Novel Measurement Systems for
Characterisation of Avalanche
Photo-Diodes
James Ernest Green
Department of Electronic and Electrical Engineering
University of Sheffield
A thesis submitted for the degree of
Doctor of Philosophy
August 2012
‘I am a Bear of Very Little Brain, and long words bother me.’
A. A. Milne
ii
Acknowledgements
I would have been unable to complete this work without the contin-
uous, unwavering support of my partner Pippa and of my parents.
I would like to thank my supervisors Richard C. Tozer and John P.
R. David. Thanks is also due to the Engineering and Physical Sci-
ences Research Council for funding the research, the EEE department
technical staff and the EEE departmental administrators.
Thanks to Stanislav I. Soloviev, Jody A. Fronheiser and Peter M.
Sandvik at The Semiconductor Technology Laboratory, part of Gen-
eral Electric Global Research who generously provided the Silicon
Carbide semiconductor devices reported in Chapter 7. Thank you
also for a considerable amount of your time, the abridged device fab-
rication method and of course dinner at the restaurant–on–the–bridge
in Nurnberg.
Further thanks to Prabhat Agarwal, Chritelle, Ervin, Fred, Micha,
Ray and Bob formerly at Philips Semiconductor. For the Design of
the Si wafers and devices reported in Chapter 6.
I would also like to thank my office colleagues and co-authors Daniel S.
G. Ong, Rajiv Gomez, Andrew Marshall, Wei Sun Loh, Yu Ling Goh,
Simon Dimler, Lionel J. J. Tan, Siew Li Tan, Peter Vines, Syahrin
Idris, Wai Mun Soong, and Kan Yeep Choo.
I owe considerable gratitude to M. Bibby, a wise and loyal friend.
Without the influence of G. F. Green, S. P. Woolley, N. A. Jaques, C.
J. Bijon, R. Lapslie, E. S. Hart, R. Hart, R. Green and S. Burton, I
would not have found myself in a suitable position to begin.
I would like to acknowledge the work of Dave Massy, Kim Foung
Li, Beng Koon Ng and Chee Hing Tan on which this work is based.
iii
Further thanks are due to Asst. Prof. Ng for generously providing
the implementation of the RPL model used in Chapter 6. I would
also like to thank Sammual Goldwasser – of Sam’s Laser Guide – for
enlightening discussions on laser pathologies.
I would like to thank Dave Stone, for providing a pleasant change of
topic.
iv
Abstract
This work is concerned with the measurement and interpretation of
avalanche noise in avalanche photo-diodes (APDs). Two new excess
noise measurement systems are described. These systems are ca-
pable of measuring photo-multiplication and excess noise in APDs
with junction capacitance up to two orders of magnitude greater
than the prior Sheffield system, and greater than any other reported.
The systems use phase sensitive recovery techniques, which allow the
avalanche noise power and photo-current to be measured unambigu-
ously from the system noise and leakage current of the test device.
The use of two independent measurement systems provides an excel-
lent method of confirming the validity of the data. The design and
optimisation of these systems and associated material is presented in
detail.
Avalanche noise measurements have been performed on a range of Sili-
con Carbide nip and separate absorption and multiplication avalanche
photo-diodes at 244 nm and 325 nm. This is the first report of pure
hole injection in SiC. The devices have i region widths, w, of 2.7 µm
and 0.57 µm. The value of α – the electron ionisation coefficient –
has been experimentally verified over a range of lower electric fields
and the extracted is significantly lower than prior reports.
A systematic experimental analysis of the effect of the dead-space
in Silicon has been undertaken using the measurement systems de-
scribed. These devices have widths, w, ranging from 0.35 µm to
31 nm. The Silicon results show that, for w < 0.4 µm, the extracted
and non-local modelled impact ionisation coefficients begin to deviate
from the bulk coefficients. In a 31 nm device at 1 MV/cm both α and
β are approximately an order of magnitude lower than the bulk value.
v
vi
Publications
Journal Papers
• James E. Green, Richard C. Tozer and John P. R. David, “A Transimpedance
Amplifier for Excess Noise Measurements of High Junction Capacitance
Avalanche Photodiodes”, Submitted Measurement Science and Technology.
• James E. Green, Richard C. Tozer, and John P. R. David, “Stability in
Small Signal Common Base Amplifiers”, Accepted IEEE Trans. Circuits
and Systems I: Regular Papers [Available] 10.1109/TCSI.2012.2209709.
• James E. Green, Richard C. Tozer and John P. R. David, “A Computation-
ally efficient circuit model of an Avalanche Photodiode”, In Prep for IEEE
Trans. Electron Devices
• James E. Green, Richard C. Tozer and John P. R. David, “An Opamp
Transimpedance Amplifier for Excess Noise Measurements of High Junction
Capacitance Avalanche Photodiodes”, In Prep. Measurement Science and
Technology
• James E. Green, Richard C. Tozer and John P. R. David, “Impact Ionization
and Dead Space in Silicon” Submitted IEEE Trans. Electron Devices.
• James E. Green, Wei Sun Loh, Andrew R. J. Marshall, Beng Koon Ng.
Richard C. Tozer, John P. R. David, Stanislav I. Soloviev and Peter M.
Sandvik, “Impact Ionization Coefficients in 4H-SiC by Ultra-Low Excess
Noise Measurement” IEEE Trans. Eletron Devices, Vol. 59, No. 4, pp.
1030 – 1036, April 2012 [Available] 10.1109/TED.2012.2185499
• S. L. Tan, W. M. Soong, J. E. Green, M. J. Steer, S. Y. Zhang, L. J. J. Tan,
Y. L. Goh, J. S. Ng, I. P. Marko, S. J. Sweeney, A. R. Adams, J. Allam, J.
vii
P. R. David, “Avalanche noise and breakdown characteristics in GaInNAs
diodes”, In Prep for Appl. Phys. Lett. or J. Appl. Phys.
Conference Papers
• James E. Green, W.S. Loh, J. P. R. David, R. C. Tozer, Stanislav I. Soloviev,
Peter M. Sandvik, “Characterisation of Low Noise 4H-SiC Avalanche Pho-
todiodes”, Materials Science Forum, 645-648 (2010) pp. 1081-1084, [Avail-
able] scientific.net
• M. J. Hobbs, S. D. Das, J. E. Green, J. P. David, C. H. Tan, S. Krishna, E.
Plis and J. Willmot, ”Uncooled GaSb/InAs type II superlattice photodiode
for radiation thermometry” in prep for IOP Photon12, Durham, UK
Book Chapters
• Daniel S. G. Ong and James E. Green (2011). “Avalanche Photodiodes in
High-Speed Receiver Systems”, in “Photodiodes - World Activities in 2011”,
Jeong-Woo Park (Ed.), ISBN: 978-953-307-530-3, InTech, 2011, [Available]
intechopen.com
viii
Contents
Acknowledgements iii
Abstract v
Publications vii
Contents ix
1 Introduction 1
1.1 A Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Photo-multiplier Tube . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Impact Ionisation and Photo-Diodes . . . . . . . . . . . . . . . . 6
1.4 Non-Local Multiplication and Noise . . . . . . . . . . . . . . . . . 14
1.5 Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Extraction of Ionisation Coefficients . . . . . . . . . . . . . . . . . 22
2 Measurement Techniques 25
2.1 Current Voltage Characteristics . . . . . . . . . . . . . . . . . . . 25
2.2 Capacitance Voltage Characteristics . . . . . . . . . . . . . . . . . 28
2.3 Photo-Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Review of Noise Measurement Systems . . . . . . . . . . . . . . . 29
3 Transistorised High Capacitance Noise Measurement System 39
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Front-End Design Methodologies . . . . . . . . . . . . . . . . . . 40
3.3 Diode Small Signal Model . . . . . . . . . . . . . . . . . . . . . . 43
ix
CONTENTS
3.4 Noise Measurement System . . . . . . . . . . . . . . . . . . . . . 44
3.5 Transimpedance Amplifier . . . . . . . . . . . . . . . . . . . . . . 45
3.6 Opamp stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7 Transistor Stage Biasing Feedback System . . . . . . . . . . . . . 49
3.8 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.9 Physical Construction . . . . . . . . . . . . . . . . . . . . . . . . 51
3.10 Noise Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.11 TIA Characterisation . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.12 Voltage Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.13 Noise Power Meter . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4 Small Signal Stability in Common Base Amplifiers 73
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Frequency Dependent Behaviour of a
Common Base Amplifier . . . . . . . . . . . . . . . . . . . . . . . 76
4.3 Addition of Parasitic Inductance in the Base Network . . . . . . . 77
4.4 The Addition of an Emitter Biasing Resistor . . . . . . . . . . . . 80
4.5 Addition of Emitter Capacitance . . . . . . . . . . . . . . . . . . 82
4.6 Addition of Load Capacitance and Collector - Base Junction Ca-
pacitance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.7 Colpitts Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Analysis of the Common Base Amplifier from an Oscillator’s Per-
spective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.9 The Effectiveness of Increasing the
Base Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.10 Practical Methods of Assessing Oscillation . . . . . . . . . . . . . 89
4.11 High Frequency Techniques . . . . . . . . . . . . . . . . . . . . . 90
4.12 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5 1 MHz Opamp Based High Capacitance Noise Measurement Sys-
tem 93
x
CONTENTS
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2 Noise Measurement System . . . . . . . . . . . . . . . . . . . . . 94
5.3 Transimpedance Front End . . . . . . . . . . . . . . . . . . . . . . 96
5.4 Bandwidth Setting Filter and ENBW . . . . . . . . . . . . . . . . 103
5.5 Voltage Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.6 Noise Power Meter . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6 Excess Avalanche Noise Thin Silicon APDs 109
6.1 Review of Impact Ionisation in Silicon . . . . . . . . . . . . . . . 109
6.2 Application: On-Chip Photonic Data
Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.3 Application: Medical/Biological Imaging . . . . . . . . . . . . . . 114
6.4 Chapter Objective . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.5 Layer Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.6 Current Voltage Characteristics . . . . . . . . . . . . . . . . . . . 116
6.7 Capacitance Voltage Characteristics . . . . . . . . . . . . . . . . . 117
6.8 Photo-multiplication . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.9 Excess Noise Measurement . . . . . . . . . . . . . . . . . . . . . . 124
6.10 Local Ionisation Coefficient Comparison . . . . . . . . . . . . . . 127
6.11 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7 Excess Noise Measurements and Impact Ionisation Coefficients
in 4H-SiC 129
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.2 Review of Impact Ionisation in Silicon Carbide . . . . . . . . . . . 135
7.3 Chapter Objective . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.4 Structure Details . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.5 Capacitance Voltage Characteristics . . . . . . . . . . . . . . . . . 140
7.6 Current Voltage Characteristics . . . . . . . . . . . . . . . . . . . 144
7.7 Photo-Multiplication Characteristics . . . . . . . . . . . . . . . . 148
7.8 Excess Noise Characteristics . . . . . . . . . . . . . . . . . . . . . 153
7.9 Ionisation Coefficient Extraction . . . . . . . . . . . . . . . . . . . 155
xi
CONTENTS
7.10 Comparison of GaN and SiC: GaN and AlGaN Detectors . . . . . 157
7.11 Comparison of GaN and SiC: GaN Power Electronics . . . . . . . 158
7.12 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8 An APD SPICE Model for Circuit Simulation 161
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.2 Characteristics of the Model . . . . . . . . . . . . . . . . . . . . . 163
8.3 Parameters of the Model . . . . . . . . . . . . . . . . . . . . . . . 164
8.4 Examples of the Model . . . . . . . . . . . . . . . . . . . . . . . . 169
8.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9 Future Work 177
9.1 Microwave Small Signal Bandwidth Measurements . . . . . . . . . 177
9.2 Microwave Noise Measurements . . . . . . . . . . . . . . . . . . . 182
9.3 Laser Noise “Cancelling” Front End . . . . . . . . . . . . . . . . . 188
9.4 Improved Squaring Circuit . . . . . . . . . . . . . . . . . . . . . . 190
9.5 An On-Wafer SPAD Setup . . . . . . . . . . . . . . . . . . . . . . 191
9.6 NLOS UV Communications with SiC . . . . . . . . . . . . . . . . 193
10 Conclusion 195
References 197
A A One Dimensional Poisson Solver 227
A.1 Two Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
A.2 Three Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
A.3 Four Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
A.4 Five Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
B APD SPICE Model Source Code 235
C Phase-Sensitive Detection 239
D High Order Nature of the Common Base Amplifier 243
xii
Chapter 1
Introduction
This thesis concerns two related topics; firstly, the development of electro-optical
measurement systems, and secondly, the characterisation of Silicon and Silicon
Carbide avalanche photo-diodes (APDs). This chapter provides a historical per-
spective before introducing the impact ionisation process; specifically, the local
modelling of multiplication and excess noise. The timeliness of this work and
the present state of knowledge of impact ionisation in Silicon and Silicon Carbide
are discussed in their respective chapters, where a selection of example industrial
uses of APD detectors are given. Approximately half of the thesis is given over
to measurement system design and half to device characterisation.
A panoply of electronic devices are presently manufactured and accurate im-
pact ionisation coefficients are required for the design and simulation of any elec-
tronic device which is expected to develop a large electric field within its structure.
Accurate impact ionisation coefficients are doubly critical in the analysis of high
speed devices developing high electric fields; the feedback nature of impact ioni-
sation, which will be discussed shortly, is intrinsically quite a ‘slow’ process when
compared to the speed of modern microwave devices and systems. It is criti-
cal that high speed device designers can design their devices such that impact
ionisation will be avoided if it is undesirable.
1
CHAPTER 1
1.1 A Historical Perspective
In the thirty years surrounding the turn of the twentieth century physicists were
struggling with problems related to the light spectrum emitted a black body. One
of the problems is sometimes called the “ultraviolet catastrophe”. The question
was posed in terms of a Helmholtz oscillator (of which a black body is an example).
The particular difficulty is with the equipartition theorem, which relates the
average energy of each oscillator mode to the temperature of the oscillator.
Classical electromagnetism, which considers energy exchange to happen con-
tinuously, can be used to compute the energy emitted by a three dimensional
black body at a given wavelength. The calculation gives u as the number of
degrees of freedom in frequency × the average energy per degree of freedom.
The average energy per degree of freedom is given by,
E =
∫∞
0
E exp (−E/k T ) dE.∫
∞
0
exp (−E/k T ) dE.
(1.1)
which reduces to k T .
The number of degrees of freedom is given by (8 π v2)/(c3). Combining these
gives the the Rayleigh-Jeans formula,
u (v, T ) =8 π v2
c3k T (1.2)
As the frequency, v increases the energy, E, increases without limit, which
is in disagreement with empirical observation. Max Planck provided a radical
change of viewpoint which resolved disagreement, he postulated the quantisation
of energy. Energy transfer is then only allowed for specific amounts of energy.
Kragh notes that Planck declared he turned to quantisation “in desperation”
having been unable to reconcile the classical approach with the observations [1].
The expression for the quantum of energy is,
ε = h v (1.3)
2
CHAPTER 1
where h is Planck’s constant and v is the frequency. The reduced Planck constant
~ = h / (2 π) is often used because it makes some expressions more elegant.
Energy, E, can only exist in integer multiples of the quantum,
E = nh v, where n = 0, 1, 2, . . . (1.4)
The quantised version of the black body expression is,
E =
∞∑
n=0
n ε exp (−n ε/k T )
∞∑
n=0
exp (−n ε/k T )
(1.5)
The ultraviolet catastrophe was not the only problem facing physics at the
beginning of the twentieth century, the observations of the interaction between
metal surfaces and light and the cathode rays that resulted - later named the
photoelectric effect - was difficult to explain using the continuous interpretation
of energy transfer. The photoelectric effect experiment requires a clean metal
surface (the cathode) which is exposed to light and yields electrons from its
surface. A second metal surface (the anode) which faces the cathode collects
electrons. Both electrodes are held in high vacuum. Light of different intensities
and wavelengths is shone onto the cathode. The cathode emits electrons (the
photoelectric current) which are collected on the anode.
Einstein’s 1921 Nobel prize was awarded “...especially for his discovery of the
law of the photoelectric effect”, which was published shortly before the, celebrated
special relativity paper of 1905. Einstein adopted a similar idea to Planck and
developed the idea that, in the photoelectric experiment, a photon (quantum of
electromagnetic energy) is absorbed by an electron which is bound to the surface
of the cathode metal. The energy imparted to the electron by the photon allows
the electron to escape the bound state and any extra energy that the photon
imparts is given to the electron as kinetic energy,
h v = Ek +W0 (1.6)
3
CHAPTER 1
where Ek is the free electron’s kinetic energy and W0 is the work function of the
metal. In the photoelectric effect experiment, a voltage may be applied between
the electrodes to cause the photoelectric current to cease. The magnitude of this
voltage must be,
e V = Ek (1.7)
An electron “cloud” may then form around the cathode. Equations 1.6 and 1.7
and may be solved for stopping voltage V ,
V =h
ev − W0
e(1.8)
If the stopping voltage is plotted as a function of frequency, a linear dependence
results where the constant of proportionality is h/e.
The photoelectric effect experiment shows that,
1. When light strikes a metal surface, a current flows immediately, irrespective
of the light intensity (assuming point 4 is satisfied).
2. For a particular frequency, the magnitude of the photoelectric current is
directly proportional to the intensity of the light.
3. The stopping potential V , and the maximum energy of the emitted elec-
trons, is a function of the frequency of the light and the work function of
the metal.
4. Each metal has a characteristic minimum frequency v0 such that
e V = h (v − vo)
5. The constant h is the same for all metals. It is the same constant used by
Planck in the black body derivation.
These experimental observations are easily explained if light is composed of
quanta. For example the second observation, is easy to explain from a quantum
paradigm. If the power of the light source incident on the surface is increased,
the number of photons incident on the cathode surface increases consequently
more electrons are released. In the classical approach, however, increasing the
intensity may be expected to increase the energy of each cathode ray as opposed
4
CHAPTER 1
to the number of cathode rays. The fourth observation is also dependent on a
quantum approach, in a classical approach cathode rays may be expected to leave
the cathode surface if light of any wavelength is incident on the cathode. The
optical energy incident on the surface may be integrated over time until enough
energy has been imparted to cause the emission of cathode rays. However this is
contrary to observation which shows that enough optical energy must be imparted
to the surface at one instant to free an electron.
1.2 Photo-multiplier Tube
Prior to the discovery of the useful optical detection capability of certain semi-
conductors, several types of thermionic valve were used to detect light including
the photomultiplier which fulfils similar application requirements to avalanche
photo-diodes. A schematic of a photomultiplier tube is shown in Fig. 1.1. It is an
evacuated glass envelope having an optically transparent faceplate behind which
is placed a photo-cathode.
The photo-cathode is a metal structure coated in an alkali metal for example
sodium-potassium-antimony or a III-V semiconductor such as Gallium Arsenide.
These materials are popular due to their low work function. Photons incident
on the photo-cathode cause the emission of electrons by the photo-electric effect.
An arrangement of focusing electrodes guide the free electrons towards a set of
dynodes. The dynodes have grids which attract the free electrons by electro-
static means. The dynodes are also metal (for example Nickel, Stainless Steel
or a Coper-Beryllium Alloy) structures which are coated in a secondary emissive
material such as Magnesium Oxide or Gallium Phosphide. A single dynode can
have a secondary emission ratio of approximately one hundred. Photo-multipliers
having up to nineteen dynodes have been commercially produced [2]. The sec-
ondary electrons travel from dynode to dynode, growing in number, until they
are collected in the last dynode behind the anode.
The key disadvantages of photo-multipliers are their bulk, and necessity for
high voltages. They are often said to be fragile, although several vidicon tubes
have visited the moon and returned pictures on more than one occasion. ‘Rugged’
photo-multipliers are also employed in oil exploration in down-hole applications.
5
CHAPTER 1
e−Directionof Light
Faceplate
Photocathode
︸ ︷︷ ︸
Dynodes Anode
Last DynodeSecondary ElectronFocusing Electrode
Vacuum ∼ 10−4 Pa
Pin
Figure 1.1 – Schematic of a Photo-multiplier tube. Redrawn from [2].
The cost of a commercial photo-multiplier and a commercial Silicon APD are
the same order of magnitude. Photo-multipliers require voltages between one
and two orders of magnitude greater than a Silicon APD when both devices are
biased to provide a similar gain.
1.3 Impact Ionisation and Photo-Diodes
In this section the generic three region device and separate absorption and mul-
tiplication layer avalanche photo diode (SAM-APD) detectors are discussed and
the impact ionisation process is qualitatively described with some simple exam-
ples. Various mathematical results relating to McIntyre’s local model [3] are also
derived.
1.3.1 Three Region Photo-Diode Detectors
A generic pin detector possessing an intrinsic region which is slightly n-type is
show schematically in Fig. 1.2 with an example electric field profile. Assuming the
device is homogeneous and uniform a one dimensional description of the device
does not represent a loss of generality. The device has been fabricated with ohmic
contacts and a precision controlled cell is applied in such a way that the device
is reverse biased. A window is etched into the top contact enabling light to enter
the p+ region.
6
CHAPTER 1
p+
n−
n+
|ξ|
x
Figure 1.2 – A generic pin / nip photo-diode with electric field shown to the right.
The reverse bias causes an electric field of greater magnitude than the diffusion
field to develop in the device. The characteristics of the field are a function
of the intrinsic region width and the doping densities in each region. At low
electric fields impact ionisation provides a negligible increase in total current.
Photons which are energetic with respect to the material band-gap may generate
carriers throughout the structure dependant on the absorption characteristics of
the material and the wavelength of the incident light. Assuming that the light is
entirely absorbed in the p+ region and all carriers are generated optically outside
of the depletion region, the minority electrons diffuse through the p+ material
until they recombine or reach the electric field. Once inside the electric field
region the electrons are swept through the device and induce a current in the
external circuit. The conditions necessary for impact ionisation are only that
the electric field be sufficient, given the material system under consideration,
that carriers drifting in an electric field can liberate further (secondary) carriers
by transferring some energy to those carriers, promoting them from the valence
band into the conduction band. Impact ionisation is shown schematically in
Fig. 1.3. The generation of the carrier is shown by the numbering. In this case
it is assumed that the primary carrier is an electron (blue) which has diffused
in to the high field region; the associate hole (red) is not shown. The primary
carrier ionises the lattice to generate a secondary e-h pair (2) the electron leaves
without taking further part. The secondary hole gains energy and ionises the
lattice producing a third generation (3) e-h pair. The second and third generation
holes leave the electric field region. The third generation electron ionises the
lattice creating a fourth generation e-h pair (4). The first three ionisation events
represent ‘carrier feedback’. This feedback is often undesirable as it causes the
7
CHAPTER 1
1
12
2
2
33
3
44
45
5
4
55
56
6
Figure 1.3 – Impact ionisation process schematically shown.
transient response of the device to increase with increasing multiplication. A
more preferable situation is one in which only electrons or holes are responsible
for impact ionisation. The fourth generation e-h pair ionises the lattice twice on
one transit of the high field region creating two fifth generation e-h pairs (5). The
hole created in the first of these fifth generation events goes on to ionise the lattice
(6) but none of the electrons from the fourth fifth or sixth generation carriers
ionise. The lack of carrier feedback results in a maximum device speed. In this
hypothetical device, in which carriers can be counted, the photo-multiplication is
seven. Seven electrons leave only one has entered. In this example the number of
holes leaving the high field region must be one less than the number of electrons,
because there is one primary carrier. If the primary carrier was a hole and the
colours of the carriers and the direction of the electric field reversed, holes would
be counted and the multiplication would be seven; one primary hole, and six
holes generated by impact ionisation. The situation described until now is that
8
CHAPTER 1
of pure injection, where all of the primary carriers are of one type and they
diffuse into the high field region. Pure injection is often desirable because it
yields the lowest possible excess noise, assuming the device is constructed such
that the more readily ionising carrier diffuses into the high field region. Use of
the less readily ionising carrier yields the largest possible excess noise. When
possible, pin and nip devices are often fabricated with identical region widths
and – as nearly as possible – doping densities such that the two devices will have
a similar electric field profile over a wide range of reverse bias. One device yields
pure electron initiated multiplication and excess noise, the other yields pure hole
initiated multiplication and excess noise. Possession of both pin and nip devices
allows cross-checking and confirmation of extracted results, including ionisation
coefficients.
Devices exhibit mixed injection when carrier pairs are photo-generated inside
the electric field region. Mixed injection may occur because the device is so
designed or by accident. Photons possessing energies only slightly larger than the
band-gap are weakly absorbed and tend to generate carriers far from the optical
window; photons that are energetic with respect to the band-gap are strongly
absorbed and tend to generate carriers close to the optical window. In the case of
mixed injection, shown in Fig. 1.4, the carrier pairs created in electron initiated
ionisation events may be thought of as possessing the multiplication and excess
noise characteristics due to pure electron injection, similar arguments apply to
hole initiated events. The multiplication and excess noise that a device yields
under mixed injection conditions is then a function of the ratio of number of
electron and hole initiated events. In the hypothetical example shown in Fig. 1.4,
β = α and there are the same number of each primary carrier type so the mixed
injection result would lay in the middle of the pure results, similar arguments
apply to noise. Note that the middle is not necessarily the mean. In the case
of β not equal to α the more readily ionising carrier type will have far more
events associated with it than the less readily ionising carrier type. When the
ionisation coefficients are disparate, special consideration should be given to the
case of very slightly mixed injection in which injection by diffusion of the less
readily ionising carrier dominates, but a few stray carriers of the more readily
ionising carrier type are also injected due to a small amount of generation inside
9
CHAPTER 1
1 1
1
2
2
1
2
2
2
3
3
2
3
3
Figure 1.4 – Mixed injection impact ionisation process shown schematically.
the e-field. In this case a very small multiplication and very high noise may be
expected. However, the very few more readily ionising carriers that do ionise the
lattice may have, on average, very long ionisation chains, whereas the very many
less readily ionising carriers may have an average chain length of only one or two
events. Hence the majority of the carrier population exists due to the few stray
more readily ionising carriers, despite the initial conditions, which may suggest
otherwise. This ensemble of carriers in which each pair of carriers may be thought
of as possessing multiplication and noise characteristics imbued to them by their
initiating carrier, would exhibit multiplication and noise characteristics similar
to that of the more readily ionising carrier. When measuring devices in which
α and β are widely separated, for example two orders of magnitude, it may be
very troublesome to produce a result exhibiting behaviour dominated by the less
readily ionising carrier type by anything other than perfect injection conditions.
1.3.2 SAM-APD
The five region structure shown in Fig. 1.5, is one possible implementation of
a separate absorption and multiplication region avalanche photo-diode (SAM-
APD). A plausible electric field profile is shown in the same figure. The operation
is fundamentally similar to the three region structure with the exception that two
further regions are added. Photons enter via the n+ layer which has an etched
10
CHAPTER 1
optical window. Assume that the majority of carriers are generated in the upper
n− region where a low electric field exists. Avalanche multiplication takes place
in the lower n− region. The benefits of a SAM structure include,
• Reduced transit time of the multiplication region. This reduces the effect
of carrier feedback over a similar width three region structure.
• Reduced breakdown voltage over a similar width three region structure.
This is a result of the high field being limited to the p+ n− and n+ charge
or grading layer. The device is approximately as thin as a three region
structure that is composed of only the p+ n− and n+ (charge) layers.
• Increased carrier collection and therefore quantum efficiency due to the
electric field reaching close to the device surface. More carriers maybe
collected by the field, as the number of carriers lost due to recombination
in the n+ cap layer is minimised.
• Increased velocity of carriers moving from the point of generation to the
high field region. In the three region structure carriers generated outside
the high field region must diffuse in. In the SAM structure the carriers
experience a low field in the absorption layer which sweeps them into the
multiplication layer.
• The possibility of pure injection is not negated by the electric filed reaching
close to the device surface. Carrier pairs generated in the absorption region
can contribute to pure injection if the carriers drifting in to the low field
towards the optical window cannot gain enough energy to ionise the lattice.
Under these conditions all the carriers entering into the multiplication re-
gion from the absorption region have been optically promoted and are not
the result of ionisation events. This is equivalent to optical carriers diffusing
into the high field region in a three region structure.
• A further benefit is the possibility of using two material systems to form
a heterostructure in which the absorption region is composed of a material
suited to absorbing light of the required wavelength and the multiplication
region is composed of a material that exhibits high gain, low dark current
11
CHAPTER 1
n+
n−
n+
n+
p+
n+
|ξ|
x
Figure 1.5 – A generic SAM-APD photo-diode with electric field shown to the right.
and bulk breakdown. An example of this is a Germanium absorber with
a Silicon multiplication region. Ge-Si SAM-APDs are in competition with
some III-V materials for use as detectors in the second and third generation
(1.3 µm and 1.55 µm) low loss fibre windows [4].
• A final advantage of the SAM structure is the lower capacitance of the device
due to the wider depletion region than in a three region structure with a
similar breakdown voltage. Assuming the device is not transit time limited,
lower capacitance translates directly into a higher bandwidth device.
1.3.3 Local Model Multiplication and Excess Noise
The ‘local’ model of photo-multiplication and excess noise in avalanche photo-
diodes is principally due to two parties. The multiplication expressions were
reported by the early Silicon workers including [5–12]. The local excess noise
model was derived by McIntyre [3]. The model is elegant, and may be solved with
closed form for some simple cases making it computationally efficient. Solutions
may be found for quite complex cases of mixed injection profile and continuous or
piecewise continuous electric fields for any number of regions. The expressions can
be evaluated by numerical approximation using an efficient quadrature algorithm
such as Simpson’s Rule or Gauss Legendre quadrature [13]. Evaluating symbolic
solutions is of limited value in all but the simplest cases where a constant electric
field is assumed. In this case the symbolic solutions may be used to extract α
and β from measurements of photo-multiplication and excess noise factor. The
12
CHAPTER 1
governing approximation of the model is that the likelihood of lattice ionisation
by any given carrier is a function of the electric filed in the vicinity of the carrier,
and is independent of the energy history of the carrier.
When invoking the local model, the ratio of ionisation coefficients is labelled k
and it is often assumed constant for a particular device under common conditions
such as pure electron and pure hole injection. k is however a function of α and β
and so changes across the depletion distance. k is sometimes quoted for a device,
this is the “effective k” and is particular to the device. Effective k is formed
from an ensemble of the position dependent k over all positions. It is however
not possible to assign a particular value of effective k to a material system as
the effective k is a function of device structure. Thicker devices are generally
less noisy than thinner ones. This generalisation is based on the premise that, in
many material systems, α and β diverge as electric field diminishes. In a thicker
device, which will break down at a lower electric field than an arbitrarily chosen
thinner device, the effective k will be lower than in the thinner structure biased
such that it produces the same ensemble multiplication.
1.3.3.1 Applicability and Limit
Dead space is a term for the distance a carrier must travel through the lattice
before the carrier can possess a non-zero probability of ionising the lattice. The
dead space is indirectly measured by expressing the energy in electron Volts
required to overcome the dead space. During the traversal of the dead space
the carrier gains energy from the electric field. It is common to assume that
carriers drift at the saturation velocity and the energy gain is in the form of
momentum. The accuracy of the local model decreases as the dead space length
becomes significant with respect to the high field region width. Put another way,
as the device is made thinner – intrinsic width is reduced – the dead space length
is a larger proportion of the intrinsic width. Under these conditions the prior
interactions between the carrier and the lattice, and the carrier and the electric
field influence the probability of ionisation in a particular region under scrutiny.
As this influence increases, the local model becomes less accurate.
In the limit, α or β → 0, single carrier multiplication exists. If the multipli-
13
CHAPTER 1
cation is large, the average multiplication chain length must be long. Since there
is little or no carrier feedback, an average carrier must ionise many times in one
pass of the high field region to produce the ensemble multiplication. There will
be a spread of multiplication chain lengths, some carriers will ionise the lattice
many times and others will will pass out of the high field region without ionising
the lattice at all. In this case a more general relationship between the average
multiplication and the dead-space is unveiled, where the parameters of interest
are the multiplication, dead-space length, and device width. The product of the
former two can not be greater than the latter. This implies that in single carrier
multiplication an upper boundary of multiplication exists for a given magnitude
of electric field. Under favourable conditions secondary carriers may be gener-
ated in spatially confined regions and then travel through the device in quasi two
dimensional ‘sheets’. This collecting of carriers – which is essentially a reduction
in the variance of multiplication values for the population of carriers – causes the
device to exhibit less noise than is supposed by the local model. Devices exhibit-
ing less noise than the minimum local model noise have been reported in InAs [14]
and HgCdTe [15] material systems. Marshall [16] has proposed an excess noise
reduction factor which provides a figure of merit proportional to the reduction of
excess noise due to the spatial confinement of the ionisation events.
1.4 Non-Local Multiplication and Noise
In this work the non local RPL model reported by Ong et al. [17] is used. The
program used to perform the required modelling is due to B. K. Ng [18], and is
used with permission. The RPL model operates by sequentially passing many
carriers into a device structure of known region widths and doping densities. The
likelihood of a carrier reaching a given distance from its prior ionisation location
(or generation point etc.) without ionising is given by a probability density
function. This is often composed of the forbidden ‘dead space’ followed by an
exponential decay. Each carrier is ‘followed’ through the device; each carrier
pair that is promoted by the original carrier is then followed. The following
of carriers continues until all the carriers produced in a particular ionisation
chain have left the electric field region. A new carrier is then inserted, and the
14
CHAPTER 1
p+ n− n+
hv hv
Figure 1.6 – Pure injection in a three region device. Both electron and hole injection areshown.
process is repeated. To provide a statistically representative result a large number
of carriers, perhaps as many as a million, are used. This modelling method
computationally intensive compared to the local model. However RPL is more
capable of accurately representing measured device behaviour under conditions
where the stochastic nature of the scattering process is limited by the prevailing
device conditions.
1.5 Derivations
In this section the local model expressions required to produce a mixed injection
model are derived. These expressions are often used with a piecewise contin-
uous approximation to the electric field, and a Poisson solver under which the
depleation approximation is observed. A suitable Poisson solver is derived in
Appendix A.
1.5.0.2 Pure Injection
The pure injection cases are simplifications of the mixed injection case, the deriva-
tion of the local model can be found in the literature [3]. Light enters the p+ or
n+ layer of a three region structure, shown in Fig. 1.6. All carriers are generated
outside the electric field.
15
CHAPTER 1
Pure Electron Injection The expression for photo multiplication is a simpli-
fication of the mixed injection case given by McIntyre [3] equations 1 - 5.
Me =1
1−(∫ w
0
α(x) exp
(
−∫ x
0
α(x)− β(x) dx.
)
dx.
) (1.9)
Note that the multiplication will become undefined when the integral in the de-
nominator reaches unity, this is regarded as a definition of breakdown. From a
perspective of considering carriers in a device, the requirement is that a single
carrier can promote an unlimited number of secondary carriers. This has destruc-
tive implications for a real device unless steps are taken to limit the current flow
and power dissipation. Breakdown is independent of the injection conditions and
incident photon energies. pin and nip devices possessing identical region widths
and doping densities and being made from the same material will break down at
the same reverse bias voltage.
The noise due to pure electron injection may be given by,
Fe = k ·Me + (1− k)
(
2− 1
Me
)
(1.10)
where k = α/β and may be considered constant over a suitably small range of
electric fields. Without invoking k it may be shown that the noise is also given by
considering the noise spectral density in a device. This expression can be derived
from [3] equation 13. Note that to arrive at this equation we have considered the
increase in hole current in a distance dx whereas McIntyre chose to consider the
increase in electron current in a small region dx. This does not represent a loss
of generality.
The noise spectral density for the pure electron case,
φ = 2 · q(
2 · Ie ·M2e + It
(
2
∫ w
0
β(x)M(x)2 dx.−M2h
))
(1.11)
The total current in the device,
It = Ie ·Me (1.12)
16
CHAPTER 1
substituting,
φ = 2 · q(
2 · Ie ·M2e + Ie ·Me
(
2
∫ w
0
β(x)M(x)2 dx.−M2h
))
(1.13)
The excess noise in terms of noise power spectral density,
F =φ
2 · q · It ·M2e
(1.14)
substituting,
F =
2 · Ie ·M2e + Ie ·Me
(
2
∫ w
0
β(x)M(x)2 dx.−M2h
)
Ie ·M2e
(1.15)
Pure Hole Injection For pure hole injection multiplication similar arguments
apply as for electrons. The expression is,
Mh =1
1−(∫ w
0
β(x) exp
(
−∫ w
x
α(x)− β(x) dx.
)
dx.
) (1.16)
The noise due to pure hole multiplication while defining k as αβis,
Fh =1
k·Mh +
(
1− 1
k
)(
2− 1
Mh
)
(1.17)
Similar arguments apply as in the prior section where the noise under pure hole
injection conditions is given by,
F =
2 · Ih ·M2h + Ih ·Mh
(
2
∫ w
0
β(x)M(x)2 dx.−M2h
)
Ih ·M2h
(1.18)
1.5.0.3 Generation of Carriers
A simple model of distributed injection has been reported by Tan et al. [19].
Their model has been used in Chapter 7. The operation of the model is shown
for p+ side injection in Fig. 1.7. Laser light is incident on the p+ surface of the
17
CHAPTER 1
p+ n− n+
G(x) = A0e−α·x
hv
Xp w XnA B
Figure 1.7 – This figure shows the relevant distances and expressions for p+ side injection.
device, and the n+ surface is backed by a substrate. The electric field lies in the
region marked by w, Xp and Xn are the distances to the layer edges from the
electric field edge. The minority electrons in the p+ layer traverse from left to
right and minority holes in the n regions traverse from right to left. Any carriers
generated outside the electric field will be captured by it. If it is desirable to
implement an approximation to the recombination of some optically generated
carriers in the p+ and n+ regions then an arbitrarily complex function of distance
may be developed to determine which carriers are collected and which recombine.
The simplest of these, in terms of implementation, is a hard threshold at a fixed
distance from A and B, all carriers generated within the threshold are collected
all others are not.
The quantity of carriers that diffuse into the p side of the electric field is given
by,
Ap = A0 (1− exp (−ϕ ·Xp)) (1.19)
where A0 is the number of carriers incident on the p+ surface per unit area per
unit time, ϕ is the absorption coefficient at the wavelength of interest and Xp is
the distance of undepleated material between the electric field and the p+ surface.
Similarly the quantity of carriers that diffuse into the n side of the electric
18
CHAPTER 1
field is given by,
An = A0 exp (−ϕ (Xp + w)) (1− exp (−ϕ ·Xn)) (1.20)
where w is the electric field distance and Xn is the distance of undepleated ma-
terial between the n side of the electric field and the n side of the device.
The total carriers generated in the device is given by,
At = Ap ·M(0)+An ·M(w)+
∫ w
0
M(x) ·A0 ·ϕ exp (−ϕXp) exp (−ϕx) dx. (1.21)
assuming that x is measured from the electric field edge at A, i.e. x = 0 at A,
x = w at B. The total quantity of optically generated carriers is,
Ag = A0 (1− exp (−ϕ (Xp + w +Xn))) (1.22)
and the ensemble multiplication for the device is given by M = At
Ag,
M =
Ap ·M(0) + An ·M(w) +
∫ w
0
M(x) · A0 · ϕ exp (−ϕXp) exp (−ϕx) dx.
A0 (1− exp (−ϕ (Xp + w +Xn)))(1.23)
The shot noise associated with the photo-current generated at x is ideally
multiplied by M2(x). Hence the shot noise generated by the multiplied primary
photo-current is given by,
Nshot = 2 · q · Ag ·M2 (1.24)
and the total device noise is,
Nt = 2 · q2 · Ap ·M(0)2 · F (0) + 2 · q2 · An ·M(w)2 · F (w)
+ 2 · q2(∫ w
0
M(x)2 · F (x) · A0 · φ exp (−φXp) exp (−φx) dx.
)
(1.25)
where,
F (x) = k ·M(x) +
(
2− 1
M(x)
)
(1− k) (1.26)
19
CHAPTER 1
p+ n− n+
G(x) = A0eα·x
hv
Xp w XnA B CZ
Figure 1.8 – Figure showing the relevant distances and expressions for n+ side injection.
and k has the usual meaning αβ.
The excess noise factor F can be expressed as F = Ntotal
Nshot,
F =1
M2 · Ag
(
Ap ·M(0)2 · F (0) + An ·M(w)2 · F (w)
+
∫ w
0
M(x)2 · F (x) · A0 · φ exp (−φXp) exp (−φx) dx.
)
(1.27)
This expression for F provides equivalent numerical solutions to an expression
for the mixed injection excess noise factor, F , which is derived without invoking
k. M as a function of x is given by the following expression (see [3] equation 5).
M(x) =
exp
(
−∫ w
x
(α(x)− β(x)) dx.
)
1−∫ w
0
α(x) · exp(
−∫ w
x
α(x)− β(x) dx.
)
dx.
(1.28)
For injection of carriers by photons entering from the n+ side,
Making the same original assumptions as Tan et al. [19],
An =
∫ C
B
A0 · exp (ϕ · x) dx. = −−A0 (exp (ϕB)− exp (ϕC))
ϕ(1.29)
20
CHAPTER 1
Ap =
∫ A
Z
A0 · exp (ϕ · x) dx. = −−A0 (exp (ϕZ)− exp (ϕA))
ϕ(1.30)
At = M(A) · Ap +M(B) · An +
∫ B
A
M(x) · A0 · exp (ϕA) · exp (ϕx) dx. (1.31)
Depending on the choice of origin in the diagram Z or A is often zero. In this
case A = 0, B = w.
1.5.0.4 Mixed Injection Noise
The noise power spectral density for the mixed injection case (1.32) is derived
by McIntyre [3]. Note that to arrive at this expression we have considered the
increase in hole current in a small distance dx whereas McIntyre chose to consider
the increase in electron current in a small region dx. This does not represent a
loss of generality.
The noise power spectral density is given by,
φ = 2 · q(
2 · Ie ·M2e + 2 · Ih ·M2
h + 2
(∫ w
0
G(x) ·M(x)2 dx.
)
+It
(
2
(∫ w
0
β(x) ·M(x)2 dx.
)
−M2h
))
(1.32)
The total current in the device,
It = Ie ·Me + Ih ·Mh +
∫ w
0
G(x) ·M(x) dx. (1.33)
substituting,
φ = 2 · q(
2 · Ie ·M2e + 2 · Ih ·M2
h + 2
(∫ w
0
G(x) ·M(x)2 dx.
)
+
(
Ie ·Me + Ih ·Mh +
∫ w
0
G(x) ·M(x) dx.
)
·(
2
(∫ w
0
β(x) ·M(x)2 dx.
)
−M2h
))
(1.34)
21
CHAPTER 1
The following expression represents total photo-current multiplied by M2 in
the noise equation,
It ·M2 = Ie ·M2e + Ih ·Mh2 +
∫ w
0
G(x) ·M(x)2 dx. (1.35)
The excess noise in terms of noise power spectral density,
F =φ
2 · q · It ·M2(1.36)
substitution of equations 1.34 and 1.35 into equation 1.36 yields the final form.
The generation profile is a relative measure of generated carriers as a function
of distance. Therefore Ie can be replaced with G(0) and Ih with G(w) assuming
the origin is the point where electrons diffuse into the high field and w to be a
similar point for holes. Me can be replaced with M(0) where M(x) is the position
dependant multiplication expression (1.28) and Mh is replaced by M(w). If one
chooses to consider the increase in electron current in a small region dx, a slightly
different form of (1.34) results, however they provide identical numerical results.
1.6 Extraction of Ionisation Coefficients
In some cases multiplication and noise due to one carrier type may not be easily
measurable. In this case the values of α and β can be extracted from measure-
ments of the other carrier type, provided both multiplication and excess noise
data are available. In the constant electric field approximation α and β are sin-
gle valued and independent of distance. An algebraic analysis is possible. The
expression for Mh (1.16) reduces to,
Mh = − α− β
β exp ((α− β)w)(1.37)
and the relation between Fh and Mh while invoking k is given in (1.17) substi-
tuting k = α/β,
Fh =α ·Mh
β+
(
2− 1
Mh
)(
1− α
β
)
(1.38)
22
CHAPTER 1
α and β may be expressed as a functions of Mh and Fh only by simultaneous
solution of (1.37) and (1.38),
α =
ln
(
− Fh −Mh
M2h − 2 ·Mh + 1
)
(Fh ·Mh − 2 ·Mh + 1)
Mh · w (Fh −Mh)(1.39)
β =
ln
(
− Fh −Mh
M2h − 2 ·Mh + 1
)
(M2h − 2 ·Mh + 1)
Mh · w (Fh −Mh)(1.40)
If Mh, Fh and the depletion width, w, are known for a particular reverse bias
voltage, or range thereof, β and α as functions of electric field can be found
directly by substitution.
23
CHAPTER 1
24
Chapter 2
Measurement Techniques
In this chapter some standard measurements are discussed including current –
voltage, capacitance – voltage and photo-multiplication measurements. These
measurements are routinely performed when a new diode sample becomes avail-
able. The excess noise measurement system after Li [20] is included. The chapter
concludes with a description and evaluation of other noise measurement systems
reported in the literature.
2.1 Current Voltage Characteristics
Forward and reverse current – voltage characteristics are one of a number of
tests performed shortly after devices are manufactured. Forward current voltage
characteristics reveal whether the structure has an exponential characteristic and
therefore obeys the Shockley equation (2.1) [21].
Ia = Is
(
expq (Va − IaRs)
n k T− 1
)
(2.1)
where I is the total current, Is represents the saturation current, q is the
electron charge, k is Boltzmann’s constant, T is the absolute temperature, and Va
is the terminal voltage, n is the ideality factor and Rs is the series resistance. The
ideality factor gives insight to the nature of the current flow across the transition
25
CHAPTER 2
Figure 2.1 – Figure showing the effect of series resistance on the characteristics of a forwardbiased diode. Is = 0.1×10−27 A; n = 2; k = 1.38×10−23 JK−1; q = 1.6×10−19 J; T = 300 K;Rs = 1 Ω to 1 MΩ in decade steps.
region. Ideality factors close to unity represent dominance of diffusion of carriers
into the transition region. Ideality factors close to two represent the opposite
situation where the current flow across the junction is dominated by generation
– recombination within the transition region. The ideality factor can be treated
as a figure of merit related to the crystal quality. The forward characteristics will
also reveal any series resistance, which may ultimately limit the device current.
Attempting a direct solution of the Shockley equation by collecting the terms
in Ia requires the use of Lambert’s Omega, and takes the form,
Ia = −−n k T W
(q Rs Is exp( q (Va+Rs Is)
n k T )nk T
)
+ q Rs Is
q Rs(2.2)
This expression can be solved numerically to produce Fig. 2.1 which demonstrates
the effect of series resistance, but (2.2) is cumbersome. The W function can be
avoided by solving for Rs and not attempting to collect terms in Ia. The resulting
expression is,
Rs = −ln( Ia+Is
Is)n k T
q− Va
Ia(2.3)
26
CHAPTER 2
The practical use of this expression is possible, by inserting a value of anode –
cathode voltage and anode current and obtaining a value for Rs. It is important
to choose a voltage at which the series resistance is apparent otherwise accuracy
is compromised. It is also necessary first to know the value of Is, which may be
found by fitting (2.1) to the data (assuming Rs = 0) or by inserting values into,
Is = − Ia
− exp(
q Va
nk T
)+ 1
(2.4)
The values of Va and Ia substituted into (2.4) must come from the region where
the series resistance does not dominate. Substitution of Is into (2.3) is the next
step, but an estimate of the ideality factor is also required. It is very much quicker
to use a trial and correction approach to find Is and Rs using only (2.1). Using a
graphical method the quality of the fitting can be grasped instantly and corrected
based on all of the data, not just a single point, without difficulty.
The reverse bias characteristics show the breakdown voltage, which can be
used in conjunction with capacitance voltage characteristics to assist modelling
the device region widths or doping densities. The nature of the breakdown may
also be apparent if several devices sizes are available, bulk breakdown is preferable.
This occurs when the maximum electric field that the material type can support
is exceeded. Ideally the current flowing when the device is reverse biased would
be the saturation current only. However in real devices the reverse leakage current
is significantly higher and is composed of the diffusion current from outside the
transition region and the generation current. There may also be an element
of surface leakage and tunnelling. Reverse IV characteristics can also exhibit
behaviour caused by other aspects of the diode structure. For example a very low
leakage current diode may exhibit a linear relationship on a log–log plot for the
first 75% of the reverse bias voltage, before undergoing bulk avalanche breakdown.
In this instance one plausible explanation is that the surface passivation has a
resistance which is constant and dominates when the diode is only exhibiting the
saturation current.
If forward and reverse characteristics can be obtained from devices having
a range of areas, the nature of the current flow may be examined. In devices
where current scales with the device perimeter, it is likely that the majority of
27
CHAPTER 2
the current flow occurs down the surface of the mesa. However a result that scales
with area is indicative of bulk current flow. If neither bulk nor surface processes
dominate the conduction, current density scales poorly with both perimeter and
area.
2.2 Capacitance Voltage Characteristics
The forward capacitance voltage (CV) characteristic yields the diffusion potential
for the device. For a one sided abrupt junction, as larger forward bias is applied
the space charge region shrinks in proportion to the square root of the inverse of
the applied voltage. The parallel plate capacitance is given by,
C = A
(e ε0 εr Nd
2 (VB + V )
) 12
(2.5)
Where e is the electron charge, ε0 is the permittivity of free space, εr is the
relative permittivity of the semiconductor material, Nd is the doping density, VB
is the diffusion or ‘built in’ voltage, and V is the applied terminal voltage.
The built in voltage may be estimated from a plot of 1/C2 versus bias. It is
given by the intersection of an extrapolation of the data and the bias axis. As
the device begins to turn on the incremental resistance reduces significantly, and
the measurement accuracy deteriorates rapidly.
Reverse capacitance voltage characteristics, are used to determine the deple-
tion width or doping density in conjunction with data provided by other experi-
ments such as secondary ion mass spectroscopy (SIMS) and the use of a Poisson
solver (see Appendix A). The depletion width and doping densities may be ad-
justed to fit a device CV profile. Breakdown voltage is also used as a guide in
determining the intrinsic region width. The reverse CV profile is required to
calibrate the new excess noise measurement systems described herein and to cal-
ibrate the excess noise measurement system after Li [20], which will be discussed
shortly. Reverse CV measurements should scale with device area.
28
CHAPTER 2
2.3 Photo-Multiplication
The simplest form of photo-multiplication measurement is similar to reverse IV
measurements with the exception that the diode is illuminated. This DC method
can not separate the dark current from the photo-current. The DC experiment
may be considered accurate when the dark current is at least two orders of mag-
nitude lower than the photo-current. In this work several combined AC multipli-
cation and excess noise measurement systems are used, these are used to obtain
multiplication and excess noise simultaneously. However from a multiplication
perspective they may be treated as any other AC method. As the reverse bias
across a pin / nip is increased, the depletion region extends into the highly doped
p+ and n+ layers on each side of the intrinsic region. As a result, minority car-
riers, generated outside the depletion region, have a greater chance of diffusing
into the depletion region. This increase in collection efficiency is manifested by an
approximately linear increase of photo-current with bias before the onset of any
real multiplication. After multiplication begins the collection efficiency continues
to increase; however, it is masked by the multiplication process. The collection
efficiency has a significant effect on the primary photo-current. It is critical to
account for this effect by using the non multiplying region data to estimate the
primary photo-current in the region where impaction ionisation does occur. The
value of multiplication is obtained from point by point division of the multiplied
photo-current by the (regressed) primary photo-current. Many data points are
taken in the non-multiplying bias region in order to yield sufficient accuracy.
2.4 Review of Noise Measurement Systems
Several excess noise measurement systems have been reported in the literature
and comparisons between the circuits described in this thesis and those previously
reported may be drawn. The figures of comparison are,
• The system signal to noise ratio, where the signal is defined as full shot
noise exhibited by 1µA
• The maximum permissible junction capacitance. In the case of multi-
29
CHAPTER 2
Laser Chopper
Chopper Driver LIA
R1
C1
Bias Tee DC Bias
AIL 13680Pre-Amplifier
AIL 1361030MHz Reciever
LIA Chart Recorder
Figure 2.2 – Measurement system after Bulman.
frequency systems the lowest available frequency is used; this produces the
most favourable result. It is assumed that the system input impedance and
diode junction capacitance form a first order low pass network.
The first reported noise measurements on photo-diodes was by Baertsch [22].
Insufficient information is provided to estimate this system’s figures of merit so
it is excluded from the comparison. Xie et al. [23] proposed a measurement
system that was substantially similar to Toivonen et al. [24]. The Xie et al.
system represents both. Bulman [25], Ando and Kanabe [26] and Lau et al. [27]
presented systems based on phase sensitive detection. Xie and Toivonen used a
DC approach.
2.4.1 Bulman’s System
Figure 2.2 shows the system reported by Bulman. It is a phase sensitive detection
(PSD) based system in which photo-current and excess noise are extracted and
read out using two lock in amplifiers. The APD is loaded by the AIL13680
through the bias tee. It may be assumed that the APD experiences a 50 Ω load.
The pre-amplifier output is fed to a receiver having a calibrated bandwidth. The
30
CHAPTER 2
resulting signal proportional to the noise power contained within the calibrated
bandwidth is passed to a lock-in-amplifier.
Bulman proposes two methods to quantify the absolute noise power mea-
sured. Firstly a pin detector is illuminated under unity gain conditions. It is
assumed that under these conditions the system will measure full shot noise. A
second calibration method is proposed in which a calibrated oscillator is used
in place of the APD. This allows the experimenter to set the power which will
be measured. Adjusting the power supplied by the oscillator permits the linear
range of the system to be estimated. Bulman reports 30 dB linear range between
−140 dBm (10−17 W) and −110 dBm (10−14 W). In several III–V semiconduc-
tors the impact ionisation coefficients are nearly equal. Under this assumption
Tager [28] has shown that excess noise is proportional to M3, when the local
model holds. Assuming Bulman’s system is used from unity gain with a noise
power of −140 dBm the maximum multiplication prior to the limit of linearity
is 10. Bulman’s report lacks some information regarding the front end amplifier.
An Analog Devices AD9618 low noise opamp in non-inverting mode is used as
a model. It achieves a gain of 100 V/V and a bandwidth of 80 MHz with 50 Ω
input impedance. The equivalent input noise voltage is 1.94 nV/√Hz. Using
this model as an approximation the signal to noise ratio for Bulmans system is,
S
N=
((100 · 50)2 · 2 · q · iph
)
(1.94× 10−9 · 100)2= −36.73 dB (2.6)
The capacitance of the diode which can be tolerated by Bulman’s system is given
to a first approximation by considering the system input impedance and the
junction capacitance as a first order RC network and assuming that all other
parasitic effects (for example diode series resistance) are negligible in comparison.
Applying the commonly known expression for the corner frequency of a first order
network RC,
C =1
2 π f R, and using f = 30 MHz and R = 50 Ω gives C = 106 pF. (2.7)
Larger capacitance will attenuate the components of noise power at the 30 MHz
measurement frequency.
31
CHAPTER 2
Light Source Chopper
Chopper Driver
A
V
Sig. Gen.
Power Meter
IF Amplifier
LIA
Bias Tee
Figure 2.3 – Measurement system after Ando and Kanbe.
2.4.2 Ando and Kanbe
Ando and Kanbe [26] reported the system shown in Fig. 2.3. It is a PSD system in
which the APD is loaded by 50 Ω due to the input impedance of the IF amplifier.
The measurement system bandwidth is defined by the IF amplifier and is 1 MHz
centred on 30 MHz. The APD is biased using a bias tee. The noise power is
read out from a lock in amplifier. A power meter and signal generator with its
output passed through a calibrated attenuator provides a means of relating the
absolute signal power to the value measured by the IF amplifier. Photo-current is
extracted by DC measurement. The measurement of devices exhibiting high dark
current is therefore difficult with this system. The various connections required
to calibrate the system and perform measurements are made using relays.
Ando and Kanbe did not report any attempts to measure shot noise on their
system. They also do not give information regarding the model numbers or
manufactures of their system components. No noise specifications for the instru-
mentation are given. Assuming that their system adds no noise other than the
32
CHAPTER 2
thermal noise of the 50 Ω input impedance then the signal to noise ratio can be
computed using,S
N=
2 q iphBR
(4 k T BR
)= −23.98 dB (2.8)
where R = 50 Ω, B = 1 MHz, T = 300 K and iph = 1 µA. The junction
capacitance which can be tolerated by Ando and Kanbe’s system is calculated in
a similar way to Bulman’s system and produces the same result C = 106 pF. The
authors claim that noise power as low as -130 dBm can be measured with 0.5 dB
accuracy. This represents a current of 6.25 µA developing full shot noise.
2.4.3 Xie
The system proposed by Xie [23] is similar to that proposed by Toivonen [24].
Figure 2.4 shows the APD is connected to a micro-strip line and DC voltage is
applied via a bias tee. The measurement is made using a CW light source and
a noise figure meter such as the Hewlett Packard 8970A. The system has two
significant advantages over PSD systems such as those of Bulman and Li. Firstly
multiple frequencies are available up to the limit of the circuits or the analyser.
Presently Agilent Technology manufactures noise figure meters capable of mea-
suring 10 MHz to 26 GHz with variable effective measurement bandwidth. Xie’s
system was limited to a maximum frequency of 1.3 GHz and 4 MHz bandwidth.
Secondly the measurement is in principle quicker than a PSD system. The oper-
ation of PSD is discussed in Appendix C but in general the time constant of a
PSD measurement may be expected to be longer than of a noise figure meter. DC
measurements have several disadvantages for example the lowest practically mea-
surable photo-current is higher than in some PSD systems using transimpedance
based amplifiers Li [20] has shown that the transimpedance amplifier reported
by Lau [27] has a more favourable noise signal to noise ratio than is theoretically
possible with a 50 Ω measurement system. The noise without illumination should
be periodically measured in order to maintain consistency. The noise without il-
lumination should be stable and sufficiently small compared to the noise with
illumination that the total noise is dominated by the photo-generated noise. If
this condition is not met the confidence of the measurement is compromised.
33
CHAPTER 2
ZT
C1
L1
−+
VB
C2
zin
Vout
Figure 2.4 – Measurement system after Xie et al.
Xie et al. reported measuring noise power as low as −190 dBmHz−1 without
difficulty using this measurement system. In a 50 Ω system this is equivalent
to the full shot noise generated by 5 µA photo-current. The capacitance which
can be tolerated by this measurement system is computed at the lowest usable
frequency, as this produces the most favourable result. By the same first order
approximation used in Bulman’s and Ando and Kanbe’s systems Xie’s system
will exhibit a -3 dB bandwidth of 10 MHz when the input is loaded with 636 pF.
In Chapter 9 an improved noise figure meter based measurement system is
proposed [29].
2.4.4 Li’s System
The system after Li [20,27] employs phase sensitive detection and a transimpedance
amplifier. A schematic diagram is shown in Fig. 2.5. The laser is chopped by
mechanical means at 180 Hz and is presented to the diode via a system of optics.
The TIA is used to convert the diode current into a voltage. This voltage is
amplified using a commercial low noise wide band amplifier module (Minicircuits
ZFL-500). A precision stepped attenuator (HP-355D) is used to vary the system
gain permitting measurement of high and low noise devices. The noise signal is
separated from the low frequency component of the photo-current by a Minicir-
cuits SBP-10.7+ LC ladder filter. After filtration, the signal resembles an am-
plitude modulated noise waveform, where periods of diode illumination produce
greater noise amplitude than periods of darkness. Further amplification follows,
34
CHAPTER 2
Chopper Driver Lock-in-amplifier
Stanford SR830
Lock-in-amplifier
Stanford SR830
Laser
He-Ne
2.5mW, 633nm
Chopper
Bias Source
Keithley 236
TIAVoltage Amp
Minicircuits
ZFL-500LN+ HP 355D Minicircuits
SBP-10.7+
Voltage Amp
ZFL-500LN+
Power Meter
VoltageAmplifier
Figure 2.5 – Measurement system after Li.
prior to a wide band squaring and averaging circuit. The output of the squaring
and averaging circuit is an approximately square voltage signal. The fundamen-
tal frequency of the signal is 180 Hz; the amplitude is proportional to the noise
power contained within the pass band of the SBP-10.7+ filter. The squaring cir-
cuit is based on an Analogue Devices AD834 analogue multiplier. The averaging
circuit is a first order RC filter with a time constant of approximately 100 µs.
The output from the squaring and averaging circuit is measured using a lock-in-
amplifier. The photo-current signal is taken from the output of the TIA where
the amplitude of the 180 Hz square wave is proportional to the photo-current.
The photo-current signal is measured on a second lock-in-amplifier.
The system after Li is superior in noise performance to prior reported systems.
The transimpedance amplifier provides a signal to noise ratio which is superior to
that theoretically possible in a 50 Ω system. Consider the connection of a photo-
diode and a 50 Ω resistor. Assume that full shot noise generated by 1 µA flows
through the resistor which exhibits thermal noise at 300 K. The noise signal to
noise ratio is then,
20 log
(√2 q iph
)
(√4 k T 50
)
= −30.15 dB (2.9)
The noise signal to noise ratio (also considering 1 µA photo-current) of Li’s
system is -25.7 dB [20]. A first order approximation considering only the volt-
35
CHAPTER 2
age noise of the Analog Devices AD9631 (7 nVHz−1/2) [30] operational amplifier
based TIA yields a signal to noise ratio of approximately -21 dB; this simple
calculation demonstrates the advantage of transimpedance amplifiers.
The dynamic range of this system is limited at the lower bound by the abil-
ity of the lock in amplifier to extract the in-phase signal from the system noise.
Practical experimentation has shown that full shot noise developed by 1 µA is
approaching the limit and the shot noise from 0.1 µA is not measurable. The
precise limit is difficult to quantify because it is affected by the prevailing electro-
magnetic conditions both passing through the experiment area and on the mains
power supply. At the upper bound the maximum attenuation of the stepped at-
tenuator provides a limitation however more attenuation could be added without
difficulty. The linearity of the transimpedance amplifier at high input current is
a second limit. Because the relationship between excess noise factor and photo-
multiplication varies between material systems it is unwise to speculate on the
maximum multiplication which can be used. Furthermore if a device is avail-
able which can be operated with a very large gain the optical illumination may
be reduced in order to reduce the multiplied photo-current and the excess noise
power. In this way higher multiplication values may be measured. In order to
measure lower multiplication values a larger primary photo-current is required.
By performing two or more measurements with differing primary photo-currents
it is possible, assuming the APD is sufficiently robust, to measure multiplication
and excess noise power over any desirable range.
The capacitance tolerated by Li’s transimpedance amplifier is lower than all
of the other systems. The interaction of the APD junction capacitance, the feed-
back resistance and the open loop gain of the operational amplifier permit the
existence of resonance in the frequency response of the transimpedance amplifier.
When the capacitance is sufficiently large oscillation breaks out and the mea-
surement system is saturated. The limit of measurable junction capacitance is
however not governed by the presence of oscillation. A result of the interaction of
the diode junction capacitance and the transimpedance amplifier is a dependence
of the effective noise power bandwidth of the system on the diode junction ca-
pacitance (which is itself dependant on the applied DC bias voltage). As a result
a correction must be applied when post processing the measurement data. The
36
CHAPTER 2
limitation of the capacitance is governed by the quality of the correction which
can be achieved. While it is known that up to 56 pF does not cause oscillation,
Li placed the limit at 28 pF [20]. This limit was obtained by calibrating the
bandwidth of the transimpedance amplifier with several values of capacitance.
Having performed the calibration, shot noise due to photo-generated carriers was
measured using a silicon photo-diode. A second data set was gathered in which
extra capacitance was placed in parallel with the photo-diode to simulate a diode
of greater capacitance. The simulated higher capacitance shot noise data was
processed using the original calibration and the quality of the fitting of the stan-
dard photo-diode shot noise and the simulated extra capacitance shot noise data
was used as a basis for defining the upper limit of the junction capacitance.
2.4.5 Summary of Prior noise measurement systems
Table 2.1 – Comparison of published noise measurement systems calculated for 1 µA signalcurrent.
AuthorNSNR [dB](cj = 1pF)
Maximum cj [pF]
Bulman et al. -36.72 106Xie et al. -31.58 636
Ando and Kanbe -23.98 106Li et al. -25.70 50
37
CHAPTER 2
38
Chapter 3
Transistorised High Capacitance
Noise Measurement System
In this chapter a versatile system for measuring excess noise and multiplication in
avalanche photo-diodes, using a bipolar junction transistor based transimpedance
amplifier front-end is reported. This system is based on phase-sensitive detection,
which permits accurate measurement in the presence of a high dark current. The
system can reliably measure the excess noise factor of devices with capacitance
up to 5 nF. It has been used to measure the thin, large area Si pin APDs which
are reported in Chapter 6. These data are in good agreement with measurements
of the same devices obtained from a second noise measurement system which is
reported in Chapter 5.
3.1 Introduction
Avalanche photo-diodes exhibit an internal gain mechanism whereby secondary
carriers are generated by impact ionisation [21]. Impact ionisation is often ex-
ploited to enhance the signal-to-noise ratio of electro-optical systems in commu-
nications, medical and military applications [31–33]. A popular model that de-
scribes avalanche multiplication was proposed by McIntyre [3]. The electron (α)
and hole (β) ionisation coefficients are usually reported as a function of electric
field. α and β are often extracted from measurements of the photo-multiplication
39
CHAPTER 3
R1
−+
VB
Voutcp
C1
Figure 3.1 – High impedance amplifier.
and excess noise factor of a particular APD structure, fabricated from a certain
material; for example in [34–36]. The ratio keff = α/β may be used as a relative
figure of merit when comparing two or more competing material systems.
Several measurement systems have been reported which evaluate the excess
noise associated with impact ionisation mechanism [23, 25–27]. Each of these
systems has some limitations with respect to the measurement bandwidth, the
minimum detectable photo-generated noise, and the maximum permissible device
junction capacitance. The merits of each system are discussed in Chapter 2.
The measurement system and front end reported in this chapter enable mea-
surements of photo-multiplication and excess noise on devices with junction ca-
pacitance, cj , up to 5 nF. For the first time wide band excess noise measurements
on thin, large area devices including forward biased devices and photovoltaic cells
are possible. It has been suggested that the noise magnitude of a solar cell is pro-
portional to the density of material defects [37]. The defect density is linked to
cell quality [38]. Noise measurements may be used to grade solar cells at the time
of production. However the measurement system is non-specific and any current
mode device or circuit – one that can be approximated by a Norton source (see
Fig. 3.4) – may be measured.
3.2 Front-End Design Methodologies
Ignoring distributed techniques, amplifiers that interface electronic systems with
electro-optical detectors can be split into three groups based on their input
impedance [39]. Each of these will be discussed briefly in order to illuminate
their relative merits. They are,
40
CHAPTER 3
ZT
C1
L1
−+
VB
C2
zin
Vout
Figure 3.2 – Impedance matched amplifier.
• High impedance (voltage) amplifiers
• Impedance matched (power) amplifiers
• Transimpedance amplifiers or current to voltage converters
3.2.1 High Impedance Amplifiers
A generic high impedance amplifier is shown in Fig. 3.1. A detector, such as a
pin diode, is connected in series with a resistance, R1, and is reverse biased. The
device is illuminated; a photo-current flows in the diode-resistor combination.
A voltage proportional to the photo-current appears across R1; it is presented
to an amplifier having a sufficiently high input impedance that it does not load
the diode – resistor network. While this topology is potentially the simplest
to design, it has considerable practical problems [40]. The bandwidth of the
system is often limited by the time constant created by the resistor, R1, and the
parasitic capacitance associated with the input node, cp. The noise performance
of the system is often dominated by the value of R1.
3.2.2 Impedance Matched Amplifiers
The impedance matched amplifier shown in Fig. 3.2 is designed with a specified
input impedance, zin, which is usually 50 Ω or 75 Ω. This topology is particularly
suited to microwave noise measurements. The measurement system, and in some
cases, the APD impedance, must be carefully designed to ensure that nearly all
41
CHAPTER 3
−+
VB C1 −
+
RF
CF
Vout
Figure 3.3 – Generic opamp transimpedance amplifier.
of the noise power generated in the detector enters the measurement system.
Xie et al. have used a matched approach to noise measurements [23]. In their
system the diode is terminated for AC signals by ZT = 50 Ω such that both ends
of the microstrip line are terminated. Assuming the APD has a large dynamic
resistance, rd, and terminating both ends of the transmission line, the necessity
of tightly controlling the APD resistance is removed. The noise power generated
by the diode is divided equally between the termination, ZT , impedance and the
measurement system input impedance, zin.
In the context of the present work, a diode capacitance, cj , of 1 nF and 50 Ω
input impedance would exhibit an input time constant of 50 ns, which corresponds
to a -3 dB frequency of 3.2 MHz. Attaining wide bandwidths from devices with
high junction capacitance requires a different approach.
3.2.3 Transimpedance Amplifiers
A generic opamp transimpedance amplifier is shown in Fig. 3.3. The shunt–
shunt feedback connection causes the circuit to present low input and output
impedance. The circuit converts the diode current into a voltage. From DC to
HF frequencies operational amplifiers are often well suited to this task, requiring
only a few external components [41]. The Analog Devices AD9631 [30] and Texas
Instruments OPA129 [42] are specifically designed for this application. Several
transimpedance amplifier topologies are reviewed by Sackinger [43]. TIAs appear
frequently in optical communications literature [44,45] and physical measurement
applications in single ended [40,46–50] and differential forms [51]. Excluding the
42
CHAPTER 3
iph rd cj
Figure 3.4 – APD small signal model.
optical communications TIAs, which generally make use of integrated circuit
technology, most measurement TIAs are developed to fulfil a specific instrumen-
tation requirement. Consequently, there is considerable application-dependent
topological variation of TIAs in the literature. At HF frequencies, operational
amplifier based TIAs are not suitable for measuring APDs with high junction
capacitances. It may be shown that an approximation of the input impedance of
an opamp based TIA, when the opamp open loop gain is much greater than unity,
is zin = RF/Av. Open loop gain, Av, decreases with increasing frequency and so
input impedance rises. Opamp based transimpedance amplifiers are also prone
to instability when the input node is loaded with significant capacitance [27].
Consider the input impedance of a transimpedance amplifier designed using an
Analog Devices AD9631 at 10 MHz with a 2.2 kΩ feedback resistor. The open
loop gain at 10 MHz is approximately 22 dB = 12.59 [30], the resulting input
impedance is approximately 175 Ω. The system reported by Lau et al. [27], which
has similar bandwidth and gain requirements to the present work, is unstable with
junction capacitance in excess of ∼50 pF. This chapter is concerned with a new
transimpedance amplifier designed using a bipolar transistor based front end.
3.3 Diode Small Signal Model
A small signal model of a reverse biased pin / nip diode is shown in Fig. 3.4.
The model is comprised of a Norton source iph and rd with a parallel capacitance,
cj . rd is the small signal dynamic resistance of the APD. cj is proportional
to the depletion width, the area of the APD and the relative permittivity of
the material. This model is used throughout the present work. It has proved
satisfactory from DC to HF frequencies. Series resistance may be added to either
terminal if necessary.
43
CHAPTER 3
Chopper Driver
Scitec Inst. 300C
PhotocurrentLock-in-amplifier
Stanford SR830
Excess NoiseLock-in-amplifier
Stanford SR830
Reference Signal
Laser
He-Ne
2.5mW, 633nm
Chopper
Bias Source
Keithley 236
TIAVoltage Amp
HP 355D
Minicircuits
SBP-10.7+
Voltage Amp
Power Meter
Figure 3.5 – Measurement System Block Diagram.
3.4 Noise Measurement System
A block diagram of the noise measurement system is shown in Fig. 3.5. The
laser is chopped by mechanical means at 180 Hz and is presented to the diode
via a system of optics. The TIA is used to convert the diode current into a
voltage. This voltage is amplified using a series of operational amplifiers with a
total terminated gain of ∼13. A precision stepped attenuator is used to vary the
system gain permitting measurement of high and low noise devices. The noise
signal is separated from the low frequency component of the photo-current by
a Minicircuits SBP-10.7+ LC ladder filter. After filtration, the signal resembles
an amplitude modulated noise waveform, where periods of diode illumination
produce greater noise amplitude than periods of darkness. Voltage gain (∼41
terminated) provided by operational amplifiers follows, prior to a wide band,
250 MHz, squaring and averaging circuit which acts as a power meter. The
squared signal is further amplified sixteen times. The output of the squaring and
averaging circuit is an approximately square voltage signal. The fundamental
frequency of the signal is 180 Hz; the amplitude is proportional to the noise power
contained within the pass band of the SBP-10.7+ filter. The squaring circuit is
based on an Analogue Devices AD835 analogue multiplier. The averaging circuit
is a first order RC filter with a time constant of approximately 100 µs. The output
from the squaring and averaging circuit is measured using a lock-in-amplifier. The
44
CHAPTER 3
photo-current signal is taken from the output of the TIA where the amplitude of
the 180 Hz square wave is proportional to the photo-current. The photo-current
signal is measured on a second lock-in-amplifier.
3.5 Transimpedance Amplifier
A simplified circuit diagram of the TIA first stage and biasing circuit is shown
in Fig. 3.6. Some circuit elements have been removed or combined to make the
circuit operation clear. There are two transistor stages which are electrically
similar and are constructed to the same physical layout. One transistor stage,
composed of T1−3, R1, R2, R3, R4 and C1, is ‘active’ and is presented with the
APD. The other, T4−6, R5, R6, R7, R8 and C2, is ‘passive’ and has only quiescent
(DC) conditions. The objective of the active transistor stage is to present a low
impedance to the APD and to convert the photo-current and noise signal into a
voltage. It is preferable to maintain the input node at ground. To enable this,
one of the power supply rails, which supplies both transistor stages, is controlled
by a feedback system composed of, R10, C3, A1 and T7. The advantages this
brings will be discussed in Section 3.7. The outputs of the two transistor stages
are subtracted using an opamp circuit to remove, as far as possible, the DC
conditions of the transistor stages from the noisy photo-current signal. This is
shown in Fig. 3.7. Further gain is provided by several more opamp stages. Output
DC offset adjustment circuitry; similar to an opamp offset null is also provided
by R11.
3.5.1 Transistor Stages
The transimpedance function of the first stage is realised by a single transistor
common base amplifier. The operation of this configuration can be thought of as
a ‘low impedance current amplifier’ [52]. It is the load resistor which converts the
current to a voltage. The common base small signal current gain (α) is slightly
less than unity; therefore the small signal current flowing in the load resistor, R1,
is approximately equal to the small signal input current. It can be shown without
difficulty that the transimpedance gain in the mid band is set by the load resistor.
45
CHAPTER 3
R2, R3 and R4 bias the transistor in a way that provides voltage and thermal
stability. The capacitances C1 and C2 provide ground at the base for AC signals.
C5 is required to make the source measure unit (SMU) a signal ground from the
viewpoint of the APD, and to significantly lessen the high frequency noise that
the SMU injects into the measurement system. The SMUs which have been used
throughout this work are unstable when the output is loaded with more than
20 nF. A small resistance R12 is added in series to isolate it from C5.
3.5.1.1 Design Requirements
The design requirements for the transistor TIA are partially derived from the
existing Li [20] measurement system.
• Transimpedance gain of approximately 2200 V/A (1100 V/A terminated).
• -3dB bandwidth of at least 10 MHz with maximum device junction capac-
itance of 5 nF.
• Lowest obtainable noise.
• DC coupling of the APD to the TIA.
• DC coupling of the output to the rest of the front end.
• Guaranteed stability, irrespective of APD capacitance.
• Operate from +/-15 V supplies.
The first three points are bound to the transistor operating conditions. The
remainder are system considerations. Design proceeds with the solution to the
bandwidth, capacitance and gain problem.
3.5.1.2 Design Process
Using a simplified hybrid–π model the common base stage may be represented
by two real poles [43]. The lower frequency pole is formed at the emitter node by
the APD junction capacitance, cj, and the impedance looking into the emitter,
re. The higher frequency pole is formed at the collector node by the load resistor
46
CHAPTER 3
50 Ω
T1−3
BF840
R1
500 Ω
R2
680 Ω
R3
2.2 k
R4
2.7 k
C1
220 µF
C5
1 µF
R12
100 Ω
VB
C2
220 µF
R7
2.2 k
R8
2.7 k
50 Ω
T4−6
BF840
R5
500 Ω
R6
680 Ω
C3
22 µF
−
+A1
TL071
R10
100 k
R11
10 k
+15V
−15V
+15V
−15V
T7
TIP32C
−15V
+15V
C4
220 µF
− +
A2AD829
C6
68 pF
R13
100 Ω
OP2
−+
A3AD829
C7
68 pF
R14
100 Ω
OP1
+15V −15V −15V +15V
Figure 3.6 – Simplified bipolar transistor transimpedance amplifier.
R1 and all stray capacitances from the collector, including the input capacitance
of the buffer opamp, which is approximately 5 pF for the Analog Devices AD829.
Assuming that the transistor stage can be approximated by a first order low
pass system, composed of only the input pole, the required time constant (τ) to
provide a -3 dB bandwidth of 10 MHz is (1/(2 π)) × 10−7 s is formed by cj =
10 nF and the small signal resistance looking into the emitter, re. The maximum
permissible value of re is approximately 1.6 Ω. re does not appear explicitly in
the hybrid–π model but it can be shown that re ≈ 1/gm if β0 >> 1. The collector
current required to achieve re = 1.6 Ω is given by IC = (k T )/(e re) ≈ 16 mA.
Where k is Boltzmann’s constant, T is the absolute temperature and e is the
electron charge.
The aim is to design an input stage with the same gain as the one used
by Li [20]. It is not possible to obtain all of the required gain from the first
(common base) stage for two reasons. Firstly, the collector to base junction
capacitance, 0.3 pF, and the input capacitance of the opamp buffer, 5 pF, form
a pole with the load resistance at the collector node. Using a load resistance of
47
CHAPTER 3
R3
1kΩ
−
+
A1AD829
R4
1kΩ
OP1
R2
1kΩ
R1
1kΩOP2
C1
68 pF
+15V
−15V
−
+
A2AD829
−15V
+15V
R5
1kΩ
R6
1kΩ
C2
68 pF
−
+
A3LM7171
−15V
+15V
R7
470Ω
R8
470Ω
R9
50ΩC3
220 nF
Noise O/P
−
+
A4TLE2141
−15V
+15V
Photocurrent O/P
Figure 3.7 – Post common base operational amplifier stages.
2.2 kΩ produces a pole at approximately 14 MHz. Stray capacitance will increase
the datasheet capacitance values, further reducing the pole frequency. Secondly,
given the collector current requirement, the DC voltage drop across 2.2 kΩ would
require the use of a supply voltage greater than +/- 15V. This is undesirable
because the circuit is used in a measurement system where +/- 15V is already
available. A 500 Ω load resistance provides an acceptable frequency response, first
stage gain and DC conditions at the collector. Additional voltage amplification
is required and is supplied using operational amplifiers after the transistor stage.
These are shown in Fig. 3.7. The transistor stage supplies approximately one
quarter of the required gain.
3.6 Opamp stages
It is preferable to DC couple the output of the common base stage as well as the
input so that the output of the front end system (Figs. 3.6 and 3.7) contains total
current as well as photo-current and noise information. Using the two common
base stages, with nearly identical operating conditions, the DC voltage added to
the output by the common base biasing arrangement can be largely cancelled by
subtraction to leave only an offset caused by the imbalance between the two halves
of the input stage. This subtraction is accomplished by the operational amplifier
based subtraction circuit shown in in Fig. 3.7. The photo-current output is a
48
CHAPTER 3
buffered version of the noise output. The buffer opamp has a lower bandwidth
than the noise measurement frequency.
3.7 Transistor Stage Biasing Feedback System
The biasing feedback system is designed to permit DC coupling of the APD to the
TIA by maintaining the active transistor stage emitter node at ground potential.
All of the components of the diode current flow into the TIA (assuming rd >> re).
Unlike AC coupling, no separate DC path is required through which noise power
could pass without entering the measurement system. The DC components of
the diode current produce a DC offset at the photo-current output (Fig. 3.7).
The bias voltage displayed on the APD biasing equipment is then equal to the
bias appearing across the device.
The feedback system, which is shown in Fig. 3.6, allows the TIA to be DC
coupled by modifying the lower supply rail voltage of both transistor stages in
order to maintain the potential of the emitters at a level close to ground potential.
This is implemented by using the second transistor stage which possesses, as
closely as practically possible, the same biasing conditions as the first, but without
an APD connected. The quiescent conditions of this ‘passive’ stage are measured
electronically and the lower supply rail of both transistor stages is adjusted in
order to bring the emitter nodes to ground potential. The opamp, A1, changes
the voltage on the lower supply rail in order to maintain the inverting input
and the non inverting input at the same potential (ground). The opamp, A1,
and the RC network of C3 and R10 form an integrator, – the capacitor is in the
feedback loop of the opamp – the time constant is a few seconds. R10 is large
in order to avoid loading the emitter node of T4−6 and in order that a long time
constant should be formed by a reasonably small value electrolytic capacitor. The
potentiometer, R11 may be used to introduce an extra DC current to the passive
front end, modifying its DC conditions slightly in order to produce an offset null
control.
49
CHAPTER 3
3.8 Stability
Unlike TIAs with feedback around more than one active device, there is no inten-
tional negative feedback in the single transistor stage. Consequently, regeneration
due to excessive phase shift, arising from useful components, is not a concern.
However, any physical realisation of a circuit possessing gain may oscillate under
sympathetic conditions. Both emitter follower and common base transistor cir-
cuits suffer from a potential stability problem when the emitter node is loaded
with capacitance [40]. SPICE simulation of the input impedance of the TIA
while connected to the APD small signal model shown in Fig. 3.4 or the model
described in Chapter 8 can provide an estimation of the likelihood of oscillation.
For the common base transistor, oscillation is likely when the real part of the
impedance looking into the emitter, ℜ(zin) is negative for any frequency below
the transistor’s transition frequency. The question of stability in the common
base amplifier is addressed from an analytical perspective in Chapter 4. In this
circuit two avenues of attack are open to combat oscillation problems both at-
tempt to prevent oscillation by affecting the transconductance of the device either
directly or by adjusting the transconductance required for oscillation to begin,
both avenues will be discussed now.
3.8.1 Parallel Transistors
Several transistors may be connected in parallel such that the total terminal
currents are shared approximately equally between the devices. Each transistor
then has lower transconductance than if a single device is used. Consequently,
the impedance looking into each transistor’s emitter is greater than for an equiv-
alently biased single transistor. The new operating conditions, under which each
transistor has an decreased gm, are less likely to be sympathetic to oscillation
for a given cj than a similarly biased stage composed of a single transistor. The
ensemble transconductance and input impedance remain unchanged. The base
spreading resistance of each transistor appears in parallel, so the ensemble rb is
reduced. For every doubling of the number of paralleled transistors, the noise due
to the base spreading resistance will be reduced by√2. The shot noise processes
are unaffected.
50
CHAPTER 3
3.8.2 Increased base resistance
A small resistance may be connected in series with the transistor base lead. A
resistance in this location worsens the effect of gain peaking and increases the
stage noise. If the resistance is made too large, it will affect the frequency response
and frequency independent gain, k, of the stage. Observation of the time constant,
τ , and frequency independent gain of the transimpedance gain in (3.1) will show
this. The use of both mechanisms provides a reasonable compromise between
noise, stability, ease of construction and frequency response.
3.9 Physical Construction
The transistors T1−3 and T4−6 which are in SOT-23 packages are placed on top
of each other. Vertically placed wires connect their electrodes together. The
transistor closest to the circuit board is soldered to the copper tracks. No other
sufficiently compact layout has been found. This layout also provides a degree
of isothermal matching. The PCB is a standard double sided 1.6 mm FR4 glass
fibre type, coated with 1 oz copper on both sides. Several board designs were
developed in order to explore the layout options and noise reducing effect of par-
alleling transistors. Up to fifteen paralleled transistors were used in various layout
implementations. The resulting conclusion is that the distance between emitter
terminals in parallel transistor front ends must be as small as is practically possi-
ble when the impedance looking into the input is low. Any parasitic components
formed between the individual transistor terminals have deleterious effects on the
frequency response, rendering the layout unusable. Similar rules apply to the
distance between the APD and the first transistor stack.
3.10 Noise Analysis
Analytical, SPICE and experimental methods which complement each other are
used to assess the front end circuit noise. This section is split into two parts,
1. Analytical noise analysis and possible optimisations
51
CHAPTER 3
2. Comparison of the analytical model with SPICE
Analytical methods are used to isolate the dominant noise source in the tran-
sistor stages. A first order analytical model shown in Fig. 3.8 has been developed
by ignoring the base emitter capacitance, cbe, and the base collector capacitance,
ccb, and the load capacitance, CL. It will be shown in the second part of this
section that the error introduced by the simplifications does not significantly af-
fect the results at the frequencies of interest. The agreement between the SPICE
noise prediction and the measured noise of the circuit is shown in section 3.11.
Optimisation of the circuit noise by component selection is explored within the
limitations of the specifications listed in section 3.5.1.1.
The APD junction capacitance, cj , acts to increase the noise contribution of
the active transistor stage as shown in Fig. 3.22 which shows measured data.
Under normal operating conditions, only the active transistor stage’s noise con-
tribution is significant.
An introduction to noise analysis techniques can be found in Leach [53] and
Motchenbacher and Connelly [54].
3.10.1 Common Base Analytical Noise Analysis
The circuit noise is most critical in the transistor stage, the signal to noise ratio
is heavily dominated by this stage. To assess the scope for improving the noise
characteristics of first transistor stage, an analysis of the relative magnitude of the
contributing noise sources, and the transfer ratio of each source to the output node
– the noise gain – has been analytically derived. An analytic solution of the noise
transfer ratios is preferable as it permits examination of the circuit components
that contribute to the frequency independent gain and time constants of the
transfer ratios. This knowledge may be used to assist the selection of component
values which minimise the circuit noise without compromising the specifications
in Section 3.5.1.1. In order to make the task tractable the frequency dependence
of the transistor β and several inter-electrode capacitances are ignored.
The small-signal model used in this analysis is shown in Fig. 3.8. Table 3.1 lists
the noise generators and their associated physical sources and Table 3.2 lists the
expressions required to compute the magnitude of the noise sources. Equations
52
CHAPTER 3
Enrb
rbInb rbe gmvbe Inc
RL
EnRL
RE
EnRE
iin cj
Figure 3.8 – Small signal noise model of common base amplifier. Definitions of the terms aregiven in Table. 3.1. The magnitudes of the noise generators are given in Table. 3.2.
3.1 – 3.6 list the noise gain of each noise generator in the noise model. The noise
gain of each noise contributor can be expressed as either a low pass or a pole-zero
form, except for the contribution of the load resistor which appears directly at
the output.
In order to ensure that the analytical expressions (3.1 – 3.6) have been derived
without error they are compared with a SPICE .AC simulation of the small signal
model (Fig. 3.8). The result of this simulation is shown by the red lines in Fig. 3.9.
From this figure it is easily seen that the base spreading resistance thermal noise
has the highest noise gain as frequency increases. The small signal parameters
are given in the figure caption.
The noise contribution at the collector node for each source is shown in
Fig. 3.10 for three parallel transistors and for twenty five parallel transistors.
In the twenty five transistor case it is assumed that the transconductance of each
transistor will be sufficiently reduced that the internal base spreading resistance
of each transistor is sufficient to make oscillation unlikely. In the three transistor
case 50 Ω is added in series as shown in Fig. 3.6. The question of stability is dis-
cussed in fully in the next chapter. The bottom right graph in Fig. 3.10 shows the
comparison of each noise source and the total noise at the collector node for three
and twenty five transistors. The solid lines represent the three transistor case,
dashed lines represent the twenty five transistor case. Note that in the twenty
five transistor case the two dominant noise mechanisms, the collector – base shot
noise and the base spreading resistance thermal noise are practically equal, and
53
CHAPTER 3
Table 3.1 – Table of analytical noise model components.
Name Description Component of
Enrb Voltage noise sourceBase spreading resistance
rb Physical resistance
Inb Current noise sourceTransistor Base – Emitter
junction
Inc Current noise sourceTransistor Base – Collector
junctionRE Physical resistance
Emitter biasing resistanceEnRE Voltage noise sourceIin AC signal current
APD small signal modelcj Physical capacitance
gmvbeVoltage controlledcurrent source
The model oftransistor action
rbe Model resistanceRL Physical resistance
Load resistanceEnRL Voltage noise source
all of the other noise sources are considerably less influential than these two. This
is a preferable situation, as uncorrelated noise combine as the root of the sum
of the squared values. Therefore having the two most dominant sources equal
is a minimum condition. The choice of twenty five transistors as an example of
‘many’ was chosen specifically because it fulfils this condition for the particular
set of small signal parameters given in the figure caption. Unfortunately it is not
possible to use so many discrete transistors. The imperfections associated with
physical circuit manifestations prevent the circuit from operating as designed. In
an integrated circuit environment it may be possible.
It is possible to enhance the noise performance of the front end by selecting
specific component values rather than attempting to adjust the noise performance
by paralleling active devices. The effects of the load resistance noise, the emitter
biasing resistance noise and the noise due to the quiescent collector current are
described now.
54
CHAPTER 3
For the input current iin
vciin
= k · 1
1 + s · τ k =RL · rbe ·RE · gm
rbe + rb +RE + gm · rbe · RE
τ =cj · RE · (rbe + rb)
rbe + rb +RE + gm · rbe · RE
(3.1)
For the load resistance EnRL
vcEnRL
= 1 (3.2)
For the Base – Collector shot noise Inc
vcInc
= k · 1 + s · τ11 + s · τ2
k =RL · (RE + rbe + rb)
RE + gm · rbe · RE + rbe + rb
τ1 =cj · RE · (rbe + rb)
RE + rbe + rbτ2 =
cj · RE · (rbe + rb)
RE + gm · rbe · RE + rbe + rb(3.3)
For the Base – Emitter shot noise Inb
vcInb
= k · 1 + s · τ11 + s · τ2
k = − RL · rbe · gm · (rb + rbe)
gm · rbe · RE +RE + rbe + rb
τ1 =cj · RE · rbrb +RE
τ2 =cj · RE · (rbe + rb)
gm · rbe · RE +RE + rbe + rb(3.4)
For the base spreading resistance, Enrb
vcEnrb
= k · 1 + s · τ11 + s · τ2
k = − RL · gm · rbegm · rbe · RE +RE + rb + rbe
τ1 = cj · RE τ2 =cj · RE · (rbe + rb)
gm · rbe · RE +RE + rbe + rb(3.5)
For the emitter biasing resistance EnRE
vcEnRE
= k · 1
1 + s · τ k =RL · rbe · gm
RE + gm · rbe ·RE + rb + rbe
τ =cj · RE · (rb + rbe)
gm · rbe · RE +RE + rb + rbe(3.6)
55
CHAPTER 3
Figure 3.9 – Transfer ratios for each noise source in the analytical model. Top left shows thenoise gain for the base spreading resistance thermal noise. Top right, base – collector shot noise,middle left, base – emitter shot noise, middle right, emitter biasing resistor thermal noise andthe bottom graph shows the transimpedance gain. Black open circles represent (3.1 – 3.6). Redsolid lines represent the SPICE .AC analysis of the small signal equivalent circuit of Fig. 3.8 inorder to check that the analytical noise equations have been derived without error. The smallsignal parameters are, rb = 20 Ω, RE = 90 Ω, gm = 0.56 A/V, rbe = 244 Ω, cj = 1 nF andRL = 500 Ω.
56
CHAPTER 3
Figure 3.10 – Collector node noise voltage due to each noise generator. From top left tobottom right the graphs represent the base spreading resistance thermal noise, the base –collector shot noise, the base – emitter shot noise, the emitter biasing resistance thermal noise,the load resistance thermal noise and all of the individual sources with the total (the rootedsum of the squared contributions) output noise voltage also. Black solid lines represent a threetransistor front end with an additional base resistance of 50 Ω. Red solid lines represent theparalleling of twenty-five transistors. In the bottom right graph solid lines represent the threetransistor front end, dashed lines represent the twenty-five transistor front end.
57
CHAPTER 3
Table 3.2 – Table of noise source expressions.
Source Description Expression [ V2
Hz, A2
Hz]
EnrbThermal noise of the base
spreading resistor4 · k · T · rb
Inb Shot noise in the B–C junction 2 · e · ICInc Shot noise in the B–E junction 2 · e · IB
EnREThermal noise of the emitter
biasing resistor4 · k · T · RE
EnRLThermal noise of the load
resistor4 · k · T ·RL
Where k is Boltzmann’s constant, e is the electron charge and T is the absolutetemperature.
3.10.2 Quiescent Collector Current
The quiescent collector current is closely bound to the frequency response of
the amplifier. From a noise perspective, an optimum quiescent collector current
exists. It is interesting to find out how far from the optimum the circuit will
operate, and if any optimisation is possible. Fig. 3.11 shows the effect on NSNR
at 10 MHz of varying the quiescent collector current. The solid line represents
RE of 680 Ω, the dashed line RE of 100 Ω. From this graph it is evident that
an optimum current exists and that the amplifier may be designed such that the
optimum can be moved towards the value of collector current required to produce
the desired input impedance. Unfortunately reducing the value of RE reduces the
NSNR so while the amplifier can operate nearer to the optimum collector current
it would do so with worse overall performance.
3.10.3 Load Resistance
The case for increasing the load resistance is relatively simple, the larger the
load resistance, the more gain the stage has. NSNR should increase with RL.
Figure 3.12 shows the effect of increasing RL on the noise signal to noise ratio
while holding all other parameters constant. The effect of the load resistance is
not to increase the NSNR without limit however as the load resistance also has
a hand in setting the noise gain of all the noise generators and after a certain
58
CHAPTER 3
Figure 3.11 – NSNR of the front-end at 10 MHz with variation in quiescent collector currenthaving RE as a parameter. RE = 680 Ω (Solid line), RE = 100 Ω (Dashed).
frequency increasing the load resistance tends to increase the transimpedance gain
and the noise at the collector similarly such that there is no further improvement
in NSNR. The practicality of passing a quiescent current of 15 mA through 10 kΩ
while maintaining power supplies of ± 15 V is neglected as is the time constant
caused by the collector node capacitance and the load resistance.
3.10.4 Emitter Biasing Resistance
Increasing the emitter biasing resistance increases NSNR (with certain small sig-
nal parameters). This may be seen by observing (3.1 – 3.6). However it is not
possible to meaningfully improve the performance of the front-end under all situa-
tions. The upper left graph in Fig. 3.13 shows the effect on NSNR with frequency
as RE is increased from 100 Ω (dashed line) to 680 Ω (solid line) while cj is 1 nF.
At 10 MHz the benefit obtained at lower frequencies by increasing RE is lost
because the higher value of RE moves the pole down in frequency. Similarly the
upper right hand graph in Fig. 3.13 shows the collector node noise voltage due
to the base spreading resistance thermal noise, Enrb, as a function of frequency
where RE is a parameter. A zero moves down in frequency by an amount just
59
CHAPTER 3
Figure 3.12 – NSNR of the front-end at 10 MHz with variation in load resistance, RL.
sufficient to negate any advantage that was obtained at low frequencies. One
example where an improvement can be obtained is in the noise voltage appearing
at the collector due to the thermal noise of the emitter biasing resistance EnRE .
This is shown in the lower graph in Fig. 3.13 as a function of frequency where
RE is a parameter. The pole still moves down in frequency but does not interfere
with the measurement bandwidth while cj is 1 nF. This is little consolation how-
ever as observation of the upper left graph in Fig. 3.13 or the lower right hand
graph in Fig. 3.10 shows that the noise due to the emitter biasing resistance is
not dominant and so no meaningful improvement in NSNR is obtained.
3.10.5 APD Junction Capacitance
The APD junction capacitance forms a pole with a combination of the emitter
biasing resistance and the input resistance of the transistor. It is therefore ex-
pected that the NSNR will worsen with increasing cj . RE is always much bigger
than the input resistance but the two appear in parallel from the perspective of
the input signal. Therefore having a large value of RE does not materially worsen
the frequency response. Under conditions where cj is small (< 100 pF, black line
in Fig. 3.14) the large value of RE does produce a material benefit in NSNR,
60
CHAPTER 3
Figure 3.13 – Exploration of the possible noise optimisation of the emitter biasing resistance.The constant small signal parameters are, rb = 56.53 Ω, RE = 680 (solid), 100 (dashed) Ω,IC = 15 mA, β = 105.8, cj = 1 nF and RL = 500 Ω.
however under the majority of conditions for which the circuit has been designed
(cj > 100 pF) the value of RE does not greatly affect the overall noise perfor-
mance (see Fig. 3.13, upper left graph). The effect cj has on NSNR is shown in
Fig. 3.14.
3.10.6 Noise Analysis Summery
The following conclusions may be drawn from the noise analysis,
• Around 10 MHz, the dominant noise contributor is the thermal noise asso-
ciated with the transistor’s base spreading resistance and the external base
resistor, Enrb,. The collector – emitter shot noise, Inc, and collector–base
shot noise, Inb, are the second and third largest contributors respectively.
Uncorrelated sources sum as the root of the contributions squared; the base
61
CHAPTER 3
Figure 3.14 – Effect on NSNR of increasing cj , all other parameters held constant: rb =56.53 Ω, RE = 680 Ω, IC = 15 mA, β = 105.8, and RL = 500 Ω. cj = 10, 100, 500 pF, 1, 2, 3,4 and 5 nF.
spreading resistance thermal noise is responsible for almost all of the output
noise.
• Paralleling transistors reduces the base spreading resistance thermal noise
contribution. Noise can be reduced without limit using this method because
the amplifier noise is greater than the source noise [55]. The diode noise
signal to front-end noise ratio is often negative (Table. 2.1). The practical
problems of layout are the limiting factor in this respect.
• Increasing the value of the emitter biasing resistor, RE , acts to reduce the
noise gain of the emitter biasing resistor’s thermal noise, EnRE , to the
collector node.
• Decreasing the time constant of RE and cj will increase the frequency at
which the noise gain of the base spreading resistance, Enrb, the dominant
noise source, begins to rise. However, decreasing the value of RE in order
to move the zero up in frequency will increase the frequency independent
gain, k, for the EnRE generator by an amount which is sufficient to negate
any improvement in noise performance.
62
CHAPTER 3
Figure 3.15 – Comparison of analytical and SPICE noise prediction. The analytical model(circles) is a superposition of the transistor noise sources listed in Table 3.2. The SPICE analysisis the solid line. The small signal parameters for both analysis are IC = 15 mA, β = 105.8,rb = 19 Ω, RE = 90 Ω and cj = 1 nF.
In general the problem of noise analysis in this case is multifaceted, it is
likely that an optimum NSNR exists which is greater than that possessed by
the implementation shown in Fig. 3.6 but noise is not the only parameter of
importance so the final circuit will always be a compromise. This circuit has
been successfully used to gather the data presented later in this thesis. A full
optimisation of this circuit may be undertaken in the future by means of a genetic
algorithm or another suitable computerised process that operates by attempting
to minimise cost while maintaining certain parameters.
3.10.7 Comparison with SPICE
To assess the usefulness of the fist order analytical model over the frequency
range of interest, it was compared with a SPICE model using identical small
signal parameters. The comparison is given in Fig. 3.15. The deviation of the
analytical model from the SPICE model occurs between 50 MHz and 60 MHz.
The error introduced by has been investigated by adding components to the
analytical model and using it to produce numerical results. These results were
then compared with SPICE analyses. A purely analytical route quickly becomes
impractical due to the large number of terms appearing in the equations as the
63
CHAPTER 3
VS
50Ω
50Ω
33kΩ
33kΩ
33kΩ
33kΩ
33kΩ
33kΩ
33kΩ
cj
it
33kΩ
33kΩ
Figure 3.16 – Circuit approximation of an APD.
circuit becomes higher order. The high frequency discrepancy between SPICE
and the first order analytical model is principally due to the effect of the Collector
– Base junction capacitance, ccb. The load capacitance, CL, and the base emitter
junction capacitance, cbe, also play some part but their effect is less dominant in
this particular design. Since the noise measurement is performed at 10 MHz, the
first order approximation does not represent a significant loss of accuracy. The
mid-band error between the SPICE transistor model and the first order analytical
model is 84 pVHz−1/2 with the analytical model producing an over estimation of
noise.
3.11 TIA Characterisation
The circuit in Fig. 3.16 is used to facilitate direct measurement of the tran-
simpedance amplifier’s frequency response and to simulate the APD capacitance.
The impedance of the resistor chain is constant within 3 dB from DC to 20 MHz.
As frequency increases, the chain impedance falls as the resistors parasitic capac-
itance becomes more significant.
3.11.1 DC Linearity
The DC linearity of the front-end should be excellent over a wide range of to-
tal APD current. It may be presumed that this measurement system will be
required to measure small photo-current (and photo-generated noise) in the pres-
ence of much larger leakage or tunnelling currents especially in very thin struc-
tures and narrow band-gap materials. The DC linearity of the front-end is shown
in Fig. 3.17. The measurement system may not be usable at 10 MHz over the full
range of input currents shown. As DC current is injected into the input it acts
64
CHAPTER 3
Figure 3.17 – The DC transfer function of the front end, including the operational amplifierswhich follow the common base stage. This measurement is performed by connecting a precisioncontrolled DC current source to the input and monitoring the DC output voltage using a DMM.
to reduce the quiescent current of the input stage, increasing re. It is therefore
more taxing to measure a high capacitance, high dark current nip device than an
equivalent pin device. In principal a device could have the position of the anode
and cathode exchanged and the biasing voltage polarity reversed. This is trivial
when measuring a packaged device, diced wafers which present p-type material
closest to the back contact may benefit from packaging.
3.11.2 Frequency Response
SPICE simulation and experimental measurements shown in Fig. 3.18 in which cj
is stepped from 0 nF to 10 nF in 1 nF steps results in a resonance phenomenon in
the amplitude response, near to the noise measurement frequency. Grey et al. [52]
have noted the necessity of considering the effect of the base spreading resistance
rb when the quiescent conditions give rise to rbe ≈ rb. The input and output
poles are weakly interacting provided rbe >> rb (Appendix D shows this). While
rbe >> rb, the transistor’s internal base node is close ground potential. If rbe is
approximately equal to rb, the transistor’s internal base node, which joins rbe and
rb, will deviate significantly from ground, and energy transfer between the emitter
and base circuits will be possible. Further analysis of the frequency response and
65
CHAPTER 3
stability of the common base circuit when rbe ≈ rb is given in Chapter 4.
It is desirable to minimise the peaking effect caused by the change in cj . The
available circuit variables that impact the magnitude of the peaking include,
• Base spreading resistance, rb.
• Collector – Base depletion capacitance, ccb.
• Base – Emitter depletion capacitance, cbe.
• Base – Emitter resistance, rbe.
Unfortunately none of these variables can be altered sufficiently to make the peak-
ing effect negligible while maintaining the gain and bandwidth requirements. It
is neither possible nor desirable to make rb zero. All of the available parame-
ters are ultimately bound to the transistor’s quiescent conditions. In this design
problem, where rb is a significant fraction of rbe, the dependence of the frequency
response on the APD junction capacitance appears to be insoluble. All transis-
tors have inter-electrode capacitances and non-zero rb, and will exhibit peaking.
Any transistor with very low rb would be unsuitable because it would oscillate.
The necessity of rb to maintain stability will be described shortly. A calibration
procedure similar to that reported by Lau [27] is used to mitigate the peaking
effect.
3.11.3 Effective Noise Power Bandwidth
The effective noise power bandwidth is obtained by connection of Fig. 3.16 to the
front end and the use of a vector network analyser with a resistive power splitter to
provide appropriate impedance matching. cj in Fig. 3.16 is incremented through a
range of values and the transimpedance gain as a function of frequency is recorded
for each capacitance. This produces a graph similar to Fig. 3.19. The area under
each curve is integrated numerically and normalised to the centre frequency,
Beff(cj) =1
G20
∫ f2
f1
|G(f, cj)|2 df. (3.7)
66
CHAPTER 3
Figure 3.18 – Measured TIA frequency response showing the variation in relative responsewith increasing APD junction capacitance. cj = 0 nF to 5 nF in 0.5 nF steps. The black solidline represents no additional capacitance. The dashed upper dashed line represents 0.5 nF.
where Beff is the ENBW, G0 is the gain at the centre frequency, G(f, cj) is the
frequency and APD capacitance dependent TIA and filter gain. f1 and f2 are
placed sufficiently far from the centre frequency, f0, that practically all of the area
under the curve in Fig. 3.19 is integrated. The result of the integration is one
value of effective noise power bandwidth per test capacitance. These values can
be visualised by plotting ENBW versus test capacitance as shown in Fig. 3.20. A
regression of this data is possible such that any value of junction capacitance can
be corrected. In the case of very high capacitance diodes the junction capacitance
tends not to change appreciably over the range of voltages for which excess noise
is measured. In most cases a constant value of junction capacitance is sufficient
to correct the measurement bandwidth.
3.11.4 Shot Noise Measurement
In practice, once a measurement of multiplication as a function of bias, and excess
noise as a function of multiplication has been recorded the computation of excess
noise is performed using,
F =Napd
NshotM· Bapd
Bcal(3.8)
67
CHAPTER 3
Figure 3.19 – Magnitude response as a function of frequency where APD junction capacitanceis a parameter. The junction capacitance is ‘simulated’ by a 1206 SMT resistor. cj = 0 to 5 nFin 0.5 nF steps. Three sets of data for 5 nF are shown to check consistency, they all lay overeach other.
where F is the excess noise factor, Napd is the value (arbitrary units) represent-
ing the noise power generated by the diode at a particular multiplication, M .
Nshot is the shot noise power (arbitrary units) generated by the primary photo-
current. Bapd and Bcal are the effective noise power bandwidths of the APD and
the calibration detector at the specific value of M , which is a function of bias
voltage. It is necessary to measure the shot noise generated by a range of photo-
currents. The result of this measurement is shown in Fig. 3.21. All unity gain
detectors should produce the same result once correction is made for the junction
capacitance. When this result is plotted on log-log axes the expected gradient is
unity. This results from the shot process and is a reasonable indicator that the
measurement system is behaving as expected.
3.11.5 Measurement of Circuit Noise
In Section 3.10 it was shown that a SPICE simulation of the common base noise
agreed with the first order analytical model for a wide range of frequency. Mea-
surement of the circuit noise produces a similar result to the SPICE analysis thus
68
CHAPTER 3
Figure 3.20 – TIA effective noise power bandwidth as a function of device capacitance. Theeffective noise power bandwidth is used in the calibration process [27]. It is given by thenormalised area under the transfer response within the bandwidth of the filter: ENBW (cj) =1G2
0
·∫∞
0 |A(f, cj)|2 · df where A is the transimpedance gain at a frequency f with a particular
value of cj . A0 is the transimpedance gain at the centre frequency of the SBP-10.7+. In practicethis integration is performed numerically on data obtained by connecting the circuit of Fig. 3.16to the TIA and recording the transfer response using a vector network analyser.
supporting the validity of the SPICE analysis and the analytical analysis are
valid. A measurement of the circuit noise as a function of frequency with junc-
tion capacitance as a parameter is shown in Fig. 3.22. Details of the measurement
are given in the caption.
3.12 Voltage Amplifiers
The voltage amplifiers required to replicate the measurement system are similar
to those shown in Fig. 5.13. The objective of this amplification is to scale the
noise signal to a suitable value to generate an output from the noise power meter
with minimal ambiguity. The gain of each set of amplifiers is given in the text
which describes the block diagram (Fig. 3.5).
The noise signal amplitude is unpredictable. The instantaneous value could
lay in the range -∞ to ∞, however practical considerations particular to this
circuit limit the range to less than the power supply voltage. An assumption
69
CHAPTER 3
Figure 3.21 – Reference data from two Si pin devices.
must be made with respect to the noise voltage values entering the power meter
so that a gain which will make the noise voltage sufficiently large to be easily
detected can be selected. The assumption places a limit on the headroom in the
power meter. Because the NSNR of the system is negative the system noise is
sufficient for calculations. If the APD noise is large enough to make the NSNR
positive, a significant attenuation will already have been applied and the system
noise will have been attenuated accordingly. Setting the maximum permissible
input voltage to the power meter at 0.7 V RMS permits a crest factor of 2. The
system noise will exceed the power meter input range approximately 4.6% of the
time. It must be noted that the circuit noise of the front end increases so greatly
with cj that the 0 dB attenuation setting is used rarely and that as attenuation is
increased the system noise drops commensurately with the attenuation because
the dominant system noise generators are placed prior to the attenuator. The
10 dB attenuation setting yields a crest factor of 6.42 and the percentage of the
waveform clipping the power meter input is negligible.
Calibration measurements on the 0 dB attenuation, shown in Fig. 3.21, using
a BPX65 (cj ∼ 4 pF) photo-diode show output noise power increases linearly
with photo-current over a range of photo-currents from 0 to 50 µA indicating
that full shot noise has been measured. Further evidence for the acceptability of
70
CHAPTER 3
Figure 3.22 – Transistor TIA noise measured using a HP4396A spectrum analyser. The TIAis connected in series with the SBP-10.7+ Filter in order to show the extent of the pass band.cj = 0 nF to 5 nF in 0.5 nF steps. SPICE simulation results for the noise voltage at the outputare also shown. The insertion loss of the filter is not more than 1.5 dB [56]; the total errorbetween SPICE and the measured performance is 2.5 dBV -approximately 13% - with SPICEunderestimating the noise.
a low crest factor is the agreement of the data gathered using this system and
those gathered using the 1 MHz system reported in Chapter 5. These data are
presented in Chapter 6. Yet more evidence in favour of proper operation is the
change in output noise voltage by one order of magnitude when the attenuator
is increased by 10 dB (the attenuator is graduated in dBV), and the squaring
function causes this to be converted into dBW.
The linearity of the system up to the point of the power meter input is exam-
ined by measuring the noise voltage using a spectrum analyser while driving the
system with a pseudo-random noise generator or by connecting and biasing an
APD into avalanche multiplication (dark or photo-generated current is accept-
able). Harmonic distortion manifests itself as a band of higher-than-background
noise around integer multiples of 10 MHz. Harmonic distortion is problematic
in this situation despite laying outside the filter pass-band because the power
meter is wide band (250 MHz) and so will act to square the unwanted extra noise
voltage caused by the distortion. Higher amplifications were used to elicit this
effect so that it could be observed. Clipping of the system noise on the 0 dB
71
CHAPTER 3
attenuation setting is not present in the reported design. It is however possible
to saturate the power meter by improper choice of attenuation.
Saturation (if it is not obvious) may be detected while performing noise mea-
surements by adding a further 10 dB attenuation and observing the power meter
output. The power meter output must reduce by an order of magnitude, if it does
not, the experiment must return to a lower photo-current (by reducing the bias
on the APD) where the required effect is observed. A proper attenuation should
then be selected and the measurement continued from this value of multiplication.
All previous higher multiplication and noise data should be discarded.
3.13 Noise Power Meter
The noise power meter used in this design is due to B. K. Ng and is described
fully in Chapter 5.
3.14 Conclusion
This chapter has presented a new transimpedance amplifier for use in measure-
ments of excess noise in avalanche photo-diodes. This measurement system has
found use in situations where the device being measured has high junction capac-
itance, up to 5 nF, and where the device has lower than usual dynamic resistance,
rd. It has been used to collect half of the data shown in Chapter 6. The measure-
ment of a much wider range of devices is now possible. This circuit design may
be used to perform measurements in modes of APD operation which increase
the capacitance and decrease the dynamic resistance such as forward bias and
photovoltaic mode. Noise measurements may yield novel methods of grading the
quality of solar cells.
72
Chapter 4
Small Signal Stability in
Common Base Amplifiers
During the design and testing of the circuitry described in Chapter 3 a persis-
tent oscillation was observed. After the circuitry was corrected, by the addition
of a judiciously placed resistance, the nature of the oscillation was unresolved.
That is to say, it was unclear which component or set of components was to
blame. Furthermore the method by which the oscillation was resolved could not
be sufficiently explained without a lot of ‘arm waving’. The analysis presented
in this chapter aims to set out why some single transistor amplifiers oscillate and
why some single transistor oscillators amplify! This chapter concentrates on the
common base connection of bipolar devices, however, similar arguments apply to
other three terminal amplifying devices and in other connections, with suitable
mathematical adjustments.
Nomenclature
In this chapter small signal parameters are lower case. Large signal parameters,
or parameters which exist in large signal and small signal scenarios, are upper
case.
IC The quiescent collector current of a transistor.
gm The conductance (transconductance) which partially models the operation
73
CHAPTER 4
of transistor. It is related to the quiescent collector current by gm =
(e IC)/(k T ) in a BJT and by gm ∝√ID in a field effect transistor un-
der strong inversion conditions.
rbe The dynamic resistance which partially models the operation of a bipolar
transistor.
cbe The capacitance which models the frequency dependence of the transistor
operation.
rb The base spreading resistance of a transistor. Both extrinsic and intrinsic
parts are lumped.
ccb The capacitance of the reverse biased collector - base junction. It is assumed
that all of this capacitance is connected to the internal base node; that is,
at the node connecting rbe and rb.
RL The load resistance; generally any resistance in the collector circuit across
which the output voltage is developed.
CL The load capacitance; generally any parasitic capacitance to ground asso-
ciated with the collector node or the circuits which load the collector.
RE The emitter biasing resistance.
CE The capacitance to ground associated with the emitter node.
LP The parasitic series inductance in the base network.
β The frequency dependent small signal relationship between ic and ib in a
bipolar transistor.
β0 The frequency independent small signal relationship between ic and ib in a
bipolar transistor.
α The frequency dependent small signal relationship between ie and ic.
α0 The frequency independent small signal relationship between ie and ic.
74
CHAPTER 4
τT The time constant associated with the unity current gain frequency, fT .
This parameter could be called τα as it is also the -3 dB point of α.
τβ The time constant associated with the -3 dB point of β.
τE The time constant of CE and RE .
τC The time constant of the load resistance, RL and the load capacitance, CL.
τL The time constant of Lp and rb.
4.1 Introduction
Several reports of high frequency stability limitations in common collector circuits
exist including [57–59]. There is comparatively less reported analysis on the
common base connection [60, 61]. This chapter considers the stability of small
signal common base bipolar transistor amplifiers in the frequency domain. This
chapter also views an unstable amplifier as a potential oscillator, and provides
a method of determining the conditions for stability by assessing the conditions
necessary for instability. Some practical guidance on the physical implementation
of video bandwidth circuits is also provided. This information is often neglected
in formal publications, or is simply addressed as ‘high frequency techniques’ with
little explanation of what is required. A particular HF technique which is well
known, but not often analysed mathematically, is the addition of base or gate
stopper resistors. Section 4.9 provides an illuminating treatment.
The arguments in this chapter are presented using a bipolar transistor because
the impetus for the work was found in a bipolar device; similar arguments apply
to FETs and comparable devices.
The hybrid-π model of the bipolar transistor is used and the usual limitation of
validity (fT )(2) is assumed throughout [62]. This chapter is divided into sections;
each section contains analysis which builds on the prior section such that the
modification to the performance of the circuit due to each addition is evident. For
each component addition the conditions sufficient for negative resistance looking
into the emitter are developed. The boundary conditions necessary for stability
are explored analytically and in terms of design rules.
75
CHAPTER 4
iin
gmvbe
RL
cberbe
rb
Figure 4.1 – Common base amplifier with base spreading resistance and frequency dependence.
Oscillation may occur in a single transistor circuit when a number of electrodes
exhibit a negative resistance over a range of frequencies. Negative resistance
may be generated due to the impedance transforming property of the transistor.
A suitable parasitic, or intentionally designed, resonant circuit must exist to
permit oscillation and to set the oscillation frequency. The oscillation described
is assumed to be unwanted or parasitic; however, several popular circuits operate
by the negative resistance method [63–67]. The negative resistance arises partly
due to passive circuit elements and partly due to the transistors intrinsic frequency
response. Analysis begins with analysis of the intrinsic frequency response of a
transistor in the common base connection.
4.2 Frequency Dependent Behaviour of a
Common Base Amplifier
The circuit of Fig. 4.1 shows a partial hybrid-π small signal equivalent circuit of a
common base transimpedance amplifier. It can be shown that a transistor in the
common base connection possesses limited bandwidth and that the limitation is
intrinsic to the transistor.
There is often a preference for the use of current gain, β, or transconductance,
gm, when performing analysis on small signal models of bipolar transistors. An
argument may be made in favour of using transconductance because it is common
to all three terminal amplifying devices. In the case of common base circuits
however, the use of α and β often produces a more elegant expression. The
equivalence of analysis using current gain and transconductance can be shown.
76
CHAPTER 4
For the transimpedance gain of 4.1, solutions are given using current gain and
transconductance. The transimpedance of the circuit in Fig. 4.1 is given by,
vciin
=β0RL
β0 + 1· 1
1 + s τT α0(4.1)
An equivalent expression without invoking α or β is,
vciin
=gm rbe RL
gm rbe + 1· 1
1 + s τT gm rbegm rbe+1
(4.2)
The impedance looking into the emitter is,
zin =rbe + rbβ0 + 1
·1 +
s τβ rbrbe+rb
1 + s τT α0(4.3)
Equation 4.3 is a first order pole – zero standard form. The zero occurs at a
lower frequency than the pole; consequently, a range of frequencies exists where
the phase shift increases from 0 at lower frequencies towards +90 before re-
turning to 0. It is instructive to observe graphically the real part of zin and the
phase of zin (Fig. 4.2) using some sensible values. The real part shows that at low
frequencies zin is given by rbe+rbβ0+1
and as frequency increases zin → rb. The phase
of zin illuminates the impedance transforming nature of the transistor. A range
of frequencies exists where the phase of zin increases from zero, reaches a maxi-
mum, and returns to zero at high frequencies. In this region the base spreading
resistance appears partially or wholly inductive when viewed from the emitter of
the transistor. It will be shown shortly that series inductance in the base network
is transformed – over a certain range of frequencies – into a negative resistance.
Similar arguments follow for capacitance transformation.
4.3 Addition of Parasitic Inductance in the Base
Network
It is assumed that the base lead possesses parasitic inductance, LP . The induc-
tance may arise intentionally by design or be formed as part of a ground plane.
77
CHAPTER 4
Figure 4.2 – Black, real part of zin, red, phase of zin, both from Fig. 4.1. Small signalparameters, rbe = 172.5 Ω, cbe = 139 pF, β0 = 100, rb = 39.66 Ω.
It may be found as a parasitic in another component such as a decoupling capac-
itor. The multiplicity of ways in which the parasitic inductance can be added to
the circuit can lead to difficulty in finding a circuit board layout and component
technology which minimises the inductance. The circuit of Fig. 4.3 differs only
slightly from that of Fig. 4.1, it may be shown that the transimpedance gain of
this amplifier is identical to that of the amplifier in Fig. 4.1. The base spreading
resistance and the base inductance form a series LR network with a time constant
τL. The impedance looking into the emitter is given by,
zin =rbe + rbβ0 + 1
·1 +
s τL rb+s τβ rbrbe+rb
+s2 τβ τL rbrbe+rb
1 + s τT α0(4.4)
iin
gmvbe
RL
cberbe
rb
LP
Figure 4.3 – Common base amplifier with base spreading resistance, frequency dependent andbase inductance.
78
CHAPTER 4
Inserting s = j ω and considering only the real parts,
ℜ(zin) =rbe + rbβ0 + 1
·
(
− τβ τL rbrbe+rb
+ rb τT α0 τLrbe+rb
+ τb rb τT α0
rbe+rb
)
ω2 + 1
1 + ω2 τT 2 α02
(4.5)
A frequency independent input resistance may be obtained if the frequency de-
pendent terms of the real part are arranged such that they cancel. The value of
τL required to fulfil this condition is,
τL =τT α0 (τT α0 rbe + rb τT α0 − τβ rb)
rb (−τβ + τT α0)(4.6)
While this condition is satisfied the input impedance takes the form,
zin = k · (1 + s τ) (4.7)
k =rbe + rbβ0 + 1
(4.8)
τ =τβ (τT α0 rbe + rb τT α0 − τβ rb)
(−τβ + τT α0) (rbe + rb)(4.9)
This circuit cannot be realised easily because there are unavoidable parasitic
elements, such as signal source capacitance, missing from Fig. 4.3, which modify
the circuit behaviour. Furthermore, the input resistance is quite sensitive to
changes in the value of τL; obtaining the precise value from a distributed set of
parasitic elements may be difficult.
A second value of τL which is of interest is the value which is just sufficient to
make the input resistance fall to zero as frequency tends to infinity. Setting (4.5)
to zero and solving in the limit ω → ∞ yields,
τL = − τβ τT α0
−τβ + τT α0
= τT (4.10)
In this circuit, where the transistor biasing current flows through the signal source
to ground, the condition for a range of frequencies over which the input resistance
is negative is τL > τT . If τL = τT , ℜ(zin) → 0 as f → ∞.
79
CHAPTER 4
iin RE
gmvbe
RL
cberbe
rb
LP
Figure 4.4 – Common base amplifier with base spreading resistance, frequency dependants,base inductance and emitter biasing.
4.4 The Addition of an Emitter Biasing Resistor
It is sometimes impractical or undesirable to have the transistor biasing current
flowing through the signal source. It is common to bias the transistor with other
components and employ level shifting, the simplest form thereof is a blocking
capacitor between the signal source and the emitter electrode. This is shown in
Fig. 4.4; the blocking capacitor is considered short circuit at all frequencies of
interest so in the small signal model RE is connected directly to the emitter of
the transistor.
The addition of RE significantly complicates the problem of finding zin, how-
ever it is still possible analytically. In the prior circuit of Fig. 4.3 if the base
network inductance is sufficiently large there will be a frequency at which the
input resistance falls below zero; at higher frequencies the input resistance re-
mains less than zero. In Fig. 4.4 a bounded range of values of τL exist that yield
a negative input resistance over a bounded range of frequencies. It is impossible
to produce a constant input resistance with this circuit. Analysis to show this
falls along similar lines to the prior circuit. The frequency dependent terms can
not be set such that they cancel and consequently no value of τL will provide
a frequency independent input resistance. The range of frequencies over which
a negative input resistance exists is of interest, as are the boundary conditions.
This analysis also falls along similar lines as that given for Fig. 4.3. An expression
is obtained for the real part of the impedance looking into the emitter (4.11); this
is set to zero and solved for τL (4.12).
80
CHAPTER 4
ℜ(zin) =(−τβ rb ω
2 τL + rbe + rb)RE (−τβ rb ω2 τL +RE + rbe + rb + β0RE)
(−τβ rb ω2 τL +RE + rbe + rb + β0RE)2 + (ωRE τβ + ω τβ rb + ω rb τL)
2
+(ω rb τL + ω τβ rb)RE (ωRE τβ + ω τβ rb + ω rb τL)
(−τβ rb ω2 τL +RE + rbe + rb + β0RE)2 + (ωRE τβ + ω τβ rb + ω rb τL)
2
(4.11)
τL =1
2 rb· τβ
(
2 rbe + β0RE − 2A± B
A
)
(4.12)
where,
A = ±((rbe + rb) (rb + rbe + β0RE +RE)
) 12 (4.13)
and,
B =((−4 (β0RE + 2 rbe)A
2
+A(β0
2RE2 + 4RE (2 rbe + rb)β0 + 4 rbe (2 rbe +RE + 2 rb)
)) 12 (4.14)
The denominator of (4.11) is a quartic polynomial in ω, it has four roots. Two
roots are complex and may be ignored, the other two are real. Setting the two
real roots equal to each other and solving the resulting expression for τL yields the
boundary conditions. The boundary conditions are found where the real part of
zin is a minimum at zero ohms. A set of values of τL is sought which satisfy both
zin = 0 and d(zin)/df = 0. This is mathematically analogous to setting the two
real roots at the same frequency. As the two roots are set at the same frequency
only one value is sought, but two values of τL are solutions because the original
expression for ℜ(zin) is quartic. For a certain set of small signal parameters,
there are two frequencies at which ℜ(zin) has only one root; a different value of
τL yields each of them. Values of τL between these boundaries will yield two
roots. In the case of two roots, the region of frequencies between the roots will
be where the real part of zin is negative. Two complex boundaries exist but they
are ignored. The boundaries are given by (4.12). Note that the sign acting on
the root in the expression for ‘A’ must be the same for both occurrences of ‘A’ in
81
CHAPTER 4
the main expression. Fig. 4.5 shows a plot of frequency versus ℜ(zin) for each of
the solution conditions using some example small signal parameters. The small
signal parameters used to produce the plot are given in the caption. The yellow
(dash dot) line represents LP ≈ 1.5 mH and is a real root. The blue (long dash)
line represents LP ≈ 11.44 nH and is the other real root. The red (dash) and
green (solid) lines represent complex roots. The value of τL that represents the
blue (long dash) line is the minimum τL that will result in instability. The yellow
(dash dot) line is the maximum τL which maintains instability. The value of τL
produces zero or more – a maximum of two – real roots of zin. If more than zero
real roots exist, the amplifier is potentially unstable. The value of the inductor,
LP , is not especially significant, but the value of the time constant of the base
spreading resistance and the inductor, τL, is important.
The problem may be approached from another perspective by finding a range
of values of gm for which the amplifier is stable or unstable for a given rb and LP .
gm may be related to the base network through β0 and rbe. A logical extension
is then to use gm = (e IC) / (k T ) to obtain a range of values of IC for which the
amplifier is stable with a given LP and rb. It is difficult to estimate values for
LP and to a lesser degree rb, several methods of measuring rb exist [68, 69]. It is
therefore preferable to estimate a worst case scenario and ensure that the circuit
design parameters that are easily controlled will not result in instability. The
transimpedance of Fig. 4.4 exhibits a second order low pass network which may
be found using standard circuit analysis.
4.5 Addition of Emitter Capacitance
The addition of capacitance at the emitter – the signal source capacitance –
further complicates the problem but it is still analytically tractable. By using
methods identical to those used in the prior section it may be shown that the
new time constant τE = CE RE has no effect on the range of frequencies over
which ℜ(zin) is negative given a certain value of τL and a particular set of small
signal parameters. The extension to the method used previously is to note that
τE is absent from the expression for the solution of the roots of the real part
of zin and so cannot influence the boundary conditions. A graphical example is
82
CHAPTER 4
Figure 4.5 – Plots of ℜ(zin) showing the values of τL that satisfy the exactly one real rootcondition. The red (dash) and green (solid) lines represent the values of τL that produce rootsof ℜ(zin) in jω. The yellow (dash dot) and blue (long dash) lines are the upper and lowerbounds of τL that produce roots of ℜ(zin) in ω. Note that both the blue (long dash) andyellow (dash dot) lines have only one root. The violet (dash space) line is an example valueof τL which lies between the blue (long dash) and yellow (dash dot) values. It has two rootsand the region of frequencies where negative input resistance exists is not limited to layingbetween the frequencies that satisfy the exactly one real root condition. To generate this figurerbe = 172.5 Ω, cbe = 139 pF, β0 = 100, RE = 680 Ω, rb = 39.66 Ω, Lp = 15 nH.
shown in Fig. 4.7, using similar small signal parameters as Fig. 4.5, which shows
that a wide range of τE has no effect on the boundary conditions. Note that all
of the lines fall through the axis at the same frequency; τE does not affect the
range of frequencies having negative input resistance. The value of CE plays a
part in tuning the frequency of oscillation in a poorly designed amplifier. The
transimpedance of the circuit in Fig. 4.6 is third order low pass and may be found
using standard circuit analysis.
4.6 Addition of Load Capacitance and Collector
- Base Junction Capacitance
Unfortunately the final example, shown in Fig. 4.8, is not analytically tractable.
Therefore Maple, a symbolic/numerical mathematics package, has been used. It
can be shown indirectly that the circuit of Fig. 4.8 has a range of frequencies over
83
CHAPTER 4
iin RE
gmvbe
RL
CE
cberbe
rb
LP
Figure 4.6 – Common base amplifier with base spreading resistance, frequency dependence,base inductance, emitter biasing and emitter capacitance.
which the real part of the input resistance is negative, and that the boundaries of
this region are still independent of τE . A numerical solution for the roots of the
input resistance can be found without requiring any information regarding the
value of τE ; therefore one may conclude that τE does not appear in the expressions
being evaluated. If τE did appear in the expressions it would also appear in the
solution, unevaluated. ccb and τC – the time constant of RL and CL – do affect
the range of frequencies over which ℜ(zin) is negative for a given τL.
A general rule for assessing the likelihood of oscillation is if τL > τT , oscillation
is likely but not certain. Observation of the real part of the impedance looking
into the emitter using a circuit simulator such as SPICE can give an indication
of the likelihood of oscillation, assuming the simulation models are accurate. The
Figure 4.7 – Plot of ℜ(zin) with frequency showing the independence of the roots of ℜ(zin)from τE . At high frequencies all lines tend to zero ohms. rbe = 172.5 Ω, cbe = 139 pF, β0 = 100,RE = 680 Ω, rb = 39.66 Ω, LP = 15 nH, CE = 0.01 pF to 100 nF in decade steps.
84
CHAPTER 4
iin RE
gmvbe
RL
CE
cberbe
rb
LP
ccb
CL
Figure 4.8 – Common base amplifier hybrid–π model neglecting Early effect.
transimpedance expression for Fig. 4.8 has five poles and four zeros.
4.7 Colpitts Oscillator
Negative resistance oscillators are found in electronic systems serving a multi-
tude of purposes. A popular negative resistance oscillator is the circuit after
Colpitts [67,70,71]. The common collector Colpitts oscillator bares strong resem-
blance to the amplifier circuits considered in this chapter. Fig. 4.9 is nearly a
standard form of the Colpitts oscillator. The base emitter diffusion capacitance
is lumped with an external capacitance into C1. C2 is the source capacitance,
RE is the emitter biasing resistance, R2 and R3 bias the transistor and LP is the
parasitic base inductance. RB is a ‘base stopper’ resistor. C3 grounds the base
for AC signals. The introduction of the collector load resistor to the oscillator
is not significant provided ccb is small. If ccb is sufficiently large that significant
coupling exists between the base and collector at the frequencies of interest then
increasing RL tends to increase the impedance looking into the emitter eventually
making it positive. For this reason a true common collector Colpitts oscillator has
no intentional elements in its collector circuit. In this chapter it is assumed that
RL is relatively low valued such that τC does not limit the high frequency per-
formance. A compatible assumption is that RL is insufficiently large to prevent
ℜ(zin) becoming negative.
85
CHAPTER 4
C3 R2
LP RB C1
C2 RE iin
RLR1
V s
Figure 4.9 – Common collector Colpitts oscillator with collector load resistance.
4.8 Analysis of the Common Base Amplifier from
an Oscillator’s Perspective
Up to this point it has been assumed that the desirable circuit function is ampli-
fication. If however we treat the common base amplifier as a common collector
Colpitts oscillator and allow the collector load resistor to remain, the circuit has
two modes of operation. The stability of the amplifier may be discovered by
considering its merits as an oscillator.
Ignoring ccb and CL, the circuit of Fig. 4.9 has the small signal representation
given in Fig. 4.6. The circuit may be analysed to find the frequency of oscillation
as a function of the circuit parameters. A further value of interest is the minimum
transconductance required for oscillation to begin.
4.8.1 Oscillation Frequency
The method of analysis can be found in a standard text such as Razavi [72]. The
principle is to provide an initial excitation and find the gain due to the excitation.
The excitation may take the form of a voltage or current source; the placement
of the source must be such that when it is removed the network is not modified.
For oscillation to begin a frequency must exist where both the real and imaginary
parts of the denominator of the transimpedance function fall to zero.
86
CHAPTER 4
The real part of the denominator of the transimpedance function of Fig. 4.6
is given by,
(−rb β0 cbe RE CE − Lpβ0 cbe − Lpgm RE CE )ω2
+ β0 + rb gm + RE gm + β0 gm RE (4.15)
The imaginary part is given by,
− ω3Lpβ0 cbe RE CE
+ (rb gm RE CE + β0 RE CE + rb β0 cbe + RE β0 cbe + Lpgm)ω (4.16)
Setting the imaginary part (4.16) to zero and solving yields,
ω2 =rb gm RE CE + β0 RE CE + rb β0 cbe + RE β0 cbe + Lpgm
Lpβ0 cbe RE CE
(4.17)
It is possible to find the value of a particular component which will yield the min-
imum transconductance required for oscillation. This is accomplished by substi-
tuting (4.17) into the real part (4.15) and differentiating the resulting expression
with respect to the component in question. For example, to find the value of CE
that requires the minimum transconductance for oscillation – all other parame-
ters being defined and held constant –substitute ω2 into the real part (4.15) and
differentiate with respect to CE. Setting the result to zero and solving for CE
yields,
C2E =
Lpβ0 cbe (rb β0 cbe + RE β0 cbe + Lpgm)
RE2(rb2β0 cbe gm + rb β0
2cbe + Lpgm2rb + Lpβ0 gm) (4.18)
For a particular set of small signal parameters the value of CE given by (4.18)
will yield oscillation with the minimum device transconductance. The frequency
of oscillation for this, minimum transconductance, value of CE is found by back
substitution into (4.17).
87
CHAPTER 4
Figure 4.10 – Plot showing the value of gm required for oscillation as a function of CE forseveral values of rb = 1 mΩ (red, dot), 0.1 Ω (green, dash), 1 Ω (violet, dot dash), 10 Ω (darkred, long dash), 39.66 Ω (blue, solid) and 100 Ω (brown, dash space). cbe is constant throughout,small signal parameters are given in Fig. 4.5, LP = 20 nH.
4.9 The Effectiveness of Increasing the
Base Resistance
A common method of stabilising an oscillating amplifier is the addition of a
resistance which increases the base spreading resistance, rb. In Fig. 4.9 this
extra base resistance is labelled RB. It is illuminating to plot the minimum
transconductance as a function of an example circuit parameter, CE , as rb –
which is now taken to be the sum of rb and RB – varies. Unfortunately, cbe
is linked to gm and the interaction between them clouds the picture, therefore
assume that the oscillator designer adds an extra capacitance in parallel with
cbe such that the combination of the two, labelled C1 in Fig. 4.9, does not vary
significantly with gm; this is common practice. The effect of increasing rb on the
necessary gm required for oscillation is shown in Fig. 4.10; small signal parameters
are given in the caption.
88
CHAPTER 4
4.10 Practical Methods of Assessing Oscillation
A spectrum analyser provides a productive method of assessing oscillation on the
test bench. A free object possessing capacitance to ground, for example a hand,
is also useful. The circuit under observation is connected to the analyser. The
background signals should be recorded prior to applying power. The circuit may
not oscillate immediately, particularly if it is cold. Oscillation in the RF region
will often result in a waveform consisting of several harmonics. As the active
device(s) increase in temperature from cold to the operating temperature a shift
in oscillation frequency may be observed. Once the oscillation frequency has sta-
bilised it is sometimes possible to lower the frequency of oscillation by bringing an
object close to the circuit. This principle underpins several musical instruments
including the Theramin [73]. The emitter node is particularly sensitive in the
common base amplifier configuration. The magnitude of the frequency variation
is dependent on the magnitude of the additional capacitance to ground formed
by the object with respect to the capacitances already present in the circuit.
It most circuit designs a common base amplifier will not drive a transmission
line directly. The output impedance of a common base amplifier is approximately
equal to RL so a buffer amplifier is often used. If the buffer has insufficient band-
width to reproduce the oscillation frequency, it will be difficult to observe it
directly using an analyser. Assuming that the buffer amplifiers have insufficient
bandwidth, their action on the oscillation signal is likely to be rectification. Con-
sequently, observing a baseband output that has a large change in DC offset when
an object possessing capacitance to ground is brought near to the RF common
base stage is suggestive of oscillation. In such situations near field probes may
be used with an analyser to observe the oscillation. Note that any wide band
active circuit’s DC offset will rise to some degree when an object is brought near
because the object acts as an antenna and background signals are capacitively
coupled into the amplifier. In the case of oscillation however the DC shift is
likely to be more pronounced as the amplitude of the oscillation will change as
the circuit capacitances increase. One may also find that quiescent DC voltages
in the RF amplifier are unusually sensitive to foreign objects and physical dis-
turbance. ‘Wiggling’ the DC power input connector on the circuit board a little
89
CHAPTER 4
may cause significant changes in quiescent conditions for a short period. If this
kind of behaviour is observed oscillations should be sought. It is unfortunately
more difficult to show that a circuit does not oscillate than to show the opposite.
4.11 High Frequency Techniques
Practical methods regarding the physical requirements of high frequency circuits
which function as designed are often neglected or summarised as ‘high frequency
techniques’. High speed transistor circuits, especially those run at higher quies-
cent currents and so larger transconductances, are more susceptible to oscillation
when inappropriate board layout and construction is employed. Below are the
points that have been found most useful in helping to achieve a stable amplifier.
• Minimise stray inductance and capacitance in the layout. Reduce the board
area taken up by RF stages. Increase track widths; this decreases self
inductance of the track. Increase distance between track and ground plane,
this decreases capacitance to ground. Only run long traces which are driven
by low impedances.
• Avoid ground planes under high speed transistor amplifiers and operational
amplifier stages. This decreases capacitance to ground.
• Ensure microwave transistor emitters are well connected; connect all the
emitter tags. This reduces the resistance and inductance in series with the
emitter.
• When decoupling nodes to ground use a range of capacitances. For exam-
ple 220 µF and 220 nF and 100 pF in parallel. Ensure that the smallest
value capacitance is closest to the node. Aluminium electrolytic capacitors
become quite inductive as frequency increases.
• Connect top and bottom ground planes together at regular intervals. Con-
nections should be spaced less than 110
th of the wavelength that corresponds
to the unity power gain frequency of the active devices.
90
CHAPTER 4
• Ground planes must not be permitted to appear as transmission lines be-
cause there will be no common ground reference for the circuits on the
board. The transistor in common base mode views the signal source with
respect to the base node, not with respect to wherever the signal source
ground enters the circuit board, or any other location in between the two.
If the ground plane possesses significant parasitic inductance over the dis-
tance between the location of the signal source ground and the location of
the transistor base ground the circuit stability will be compromised.
• Decouple all power supplies at the terminals of all active devices and circuit
stages.
Sometimes ensuring the layout is well conceived is not sufficient to prevent
oscillation. In these cases extra components must be added to: 1) make the
impedances larger at high frequencies, 2) damp resonances or 3) increase the
transconductance necessary for oscillation. Many methods of achieving these
aims exist, but the additional general guidance below has proved useful.
• Use extra base resistance sparingly, increasing the base resistance reduces
gain, reduces bandwidth and carries a high noise penalty [74]. If frequencies
are sufficiently low, a ferrite bead may be used instead.
• Use several active devices in parallel to lower the individual transconduc-
tance, but maintain the ensemble transconductance. This carries a noise
benefit; multiple devices may be used in conjunction with an extra base
resistance such that the noise penalty of the extra resistance is partially
mitigated [74].
• In some cases a Zobel network (image filter) is a viable alternative to in-
creasing the base spreading resistance without suffering the same speed,
gain or noise penalty. For example a Zobel network connected from the
gate to drain of a power FET operating as a source follower can prevent
the impedance looking into the gate becoming negative at high frequencies,
due to a capacitive load on the source.
91
CHAPTER 4
4.12 Conclusion
Analytical results have been developed for a series of progressively more complex
common base small signal amplifier models which address the nature of, and
conditions necessary for, instability. The common base amplifier has been treated
as a Colpitts oscillator and the conditions necessary for oscillation established by
a second method. The common practice of using a ‘base stopper’ resistor has
been explored. A general design rule for avoiding oscillation (τL < τT ) has been
given. Practical approaches to observing oscillation and recovering an oscillating
amplifier have been discussed.
92
Chapter 5
1 MHz Opamp Based High
Capacitance Noise Measurement
System
This chapter reports a second versatile system for measuring excess noise and
multiplication in avalanche photo-diodes, using an operational amplifier based
transimpedance amplifier (TIA) front-end and operating by phase-sensitive de-
tection, which facilitates accurate measurement of multiplication and excess noise
in the presence of a high dark currents. The system can reliably measure the ex-
cess noise factor of devices with capacitance up to 5 nF at a centre frequency of
1 MHz. This system has recently been used to measure the thin, large area Si
pin avalanche photo-diodes which are reported in Chapter 6. The data obtained
from this measurement system is in good agreement with that obtained from the
10 MHz system described in Chapter 3.
5.1 Introduction
The problems of measuring the noise associated with large capacitance photo-
diodes, or photodiodes that for some reason must be connected to the TIA by
long cable lengths, have been discussed in detail at the start of Chapter 3. In
Chapter 3 a discrete TIA design was described with special attention being given
93
CHAPTER 5
Chopper Driver
Scitec Inst. 300C
Photo-currentLock-in-amplifier
Stanford SR830
Reference Signal
Excess NioseLock-in-amplifier
Stanford SR830
Laser
He-Ne
2.5mW, 633nm
Chopper
Bias Source
Keithley 236
TIAVoltage Amp
Minicircuits
ZFL-500LN+ HP 355D
f0 = 1 MHz
Friend, N = 8
Voltage Amp
ZFL-500LN+
Power Meter
VoltageAmplifier
Figure 5.1 – A system diagram of the 1 MHz operational amplifier based noise measurementsystem.
to the trade-off between, input impedance, noise, bandwidth and stability. In
this chapter a noise measurement system which has similar goals to the system
described in Chapter 3 but having differing specifications is described. The objec-
tive of this measurement system is to corroborate the data presented in Chapter 3
by excluding the possibility of measuring 1/f effects and RC limiting. By relin-
quishing an order of magnitude in frequency range the design problem becomes
much more tractable. Operation amplifiers are now viable front end devices. This
measurement system also benefits from better system noise performance then the
transistorised system under low capacitance conditions.
5.2 Noise Measurement System
The principle difference between this system and Li’s system [20] is the change
in the measurement bandwidth and centre frequency. The open loop gain of
the AD9631 at 10 MHz is approximately 12. It can be shown that the input
impedance is the feedback impedance reduced by this factor and is approximately
180 Ω. The input impedance and the APD capacitance form a filter which must
not limit the measurement bandwidth. In the 10 MHz case the maximum value
of APD capacitance due to RC limiting is approximately 70 pF. However in
the 1 MHz case the open loop gain is increased to approximately 112 and the
94
CHAPTER 5
input impedance falls to approximately 19.6 Ω yielding an RC limit of 8 nF. The
limitation on the APD capacitance due to stability problems that result from the
interaction of the APD capacitance and the open loop transfer function of the
opamp will be discussed shortly.
The structure of the measurement system is shown in Fig. 5.1. The laser
is mechanically chopped by a rotating, slotted, disk at 180 Hz and is incident
on the diode via a set of optics. The transimpedance amplifier (TIA) is shown
schematically in Fig. 5.2. It converts the diode current into a voltage. The voltage
is amplified using a commercial low noise, wide band communications amplifier
with a terminated gain of 25 dB. A precision stepped attenuator is used to select
the system gain permitting measurement of devices under injection conditions
which elicit both high and low excess noise. The noise signal is separated from the
low frequency component of the photo-current by a cascade of single tuned Friend
bandpass filter sections [75]. The filter sets the measurement bandwidth of the
system. The circuit diagram is shown in Fig. 5.10. The filtered signal resembles
an amplitude modulated noise waveform. Periods of device illumination produce
greater noise amplitude than periods of darkness. Voltage gain is provided by a
second commercial amplifier module, which is identical to the first, and follows
the filter. Further voltage gain is rendered by operational amplifiers (shown in
Fig. 5.13) prior to a wide band (250 MHz) squaring and averaging circuit. The
squared, averaged signal is amplified a further sixteen times. After squaring
and averaging the signal is approximately square with a fundamental frequency
of 180 Hz. The signal amplitude is a function of the noise power contained
within the pass band of the filter. The squaring circuit makes use of an Analogue
Devices AD835 analogue multiplier and is shown schematically in Fig. 5.14. The
averaging circuit is a first order RC network with a time constant of approximately
100 µs. The output from the squaring and averaging circuit is measured using
a lock-in-amplifier. The photo-current signal is taken from an additional output
of the TIA where the amplitude of the 180 Hz square wave is proportional to
the photo-current. The photo-current signal is measured using a second lock-in-
amplifier.
95
CHAPTER 5
R1
100 ΩAPD Bias Voltage
C1
1 µF
−
+A1
AD9631
C2
70 pFR3
33 Ω
R2
2.2 k
R4
22 Ω
+5V
C3
100 pFC4
220 nF
C5
100 pFC6
220 nF
R5
22 Ω
−5V
−
+
A2
TLE2141
+5VC8
220 nF
-5V C9
220 nF
R7
50 Ω
Photocurrent Output
−
+
A2
TLE2141
+5VC10
220 nF
-5V C11
220 nF
R8
50 Ω
Noise Output
Figure 5.2 – Opamp based transimpedance amplifier showing all components.
5.3 Transimpedance Front End
A schematic diagram of the TIA is shown in Fig. 5.2. In this figure A1 is the
transimpedance amplifier. The opamp is the same model reported by Li [20,27].
R2 sets the transimpedance gain at all frequencies of interest. C2 is computed
using the Analog Devices data sheet [30]. Together C2 and R3 add a zero to
the compensation network around 70 MHz and so provide leading phase shift
enhancing the phase margin of the feedback system and maintaining stability.
R4 and R5 are data sheet recommendations. C3−6 decouple the opamp power
supplies to ground. C1 grounds the biased terminal of the APD for AC signals
and filters any HF noise that would be injected into the APD terminal from the
SMU. R1 is used to isolate the output of the biasing source (Kiethley 236) from
C1 by raising the biasing source output resistance sufficiently to prevent it from
oscillating when presented with the capacitive load.
96
CHAPTER 5
5.3.1 Front-End Bandwidth
Considering the schematic in Fig. 5.2 and employing the same diode small sig-
nal model as in Chapter 3 (Fig. 3.4) and assuming the opamp is perfect, the
transimpedance in the mid-band is given by,
Vo = −(iin R2) (5.1)
Unfortunately the junction capacitance of the APD and the open loop transfer
function interact in such a way as to promote instability. A more realistic approx-
imation is that A1 is a first order low pass system having a frequency independent
gain A0 of 560 and time constant τ0 of 1.6 µs [30]. The transimpedance gain stage
formed by A1 and the surrounding components forms a system with three poles
and one zero.
A0R2 · (sR3C2 + 1)
(1 + s τ0)(
1 + A0
1+s τ0
)(
1 +cj R2+R2 C2+
A0 R2 C21+s τ0
+A0 R3 C21+s τ0
+R3 C2
1+A0
1+s τ0
s+cj R2 R3 C2
1+A0
1+s τ0
s2)
(5.2)
Using parameters from Fig. 5.2 and datasheet values for A0 and τ0, it can be
shown numerically that the form changes from three real poles when cj less than
approximately 1.5 nF to a conjugate pair and one real pole when cj is greater than
approximately 1.5 nF. Eventually when cj becomes very large the conjugate poles
tend towards the real axis. This is shown graphically in the pole zero diagram of
Fig. 5.3.
Example plots of the circuit response near to the measurement bandwidth
are shown in Fig. 5.4. In comparison an example uncompensated response is
shown in Fig. 5.5. A practical implementation of this circuit may be expected
to oscillate. Unlike the Li system in which R3 = 0 Ω [20], A0 divides C2 in all
terms of the denominator, consequently C2 must be considerably larger to have
the desired effect in this design. Furthermore the bandwidth of the circuit is a
much weaker function of cj than in Li’s design.
97
CHAPTER 5
Figure 5.3 – Pole-Zero plot showing the variation of pole location with cj = 0, 10, 100 and500 pF, 1, 1.5, 2, 3, 4, 5, 10, 20, 50, 200 nF, 1 µF. The colour order is a repeated pattern ofblue, red, green, purple, orange, brown and black. rd = 1 kΩ, R3 = 33 Ω, R2 = 2.2 kΩ, C2 =70 pF. A0 = 562, τ0 = 1.5915 µs.
Figure 5.4 – Approximation of the 1 MHz operational amplifier based noise measurementsystem front end response. A0 = 562. τ0 = 1.5915 µs. rd = 1 kΩ, C2 = 70 pF, R2 = 2.2 kΩ,R3 = 33Ω and cj = 1 pF, 100 pF and 1 to 5 nF in 1 nF steps.
98
CHAPTER 5
Figure 5.5 – Approximation of the 1 MHz operational amplifier based noise measurementsystem front end response assuming the TIA opamp is first order. A0 = 562. τ0 = 1.5915 µs.rd = 1 kΩ, C2 = 0 pF, R2 = 2.2 kΩ, R3 = 0 Ω and cj = 1 pF, 100 pF and 1 – 5 nF in 1 nFsteps.
iin cj rd I2n
V 2n
−
+ A1
AD9631
R2
InR22
C2R3
EnR23
R8Vout
R9
Figure 5.6 – Schematic showing the noise model of the TIA showing all of the noise generatorsand the photo-current. R8 and R9 are 50 Ω, and are not included in the noise analysis, it canbe shown that the contribution of these resistors is insignificant.
99
CHAPTER 5
5.3.2 TIA Noise Performance
The noise performance of the transimpedance amplifier can be assessed using the
standard opamp noise model which makes use of equivalent noise voltage and
noise current generators, V 2n and I2n. The equivalent circuit of the TIA from a
noise perspective is shown in Fig. 5.6. The approach is similar to that used in
Chapter 3. The total output noise voltage is the superposition of the value of each
source multiplied by the noise gain from the source to the output. The resulting
values are combined as the rooted sum of the squared values. The pertinent
results are shown in Fig. 5.7 and will be discussed shortly.
The output noise voltage resulting from the opamp input current noise gener-
ator, In, and the thermal noise of R2, InR2 are both identical to the photo-current
generator expression (5.2). The output noise voltage due to the opamp voltage
noise generator (Vn) is,
−A0 (rd +R2)
(
1 +R2 R3 C2 rd cj s
2
rd+R2+
(rd R3 C2+rd R2 C2+R2 rd cj R2 R3 C2) s
rd+R2
)
(1 + s τ0)(
rd +A0 rd1+s τ0
+R2
)(
1 +R2 R3 C2 rd cj s2
rd+A0 rd1+s τ0
+R2+ Qs
rd+A0 rd1+s τ0
+R2
) (5.3)
where,
Q =
((
rd +A0 rd1 + sτ0
+ R2
)
R3 +
(
rd +A0 rd1 + sτ0
)
R2
)
C2 + R2 rd cj (5.4)
The noise voltage at the output resulting from the thermal noise of R3, EnR3
is,
A0 rdR2C2 s
(1 + s τ0)(
rd +A0 rd1+s τ0
+R2
)(
1 +R2 R3 C2 rd cj s2
rd+A0 rd1+s τ0
+R2+ Qs
rd+A0 rd1+s τ0
+R2
) (5.5)
5.3.3 Characterisation of the TIA
This transimpedance amplifier is characterised by similar means to the 10 MHz
TIA reported in Chapter 3. The details of the circuit used to approximate a
current source are described in Section 3.11. The DC characteristic of the TIA
100
CHAPTER 5
Figure 5.7 – Top left: graph showing the equivalence of the noise gain between the analyticalmodel (symbols) and a SPICE small signal (.AC) simulation (lines) for each source in theequivalent circuit. EnR3 - black, EnR2 - red, Vn - green, In - blue and the transfer functionis shown in yellow. Top right: Analytical model output noise voltage versus frequency foreach source with cj as a parameter. cj = 1 pF (solid lines), 5 nF (dashed lines). Black linesrepresent shot noise from 1 µA, EnR3 - red. EnR2 - green. In - blue. Vn - yellow. Thetotal is shown in pink. Middle left: Measured circuit noise c.f analytical model. Solid linesrepresent measured data from 0 nF to 5 nF in 1 nF steps. Middle right: Measured circuit noisec.f SPICE noise simulation (.NOISE) using an Analog Devices AD9631 model. Solid linesrepresent measurements, 0 nF to 5 nF in 1 nF steps. The bottom graph shows the analyticalmodel noise signal to noise ratio (NSNR) against frequency with cj as a parameter. cj = 0, 10,100 and 500 pF then 1 nF – 5 nF in 1 nF steps for a 1 µA signal current.
101
CHAPTER 5
Figure 5.8 – DC response of the TIA showing the onset of non-linearity around 2 mA inputcurrent.
is shown in Fig. 5.8 and is linear until the opamp saturates against the power
supply rail. This occurs around at around 4 Volts and is partially due to the use
of R4 and R5, but dominantly due to the voltage drop across circuits internal to
the opamp.
Figure 5.9 shows the measured frequency response of the TIA front end as a
function of increasing APD capacitance. The agreement with the model which
assumes the opamp is first order is very reasonable, with the resonance effect
being slightly more pronounced in the model. The inset of Fig. 5.9 shows the
difference in roll off due to increasing cj more clearly.
Graphs depicting the measured and simulated noise characteristics of the TIA
are shown in Fig. 5.7. From the top right graph in this figure it is evident that the
voltage noise of the opamp is by between one half and two orders of magnitude
greater than any other source, depending on the junction capacitance. The noise
gain of this source is shown in the the top left graph, and is 15.2 dB (5.75 lin) at
1 MHz when cj is 1 nF and rd is 1 kΩ. As rd increases k, the frequency independent
gain for the Vn source, decreases reaching approximately 1.2 (lin) when rd is 10 kΩ.
Ultimately, however, the voltage noise of the opamp is the determining factor in
the noise of this TIA, and it can only be adjusted by selecting a different opamp.
The near equivalence of the analytical model and the SPICE model is shown
102
CHAPTER 5
Figure 5.9 – Frequency response of TIA. Input capacitance increases from 0 nF to 5.5 nF in0.5 nF steps.
by the middle graphs of Fig. 5.7. The early roll off of the analytical model could
be corrected by adding poles and zeros to the transfer function of the opamp to
force it to follow the open loop response published by Analog Devices more closely.
However this added complexity would cloud an already complicated picture and
add little overall.
The bottom graph in Fig. 5.7 shows the NSNR as a function of frequency for
several values of cj . Up to approximately 100 pF, the NSNR of this front end
is nearly equal to Li’s system [20]. Thereafter NSNR deteriorates due to the Vn
peaking, at 5 nF the NSNR is approximately -52 dB. By comparison the 10 MHz
system described in Chapter 3 also has a low frequency NSNR of just over -25 dB,
falling to approximately -55 dB for cj = 5 nF.
5.4 Bandwidth Setting Filter and ENBW
The effective noise power bandwidth (ENBW) of the measurement system is
defined by a cascade of single tuned Friend filter sections [75]. A simplified
schematic is shown in Fig. 5.10. The ENBW is in approximately the same
relative proportion to the centre frequency as in the system after Li [27] and as in
103
CHAPTER 5
R1
50 Ω
Input
−
+
A1
AD829
R2
2.7 k
R3
249 Ω
C1
100 pF
C2
100 pFR4
10 k
−
+ A2
AD829
x7
R5
2.7 k
R6
249 Ω
C3
100 pF
C4
100 pFR7
10 k
−
+ A3
LM7171
C5
220 nFR8
50 ΩOutput
Figure 5.10 – 1 MHz Single tuned 16th order bandpass filter.
the system described in Chapter 3. The frequency response of the combination
of the filter and front end is shown in Fig. 5.11 as a function of APD capacitance.
The effective noise power bandwidth of the system is computed by normalized
numerical integration of Fig. 5.11. The ENBW of the system as a function of
device junction capacitance is shown in Fig. 5.12 and is given by
Beff(cj) =1
G20
∫ f2
f1
|G(f, cj)|2 df. (5.6)
where Beff is the ENBW, G0 is the gain at the centre frequency, G(f, cj) is the
frequency and APD capacitance dependent TIA and filter gain. f1 and f2 are
placed sufficiently far from the centre frequency, f0, that practically all of the
area under the curve in Fig. 5.11 is integrated.
5.5 Voltage Amplifiers
The voltage amplifier schematic required to replicate the measurement system is
shown in Fig. 5.13. The objective of this amplification (and that provided by the
Minicircuits Amplifiers) is to scale the noise signal to a suitable value to generate
an output from the noise power meter with minimal ambiguity. The order and
gain of each amplifier is shown in the block diagram of Fig. 5.1.
Similar arguments apply to the nature of the noise signal and the design of
crest factor in this measurement system as for the 10 MHz system described in
Chapter 3.
104
CHAPTER 5
Figure 5.11 – Frequency response of TIA and ENBW setting filter with input capacitance.Input capacitance increases from 0 nF to 5.5 nF in 0.5 nF steps. Solid line 0 nF, medium dash5.5 nF.
Figure 5.12 – The effective noise power bandwidth as a function of APD capacitance. Circlesand triangles represent independent data sets.
105
CHAPTER 5
C1
220 nF
R1
50 Ω
From ZFL-500LN+
−
+
A1
LM7171
R2
470 ΩR3
220 Ω
R4
50 ΩC2
220 nF
To Power Meter
3x
Figure 5.13 – Opamp voltage amplifier.
Calibration measurements on the 0 dB attenuation, using a BPX65 (cj ∼ 4 pF)
photo-diode show output noise voltage increases with the root of photo-current
with excellent agreement (R2 > 0.99) over a range of photo-currents from 0 to
200 µA indicating that full shot noise has been measured. Further evidence for
the acceptability of a low crest factor is the agreement of the data gathered using
this system and that gathered using the 10 MHz system reported in Chapter 3.
This data is presented in Chapter 6. The likelihood of agreement between the two
systems while both are producing erroneous data is extremely small. While both
systems operate under the same assumptions slight differences in implementation
(for example the difference in background noise floor and the difference in total
gain) would have to conspire to produce a consistent and repeatable error in
both measurement systems. Moreover the consistency of data between attenuator
settings on individual systems is excellent.
The linearity of the system prior to the power meter can be examined by
similar methods to those described in Chapter 3 and by connecting a sinusoidal
signal generator to the TIA (Fig. 5.2) via the resistor chain (Fig. 3.16) and ex-
amining the output from the final amplifier prior to the power meter using a
spectrum analyser. This method allows the measurement of the majority of the
system gain. Measurement of the power meter transfer function is obtained sep-
arately. Signal distortion is manifested as extra peaks on the spectrum which are
not present in the generator output. Clipping of the system noise on the 0 dB
attenuation setting is not present in the reported design.
106
CHAPTER 5
R1
50 Ω
Input
−
+
−
+
Σ
+
++1
R2
330 Ω
R3
100 ΩAD835
R4
1 k
C1
0.1 µF
−
+
A1
TL071
R5
15 kR5
1 k
Output
Figure 5.14 – AD835 based power meter.
5.6 Noise Power Meter
The power meter design shown here was developed by B. K. Ng, and is based
on the Analog Devices AD835 wide band analogue multiplier. A representative
schematic is shown in Fig. 5.14, where the AD835 is shown in a block diagram
form. The inputs of the multiplier are connected together in order that the output
voltage is dominantly proportional to the square of the input. A small linear term
and a DC offset remain. The DC offset is lessened by trimming the offset null of
the TLO71 and finally removed by AC coupling the power meter output to the
lock-in-amplifier. The linear term is partially resultant from imperfection in the
squaring process, but it is likely that the majority of the linear term is a result
of regressing manually collected data possessing some uncertainty. The squared
noise voltage signal is applied to a first order low pass filter, R4 and C1, with a
time constant of approximately 100 µs. The filter is required to pass the 180 Hz
chopping signal such that the LIA can differentiate between light and dark noise,
but must filter the noise signal (around 1 MHz) to produce a signal representing
the average value of the squared noise voltage.
The transfer function for the multiplier circuit after Ng is,
Vo = 3.53 V 2i − 0.03 Vi + 0.8850 (5.7)
where Vo is the mean squared output voltage and Vi is the RMS input voltage
of a deterministic test signal.
107
CHAPTER 5
Figure 5.15 – Measured power meter transfer characteristic.
5.7 Conclusion
In this chapter a new transimpedance amplifier based measurement system for use
in measurements of excess noise in avalanche photodiodes has been documented.
This amplifier has found use in situations where the device being measured has
high junction capacitance, up to 5 nF, and where the device has lower than usual
dynamic resistance, rd. The measurement of a much wider range of devices is now
possible than was previously. Noise measurements on devices held in cryogenic
chambers or high temperature ovens is, in principle, possible by connection of the
device and TIA by coaxial cable. This measurement system has been used to con-
firm measurements made using the 10 MHz transistor based noise measurement
system described in Chapter 3.
An area of particular interest at present is the use of narrow band-gap materi-
als including InAs and HgCdTe in X-ray detection applications. X-ray detectors
are often relatively large area devices, and when manufactured from narrow band-
gap materials may exhibit significant tunnelling currents. These devices would be
unmeasurable on prior systems due to bandwidth constraints or system instabil-
ity. The present work and the noise measurement system reported in Chapter 3
overcome these obstacle, opening new avenues of investigation.
108
Chapter 6
Excess Avalanche Noise Thin
Silicon APDs
This chapter reports multiplication and excess noise factor measurements, and
non-local modelling results, in a set of Silicon pin diodes over a range of optical
injection wavelengths. A local model ionisation coefficient extraction is performed
to highlight the effect of the dead-space on the ionisation coefficients when com-
pared to the bulk coefficients. The data has been gathered using the two new
measurement systems, described in Chapters 3 and 5, specifically designed to
measure excess noise and multiplication in avalanche photo-diodes possessing
high junction capacitance and large tunnelling current.
6.1 Review of Impact Ionisation in Silicon
Silicon was one of the first semiconductor materials to have its bulk ionisation
properties investigated systematically. A portion of this systematic investigation
was undertaken by Bell Systems, where several of the early workers were based.
The first report of ionisation coefficients in Silicon was by McKay and McAffee [7]
both authors were with Bell. Impact ionisation was treated with the pre-existing
theory of gas discharge as a framework, which is also called ‘avalanche theory’ but
is also referred to as ‘Townsend’s beta theory’. They assumed α and β were equal
and used pn junctions, so the ionisation coefficient was reported as a function of
109
CHAPTER 6
the peak electric field. The authors fitted the data to,
α0 =(1−M−1)
w(6.1)
where α0 is the ionisation coefficient, and has the units cm−1, M is the photo-
multiplication and w is the effective width of the device assuming a uniform
field. Microcomputers were not available to assist the early workers with taxing
integrals, and that the uniform field approximation produces closed form solutions
which can be derived, with some effort, by hand.
McKay published a separate work on the breakdown of Silicon diodes in 1954
[6] still assuming that the coefficients were equal. This work is significant, it is
the first work in which impact ionisation is considered in a similar way to the
present methods. It is also insightful, predicting the non-local nature of avalanche
multiplication – its dependence on carrier history, most evident in thin devices.
McKay gives a solution to a multiplication integral, which is derived from first
principles, for several junction geometries. The usefulness of the experimental
data appears to be limited by the device material quality.
P. A Wolff, also with Bell, and a collaborator of McKay and McAffe solved
the Boltzmann transport equation in a later publication of 1954 [8]. This was
done by separating the problem into three regions, firstly carriers with energies
appreciably less than the pair creation energy were solved, then carriers with
energies appreciably more than the pair creation energy, finally the third region
in which the carriers have energies approximately equal to the pair creation en-
ergy was resolved by extrapolating the solutions in the other two regions until a
smooth function resulted. This paper ends with a strict definition of ‘high field’.
The field is considered ‘high’ when a carrier can move from zero energy to the
pair creation energy in less distance than the average distance between phonon
scattering events. It is in circumstances where the electric field strength is suf-
ficiently high that the effect of phonon scattering on the ionisation process may
be neglected.
The first work that showed α > β was by Miller in 1957 [76]. Miller was also
working for Bell. His 1957 paper is the first in which note has been made of the
importance of the injection purity, this is not surprising when it is considered that
110
CHAPTER 6
the prior workers assumed α and β to be equal so the initiating carrier type was
irrelevant. This work is the first Silicon paper in which the following expression
appeared,
1− 1
M= −
∫ w
0
αi (E) exp
[
−∫ x
0
[αi (E)− βi (E)] dx′.
]
dx. (6.2)
Miller stated his empirical formula for fitting multiplication data (8.2) in a
1955 Germanium paper [5]. The diodes reported by Miller [76] were pn step
junctions, although Miller preferred transistors for providing carriers of one type
(i.e. pure injection).
Chynoweth and McKay published two papers in 1957 [10, 77] expanding the
volume of available data. Chynoweth published again in 1958 [11] this publication
contains more data, and the first example of α and β plotted as functions of inverse
electric field. The empirical expression which bares his name is also published.
However it is clear that the expression is borrowed from gas discharge theory and
was proposed as early as 1936 [78].
α (β) = A exp
(B
ξ
)C
(6.3)
In the 1958 paper the power term does not appear.
Shockley entered the debate with a publication in 1961 [12] which argued
that the multiplication effect was produced by those carriers lucky enough not to
undergo collisions which randomise the motion of the carriers. Baraff provided
a theoretical paper which applied boundaries to both Shockley’s and Wolffe’s
solutions of the Boltzmann transport equation [9]. It was shown that Shockley’s
solution was more appropriate at low electric fields and that Wolffe’s was more
appropriate at high fields.
Throughout the 1960s and 1970s several authors published more data regard-
ing the values ionisation coefficients [79–90]. The coefficients of Van Overstraeten
and De Man [85] and Grant [87] are preferred due to the large number of samples
that their study undertook.
Other than the work of Robbins [91, 92] which showed the isotropic nature
of impact ionisation coefficients in Silicon, no significant work was carried out
111
CHAPTER 6
thereafter until Pauchard [93] and Tan [94] both in 2000. These papers deal with
sub-micron width devices. Pauchard found no evidence of dead space effects in
a 0.8 µm structure. Tan et al. showed that excess noise is reduced in devices as
thin as 0.1 µm and that the choice of carrier injection becomes less significant as
depletion width decreases. It is however still preferable to select the more readily
ionising carrier to produce a low noise device.
Recent work has been reported by Massey et al. [36]. This data is in good
agreement with that of Van Overstraten and DeMan [85]. Massey concludes that
the dead space effect is less pronounced in Silicon than in GaAs and as a result
the local model produces a reasonable approximation of impact ionisation for low
values of multiplication even when the dead space is significant. Agrawal [95]
showed data from sub-micron structures in agreement with that of Massey.
6.2 Application: On-Chip Photonic Data
Transfer
The principle benefits of optical interconnects between cells of a VLSI design
were reported in a seminal paper by Goodman et al. [96] and further expanded
and updated by Miller [97]. This topic has received attention in the professional
electronics press also, for example [98, 99].
The principle of operation has been widely reported and may be found de-
scribed in many of the references herein. Briefly, two or more cells in a digital
chip are to communicate at a high data rate. An alternate scenario may be that a
clock signal having very low delay around the chip is to be transmitted to several
cells simultaneously. In any case the transmitting cell either generates a series of
NIR optical pulses or modulates an external NIR optical source [100, 101] which
is passed into the chip by some means, for example [102]. The modulated light
is coupled into a Silicon dioxide wave guide structure [103] which is formed by
standard lithographic processes on top of standard metallised interconnects. The
wave guide provides an equivalent function to a fibre-optic cable. The modu-
lated light arrives at a detector fabricated from a material suited to its detection,
InGaAs and Ge have been reported, [104–107] This detector is either grown epi-
112
CHAPTER 6
taxially onto the Silicon structure or is produced separately and is mechanically
bonded both methods have been reported [108,109]. MSM and pin detectors have
been reported [110]. The state of the art appears to lie with a wave-guide pin
device which combines high -3 dB bandwidth of 47 GHz with good responsivity
0.8 A/W, low dark current, 0.072 nA at -1 V and a CMOS compatible operating
voltage of -3 V [111].
The principle advantages of intra-chip optical communications are,
• Unlike electrical interconnects, the maximum bandwidth of an optical in-
terconnect is unrelated to the number of nodes to which it is connected.
That is to say there is no optical effect which is equivalent to capacitive
loading of an electrical interconnect.
• Optical interconnects run in close proximity are incapable of interacting
with each other by capacitive coupling.
• Wave-guides are not confined to being planar structures. Unlike electrical
interconnects two wave-guides can pass through each other without the
information each is carrying being distributed into both.
• The possibility of adapting the route of an optical interconnect – on the fly
– based on the data being sent has been demonstrated [112].
• Latency in clock distribution can be avoided by passing the clock into several
areas of the chip simultaneously from a common optical source.
• The data rate that can be achieved in optical communications is signifi-
cantly greater than in electrical interconnects because the carrier frequency
is higher.
Ge on Si pn, unity gain pin and APD detectors have been reported in recent
times including [4, 106, 113–115]. These attempts to produce a NIR photodiode
on a Silicon wafer can be broadly categorised into three groups,
• Bonding structures composed of narrower band-gap materials including Ge
and InGaAs to a Si wafer [107, 109].
113
CHAPTER 6
• Hetero-epitaxial growth for example Ge on Si [106, 116, 117]
• Homo-epitaxial growth of Si to produce a visible light interconnect [118].
The use of a thin high speed, low noise, Silicon multiplication region APD
is briefly considered by Chen et al. [118]. However, APDs are yet to receive as
much attention in this application as pin detectors.
APDs considered for this application must be very high speed, data rates of
100 Gbits−1 are desirable. The APD should also operate at CMOS voltages. Both
speed and voltage demands require a thin device which is likely to exhibit quan-
tum mechanical tunnelling. The device area will be determined by the necessity
of obtaining a high coupling efficiency between the wave guide and the device
surface. Operating voltage (or electric field), gain, area, speed and tunnelling
are contended against each other to find an optimum device design. Tunnelling
current scales with device area, the thinnest device in this chapter can provide a
useful estimate of the tunnelling, gain and voltage trade off. For example, based
on the data in this chapter, a 31 nm thick 5 µm x 10 µm device may be expected
to exhibit 5 nA tunnelling current at 4 V reverse bias and provide a multiplication
of ∼3.5 for 850 - 1064 nm light.
6.3 Application: Medical/Biological Imaging
Silicon is the principle material of choice when optical detection in the visible
region is required. A great variation in devices exists including CCD arrays
CMOS arrays, linear detector arrays for for photographic and video imaging and
position sensing or machine vision; as well as the pin / nip diodes which are of
greatest interest here. In medical fluorescence imaging (MFI), the detection of
the florescence wavelength is often important but in other cases more information
is available by sensing the fluorescence decay [119–121]. Capturing a spatially
resolved image from very little incident light is often required and high frame rate
is desirable. The case of medical fluorescence imaging is particularly interesting.
Early systems used vidicon tube detectors and gas lasers. They were somewhat
unwieldy and consequently fluorescence was observed in vitro [122]. The advent
114
CHAPTER 6
of reliable semiconductor lasers that produce near UV lines, and the use of fibre
optics allows in vivo studies of tissue [123, 124].
The present clinical uses of MFI are,
• Early cancer diagnostics
• Identification of tumour boundaries
• Assessment of blood vessels
• Visualisation of lymph vessels
• Treatment response assessments
• Interactive dosimetry [125, 126]
Presently photon counting directly in CMOS [127–129] is generating consid-
erable interest as a suitable technology for florescence imaging and positron emis-
sion tomography [130] and chemical sensing [131] and as a replacement for LIDAR
systems such as [132, 133].
6.4 Chapter Objective
The objective of this chapter is to provide a systematic investigation of the extent
of the dead-space effect in thin Silicon avalanche photo-diodes by measurement of
excess noise as well as photo-multiplication. This work also provides experimental
confirmation that the RPL model and ionisation coefficients can effectively predict
the effects of impact ionisation in Si devices possessing dimensions approximately
equal to present FET gate lengths.
Excess noise measurements on five APD structures having intrinsic widths
over the range of interest for investigating dead-space are reported. Photo-
multiplication is measured at five wavelengths, providing a range of injection
conditions including, for the first time, pure injection in the thinnest device re-
ported to date. Excess noise measurements are obtained at 633 nm. The mea-
surement systems used to collect this data are described in Chapters 3 and 5. A
comparison of bulk coefficients and extracted coefficients (ignoring dead-space)
will be provided in order to assess the possible benefit provided by the dead-space
in thin structures.
115
CHAPTER 6
6.5 Layer Details
Five p+in+ layers were grown epitaxially on As doped n+ substrates using chemi-
cal vapour deposition. Intrinsic layers with nominal region thickness w = 0.031,
0.079, 0.132, 0.223 and 0.347 µm were grown and capped with 0.2 µm thick p+
cladding layers. Square mesas of areas from 0.08 mm2 to 1.65 mm2 were fab-
ricated by reactive ion etching and their edges were chemically passivated with
a tetraethoxysilane layer. A top electrical contact was formed on the p+ layer
via sputtered AlSi and a small hole was etched into this contact to allow optical
access. SIMS measurements were performed, confirming the nominal growth val-
ues of w and the top p+ cladding layer and showing steep doping profiles at the
interfaces in each of the structures.
Table 6.1 – Layer details for thin Silicon layers.
Layer w [µm] Vbd [V]
8 0.031 ∼511 0.079 6.512 0.132 8.213 0.223 11.214 0.347 14.7
6.6 Current Voltage Characteristics
Current voltage (IV) measurements were performed without room lighting. For-
ward and reverse IV measurements were taken using an HP 4140B Picoammeter.
Forward and reverse bias measurements were undertaken on one device of each
size on each wafer, however for clarity only the 1 mm2 data are shown in Fig. 6.2.
Fitting of each characteristic to (2.1) is also shown in this figure. The emission
coefficient for all wafers except #12 is 1.15. In wafer #12 n = 1.1. Saturation
currents are between 25 fA and 50 fA for all wafers. The emission coefficient close
to unity indicates the dominant transport mechanism is diffusion. There is no
sign of series resistance in these layers.
In reverse, all of the layers except #8 show bulk breakdown and low leakage
currents. Layer #8 exhibits considerable tunnelling current, and does not break-
down prior to current limiting. However the measurement techniques used allow
116
CHAPTER 6
the photo-generated current to be extracted from the tunnelling current even un-
der conditions where the tunnelling current is up to 60 dB greater in magnitude.
An example is shown in the Fig. 6.3 where 325 nm photo-current measurements
are shown overlaid on the dark current plots.
6.7 Capacitance Voltage Characteristics
Junction capacitance as a function of reverse bias voltage is shown Fig. 6.1.
These data were gathered using an HP4275A multi-frequency LCR meter. A
1-D Poisson solver is used to fit the doping densities and intrinsic region width
to the measured data. The fitted intrinsic region widths yield the data shown
in Table 6.1. The Poisson solver is described in Appendix A. The diffusion
voltage is obtained from extrapolation of the forward bias capacitance voltage
characteristics towards the bias axis. The crossing point represents the diffusion
voltage, and is 1 V.
Figure 6.1 – Capacitance versus reverse bias voltage characteristics for 1 mm2 devices. #8,red circles #11, green circles, #12, blue circles, #13, pink circles, #14, grey circles. Black linesrepresent fitting to a 1-D Poisson solver.
117
CHAPTER 6
Figure 6.2 – Forward bias current versus voltage characteristics (open symbols). Dashed linesrepresent a numerical, trial and correction, fitting to (2.1).
118
CHAPTER 6
Figure 6.3 – Reverse bias dark current versus voltage characteristics (solid lines) and photo-current versus bias voltage (open circles) taken at 325 nm.
119
CHAPTER 6
Table 6.2 – Absorption coefficients at selected wavelengths in Silicon.
Wavelength[nm]
α [cm−1]Light reaching i region
[%]
325 1.30× 106 50× 10−9
442 27.7× 103 57532 9.00× 103 84633 4.00× 103 921064 26.86 99.95
6.8 Photo-multiplication
Photo-multiplication was measured at 325, 442, 532, 633 and 1024 nm. Excess
noise was measured as a function of reverse bias at 633 nm and will be discussed
shortly. 325 and 442 nm light was provided by a He-Cd laser, a diode pumped
solid state laser provided 1064 and 532 nm. 633 nm was obtained from a HeNe
laser making it suitable for noise measurement. The laser light is focused to a
spot on the device p+ layer through the optical window. The window dimensions
are considerably greater than the spot diameter. The absorption coefficient of Si
at these wavelengths is given in Table 6.2 [134].
The photo-multiplication versus reverse bias voltage data is shown in Fig. 6.4
and M-1 against reverse bias voltage is shown in Fig. 6.5 which exposes the lower
multiplication values. A non-local model using the parameters in Table 6.3 is used
to generate the lines in these figures. The modelling results are obtained using an
extended version of the RPL model [17], the particular implementation of which
is due to B. K. Ng [18]. This model accounts for photo-generation throughout
the structure, diffusion of photo-generated carriers and surface recombination.
The parameters associated with the generation of the modelling data are given in
Table 6.3. The ionisation coefficients used are due to Rang [90]. The ionisation
threshold energies are due to Tan et al. [94]. The hole diffusion length is after
Tyagi and Van Overstraeten [135]. The electron diffusion length is taken from del
Alamo and Swanson [136]. The model makes use of a flat electric field approx-
imation, which requires slight adjustment of the intrinsic region width to yield
the correct breakdown voltage. Not withstanding this adjustment, the modelling
120
CHAPTER 6
Table 6.3 – RPL modelling parameters.
Parameter Value
Electron Ionisation coefficient (α)1.286×
106(
−1.4×106
ξ
)
cm−1
Hole Ionisation coefficient (β)1.438×
106(
−2.02×106
ξ
)
cm−1
Electron threshold energy (Ethe) 1.7 eV
Hole threshold energy (Ethh) 2.4 eV
Diffusion Voltage (Vd) 1 V
Top Cladding
Minority carrier diffusion length (Le) 1.8 µm
Mobility (µe) 800 cm2
V s
p+ cladding thickness 0.2 µmRecombination Velocity (S) 1200000 cm
s
Back Cladding
Minority carrier diffusion length (Lh) 1 µm
Mobility (µh) 100 cm2
V s
n+ cladding thickness 99.5 µmRecombination Velocity (S) 1200000 cm
s
data fits well in all layers except layer #8 when the photo-multiplication is less
than two, where the RPL model overestimates the photo-multiplication slightly.
The likely cause of this error is that the very small intrinsic region distance in
layer #8 requires a proportionally larger adjustment to fit the correct break-
down voltage in the flat field approximation. Despite the very high doping in the
cladding layers, the electric field must extend a (relatively) large distance into the
cladding layer compared to the intrinsic width. This yields a more trapezoidal
electric field profile than in the other layers which also have very high doping
in the cladding layers, but also have a much thicker intrinsic width, making the
error proportionally smaller.
121
CHAPTER 6
Figure 6.4 – Photo-multiplication, M as a function of reverse bias voltage, V, for each wafer;obtained from the 10 MHz measurement system. Open symbols represent independent mea-surement datasets where the symbol colour represents 325 (blue), 442 (cyan), 532 (green), 633(red), 1064 nm (dark grey). Black lines represent non-local modelling data for 325 (solid),442 (long dashed), 532 (medium dashed), 633 (short dashed), 1064 nm (dotted) and pure holeinjection (dash – dot) using doping and intrinsic region widths obtained from 1-D Poisson fit-ted capacitance voltage characteristics then adjusted to provide the correct breakdown voltageusing a flat electric field approximation.
122
CHAPTER 6
Figure 6.5 – Photo-multiplication, M-1 as a function of reverse bias voltage, V, for eachwafer; obtained from the 10 MHz measurement system. Open symbols represent independentmeasurement datasets where the symbol colour represents 325 (blue), 442 (cyan), 532 (green),633 (red), 1064 nm (dark grey). Black lines represent non-local modelling data for 325 (solid),442 (long dashed), 532 (medium dashed), 633 (short dashed), 1064 nm (dotted) and pure holeinjection (dash – dot) using doping and intrinsic region widths obtained from 1-D Poisson fittedcapacitance voltage characteristics then adjusted to provide the correct breakdown voltage usinga flat electric field approximation.
123
CHAPTER 6
6.9 Excess Noise Measurement
For the 633 nm case, 92% of the light reaches the depletion region through the 0.2
µm p+ cap layer. The excess noise results obtained from the 10 MHz measurement
system, described in Chapter 3, are shown in Fig. 6.6. The same flat electric
field model is used to generate the coloured lines in this figure as for the photo-
multiplication data.
With the exception of #8, the agreement is good at lower multiplications in all
cases. The curvature of the noise data which is particularly apparent in #11 is due
to the falling dynamic resistance of the diode as voltage and, especially, current
increases. For the 10 MHz system the limit of measurement is approximately
rd = 200 Ω. In devices having low dynamic resistance and high series resistance
the limit is considerably increased as the device series resistance is effectively
added to the measurement system input impedance. For the red triangles in #11
at 6.3 V reverse bias rd = 236 Ω. All data after this voltage could, legitimately,
be discarded. To avoid this effect a lower primary photo-current may be used,
however this worsens the NSNR increasing the measurement uncertainty.
In #8 the data is in plausible agreement, if a little high. Measurement of
this layer is particularly taxing, as several factors that deteriorate the measure-
ment accuracy are present. The device has particularly high capacitance. This
increases the noise floor of the measurement system worsening the (already neg-
ative) NSNR. To improve the NSNR, a larger primary photo-current is desirable.
Since there is little multiplication available in this layer, a large primary photo-
current is generally desirable but, the increased photo-current reduces the device
dynamic resistance. The dynamic resistance is further reduced by the very large
tunnelling current which flows as ‘breakdown’ is approached. In spite of the
taxing measurement the #8 data is in plausible agreement with the modelling.
The excess noise results obtained from the 1 MHz measurement system, de-
scribed in Chapter 5, are shown in Fig. 6.7. These data are in good agreement
with the 10 MHz excess noise data. Confirmation of the data with two separate
measurement systems and their general agreement with the non-local modelling
forms a robust argument in favour of the accuracy of the data. The agreement
between the 10 MHz and 1 MHz data shows no bandwidth limiting or 1/f effects
124
CHAPTER 6
in the devices.
Figure 6.6 – Excess noise factor versus photo-multiplication at 10 MHz. Closed symbolsrepresent independent datasets at 633 nm. Coloured lines represent non-local modelling datafor 325 (blue, solid), 442 (cyan, long dashed), 532 (green, medium dashed), 633 (red, shortdashed), 1064 nm (dark grey, dotted) and pure hole injection (black, dash – dot) using dopingand intrinsic region widths obtained from 1-D Poisson fitted capacitance voltage characteristicsthen adjusted to provide the correct breakdown voltage using a flat electric field approximation.Short dashed grey lines represent McIntyre’s lines of constant. k = 0 to 1 in 0.1 steps.
125
CHAPTER 6
Figure 6.7 – Excess noise factor versus photo-multiplication at 1 MHz. #8 red closed symbolsrepresent the lower measurement bound, dark red closed symbols represent the upper measure-ment bound. In all other graphs closed symbols represent independent datasets. Red closedsymbols represent 633 nm measurements, green closed symbols represent 543 nm measurements.Coloured lines represent non-local modelling data for 325 (blue, solid), 442 (cyan, long dashed),532 (green, medium dashed), 633 (red, short dashed), 1064 nm (dark grey, dotted) and pure holeinjection (black, dash – dot). Short dashed grey lines represent McIntyre’s lines of constant. k= 0 to 1 in 0.1 steps.
126
CHAPTER 6
Figure 6.8 – Extracted α (left) and β (right) using local model, flat field approximation.
6.10 Local Ionisation Coefficient Comparison
To assess the effect of the dead-space in this set of devices, the method after
Stillman and Wolfe is used [137]. Ionisation coefficients are extracted from these
devices assuming that the dead-space is negligible and the local model applies.
A flat electric field profile is also assumed so that (1.9) and (1.16) can be used in
conjunction with,
ξ =Vt + Vd
w(6.4)
where Vt is the terminal voltage, Vd is the diffusion voltage and w is the intrinsic
region width obtained from CV fitting.
Figure 6.8 shows the extracted α and β respectively for all layers. The bulk
coefficients of Van Overstraeten and De Man [85] are shown to agree well with
the extracted coefficients as intrinsic width increases.
6.11 Conclusion
A systematic empirical investigation of the effect of dead-space in Silicon has
been reported in this chapter. Measurements of photo-multiplication and excess
noise factor on a series of five Silicon pin photo-diode structures including, the
thinnest intrinsic width, widest area, APD structure reported to date. These
results have been confirmed using the two new measurement systems reported in
127
CHAPTER 6
Chapters 3 and 5. It has been shown that the RPL model is sufficiently accurate
to predict the effective reduction in ionisation coefficients over a wide range of
nominal intrinsic region width. From this work it is clear that the dead-space
starts to affect the device characteristics when the region width is approximately
0.4 µm.
The intrinsic width of the thinnest structure in this work is similar in di-
mension to present FET gate lengths in VLSI designs and microprocessors. In
circumstances where impact ionisation is a possibility, the use of one or more thin
regions, if they can be accommodated by the design, allows much higher electric
fields to be developed in those regions while the ionisation coefficients remain
considerably below the bulk value. This effectively increases device speed and re-
duces power dissipation (assuming no significant tunnelling or carrier feedback)
when compared to a thicker region in an otherwise similar device. For example in
a 71 nm long MOS transistor channel, sustaining an electric field of 0.25 MV/cm,
α is reduced by a factor of 45 over the bulk value that a 400 nm channel length
transistor experiences. β is negligible, so carrier feedback is unlikely to have a
significant effect.
An area of present interest is the use of optical on-chip interconnects as a
replacement for electrical interconnects. Used in this way, optical interconnects
offer several reported advantages over electrical interconnects [96, 112]; the most
important arguably being higher bandwidth and lack of crosstalk. Several ap-
proaches for detection of light in on-chip optical interconnects exist, including
the use of hetro-structures of Ge grown on Si, and separately fabricated detectors
bonded to Si wafers [104–106, 108–111, 113–115]. The use of Si as an absorption
and multiplication region [118] has also been evaluated. It is expected that thin
(high speed, low noise) small area Si avalanche photo-diodes may be integrated
into VLSI designs at the transistor level, where Silicon will form either the en-
tire device, or the multiplication region of a device possessing a separate NIR
absorber, Ge or InGaAS for example.
128
Chapter 7
Excess Noise Measurements and
Impact Ionisation Coefficients in
4H-SiC
This chapter reports multiplication and excess noise factor measurements and
local modelling results in a thick (2.7 µm) Silicon Carbide nip APD and a thick
(0.57 µm) SAM-APD. Multiplication and excess noise data has been gathered for
optical injection at 244 and 325 nm, where 244 nm provides pure hole injection
in the thicker device. Ionisation coefficient extraction is performed using a local
model to verify experimentally the ionisation coefficient at lower electric fields.
The data have been gathered using the measurement system first reported by Li
and described in Chapter 2.
7.1 Introduction
4H Silicon Carbide is a wide, indirect band gap (3.23 eV) [138] visible blind,
radiation tolerant [139] semiconductor. It is chemically quite inert, pressure tol-
erant, has a large bulk breakdown field, and is tolerant of high power densities. It
also has a high carrier saturation velocity ∼ 8.0 x 106 cms−1 for 8 off the c-axis
towards <1120> [140]. The military and civilian potential of Silicon Carbide
detectors has been noted [141,142] in the following areas: Missile/rocket exhaust
129
CHAPTER 7
plume detection, tracking and chemical composition analysis [143], chemical and
biological battlefield reagent detectors, medical fluorescence imaging, fire/smoke
detection, furnace control, water purification, internal combustion engine moni-
toring, UV radiation dose monitoring, pollution monitoring, short range non line
of sight communications, autocorrelation of pulse lasers and covert space-to-space
communications. SiC is presently a multi-billion dollar industry in the combined
areas of: High power microwave transistors, power semiconductor rectifiers and
switches, high temperature linear devices, super bright blue and green LEDs,
blue and UV laser diodes (as a substrate material) and synthetic gemstones. SiC
growth and fabrication is considerably more mature than GaN. SiC power de-
vices have been demonstrated at temperatures exceeding 600C. NASA Glenn
Research Center has published several papers regarding characterisation of SiC
high voltage, high power, devices at 500C for up to 2000 hours [144, 145].
7.1.1 Bulk Growth
For Silicon Carbide semiconductor manufacturing to flourish, a high level of inte-
gration with pre-existing Silicon device production technology is required. This
implies the growth of large area wafers. A modification to the Lely process was
developed and reported by Tairov and Tsvetkov [146,147]. A seed crystal is held
in a cavity, polycrystalline SiC is vaporised at 2400C and then is allowed to
condense onto the seed crystal. The majority of SiC growth is done on or up to
8 off the c-axis <0001>. Other growth directions have been attempted. The
bulk growth rate is in the order of a few millimetres per hour.
7.1.2 Epitaxy
Chemical Vapour Deposition (CVD) is a favourable process for SiC epitaxy. It
requires little adaptation from Si to SiC processes. Growth rates of several mi-
crons per hour are reported for a temperature of 2000C [141,148]. The addition
of HCL to the process, and the use of ‘warm wall’ CVD reactors can increase
the growth rate to a few hundred microns per hour [149]. This makes the epi-
taxial growth of substrates financially viable; which is attractive because defects
in the substrate can cause defects in the epitaxial layers, thereby affecting device
performance.
130
CHAPTER 7
Micropipes were, until recently, a significant problem. A micropipe is thought
to develop when several screw dislocations occur in close proximity, if the Burgers
vector is sufficiently high, the volume enclosed by the dislocations will be empty,
hence a pipe [150]. The diameter is in the region of a micrometer. The epitaxial
layers are of considerably higher quality, in terms of defects/cm2, than substrates.
Electronic devices are manufactured by standard methods from the epitaxial lay-
ers i.e. photo-lithography, dry etching and evaporation of metal contacts [151].
Wet etching is troublesome because SiC is chemically quite inert.
7.1.3 Doping
Doping of semiconductors is often achieved by diffusion of dopants. SiC dopants
tend not to diffuse easily below 1800C [142], which is favourable for high tem-
perature operation but a hindrance for device production. Wide bandgap semi-
conductors can be difficult to dope. This is due to the large ionisation energies,
and compensation effects of most substitutional impurities. The main n and p-
type dopants in SiC are nitrogen and aluminium respectively. Nitrogen creates
a relatively shallow donor level in the band-gap (50 ± 20 meV) [152], and is an
excellent donor atom. However aluminium creates a deeper acceptor level (220
± 20 meV) [152]. Aluminium doping provides acceptable p type layers. A wide
doping range is possible for both n and p type layers. Doping of a whole epi-
layer is achieved by adding dopants during CVD. Commercial layers produced
by CVD are available with doping in the range 9 x 1014 to 1 x 1019 cm−3. It is
not possible to selectively dope parts of the substrate during growth. Therefore,
ion implantation is sometimes used, which causes damage to the crystal struc-
ture. The damage is partially repaired by high temperature annealing. When the
temperature used for the implantation is high the lattice can self heal to some
degree. Ion implantation is only possible to a depth of ∼1 µm. Selective doping
by ion implantation presently produces inferior quality layers.
7.1.4 Commercial Wafers
In 1989 Cree Inc. announced the commercialisation of 2.5” SiC substrates.
150 mm SiC wafers are presently available with epitaxial layers laid down to
131
CHAPTER 7
the purchasers’ specification. In May 2007 Cree announced the commercial pro-
duction of 4” wafers with zero micropipes [153]. Other manufacturers include
Dow Corning (formally Sterling Semiconductor), II-VI, SiCrystal and Nippon
Steel, SemiSouth and United Silicon Carbide.
7.1.5 Application: Covert NLOS Communications
Non line of sight communication schemes have been considered for at least forty
years [154]. Some advantages are,
• Immunity to RF propagation effects which hinder or preclude the use of
short range RF links in built up areas.
• A very large spectrum is available, without licensing requirements.
• The possibility of considerably higher channel capacity than presently used
RF systems.
• Undetectable at ‘stand off’ ranges, due to atmospheric absorption.
• Negligible solar background in the UV-C band (200 nm – 280 nm).
• Low susceptibility to ‘jamming’.
Figure 7.1 shows a schematic of an example channel geometry. The principle of
operation is as follows: a transmitter, sighted at distance from a receiver, emits
UV radiation into the atmosphere. Some of this radiation interacts with the
molecules and aerosols in the atmosphere, by Rayleigh scattering and absorption
respectively. A portion of the scattered radiation falls onto the receiving device
and is converted into an electrical signal. A ‘single scatter model’, such as the
one proposed by Luettgen [155], limits a photon to one scattering event, during
its transit from transmitter to receiver.
The number of scattering events encountered by a photon is not limited.
However the simplifying assumption, that each photon is only scattered once, is
valid when the product of the transmitter receiver distance, and the scattering
coefficient, ks, is less than or equal to 0.1 [156]. The scattering coefficient is a
132
CHAPTER 7
X
ZF1
ReceiverF2
Transmitter
βRθR βT
θR
ξmin
ξmaxScatteringVolume
Figure 7.1 – A schematic of an NLOS transmission channel, redrawn from [155].
weak function of wavelength. Taking 274 nm as an example ks = 5 x 10−4 m−1,
the valid range of the model is 200 meters.
Although several NLOS ‘test beds’ have been reported including [157, 158],
Silicon Carbide has been largely ignored as a detector technology. No practical
work has been reported on NLOS systems using a Silicon Carbide detector; such
work would be of considerable benefit.
7.1.6 Application: Water Purification
Water purification systems based on UV have been in use for at least one hundred
years [159]. Methods of purification may take several forms in addition to UV
irradiation, including application of gaseous chlorine or hypochlorite to the water
supply. UV purification systems based on vacuum device technology are com-
mercially available and are in widespread use [160]. Recently, several researchers
have published test-bed systems in which the UV lamp is replaced by an array of
UV LEDs [161–164]. UV LEDs have the usual advantages that semiconductors
have over vacuum devices, most notably,
• Reduced power consumption / increased efficiency
• Mechanical ruggedness
133
CHAPTER 7
• Longer operating life
• Simpler electronics required to drive the device
The ruggedness combined with the long operating life make the LED a par-
ticularly suitable choice for field operation, in developing countries, for example.
It has been estimated that 1.6 million deaths were caused throughout the world
by water-borne diseases in 1999 alone [165]. The application of Silicon Carbide
detectors to the design of these systems is in an equivalent role to the photo-diode
in light output stabilisation of a laser. The photo-diode monitors the UV radi-
ance in the treatment chamber and controls the emitter characteristics – current
is an obvious choice – to produce the desired dose (energy per unit surface area).
It is also possible to prevent water leaving the device untreated in the event of
emitter failure. In standard UV lamp systems a photo-diode may be used to
similar effect, however because the lamp life is generally short – they are replaced
yearly – there may be no monitoring of the lamp output at all.
7.1.7 Application: SiC Power Electronic Devices
Silicon Carbide has been lauded for some time as a replacement for Silicon in
power semiconductor applications [166]. SemiSouth has significant interests in the
power SiC market and has obtained at least eleven patents (for example [167]) for
device designs and processing techniques during the last decade. These have been
awarded in areas related to the production of SiC SBDs and JEFTs. Recently -
February 2011 [168,169] - SemiSouth began shipping normally-off power JFETs,
which are capable of greater power dissipation and higher operating temperatures
than a similarly sized Silicon MOSFET. However the potential of SiC is yet to be
fully exploited. Close inspection of the published characteristics [170] shows that
the material is imperfect, the drain - source leakage currents, especially at elevated
temperatures is significant ∼ 200 µA max at 1.7 kV. Furthermore the trench
JFET design appears to make heat removal from the device problematic. In the
high blocking voltage design TJC = 2.6C/W. The 1.2 kV blocking voltage design
is similarly specified to a Si Power MOSFET. The device package – TO247 – is
134
CHAPTER 7
not a limiting factor of heat removal in this design at present1, lending weight to
the argument that there is still room for improvement in the commercial devices.
Cree are noted as field leaders in the production of SiC wafers, optoelectronic
emitters based on SiC substrates and power and RF devices. Cree have pro-
duced zero micropipe 4” wafers since 2007 [153], SemiSouth presently market
a ‘premium’ grade 3” wafer with 15 micropipes / cm2 [171]. Cree also has in-
terests in RF and Power SiC products. They presently produce more than 20
SBDs for power electronics applications. A recent - January 2011 - addition to
the Cree product line is an enhancement mode SiC MOSFET [172]. Some of
the early power devices made in SiC have been Trench JFETs [173] however the
appearance of a commercial MOS transistor not surprising when it is realised
that Cree patented a method of establishing an oxide layer on Silicon Carbide
in 2005 [174] and several methods for forming a MOS device in SiC, the most
relevant of which appears to be [175]. The Cree device is generally of higher
specification than the SemiSouth devices. Transconductance is similar to Silicon
MOSFETs of lower specifications so large (>20V) – by Si standards – gate drive
voltages are required, the gate charge is small by Si standards however. The
transconductance is approximately two orders of magnitude less than a Si BJT
at a collector / channel current of 20 A. Several reports of SiC BJTs exist in the
academic literature [176, 177].
7.2 Review of Impact Ionisation in Silicon Car-
bide
Work on impact ionisation in Silicon Carbide was initially hampered by material
problems [178–180]. New device fabrication techniques have been developed to
permit the manufacture of devices which are capable of bulk breakdown for exam-
ple, positive edge termination [181], the use of a guard ring [182] and multi-step
junction edge termination [183]. It has been demonstrated that micropipes and
dislocations can adversely affect the electrical properties of devices [184, 185].
The earliest measurements on 4H-SiC, conducted on devices exhibiting uni-
1A sample of package vendors give the minimum TJC for the T0247 package as 0.4C/W
135
CHAPTER 7
form breakdown characteristics was reported by Konstantinov [186, 187]. Mul-
tiplication characteristics of three thick n−p+ mesa diodes with doping varying
from 5.2 x 1016 cm−3 to 1.2 x 1018 cm−3 were reported. DC photo multiplication
was measured with the device excited by 325 nm light. The ionisation coefficients
were obtained in the form proposed by Thornber [188] by fitting a local model
to each of the multiplication characteristics. Konstantiov et al. showed that the
impact ionisation in holes is considerably greater than for electrons.
A pulsed electron beam induced current and phase sensitive detection was
used by Raghunathan et al. [189, 189] to obtain a value for beta over a range
of temperatures from Shottky barrier diodes. Raghunathan et al. assumed α is
negligible. In the range of electric fields close to breakdown carrier feedback is an
essential process, this is evidenced by the super–linear increase in multiplication
with bias close to breakdown.
Yan et al. [190] reported noise measurements in which a p+pnn+ reach through
photo-diode was excited with 325 nm yielding k = 0.1. Ng et al. [191] reported
short (230 nm) and long (365 nm) wavelength multiplication and excess noise
results on two thin devices (w of 0.105 µm and 0.285 µm). The light was injected
from the p side of both samples. The ionisation coefficients were obtained by fit-
ting the data to a non-local RPL model [18]. Non uniform optical generation and
non-uniform electric fields were accommodated in this model. Agreement with
the experimental data was found by using electron and hole ionisation thresh-
old energies of 12 eV and 8 eV respectively. Multiplication increased and excess
noise decreased as the wavelength increased and carrier injection became more
hole dominated. The lowest excess noise of k = 0.1 and k = 0.15 for device
widths of 0.105 µm and 0.285 µm respectively was observed at 365 nm.
Hatakeyama et al. [192] reported the anisotropy of the impact ionisation co-
efficients on the <0001> and <1120 > faces. They used CW optical excitation
at 350 nm and so measured DC photo-multiplication. The impact ionisation co-
efficients were established by fitting the multiplication verses bias voltage, and
breakdown voltage versus doping density.
Guo et al. [193] measured a n+n−n−p+ SAM-APD at 351/363 nm and 270 nm
achieving a k of 0.2 and 0.3 respectively. The device thickness in this case was
120 nm. Guo et al [194] also reported a p+ppn+ SAM-APD having an intrinsic
136
CHAPTER 7
region of 180 nm, which was measured using 363 nm light. The authors report k
of 0.12 for this device.
Loh et al. [195] reported photo-multiplication measurements and impact ion-
isation coefficients based on three structures including one sample which is used
in this work. The measurements were carried out using 244 nm and 325 nm light.
A local model was used to extract the ionisation coefficients from multiplication
versus bias data. By accounting for the mixed injection of the 325 nm light in
several nip structures The authors were able to estimate the aggregate value of
α as approximately twenty five times less than β, k = 0.04.
The temperature dependence of the impact ionisation mechanism in 4H-SiC
has been investigated by Konstantinov et al. [187]. They show that α increases
whilst β decreases with temperature. A similar effect is seen in the 6H polytype.
The movement of the breakdown voltage as a function of temperature in 4H-
SiC is determined by holes, not electrons as in 6H-SiC. This is a result of the
greater enhancement of β with temperature than α. The hole impact ionisation
coefficients in 4H-SiC and 6H-SiC SBDs were obtained by Raghunathan et al.
[196] for temperatures over the range 300 to 450 K using their P-EBIC technique.
The positive temperature coefficient of the breakdown voltage in 4H-SiC has been
confirmed by the work of Neudeck et al. [197] Lee et al. [198], Ozpineci et al. [199]
and Vassilevski [140]. The breakdown voltage of 4H-SiC increases linearly at a
rate of 16 mV/C up to 257C in Yan et al.’s work [200] and at 10 mV/C up to
200C in Guo et al.’s work [193]. This suggests that 4H-SiC has thermally stable
avalanche breakdown behaviour.
Despite the wide variation in the published impact ionisation coefficients in
4H-SiC as shown in Fig. 7.21, all the authors have demonstrated that β is much
greater than α especially at low fields. Konstantinov et al. [186, 187] have sug-
gested that the asymmetry in the impact ionisation coefficients, is a property
of devices in which the electric field is parallel to the c-axis direction (E‖c).Hatakeyama et al. [192] measured breakdown voltage in a <1120> face p+n diode
that was 60% of that on a <0001> face p+n diode confirming the large anisotropy
of the extracted impact ionisation coefficients. Anomalous breakdown fields were
also demonstrated by Nakamura et al. [201]. The breakdown fields along the
<1120> and <0338> plane were about 75% of that measured in the <0001>
137
CHAPTER 7
plane for 4H-SiC. The breakdown field exhibits a positive temperature coefficient
of 200 V/cmK, 600 V/cmK and 400 V/cmK for <0001>, <1120> and <0338>
plane diodes, respectively. Tanaka et al. [202] have observed larger leakage cur-
rents and lower breakdown voltages along the <1120> direction compared with
<0001> direction.
The anisotropy impact ionisation in 4H-SiC has been ascribed to the con-
duction band splitting into a number of mini bands. This occurs only along the
<0001> direction (c-axis) [203]. A second process thought to be responsible is
the high energy loss of hot electrons due to the Bragg reflections from the zone
edges inherent to non-cubic SiC. The splitting between the first and second mini-
conduction bands is sufficiently large to assume very few electrons will occupy the
higher mini-band. The electrons in the first mini-band can only initiate impact
ionisation by means of scattering by phonons or crystal imperfections [186, 187].
The impact ionisation initiated by electrons along the <0001> direction is greatly
suppressed, leading to a much larger breakdown field along <0001> in 4H-SiC.
A theoretical study of β in 4H-SiC was performed by Bellotti et al. [204]
in which a full band Monte Carlo simulation was reported. Their simulation
showed a strong anisotropy in hole ionisation. The range of values of β in the
direction perpendicular to c-axis (E⊥c) were found to be about 5 to 10 times
higher than those parallel to c-axis (E‖c) for particular field. This is a consequenceof the anisotropy of the Brillouin zone [204]. These differences in β introduce
considerable differences in the transport characteristics in different directions.
Agreement with Raghunathan et al.’s experimental results [196] was achieved
when inter-band tunnelling effects near the crossing and mixing points in the
valence bands was included in the model to account for the transport parallel to
the c-axis. Bertilsson et al. [205] have reported a two-dimensional drift-diffusion
model which was used in conjunction with the anisotropy in the hole ionisation
coefficients reported by Bellotti et al. [204] to explain the discrepancy in the
experimental results of Konstantinov et al. [187] and Raghunathan et al. [196].
The authors also suggest that the anisotropy of β is greater at lower fields and
reduces with increasing electric field strength.
138
CHAPTER 7
7.3 Chapter Objective
In this work confirmation of the published ionisation coefficients is sought by
measuring photo multiplication and excess noise on two thick 4H-SiC structures.
There is considerable disparity between the published ionisation coefficients of 4H-
SiC [186, 189, 191, 192, 195, 204].The variation is most pronounced at low electric
fields such as those more typically found in power electronic devices. The diodes
used in this work were diodes were produced primarily for other applications but
were generously provided by The Semiconductor Technology Laboratory which
is part of General Electric Global Research. All of the wafers and epitaxial
layers were grown 8 off the c-axis <0001> using metal-organic chemical vapour
deposition (MoCVD), the layers were purchased from Cree Research (Durham,
NC, USA).
7.4 Structure Details
Two samples are used in this work the schematic cross sectional view of sample
B, a 4H-SiC non-reach-through SAM-APD, is shown in Fig. 7.2. The structure
consists of a 0.2 µm n+ cap layer (ND = 4 x 1018 cm−3), a 1.35 µm n− absorption
layer (ND = 7.5 x 1015 cm−3), a 0.45 nm n+ charge layer (ND = 3 x 1017 cm−3),
a 2.69 µm n− multiplication layer (ND = 7.5 x 1015 cm−3), and a 2 µm p+ layer
(NA = 2 x 1018 cm−3) grown on an n-type substrate. These devices are effectively
0.66 mm x 0.66 mm square mesa diodes, and have a positively bevelled edge. The
diodes were etched using an inductively coupled plasma (ICP) etching process and
Figure 7.2 – Sample B layer details.
139
CHAPTER 7
Figure 7.3 – Sample A layer details.
passivated using a thermal SiO2 film with an additional plasma enhanced chemical
vapour deposition SiO2 film on top. Optical access windows were formed in the
anode contact. Cathode contacts were fabricated by sputtering Ni, whilst the
anode contacts were formed by depositing a layered stack of Ti/Al/Ti/Ni. An
earlier version of the fabrication process is described in more detail by Yan et
al. [206]. Similar methods were employed in the production of sample A. A
schematic cross-sectional view of sample A is shown in Fig. 7.3. The structure
is composed of a 0.19 µm n+ cap layer (ND = 3.5 x 1018 cm−3), a 0.66 µm n−
absorption layer (ND = 2 x 1016 cm−3), a 0.197 µm n+ charge layer (ND = 6.7
x 1017 cm−3), an n− 0.582 µm multiplication layer (ND = 5 x 1015 cm−3), a
1.87 µm p+ layer (NA = 2 x 1018 cm−3), a 0.457 µm n+ buffer layer (ND =
1 x 1018 cm−3) grown on an n+ substrate. These devices are fabricated as a
square array of 16 circular mesa diodes. The effective diameter of the junction is
106.6 µm.
7.5 Capacitance Voltage Characteristics
Capacitance voltage characteristics were measured for sample A using a HP4275A
LCR meter coupled to a Keithly 237 SMU. Sample B was measured using an
Agilent B1500A semiconductor analyser. The layer widths and doping densities of
each layer were established by fitting capacitance voltage measurements to a one
dimensional Poisson model which is described in Appendix A. The capacitance
140
CHAPTER 7
Figure 7.4 – Capacitance voltage fitting of Sample A. Black circles represent measured datawhile the red line is developed from a one dimensional Poisson fitting algorithm.
Figure 7.5 – Capacitance voltage fitting of Sample B. Black circles represent measured datawhile the green line is developed from a one dimensional Poisson fitting algorithm.
141
CHAPTER 7
voltage fitting of samples A and B are shown in Figs. 7.4 and 7.5 respectively.
Sample B does not punch through its charge layer from 0 to 600 V bias. Be-
tween 600 V and breakdown at approximately 625 V the collection efficiency of
244 nm does not change by a measurable amount. Sample B is modelled using a
three region approximation. The fitted doping densities and region widths sug-
gest that the electric field will not punch through the charge layer before 772 V,
considerably above breakdown. The n− region is fully depleted by 50 V. An ac-
ceptable fit to the CV curve of sample B is given by assuming an n+n−p+ structure
0.45 nm n+ charge layer (ND = 3 x 1017 cm−3), a 2.6935 µm n− multiplication
layer (ND = 7.5 x 1015 cm−3), and a 2 µm p+ layer (NA = 2 x 1018 cm−3).
In sample A at least four, perhaps five, regions are partially depleted prior
to breakdown at approximately 193 V. Punch through of the multiplication layer
occurs at a few Volts. Depletion of the absorption layer begins around 160 Volts
and appears to be complete by 190 V however fitting doping densities and re-
gion widths under this interpretation leads to an anomalous breakdown voltage
value when using the coefficients of Loh et al. [195] and Ng [191]. Fitting accu-
rately in this high voltage region also leads to doping densities and/or depletion
widths which are incompatible with SIMS data. The high series resistance of this
layer coupled with the small depletion capacitance (∼ 0.5 pF) provides sufficient
uncertainty to overlook the lack of perfect agreement in the CV fitting close to
breakdown. It will be shown shortly that the series resistance in this layer is
approximately 10 kΩ. The overall error of the last few measured data points is
a maximum of 43 fF greater than the modelled result. Errors of this order can
be ascribed to the movement of the probe towards the copper plate which holds
the device after calibration of the CV test setup and similar usually negligible
systematic errors.
The electric field associated with samples A and B at 192.3V and 620V re-
spectively are shown in Figs. 7.6 and 7.7. In these figures the optical generation
and ionisation coefficient profiles are also shown, these are discussed throughout
the chapter.
142
CHAPTER 7
Figure 7.6 – Several parameters of Sample A as a function of distance at 192.3 V reverse bias.Electric field (black, solid), electron ionisation coefficient α (red, dash–dot), hole ionisationcoefficient β (violet dash–dot–dot), absorption of 244 nm light shone onto the surface of the n+
capping layer (blue, long dash) and absorption of similarly injected 325 nm light (dark green,medium dash). Light enters the n+ layer from the right of the figure.
Figure 7.7 – Several parameters of Sample B as a function of distance at 620 V reverse bias.Electric field (black), electron ionisation coefficient α (red, dash–dot), hole ionisation coefficientβ (violet dash–dot–dot), absorption of 244 nm light shone onto the surface of the n+ cappinglayer (blue, long dash) and absorption of similarly injected 325 nm light (dark green, mediumdash).
143
CHAPTER 7
Figure 7.8 – Sample A forward bias for the sixteen samples in the array (solid lines). Longdashed line shows an ideal diode with ideality factor of two. The medium dashed line shows asimilar diode with Rs = 10 kΩ.
7.6 Current Voltage Characteristics
Current voltage (IV) measurements were performed without room lighting (flo-
rescent tubes), however no difference could be found between IV curves taken
under room lighting and those obtained in darkness. Forward IV measurements
were taken using an HP 4140B Picoammeter. Reverse IV measurements on sam-
ple A were taken using an Agilent B1500A. Sample B reverse characteristics were
obtained with a Keithly 237.
Forward bias measurements were undertaken on all devices in both sample
sets; sixteen devices of sample A and twelve devices of sample B. Data relating
to samples A and B are shown in Figs. 7.8 and 7.9. The measurement range
is limited to a maximum current of 100 µA. Currents greater than this value
are known to deleteriously reduce the reliability and lifetime of these devices.1
Because the series resistance of sample B is modest, one device was sacrificed in
order to measure higher currents and approximate the series resistance. Series
resistance in sample A fits approximately 10 kΩ, in sample B, 130 Ω is the ap-
propriate value, see Figs. 7.8, 7.9 and 7.10. The analytic method for modelling
1P. M. Sandvik, private communication 2009. Also see US Patent No. 7,002,156.
144
CHAPTER 7
Figure 7.9 – Sample B forward Bias for all samples. Other than the two devices whichwere dead on arrival (pink and dark yellow) the devices show highly consistent forward biascharacteristics. The black line represents an ideality factor of two.
a forward biased diode and series resistance yields a solution which requires a
numerical analysis because it contains a Lambert W function. The two stage
fitting procedure described in Chapter 2 was used to obtain the saturation cur-
rent, Is, emission coefficient, n, and series resistance, Rs. The fitted emission
coefficient of two indicates that majority carrier recombination within the de-
pletion region dominates the transport [179, 207]. It is impossible to establish
the relative distribution of current between bulk and surface recombination in
these layers because all devices are of the same area. The reverse bias current
voltage characteristics of sample A are shown in Fig. 7.12, and are measurement
system limited up to breakdown for twelve devices. All devices show generally
abrupt breakdown characteristics. Current rises from the near the measurement
limit by at least two orders of magnitude over a few volts reverse bias. The
inconsistency in breakdown observed in sample A may be ascribed to the large
and quite variable series resistance of each diode. Diodes with negligible series
resistance fabricated on a localised area of a wafer may be expected to exhibit
identical breakdown behaviour under ideal conditions. Several diodes exhibit real
dark current levels above the measurement floor prior to breakdown (for example
the grey line in Fig. 7.11), this may be ascribed to higher levels of trap assisted
145
CHAPTER 7
Figure 7.10 – Sample B high current forward bias characteristic. Black circles represent data,red line represents an ideal diode with 130 Ω series resistance.
thermal generation in these devices than in those surrounding them.
The reverse bias current voltage characteristic of sample B is shown in Fig. 7.12,
in which two 244 nm photo-current plots are also presented. The dark current
is measurement limited up to breakdown. The true dark current is much lower
than this measurement suggests. The curve shown does not appear to exhibit
impact ionisation, unless the impact ionisation is purely electron initiated, in
which case it would not be apparent until very close to breakdown. However it
has been shown by several workers including Konstantinov [186,187], Ng [18,191]
and Loh [195] that β > α in 4H-SiC hence a generally hole dominated dark current
multiplication curve may be expected. The two pure hole injection photo-current
plots show the likely shape of the real dark current prior to breakdown. Measure-
ments taken with a HP 4140B Picoammeter up to 50 Volts reverse bias which is
the maximum for this measurement equipment, show dark currents near to the
measurement limit of 1 pA for most devices. It is possible that this too is an
overestimation of the dark current. Generally extremely low dark currents are
expected in SiC due to the very low intrinsic carrier density (∼ 10−9 cm−3 [152])
which is a function of band gap. The ideal generation current would be negligible.
146
CHAPTER 7
Figure 7.11 – Sample A reverse current voltage characteristics for all 16 devices.
Figure 7.12 – Sample B reverse current voltage characteristics for one device (black, solid).Photo-current due to 244 nm injection is shown for two optical intensities (red, long dashand blue, medium dash). The optical intensity of the blue line is approximately an order ofmagnitude greater than the red line.
147
CHAPTER 7
Figure 7.13 – Light absorption as a function of wavelength [208,209] for 4H-SiC (black circles),6H-SiC (blue circles) and the extrapolation due to Ng [18].
7.7 Photo-Multiplication Characteristics
The multiplication and excess noise characteristics were measured simultaneously
using the measurement system after Li et al. [27] which is described in Chapter 2.
The multiplication is measured by an AC (LIA) measurement of photo-current
passed through a transimpedance amplifier. The uniformity of several diodes
were checked prior to multiplication and noise measurements. This is accom-
plished by scanning the devices with 325 nm light with a low reverse bias (-5V)
applied. The spot size is focused such that it is significantly smaller than the
device area. Under these conditions a uniform device presents a ‘flat top’ re-
sponse. The method is similar to that described by Guo et al. [193]. Uniformity
between devices is confirmed by the grouping of the excess noise measurements.
Excess noise is approximately proportional to M2 or M3 when β = α [28]. Mul-
tiplication differences due to non-uniformity may be expected to exhibit a strong
spread in excess noise results. However, the spread in excess noise is small for
several devices for multiplication values up to approximately two hundred and
fifty. A typical reverse bias photo-current and dark current curve of sample B
is shown in Fig. 7.12. The photo-current is several orders of magnitude greater
than the measured dark current, the multiplication due to the two levels of light
148
CHAPTER 7
Figure 7.14 – Relative light absorption in 4H-SiC at several wavelengths. Black – 230 nm,red – 244 nm, blue – 280 nm, pink – 325 nm, green – 360 nm.
intensity shown are similar for a prescribed reverse bias voltage. This implies that
heating and space charge effects have not influenced the results. The multiplica-
tion characteristics of samples A and B are shown in Figs. 7.16 and 7.18. The
multiplication factor increases substantially as breakdown is approached. The
breakdown voltage is fitted using Miller’s empirical expression [5],
M =1
1−(
VVbd
)n (7.1)
where M is the multiplication at a reverse bias voltage V . Vbd is the breakdown
voltage and n is a fitting parameter which is proportional to the ‘sharpness’ of
the breakdown. The breakdown voltage of Sample A is 193.78 V and sample B is
624.60 V. The measured breakdown voltage of sample A is hampered by the large
series resistance, however agreement is adequate. The agreement for sample B is
good. In this work the 4H-SiC absorption coefficients of Sridhara et al. [208,209]
are used and the extrapolation due to Ng [18] is also employed. Both are shown
with 6H-SiC data in Fig. 7.13. In sample A approximately 97% of 244 nm light
generates e-h pairs between the window and the high field region contributing to
pure hole injection. Practically all of the remaining 3% generates carriers inside
the depletion region contributing a mixed injection component, the proportion of
149
CHAPTER 7
Figure 7.15 – Sample A multiplication characteristics as a function of reverse bias for 244 nm(circles) and 325 nm (squares). The black solid line represents pure hole multiplication, blue– 244 nm mixed injection, green – 325 nm mixed injection and red represents pure electroninjection.
244 nm generating carriers in the p+ layer is negligible. At 325 nm the proportion
of light generating carriers prior to the electric field is approximately 12%, a
further 9% generates carriers inside the depletion region. At the end of the p+
layer nearly 38% of all the incident light will have been absorbed. Under 325 nm
injection the primary carrier conditions are split quite evenly between generation
in the high field region and diffusion/drift of electrons and holes into the high field
regions from the p+ layer and n layers respectively. In performing this calculation
it has been assumed that carriers generated within the n+ capping layer, the
absorption region and half of the charge layer contribute to pure hole injection.
This is plausible because the electrons which are produced in the generation of
carriers in this region have practically no chance of ionising the lattice prior to
leaving the electric field region because they will not experience the high field
of the multiplication layer. Holes generated in the n+ cap, absorber and half of
the charge layer will travel towards the multiplication region and so contribute to
pure hole injection. Despite the relatively even injection conditions in this device
is clear from the measured data that holes dominate the ionisation process in this
device at 244 nm and at 325 nm. In sample B at 244 nm >99.9% of the light is
absorbed in the cladding layers outside the electric field. The remaining <0.1%
150
CHAPTER 7
Figure 7.16 – Sample A multiplication characteristics as a function of reverse bias for 244 nm(circles) and 325 nm (squares). The black solid line represents pure hole multiplication, blue– 244 nm mixed injection, green – 325 nm mixed injection and red represents pure electroninjection.
is absorbed within the depletion region. This device may be assumed to exhibit
pure hole injection at 244 nm. At 325 nm carriers are generated throughout the
structure. In the n+ cladding layers approximately 23% of the light is absorbed,
in the depletion region approximately 23% is absorbed. By the end of the p+
layer around 60% of the 325 nm light is absorbed.
Multiplication data at 244 nm and 325 nm for samples A and B versus reverse
bias voltage are shown in Figs. 7.18 and 7.16 where M − 1 is plotted to expose
the low multiplication values. Figs. 7.17 and 7.15 show the multiplication versus
reverse bias in the region near to breakdown.
While mixed injection prevails, the difference in multiplication characteristics
under short and long wavelength injection conditions is not very great. This is
due to the overwhelming propensity of holes to ionise the lattice compared to
electrons. Even a small quantity of holes ionising the lattice can produce long
ionisation chains which cause electron dominated optical generation conditions
to produce hole dominated multiplication characteristics. This is demonstrated
by the positioning of the 325 nm curve close to the pure injection 244 nm curve
compared to the predicted pure electron curve in both devices.
151
CHAPTER 7
Figure 7.17 – Sample B multiplication characteristics as a function of reverse bias for 244 nm(circles) and 325 nm (squares). The black solid line represents pure hole multiplication, blue– 244 nm mixed injection, green – 325 nm mixed injection and red represents pure electroninjection.
Figure 7.18 – Sample B multiplication characteristics as a function of reverse bias for 244 nm(circles) and 325 nm (squares). The black solid line represents pure hole multiplication, blue– 244 nm mixed injection, green – 325 nm mixed injection and red represents pure electroninjection.
152
CHAPTER 7
7.8 Excess Noise Characteristics
The excess noise characteristics are of considerable interest primarily because
pure injection data is available for the first time, secondly because sample B
is relatively thick compared to the majority of devices for which excess noise
results have been published. The thickest device published thus far is by Ng et
al. [18, 191], the intrinsic width is 285 nm; sample B is nearly ten times thicker.
Figs. 7.19 and 7.20 show excess noise characteristics produced by injection of
244 nm and 325 nm light into samples A and B respectively. The excess noise
data corresponds to the multiplication data shown in Figs. 7.15, 7.16, 7.17 and
7.18. Due to the large series resistance of sample A excess noise cannot be ac-
curately measured while the multiplied photo-current causes the series resistance
to significantly affect the device multiplication characteristics. Good agreement
can be seen between the data taken for both wavelengths especially when the
low quantity of noise is considered. In sample A 244 nm injection corresponds
to McIntyre’s effective k of 0.1, and 325 nm injection corresponds to 0.45. In
sample B 244 nm injection corresponds to McIntyre’s effective k of 0.007 while
325 nm injection corresponds to 0.045. The 244 nm injection excess noise data
for sample B falls below F = 2 in some cases. This is especially noted in the
black circles and green circles where higher starting photo-current was used to
ensure an accurate measurement of noise. It is possible that there is a slight cal-
ibration error, F axis offset is a function of the disparity between the calibration
shot noise and the APD noise either due to uncertainty in the calibration noise
or APD noise measurement or to uncertainty in the bandwidth calibration of the
measurement system for the calibration device capacitance or the APD capaci-
tance. To mitigate this as far as possible sample B was tested at unity gain with
varying optical power in order to produce a shot noise calibration. This calibra-
tion was compared and found to be very nearly identical to a UV enhanced Silicon
reference diode (Perkin Elmer UV–BQ–40) when adjusted for capacitance effects.
Consequently sample B shot noise data was used as the calibration for Figs. 7.19
and 7.20, and as a result no bandwidth corrections were necessary. Despite this
careful checking, a slight error remains. It may be suggested that, because α is
small, there must be many hole ionisations in each transit of the device, if the
153
CHAPTER 7
Figure 7.19 – Sample A excess noise characteristics as a function of multiplication. Circlesrepresent 244 nm data, squares – 325 nm. Black solid line – pure hole injection, blue solid line– 244 nm mixed injection, green medium dashed line – 325 nm mixed injection and the red longdashed line represents pure electron injection. Grey lines represent McIntyre’s lines of constanteffective k from 0 to 0.5 in 0.05 steps.
Figure 7.20 – Sample B excess noise characteristics as a function of multiplication. Circlesrepresent 244 nm data, squares – 325 nm. Black solid line – pure hole injected, blue solid line– 244 nm mixed injection, green medium dashed line – 325 nm mixed injection and the redlong dashed line represents pure electron injection. The grey lines represent McIntyre’s lines ofconstant effective k from 0 to 0.1 in 0.01 steps.
154
CHAPTER 7
number of ionisations required per transit forces each carrier to ionise directly
after the expiration of the dead space some grouping of carriers would result and
F < 2 is possible. This effect is unlikely to be seen at higher multiplication be-
cause α and β tend to converge as electric field increases. The increasing carrier
feedback would permit long hole ionisation chains to begin at random positions
in the lattice due to ionisation of the lattice caused by relatively few electrons,
thus breaking up the grouping of carriers. It is just as likely, however, looking
at the red circles, that the true curve falls on F = 2 at multiplications less than
forty and that the slightly low result is a systematic error.
7.9 Ionisation Coefficient Extraction
Ionisation coefficients have been extracted from the multiplication and excess
noise data gathered from samples A and B using the local framework proposed
by McIntyre [3]. This extraction was performed partially by extraction from mul-
tiplication and excess noise data and thereafter by trial and correction approach
the main steps of the extraction method are outlined below.
• Using the ionisation coefficients of Ng and Loh, fit the measured breakdown
voltage of a nip / pin (sample B) to that of a device possessing a flat electric
field profile by adjusting the intrinsic region width, w, of the device.
• Using this fitted intrinsic width, extract ionisation coefficients from the data
by use of the expressions given in Section 1.6.
• By trial and correction adjust the coefficients such that they fit the multi-
plication and excess noise data of Sample B using the CV fitted (piecewise
linear) region widths and doping densities.
• Use the extracted coefficients to model Sample A.
• Iteratively adjust the coefficients until both samples A and B fit the multi-
plication and excess noise data within permissible tolerances.
This process is quite slow and the fitting shown in this chapter represents ap-
proximately forty hours work completed over three days. The fitted coefficients
155
CHAPTER 7
are shown plotted in Fig. 7.21 with the coefficients of Ng [191] Loh et al. [195]
and Konstantinov et al. [186] Multiplication and excess noise data produced us-
ing these coefficients is shown as lines in Figs. 7.15, 7.16, 7.19, 7.17, 7.18 and
7.20. The ionisation coefficients can be expressed using Chynoweth’s expression.
The ionisation coefficient for electrons, α is separated into a low field and high
field regions to improve the fitting of Chynoweth’s expression to the extracted
coefficients. The low field and high field α expressions overlap across a relatively
broad range of electric filed values around 2.5 MV/cm. The expressions for alpha
and beta are given in (7.2), (7.3) and (7.4).
αl = 1.933× 104 · exp[
−(2.888× 106
ξ
)]4.828
(7.2)
αh = 1.878× 106 · exp[
−(9.134× 106
ξ
)]1.459
(7.3)
β = 6.000× 106 · exp[
−(1.387× 106
ξ
)]0.9600
(7.4)
Figure 7.21 – Ionisation coefficients of 4H-SiC versus inverse electric field published by variousworkers including those extracted in this work. Bellotti et al. [204] (black squares), Konstantinovet al. [186] (black, short dashed lines), Ng et al. [191] (β – dark red triangles, α – dark bluetriangles), Loh et al. [195] (green, long dashed lines) and this work (β – red circles, α – bluecircles).
156
CHAPTER 7
7.10 Comparison of GaN and SiC: GaN and Al-
GaN Detectors
Wide band gap semiconductors promise mechanical ruggedness, high breakdown
fields and solar blindness. The main contenders in wide band gap detectors are
summarised in [210]. Gallium Nitride is a wide, direct band gap semiconductor
(3.4 eV), with high breakdown field and high electron mobility. This offers the
possibility of very wide bandwidth devices, with high responsivity (due to the
direct band gap). GaN has excellent mechanical properties and is chemically quite
inert. It is sometimes reported that GaN is intrinsically solar blind. However,
there is significant background radiation at sea level, around 362 nm, which is
the cut off wavelength of GaN. The ternary, AlGaN, can be made solar blind,
however, operating from 200 nm to 362 nm. The cut off wavelength can be
tuned by adjusting the fraction of the ternary alloy. A review of wavelength of
detection with composition is given in [139]. The following table shows a selection
of published work on AlGaN and GaN photo-diodes. The dark current density is
a figure of merit related to the quality of the material.
Dark Current Density Authors
2.33 to 4.65 mA/cm−2 at -150 V K. A. McIntosh et al., 1999 [211]
3.54 µA/cm−2 at -90 V B. Yang et al., 2000 [181]
5.556 mA/cm−2 at -30 V Carrano et al.,2000 [212]
20 mA/cm−2 at -100 V S. Verghese et al., 2001 [213]
230 uA/cm−2 at -5 V R. McClintock et al., 2004 [214]
1 mA/cm−2 at -70 V R. McClintock et al., 2005 [215]
10 µA/cm−2 at -45 V S-C. Shen et al., 2007 [216]
10 µA/cm−2 at -100 V Y. Yoshizumi et al., 2007 [217]
400nA/cm−2 at -15 V H. Jaing et al., 2007 [218]
24 µA/cm−2 at -60 V T.Tut et al., 2007 [219]
35.4 µA/cm−2 at -80 V J.B. Limb et al., 2008 [220]
45 µA/cm−2 at -80 V J. L. Pau et al., 2008 [221]
6.37 µA/cm−2 at -120V T. Tut et al., 2008 [222]
7.87 mA/cm−2 at -100 V Bayram et al., 2008 [223]
157
CHAPTER 7
Dark Current Density Authors
88 mA/cm−2 at -80 V Bayram et al., 2008 [224]
9.55 µA/cm−2 at 55 V L. Sun et al., 2010 [225]
133 µA/cm−2 at 100 V E. Cicek et al., 2010 [226]
In comparison 4H-SiC is mature and in high quality devices the dark current
is often hidden below the instrument limit [227], often less than 1 pA until very
close to breakdown. While GaN/AlGaN is presently well suited to production of
optoelectronic emitters it is unsuitable for production of devices which operate
under large electric fields such as APDs. High defect densities caused by the lack
of a native substrate limits the electric field to values below the bulk breakdown
level. At the present time usefulness of GaN/AlGaN as a detector remains limited
to the research environment.
7.11 Comparison of GaN and SiC: GaN Power
Electronics
It is predicted that GaN will eventually outperform Si and SiC power devices
[210, 228]. This is principally due to its larger critical electric field, low intrinsic
carrier concentration and high carrier mobility. However, the breakdown voltage
of physically realisable GaN devices has, thus far, been less than the predicted
value. The realisation of GaN’s bulk properties in working devices have been ham-
pered by the material problems and semiconductor processing [211,212,216,229]
The high defect density and number of dislocations found in the epitaxial layers
grown on lattice-mismatched Sapphire or SiC substrates leads to high leakage
current and premature microplasmic breakdown before the bulk breakdown field
is reached. Dmitriev et al. [229] have reported the breakdown of linearly graded
pn junctions in the electric field range of 1.5 to 3 MV/cm. Their devices have
reverse leakage current of 10 mA/cm−2. The pn junction breakdown exhibited a
positive temperature coefficient, but the breakdown was of a localised microplas-
mic nature due to defects. The progress of commercial GaN power switching
158
CHAPTER 7
devices compared to SiC can be loosely approximated by observing the patents
sought by Cree and their competitors as well as in the academic literature, for
example [230].
Irrespective of which material GaN or SiC wins the high power/high voltage
electronics ‘battle’, in which SiC presently has the upper hand, the two materials
are for the moment at least, bound together. SiC is a favourable substrate mate-
rial for growth of GaN epilayers. Cree, arguably the leader in SiC production, has
significant interests in the production of SiC wafers to manufacture GaN LED
products besides their SiC on SiC wafer/epi-layer production, power and RF de-
vice lines. If the outstanding GaN material problems can be solved it is possible
that the research effort expended on SiC growth will not have been wasted.
7.12 Conclusion
The avalanche multiplication and excess noise characteristics of two thick 4H-SiC
APDs with intrinsic region widths of 0.59 µm and 2.7 µm have been investigated
using 244 nm and 325 nm injection. For the first time pure hole injection multipli-
cation and excess noise data have been obtained in a SiC APD. The excess noise
characteristics of the 244 nm injection correspond to an effective k of 0.007 in the
case of sample B and 0.1 in the case of sample A. This unexpectedly low noise
in thicker devices has been used to extend the range of experimentally verified
ionisation coefficients into the low field region where these devices operate.
A local model has been used in conjunction with a principally trial and correc-
tion fitting approach to extract new ionisation coefficients which, while following
the work of Ng et al. [191] and Loh et al. [195] at high fields, properly reflect the
multiplication and very low noise obtained at lower electric fields.
159
CHAPTER 7
160
Chapter 8
An APD SPICE Model for
Circuit Simulation
This chapter presents a simple SPICE model for use in simulation of detector
circuitry. The model may be used to represent a real or fictitious device and
operates in large signal and small signal simulations. The source code is presented
in Appendix B.
8.1 Introduction
When simulating the electrical characteristics of new transimpedance amplifier
designs or other electro-optical systems it is often assumed that the photo-diode
detector is simply a perfect current generator. It is often desirable to ensure
that certain characteristics of real photo-diodes do not detrimentally influence
the circuit behaviour. In these cases the required parasitic components are often
added to the simulation, as deemed necessary, depending on the situation. A
more universal approach is desirable; to that end a - nearly - universal model of
a linear mode APD has been developed.
The main requirement of this model is to reproduce, in a SPICE simulator,
characteristics of a pre-existing semiconductor detector such that the predicted
performance of a proposed circuit may be evaluated when used with the given
photo-diode. The photo-diode being modelled will have been measured using
161
CHAPTER 8
standard characterisation tests including,
• Reverse bias current voltage characteristics (IV)
• Reverse bias capacitance voltage characteristics (CV)
• Reverse bias photo-multiplication (Multiplication)
• Excess noise factor (Noise)
• Responsivity over a range of wavelengths (responsivity)
Not all of these tests are necessary; it is perfectly valid to develop a model of a
fictitious diode. For example the instrumentation designer might like to simulate
how a new measuring system will respond to use with a diode possessing very high
tunnelling currents. In which case, some tunnelling characteristics which expose
the new measuring systems weaknesses may be contrived without reference to
a real photo-diode. In this discussion it is assumed that a real device is being
modelled and that it is desirable to replicate this device’s characteristics with
some accuracy.
The model must be efficient with the simulators resources, as the majority
of the memory and processing power should be available for the simulation of
the circuit not the APD. Several circuit based models are described in the liter-
ature which attempt to represent accurately the behaviour of an APD at high
frequencies [231–238]. This is challenging when – as in SPICE – only lumped cir-
cuit models are available. The number of lumped elements required for accurate
device modelling is often quite large. These models are separate from the local
model and from the RPL, recurrence and Monte-Carlo models which attempt
to model the semiconductor statistically. A phenomenological approach, which
lends itself to the type of model required has been sought.
162
CHAPTER 8
8.2 Characteristics of the Model
The model includes,
• Capacitance as a function of reverse bias
• Multiplication as a function of reverse bias
• Excess noise as a function of multiplication (local model)
• Series resistance
• Responsivity
• Bulk leakage current as a function of reverse bias
• Surface leakage current
• Tunnelling current as a function of reverse bias
• The gain bandwidth product of the APD
• The noise bandwidth of the APD (experimental)
All of these functions can have their variables parametrised by the calling
script. For example it is possible to supply a list of series resistances and observe
the effect these have on noise as a function of multiplication by running several
simulations (one at each resistor value) and collecting the data into one graph.
Similarly a table of responsivity, VBD, nm and k values could be provided to
model differing injection conditions which elicit different multiplication noise and
responsivity profiles. Examples of these methods are shown in section 8.4.
The model does not include,
• Geiger mode (SPAD) operation
• Transit time effects
• Stray inductance or capacitance and packaging parasitic components
163
CHAPTER 8
Table 8.1 – Node connections for the APD model.
Node Function
LIrradiance falling on the device in Watts. Use for illuminating the
device in .AC .TRAN .DC .OP .NOISEA Anode terminalC Cathode terminal
MA voltage representing the large signal multiplication. Use forobtaining a voltage which represents the photo-multiplication in.OP .DC .TRAN. This node is meaningless in .AC and .NOISE
FA voltage representing the large signal excess noise factor. Use forobtaining a voltage which represents the excess noise factor in.OP .DC .TRAN. This node is meaningless in .AC and .NOISE
CapA voltage representing the large signal parallel plate capacitanceof the device. Use for obtaining the same in .OP, .DC, .TRAN.
Meaningless in .AC and .NOISE
This model is not appropriate for Geiger mode operation; convergence problems
abound. Furthermore ‘M’ and ‘F’ are ill-defined in Geiger mode. Transit time
effects are not included. At the frequencies where these effects begin to dominate,
SPICE is unlikely to be an appropriate choice of circuit simulator because it is
parameter based, and is incapable of properly representing distributed effects.
More appropriate simulators exist including Microwave Office, Ansys HFSS and
Agilent ADS, which better represent the distributed high frequency effects using
methods such as numerical electromagnetic solvers. Similar arguments apply to
stray inductance and capacitance and packaging parasitics. The package para-
sitics can be added easily if desired. A simple transit time model which could be
integrated into this model was proposed by Jou [232].
8.3 Parameters of the Model
The node connections for the APD model are given in Table 8.1, the parameters
of the model are given in Table 8.2. The parameters and methods by which they
may be obtained are described below. It is necessary to supply all parameters.
164
CHAPTER 8
RL
L
−+
V1
B1C1 B2
D1 B5 B3 B4 B10
R2C
A
Q2
−+
V2
Q1
B6
−+
B7
Cap
−+
B8
M
−+
B9
F
100
Ci
200
201
Figure 8.1 – A schematic representation of the APD model. Node labels are internal to thesub-circuit and can be reused in a calling circuit. B1 represents the bulk dark current, B2
represents the primary photo-current and B5 represents the multiplied bulk dark and primaryphoto-current. B3 represents the tunnelling current, B4 provides the surface leakage currentand B10 provides the small signal noise. B7−9 supply the large signal and DC reverse biascapacitance, multiplication and excess noise factor. B6 and surrounding components producethe value for small signal noise. V1 is a zero volt DC source used to sum the currents from B1
and B2.
If, for example, tunnelling current does not apply to a particular device, set ITM
= 0 and ITC = -30. This will render the tunnelling current effectively switched
off for all practical currents.
8.3.1 Series Resistance
The device series resistance (RS) may be fitted to,
Ia = Is
(
expq (Va − IaRs)
n k T− 1
)
(8.1)
165
CHAPTER 8
Figure 8.2 – APD model parameters. Parameter names are typeset in agreement with thesource code in Appendix B.
Parameter Function
RS
Device series resistance. Obtain this by performing forward IVs upto high currents (100 mA) on a few sacrificial devices, and then fitthe data for RS in series with a diode approximated by Shockley’sdiode equation.
RResponsivity Amps/Watt. This can be obtained by measuring thedevice being modelled and a reference device using a monochroma-tor and white light source.
VBDThe breakdown voltage of the device. Obtain this by regressingphoto-multiplication data to Miller’s empirical expression [5].
NMThe multiplication exponent of Miller’s empirical expression. Ob-tained simultaneously with VBD.
CJMDevice reverse bias capacitance. This is the linear coefficient of1/C2(V) when using the linear regression method.
CJC y-axis intercept of 1/C2(V) in the linear regression method.
C0y-axis intercept of C(V) in the inverse second order polynomialregression of C(V).
aReciprocal coefficient of C(V) in the inverse second order polyno-mial regression of C(V).
bReciprocal squared coefficient of C(V) in the inverse second orderpolynomial regression of C(V).
IBULKBulk dark current at unity gain. Bulk dark current experiencesmultiplication.
ISURFSurface leakage current. Surface leakage does not experience mul-tiplication.
ITMLinear coefficient of the linear regression of the common logarithmof the tunnelling current.
ITCY-axis intercept of linear regression of the common logarithm ofthe tunnelling current.
k McIntyre’s effective k [3].
166
CHAPTER 8
Where Ia is the diode forward bias dark current, Is is the saturation current, q
is the electron charge, n is the emission coefficient, k is Boltzmann’s constant, T
is the absolute temperature of the junction, Rs is the series resistance and Va is
the voltage across the diode. The appearance of Ia on both sides of the equation
makes fitting analytically slightly complex, however it is possible by resorting to
Lambert’s W. An iterative, numerical, trial and correction method is often used.
8.3.2 Multiplication
Photo-multiplication is modelled by regressing measured multiplication data to
Millers empirical expression [5],
M =1
1−(
VVbd
)n (8.2)
where M is the photo-multiplication, V is the reverse bias voltage, Vbd is the
reverse breakdown voltage and n is a coefficient of curvature of the multiplication
profile. nm and Vbd are extracted from the regression.
8.3.3 Device Capacitance
Two methods are provided for modelling the reverse bias device capacitance as a
function of reverse bias voltage the first requires a linear regression of the reverse
bias capacitance voltage characteristic to,
C(V ) = cjm · V + cjc (8.3)
Where cjm and cjc are model parameters, V is the reverse bias voltage and C(V )
is the device capacitance as a function of reverse bias voltage. This regression is
often sufficient as the regressed capacitance need only fit the device capacitance
at higher reverse bias voltages.
The second method is a regression of C(V ) to a second order inverse polyno-
mial,
C(V ) =a
V 2+
b
V+ C0 (8.4)
167
CHAPTER 8
where C0, b and a are the parameters being extracted, and C(V ) is given above.
This regression requires an additional parameter but often fits a wider range of
the reverse bias capacitance – voltage characteristic.
8.3.4 Tunnelling Current
Band to band tunnelling current, which is the only tunnelling current modelled,
is implemented by regressing the common logarithm of the measured tunnelling
current to a linear expression,
IT = ITm · V + ITc (8.5)
where ITm and ITc are the model parameters being found, V is the reverse bias
voltage and IT is the measured tunnelling current. Tunnelling current does not
experience avalanche multiplication.
8.3.5 Bulk and Surface Leakage Currents
The bulk, Ibulk, and surface, Isurf , leakage currents can be estimated by observa-
tion of dark reverse current voltage characteristics. The bulk and surface currents
can only be separated if devices of differing area are available. Masks available
to the EPSRC National Centre for III-V Technologies often have 25, 50, 100 and
200 µm radius mesa diodes fabricated in close proximity to each other. For de-
vices of differing area, leakage current which is dominated by bulk processes is a
function of normalised device area. Therefore if the device dark current density
normalised to device area is constant the leakage current is bulk dominated. Sim-
ilarly surface leakage current dominates if the dark current density is constant
when normalised to device perimeter. Generally both processes are found in real
devices and these two terms may be fitted iteratively. In this model bulk leak-
age is bias dependant and experiences avalanche multiplication; surface leakage
current does not experience avalanche multiplication.
168
CHAPTER 8
8.4 Examples of the Model
In this section examples of the model are shown in LTSpice. The simulations
reproduce the characterisation experiments that were used to gather the data
necessary to produce the model. This is not representative of how the model is
intended to be used but shows example characteristics of a modelled device. The
model is principally that of a thin silicon pin diode, however to show tunnelling
in the model the tunnelling is taken from a similar thickness AlInAs device, as is
the device capacitance. The responsivity is not representative of any particular
device.
169
CHAPTER 8
8.4.1 DC Characteristics
Example DC characteristics are shown in Fig. 8.3. In this simulation, L, M, F,
Cap, the anode and cathode voltages, currents and device power dissipation can
be meaningfully interrogated.
Figure 8.3 – Top left: LTSpice schematic used to produce the DC characteristics. The APDis connected in a photo-current experiment where V2 is an SMU and V1 represents the opticalpower (Watts). V1 is parametrised between 0 and 100 V by the .step command. V2 is sweptfrom 0 to 13.5 V by the .dc command. Top right: Total current as a function of reverse biasvoltage with optical intensity as a parameter. It is the result of the DC sweep shown on theleft. Multiplication can be observed above 7 V. Tunnelling can be observed in the lower fourtraces. Bottom: Capacitance as a function of reverse bias voltage, obtained by plotting the‘voltage’ on the Cap node.
170
CHAPTER 8
8.4.2 Optical Characteristics
Disparate injection conditions can be simulated by the use of tables of parameters.
This is shown in Fig. 8.4
Figure 8.4 – Top left: An LTSpice schematic showing the use of tables of parameters to sim-ulate sets of several parameters. This permits simulation of several wavelengths illuminatingthe device. Reponsivity (R), the shape of the multiplication curve (nm) and the effective k(k) are parametrised and selected from three tables by the .step command. The APD param-eters are shown in blue text. Top Right: M-1 versus reverse bias voltage for several injectionconditions. Bottom Left: Excess noise factor, F as a function of multiplication, M for severalinjection wavelengths. Bottom Right: The diode current may be used to observe the change inresponsivity with wavelength – incident optical power is constant for this simulation.
171
CHAPTER 8
8.4.3 Small Signal Characteristics
An example of small signal characteristics can be obtained by .ac analysis and
a schematic similar to Fig. 8.5. The excess noise characteristics are available in
small signal simulation by the use of the .NOISE command. It can be shown that,
in the mid-band, the small signal noise of the APD and the DC or large signal
(.TRAN) excess noise factor (available on the ‘F’ node) are equivalent. Relatively
little information is available regarding the frequency dependence of excess noise.
In RC limited devices it seems likely that the excess noise will roll off by the same
means as a periodic signal. In transit time limited devices however, and at optical
modulation frequencies where the transit time of the device and the period of the
modulating signal are similar, it is unclear if the full excess noise will be available
at the device terminals. An experiment that may begin to provide some data is
discussed in Chapter 9.
172
CHAPTER 8
Figure 8.5 – Left: An LTSpice schematic showing the analysis of small signal response (.AC).Right: The resulting graph of cathode current as a function of frequency with reverse biasvoltage as a parameter. Note the inherent gain bandwidth product. This model is RC limitedunder all circumstances. The lowest, blue, line does not follow the others because the devicecapacitance changes markedly over the reverse bias range.
Figure 8.6 – Left: In this simulation excess noise is treated with a small signal model whereeach (successively larger) DC bias represent a change in ‘operating point’. This is useful whenthe SNR of a receiver system is required under certain conditions, for example simulatingthe SNR over a range of multiplications to evaluate the optimum combination of TIA gainand device multiplication as a function of the SNR at the TIA output. Right: Small signalresult (.NOISE) of excess noise factor of an APD. the roll-off of this characteristic may notbe physically representative as very little information exists regarding noise gain bandwidthproduct. However the low frequency values follow McIntyre’s model [3]. It can be shown thatthe small signal noise shown here leads to the large signal noise shown in Fig. 8.4.
173
CHAPTER 8
8.4.4 Large Signal Characteristics
Large signal characteristics are available via the anode and cathode voltages and
currents as well as the M, F and Cap nodes. An example of large signal APD
parameters in a 1 MHz operational amplifier based TIA is shown in Fig. 8.7. In
this figure one aspect of APD operation is exposed which can not be observed
easily in a laboratory. Namely that a large enough photo-current coupled with
a large enough multiplication and sufficient series resistance may cause the dark
multiplication and the light multiplication to differ in value. This effect applies
to excess noise also and represents a region of operation where the F versus M
characteristics may be expected to ‘bend over’ – become more nearly parallel
to the bias axis. An example of this type of bending is seen in the middle left
graph of Fig. 6.6, in the case of this data it is the low dynamic resistance of the
diode which causes some of the noise power to pass to ground without entering
the TIA. The shape of the graphs for these two measurement shortcomings is
indistinguishable.
8.5 Conclusion
This chapter has presented a simple SPICE model of an APD intended for the use
in simulating the interaction between real APDs and front-end circuits. Examples
of the model characteristics have been provided. The source code is presented in
Appendix B.
174
CHAPTER 8
Figure 8.7 – Top Left: This figure shows an opamp TIA with bandwidth of approximately1 MHz driven by the APD model, model data is shown in the blue text. Transient analysis(.TRAN) is preformed to examine the behaviour of the TIA and APD combination in the timedomain. The diode reverse bias voltage is a parameter. Top Right: Transient response of thetransimpedance amplifier under several APD bias voltages. Bottom Left: This figure is noteasily produced in a laboratory. It shows the time dependence of the photo-multiplication. Athigher gains the photo-current causes a non-negligible IR drop across the device series resistancechanging the junction bias sufficiently to influence the multiplication (and noise). Bottom Right:This is a plot of TIA output voltage on with a logarithmic y-axis. It allows the effect of darkcurrent (in this case tunnelling current) mechanisms on the time averaged output voltage to beseen more easily.
175
CHAPTER 8
176
Chapter 9
Future Work
One aspect of this thesis is the development of measurement systems for charac-
terisation of electro-optical devices. Many thesis have discussed the future work
regarding specific material systems for a range of purposes but often when III-V
materials are concerned telecommunications is the key market. In this section
some ideas are presented regarding improvements to the measurement methods in
use. The ideas in this chapter are not necessarily original, but most of them were
developed independently (i.e. without prior knowledge of their pre-existence).
These ideas have been collecting over the course of this work, and several have
already been implemented at Sheffield in some form with varying degrees of suc-
cess.
9.1 Microwave Small Signal Bandwidth Mea-
surements
The basis for this idea is that a measurement tool with comparable functionality
to a ‘lightwave component analyser’ – such as the Agilent N4373C – is desirable.
However the LCA is expensive and beyond the means of many research groups.
Consequently a vector network analyser (VNA) may be substituted with the
addition of some external components. Not only is the VNA less expensive, but
its usefulness is not limited to electro-optical problems. The cost may be shared
by several research groups, as is the case at Sheffield.
177
CHAPTER 9
CW Laser
1330/1550nm
EAM
CIP 40G-SR-EAM-1550CIP 40G-IR-EAM-1300
Bias TeePicosecond 5544
SMU
Bias TeePicosecond 5544
SMU
VNA
Figure 9.1 – Block diagram of VNA system.
9.1.1 Measurement System
A block diagram of one implementation is shown in Fig. 9.1. The CW laser is
modulated by an electro-absorption modulator (EAM), which is biased by an
SMU via a bias tee. The VNA output signal is also applied to the EAM via the
bias tee. A second tee is used to apply DC bias to the device from a second SMU.
The modulated current signal drops a voltage across the tee, and this voltage is
applied to the input of the VNA.
This implementation has several flaws relating to the impedance matching of
the various components, but most importantly the EAM and the APD. These
flaws yield results with unexplainable low frequency attenuations, various reso-
nances, and oddly shaped high frequency roll-offs. Calibration of the VNA by
the use of a calibration substrate can not help at all as the impedances when
the calibration substrate is in use are different from when the APD is in use.
The principle difficulty arises because the APD (essentially an un-matched high
impedance source) is connected to the surrounding equipment by a 50 Ω a coaxial
cable which is electrically very long. “Never put a photo-detector on the end of
an [electrically] long cable” [40].
This does not mean that transimpedance amplifiers, which are unmatched by
178
CHAPTER 9
Iph cj rd
RS
Transmisson LineZ0 = 50 Ω
VNAZin = 50 Ω
Figure 9.2 – Diagram of a series matched APD – transmission line – VNA system.
design, are useless at microwave frequencies! Practically all optical communica-
tions systems use IC transimpedance amplifiers. Reviewing some datasheets it
is easy to see that while the input impedance is not always low, 100 Ω is not
uncommon, the electrical distance between the unmatched APD and the TIA is
often specified to an exacting degree. It is common to see a graph of frequency
response as a function of the length and thickness of the bond-wire that connects
the APD to the TIA. The selection is often made based on the capacitance of
the detector. It is possible to make use of a TIA approach in ultra wide-band
measurements too, and some ideas on this topic will be discussed shortly.
9.1.2 Matching
The difficulty arising in ‘ultra wide-band’ measurements (for example 10 MHz –
50 GHz) is that all wide band impedance matching techniques have some asso-
ciated difficulty or trade-off. For the style of measurement proposed impedance
matching is required, unless the TIA approach, which will be addressed shortly, is
being followed. There are many ways of approaching matching an APD to a line,
and many ways of practically implementing that matching. Three possibilities
will be discussed which are representative of the major obstacles.
A diagram showing a series matching arrangement is shown in Fig. 9.2. In
this figure several key assumptions are made including that the series resistance
of the device is less than 50 Ω and that the series resistance can be supplemented
with an external resistor RS to force a series match to the line. This matching
is difficult to implement practically. Ideally, a set of APDs would be fabricated
and tested for series resistance. Having obtained the value of series resistance
internal to the diode the correct value may be added to another production run
by deposition and laser trimming of a thin film resistor on the substrate, close to
179
CHAPTER 9
ZT
C1
T2
Z0 = 50 Ω RS
iph cj rd
L1
−
+VB
T1
Z0 = 50 Ω
C2
VNAZin = 50 Ω
Figure 9.3 – A variation of the parallel matching arrangement after Xie [23] adapted forbandwidth measurements.
the diode. Not only is this process expensive, it depends on the devices having
similar series resistance across production runs and the availability of the various
technologies. A second (inferior) possibility is the use of a very small (0201)
microwave resistor epoxied onto the header or ceramic plate carrying the wafer
and connected to the device by usual wire bonding methods.
This method can only work when the transit time is large or the capacitance
is extremely small (∼ 10s fF). The necessity of a small capacitance is due to the
low pass filter formed by RS and cj . It is possible to terminate the VNA end of
the transmission line with an open circuit, no voltage division is required there
to provide a wideband match. This is because the reflected signal will be com-
pletely absorbed when it returns to the diode end of the transmission line, hence
resonances are impossible as standing waves can not build up. In practice the
transmission line will almost always be terminated however. A further assump-
tion is that the dynamic resistance of the diode is always very large compared to
50 Ω.
In the next solution it is assumed that, since the bandwidth is being mea-
sured, the quantum efficiency is a secondary concern (it can be measured by
other methods if desired) and it is permissible to introduce a frequency indepen-
dent attenuation in the measurement. This method borrows heavily from the
noise measurement after Xie [23]. The system diagram is shown in Fig. 9.3. The
APD is connected to a line, T2, which is electrically short (< λ/10 at the max-
180
CHAPTER 9
iph cj rd
RS
R1 C1
R2
L1
Figure 9.4 – Passive components are used to form an all–pass filter with a frequency indepen-dent impedance of 50 Ω.
imum frequency of interest e.g. 50 GHz). The line is then terminated by ZT
which is 50 Ω. C1 is necessary to prevent the biasing voltage VB driving a DC
current through ZT . Similar arguments apply for the use of C2 and the VNA
input impedance. The purpose of T2, C1 and ZT is to provide a termination at
the APD end of T1. The other end of T1 is terminated by the input impedance of
the VNA. Provided RS is much less than 50 Ω and rd much larger than 50 Ω the
system will be matched at all nodes. There is still the RC limit of the APD and
the characteristic impedance. If the transit time limit occurs after the RC limit
the system will measure an artificially low bandwidth. ZT , C1 and T2 would, ide-
ally, be fabricated with the APD on the semiconductor substrate. However this
may be impractical. It should be possible however to epoxy a wafer to a ceramic
header or disk and epoxy several sets of resistors and capacitors (assuming 0201
or 0402 size) around the wafer and then use standard wire bonding techniques to
make the connections.
The prior examples have chosen to neglect the capacitive nature of the APD
and assume that at all frequencies of interest the APD is dominantly resistive
(because cj is very small ∼200 fF or less) while this is not an unreasonable as-
sumption in some cases it can not be guaranteed, and even 200 fF yields an
impedance of 16 Ω at 50 GHz. As frequency increases the above methods, while
better than nothing, will eventually fail to operate as designed. The final method
attempts to make the APD look resistive at all frequencies. The idea for this
method, while is is probably not original, came partly from a particularly taxing
undergraduate tutorial sheet in which the proof of a conjecture about the rela-
181
CHAPTER 9
tionship between the circuit components is sought, and partly from audio power
amplifiers where some image filters are also called Zobel networks.
The circuit diagram is shown in Fig. 9.4. In this diagram the APD is shown as
cj , rd, iph and RS. And it is assumed that RS is less than 50 Ω. It is also assumed
that across the range of bias voltages which the bandwidth will be measured the
change in cj is negligible. High speed detectors tend to have small depletion
capacitances. R1 is ideally a laser trimmed semiconductor resistor, but it could
be a 0201 microwave component, or otherwise fabricated into a MMIC or flip-
chip arrangement with the detector substrate. Its purpose is to increase the total
value of R1 +RS to 50Ω. C1 is ideally an IC capacitor. The purpose of which is
to block the biasing voltage from passing a large current through L1 and R2. A
suitable value for a 10 MHz -3 dB point in a 50 Ω system is 1 nF. The capacitor
must be able to sustain the biasing voltage, so it can not be arbitrarily thin. This
value will require a considerable area of semiconductor and it may be necessary to
relinquish the 10 MHz – 100 MHz frequency range and use 100 pF instead. The
value of R2 should be equal to R1+RS, this causes the network to be resonant at
all frequencies. It can be shown that when R2 = R1 +RS the value of L required
to provide a frequency independent response is L = C R2. Ideally L is a spiral
inductor fabricated near to the APD on the semiconductor substrate by normal
metallisation methods and a small air bridge. For an APD with cj = 1 pF the
required value of L is 2.5 nF. It is important to note that the objective is to
provide an impedance looking into the network of 50 Ω at all frequencies, it does
not remove the bandwidth limit of the APD formed by cj and the surrounding
resistors.
9.2 Microwave Noise Measurements
In this section the measurement of noise above 10 MHz will be discussed. This
area is interesting both from a measurement design perspective and from a device
perspective. The device ideas presented in the following discussion are due to
Marshall1. Consider an APD with a modulated optical source incident on the
window. The period of the modulation is less than the transit time of the device.
1Andrew R. J. Marshall private communication 2010.
182
CHAPTER 9
For example in InGaAs the saturated electron drift velocity is ∼ 1 × 107 cm/s.
A 2.56 µm device would be required for a 40 GHz signal to have the same period
as the transit time. It may be hypothesised that the optically generated carriers
would be generated at regular intervals of the wavelength throughout the device
and that assuming impact ionisation was not taking place, or that one of the
ionisation coefficients was zero. The carriers would travel together, this represents
a reduction in the variance of the position of the carrier population which may
lead to a sub shot noise result. The effect, if it exists, should be detectable prior
to the frequency where an integer number of wavelengths fits into a device i.e.
as the wave period approaches the transit time. To approach this work, a thick
(∼ 5 µm) three region structure, manufactured from one of the III-V materials
that is commonly used in telecommunications is preferable. If the objective is
to show that the effect exists is would be prudent to choose the material with
the slowest saturation velocity as this should yield any observable effect at the
lowest frequency. This work would be interesting even if the noise is shown to
follow a shot process. For many years excess avalanche noise has been reported in
telecommunications material systems at 10 – 30 MHz. No work has been reported
showing the noise of a telecommunications APD at the bandwidths it is intended
to operate. While BER measurements are routinely reported, these deal directly
with the quality of the detector within a communications systems perspective.
The underlying physical processes require direct investigation.
9.2.1 Measurement Systems
The measurement systems discussed below are original and have been published
[29]. The systems are an improvement over the Xie system [23]. While the
systems appear similar to that of Xie, the measurement method is somewhat
different, and the results potentially more stable and repeatable.
There are two possible improvements to the design proposed by Xie et al. [23].
Both are essentially improvements to the method by which the instrumentation
is calibrated. The introduction of the calibrated noise source (HP346B) permits
the use of direct noise figure measurement as opposed to hot/cold measurements,
which is a considerable improvement. The noise figure meter (N8973A, or an older
183
CHAPTER 9
model such as the N8970) is designed such that the noise source is connected
to the device (for example an LNB) under test. Of course if the device is an
electro-optical transducer this is impossible as there is no place to attach the
noise source. This leads to the use of a pre-test calibration followed by hot/cold
measurements. It would be preferable to use the noise figure analyser according
to its design principle i.e. with the noise source in the measurement. The NFA
is provided with prior calibration – by the manufacturer – of the noise sources
contribution to the system. The system gain is also computable by measuring
the effect on the noise output when the noise source is switched on and off – it
is pulsed by the noise figure meter. The time average of the change in noise level
can provide the gain from the noise input port to the NFA input port. The prior
knowledge of the known noise input from the calibrated source (HP346B) allows
the NFA to compute the gain and noise figure nearly instantly, a considerable
improvement in measurement speed, accuracy, precision and repeatability. The
use of real time signal averaging is also possible, effectively increasing the system
signal to noise ratio. The noise source can not be directly applied to the detector.
However, a secondary port can be created which permits the connection of an
APD and the noise source to the NFA simultaneously. Two example designs
are provided, the first uses a 50 Ω matched topology similar to that of Xie et
al. [23] The second describes a similar overall structure but using a commercial
transimpedance amplifier.
The APD multiplication, excess noise factor and noise power bandwidth can
be established simultaneously in one measurement. The limitation of the sys-
tem bandwidth can be alleviated by two methods. Firstly a higher maximum
frequency noise figure meter can be obtained. Agilent Technologies presently
manufactures noise figure meters/analysers capable of directly measuring up to
26 GHz. The use of heterodyne techniques could extend this considerably. How-
ever a relatively inexpensive alternative is to use a lower bandwidth noise figure
meter but begin measuring bandwidth once the APD has been biased to achieve
a high gain. The high frequency roll off due to a finite gain bandwidth product
can be observed at lower frequencies; the unity noise gain bandwidth product can
then be inferred. A potential problem with this method is that carrier feedback
will randomise the ionisation position for any given carrier so noise reduction, if it
184
CHAPTER 9
exists, will only be observable at relatively low gains. The importance of correct
impedance matching cannot be overemphasized.
9.2.1.1 50 Ω System
The system diagram shown in Fig. 9.5 shows the structure of the measurement
setup. An SMU drives a bias tee composed of L1 and C1. An example of a
suitable tee is PicoSecond Model 5541A. The APD is connected to a microwave
DC block, C1, and this is in turn connected to a termination (50 Ω). The DC block
and the termination must be electrically close to the APD even at the highest
measurement frequency. From the point of view of the first amplifier the APD is
a Norton source coupled to the end of a properly terminated transmission line.
Approximately half of the noise power will escape to ground via R1, the rest will
enter the measurement system. It is possible to calibrate the measurement system
either manually (i.e. use a 50 Ω signal generator to list a table of adjustments
for each frequency and post process the measured device data based on these
reading) or automatically by using the HP 346B Noise source hooked to the first
amplifier input instead of the APD. The attenuator setting must be noted down
when the calibration is carried out. The first amplifier in the chain must be of
the lowest possible noise. Examples include Minicircuits ZFL-1000LN+, ZX60-
33LN+ and Pasternack PE1513. The ZFL-1000 has low noise and a reasonably
flat gain versus frequency profile from 100 kHz to 1 GHz however bandwidth is
limited to 1 GHz. The ZX60-33LN+ has exceptionally low noise, and reasonable
gain versus frequency characteristics from 50 MHz to 3 GHz. The PE1513 has
relatively poor noise especially as frequency increases, the gain versus frequency
profile is not ideal either; however it is the only device which covers the whole
frequency range of the NFA. Unless APDs possessing bandwidths below 50 MHz
are to be routinely measured the author’s preferred choice would be the ZX60-
33LN.
The specifications of the second and third amplifiers are considerably less
critical than the first. Any microwave device with reasonable noise and gain versus
frequency characteristics will be acceptable. The stepped attenuator should be of
the precision type for example the Trilithic RSA35-100 (0 dB to 100 dB in 10 dB
185
CHAPTER 9
SMU
L1
C1
R1 APD
C2
50Ω Amplifier
10MHz to 3GHz
50Ω Amplifier
10MHz to 3GHz Stepped Attenuator0 – 100dB
50Ω Amplifier
10MHz to 3GHz
PowerCombiner
HP 346BNoise Source
Agilent N8973A
Figure 9.5 – 50 Ω excess noise measurement system.
steps) would be ideal. The power combiner may be of any type which covers
the required bandwidth. A suitable resistive splitter/combiner is the Minicircuits
ZX10E-14-S+.
The maximum device capacitance is approximately 2 pF to obtain a 3 dB point
of approximately 3 GHz. R1 must be electrically close to the APD, consequently
it is unlikely that the noise contribution of this resistor could be minimised by
cooling as was reported by Xie et al. [23]. If the APD was measured at low tem-
perature however it would be plausible to place R1 and C1 in the cryostat cham-
ber with the APD, hence obtaining a noise advantage at lower temperatures. A
laser is often used to excite electro-optical transducers in characterisation experi-
ments. In this case the laser should be a gas laser possessing a single longitudinal
mode, preferably frequency and amplitude stabilised. Noise characterisation ex-
periments using semiconductor lasers have always proved impossible, the laser
relative intensity noise (RIN) is too great to permit measurement of the detec-
tor noise. Multi-longditudinal mode lasers often have baseband mode beating
effects [40]1, and are generally unsuitable in this experiment.
1Samual Goldwasser private communication 2011
186
CHAPTER 9
SMU
L1
C1
R1 APD
C2
50Ω Amplifier
10MHz to 3GHz
50Ω Amplifier
10MHz to 3GHz Stepped Attenuator0 – 100dB
50Ω Amplifier
10MHz to 3GHz
PowerCombiner
HP 346BNoise Source
Agilent N8973A
Figure 9.6 – Transimpedance amplifier excess noise measurement system.
9.2.1.2 TIA CW Noise Measurement System
The structure of this measurement system is nearly identical to the 50 Ω system
previously described. The principle difference is the use of a transimpedance
amplifier front end instead of a 50 Ω system. Fig. 9.6 shows the system diagram.
C1 provides an AC ground for the APD such that the very great majority
of the noise current flows into the TIA. Example TIAs are given in the figure.
Commercial TIAs often have input impedance which is not a good approximation
to a virtual earth. As a result the maximum permissible device capacitance is
often lower than in the 50 Ω system case and is dependent on the particular
TIA in use. The MAX3910 provides ∼9 GHz small signal bandwidth and nearly
linear output voltage to input current relationship for photo-currents in the range
0 to 900 µApk−pk. The small signal gain is approximately 1.6 kV/A in the linear
region.
Unlike the 50 Ω system it is not possible to connect the noise source to the TIA
input for calibration purposes. This is a major limitation of the TIA measurement
compared with the 50 Ω measurement. Calibration of the TIA signal path with
the noise source is only possible at the TIA output. A plausible method of
calibration is to use a unity gain wide band (much wider than the device that is
being measured) pin diode which is known to exhibit shot noise. Any deviation
187
CHAPTER 9
Iph cj rd
−
+
A1AD829
C3
68 pF
−5V
+5V
R5
2.2 kΩ
C1
1.5 pF
R2
1 kΩ
R1
1 kΩ
R3
50Ω
Agilent/HP 346B −
+
A2AD829
C4
22 pF
−5V
+5V
R4
1 kΩ
R6
50Ω
C2
220 nF
Output
Figure 9.7 – Implemented 10 – 20 MHz transimpedance based excess noise measurment frontend with noise source port.
from shot noise can be calibrated out.
Fig. 9.7 shows a simple circuit which accomplishes the summing operation
with a TIA. The bandwidth is limited to 20 MHz, and the noise figure of the
circuit prior to any photo-generated noise is +23 dB (measured) consequently a
large (> 10 µA) photo-current is required to affect a significant increase in the
noise figure. The circuit does however possess a stable (if high) noise floor within
the operating bandwidth and the measured noise is repeatable over a period of
several hours and is nearly identical across the available measurment bandwidth.
These three features are lacking from the hot/cold measurements presently in use.
This particular implementation is of little practical value beyond being a proof
of concept however. A much lower noise, wider band circuit preferably avoiding
opamps is required to perform meaningful research.
9.3 Laser Noise “Cancelling” Front End
The work in this section involves the application of Hobbs’ noise cancelling circuit
to the PSD style measurment systems that are in use at Sheffield. The objective
is to make any laser usable in excess noise measurements. Presently only gas
lasers and often only He-Ne lasers have noise characteristics that are suitable for
excess noise measurements.
188
CHAPTER 9
I1
+VS
Q1Q2
−+
-15V+15V R12
R4
−+
C2-15V+15V
R2
Q5 Q4
R3
Q6
Q7
I2
−VS
Q8
Q3
−+
-15V+15V R1
−
+ V1
Output
Figure 9.8 – Noise “cancelling” front end modified after Hobbs [40]. This version has the DCservo detached from the output node. I1 is composed of the AC photo-current, DC photo-current + laser RIN. I2 is composed of the DC photo-current + laser RIN all multiplied by ∼1.01 – 2. The exact ratio is not important.
This work follows on from that of Hobbs [40] pp. 721 – 731. The principle
of operation relies on a circuit which allows division of two currents in an ar-
bitrary ratio, and that an analogue of this operation is the combination of two
currents. If these currents share a common noise component, which is not ran-
dom (for example laser RIN) it may be made to cancel such that a much smaller
signal component that was previously heavily corrupted is unmasked. The cir-
cuits Hobbs presents perform this function, unfortunately, do not permit the use
of PSD because the photo-current must have a DC component.
With reference to Fig. 9.8, the laser is tapped by a beam splitter, the ratio of
the two output beams is not significant, there is no requirement for them to be
of equal intensity but one must be equal to or greater than the other. One of the
output beams is chopped by some means which does not completely extinguish
189
CHAPTER 9
the signal in the off period, the other is CW. I1 is the photo-diode under test. It
is illuminated by the chopped laser light which includes the RIN. I2 is a second
photo-diode which is not necessarily identical to the first (for example a BPX65).
It is illuminated by the CW output beam which also contains the RIN but not the
phase information associated with the chopping process. The servo ensures that
the DC conditions across the two differential pairs are balanced the cancellation
occurs due to the addition of the current in the collector of Q4 and the collector
of Q1. The collector of Q1 passes the PSD signal the (wanted) shot and excess
(if multiplication is present) noise of the APD represented by I1, the Laser RIN
and a DC component of photo-current. Q4’s collector passes the laser RIN and a
DC component of photo-current. It is critical that the ratio of the RIN derived
current in the two photo-diodes and the ratio of the DC components be identical.
This is the reason why there must always be a DC component of photo-current.
If the restriction on the ratio of the RIN currents and the DC currents did not
exist, a pair of current sources in parallel with I1 and I2 would be sufficient. It
may be possible to alleviate this necessity by measuring the DC component of
current in the two diodes and using a pair of controlled current sources (easily
manufactured from an opamp and a JFET) to maintain a DC current of the
correct ratio. This would also allow high dark current and photo-multiplication
to be appropriately handled. No method of achieving the measurement of the
DC component has been forthcoming which does not also disturb the balance
of the currents sufficiently to make their cancellation only partially effective. If
a solution can be found it would make noise measurements readily available at
many more wavelengths.
9.4 Improved Squaring Circuit
When using the two measurement systems reported in Chapters 3 and 5 the
system noise floor is strongly affected by the junction capacitance of the APD.
The output of the noise power meter has a DC component proportional to its
input noise. When using large capacitance APDs the output of the power meter
may saturate due to the DC component. The appropriate action is to reduce the
system gain by adjusting the attenuator, this may be necessary even if the power
190
CHAPTER 9
P.G.50 Ω C1
L1
SMU
16 Ω
16 Ω
SPAD
220 nF
Ch1
16 Ω
220 nF
Ch2
CT
Figure 9.9 – An on-wafer SPAD experiment.
meter did not posses a DC offset proportional to its input noise. It would be
possible to detect this non-linearity by checking that two adjacent attenuation
settings do produce a factor of ten change in noise output voltage. However,
since the output is saturated it is impossible to continue without increasing the
attenuation. In situations where the output saturates due to the DC but the
noise power meter would otherwise operate correctly, system sensitivity is lost
due to the design of the power meter. A design of power meter having no DC
component in its output is desirable.
9.5 An On-Wafer SPAD Setup
This is an extremely simple system to produce a SPAD measurement system for
devices still in wafer form. The underlying principle is borrowed from Dimler
et al. [239]. It’s practical usefulness may be limited as it breaks Hobbs’ ‘rule’
of electro-optical design – diodes are connected by necessarily on long cables.
Nevertheless, a setup similar to the one shown in Fig. 9.9 was constructed and
initial results appeared promising, however its use was discontinued and it was
dismantled without a proper assessment of its suitability.
A preferred embodiment is shown in Fig. 9.10. The box is conducting (e.g
Aluminium). The impedance matching and balance capacitors are SMT compo-
nents mounted on an FR4 PCB as close together as physically possible, ground
191
CHAPTER 9
P.G.
Seriestermination
50 Ω Coax
Wide BandBias Tee
SMU
Triax
50 Ω Coax 50 Ω Coax
SPAD
This coax and diode combinationrepresents a probe and probestation
16 Ω 16 Ω 220 nF
16 Ω
20 pF 190 pF
220 nF
50 Ω Coax
CRO CH150 Ω termination
50 Ω Coax
CRO CH250 Ω termination
Shield
Core
It may be necessary to addan LNA here if one is usedin CH1
In may be necessary to have an LNAhere in series with a continuouslyvariable attenuator
Figure 9.10 – An on-wafer SPAD experiment – a preferred implementation.
planes are essential. ∼16 Ω may be manufactured with two 33 Ω in parallel. It
is assumed that the diode series resistance is much lower than 50 Ω and that the
diode dynamic resistance never falls as low as 50 Ω. If these conditions are met
it is likely (although it has not been proved) that all nodes see 50 Ω in all direc-
tions. Balance is found when the sum of the trimmer capacitor and the 190 pF
capacitor (Fig. 9.10) equal the sum of the SPAD junction capacitance and the
cabling capacitance from the aluminium box to the probe-station including the
probe and all strays. The CRO is set up to display channel one subtracted from
channel two.
192
CHAPTER 9
9.6 NLOS UV Communications with SiC
In Chapter 7 an application of SiC detectors to NLOS communication was de-
scribed. Presently, there are no reported field trials of NLOS communication
systems using SiC detectors. Several ‘test beds’ exist using PMTs and GaN
photo-diodes for example [157,158,240–242]. While the idea of NLOS communi-
cations is not a new idea a comparison of SiC and GaN in a real world application,
a pair of personal radios for example, using a GaN/AlGaN transmitter would be
of interest. Comparing dark current densities tends not to yield much practically
useful information at the system level. BER should be one of the metrics by
which the systems are measured, as well as (accelerated) lifetime testing, and
maximum range.
193
CHAPTER 9
194
Chapter 10
Conclusion
The objectives of this work were threefold, firstly to develop new excess noise mea-
surement systems that increase the range of measurable junction capacitances of
an APD. Secondly to verify the impact ionisation coefficients in 4H-SiC, especially
at lower electric fields, where considerable disagreement exists in the literature.
The final objective was to systematically explore the dead-space effect in Silicon
by providing multiplication and excess noise measurements on a series of differing
thickness devices.
The design of two measurement systems capable of measuring APDs with
junction capacitance up to two orders of magnitude greater than the prior Sheffield
system, and greater than any other reported, has been described. The design and
optimisation of these systems and associated material has been presented in de-
tail. The systems use phase sensitive recovery techniques allowing the avalanche
noise power and photo-current to be measured unambiguously from the system
noise and leakage current of the test device. Both measurement systems systems
have been used to confirm the validity of the data in the high capacitance devices.
Avalanche noise measurements have been performed on a range of Silicon Car-
bide nip and separate absorption and multiplication avalanche photo-diodes with
i region widths, w, of 2.7 µm and 0.57 µm. with the objective of experimentally
verifying the value of α over a range of lower electric fields. This has been ac-
complished, and a new values of α and β have been published based on the data.
This work also represents the first reported (and lowest noise) measurement of
pure hole injection in 4H-SiC.
195
CHAPTER 10
Using the systems described, avalanche noise measurements have been per-
formed on a range of Silicon pin diodes. These devices have widths, w, ranging
from 0.35 µm to 0.031 µm. The Silicon results show that, for w < 0.4 µm, the
extracted and non-local modelled impact ionisation coefficients begin to deviate
from the bulk coefficients. This is the first systematic measurement of the effect
of the dead-space in Silicon and includes the thinnest reported device to date.
Using the experimental data and analysis herein it is possible to estimate, for
example, the length a transistor channel must be reduced to in order to support
a certain electric field while yielding a desired maximum value of current multi-
plication. Such information is of considerable use to IC designers. This work has
also verified that the RPL model accurately represents characteristics in devices
as thin as 71 nm, and provides acceptable analysis at 31 nm.
196
References
[1] H. Kragh, Quantum Generations: A History of Physics in the Twentieth
Century. Princeton University Press, 2002. 2
[2] Hamamatsu Photonics K. K., Photomultiplier Tubes: Basics and Applica-
tions, 3rd ed. Hamamatsu Photonics K. K., 2006. 5, 6
[3] R. McIntyre, “Multiplication noise in uniform avalanche diodes,” IEEE
Trans. Electron Devices, vol. 13, no. 1, pp. 164–168, Jan. 1966. doi:
10.1109/T-ED.1966.15651 6, 12, 15, 16, 20, 21, 39, 155, 166, 173
[4] M. S. Carroll, K. Childs, R. Jarecki, T. Bauer, and K. Saiz, “Ge-Si separate
absorption and multiplication avalanche photodiode for geiger mode single
photon detection,” Appl. Phys. Lett., vol. 93, no. 18, p. 183511, 2008. doi:
10.1063/1.3020297 12, 113
[5] S. L. Miller, “Avalanche breakdown in Germanium,” Phys. Rev., vol. 99,
pp. 1234–1241, Aug. 1955. doi: 10.1103/PhysRev.99.1234 12, 111, 149, 166,
167
[6] K. G. McKay, “Avalanche breakdown in Silicon,” Phys. Rev., vol. 94, pp.
877–884, May 1954. doi: 10.1103/PhysRev.94.877 12, 110
[7] K. G. McKay and K. B. McAfee, “Electron multiplication in Silicon
and Germanium,” Phys. Rev., vol. 91, pp. 1079–1084, Sep. 1953. doi:
10.1103/PhysRev.91.1079 12, 109
[8] P. A. Wolff, “Theory of electron multiplication in Silicon and Ger-
manium,” Phys. Rev., vol. 95, pp. 1415–1420, Sep. 1954. doi:
10.1103/PhysRev.95.1415 12, 110
197
REFERENCES
[9] G. A. Baraff, “Distribution functions and ionization rates for hot electrons
in semiconductors,” Phys. Rev., vol. 128, pp. 2507–2517, Dec. 1962. doi:
10.1103/PhysRev.128.2507 12, 111
[10] A. G. Chynoweth and K. G. McKay, “Threshold energy for electron-hole
pair-production by electrons in Silicon,” Phys. Rev., vol. 108, pp. 29–34,
Oct. 1957. doi: 10.1103/PhysRev.108.29 12, 111
[11] A. G. Chynoweth, “Ionization rates for electrons and holes in Silicon,” Phys.
Rev., vol. 109, pp. 1537–1540, Mar. 1958. doi: 10.1103/PhysRev.109.1537
12, 111
[12] W. Shockley, “Problems related to p-n junctions in Silicon,” Czechoslovak
Journal of Physics, vol. 11, pp. 81–121, 1961. doi: 10.1007/BF01688613 12,
111
[13] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Nu-
merical Recipies in C. Foundation Books, 2007. 12
[14] A. Marshall, C. H. Tan, M. Steer, and J. David, “Extremely low excess noise
in InAs electron avalanche photodiodes,” IEEE Photon. Technol. Lett.,
vol. 21, no. 13, pp. 866–868, Jul. 2009. doi: 10.1109/LPT.2009.2019625
14
[15] J. Beck, C. Wan, M. Kinch, J. Robinson, P. Mitra, R. Scritchfield,
F. Ma, and J. Campbell, “The HgCdTe electron avalanche photodiode,”
in LEOS Summer Topical Meetings, 2006 Digest of the, 2006, pp. 36–37.
doi: 10.1109/LEOSST.2006.1694056 14
[16] A. R. J. Marshall, “The InAs electron avalanche photodiode and the influ-
ence of thin avalanche photodiodes on receiver sensitivity,” Ph.D. thesis,
University of Sheffield, Sheffield, UK, Aug. 2009. 14
[17] D. S. Ong, K. F. Li, G. J. Rees, J. P. R. David, and P. N. Robson, “A simple
model to determine multiplication and noise in avalanche photodiodes,” J.
Appl. Phys., vol. 83, pp. 3426–3429, 1998. doi: 10.1063/1.367111 14, 120
198
REFERENCES
[18] B. K. Ng, “Impact ionization in wide band gap semiconductors:
AlxGa1−xAs and 4H-SiC,” Ph.D. thesis, Department of Electronic and Elec-
trical Engineering, University of Sheffield, Jul. 2002. 14, 120, 136, 146, 148,
149, 153
[19] C. H. Tan, J. P. R. David, G. J. Rees, R. C. Tozer, and D. C. Herbert,
“Treatment of soft threshold in impact ionization,” J. Appl. Phys., vol. 90,
no. 5, pp. 2538–2543, 2001. doi: 10.1063/1.1388865 17, 20
[20] K. F. Li, “Avalanche noise in submicron GaAs and InP structures,” Ph.D.
thesis, University of Sheffield, Sheffield, UK, 1999. 25, 28, 33, 34, 35, 37,
46, 47, 94, 96, 97, 103
[21] S. M. Sze and K. K. Ng, Physics of Semiconductor Devices, 3rd ed. Wiley-
Blackwell, 2006. 25, 39
[22] R. Baertsch, “Low-frequency noise measurements in Silicon avalanche pho-
todiodes,” IEEE Trans. Electron Devices, vol. 13, no. 3, pp. 383–385, Mar.
1966. doi: 10.1109/T-ED.1966.15698 30
[23] F. Z. Xie, D. Kuhl, E. H. Bottcher, S. Y. Ren, and D. Bimberg, “Wide-
band frequency response measurements of photodetectors using low-level
photocurrent noise detection,” J. Appl. Phys., vol. 73, no. 12, pp. 8641–
8646, 1993. doi: 10.1063/1.353397 30, 33, 40, 42, 180, 183, 184, 186
[24] M. Toivonen, A. Salokatve, M. Hovinen, and M. Pessa, “GaAs/AlGaAs
delta-doped staircase avalanche photodiode with separated absorption
layer,” Electronics Letters, vol. 28, no. 1, pp. 32–34, Jan. 1992. doi:
10.1049/el:19920020 30, 33
[25] G. E. Bulman, “The experimental determination of impact ionisation coef-
ficients in GaAs and InP,” Ph.D. thesis, University of Illinois, USA, 1983.
30, 40
[26] H. Ando and H. Kanbe, “Ionization coefficient measurement in GaAs by
using multiplication noise characteristics,” Solid-State Electronics, vol. 24,
no. 7, pp. 629–634, 1981. doi: 10.1016/0038-1101(81)90191-X 30, 32, 40
199
REFERENCES
[27] K. S. Lau, C. H. Tan, B. K. Ng, K. F. Li, R. C. Tozer, J. P. R. David,
and G. J. Rees, “Excess noise measurement in avalanche photodiodes using
a transimpedance amplifier front-end,” Meas. Sci. Technol., vol. 17, no. 7,
pp. 1941–1946, 2006. doi: 10.1088/0957-0233/17/7/036 30, 33, 34, 40, 43,
66, 69, 96, 103, 148
[28] A. S. Tager, “Current fluctuations in a semiconductor under the conditions
of impact ionization and avalanche breakdown,” Fiz. Tver. Tela, vol. 6, pp.
2418–2427, 1964. 31, 148
[29] D. S. G. Ong and J. E. Green, “Avalanche photodiodes in high-speed
receiver systems,” in Photodiodes - World Activities in 2011, J.-W. Park,
Ed. InTech, Jul. 2011. Available: http://www.intechopen.com/articles/
show/title/avalanche-photodiodes-in-high-speed-receiver-systems 34, 183
[30] Analog Devices Ultralow Distortion, Wide Bandwidth Voltage Feedback Op
Amps: AD9631/AD9632 Rev. C, Analog Devices, Jul. 2003. 36, 42, 43, 96,
97
[31] L. Yin, Q. Chen, S. Kou, and J. Qin, “Research on avalanche pho-
todiode based photon imaging system,” in Photonics and Optoelec-
tronics, 2009. SOPO 2009. Symposium on, Aug. 2009, pp. 1–4. doi:
10.1109/SOPO.2009.5230196 39
[32] R. G. Smith and S. D. Personick, Receiver design for optical fibre com-
munication systems in Semiconductor Devices for Optical Communication,
Chapter 4. New York: Springer-Verlag, 1980. 39
[33] C. E. Talley, G. A. Cooksey, and R. C. Dunn, “High resolution fluorescence
imaging with cantilevered near-field fiber optic probes,” Appl. Phys. Lett.,
vol. 69, no. 25, pp. 3809–3811, 1996. doi: 10.1063/1.117112 39
[34] L. Tan, J. Ng, C. Tan, and J. David, “Avalanche noise characteristics in
submicron InP diodes,” IEEE J. Quantum Electron., vol. 44, no. 4, pp.
378–382, 2008. doi: 10.1109/JQE.2007.914771 40
200
REFERENCES
[35] Y. Goh, A. Marshall, D. Massey, J. Ng, C. Tan, M. Hopkinson,
J. David, S. Jones, C. Button, and S. Pinches, “Excess avalanche noise
in In0.52Al0.48As,” IEEE J. Quantum Electron., vol. 43, no. 6, pp. 503–507,
2007. doi: 10.1109/JQE.2007.897900 40
[36] D. J. Massey, J. P. R. David, C. H. Tan, B. K. Ng, G. J. Rees, D. J. Robbins,
and D. C. Herbert, “Impact ionization in submicron Silicon devices,” J.
Appl. Phys., vol. 95, no. 10, pp. 5931–5933, 2004. doi: 10.1063/1.1691177
40, 112
[37] L. Vandamme, “Noise as a diagnostic tool for quality and reliability of
electronic devices,” IEEE Trans. Electron Devices, vol. 41, no. 11, pp. 2176–
2187, 1994. doi: 10.1109/16.333839 40
[38] R. Macku and P. Koktavy, “Study of solar cells defects via noise measure-
ment,” in Electronics Technology, 2008. ISSE ’08. 31st International Spring
Seminar on, 2008, pp. 96–100. doi: 10.1109/ISSE.2008.5276442 40
[39] S. B. Alexander, Optical communication receiver design, ser. SPIE tutorial
texts in optical engineering and IEEE telecommunications series. Belling-
ham, Wash. : London : SPIE Optical Engineering Press; Institution of
Electrical Engineers, 1997. 40
[40] P. C. D. Hobbs, Building Electro-optical Systems: Making it All Work,
2nd ed. Wiley-Blackwell, 2009. 41, 42, 50, 178, 186, 189
[41] B. Pease, “What’s all this transimpedance amplifier stuff, anyhow?” Elec-
tronic Design, vol. 49, no. 1, p. 139, Jan. 2001. 42
[42] OPA129 Datasheet, Texas Instruments, Texas Instruments, Post Office Box
655303, Dallas, Texas 75265, 2007. 42
[43] E. Sackinger, “The transimpedance limit,” IEEE Trans. Circuits Syst. I,
vol. 57, no. 8, pp. 1848–1856, Aug. 2010. doi: 10.1109/TCSI.2009.2037847
42, 46
201
REFERENCES
[44] C. Wu, E. Sovero, and B. Massey, “40-GHz transimpedance amplifier with
differential outputs using InP-InGaAs heterojunction bipolar transistors,”
IEEE J. Solid-State Circuits, vol. 38, no. 9, pp. 1518–1523, Sep. 2003. doi:
10.1109/JSSC.2003.815927 42
[45] Z. Lu, K. S. Yeo, J. Ma, M. A. Do, W. M. Lim, and X. Chen,
“Broad-band design techniques for transimpedance amplifiers,” IEEE
Trans. Circuits Syst. I, vol. 54, no. 3, pp. 590–600, Mar. 2007. doi:
10.1109/TCSI.2006.887610 42
[46] G. Ferrari and M. Sampietro, “Wide bandwidth transimpedance ampli-
fier for extremely high sensitivity continuous measurements,” Rev. Sci. In-
strum., vol. 78, no. 9, p. 094703, 2007. doi: 10.1063/1.2778626 42
[47] R. M. Howard, “Ultralow noise high gain transimpedance amplifier for char-
acterizing the low frequency noise of infrared detectors,” Rev. Sci. Instrum.,
vol. 70, no. 3, pp. 1860–1867, 1999. doi: 10.1063/1.1149681 42
[48] D. Micusik and H. Zimmermann, “Transimpedance amplifier with 120 dB
dynamic range,” Electronics Letters, vol. 43, no. 3, pp. 159–160, Jan. 2007.
42
[49] B. Wilson and I. Darwazeh, “Transimpedance optical preamplifier with a
very low input resistance,” Electronics Letters, vol. 23, no. 4, pp. 138–139,
Dec. 1987. doi: 10.1049/el:19870097 42
[50] C. Ciofi, F. Crupi, C. Pace, G. Scandurra, and M. Patane, “A new cir-
cuit topology for the realization of very low-noise wide-bandwidth tran-
simpedance amplifier,” IEEE Trans. Instrum. Meas., vol. 56, no. 5, pp.
1626–1631, Oct. 2007. doi: 10.1109/TIM.2007.904566 42
[51] M. Jalali, A. Nabavi, M. Moravvej-Farshi, and A. Fotowat-Ahmady, “Low-
noise differential transimpedance amplifier structure based on capaci-
tor cross-coupled gm-boosting scheme,” Microelectronics Journal, vol. 39,
no. 12, pp. 1843–1851, 2008. doi: 10.1016/j.mejo.2008.08.005 42
202
REFERENCES
[52] P. R. Gray, P. J. Hurst, S. H. Lewis, and R. G. Meyer, Analysis and Design
of Analog Integrated Circuits, 4th ed. Wiley, Feb. 2001. 45, 65
[53] W. M. J. Leach, “Fundamentals of low-noise analog circuit design,” Proc.
IEEE, vol. 82, no. 10, pp. 1515–1538, Oct. 1994. doi: 10.1109/5.326411 52
[54] C. D. Motchenbacher and J. A. Connelly, Low-Noise Electronic System
Design. Wiley-Interscience, Jun. 1993. 52
[55] W. M. Leach, Jr., “Noise relations for parallel connected transistors,” J.
Audio Eng. Soc., vol. 47, no. 3, pp. 112–118, 1999. 62
[56] SBP-10.7+ Coaxial Bandpass Filter, Mini-Circuits, P.O. Box 350166,
Brooklyn, NY 11235 U.S.A. 71
[57] J. Kozikowski, “Analysis and design of emitter followers at high frequen-
cies,” IEEE Trans. Circuit Theory, vol. 11, no. 1, pp. 129–136, Mar. 1964.
doi: 10.1109/TCT.1964.1082247 75
[58] R. Pritchard, “Frequency variations of current-amplification factor for junc-
tion transistors,” Proc. IRE, vol. 40, no. 11, pp. 1476–1481, Nov. 1952. doi:
10.1109/JRPROC.1952.273982 75
[59] J. Nerad and R. C. Tozer, “Preventing inductive output impedance of high
frequency emitter follower stages,” Radioengineering, vol. 6, no. 1, pp. 23–
28, Apr. 1997. 75
[60] I. Darwazeh and B. Wilson, “Hybrid-π common base analysis,” IEEE
Trans. Circuits Syst. I, vol. 37, no. 5, pp. 655–656, May 1990. doi:
10.1109/31.55013 75
[61] J. Gibbons, “An analysis of the modes of operation of a simple transis-
tor oscillator,” Proc. IRE, vol. 49, no. 9, pp. 1383–1390, Sep. 1961. doi:
10.1109/JRPROC.1961.287896 75
[62] J. Millman and A. Grabel, Microelectronics, 2nd ed., S. W. Director, Ed.
McGraw Hill, 1987. 75
203
REFERENCES
[63] E. H. Colpitts, “Oscillation generator,” United States of America Patent
1 624 537, Feb., 1927. 76
[64] L. Hartley, “Oscillation-generator,” United States of America Patent
1 356 763, Jun., 1920. 76
[65] J. Clapp, “An inductance-capacitance oscillator of unusual frequency
stability,” Proc. IRE, vol. 36, no. 3, pp. 356–358, Mar. 1948. doi:
10.1109/JRPROC.1948.233920 76
[66] E. H. Armstrong, “Wireless receiving system,” United States of America
Patent 1,113,149, 1914. 76
[67] R. W. Rhea, Oscillator Design and Computer Simulation, 2nd ed. Noble
Publishing Corporation, 1997. 76, 85
[68] B. Cordell, Designing Audio Power Amplifiers. McGraw Hill, 2010. 82
[69] I. E. Getreu, Modeling the Bipolar Transistor (Computer-aided design of
electronic circuits). Elsevier Science, 1978. 82
[70] W. A. Edson, Vacuum Tube Oscillators. Chapmann Hall, 1953. 85
[71] D. M. Pozar, Microwave Engineering, 2nd ed. Wiley, 1998. 85
[72] B. Razavi, Design of Analog CMOS Integrated Circuits. McGraw Hill,
2001. 86
[73] L. S. Theremin, “Method and aparatus for the generation of sounds,”
United States of America Patent 1 661 058, 1928. 89
[74] J. E. Green, R. C. Tozer, and J. P. R. David, “A transimpedance ampli-
fier for excess noise measurements of high junction capacitance avalanche
photodiodes,” IEEE Trans. Circuits Syst. I, In Review. 91
[75] T. Deliyannis, “High-Q factor circuit with reduced sensitivity,” Electronics
Letters, vol. 4, no. 26, pp. 577–579, Dec. 1968. doi: 10.1049/el:19680451 95,
103
204
REFERENCES
[76] S. L. Miller, “Ionization rates for holes and electrons in Silicon,” Phys.
Rev., vol. 105, pp. 1246–1249, Feb. 1957. doi: 10.1103/PhysRev.105.1246
110, 111
[77] A. G. Chynoweth and K. G. McKay, “Internal field emission in Silicon
p − n junctions,” Phys. Rev., vol. 106, no. 3, pp. 418–426, May 1957. doi:
10.1103/PhysRev.106.418 111
[78] K. G. Emeleus, R. W. Lunt, and C. A. Meek, “Ionization, excitation, and
chemical reaction in uniform electric fields. I. the townsend coefficient of
ionization,” Proc. R. Soc. Lond. A, vol. 156, pp. 394–411, Aug. 1936. doi:
10.1098/rspa.1936.0156 111
[79] J. Moll and R. V. Overstraeten, “Charge multiplication in Silicon p-n
junctions,” Solid-State Electronics, vol. 6, no. 2, pp. 147–157, 1963. doi:
10.1016/0038-1101(63)90009-1 111
[80] C. A. Lee, R. A. Logan, R. L. Batdorf, J. J. Kleimack, and W. Wiegmann,
“Ionization rates of holes and electrons in Silicon,” Phys. Rev., vol. 134, pp.
A761–A773, May 1964. doi: 10.1103/PhysRev.134.A761 111
[81] T. Ogawa, “Avalanche breakdown and multiplication in Silicon pin junc-
tions,” Jpn. J. Appl. Phys., vol. 4, no. 7, pp. 473–484, 1965. doi:
10.1143/JJAP.4.473 111
[82] S. M. Sze and G. Gibbons, “Avalanche breakdown voltages of abrupt and
linearly graded p − n junctions in Ge, Si, GaAs, and GaP,” Appl. Phys.
Lett., vol. 8, no. 5, pp. 111–113, Mar. 1966. doi: 10.1063/1.1754511 111
[83] C. R. Crowell and S. M. Sze, “Temperature dependence of avalanche multi-
plication in semiconductors,” Appl. Phys. Lett., vol. 9, no. 6, pp. 242–244,
Sep. 1966. doi: 10.1063/1.1754731 111
[84] W. Fulop, “Calculation of avalanche breakdown voltages of Silicon p-n
junctions,” Solid-State Electronics, vol. 10, no. 1, pp. 39–43, 1967. doi:
10.1016/0038-1101(67)90111-6 111
205
REFERENCES
[85] R. V. Overstraeten and H. D. Man, “Measurement of the ionization rates
in diffused Silicon p-n junctions,” Solid-State Electronics, vol. 13, no. 5, pp.
583–608, 1970. doi: 10.1016/0038-1101(70)90139-5 111, 112, 127
[86] P. Mars, “Temperature dependence of avalanche breakdown voltage in
p-n junctions,” Int. J. Electronics, vol. 32, no. 1, pp. 23–37, 1971. doi:
10.1080/00207217208938266 111
[87] W. N. Grant, “Electron and hole ionization rates in epitaxial Silicon at high
electric fields,” Solid-State Electronics, vol. 16, no. 10, pp. 1189–1203, 1973.
doi: 10.1016/0038-1101(73)90147-0 111
[88] D. Decker and C. Dunn, “Temperature dependence of carrier ionization
rates and saturated velocities in Silicon,” Journal of Electronic Materials,
vol. 4, pp. 527–547, 1975. doi: 10.1007/BF02666234 111
[89] N. N. Kryukova, A. S. Kyuregyan, T. T. Mnatsakanov, V. A. Kuzmin,
and V. B. Shuman, “Impact ionization coefficients of electrons and holes in
Silicon,” Sov. Phys. Semicond., vol. 9, p. 481, 1975. 111
[90] T. Rang, “Carrier impact ionization rates and their temperature-
dependence in Silicon,” Izvestiya Vysshikh Uchebnykh Zavedenii Radioelek-
tronika, vol. 28, no. 5, pp. 83–85, 1985. 111, 120
[91] V. Robbins, T. Wang, J. Tang, K. Hess, G. Stillman, R. McIntyre, and
P. Webb, “IVB-6 impact ionization coefficients in <100> and <111> Sil-
icon,” IEEE Trans. Electron Devices, vol. 31, no. 12, pp. 1977–1978, Dec.
1984. doi: 10.1109/T-ED.1984.21868 111
[92] V. M. Robbins, T. Wang, K. F. Brennan, K. Hess, and G. E. Stillman,
“Electron and hole impact ionization coefficients in (100) and in (111) si,”
J. Appl. Phys., vol. 58, no. 12, pp. 4614–4617, 1985. doi: 10.1063/1.336229
111
[93] A. Pauchard, P.-A. Besse, and R. Popovic, “Dead space effect on the wave-
length dependence of gain and noise in avalanche photodiodes,” IEEE
206
REFERENCES
Trans. Electron Devices, vol. 47, no. 9, pp. 1685–1693, Sep. 2000. doi:
10.1109/16.861578 112
[94] C. H. Tan, J. C. Clark, J. P. R. David, G. J. Rees, S. A. Plimmer,
R. C. Tozer, D. C. Herbert, D. J. Robbins, W. Y. Leong, and J. Newey,
“Avalanche noise measurement in thin si p+-i-n+ diodes,” Appl. Phys. Lett.,
vol. 76, no. 26, pp. 3926–3928, 2000. doi: 10.1063/1.126823 112, 120
[95] P. Agarwal, M. J. Goossens, V. Zieren, E. Aksen, and J. W. Slotboom,
“Impact ionization in thin Silicon diodes,” IEEE Electron Device Lett.,
vol. 25, no. 12, pp. 807–809, Dec. 2004. doi: 10.1109/LED.2004.838557 112
[96] J. Goodman, F. Leonberger, S.-Y. Kung, and R. Athale, “Optical intercon-
nections for VLSI systems,” Proc. IEEE, vol. 72, no. 7, pp. 850–866, Jul.
1984. doi: 10.1109/PROC.1984.12943 112, 128
[97] D. Miller, “Rationale and challenges for optical interconnects to elec-
tronic chips,” Proc. IEEE, vol. 88, no. 6, pp. 728–749, Jun. 2000. doi:
10.1109/5.867687 112
[98] C. Gunn, “CMOS photonics for high-speed interconnects,” IEEE Micro,
vol. 26, no. 2, pp. 58–66, Mar. 2006. doi: 10.1109/MM.2006.32 112
[99] Y. Kang and M. Morse, “Avalanche photodiodes: Silicon photonics rein-
vents avalanche photodetectors,” Laser Focus World, Oct. 2009. 112
[100] Q. Xu, B. Schmidt, S. Pradhan, and M. Lipson, “Micrometer-scale Silicon
electro-optic modulator,” Nature, vol. 435, pp. 325–327, May 2005. doi:
10.1038/nature03569 112
[101] L. Liao, D. Samara-Rubio, M. Morse, A. Liu, D. Hodge, D. Ru-
bin, U. Keil, and T. Franck, “High speed Silicon Mach-Zehnder mod-
ulator,” Opt. Express, vol. 13, no. 8, pp. 3129–3135, Apr. 2005. doi:
10.1364/OPEX.13.003129 112
[102] D. Taillaert, W. Bogaerts, P. Bienstman, T. Krauss, P. Van Daele, I. Mo-
erman, S. Verstuyft, K. De Mesel, and R. Baets, “An out-of-plane grating
207
REFERENCES
coupler for efficient butt-coupling between compact planar waveguides and
single-mode fibers,” IEEE J. Quantum Electron., vol. 38, no. 7, pp. 949–
955, Jul. 2002. doi: 10.1109/JQE.2002.1017613 112
[103] R. Soref and B. Bennett, “Electrooptical effects in Silicon,” IEEE
J. Quantum Electron., vol. 23, no. 1, pp. 123–129, Jan. 1987. doi:
10.1109/JQE.1987.1073206 112
[104] K.-W. Ang, J. W. Ng, A.-J. Lim, M.-B. Yu, G.-Q. Lo, and D.-L. Kwong,
“Waveguide-integrated Ge/Si avalanche photodetector with 105 GHz gain-
bandwidth product,” in Optical Fiber Communication (OFC), collo-
cated National Fiber Optic Engineers Conference, 2010 Conference on
(OFC/NFOEC), Mar. 2010, pp. 1–3. 112, 128
[105] M. Morse, O. Dosunmu, T. Yin, Y. Kang, G. Sarid, E. Ginsburg, R. Co-
hen, and M. Zadka, “Progress towards competitive Ge/Si photodetectors,”
in Silicon photonics and photonic intergrated circuits, ser. Proceedings of
SPIE, G. C. Righini, S. K. Honkanen, L. Pavesi, and L. Vivien, Eds., vol.
6996, no. 699614, 2008, pp. 699 614–1–699 614–11. doi: 10.1117/12.787034
112, 128
[106] Y. Kang, H.-D. Liu, M. Morse, M. J. Paniccia, M. Zadka, S. Litski,
G. Sarid, A. Pauchard, Y.-H. Kuo, H.-W. Chen, W. S. Zaoui, J. E. Bow-
ers, A. Beling, D. C. McIntosh, X. Zheng, and J. C. Campbell, “Mono-
lithic Germanium/Silicon avalanche photodiodes with 340 GHz gainband-
width product,” Nature Photonics, vol. 3, pp. 59–63, Jan. 2009. doi:
10.1038/nphoton.2008.247 112, 113, 114, 128
[107] A. R. Hawkins, T. E. Reynolds, D. R. England, D. I. Babic, M. J.
Mondry, K. Streubel, and J. E. Bowers, “Silicon heterointerface pho-
todetector,” Appl. Phys. Lett., vol. 68, no. 26, pp. 3692–3694, 1996. doi:
10.1063/1.115975 112, 113
[108] C. Masini, L. Calace, G. Assanto, H.-C. Luan, and L. Kimerling, “High-
performance p-i-n Ge on Si photodetectors for the near infrared: from model
208
REFERENCES
to demonstration,” IEEE Trans. Electron Devices, vol. 48, no. 6, pp. 1092–
1096, Jun. 2001. doi: 10.1109/16.925232 113, 128
[109] H. Kanbe, M. Miyaji, M. Hirose, N. Nitta, and M. Taniwaki, “Analysis of a
wafer bonded Ge/Si heterojunction by transmission electron microscopy,”
Appl. Phys. Lett., vol. 91, no. 14, pp. 142 119–142 119–3, Oct. 2007. doi:
10.1063/1.2795797 113, 128
[110] J. Wang and S. Lee, “Ge-photodetectors for Si-based optoelectronic integra-
tion,” Sensors, vol. 11, no. 1, pp. 696–718, 2011. doi: 10.3390/s110100696
113, 128
[111] D. Suh, J. Joo, S. Kim, and G. Kim, “High-speed RPCVD Ge
waveguide photodetector,” in Group IV Photonics, 2009. GFP ’09.
6th IEEE International Conference on, Sep. 2009, pp. 16–18. doi:
10.1109/GROUP4.2009.5338306 113, 128
[112] A. Shacham, K. Bergman, and L. Carloni, “Photonic networks-on-chip for
future generations of chip multiprocessors,” IEEE Trans. Comput., vol. 57,
no. 9, pp. 1246–1260, Sep. 2008. doi: 10.1109/TC.2008.78 113, 128
[113] J. Joo, S. Kim, I. G. Kim, K.-S. Jang, and G. Kim, “High-sensitivity
10 Gbps Ge-on-Si photoreceiver operating at λ ∼ 1.55µm,” Opt. Express,
vol. 18, no. 16, pp. 16 474–16 479, Aug. 2010. 113, 128
[114] J. E. Bowers, D. Dai, Y. Kang, and M. Morse, “High-gain high-sensitivity
resonant Ge/Si APD photodetectors,” in Infrared Technology and Applica-
tions XXXVI, B. F. Andresen, G. F. Fulop, and P. R. Norton, Eds., vol.
7660, no. 1. SPIE, 2010, p. 76603H. doi: 10.1117/12.855030 113, 128
[115] J. A. Seamons and M. S. Carroll, “High bit rate Germanium single photon
detectors for 1310nm,” in Quantum Information and Computation VI, E. J.
Donkor, A. R. Pirich, and H. E. Brandt, Eds., vol. 6976, no. 1. SPIE, 2008,
p. 697607. doi: 10.1117/12.778125 113, 128
209
REFERENCES
[116] S. Assefa, F. Xia, and Y. A. Vlasov, “Reinventing Germanium avalanche
photodetector for nanophotonic on-chip optical interconnects,” Nature, vol.
464, pp. 80–85, Mar. 2010. doi: 10.1038/nature08813 114
[117] J. Liu, J. Michel, W. Giziewicz, D. Pan, K. Wada, D. D. Cannon, S. Jongth-
ammanurak, D. T. Danielson, L. C. Kimerling, J. Chen, F. O. Ilday, F. X.
Kartner, and J. Yasaitis, “High-performance, tensile-strained Ge p-i-n pho-
todetectors on a Si platform,” Appl. Phys. Lett., vol. 87, no. 10, p. 103501,
2005. doi: 10.1063/1.2037200 114
[118] L. Chen, K. Preston, S. Manipatruni, and M. Lipson, “Integrated GHz
Silicon photonic interconnect with micrometer-scale modulators and de-
tectors,” Opt. Express, vol. 17, no. 17, pp. 15 248–15 256, Aug. 2009. doi:
10.1364/OE.17.015248 114, 128
[119] D. D.-U. Li, S. Ameer-Beg, J. Arlt, D. Tyndall, R. Walker, D. R. Matthews,
V. Visitkul, J. Richardson, and R. K. Henderson, “Time-domain fluores-
cence lifetime imaging techniques suitable for solid-state imaging sensor ar-
rays,” Sensors, vol. 12, no. 5, pp. 5650–5669, 2012. doi: 10.3390/s120505650
114
[120] D. Tyndall, B. Rae, D. Li, J. Richardson, J. Arlt, and R. Henderson, “A
100Mphoton/s time-resolved mini-silicon photomultiplier with on-chip fluo-
rescence lifetime estimation in 0.13 µm cmos imaging technology,” in Solid-
State Circuits Conference Digest of Technical Papers (ISSCC), 2012 IEEE
International, Feb. 2012, pp. 122–124. doi: 10.1109/ISSCC.2012.6176946
114
[121] D. D.-U. Li, J. Arlt, D. Tyndall, R. Walker, J. Richardson, D. Stoppa,
E. Charbon, and R. K. Henderson, “Video-rate fluorescence lifetime imag-
ing camera with cmos single-photon avalanche diode arrays and high-speed
imaging algorithm,” Journal of Biomedical Optics, vol. 16, no. 9, p. 096012,
2011. doi: 10.1117/1.3625288 114
210
REFERENCES
[122] P. Andersson, S. Montan, and S. Svanberg, “Multispectral system for med-
ical fluorescence imaging,” Quantum Electronics, IEEE Journal of, vol. 23,
no. 10, pp. 1798–1805, Oct. 1987. doi: 10.1109/JQE.1987.1073216 114
[123] A. Ehrhardt, H. Stepp, K.-M. Irion, W. Stummer, D. Zaak, R. Baum-
gartner, and A. Hofstetter, “Fluorescence detection of human malignancies
using incoherent light systems,” Medical Laser Application, vol. 18, no. 1,
pp. 27–35, 2003. doi: 10.1078/1615-1615-00084 115
[124] R. Rydell, C. Eker, S. Andersson-Engels, A. Krogdahl, P. Wahlberg, and
K. Svanberg, “Fluorescence investigations to classify malignant laryngeal
lesions in vivo,” Head & Neck, vol. 30, no. 4, pp. 419–426, Apr. 2008. 115
[125] A. Johansson, J. Svensson, N. Bendsoe, K. Svanberg, E. Alexandratou,
M. Kyriazi, D. Yova, S. Grafe, T. Trebst, and S. Andersson-Engels, “Fluo-
rescence and absorption assessment of a lipid mTHPC formulation following
topical application in a non-melanotic skin tumor model,” J. Biomed. Opt.,
vol. 12, no. 3, p. 034026, 2007. doi: 10.1117/1.2743080 115
[126] S. Palsson, U. Stenram, M. S. Thompson, A. Vaitkuviene, V. Poskiene,
R. Ziobakiene, J. Oyama, M. J. DeWeert, U. Gustafsson, N. Bendsoe,
S. Andersson-Engels, S. Svanberg, and K. Svanberg, “Methods for detailed
histopathological investigation and localization of biopsies from cervix uteri
to improve the interpretation of autofluorescence data,” J Environ Pathol,
Toxicol & Oncol., vol. 25, no. 1-2, pp. 321–40., 2006. 115
[127] M. Gersbach, Y. Maruyama, R. Trimananda, M. Fishburn, D. Stoppa,
J. Richardson, R. Walker, R. Henderson, and E. Charbon, “A time-resolved,
low-noise single-photon image sensor fabricated in deep-submicron cmos
technology,” IEEE J. Solid-State Circuits, vol. 47, no. 6, pp. 1394–1407,
Jun. 2012. doi: 10.1109/JSSC.2012.2188466 115
[128] J. Richardson, E. Webster, L. Grant, and R. Henderson, “Scaleable single-
photon avalanche diode structures in nanometer CMOS technology,” IEEE
Trans. Electron Devices, vol. 58, no. 7, pp. 2028 –2035, Jul. 2011. doi:
10.1109/TED.2011.2141138 115
211
REFERENCES
[129] R. Henderson, E. Webster, R. Walker, J. Richardson, and L. Grant, “A
3 x 3, 5 µm pitch, 3-transistor single photon avalanche diode array with
integrated 11V bias generation in 90 nm cmos technology,” in Electron
Devices Meeting (IEDM), 2010 IEEE International, Dec. 2010, pp. 14.2.1–
14.2.4. doi: 10.1109/IEDM.2010.5703359 115
[130] L. Braga, L. Pancheri, L. Gasparini, M. Perenzoni, R. Walker, R. Hender-
son, and D. Stoppa, “A cmos mini-sipm detector with in-pixel data com-
pression for pet applications,” in Nuclear Science Symposium and Medical
Imaging Conference (NSS/MIC), 2011 IEEE, Oct. 2011, pp. 548–552. doi:
10.1109/NSSMIC.2011.6154110 115
[131] Y. Wang, B. R. Rae, R. K. Henderson, Z. Gong, J. Mckendry, E. Gu, M. D.
Dawson, G. A. Turnbull, and I. D. W. Samuel, “Ultra-portable explosives
sensor based on a CMOS fluorescence lifetime analysis micro-system,” AIP
Advances, vol. 1, no. 3, p. 032115, 2011. doi: 10.1063/1.3624456 115
[132] J. Degnan, R. Machan, E. Leventhal, G. Jodor, C. Field, and J. Sirota,
“Cryosphere and biomass measurements using a photon-counting 3d imag-
ing lidar,” in Lasers and Electro-Optics (CLEO), 2011 Conference on, may
2011, pp. 1–2. 115
[133] C. Niclass, M. Sergio, and E. Charbon, “A single photon avalanche diode
array fabricated in 0.35 µm CMOS and based on an event-driven read-
out for TCSPC experiments,” in Advanced Photon Counting Techniques,
ser. Proceedings of the Society of Photo-Optical Instrumentation Engineers
(SPIE), W. Becker, Ed., vol. 6372, SPIE; Boston Elect Corp; Becker & Hickl
GmbH. SPIE-INT SOC Optical Engineering, 2006, pp. U216–U227. doi:
10.1117/12.685974 115
[134] D. E. Aspens, Properties of Silicon, ser. EMIS Datareviews Series. IN-
SPEC, The Institution of Electrical Engineers, 1988, no. 4. 120
[135] M. Tyagi and R. V. Overstraeten, “Minority carrier recombination in
heavily-doped Silicon,” Solid-State Electronics, vol. 26, no. 6, pp. 577–597,
1983. doi: 10.1016/0038-1101(83)90174-0 120
212
REFERENCES
[136] J. A. del Alamo and R. M. Swanson, “Modelling of minority-carrier trans-
port in heavily doped Silicon emitters,” Solid-State Electronics, vol. 30,
no. 11, pp. 1127–1136, 1987. doi: 10.1016/0038-1101(87)90077-3 120
[137] G. E. Stillman and C. M. Wolfe, Avalanche Photodiodes, ser. Semiconduc-
tors and Semimetals. Academic Press, 1977, vol. 12, p. 291. 127
[138] Y. U. Goldberg, M. E. Levinshtein, and S. L. Rumyantsev, Eds., Properties
of Advanced Semiconductor Materials GaN, AlN, SiC, BN, SiGe. John
Wiley & Sons, Jan. 2001. 129
[139] M. Razeghi, “Short-wavelength solar-blind detectors-status, prospects, and
markets,” Proc. IEEE, vol. 90, no. 6, pp. 1006–1014, Jun. 2002. doi:
10.1109/JPROC.2002.1021565 129, 157
[140] K. Vassilevski, K. Zekentes, A. Zorenko, and L. Romanov, “Experimen-
tal determination of electron drift velocity in 4H-SiC p+-n-n+ avalanche
diodes,” IEEE Electron Device Lett., vol. 21, no. 10, pp. 485–487, Oct.
2000. doi: 10.1109/55.870609 129, 137
[141] A. Powell and L. Rowland, “SiC materials-progress, status, and poten-
tial roadblocks,” Proc. IEEE, vol. 90, no. 6, pp. 942–955, Jun. 2002. doi:
10.1109/JPROC.2002.1021560 129, 130
[142] P. G. Neudeck, The VLSI Handbook, 2nd ed. CRC Press, Dec. 2006, ch.
Silicon Carbide Technology. 129, 131
[143] A. Blanc, L. Deimling, and N. Eisenreich, “UV- and IR-signatures of rocket
plumes,” Propellants, Explosives, Pyrotechnics, vol. 27, no. 3, pp. 185–189,
2002. 130
[144] R. S. Okojie, D. Spry, J. Krotine, C. Salupo, and D. R. Wheeler, “Stable
Ti/TaSi2/Pt ohmic contacts on N-Type 6H-SiC epilayer at 600C in air,” in
Materials Research Society Symposia Proceedings, ser. Wide-Bandgap Elec-
tronic Devices, R. J. Shul, F. Ren, M. Murakami, and W. Pletschen, Eds.,
vol. 622. Materials Research Society, 2000. doi: 10.1557/PROC-622-T8.3.1
130
213
REFERENCES
[145] P. G. Neudeck, D. J. Spry, L.-Y. Chen, C. W. Chang, G. M. Beheim, R. S.
Okojie, L. J. Evans, R. Meredith, T. Ferrier, M. J. Krasowski, and N. F.
Prokop, “Long-term characterization of 6H-SiC transistor integrated cir-
cuit technology operating at 500C,” in Silicon Carbide 2008 - Materials,
Processing and Devices, ser. Materials Research Society Symposium Pro-
ceedings, M. Dudley, A. R. Powell, C. M. Johnson, and S.-H. Ryu, Eds.,
vol. 1069. Materials Research Society, 2008, pp. 209–214. 130
[146] Y. Tairov and V. Tsvetkov, “Investigation of growth processes of ingots of
Silicon Carbide single crystals,” Journal of Crystal Growth, vol. 43, no. 2,
pp. 209–212, 1978. doi: 10.1016/0022-0248(78)90169-0 130
[147] Y. M. Tairov and V. F. Tsvetkov, “General principles of growing
large-size single crystals of various Silicon Carbide polytypes,” Jour-
nal of Crystal Growth, vol. 52, Part 1, no. 0, pp. 146–150, 1981. doi:
10.1016/0022-0248(81)90184-6 130
[148] A. R. Powell, J. J. Sumakeris, R. T. Leonard, M. F. Brady, S. G. Muller,
V. F. Tsvetkov, H. M. Hobgood, A. A. Burk, M. J. Paisley, R. C. Glass,
and C. H. J. Carter, “Status of 4H-SiC substrate and epitaxial materials
for commercial power applications,” inMaterials Research Society Symposia
Proceedings, ser. Silicon Carbide 2004 – Materials, Processing and Devices,
R. J. Shul, F. Ren, M. Murakami, and W. Pletschen, Eds., vol. 815. Ma-
terials Research Society, 2004, pp. 3–14. doi: 10.1557/PROC-622-T8.3.1
130
[149] A. Burk, “Development of multiwafer warm-wall planetary vpe reactors for
sic device production,” Chemical Vapor Deposition, vol. 12, no. 8-9, pp.
465–473, 2006. doi: 10.1002/cvde.200500028 130
[150] F. C. Frank, “Capillary equilibria of dislocated crystals,” Acta
Crystallographica, vol. 4, no. 6, pp. 497–501, Nov. 1951. doi:
10.1107/S0365110X51001690 131
[151] B. G. Streetman and S. Banergee, Solid State Electronic Devices, 5th ed.
Prentice Hall, 2001. 131
214
REFERENCES
[152] Y. S. Park, SiC Material and Devices, ser. Semiconductors and Semimetals.
Academic Press, 1998, vol. 52. 131, 146
[153] D. Lovig. (2007, May) Cree demonstrates 100-mm zero-micropipe Silicon
Carbide substrates. Internet. Available: http://www.cree.com/press/
press detail.asp?i=1179923837641 132, 135
[154] R. S. Kennedy, “Communication through optical scattering channels: An
introduction,” Proc. IEEE, vol. 58, no. 10, pp. 1651–1665, Oct. 1970. doi:
10.1109/PROC.1970.7987 132
[155] M. R. Luettgen, J. H. Shapiro, and D. M. Reilly, “Non-line-of-sight single-
scatter propagation model,” J. Opt. Soc. Am., vol. 8, pp. 1964–1972, 1991.
doi: 10.1364/JOSAA.8.001964 132, 133
[156] H. C. van de Hulst, Light Scattering by Small Particles. New York: Wiley,
1957. 132
[157] G. Chen, F. Abou-Galala, Z. Xu, and B. M. Sadler, “Experimental evalua-
tion of LED-based solar blind NLOS communication links,” Opt. Express,
vol. 16, no. 19, pp. 15 059–15 068, Sep. 2008. doi: 10.1364/OE.16.015059
133, 193
[158] G. Shaw, A. Siegel, and M. Nischan, “Demonstration system and appli-
cations for compact wireless ultraviolet communications,” in Sensors, and
command, control, communications, and intelligence (C3I) technologies for
homeland defence and law enforcement II, ser. Proc. SPIE, E. Carapezza,
Ed., vol. 5071. SPIE, 2003, pp. 241–252. doi: 10.1117/12.500861 133, 193
[159] R. L. Wolfe, “Ultraviolet disinfection of potable water,” Environmen-
tal Science & Technology, vol. 24, no. 6, pp. 768–773, 1990. doi:
10.1021/es00076a001 133
[160] M. R. DeVries. (2005, Nov.) Trojan technologies wins new york city
drinking water UV project. Internet. Available: http://www.trojanuv.com/
resources/trojanuv//News/Press/New York City Press Release.pdf 133
215
REFERENCES
[161] C. Chatterley and K. Linden, “Demonstration and evaluation of germicidal
UV-LEDs for point-of-use water disinfection,” Journal of Water and Health,
vol. 8, no. 3, pp. 479–486, 2010. doi: 10.2166/wh.2010.124 133
[162] S. A. Brownell, A. R. Chakrabarti, F. M. Kaser, L. G. Connelly, R. L.
Peletz, F. Reygadas, M. J. Lang, D. M. Kammen, and K. L. Nelson, “As-
sessment of a low-cost, point-of-use, ultraviolet water disinfection technol-
ogy,” Journal of Water and Health, vol. 6, no. 1, pp. 53–65, Mar. 2008. doi:
10.2166/wh.2007.015 133
[163] A. Hamamoto, M. Mori, A. Takahashi, M. Nakano, N. Wakikawa,
M. Akutagawa, T. Ikehara, Y. Nakaya, and Y. Kinouchi, “New wa-
ter disinfection system using UVA light-emitting diodes,” Journal of
Applied Microbiology, vol. 103, no. 6, pp. 2291–2298, Dec. 2007. doi:
10.1111/j.1365-2672.2007.03464.x 133
[164] M. D. Sobsey, C. E. Stauber, L. M. Casanova, J. M. Brown, and M. A.
Elliott, “Point of use household drinking water filtration: A practical, ef-
fective solution for providing sustained access to safe drinking water in the
developing world,” Environmental Science & Technology, vol. 42, no. 12,
pp. 4261–4267, Jun. 2008. doi: 10.1021/es702746n 133
[165] A. Pruss, D. Kay, L. Fewtrell, and J. Bartram, “Estimating the burden of
disease from water, sanitation, and hygiene at a global level,” Environmen-
tal Health Perspectives, vol. 110, no. 5, pp. 537–542, May 2002. 134
[166] B. Baliga, “The future of power semiconductor device technology,” Proc.
IEEE, vol. 89, no. 6, pp. 822–832, Jun. 2001. doi: 10.1109/5.931471 134
[167] M. S. Mazzola and J. N. Merrett, “Monolithic vertical junction field ef-
fect transistor and schottky barrier diode fabricated from Silicon Carbide
and method for fabricating the same,” United States of America Patent
7 294 860, Nov., 2007. 134
[168] Semisouth announces two new 1700V and 1200V SiC JFET products.
Available: http://semisouth.com/archives/191 134
216
REFERENCES
[169] (2011, Feb.) Semisouth ships first normally-off SiC JFET for up to 50kW.
Available: http://www.semiconductor-today.com/news items/2011/FEB/
SEMISOUTH 280211.html 134
[170] Datasheet: SJEP170R550. Available: http://semisouth.com/wp-content/
uploads/2011/05/DS SJEP170R550 rev1.4.pdf 134
[171] Datasheet: Semisouth 76.2mm Silicon carbide epitaxial wafer (4H polytype,
4off-axis). Available: http://semisouth.com/wp-content/uploads/2011/
05/76.2mmEpiwaferSpecificationSheetRevB.pdf 135
[172] (2011, Jan.) Cree launches industry’s first commercial Silicon carbide
power MOSFET; destined to replace Silicon devices in high-voltage
(>1200-V) power electronics. Internet. Available: http://www.cree.com/
press/press detail.asp?i=1295272745318 135
[173] R. Malhan, Y. Takeuchi, M. Kataoka, A.-P. Mihaila, S. Rashid, F. Udrea,
and G. Amaratunga, “Normally-off trench JFET technology in 4H Silicon
Carbide,” Microelectronic Engineering, vol. 83, no. 1, pp. 107–111, 2006, the
Symposium K Proceedings of the 3rd International Conference on Materials
for Advanced Technologies (ICMAT 2005). doi: 10.1016/j.mee.2005.10.035
135
[174] “Methods of fabricating oxide layers on Silicon Carbide layers utilizing
atomic oxygen,” United States of America Patent 7,572,741 B2, Aug., 2009.
135
[175] C. Harris, K. Bertilsson, and A. Konstantinov, “Silicon Carbide self-aligned
epitaxial MOSFET for high powered device applications,” United States of
America Patent 7,629,616, Dec., 2009. 135
[176] J. Zhang, Y. Luo, P. Alexandrov, L. Fursin, and J. Zhao, “A high current
gain 4H-SiC NPN power bipolar junction transistor,” IEEE Electron Device
Lett., vol. 24, no. 5, pp. 327–329, May 2003. doi: 10.1109/LED.2003.812531
135
217
REFERENCES
[177] J. Zhang, P. Alexandrov, T. Burke, and J. Zhao, “4H-SiC power bipolar
junction transistor with a very low specific ON-resistance of 2.9 mΩ·cm2,”
IEEE Electron Device Lett., vol. 27, no. 5, pp. 368–370, May 2006. doi:
10.1109/LED.2006.873370 135
[178] P. Neudeck and J. Powell, “Performance limiting micropipe defects in Sili-
con Carbide wafers,” IEEE Electron Device Lett., vol. 15, no. 2, pp. 63–65,
Feb. 1994. doi: 10.1109/55.285372 135
[179] T. Kimoto, N. Miyamoto, and H. Matsunami, “Performance limiting sur-
face defects in SiC epitaxial p-n junction diodes,” IEEE Trans. Electron
Devices, vol. 46, no. 3, pp. 471–477, Mar. 1999. doi: 10.1109/16.748864
135, 145
[180] P. Neudeck and C. Fazi, “Study of bulk and elementary screw dislocation
assisted reverse breakdown in low-voltage (<250 V) 4H-SiC p+-n junction
diodes. II. dynamic breakdown properties,” IEEE Trans. Electron Devices,
vol. 46, no. 3, pp. 485–492, Mar. 1999. doi: 10.1109/16.748866 135
[181] B. Yang, T. Li, K. Heng, C. Collins, S. Wang, J. Carrano, R. Dupuis,
J. Campbell, M. Schurman, and I. Ferguson, “Low dark current GaN
avalanche photodiodes,” IEEE J. Quantum Electron., vol. 36, no. 12, pp.
1389–1391, Dec. 2000. doi: 10.1109/3.892557 135, 157
[182] H. Felsl and G. Wachutka, “Performance of 4H-SiC schottky diodes
with Al-doped p-guard-ring junction termination at reverse bias,” in Sil-
icon Carbide and Related Materials 2001, PTS 1 and 2, Proceedings,
ser. Materials Science Forum, S. Yoshida, S. Nishino, H. Harima, and
T. Kimoto, Eds., vol. 389-3. Trans. Tech., 2002, pp. 1153–1156. doi:
10.4028/www.scientific.net/MSF.389-393.1153 135
[183] X. Li, K. Tone, L. Fursin, J. Zhao, T. Burke, P. Alexandrov, M. Pan,
and M. Weiner, “Multistep junction termination extension for SiC power
devices,” Electronics Letters, vol. 37, no. 6, pp. 392–393, Mar. 2001. doi:
10.1049/el:20010258 135
218
REFERENCES
[184] T. Kimoto, T. Urushidani, S. Kobayashi, and H. Matsunami, “High-
voltage (>1 kV) SiC schottky barrier diodes with low on-resistances,”
IEEE Electron Device Lett., vol. 14, no. 12, pp. 548–550, Dec. 1993. doi:
10.1109/55.260785 135
[185] A. O. Konstantinov, “The temperature dependence of impact ionization in
Silicon Carbide and related effects,” in Amorphous and Crystalline Silicon
Carbide III, ser. Springer Proceedings in Physics, vol. 5. Springer, 1992,
pp. 213–219. 135
[186] A. O. Konstantinov, Q. Wahab, N. Nordell, and U. Lindefelt, “Ionization
rates and critical fields in 4H Silicon Carbide,” Appl. Phys. Lett., vol. 71,
no. 1, pp. 90–92, 1997. doi: 10.1063/1.119478 136, 137, 138, 139, 146, 156
[187] A. Konstantinov, Q. Wahab, N. Nordell, and U. Lindefelt, “Study
of avalanche breakdown and impact ionization in 4H Silicon Car-
bide,” Journal of Electronic Materials, vol. 27, pp. 335–341, 1998. doi:
10.1007/s11664-998-0411-x 136, 137, 138, 146
[188] K. K. Thornber, “Applications of scaling to problems in high-field elec-
tronic transport,” J. Appl. Phys., vol. 52, no. 1, pp. 279–290, 1981. doi:
10.1063/1.328490 136
[189] R. Raghunathan and B. Baliga, “Measurement of electron and hole impact
ionization coefficients for SiC,” in Power Semiconductor Devices and IC’s,
1997. ISPSD ’97., 1997 IEEE International Symposium on, May 1997, pp.
173–176. doi: 10.1109/ISPSD.1997.601463 136, 139
[190] F. Yan, C. Qin, J. Zhao, M. Weiner, B. Ng, J. David, and R. Tozer, “Low-
noise visible-blind UV avalanche photodiodes with edge terminated by 2
positive bevel,” Electronics Letters, vol. 38, no. 7, pp. 335–336, Mar. 2002.
doi: 10.1049/el:20020216 136
[191] B. Ng, J. David, R. Tozer, G. Rees, F. Yan, J. Zhao, and M. Weiner,
“Nonlocal effects in thin 4H-SiC UV avalanche photodiodes,” IEEE
Trans. Electron Devices, vol. 50, no. 8, pp. 1724–1732, Aug. 2003. doi:
10.1109/TED.2003.815144 136, 139, 142, 146, 153, 156, 159
219
REFERENCES
[192] T. Hatakeyama, T. Watanabe, T. Shinohe, K. Kojima, K. Arai, and
N. Sano, “Impact ionization coefficients of 4H Silicon Carbide,” Appl. Phys.
Lett., vol. 85, no. 8, pp. 1380–1382, 2004. doi: 10.1063/1.1784520 136, 137,
139
[193] X. Guo, L. Rowland, G. Dunne, J. Fronheiser, P. Sandvik, A. Beck, and
J. Campbell, “Demonstration of ultraviolet separate absorption and mul-
tiplication 4H-SiC avalanche photodiodes,” IEEE Photon. Technol. Lett.,
vol. 18, no. 1, pp. 136–138, Jan. 2006. doi: 10.1109/LPT.2005.860384 136,
137, 148
[194] X. Guo, A. Beck, Z. Huang, N. Duan, J. Campbell, D. Emerson, and
J. Sumakeris, “Performance of low-dark-current 4H-SiC avalanche photodi-
odes with thin multiplication layer,” IEEE Trans. Electron Devices, vol. 53,
no. 9, pp. 2259–2265, Sep. 2006. doi: 10.1109/TED.2006.879677 136
[195] W. Loh, B. Ng, J. Ng, S. Soloviev, H.-Y. Cha, P. Sandvik, C. John-
son, and J. David, “Impact ionization coefficients in 4H-SiC,” IEEE
Trans. Electron Devices, vol. 55, no. 8, pp. 1984–1990, Aug. 2008. doi:
10.1109/TED.2008.926679 137, 139, 142, 146, 156, 159
[196] R. Raghunathan and B. Baliga, “Temperature dependence of hole impact
ionization coefficients in 4H and 6H-SiC,” Solid-State Electronics, vol. 43,
no. 2, pp. 199–211, 1999. doi: 10.1016/S0038-1101(98)00248-2 137, 138
[197] P. Neudeck and C. Fazi, “Positive temperature coefficient of breakdown
voltage in 4H-SiC pn junction rectifiers,” IEEE Electron Device Lett.,
vol. 18, no. 3, pp. 96–98, Mar. 1997. doi: 10.1109/55.556092 137
[198] Y. S. Lee, M. K. Han, and Y. I. Choi, “The temperature dependent break-
down voltage for 4H- and 6H-SiC diodes,” MRS Proceedings, vol. 622, p.
T6.4.1, 2000. doi: 10.1557/PROC-622-T6.4.1 137
[199] B. Ozpineci and L. Tolbert, “Characterization of SiC schottky diodes at
different temperatures,” IEEE Power Electron. Lett., vol. 1, no. 2, pp. 54–
57, Jun. 2003. doi: 10.1109/LPEL.2003.821026 137
220
REFERENCES
[200] F. Yan, J. H. Zhao, and G. H. Olsen, “Demonstration of the first 4H-SiC
avalanche photodiodes,” Solid-State Electronics, vol. 44, no. 2, pp. 341–346,
2000. doi: 10.1016/S0038-1101(99)00240-3 137
[201] S. Nakamura, H. Kumagai, T. Kimoto, and H. Matsunami, “Anisotropy in
breakdown field of 4H-SiC,” Appl. Phys. Lett., vol. 80, no. 18, pp. 3355–
3357, 2002. doi: 10.1063/1.1477271 137
[202] Y. Tanaka, K. Fukuda, K. Arai, K. Kojima, T. Suzuki, and T. Yat-
suo, “Reverse characteristics of pn diodes on 4H-SiC (000-1)C and (11-
20) face,” Appl. Phys. Lett., vol. 84, no. 10, pp. 1774–1776, 2004. doi:
10.1063/1.1652235 138
[203] L. V. Keldysh, “Effect of ultrasonics on the electron spectrum of crystals,”
Soviet Physics-Solid State, vol. 4, no. 8, pp. 1658–1659, 1963. 138
[204] E. Bellotti, H.-E. Nilsson, K. F. Brennan, P. P. Ruden, and R. Trew, “Monte
carlo calculation of hole initiated impact ionization in 4H phase SiC,” J.
Appl. Phys., vol. 87, no. 8, pp. 3864–3871, Apr. 2000. doi: 10.1063/1.372426
138, 139, 156
[205] K. Bertilsson, H.-E. Nilsson, and C. Petersson, “Simulation of anisotropic
breakdown in 4H-SiC diodes,” in Computers in Power Electronics,
2000. COMPEL 2000. The 7th Workshop on, 2000, pp. 118–120. doi:
10.1109/CIPE.2000.904702 138
[206] F. Yan, C. Qin, J. Zhao, M. Bush, G. Olsen, B. Ng, J. David, R. Tozer,
and M. Weiner, “Demonstration of 4H-SiC avalanche photodiodes linear
array,” Solid-State Electronics, vol. 47, no. 2, pp. 241–245, 2003. doi:
10.1016/S0038-1101(02)00201-0 140
[207] C.-T. Sah, R. Noyce, and W. Shockley, “Carrier generation and recombina-
tion in P-N junctions and P-N junction characteristics,” Proc. IRE, vol. 45,
no. 9, pp. 1228–1243, Sep. 1957. doi: 10.1109/JRPROC.1957.278528 145
221
REFERENCES
[208] S. G. Sridhara, R. P. Devaty, and W. J. Choyke, “Absorption coefficient of
4H Silicon Carbide from 3900 to 3250 A,” J. Appl. Phys., vol. 84, no. 5, pp.
2963–2964, 1998. doi: 10.1063/1.368403 148, 149
[209] S. G. Sridhara, T. J. Eperjesi, R. P. Devaty, and W. J. Choyke, “Penetration
depths in the ultraviolet for 4H, 6H and 3C Silicon Carbide at seven common
laser pumping wavelengths,” Materials Science and Engineering: B, vol.
61-62, pp. 229–233, 1999. doi: 10.1016/S0921-5107(98)00508-X 148, 149
[210] M. Razeghi and A. Rogalski, “Semiconductor ultraviolet detectors,” J.
Appl. Phys., vol. 79, no. 10, pp. 7433–7473, 1996. doi: 10.1063/1.362677
157, 158
[211] K. A. McIntosh, R. J. Molnar, L. J. Mahoney, A. Lightfoot, M. W. Geis,
K. M. Molvar, I. Melngailis, R. L. Aggarwal, W. D. Goodhue, S. S. Choi,
D. L. Spears, and S. Verghese, “GaN avalanche photodiodes grown by hy-
dride vapor-phase epitaxy,” Appl. Phys. Lett., vol. 75, no. 22, pp. 3485–
3487, Nov. 1999. doi: 10.1063/1.125363 157, 158
[212] J. C. Carrano, D. J. H. Lambert, C. J. Eiting, C. J. Collins, T. Li, S. Wang,
B. Yang, A. L. Beck, R. D. Dupuis, and J. C. Campbell, “GaN avalanche
photodiodes,” Appl. Phys. Lett., vol. 76, no. 7, pp. 924–926, Feb. 2000. doi:
10.1063/1.125631 157, 158
[213] S. Verghese, K. McIntosh, R. Molnar, L. Mahoney, R. Aggarwal, M. Geis,
K. Molvar, E. Duerr, and I. Melngailis, “GaN avalanche photodiodes oper-
ating in linear-gain mode and geiger mode,” IEEE Trans. Electron Devices,
vol. 48, no. 3, pp. 502–511, Mar. 2001. doi: 10.1109/16.906443 157
[214] R. McClintock, A. Yasan, K. Mayes, D. Shiell, S. R. Darvish, P. Kung,
and M. Razeghi, “High quantum efficiency AlGaN solar-blind p-i-n photo-
diodes,” Appl. Phys. Lett., vol. 84, no. 8, pp. 1248–1250, Feb. 2004. doi:
10.1063/1.1650550 157
[215] R. McClintock, A. Yasan, K. Minder, P. Kung, and M. Razeghi,
“Avalanche multiplication in AlGaN based solar-blind photodetectors,”
222
REFERENCES
Appl. Phys. Lett., vol. 87, no. 24, pp. 241 123–241 123–3, Dec. 2005. doi:
10.1063/1.2140610 157
[216] S.-C. Shen, Y. Zhang, D. Yoo, J.-B. Limb, J.-H. Ryou, P. Yoder, and
R. Dupuis, “Performance of deep ultraviolet GaN avalanche photodiodes
grown by MOCVD,” IEEE Photon. Technol. Lett., vol. 19, no. 21, pp.
1744–1746, Nov. 2007. doi: 10.1109/LPT.2007.906052 157, 158
[217] Y. Yoshizumi, S. Hashimoto, T. Tanabe, and M. Kiyama, “High-
breakdown-voltage pn-junction diodes on GaN substrates,” Journal of
Crystal Growth, vol. 298, no. 0, pp. 875–878, 2007, thirteenth International
Conference on Metal Organic Vapor Phase Epitaxy (ICMOVPE XIII). doi:
10.1016/j.jcrysgro.2006.10.246 157
[218] H. Jiang and T. Egawa, “High quality AlGaN solar-blind schottky photo-
diodes fabricated on AIN/sapphire template,” Appl. Phys. Lett., vol. 90,
no. 12, p. 121121, 2007. doi: 10.1063/1.2715479 157
[219] T. Tut, M. Gokkavas, A. Inal, and E. Ozbay, “AlxGa1−xN-based avalanche
photodiodes with high reproducible avalanche gain,” Appl. Phys. Lett.,
vol. 90, no. 16, p. 163506, 2007. doi: 10.1063/1.2724926 157
[220] J. Limb, D. Yoo, Y. Zhang, J. Ryou, S. Shen, and R. Dupuis, “GaN ul-
traviolet avalanche photodiodes grown on 6H-SiC substrates with SiN pas-
sivation,” Electronics Letters, vol. 44, no. 4, pp. 313–314, Feb. 2008. doi:
10.1049/el:20082830 157
[221] J. L. Pau, C. Bayram, R. McClintock, M. Razeghi, and D. Silversmith,
“Back-illuminated separate absorption and multiplication GaN avalanche
photodiodes,” Appl. Phys. Lett., vol. 92, no. 10, p. 101120, 2008. doi:
10.1063/1.2897039 157
[222] T. Tut, T. Yelboga, E. Ulker, and E. Ozbay, “Solar-blind AlGaN-based
p-i-n photodetectors with high breakdown voltage and detectivity,” Appl.
Phys. Lett., vol. 92, no. 10, p. 103502, 2008. doi: 10.1063/1.2895643 157
223
REFERENCES
[223] C. Bayram, J. L. Pau, R. McClintock, and M. Razeghi, “Performance en-
hancement of GaN ultraviolet avalanche photodiodes with p-type delta-
doping,” Appl. Phys. Lett., vol. 92, no. 24, p. 241103, 2008. doi:
10.1063/1.2948857 157
[224] C. Bayram, J. L. Pau, R. McClintock, M. Razeghi, M. P. Ulmer, and
D. Silversmith, “High quantum efficiency back-illuminated GaN avalanche
photodiodes,” Appl. Phys. Lett., vol. 93, no. 21, p. 211107, 2008. doi:
10.1063/1.3039061 158
[225] L. Sun, J. Chen, J. Li, and H. Jiang, “AlGaN solar-blind avalanche photo-
diodes with high multiplication gain,” Appl. Phys. Lett., vol. 97, no. 19, p.
191103, 2010. doi: 10.1063/1.3515903 158
[226] E. Cicek, Z. Vashaei, R. McClintock, C. Bayram, and M. Razeghi, “Geiger-
mode operation of ultraviolet avalanche photodiodes grown on sapphire and
free-standing GaN substrates,” Appl. Phys. Lett., vol. 96, no. 26, p. 261107,
2010. doi: 10.1063/1.3457783 158
[227] W. S. Loh, “Studies of avalanche breakdown in 4H-SiC and reliability of
Al wire bonds for high temperature and power applications,” Ph.D. thesis,
The University of Sheffield Dept. of Electronic and Electrical Engineering,
Oct. 2008. 158
[228] E. Monroy, F. Omnes, and F. Calle, “Wide-bandgap semiconductor ultravi-
olet photodetectors,” Semicond. Sci. Technol., vol. 18, no. 4, pp. R33–R51,
2003. doi: doi:10.1088/0268-1242/18/4/201 158
[229] V. A. Dmitriev, K. G. Irvine, J. C. H. Carter, N. I. Kuznetsov, and E. V.
Kalinina, “Electric breakdown in GaN p-n junctions,” Appl. Phys. Lett.,
vol. 68, no. 2, pp. 229–231, 1996. doi: 10.1063/1.116469 158
[230] Y. Wu, P. Parlkh, and V. Mishra, “High voltage GaN transistors,” United
States of America Patent 7,692,263, 2010. 159
[231] A. Bandyopadhyay, M. Deen, L. Tarof, and W. Clark, “A simplified ap-
proach to time-domain modeling of avalanche photodiodes,” IEEE J. Quan-
224
REFERENCES
tum Electron., vol. 34, no. 4, pp. 691–699, Apr. 1998. doi: 10.1109/3.663452
162
[232] J.-J. Jou, C.-K. Liu, C.-M. Hsiao, H.-H. Lin, and H.-C. Lee, “Time-delay
circuit model of high-speed p-i-n photodiodes,” IEEE Photon. Technol.
Lett., vol. 14, no. 4, pp. 525–527, Apr. 2002. doi: 10.1109/68.992599 162,
164
[233] A. Zarifkar and M. Soroosh, “Circuit modeling of separate absorp-
tion, charge and multiplication avalanche photodiode (SACM-APD),” in
Laser and Fiber-Optical Networks Modeling, 2004. Proceedings of LFNM
2004. 6th International Conference on, Sep. 2004, pp. 213–219. doi:
10.1109/LFNM.2004.1382469 162
[234] Y. El-Batawy and M. Deen, “Analysis and circuit modeling of waveguide-
separated absorption charge multiplication-avalanche photodetector (WG-
SACM-APD),” IEEE Trans. Electron Devices, vol. 52, no. 3, pp. 335–344,
Mar. 2005. doi: 10.1109/TED.2005.843884 162
[235] X.-X. Fang, Xie-Sheng, C.-X. Hong, and Chen-Chao, “Circuit
modeling of SAGCM-APD,” in Nanoelectronics Conference, 2008.
INEC 2008. 2nd IEEE International, Mar. 2008, pp. 689–694. doi:
10.1109/INEC.2008.4585579 162
[236] J. Wu and G. Wang, “A novel equivalent circuit model for separate ab-
sorption grading charge multiplication avalanche photodiode (APD)-based
optical receiver,” J. Lightw. Technol., vol. 28, no. 5, pp. 784–790, Mar.
2010. doi: 10.1109/JLT.2009.2037594 162
[237] D. Dai and J. Bowers, “Simple matrix-method modeling for avalanche pho-
todetectors with arbitrary layer structures and absorption/multiplication
coefficients,” J. Lightw. Technol., vol. 28, no. 9, pp. 1404–1413, May 2010.
doi: 10.1109/JLT.2010.2042279 162
[238] M. Jalali, M. K. Moravvej-Farshi, S. Masudy-Panah, and A. Nabavi,
“An equivalent lumped circuit model for thin avalanche photodiodes with
225
REFERENCES
nonuniform electric field profile,” J. Lightw. Technol., vol. 28, no. 23, pp.
3395–3402, Dec. 2010. doi: 10.1109/JLT.2010.2084987 162
[239] S. Dimler, J. Ng, R. Tozer, G. Rees, and J. David, “Capacitive quenching
measurement circuit for geiger-mode avalanche photodiodes,” Selected Top-
ics in Quantum Electronics, IEEE Journal of, vol. 13, no. 4, pp. 919–925,
Jul. 2007. doi: 10.1109/JSTQE.2007.903595 191
[240] J. Puschell and R. Bayse, “High data rate ultraviolet communication sys-
tems for the tactical battlefield,” in Tactical Communications Conference,
1990. Vol.1. Tactical Communications. ‘Challenges of the 1990’s’., Proceed-
ings of the, Apr. 1990, pp. 253–267. doi: 10.1109/TCC.1990.177806 193
[241] G. A. Shaw, M. L. Nischan, M. A. Iyengar, S. Kaushik, and M. K. Griffin,
“NLOS UV communication for distributed sensor systems,” P. Hamburger,
Ed., vol. 4126, no. 1. SPIE, 2000, pp. 83–96. doi: 10.1117/12.407519 193
[242] A. M. Siegel, G. A. Shaw, and J. Model, “Short-range communication
with ultraviolet LEDs,” I. T. Ferguson, N. Narendran, S. P. DenBaars,
and J. C. Carrano, Eds., vol. 5530, no. 1. SPIE, 2004, pp. 182–193. doi:
10.1117/12.582112 193
226
Appendix A
A One Dimensional Poisson
Solver
This appendix lists the expressions required to model a 2, 3, 4 or 5 region semi-
conductor. This model is used to model the SiC devices in Chapter 7. Figure A.1
shows the progression of depletion through a five region structure as the magni-
tude of reverse bias voltage is increased. When fitting the CV profile of a diode
structure with several regions, there will be a set of voltages where the diode
behaves like a two region structure, and a further set of voltages where it be-
haves like a three region structure and so on. It is necessary to know what the
terminal voltage will be at each transition such that the correct set of equations
can be used to find the relevant depletion distances. The transition voltages
are obtained by considering the area under the electric field curve and the one
dimensional Poisson equation.
227
APPENDIX A
N1 N2 N3 N4 N5
P-Type
Capping Layer
N-Type
Multiplication
Layer
N-Type
Charge Layer
N-Type
Absorption Layer
N-Type
Capping Layer
ξ
x
ξ
x
ξ
x
ξ
x
E1
E1
E2
E1
E2
E3
E1
E2
E3
E4
A1A2
X1
X2
X1
X2
X3
X1
X2
X3
X4X1
X2
X3
X4
X5
Figure A.1 – Top: Schematic diagram of a reverse biased SAM-APD. Upper Middle: Twopartially depleted regions (p+–n). Middle: One fully depleted region and two partially depletedregions (p+–n−–n+). Lower Middle: Two fully depleted regions and two partially depletedregions (p+–n−–n+–n−). Bottom: Three fully depleted regions and two partially depletedregions (p+–n−–n+–n−–n+).
228
APPENDIX A
A.1 Two Regions
A1 =1
2X2E1 (A.1)
A2 =1
2X1E1 (A.2)
E1 = −qN2 X2
ε=
qN1 X1
ε(A.3)
by appropriate substitutions it may be shown that,
A2 =1
2
qN22X2
2
εN1
(A.4)
A1 = −1
2
X22 qN2
ε(A.5)
The terminal voltage is then,
Vt = −1
2
qN2X22
ε+
1
2
qN22X2
2
εN1
(A.6)
Setting the terminal voltage to zero and solving for X2,
X2 =
(−2N1
2Vt εN2 q + 2N1 Vt εN22 q
) 12
N2 q (−N1 + N2 )(A.7)
X1 is then,
X1 = −N2 X2
N1
(A.8)
229
APPENDIX A
A.2 Three Regions
By The same approach as for two regions, expressions for E1, E2, X1 and X3 may
be written. X2 is known (from for example SIMS or the growth order sheet)
because the multiplication region is fully depleted.
E2 = −qN3 X3
ε(A.9)
E1 = E2 − qN2 X2
ε(A.10)
E1 =qN1 X1
ε(A.11)
Solving for X1,
X1 = −N3 X3 + N2 X2
N1
(A.12)
The terminal voltage is,
Vt =1
2E1 X1 +
1
2E2 X3 +
1
2(E1 + E2 )X2 (A.13)
Setting (A.13) to zero and solving for X3 yields,
X3 =1
2
−b+√b2 − 4 a c
a(A.14)
a = −N3 +N3
2
N1
(A.15)
b =
(
−2X2 + 2N2 X2
N1
)
N3 (A.16)
c =N2
2X22
N1
− N2 X22 − 2
εVt
q(A.17)
230
APPENDIX A
A.3 Four Regions
Similarly for four regions, expressions for E1, E2, E3, X1 and X4 may be written.
X2 and X3 are known because the multiplication and charge regions are fully
depleted.
E1 =qN1 X1
ε(A.18)
E3 = −qN4 X4
ε(A.19)
E2 = E3 − qN3 X3
ε(A.20)
E1 = E2 − qN2 X2
ε(A.21)
(A.22)
Solving for X4,
X1 = −N4 X4 + N3 X3 + N2 X2
N1
(A.23)
The terminal voltage is,
Vt =1
2X1 E1 + X2
(1
2E1 +
1
2E2
)
+ X3
(1
2E2 +
1
2E3
)
+1
2X4 E3 (A.24)
Setting (A.24) to zero and solving for X4 yields,
X4 =1
2
−b+√b2 − 4 a c
a(A.25)
a = −N4 +N4
2
N1
(A.26)
b = 2N4 N3 X3
N1
+ 2N4 N2 X2
N1
− 2X3 N4 − 2X2 N4 (A.27)
231
APPENDIX A
and
c = 2N3 X3 N2 X2
N1
+N2
2X22
N1
− N3 X32 +
N32X3
2
N1
− N2 X22
− 2X2 N3 X3 − 2εVt
q(A.28)
A.4 Five Regions
Similarly for five regions, expressions for E1, E2, E3, E4, X1 and X4 may be writ-
ten. X2, X3 and X4 are known because the multiplication, charge and absorption
regions are fully depleted.
E4 = −qN5 X5
ε(A.29)
E3 = E4 − qN4 X4
ε(A.30)
E2 = E3 − qN3 X3
ε(A.31)
E1 = E2 − qN2 X2
ε(A.32)
E1 =qN1 X1
ε(A.33)
Solving for X5,
X1 = −N5 X5 + N4 X4 + N3 X3 + N2 X2
N1
(A.34)
The terminal voltage is,
Vt =1
2X1 E1 + X2
(1
2E1 +
1
2E2
)
+ X3
(1
2E2 +
1
2E3
)
+
X4
(1
2E3 +
1
2E4
)
+1
2X5 E4 (A.35)
232
APPENDIX A
By the same process as in the prior problems, setting to zero and solving for X5,
X4 =1
2
−b+√b2 − 4 a c
a(A.36)
a =N5 2
N1− N5 (A.37)
b = 2
(
−X2 − X3 − X4 +N4 X4 + N2 X2 + N3 X3
N1
)
N5 (A.38)
and,
c = −2X3 N4 X4−N3 X32−N2 X2
2−N4 X42−2X2 N4 X4−2X2 N3 X3−2
εVt
q+
N42X4
2 + 2N4 X4 N3 X3 + 2N4 X4 N2 X2 + N32X3
2 + 2N3 X3 N2 X2 + N22X2
2
N1
(A.39)
233
APPENDIX A
234
Appendix B
APD SPICE Model Source Code
This appendix lists the SPICE model for the APD discussed in Chapter 8.
.subckt APD L A C M Cap F
* Light entering the APD. V(L) is External Pin for Light entering the
device Units: Volts represents Watts
R L 0 1G
* Series Resistance
Rs C 100 Rs
* A sensing voltage source to properly execute multiplication on the
* bulk dark current and primary photocurrent
V1 100 Ci 0
* Primary Photocurrent
BPR Ci A i=R*V(L)
* Multiplication value computed using Millers Empirical Approximation
* having regressed some real multiplication data
BM M 0 v=1/(1-(V(Ci,A)/VBD)**nm)
* Capacitance due to physical dimensions. Capacitance is obtained by
* linear regression of 1/C(Bias)^2 data. Comment out this line or the
* next as appropriate.
* BC Cap 0 v=(1/(Cjm*V(Ci,A)+CJc))**(0.5)
* Capacitance due to physical dimensions. Capacitance is obtained by
* inverse second order polynomial regression of [C(Bias) C = C0 +
* a/bias + b/bias^2]
*BC Cap 0 v=(C0 + a/V(100,A) + b/(V(100,A)**2))
235
APPENDIX B
* Capacitance due to physical dimensions. Capacitance is obtained by
* regression of an exponential decay of the form
* y = y0 + a*exp(-b*bias)
BC Cap 0 v=(C0 + a*exp(-b*V(100,A)))
* A Non Linear Capacitor. ‘x’ represents the voltage across the
* non-linear capacitance.
CJ A Ci Q=V(Cap)*x
* Bulk Dark Current (is Multiplied).
Bgen Ci A i=Ibulk
* Surface Dark Current (is Not Multiplied).
Bsuf 100 A i=Isurf
* This line multiplies the Bulk dark current and the primary
* photocurrent.
BI 100 A i=I(V1)*(V(M)-1)
* Tunnelling current. The tunnelling current is modelled by performing
* the linear regression of the common log of some tunnelling current
* data ie log10(Itunnel) = m*Bias + c.
* This is generally a good enough fit for a first order model.
* Tunnelling current is not multiplied.
BT 100 A i=10**(itm*V(Ci,A) + itc)
* Excess Noise Factor Using the Local Expression for F(M) where pure
* carrier injection prevails.
* The modeling of mixed injection is possible by adjusting k and
* providing new coefficients for Millers approximation of M
BF F 0 v=k*V(M) + (2-(1/V(M)))*(1-k)
* two times the total current that will be multiplied by the shot
* formula. i.e. N^2 = 2*q*BTI
BTI 201 200 i=2*(I(BI)*V(M)*V(F)+Isurf+I(BT))
* Two Common Base transistors share the photocurrent and each generate
* 1/2 of the required noise
Q1 0 202 200 pnp
Q2 0 202 201 npn
.model npn npn (IS=0.01f BF=1 cje=1p)
.model pnp pnp (IS=0.01f BF=1 cje=1p)
* A voltage source senses the noise current flowing from Base to Ground
236
APPENDIX B
V2 202 0 0
* A current source in the main diode circuit develops the appropriate
* noise current
BN 100 A i=I(V2)
* A diode is used to model forward Bias
D1 A 100 diode
.model diode D(is=0.1f rs=1)
.ends
237
APPENDIX B
238
Appendix C
Phase-Sensitive Detection
Phase sensitive or “Lock-in” detection is a method increasing the signal to noise ratio
of an electronic system by decreasing the bandwidth. A complimentary viewpoint is
that the time taken to make an observation using the system is increased in order
to gather more information. The method requires the use of a linear phase detector,
an analogue circuit function which, more recently, has been implemented using DSP
techniques, followed by a filter. A simplified schematic of a linear phase detector and
RC filter is shown in Fig. C.1. The linear phase detector has two inputs labelled ‘input
signal’ and ‘reference signal’ the output of the linear phase detector is connected to R3
which forms part of an RC filter.
The ‘input signal’ is an appropriately scaled voltage taken from an experimental
setup. The signal is often heavily corrupted by noise. The reference signal is a periodic
signal which has its phase information imparted to the input signal. In the case of
the experimental work described in this thesis a slotted disk is placed in the path of
the light source and rotated at a known frequency (the frequency is measured by a
LED/photo-diode arrangement on the chopper assembly). The APD biasing voltage
could be modulated and a CW light source used instead. This would be advantageous
because mechanical choppers are affected by ‘jitter’ which is caused by mechanical
deficiencies.
The linear phase detector operates by inverting the gain of the amplifier on the
transitions of the reference signal. The first order filter represents any arbitrarily com-
plex form of signal averaging. The time constant of R3 and C1 must be much longer
than the period of the reference signal.
239
APPENDIX C
−
+
R1
R2
R3
C1
Phase DetectorOutput
ReferenceSignal
Input
Signal
Figure C.1 – Simplified Schematic of a linear phase detector.
For the purposes of simplified analysis assume the input signal is of the form,
A cos (ω t+ φ) (C.1)
and the reference signal is a square and changes state at 0, πω ,
2πω etc. The period of
the signal is T = 2πω . The output voltage is then,
〈A cos (ω t+ φ)〉|π/ω0 − 〈A cos (ω t+ φ)〉|2π/ωπ/ω (C.2)
Integrating to perform the averaging,
〈Vout〉 =ω
2π
∫ π/ω
0A cos (ω t+ φ) dt.− ω
2π
∫ 2π/ω
π/ωA cos (ω t+ φ) dt. (C.3)
which yields a DC signal proportional to φ and A,
Vout = −A sin (φ)
π− A sin (φ)
π(C.4)
Vout = −2A sin (φ)
π(C.5)
A corrupting signal component may be considered by adding another frequency to the
input signal which is not equal to the reference frequency,
cos (ω t+∆ω t) (C.6)
The phase shift of the original signal, φ, is now varying at the difference in frequency of
the corrupting signal and the input signal. The resulting output is a sinusoidal signal
240
APPENDIX C
Figure C.2 – Black: In-phase input signal, Red: Corrupted input signal, Blue: Output signalresulting from the corrupted input signal.
with a period equal to the difference in frequency,
Vout = −2A sin (∆ω t)
π(C.7)
An example is shown in Fig C.2 in which the nominal angular frequency is 2π rads−1,
the corrupting signal is 2.2π rads−1. The output signal is therefore 0.2 rads−1.
241
APPENDIX C
242
Appendix D
High Order Nature of the
Common Base Amplifier
This appendix gives a brief analysis which shows that a hybrid-π transistor model which
includes the frequency dependence of β and the APD junction capacitance (Fig. D.1)
is second order, and that the second order behaviour becomes apparent when rb ≈ rbe
Summing currents at the base node (b)
− vb
rb− vb − ve
rbe− (vb − vc) · s · ccb = 0 (D.1)
summing currents at the collector (c)
− vc
RL− gm · (vb − ve) + (vb − vc) · s · ccb = 0 (D.2)
rb
b
cbe rbe
iin
e
cj
gmvbe
c
RL
Figure D.1 – Common base amplifier hybrid-π small signal model
243
APPENDIX D
and summing currents at the emitter
− iin + gm · (vb − ve) +vb − ve
rbe− ve · s · cj = 0 (D.3)
To include the base emitter junction capacitance (cbe) let the rbe term in Eq. D.1
become
rbe =r′
be
1 + s · cbe · r′
be
(D.4)
The combination of the following stage input capacitance and the parasitic capacitance
associated with the load resistor can be included by letting the RL term in Eq. D.2
become
RL =R
′
L
1 + s · CL ·R′
L
(D.5)
where CL is the combined capacitance. To include the emitter biasing resistor RE , let
Eq. D.3 become
− iin + gm · (vb − ve) +vb − ve
rbe− ve · s · cj −
ve
RE= 0 (D.6)
Solving the node equations Eq. D.1–D.3 without CL, cbe or RE yields a second order
response of the formvc
iin= k · (1 + s · τ1)
( s2
ω20+ s
ω0·q+ 1)
(D.7)
where
k = −gm · rbe ·RL
gm · rbe + 1(D.8)
τ1 =ccb · (rb · gm · rbe + rb)
gm · rbe(D.9)
1
ω20
=((rb + gm · rb · RL +RL) · rbe +RL · rb) · cj · ccb
gm · rbe + 1(D.10)
1
ω0 · q=
(RL · gm · rbe +RL + gm · rb · rbe + rb) · ccbgm · rbe + 1
+(rbe + rb) · cjgm · rbe + 1
(D.11)
244