
peak-to-peak voltage at the device that is less than 30 mV). An Agilent 4155B semi-

conductor parameter analyzer was connected through the network analyzer’s bias

tee to provide the dc bias from 0 V to 0.81 V. A two-port LRM (line-reflect-match)

calibration was performed using a Cascade Microtech 104-783A W-band impedance

standard substrate (ISS). A Cascade Microtech 116-344 absorber was used to reduce

undesired mode content in the transmission lines and minimize external reflection.

As an example, Fig. 2.5 shows the measured S-parameter S11 and simulated S11

based on the proposed equivalent circuit at the bias V = 0.15 V (pre-peak region),

and 0.45 V (near the center of the NDR region). The simulation is in close agreement

with the measurement throughout the entire bias range.

22
 -0.4 0.06

0.05
-0.42
0.04
-0.44
Real S11

Imag S11
0.03

0.02
-0.46
0.01
-0.48
0
(a)
-0.5 -0.01
0 5 10 15 20 25 30
Frequency (GHz)

-0.2 0.02

0.01
-0.22
0
-0.24
Real S11

Imag S11
-0.01
-0.26
-0.02

-0.28 -0.03
(b)
-0.3 -0.04
0 5 10 15 20 25 30
Frequency (GHz)

Figure 2.5. Comparison of measured (circles) and simulated (line) S-parameter at

two selected biases. (a) 0.15 V (pre-peak region); (b) 0.45 V (near the center of the
NDR region). To avoid being indistinguishable between the measured and simulated
curves, only 20% of the total 801 measured data points in each curve are shown.

23
 Equivalent circuit parameters were extracted by fitting the measured S-parameter

data over the entire frequency and bias range. Fig. 2.6 compares the differential

conductance of the RTD, extracted from the S-parameter measurement, with the

derivative of the dc I-V curve. The differential conductance extracted from the

S-parameter measurement closely agrees with that extracted from dc I-V measure-

ment.

20
Differential Conductance, G D (mS)

15

10

-5

-10
0 0.2 0.4 0.6 0.8
Voltage (V)

Figure 2.6. Comparison of the differential conductance extracted from the dc I-V
measurement (line) and S-parameter measurements (circles).

24
 The differential conductance in the pre-peak region in Fig. 2.6 shows a plateau

just below 0.2 V. This can be explained using Fig. 2.7, where the quantized state

E0 has a non-zero resonance width. The electrons in the emitter can tunnel into the

quantum well with a probability given by Eq. 2.3, drawn in Fig. 2.7. The shaded

area represents no corresponding electrons available in the emitter can tunnel into

the quantum well. Since the change in the differential conductance is proportional

to the change in the shaded area in Fig. 2.7(a), a higher differential conductance

occurs near zero bias, shown in Fig. 2.6. As the bias voltage, VA , increases and most

part of the probability curve would align below the Fermi level in the emitter, shown

in Fig. 2.7(b), the change in the shaded area would approach zero, resulting in a

constant differential conductance, i.e. a plateau in the I−V characteristic in Fig. 2.6.

Emitter Well Collector Emitter Well Collector

Probability

EF EF
1 E0 E0

qVA
(a) (b)

Figure 2.7. Schematic band diagram of the RTD at (a) equilibrium, and (b) an
applied bias voltage of VA .

25
 Fig. 2.8 shows a comparison of the quantum inductance, LQ , extracted from the

S-parameter measurements and calculated from Eq. 2.21. Good agreement between

calculation and measured data is obtained with the greatest deviation observed

near the center of the NDR region. Since lifetime might be expected to vary with

bias, some deviations may be expected as in this model lifetime is assumed to be

constant. The extracted lifetime is 2.58 ps, which is comparable to the calculated

lifetime using Frensley’s Schrödinger-Poisson solver BandProf for the structure of

Fig. 2.3 with a tunneling barrier of 1 nm and quantum well width of 3 nm.

8
Reciprocal of Inductance, L Q (nH )
-1

6
4 τ = 2.58 ps
2
0
-2
-4
-6
0 0.2 0.4 0.6 0.8
Voltage (V)

Figure 2.8. The reciprocal of the quantum inductance, LQ , vs. bias showing good
agreement between extracted inductance from S-parameter measurements (circles)
and the unified ac model (line).

26
 The total capacitance, CP , extracted from the S-parameter measurements and

calculated from Eq. 2.24 is plotted in Fig. 2.9. A similar close agreement is seen

in Fig. 2.9, where the escape rate, νC , is also assumed to be bias-independent, and

extracted to be (0.79 ps)−1 . The escape rate from the emitter to the quantum well

can be calculated from (1/τ −νC ) to be −0.88 ps−1 . This negative value indicates the

charge transfer direction is opposite to what has been assumed in the model. This

also indicates that the time constant which controls the charging and discharging

of the quantum well is longer than the electron lifetime in the quantum well. The

electron lifetime is typically associated with the process of electrons tunneling out

of the quantum well and results in a decrease in the quantum well charge. In this

model, the quantum well charge is increased by the tunneling-in process, thus the

change in the quantum well charge is controlled by the difference in the tunneling-in

and tunneling-out processes.

27
 40

35
Capacitance, C P (fF)

30

25
C0 = 29.3 fF
20
1/ν = 0.79 ps
C

15
0 0.2 0.4 0.6 0.8
Voltage (V)

Figure 2.9. The total capacitance, CP , vs. bias showing close agreement between
extracted capacitance from S-parameter measurements (circles) and the unified ac
model (line).

28
2.1.4 RTD Maximum Frequency of Oscillation

To explore the RTD in high-speed applications, high-frequency performance of

the RTD must be characterized. The upper limit for the frequency at which the

RTD can be operated is given by the maximum frequency of oscillation, fmax , which

is defined as the frequency above which the magnitude of the diode S11 is below 0

dB. This definition indicates that at the frequency above the maximum frequency of

oscillation, the power gain of the RTD is below one, i.e. the RTD can no longer be

treated as an active device. An equivalent definition can also be found in Ref. [55],

where the maximum frequency of oscillation is defined as the frequency at which

the real part of the impedance of the RTD becomes zero.

Given the new equivalent circuit of the RTD in Fig. 2.2(d), the impedance of

the RTD can be expressed as

Z(ω) = RS + 1/[1/(RD + jωLQ ) + jωCP ], (2.25)

where RD is the differential resistance, RS is the series resistance, CP is the total

capacitance including geometrical depletion capacitance and quantum capacitance,

and LQ is the quantum inductance, given by τ RD , and τ is the electron lifetime in

the quantum well. Letting the real part of the impedance equals to zero yields

4 2
ωmax RD CP2 τ 2 RS + ωmax
2 2
(RD CP2 RS − 2RD RS CP τ ) + RS + RD = 0. (2.26)

The maximum frequency of oscillation can be obtained by solving Eq. 2.26 for ωmax

(fmax = ωmax /2π),

s p
2
1 2RS τ − RD RS CP − RD RS2 CP2 − 4RD RS τ (RS CP + τ )
fmax = . (2.27)
2π 2RD RS CP τ 2

Similar expression can be found in Ref. [55], where Brown model [21] was used and

the quantum capacitance was neglected.

29
 Equation 2.27 indicates that the maximum frequency of oscillation of the RTD

can be increased by reducing the election lifetime in the quantum well and the

capacitance. The lifetime can be reduced by thinning the barrier to provide an

energetically wider transmission resonance and a leakier quantum well. The ca-

pacitance can be reduced by increasing the depletion region width, which can be

done by employing a thicker spacer layer on the collector side of the RTD. However,

as the depletion region width increases, the transit time of the electron across the

depletion region will increase, and which can be estimated to be 10 fs/nm for the

InGaAs collector. For the structure in Table 2.1, the space layer is 2 nm and the

estimated collector depletion region transit time is 20 fs, therefore the transit time

effect is negligible.

To make it clear how the differential resistance affects the maximum frequency

of oscillation, a plot of the maximum frequency of oscillation as a function of the

differential resistance has been plotted, based on Eq. 2.27, in Fig. 2.10, where RS ,

CP , and τ (see inset of Fig. 2.10) are assigned constant values which can be achieved

in InP technology. Figure 2.10 indicates that the maximum frequency of oscillation

can be increased by reducing the value of the negative differential resistance. To

reduce the negative differential resistance, it is necessary to achieve a high peak

current, which can be achieved by reducing the barrier thickness. Further, the

differential resistance is dependent on the applied voltage. The minimum value of

the negative differential resistance can be obtained by biasing the RTD near the

center of the NDR region.

30
 500
R =1Ω
S
400 C = 30 fF
P

(GHz) τ = 1 ps
300
max
f

200

100
-1000 -800 -600 -400 -200 0
R (Ω)
D

Figure 2.10. The calculated RTD maximum oscillation frequency, fmax , vs. the
negative differential resistance, RD , where RS , CP , and τ are assumed to be constant.

Figure 2.11 shows the maximum frequency of oscillation as a function of the se-

ries resistance, where RD , CP , and τ are assigned constant values (see inset) which

can be achieved in InP technology. From Fig. 2.11, the maximum frequency of oscil-

lation is achieved by minimizing the series resistance. The series resistance consists

of the metal-semiconductor contact resistances, the spreading resistance under the

contact, and the sheet resistance in the semiconductor between the emitter mesa

edge and the collector contact. For the structure in Table 2.1, the RTD collector

layer sheet resistance in the extrinsic collector region is the dominant component.

This extrinsic collector resistance can be reduced by decreasing the RTD emitter-

collector contact separation spacing.

31
 600
R = -50 Ω
D
C = 30 fF
P
400
(GHz)

τ = 1 ps
max

200
f

0
0 2 4 6 8 10
R (Ω)
S

Figure 2.11. The calculated RTD maximum oscillation frequency, fmax , vs. the
series resistance, RS , where RD , CP , and τ are assumed to be constant.

32
2.2 Heterojunction Bipolar Transistor

2.2.1 HBT Heterostructure Design

The InP/InGaAs HBT has currently recorded the highest current gain cut-off

frequency, fT , and simultaneously measured maximum oscillation frequency, fmax ,

over 450 GHz, of any bipolar transistors [49]. This is not only attributed to the

high electron velocity in InGaAs, but is also a consequence of reduction of device

parasitics in the technology. The cut-off frequency, fT , is given by [60]

1
fT = , (2.28)
2π(τE + τB + τDC + τC )

where τE is the emitter charging time, τB is the base transit time, τDC is the collector

depletion layer transit time, and τC is the collector charging time. Using InGaAs

as a base material provides low base transit time, due to the high minority electron

mobility (2000−3000 cm2 /V·s) [63]. Further, the energy separation between the Γ

and L valley of InGaAs (∼ 0.55 eV ) is larger than the conduction band discontinuity

of the emitter-base junction (∼ 0.25 eV). The electrons, thus, can traverse the base

with a high injection velocity without intervalley scattering. The drawback of the

InP/InGaAs/InP double heterostructure is the “type I” straddling band lineup at

the base-collector junction. The conduction band discontinuity at the base-collector

junction blocks electrons entering the collector from the base, unless this barrier

is removed as for example by the incorporation of a set-back layer [61]. Quater-

nary step-grading of the base-collector has also been proposed and demonstrated to

suppress the blocking effect of the base-collector heterojunction [47, 62].

The band diagram of the InP/InGaAs DHBT is shown in Fig. 2.12. A lightly-

doped (3×1016 cm−3 ) InGaAs set-back layer and a compositionally graded InAlGaAs

layer are introduced to eliminate the conduction band barrier at the base-collector

junction. The InGaAs base layer is also compositionally graded to form a quasi-

33
 Emitter Base Collector Subcollector
1
EC
0.5

Energy (eV) 0
EF
-0.5

-1
E
V
-1.5
0 100 200 300 400 500 600
Position (nm)

Figure 2.12. Calculated energy band diagram of InP/InGaAs double heterojunction

bipolar transistor (computed using the Schrödinger-Poisson solver, BandProf, of W.
R. Frensley).

electric field, which increases the average electron drift velocity in the base, thereby

decreasing the base transit time. This graded base and collector structures shown

in Fig. 2.12 were designed by Intelligent Epitaxy Technology, Inc., Richardson, TX.

As an alternative, the GaAsSb/InP heterostructure offers a “type II” staggered

band lineup, shown in Fig. 2.13, which introduces no base-collector barrier. This

staggered band lineup requires no grading at the base-collector junction and pro-

vides a higher hole blocking barrier at the emitter-base junction. Compared with

InGaAs, the minority electron mobility in GaAsSb is low, 600−800 cm2 /V·s com-

pared to 2000−3000 cm2 /V·s [63]. As a consequence, the GaAsSb base layer must

be thinner than the InGaAs base to achieve a comparable base transit time. This

thinner base may be expected to yield a higher base resistance, thus degrading the

34
 EmitterBase Collector Subcollector
1
EC
0.5

Energy (eV) 0
EF
-0.5

-1
EV
-1.5
0 100 200 300 400 500 600
Position (nm)

Figure 2.13. Calculated energy band diagram of InP/GaAsSb double heterojunction

bipolar transistor (computed using the Schrödinger-Poisson solver, BandProf, of W.
R. Frensley).

maximum oscillation frequency, fmax , since fmax is inversely proportional to the

square root of the base resistance.

2.2.2 HBT SPICE Model

The Gummel-Poon (GP) model [64] has been used for a long time in the IC

industry for circuit simulation of the bipolar transistor, and it has been built into

commercial CAD software, such as HSPICE and ADS. The small-signal Gummel-

Poon model is shown in Fig. 2.14, where rb , re , and rc are the base resistance, emitter

resistance, and collector resistance, respectively, rπ is the input resistance, rµ is the

base-collector resistance, Cπ is the base-emitter capacitance, Cµ is the base-collector

capacitance, and Csub is the collector-substrate capacitance. A detailed description

35
 + vµ

rµ
rb rc
B C
+ Cµ
vπ rπ Cπ Csub
gmπvπ-gmµvµ

re

Figure 2.14. The small-signal Gummel-Poon model for an InP HBT.

of the Gummel-Poon model can be found in Getreu’s book [65].

The Gummel-Poon model is used here in the simulation of TDT circuits; and

model parameters are summarized in Table 2.3, where the dc model parameters

were fit from an InAlAs/InGaAs HBT from the 1998 UCSB Ph.D. thesis of B.

Agarwal. The junction depletion capacitances were calculated based on the device

geometry and the layer structures shown in Fig. 2.13, and the calculation equations

are discussed in Appendix B.

36
 TABLE 2.3

InP HBT SPICE MODEL PARAMETERS FOR 1 µm2 DEVICE AREA

PARAM. VALUE UNITS DESCRIPTION

BF 50 Ideal maximum forward current gain
BR 1 Ideal maximum reverse current gain
CJC 0.5 fF Zero bias base/collector depletion capacitance
CJE 2.8 fF Zero bias base/emitter depletion capacitance
CJS 0.0001 fF Zero bias collector/substrate capacitance
EG 0.76 eV Energy gap
FC 0.9 Forward bias depletion capacitance coefficient
IS 1.8 fA Saturation current
ISC 0.14 fA Base-collector leakage saturation current
ISE 0.36 fA Base-emitter leakage saturation current
ITF 0 A High current parameter for effect on TF
MJC 0.1 Base-collector junction grading coefficient
MJE 0.3 Base-emitter junction grading coefficient
MJS 0 Substrate-junction exponential factor
NC 1.1 Base-collector leakage current ideality factor
NE 1.2 Base-emitter leakage current ideality factor
NF 1 Base-emitter ideality factor
NR 1.1 Base-collector ideality factor
RB 100 Ω Zero bias base resistance
PTF 2 Excess phase at f=1/2πf
RBM 100 Ω Minimum base resistance at high currents
RC 6 Ω Collector resistance
RE 10 Ω Emitter resistance
TF 0.8 ps Forward transit time
o
TNOM 27 C Temperature
TR 10 ps Reverse transit time
VAR 1 V Reverse Early voltage
VAF 20 V Forward Early voltage
VJC 0.7 V Base-collector built-in potential
VJE 0.725 V Base-emitter built-in potential
VJS 100 V Substrate-junction built-in potential
XCJC 1 Fraction of Cjc goes to internal base pin
XTF 0 Coefficient for bias dependence of TF
XTI 1 Saturation current temperature coefficient

37
 Shown in Fig. 2.15 are the dc model fits to the collector and base current data of

Agarwal. The emitter, base, and collector resistance are computed using a contact

resistivity of 3 × 10−7 Ωcm2 , a base sheet resistance of 1300 Ω/square, and base-

emitter separation of 0.3 µm. The calculation equations are discussed in Appendix

B.

-2
10

-4 SPICE model (line)

10 V =0V
CB
Current (A/µm2)

-6
IC
10
I
B

-8
10

-10
10 data points from Agarwal
UCSB Ph. D. thesis 1998
Fig. 4.6
-12
10
0 0.2 0.4 0.6 0.8
Base-Emitter Voltage (V)

Figure 2.15. SPICE simulated Gummel plot and comparison with measured data.
The parameters used in the SPICE simulation are shown in Table 2.3.

38
 The simulated common-emitter characteristic for the transistor model is shown

in Fig. 2.16. At the collector current density of 0.8 mA/µm2 , the dc current gain

is 16 and the output resistance is 8.8 kΩ. This small output resistance, indicated

by the large slope of the current-voltage characteristic in Fig. 2.16, arises from the

small early voltage in the model.

IB = 50 µA
Collector Current (mA/µm2)

0.8
40 µA
0.6
30 µA

0.4 20 µA

0.2 10 µA

0
0 1 2
Collector-Emitter Voltage (V)

Figure 2.16. SPICE simulated common-emitter characteristics. The parameters

used in the SPICE simulation are shown in Table 2.3.

39
 The high frequency performance of the device is shown in Fig. 2.17. At the

current density of 1.35 mA/µm2 and the collector-emitter voltage of 2 V, the sim-

ulated cut-off frequency and maximum oscillation frequency are 165 GHz and 360

GHz, respectively. In the absence of experimental results at the beginning of this

work, these models were used to estimate the expected performance of the InP-HBT.

30

20 VCE = 2 V
Gain (dB)

VBE = 0.76 V
fmax = 360 GHz
2
J = 1.35 mA/µm
10 C

fT = 165 GHz
0
107 108 109 1010 1011 1012 1013
Frequency (GHz)

Figure 2.17. SPICE simulated small signal current gain and power gain frequency
dependence for a 1 µm2 device. The parameters used in the SPICE simulation are
shown in Table 2.3.

40
2.2.3 HBT Cut-Off Frequency and Maximum Oscillation Frequency

Useful figures of merit for the HBT are fT , the cut-off frequency, and fmax , the

maximum oscillation frequency. The cut-off frequency, Eq. 2.28, is defined as the

frequency at which the current gain is 0 dB. The four time constants determining

fT are τE , τB , τDC , and τC , and these ae summarized by Sunderland and Dapkus

[60]. The emitter charging time, τE , is given by

VT
τE = (re + )(Cπ + Cµ ), (2.29)
JE WE LE

where WE is the emitter width, and LE is the emitter length. The base transit time,

τB , is given by
XB2
τB = , (2.30)
2VT µnB
where XB is the base layer thickness and µnB is the minority electron mobility in

the base. The collector depletion layer transit time, τDC , is given by

XC
τDC = , (2.31)
2vsat

where XC is the collector depletion region width and vsat is the electron saturation

velocity in the collector. The last time constant, τC , is the collector charging time

and is given by

τC = rc Cµ . (2.32)

The maximum oscillation frequency of the HBT is defined as the frequency at which

the unilateral power gain of the HBT equals unity, and is given by [66],
s
fT
fmax = , (2.33)
8πrb Cµ

where rb is the base resistance. The detailed discussions of calculating the cut-

off frequency and maximum oscillation frequency of the HBT based on the device

geometry and layer structures are shown in Appendix B.

41
 From Eq. 2.29−2.32, the cut-off frequency can be increased by lateral scaling

to decrease the emitter and collector charging time. It is also important to reduce

the base and collector depletion region transit time by reduce the base and collector

thickness. However, reducing the collector thickness will be traded off by lower-

ing the breakdown voltage, and reducing the base thickness will increase the base

resistance, thereby decreasing the maximum oscillation frequency.

A reduction of the base resistance can be achieved by reducing the base-emitter

spacing, shown in Fig. 2.18, where the base resistance consists of three components:

the metal-semiconductor contact resistance, RBB , the extrinsic base resistance, Rbx ,

and the intrinsic base resistance, Rbi . Since the base is usually heavily-doped in het-

erojunction bipolar transistors, the metal-semiconductor contact resistance, RBB ,

can be small, as quantified in the upcoming example. Further, the base layer is de-

signed thin to reduce the base transit time, thereby increasing the current gain and

cut-off frequency. Therefore the spreading resistances in the base semiconductor,

Rbx and Rbi , are usually dominant parts of the total base resistance. For example,

a 2 × 2 µm2 emitter InP/GaAsSb HBT with a 40 nm base doped to 4 × 1019 cm−3 ,

the intrinsic base resistance and the base contact resistance can be estimated to be

54 Ω and 4 Ω, respectively. For a base-emitter spacing of 1 µm, the extrinsic base

resistance is estimated to be 163 Ω. Decreasing the base-emitter spacing to 0.1 µm,

the extrinsic base resistance can be reduced to 16 Ω, and the total base resistance

will decrease by 66%.

42
 XBE WE

Emitter
Base Contact
contact Emitter
RBB Base
Collector Rbx Rbi
contact Collector

Subcollector

Figure 2.18. Schematic cross section of a heterojunction bipolar transistor.

43
 CHAPTER 3

CIRCUIT DESIGN AND SIMULATION

This chapter summarizes the TDT differential comparator and the TDT fre-

quency translator design and simulation. Tunnel diode circuit design using the

negative differential resistance and multi-stable I−V characteristics of the TD is

introduced in Section 3.1. A TDT differential comparator, utilizing the bi-stability

characteristic and the high switching speed of the tunnel diode, was invented in

this research. Circuit analysis and simulations show that the TDT differential com-

parator is faster and dissipates less power than its transistor-only counterpart. A

TDT frequency translation circuit based on Cellonics technology [41] is discussed in

Section 3.3. Adding a transistor to the frequency translation circuit increases the

frequency and the magnitude of the output signal, but decreases the input immunity.

3.1 Circuit Design Using Tunnel Diodes

The two unique attributes of the TD, the negative differential resistance and

the multi-valued current-voltage characteristic, along with its high-speed index are

attractive for circuit applications. The negative differential resistance enables the

creation of high-speed oscillators [68, 69]. The multi-valued I-V characteristic en-

ables tunnel diodes to be used in logic circuits and memories [67, 25]. The basic

design concepts of these two categories of circuits are introduced as follows.

44
 Tunnel diode oscillators can simply be obtained by biasing the tunnel diode

in the NDR region, and connecting the tunnel diode to a resonator. This type of

oscillator has been reported at frequencies as high as 712 GHz with an output power

density of 15 W/cm2 [68]. To increase the power, tunnel diode oscillator arrays can

be used. It has been reported that a 16-element InP-based Schottky-collector RTD

oscillator array produced a power density of 440 W/cm2 at 290 GHz [69].

The multi-stability characteristics of the tunnel diode can be exploited in high-

speed digital application. In this case, the tunnel diode is biased out of the NDR

region, and the voltage switches between the stable states with high switching speed

due to the high speed index of the tunnel diode. A commonly used topology is the

clocked tunnel diode pair, Fig. 3.1, where the controlled-current source, ICT L , can

be implemented using a FET [9], an HBT [11], a photodiode [70], or other diodes

[67].

VCK

ICTL D2
VO
D1

Figure 3.1. Current-controlled clocked tunnel diode pair.

45
 The operation of this tunnel diode bistable circuit can be understood through

the load lines, shown in Fig. 3.2. When the clock, VCK , is at a low level, the tunnel

diode pair is in a monostable state and the output voltage is low. When the clock

is high, the output node between the tunnel diodes is bistable, latching to either

a high or low voltage state, as determined by the tunnel diode peak currents and

the control current occurring at the rising edge of the clock. The peak current,

IP 2 , of the load diodes, D2 , is designed to be greater than the peak current, IP 1 ,

of the driver diodes, D1 , with the following relationship IP 2 < IP 1 + ICT L . This

relationship is set by the choice of device areas.

Driver Diode D
1
Load Diode D2
Current

ICTL

Load Diode D2

Driver Diode D1

Voltage V VCK VO Voltage VCK

O
(a) (b)

Figure 3.2. Load lines of the tunnel diode pair at two conditions: (a) ICT L is off;
and (b) ICT L is on.

46
 In series-connected tunnel diodes, the diode with the lowest peak current switches

first in response to an applied voltage. Assuming the control current is off, Fig. 3.2a,

and the clock switches from low to high, the driver diode D1 switches because the

peak current of driver diode D1 is less than the peak current of load diode D2 . The

resulting output voltage, VO , is then in a high voltage state. If, on the other hand,

Fig. 3.2(b), the control current is on, and the clock switches from low to high, the

added control current through diode D2 causes load diode D2 to switch instead of D1 .

Diode D2 switches instead of D1 because of the design condition IP 2 < IP 1 + ICT L ,

and that the lowest current branch is the one to switch first. As a result, the output

remains in a low voltage state, Fig. 3.2(b).

3.2 Differential Comparator

Current-mode-switching differential comparators are widely used in high-speed

logic and mixed analog-digital circuits [71, 72, 73]. A conventional transistor-only

differential comparator is shown in Fig. 3.3 and consists of a differential transistor

pair, Q1 and Q2 , and output followers, Q3 and Q4 , which are used to buffer the

output signal. According to this circuit topology, there are limited options for

increasing the circuit speed. First of all, all the parasitics must be minimized. The

switching time of the differential pair is determined by the time delay at the input

node and output node, X. The time delay at the input node is inversely proportional

to the transistor’s transconductance. An increase in the tail current, IT AIL , can be

used to maximize transconductance. The transconductance maximum is limited by

the maximum power density of the technology. The time delay at the output node

is proportional to the collector resistance, RC , which can be reduced at the expense

of the output voltage swing which is equal to the product of the collector resistance

47
 VCC
RC RC
Q3 X Q4

VOUT VIN VIN VOUT

Q1 Q2
RL RL
I1 I2
ITAIL
−VEE

Figure 3.3. Schematic of a conventional transistor-only differential comparator.

and the tail current.

A new approach using tunnel diodes in a differential comparator is shown in Fig.

3.4, where two tunnel diode pairs D1 −D3 and D2 −D4 are connected to the collector

outputs of the input differential transistor pair Q1 and Q2 . Emitter followers Q3

and Q4 are used to buffer the output signal.

The operation of this TDT differential comparator can be understood in terms

of the current-controlled clocked tunnel-diode pair discussed in Section 3.1. The

collector current of the input transistors, Q1 and Q2 , can be treated as the control

current. When the input to transistor Q1 is low, the tail current flows through

transistor Q2 . In this case, when the clock switches from low to high, driver diode

D1 switches because of the design condition IP 1 < IP 3 and the collector current of

transistor Q1 is off, resulting in a high voltage state at node X. If the input to

transistor Q1 is high, the tail current flows through Q1 . When the clock switches

48
 VCC

D3 CK D4
Q3 X
Q4
D1 D2

VOUT VOUT
VIN VIN
RL Q1 Q2 RL
I1 I2

ITAIL
−VEE

Figure 3.4. Schematic of a tunnel diode/transistor differential comparator.

from low to high, the load diode D3 switches because of the design condition IP 3 <

IP 1 + IT AIL and as a result, node X remains at a low voltage state. This circuit

provides a return-to-zero (RZ) format output, which means the output is reset to

zero before the next pulse is generated.

The simulated output waveform of the TDT and transistor-only circuits of Figs.

3.3 and 3.4 are compared in Fig. 3.5. Each circuit has a bitstream applied as

indicated by the pattern of ones and zeros along the upper horizontal axis of the

figure. This bitstream has an amplitude of 400 mV with rising and falling time of

1 ps. The transistor-only circuit, which lacks the RZ format, is just sufficient in

49
 0.2 1 1 0 1 0 1 0 1 1 1

0 (b)
(V)
-0.2
OUT

-0.4
V

-0.6
(a)
-0.8
0 20 40 60 80 100
t (ps)

Figure 3.5. Simulated output waveforms for the two comparator circuits of (a) tunnel
diode/transistor comparator, and (b) conventional bipolar transistor comparator.

speed to follow this waveform at 100 GHz. The TDT circuit which is clocked by a

100 GHz sinusoidal source achieves the RZ output. To drive the tunnel diode pair,

the off-chip clock may need to be buffered, e.g. using emitter followers, to the node

CK.

This TDT differential comparator circuit of Fig. 3.4 is faster and dissipates less

power than the conventional transistor-only differential comparator, Fig. 3.3. This

can be understood as follows. The open-circuit time constant at node X of the TDT

circuit of Fig. 3.4 is given by

τT DT ∼
= (RD1 //RD3 )(CBC1 + CBC3 + CD1 + CD3 ), (3.1)

where, RD1 , RD3 and CD1 , CD3 are the resistances and capacitances of D1 and D3 ,

50
respectively, and CBC1 and CBC3 are the base-collector capacitances of Q1 and Q3 ,

respectively. The open-circuit time constant at node X of the transistor-only circuit,

Fig. 3.3, is given by

τT ∼
= RC (CBC1 + CBC3 ), (3.2)

where RC is the collector resistance. For an equivalent output voltage swing at

node X of 600 mV and with a tail current of 1.5 mA, the transistor-only circuit

requires an RC of 400 Ω. With base-collector capacitance of approximately 0.5

fF/µm2 , the open-circuit time constant of the transistor-only circuit, τT , has a

magnitude of approximately 0.6 ps (CBC3 is typically larger in the transistor-only

circuit because a larger area output transistor is needed, as will be explained in the

next paragraph). In the TDT circuit, the parallel combination of the tunnel diode

resistance, RD1 //RD3 , is 60 Ω, and the tunnel diode capacitances (2 fF/µm2 ) added

to the base-collector capacitance at node X, result in an open-circuit time constant

of approximately 0.4 ps.

The total time delay of the differential pair can be expressed by

q
τD ∼
= 2
τIN 2
+ τOU T, (3.3)

where τIN is the time delay at the input and τOU T is the time delay at the output and

is equal to τT DT for the TDT circuit and τT for the transistor-only circuit. When the

TDT and transistor-only circuit operate at the same tail current, the input delay

for both circuits are identical and can be estimated by

CBE (Av + 1)CBC ∼ CBE

τIN ∼
= + = + τOU T , (3.4)
gm gm gm

where Av is the voltage gain, gm is the transconductance and equal to dIC /dVBE ,

and CBE is the base-emitter capacitance. Since the base-emitter capacitance is dom-

inated by the diffusion capacitance as the transistor operates in the active region,

51
it can be estimated by
d(τf IC )
CBE ∼
= = τf gm , (3.5)
d(VBE )
where τf is the forward transit time and is 0.8 ps from Table 2.3. From Eq. 3.3−3.5,

the total time delay of the differential pair is approximately 0.9 ps and 1 ps for the

TDT and transistor-only circuits, respectively. Therefore, the TDT circuit is faster

than the transistor-only circuit when they operate at the same tail current. In the

transistor-only circuit, speed improvements to equal the TDT circuit are obtained

by trading-off power and gain, i.e. by increasing the tail current and decreasing the

pull-up collector resistor.

Next consider the voltage gain of the output stage. The output of the differential

pair at node X in Fig. 3.4 can be replaced by a Thevenin equivalent voltage source,

VT H , in series with a Thevenin equivalent resistor equal to the output resistance,

RX , of the differential pair. The voltage gain of the output stage is given by

VOU T 1 ∼ 1
AV = = RX +rπ = RX +rπ
, (3.6)
VT H 1+ (β+1)(RL //ro )
1+ (β+1)RL

where rπ is the input resistance of Q3 , ro is the output resistance of Q3 , RL is the load,

β is the ac current gain of Q3 , and ro À RL in the general case. From the previous

discussion, the output resistance is different in the two circuits, equaling RD1 //RD3

in the TDT circuit and RC in the transistor-only circuit and RD1 //RD3 < RC . To

obtain the same output voltage swing (voltage gain) in the two followers, the input

resistance, rπ , of Q3 in the transistor-only circuit has to be smaller than that in

the TDT circuit. Since rπ = βVT /IC , where VT is the thermal voltage, and IC is

the collector current of Q3 , the emitter follower in the transistor-only circuit has

to be biased at higher current, resulting in more power dissipation. Further, the

high collector current requires the transistor size to be increased, resulting in a

capacitance increase and speed reduction in the conventional circuit.

52
 This TDT differential comparator is of special interest for use as a switching

element in direct digital synthesizers (DDS). A recently developed algorithm [74,

75], based on list decoding, for DDS provides an improved signal-to-noise ratio

(SNR), compared with conventional Σ∆ approaches. In both list decoding and Σ∆

approaches, a specially designed digital bitstream is output through a single-bit

digital-to-analog converter (DAC) and a reconstruction filter to form the analog

signal. The Fourier spectrum of the desired signal is embedded in the pattern of the

bitstream. A high-speed and high-linearity single-bit DAC is required to achieve

high SNR, because the severe distortion of the output signal is usually generated

from the nonlinearity of the DAC. The comparator circuits, proposed in this thesis,

are designed for use in such DDS applications.

To compare the power dissipation and linearity of both TDT and transistor-only

circuits, both circuits (Fig. 3.3 and 3.4) were simulated in Agilent ADS. The input

bitstream has a 100 Gbps bit rate, and an amplitude of 400 mV with rising and

falling time of 1 ps. The synthesized passband signal frequency is approximately

37.3 GHz. The simulated spectrums of the synthesized signal in both TDT and

transistor-only circuits are shown in Fig. 3.6. Approximately 60 dBc spur free dy-

namic range (SFDR) is obtained for both circuits, showing these two circuits have

approximately the same linearity. The power dissipation is reduced by approxi-

mately 4× in the TDT differential pair and 1.6× in the full TDT circuit. It is

reported that a conventional DDS utilizing a high speed, multi-bit DAC achieved

30 dBc SFDR, with a clock rate of 9.2 GHz, an output frequency of 4.56 GHz and

a power dissipation of 15 W [76]. The single-bit approach coupled with the TDT

comparator offers techniques to significantly lower power, improve speed, and extend

SFDR.

53
 dB
100

80

60

40

20
(a)

32 34 36 38 40 42
Frequency (GHz)
dB
100

80

60

40

20
(b)

32 34 36 38 40 42
Frequency (GHz)

Figure 3.6. Simulated output spectrum of the synthesized passband signal for (a)
tunnel diode/transistor and (b) transistor-only differential comparator showing 60
dBc SFDR around 37.3 GHz. This simulation uses high speed InP-based HBT and
RTD models.

The TDT differential comparator has been layed out, as shown in Fig. 3.7. The

pads labeled IN and IN BAR were designed for the ground-signal-ground-signal-

ground coplanar probe with 150 µm pitch size to provide the differential input

bitstream. The pads labeled OUT and OUT BAR were designed for the 150 µm

pitch ground-signal-ground (GSG) coplanar probes to output the synthesized sig-

54
nals. The pads labeled VEE, CLOCK, VCC, BIAS1, and BIAS2 were designed for

a special probe card, which has one 150 µm pitch GSG coplanar probe and seven

dc probes. The GSG probe will be connected to the pad labeled CK to provide the

clock signal and other dc probes will be connected to the dc pads to provide dc bias

for the circuit. The differential amplifier in the circuit was layed out symmetrically

to achieve equal phase delay in the two differential signals. Each dc pad was con-

nected to a metal-dielectric-metal capacitor to obtain an ac ground. The resistors

were layed out assuming the resistivity of 25 Ω/2.

Figure 3.7. Layout for the RTD/HBT differential comparator. The circuit schematic
is shown in Fig. 3.4. Layout area is 1.8×1.1 mm2 .

55
 Recently, it is noted that Yang’s group was working independently on the RTD/HBT

flip-flops utilizing the same circuit topology as in Fig. 3.4 for logic applications, and

demonstrated a 20-GHz single-ended input/output flip-flop [11] and a 10-GHz dif-

ferential output flip-flop [77]. Their circuit was fabricated with InP/InGaAs single

heterojunction bipolar transistors (SHBTs) and AlAs/InGaAs/InAs RTDs using

optical lithography and the wet etching technique. The SHBT showed maximum fT

and fmax of 89 GHz and 133 GHz, respectively, and the RTD showed a peak current

density of 1 mA/µm2 and a peak-to-valley current ratio of 13 [77].

The demonstrated circuit [77] operated at 10 GHz with a power dissipation of

72.5 mW. The power dissipation is much higher than the simulation results in this

research [40], which might because the circuit is biased at high tail current. It is the

advantage of this TDT circuit that it can operate at a relative low tail current and

achieve the same speed compared to its transistors-only counterpart. Therefore, it

is expected to reduce the power dissipation by optimizing the circuit and device

design.

56
3.3 Frequency Translator

In 1918, Armstrong invented the superheterodyne radio receiver. To this day,

almost all radio and TV receivers being made are of this type. The receiver, Fig.

3.8(a), consists of an antenna, a radio-frequency (RF) section, a mixer and local

oscillator, an intermediate-frequency (IF) section, and a demodulator [78]. The in-

coming modulated wave is picked up by an antenna and amplified in the RF section.

The mixer and local oscillator convert the incoming signal to a predetermined in-

termediate frequency, which is lower than the incoming carrier frequency. The IF

section provides the amplification and selectivity. The output of the IF section is

applied to the demodulator to recover the baseband signal.

Antenna

RF IF
Mixer Demodulator
Section Section

Local
~ Oscillator
(a)
Antenna

RF Frequency Translator/ Counter/

Section Pulse Generator Decision

(b)

Figure 3.8. Schematic architectures of two types of receivers: (a) conventional

superheterodyne receiver [78] and (b) Cellonics receiver [41].

57
 Cellonics Inc., Singapore, has recently invented a new demodulation technique

based on pulse generating and counting [41]. In the Cellonics receiver, Fig. 3.8(b),

a nonlinear circuit is used to generate pulses from the received signal and the de-

modulation is performed by counting the generated pulses. In the Cellonics circuit,

a slow time-varying input signal is converted to fast pulse trains. Cellonics calls this

process frequency translation.

The Cellonics receiver has several advantages over the conventional superhetero-

dyne receiver [41]. First, no mixers, local oscillators or phase-locked loops (PLLs)

are needed in the Cellonics receiver, resulting in smaller chip area and lower power

consumption. In conventional receivers, thousands of carrier cycles are required to

extract one symbol because the receiver requires time to synchronize with the car-

rier signal. With Cellonics technology, information can be extracted in every carrier

cycle, which maximizes the transmission throughput.

The Cellonics frequency translation technology can also be used in the trans-

mitter. Using the frequency translation circuit, the baseband signal is up-converted

to the RF frequency pulses with a tuned certain center frequency which is fit into

the FCC (Federal Communications Commission) specified band [79]. Similarly as

the Cellonics receiver, no mixers, local oscillators and PLLs are required in the

transmitter resulting lower power dissipation.

The Cellonics frequency translation circuit using the tunnel diode is shown in

Fig. 3.9 [79]. This circuit is, in essence, a tunnel diode oscillator circuit biased into

oscillation by a digital bitstream. When the input digital bit is at the low level, the

voltage applied to the tunnel diode, D1 , is less than the tunnel diode’s peak voltage,

resulting in an output low voltage. When the input digital bit is at a high level,

D1 will be biased in its NDR region, resulting in an oscillating output signal. The

oscillation frequency is determined by the inductor, L, and the capacitance of D1 .

58
 D1
INPUT OUTPUT

Figure 3.9. Cellonics tunnel diode frequency translation circuit diagram [79].

Since the tunnel diode is a two-terminal device, there is no isolation between

the input and output signals in the Cellonics frequency translation circuit. Here a

transistor is added to isolate the input and output signals. This circuit is shown in

Fig. 3.10 and works similarly to the Cellonics circuit. The simulated input-output

waveform of this TDT frequency translator is shown in Fig. 3.11. As the input

pulse signal voltage is at a low level, the output voltage is at a low level because the

transistor, Q1 , is off. As the input pulse signal switches to a high level, the tunnel

diode, D1 , is biased into its NDR region, resulting in an oscillating output signal. In

the simulation, an emitter follower is connected to the output of the TDT frequency

translator to drive the 50 Ω load, and the input voltage source resistance is 50 Ω.

59
 VCC

L
INPUT Q1
OUTPUT
D1

Figure 3.10. Notre Dame tunnel diode/transistor frequency translation circuit. The
circuit component and source values in the simulation are: L = 1 nH, VCC = 1 V,
and the HBT and RTD areas are both 2 × 2 µm2 .

1.6 1.6
Input

0.8 0.8

V
V (V)

out
in

Output (V)
0 0

-0.8 -0.8
0 200 400 600
Time (ps)

Figure 3.11. Simulated output waveform for the tunnel diode/transistor frequency
translation circuit. Circuit component values are given in the caption of Fig. 3.10.

60
 Since the negative differential resistance and the device capacitances are all bias

dependent, the frequency and magnitude of the output signal are both depended on

the input voltage. For the use of frequency translation, it is necessary to minimize

the frequency fluctuation according to the input variation. The simulated frequency

and magnitude of the output signal vs. the input voltage for both the Cellonics and

TDT frequency translation circuits are shown in Fig. 3.12. From Fig. 3.12, the

input voltage in the TDT circuit is approximately 0.7 V higher than that in the

Cellonics circuit, which is due to the transistor turn-on voltage is approximately 0.7

V.

By adding the transistor into the circuit, the frequency of the output signal

increases approximately by a factor of 2 and the output voltage magnitude increases

approximately 10%. The frequency is determined by the resonant in the circuit,

i.e. the inductance and capacitance. In the Cellonics circuit, the frequency is

determined by that series inductor, L, and the tunnel diode junction capacitance,
√
and is proportional to 1/ L. In the TDT circuit, the inductor is connected to the

base of the transistor, the equivalent inductance in the resonator should be equal to
p
L/(β + 1). Therefore, the output frequency is proportional to 1/ L/(β + 1), and

is higher than that in the Cellonics circuit for the same inductance.

From Fig. 3.12, it also can be seen that when the input voltage varies by 1%,

the maximum change in the output frequency is 0.7% and 0.5% in the TDT and

Cellonics circuit, respectively. In summary, adding a transistor into the Cellonics

frequency translation circuit provides isolation and power gain. The drawback is

that it degrades the immunity to the input voltage variation and increases the input

voltage.

61
 20 800

V
Frequency (GHz)

out, peak-to-peak
15 600

10 400

5 200

(mV)
(a)
0 0
0.3 0.4 0.5 0.6
Vin (V)
40 800

V
Frequency (GHz)

out, peak-to-peak
30 600

20 400

10 200

(mV)
(b)
0 0
1 1.1 1.2 1.3
Vin (V)

Figure 3.12. Simulated frequency and magnitude of the output signal vs. the
input voltage for (a) the Cellonics frequency translation circuit of Fig. 3.9, and (b)
the tunnel diode/transistor frequency translation circuit of Fig. 3.10. The circuit
component values in the simulation are: L = 1 nH, VCC = 1 V, and the RTD area
is 2 × 2 µm2 .

62
 It is noted that the circuit topology proposed by De Los Santos [29] can also

be used for the frequency translation, and the output frequency is proportional
√
to 1/ L. Therefore, in this research, connecting the inductor to the base of the

transistor can achieve higher output frequency compared to Cellonics and De Los

Santos circuits when they use the same inductance.

The TDT frequency translator has been layed out, as shown in Fig. 3.13. The

pads labeled OUT are designed for the 150 µm pitch GSG coplanar probes to output

the signals. The pads labeled VEE, VCC, and BIAS were designed for connecting

the dc probes to provide bias for the circuit. Each dc pad was connected to a

metal-dielectric-metal capacitor to obtain an ac ground. The resistors were layed

out assuming the resistivity of 25 Ω/2. The unlabeled pad beside the BIAS pad

was connected to the base of the transistor, and the wire inductor will be bonded

between this pad and the BIAS pad.

63
Figure 3.13. Layout for the RTD/HBT frequency translator. The circuit schematic
is shown in Fig. 3.10. Layout area is 0.8×0.5 mm2 .

64
 CHAPTER 4

SELF-ALIGNED NITRIDE SIDEWALL PROCESS

To fabricate the TDT circuits, a new self-aligned contact process was demon-

strated for the first time. The novelty of this process is using silicon nitride sidewalls

and a BCB etchback to obtain self-aligned emitter-base contacts. In this Chapter,

prior art in self-aligned emitter-base contact formation is first reviewed, Section 4.1.

The process developed in this research is outlined in Section 4.2. Two key steps of

this new process are the nitride sidewall formation and the BCB etchback process

and these are shown in detail in Section 4.3. Section 4.4 discusses the fabrication of

the InP/InGaAs DHBTs. Device test results prove the feasibility of this new side-

wall and etchback process. Fabrication and testing of AlAs/InGaAs/InAs RTDs is

shown in Section 4.5

4.1 Process Overview

As discussed in Chapter 2, it is necessary to reduce the extrinsic base resistance

to increase the maximum oscillation frequency of the device, and the extrinsic base

resistance is proportional to the base-emitter contact spacing, therefore a self-aligned

contact process is required to minimize the emitter-base contact spacing. Prior

published self-aligned base-emitter contact processes are often based on either wet-

etching and lift-off [46, 49], or emitter and base regrowth [80, 81], and are considered

65
to have process yield limitations [82]. In the wet-etching and lift-off process, the

base-emitter separation, XBE , is critically depended on the emitter mesa lateral

etch from the wet chemical etchant, a process which is difficult to precisely control.

Further in the lift-off process, metal stringers are easily left which can short the base-

emitter junction [73]. Therefore, the wet-etching and lift-off process is undesirable

for volume manufacturing of large-scale circuits consisting of thousands of submicron

devices with high yield [82]. As for the emitter or base regrowth process, additional

masks and surface cleaning before the regrowth are introduced, which increases the

process complexity and expense.

Recently, Vitesse [48] and Rockwell Scientific [83] have demonstrated high per-

formance HBTs using dielectric-sidewall processes for the self-aligned emitter-base

contacts. Dielectric sidewalls allow the emitter mesa undercut to be minimized

and eliminates the lift-off process for the base metallization, resulting in high yield.

In Vitesse’s process, dielectric sidewalls are used for the emitter-base contacts, and

their Inp DHBT has demonstrated fT and fmax both over 300 GHz [48]. In Rockwell’s

process, the dielectric sidewall is also used for the emitter-base contact separation,

and a selective-base electroplating is required for the base contact. HBTs fabricated

in the Rockwell process have demonstrated 326 GHz fT and 305 GHz fmax [83].

In this work, a new self-aligned contact process using silicon nitride sidewalls

and a BCB etchback has been developed and is outlined as Figs. 4.1 and 4.2.

The emitter contact lithography and metallization uses in a lift-off process using

Ti/Pt/Au as the emitter contact metals, Fig. 4.1(a). The emitter mesa is etched in

a wet chemical etchant using the emitter contact as the etch mask, Fig. 4.1(b). The

silicon nitride sidewall is formed by depositing a PECVD Si3 N4 film, followed by an

anisotropic RIE etch in an SF6 /Ar plasma, Fig. 4.1(c). To form the base contact,

tungsten is blanket sputtered on the wafer, Fig.4.1(d). The surface is planarized by

66
 200 nm

Emitter
Metallization
(a) Ti/Pt/Au

Emitter

Base Collector

Emitter Etch Ti/Pt/Au

Metal as Etch Mask Emitter
(b)
Base Collector

Si3N4 sidewall

Silicon Nitride
Sidewall Formation Ti/Pt/Au
(c)
Emitter

Base Collector

W
Base Metal
Deposition Ti/Pt/Au
(d)
Si3N4 Emitter Si3N4

Base Collector

Figure 4.1. Scale drawings of self-aligned emitter-base contact formation using a

nitride sidewall spacer. These drawings describe: (a) the emitter metallization, (b)
the emitter mesa etch, (c) the silicon nitride sidewall formation, and (d) the base
metal deposition.

67
 Benzocyclobutene
(BCB)

BCB Deposition
(a)
W

Ti/Pt/Au
Si3N4 Emitter Si3N4

Base Collector

W
BCB BCB
BCB Etchback
(b) Ti/Pt/Au
Si3N4 Emitter Si3N4

Base Collector

BCB BCB
W RIE Etch
BCB as Etch Mask Ti/Pt/Au
(c) W W
Si3N4 Emitter Si3N4

Base Collector

BCB Removal
Ti/Pt/Au
(d) W W
Si3N4 Emitter Si3N4

Base Collector

Figure 4.2. Scale drawings of self-aligned emitter-base contact formation using a

nitride sidewall spacer. These drawings describe: (a) the BCB deposition, (b) the
BCB etchback, (c) the tungsten RIE etch, and (d) the removal of the BCB.

68
spinning on BCB followed by a reflow at 250 ◦ C for one hour, Fig. 4.2(a). The BCB

is then RIE etched in an SF6 /O2 plasma and the etch stops as soon as the tungsten

on top of the emitter mesa appears, Fig. 4.2(b). This part of the tungsten which

is on the top of the emitter mesa is etched off in an SF6 /Ar plasma and the base

contact metal is protected by the remaining BCB, 4.2(c). After removing the BCB,

the self-aligned emitter-base contact is obtained, Fig. 4.2(d). The collector contact

is formed by a lift-off process. The detailed description of the sidewall formation

and the BCB etchback is shown in the next section.

Optical micrographs of a fabricated 4×4 µm2 emitter InP/InGaAs HBT with

the sidewall process are shown in Fig. 4.3. It can be seen that the base metal wraps

around the emitter metal. The base and collector contacts are non-self-aligned and

the spacing is 2 µm. The opening in the collector metal was designed to assist the

acetone flowing and ease the metal lift-off process.

In addition to the device formation, the circuits of interest have required on-chip

resistors and capacitors. A 30 nm titanium thin film is deposited in a lift-off process

as the on-chip resistor. The on-chip metal-dielectric-metal capacitor is formed by

metal lift-off processes and a 300 nm PECVD silicon nitride deposition. The detailed

process flow is summarized in Appendix A.

69
 Emitter

Emitter Base

Emitter Base

Collector

Figure 4.3. Optical micrographs of a 4x4 µm2 emitter HBT. The graphs show the
step-by-step HBT emitter, base and collector contact formation.

70
4.2 Nitride Sidewall Formation and BCB Etchback Process

The silicon nitride sidewall spacer is formed using an anisotropic etch for silicon

nitride. Different etch recipes, using SF6 , CF4 , or CHF3 , have been investigated.

The best sidewall process was achieved using an anisotropic etch for silicon nitride

in an SF6 /Ar plasma.

The silicon nitride film is deposited in a Unaxis 790 PECVD system with the

following recipe established by Shishir Rai: SiH4 /NH3 (40/4 sccm), 500 mTorr, 200

W, 250 o C. The nitride deposition rate is ∼ 20 nm/min.

To obtain a well-defined nitride sidewall, an anisotropic RIE process with a well-

controlled etch rate is required. Silicon nitride etching is often performed using a

fluorine-producing gas, such as CF4 , SF6 , or CHF3 [84, 85, 86], and all three of

these gases were investigated; silicon nitride sidewalls were successfully observed

producing using CHF3 and SF6 , but not with CF4 .

Mele et al. [84] discussed anisotropic etching profile using CHF3 as the reactive

ion etchant. They suggest that CHF3 decomposes in the plasma to form a polymer

which deposits on the sidewall to suppress lateral fluorine etching [84]. However,

this polymer deposition on the surface causes the nitride etch rate to decrease as a

function of time [84], which increases the difficulties of etch time control.

To investigate the etch rate of the RIE etch for silicon nitride in the CHF3 plasma,

a 300 nm thick silicon nitride film was patterned with AZ1813 photoresist as the

etch mask. Samples were etched in a Plasma-Therm RIE 790 system for different

times, and after removing the photoresist, step-profiling was used to obtain the etch

depths vs. etch times, Fig. 4.4. The etch rate is approximately 6 Å/s and decreases

as a function of time.

To investigate the nitride sidewall formation using CHF3 , Ti/Au patterns were

formed using a lift-off process on a GaAs substrate, followed by depositing and

71
 300
y = 38.625x R= 0.9898
250 Etch Rate: 0.6 nm/s

Etch Depth (nm)

200

150

100

50
CHF 20 sccm, 15 mT, 250 W
3
0
0 2 4 6 8 10
Etch Time (min)

Figure 4.4. RIE etch characteristic of silicon nitride in CHF3 .

etching a PECVD silicon nitride film in a CHF3 plasma. The deposited silicon

nitride thickness was 300 nm, and an over-etch factor of 1.4 was used. An SEM

micrograph, Fig. 4.5, shows the presence of the sidewall. However, the surfaces of

the GaAs substrate and the Ti/Au mesa are rough, which may because of a polymer

or silicon nitride residue. Since the etch rate is dependent on the polymer deposition

so that it decreases with time and is not repeatable, it is difficult to control the etch

time to obtain sidewalls with smooth surface using CHF3 .

Anisotropic etching of silicon nitride can also be achieved using an SF6 plasma,

which yields a more linear etch rate compared to using CHF3 , because no carbon

polymer is formed in the plasma. The same procedure, as described above, was

performed to determine the silicon nitride etch rate in the SF6 /Ar plasma, and the

result is shown in Fig. 4.6. Argon is used to improve the anisotropy of the etching.

The etch rate is estimated to be 17 Å/s, and is more constant and faster than using

72
 SiNx Ti/Au

GaAs

Figure 4.5. SEM micrograph of the silicon nitride sidewall on a GaAs substrate.
Silicon nitride is RIE etched in CHF3 plasma.

CHF3 .

An SEM micrograph of the nitride sidewall formed using SF6 /Ar is shown in Fig.

4.7. The deposited silicon nitride was 300 nm, an over-etch factor of 1.2 was used,

and the sidewall thickness was about 150 nm. The ratio of the vertical etch rate to

the lateral etch rate is approximately 2:1, which means to obtain a certain thickness

of sidewall, a doubled sidewall thickness of deposition is indicated. Compared to

Fig. 4.5, the GaAs and Ti/Au surfaces are smoother after the sidewall etch using

SF6 /Ar even with a smaller over-etch factor. Therefore, for this work, SF6 /Ar was

selected for the RTD/HBT process development.

73
 250
SF6/Ar: 2/18 sccm
200 15 mTorr, 150 W

Etch Depth (nm)

150

100

50
Etch Rate: ~ 1.7 nm/s
0
0 50 100 150
Etch Time (s)

Figure 4.6. RIE etch characteristic of silicon nitride in SF6 and Ar.

Figure 4.7. SEM micrograph of the silicon nitride sidewall on a GaAs substrate.
Silicon nitride is RIE etched in SF6 /Ar plasma.

74
 To investigate the BCB etchback process, first the BCB RIE etch rate was cal-

ibrated. The BCB was spun on the wafer with a spin speed of 5000 rpm for 30 s,

followed by a re-flow process in a Unaxis 790 PECVD system in nitrogen environ-

ment at 250 o C with the nitrogen pressure of 1 Torr and the flow rate of 500 sccm

for 1 hour. The BCB film thickness was measured to be approximately 1.3 µm.

The BCB was patterned with AZ1813 photoresist as the etch mask. Samples were

etched in a SF6 /O2 plasma for different times, and after removing the photoresist,

were step-profiled to obtain the etch depths for different etch times, Fig. 4.8. The

etch rate was found to be 3 nm/s.

1200
SF /O : 3.33/30 sccm,
1000 6 2
300 mTorr, 60 W
Etch Depth (nm)

800

600

400

200
Etch Rate: ~ 3 nm/s
0
0 1 2 3 4 5 6 7
Time (min)

Figure 4.8. RIE etch characteristic of BCB in SF6 and O2 .

75
 The BCB etchback process was investigated on a AlGaAs/GaAs HBT pilot wafer.

The Ti/Au was formed by a lift-off process. The emitter mesa was defined by a wet

chemical etch in 1H2 SO4 :8H2 O2 :160H2 O using the emitter metal as etch mask. The

silicon nitride sidewall was formed by depositing and etching a nitride film in SF6 /Ar.

The silicon nitride deposition and etch conditions are the same as described above.

300 nm titanium and tungsten were sputtered for the base metal contact. BCB was

then deposited and etched back in SF6 /O2 using the same conditions as described

above. The etch stopped as the tungsten on top of the gold appeared, which can

be determined by step profiling of the emitter mesa. The SEM micrograph of this

etchback process is shown in Fig. 4.9. From Fig. 4.9, it can been seen that the

emitter mesa is lateral etched. It also can be seen that the emitter and base contacts

are separated by the silicon nitride sidewall, and the BCB surface is flat and can be

used as the etch mask in the following step of RIE etching off Ti/W on the top of

the emitter mesa.

Ti/Au
Ti/W
BCB
emitter mesa

SiNx
1.5 µm

Figure 4.9. SEM micrograph of the BCB etchback process on an AlGaAs/GaAs

HBT wafer.

76
4.3 DHBT Fabrication

The DHBT structures were grown by Intelligent Epitaxy Technology, Inc., Richard-

son, TX. The structures consist of a 50 nm InP emitter (Si: 3×1017 cm−3 ), a 40 nm

graded InGaAs base (C: 5×1019 cm−3 ), a 110 nm InP collector (Si: 3×1016 cm−3 ),

and a 200 nm InP subcollector (Si: 3×1019 cm−3 ). A 20 nm InGaAs setback layer (Si:

3×1016 cm−3 ) and a 24 nm compositional graded InAlGaAs layer (Si: 3×1016 cm−3 )

are used to suppress the current blocking at the base-collector junction. Heavily

doped InGaAs ohmic contact layers (Si: 3×1019 cm−3 ) are used for the emitter and

subcollector contacts.

The silicon nitride sidewall and BCB etchback process was used to fabricate the

InP/InGaAs DHBT. A representative measured Gummel plot is shown in Fig. 4.10.

From Fig. 4.10, the ideality factors of the base current and the collector current

are extracted to be 1.4 and 1.1, respectively, and which are close to those shown in

[49]. The dc current gain is extracted to be 51 at the base-emitter voltage of 0.8 V,

which achieves the expectation from the model that the maximum current gain is

50. The saturation current can also be extracted to be 34 fA, which is 20x larger

than that in the model.

77
 (20) µm
-3 2 2
10
I
C
V =0
BC
10-5
I , I (A)

I
B

-7 B
10
C

-9
10 n = 1.1
C
n = 1.4
B
-11
10
0.2 0.4 0.6 0.8
V (V)
BE

Figure 4.10. Measured Gummel plot of an InP/InGaAs double heterojunction bipo-

lar transistor with nitride sidewall process.

78
 A measured dc IC −VCE characteristic is shown in Fig. 4.11. The low slope

in the saturation region indicates a large emitter or collector resistance, which is

approximately 30 Ω. Since during the fabrication, the thin InGaAs subcollector

(10 nm) was unexpectedly exposed to the O2 plasma, this InGaAs layer may be

damaged and result in a large contact resistance.

Micro-scale devices were not yielded in this process because of the difficulties in

the bondpad process. Devices with emitter dimensions of 2 µm and 4 µm shorted

due to an excessive lateral etch in the via process (approximately 2.6 µm more than

expected). Increasing the plasma power and reducing the SF6 pressure is expected

to reduce the lateral etch. DC test results on larger area transistors have demon-

strated the feasibility of this nitride sidewall and BCB etchback process.

79
 30
2 2 400 µA
(20) µm
25
300
20
I (mA)

15 200
C

10
100
5
I =0
B
0
0 0.5 1 1.5 2 2.5 3
V (V)
CE

Figure 4.11. Measured common-emitter IC −VCE characteristic of an InP/InGaAs

double heterojunction bipolar transistor with nitride sidewall process.

80
4.4 RTD Fabrication

To establish a suitable high current density AlAs/InGaAs/InAs RTD growth

process, RTDs were grown at IntelliEPI, and fabricated and tested at Notre Dame.

For this RTD development, a non-self-aligned process was performed. A 200 nm

Ti/Au was formed by a lift-off process for the RTD emitter contact. The RTD

structure was etched in 1H2 SO4 :8H2 O2 :160H2 O using the emitter metal as a mask.

The etch is nonselective and was terminated by dead reckoning just pass the double

barrier. DC I−V measuremenst were performed to determine the RTD peak current

density. Shown in Fig. 4.12 is an example of the RTD DC test structure, where

two probes were placed on two adjacent pads with different areas. This test config-

uration is equivalent to two RTDs with different areas back-to-back connected, and

most of the voltage drop appears across the smaller emitter area device.

Probe
InGaAs

(20)2 µm2
(10)2 µm2 (10)2 µm2 80 x 475 µm2
Au Ti/Au Ti/Au
Au

RTD RTD
80 x 475 µm2
InGaAs
Au
InP HBT Emitter

Top view Side view

Figure 4.12. Resonant tunneling diode DC test structure, (a) top view, (b) side
view.

81
 The measured I−V characteristics of two RTDs with different barrier thickness

are shown in Fig. 4.13 (Other layer structures are shown in Table 2.1). A double

sweep technique (forward and backward sweep) was used due to the high series

resistance which originates from the high sheet resistance between the two test pads.

The estimated sheet resistance is approximately 25 Ω, which is close to the measured

result of 30 Ω. The forward bias corresponds to positive voltage on the smaller area

emitter pad. The peak voltages and currents for forward and reverse biases differ,

which indicates that the two barriers are asymmetric in thickness, composition or

strain. If the difference is solely due to thickness, since the barrier close to the RTD

emitter will have the greatest barrier height and will therefore limit the current the

most, the barrier close to the RTD collector, for both diodes, is thicker than the

barrier close to the RTD emitter. The peak current density increases by a factor of 4

between the two RTDs of Fig. 4.13 and the peak-to-valley current ratio degrades by

a factor of 4.5 when the barrier thickness reduces from 1.7 nm to 1.6 nm because the

tunneling current exponentially depends on the barrier width. For the TDT circuit,

a peak current density of 1 mA/µm2 and a peak-to-valley current ratio greater than

2 are needed, which are close to the test results on wafer 060745A3AA01. It is

reported that AlAs/InGaAs/InAs RTDs can achieve a peak current density of 1

mA/µm2 with a peak-to-valley current ratio of 13 [77].

82
 0.04
0.04
Wafer ID: 060729A2BA02 Wafer ID: 060745A3AA01
Barrier Thickness: 1.7 nm Barrier Thickness: 1.6 nm
0.02
(10) µm Emitter
2 2
0.02 (4) µm Emitter
2 2

Current (A)
Current (A)

0
0
-0.02

-0.02 PVR = 1.8

-0.04 PVR = 8.2
2 J = 1.8 mA/µm2
JP = 0.46 mA/µm P

-0.06 -0.04
-2 -1 0 1 2 -2 -1 0 1 2
Voltage (V)
Voltage (V)
(a) (b)

Figure 4.13. Measured current-voltage characteristics of the resonant tunneling

diodes on wafers: (a) 060729A2BA02, (b) 060745A3AA01.

83
 CHAPTER 5

SUMMARY AND CONCLUSIONS

5.1 Summary of Achievements

Since 1990s, TDT technology has demonstrated potential in high-speed and low-

power applications, such as logic, signal processing and communications. In this

research, two new ways of using tunnel diodes have been investigated. In the course

of this investigation, the TDT differential comparator and the TDT frequency trans-

lator were conceived, designed, and simulated.

In the TDT differential comparator circuit, two clocked tunnel diode pairs were

connected to the output ports of the differential amplifier, which enables the TDT

comparator to operate at a lower tail current while achieving the same specifications

compared to the transistor-only comparator. InP-based RTDs and HBTs were used

in the mode simulation. Simulation results showed power dissipation reduced by

approximately 1.6× with the TDT approach at 100 GHz clock frequency. In this

research, the designed TDT comparator was of special interest for the single-bit

oversampling DAC for DDS applications. The circuit topology can also be used in

other low-power applications, e.g., in flip-flops and logic gates.

The TDT frequency translation circuit, consisting of a tunnel diode, a transis-

tor, and an inductor, was designed for use in communication systems to upconvert

digital signals. This circuit functions as a frequency translator without a separate

mixer, local oscillator or PLLs, which can lead to smaller chip area and lower power

84
dissipation. Compared to the prior art based on tunnel diode alone, the added

transistor provides input/output isolation and power gain.

To fabricate these TDT circuits, a new self-aligned contact process utilizing

silicon nitride sidewalls and a BCB etchback was developed. This process allows

scaling of the transistor access resistance to achieve the frequency response needed

for circuit demonstrations and beyond. A baseline silicon nitride sidewall process

was established with an anisotropic etch for silicon nitride using an SF6 /Ar plasma,

and a baseline BCB etchback process was also developed. InP/InGaAs DHBTs were

fabricated using this sidewall process, and results from dc current-voltage measure-

ments demonstrated the feasibility of this new process. AlAs/InGaAs/InAs RTDs

were also fabricated in a non-self-aligned process and demonstrated a peak cur-

rent density of 1.8 mA/µm2 and a peak-to-valley current ratio of 1.8, close to the

requirements for circuit demonstrations.

In addition, a unified RTD small-signal equivalent circuit model was developed

to enable the circuit designs of this research. Analytic expressions for both the

quantum inductance and quantum capacitance were derived by a new approach.

The equivalent circuit model was verified by both DC I-V and microwave frequency

S-parameter measurements on Raytheon’s AlAs/InGaAs/InAs RTDs from 45 MHz

to 30 GHz. Based on this new finding, the high frequency response of the RTD

was investigated, which allowed a better understanding of the highest operation

frequency that the RTD can achieve.

As shown from the results of this investigation, TDT technology has advantages

in increasing circuit speed, reducing power dissipation and reducing circuit compo-

nents for specific applications. However, there remain challenges and difficulties to

be overcome for the wider implementation of TDT technology. First, the operating

point of the TDT circuit critically depends on the peak current of the tunnel diode.

85
Since the peak current depends exponentially on the barrier thickness for RTDs and,

if one considers the possible extension of this technology to Si, the junction width

for the Esaki diode, it remains difficult to obtain reproducible device characteristics.

Accurately controlled uniform barrier thickness or junction width must be achieved.

In III-V technology, RTDs grown by MBE have shown well controlled and uniform

barriers, however these must be matched to the transistor currents in order to yield

circuits. In Si technology, no CMOS-compatible process has yielded characteristics

for Esaki diodes suitable for other than small scale circuit development. Further-

more, compared to CMOS digital circuits, the off current in TDT logic circuits will

be a few orders of magnitude higher because the tunnel diode cannot be completely

shut off. For these reasons, the integration of the tunnel diode with CMOS for

digital applications appears less attractive.

Reported high-speed III-V HBTs with fT and fmax over 300 GHz all operate at

a high current densities (> 4 mA/µm2 ) [46, 47, 48, 49]. Tunnel diode will need to

have a comparable peak current density, otherwise, the transistor has to operate at

a lower current density which will slow down the entire circuit speed. For silicon

tunnel diodes, the highest reported peak current density is 1.5 mA/µm2 , grown by

MBE [87]. For III-V RTDs, the highest reported peak current density is 6.8 mA/µm2

[43]. As the RTD operates at such high current density, the thermal stability needs

to be investigated because the RTD is usually grown on top of a transistor, and

there are several layers in the transistor which have poor thermal conductivities.

The thermal reliability of the tunnel diode is not well explored to date.

Overall, the TDT technology in III-V compound semiconductor is close to pro-

duction for specific applications which require simultaneously high-speed and low-

power. In silicon, until a manufacturable process is developed for integrating the

tunnel diode with CMOS, CMOS technology will remain preferable.

86
5.2 Recommendations for Future Research

The Gummel-Poon model is used for the circuit simulation in this research.

Since the GP model is developed for Si BJTs, it can not accurately predict some

characteristics of InP HBTs. For examples, the base-collector capacitance are not

bias dependant as the collector is fully depleted. The conduction band barriers

at the base-emitter and base-collector junction of the InP/InGaAs DHBT are bias

dependant and will affect the base and collector current-voltage characteristics. A

more accurate SPICE model for the InP HBTs needs to be investigated for circuit

simulation and the model parameters should be extracted from the measurements

of HBTs.

The via process needs to be improved to yield the TDT circuits. During the via

etch process, the lateral etch needs to be minimize, thereby preventing a shorted

emitter-base contacts formed in the bondpad process. For the 2 µm emitter transis-

tor, the lateral etch should be less than 100 nm and for the 4 µm emitter transistor,

the lateral etch should be less than 1 µm. To increase the transistor speed, further

scaling of the transistor size is required. The sidewall process for submicron devices

can be investigated using the e-beam lithography.

The process developed in this work self-aligns the base-emitter junction. By

extension it is possible to self-align both base and collector to the emitter metal-

lization. Here, a new fully self-aligned process using silicon nitride-sidewalls and

metal reactive ion etching has been conceived for the submicron HBT fabrication.

The fabrication sequence is outlined as follows, shown in Fig. 5.1 and 5.2. The pro-

cess starts with a blanket deposition of tungsten. The tungsten is patterned with

photoresist as the etch mask. The tungsten is then RIE etched to form the emitter

contact, Fig. 5.1(a). The emitter mesa is etched using the emitter contact as the

etch mask, Fig. 5.1(b). A silicon nitride sidewall is formed by depositing a PECVD

87
Si3 N4 film, followed by an anisotropic RIE etching, Fig. 5.1(c). To form the base

contact, tungsten is first blanket sputtered on the wafer, Fig. 5.1(d), followed by an

anisotropic RIE etching to form the tungsten base contact, Fig. 5.1(e).

Next the base and collector are etched using the tungsten as the etch mask,

Fig. 5.2(a). Forming a silicon nitride sidewall around the base mesa using a similar

process as that for the emitter sidewall, Fig. 5.2(b). Finally, blanket sputtering, Fig.

5.2(c), and anisotropically RIE etching of tungsten to form the collector contact,

Fig. 5.2(d).

This proposed process has several notable attributes. Only one mask is required

in the front-end process, which reduces lithography-induced errors. The fully self-

aligned contacts enable the extrinsic resistances to be minimized. The metal lift-off

process at the emitter, base, and collector levels is eliminated to improve the yield.

88
 400 nm

Emitter
Metallization Base
W
(a)
Emitter
Collector
Subcollector

Base
W
Emitter Etch
Metal as Etch Mask Emitter
(b) Collector
Subcollector

Si3N4 sidewall

Silicon Nitride Base

W
Sidewall Formation
(c) Emitter
Collector
Subcollector

Base Metal
Deposition Base
(d)
Emitter
Collector
Subcollector

Si3N4 sidewall

Base
Tungsten RIE Etch W
W
W
(e) Emitter
Collector
Subcollector

Figure 5.1. Scale drawings of a fully self-aligned HBT process. These drawings
describe the emitter metallization, the emitter mesa etch, the silicon nitride sidewall
formation, and the base metal deposition and etch.

89
 Si3N4 sidewall

Base and Collector

W
Mesa Etch W W
(a) Emitter
Collector
Subcollector

Si3N4 sidewall

Silicon Nitride
W
Sidewall Formation W W
(b) Emitter
Collector
Subcollector

Collector Metal
Deposition
(c)
Emitter
Collector
Subcollector

Si3N4 sidewall

Tungsten RIE Etch W

W
W
(d) W Emitter W

Collector
Subcollector

Figure 5.2. Scale drawings of a fully self-aligned HBT process. These drawings
describe the base and collector mesa etch, and the collector metal deposition and
etch.

90
 APPENDIX A

FABRICATION PROCESS FLOW

Purpose: Fabrication of TDT circuits using the HBT RTD mask.

RTD Emitter Metallization: RTD Emitter Mask − Layer 1

• Inspect the wafer under microscope

• Check the resistivity of the DI water MΩ
• Solvent clean
◦ Soak in hot acetone 3 min - hot plate 60 ◦ C
◦ Soak in hot methanol 3 min - hot plate 60 ◦ C
• Prepare 2-1 etchant 2H2 O:1HCl - add acid to water - wait to cool
◦ Removes metals, lifts off organics, leaves oxide rich surface
• Rinse in DI 1 min
• Soak in 2-1 15 s
• DI rinse 10 s
• Blow dry with N2

• AZ5214E photoresist process

◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C
◦
◦ Hot plate reversal bake 110 C 60 s
◦ Measure intensity W/cm2

91
 ◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Soak in 2-1 5 s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Load without delay

• Evaporate
◦ nm Ti (2 Å/s)
◦ nm Pt (2 Å/s)
◦ nm Au (5 Å/s); Au spitting

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, RTD emitter metal height Å

RTD Emitter Mesa Etching:

• Solvent clean if necessary
• Mix 1H2 SO4 :8H2 O2 :160H2 O etchant (add acid to water). This needs to cool
before etching (InGaAs etch rate: 40-60 Å/s)
• Rinse 1 min in DI
• Etch in 1-8-160 s
◦ Ideal etch depth Å. Ok if between and Å
• Rinse 1 min in DI
• Blow dry with N2
• Step profile, RTD mesa height Å

92
HBT Emitter Metallization: HBT Emitter Mask − Layer 2

• Solvent clean if necessary

• Prepare 2-1 etchant 2H2 O:1HCl - add acid to water - wait to cool

• AZ5214E photoresist process

◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C
◦
◦ Hot plate reversal bake 110 C 60 s
◦ Measure intensity W/cm2
◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Soak in 2-1 5 s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Load without delay

• Evaporate
◦ nm Ti (2 Å/s)
◦ nm Pt (2 Å/s)
◦ nm Au (5 Å/s); Au spitting

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, HBT emitter metal height Å

93
RTD Mesa Etching: RTD Mesa Etching Mask − Layer 3

• Solvent clean if necessary

• Wet chemical etching
◦ Mix 1H2 SO4 :8H2 O2 :160H2 O etchant (add acid to water). This needs to
cool before etching (InGaAs etch rate: 40-60 Å/s)
◦ Mix 1H3 PO4 :1HCl:1H2 O etchant (add acid to water). This needs to cool
before etching (InP etch rate: 50-70 Å/s)
◦ Rinse 1 min in DI
◦ Etch in 1-8-160 to a total depth of Å ( s), selectivity stops
on InP
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Step profile, HBT emitter height Å
◦ Etch in 1-1-1 to a total depth of Å ( s), selectivity stops on
the base
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Step profile, HBT emitter height Å

Sidewalls and BCB Process:

• Solvent clean if necessary

• Deposit SiNx 3000 Å (Recipe: SiH4 40 sccm, NH3 4 sccm; 200 W,
500 mTorr. Deposition rate: ∼ 20 nm/min)
• RIE etch
◦ O2 plasma clean if necessary
◦ Condition chamber (SF6 /Ar 2/18 sccm @ 15 mTorr, 150 W for 2 min)
◦ Etch (SF6 /Ar 2/18 sccm @ 15 mTorr, 150 W for min,
DC V); (SiNx etch rate: ∼ 17 Å/s)
◦ Selectivity stops on the HBT base
◦ Inspect under SEM if necessary

• Tungsten sputtering
◦ Base pressure Torr; table spacing
◦ W sputtering for min: Ar: 148 sccm @ 20 mTorr, forward
power: W, reflect power: W, DC: V

94
 ◦ Step profile, deposition thickness Å
◦ Step profile, HBT emitter height Å

• Solvent clean if necessary

• Spin ADP, 1000 rpm for 10s, followed by 5000 rpm for 30 s
• Spin BCB 3022-35, 1000 rpm for 10 s, followed by 5000 rpm for 30s (∼ 1.3
µm)
• Cure in PECVD system
◦ N2 500 sccm @ 1000 mTorr
◦ 250 ◦ C for 1 hour
• Step profile, HBT emitter height Å
• BCB RIE etch
◦ Condition chamber (SF6 /O2 3.33/30 sccm @ 300 mTorr, 60 W for 2 min)
◦ Etch (SF6 /O2 3.33/30 sccm @ 300 mTorr, 60 W for min,
DC V); (BCB etch rate 1747 Å/m)
◦ Etch stops when W appears
◦ Step profile, HBT emitter height Å
• W RIE etch
◦ Condition chamber (SF6 /Ar 10/35 sccm @ 190 mTorr, 150 W for 2 min)
◦ Etch (SF6 /Ar 10/35 sccm @ 190 mTorr, 150 W for min,
DC V); (Etch rate ∼ 500 Å/s; Etch stops when Au appears)
• BCB removal
◦ Condition chamber (SF6 /O2 2/38 sccm @ 300 mTorr, 60 W for 2 min)
◦ Etch (SF6 /O2 2/38 sccm @ 300 mTorr, 60 W for min,
DC V)
◦ Step profile, HBT emitter height Å
• Measure HBT emitter-base resistance, make sure emitter and base are not
shorted (base-emitter junction diodes)
• Inspect − optical, SEM test structures if necessary

95
HBT Base Contacts: HBT Base Mask − Layer 4

• Solvent clean if necessary

• Photoresist process
◦ Spin PR 1813, 800 rpm for 3s, followed by 2500 rpm for 30 sec
◦ Soft bake 90 o C for 1 min
◦ Expose s
◦ Develop in AZ327 for 30 sec, redevelop as necessary s
◦ Rinse in DI water
◦ Blow dry with N2
◦ Bake 120 o C for 1 min

• W RIE etch
◦ Condition chamber (SF6 /Ar 10/35 sccm @ 190 mTorr, 150 W for 2 min)
◦ Etch (SF6 /Ar 10/35 sccm @ 190 mTorr, 150 W for min,
DC V); (Etch rate ∼ 500 Å/s; Etch stops at the base)

• Wet chemical etching − collector definition etch

◦ Mix 1H2 SO4 :8H2 O2 :160H2 O etchant (add acid to water). This needs to
cool before etching (InGaAs etch rate: 40-60 Å/s)
◦ Mix 1H3 PO4 :1HCl:1H2 O etchant (add acid to water). This needs to cool
before etching (InP etch rate: 50-70 Å/s)
◦ Rinse 1 min in DI
◦ Etch in 1-8-160 to a total depth of Å ( s), selectivity stops
on InP collector
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Etch in 1-1-1 to a total depth of Å ( s), selectivity stops on
InGaAs subcollector
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Remove the photoresist by soaking in acetone/methanol
◦ Step profile, measure etch depth Å

96
Isolation Etching: Isolation Etching Mask − Layer 5

• Solvent clean if necessary

• Photoresist process
◦ Spin PR 1813, 800 rpm for 3s, followed by 2500 rpm for 30 sec
◦ Soft bake 90 o C for 1 min
◦ Expose s
◦ Develop in AZ327 for 30 sec, redevelop as necessary s
◦ Rinse in DI water
◦ Blow dry with N2
◦ Bake 120 o C for 1 min

• Wet chemical etching

◦ Mix 1H2 SO4 :8H2 O2 :160H2 O etchant (add acid to water). This needs to
cool before etching (InGaAs etch rate: 40-60 Å/s)
◦ Mix 1H3 PO4 :1HCl:1H2 O etchant (add acid to water). This needs to cool
before etching (InP etch rate: 50-70 Å/s)
◦ Rinse 1 min in DI
◦ Etch in 1-8-160 to a total depth of Å ( s), selectivity stops
on InP subcollector
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Etch in 1-1-1 to a total depth of Å ( s), selectivity stops on
InGaAs etch stop layer
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Etch in 1-8-160 to a total depth of Å ( s), selectivity stops
on InP substrate
◦ Rinse 1 min in DI
◦ Blow dry with N2
◦ Remove the photoresist by soaking in acetone/methanol
◦ Step profile, measure etch depth Å

97
Resistor Process: Resistor Mask − Layer 6

• Solvent clean if necessary

• AZ5214E photoresist process
◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C
◦ Hot plate reversal bake 110 ◦ C 60 s
◦ Measure intensity W/cm2
◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Load without delay

• Evaporate
◦ 30 nm Ti (2 Å/s)

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, resistor metal thickness Å

HBT Collector Metallization: HBT Collector Mask − Layer 7

• Solvent clean if necessary

• AZ5214E photoresist process
◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C

98
 ◦ Hot plate reversal bake 110 ◦ C 60 s
◦ Measure intensity W/cm2
◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Load without delay

• Evaporate
◦ 20 nm Ti (2 Å/s)
◦ 10 nm Pt (2 Å/s)
◦ 150 nm Au (5 Å/s); Au spitting

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, HBT collector metal height Å

Capacitor Process: Capacitor Mask − Layer 8

• Solvent clean if necessary

• Deposit SiNx 3000 Å (Recipe: SiH4 40 sccm, NH3 4 sccm; 200 W,
500 mTorr. Deposition rate: ∼ 20 nm/min)

• AZ5214E photoresist process

◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C
◦
◦ Hot plate reversal bake 110 C 60 s
◦ Measure intensity W/cm2

99
 ◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Load without delay

• Evaporate
◦ 20 nm Ti (2 Å/s)
◦ 80 nm Au (5 Å/s); Au spitting

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, capacitor top metal height Å

Via Process: Via Mask − Layer 9

• Solvent clean if necessary

• Photoresist process
◦ Spin PR 1813, 800 rpm for 3s, followed by 2500 rpm for 30 sec
◦ Soft bake 90 o C for 1 min
◦ Expose s
◦ Develop in AZ327 for 30 sec, redevelop as necessary s
◦ Rinse in DI water
◦ Blow dry with N2

• Via etching
◦ O2 plasma clean if necessary
◦ Condition chamber (SF6 20 sccm @ 20 mTorr, 50 W for 2 min)

100
 ◦ Etch (SF6 20 sccm @ 20 mTorr, 50 W for min,
DC V); (Etch rate: 18 Å/s)
◦ Etch stops when Au contacts appear
◦ Remove the photoresist by soaking in acetone/methanol
◦ Step profile, etch depth Å

Bondpad Process: Bondpad Mask − Layer 10

• Solvent clean if necessary

• AZ5214E photoresist process
◦ Spin AZ5214E 5000 rpm 30 s (1.2 µm)
◦ Hot plate soft bake 105 ◦ C 30 s
◦
◦ Expose s, temperature C
◦
◦ Hot plate reversal bake 110 C 60 s
◦ Measure intensity W/cm2
◦ Expose 300 mJ/cm2 , s
◦ Develop in AZ327 30 s, redevelop as necessary s
◦ DI rinse 1 min
◦ Blow dry with N2
◦ Inspect
◦ UVO clean 2 min
◦ Load without delay

• Evaporate
◦ 20 nm Ti (2 Å/s)
◦ 300 nm Au (5 Å/s); Au spitting

• Lift-off
◦ Soak in hot acetone − adjust to near boil
◦ Vigorous spray with acetone
◦ Rinse/spray with methanol
◦ Blow dry with N2
◦ Microscope inspect
◦ Step profile, deposition thickness Å

101
 APPENDIX B

ESTIMATION OF THE HETEROJUNCTION BIPOLAR TRANSISTOR

CUT-OFF FREQUENCY AND MAXIMUM OSCILLATION FREQUENCY

The cut-off frequency, fT , and maximum oscillation frequency, fmax , are given by
q
1 1 fT
fT = 2π τE +τB +τDC +τC
, and f max = 8πRB CC
, respectively, where τE is the emitter

charging time, τB is the base transit time, τDC is the collector depletion layer transit

time, τC is the collector charging time, RB is the base resistance, and CC is the

collector capacitance. The time constants τE , τB , τDC , and τC are given as [60]
VT
τE = (RE + )(CE + CC ) (B.1)
JE WE LE
XB2
τB = (B.2)
2VT µnB
XC
τDC = (B.3)
2νsat
τC = RC CC (B.4)

where RE is the emitter resistance, VT is the thermal voltage, JE is the emitter

current density, WE is the emitter width, LE is the emitter length, CE is the emitter

capacitance, XB is the base thickness, µnB is the minority electron mobility, XC is

the collector thickness, νsat is the electron saturation velocity, and RC is the collector

resistance. The collector is so thin that we assume it is fully depleted.

The base-emitter junction depletion width is given by [66]

r
2²r InP
XdepE = (ΦBE − VBE ) (B.5)
qNDE

102
where ²r InP is the dielectric constant of InP, NDE is the emitter doping concen-

tration, ΦBE is the base-emitter junction built-in potential, and VBE is the base-

emitter bias voltage. The base-emitter junction capacitance, CE , is estimated to

be CE = ²r InP WE LE /XdepE , where WE and LE are the width and length of the

emitter, respectively. The base-collector junction capacitance, CC , is estimated to

be CC = ²r InP WB LB /XC , where we assume the collector is fully depleted, and WB

and LB are the width and length of the base, respectively.

The emitter, base, and collector resistance are given by [66]

RE = REepi + REE (B.6)

RB = Rbi + Rbx + RBB (B.7)

RC = RCepi + RSCepi + RSCx + RCC (B.8)

where REepi is the emitter epitaxial resistance, REE is the emitter contact resistance,

Rbi is the intrinsic base resistance, Rbx is the extrinsic base resistance, RBB is the

base contact resistance, RCepi is the collector epitaxial resistance, RSCepi is the sub-

collector epitaxial resistance, RSCx is the extrinsic sub-collector resistance, and RCC

is the collector contact resistance. Above resistances are given by [66]

Xcap (XE − XdepE )

REepi = ρcap + ρE (B.9)
WE LE WE LE
ρσE
REE = (B.10)
WE L E
ρB
Rbi = (B.11)
8XB (WE /LE + LE /WE )
ρB
Rbx = (B.12)
XB nbx
r µ r ¶
1 ρB ρσB ρB
RBB = coth WBM (B.13)
LBM XB XB ρσB
XC
RCepi = ρC (B.14)
WB LB

103
 XSC
RSCepi = ρ( SC) (B.15)
WB L B
ρSC
RSCx = (B.16)
XSC nSCx
r µ r ¶
1 ρSC ρσSC ρSC
RCC = coth WCM (B.17)
LCM XSC XSC ρσSC
where ρcap , ρE , ρB , ρC , and ρSC are the resistivity of the emitter cap, the emitter, the

base, the collector, and the sub-collector, respectively, ρσE , ρσB , and ρσSC are the

contact resistivity of the emitter, the base, and the sub-collector, respectively. Xcap ,

XE , XC , and XSC are the thickness of the emitter cap, the emitter, the collector,

and the sub-collector, respectively. nbx and nSCx are the number of squares in the

extrinsic base and in the sub-collector extrinsic region, respectively. WEM , LEM

and WCM , LCM are the width and length of the base metal and the collector metal,

respectively.

104
 APPENDIX C

DESIGN APPROACH USING TUNNEL DIODES FOR LOWERING POWER

IN DIFFERENTIAL COMPARATORS

The following article is the reprint from IEEE Transactions on Circuits and

Systems − II: Express Briefs, vol. 52, no.9, pp. 572-575, Sep. 2005.

I would like to express special gratitude to the co-author A. Seabaugh.

105
106
107
108
109
 APPENDIX D

TUNNEL DIODE/TRANSISTOR DIFFERENTIAL COMPARATORS

The following article is the reprint from International Journal of High Speed

Electronics and Systems, vol. 14, no. 3, pp. 640-645, 2004.

I would like to express special gratitude to the co-authors S. Sutar and A.

Seabaugh.

110
111
112
113
114
115
116
 APPENDIX E

UNIFIED AC MODEL FOR THE RESONANT TUNNELING DIODE

The following article is the reprint from IEEE Transactions on Electron Devices,

vol. 51, no. 5, pp. 653-657, 2004.

I would like to express special gratitude to the co-authors A. Seabaugh, P. Cha-

hal, and F. Morris.

117
118
119
120
121
122
 APPENDIX F

LOW POWER, HIGH SPEED, AND MIXED-SIGNAL TUNNELING DEVICE

TECHNOLOGY

The following article is the reprint from Proceedings of International COE Work-

shop on Nano Processes and Devices, and their Applications, pp. 37-38, 2005.

I would like to express special gratitude to the co-authors A. Seabaugh, S. Sutar,

Q. Zhang, W. Zhao, J. Zhao, Y. Yan, D. Wheeler, B. Wu, S. Kabeer, Z. Racz, and

P. Fay.

123
124
125
 APPENDIX G

VERTICAL SILICON TUNNEL DIODE ON HIGH RESISTIVITY SILICON

The following article is the reprint from 62nd Device Research Conference Digest,

pp. 27-28, 2004.

I would like to express special gratitude to the co-authors Y. Yan, J. Zhao, W.

Zhao, and A. Seabaugh.

126
127
128
 APPENDIX H

UNIFIED PHYSICS-BASED AC MODEL FOR THE RESONANT TUNNELING

DIODE

The following article is the reprint from 61st Device Research Conference, Late

News Papers, pp. 12-13, 2003.

I would like to express special gratitude to the co-author A. Seabaugh.

129
 Unified Physics-Based AC Model for the Resonant Tunneling Diode

Qingmin Liu and Alan Seabaugh

Department of Electrical Engineering, University of Notre Dame, IN 46556-5637
Phone: (574) 631-4473, Email: seabaugh.1@nd.edu

The resonant tunneling diode (RTD) stands as the fastest, large-signal semiconductor switching
device with measured slew rates as high as 300 mV/ps [1]; The RTD has long been explored for
use in triggers, quantizers, oscillators, memory cells, and A/D converters [e.g. 2]. Circuit designs
using RTDs require an accurate small-signal equivalent circuit model, which is suitable and easy
to incorporate into computer-aided design (CAD) software. Two equivalent circuit models for the
RTD are commonly used: a series-inductance model [3] and a parallel-inductance model [4]. In
this paper, we present a new physics-based approach, which provides the form of the circuit and
analytic expressions for the bias dependent quantum inductance and capacitance. This model
unifies the previous models by Brown, et al. for the quantum inductance [4] and by Lake and
Yang [5] for the quantum capacitance, and extends the RTD SPICE model of Broekaert, et al. [2].

Our derivation parameterizes the sequential tunneling process between emitter and quantum well,
and quantum well and collector in terms of the emitter and quantum well charges and the
tunneling rates. We show that the change in the quantum well charge lags behind the bias
change, resulting in the existence of a quantum inductance in the tunneling current path, which is
given by LQ = τ / G D , where τ is the electron lifetime in the quantum well, and GD is the
differential conductance of the RTD. The change in the quantum well charge also induces a
change in the image charge in the collector that results in an additional quantum capacitance,
which adds to the geometrical capacitance. The total capacitance is derived to be
-1
C P = C 0 − G D / v C , where C0 is the geometrical capacitance, and vC is the electron escape rate (s )
from the quantum well to the collector.

Both dc current-voltage (I-V) and microwave frequency S-parameter measurements were made
on 1.6×1.6 µm2 AlAs/InGaAs/AlAs RTDs. Bias dependent (0 − 0.81 V) S-parameters were
measured from 45 MHz to 30 GHz using an Agilent 8510XF vector network analyzer with port
power of –33 dBm. Equivalent circuit parameters were extracted by fitting the measured S-
parameter data over the entire frequency and bias range. Close agreement between calculation
and measured data is obtained. The dc and microwave frequency measurement and
characterization of AlAs/InGaAs/AlAs RTDs support the model theory.

We would like to thank Prem Chahal, Frank Morris, and Gary Frazier (Raytheon) for supplying
the RTDs and Patrick Fay (Notre Dame) for valuable discussions. This work was sponsored in
part by a Raytheon University Research Grant and the Office of Naval Research.

[1] E. Özbay, et al., IEEE Electron Dev. Lett. 14, 400-402 (1993).
[2] T. Broekaert, et al., IEEE J. Solid State Circ. 33, 1342-1349 (1998).
[3] J. M. Gering, et al., J. Appl. Phys. 61, 271-276 (1987).
[4] E. R. Brown, et al., Appl. Phys. Lett. 54, 934-936 (1989).
[5] R. Lake, et al., IEEE Trans. Electron Dev., 50, 785-789 (2003)

130
 2
Quantum 20

Differential Conductance G d (mS)

Emitter Collector
1.5 Well
15
1
Energy (eV)
0.5 J J 10
E C
E
F
0 E
0 5
E − eV
-0.5 F

InGaAs

InGaAs

InGaAs

InGaAs
0

AlAs

AlAs
InAs
-1

-1.5 -5
0 5 10 15 20
Position (nm) -10
Fig. 1. Resonant tunneling diode computed energy 0 0.2 0.4 0.6 0.8
band diagram. Voltage (V)
Fig. 5. Comparison of the differential conductance
3 extracted from the dc I−V measurement (solid line)
with S-parameter measurement (circles).
2
8

Reciprocal of Inductance L Q (nH-1)

1
Current (mA)

6
0
4 τ = 2.58 ps
-1
2 2 2
-2 (1.6) µm

0
-3

-0.8 -0.4 0 0.4 0.8 -2

Voltage (V)
-4
Fig. 2. AlAs/InGaAs/InAs/InGaAs/AlAs resonant
tunneling diode current-voltage characteristic. -6
0 0.2 0.4 0.6 0.8
LQ Gd Voltage (V)
Rs
Fig. 6. The reciprocal of the quantum inductance,
Cp LQ, vs. bias showing good agreement between
measurement (circles) and calculation (solid line).
Fig. 3. Parallel-inductance equivalent circuit. 40
-0.2 0.02
35
0.01
Capacitance C p (fF)

-0.22
0 30
-0.24
Real S11

Imag S11

-0.01
25
-0.26
-0.02 C = 29.3 fF
0
20 1/ν = 0.79 ps
-0.28 -0.03 C

-0.3 -0.04 15
0 5 10 15 20 25 30 0 0.2 0.4 0.6 0.8
Voltage (V)
Frequency (GHz)

Fig. 4. Comparison of the measured (circles) and Fig. 7. The total capacitance CP vs. bias showing
simulated (solid line) S-parameters at V = 0.45 V close agreement between measurement (circle) and
(near the center of the NDR region). calculation (solid line).

131
 APPENDIX I

PERFORMANCE-AUGMENTED CMOS USING BACK-END UNIAXIAL

STRAIN

The following article is the reprint from 60th Device Research Conference Digest,

pp. 41-42, 2002.

I would like to express special gratitude to the co-authors R. Belford, W. Zhao,

J. Potashnik, and A. Seabaugh.

132
133
134
 APPENDIX J

SILICON-BASED TUNNEL DIODES AND INTEGRATED CIRCUITS

The following article is the reprint from 4th International Workshop on Quantum

Functional Devices, pp. 5-8, 2000.

I would like to express special gratitude to the co-authors A. Seabaugh, Z Hu,

D. Rink, and J. Wang.

135
136
137
138
139
 REFERENCES

[1] P. M. Thompson, “The place of the tunnel diode in solid-state circuits,” 1962
IEEE International Solid-State Circuits Conference Technical Digest, pp. 66-67,
Feb. 1962.
[2] W. F. Chow, “Principles of Tunnel Diode Circuits,” John Wiley, New York,
1964.
[3] J. O. Scanlan, “Analysis and Synthesis of Tunnel Diode Circuits,” John Wiley,
New York, 1966.
[4] L. L. Chang, L. Esaki, and R. Tsu, “Resonant tunneling in semiconductor
double barriers,” Applied Physics Letters, vol. 24, pp. 593-595, Jun. 1974.
[5] R. G. Swartz, “Technology commentary-In perspective: The tunnel diode,”
1986 IEEE International Solid-State Circuits Conference Technical Digest, pp.
277-280, Feb. 1986.
[6] R.W. Keyes, “Physics of digital devices,” Reviews of Modern Physics, vol. 61,
pp. 279-289, Apr. 1989.
[7] Y. Watanabe, Y. Nakasha, K. Imanishi, and M. Takikawa, “Monolithic in-
tegration of InGaAs/InAlAs resonant tunneling diode and HEMT for single-
transistor cell SRAM application,” in 1992 IEDM Technical Digest, pp. 475-478,
Dec. 1992.
[8] A. C. Seabaugh, E. A. Beam, III, A. H. Taddiken, J. N. Randall, and Y.
Kao, “Co-integration of resonant tunneling and double heterojunction bipolar
transistors on InP,” IEEE Electron Device Letters, vol. 14, pp. 472-474, Oct.
1993.
[9] K. Maezawa, H. Matsuzaki, M. Yamamoto, and T. Otsuji, “High-speed and low-
power operation of a resonant tunneling logic gate MOBILE,” IEEE Electron
Device Letters, vol. 19, pp. 80-82, Mar. 1998.
[10] A. Seabaugh, B. Brar, T. Broekaert, F. Morris, P. van der Wagt, and G. Frazier,
“Resonant-tunneling mixed-signal circuit technology,” Solid-State Electronics,
vol. 43, pp. 1355-1365, Aug. 1999.
[11] S. Choi, B. Lee, T. Kim and K. Yang, “CML-type monostable bistable logic
element (MOBILE) using InP-based monolithic RTD/HBT technology,” Elec-
tronics Letters, vol. 40, pp. 792-793, Jun. 2004.

140
[12] M. W. Dashiell, R. T. Troeger, S. L. Rommel, T. N. Adam, P. R. Berger, C.
Guedj, J. Kolodzey, A. C. Seabaugh, and R. Lake, “Current-voltage charac-
teristics of high current density silicon Esaki diodes grown by molecular beam
epitaxy and the influence of thermal annealing,” IEEE Transactions on Elec-
tron Devices, vol. 47, pp. 1707-1714, Sep. 2000.
[13] U. Auer, W. Prost, M. Agethen, F.-J. Tegude, R. Duschl, and K. Eberl, “Low-
voltage MOBILE logic module based on Si/SiGe interband tunneling diodes,”
IEEE Electron Device Letters, vol. 22, pp. 215-217, May 2001.
[14] J. Wang, D. Wheeler, Y. Yan, J. Zhao, S. Howard, and A. Seabaugh, “Silicon
tunnel diodes formed by proximity rapid thermal diffusion,” IEEE Electron
Device Letters, vol. 24, pp. 93-95, Feb. 2003.
[15] L.-E. Wernersson, S. Kabeer, V. Zela, E. Lind, J. Zhang, W. Seifert, T. Kosel,
and A. Seabaugh, “SiGe Esaki tunnel diodes fabricated by UHV-CVD growth
and proximity rapid thermal diffusion,” Electronics Letters, vol. 40, pp. 83-85,
Jan. 2004.
[16] S. Sudirgo, R.P. Nandgaonkar, B. Curanovic, J. Hebding, K.D. Hirschman, S.S.
Islam, S.L. Rommel, S.K. Kurinec, P.E. Thompson, N. Jin, and P.R. Berger,
“Monolithically integrated Si/SiGe resonant interband tunneling diodes/CMOS
MOBILE latch with high voltage swing,” Proceedings of 2003 International
Semiconductor Device Research Symposium. pp. 22-23, Dec. 2003.
[17] H. Mizuta and T. Tanoue, “The physics and applications of resonant tunneling
diodes,” Cambridge University Press, New York, pp. 142-143, 1995.
[18] J. N. Schulman, H. J. De Los Santos, and D. H. Chow, “Physics-based RTD
current-voltage equation,” IEEE Electron Device Letters, vol. 17, pp. 220-222,
May. 1996.
[19] T. P. E. Broekaert, B. Brar, J. P. A van der Wagt, A. C. Seabaugh, F. J.
Morris, T. S. Moise, E. A. Beam, and G. A. Frazier, “A monolithic 4-Bit
2-Gsps resonant tunneling analog-to-digital converter,” IEEE Journal of Solid-
State Circuits, vol. 33, pp. 1342-1349, Sep. 1998.
[20] J. M. Gering, D. A. Crim, D. G. Morgan, P. D. Coleman, W. Kopp, and H.
Morkoc, “A small-signal equivalent-circuit model for GaAs-Alx Ga1−x As reso-
nant tunneling heterostructures at microwave frequencies,” Journal of Applied
Physics, vol. 61, pp. 271-276, Jan 1987.
[21] E. R. Brown, C. D. Parker, and T. C. L. G. Sollner, “Effect of quasibound-state
lifetime on the oscillation power of resonant tunneling diodes,” Applied Physics
Letters, vol. 54, pp. 934-936, Mar. 1989.
[22] Q. Liu, A. Seabaugh, P. Chahal, and F. Morris, “Unified ac model for the
resonant tunneling diode,” IEEE Transactions on Electron Devices, vol. 51,
pp. 653-657, May 2004.
[23] R. Lake and J. Yang, “A physics based model for the RTD quantum capac-
itance,” IEEE Transactions on Electron Devices, vol. 50, pp. 785-789, Mar.
2003.

141
[24] A. Seabaugh, X. Deng, T. Blake, B. Brar, T. Broekaert, R. Lake, F. Morris,
and G. Frazier, “Transistors and tunnel diodes for analog/mixed-signal circuits
and embedded memory,” 1998 IEDM Technical Digest, pp. 429-432, Dec. 1998.
[25] P. van der Wagt, A. Seabaugh, and E. Beam, III, “RTD/HFET low standby
power SRAM gain cell,” IEEE Electron Device Letters, vol. 19, pp. 7-9, Jan.
1998.
[26] J. N. Schulman, “Extension of Tsu-Esaki model for effective mass effects in
resonant tunneling,” Applied Physics Letters, vol. 72, pp. 2829-2831, Jun. 1998.
[27] H. Matsuzaki, J. Osaka, T. Itoh, K. Sano, and K. Murata, “Monolithic integra-
tion of resonant tunneling diodes, schottky barrier diodes and 0.1-m-gate high
electron mobility transistors for high-speed ICs,” Japanese Journal of Applied
Physics, vol. 40, pp. 2186-2190, Apr. 2001.
[28] W. Otten, P. Glösekötter, P. Velling, A. Brennemann, W. Prost, K. F. Goser,
and F.-J. Tegude, “InP-based monolithically integrated RTD/HBT mobile for
logic circuits,” in Proceedings of 2001 IEEE International Conference on In-
dium Phosphide and Related Materials, pp. 232-235, May 2001.
[29] H. J. De Los Santos, K. K. Chui, D, H, Chow, and H. L. Dunlap, “An efficient
HBT/RTD oscillator for wireless applications,” IEEE Microwave and Wireless
Components Letters, vol. 11, pp. 193-195, May. 2001.
[30] S. Choi, B. Lee, T. Kim, and K. Yang, “A varactor-tuned MMIC VCO using
InP-based RTD/HBT technology,” Microwave and Optical Technology Letters,
vol. 45, pp. 408-411, Jun. 2005.
[31] Y. Kawano, Y. Ohno, S. Kishimoto, K. Maezawa, T. Mizutani, and K. Sano,
“88 GHz dynamic 2:1 frequency divider using resonant tunneling chaos circuit,”
Electronics Letters, vol. 39, pp. 1546-1547, Oct. 2003.
[32] K. Maezawa, Y. Kawano, Y. Ohno, S. Kishimoto, and T. Mizutani, “Direct
observation of high-frequency chaos signals from the resonant tunneling chaos
generator,” Japanese Journal of Applied Physics, vol. 43, pp. 5235-5238, 2004.
[33] J. P. A. Wagt, “Tunneling-based SRAM,” Proceedings of the IEEE, vol. 87, pp.
571-595, Apr. 1999.
[34] K. Maezawa and T. Mizutani, “A new resonant tunneling logic gate employing
monostable-bistable transition,” Japanese Journal of Applied Physics, vol. 32,
pp. L42-L44, Jan. 1993.
[35] C. Samori, S. Levantino, and V. Boccuzzi, “A -94 dBc/Hz @ 100 KHz, fully-
integrated, 5-GHz, CMOS VCO with 18% tuning range for Bluetooth applica-
tions,” Proceedings of 2001 IEEE Custom Integrated Circuits Conference, pp.
201-204, 2001.
[36] T. Kim, A. Adams, and N. Weste, “High performance SOI and bulk CMOS
5 GHz VCOs,” Proceedings of 2003 IEEE Radio Frequency Integrated Circuits
Symposium, pp. 93-96, 2003.

142
[37] Y. Kawano, S. Kishimoto, K. Maezawa, and T. Mizutani, “Resonant tunnel-
ing chaos generator for high-speed/low power frequency divider,” vol. 38, pp.
L1321-L1322, Nov. 1999.
[38] T. Ohira, T. Hiraoka, and H. Kato, “MMIC 14-GHz VCO and Miller frequency
divider for low-noise local oscillators,” IEEE Transactions on Microwave The-
ory and Techniques, vol. MTT-35, pp. 657-662, Jul. 1987.
[39] S. Kudszus, W. H. Haydl, M. Neumann, and M. Schlechtweg, “94/47-GHz
regenerative frequency divider MMIC with low conversion loss,” IEEE Journal
of Solid-State Circuits, vol. 35, pp. 1312-1317, Sep. 2000.
[40] Q. Liu and A. Seabaugh, “Design approach using tunnel diode for lowering
power in differential comparators,” IEEE Transactions on Circuits and Sys-
tems, II, vol. 52, pp. 572-575, Sep. 2005.
[41] Cellonics Inc., “CellonicsTM technology technical white paper,” Cellonics.com,
14 Mar. 2004, <http://www.cellonics.com/downloads/files/TWP.pdf>.
[42] N. Waldron and J. A. del Alamo, “Self-aligned InP HEMT technology for 100
Gbits/s photonic systems,” 2004 SRC Review Presentation, 10 Jan. 2005, email
communication.
[43] N. Shimizu, T. Nagatsuma, T. Waho, M. Shinagawa, M. Yaita, and M. Ya-
mamoto, “In0.53 Ga0.47 As/AlAs resonant tunnelling diodes with switching time
of 1.5 ps,” Electronics Letters, vol. 31, pp. 1695-1697, Sep. 1995.
[44] T. P. E. Broekaert and C. G. Fonstad, “Extremely high current density,
low peak voltage, pseudomorphic In0.53 Ga0.47 As/AlAs/InAs resonant tunnel-
ing diodes,” in 1989 IEDM Technical Digest, pp. 559-562, 1989.
[45] J. H. Smet, T. P. E. Broekaert, and C. G. Fonstad, “Peak-to-valley
current ratio as high as 50:1 at room temperature in pseudomorphic
In0.53 Ga0.47 As/AlAs/InAs resonant tunneling diodes,” Journal of Applied
Physics, vol. 71, pp. 2475-2477, Mar. 1992.
[46] M. W. Dvorak, C. R. Bolognesi, O. J. Pitts, and S. P. Watkins, “300 GHz
InP/GaAsSb/InP double HBTs with high current capability and BVCEO ≥ 6
V,” IEEE Electron Device Letters, vol. 22, pp. 361-363, Aug. 2001.
[47] M. Ida, K. Kurishima, and N. Watanabe, “Over 300 GHz fT and fmax
InP/InGaAs double heterojunction bipolar transistors with a thin pseudomor-
phic base,” IEEE Electron Device Letters, vol. 23, pp. 694-696, Dec. 2002.
[48] G. He, J. Howard, M. Le, P. Partyka, B. Li, G. Kim, R. Hess, R. Bryie, R. Lee,
S. Rustomji, J. Pepper, M. Kail, M. Helix, R. B. Elder, D. S. Jansen, N. E.
Harff, J. F. Prairie, E. S. Daniel, and B. K. Gilbert, “Self-aligned InP DHBT
with fT and fmax over 300 GHz in a new manufacturable technology,” IEEE
Electron Device Letters, vol. 25, pp. 520-522, Aug. 2004.

143
[49] Z. Griffith, M. J. W. Rodwell, X. Fang, D. Loubychev, Y. Wu, J. M. Fastenau,
and A. W. K. Liu, “InGaAs/InP DHBTs with 120-nm collector having simul-
taneously high fT , fmax ≥ 450 GHz,” IEEE Electron Device Letters, vol. 26,
pp. 530-532, Aug. 2005.

[50] S. K. Diamond, E. Özbay, M. J. W. Rodwell, D. M. Bloom, Y. C. Pao, and

J. S. Harris, “Resonant tunneling diodes for switching applications,” Applied
Physics Letters, vol. 54, pp. 153-155, Jan. 1989.
[51] T. Inata, S. Muto, Y. Nakata, T. Fujii, H. Ohnishi, and S. Hiyamizu, “Excellent
negative differential resistance of InAlAs/InGaAs resonant tunneling barrier
structures grown by MBE,” Japanese Journal of Applied Physics, vol. 25, pp.
L983-L985, Dec. 1986.
[52] T. Inata, S. Muto, Y. Nakata, S. Sasa, T. Fujii, and S. Hiyamizu, “A pseudo-
morphic In0.53 Ga0.47 As/AlAs resonant tunneling barrier with a peak-to-valley
current ratio of 14 at room temperature,” Japanese Journal of Applied Physics,
vol. 26, pp. L1332-L1334, Aug. 1987.
[53] S. Muto and T. Inata, “InGaAs-based resonant tunneling barrier structures
grown by MBE,” Semiconductor Science and Technology, vol. 9, pp. 1157-1170,
Jun. 1994.
[54] T. P. E. Broekaert, W. Lee, and C. G. Fonstad, “Pseudomorphic
In0.53 Ga0.47 As/AlAs/InAs resonant tunneling diodes with peak-to-valley cur-
rent ratios of 30 at room temperature,” Applied Physics Letters, vol. 53, pp.
1545-1547, Oct. 1988.
[55] H. Mizuta and T. Tanoue, “The physics and applications of resonant tunneling
diodes,” Cambridge University Press, New York, pp. 135, 1995.
[56] R. Tsu and L. Esaki, “Tunneling in a finite superlattice,” Applied Physics
Letters, vol. 22, pp. 562-564, Jun. 1973.
[57] T. Wei, S. Stapleton, and O. Berolo, “Scattering parameter measurements of
resonant tunneling diodes up to 40 GHz,” IEEE Transactions of Electron De-
vices, vol. 42, pp. 1378-1380, Jul. 1995.
[58] K. Huang, M. Carroll, G. Starnes, R. Lake, D. Janes, K. Webb, and M. Mel-
loch, “Numerically generated resonant tunneling diode equivalent circuit pa-
rameters,” Journal of Applied Physics, vol. 76, pp. 3850-3857, Sep. 1994.
[59] P. Zhao, H. L. Cui, D. L. Woolard, K. L. Jensen, and F. A. Buot, “Equivalent
circuit parameters of resonant tunneling diodes extracted from self-consistent
Wigner-Poisson simulation,” IEEE Transactions on Electron Devices, vol. 48,
pp. 614-626, Apr. 2001.
[60] D. Sunderland and P. Dapkus, “Optimizing N-p-n and P-n-p heterojunction
bipolar transistors for speed,” IEEE Transactions on Electron Devices, vol.
ED-34, pp. 367-377, 1987.

144
[61] L. M. Su, N. Grote, R. Kaumanns, and H. Schroeter, “NpnN double-
heterojunction bipolar transistor on InGaAsP/InP,” Applied Physics Letters,
vol. 47, pp. 28-30, Jul. 1985.
[62] M. Dahlström, Z. Griffith, M. Urteaga, and M. J. W. Rodwell, “InGaAs/InP
DHBT’s with > 370 GHz ft and fmax using a graded carbon-doped base,” 61st
Device Research Conference, Late News Papers, pp. 6-7, Jun. 2003.
[63] M. W. Dvorak, O.J. Pitts, S. P. Watkins, and C. R. Bolognesi, “Abrupt junction
InP/GaAsSb/InP double heterojunction bipolar transistors with FT as high as
250 GHz and BVCEO > 6 V,” 2000 IEDM Technical Digest, pp. 178-181, Dec.
2000.
[64] H. K. Gummel and H. C. Poon, “Integral charge control model of bipolar
transistors,” The Bell System Technical Journal, vol. 49, pp. 827-852, May
1970.
[65] I. E. Getreu, “Modeling the Bipolar Transistor,” Tektronik, Inc., Beaverton,
Oregon, pp. 242-245, 1976.
[66] W. Liu, “Fundamentals of III-V Devices: HBTs, MESFETs, and
HFETs/HEMTs,” John Wiley & Sons, New York, pp. 220-280, 1999.
[67] W. Williamson, S. B. Enquist, D. H. Chow, H. L. Dunlap, S. Subramanian,
P. Lei, G. H. Bernstein, B. K. Gilbert, “12 GHz clocked operation of ultralow
power interband resonant tunneling diode pipelined logic gates,” IEEE Journal
of Solid-State Circuits, vol. 32, pp. 222-231, Feb. 1997.
[68] E. R. Brown, J. R. Söderström, C. D. Parker, L. J. Mahoney, K. M. Molvar,
and T. C. McGill, “Oscillations up to 712 GHz in InAs/AlSb resonant-tunneling
diodes,” Applied Physics Letters, vol. 58, pp. 2291-2293, May 1991.
[69] M. Reddy, S. C. Martin, A. C. Molnar, R. E. Muller, R. P. Smith, P. H. Siegel,
M. J. Mondry, M. J. W. Rodwell, H. Kroemer, and S. J. Allen, Jr., “Monolithic
schottky-collector resonant tunnel diode oscillator arrays to 650 GHz,” IEEE
Electron Device Letters, vol. 18, pp. 218-221, May 1997.
[70] K. Sano, K. Murata, T. Otsuji, T. Akeyoshi, N. Shimizu, and E. Sano, “An 80-
Gb/s optoelectronic delayed flip-flop IC using resonant tunneling diodes and
uni-traveling-carrier photodiode,” IEEE Journal of Solid-State Circuits, vol.
36, pp. 281-289, Feb. 2001.
[71] Y. Suzuki, Z. Yamazaki, Y. Amamiya, S. Wada, H. Uchida, C. Kurioka, S.
Tanaka, and H. Hida, “120-Gb/s multiplexing and 110-Gb/s demultiplexing
ICs,” IEEE Journal of Solid-State Circuits, vol. 39, pp. 2397-2402, Dec. 2004.
[72] M. Meghelli, “A 132-Gb/s 4:1 multiplexer in 0.13-µm SiGe-bipolar technology,”
IEEE Journal of Solid-State Circuits, vol. 39, pp. 2403-2407, Dec. 2004.
[73] Z. Griffith, Y. Dong, D. Scott, Y. Wei, N. Parthasarathy, M. Dahlström, C.
Kadow, V. Paidi, M. J. W. Rodwell, M. Urteaga, R. Pierson, P. Rowell, B. Brar,

145
 S. Lee, N. X. Nguyen, and C. Nguyen, “Transistor and circuit design for 100-
200-GHz ICs,” IEEE Journal of Solid-State Circuits, vol. 40, pp. 2061-2069,
Oct. 2005.
[74] A. K. Gupta and O. M. Collins, “A new interpretion and extension of Σ∆
modulation”, in Proceedings of IEEE International Symposium on Information
Theory, pp. 194, 2001.
[75] A. K. Gupta and O. M. Collins, “Viterbi decoding and Σ∆ modulation”, in
Proceedings of IEEE International Symposium on Information Theory, pp. 292,
2002.
[76] A. Gutierrez-Aitken, J. Matsui, E. N. Kaneshiro, B. K. Oyama, D. Sawdai, A.
K. Oki, and D. C. Streit, “Ultrahigh-speed direct digital synthesizer using InP
DHBT technology,” IEEE Journal of Solid-State Circuits, vol. 37, pp. 1115-
1119, Sep. 2002.
[77] T. Kim, B. Lee, S. Choi, and K. Yang, “Resonant tunneling diode/HBT D-
flip flop ICs using current mode logic-type monostable-bistable transition logic
element with complementary outputs,” Japanese Journal of Applied Physics,
vol. 44, pp. 2743-2746, Apr. 2005.
[78] S. Haykin, “Communication Systems,” John Wiley & Sons, New York, pp.
128-129, 2001.
[79] J. Joe, “Cellonics UWB pulse generator,” 2003 International Workshop on
Ultra Wideband Systems, paper 1029, Jun. 2003.
[80] S. H. Park, T. P. Chin, Q. Z. Liu, S. L. Fu, T. Nakamura, P. K. L. Yu, and
P. M. Asbeck, “Submicron self-Aligned HBTs by selective emitter regrowth,”
IEEE Electron Device Letters, vol. 19, pp. 118-120, Apr. 1998.
[81] Y.M. Hsin, N.Y. Li, C.W. Tu, and P.M. Asbeck, “AlGaAs/GaAs HBTs with
extrinsic base regrowth,” Journal of Crystal Growth, vol. 188, pp. 355-358, Jun.
1998.
[82] M. Le, “New processes allow InP HBTs to top 150 GHz,” Solid State Technol-
ogy, vol. 47, pp. 55-58, Jul. 2004.
[83] M. Urteaga, P. Rowell, R. Pierson, B. Brar, M. Dahlström, Z. Griffith, M.
Rodwell, S. Lee, N. Nguyen, and C. Nguyen, “Deep submicron InP DHBT
technology with electroplated emitter and base contacts,” in Proceedings of
62nd Device Research Conference, pp. 239-240, Jun. 2004.
[84] T. C. Mele, J. Nulman and J. P. Krusius, “Selective and anisotropic reactive
ion etch of LPCVD silicon nitride with CHF3 based gases,” Journal of Vacuum
Science and Technology B, vol. 2, pp. 684-687, Oct.-Dec. 1984.
[85] S. E. Hicks, S. K. Murad, I. Sturrock and C. D. W. Wilkinson, “Improving the
resistance of PECVD silicon nitride to dry etching using an oxygen plasma,”
Microelectronic Engineering, vol. 35, pp. 41-44, Feb. 1997.

146
[86] J. Lim, H. Ahn, H. Ji, W. Chang, J. Mun and H. Kim, “Fabrication and
characteristics of 0.12 µm AlGaAs/InGaAs/GaAs pseudomorphic HEMT using
a silicon nitride assisted process,” Semiconductor Science and Technology, vol.
19, pp. 1416-1421, 2004.
[87] N. Jin, S. Chung, A. Rice, P. Berger, R. Yu, P. Thompson, and R. Lake, “151
kA/cm2 peak current densities in Si/SiGe resonant interband tunneling diodes
for high-power mixed-signal applications,” Applied Physics Letters, vol. 83, pp.
3308-3310, Oct. 2003.

147