close

Вход

Забыли?

вход по аккаунту

?

Microwave and low-frequency noise characterization of Npn aluminum gallium arsenide/gallium arsenide heterojunction bipolar transistors

код для вставкиСкачать
INFORMATION TO USERS
This m anuscript has been reproduced from the microfilm master. U M I
films the text directly from the original o r copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may
be from any type of com puter printer.
The quality of this reproduction is dependent upon the quality of the
copy subm itted. B roken or indistinct print, colored or p o o r quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event th at th e author did not send U M I a com plete
m anuscript and th ere are missing pages, these will be noted. Also, if
unauthorized copyright m aterial had to be removed, a note will indicate
the deletion.
O versize m aterials (e.g., maps, drawings, charts) are rep ro d u ced by
sectioning th e original, beginning at the upper left-hand corner and
continuing from left to right in equal sections with small overlaps. Each
o riginal is also p h o to g ra p h e d in one exposure an d is in clu d ed in
reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. H igher quality 6" x 9" black and white
photographic prints are available for any photographs o r illustrations
appearing in this copy for an additional charge. Contact U M I directly
to order.
U n iversity M icrofilm s International
A B ell & H ow ell Inform ation C o m p a n y
3 0 0 N orth Z e e b R o a d . A nn Arbor, Ml 4 8 1 0 6 - 1 3 4 6 U SA
3 1 3 /7 6 1 - 4 7 0 0
8 0 0 /5 2 1 - 0 6 0 0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Order N um ber 9217806
Microwave and low-frequency noise characterization o f Npn
A lG aA s/G aA s heterojunction bipolar transistors
Costa, Damian, Ph.D.
Stanford University, 1992
UMI
300 N. ZeebRd.
Ann Aibor, MI 48106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
MICROWAVE AND LOW-FREQUENCY NOISE
CHARACTERIZATION OF Npn AIGaAs/GaAs
HETEROJUNCTION BIPOLAR TRANSISTORS
A DISSERTATION
SUBMITTED TO TH E DEPARTMENT OF ELEC TR IC A L ENGINEERING
AND THE COM M ITTEE ON GRADUATE STUDIES
O F STANFORD UNIVERSITY
IN PARTIAL FULFILLM ENT O F THE REQUIREM ENTS
FOR THE DEGREE O F
DOCTOR OF PH ILO SO PH Y
By
Damian Costa
D ecem ber 1991
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
© Copyright by Damian Costa 1991
All Rights Reserved
ii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
I certify that I have read this dissertation and that in
my opinion it is fully adquate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
mes S. Harris, Jr. (Principal Advisor)
I certify that I have read this dissertation and that in
my opinion it is fully adquate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
Robert Dutton
I certify that I have read this dissertation and that in
my opinion it is fully adquate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy ^
Jim Plummer
Approved for the University Committee
on Graduate Studies:
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Abstract
Abstract
The AlGaAs/GaAs heterojunction bipolar transistor (HBT) is emerging as a versatile
device technology, which is attractive for analog/microwave and digital applications. For
nonlinear circuit applications, such as mixers and oscillators, a combination of high
operational frequency and reduced low-frequency noise is required. Accurate, physicallybased equivalent circuits are useful for understanding the speed limitations and potential of
devices. Likewise, low-frequency noise measurements can be used as a diagnostic tool for
identifying the dominant noise mechanisms of a device. In this thesis, the fabrication,
microwave characterization and modeling, and low-frequency noise characterization of Npn
AlGaAs/GaAs HBTs are discussed.
To examine the microwave performance of AlGaAs/GaAs HBTs; a self-aligned base,
proton-isolated HBT process was designed and implemented. For a HBT with a base doping
of 1019/cm3 simultaneous ft and fmax above 30 GHz were obtained.
A technique for directly determining the small-signal equivalent circuit of
AlGaAs/GaAs HBTs was developed. With this technique most of the parasitics are extracted
from measurements of special test patterns. The intrinsic circuit element values are
calculated at any given frequency from y-parameter data, which has been de-embedded from
the previously known parasitics. The use of the equivalent circuit as a vehicle for optimizing
the high-frequency performance of HBTs is demonstrated.
The low-frequency noise characteristics of Npn AlGaAs/GaAs HBTs were measured
as a function of bias current, device geometry, extrinsic-base-surface condition, A1 mole
fraction in the emitter, and temperature.
The origins of various noise components are
interpreted in terms of specific physical parameters, such as the surface recombination
velocity and the deep donor DX center concentration. The role of an AlGaAs passivation
ledge (covering the extrinsic base surface) in improving the 1/f noise behavior of
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Abstract
AlGaAs/GaAs HBTs is demonstrated. The effect of the A1 mole fraction on the burstnoise behavior of AlGaAs/GaAs HBTs is shown. Based on these results, the important
issues for further improvement of the low-ffequency noise of 13I-V HBTs are examined.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Acknowledgements
Acknowledgements
This dissertation would not have been possible without the assistance of a large
number of people. Thanks must begin with my advisor, Professor James Harris. He has
given me the academic freedom to pursue the research topics of my choosing. His
"coaching" on both technical and non-technical issues has been extremely valuable: his
insight on all types of semiconductor devices has been inspiring, and his great sense of
humor has been an enormous comfort in times of need.
I am indebted to the members of my reading committee, Professors Robert Dutton
and Jim Plummer. I would also like to thank Professor Simon Wong for chairing my orals
committee.
Thanks go to a number of students and visiting scholars in the Harris Group. I
would like to thank Daisuke Ueda for patiently teaching me the fine art of GaAs processing
and giving me a new appreciation for Diet Dr. Pepper. Muchas gracias to Elias Munoz for
invaluable discussions about low-frequency noise and the DX center.
I am grateful to
Won-Seong Lee for his tremendous background work on AlGaAs/GaAs HBTs. I would
like to thank Tony Ma for discussions on HBT processing, for maintaining the MBE
machine, and his lessons in life. I am also indebted to Darrell Hill for his analytical model
of the band bending of AlGaAs ledges and for his constructive criticism. I owe a
tremendous thank you to William Liu for growing all of my wafers and for his assistance in
developing the high-frequency HBT process. His "Just Do It" attitude and late-night
processing to Live 105 at full volume have set the standard for all future Harris Groupies.
I would also like to thank the other members of the Harris Group, who have helped me out
in one way or another over the years and have made the lab an enjoyable place to work.
This work has also benefitted greatly from collaborations with several people
outside the Harris Group. I wish to thank Rob Marsland and Chris Madden of Professor
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Acknowledgements
Dave Bloom's group for assistance in making the s-parameter measurements, Brian
Hughes of Hewlett-Packard for useful discussions about equivalent-circuit modeling, and
Don D'Avanzo and Noel Fernandez of Hewlett-Packard for making the low-frequency
noise measurement system available for my use.
Final thanks go to our dynamic duo, Gail Chun-Creech and Susie Bums, for their
cheerful dispositions, and their uncanny knack at maneuvering through Stanford's
bureaucracy and, of course, for Gail's "penmanship".
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table of Contents
1.
In trod u ction
1.1
1 .2
2.
In tro d u c tio n
........................................................................................ 1
M o tiv atio n a n d O u tlin e o f T h esis ........................................... 3
B ackground
2 .1
................................................................................... 1
.................................................................................... 7
M olecular Beam E p itax y (M BE)
2.1.1
MBE Growth Considerations
................................................ 8
............................................... 9
2.1.2 Compositional Grading with MBE
2 .2
G aA s/A lG aA s
2 .3
DX
M a te ria l
C e n te rs
2.3.1
S y stem
...................................... 10
...............................................13
.................................................................................... 14
Electrical Properties of DX Centers in A ^ G a ^ A s ................ 15
2.3.2 Role of DX Centers in AlGaAs/GaAs HBTs ........................ 17
2.3.3 DX Centers in other III-V Alloys
2 .4
S u rfa c e
2.4.1
3.
P r o p e r tie s
of
G aA s
....................................... 19
....................................
21
Surface Effects in AlGaAs/GaAs HBTs .............................. 23
2 .5
N etw ork C h a ra c te riz a tio n w ith S -P a ra m ete rs ....................... 23
2 .6
N oise
D evice
S p e c tra l
O peration
D e n s ity
............................................................ 27
.....................................................................28
3 .1
3 .2
O peration of H om ojunction B ipolar T ran sisto r .................
29
O peration of H etero ju n ctio n B ip o lar T ra n sisto r ................... 30
3 .3
C u rre n t
G a in
Is s u e s
.................................................................33
3.3.1
Base Current Components
..................
33
3.3.2
Effect of AlGaAs Surface Passivation Ledge ......................... 37
3 .4
A b ru p t versus G ra d ed E m itte r-B ase J u n c tio n s ......................40
3 .5
H igh-F requency P erfo rm an ce F ig u res of M erit ....................43
3.5.1
Charge Control Model Derivation of ft ............................... 43
3.5.2
Derivation of ft from Equivalent-Circuit Model ..........
3.5.3
Derivation o f fmax
47
............................................................. 51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.
Device Design, Fabrication, and Experimental DC and
M icrowave Results ............................................................
4.1
4.2
4.3
4.4
5.
Epitaxial Layer Structure Design
........................................ 56
....................................................................... 58
D ev ice
L ayout
4.2.1
Layout of DC/Low-Frequency Devices
4.2.2
Layout of High-Frequency Devices
D ev ice
Fabrication
............................... 59
.................................... 60
.................................................................. 67
4.3.1
Fabrication of DC/Low-Frequency Devices ......................... 67
4.3.2
Fabrication of High-Frequency Devices
4.3.3
Proton Isolation Implant
E x p e ri m en ta l
R e su l ts
4.4.1
DC Results
4.4.2
Microwave Results
................................70
.......................................................73
............................................................... 74
............................................................................ 74
............................................................... 78
Direct Extraction of the AlGaAs/GaAs HBT Small-Signal
Equivalent Circuit
5.1
Introduction
................................................................................... 83
5.2
D e -e m b e d d i n g
Procedure
5.3
Measurement
5.4
5.5
6.
55
of
..........................................................84
Parasitics
.................................................... 87
5.3.1
Open Test Structure
5.3.2
Shorted Test Structure
5.3.3
Series Emitter and Collector Resistances .............................. 90
5.3.4
Extrinsic Base-Collector Test Structure
Results
and
............................................................. 87
D is cu ss io n
....................................................... 87
................................ 91
.......................................................... 95
5.4.1
Intrinsic Element Extraction
................................................ 95
5.4.2
Broad-Band S-parameters
5.4.3
Calculation of Power and Current Gains .............................. 99
5.4.4
Feedback for Device Design
.................................................... 96
............................................. 99
S u m m a r y .............................................................................................. 100
Low-Frequency Noise Characterization of Npn
A lG aA s/G aA s HBTs ............................................................. 102
6.1
Introduction
.................................................................................103
6.2
Measurement
S e t- u p
6.3
Theory:
Sources
of
............................................................ 105
Noise
.............................................106
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.3.1 W hite Noise
6.3.2
1/f Noise
6.3.3 Burst Noise
6.4
Results
........................................................................ 107
............................................................................. 107
...........................................................................110
...............................................................................................112
6.4.1 Current Dependence
........................................................... 112
6.4.2 Device Geometry Dependence
........................................... 114
6.4.3 Effect of AlGaAs Surface Passivation Ledge ...................... 117
6.4.4 Effect of A1 Mole Fraction in the Emitter ........................... 122
6.4.5 Temperature Dependence ..................................................... 123
6.5
6.6
D iscussion
Prospects for Reduced Low-Frequency Noise in III-V HBT
Design
6.7
7.
......................................................................................125
Summary
............................................................................................ 135
....................................................................................... 138
Conclusions and Suggestions for Future Research ..........139
7 .1
7.2
Conclusions
Suggestions for
............................................................................. 139
Future Research
..................................... 141
A.
High-Frequency HBT Fabrication Sequence ................... 146
B.
Derivation of Lumped Circuit Model for the Base
C ontact ....................................................................................... 150
C.
Definition of Equivalent Input Base Noise Current
Spectral D ensity ....................................................................154
R e fe r e n c e s .............................................................................................. 161
X
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
List of Tables
Table 4.1
Typical HBT epitaxial layer structure .................................................56
Table 4.2
Comparison of fma* values calculated with the classical expression
(equation 4.2) and experimental values ............................................. 82
Table 5.1
Comparison of the series emitter and collector resistances.................... 91
Table 5.2
Equivalent Circuit Elements (Vcb = 0.5 V) ........................................ 97
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Figures
Figure 1.1
Frequency response o f the gain and the low-frequency noise spectral
density of a bipolar transistor ............................................................. 3
Figure 2.1
Schematic diagram of MBE growth chamber ..........................................8
Figure 2.2
(a) Aluminum profile across base-emitter layer implemented with chirp
grading technique (b) Corresponding conduction-band edge diagram (at
flatband)
Figure 2.3
...................................................................................................11
Equilibrium energy band diagram of chirp-graded base-emitter
heterojunction
Figure 2.4
......................................................................................... 12
Conduction-band minima, deep-donor DX level, and hydrogenic donor
levels of AlxG aj.xAs as a function of A1 mole fraction........................ 14
Figure 2.5
Fractional occupation of DX centers in Si-doped AlGaAs vs A1 mole
fraction
Figure 2.6
................................................................................................... 17
Location of the conduction-band minimum for eighteen ternary III-V
alloys
........................................................................................................ 20
Figure 2.7
Two-port representation of a bipolar tran sisto r....................................24
Figure 3.1
Energy band diagram of npn homojunction bipolar transistor................ 29
Figure 3.2
Energy band diagram of Npn heterojunction bipolar transistor...............31
Figure 3.3
(a) Schematic cross section of mesa-type HBT and the various base
current components (b) Energy band diagram of intrinsic HBT and the
corresponding
Figure 3.4
base current components ........................................... 34
(a) Schematic cross section of emitter and base layers of AlGaAs/GaAs
HBT with an AlGaAs ledge covering the extrinsic base (b) Corresponding
energy band diagram of a section through the AlGaAs ledge and base
layer
Figure 3.5
.......................................................................................................... 38
Comparison of energy band diagrams of abrupt and graded
heterojunctions ....................................................................................
Figure 3.6
40
Idealized cross section of bipolar transistor for illustrating the various
delays comprising xEC .........................................................................44
xii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.7
First order hybrid-p equivalent-circuit model of a bipolar transistor
for calculating ft ....................................................................................48
Figure 4.1
Layout of the DC/Low-Frequency devices .........................................59
Figure 4.2
Layout of ohmic-metal layers and isolation mask of high-frequency
HBT
Figure 4.3
...........................................................................................................61
Bottom (a) and side (b) views o f a Cascade-MicroTech probe in the
ground-signal-ground configuration ................................................... 62
Figure 4.4
Coplanar waveguide in the ground-signal-ground configuration formed
by metal conductors on top of a substrate ..........................................63
Figure 4.5
Layout of high-frequency HBT ............................................................64
Figure 4.6
Fabrication process flow of DC/Low-Frequency H B T s .........................69
Figure 4.7
Fabrication process o f microwave HBTs ............................................. 72
Figure 4.8
Distribution of damage density vs depth (solid line) and
piece wise-linear approximation (dashed line) ..................................... 74
Figure 4.9
Comparison of Gummel plots of 4 um x 10 um emitter
AIq 3GaQ 7As/GaAs HBTs with and without the AlGaAs surface
passivation ledge
Figure 4.10
................................................................................ 75
Inverse DC current gain vs Pg/Ae for different size
AIq 3GaQ 7As/GaAs HBTs with and without the AlGaAs surface
passivation ledge .................................................................................. 76
Figure 4.11
Comparison of the collector current dependences of the DC current gain
of 4 um x 10 um emitter HBTs with and without the AlGaAs
passivation ledge for A1 mole fractions x = 0.2 and x = 0 .3 ..................77
Figure 4.12
Unilateral power gain (U) and common-emitter current gain(hfe)v s
frequency of a HBT biased at Ic = 15 mA and Vc{, = 0.5 V...................79
Figure 4.13
Collector current dependence of ft and fmax of device described in
figure
Figure 4.14
4.12
............................................................................................. 80
Comparison of the total emitter-to-collector delay (^ec) derived from
the measure ft values with the values computed from the
charge-control model and the calculated individual delay components.....81
Figure 5.1
Complete HBT small-signal equivalent c irc u it..................................... 85
xiii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5.2
Open test structure and corresponding circuit m o d e l.............................88
Figure 5.3
Frequency dependences of parasitic capacitances.................................88
Figure 5.4
Shorted test structure and corresponding circuit m o d e l......................... 89
Figure 5.5
Frequency dependences of parasitic inductances..................................89
Figure 5.6
(a) Schematic cross section of the extrinsic base-collector test structure
and the corresponding circuit model, (b) Simplified circuit m odel.........92
Figure 5.7
Frequency dependences o f the computed and the experimental values of
the extrinsic base-collector resistance (Rbc) ....................................... 94
Figure 5.8
Frequency dependence of the extrinsic base-collector junction
capacitance (Cjc ,ext) ........................................................................... 95
Figure 5.9
Frequency dependences of the base-emitter capacitance (Cn) and
the intrinsic base-collector junction capacitance (Cjc ,int)...................... 96
Figure 5.10
Comparison of the measured (solid lines) and the calculated (dashed lines)
s-parameters over the frequency range of 1 to 18 GHz of a HBT
operating at Ic = 2 mA and Vcb = 0.5 V............................................. 98
Figure 5.11
Comparison of the measured and model-produced unilateral power
gain (U) and common-emitter current gain (hfe) of a HBT biased at
Ic = 8 mA and Vcb = 0.5 V............................................................... 99
Figure 6.1
Schematic cross sections of HBTs (a) without and (b) with the
AlGaAs surface passivation ledge covering the extrinsic base surface... 104
Figure 6.2
Experimental setup for low-frequency noise measurements of H B Ts... 106
Figure 6.3
Equivalent input base noise current spectral densities and the various
noise components for a 4 inn x 10 um emitter Alg 3Gag 7As/GaAs
HBT without the AlGaAs passivation ledge and operating in the
linear region at three different bias currents ......................................112
Figure 6.4
Collector current dependence of the equivalent input base noise
current spectral density at 100 Hz, Sjb(100Hz); the magnitude of
the extracted low-frequency burst-noise plateau, Siburs[ (0 Hz); and
the burst-noise time constant, I, of the 4 um x 10 um
emitter Al0 3G ag7As/GaAs HBT described in figure 6 .3 ............. 113
xiv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 6.5
Comparison of the equivalent input base noise current spectral densities
of Al0 3Gao 7As/GaAs HBTs without the AlGaAs passivation ledge
for two different emitter perimeter-to-area ratios at a fixed collector
current o f 20 mA .............................................................................. 115
Figure 6.6
Comparison of the equivalent input base noise current spectral densities
of AIq 3Gag 7As/GaAs HBTs without the AlGaAs passivation ledge
for four different emitter perimeters (areas) at a fixed collector
current density of 5 x 103 A/cm2 ...................................................... 116
Figure 6.7
Emitter perimeter dependence of the equivalent input base noise
current spectral density at 100 Hz of Al0 3GaQ 7As/GaAs HBTs without
the AlGaAs passivation ledge at a fixed collector current density of
5 x 103 A/cm2........................................................................................117
Figure 6.8
Comparison of the equivalent input base noise current spectral densities
of Alo^Gao 7As/GaAs HBTs with and without the AlGaAs ledge for
two different emitter areas at a fixed collector current density of
5 x 103 A/cm2 ........................................................................................118
Figure 6.9
Comparison of collector current dependence of the equivalent input
base noise current spectral densities at 100 Hz of Al0 3GaQ 7As/GaAs
HBTs with and without the AlGaAs passivation ledge for two
different emitter sizes....... .................................................................. 120
Figure 6.10
AlGaAs passivation ledge width dependence of the equivalent input
base noise current spectral density at 100 Hz and the DC current gain for
a4
um x 10 um emitter AI q 3GaQ 7As/GaAs HBT at a fixed
collector current density
Figure 6.11
of 5
x
103A/cm2.......................121
Comparison of the equivalent input base noise current spectral densities
of 4 um x 10 um emitter HBTs with and without the AlGaAs
passivation ledge for two different A1 mole fractions (x) at a fixed
collector current density
of 5
x
103 A /cm 2....................122
xv
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Figure 6.12
Temperature dependence of the equivalent input base noise current
spectral density of a 4 um x 10 um emitter Al0^Ga^ 7As/GaAs
HBT without the AlGaAs passivation ledge at a fixed collector current
of 2 mA
Figure 6.13
.............................................................................................. 123
Equivalent input base noise current spectral density at T = 14 C multiplied
by frequency and the various noise components fitted to the data for the
4 um x 10 um emitter AIq 3Gag 7As/GaAs HBT described in
figure
Figure 6.14
6.12
...........................................................................................125
Effective potential barrier for thermal capture and emission of electrons
by the DX center................................................................................ 126
Figure B. 1
Schematic cross section of a contact and the corresponding transmission
line
Figure C. 1
model
............................................................................................150
Representation of device noise sources by a noiseless two-port network and
equivalent input voltage and current generators................................. 153
Figure C.2
Simplified noise model and definition of equivalent input (base) noise
current, Inj ........................................................................................... 154
Figure C.3
Low-frequency, equivalent-circuit model o f H B T ..............................154
Figure C.4
General, low-frequency HBT equivalent-circuit model with noise
sources
Figure C.5
Alternate, low-frequency HBT equivalent-circuit model with noise
sources
Figure C.6
................................................................................................... 156
................................................................................................. 157
Simplified, low-frequency HBT equivalent-circuit model with noise
sources for calculating the equivalent input base noise current
spectral
density......................................................................................158
xvi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
Introduction
1
Chapter 1
Introduction
One way ...to understand nature is to imagine the gods are playing some great game and
you don't know the rules of the game, but you're allowed to look at the board, at least
from time to time. Andfrom these observations, you try to figure out what the rules are of
the game.
Richard Feymann
1.1
Introduction
The impetus for growth in the contemporary electronics industry has been the quest
for faster semiconductor devices, packed more densely, and operating with ever smaller
amounts of power. Scaling of devices to reduce vertical and lateral transit times,
implementation of materials with superior carrier transport properties, and the minimization
of device parasitics, such as capacitances and resistances, are critical for realizing high­
speed devices. The maturity of epitaxial-growth technologies, such as molecular beam
epitaxy (MBE) and metal-organic chemical vapor deposition (MO-CVD) has made possible
the growth of ultra-thin, nearly arbitrarily doped compound semiconductor layers and
heterojunctions (the juxtaposition of two materials with different energy bandgaps).
Electrons thus require a shorter time to traverse these thin layers, and highly-doped layers
reduce parasitic resistances and permit a higher current-drive capability. Heterojunctions
provide a means for confining charge to specific regions of the device, and a gradual
variation of the energy bandgap with position can be used to establish quasi-electric fields,
which propel selected carriers, possibly in a near-ballistic fashion, across these regions.
The added flexibility in device design with this so-called "band-gap engineering" has
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
2
Introduction
allowed a re-optimization of doping levels and geometries, leading to unprecedented high­
speed performance.
The AlGaAs/GaAs heterojunction bipolar transistor (HBT), the subject of this
thesis, is emerging as a veiy promising high-speed device for a wide variety of applications
[1-1]. Discrete AlGaAs/GaAs HBTs capable of operating well above 100 GHz have been
demonstrated [1-2, 1-3]. The AlGaAs/GaAs HBT combines the advantages of the
properties of G a A s[l-4 ] with the benefits resulting from the design freedom of
heterostructures [1-5]. In comparison with a homojunction bipolar transistor, due to the
wide bandgap emitter (AlGaAs), a heavy base doping to reduce the base resistance, and a
low emitter doping to reduce the emitter-base junction capacitance can simultaneously be
incorporated in a HBT, without sacrificing emitter injection efficiency. Compared to Si
devices, GaAs-based HBTs have a higher electron mobility and the potential of overshoot
velocity. The availability of a semi-insulating GaAs substrate is also crucial in reducing
pad parasitics and simplifying device isolation.
While the speed of semiconductor devices (in particular, the unity current gain
cutoff frequency, ft , and the maximum oscillation frequency, fmax) is often accepted as the
"holy grail" of device designers; other figures of merit, such as dc current gain, breakdown
voltage, output power, noise, and reliability can be equally important depending on the
specific circuit application. The simultaneous optimization of more than one of these
figures of merit usually imposes some tradeoff. For example, as the emitter size of a
conventional AlGaAs/GaAs HBT is shrunk in order to improve the high-frequency
performance, the DC current gain [1-6, 1-7] and the low-frequency noise [1-8] are
degraded. Johnson [1-9] has shown from fundamental principles that high-frequency
performance and output power are inversely related. In order to achieve very high fmax
values, the base layer of a HBT must be doped very heavily.
However, at high
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
3
Introduction
concentrations, the diffusion of Be (the most common p-type dopant of GaAs grown by
MBE) during growth or under forward-bias stress can result in severe reliability problems
[ 1- 10].
1.2
Motivation and Outline of Thesis
The motivation of this thesis is summarized in figure 1.1. Two performance
Gain
Noise Spectral Density
log (Frequency)
Figure 1.1 Frequency response of the gain and the low-frequency noise spectral density
of a bipolar transistor.
criteria for bipolar transistors are gain and noise. In many applications it’s useful to extend
the gain of the device to as high a frequency as possible. Likewise, in certain applications
the noise that exists at low frequencies needs to be minimized. In linear applications, the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
Introduction
4
presence of low-frequency noise does not really have an impact on the characteristics of the
high frequency signal applied to the transistor. However, in nonlinear applications, such
as mixers and oscillators, due to unintentional mixing, the low-frequency noise can be upconverted as noise sidebands around the microwave carrier signal. Therefore, a clear
understanding of the both the limitations of the high-frequency performance and the
dominant low-frequency noise mechanisms of AlGaAs/GaAs HBTs are necessary.
Further improvement in the high-frequency performance of HBTs will require very
careful optimization of the epitaxial structure, processing, and device geometry.
Equivalent-circuit models are a useful tool for understanding the speed potential of HBTs,
particularly when the device must simultaneously satisfy a number of performance criteria.
Accurate, physically-based device models serve as a bridge between the known
design/process variables and the desired high-frequency response [1-11]. Since any model
is only as accurate as the accuracy of its element values, the issue of model-parameter
extraction is of paramount importance. In this dissertation, a new technique for directly
extracting the small-signal equivalent circuit model of AlGaAs/GaAs HBTs is described.
Low-frequency noise can limit the bandwidth and stability of a wide variety of
integrated circuits [1-12]. It also degrades the spectral purity of nonlinear microwave
circuits, such as oscillators and mixers [1-13]. Since oscillators and mixers are integral
components for the frequency-conversion-multiplexing processes that occur in
telecommunication systems, these systems are an important example where high-speed
devices with reduced low-frequency noise will be increasingly in high demand. As the
standard radio frequencies become too crowded, personal communication networks and
local area networks are being pushed into the GHz regime. For these applications, devices
must not only have gain above 2 GHz, but also their low-frequency noise must be
minimized.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
5
Introduction
This thesis presents the first in-depth study of the low-frequency noise properties
of Npn AlGaAs/GaAs HBTs. The low-frequency noise characteristics are examined as a
function of various device-structure parameters under different operating conditions. In
contrast to many previous studies, the origin of various noise components are interpreted in
terms of specific physical quantities, such as the surface recombination velocity and deep
donor DX center concentration [1-14]. Simple modifications in the epilayer design and
fabrication process that substantially improve the low-frequency noise characteristics of our
AlGaAs/GaAs HBTs are demonstrated. The prospects for further reduction in the lowfrequency noise of HI-V HBTs through optimized design are also discussed.
This dissertation is divided into 7 chapters. Chapter 2 discusses the crystal growth
technique of MBE, by which the epitaxial layer structures of the AlGaAs/GaAs HBTs used
in this thesis work were grown. It includes the MBE growth conditions and a technique
for compositional grading with MBE. This chapter also describes the GaAs/AlGaAs
system and compares it to silicon. A significant portion of this chapter is devoted to the
deep donor "DX center" in AlGaAs. In addition, this chapter introduces s-parameters and
the noise spectral density, which are two important quantities for characterizing the
microwave and low-frequency noise performance of AlGaAs/GaAs HBTs, respectively.
Chapter 3 explains the basic operating principles of HBTs and illustrates how they
differ from those of a homojunction bipolar transistor.
The various base current
components are identified and their physical origins are described. This chapter also deals
with the effects of heterojunction grading. Expressions for ft and fmax are also derived.
Chapter 4 is concerned with design of the HBT epitaxial layer structure. A
thorough description of the proper layout needed for DC and high-frequency devices is
given. This chapter also illustrates the important steps in the fabrication sequence of HBTs
intended to operate both at DC/Low-frequencies and at microwave frequencies. This
chapter concludes with a presentation of experimental DC and high-frequency results.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
6
Introduction
Chapter 5 is devoted a the discussion of a method for determining the small-signal
equivalent-circuit model of AlGaAs/GaAs HBTs. Most of the parasitic elements are
independently extracted from measurements of test structures. The intrinsic element values
are calculated directly at any given frequency from y-parameter data, which is de-embedded
from the previously determined parasitics.
The use of the equivalent-circuit as a
diagnostic tool for improving device performance is demonstrated.
Chapter 6 presents low-frequency noise measurements of AlGaAs/GaAs HBTs as a
function of bias current, device geometry, extrinsic-base-surface condition, aluminum mole
fraction in the emitter, and temperature. Based on these results, simple modifications in the
epitaxial structure and fabrication process that significantly reduce the low-frequency noise
of our AlGaAs/GaAs HBTs are demonstrated. This chapter ends by highlighting the
prospects for designing m -V HBTs with even better low-frequency noise performance.
Chapter 7 summarizes the contributions of this thesis work and makes suggestions
for future work on HBTs.
Appendix A contains a detailed step-by-step process sequence for fabricating
microwave HBTs.
Appendix B deals with the derivation of the lumped, parallel RC
combination model of the base contact, which is used in the HBT equivalent circuit.
Appendix C defines the equivalent input base noise current spectral density.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
7
Background
Chapter 2
Background
Good order is the foundation of all good things.
Edmund Burke
This chapter provides background information necessary for understanding the
growth of epitaxial layers found in AlGaAs/GaAs heterojunction bipolar transistor (HBT)
structures, the transport-related properties of these layers, and the characterization of
HBTs.
Section 2.1 describes the crystal growth technique of molecular beam epitaxy
(MBE), MBE growth issues, and a technique for compositional grading with MBE.
Section 2.2 presents the GaAs/AlGaAs material system and its advantages relative to
silicon. Section 2.3 introduces the deep donor-related "DX center" commonly found in ntype AlGaAs, reviews the electrical properties of DX centers in AlxG aj.xAs, suggests what
role(s) the DX center may play in the performance of AlGaAs/GaAs HBTs, and indicates
other III-V alloys in which DX centers may exist. Section 2.4 is concerned with the
surface properties of GaAs and their effects on the electrical characteristics of
AlGaAs/GaAs HBTs. Section 2.5 explains two-port network theory and scattering
parameters, which are useful for characterizing the high-frequency response of HBTs.
Section 2.6 introduces the noise spectral density.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
2.1
8
Background
Molecular Beam Epitaxy (MBE)
All of the device epitaxial layers used in this work were grown using a Varian Gen
II MBE system. MBE is basically a sophisticated evaporation method for precisely
growing thin, single-crystal layers in an ultra-high vacuum ( ~ 10 ' 10 torr) [2-1]. Figure
2.1 shows a schematic diagram of the growth chamber used in this study.
RHEED gun
Mass Spectrometer
crucible
Cryo
shroud
Beam Flux
Monitor
gate valve
source
Shutter
Substrate
heater
Furnace
Phosphor Screen
Figure 2.1 Schematic diagram of MBE growth chamber (courtesy W. Liu).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
9
The film constituents are deposited on a substrate from beams of atoms or molecules. A
beam is formed by heating an effusion cell or furnace to the temperature at which the
material it contains inside begins to evaporate. The system employs a furnace, with a
shutter covering the aperture, for every element that is to be introduced into the wafer. The
typical source materials are As, Ga, Al, In, Si (n-type dopant) and Be (p-type dopant). If
the shutter is open, atoms escape into the chamber and travel in straight lines (without
colliding with any other species) due to the high vacuum inside the chamber. The ultrahigh vacuum also prevents the incorporation o f unwanted impurities, and when combined
with clean sources can provide high quality material. The substrate is heated so that the
atoms or molecules impinging on the substrate have sufficient mobility to find the
appropriate sites to maintain the crystal structure of the material below. During growth, the
wafer is rotated to improve the epitaxial layer uniformity across the wafer. The flux of a
beam, and therefore the growth rate is controlled by varying the temperature of the furnace.
A typical growth rate for m -V semiconductors is one monolayer per second. Since the
actuating time of the shutters is short compared to the time to grow one atomic layer, a
precise number of monolayers can be achieved.
2 .1 .1
MBE Growth Considerations
The quality of the epitaxial material depends on the substrate temperature and
relative fluxes of the constituent beams. The typical growth rates for InGaAs, AlGaAs,
and GaAs layers were between 0.5 to 1 um/hr. The GaAs and AlGaAs layers were grown
at 600 °C, while the InGaAs layers were grown at 500 °C to prevent the the desorption of
In [2-2]. The substrates used for growth were oriented (100) GaAs. Recent results
suggest that tilting the (100) GaAs substrate 3 ° towards <111>A improves the quality of
bulk GaAs and the AlGaAs/GaAs hetero-interface[2-3].
When growing GaAs, AlGaAs, or InGaAs; the As flux is usually several times
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
10
that of the Ga flux. Under this arsenic over-pressure environment, As atoms striking the
surface instantaneously stick to the Ga atoms already present, until an atomic layer of As
atoms is created. Additional As atoms, which arrive at the surface, do not stick and simply
desorb from the surface. The exact As/Ga flux ratio modifies the relative concentrations of
As and Ga vacancies at the growth surface, which in turn, influences the incorporation and
redistribution of impurities. Thus, the As/Ga flux ratio affects the material quality of GaAs
and AlGaAs films. A relatively low As/Ga flux ratio tends to minimize the concentrations
of impurities originating from the arsenic source [2-4] and the concentrations of impurities
related to aluminum-oxygen complexes [2-5]. Such a low As/Ga flux ratio is desirable for
reducing the concentration of recombination centers ( and therefore the associated
recombination current) in the emitter-base space charge region of AlGaAs/GaAs HBTs .
On the other hand, for heavily p-doped GaAs base layers, as needed for high fmax Npn
transistors, a high As/Ga flux ratio suppresses the diffusion and carry-forward movement
of the p-type dopant Be, since it is substitutional on the Ga sublattice. [2-6, 2-7]. As a
compromise between these two constraints, the typical As4/Ga beam equivalent flux ratio
in this study was approximately 20 for the growth of devices with a base doping of of 5 x
1018 Be/cm3 . For the growths of more heavily doped base layers (1-2 x 1019 Be/cm3),
this ratio was raised to about 30. A common method for taking into account the movement
of Be is to insert an intentionally undoped spacer layer between the base and emitter layers
to absorb the Be doping "tail".
2 .1 .2
Compositional Grading with MBE
In all of the HBT epitaxial layers grown in this study, the A1 composition was
graded across the AlGaAs emitter - GaAs base heterojunction to remove the discontinuities
in the conduction and valence bands at the interface. With "thermal grading", the A1 mole
fraction (x) is adjusted from 0 to x by gradually increasing the A1 furnace temperature as the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
11
Background
layer is grown. Because the grading occurs over a short distance (typically 300 A) and flux
transients are associated with rapid heating of the A1 furnace, accurate and reproducible
variation of the A1 composition is quite difficult to achieve. In this study, the "chirp
grading" technique [2-8] was used to grade the emitter-base junction by growing a
superlattice of alternating AlGaAs (A1 mole fraction being fixed) and GaAs layers of
varying thicknesses.
Al
Com position
chirped approximation
intended profile
Em itte r
Grading Distance
Base
(a)
Conduction-Band Edge
A lG aA s
both Al & Ga shutters open
G aA s
only Ga shutter open
(b)
Figure 2.2 (a) Aluminum profile across base-emitter grading layer implemented with chirp
grading technique (b) Corresponding conduction-band edge diagram (at flatband).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
12
Background
Figure 2.2 illustrates (a) the implementation of compositional grading using the "chirp
grading" technique and (b) the conduction-band edge of the resulting superlattice (at
flatband). The grading is accomplished by opening and closing the Al shutter with the
furnace temperature held constant, thereby avoiding the flux transient problems. As the
growth proceeds from the base to the emitter, the Al shutter is opened for progressively
longer durations. By dividing the grading distance into sufficiently narrow subregions
(typically twenty 15 A subregions = width o f AlGaAs barrier + GaAs well), we can
achieve an accurate step-wise approximation to the intended Al profile (see Figure 2.2 (a)).
Figure 2.3 shows the resulting energy band diagram of the chirp-graded emitter-base
junction of a HBT at equilibrium.
2.0
conduction band
0.5
0.0
^
01
I
c
■Base
Emitter
-0.5
valence band
-
2.0
-200
200
400
Depth
600
O
800
1000
(A)
Figure 2.3 Equilibrium energy band diagram of chirp-graded base-emitter heterojunction
(courtesy P. van der Wagt).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
13
Background
The GaAs wells are sufficiently thin that the quasi-bound quantum levels between each
AIq 3Gag 7A s layer do not trap electrons. Thus, for electrons the superlattice behaves like a
continuously graded layer.
Because of the heavier effective mass of holes, each
AIq 3GaQ 7A s layer behaves as a barrier through which holes can not tunnel freely.
2 .2
GaAs/AlGaAs Material System
All of the device structures were grown on GaAs substrates by molecular beam
epitaxy (MBE). Despite the lack of a native oxide and a higher defect density compared to
Si, GaAs offers several fundamental advantages over Si. GaAs has an electron mobility
that is about five times larger than that of Si (the hole mobilities are similar). This higher
mobility decreases the transit time of electrons, as well as the resistivity of n-type neutral
regions. While the steady-state, high field drift velocities of GaAs and Si are both
approximately 107 cm/s; on a transient basis, the drift velocity of GaAs can "overshoot" the
steady state value by approximately five times compared to Si. Moreover, this transient
velocity-overshoot behavior can be maintained over a distance approximately one micron in
GaAs, but only 1000 A in Si [2-9]. In addition, the wider bandgap of GaAs makes semiinsulating GaAs achievable, and such substrates are readily available. The semi-insulating
nature of the substrate reduces the capacitance between devices and the substrate and
interconnects and the substrate to negligible values. Reduction of the parasitic substrate
capacitance not only permits higher device and circuit speeds, but also simplifies the
corresponding equivalent-circuit modeling.
Due to the close lattice-match, the significant energy bandgap difference, and the
isoelectronic nature of GaAs and ALAs, heterostuctures consisting of GaAs and AlxGalo[As
alloys have been extensively used as a means of manipulating the electron confinement and
transport to improve device performance. In AlxGaj_xAs alloys, the energy bandgap can
be varied continuously by increasing the Al mole fraction, x, from 0 (GaAs) to 1 (ALAs).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
14
Background
Figure 2.4 shows the the local energy minima of the different conduction-band valleys as a
function of x [2-10], along with the shallow, hydrogenic donor levels associated with the
r and L conduction-band minima [2-11] and the so-called "DX center" [2-12] (See section
2.3).
As shown, for x < 0.45, the T minimum remains the lowest energy conduction
band, thus maintaining a direct bandgap. For x > 0.45, the X minimum becomes the
lowest energy conduction band, creating an indirect bandgap material.
2.4
2.2
✓-N
^
2
eu
&
c
1.8
DX
conduction-band minima
- - DX Level
Hydrogenic Levels
1.6
1.4
0
0.2
0.4
0.6
0.8
1
Al mole fraction, x
Figure 2.4 Conduction-band minima, deep-donor DX level, and hydrogenic donor levels
of A ^ G a ^ A s as a function of Al mole fraction.
2.3
DX Centers
It's well known that AlGaAs has a number of electron traps [2-4]. Donor-related
deep levels in AlGaAs have been studied extensively over the last twenty years because of
their unusual and complex properties, such as the persistent photoconductivity effect [213]. Because such peculiar behavior was not expected for a simple donor, Lang et al [2-
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
15
14] attributed the deep level to a complex defect consisting of a donor (D) and an unknown
defect (X), hence the name DX center. Subsequent experiments strongly suggested that the
DX center is simply the donor itself [2-15]. The effect o f DX centers on the electrical
characteristics o f modulation-doped field effect transistors (MODFETs) has been
investigated in great detail [2-16,2-17,2-18], and more recently work has begun to relate
the thermal capture and emission properties of DX centers to the DC [2-19, 2-20] and
switching [2-21] behavior of heterojunction bipolar transistors (HBTs). Even though these
studies have enhanced tremendously the phenomenological understanding of DX centers
and their influence on device performance, a conclusive microscopic model for the DX
center has yet to be formulated. While a comprehensive discussion of the DX center is
beyond the scope of this thesis, the aim of this section is to establish a foundation for
understanding how the noise performance of AlGaAs/GaAs HBTs may be affected by the
presence of DX centers in the AlGaAs emitter.
2 .3 .1
Electrical Properties of DX centers in AlxG a i.xA s
The current view of donors in AlxGai_xAs is that each donor exhibits two types
of electronic states. One state is a shallow, delocalized effective-mass level pinned to the
lowest conduction-band minima (either T or X) and related to the substitutional site
configuration of the donor. The second state is a deep, more localized level, the DX
center, produced by a lattice distortion around the donor atom and roughly correlated to
the L conduction-band minimum [2-22]. Moreover, it has been reported that the emission
and capture rates of the DX center are sensitive to variations in the local environment,
caused by a distribution in the number of Al nearest-neighbor atoms [2-23]. Since the DX
center and the shallow center coexist, the sum of their concentrations is equal to the total
dopant concentration. The occupancy of the shallow, hydrogenic donors and the deep DX
centers is determined by their thermal equilibrium with the lowest conduction band as
dictated by the relative positions of the Fermi level, the lowest conduction-band minima,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
16
and the donor levels.
A schematic diagram of the binding energy of the DX center in Si-doped
AlxGai_xAs, as a function of Al mole fraction (x), is shown in figure 2.4 (in the previous
section). As can be seen, the DX center lies approximately at a constant energy below the
L conduction-band minimum. For x < 0.2, the DX center is above the T conduction-band
minimum ( resonant with the conduction band). In this composition range, the shallow,
hydrogen-like donor associated with the T conduction-band minimum is the lowest energy
state of the donor atom, and therefore, it controls the free-electron concentration ( except in
the case of highly degenerate material, where the Fermi level has been pushed well above
the T conduction-band minimum). For 0.2 < x < 0.8, the DX center becomes the lowest
energy state of the donor atom, and thus, it controls the conductivity of the material. In
the composition range of 0.2 < x < 0.4 (typical compositions for AlGaAs emitters in
HBTs), the DX binding energy (Er min - Epx) increases with x up to a maximum of about
160 meV.
The binding energy of the DX center is considerably larger than that of the
shallow, r-m inim um donor's binding energy of 5 meV. As a result of the large binding
energy of DX, the conductivity of the material in this composition range is reduced at room
temperature and is highly temperature sensitive. For x > 0.8, the shallow, X-minimum
donor becomes the lowest energy state of the donor. This implies that for x > 0.8, the
effects of DX centers may again be diminished.
Figure 2.5 shows the relative occupation of DX centers in Si-doped AlGaAs as a
function of x. As x increases from 0.2 to 0.4, the relative fraction of occupied DX centers
increases quickly. This increase in the occupancy of DX centers is a direct consequence
of the large binding energy of the DX center. Two key features which characterize the
DX center are its capture and emission barriers, which determine the flow of electrons to
and from the DX center [2-23]. The thermal emission of electrons from the DX center to
the conduction band was found to depend only on the donor species and not the Al mole
fraction [2-24]. Conduction band electrons must also surmount a repulsive energy barrier
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
17
Background
before being captured by the localized DX center.
1.0
0.8
0.6
Ss
o u
73*
eO
0.4
0.2
0.0
0.0
0.2
0.4
0.6
0.8
1.0
Al mole fraction, x
Figure 2.5 Fractional occupation of DX centers in Si-doped AlGaAs vs Al mole fraction
(Reference [2-26]).
It has been suggested that the capture occurs via an intermediate state (probably a
L-valley state) which is resonant with the conduction band and lies at a constant energy
above the DX level [2-25]. As a result, the thermal activation energy for capture depends
on both the capture barrier of each donor species with respect to this higher lying (L-valley)
state and the energy difference between the L and T conduction-band minima, and thus on
the alloy composition [2-26].
2 .3 .2
Role of DX Centers in AlGaAs/GaAs HBTs
Tiwari et al [2-19] have speculated on the effect of DX centers as generation-
recombination (g-r) centers in the emitter-base space charge region of a HBT. As shown in
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
18
figure 2.5, the relative fraction of occupied DX centers at Al mole fractions of 0.25 to 0.30
is quite high. Because the DX level is much closer to the conduction-band edge than to the
energy midgap, the DX center behaves as an effective electron trap. As a result of the DX
center's position in the forbidden bandgap, it's unlikely that the DX center can be equally
efficient in capturing holes, unless the DX center has a large hole capture cross-section.
However, given the repulsive electron capture barrier of the DX center, one might conceive
the DX center as having a stronger coupling to the valence band than a normal donor.
Watanabe et al [2-27] determined that the hole capture cross-section of the DX center at
room temperature is actually four orders of magnitude larger than that of electrons. As a
result of this large discrepancy in electron and hole capture cross-sections, they also
concluded that the DX level is not a very efficient g-r center, although their reasoning is
quite different from that of Tiwari et al.
At low temperatures, the presence of Si deep donors does not affect the DC
performance of AlGaAs/GaAs HBTs [2-28], Upon cooling to low temperature ( KT «
DX capture energy barrier), electrons in the AlGaAs emitter will initially be frozen out at
the DX centers. When the emitter-base junction is forward biased, electrons from the
GaAs contact layer will spill over into the emitter and be injected into the base, making
normal transistor operation possible. The release of photons as electrons undergo band-toband recombination in the neutral base, photoionizes the DX centers. Because of the
thermal capture barrier, the DX centers will remain empty and behave as ionized shallow
donors.
At room temperature, the current transients cause by the capture and emission of
electrons at DX centers are negligible compared to the currents usually found in HBTs. As
a result, Nathan et al [2-21] argued that the existence of DX centers is not expected to have
a major influence on the switching characteristics of AlGaAs/GaAs HBTs.
However, these small current fluctuations by definition constitute noise. Thus, the
capture and emission of electrons by DX centers are expected to impact the noise behavior
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
19
o f AlGaAs/GaAs HBTs (see chapter 6).
2 .3 .3
DX Centers in other III-V Alloys
Deep donor-related levels with DX center-like properties have been reported in
GaPxAs!_x [2-29, 2-30], AlxGaj.xSb [2-31,2-32], and A lJ n ^ A s [2-33,2-34] (although
not conclusively). Because HBTs consisting o f InP/InQ 53GaQ 47As [2-35], and
AlQ48In0 52As/In05 3GaQ47As [2-36] heterojunctions lattice-matched to InP are attractive
alternatives to AlGaAs/GaAs HBTs due to the superior properties of InGaAs base layers
compared to GaAs; knowledge of the composition ranges (if any) over which DX centers
are active in n-type layers of these other alloys, is quite useful. Assuming the DX center is
correlated with the L conduction-band minimum, Tachikawa et al [2-37] have proposed a
simple method of predicting the presence of DX centers in other ni-V alloys by examining
the composition ranges where the L conduction-band minimum lies close to the bottom of
the conduction band. Figure 2.6 shows the location of the simplified conduction-band
minima for eighteen ternary ni-V alloys. The alloy systems which could possibly include
the DX center have been circled. From the point of view of the most promising n-type
emitters in HBTs, figure 2.6 suggests that InP should be free of DX centers,while A ^Inj.
xAs (0.55 < x < 0.85) could potentially contain DX centers. It should be pointed out that
the composition Al0 48In0 52As necessary for lattice-matching to InQ53GaQ 47As is below,
yet close to the composition range where the DX center may exist. Given that the bowing
parameters of the composition-dependent conduction-band minima have been neglected in
figure 2.6, n-type Al0 4gln0 52As is certainly a candidate for having DX centers, and I
suggest that Al0 48Irio 52As be studied carefully for the presence of the DX center.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
C hapter 2
20
Background
>•»
w
V
o
tc
n
u>
0.
<
o
s
<
m
AlSb
MAS
AlP
AlSb
mp
GoP
AlP
>
e
>•
o
a
*
UJ
-X
0.
<
O
§«
o
GaSb
GaAs
GoP
GaSb InAs
GaAs
AlAs
inAs
inAs
mP
inSb
GaSb
AlSb
mSb
CD 0
inSb
fnSb
Figure 2.6 Location of the conduction-band minimum of eighteen
ternary III-V alloy systems (reference [2-37]).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
2 .4
Background
21
Surface Properties of GaAs
The unpassivated GaAs surface contains a large number (1013-1014 cm '2) of
electronically active acceptor and donor levels in the forbidden bandgap, probably due to
Ga and As vacancies. These states may behave as charge traps or recombination centers.
The high surface charge density requires a redistribution of charge near the semiconductor
surface in order to preserve charge neutrality. Since a small fluctuation in the surface Fermi
level alters the charge density significantly, the Fermi level is essentially pinned near
midgap, and charge neutrality is achieved by bending of the energy bands. In p-type
GaAs, the Fermi level is pinned approximately 0.5 eV above the valence-band edge due to
the Ga vacancy donor level, and in n-type GaAs, the Fermi level is pinned approximately
0.75 eV above the valence-band edge due to the As vacancy acceptor level [2-38]. The
band bending at the surface creates an electric field which attracts minority carriers toward
the surface, where they can interact with the high concentration of surface defects. This
large density of surface states, coupled with the band bending at the surface, makes surface
recombination perhaps the single most critical parasitic effect in GaAs minority-carrier
devices.
As with bulk recombination, surface recombination can be modeled with ShockleyRead-Hall recombination through deep-level traps [2-39]. At steady state, the net surface
recombination rate, Us, can be expressed as
,2
nsPs - ni
Us =
(2.1)
^ > s + nt) +^Ps+Pt)
where
Sn=»nOnNt
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2‘2)
Chapter 2
22
Background
and
Sp =Up(JpNt
(2.3)
where ns and ps are the surface electron and hole concentrations, nt and pt are the electron
and hole concentrations if the Fermi-level coincided with the trap level, n, is the intrinsic
carrier concentration, sn and Sp are the electron and hole surface recombination velocities,
an and o p are the electron and hole capture cross-sections, v„ and v p are the electron and
hole thermal velocities, and Nt is the trap density. Recombination at the surface results in a
net flow of minority carriers to the surface, and therefore, a surface current. The surface
recombination current density for minority carrier electrons, Jn surf , is commonly
expressed in terms of an effective surface recombination velocity, s, and takes the form
Jn,surf
- qsn' —qUs —qtinE + qDn
(2.4)
where q is the electronic charge, n' is the excess electron concentration value at the edge of
the surface space charge region, p n is the electron mobility, E is the electric field in
surface space charge region, Dn is the electron diffusivity, n is the electron concentration,
and x is a position coordinate perpendicular to the surface. The effective surface
recombination velocity is defined as
(2.5)
[2-40]. Thus, the effective surface recombination velocity depends on the product of the
electron and hole concentrations at the surface, the excess minority carrier concentration at
the bulk edge of the surface depletion region, and the surface potential (amount of band
bending). The effective surface recombination velocity is also sensitive to changes in the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Background
23
surface condition induced by processing steps, such as exposure to some types of plasmas
[2-41]. In general s is not a constant which is decoupled from n'. Because of the large
density of surface traps and the band bending at the surface, the effective surface
recombination velocity of a bare GaAs surface is extremely high, on the order of 10^ cm/s.
2.4.1
Surface Effects in AlGaAs/GaAs HBTs
The conventional mesa-type structure of AlGaAs/GaAs exposes carriers to the
extrinsic GaAs base surface, as well as the surface of the emitter-base junction.
Recombination of carriers at these surfaces is the origin of additional base current and
degrades the current gain of small-geometry AlGaAs/GaAs HBTs [2-42, 2-43]. The
surface recombination current is examined in section 3.3.1. Fluctuations in the base
surface recombination current produce excess 1/f noise in bipolar transistors [2-44]. This
1/f surface noise is described in detail in Chapter 6. One technique for minimizing the
surface recombination effect is the use of a thin, depleted AlGaAs ledge covering the
extrinsic base [2-45]. This method is discussed in section 3.3.2.
2.5
Network Characterization with S-Parameters
A transistor can be viewed as a two-port device linking input and output circuits.
Such a two-port network ( figure 2.7) can be characterized by a number of parameter sets,
such as hybrid (or h - ) parameters, impedance ( or z - ) parameters and admittance (or y - )
parameters which relate the input terminal voltage (V j) and current (Ij) to the output
terminal voltage (V2) and current (I2) [2-46].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
24
Background
2
11
n
+
+
V1
1
V2
Figure 2.7 Two-port representation of a bipolar transistor.
The defining equations are:
h-parameters
Vi = h u ll + I112V2
( 2 .6 )
12 = h2 ili + I122V2
(2.7)
z-parameters
Vi = z n Ii + z12l 2
( 2 .8 )
V2 = Z21I 1 + Z22I2
(2.9)
v-parameters
11 = y n V i + y 12V2
12 = y21v l +Y22V2
( 2 . 10 )
( 2 . 11)
The only difference in these parameters is the selection of independent and dependent
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
25
Background
variables. The parameters are the constants used to relate these variables.
The parameters are determined from measurements at the terminals of the two-port
network (box), in such a way as to eliminate one o f the independent voltages or currents.
For example,
is determined by shorting the output of the box (V2 = 0) and measuring
the respective voltage and current at the input of the box. From equation (2.10), y ^ is
easily obtained:
y n - ^
(V2 - 0 )
(2 . 12)
In principle, any of these parameter sets can be used at any frequency; however, at high
frequencies, total voltage and current at the ports of the network can not be easily
measured, broadband short and open circuits are difficult to achieve, and active devices
( such as transistors) may not be stable when open or short circuit terminated.
At high frequencies, the power of traveling waves and the ratio of incident and
reflected waves are the quantities that are most easily measured. S-parameters are just
another set of two-port parameters that relate the traveling voltage waves, rather than total
voltages and currents. Since these parameters relate the incident to the reflected ( or
scattered) wave, these parameters are named scattering ( or s - ) parameters. The sparameters relate the traveling waves as follows:
bi = s n a i + s12a2
(2-13)
b2 = s2jai + s22a2
(2.14)
where a} and a2 are the normalized (to the characteristic impedance of the transmission line)
incident voltage waves at the input and output p o rts, and bj and b2 are the normalized
reflected voltage waves at the respective ports. Physically, laj!2 is the incident power on
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
26
Background
the input port, and Ibjl2 is the reflected power on the input port.
The s-parameters are easily found by simply terminating the appropriate port with
the characteristic impedance of the transmission lines (Z0) used. Thus, regardless of the
length of the transmission line, there will be no reflected wave from the load (and therefore,
no incident wave at the terminated port). For example, by terminating the output port with
Z0, we can solve equations (2.13) and (2.14) to obtain:
S11 = bi/aj = T
(2.15)
S21 = b2/ai
(2.16)
Therefore, Sjj is the input reflection coefficient (O with the output of the network
terminated in ZG. In general, given a reflection coefficient at some position along a
transmission line and the characteristic impedance, one can calculate the impedance at that
same position, sj 1 and s22 are sometimes "loosely" referred to as impedances, actually
meaning the impedance at the plane where reflection coefficients are Sjj and S22Although s-parameters are the parameters of choice at high frequencies because of
their ease of measurement, equivalent-circuit modeling is more difficult with s-parameters
because s-parameters can not immediately be interpreted as circuit element values. Since
the total voltages and currents are related to the incident and reflected voltage waves, the
other parameter sets can be derived from the s-parameters [2-46].
More intuitive device
modeling and de-embedding techniques are accomplished by converting the measured sparameters to z-parameters or y-parameters. Z-parameters are used to extract series
elements and y-parameters are used to extract shunt elements. The equivalent circuit
modeling of AlGaAs/GaAs HBTs is described in Chapter 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
2.6
Background
27
Noise Spectral Density
Noise, in the electrical sense, refers to fluctuations in current or voltage.
Because noise is random in nature, it can not be described by an instantaneous value: at
some specific instant, the noise signal could take on an infinite value. Instead, noise is
analyzed with statistical averages. One commonly used average to describe noise is the
mean-squared average. For example, the mean-squared noise current can be expressed as
either an integral in the time domain or the frequency domain according to:
i2(t) = limT _>oo i |
i(t) i(t) dt = J Si(f)df
(2.17)
Jo
where i2(t) is the mean-squared noise current in units A2, i(t) is the instantaneous noise
current, and Sj(f) is the noise current spectral density in units A2/Hz. Sj(f) describes the
frequency dependence of the mean-squared noise current per unit bandwidth. The noise
current spectral density is the quantity that is typically measured experimentally.
Analytically, Sj(f) can be found from the Fourier transform of a correlation function [2-47],
Si(f) = 2 1
i(t) i(t+v) exp (-jcov) dv
Jo
where to is the angular frequency and v is a variable of integration.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
(2.18)
Chapter 3
28
Device Operation
Chapter 3
Device Operation
What admits no doubt in my mind is that the Creator must have known a great deal of
wave mechanics and solid state physics, and must have applied them.
A. Szent-Gyorgi
This chapter covers the operation of heterojunction bipolar transistors (HBTs).
Section 3.1 describes the fundamental operating principles o f homojunction bipolar
transistors. Section 3.2 compares the operation of homojunction transistors with that of
heterojunction bipolar transistors. Section 3.3 is concerned with current gain issues. The
various base current components are identified and their physical origins are discussed. A
technique for reducing the base surface recombination component, which is usually
dominant in small-geometry AlGaAs/GaAs HBTs, is reviewed. Section 3.4 deals with
the effects of heterojunction grading. Section 3.5 examines the microwave performance of
HBTs. Expressions for two commonly-used figures of merit, ft (short-circuit commonemitter current-gain cutoff frequency) and fmax (maximum frequency of oscillation), are
derived from both charge control and equivalent circuit models.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
3.1
29
Device Operation
Operation of Homojunction Bipolar Transistor
A bipolar junction transistor consists of two closely spaced, "back-to-back" pn
junctions. Figure 3.1 shows the energy band diagram of a npn homojunction bipolar
transistor.
electron
-H
Base
Emitter
Collector
Figure 3.1 Energy band diagram of npn homojunction bipolar transistor.
Under normal operation, the base-emitter junction is forward biased and the base-collector
junction is reverse biased. The application of forward bias reduces the conduction-band
potential barrier (AVn) for electron flow (In) from emitter to base. Most of the electrons
that surmount this barrier diffuse across the base without recombining, since the base is
made sufficiently thin. Upon reaching the reverse biased base-collector junction, electrons
are swept into the collector by the large electric field within the junction. The collector
current (Ic) of the transistor consists of these collected electrons (Ic ~ In).
At the same time, the application of forward bias reduces the valence-band potential
barrier (AVp) for hole flow (Ip) from base to emitter. This back-injected hole current is the
dominant component of base current in a homojunction device (Ib ~ Ip). Since the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
30
Device Operation
potential barriers for electrons and holes are the same (AVn = AVp) in a homojunction
device, the ratio of injected electron current to injected hole current (I„ / Ip) is controlled by
the relative doping densities in the emitter and base. In order to have appreciable current
gain, P = Ic / Ijj ~ In / Ip, the emitter must be more heavily doped than the base in a
homojunction transistor. This inherent restriction on the base doping concentration limits
the achievable value of fmax.
3.2
Operation of Heterojunction Bipolar Transistor
As illustrated in the previous section, in a homojunction bipolar transistor (BJT),
the emitter and base regions are composed o f the same energy-bandgap material.
As a
result, the conduction-band potential barrier for electron flow from emitter to base and the
valence-band potential barrier for hole flow from base to emitter are constrained to be the
same in a BJT. To insure sufficient current gain, the device design must incorporate a
higher doping in the emitter than in the base.
In contrast to the BJT, in a heterojunction bipolar transistor (HBT), the emitter and
base regions are made of different energy-band gap materials.
For the case of
AlGaAs/GaAs HBTs, the emitter (AlGaAs) is made of larger bandgap material than the
base (GaAs) (see figure 2.4). This bandgap difference between the emitter and base gives
the HBT a tremendous advantage over the BJT. Figure 3.2 shows an approximate energy
band diagram of a HBT with a graded heterojunction (the impact of heterojunction grading
is discussed in section 3.4). Unlike the BJT, the potential barriers for electron and hole
injection , AVn and AVp, differ by the bandgap potential difference between the emitter
and base, AE~/q. Electrons injected from the emitter into the base "see” a lower barrier
o
than holes injected from the base into the emitter.
The consequences of this difference in potential barriers for electron and holes can
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
31
Device Operation
be best understood by examining the ratio of the desired electron injection current from the
emitter to the base, I„, to the undesired hole back-injection current from the base into the
emitter, Ip.
AV.
electron
Base
Emittei
A V,
Collector
-x n
xp
w,b
Figure 3.2 Energy band diagram of Npn heterojunction bipolar transistor.
Neglecting degeneracy effects (using the Boltzmann approximation) and assuming that
current flow is only by diffusion, we can express the injected electron and hole currents as
In = qAeb nb(xp)
= qAebNe expj1^ ^ - ) ^
(3.1)
and
Ip = qAebPe(-Xn) ^
“ q A et>Nb ex p ( '^ f E) ^
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(3 ‘2)
Chapter 3
Device Operation
32
where q is the electronic charge, k is the Boltzmann constant, T is the temperature, Aeb is
the area of the emitter-base junction, nb(xp) is the election concentration at the base edge of
the emitter-base junction, D„b is the electron diffusivity in the base, Wb is the base width,
Ne is the emitter doping, pe(-xn) is the hole concentration at the emitter edge of the emitterbase junction, Dpe is the hole diffusivity in the emitter, Lp is the hole diffusion length in
the emitter, Nb is the base doping, and the slight differences in the hole and electron
effective density of states between the wide bandgap and narrow bandgap materials have
been neglected. Dividing equation (3.1) by (3.2), we can write the injection ratio as
In._ D n ^ N e j A E g )
Ip DPeWbNb * kT I
( ' }
where AEg = q(AVp -AVn). For Al02GaQ8As/GaAs HBTs, AEg ~ 10 kT and expj^jp-)
Ia e \
~ 2.2 x 104; and for Al0 3GaQ 7As/GaAs HBTs, AEg ~ 14.5 kT and exp|-jp^j ~ 2.0 x 106.
Thus, for A1 mole fractions > 0.2, the difference in bandgap between the emitter and the
base provides a factor of over 104 improvement in 1,,/Ip over the homojunction case ( AEg
= 0). As a result, a HBT designer has additional flexibility in selecting the optimum
emitter and base doping levels, without seriously degrading the current gain due to the
back-injection of holes into the emitter.
This expanded freedom in device design allows the implementation of a heavily
doped base and a lightly doped emitter, while still maintaining adequate current gain. A
high doping in the base lowers the base sheet resistivity and base contact resistance, leading
to an improved maximum frequency of oscillation. Simultaneously, the low emitter doping
reduces the emitter-base junction capacitance, which influences the current-gain cutoff
frequency, as well as maximum frequency of oscillation. These added benefits to the speed
performance of HBTs are discussed in greater detail in section 3.5
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 3
3 .3
Device Operation
33
Current Gain Issues
The most important figure of merit describing the dc behavior of bipolar transistors
is the dc current gain, P, which is simply defined as Ip/If,, where Ic is the collector current
and Ib is the total base current. For most applications, a dc current gain on the order of 10
to 100 is needed. A primary impetus in HBT design is to minimize the base current in
order to maintain this gain. The base current is the sum of various components, which
originate in different regions of the device structure.
3 .3 .1
Base Current Components
As shown in figure 3.3, there are six base current components (arrows in figure 3.3
(a) point to the regions where the recombination takes place and not necessarily in the
direction of hole current flow) in a conventional mesa-type HBT: three associated with the
bulk (1^, 1^, 1^) and three associated with the surface (Ib , Ib , 1^) (not shown in the
one-dimensional plot in figure 3.3 (b)). In this section, the origins of these base current
components (assuming low level injection) and their ideality factors ( I «= exp(Vbe/nkT),
where
is the base-emitter voltage and n is the ideality factor) are described.
Ibj is the back-injected hole current from the base into the emitter. Since 1^ is
proportional to the injected hole concentration at the emitter edge of the emitter-base
junction (equation 3.2), Ib « expCV^/kT). With the presence of the heterojunction ,
which significantly increases the potential barrier for hole injection, the hole back-injection
base current component is virtually eliminated in AlGaAs/GaAs HBTs.
1 ^ arises to replenish the holes lost from direct band-to-band recombination inside
the quasi-neutral base, where minority carrier electrons only have a finite lifetime. Since
the recombination rate is directly proportional to the excess injected electron concentration,
Ib “= exp(Vbe/kT) [3-1]. At very high doping levels, I^ is most likely due to Auger
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
34
Device Operation
recombination [3-2].
lu
Emittei
extrinsic base surface
t
/
t
'• /
base contact
F
J ly
1
*
Base
x
Collector
(a)
Figure 3.3 (a) Schematic cross section of mesa-type HBT and the various base current
components (b) Energy band diagram of intrinsic HBT and the corresponding base
current components.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
35
Device Operation
Ib3 is produced by recombination at centers in the bulk of the emitter-base space charge
region in response to the large increase in carriers within the emitter-base junction under
forward bias.
With some simplifying assumptions, it has been shown that Ib «=
njexp(Vbe/2kT), where nj is the intrinsic carrier concentration [3-3]. This dependence
results because on the average, each carrier is only required to overcome half of the
potential barrier before recombination takes place. Because of the suppression of hole
injection into the emitter, the maximum space-charge recombination occurs in the wide-gap
emitter near the hetero-interface.
Ib results from recombination at the surface of the emitter-base space-charge
region. Since holes and electrons are supplied from either side of the heterojunction,
surface recombination within the emitter-base space charge region is generally assumed to
behave similarly to conventional bulk recombination in the space charge region, which
suggests that Ib «= exp(Vbe/2kT).
Ibs is due to electrons injected into the base which diffuse laterally and recombine
with majority carrier holes at the extrinsic base surface . Since this surface is outside the
emitter-base junction, this would suggest that recombination at the extrinsic base surface
behaves much like recombination in the quasi-neutral base, and hence Ib should exhibit an
ideality factor of one. However, the ideality
factor of the extrinsic base surface
recombination current remains controversial as theories and experimental results support
both an ideality factor of two [3-4] and an ideality factor of one [3-5, 3-6]. As described in
section 2.4, the surface recombination current density for minority carrier electrons,
Jn surf, is commonly expressed in terms of the surface recombination velocity, s. For
convenience, equation (2.4) is again written as
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Device Operation
Jn, surf —
36
(3*4)
where n' is the excess electron concentration at the edge of the base surface space charge
region. If s is a constant, then Jn> surf «= n' «= exp(Vbe/kT) and the ideality factor of the
extrinsic base surface recombination current is unity. However, reference [3-4] argues that
s is not constant and decoupled from n \ Based on a number of assumptions, they
demonstrate that the ideality factor of the extrinsic base surface recombination current is
two.
However, experimentally, they examined a laser structure which itself has an
exp(V/2kT) dependence and thus not the most appropriate structure for examining this
phenomenon.
Tiwari et al [3-7] have attempted to reconcile these two opposing views by showing
that both approaches may be true, depending on the rate-limiting step in the recombination
process. Their simulations suggest that if the surface recombination velocity has a low to
moderate value, then the np product at the surface limits the recombination process (s is
not a constant but depends on the electron and hole concentration at the surface) and an
ideality factor of two results for the extrinsic base surface recombination current.
However, when s is high, the recombination process depends on the rate that carriers can
be supplied to the surface. As long as s is high, the exact value does not matter and
therefore s can be viewed as a constant. Under this condition, an ideality factor of one
results for the extrinsic base surface recombination current. As a consequence of these
two possibilities for the rate-limiting steps, they showed that the ideality factor of the
extrinsic base surface recombination current varies with collector current density, Jc. For s
= 2 x 106 cm/s (an appropriate value for bare GaAs) and Jc > 102 A/cm2, the ideality
factor was nearly unity. For s = 2 x 103 cm/s (an appropriate value for InGaAs), the
ideality factor was close to two over the entire simulated range of collector current densities
( 10'6 < Jc < 105 A/cm2 ).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Device Operation
37
Ibg is the interface recombination current at the base contact. Because the contact
metal has a very short minority carrier (hole) lifetime, it is characterized by a high interface
recombination velocity, much like the GaAs surface. If the spacing between the emitter
mesa and the base contact is much larger than the lateral diffusion length of electrons [3-8],
then most of the electrons injected into the base will not reach the base contact, but rather
they will either diffuse across the base to the collector or recombine at the extrinsic base
surface between the emitter mesa and the base contact. However, if the separation between
the emitter-mesa and the base contact is comparable to the base width (as might occur in a
self-aligned base process), then this component can be large. Invoking the same argument
used for the case of the extrinsic surface recombination current (Ib ) when s is high, I
postulate that the recombination current at the base contact should have the same ideality
factor- namely one.
3 .3 .2
Effect of AlGaAs Surface Passivation Ledge
As explained in section 2.4, the combination of a high density of recombination
sites at the GaAs surface and the electric field within the surface depletion region, which
propels minority carriers toward the surface, leads to a very high surface recombination
velocity. As a result, the base current component ( 1 ^ ) due to the lateral diffusion and
recombination of injected electrons at the exposed extrinsic-base surface (predominantly
near the edge of the emitter mesa, where the electron concentration is the highest) is the
dominant base current component in AlGaAs/GaAs HBTs with large emitter perimeter-toarea ratios.
Lin and Lee [3-9] proposed a technique for minimizing this surface recombination
by implementing a thin AlGaAs layer (AlGaAs surface passivation ledge) surrounding the
emitter mesa. In this method under ideal conditions, the AlGaAs ledge is fully depleted due
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 3
38
Device Operation
to the overlapping depletion regions from the free surface and the emitter-base junction.
Since a properly designed AlGaAs ledge is wholly depleted, it decreases the electron
concentration that can recombine at the base surface.
A more lucid understanding of the role of the AlGaAs surface passivation ledge in
suppressing the extrinsic base surface recombination current can be gained by examining
how the AlGaAs ledge modifies the band bending at the extrinsic base surface.
Emitter
AlGaAs
with Ledge
Base
Surface
X
No Ledge
(b)
Figure 3.4 (a) Schematic cross section of emitter and base layers of AlGaAs/GaAs HBT
with an AlGaAs ledge covering the extrinsic base (b) Corresponding energy band diagram
of a section through the AlGaAs ledge and base layer.
Figure 3.4 shows a schematic cross section of the emitter and base layers of an
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
39
Device Operation
AlGaAs/GaAs HBT with an AlGaAs ledge covering the extrinsic base surface and the
corresponding energy band diagram of a section through the AlGaAs ledge and base layer
below. For illustrative purposes, the band diagram of the extrinsic base region for the case
where the base surface is left bare, is also shown. As can be seen from figure 3.4, the
electric field (proportional to the slope of the conduction band), which drives injected
minority carrier electrons toward the surface, is smaller and does not penetrate as deeply
into the base in the presence of an AlGaAs ledge compared to the case of the exposed base
surface. By decreasing the concentration of minority carrier electrons that reach the
surface, the AlGaAs ledge reduces the effective surface recombination velocity and the
extrinsic base surface recombination current. Experimental results demonstrating the role
of the AlGaAs surface passivation ledge in improving the DC current gain and 1/f noise of
AlGaAs/GaAs HBTs are shown in sections 4.4.1 and 6.4, respectively.
Experimental [3-10] and analytical [3-11] studies have shown that the effectiveness
of the AlGaAs ledge depends strongly on the doping, composition, and thickness o f the
layer. For an emitter doping of 5 x 1017 cm-3, the AlGaAs ledge of the thin emitter HBT
epitaxial structures used in this study (see chapter 4) is about 700 A.
Consistent ledge
thicknesses are defined by using selective etches. For a given doping, composition, and
base-emitter voltage; if the AlGaAs ledge is too thick, such that the depletion regions from
the free surface and the pn junction don’t interact, a thin conducting layer is formed. This
existence of parasitic conduction through the ledge is of particular concern when the baseemitter junction is strongly forward-biased, as the width of the base-emitter depletion
region shrinks significantly. An undepleted portion of the ledge acts as an extension of the
emitter and the parasitic emitter current that flows through the ledge is injected into the
base. Since these electrons are actively injected closer to the base contact, the base current
component due to recombination at the base contact ( Ij,6) increases [3-12]. Thus, the
current-gain benefit from the AlGaAs ledge is reduced.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
3 .4
40
Device Operation
Abrupt versus Graded Emitter-Base Junctions
One of the most important features of molecular beam epitaxy (MBE) is the ability
to grow structures with abrupt transitions between materials with different energy
bandgaps. The energy band diagram of such abrupt junctions reveals discontinuities in the
conduction and valence bands at the hetero-interface. An alternative approach is to
gradually vary the composition between the two different materials in order to smooth out
these band discontinuities and thus form a graded junction. Figure 3.5 compares the
energy band diagrams of abrupt and graded emitter-base junctions of AlGaAs/GaAs HBTs.
Examination of figure 3.5 brings out the respective advantages of each type of junction.
AqV
A qV r
A qV
A qV
Abrupt Junction
Graded Junction
Figure 3.5 Comparison of energy band diagrams of abrupt and graded heterojunctions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
41
Device Operation
Compared to a homojunction (shown as dashed lines in figure 3.5) the difference
between the energy barriers for holes and electrons (qAVp - qAVn) is AEg for a graded
junction and AEVfor an abrupt junction. It should be pointed out that this difference occurs
because the energy barrier for electrons of an abrupt junction it much larger than that of a
graded junction (qAVn>abrupt = AEC+ qAVn
), although qAVp>abrupt ~ qAVp) g ^ .
Since for AlGaAs/GaAs heterojunctions A Ev ~ 0.35 A Eg , the injection factor (equation
(3.3)) for an abrupt AlQ 3Ga0?As/GaAs HBT is j0-** exp ( ^ j ) “ expj^ ' ^ f i j = 160
(as opposed to 2 x 106 for a graded junction). Therefore, despite the fact that the backinjected diffusion hole current in the emitter of an abrupt junction is comparable to that of a
graded junction, the injection ratio is decreased for an abrupt junction device. However,
as described in section 3.3, the base current of a HBT consists of several components,
including the emitter-space charge recombination current which is directly proportional to
the intrinsic carrier concentration (nj). Because the heterojunction suppresses the back
injection of holes, most of the recombination inside the emitter-base space charge occurs
near the hetero-interface. As can be seen from figure 3.5, the energy bandgap near the
hetero-interface is significantly larger (nj is smaller) for the abrupt junction, hence the
emitter-base space charge current both in the bulk and the surface will be smaller for an
abrupt junction device relative to a graded junction device. In practice for an abrupt
AlGaAs/GaAs HBT, the decrease in injection ratio is compensated by the decrease in the
emitter space charge current.
Another consequence of the larger energy barrier for electron flow in the case of the
abrupt junction is that the built-in voltage and therefore the tum-on voltage of abrupt
heterojunctions is larger than that of graded junctions. While an increase in the tum-on
voltage is usually not important in microwave applications, especially in power applications
where the devices operate at high collector-emitter voltages, it leads to larger power
dissipation in digital circuits, especially in high speed-logic circuits, where the devices
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Device Operation
42
operate over small voltage swings.
An additional property of abrupt junctions is that the conduction band spike may act
as a "ramp" to launch hot electrons into the base. Since these energetic electrons may
possess a higher than average velocity as they traverse the base region, both the neutral
base recombination and the base transit time are possibly reduced. The actual resulting
benefit depends on how much excess energy these launched electrons receive from the
electric field inside the junction. If the electric field at the interface is too high, some
fraction of these electrons will scatter into the lower mobility (and therefore lower velocity)
satellite conduction-band valleys (L and X valleys), resulting in an insignificant overall
enhancement in transport through the base.
In spite of these issues of ballistic transport, recombination, etc.; in practice, the
most important issue of abrupt versus graded junctions is that an abrupt junction is more
susceptible to Be diffusion from the base into the emitter. Since the base doping is
typically 10 to 100 times larger than the emitter doping in a high-frequency HBT, even a
slight penetration of Be into the AlGaAs easily counterdopes the exposed n-type region,
forming an AlGaAs pn homojunction. Thus, all the advantages associated with the
heterojunction are lost. On the other hand, a similar amount o f Be movement into a graded
junction HBT has almost no effect (the turn on voltage is increased slightly). Not only is
the growth of heavily doped HBTs with abrupt junctions more difficult, abrupt HBTs may
be also be more susceptible to reliability problems caused by the Be movement during
operation [3-13].
For this reason, all of the HBTs used in this study have graded
junctions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
3 .5
Device Operation
43
High-Frequency Performance Figures of Merit
The two most popular figures of merit to quantify the speed performance of
microwave transistors are the short-circuit common-emitter cuirent-gain cutoff frequency,
ft, and the maximum frequency of oscillation, fmax. ft is defined as the frequency at which
the incremental current gain of the transistor in the common-emitter configuration (with the
output short circuited) drops to unity. fmax is defined as the frequency at which the
unilateral power gain drops to zero. Since to first order ft depends on the delays
associated with the modulation of stored charge resulting from one-dimensional electron
flow through the intrinsic transistor, it can be estimated from a charge control formalism.
ft can also be derived from an equivalent circuit: a combination of electric-circuit elements
arranged in a specific configuration to portray the electrical characteristics of the device to
which the "circuit is equivalent", f , , ^ depends on the charging of the base-collector
circuit, which is distributive and two-dimensional in nature; it is most easily computed from
an equivalent circuit or two-port parameters.
3 .5 .1
Charge Control Model Derivation of ft
Applied to bipolar junction transistors, charge-control modeling relates the overall
time delay in terms of the controlling charge (supplied by the base lead) and the controlled
output (collector ) current.
More specifically, the incremental common-emitter current
gain, hfe, at high frequencies can be expressed in terms of the incremental base charge ,
dQb , associated with an incremental input voltage and the corresponding increment in
collector current dlc [3-14]:
die
dlb
die 1 _ ft
f
dQb jco
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
44
Device Operation
where f is the frequency and to is the angular frequency.
The total emitter-to-collector
delay, XEC, can then be expressed as
Tec = --77^ =
die
2 JC ft
= t E + Tb + xcsc + xc
(3.5)
where XEC is comprised of various time delays discussed below. Figure 3.6 illustrates
these various charging and transit times across each region.
Emitter
Collector
Base
►
W.
B
W.
be
Figure 3.6 Idealized cross section of bipolar transistor for illustrating the various delays
comprising xEC.
XE is the emitter delay. It corresponds to the time required to charge the baseemitter junction capacitance, Cjbe, when the base-emitter junction potential (Vbe) is
changed to the desired operating value. Physically, XE is associated with the movement of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
45
Device Operation
charge at the edges of the base-emitter depletion region, dQEj. XE is given by
TE= 1 f =d-s!fCjbe
(3-6)
where r'e is the incremental resistance o f the base-emitter junction.
Equation (3.6)
assumes that the delay associated with holes that are stored in the quasi-neutral emitter is
negligible compared to the other delay components. This is generally true for HBTs, but
not for homojunction bipolar transistors, in which the storage of holes in the quasi-neutral
emitter can constitute a significant portion of the total delay. It should also be pointed out
that the equivalent (Thevenin) charging resistance of
is r'e, not r'e plus the emitter
series resistance.
Tb is the base transit time or the base delay due to the storage of holes in the quasi­
neutral base in order to neutralize the charge associated with electrons crossing the base
region. In the limit of electron flow by diffusion only (for devices with no doping
gradients or energy gap variations in the base), Tb can be written as
(3-7)
where Wb is the neutral base width and D„b is the electron diffusivity in the base.
Equation (3.7) is valid under low or moderate injection conditions. In the case of highlevel injection in base (which is not likely in HBTs since the base doping »
emitter
doping) the base transit time is actually reduced from that in equation (3.7) due to the
Webster effect [3-15], and in the case of high level injection in of the collector (when free
electron concentration in base-collector depletion region is comparable to the ionized
collector donor concentration), the base transit time is larger than that predicted by equation
(3.7) due to base "push-out" or the Kirk effect [3-16]. As a result, when the bipolar
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
46
Device Operation
transistor approaches these high level injection conditions, one can not assume that the base
transit time is a constant, independent of collector current.
Tcsc is the transit time across the base-collector space charge region or the time delay
due to hole charge stored in the base-collector space charge region. It is given by
(3.8)
where x is the spatial coordinate,
is the base-collector depletion width, and t)s is the
saturation velocity. Although not intuitive, the factor of two arises because only half of the
charge needed to neutralize the moving charge in the current stream is associated with hole
charge in the base (the other half is neutralized by extra ionized donors in the collector) [317]. For simplicity, equation (3.8) also assumes that all the electrons traversing the basecollector space charge region are traveling at the saturated velocity. Equation (3.8) is also
not strictly valid at high collector currents with the onset of the Kirk effect (Jc crit=
= 8 x 104 A/cm2 for a collector doping of Nc = 5 x 1016 cm-3). Under this condition, the
electric field inside the base-collector depletion region begins to collapse. If the electric
field is reduced sufficiendy, then the carriers traversing the space charge region will not be
accelerated to the saturation velocity, and the transport may even become part diffusion
rather than purely drift as assumed in equation (3.8). Accurate extraction of the individual
time delays from the measured ft requires consideration of the bias dependence of Tb and
Tcsc at high collector current densities.
Tc is the RC time constant for charging the base-collector junction capacitance in
response to an incremental input voltage with the output terminals short-circuited. Tc is
given by
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 3
47
Device Operation
Tc - (Re + Rc + r'e) Cjbc - |Re + Rc +
Cjbc
(3.9)
where RE is the series emitter resistance, Rc is the series collector resistance, and C j^ is the
base-collector junction capacitance (including both intrinsic and extrinsic base regions).
Equation (3.9) shows the strong impact of the series collector and emitter resistances
(including the contact resistances) on the time constant associated with charging the basecollector junction capacitance.
The overall transit time is the sum of individual time delays and can be expressed as
2
XEC = F ( Cjbe + Cjbc) + ^
q ic
^u nb
+^
£ + ( R e + R c) Cjbc .
zvs
(3.10)
A few comments about equation (3.10) can be made. First, for a given device geometry,
the transistor must be operated at high collector current levels in order to minimize the first
term in equation (3.10). For digital circuits, in which power dissipation is an important
issue, then the base-emitter and base-collector junction areas must be reduced as much as
possible in order to allow high-speed operation at lower collector currents. Tailoring of the
electric field profiles in the base and collector regions can used to reduce the second (base
transit time) and third (transit time across base-collector space charge region) terms in
equation (3.10) [3-18, 3-19, 3-20]. These modifications in the epitaxial structure were not
incorporated in this thesis work. The last term in equation (3.10) emphasizes the need for
an advanced fabrication technology. A low-resistance ohmic contact technology and
process steps to minimize the extrinsic base-collector junction capacitance, such as ion
bombardment [3-21] and self-alignment of the base and collector contacts [3-22,3-23] are
critical for realizing ultrafast HBTs.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
3 .5 .2
48
Device Operation
Derivation of ft from Equivalent Circuit Model
The short-circuit common-emitter current-gain cutoff frequency can also be
calculated from a small-signal equivalent circuit of the bipolar junction transistor. ft is
defined as the frequency at which the incremental common-emitter current gain, with output
of the transistor shorted, becomes unity. The hybrid-7t topology [3-24. 3-25, 3-26]
provides a convenient representation for bipolar transistors measured in the commonemitter configuration. Figure 3.7 shows a first order hybrid-7t model of a bipolar
transistor with the collector shorted to ground.
b
jbc
Figure 3.7 First order hybrid-Tt equivalent circuit model of a bipolar transistor for
calculating ft.
The circuit elements are defined as follows: rn is the incremental emitter-base diode
resistance referred to the base ( rff ~ pr'e = p/Gm), Cn is the base-emitter capacitance ( =
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
49
Device Operation
base-emitter junction capacitance ( C ^ ) + base-charging capacitance (Cb)), Cjbc is the total
base collector junction capacitance; Gm is the intrinsic transconductance (= qlg/KT), Rb is
the total base resistance (defined in equation (4.3)), RE is the emitter series resistance, and
Rc is the collector series resistance. In this simple model, the distributed nature of the
base-collector circuit is neglected, and the base-collector circuit is represented by two
lumped terms Rb and Cjbc •
The current-gain cutoff frequency can be calculated from figure 3.7 using the expression
hfe<to) = h2i(Co ) = i |v<jui, 0
(3.11)
where ic and ib are the incremental collector and base currents, respectively and to is angular
frequency. Using nodal analysis, we can write the collector and base currents as
ic = (vc - vb') s Cjbc + GmVji
(3 •12)
i \ > = ^ + (vb-- vc)s Cjbc
(3-13)
and
- U i + s Q
Zjr
(3.14)
r7t
where vb- and vc are the voltages with respect to ground from nodes b' and c', respectively
and s is the complex phase angle (= jco). After some manipulation, it can be shown that
_
Gm - s C jbc[ l + ( ^ + Gm)Re] - s2 Cn Cjbc Re
+ s |Q t + Cjbcjl + (j^- + Gm)( Re + Rc)]} + s2 Cn Cjbc( Re + Rc)
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
^
Chapter 3
50
Device Operation
It should be pointed out that to first order (base resistance is lumped into one resistance
connected to the input lead), the current gain is independence o f base resistance.
Neglecting the second order terms in s, we can write the magnitude squared of the current
gain as
„
Gm^ +
( Cjbcf 1 + (jT“ + Gm) R e]}
I hfeO'co) | 2 = ----------------- ^
--------------------- —
( i ^ + Gm)(
rji2 to^JCjc
" I — + Cjb<|l + 1%
"T R“e + Rc)]}
~'J)
(} 2 + g)2q 2
= UI"
” 7
J - + co2d
rrt2
(3.16)
For a typical microwave AlGaAs/GaAs HBT ( Ic = 8 mA, Gm = 307 mS, p = 20, r^ = 65
O, CK = 650 fF, Cjbc = 60 fF, RE = 15 Q, and Rc = 10 O, which gives Cj = .25 pF
and C2 = 1.2 pF), Gm2 » co2Cj and — « to2C2 at f = 26 GHz. Therefore, equation
ijt2
(3.16) simplifies to
1 hfe(j0))l
(o{CK + Cjbcf 1 + G m(R E + Rc)])
where the approximation Gm »
(3' 1?)
1/ rn has also been used. The total emitter-collector delay
is determined from the frequency at which the magnitude of the current gain becomes unity
or I hfe(co = 2 7t ft = — ) I = 1. Solving for
Tec
from equation (3.17) we find
Tec = pJ-(C * +Cjbc) + ( R e + Rc)Cjbc.
'J,m
(3-18)
Substituting the identities Cn = C ^ + Cb = C j^ + GmXF = Cjbe + Gm (Tb + Tcsc) and Gm
= qIc/KT into equation (3.18), we arrive at the expression
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Tec -
Device Operation
(Cjbe + Cjbc) + Tb + Tcsc + ( Re + Rc) Cjbc
51
(3.19)
which is identical to the charge-control equation (3.10) when Tb and Tcsc are defined as in
equations (3.7) and (3.8).
While this result is perhaps surprising at first glance to both the device physicist
(who prefers to use the charge control model) and the circuit designer (who prefers to use
the equivalent-circuit model), it is actually quite reasonable since the hybrid-7t model is
also based on the physics of the device. Examination of the incremental distribution and
flow of carriers in the neutral regions of the device leads to relationships between the
junction voltages and currents. The linearized relationships between the incremental
voltages and currents are then represented by the hybrid-7t equivalent circuit.
3 .5 .3
Derivation of fmax
The second commonly quoted figure of merit, fmax, is associated with the power
gain of the device.
Unlike the current gain, the power gain o f the transistor is quite
sensitive to the source and load terminations. In order to transmit power from the source
to the load most efficiendy (deliver the maximum power to the load), the input and output
of the transistor must be conjugately matched to the source and load, respectively.
However, since the input and output impedances of the transistor are complex at high
frequencies, the simultaneous conjugate-match condition at both ports may cause the device
to become unstable. Based on the stability of the device, two power gains are defined as
follows: the maximum stable gain (MSG) and the maximum available gain (MAG) are the
maximum power gains when the input and output are simultaneously conjugately matched,
with MSG being defined when the device is conditionally stable and MAG being defined
when the device is unconditionally stable.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
52
Device Operation
These power gains are most readily calculated from two-port parameters by
straightforward analysis. Since the various two-port parameter sets can all be expressed in
terms of each other, the power gains can be measured or computed from the most
convenient parameter set. In real applications, MSG tends to be more useful than MAG
because most transistors are not unconditionally stable over the entire frequency range. In
terms of h-parameters, the maximum stable (power) gain can be expressed as [3-27],
/a 2 (V)
K
’
MSG = ___________I **21! 2___________
" 4 Reflin) Re(h22) - 2 Re(hi2h2l) '
One additional power gain is the unilateral power gain, U. This is the maximum
available power gain when the two-port has been simultaneously matched and the feedback
parameter has been neutralized to zero.
This neutralization or unilateralization is
accomplished by adding a network such that a part of the output signal is fed back to the
input circuit in order to balance out a portion of the feedback inherent in the transistor. This
gain is the highest possible gain that an active two-port network can ever attain. The
frequency where the unilateral power gain drops to unity defines the boundary between an
active and a passive network. This frequency is usually referred as fmax> the maximum
frequency of oscillation.
In terms of y-parameters, the unilateral power gain is given by [3-28],
__________ 1y2i - y d 2__________
(3.21)
4 { R e (y ii)R e (y 2 2 )- R e(y12)Re(y2i)) '
Experimentally, U is calculated from the measured s-parameters (s-parameters are
converted to y-parameters and equation (3.21) is used), and fmax is defined as the
frequency at which U becomes unity. While MAG, MSG, and U are easily determined
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Device Operation
53
from two-port parameters, equations (3.20) and (3.21) provide limited insight into the
physical parameters, such as resistances and capacitances, which influence the power gain
of the device.
A first order analysis to determine the physical parameters that limit the power gain
can be done by using equation (3.20) in conjunction with the equivalent circuit in figure
3.7. Neglecting the effects of emitter series resistance (RE) and collector series resistance
(Rc) on power gain, assuming the reverse-current gain (h12) is zero, and taking the
f
magnitude of the forward-current gain at high frequencies to be | h2i | = -j- , it can be
shown by circuit analysis that
R e( hii )=R b
Re(h22) - Gm C,^ j^
(3.22)
= 2 7t ft Cjbc
(3-23)
and
MSG »
— J-.
8 7ERb Cjbc f2
(3.24)
The maximum frequency of oscillation can be approximated from the frequency at which
MSG becomes unity. It is therefore given by the classical expression,
(3-25)
Note that in addition to the assumptions listed above in deriving equation (3.25),
the use of equivalent circuit in figure 3.7 assumes that the base resistance is purely resistive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Device Operation
54
and can be lumped into a single element. As will be shown in chapter 5, the base contact
impedance in the case of moderately doped bases is complex, not purely resistive.
Moreover, the base-collector junction capacitance has also been lumped into a single
element.
More exact expressions for fmax, which attempt to take into account the
distributive nature of the base-collector circuit, have been derived by replacing RbCjbc with
an effective RC time constant (Rjj^jbc^eff w^ere the extrinsic and intrinsic components of
Rb and Cjbc are more properly weighted [3-29]. Alternatively, more sophisticated
equivalent circuits (than that shown in figure 3.7), which more accurately reflect the actual
device structure, have been used to calculate the y-parameters and compute U with equation
(3.21) [3-30]. Chapter five describes the modeling of the high-frequency response of
AlGaAs/GaAs HBTs with a second order equivalent circuit that models the base contact
with a complex impedance and divides the base-collector junction capacitance into intrinsic
and extrinsic portions.
Despite its limitations, equation (3.25) illustrates a few general points about the
power performance of bipolar transistors at high frequencies. In order to achieve the
highest fmax, both the base resistance and the base-collector junction capacitance must be
minimized. Equation (3.25) demonstrates the advantage o f HBTs over homojunction
bipolar transistors, as a consequence of a higher allowable base doping ( and thus lower
base resistance) in HBTs as compared to homojunction bipolar transistors. While the
intrinsic base-collector junction capacitance can not really be modified by the fabrication
process, the extrinsic base-collector junction capacitance is amendable to processing steps,
such as self-alignment and ion bombardment. A combination of these techniques has
allowed the realization of AlGaAs/GaAs HBTs with fmax = 218 GHz [3-31].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4
Device Design, Fabrication, and Experimental DC and Microwave Results
55
Chapter 4
Device Design, Fabrication, and Experimental DC
and Microwave Results
Freedom has a thousand charms to show
That slaves, howe'er contented, never know.
William Cowper
This chapter discusses the design aspects, fabrication process, and measured DC
and microwave characteristics of HBTs. Section 4.1 describes the design of the HBT
epitaxial layer structure. Section 4.2 is concerned with the device layout for DC and highfrequency operation.
Section 4.3 examines the process sequences used in device
fabrication. Section 4.4 presents experimental DC and high-frequency results.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
4.1
56
Epitaxial Layer Structure Design
The availability o f heterojunctions in the AlGaAs/GaAs and GaAs/InGaAs
systems provides additional freedom in the design of the epitaxial layer structure of
HBTs. Table 4.1 shows the typical epitaxial layer structure o f a Npn AlGaAs/GaAs
HBT. Si is used as the n-type dopant and Be is used as the p-type dopant. In this
section, the purpose of each layer is highlighted.
Material
ThicknessfA)
Composition
_3
Doping (cm )
Contact Cap
n-InyGaj ^As
500
0 —> 0.5
2xl019
Cap
n-GaAs
1000
Grading
N-AlxGaj ^As
300
x —> 0
5xl017
Emitter
N-AlxGaj_xAs
300
X
5xl017
Grading
N-AlxGaj xAs
300
0 —> x
5xl017
Base
p-GaAs
800,1400
5x 1018,1019
Collector
n-GaAs
4000
5xl016
Sub Coll*
n-GaAs
6500
5xl018
Substrate*
semi-insulating
Laver
5xl017
GaAs
Table 4.1 Typical HBT epitaxial layer structure. *A heavily doped subcollector and a semiinsulating substrate are used in the high-frequency devices, while in the DC /low-frequency
devices, the subcollector layer is omitted and the substrate is n+.
The purpose of the contact layer is to provide a low emitter resistance. InGaAs is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
57
superior to GaAs for forming ohmic contacts. First, a higher n-type doping is achievable
in MBE grown InGaAs compared to GaAs. Second, the Fermi level of InGaAs is pinned
in the conduction band, rather than at midgap as with GaAs. Thus, the barrier for electron
flow at the emitter-contact metal/ semiconductor interface is negligible for InGaAs. In
biyGaj.yAs, both the maximum possible n-type doping and the position of the Fermi-levelpinning with respect to the conduction-band edge, increase with indium composition (y).
Typically, the top 300 A of the contact layer is In0 5GaQ 5As.
A contact n-GaAs layer is inserted between the InGaAs contact-cap layer and the
AlGaAs emitter. This GaAs layer is essential because AlGaAs and InGaAs are grown at
different temperatures (see section 2.1). This GaAs layer acts as a buffer, preventing the
mixing of In and Al, and it serves as a supply layer of electrons to the emitter.
The wide bandgap n -A ^G a^A s emitter provides high electron injection efficiency
into the base. The aluminum mole fraction, x, is chosen so that the bandgap difference
between the emitter and the base is such that AEg »
KT. Traditionally, an aluminum
mole fraction of x = 0.3 has been used in AlGaAs/GaAs HBTs; but as shown in section
3.2, an aluminum mole fraction as low as x = 0.2 still allows AEg »
KT. The relative
occupation of deep donor levels in AlGaAs decreases as the aluminum mole fraction is
decreased from x = 0.3 to x = 0.2 (see section 2.3).
In this thesis work, both
AIq 3Ga0 7As/GaAs and Al0 2Ga0 8As/GaAs HBTs were fabricated. As discussed in
Chapter 6, this reduction of the aluminum mole fraction virtually eliminates anomalous
low-frequency noise in our AlGaAs/GaAs HBTs [4-1].
In order to facilitate device
fabrication (see section 4.3), the fixed x, AlxGaj_xAs emitter is made very thin [4-2]. Such
a thin emitter (300 A) also makes a negligibly small contribution to the overall series emitter
resistance. Bandgap grading regions are inserted on either side of the AlxGaj_xAs emitter.
The selection of a graded heterojunction over an abrupt heterojunction is explained in
section 3.4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
58
The thickness and doping of the base layer are a trade-off between base resistance,
transit time, and transport factor. As a compromise between these parameters, the base
width is usually shrunk as the base doping is increased. A high base doping is critical for
attaining a low sheet resistance for both low intrinsic and extrinsic base resistances and low
contact resistance. However, as mentioned in section 2.1, heavily Be-doped GaAs base
layers are difficult to grow without significant Be diffusion into the emitter. In the highfrequency structures, the base width was 800 A and base doping was 1019 Be/cm3. In
addition, a 100 A undoped GaAs setback layer (not shown in Table 4.1) was inserted
between the base and emitter-base grading layers to absorb any movement of the Be during
growth. In the DC/low-frequency structures, the base width was 1400 A and the base
doping was 5 x 1018 Be/cm3 with no setback layer.
The purpose of the n‘-collector is to collect the electrons which traverse the base.
The doping and thickness of this layer are a trade-off between collector-base junction
capacitance and breakdown voltage versus the transit time across the base-collector space
charge region and prevention of the base push-out effect. The doping in the n'-collector
was 5 x 1016 Si/cm3. As shown in section 3.5.1 for this value of collector doping, the
onset of the base push-out effect should be prevented up to collector current densities close
to 105 A/cm2.
In the high-frequency structures, a heavily doped subcollector (5 x 1018 Si/cm3) is
used to reduce the collector series and contact resistances. The thickness of this layer is a
tradeoff between sheet resistance and device isolation.
4.2
Device Layout
Proper design of the device layout (mask design) is critical in order to access the
various layers in the epitaxial structure during fabrication, define the active device
geometry, and make contact to the active devices for electrical characterization.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
4 .2 .1
59
Layout of DC/Low-Frequency Devices
Figure 4.1 shows the layout of the DC/Low-Frequency (noise) devices. The
Isolation Mask
Y
Surface Passivation Mask
— -V —
Emitter
Base
Figure 4.1 Layout of the DC/Low-Frequency devices.
important features of the mask design are 1) device isolation by means of a reversed biased
base-collector pn junction 2) a backside collector contact 3) a base ohmic metal nearly
surrounding the emitter mesa in order to reduce the base-contact resistance [4-3] ( a
continuous base ohmic contact metal totally surrounding the emitter mesa results in an even
lower base-contact resistance, but liftoff of the base-contact metal is more difficult) and 4)
a minimum emitter width of 4 um so that the contact hole to the emitter can be opened
directly on top of the emitter mesa. The interconnection metal, which forms the device
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
60
feeds and the contact pads (not shown in figure 4.1), is added to facilitate electrical probing
of the devices. The contact pads should be sufficiently large compared to the diameter of
the probe in order to facilitate electrical measurement. The exact geometry of the device
feeds and contact pads are not important for DC/Low-Frequency Devices.
4 .2 .3
Layout of High-Frequency Devices
The layout of high-frequency transistors is more complicated than that of transistors
intended for DC and low-frequency characterizations: defining and contacting the small
active regions of high-frequency devices are more difficult than in larger DC transistors,
and the layout of the contact pads of microwave devices requires careful attention in order
to minimize inductive and capacitive parasitics and make the layout compatible with onwafer GHz probes.
The mask layout of the ohmic contact metals to the emitter, base, and collector
layers is shown in figure 4.2. The interconnect metallization is not shown in figure 4.2.
The active device is defined by proton implantation everywhere outside of the isolation
mask. For a typical 2 um wide emitter finger, a contact hole directly on top of the emitter
mesa is extremely difficult to open with standard optical lithography. Instead, the emitter
finger extends outside of the active region where a larger emitter contact hole can be
defined. The arrangement of the device feeds from the active device to the contact holes
facilitates the measurement of the transistor in the common-emitter configuration. The
contact hole to the base (input) and the contact hole to the collector (output) are positioned
opposite to each other with the contact hole to the emitter (common) located between the
two.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
B a se
o h m ic
m e ta l
c o n ta c t h o le
C o lle c t o r
o h m ic
A c tiv e
D e v ic e
m e ta l
lllllllll llill lllll II
X
~1
I s o la t io n M a s k
/
E m itt e r o h m i c M e t a l
Figure 4.2 Layout of ohmic-metal layers and isolation mask of high-frequency HBT.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
62
In order to measure the frequency response of microwave transistors, contact pads
must be added. The design aspects of the contact pads can be best understood by first
briefly reviewing the physical make-up of on-wafer coplanar GHz probes. Figure 4.3
shows the (a) bottom and (b) side views of a Cascade-MicroTech probe with a groundsignal-ground configuration (coplanar waveguide configuration).
Ground
Ground
Figure 4.3 Bottom (a) and side (b) views of a Cascade-MicroTech probe in the groundsignal-ground configuration.
In order to make the device layout compatible with on-wafer probes, the contact pads must
also be arranged in the coplanar waveguide configuration. S-parameter measurements (see
section 2.5) of the transistor are made by using GHz probes (one probe is connected to the
base and the other to the collector for common-emitter measurements) in conjunction with a
network analyzer (50 Q. system) after calibration to the probe tips. Calibration is performed
using an open, short, load, and thru on the well understood impedance standard substrate
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
63
provided by Cascade-MicroTech.
Before examining the actual device layout for high-frequency characterization, let's
point out some features of a standard coplanar waveguide as shown in figure 4.4.
Ground
Signal
Ground
Conductors
Substrate
Figure 4.4 Coplanar waveguide in the ground-signal-ground configuration formed by
metal conductors on top of a substrate.
Under single mode operation, the coplanar waveguide supports a quasi-TEM mode and can
thus be modeled simply as a transmission line with an inductance and capacitance per unit
length [4-4]. The characteristic impedance of the coplanar waveguide transmission line is
determined by the ratio of the center-conductor width (w) to the ground-signal spacing (s).
For a GaAs substrate, a ratio of
SL = lQQun?- yields a characteristic of 50 Q [4-5].
s
65 um
Figure 4.5 shows the layout of a high-frequency H B T . To obtain reproducible
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
Emitter
(ground)
Common-ground inductive loops
9
/
*
Base
Collector
(signal)
(signal)
Active Device
Probe Placement Mark
Emitter
(ground)
Figure 4.5 Layout of high-frequency HBT.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
65
measurements the probes have to be placed accurately during measurement o f the
transistors and the special test patterns for extracting the parasitics associated with the pads
and the device feeds (see chapter 5). Alignment marks extend out from the ground planes
to facilitate accurate probe placement. Since everything beyond the probe tips is measured,
the measured s-parameters are actually the s-parameters of the active transistor embedded
within a connection structure. In order to minimize the effects of the contact pads and the
device connections, several precautions must be observed.
The dimensions of the contact pads are designed to make probing easy. Since the
probe tips themselves have dimensions of 50 um x 50 um, the contact pads are 100 um x
100 um. Given the center-conductor width (w) is 100 um, the ground-signal separation (s)
is 65 um to make the characteristic impedance o f the transmission lines 50 £2. This helps
to minimize reflections at the transition between the probe tips and the contact pads. The
series parasitic inductance and shunt parasitic capacitance depend on the dimensions of the
contact pads and signal-ground separation. The width of the contact pad and the signalground separation are not really adjustable since they are constrained by the need to
facilitate probing and make the characteristic impedance 50 £2. The length of the contact
pads (the transmission line length) is made as short as possible in order to reduce the series
inductance [4-6] and radiation losses due to the open nature of the transmission lines [4-7].
The minimum contact pad length is limited again by the dimensions of the probe tips; the
length of the contact pads is 100 um for easy probing. It should be pointed out that contact
pads as large as 100 um x 100 um are possible for GaAs-based devices because the
substrate is semi-insulating. In addition to the shunt capacitance of the transmission line
(between the signal and ground lines), there is also the parasitic capacitance between pad
and the substrate. This capacitance is much smaller with semi-insulating GaAs substrates
than for conductive silicon substrates. For silicon devices, the pad area is reduced as much
as possible (typically 50 um x 50 um) at the expense of more difficult probing.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
66
A connection tying the common lead (emitter) of the active transistor to the ground
lines is also necessary. The two grounds must be connected to suppress the propagation of
unbalanced ("slotline") modes [4-8]. Such slotline modes are undesirable since they
produce additional losses, as well as complicating the simple transmission-line treatment of
the coplanar waveguide. Current from both the input and output will flow through this
common connection.
The resulting ground loops give rise to a common-ground
inductance, which is proportional to the area of the loop. The common-ground inductance
is minimized by squeezing together the ground and signal lines [4-9]. This can not be done
arbitrarily since the aspect ratio of the signal/ground lines must be kept constant to maintain
the characteristic impedance of the transmission line. By tapering the width of the signal
lines in the direction of the active device and angling the ground lines inward, the aspect
ratio of the signal/ground lines is maintained approximately constant and the commonground loop area is reduced. This design also reduces the abruptness of the discontinuities
between the contact pads and the ohmic device feeds as a consequence of the large
discrepancy in size between the active device and the contact pads. At microwave
frequencies, such discontinuities modify the electric and magnetic field patterns, producing
additional parasitic capacitances and inductances.
Although a wide connection between the ground lines and the emitter contact hole is
desirable to minimize the common-ground inductance, a narrow connection is desirable to
reduce the crossover capacitance between this connection and the ohmic collector device
feed below (the two are separated by 6000 A of polyimide). As a compromise, the
common-ground connection is wide over most of the spacing between the ground line and
the emitter contact hole, but in the region near the active device, the width is narrowed to 10
um. Using a simple parallel-plate capacitor approximation, we find for a crossover area of
40 um2 that this crossover capacitance is about 1 fF. Over the 1 to 26.5 GHz frequency
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
67
range, the effect of this crossover capacitance is not important. An alternative to eliminate
this crossover capacitance is to use a common-collector, rather than a common-emitter
layout.
4 .3
4 .3 .1
Device Fabrication
Fabrication of DC/Low-Frequency Devices
The mask layout used in this fabrication process in shown in figure 4.1. Figure
4.6 illustrates the schematic cross sections of a HBT after each critical step in the
fabrication sequence. The process starts with a sputter deposition of Tungsten (W) over
the whole wafer surface. Ti/Al is evaporated and patterned by lift off to form the emitter
ohmic metal, which acts naturally as a mask for removing the undesired W by Reactive
Ion Etching (RIE) with C2F6 and SF6. The W is slightly overetched to leave a ~ 0.2 um
undercut beneath the Ti/Al pattern.
The Ti/Al pattern also serves as a wet-etch mask for defining the emitter mesa.
First, H3PC>4:H202:H20 = 3:1:50 at 20 °C is used to remove the InGaAs contact cap layer.
Then, NH4OH:H202 = 1:200 at 10 °C is used to selectively etch the GaAs cap layer with
respect to the AlGaAs emitter [4-10].
Part of the top GaAs/AlGaAs grading layer is
consumed before the etching stops. This selective etching step exposes a thin ( ~ 700 A)
AlGaAs layer above the p+ base layer.
The effect of this thin AlGaAs layer (AlGaAs
surface passivation ledge) is explained in section 3.3.2.
In order to protect the AlGaAs surface passivation ledge while etching with
H3PC>4:H202:H20 = 3:1:50 at 20 °C to contact the base, a photoresist mask is defined.
Since this etchant is not selective, the etching relies solely on timing and a known etch rate.
After etching to the base, the base-contact metal (Ti/Pt/Au) is evaporated. The active device
region is then defined by using an isolation mask and etching the area outside this masked
region well into the collector layer. The collector-contact metal (Au/Ge/Ni/Au) is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
evaporated on the backside of the substrate and alloyed at 440 °C for 15 sec. Next, a
GaAs
3aAs
AlGaAs
Emitter Metal
7
f
InGaAs
Sputter Tungsten
Emitter Metal Evaporation
and Liftoff
RIE Tungsten
T5/A1
AlGaAs
^
a
Emitter Mesa Definition
Etch to 700A above Base
Photoresist
®ase
Metal
m iii il l J J
AlGaAs
Ledge
Application of Surface
Passivation Mask
Formation of AlGaAs
Surface Passivation Ledge
Base Metal Evaporation
and Liftoff
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
69
Application of Isolation
Mask
Collector
Etch well into collector
n + Substrate
Evaporate Backside
Collector Metal
Collector
n + Substrate
ezzzzzzz:
Collector Metal
Interconnect Metal
Polyimide
Spin Coat Polyimide
Define and RIE contact
holes
Collector
n + Substrate
Evaporate Interconnect
Metal
z
Collector Metal
Figure 4.6 Fabrication process flow of DC/Low-Frequency HBTs.
polyimide layer ( ~ 6000 A) is spun on and baked to form an isolation dielectric. After the
contact holes are opened, the exposed polyimide is removed by RIE with oxygen. Finally,
the interconnect metal (Ti/Au) is evaporated and lifted off.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
4 .3 .2
70
Fabrication of High-Frequency Devices
Figures 4.2 and 4.5 show the mask layout for the fabrication of microwave HBTs.
Figure 4.6 highlights the process sequence. The process starts with a dual proton implant
to isolate the active device defined by the isolation mask. The masking material for the
implantation consists of a 1.5 um aluminum layer on top of a 7000 A SiC>2 layer deposited
by Plasma Enhanced Chemical Vapor Deposition (PECVD). The isolation damage is
created with an implant of dose = 2 x 1015/cm 2 at 200 KeV, followed by an implant of
dose = 1 x 1015/cm2 at 70 KeV. The substrate is rotated 45 ° from the major flat and tilted
7° to avoid channeling. Proper selection of the implant parameters is critical for insuring
proper device isolation all the way down to the semi-insulating substrate without causing
excessive leakage currents in the emitter-base junction.
Section 4.3.3 discusses the
important considerations for using proton implantation as a means for isolating
AlGaAs/GaAs HBT epitaxial layers.
After isolation bombardment, the aluminum and oxide masking materials are
removed by wet etching. H jP O ^ ^ O = 1:1 at 70 °C is used to remove the aluminum layer
and HF:H20 = 1:6 at room temperature is used to remove the oxide. The formation of the
emitter-contact metal and definition of the emitter mesa down to approximately 700 A above
the base are done in the same fashion as in the DC/Low-frequency process described in the
previous section.
Next, a 1000 A nitride layer is deposited by PECVD over the entire wafer, and
subsequently etched by RIE with C^Fg. As a result of the directional nature of RIE, the
nitride layer beneath the Ti/Al remains and forms a thin isolating sidewall against the edges
of the emitter mesa. This nitride sidewall allows the definition of a short AlGaAs
passivation ledge ( ~ 0.2 um) during the etch down to the base. The base-contact metal
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
(Ti/Pt/Au) is then evaporated and lifted off. The combination of the undercut beneath the
AAA
___________________________ AAA
AAA
A A
A A
A A
Proton Implantation
to Define Active
D evice
Subcollector
Semi-insulating
Substrate
Ti/Al
[rgrrm j^
-W
AAA
AAA
Sputter Tungsten
A A <
A A
A A .
A A
Emitter Metal Evaporation
and Liftoff
RIE Tungsten
Nitride
Sidewall
i m m rrn
Emitter Mesa Definition
Nitride Deposition and
Formation of Nitride
Sidew all
Etch Down to Base
Base
Metal ^
j»flL
Base Metal Evaporation
and Liftoff
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
72
Define Collector
Etch to Subcollector
Subcollector
Collector Metal
Evaporation and
L ifto ff
Spin Coat
Polyim ide
Figure 4.7 Fabrication process of microwave HBTs.
Ti/Al emitter contact and nitride sidewall allows the base metal to be self-aligned to the
emitter without any shorting between the base metal and the emitter mesa.
Contact to the subcollector is made by etching with
= 3:1:50 at
20 °C, evaporating Au/Ge/Ni/Au, and alloying at 440 °C for 15 sec. Finally, polyimide is
spun on to planarize the device and the interconnect metal (Ti/Au) is deposited to contact the
different ohmic metals via contact holes. A detailed step by step process description is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
73
compiled in Appendix A.
4 .3 .3
Proton Isolation Implant
Proper implant isolation can convert doped layers into highly resistive layers by
creating defects that trap electrons and holes, and thus reduce the conductivity.
The
damage profile of a finished device depends on the implant parameters, dose and energy,
and the thermal stability of the implant species during device fabrication. Since the HBT
epi-layer structure is about 1.1 to 1.2 um thick, protons were chosen as the implant species
because of the availability of equipment capable of implanting this species all the way
through the conducting epitaxial layers to the semi-insulating substrate. Based on the
proton damage range statistics from the TRIM Monte-Carlo ion-implantation simulation
code [4-11], an implant energy of 200 KeV was chosen to center the damage peak of the
first implant in the subcollector layer, and an implant energy of 70 KeV was selected to
place the damage peak of the second implant at the back edge of the base layer.
As shown in figure 4.8, the damage profile caused by proton implantation takes the
form of a skewed Gaussian [4-12], where Rp is the projected range and ARp is the
standard deviation. By approximating the actual linear-Gaussian profile with a linear-linear
distribution peaking at Rp and reaching zero at the surface and at a depth = Rp + 2ARp
[4-13] and finding the area under the curve, we can calculate the required dose to create the
desired maximum damage density, D'max. Based on TRIM simulations and experimental
data [4-12], 3 electrons are trapped per implanted proton. Thus, for an as-implanted
structure, we would select the dose to give a D'max which is three times the doping
concentration. However, since the generated damage is exposed to the alloying of the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
74
Implant Damage Density
D'max
Dmax
Rn + 2 A R
Depth
Figure 4.8 Distribution of damage density vs depth (solid line) and piecewise-linear
approximation (dashed line).
collector-contact metal at 440 °C, which tends to anneal the as-implanted damage, a much
larger ratio of D'max/Doping = 30 to 40 was empirically found necessary in order to retain
the damage after the collector anneal.
4 .4
4 .4 .1
Experimental Results
DC Results
Figure 4.9 is a comparison of typical gummel plots of 4 um x 10 um emitter
Al0 3GaQ 7As/GaAs HBTs with and without the AlGaAs surface passivation ledge. While
the two devices have nominally identical collector currents, the magnitude of the base
current of the device with the AlGaAs ledge is lower than that of the device without the
AlGaAs ledge (with a bare extrinsic base surface). The base current of the device without
the AlGaAs ledge is dominated by recombination at the extrinsic base surface due to the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
75
no AlGaAs ledge
with AlGaAs ledge
b, no ledge
= 1.19
b, ledge
= 1.40
A = 4 um x 10 im
0.7
0.8
0.9
1.2
1.4
V be W
Figure 4.9 Comparison of Gummel plots of 4 um x 10 um emitter Al0 3Ga0 7As/GaAs
HBTs with and without the AlGaAs surface passivation ledge.
high surface recombination velocity of GaAs. The experimentally observed base current
ideality factor of 1.19 for the device with an exposed extrinsic base surface is close to the
unity dependence predicted in section 3.3. Incorporation of the AlGaAs ledge reduces the
magnitude of the total base current by decreasing the extrinsic base surface recombination
current component (see section 3.3.2). However, the base current ideality factor of the
device with AlGaAs ledge (= 1.40) is larger than that of the device without the ledge. This
occurs because with the suppression of the extrinsic-base surface recombination
component, the base current component due to recombination in the emitter-base space
charge region (ideality factor of 2) becomes more important, which is reflected by an
increase in the ideality factor of the overall base current
Recombination at the extrinsic base surface is the main degradation mechanism of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
76
the DC current gain in AlGaAs/GaAs HBTs, particularly as the emitter perimeter-to-area
ratio (Pg/Ag) increases. Lee et al [4-14] showed that the inverse of the DC current gain is a
linear function of Pg/Ag and takes the following form:
jbulk
jsurface n
J-=V+
V~r
J3
Jc Ac
where (3 is the dc current gain, JbbuIk is the bulk base current density, Jbsurface is the base
surface recombination current density, and Jc is the collector current density. Thus, at a
fixed collector current density, the slope is a measure of the excess base current from
surface recombination. Figure 4.10 shows a typical plot of the inverse of the current gain
as a function of Pg/Ag for different size Al0 3Gao 7As/GaAs HBTs used in this study.
0.05
J c = 5 x 10 A/cm
68
0.04
x = 0.3
0.03
No AlGaAs ledge
4)
CQ
^
0.02
0.01
With AlGaAs ledge
0.2
0.4
0.6
0.8
Pe/Ae (1/um)
Figure 4.10 Inverse DC current gain vs Pg/Ae for different size Al0 3Gao 7As/GaAs HBTs
with and without the AlGaAs surface passivation ledge.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
77
As expected, the slope of the devices with the AlGaAs passivation ledge is smaller than
that of devices without the ledge. For emitter dimensions of 4 um x 10 um, the current
gain of the device with the AlGaAs ledge is approximately 3.5 times higher than that of the
device without the AlGaAs ledge. For emitter dimensions of 20 um x 20 um, the current
gain difference between devices with and without the ledge is about a factor of 1.5.
Figure 4.11 is a comparison of the collector current dependences of the DC current
gain of 4 um x 10 um HBTs with and without the AlGaAs passivation ledge for A1 mole
fractions x - 0.2 and x = 0.3.
200
100
x = 0.3, with ledge
x = 0.2, with ledge
(0
a>
m
x = 0.2, No ledge
x = 0.3, No ledge
A = 4 x 10 um
Collector
Current
(A)
Figure 4.11 Comparison of the collector current dependences of the DC current gain of 4
um x 10 um emitter HBTs with and without the AlGaAs passivation ledge for A1 mole
fractions x = 0.2 and x = 0.3.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
78
For both cases with and without the AlGaAs ledge, the gain of the x = 0.3 device is higher
than that of the x = 0.2 device, particularly at low collector currents.
If Ic = Ico
exp(Vbe/KT) and Ib = 1 ^ expO /^nK T), then p «= Ic 1' 1^n. Thus, the steeper slopes of the
x = 0.2 gain curves indicate a higher base current ideality factor (n). The base current
ideality factor of a x = 0.2 device is higher than that of a x = 0.3 device because of the
smaller energy gap within the emitter-base space charge region, which allows more spacecharge recombination (n = 2). Thus, at low collector currents the x = 0.2 devices will have
a smaller current gain than the x = 0.3 devices, despite the fact that the injection efficiency
Alo 2Gao 8-As/GaAs heterojunction is more than adequate (see section 3.2). However, at
moderate collector currents the current gains of the x = 0.2 devices approach the gains of
the x = 0.3 devices. At a collector current density of 5 x 103 A/cm2 (Ic = 2 mA), the gains
of x = 0.2 devices are only 10 to 15% lower than that of the x = 0.3 devices.
4.4.2
Microwave Results
Figure 4.12 shows the frequency response of a typical high-frequency HBT with
an emitter-base junction area of approximately 16 um2 and a base-collector junction area of
96 um2. The device is biased at a collector current of Ic =15 mA and a collector-base
voltage of Vcb = 0.5 V. By extrapolating the unilateral power and current gains to unity (0
dB) at 20 dB/decade, we obtain ft = 31 GHz and fmax = 33 GHz.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
30
Ic = 15 mA,’ Vcb, = 0.5 V
»0o
25
79
- V
■V %
h
21
= 16um2
2
A ^^um 2
20
tt
■o
'w'
e
'53
O
" u
V
\
15
n.
V L
20dB/decade
N.
10 —
\
5
l
0
1
.
l.i
.1
f
= 33 GHz
max
1 1 1 111
f( = 31 GHz X
1 1 I 11
1
10
30
100
Frequency (GHz)
Figure 4.12 Unilateral power gain (U) and common-emitter short-circuit current gain (h2 i)
vs frequency of a HBT biased at Ic = 15 mA and Vcb = 0.5 V.
Figure 4.13 shows the collector current dependences of ft and fmax of the same
device described in figure 4.12. It should be pointed out that these results are obtained
from a structure with a relaxed geometry. Furthermore, relatively high values of ft and
fmax 3X6 achieved simultaneously: the device structure has not been geared to maximize
one of these figures of merit at the expense of the other.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
80
35
s
S
O
cn
’5
max
30
e
s
a"
<
uu
25
©
20
fa
3
U
15
10
Collector Current (mA)
Figure 4.13 Collector current dependence of ft and fmax of device described in figure 4.12.
In order to interpret these results, figure 4.14 compares the total emitter-to-collector delay
(Tec) derived from the measure ft values with the values computed from the charge-control
model (described in section 3.5.1) using the known geometrical and measured material
parameters. As can be seen, the charge-control model predicts the experimental results
fairly well. In order to expose the dominant time delay, each o f the individual delays (see
section 3.5.1) is also plotted. This comparison of the constituent time delays reveals that
the dominant component is the RC time constant associated with charging the base-collector
junction capacitance. This results because of the large parasitic emitter and collector
resistances. In an optimized design, each delay contributes approximately equally to ft.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
10.0
81
EC, cal
a
o-
—
"a5
1.0
CSC
s
H
e' jeb
slope =-1
0.1
10
1
Collector Current (mA)
Figure 4.14 Comparison of the total emitter-to-collector delay (TEC) derived from the
measure ft values with the values computed from the charge-control model and the
calculated individual delay components.
Table 4.2 compares the values of fmax calculated from the classical expression
(equation 3.25) with the experimental values at two different values of collector current.
For convenience the classical expression for fmax is shown below
w
”
•
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
< 4 ' 2 )
Chapter 4 Device Design, Fabrication, and Experimental DC and Microwave Results
82
The total base resistance is calculated from the expression [4-15]
r» - r “ +
-& i
t
*2 W
'
v * )
<4-3)
where Rbi is the intrinsic base resistance, R'b con is the dc base contact resistance, ps is the
base sheet resistance, se is the active emitter width, Le is the active emitter length, Lb is
active base-contact length, sb is the active base contact width, and rc is specific base
contact resistivity. Using the measured values o f rc = 8 x 10'5 Q-cm2 and ps = 850 £2/sq;
the known geometry se = 2 um, Le = 8 um, sb = 2 um ,
= 8 um, and sbc = 3 um
(base contact/collector contact misalignment); and a calculated base-collector junction
capacitance per unit area, we computed values of fmax from equation 4.2.
Ic(mA)
^t^neas (GHz)
^max, meas (GHz)
^max, cal ('-’Hz)
2
22
23
6.8
15
31
33
8.2
Table 4.2 Comparison of fmax values calculated with the classical expression (equation
4.2) and experimental values.
As can be seen from Table 4.2, the measured values of fmax are 3 to 4 times larger than
the values predicted from the classical expression.
In order to understand this
discrepancy and better understand the limitations of our device structure, an equivalentcircuit model of the device at high frequencies was extracted. This work is described in
the next chapter.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 83
Chapter 5
Direct Extraction of the AlGaAs/GaAs HBT SmallSignal Equivalent Circuit
The morality of art consists in the perfect use of an imperfect medium.
Oscar Wilde
5.1
Introduction
Accurate, physically-based equivalent circuits are very useful tool for designing
devices with reduced parasitics and optimized microwave performance.
Although
numerical optimization is often used to produce a best fit for model-generated s-parameters
to the measured s-parameters, the resulting element values depend on the starting values
and may be non-physical. This uncertainty in the element values has been addressed by
several authors. A new de-embedding method for determining FET model parameters was
proposed by Dambrine et al [5-1]. Trew et al [5-2] described a parameter extraction
technique for a HBT that makes use of the emitter-to-collector time delay to constrain the
element values used in their optimization.
This chapter discusses a novel de-embedding technique for determining the smallsignal equivalent circuit of an AlGaAs/GaAs HBT. This technique is applied to the self­
aligned base AlGaAs/GaAs HBT (NB = 1019/cm3) described in chapter 4. Section 5.2
shows the device model and outlines the de-embedding procedure. As explained in section
5.3, most of the parasitics are first obtained from measurements of independent test
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 84
structures and those remaining are determined from measurements of the transistor.
Knowledge of these parasitic elements permits us to de-embed the intrinsic HBT from the
terminal measurements of the transistor and to compute the intrinsic element values directly
at any given frequency. Section 5.4 presents the results of the intrinsic element extraction
at various collector currents, compares the measured and the model-produced s-parameters,
and compares the measured and model-generated power and current gains; and discusses
how the device model provides feedback for improving the device performance.
5.2
De-embedding Procedure
The complete HBT small-signal, equivalent-circuit model is shown in figure 5.1.
This equivalent circuit consists of an intrinsic device surrounded by shells of parasitics. We
use the hybrid-7t representation (see section 3.5.2) to model the intrinsic device [5-3, 5 4], The intrinsic elements are r ^ , the dynamic base-emitter diode resistance referred to the
base; Cn , the base-emitter capacitance (base-emitter junction capacitance + base-charging
capacitance); Cjc int, the intrinsic base-collector capacitance; Gm, the transconductance; and
r0, the output resistance. The parasitics associated with the contact pads and the device
connections are modeled as a shunt network ( Cpbe, Cpbc,
) followed by a series
network ( Lpb, Rpb, Lpe, Rpe, LpC, Rpc ). The base-collector junction capacitance is
divided into intrinsic (Cjc in[) and extrinsic (Cjc ext) parts. RE and Rc are the series emitter
resistance and the series collector resistance. The base-contact impedance is represented by
a parallel RC combination (Rbcon II C5 con) in series with the lateral resistance of the base
semiconductor underneath the base contact. This resistance and the base semiconductor
resistance underneath the emitter are combined into one base-spreading resistance (Rbb).
Since the parasitics are independently measured, we can de-embed the intrinsic
device from the parasitics with a few matrix manipulations. This procedure can be
summarized as follows:
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 85
Shunt Parasitics
E xtrinsic Base-Collector Parasitics
jc.ext
Rb.con
HH
b*c*
■jc.int
b.con
pbe
G
V
gm
pee
c '» '
Intrinsic HBT
Series P arasitics
Figure 5.1 Complete HBT small-signal equivalent circuit.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 86
1)
measure the s-parameters of the transistor embedded in its parasitic environment
2)
convert the s-parameters to y-parameters and subtract the parasitic shunt elements
^■pbe’ ^pbc’ an<^ Cpce
3)
convert the new y-parameters to z-parameters and subtract the parasitic series
elements Lpb, Rpb, Rb con II Cb con, L ^ , R ^ , RE> L ^ , R ^ , and Rc
4)
convert the new z-parameters to y-parameters and subtract the parasitic shunt
element Cjc ext
5)
convert the new y-parameters to z-parameters and subtract the parasitic series
element Rbb
6)
convert the new z-parameters to y-parameters that correspond to the intrinsic HBT
y-parameters.
Applying simple circuit analysis, we can directly compute the intrinsic element
values from the intrinsic HBT y-parameters with the following expressions [5-5]:
Yb'e' = Y ll + Yi2
(5.1)
Yb-C*= -Y i2
(5.2)
Yc'e' = Y22 + Y i2
(5 .3 )
Y g m = Y 2i - Y i 2
(5.4)
„
_ im(Yb'e')
'-"ft-
(5.5)
(5.6)
(5.8)
Gm —r^Ygm)
(5.9)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 87
where b', e’, and c' denote the terminals of the intrinsic HBT.
5 .3
Measurement of Parasitic Elements
The s-parameters from HBTs and test patterns on the same die were measured over
the frequency range of 1 to 18 GHz by means of on-wafer Cascade MicroTech Probes
(model # WPH-105-150 with SMA connector elbows (see section 4.2)) and a HP 8510A
network analyzer.
5 .3 .1
Open Test Structure
The parasitics associated with the pads were determined by measuring a test pattern
which consisted of only the pads [5-6]. Measurements of the open test structure were
modeled as a pi-network of capacitors. Figure 5.2 shows the open test structure and the
corresponding circuit model. The simplicity of this model is a direct consequence of a
semi-insulating GaAs substrate and proper device isolation down to the semi-insulating
substrate. In contrast, suitable pad-parasitic correction for a silicon device is more
complex since the modeling of a conducting substrate requires several additional elements
[5-7]. The pad capacitance values were Cpbe = 13 fF and Cpce = 20 fF. The isolation
between the pads was 38 dB, which corresponds to a capacitance Cpbc of 3 fF. Figure 5.3
shows the the frequency dependences of the parasitic capacitances. Nearly constant values
were observed from 1 to 18 GHz with deviations from the mean values being less than 5%.
5 .3 .2
Shorted Test Structure
The parasitic device-connection impedances were determined by measuring a test
pattern which consists of the pads, the device feeds, and a short replacing the transistor .
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit
Base
Emitter
Collector
Emitter
Cpbc
B
I ~
C pbe
k Cpee i
_ 0
c
E
Figure 5.2 Open test structure and corresponding circuit model.
30
pee
Q>
e
CQ
20
pbe
'3
M
B.
n
U
10
pbe
0
0
5
10
15
20
Frequency (GHz)
Figure 5.3 Frequency dependences of parasitic capacitances.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit
Base
Short
Emitter
Collector
Emitter
Figure 5.4 Shorted test structure and corresponding circuit model.
60
50
SB
a
oo
c
a
o
3
■o
c
40
30
20
10
0
0
5
10
15
20
Frequency (GHz)
Figure 5.5 Frequency dependences of parasitic inductances.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 90
The shorted test structure is modeled as a T-network of series resistors and inductors.
Figure 5.4 shows the shorted test structure and the corresponding circuit model. The
parasitic inductance values were Lpb = 47 pH, Lpe = 12 pH, and Lpc = 50 pH. The
parasitic resistance values were Rpb = 4 Q, Rpe = 5.5
and RpC = 5 Q. Figure 5.5
shows the frequency dependences of the parasitic inductances. The constant values again
show that the assumptions used are valid.
5 .3 .3
Series Emitter and Collector Resistances
The series emitter resistance was determined by means of the open-collector method
[5-8]. The open-collector method gives the series emitter resistance (RE) in the form
VceSac = 0) = £ £ ln ( a i) + (RE + RM)lB
(5.10)
where Vces is the collector-emitter saturation voltage, Ic is the collector current, otj is the
inverse common-base current gain, Ig is the base current, and RM is the probe contact
resistance. The measured value determined by the open-collector method agrees with the
emitter resistance calculated from the geometry, the measured emitter sheet resistance, and
the measured emitter specific contact resistivity. The collector resistance was calculated
from known geometrical and material parameters. In addition, the sum of the emitter
resistance and the collector resistance was obtained from RF measurements of the
resistance between the collector and the emitter (Rce) of a HBT whose base was opencircuited. Rce agrees quite well with the sum of the values obtained from calculation and the
open-collector method. Table 5.1 summarizes the resistance values.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 91
Table 5.1
RE,cal
R E,oc ^
Rc,cal ^
R ce ^
22
24.5
18
45
Table 5.1 Comparison of the series emitter and collector resistances. The resistances are
defined as follows:
RE cal is the calculated series emitter resistance; RE oc is the series
emitter resistance determined by the open-collector method; Rc ca] is the calculated series
collector resistance; and Rce ( ~ RE + Rc ) is the measured RF resistance between the
collector and the emitter with the base open-circuited. The measured resistance values have
been corrected for the resistance of the device feeds (Rpe and R ^ ).
5 .3 .4
Extrinsic Base-Collector Test Structure
The extrinsic base-collector parasitics were determined from RF measurements of a
HBT structure whose emitter layer and base layer underneath the emitter had been removed
by etching. Figure 5.6 shows the schematic cross section and the corresponding circuit
models after the capacitances and inductances associated with the pads and the device
connections have been removed. The lumped, parallel RC combination model for the base
contact is derived from Berger's [5-9] lossy transmission line model (see Appendix B).
Because the structure is symmetric, the equivalent circuit in figure 5.6 (a) can be replaced
with the simplified circuit shown in figure 5.6 (b).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit
B
B
2R ,
2R
b.con
2 R b.con \
2 R b.con
p+ b a s e
b.con/ 2
n-
c o lle c to r
n+
c o lle cto r
c
(a)
b.con
1
^jc.ext
R pb
B
R pc
2
H
H H
^b.con
2 be =
1/Y2 2 = " b e *
' X bb
(b)
Figure 5.6 (a) Schematic cross section of the extrinsic base-collector
test structure and the corresponding circuit model, (b) Simplified
circuit model.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 93
The impedance between the base and collector (Z ^) can be readily obtained from the
simplified circuit in figure 5.6 (b) and is expressed as;
Zbe —
— R b c "*■j X b c
( 5 .1 1 )
where
Rbc = Rpb+
1
;
rr+ Rpc + Rc
+ (C0Rb,con Cb.conf
(C 0R b,con P C b .c
X bc = - 1
CO 1 + (coRb.con Cb,con)^J
COCjc,ext
(5.12)
ii i
W\Cb,con
CjCiext
J___ (5.13)
coCjc .e x t
Rb con is the shunt base-contact resistance and Cb con is the shunt base-contact capacitance.
This proposed base-contact model was verified by comparing the computed and the
experimental values of R ^ . Figure 5.7 shows frequency dependences of the measured
Rj^ and the calculated Rbc. The good agreement between the experimental value and the
computed value supports the validity of our base-contact model.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 94
£
o
120
2
80
o
U
<
u
cC
/>
Z
CO
o
40
c
•S
X
w
20
Frequency (GHz)
Figure 5.7 Frequency dependence of the computed and the experimental values of the
extrinsic base-collector resistance (R^).
Because the extrinsic base-collector capacitance (Cjc>ext) is effectively in series with a much
larger base-contact capacitance (Cjj Con >10 Cjc ex t), Cjc ext dominates X^c and thus it can
be calculated from equation (13). Figure 5.8 shows the frequency dependence of CJc ext
of the same test structure . Cjc ext deviates less than 3% from 60 fF over the entire
measured frequency range.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 95
100
'.
u
V 0)
— u
e
;jo es
*
tm
V Km
J
2 «
ea «
uu
‘5s .2£
'£ z
* 5=
w
80
jc.ext
A#
60
40
20
0
0
5
10
15
20
Frequency (GHz)
Figure 5.8 Frequency dependence of the extrinsic base-collector junction capacitance
(^jc.ext)-
5.4
5.4 .1
Results and Discussion
Intrinsic Element Extraction
The validity of the equivalent circuit was tested by examining the frequency
dependence of the intrinsic circuit elements. Figure 5.9 shows the intrinsic capacitances Cn
and Cjc int as a function of frequency of a device operating at Ic = 1 mA, Vcb = 0.5V. Cn
deviates less than 5% from 130 fF and Cjc jnt deviates less than 8% from 11 fF across the
entire frequency range. The small deviations of these intrinsic capacitances verify that both
the circuit topology and the de-embedding procedure are accurate and reliable.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 96
200
40
160
30
fe
a.
U
120
20
80
40
0
5
10
15
20
F req u en cy (GHz)
Figure 5.9 Frequency dependence of the base-emitter capacitance (C^) and the intrinsic
base-collector junction capacitance (Cjc int).
The de-embedding procedure is executed on a Sun-3/60 Workstation in less than one
minute. This method has been used to extract the circuit elements at many bias points.
Table 5.2 shows the intrinsic circuit elements of a HBT operating at collector currents from
1 to 8 mA and the bias-independent parasitic elements.
5 .4 .2
Broad-Band S-parameters
Figure 5.10 compares the measured s-parameters with the model-produced s-
parameters of an HBT operating at Ic = 2 mA, Vcb = 0.5V. Throughout the entire 1-18
GHz frequency range, the computed s-parameters agree well with the experimental data.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 D irect Extraction o f the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 97
2
Geomet r y:
Intrinsic
A
16 u m 2
11 2 um
cb
Elements:
current
(mA)
Cn
(ft)
G m (mS)
C j c .in , ( f R)
428
130
11
2
291
240
11
77
4
186
420
11
154
133
670
12
307
R 0 > 0. 6 M
37
Elements
Rb,con = 1 6 0
C j c ,e X. = 8 8
R
13 fF
Lpb = 47 p h
4 «
c
=
ft
18
=
C pbc
L p®
■
Rpe
"
ft
Re
3fF
12 pH
5.5
C b,con -
ft
= 24.5
= 20
C pce
L pc
R pc
0- 8 8 PF
ft
fF
=
5 0 pH
£
R bb = 8 6 n
=
( f F)
II
Parasitic
R*
1
8
R pb =
=
cn
collector
A eb =
Table 5.2 Equivalent Circuit Elements (Vcb = 0.5 V)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ft
Chapter 5 D irect Extraction o f the AlGaAs/GaAs H B T Small-Signal Equivalent Circuit
90
S21/2
S12 X 2
180
0.2 \
50
0 .4 )
0 .6 1 0.8
1.0
100
S22
-J io
$11
-J50
Figure 5.10 Comparison of the measured (solid lines) and the calculated
(dashed lines) s-parameters over the frequency range o f 1 to 18 GHz of a
HBT operating at Ic = 2 mA and Vcb = 0.5 V.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 99
5 .4 .3
Calculation of Power and Current Gains
The equivalent-circuit model can also be used to calculate the unilateral power and
common-emitter current gains of the device and predict ft and fmax. Figure 5.11 compares
the measured unilateral power and common-emitter current gains with the experimental
values. As can be seen from figure 5.11, a good fit is obtained.
30
25
20
Model = solid lines
15
10
5
0
1
100
10
Frequency
(GHz)
Figure 5.11 Comparison of the measured and model-produced unilateral power gain (U)
and common-emitter short-circuit current gain (h21) of a HBT biased at Ic = 8 mA and Vcb
= 0.5 V.
5 .4 .4
Feedback for Device Design
Equivalent circuits are valuable diagnostic tools that provide information for
improving the microwave performance of HBTs. The most obvious information obtained
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 100
from this modeling work is that the series emitter and collector resistances need to be
reduced. We have also learned that the layout of the device in the coplanar waveguide
configuration and the device isolation are both quite suitable, since the associated parasitic
capacitances and inductances are very small and have a negligible effect on ft and fmax
(although this was not explicitly shown). Since the extrinsic portion of the base-collector
junction capacitance is five to six times larger than the intrinsic portion, the device
fabrication must focus on reducing the extrinsic part. The most illuminating information
resulting from this modeling work is that for AlGaAs/GaAs HBTs with moderately (by
GaAs standards) doped bases (NB = 1019 /cm3) the base-contact impedance decreases at
high frequencies because the base-contact metal/semiconductor capacitance (Cb con) shunts
the interfacial contact resistance (Rb con). This explains the discrepancy between the
measured values of fmax and the values predicted with the classical expression (Table 4.2)
where the base contact is assumed to present a purely resistive impedance. It should be
pointed out that the shunting action of Cb con is only important for moderately doped
bases. For more heavily doped bases (NB = 5 x 1019 /cm3), the specific contact resistivity
(and therefore, Rb con (see Appendix B)) is over an order of magnitude lower than that of
bases doped at NB = 1019 /cm3.
In AlGaAs/GaAs HBTs with heavily doped bases,
R b con shunts Cb con , rather the vice-versa, and thus, the existence of Cbcon is
irrelevant.
5.5
Summary
This chapter has described a new, direct technique for determining the small-signal
equivalent circuit of a HBT. The parasitic elements are largely determined from
measurements of test structures, reducing the number of elements determined from
measurements of the transistor. The intrinsic circuit elements are evaluated from yparameter data, which is de-embedded from the known parasitics. The equivalent-circuit
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 5 Direct Extraction of the AlGaAs/GaAs HBT Small-Signal Equivalent Circuit 101
elements are uniquely determined at any frequency. The validity of this technique is
confirmed by showing the frequency independence of the extracted circuit elements. The
equivalent circuit models the HBT s-parameters over a wide range of collector currents.
The accuracy of this technique is high, and the computer time is negligible compared to
numerical optimization without independent extraction of the major parasitic elements. The
physical basis of the equivalent circuit and the accurate determination of the element values
provides feedback for intelligently modifying the device design and fabrication for
improved high-frequency performance.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
102
Chapter 6
Low-Frequency Noise Characterization of Npn
AlGaAs/GaAs HBTs
On the surface there is infinite variety of things; at the base
a simplicity of cause.
Ralph Waldo Emerson
This chapter is concerned with the low-frequency noise properties of Npn
A lxGaj_xAs/GaAs HBTs. Section 6.1 briefly reviews the present state of research
involving the low-frequency noise of AlGaAs/GaAs HBTs, provides the motivation for
this work, and illustrates the device structure. Section 6.2 describes the arrangement for
noise measurements. Section 6.3 outlines the theoretical background for understanding the
possible sources of noise in HBTs. Section 6.4 presents the results of the noise
measurements as a function of the various device parameters and measurement conditions.
These results are discussed in section 6.5, and section 6.6 highlights the prospects for
further reduction of low-frequency noise in HBTs.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
6.1
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
103
Introduction
Low-frequency noise can limit the bandwidth and stability of a wide variety of
integrated circuits [6-1]. It also degrades the spectral purity of nonlinear microwave
circuits, such as oscillators and mixers where the low-frequency, base-band noise upconverts as noise sidebands around the RF carrier signal [6-2]. While the high-frequency
performance of AlGaAs/GaAs heterojunction bipolar transistors (HBTs) is well-established
[6-3], the noise properties have largely been overlooked. Jue et al [6-4] measured the
the low-frequency noise of large-area Al0 3Ga0 7As/GaAs double heterojunction bipolar
transistors (DHBTs). They observed a noise spectra that was 1/f in nature from 1 to 100
Hz, but exhibited variations in dependence with both temperature and bias level up to 100
KHz. Recent experimental results of small-geometry Al0 2sGaQ 75As/GaAs HBTs [6-5],
indicate that a HBT incorporating a thin, depleted AlGaAs layer ( AlGaAs surface
passivation ledge) over the extrinsic base region shows dramatically lower 1/f noise over
the frequency range of 10 Hz to 10 KHz as compared to a device without the AlGaAs
ledge. However, these devices still exhibited a large anomalous noise "bump" over the
frequency range of 10 KHz to 1 MHz.
In this chapter, we present an in-depth study of the low-frequency noise
properties of AlGaAs/GaAs HBTs. Low- frequency noise measurements (100 Hz to 10
MHz) of Npn AlGaAs/GaAs HBTs with and without AlGaAs ledges were made as a
function of bias current, device geometry, aluminum mole fraction in the emitter, and
temperature. These measurements were performed on HBTs from two separate wafers.
The epitaxial layers in the HBT structures were defined in Table 4.1 (section 4.1). The
first wafer had an aluminum mole fraction of x = 0.2 in the emitter. The second wafer had
an aluminum mole fraction of x=0.3, but was otherwise identical to the first. The device
layout and fabrication process were also described in sections 4.2.1 and 4.3.1,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
104
respectively. Figure 6.1 shows schematic cross sections of HBTs without (a) the AlGaAs
passivation ledge and with (b) the AlGaAs passivation ledge. The mask design was such
that for a given device geometry, devices with and without the ledge were adjacent to each
other.
Exposed
Base
Surface
rs
Emitter
Emitter
r r ? ? /7 V
Collector Metal
/t t
H
Base
Base
v
Depleted
AlGaAs
Ledge
; v / T r y
Collector Metal
(a)
; ; '7S\
(b)
Figure 6.1 Schematic cross sections of HBTs (a) without and (b) with the AlGaAs surface
passivation ledge covering the extrinsic base surface.
The noise measurements show the existence of three distinct regions in the noise
spectra: a 1/f Iine-shape at the lower end of the measured frequency range, a Lorenztian
spectrum (noise "bump") at intermediate frequencies, and a white noise region at the higher
end of the measured frequency range. Most of the previous noise studies of AlGaAs/GaAs
HBTs have tended to follow a phenomenological approach, with no clear understanding of
the precise physical origin of the noise sources. We have strived to interpret our results in
terms of a specific physically-based model. The origin of the 1/f noise of small-geometry
AlGaAs/GaAs HBTs is clearly demonstrated to result primarily from fluctuations in the
extrinsic-base surface recombination velocity, as explained originally by Fonger [6-6]. Our
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
105
measurements also suggest that the anomalous noise "bump" is generated by a trap,
possibly the DX center [6-7], in the AlGaAs near the surface of the emitter-base junction.
It is also shown that the use of an AlGaAs ledge and the reduction of the Al mole fraction
(x) from x = 0.3 to x = 0.2 significantly reduces the low-frequency noise of our
AlGaAs/GaAs HBTs without adversely affecting the current gain. In addition to the simple
modifications to the epitaxial structure and fabrication process that were implemented in this
study, we briefly examine some of the important issues for further improvement of the
low-frequency noise performance of HBTs.
6.2
Measurement Set-up
The arrangement for noise measurements of HBTs is shown in figure 6.2. The
device noise measurements were preceded with a system calibration to measure the residual
noise of the system. With the transistor in the common-emitter configuration, the noise
power at the collector was measured from 100 Hz to 10 MHz with a selective level meter
(HP 3586). Using a simple low-frequency circuit model for the transistor and the gain of
the system , we refer the noise at the output back to the input and define an equivalent
input base noise current [6-8]. Calculation of the equivalent input base noise current is
described in Appendix C. Over 100 measurements were made to verify the consistency of
the measured data. To compare the two wafers, devices from approximately the same
spot
of each wafer were chosen to minimize variations due to possible growth
nonuniformities.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
106
+ V
Low-frequency
Low-frequency
F ilte r
F ilte r
Amp
Signal Source
0 ase
HP 3336
Bias
j
Collector
Signal Analyzer
Device
Bias
HP 3586
Under
T
Test
Figure 6.2 Experimental setup for low-frequency noise measurements of HBTs.
6 .3
Theory:
Sources of Noise
The low-frequency noise of bipolar transistors arises from several independent
noise sources [6-9].
This section describes the most probable noise sources in
AlGaAs/GaAs HBTs according to the shapes of the respective noise spectral densities. The
noise spectral density is defined in section 2.6.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
6 .3 .1
White Noise
107
White noise refers to noise whose spectral density is independent of frequency.
The major part of the white noise of the equivalent input base noise current is due to shot
noise from the dc base current. Shot noise arises from the discrete nature of electronic
charge and the fact that the current flow consists of individual pulses, which flow with
random spacing. The potential barrier of a pn junction establishes a spatial dispersion of
the current pulses exiting the space charge region, as well as within the space charge
region. For those frequencies which have periods much less than the width of each current
pulse, van der Ziel [6-10] has shown that the current spectral density of shot noise
associated with a dc current, I, is the well known expression
S,!hol(f) = 2ql
(6.1)
where f is the frequency and q is the electronic charge. This random dispersion of carrier
flow through the various regions of the bipolar transistor produces full shot noise in each
distinct component of the dc base current, as well as the dc collector current [6-11].
6 .3 .2
1/f Noise
This type of noise derives its name from the fact that its spectral density is roughly
inversely proportional to frequency. Although many mechanisms (both bulk and surface
related) for 1/f noise have been suggested, its physical origins have not been conclusively
determined, even in conventional homojunction silicon bipolar transistors. Many theoretical
studies have associated 1/f noise with surface states [6-6, 6-12, 6-13, 6-14]. Numerous
experimental results support this surface noise theory [6-15, 6-16, 6-17, 6-18, 6-19, 6-5].
Using a simplified parallel-plane geometry representation of a bipolar transistor (quite
similar to the actual mesa structure of a modem AlGaAs/GaAs HBT), Fonger originally
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
108
attributed 1/f noise to fluctuations in the current recombining at the extrinsic base surface.
More specifically, fluctuations in the occupancy of surface traps perturb the surface
recombination velocity, which in turn produces fluctuations in the base surface
recombination current, I^s. Assuming the effective surface recombination velocity (s) is a
constant, independent of position and excess minority carrier density at the base surface
(n’), Fonger expressed the base surface recombination current as
( 6 -2)
where A is the surface area. In general, s is not a constant that can be decoupled from n ';
however, when s is large ( s ~ 106 cm/s for a bare GaAs surface ), it can be treated as a
constant because the net recombination rate is limited by the supply of minority carries from
the quasi-neutral base, rather than the np product at the surface [6-20, section 3.3.1].
Thus, the assumption of a constant s in equation (6.2) is valid.
A fluctuation in s, 8s, gives rise to a fluctuation in I|jS, 8lbS, of the form
5lbs = ^ 5 s
(6.3)
Averaging and making a Fourier expansion, Fonger showed that the spectral density
associated with a fluctuation in I(jS could be expressed as
S i b s O ^szV 8^
(6.4)
where Ss(f) is the spectral density associated with a fluctuation in s and characterizes the
noise contribution of the surface. Fonger empirically determined Ss(f) to vary as 1/f and
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
109
later, van der Ziel [6-14] demonstrated that
(6.5)
where Nt is the number of surface states.
In order to carry out the integration in equation (6.2), the excess electron profile
must be found, which requires solving a two-dimensional current continuity equation in the
base with the appropriate boundary condition at the base surface. Such an exact solution
requires numerical analysis [6-21], Fonger approximated equation (6.2) with the following
expression
Ibs —qsn (0)ASiCff
( 6 . 6)
where n'(0) is the excess injected minority electron concentration at the base edge of the
emitter-base space charge region and As eff is an "effective" base surface area where
electrons recombine. Equation (6.6) reflects the fact that most of the recombination at the
exposed extrinsic base surface occurs close to the perimeter of the emitter. Since n’(0) can
be expressed in terms of the collector current ( Ic = qAeDnn'(0)/Wb ), the spectral density
can be rewritten as
(6.7)
where Ic is the collector current, Wb is the base width, Dn is the electron diffusivity in the
base, Ae is the emitter area, and s has been assumed to be independent of bias level, as
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
110
mentioned before. For a rectangular shaped emitter mesa, As eff can be approximated as
As eff = LdPe, where Ld is the lateral electron diffusion length in the base, and Pe is the
emitter perimeter length. Substituting for As eff in equation (6.7), we can express the
spectral density associated with recombination at the extrinsic base surface as
(6 .8)
The quadratic dependence of the spectral density on collector current results because Ibs «=
n'(0)
e(lVbe/KT « Ic. The recombination at the extrinsic base surface occurs after
electrons have surmounted the entire pn junction barrier. Simulations also indicate that for
a collector current density greater than about 102 A/cm2 and at high values of s, the ideality
factor of the extrinsic base surface recombination current of AlGaAs/GaAs HBTs is nearly
unity [6-20]. The ideality factors of the base current of our 4 um x 10 um emitter HBTs
without the AlGaAs ledges were typically in the range of 1.15 to 1.20.
Fonger’s own
measurements of transistors showed a spectral density proportional to Ic2/n , where the
ideality factor n ranged from 1 to 1.33.
Recombination of carriers at the surface of the emitter-base space charge region also
gives rise to a current Ibes . Fluctuations in Ibes also cause 1/f noise [6-9]. Since the
recombination in this case occurs on the average before carriers have climbed the full pn
junction barrier, Ibes0-1 eqVbe/,nkT
Ic^ n, where 1 < n < 2. Therefore, s I t e (f) should
vary as \ %n. Several workers have observed this dependence in devices [6-16, 6- 22].
6 .3 .3
Burst Noise
Electrical noise in the form o f random "bursts", caused by a switching between
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
111
two states (in the simplest case), is sometimes observed in semiconductor devices Burst
noise in bipolar transistors is often associated with traps or generation-recombination (g-r)
centers near the emitter-base space charge region. The origin of burst noise has been
attributed to dislocations in the bulk near the emitter-base junction [6-23], dislocations at
the surface of the emitter-base space charge region [6-24], emitter-edge dislocations [625], g-r centers associated with metallic precipitates in the vicinity of the pn junction [626], and defects near the surface of the emitter-base junction [28], Burst noise has also
been found in AlGaAs/GaAs HBTs [6-4, 6- 5, 6-28, 6-29].
Most of these previous studies resulted in phenomenological models, which
describe burst noise as a function of bias conditions and temperature. The results of these
investigations have shown that the spectral density of burst noise for a single time constant
process, takes the form of a Lorenztian
Sfburst^ “ T
(6.9)
where T is the time constant of the trap involved and m is a constant in the range of 0.5 to
2. The exact value of m is likely related to the spatial location of the dominant noiseproducing trap in the emitter-base space charge region. The thermally activated time
constant can be expressed as
( 6 . 10)
For a small distribution of time constants, the Lorenztian spectrum in equation (6.9)
becomes somewhat rounded.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
6 .4
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
112
Results
6 .4 .1
Current Dependence
The low-frequency noise was measured as a function of bias current. These results
are important because noise theories predict specific bias dependences. Figure 6.3 shows
the typical equivalent input base noise current spectral densities and the various components
of noise current for a 4 um x 10 um emitter Alg^Gag 7As/GaAs HBT operating in the linear
region (Vce = 3V) at three different bias currents .
N
-17
EC
<
A = 4 x 10 um , No ledge
,-18
sum(Ic=lmA)
,-19
Ic=lmA,Ib=48uA
Ic=2mAJb=86uA
Ic=4mA,Ib=158uA
,-20
1/f Noise
,-21
,-22
c
V
ula
s;
,-23
Shot Noise = 2ql.
3
u
Burst Noise
,-24
Frequency (Hz)
Figure 6.3 Equivalent input base noise current spectral densities and the various noise
components for a 4 um x 10 um emitter A1q 3Gag 7As/GaAs HBT without the AlGaAs
passivation ledge and operating in the linear region at three different bias currents.
This device does not have an AIGaAs passivation ledge. While the noise spectrum changes
considerably as the bias current is varied, almost no change was observed when VCb was
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
113
varied over the range of l V < V c b< 4 V. Each noise spectrum has been resolved into
three components: 1/f noise, burst noise, and white noise. Figure 6.4
shows the
equivalent input base noise current spectral density at 100 Hz, Sjb(100Hz), as a function
of collector current.
,-17
,-18
S., (100Hz), slope = 2.1
Vi
C
V
Q
es
,-19
slope = - 0.84
,-20
e
,-21
u
3
u
,-22
I,burst(0
Hz)l sl°Pe = 1,88
0.5
Collector Current (mA)
Figure 6.4 Collector current dependence of the equivalent input base noise current spectral
density at 100 Hz, SIb(100Hz); the magnitude of the extracted low-frequency burst-noise
plateau, Sjburst (0 Hz); and the burst-noise time constant, T, of the 4 um x 10 um emitter
AIq3GaQ 7As/GaAs HBT described in figure 6.3.
SIb(100Hz) varies as lc2A over the measured collector current range, very close to the
collector dependence predicted by equation (6.8). At intermediate frequencies ( 104 - 106
Hz ) each noise spectrum exhibits a striking "bump", characteristic o f burst noise. It
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
114
should be pointed out that the magnitude of the noise "bump" is considerably higher than
the shot noise level. This "bump" was fitted to a Lorenztian spectrum of the form given by
equation (6.9), although the slight smearing of the Lorenztian-like data suggests that a few
closely spaced time constants exist, rather than a single one. The magnitude of the lowfrequency plateau, Siburst(f=0Hz), increases and the time constant, T, decreases with
increasing collector current. The variations of Siburs[(f=0Hz) and X with collector current
are also shown in figure 6.4. The functional dependences were found to be 5>iburst( f = 0 H z )
Ic 1-88 and X
Ic"0-84. The white noise at the high end of the measured frequency range
is approximately equal to the shot noise, Sjb shot = 2 q ljj, where I5 is the total dc base
current.
6 .4 .2
Device Geometry Dependence
Examination of the geometry dependence of the low-frequency noise assists in
separating surface and bulk effects. At a constant collector current, equation (6 .8 ) predicts
a quadratic dependence of the 1/f noise on emitter perimeter-to-area ratio (Pg/Ag). Figure
6.5 is a comparison of the equivalent input base noise current spectral densities of
Alo.3Gao.7 As/GaAs HBTs, without the AlGaAs ledge, for two different values of Pg/Ag
at a fixed collector current of 20 mA. As the Pg/Ae increases from 0.2 to 0.33, Sjb(100Hz)
increases by a factor of 2.5. This result agrees closely with the predicted increase of 2.7,
calculated from the (Pe/A e)2 dependence in the 1/f surface theory of equation (6 .8 ).
Khajezadeh [6-17] also observed a similar dependence of 1/f noise on emitter geometry in
silicon bipolar transistors. Figure 6.5 shows that the noise "bump" , in the frequency
range of 10 KHz to 1 MHz, becomes more pronounced and increases in magnitude as the
Pg/Ae increases.
The constant collector current condition limits the range of emitter areas, which can
be examined, because of the likelihood of appreciable self-heating of the junction as the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
115
emitter area is reduced. Because noise-producing processes can be thermally activated (as
in equation 10), variations in the junction temperature can affect the noise measurements. A
N
E
<
,-16
Ic = 20 mA
,-17
No AlGaAs ledge
x = 0.3
,-18
A = 12 x 12 um
c
n
s
,-19
V .= “
*3
E D.
c
vu
u
S
U
,-20
Ae = 20 x 20 um
P /A =0.2
,-21
Frequency (Hz)
Figure 6.5 Comparison of the equivalent input base noise current spectral densities of
AIq 3GaQ 7As/GaAs HBTs without the AlGaAs passivation ledge for two different emitter
perimeter-to-area ratios at a fixed collector current of 20 mA.
more useful measurement condition for comparing the effect of device geometry is a
constant collector current density. Figure 6.6 compares the equivalent input base noise
current spectral densities of four Alo.3Gao.7 As/GaAs devices with emitter perimeters
ranging from 28 to 80 um (emitter areas ranging from 40 to 400 um 2 without AlGaAs
ledges) at a fixed collector current density of 5 x 10^ A/cm2 . These curves show that at
a
constant collector current density, larger-perimeter (larger-area) devices have inferior
low-frequency noise performance. These results can be compared to Fonger's theory by
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
116
Pe=28um
Pe=32um
Pe=48um
Pe=80um
J = 5 xlO A/cm
C
No AlGaAs ledge
x = 0.3
'
i 11 " " I
i
' " " "I
i
i i ’""I
V
i
i i m ill
\
n;
i
'
""“1
^
=
I
-=
=
__
i
' 11’" '!
i
i I ’" "
Frequency (Hz)
Figure 6.6 Comparison of the equivalent input base noise current spectral densities of
AIq 3GaQ 7As/GaAs HBTs without the AlGaAs passivation ledge for four different emitter
perimeters (areas) at a fixed collector current density of 5 x 103 A/cm2.
substituting Ic = JcAe into equation (6 . 8). This substitution yields the expression
Sibsff)= Jc p ^ ) 2 Pe2 Ss(f)
(6.11)
Thus, the spectral density associated with recombination at the extrinsic base surface varies
as the square of the emitter perimeter when the collector current density is held constant.
Figure 6.7 shows Sjb(100Hz) as a function of emitter perimeter for various size
AlojGaojAs/GaAs devices without the AlGaAs ledge at a fixed collector current density of
5 x 103 A/cm2. A best fit slope of 2.46 is in reasonable agreement with the quadratic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
117
perimeter dependence predicted by equation (6 . 11).
i-17
J =5x10
A /c m
Sj^lOOHz), Slope = 2.46
r 18
,-19
Emitter Perimeter (um)
Figure 6.7 Emitter perimeter dependence of the equivalent input base noise current
spectral density at 100 Hz of A1q 3Ga 0 7 As/GaAs HBTs without the AlGaAs passivation
ledge at a fixed collector current density of 5 x 103 A/cm2.
6 .4 .3
Effect of AlGaAs Surface Passivation Ledge
Recombination at the extrinsic base surface is a major mechanism of gain
degradation in small-geometry AlGaAs/GaAs HBTs [6-30]. The band bending at the
surface, which results from Fermi-level pinning, creates an electric field which attracts
minority carriers toward the surface. For GaAs, the combination of a high surface state
density ( on the order of ~ 1014 cm'2), and the band bending at the surface results in a very
large effective surface recombination velocity. The use of a thin, depleted AlGaAs layer
over the extrinsic base region changes the band bending at the surface and thus suppresses
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
118
the concentration of minority carrier electrons that reach the base surface, particularly near
the edge of the emitter (see section 3.3.2). In this way, the effective surface recombination
velocity is reduced. The use of this AlGaAs surface passivation ledge surrounding the
emitter mesa has been shown to dramatically improve the current gain of HBTs [6-21, 6 31, 6-32].
In this study, for emitter dimensions of 4 um x 10 um, the current gain of a device
with the AlGaAs ledge is approximately 3.5 times higher than that of a device without the
AlGaAs ledge. For emitter dimensions of 20 um x 20 um the current gain difference
between devices with and without the ledge is about a factor of 1.5 (see section 4.4.1).
Figure 6.8 is a comparison of the equivalent input base noise current spectral densities of
N
X
© <
i-17
A = 20 um x 20 um
r 18
solid = No AlGaAs ledge
i-19
open = With AlGaAs ledge
r20
,-21
A = 4 um x 10 um
,-22
e<u
tm
u
S
U
,-23
Jc = 5 X Hr A/cm'
x = 0.3
,-24
Frequency (Hz)
Figure 6.8
A Iq
Comparison of the equivalent input base noise current spectral densities of
3GaQ7As/GaAs HBTs with and without the AlGaAs ledge for two different emitter
areas at a fixed collector current density of 5 x 103 A/cm2.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
119
AlojGaQ 7As/GaAs HBTs with and without the AlGaAs ledge for two different emitter
areas at a fixed collector current density of 5 x 103 A/cm2. For emitter dimensions of 4
um x 10 um, Sib(100Hz) of the device with the AlGaAs passivation ledge is approximately
a factor of ten (10 dB) lower than that of a similar size HBT without the ledge. For emitter
dimensions of 20 um x 20 um, the 1/f noise difference between devices with and without
the ledge is about a factor of two. These results indicate that a reduction in the number of
electrons reaching the base surface (where they interact with surface states) by an AlGaAs
ledge plays an increasingly important role in improving the 1/f noise behavior as emitter
dimensions are scaled downward. Since the reduction of the surface recombination current
decreases the total base current, the corresponding shot noise level is also decreased. It
should also be pointed out that the magnitude and the shape of the noise "bump" changes
with the incorporation of the AlGaAs ledge.
Figure 6.9 compares the collector current dependences of Sjb(100Hz) of
AlojGao 7As/GaAs HBTs with and without the AlGaAs ledge for two emitter dimensions
of 4 um x 10 um and 20 um x 20 um. For emitter dimensions of 4 um x 10 um,
Sjb(100Hz) of devices both with and without the AlGaAs ledge varies approximately as the
square of the collector current. These dependences suggest that for the 4 um x 10 um
device, even though the AlGaAs ledge reduces the surface noise, the ledge does not
completely eliminate the surface effect and some residual surface noise still remains as the
dominant 1/f noise source. On the other hand, for emitter dimensions of 20 um x 20 um,
while Sjb(100Hz) of the device without the AlGaAs ledge exhibits a near quadratic
dependence on collector current ( again due to the base surface effect), Sjb(100Hz) of the
1 CQ
device with the AlGaAs ledge shows a weaker dependence on collector current ( Ic1--3 ).
This dependence of the smaller Pg/Ag device with the ledge indicates that some other 1/f
source (perhaps associated with a bulk process [6-33]) is emerging, as the extrinsic base
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
120
surface 1/f noise is minimized.
C
N
,-16
2 X
u S
V
N
o s
Z 2
<u
V)
« ss
QQ
>>
S ’w
a B
B
solid = No AlGaAs ledge
open = With AlGaAs ledge
i-17
slope = 1.91
J
•-
slope = 1.58
i-18
slope = 1.96
a
B
—
um x 10 um
,-20
L.
U
a>
‘3 c.
o* c«
U
A = 20 um x 20 um
,-19
1
/
slope = 2.13
1
,0
,2
Collector Current (mA)
Figure 6.9 Comparison of collector current dependence of the equivalent input base noise
current spectral densities at 100 Hz of Al0 3Ga 0 7 As/GaAs HBTs with and without the
AlGaAs passivation ledge for two different emitter sizes.
While most of the results presented thus far have been for devices with 5 um wide
ledges, HBTs with ledge widths (L) of
L = 0.2, 1, and 2 um were also measured.
Figure 6.10 shows the ledge width dependence of the DC current gain and Sjb(100Hz).
For L > 1 um, both the current gain and the 1/f noise are approximately constant. As long
as the base-contact metal is spaced from the emitter mesa by a distance much greater than
the electron lateral diffusion length in the base (Lj ~ 0.5 um [6-34, 6-35]), most of the
electrons injected into the base will not reach the base contact so that the base metal plays
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
121
no role in the recombination process. However, when the distance between the emitter
100
50
O
o>
u
l_
3
U
u
Q
JC = 5 x l 0 3 Aycm2
A = 4 um x 10 um, x = 0.3
L_f
Ledge
I_____
10
Width (um)
Figure 6.10 AlGaAs passivation ledge width dependence of the equivalent input base noise
current spectral density at 100 Hz and the DC current gain for a 4 um x 10 um emitter
AIq 3GaQ 7As/GaAs HBT at a fixed collector current density of 5 x 103 A/cm2.
mesa and the base-metal is on the order of L j (as with the self-aligned base HBT whose
emitter-mesa to base-contact spacing is ~ 0.2 um), the high recombination velocity of the
base-contact metal influences the recombination at the surface and reduces the benefit of the
passivation ledge both in terms of current gain and 1/f noise. These results suggest a
performance tradeoff in ultimate speed (requiring a self-aligned base HBT) versus current
gain and 1/f noise.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
6 .4 .4
122
Effect of Aluminum Mole Fraction in the Emitter
As is well-known, AlGaAs has a number of electron traps because A1 is more
reactive than Ga [6-36]. Because burst noise in bipolar transistors is usually attributed to
defects in the emitter-base space charge region and the depletion region mainly extends into
the more lightly doped AlGaAs emitter, we measured the low-frequency noise of HBTs
with A1 mole fractions of x = 0.2 and x = 0.3 in the emitters. Figure 6.11 compares the
equivalent input base noise current spectral densities of 4 um x 10 um emitter
AlgjGaQ ?As/GaAs HBTs with those of similar size A1q^Gap 8 As/GaAs HBTs at a
collector current density of 5 x 103 A/cm2.
,-18
■nm
-*— x=0.3, No ledge
N
,-19
<
,-20
a
'5
s
v
Q
s &
—•
■— x=0.2, No ledge
■a— x=0.3, with ledge
ci— x=0.2, with ledge
,-21
,-22
u
■e£ °v
v e.
a
,-23
= 4 um x 10 um
J = 5 x 1(T A/cm
,-24
uni
Frequency (Hz)
Figure 6.11 Comparison of the equivalent input base noise current spectral densities of 4
um x 10 um emitter HBTs with and without the AlGaAs passivation ledge for two different
A1 mole fractions (x) at a fixed collector current density of 5 x 103 A/cm2.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
123
Both devices with and without the AlGaAs ledge were compared. While the x = 0.2 and x
= 0.3 devices exhibited comparable 1/f noise and white noise levels, the noise "bump" was
virtually eliminated for the x = 0.2 devices. For both devices with and without the AlGaAs
ledge, Sjb (50 KHz) of the x = 0.2 devices was roughly three times lower than that of the x
= 0.3 devices. These results are very strong evidence that the noise "bump" is associated
with traps in the AlGaAs emitter.
6.4.5
Temperature Dependence
The temperature variation of low-frequency noise provides a diagnostic tool for
identifying traps, which produce Lorenztian-shaped spectra [6-37,6-38,6-39]. Figure 6.12
N
4>
05
18
S
'o <
w
Z
V
05
«
10
T= 0 C
10,-19
T =14 C
>>
T = 30 C
-20
05 c
10
Q
s
a
,G
fi
a
>
T = 45 C
T = 62 C
u
*CVJ
1C'21
a
in
-22
G
) 10
‘3 O
lm
O'
w u
9
u 10 23
io 1
x = 0.3
No AlGaAs ledge
A =4umxl0um
e
I t n m l
i( r
io3
io4
io 3
io°
I
I I m i l l
io '
io°
Frequency (Hz)
Figure 6.12 Temperature dependence of the equivalent input base noise current spectral
density of a 4 um x 10 um emitter Al0 3 Ga 0 7 As/GaAs HBT without the AlGaAs
passivation ledge at a fixed collector current of 2 mA.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
124
shows the temperature dependence of the equivalent input base noise current spectral
density of a typical 4 um x 10 um Al0 3Gao 7As/GaAs HBT without the AlGaAs ledge at
a fixed collector current of 2 mA. Both the 1/f noise and the shot noise (as expected) are
nearly invariant with temperature. From equation (6 .8 ), assuming Ss(f) is independent of
temperature, we can see that at a constant collector current, the 1/f noise only has a weak
temperature dependence via the electron diffusivity in the base. Fonger's experimental
results of bipolar transistors also show that the 1/f surface noise is insensitive to
temperature [6 -6 ].
On the other hand, the noise "bump", which has the form of a slightly broadened
Lorenztian spectrum, has a large temperature dependence, indicating that the burst-noise
"bump" arises from a thermally activated process, such as trapping/detrapping. For
increasing temperature, the magnitude of the low-frequency plateau decreases and the
characteristic -3dB frequency (inversely related to the time constant) shifts toward higher
frequencies, because the time constant decreases with increasing temperature as predicted
by equation (6 . 11 ).
In order to fit the noise "bump" at each temperature to Lorenztian components, the
equivalent input base noise current spectral density was multiplied by frequency [6-39].
This operation expands the detailed structure of the "bump" and allows a more accurate fit
for extracting the time constant(s). Figure 6.13 shows the equivalent input base noise
current spectral density at T = 14 °C multiplied by frequency and the 1/f noise and two
Lorenztian components fitted to the data. The time constant is simply determined from the
-3 dB characteristic frequency now located at the peak of the modified Lorenztian spectrum
from the expression
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
125
where fp is the peak frequency. In figure 6.13, the peak in the "bump" at approximately
100 kHz was fitted with two Lorenztian spectra having time constants less than one decade
apart. Arrhenius plots of both time constants were linear with inverse temperature and
Sum
1/f noise
TL
- Trap 1
Trap 2
101
102
103
104
white noise
105
106
107
108
Frequency (Hz)
Figure 6.13 Equivalent input base noise current spectral density at T = 14 C multiplied by
frequency and the various noise components fitted to the data for the 4 um x 10 um emitter
Al0 3GaQ 7As/GaAs HBT described in figure 6.12.
activation energies of 0.18 eV and 0.20 eV were estimated from the plot.
6.5
Discussion
The results in the previous sections clearly demonstrate that the 1/f noise of
AlGaAs/GaAs HBTs originates primarily at the extrinsic base surface. The observed white
noise is consistent with the shot noise of the dc base current. However, the origin of the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
126
burst-noise "bump" requires some discussion.
While the exact nature of the burst-noise "bump" is not known , its dependence
on A1 composition suggests that the noise "bump" is associated with an AlGaAs trap in the
neutral emitter or the emitter-base space charge region. The observed time constants are
close to the capture and emission times of the DX center in Si-doped AlGaAs [6-40] and
suggest it as a possible candidate for the noise-"bump"-producing trap.
In order to relate the burst-noise "bump" in terms of the DX center, the pertinent
properties of DX centers are briefly reviewed (see also section 2.3). The DX center is a
deep donor level which roughly follows the L conduction-band minimum. In AlxGaj.xAs
the DX center is the lowest-energy state of the donor atom when x > 0.22 and thus it
determines the free-electron concentration. The binding energy of the DX center (Ed = Er
- EDx , where Er is the T conduction-band edge and EDX is the energy level of the DX
center, both referenced to the valence-band edge) changes with A1 mole fraction and the
electron emission (E ^) and capture energy (Ec^) barriers affect the movement of electrons
to and from the DX center [6-41]. Figure 6.14 summarizes these energies.
ET
Ef
Figure 6.14 Effective potential barrier for thermal capture and emission of electrons by the
DX center.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
127
With respect to the phenomenological capture barrier of Figure 6.14, the lifetime of an
empty trap before it captures an electron, T.''n , takes the form
'n
(6.13)
where a (=
exp(-Ec /kT)) is the capture cross section, (u) is the average carrier
velocity, n is the free-electron concentration, and Ep is the Fermi-level. For Si-doped
AlxG aj.xAs over the composition range of 0.2 < x < 0.4, Ec decreases from about 0.44
to 0.20 eV [6-41]. The average lifetime of an occupied trap before it emits an electron, Te ,
n
can be expressed as
(6.14)
where en is the emission probability.
E.
n
(= 0.43 eV for Si-doped AlxG a j.xAs) is
independent of A1 mole fraction [6-42].
Burst noise is usually analyzed with a random switching model [6-43]. In the
simplest case, a current fluctuation is produced by the switching between two states.
Fluctuations in the occupancy of DX centers are possible both in the neutral emitter and the
emitter-base space charge region. The switching model for burst noise is most easily
applied in the charge neutral emitter. Because the heterojunction suppresses the backinjection of holes, only electron capture and emission are significant in the neutral emitter.
Copeland [6-44] has shown that the noise current spectral density, Si(f)> caused by the
alternating emission and capture of electrons in uniform, neutral n-type material is given by
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
128
(6.15)
and
(6.16)
where I is the dc current flowing through the material under an applied voltage, nt is the
trapped electron concentration, Sn[(f) is the noise spectral density associated with a
fluctuation in the trapped electron concentration, nD is the steady state free electron
concentration (nQ= n - 5n), V is the volume, and Nt is the density of traps. In equation
(6.15), Copeland incorporated the relationship between the dc current and the free-electron
density in a semiconductor resistor: I = qnv, where v is the electron drift velocity.
This
leads to the simple expression -3l/3nt = 31/3n ~ I/n0. Physically this means that when a
single electron is emitted (AN = 1), the dc current increases (= magnitude of noise current
pulse) by A 1 = ^
AN ^
Several comments regarding equation (6.15) are in order. Since the activation of T
is a mixture of capture and emission, the activation energy determined from the noise
spectrum (Ea) can not, in general, be deconvolved into capture and emission components.
Thus, the interpretation of Ea is not straightforward. It is generally assumed that
fluctuations in the occupancies of traps contribute significant noise only when the Fermilevel is close to the energy level of the trap because trap levels more than a few kT below
the Fermi-level are permanently filled and those more than a few kT above the Fermi level
are permanently empty. Based on this reasoning it has been shown by Hsu [6-45] that
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
129
Sn (f) is a maximum when X. = 2 Te .B y examination of equations (6.13) and (6.14)
t
n
n
(with the prefactors of Tc and Te being the same), Sn (f) is a maximum when EDX = Ep
n
n
t
+ kT ln2, close to the commonly believed condition that the maximum noise contribution
occurs when the Fermi-level and the trap level are the same. When the concentration of
traps is much less than the doping density, n0 does not depend on the trap occupancy, and
(
91
2
— j = (^-) is a constant so that Sj(f) is
maximized when S„t(f) is a maximum. However, in the case of the DX center, the DX
concentration is equal to the donor concentration (Np>) so that n0 depends strongly on the
occupancy of DX centers. Since nc decreases as the occupancy of DX centers exceeds one
half, Sj(f) is not a maximum when
ef
“ edx-
This implies that in the special case of
current fluctuations in a semiconductor resistor with Nt ~ ND and a fixed current flowing
through it, Sj(f) is not a maximum when the trap level and Fermi level coincide. This
results as a compromise between the maximum number of random noise current pulses
generated when the occupancy is one half and a larger noise current pulse amplitude (at a
constant dc current) as the occupancy increases (n0 decreases).
For our AlGaAs/GaAs HBT structure in thermal equilibrium, the boundary
conditions of the degenerate n-GaAs contact and the p-base layers set the Fermi-level about
0.5 kT above the conduction-band edge in the neutral emitter. Assuming a degeneracy
factor of 2, we find that over 97% of the DX centers in the neutral A l^ G a ^ A s emitter (Ed
= 90 meV) are filled and slightly less than half of the DX centers in the neutral
Al0 2GaQ gAs emitter (Ed = 10 meV) are occupied. Under normal operation, a small electric
field is dropped across the neutral emitter to support the drift current through the neutral
emitter, analogous to the structure considered by Copeland. Based on the arguments in the
previous paragraph, it's not clear whether the x= 0.2 or x = 0.3 neutral emitter makes a
larger noise contribution, despite the fact that in the neutral AIq^Ga^ gAs emitter, the DX
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
and Fermi levels are quite close .
130
Since the free-electron concentration in the neutral
emitter is essentially constant ( n ~ n0, since 8 n is small), T = f(Tc ) = f(n) is fixed. This
contradicts our results (shown in figures 6.3 and 6.4) that the time constant of the burstnoise "bump" decreases with increasing current. Thus, DX centers in the neutral emitter
can not be primarily responsible for the observed noise "bump".
Because the free-electron concentration (and the hole concentration) and hence the
time constant do vary with collector current in the emitter-base space charge region, a
fluctuation in the occupancies of traps on the emitter side of the transition region is more
likely the origin of the burst-noise "bump". For the typical bias conditions used in this
study, the emitter-base space charge region is mainly within the compositionally graded
emitter-base matching layer. Analysis of the noise current produced within the emitterbase space charge region is more complicated than in the neutral region for several reasons.
First, since the carrier concentrations change with position across the emitter-base space
charge region, T also varies with position, and in general, a single time constant does not
exist.
Second, although the DX center is typically considered an electron trap (because of
the level's proximity to the conduction band), Watanabe et al [6-46] found that at room
temperature, the hole capture cross section of the DX center in Si-doped AlxGaj_xAs is
about three to four orders of magnitude larger than the electron capture cross section of the
DX center. The hole capture energy barrier was determined to be 0.14 eV. Thus, despite
the fact that the hole concentration decreases rapidly inside the emitter-base space charge
region with distance away from the hetero-interface, the interaction of holes and DX centers
could be important inside the emitter-base grading layer due to the DX center's strong
affinity for holes. Including the possibility of hole capture and emission processes, we
can modify equation (6.16) to express the noise time constant as
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
=
131
( 6 1 7)
TCn
Xen
XCp
x ep
where Tc is the lifetime of an empty trap before it captures an electron, Tc is the lifetime
of a filled trap before it captures a hole, Te is the lifetime of a filled trap before it emits an
n
electron , and Te is the lifetime of an empty trap before it emits a hole. With the inclusion
p
of hole processes, the interpretation of the activation energy extracted from the noise
measurements becomes even more difficult. Depending on the noise current generating
mechanism, fluctuations in both the electron current and the hole recombination current in
the emitter-base space charge region are possible. It should be pointed out that many of
the simplifying assumptions often used in calculating recombination rates (currents), such
as the trap level being at midgap and equal hole and electron capture cross sections, are not
valid in the case of the DX center.
Finally, at thermal equilibrium, the built-in voltage of the emitter-base junction
thermally ionizes all of the DX centers within the depletion region, driving the freeelectrons into the neutral emitter. In contrast to the DX center occupancy which controls
the steady state free-electron concentration in the neutral emitter, the steady state electron
concentration inside the emitter-base space charge region under forward bias is virtually
unaffected (recombination via DX centers may affect the electron profile slightly) by the
occupancy of DX centers. Based on the arguments about equation (6.15), the amplitude of
a noise current pulse produced when a DX center within the emitter-base space charge
region emits or captures an electron (or a hole) should be independent o f the DX center
occupancy. Therefore, the maximum noise current due to fluctuations in the occupancy of
DX centers within the emitter-base space charge region should occur when EDX = Ep.
If one assumes that most of the noise contribution comes within a small region in
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
132
the emitter-base space charge region centered about the position where EDX ~ EF, then, to
first order, the noise spectra can be characterized by a small distribution of time constants.
Even for neutral n-type A ^ G a ^ A s of a fixed A1 mole fraction, a finite range of noise time
constants is expected since the number and distance of the nearest neighbor A1 atoms
influence the DX center emission and capture rates [6-40]. As a result, a smearing of the
DX center noise time constants is quite reasonable, particularly from DX centers in the
emitter-base space charge region where the A1 composition is changing.
For the AIojGag 7As/GaAs HBTs, the activation energies ( Ea = 0.18 and 0.20
eV) extracted from our noise measurements are close to the 0.162 eV trap determined by
Tutt et al [6-29] from low-frequency noise measurements of AlGaAs/GaAs HBTs. We
were unable to extract an activation energy from the noise measurements of the
A1q,2Ga0 8 As/GaAs HBTs (presumably because the noise "bump' is masked by the 1/f
noise over the measured temperature range). Our activation energies are considerably
lower than the activation energy for thermal emission of electrons typically found by
DLTS (Eej) « 0.43 eV) in Si-doped A ^ G a ^ A s .
explained by assuming that the activation of Xr
n
, Xr
This discrepancy can possibly be
p
,Te
n
, and
X„
p
all contribute to the
activation of the noise time constant, while in DLTS measurements of n-type AlGaAs, only
X. determines the measured activation energy. For the case of E nx = EF ( X
cn
equations (6.13) and (6.14) and assuming that X
=X
p
n
= Xe from
n
), equation (6.17) can be
P
rewritten as
Xen
Tcp.
where Ee eff is an effective electron emission energy barrier. The capture of holes by DX
centers acts to decrease the lifetime of electrons trapped on DX centers, which can be
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
133
viewed as a lowering of the electron emission energy barrier. This causes Ee eff < Ee as
determined from our results.
Our results (figure 6.11) showed that Sjb (50 KHz) of the AIq^Gag gAs/GaAs
HBT was roughly three times lower than that of the Alg^Gag 7As/GaAs HBT. One
possible explanation for this behavior is now examined. Because the binding energy of the
DX center is small in Alg^Gag gAs (Ed = 10 meV), the DX level becomes resonant with the
conduction band a very short distance into emitter-base grading layer. Under bias
conditions used in this study, the edge of the emitter-base space charge region pushes into
the grading layer, so that the quasi Fermi-level crosses the DX level inside the neutral
emitter rather than inside the emitter-base space charge region. Since for Alo^Gag 7As the
binding energy of the DX center is nearly an order of magnitude larger (Ed = 90 meV), the
quasi-Fermi level still crosses the DX level inside the emitter-base space charge region
under these same bias conditions, and therefore, the noise contribution from inside the
emitter-base space charge region of the x = 0.3 device is expected to be greater than that of
the x = 0.2 device.
Our extracted activation energies can also be compared with those determined from
low-frequency noise measurements of AlxGaj_xAs/GaAs HEMTs. Loreck et al [6-37]
found several deep levels with activation energies near 0.4 eV from low-frequency noise
measurements of Al0 3Gao 7As/GaAs HEMTs. From low-frequency noise measurements
of AlxG aj.xAs/GaAs HEMT-like structures, Kirtley et al [6-47] determined activation
energies of 0.433 eV (x=.14), 0.352 eV (x=0.34), and 0.381 eV (x=0.37). These
activation energies are close to the values of thermal electron emission energy found by
DLTS. Under normal HEMT operation most of the n-AlGaAs layer is depleted with only
a narrow flat-band region where parallel conduction occurs. Assuming that most of the
noise originates within this flat-band region, we can apply Copeland's model for the special
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization of Npn AlGaAs/GaAs HBTs
134
case of the DX center. Since only electron capture and emission needs to be considered
inside heavily doped n-type AlGaAs, the activation energy determined from the noise
measurements of HEMTs is more easily interpreted than that of HBTs. For the devices
with x > 0.3, the Fermi-level is pinned roughly at the DX level in the flat-band region.
From equations (6.13) and (6.14) when Ep =
Te = Tc^ so that T = 0.5 Te and hence
Ea = Ee . For the x = 0.14 device, the DX level is considerably above the Fermi-level in
the flat-band region, both levels being resonant with the conduction band. This implies
Te -1 »
Tc -1
so that T = Te
and Ea = ECn.
Thus, in the electron majority carrier
HEMT, where the Fermi-level is almost always coincident with or below the DX level,
low-frequency noise generation reflects the electron emission process.
A complete discussion of the possible mechanisms producing the noise "bump"
should also address the data that relate the noise "bump" to the device geometry. From
figure 6.5, we can see that at a fixed collector current, the noise "bump" is more
pronounced as the Pe/Ae increases (Ae decreases). From figure 6 . 8 , we can see that the
AlGaAs ledge affects the magnitude and shape of the noise "bump". Equation (6.15)
shows that for the simple case of noise caused by the switching between electron capture
and emission at a fixed dc current, Sj(f) is inversely proportional to the semiconductor's
volume and hence to the area perpendicular to the current flow. If we assume that for the
noise process inside the emitter-base space charge region (even though more than just two
switching states may exist) Sj(f) is also inversely proportional to Ae at a fixed emitter
current, this model predicts an increase in the magnitude of the noise "bump" as the emitter
area decreases, and thus would agree with the data in figure 6.5.
Another possibility is that the dominant noise-producing traps are located near the
surface of the emitter-base space-charge region. With this model, the noise "bump" is
more pronounced as the Pg/Ae increases. This model, then, is also consistent with the data
in figure 6.5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
135
If the noise bump is indeed due to fluctuations in the electron current, this would be
reflected in the noise generator i ^ (see figure C .6 in Appendix C). From equation (C .7),
2
we see that the contribution of i ce to the equivalent input base noise current spectral
.2
nO
density is i ce /p . Since the AlGaAs ledge increases the current gain, the noise "bump" in
the equivalent input base noise spectral density would be shifted down with the
incorporation of the ledge, regardless of whether the "bump" is associated with traps at the
surface or in the bulk of the emitter-base space charge region. If the noise "bump" is
actually due to fluctuations in the hole current inside the emitter-base space charge region (
2
reflected by i b-e in figure C.6 ), then the presence of the ledge would only have an effect
on the equivalent input base noise current spectral density if the traps were located at the
surface of the emitter-base space charge region. The fact that the presence of the ledge also
affects the shape of the noise "bump", as well its magnitude, suggests that the noise
"bump" is due to fluctuations in the occupancies of AlGaAs traps near the surface of the
emitter-base junction, although this has not been demonstrated unequivocally. One method
to distinguish whether the noise "bump" originates at the surface or in the bulk of the
emitter-base space charge region is to measure the noise of HBTs with a fixed Ae, but a
variable P^/A^
If the traps at the surface of the emitter-space charge region are the source of the
burst noise "bump", then constructing a quantitative model for the burst noise is extremely
difficult because the surface potential is quite a complicated function of position through the
emitter-base space charge region as a consequence of Fermi-level-pinning, which bends the
bands upward in the neutral emitter and downward in the extrinsic base region [6 -20 ].
6.6
Prospects for Reduced Low-Frequency Noise in III-V
HBT Design
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
136
The key factor in reducing the low-frequency noise of HBTs (or any device) is to
minimize fluctuations in the occupancies of trap/recombination centers, either by decreasing
the number of active states (as with the reduction of the Al mole fraction in AlGaAs/GaAs
HBTs) or by reducing the number of carriers reaching these states (as with the surface
passivation ledge in AlGaAs/GaAs HBTs).
The resulting implications for epitaxial layer
design, material selection, and device configurations are highlighted.
The 1/f surface noise performance of AlGaAs/GaAs HBTs is expected to benefit
from the same schemes that have been incoiporated in epitaxial-structure design to improve
current gain by reducing the flux of carriers to the extrinsic base surface [6-20,6-31,6-48].
A thinner base region ( as seen from equation (6 . 8 ) 1 ) allows a smaller flow of electrons
to the extrinsic base surface. This occurs because the flow of minority carrier electrons in
the base is governed by two competing boundaries, the high-recombination base surface
and the reversed biased collector-base junction. A vertical quasi-electric field in a
compositionally graded base layer bends the lateral flux of electrons toward the collectorbase junction. An abrupt emitter-base junction forms a larger potential barrier at the surface
than a graded junction, and therefore produces a smaller flux of carriers. Futhermore, an
abrupt junction can be implemented to "launch" electrons into the base with additional
kinetic energy in the vertical direction.
The choice of the epitaxial material system and donor species in the emitter provides
a means for enhancing both the 1/f and burst-noise behavior of HBTs. HBTs with base
layers composed of materials with lower surface recombination velocities than GaAs, such
as InGaAs, should exhibit lower 1/f surface noise2. Assuming the dominant source of
1Equation (6.8) predicts SIbs(0 “ Wb2Ld2. Since roughly Ld « W b, then the 1/f base surface noise
should show a very strong dependence on neutral base width, Sjbs( f ) « Wb4.
E quations (6.2), (6.6), (6.7), and (6.8) were derived based on the assumption that s is a constant The
validity of this assumption was justified for the case when s is very large. However, when s takes on a low
or moderate value, such as with InGaAs, the above equations may not be strictly valid.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
137
burst noise in AlGaAs/GaAs HBTs is associated with the donor-related DX center in the ntype AlGaAs, we can minimize the trapping effects of DX centers by selecting other III-V
alloys where the presence of DX centers is diminished or eliminated. Using a simple
model, which correlates the DX center with the L conduction-band minimum, Tachikawa
[6-49, see figure 2.6] has predicted that InP should be free of DX centers, while
Al0 48 In0 52As is close, but below, the composition where DX centers may possibly be
stable. This suggests that InP/InQ 53Gao.47As and Al0 4gInQ 52 As/In0 53Gao 47AS HBTs
show promise as low 1/f surface noise and DX-related burst-noise devices.
Because the binding energies of the DX states of Te, Se, and Sn donors in AlGaAs
are slightly smaller than that of Si, the active DX center concentration at a given Al mole
fraction can be reduced somewhat by using Te, Se, or Sn instead of Si. [6-50]. In
addition, since the emission and the capture activation energies of the DX states of Te, Se,
and Sn are considerably different than those of Si, one would expect the trapping time
constant of the burst noise to depend on the donor species [6-42]. Thus, variation of the
donor species in the AlGaAs emitter (and the resulting shift in -3dB frequency of the
Lorenztian spectrum) could prove to be a useful tool for confirming the role of the DX
center in the low-frequency behavior of AlGaAs/GaAs HBTs.
The most common AlGaAs/GaAs HBTs are of the Npn, emitter-up configuration.
It's interesting to speculate on the low-frequency noise behavior of other configurations for
AlGaAs/GaAs HBTs. Because the DX center is a donor state in AlGaAs, Pnp devices
which have acceptors, not donors, in the AlGaAs should be free of DX centers and hence
DX center related burst noise. Optimized Pnp AlGaAs/GaAs HBTs also incorporate
thinner bases than Npn devices [6-51], which should result in lower 1/f surface noise.
The use of an inverted collector-up/emitter-down (C-up) structure has been
speculated to give a substantial increase in the high-frequency performance of
AlGaAs/GaAs HBTs [6-52]. Under ideal conditions with a C-up Npn structure operating
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Low-Frequency Noise Characterization o f Npn AlGaAs/GaAs HBTs
138
in the active mode, the injection of minority carrier electrons into the base from the emitter
is buried away from surfaces. Electrons which diffuse and arrive at the collector-base
junction are exposed to surfaces, but the electron concentration at the base edge of the
collector-base junction is many orders of magnitude smaller than the electron concentration
at the base edge of the emitter-base junction. This suggests that the 1/f surface noise of Cup AlGaAs/GaAs HBTs should be better than that of conventional E-up devices.
6 .7
Summary
The dominant source of 1/f noise in our AlGaAs/GaAs HBTs was shown to be
associated with states at the extrinsic base surface. This finding is consistent with Fonger's
model which attributes the 1/f noise to fluctuations in the surface recombination velocity
and hence fluctuations in the extrinsic base surface recombination current.
Our
measurements also suggest that the origin of the anomalous noise "bump" observed at
intermediate frequencies is due to fluctuations in the occupancies of AlGaAs traps near the
surface of the emitter-base junction. The DX center was identified as a possible candidate
for this trap. A qualitative model based on the electron and hole emission and capture
processes of the DX center has been presented in order to interpret the activation energy of
the time constant extracted from the noise "bump". The results of this study suggest
several modifications in the epitaxial structure and the fabrication process that substantially
reduce the low-frequency noise of AlGaAs/GaAs HBTs. We have demonstrated that the
use of a protective AlGaAs ledge over the extrinsic base surface and a reduction of the Al
mole fraction in the emitter from 0.3 to 0.2 significantly improves the low-frequency-noise
performance of our AlGaAs/GaAs HBTs. Based on these observations, we have
suggested structures and materials to design III-V HBTs with even better low-frequency
noise performance.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
Conclusions and Suggestions for Future Research
139
Chapter 7
Conclusions and Suggestions for Future Research
Where there is no vision, people perish.
Proverbs 29:18
7.1
Conclusions
The research described in this thesis has concentrated on the microwave and low-
frequency noise characterization of Npn AlGaAs/GaAs heterojunction bipolar transistors.
AlGaAs/GaAs HBTs with a relaxed geometry can easily operate at higher frequencies than
silicon bipolar transistors. Further improvement in the microwave performance of
AlGaAs/GaAs HBTs requires very careful optimization of the epitaxial layer structure,
fabrication process, and device geometry. The equivalent circuit has proven to be an
invaluable tool for providing insight concerning the frequency response of our
AlGaAs/GaAs HBTs. In AlGaAs/GaAs HBTs, the presence of a mesa-type structure
(which exposes the base surface) and a high defect density in the AlGaAs emitter produces
additional low-frequency noise sources. Identification of these noise sources has been the
necessary first step for incorporating modifications in the epitaxial layer structure and
fabrication process that minimize these noise sources.
In the course of this study, a self-aligned base, implant-isolated AlGaAs/GaAs
HBT fabrication process was developed together with William Liu. Implementation of this
fabrication process, along with the proper device layout, has yielded devices with both ft
and fmax above 30 GHz. These are impressive results considering that a relaxed geometry
was used, no special layers in the epitaxial structure to tailor the electric field profile in
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
Conclusions and Suggestions for Future Research
140
order to provide ballistic transport were included, and the base doping was limited to 1019
Be/cnA The microwave performance of these devices is limited by relatively high parasitic
resistances, especially the emitter contact resistance.
As a means for understanding the speed limitations and potential of our HBTs, a
technique for extracting the small-signal equivalent circuit of HBTs was conceived. The
direct determination of most of the parasitic and intrinsic element values provides the model
with a sound, physical basis. The validity of this technique was confirmed by showing
the frequency independence of the circuit elements. The use of the equivalent circuit model
as a diagnostic tool for improving device design was demonstrated. It was shown that the
shunt capacitance associated with the base contact greatly reduces the base-contact
impedance at high frequencies. Thus, this modeling work was crucial in explaining the
discrepancy between the higher fmax values observed experimentally and those predicted
from the classical expression based upon the DC base-contact resistance.
Despite the success of the equivalent circuit, it contains certain assumptions. This
method assumes the common features of the test structures and the actual transistor are
nominally identical. Accurate values of some geometrical and material parameters are
required to determine and/or confirm some of the parasitic elements. These aspects may be
regarded as a limitation of this method. This technique also assumes that all the parasitic
elements, including the base semiconductor spreading resistance (Rbb) and the base-contact
capacitance (Cb con), are bias independent. This assumption is valid up to moderately
high collector current densities. As the collector current is increased, the base current also
increases. At sufficiently high base currents, the voltage drop across the base-contact
metal/base-semiconductor depletion region is no longer an insignificant fraction of the
Schottky built-in voltage so that the depletion region width and the corresponding
capacitance are no longer bias-independent constants. The related current-crowding effect
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
Conclusions and Suggestions for Future Research
141
at high base currents also affects Rbb . In addition, the base "push-out" effect influences
Rbb at high collector currents. In practice, it was found that our equivalent circuit, with the
assumption of bias-independent parasitic elements, no loner accurately described the sparameters of the transistor at collector current densities > 5 x 104 A/cm2.
This dissertation has provided the first in-depth study of the low-frequency noise
characteristics of Npn AlGaAs/GaAs HBTs. The dominant sources of low-frequency
noise in AlGaAs/GaAs HBTs were identified. The 1/f noise was mainly associated with
the extrinsic base surface. The role of an AlGaAs/GaAs surface passivation ledge in
improving the 1/f noise of AlGaAs/GaAs HBTs was demonstrated. The burst noise was
attributed to an AlGaAs trap, possibly the DX center, near the surface of the emitter-base
junction. It was shown that a reduction of the Al mole fraction from x = 0.3 to x = 0.2
virtually eliminated the burst noise.
Based on these results, the important issues for
further improvement of the low-frequency noise performance of III-V HBTs were
examined.
Several questions about the low-frequency noise of AlGaAs/GaAs HBTs remain.
Although the AlGaAs passivation ledge reduces the 1/f surface noise, the effect of the
extrinsic base surface was not completely eliminated. The origins of residual 1/f noise
(once the extrinsic base surface is passivated) need to be identified. The schemes for
reducing the 1/f surface noise proposed in section 6 .6 (and outlined in the next section)
should be attempted. In addition, the exact nature of the burst-noise "bump" is unknown.
Additional work needs to be done to sort out the precise role of the DX center in producing
this noise "bump".
7.2
Suggestions for Future Research
Based upon the results of this thesis study, several opportunities for further
investigation are worthwhile.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
142
Conclusions and Suggestions for Future Research
High-Frequency Response Related Topics:
•
The emitter contact resistance is one of the most important parasitics limiting the
microwave performance of our AlGaAs/GaAs HBTs. We speculate that the high
emitter contact resistance results from damage to the InGaAs surface during the
sputtering of Tungsten. Modification of the fabrication sequence to eliminate the
sputtering o f W while retaining the self-alignment of the base contact to the emitter
mesa is critical for realizing faster devices.
•
The use of carbon as the base dopant should be investigated to overcome the diffusion
problems of Be (see chapter 2) at high concentrations, as needed to achieve very high
values of fmax. While the diffusion of carbon at high concentrations is considerably
less than that of Be, issues regarding the activation of carbon, the quality of carbondoped base layers, and the reliability of HBTs incorporating carbon-doped bases needs
to be addressed.
•
HBTs
based
in
other
m aterial
system s
should
be
explored.
A10.48In 0.52A s/ In 0.53G a0.47As t7-1] and InP/!n0.53G a 0.47As t7 ' 2] HBTs
lattice-matched to semi-insulating InP substrates show great promise as high-frequency
devices. An InGaAs p-type base yields a lower contact resistance, higher electron
mobility, and higher velocity overshoot than GaAs. The degree of improvement over
AlGaAs/GaAs should be studied. The strained Si/SiGe HBT with its potential
compatibility with the vastly more mature Si bipolar transistor technology is also an
excellent candidate for high-frequency operation.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
•
Conclusions and Suggestions for Future Research
143
The equivalent-circuit extraction technique should be extended to model device
operation at very high collector cunrent densities.
Low-Frequency Noise Related Topics:
•
Identification of the residual 1/f noise sources in surface passivated AlGaAs/GaAs
HBTs should be investigated. Fluctuations in the surface recombination velocity at the
extrinsic base surface and the emitter-base junction surface, fluctuations in the interface
recombination velocity at the hetero-interface, and carrier velocity and mobility
fluctuations due to scattering in the bulk of the neutral base and emitter-base junction
(quantum 1/f noise [7-3]) are possible sources of the residual 1/f noise.
•
The role of a thinner base, a quasi-electric field in the base, and an abrupt
heterojunction in improving the 1/f extrinsic base surface noise should be tested
experimentally (see section 6 .6 ).
•
The correlation between base layers composed of materials with lower surface
recombination velocities than GaAs ( s = 106 cm/s) and lower 1/f bipolar transistor
noise should be studied. The 1/f noise of bipolar transistors with InGaAs ( s = 103
cm/s), Si
( thermally oxidized surface,
s = 10 cm/s), and SiGe (s =
unknown) bases should be compared (see section 6 .6 ). In particular, the surface
recombination velocity (s) of SiGe is of considerable interest. Since SiGe base layers
are metastable and can not be exposed to the elevated temperatures required for
standard thermal oxidation, the quality of the passivation oxide on top of the SiGe base
surface is inferior in quality compared to a standard thermal oxide on Si. As a result,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
144
Conclusions and Suggestions for Future Research
one might expect that the surface recombination velocity of a low-temperature oxide
passivated SiGe surface would be higher than that of a thermally oxidized Si surface.
• Since the interface recombination velocity is proportional to the relative lattice
mismatch, the effect of misfit dislocations and strain relaxation on the 1/f noise of
bipolar transistors incorporating strained layers, such as in A^Gaj^As/InyGa^yAs and
Si/Sii.xGex HBTs, should be considered.
• The role of the DX center in producing the burst-noise "bump"
observed at
intermediate frequencies of the noise spectra of AlGaAs/GaAs HBTs needs to be
clarified.
DLTS measurements should be performed on large area AlGaAs/GaAs
HBTs to confirm the presence of the DX center. Low-frequency noise measurements
of HBTs with n-type emitters composed of materials known to be free of DX centers,
such as InP and possibly InAlAs, should be done.
Low-frequency noise
measurements of Pnp AlGaAs/GaAs HBTs should also shed light on the role of the DX
center, since DX centers do not exist in p-type AlGaAs (see section 6 .6 ).
•
The 1/f noise of inverted collector-up/emitter down HBTs should be studied to verify
whether burying the base-emitter junction away from surfaces results in lower 1/f
noise (see section 6 .6 ).
•
As shown in figure 6 .10, the high recombination velocity of the base-contact metal in
the self-aligned base HBT ( AlGaAs ledge width of 0.2 um) degrades the 1/f noise
performance. The performance tradeoff in speed (requiring a self-aligned base HBT
versus 1/f noise (requiring an AlGaAs ledge width of > 1 um) should be explored. Ft
and fmax and the 1/f noise of a high-frequency Alo^Gao.sAs/GaAs HBT with the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
Conclusions and Suggestions for Future Research
145
conventional electrode topology should be measured as a function of AlGaAs ledge
width. As alternatives to this conventional single emitter contact (E), dual base contact
(B), and dual collector contact (C) topology (CBEBC), other doubly self-aligned
structures with 2 emitter contacts (half the length of the conventional structure) and one
or two base and collector contacts (CEBEC and BECEB topologies) have been
proposed to investigate the tradeoff between base resistance and base-collector
junction capacitance [7-4],
The 1/f noise behavior of these doubly self­
aligned structures should be measured to confirm the role of the base-contact metal
geometry.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix A
High-Frequency HBT Fabrication Sequence
146
Appendix A
High-Frequency HBT Fabrication Sequence
This appendix describes the step by step fabrication sequence of the high-frequency
AlGaAs/GaAs HBTs used in this study. Note all metallizations are immediately followed
by liftoff.
1.
Wafer Degrease
TCA, acetone, methanol, IPA, blow dry
IM PLANT ISOLATION MASK
2.
Deposit oxide (7000 A)
SiH4/He flow (200 seem), dial (150)
N 20 flow (350 seem), dial (512)
He dial(158)
base pressure after loading (240-280)
auto @ 900mtorr
25 W, 17 min
3.
Photolithography (ISOL)
use primer, 30s, 3.5krpm
AZ 4330 30s, 3.5 krpm
bake 85C, 20 min
chlorobenzene 15 min
expose 19 sec
bake 85C, 5 min
4.
Isolation Metal
evaporate Al, 1.5 um
5.
RIE oxide
C2F6 (Freon 116), etch rate 270 - 500A/min
C2F6 flow (20 seem), dial (928)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix A
High-Frequency HBT Fabrication Sequence
base pressure before loading (18mtorr)
auto @ 4 0 , etch for 32 min
ISOLA TIO N IM PLANATION
6.
H+ dose = 2el5/cm2, 200 KeV
followed by dose = lel5/cm 2, 70 KeV
7.
Remove Al/oxide mask
etch alignment mark H3P04:H202:H20=3:1:50@ 20C, lmin
Etch Al
Phosphoric:H20=l:l @ 70 C, Stop etching when Al clears ( ~ 8 min)
Etch oxide
BOE 6:1 (etch rate 2000A/min), 8 min
8.
Sputter W
8-10 min
EM ITTER OHM IC METAL
9.
Photolithography (EMIT)
AZ 1370 30s, 3.5 krpm
bake 85C, 20 min
chlorobenzene 10 min
expose 10.5 sec
bake 85C, 5 min
10.
Emitter Metallization
Ti/Al =700/1000 A
11.
RIEW
C2F6 flow (17 seem), dial (804)
SF6 flow (3 seem), dial (288)
etch rate approx 1300A/min
base pressure (20 mtorr)
250W, etch time lmin 45 sec
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix A
High-Frequency HBT Fabrication Sequence
EMITTER MESA DEFINITION AND FORMATION OF AlGaAs LEDGE
12.
Etch ~ 700A before base
H3P04:H2O2:H20=3:1:50 @ 20C, 1 min 15 sec
NH40H:peroxide =1:200 @ 10C, 40 sec
13.
Deposit Nitride, 1000 A
N2 dial (241)
Si/N2 flow (395 seem), dial(288)
NH3 flow (0.7 seem), dial (32)
auto @ 650 mtorr
25 W, 11 min
14.
RIE Nitride
C2F6 flow (20 seem), dial (928)
etch rate, 300A/min
base pressure (15.7 mtorr)____
auto @40 mtonr
200 W, etch for 3 min 45 sec
ETCH DOWN TO BASE
H3P04:H2O2:H20=3:1:50 @ 20C, roughly 1 min
15.
Photolithography (BASE)
Same as with EMIT photolithography
16.
Base Metallization
Ti/Pt/Au = 280/280/840 A
DEFINITION OF COLLECTOR
17.
Photolithography (COLL)
Same as with EMIT photolithography
Oxygen plasma
Pressure =100 mTorr, 0 2 10 seem
Power = 100 W, 15 sec
18.
Etch to Collector
H3P04:H2O2:H20=3:1:50 @ 20C, 7-8 min
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix A
High-Frequency HBT Fabrication Sequence
COLLECTOR OHMIC METAL
19.
Collector Metallization
Au/Ge/Ni/Au = 280/70/88/840 A
20.
RTA
440 °C for 15 sec
POLYIMIDE SPIN-ON
21.
spin on promoter, QZ3289:QZ 3290 = 1:9
5 krpm for 20 sec
spin on polyimide 285
4 krpm for 1 min
22.
bake polyimde
85 °C for 30 min
70 °C for 15 min
240 °C for 15 min
CONTACT HOLE DEFINITION
2 3.
Photolithography (CONT)
AZ 1370 30s, 5 krpm
bake 85C, 20 min
expose 6.5 to 7 sec
24.
RIE polyimide
0 2 plasma, 50 seem
auto @ lOOmtorr
100 W, 12 min
25.
strip resist mask
acetone 1 min, methanol 1 min, D I 1 min
INTERCONNECTS
26.
Photolithography (INTE)
same as with EMIT photolithography
27.
Interconnect Metallization
Ti/Au = 400/5000 A
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix B
Derivation of Lumped Circuit Model for Base Contact
150
Appendix B
Derivation of Lumped Circuit Model for Base Contact
This appendix derives the lumped model for the base contact used in the HBT
small-signal equivalent circuit in Chapter 5. Figure B .l shows a schematic cross section
of a contact and the corresponding transmission line model. The contact impedance arises
CONTACT METAL
INTERFACE LAYER
/
SEMICONDUCTOR LAYER
rc
Ps
Figure B .l Schematic cross section of a contact and the corresponding transmission line
model.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix B
151
Derivation of Lumped Circuit Model for Base Contact
from three parts: an interfacial conductance (G), a metal-semiconductor capacitance (C),
and a lateral semiconductor resistance (R).
G, C, and R are per unit length. The
parameters of the transmission line in terms of material parameters are as follows:
R = £-
(B .l)
G=^
Tc
(B.2)
Lb
where ps is the base sheet resistance (in Q/sq), rc is the specific contact resistivity in (Qcm2), Lj, is the base-contact length, q is the electronic charge, es is the semiconductor
dielectric constant, NB is the base doping, <hb is the metal-semiconductor barrier height, K
is Boltzmann's constant, and T is the temperature.
The base-contact impedance (Z) is equivalent to the input impedance of the
transmission line. By conventional analysis, we obtain for the base-contact impedance
IG +jcoC
cot
"1
jthjRsb-^
Z(x=0) = Rsb
----- .¥
R-^- '■
Rs
(B.4)
where x is the lateral distance along the base-contact width, sb is the base-contact width,
and co is the angular frequency.
Applying the properties of complex numbers [B-l, B-2], we can rewrite equation
(B.4) as
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix B
Z - RSb
Derivation of Lumped Circuit Model for Base Contact
coth(A+jB)
(A +jB )
152
(B.5)
where
_ Rsb J
n
g , . / g 2+ c ¥
V R
V
R2
(B.6 )
and
p _ Rsb y,
G | . / G 2 + C2©2
(B.7)
We then use the series expansion of the coth function,
V (A + jB )211
coth(A + jB) =
n
n=0
(2n)!
■ ------
(B.8 )
y (A +jB
(2n+1)!
and make the second-order Taylor series approximation,
, | (A + jB ) 2
coth(A + jB) s
a + .2b
Substituting equations (B.6 ), (B.7), and (B.9) into equation (B.5), we have
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(B.9)
Appendix B
Z =^
2
Derivation of Lumped Circuit Model for Base Contact
+ ----- \ — - .
SbG + jcost£
153
(B.10)
Equation (B.10) amounts to replacing the distributed base-contact model with a lumped
representation, which consists of resistance sbR/2, in series with the parallel combination
of resistance, l/sbG, and capacitance, sbC. From equation (B.10) , we define the
equivalent-circuit elements in figure 5.1 as follows:
R b .c o n ^ ^ V
(B.l 1)
2 S tG
Cb,con=
2SbC
R b b ^ i l 5# - )
2A 2 /
(B.12)
+
bV P sr1
12
Le
<B 1 3 >
where se is the active emitter width and Le is the active emitter length. The additional
factors of 2 in equations (B .ll), (B.12), and (B.13) reflect the fact that there are two base
contacts.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
154
Appendix C
Definition of Equivalent Input Base Noise Current Spectral
Density
This appendix derives the equivalent input base noise current spectral density. As
shown in figure C .l, any noisy two-port network can be modeled by the same noiseless
network (all internal noise sources removed) and equivalent noise voltage (En) and noise
current (In) generators connected at the input [C-l].
■n
0
- o
Noise-free
Network
o
Figure C. 1 Representation of device noise sources by a noiseless two-port network and
equivalent input voltage and current generators.
A further simplification is to represent both noise sources with an equivalent input noise,
Enj [C-2]. The equivalent input noise power, Pnj, is defined as the power that a 50 Q
noise source, En;, connected at the input of the noise-free device would have to be set to in
order to produce the same power at the output that the device noise sources produce (see
figure C-2). The equivalent input (base) noise current,
is defined as the current driven
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
155
by Ejy into the noiseless device.
Ini is determined by first measuring the transducer (power) gain of the system.
The gain of the very low-noise post amplifier (Zjn = 50 Q) is subtracted to yield the
_ni
O
Figure C.2 Simplified noise model and definition of equivalent input (base) noise current,
I m-
transducer gain of the device, G j. From the simple low-frequency, equivalent-circuit
m
50 Q
Figure C.3 Low-frequency, equivalent-circuit model of HBT.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
156
model of the transistor shown in figure C-3, we derive the following expression for the
transducer gain
qt _
Power delivered to load _ *o (50Q) _ f e n
Power available from source
|Vsj2
’ Rin I
(C D
so n
and
Rin = 50 + Rb + rrt +Re (Po +1)
(C.2 )
where i0 is the output current, vs is the source voltage, P0 ( = gmr^) is the incremental
common-emitter current gain, Rb is the base resistance, rn is the dynamic emitter-base
diode resistance referred to the base, and RE is the emitter series resistance, and Rin is the
input resistance. Since both Gx and p 0 are known from measurements, Rin can easily be
calculated from equation (C.l).
With the signal sourceoff, the output noise power, Pno, is measured. Pn; is
computed by dividing by the transducer gain
P n i-J K
(jx
(C-3)
The equivalent input (base) noise current is then simply calculated as
r .2 _ Enj2 _ 200 Pnj
01
R2
^•in
R2
^ in
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(C 4 )
157
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
where the definition of equivalent input noise power
has been used.
Figure C.4 shows a general, low-frequency equivalent-circuit model of the
transistor with noise sources. As proposed by Fonger, an alternative representation which
. 2
I .
. 2
m
r%
no
1 b ’c
O 1,1I
>-
W--------- - -------------
oG
m
V
k
50
f 501*
e
-o—
v:
en
i
R,
Figure C.4 General, low-frequency HBT equivalent-circuit model with noise sources.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
158
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
better facilitates the locating of noise sources is shown in figure C.5. In figure C.5, the
2
i b'c noise source has been split into two generators: one in the b'e position, which has
been absorbed into the existing i\< e, and one in the ce position. The noise sources in
2
figure C.5 are defined as follows: v bn (= 4KTRb) is the thermal noise from the base
no
b ’e
en
Figure C.5 Alternate, low-frequency HBT equivalent-circuit model with noise sources.
2
resistance, i b'e is the superposition of the low-frequency noise and shot noise sources
2
.
.
.
associated with base current, v en (= 4KTRg) is the thermal noise from the emitter series
2
resistance, and i ce is the superposition of the low-frequency noise source and shot noise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix C Definition of Equivalent Input Base Noise Current Spectral Density
159
sources associated with the collector current. Because of the emitter-follower action of the
transistor, at low frequencies RE is reflected back to the base and contributes an effective
resistance of RE( P0 + 1) to the input resistance of the transistor. This effect is taken into
account in the simplified circuit in figure C.6 .
From figure C.6 , the equivalent input base noise current can be expressed in terms
of the device noise sources. First, the output noise current is calculated as the
Figure C .6 Simplified, low-frequency HBT equivalent-circuit model with noise sources
for calculating the equivalent input base noise current spectral density.
superposition of the contributions from the various noise sources.
After some
manipulation, we find that the output noise current can be expressed as:
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix C Definition o f Equivalent Input Base Noise Current Spectral Density
i2 _ ;2 , /-i2..2 . ;2 , q2 Rj ;2. , r 2 (ven>+ V^n)
Ino - Ice + G mv^ = Ice + Po ~ r b'e + P o
R2in
~--------
160
(n c.\
((~.t>)
R2in
2
where V en- = 4KTRE(P0 +1), Rj = 50 + Rb + Re(P0 +1) and the thermal noise from the
5012 resistances is omitted since this is taken into account during the system calibration.
Combining equations (C.4), (C.5), and (C.l), the equivalent input base noise current can
be written in the form:
& = % “ % + &>+ ^
Po
Po
(C.7)
4m
under the condition Rin ~ Rj
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
References
161
C hapter 1 References
[1-1]
M. E. Kim, A. K. Oki, G. M. Gorman, D. K. Umemoto, and J. B. Camou,
"GaAs heterojunction bipolar transistor device and IC technology for highperformance analog and microwave application", IEEE Trans. Microwave Theory
and Tech., vol. 37, pp. 1286-1303, 1989.
[1-2]
T. Ishibashi, and Y. Yamauchi, "A possible near-ballistic collection in an
AlGaAs/GaAs HBT with a modified collector structure," IEEE Trans. Electron
Devices, vol. 35, pp. 401-404, 1988.
[1-3]
P. M. Asbeck, M. F. Chang, K. C. Wang, D. L. Miller, G. J. Sullivan, N. H.
Sheng, E. Sovero, and J. A. Higgins, "Heterojunction bipolar transistors for
microwave and millimeter-wave integrated circuits," IEEE Trans. Electron
Devices, vol. 34, pp. 2571-2579, 1987.
[1-4]
S. M. Sze, Semiconductor Devices: Physics and Technology, New York: John
Wiley & Sons, 1985.
[1-5]
H. Kroemer, "Heterostructure bipolar transistors and integrated circuits," IEEE
Proc., vol. 70, pp. 13-25, 1982.
[1-6]
H. H. Lin and S. C. Lee, "Super-gain AlGaAs/GaAs heterojunction bipolar
transistors using an emitter edge-thinning design", Appl. Phys. Lett., vol. 47, pp.
839-841 (1985).
[1-7]
W. S. Lee, D. Ueda, T. Ma, Y.C. Pao, and J. S. Harris, Jr., "Effect of emitterbase spacing on the current gain of AlGaAs/GaAs heterojunction bipolar
transistors", IEEE Electron Device Lett., vol. 10, pp. 200-202, 1989.
[1-8]
D. Costa, W. Liu, and J. S. Harris, J r .," Reduction o f Low-Frequency Noise in
Npn AlGaAs/GaAs HBTs", Device Research Conference Digest, June 1991.
[1-9]
E. O. Johnson, "Physical Limitations on Frequency and Power Parameters of
Transistors", RCA Review, pp. 163 - 177, 1965.
[1-10] M. E. Hafizi, L. M. Pawlowicz, L. T. Tran, D. K. Umemoto, D. C. Streit, A. K.
Oki, M. E. Kim, and K. H. Yen, "Reliability Analysis of GaAs/AlGaAs HBTs
under Forward Current/Temperature Stress", GaAs IC Symposium Digest, p.
329, 1990.
[1-11] D. Costa, W. Liu, and J. S. Harris, Jr., "Direct Extraction of the AlGaAs/GaAs
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission
References
162
Heterojunction Bipolar Transistor Small-Signal Equivalent-Circuit", IEEE Trans.
Electron Dev., Sept. 1991.
[1-12] C.Y. Chen, J. Bayruns, R.J. Bayruns, and N. Scheinberg, "Reduction of Lowfrequency Noise in a DC-2.5 GHz GaAs Amplifier, ", GaAs IC Symposium
Digest, p. 289, 1988.
[1-13] H.J. Siweris and B. Schiek, "Analysis of Noise Upconversion in Microwave FET
oscillators", IEEE Trans. Microwave Theory Tech., vol. 33, p.233, 1985.
[1-14] D.V. Lang, R.A. Logan, and M. Jaros, "Trapping characteristics and a donorcomplex (DX) model for the persistent-photoconductivity trapping center in Tedoped AlxGai_xAs", Phys. Rev. B, vol. 19, No. 2, p. 1015, 1979.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
163
C hapter 2 References
[2-1]
E. H. C. Parker, The Technology and Physics o f Molecular Beam Epitaxy, New
York: Plenum Press, 1985.
[2-2]
J. M. Woodall, J. L. Freeouf, G.D. Pettit, T. Jackson, and P. Kirchner, "Ohmic
contacts to n-GaAs using graded band gap layers of G aj.xInxAs grown by
molecular beam epitaxy", J. Vac. Sci. Tech., vol. 19, p. 626, 1981.
[2-3]
N. Chand, P. R. Berger, and N. K. Dutta, "Effects of Substrate Tilting in
Substantial Improvement of DC Performance of AlGaAs/GaAs NpN DHBTs
grown by MBE", Device Research Conference Digest, p. VIA-1, June 1991.
[2-4]
K. Yamanaka, S. Naritsuka, K. Kanamoto, M. Mihara, and M. Ishii, "Electron
traps in AlGaAs grown by molecular-beam-epitaxy", J. Appl. Phys., vol. 61, No.
11, p. 5062, 1987.
[2-5]
K. Akimoto, M. Kamada, K. Taira, and N. Watanabe, "Photoluminescence killer
in AlGaAs grown by molecular beam epitaxy", J. Appl. Phys., vol. 59, p. 2833,
1986.
[2-6]
Y.C. Pao, T. Hieri, and T. Cooper, "Surface effect-induced fast Be diffusion in
heavily doped GaAs grown by molecular beam epitaxy" J. Appl. Phys., vol. 60,
pp. 201-204, 1986.
[2-7]
J. M. Miller, D. M. Collins, and N. J. Moll, "Control of Be diffusion in molecular
beam epitaxy G aA s", Appl. Phys. Lett., vol, 46, no. 10, p. 960, May 1985.
[2-8]
K. Nakagawa, N. J. Kawai, and H. Ohta, "Design Principles for CHIRP
Superlattice Devices", Superlattices and Microstructures, vol. 1, pp. 187-192,
1985.
[2-9]
J. G. Ruch, IEEE Trans. Electron Dev., vol. 19, p. 5, 1972.
[2-10] H. G. Casey, Jr. and M. B Panish, Heterostructure Lasers, Academic, New York,
1978.
[2-11] N. Lifshitz, A. Jayaraman, and R. A. Logan, " Pressure and compositional
dependences of the Hall coeficient in Alx G a ^ A s and their significance", Phys.
Rev. B, vol 21, No. 2, p.670, 1980.
[2-12] N. Chand, T. Henderson, J. Clem, W. T. Masselink, R. Fischer, Y. C. Chang,
and H. Morkoc, "A Comprehensive Analysis o f Si doped Alx Gaj.xAs (x = 0 to 1):
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
164
Theory and Experiments", Phys. Rev. B, vol 30, p. 4481, 1984.
[ 2-13] R.J. Nelson, "Long-lifetime photoconductivity effect in n-type GaAlAs", APL,
vol. 31, 5, Sept. 1977.
[2-14] Same as reference [1-14].
[2-15] M. Mizuta, M. Tachikawa, H. Kukimoto, and S. Minomura, "Direct Evidence for
the DX center being a Substitutional Donor in AlGaAs Alloy System", Jpn. J.
Appl. Phys., Vol. 24, 2, pp. L143-L146, Feb. 1985.
[2-16] A.I. Valois, G.Y. Robinson, K. Lee, and M.S. Shur, "Temperature dependence of
the I-V characteristics of modulation doped FETs", J. Vac. Science Technol.
B l(2), pp. 190-195, Apr.-June 1983.
[2-17] P.M. Mooney, P.M. Solomon, and T.N. Theis, "Charge Trapping in
GaAs/AlGaAs Modulation-Doped FETs", International Symposium on GaAs and
Related Compounds, Sept. 1984.
[2-18] Alex Katalsky and Richard A. Kiehl, "On the Low-Temperature Degradation of
(AlGa)As/GaAs Modulation-Doped Field-Effect Transistors", IEEE Trans.
Electron Dev., vol. 33, pp. 414-423, March 1986.
[2-19] S. Tiwari, S.L. Wright, and A. W. Kleinhausser, "Transport and Related
Properties of (Ga,Al)As/GaAs Double Heterojunction Bipolar Junction
Transistors", IEEE Trans. Electron Dev., vol. 34, pp. 185-197, Feb. 1987.
[2-20] I. Izpura, E. Munoz, and E. Calleja, Inst. Conf. on the Science and Technology of
Defect Control in Semiconductors, Yokohama, Japan, Sept. 1989.
[2-21] M.I. Nathan, S. Tiwari, P.M. Mooney, and S.L Wright, "DX centers in AlGaAs
p-n heterojunctions and heterojunction bipolar tmsistors", J. Appl. Phys, Vol. 62,
8 , pp. 3234-3236, Oct. 1987.
[2-22] D.J. Chadi and K.J. Chang, "Energetics of DX-center formation in GaAs and
AlxGa!_xAs Alloys", Phys. Rev. B, vol. 39, 14, pp. 10063-10074, May 1989.
[2-23] P. M. Mooney, N. S. Caswell, P. M. Solomon, and S. L. Wright, "Charge
Trapping by Deep Donors in Si-doped AlxGai_xAs", Materials Research Society
Meeting, San Francisco, April 1985.
[2-24] E. Calleja, A. Gomez, and E. Munoz, Appl. Phys. Lett., 52, p. 383, 1988.
[2-25] T. N. Theis, "The DX center in GaAs and A ^ G a ^ A s " , Int. Conf. Shallow
Impurities in Semiconductors, p. 307, 1988.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
165
[2-26] P. M. Mooney, "Deep donor levels (DX centers) in III-V semiconductors", J.
Appl. Phys., vol. 67, No. 3, p. R l, 1990.
[2-27] M. Watanabe, Y. Ahizawa, N. Sugiyama, and T. Nakanisi, "Nonradiative
recombination process at deep levels in AlGaAs grown by MBE", Int. Symp.
GaAs and Related Compounds, p. 105,1986.
[2-28] I. Izpura.E. Munoz, and E. Calleja, Int. Conf. on the Science and Technology of
Defect Control in Semiconductors, Sept. 1989.
[2-29] I. D. Henning and H. Thomas, Solid State Electron., vol. 25, p. 325, 1982.
[2-30] E. Calleja, E. Munoz, and F. Garcia, Appl. Phys. Lett., vol. 42, p. 528, 1983.
[2-31] L. Konczewicz, E. Litwin-Stazewska, S. Porowski, A. Iller, R. L. Aulombard, J.
L. Robert, and A. Joullie, Physica, vol. 117B &118B, p. 92, 1983.
[2-32] Y. Takeda, Y. Zhu, and A. Sasaki, Inst. Phys. Conf. Ser., vol. 83, p. 203, 1987.
[2-33] K. Nakashima, S. Nijima, Y. Kawaura, and H. Asahi, Phys. Status Solidi, vol.
A 103, p. 511, 1987.
[2-34] W. P. Hong, S. Dhar, P. K. Bhattacharya, and A. Chin, J. Electron. Mater., vol.
16, p. 271, 1987.
[2-35] R. N. Nottenburg, Y. K. Chen, M. B. Panish, D. A. Humphrey, and R. Hamm,
"Hot-electron InGaAs/InP Heterostructure Bipolar Transistors with ft of 110
GHz", IEEE Electron Dev. Lett., vol. 10, No. 1, p. 30, Jan. 1989.
[2-36] U. K. Mishra, J. F. Jensen, D. B. Rensch, A. S. Brown, W. E. Stanchina, R. J.
Trew, M. W. Pierce, and T. V. Kargodorian, "Self-aligned AlInAs-GalnAs
Heterojunction Bipolar Transistors and Circuits", IEEE Electron Dev. Lett., vol.
10, No. 10, p. 467, Oct. 1989.
[2-37] M. Tachikawa, M. Mizuta, H. Kukimoto, and S. Minomura, "A Simple
Calculation of the DX Center Concentration Based on a L-Donor Model", Jap. J.
Appl. Phys., vol. 24, No. 10, pp. L821-L823, Oct. 1985.
[2-38] W. E. Spicer, P.W. Chye, P. R. Skeath, C. V. Su, and I. Lindau, "New and
unified model for Schottky barrier and III-V insulator interface states formation",
J. Vac. Sci. and Tech., vol. 16, pp. 1422-1433, 1979.
[2-39] C. T. Sah, R. N. Noyce, and W. Shockley, "Carrier generation and recombination
in p-njunctions and p-n junciton characteristics", Proc. IRE, vol. 45, pp. 1228-
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
166
1243, Sept. 1957.
[2-40] D. E. Aspnes, "Recombination at semiconductor surfaces and interfaces", Surface
Science, vol. 132, pp. 406-421, 1983.
[2-41] S. Tiwari, private communication, 1991.
[2-42] S. Tiwari and D. J. Frank, "Analysis of the Operation of GaAlAs/GaAs HBTs",
IEEE Trans. Electron Dev., vol. 36, p. 2105, Oct. 1989.
[2-43] W. S. Lee, Thin Barrier Type AlGaAs/GaAs Heterojunction Bipolar Trnsistor
Structure with Reduced Surface Effects, Ph.D Thesis, Stanford University, p. 90
May 1990.
[2-44] W. Fonger, " A determination of 1/f noise sources in semiconductor diodes and
transistors", in Transistors I, Princeton, NJ: RCA Labs, p. 239, 1956.
[2-45] Same as reference [1-6].
[2-46] R. W. Anderson, "S-parameter Techniques for Faster, More Accurate Network
Design", HP Application Note 95-1.
[2-47] N. Wiener, Acta Math., vol. 55, p. 117, 1930.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
167
C h ap ter 3 References
[3-1]
R. S. Muller and T. I. Kamins, Device Electronics for Integrated Circuits, New
York: John Wiley and Sons, p. 212, 1977.
[3-2]
S. Tiwari and S. L. W right, "Material properties of p-type GaAs at large dopings",
Appl. Phys. L e tt., vol. 56, no. 6 , p. 563, Feb. 5, 1990.
[3-3]
Same as reference [2-40].
[3-4]
C. H. Henry, R. A. Logan, and F. R. Merrit, "Effect of surface recombination
current in AlxG aj,xAs heterojunctions", J. Appl. Phys., vol. 49, no. 6 , pp. 35303542, June 1978.
[3-5]
Same as reference [2-45].
[3-6]
W. Liu, Microwave and DC Studies o f Npn and Pnp AlGaAs/GaAs
Heterojunction Bipolar Transistors, Ph.D Thesis, Stanford University, p. 122,
Feb. 1991.
[3-7]
Same as reference [2-42].
[3-8]
Same as reference [1-7].
[3-9]
Same as reference [1-6].
[3-10] Same as reference [2-43].
[3 -11] D. G. Hill, Pnp Heterojunction Bipolar Transistors in AlGaAs/InGaAs/GaAs,
Ph.D Thesis, Stanford University, p. 74, June 1990.
[3-12] W. Liu, p. 95.
[3-13] Same as reference [1-10].
[3-14] S. M. Sze, High-Speed Semiconductor Devices, New York: Wiley & Sons, p.
335, 1990.
[3-15] Muller and Kamins, p. 260.
[3-16] C. T. Kirk, Jr., " A Theory of Transistor Cutoff Frequency Fall Off at High
Current Densities", IRE Trans., pp. 164-174, March 1962.
[3-17] R. G. Meyer and R. S. Muller, "Charge-control Analysis of the Collector-Base
Space-charge Region Contribution to Bipolar Transistor Time Constant", IEEE
Trans. Electron. Dev., vol 34, no. 2, pp. 450-452, Feb. 1987.
[3-18] D. L. Miller, P. M. Asbeck, R. J. Anderson, and F. H. Eisen, "(GaAl)As/GaAs
Heterojunction Bipolar Transistor with Graded Composition in the Base", Electron
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
168
Lett., vol. 19, No. 10, pp. 367 - 368, 1983.
[3-19] C. M. Mazier, M.G. Klausmeier-Brown, Mark Lundstrom, "Proposed Structure
for Collector Transist Time Reduction in AlGaAs/GaAs Bipolar Transistor",
Electron Dev. Lett., vol. 7, pp. 483-485, Aug. 1986.
[3-20] Same as reference [1-7].
[3-21] O. Nakajima, K. Nagata, Y. Yamauchi, H. Ito, and T. Ishibashi, "High
Performance AlGaAs/GaAs HBTs Utilizing Proton-Implanted Buried Layers and
Highly Doped Base Layers", IEEE Trans. Electron. Dev., vol 34, pp. 2393 2397, Dec. 1987.
[3-22] H. Hayama, M. Madihian, A. Okamoto, H. Toyoshima, and K. Honjo, "Fully
Self-Aligned AlGaAs/GaAs Heterojunction Bipolar Transistors for High-Speed
Integrated-Circuits", IEEE Trans. Electron. Dev., vol 35, no. 1, pp. 1771-1777,
Nov. 1988.
[3-23] W. Liu, D. Costa, and J. S. Harris, Jr., "Novel Doubly Self-Aligned
AlGaAs/GaAs HBT", Electron Lett., vol. 26, pp. 1361-1362, Aug. 1990.
[3-24] Muller and Kamins, p. 262.
[3-25] D. LeCroisette, Transistors, Englewood Cliffs, NJ : Prentice Hall, Inc., 1963.
[3-26] P. Gray and C. Searle, Electronic Principles: Physics, Models, and Circuits, New
York: Wiley, pp. 373, 1969.
[3-27] R. L. Pritchard, Electrical Characteristics of Transistors, New York: McGrawHill, p. 498, 1969.
[3-28] Pritchard, p. 502.
[3-29] Pritchard, p. 393.
[3-30] M. B. Das, "High Frequency Limitations of Millimeter-wave Heterojunction
Bipolar Transistors, IEEE Trans. Electron. Dev., vol 35, pp. 604 - 614, May
1988.
[3-31] P. M. Asbeck, M. F. Chang, J. A. Higgins, N. H. Sheng, and G. J. Sullivan, and
K. C. Wang, "GaAlAs/GaAs Heterojunction Bipolar Transistors: Issues and
Prospects for Application", IEEE Trans. Electron. Dev., vol 36, pp. 2032-2041,
Oct. 1989.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
169
C hapter 4 References
[4-1]
D. Costa and J. S. Harris, Jr., "Low-Frequency Noise Characterization of Npn
AlGaAs/GaAs Heterojunction Bipolar Transistors", Int. Symp. GaAs and Related
Compounds, 1991.
[4-2]
W. S. Lee, Thin Barrier Type AlGaAs/GaAs Heterojunction Bipolar Transistor
Structure with Reduced Surface Effects, Ph.D Thesis, Stanford University, p. 32,
May 1990.
[4-3] A. B. Phillips, Transistor Engineering and Introduction to Integrated
Semiconductor Circuits, New York: McGraw-Hill, 1962, p. 216.
[4-4] K. C. Gupta, R. Gang, and I. S. Bahl, Microstrip lines and Slotlines, New York:
Artech, p.292.
[4-5] D. Ferry, GaAs Technology, Indianapolis: Howard W. Sams & Co., Inc., p.
211, 1985.
[4-6] M. J. W. Rodwell, Picosecond Electrical Wavefront Generation and Picosecond
Optoelectronic Instrumentation, Ph.D Thesis, Stanford University, p. 70, Dec.
1987.
[4-7] Gupta, Gang, and Bahl, p. 288.
[4-8] M. J. Howes and D. V. Morgan, Gallium Arsenide, New York: Wiley & Sons, p.
458, 1985.
[4-9] Cascade MicroTech Application Note, "Layout Rules for GHz Probing", p. 8 ,
1988.
[4-10] R. A. Logan and F. K. Reinhart, "Optical Waveguides in GaAs/AlGaAs Epitaxial
Layers", J. Appl. Phys., vol. 44, p. 4172, 1973.
[4-11] R. Anholt, "TRIM Calculations of Ion Implantation into GaAs", 1985.
[4-12] D. Avanzo, "Proton Isolation for GaAs Integrated Circuits", IEEE Trans. Electron
Dev., vol. 29, no. 7, p. 1053, July 1982.
[4-13] Rodwell, p. 80.
[4-14] Lee, p. 89.
[4-15] Same as reference [3-22].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
170
C hapter 5 References
[5-1]
Gilles Dambrine, Alain Cappy, Frederic Heliodore, and Edourad Playez, "A New
Method for Determining the FET Small-Equivalent Circuit," IEEE Trans.
Microwave Theory Tech., vol. 36, pp. 1151-1159, July 1988.
[5-2]
R J . Trew, U.K. Mishra, W.L. Pribble, and J.F. Jensen, "A Parameter Extraction
Technique for Heterojunction Bipolar Transistors", IEEE Microwave Theory
Tech. Digest, pp. 897-899, 1989.
[5-3]
P. Gray and C. Searle, Electronic Prinicples: Physics, Models, and Circuits, pp.
373-385, 1969.
[5-4]
R.L. Pritchard, J.B. Angell, R.B. Adler, J.M. Early, and W.M. Webster,
"Transistor Internal Parameters for Small-Signal Representation", Proc. IRE,
pp.725-738. 1961.
[5-5]
R.L. Pritchard, Electrical Characteristics of Transistors, p.213,1969.
[5-6]
P.J. van Wijnen, H.R. Claessen , and E.A. Wolsheimer, "A New Straighforward
Calibration and Correction Procedure for 'On Wafer' High Frequency S-parameter
Measurements (45 MHz - 18 GHz), IEEE Bipolar Circuits and Technology
Meeting Digest, pp. 70-73, 1987.
[5-7]
A. Fraser, R. Gleason, and E.W. Strid, "GHz On-Silicon-Wafer Probing
Calibration Methods", Cascade Microtech Application Note, 1988.
[5-8]
I.E. Getreu, Modeling the Bipolar Transistor, pp. 140-150,1978.
[5-9]
H.H. Berger, "Models for Contacts to Planar Devices," Solid-State Electronics,
15, pp. 145-157, June 1971.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
171
C h ap ter 6 References
[6 -1 ]
Same as reference [1-12].
[6-2]
Same as reference [1-13].
[6-3]
Same as reference [1-14].
[6-4]
S. C. Jue, D. J. Day, A. Margittai, and M. Svilans, "Transport and Noise in
GaAs/AlGaAs Heterojunction Bipolar Transistors - Part II: Noise and Gain at Low
Frequencies", IEEE Trans. Electron Devices, ED-36, p, 1020,1989.
[6-5]
N. Hayama, S. Tanaka, and K. Honjo, "1/f Noise Reduction for Microwave Self­
aligned AlGaAs/GaAs HBTs with AlGaAs Surface Passivation Layer", Third
Asia-Pacific Microwave Conference Proceedings, p. 1039, 1990.
[6 - 6 ]
Same as reference [2-44].
[6-7]
Same as reference [1-14]
[6 -8 ]
C. D. Motchenbacher and F. C. Fitchen, Low-Noise Electronic Design, John
Wiley
& Sons, p. 66 , 1973.
[6-9]
A. van der Ziel, X. Zhang,and A. H. Pawlikiewicz, "Location of 1/f Noise
Sources in BJT's and HBJT's - 1 Theory", IEEE Trans.Electron Devices, ED-33,
p. 1371, 1986.
[6-10] A. van der Ziel, "Theory of Shot Noise in Junction Diodes and Junction
Transistors", Proc. IRE, 43, p. 1639, 1955.
[6-11] R. D. Thorton, D. DeWitt, E.R. Chenette, and P.E. Gray, Characteristics and
Limitations of Transistors, p. 147.
[6-12] A. L. McWhorter, Semiconductor Surface Physics, Philadelphia, p.169,1957.
[6-13] O. Jantsch, "Flicker (1/f) Noise Generated by a Random Walk of Electrons in
Interfaces" JEEE Trans. Electron Devices, ED-34, p. 1100, 1987.
[6-14] A. van der Ziel, "Formulation of Surface 1/f Noise Processes in Bipolar Junction
Transistors and in p-n Diodes in Hooge-Type Form", Solid State Elect, vol. 32,
No. 1, p. 91, 1989.
[6-15] C. T. Sah and F. M. Heilscher, "Evidence of Surface Origin of the 1/f Noise",
Phys. Rev. Lett, vol. 17, p. 956, 1966.
[6-16] S. T. Hsu, "Surface State Related 1/f Noise in p-n Junctions", Solid State Elect,
vol. 13, p. 843, 1970.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
172
[6-17] H. Khajezadeh and T.T. McCaffery, "Materials and Process Considerations for
Monolithic Low-l/f-Noise Transistors", Proc. IEEE, vol.57, No.9, p.1518,
1969.
[6-18] H. K. Chung, M.A. Rosenberg, and P.H. Zimmerman, "Origin of 1/f nosie in
HgCdTe Photodiodes", J. Vac. Sci. Tech. A, vol. 3, No. 1, p. 189, 1985.
[6-19] W. A. Radford and C. E. Jones, "1/f noise in Ion-Implanted and Double-Layer
Epitaxial HgCdTe Photodiodes", J. Vac. Sci. Tech. A, vol. 3, No. 1, p. 183,
1985.
[6-20] S. Tiwari and D J. Frank, "Analysis of the Operation of GaAlAs/GaAs HBT's",
IEEE Trans. Electron Devices, ED-36, p. 2105, Oct. 1989.
[6-21] W. U. Liu, Microwave and DC Studies of Npn and Pnp AlGaAs/GaAs
Heterojunction Bipolar Transistors, Ph.D. Thesis, Stanford University, p. 76,
Feb. 1991.
[6-22] X. C. Zhu and A. van der Ziel, "Hooge parameters of n+-p-n and p+-n-p
transistors"/£’£’£’Trans. Electron Devices, ED-32, p. 658, 1985.
[6-23] S. R. Morrison, "Recombination of Electrons and Holes at Dislocations, " Phys.
Rev.,vol 104, No. 3, p. 619, 1956.
[6-24] D. Green and A. G. Jordan, "Effects of dislocations on the noise of planar pn
junctions", Int. J. Electronics, vol. 27, No. 2, p. 159, 1969.
[6-25] M. Mihaila, K. Amberiadis, and A. van der Ziel, "Low-Frequency Noise Due to
Emitter-Edge Dislocations in npn Transistors", Noise in Physical Systems and 1/f
Noise,p.219, 1983.
[6-26] S. T. Hsu, R.J. Whittier, and C.A. Mead, "Physical Model for Burst Noise in
Semiconductor Devices", Solid State Elect, vol. 13, p. 1055,1970.
[6-27] R. C. Jaeger and A.J. Broderson, "Low-Frequency Noise Sources in Bipolar
Junction Transistors", IEEE Trans. Electron Devices, ED-17, p. 128,1970.
[6-28] X. N. Zhang, A. van der Ziel, K.H. Duh, and H. Morkoc, "Burst and LowFrequency Generation-Recombination Noise in Double-Heterojunction Bipolar
Transistors'VFFF Electron Dev Lett, EDL-5, p. 2 1 1 ,1984.
[6-29] M. N. Tutt, D. Pavlidis, and B. Bayraktaroglu, "An assessment of noise sources
and charactersistics of AlGaAs/GaAs heterojunction bipolar transistors", Int.
Symp. GaAs and Related Compounds, p. 701,1989.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
173
References
[6-30] S. Tiwari, D. J. Frank, and S. L. Wright, "Surface Recombination in
GaAs/AlGaAs Heterostructure, /. Appl. Phys., vol. 64, No. 10, p.5009, 1988.
[6-31] W. S. Lee, Thin Barrier Type AlGaAs/GaAs Heterojunction Bipolar Transistor
Structure with Reduced Surface Effects, Ph.D. Thesis, Stanford University, p.
96, May 1990.
[6-32] Same as reference [1-6].
[6-33] G. Blasquez and D. Sauvage, "1/f Bulk Current Noise in Short Diodes and Bipolar
T ransistors", Noise in Physical Systems and 1/f Noise, Elsevier Science
Publishers B.V., p. 275, 1983.
[6-34] Y. S. Hiraoka, J. Yoshida, and M. Azuma, "Two-Dimesional Analysis ofEmittersize Effect on Current Gain of GaAlAs/GaAs HBT's", IEEE Trans. Electron
Devices, ED-34, pp.721-725, 1987.
[6-35] Y. S. Hiraoka, and J. Yoshida,
"Two-Dimesional Analysis of Surface
Recombination Effect on Current Gain for GaAlAs/GaAs HBT's", IEEE Trans.
Electron Devices, ED-35, pp. 857-862, 1988.
[6-36] Same as reference [2-4].
[6-37] L. Loreck, H. Dambkes, K. Heime, K. Ploog, and G. Weimann, " Deep-Level
Analysis in (AlGa)As-GaAs 2-D Electron Gas Devices by Means of LowFrequency Noise Measurements", IEEE Electron Dev Lett, EDL-5, p. 9,1984.
[6-38] F. Scholtz, J. M. Hwang, and D. K. Schroder, "Low Frequency Noise and DLTS
as Semiconductor Device Characterization Tools", Solid State Elect, vol. 31, No.
2, p. 205, 1988.
[6-39] B. Hughes, N. G. Fernandez, and J. M. Gladstone, "GaAs FETs with a FlickerNoise Comer Below 1 MHz", IEEE Trans. Electron Devices, ED-34, p. 733,
1987.
[6-40] Same as reference [2-23].
[6-41] Same as reference [2-26].
[6-42] E. Munoz, "DX Centers and III-V Device Performance", Mat. Res. Soc. Symp.
Proc.,Yol. 184, p. 49, 1990.
[6-43] S. Machlup, J. Appl. Phys., vol. 25, p. 341, 1954.
[6-44] J. A. Copeland, "Semiconductor Impurity Analysis from Low-Frequency Noise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
174
Spectra", IEEE Trans. Electron Devices, ED-18, pp. 50-53, Jan. 1971.
[6-45] S. T. Hsu, "Electron Trapping Noise in SOS Field-Effect Transistors Operating in
the Linear Region", RCA Review, vol. 38, pp. 226-237, June 1977.
[6-46] Same as reference [2-27].
[6-47] J. R. Kirtley, T. N. Theis, P. M. Mooney, and S. L. Wright, " Noise
Spectroscopy of Deep donor levels (DX) centers in GaAs-AlxGa|_xAs", J. Appl.
Phys., vol. 63, No. 5, p. 1541, Mar. 1988.
[6-48] O. Nakajima, K. Nagata, H. Ito, T. Ishibashi, and T. Sugeta, "Suppression of
Emitter Size Effect on Current Gain in AlGaAs/GaAs HBTs", J. Appl. Phys., vol.
24, No. 10, p. 1368, 1985.
[6-49] Same as reference [2-37].
[6-50] T. Yokoyama, M. Suzuki, T. Maeda, T. Ishikawa, T. Mimura, and M. Abe, "Sedoped AlGaAs/GaAs HEMTs for Stable Low-Temperature Operation", IEEE
Electron Dev Lett, EDL-11, p. 193, May 1990.
[6-51] D. A. Sunderland and P.D. Dapkus, "Optimizing N-p-n and P-n-p Heterojunction
Bipolar Transistors for Speed", IEEE Trans. Electron Devices, ED-34, Feb.
1987.
[6-52] Same as reference[l-5].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
References
Chapter 7 References
[7-1]
Same as reference [2-37].
[7-2]
Same as reference [2-35].
[7-3]
A. van der Ziel, Noise in Solid State Devices and Circuits, New York: Wiley &
Sons, p. 254, 1985.
[7-4]
Same as reference [3-23].
Appendix B References
[B-l] R. V. Churchhill and J. W. Brown, Complex Variables and Applications, p.21,
1984.
[B-2] J. E. Schutt-Aine, "Determination of a Small-Signal Model for Ion-Implanted
Microwave Transistors," IEEE Trans Devices, Vol. ED-30, No.7, pp. 750-758,
1983.
Appendix C References
[C-l] P. Gray and R. Meyer, Analysis and Design of Analog Integrated Circuits, John
Wiley &Sons, p. 657, 1984.
[C-2] Same as reference [6 - 8 ].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
175
Документ
Категория
Без категории
Просмотров
0
Размер файла
7 071 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа