вход по аккаунту


High performance guard ring P-i-N photodetectors for microwave subcarrier wavelength division multiplexed networks

код для вставкиСкачать
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter free; while others may be
from any type o f computer printer.
The quality of this reproduction is dependent upon th e qu ality of the
copy subm itted. Broken o r indistinct print, colored o r poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UM I a complete
manuscript and there are missing pages, these will be noted.
Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and
continuing from left to right in equal sections with small overlaps. Each
original is also photographed in one exposure and is included in reduced
form at the back o f the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6” x 9” black and white
photographic prints are available for any photographs o r illustrations
appearing in this copy for an additional charge. Contact U M I directly to
A Bell & Howell Information Company
300 North Zeeb Road, Ann Aibor MI 48106-1346 USA
313/761-4700 800/321-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
High Performance Guard Ring P-i-N Photodetectors for Microwave
Subcarrier Wavelength Division Multiplexed Networks
Babatunde Odubanjo
Summitted in partial fulfillment of the
requirements for the degree
of Doctor of Philosophy
in the Graduate School of Arts and Sciences
Columbia University
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 9809756
UMI Microform 9809756
Copyright 1997, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
300 North Zeeb Road
Ann Arbor, MI 48103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
High Performance Guard Ring P-i-N Photodetectors for Microwave
Subcarrier Wavelength Division Multiplexed Networks
Babatunde Odubanjo
This dissertation is concerned with the design and fabrication o f high performance
guard ring photodetectors suitable for use in very dense WDM microwave subcarrier
multiplexed (SCM) networks and Passive Optical Networks (PONs). Two issues are
addressed. The first obvious quantity o f concern is the problem o f nonlinearity distortion
characteristics o f the detector arising from - internal electric field perturbation due to
space charge and external loading induced effects - intermodulation effects due to the
mixing between different microwave subcarriers - and optical beat notes interference
issue. The second is the problem o f slower responses due to diffusion currents.
Our investigation begins with an analytic study o f the transient response o f a typical p i-n and avalanche photodetectors (APDs) to determine the origin o f their nonlinearities
and slow response and their suitability for use in a narrow-band receiver required for
SCM networks. We found that for a single channel link where the total receiver
photocurrent is o f the same magnitude as the signal photocuirent, avalanche
photodetectors can dramatically increase receiver sensitivity. However in microwave
subcarrier multiplexed networks the photodiodes in each receiver detects all subcarrier
channels therefore the total received photocurrent is greater than that of a single channel
by many orders. Thus under these criteria, APDs cannot provide any improvement in the
receiver sensitivity. Infact avalanche gain noise mechanism degrades the expected S/N
permission of the copyright owner. Further reproduction prohibited without permission.
ratio due to its intrinsic excess noise factor. Furthermore, APDs typically require large
bias voltage. There is also the problem o f limited gain-bandwidth product which makes
avalanche photodetectors unsuitable for very dense WDM networks. The clear choice for
SCM-based networks is therefore the p -i-n photodiode which is shot-noise lim ited in high
speed applications.
We use our analysis to optimize a typical p -i-n photodiode. The result is a novel guard
ring photodiode that incorporates an additional diffused p* region around the active layer
(and active coupling o f the detector while monitoring the slow tail and the harmonics) to
circumvent the problems o f nonlinearity distortions and slower responses due to diffusion
currents. The optimal device involves a tradeoff o f many crucial parameters such as the
quantum efficiency and speed.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Table of Contents
Chapter 1 Introduction............................................................................ l
1.1 Principles o f Subcarrier WDM networks.......................................................................3
1.2 Nonlinear Mechanisms in Photodetectors......................................................................7
1.3 Linear Photodetectors .................................................................................................16
1.4 Outline o f the Dissertation .......................................................................................... 19
Chapter 2 Field Profiles of Electrically Isolated Guard Ring
2.1 Introduction ................................................................................................................ 22
2.2 Formulation o f the Discrete Poisson E quation.......................................................... 24
2.3 Results and Discussion................................................................................................ 29
2.4 Conclusion ..................................................................................................................35
Chapter 3 Separate Absorption and Multiplication APDs
.................... 36
3.1 Introduction.................................................................................................................36
3.2 Two-dimensional design and analysis........................................................................40
3.3 Device Fabrication...................................................................................................... 43
3.4 Device structure and architecture ............................................................................46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.5 Conclusion
Chapter 4 Experimental Study of Quarternary Weil Superlattice
SAM-Avalanche Photodetectors........................................................ 50
4.1 Introduction...............................................................................................................50
4.2 Device Structure........................................................................................................51
4.3 Energy Band Calculation..........................................................................................56
4.4 Conclusion................................................................................................................ 59
Chapter S High Performance Guard Ring P-i-N Photodetector for
Digital Fiber Optics Communications................................................ 60
5.1 Introduction...............................................................................................................60
5.2 Analysis.....................................................................................................................63
5.3 Results and Discussion..............................................................................................67
5.4 Conclusion.................................................................................................................75
Chapter 6 Future Work - Mid Inlira Red (MIR) Photodetectors...... 76
i i
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List Illustrations
1-1 A microwave subcarrier multiplexed WDM network................................................. 5
2-1 An illustration o f the stimulated segment.................................................................. 25
2-2 Potential (a) and E-field (b) profiles o f a planar device without guard rings........... 31
2-3 Potential (a) and E-field (b) profiles o f a planar device with a pair o f guard ring...32
2-4 Device structures (a) unguarded (b) with one guard ring......................................... 33
3-1 A reverse p-n junction for better electric confinement............................................. 38
3-2 An avalanche photodiode wafer spread.................................................................... 40
3-3 Scans across the active area o f the device at biases near V b d ..................................46
4-1 Epitaxial wafer structure #1.......................................................................................54
4-2 Epitaxial wafer structure #11...................................................................................... 55
4-3 Energy band diagram
5-1 Schematic o f a typical p-i-n photodiode under illumination....................................62
5-2 Cross-section o f the (75pm) guard ring p-i-n photodiode....................................... 66
5-3 Typical response o f a guard ring photodiode to a 20ns pulse (a) and slowtail....... 70
5-4 Optical response (a) unguarded (b) guarded
5-5 Capacitance vs reverse voltage and bandwidth (a) unguarded (b) guarded............ 73
6-1 Schematic o f a InGaAsSb/GaSb-based mesa back-illuminated photodiode............78
6-2 A blind-diode dark current -voltage characteristics................................................. 79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
The exponential growth o f global communications has placed strenuous demand on
telecommunications and computer communication infrastructures. There is currently a
concerted effort underway to determine if Wavelength Division Multiplexed (WDM)
lightwave networks can meet the expected demand. WDM allows multiple users to share
a fiber optic network in a manner similar to radio communications except that the carrier
frequencies are now at lOOTHz. WDM exploits the abundance o f optical bandwidth
which is o f the order o f 35THz in the low-Ioss zero-dispersion region between 1.2 and
1.6pm. By using wavelength to perform such system and network tasks such as
switching, routing, diagnostics and control, it is possible to take advantage o f the
enormous bandwidth. To fully utilize such a bandwidth would require individual optical
pulse widths o f few tens of femtoseconds. Technologies do exist for generating such
ultra-short pulses in research laboratories but systems employing them are not yet
practical. In practice the usable bandwidth in a fiber link or network is much less than the
stated 35THz and its use is determined by complex interplay between the components
and devices and the relevant network architecture.
The transmission rate at which a single channel can be operated is constrained by the
speed at which the electronics components can be driven. This naturally suggests the use
o f a multi-access shared-wavelength scheme. For growth requirements the network
should be modular and scalable to counter capacity saturation. This is the principle at the
heart o f the WDM architecture. While the question o f a scalability is a complex subject, it
can be simply described as the ultimate constraints imposed on the network throughput
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
due to namely - wavelength selective devices. O f paramount importance to the
achievement o f scalability are the notions o f wavelength re-use and wavelength
translation. With this approach, each constellation o f users is attached to the network
through a generic access station equipped with a battery o f microwave transmitters and
receivers. The minimum spacing is limited by crosstalk mechanisms such as optical and
electrical nonlinearities. Optical nonlinearities tend to become crucial as wavelength
spacing becomes very critical. The photodiode detects the signal and noise as both
additive and as nonlinear mixing o f the amplifier and the desired signal noise in a squareIaw process. Degradation o f the photodetector’s signal-to-noise ratio (SNR) will
inevitably lead to channel outages. This suggests time for network upgrading and the
combination of WDM and microwave SubCarrier Multiplexing (SCM) appear to be
promising. Microwave subcarrier multiplexing is a scheme for coupling multiple
microwave channels onto a angle optical carrier. The term subcarrier multiplexing is
used to distinguish this technique which uses microwave subcarriers and optical carriers
from traditional frequency division multiplexing. In lightwave netw ork the to m FDM is
restricted to direct modulation o f optical carrier with data. The SCM-WDM scheme is
analog by nature and uses m ature microwave technology to partition the modulation
bandwidth o f semiconductor lasers to carry multiple users. The microwave multiplexing
is done in the frequency domain. The photodetector senses all the subcarriers over the
entire usable bandwidth but only the desired narrow-band channel has to be demodulated
using conventional microwave technique. The process o f determining the most suitable
photodetector for the task o f extracting intelligible signals from the SCM-WDM network
is the core subject o f this thesis.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Efforts are underway to standardize optimized photodetectors for specific traffic
patterns and network requirements in the optoelectronics industry. For example among
other requirements, photodetectors for microwave signal reception under very high speed
modulation need a small active region. Under this condition we can expect significant
perturbation o f the device electric field due to large optical power density. The distortion
o f the electric field which governs the drift velocity may now lead to nonlinear response
in photodetectors. Another objective is the achievement o f large dynamic range which
can limit intermodulation distortions in narrow-band channels that are compatible with
SCM-WDM lightwave networks and explore in detail their photodetection specification
requirements at the device level. In particular this chapter will deal with design issues
concerning the realization o f practical and reliable photodetectors for this type o f
networks. We will investigate the sources o f nonlinearities in photodetectors and how to
control and manage them and we conclude with an outline o f the dissertation.
1.1 Principles of Subcarrier WDM Networks
The conventional concept o f WDM networks requires packing one to eight channels in
each of the usable transmission wavelength windows which are then combined and
broadcast via a star coupler. The problem with this approach is that the architecture
requires fast tunable optical filters and lasers. The tuning time o f the state o f the art
research grade WDM network is about 25milliseconds between wavelengths. The tuning
time of commercially available optical network components is too long for efficient
traffic transmission. There are two commonly used physical topologies - broadcast-andselect and wavelength routing networks. In broadcast-and-select type of network, traffic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
transmissions are received by all the outputs. In a particular design, the input sources
have to be tunable and all the output receivers are tuned to fixed wavelengths. This is
primarily a space-division switch. The m ajor disadvantage o f this switch is that it exhibits
contention problems. Specifically, output port contention exists in this type o f network.
Therefore a contention-resolution scheme has to be provided. This architecture uses
wavelength addressing o f the output ports, but it has the capability o f selecting only a
single wavelength at each output. Thus only point-to-point connections are possible and
multicast connections are virtually impossible. It is however possible to circumvent this
restriction by using an acousto-optic filter [Kaminow I.P, 1988; Cheung K., et al, 1989]
to simultaneously select more than one wavelength. In addition the broadcast-and-select
architecture can be made to support multicast connections by providing tunable receivers
at the output and fixing the input lasers at unique wavelengths. Output contention also
exists in this format and can be made worse by the multicasting. The performance o f such
a network can be severely limited by means o f contention resolution. Several proposed
architectures for this type o f network can be found in [Brackett C.A., 1988]. The second
major architecture is the wavelength routing network. These networks feature one or
more wavelength-selective elements and they have the property that the path signal takes
through the network is uniquely determined by the wavelength o f the signal and the input
port. To illustrate, an N x N network with N tunable sources can be interconnected with
N photodetectors and receiving elements. By tuning to a given selected wavelength, the
signal from a particular laser can now be routed to a unique output port. This network can
be realized by using only 0(N ) wavelengths. ATinputs can be interconnected concurrently
with # outputs.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In all o f the WDM architectures mentioned above we need at least as many
wavelengths as the number o f users - 0 (N ). When the number o f wavelengths ATbecomes
large, inter-channel cross-talk becomes a critical issue. In order to alleviate the tight
channel-spacing specifications required o f these networks, it is apparent that the above
networks may be combined with each other and with more traditional architectures to
generate robust networks.
An approach [Choy M M Tong F.F., and Odubanjo, T.; 1991] proposed expanding
WDM by providing a number o f microwave subcarriers within each wavelength. This
scheme (illustrated in figure 1-1) avoids the need for high-speed receivers and inter­
carrier stabilization. In this scheme a station transmits on a fixed wavelength and a fixed
subcarrier frequency on that wavelength and then uses a tunable filter followed by a
tunable subcarrier
Constellation • 2
Constellation (1
Star Coupler
Figure 1-1 A Microwave SubcarrierMultiplexed-WDM Network
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
receiver to receive any channel. Alternately, each subcarrier can be combined optically.
In either case the multiplexed optical signals are photodetected and the channels are
demultiplexed at microwave frequencies using inexpensive microwave components. The
immediate scheme above applies to the broadcast topology and full connectivity is
usually achieved by using a m ultihop architecture. This will remove the slow-tuning
bottleneck present in the optical domain. This version o f m ultihop uses a broadcast star
physical topology. Each participating station has a fixed tuned set o f transmitters and
fixed tuned receivers. Thus by imposing a carefully selected connectivity pattern on the
broadcast physical topology we can obtain a logical topology that provides stations with
unique transmission paths to all stations in the network. A drawback to the above
approach is that all lasers in the network have to be at distinct wavelengths.
One practical approach to obtaining a 2-dimensional network (electronics and optics) is
to use the broadcast physical topology in a passive star architecture combining subcarrier
and wavelength division multiplexing as a unit. Then a m ultihop (such as the shufflenets
or de Bnrijn graphs [de Bruijn N.G. and van Aardenne-Ehrenfest T., 1951; and Ghartrand
G., and Lesniak L., 1979]) logical topology is superimposed on this physical layout to
realize a practical network. This scheme exploits the fact that the technology required for
rapid microwave tuning is mature and well developed. In fact current state o f the art
microwave voltage-controlled-oscillators can be switched in tens o f nanoseconds. In this
network each user is provided with a fast-tunable VCO - generated subcarrier modulator
and a fixed-tuned transmitter to allow it to send traffics upstream on any o f the available
subcarrier channel on that wavelength. A more desirable implementation is to have many
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
stations o r users grouped together in constellations or clusters to share a single laser (to
prevent optical beat notes or interference). A scheduler in the cluster controller is added
for preventing traffic collision. Since only one laser is allowed per cluster, we should
expect no optical interference problem if lasers wavelengths are well spaced, h i case o f
wavelength drifting however the photodetector at the sink on the downstream should be
designed and optimized to
minimize optical beat notes and the anticipated
intermodulation distortions due to mixing between the different microwave subcarriers.
1.2 Nonlinear Mechanisms in Photodetectors
A complete understanding o f the sources o f nonlinear response in PDs (photodetectors)
is essential for designing an optimum device. Most o f the effort to determine the origin o f
the nonlinear behavior o f lightwave communication systems has focused on sources
[Schaf&er J.H & Bridges W.B 1993; Okuda et al, 1994]. However, photodetectors have
been found to be quite nonlinear [Humphreys D .A , and Lobbett R .A ; 1988]. This
nonlinear characteristics can define the fundamental limiting factor in high performance
and advanced networks such as the one discussed above in section 1.1. The photodetector
is a fundamental component in that it ultimately determines the fidelity o f the network
(i.e the SNR). Light incident on a photodetector consists o f photons that arrive randomly
and may be described by a Poisson process with the rate parameter being the
instantaneous optical power. Hence, the generation o f electron-hole pairs in the detector
is also a Poisson process, and the photocunent is a shot noise process. The resulting
error-rate performance, in the absence o f other noise sources, yields the quantum limit.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The origin o f nonlinear properties o f PDs can be traced to electric field dependence o f
diffusion constants, trap centers, highly-doped undepleted
absorbing regions,
recombination, scattering and non-short-circuit loading. Indeed a large optical power
absorption in a relatively small device can perturb by space charge effects o f the ambient
electric field which governs carriers transport and thus induce a nonlinear electrical
response. The photodetector can also be depolarized as a result o f external current flow in
the load circuit and current crowding (lateral shunting o f photocurrent to the contact).
This can lead to perturbation o f the internal electric field. These types o f nonlinearities
induce harmonics. At higher operating frequencies, the field dependence o f the carrier
velocity in photodetectors generates harmonics and intermodulation products that can
degrade the dynamic range o f microwave-fiber optic links. The other major concern
source o f nonlinearity in PDs is the problem o f optical mixing. The SNR o f microwave
SCM-WDM networks has been analyzed [Darcie T.E., 1987] where the maximum
number o f channels is evaluated for a given transmission bandwidth. However when
multiple sources are present the mixing o f the optical fields must be considered. This
mixing takes place at the shared photodetector giving rise to optical interference terms in
form o f beat notes [Desem C., 1990 and 1988]. These beat notes can effectively limit the
number o f optical carriers that can be carried by each optical source or can ultimately
result in microwave channel outages in the downstream direction. In this section, the
origin o f the principal sources o f nonlinearity will be briefly reviewed and solutions will
be presented.
We begin our investigation into the origin o f nonlinearities in PDs by examining a
typical p-i-n photodetector. Carrier transport (and dynamics) is analytically described by
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
three coupled nonlinear differential equations, Poisson’s equation and the electron and
hole continuity equations. These equations are solved to determine and identify the major
sources o f nonlinear mechanisms in the PD. The p-i-n under examination is a single
heterojunction device and is illustrated below in Fig. 5-1. This device consists o f diffused
p + region in an InP cap followed by an unintentionally doped ri~ InGaAs and a highly
doped »+ substrate. The pertinent basic equations are given [A. S. Grove, 1967; and S.
M. Sze, 1969]
3p/3t = G - R - vp(5p/3x) - p(5vp/dx) - l/q(dJpW 3 x ),............................................. (1.1)
dn/dt = G - R + v„(dn/dx) + nKjdvJdx) + l/q(0J„<fifift5x)...........................................
The continuity equations above are linear only strictly if v„ and vp are completely
independent o f the carrier densities. However the carrier velocities are in fact related to
the electric field [Windhom T IL , et al; 1982] via space charge effects as demonstrated
by the empirical equations below [M. Dentan and B. de Cremoux; 1990]
vn(E) = (EOlo + VouP|E|)y(l + PE3) ............................................................................ 0 -3 )
vp(E) = ftlpVpwEyCv,' + (Ip’E O '^yO + PE) ...............................................................(1.4)
The electric field in equations 1.3 & 1.4 above can be further broken down into the sum
o f - equilibrium electric field and space charge field electric field as a result o f excess
carriers from light illumination, as follows
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
V*Edark = qfe(Po-na+ N j - N J
( 1.6)
q /s ( p r -n ' )
p0 and no are the equilibrium carrier concentration and p ' and n ' are the excess
photogenerated carriers, h i the high injection regime, the diffusion constants and the
carrier velocities are both dependent on carrier densities. As a result the space-charge
fields may produce changes in the electron and hole velocities in addition to its effect on
diffusion constants. This field may also modify the scattering properties o f the carriers.
There is also carrier bleaching problem, hi all, the combination o f these problems may
lead to potential drop within the photodetector as result o f partial internal electric field
collapse and flow o f current in the external circuit. Thus in the high-injection regime
photocarriers may perturb space-charge field and induce carrier velocity variations
resulting in nonlinear behaviors in photodetectors. Other nonlinear mechanisms o f
significance are due to dynamic o r transit time effects and nonlinearities associated
absorption in undepleted regions within the device. O ur investigation revealed that
nonlinearities as a result o f transit tim e effects are frequency dependent. This frequency
dependence has been studied by (Hayes R.R. and Persechini D.L., 1993]. Their studies
suggest that the transit-time effect frequency dependence may be related to the change in
transit time imposed by different carrier velocities compared to the period o f the signal.
As for nonlinear effects associated w ith absorptions outside the active region, our studies
show this to be linked to the lifetime o f the photo-generated carriers. That is photo­
generated carriers which live long enough to diffuse into the active region. Therefore this
nonlinearity is a function o f low-field carrier mobilities [Wake D., et al 1993] device
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
physical parameters, doping level in the region and the diffusion constant o r carrier
lifetime. The role high optical power density plays in generating nonlinear distortions in
PDs has not been well documented in literature. However our investigations suggest that
at high optical power densities nonlinear distortions can degrade the performance o f PDs.
For example a reduction in the diode responsivity and significant phase shift can occur.
Under high optical power illumination the PD photocurrent increases, the electric field
within the space charge region starts to collapse, the carriers slow down, the photocarrier
collection time increases and subsequently the output photocurrent is delayed. The field
profile will then recover at the end o f the cycle. The delay in carrier collection is a
confirmation o f the transient charge storage effect. And the onset o f the electric field
collapse marks the threshold o f high optical power induced nonlinear distortion. The
primary mechanism responsible for this field collapse may be related to current
crowding. In the high injection regime large amount o f photogenerated currents will be
shunted laterally through the typically thin p-layer o f a p -i-n detector. Since the p-region
must be thin for efficient transport the sheet resistance is high and the bias potential
across the junction is dependent on the distance from the p-layer contact, causing most o f
the current to flow under or near the contact. The voltage and current profiles are
therefore not uniform leading to field collapse in the active region and markedly
nonlinear response. Finally another strong source o f nonlinearity in PDs is band
discontinuities at the heterojunction interfaces. Band discontinuities may lead to the
creation o f quasi-stable mini-states in which carriers can be trapped. Heterojunctions in
PDs can cause carrier trappings in these mini-energy states at conduction and valence
band discontinuities. Hole trapping problem is significant in InP/InGaAs because o f the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
large valence band discontinuity. The total emission rate is the sum o f the two emission
rates for holes and electrons. If the interfacial deep-level recombination rate is significant,
the lifetimes o f electrons and holes are now different because o f the traps (
response o f the trapped carriers can be obtained by convolving an intrinsic current source
with the emission function. The result is that the carrier-trapped current response is
primarily a nonlinear Arrhenius function type.
We now examine the source o f nonlinearity arising from optical beat notes. Our
approach is to treat the two interfering sources as nonstationary and mutually incoherent.
We can express the electric fields in term s o f the analytic signals and complex envelopes
E i( t) - R e ( i m ) = R e { 6 t (t)ep * v‘‘>
( 1.8)
The optical intensity incident on the photodetector is given by
E i(t)
E2 (t)
2El (0E2(0
The photocurrent at the receiver is obtained by low-pass filtering the mean square total
electric field down to the detector bandwidth [Davenport W.B., and Root W.L.; 1958]:
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
where ho(t) is the impulse response o f the detector and i*j(t) can be broken down into its
constituent components as follows
id fl) = idlQ )
id tf )
id x fi)
................................................................................................................................................................................................................................................. (
1 11
In addition to the heterodyne cross-term i ^ t ) above, the total photocurrent includes the
direct detection terms
*dl,d2(t) = 91/2 1E i 2(t) 12 ............................................................................................ ( 1. 12)
The term due to optical mixing is
idcft) - 91/2 Re (E i(t)E 2* (t)} ................................................................................. (1.13)
The composite autocorrelation term is now
Rdfrd) = ridl(r) +
+ 2 ( id l) ( id2> + fid xt* )...........................................(1-14)
Taking the Fourier Transform o f this yields the power spectral density o f the total
intensity by the Wiener-Khintchine theorem [Papoulis A., 1984 and Rice S.O., 1944]
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
= S n (f) + S j2 0 + 2 (h ) d 2 m + 4Se i (0 &SE20)
angular brackets denotes ensemble averaging over Langevin noise sources.
Therefore the power spectrum o f the photocurrent is
Sid CO=
Sjd(f) = SfdiCO + Sid20) + 2 (id i) ( id2>S(J) + { 4Se i (D ® S e 20)}D
From the microscopic view, our analysis o f the photodetector’s response assumes that its
output electrical photocurrent ij[t) depends only upon the instantaneous number of
photogenerated carriers n(t). For small departure from linearity the response can be
expanded in Taylor series, retaining term s only to second order. The illuminating pulses
are assumed to be from two stationary uncorrelated optical sources which generate carrier
densities n j and «2- The Taylor expansion is given as
I(n) = n(di/dn) + (n /2) (d*i/dn) = i(nj
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The last term is the nonlinear component which depends on the interaction o f the carrier’s
generated by the two optical sources. This is detected by the photodetector. If the pulses
which generate « / and n 2 are separated by a time t but are otherwise identical and if each
pulse generates no carriers, the detected autocorrelation signal 5Q (r) is
SQ ( t ) = ( i l / d t n )
n0 (t) n0 ( t + r ) d t ........................................................(1.20)
where ijft) = I(t) ®hD(t)
Thus we have shown that the power spectral density o f the photodetector photocurrent is
also carrier-dependent. The nonlinearity will be severe in the high injection regime.
Under this condition, the photogenerated carrier concentration is very high and not all o f
them will be collected in the external circuit. A higher proportion will recombine at a rate
not proportional to the carrier concentration leading to some degree o f carrier bleaching
(due to a decrease of the absorption coefficient) and therefore nonlinear distortion sets in.
Carrier bleaching may result from near population inversion due to the excess
photogenerated carriers thus making the active layer transparent. Whereas in the low
injection regime the density o f the photo-electron hole pairs is relatively small (although
some will also recombine) but the recombination rate will be roughly proportional to the
photogenerated electron-hole pairs. Therefore the photoresponse is linear in this regime.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.3 Linear Photodetectors
In the preceding sections we have surveyed the m ajor sources o f nonlinearity in a
typical p-i-n photodetector. In this section, the characteristics and the features o f a linear
photodetector is presented. We will focus on the physics and technology o f high-speed
linear photodetectors. hi section 1.2 we found that saturated carrier velocities are only
one o f the many assumptions that are required to linearize the transport equations. There
are many other nonlinear transport equation terms, including electric field dependence o f
diffusion constants, highly-doped undepleted absorbing regions, heterojunction traps,
nonzero load resistance, recombination, current crowding, space-charge induced field
perturbation, carrier bleaching and scattering.
The phenomenological argument may be used to linearize the transport equations. For
simplicity we may assume that the field in the intrinsic region o f the reverse-biased p-i-n
junction is Eo, due solely to the external bias. Therefore the diode has basically the
characteristics o f a parallel plate capacitor with plate separation equivalent to the
thickness o f the intrinsic region L. The charge stored on one plate o f the capacitor is
given as follows
Q = eoeE0 ........................................................................................(1.21)
Assuming an illumination intensity o f Ip photons/sec.m2 be incident on the p-i-n, and let a
fraction P of this light be absorbed in the intrinsic region o f the diode. If the absorption
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
constant a for the incident light is o f the order o f l/L , and the light is uniformly absorbed,
where each photon creates an electron-hole pair. A uniform volume pair generation rate *
go may therefore be defined by the equation
( 1.22)
In the special case o f equal mobilities for electrons and holes, we may assume an average
drift velocity, v = (jE q. The photogenerated carrier can transverse the intrinsic region
within a time L/2v. Therefore
L/2/jE q £ L/2vsaturated
In order for the perturbation on the applied field be small, it is required that the charge in
1 .2 1
above be much greater than the number o f electron-hole pairs generated in
the junction at equilibrium. This requirement precludes any recombination in the intrinsic
region. Therefore the recombination current terms (p-po)/xb and (n-no)/rb
in the
continuity equations 1.1 & 1.2 o f page 9 can be neglected if the average time which a
photogenerated carrier spends in the intrinsic region is less than the bulk lifetime. For
high-resistivity intrinsic semiconductor region are expected to be swept out o f this region
before any appreciative recombination can take place. For a time-independent generation
rate, the current density in the intrinsic region is given as
J = Jp + Jn = qp(p+ n)E - qD V (p -n )
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Assuming no recombination in the intrinsic region then the current density must be equal
to the light current density, Lqga (the product o f the thickness o f the /-region, the electron
unit charge q and the generation rate g0). If the magnitude o f the diffusion current is
small compared to that o f the total current density then w e may neglect the diffusion
contribution if the applied voltage is much greater than kT/q. Therefore the above
equation may be rewritten as
J = LS<fl = q tfp + n )E ...................................................................................... (1.25)
p(x) = (g/pEo) •( x+L/2 ) ...................................................................................(1.26a)
n(x) = -(gZ fiE oW x-L /2).................................................................................. (1.26b)
For this linear variation o f carrier concentrations, it is obvious that the diffusion terms in
the continuity equations, will vanish. The nonlinear terms in the continuity equations may
also be neglected, if the condition below is satisfied
yqp(p-n)/eoepE (dp/dx) = (L /psoe)»(Ji/E o ) « 1 ....................................... (1.27)
The condition that this ratio be negligibly small compared to 1 is equivalent to requiring
that there is no recombination o f the photogenerated electron-hole pairs (EHPs) in the
intrinsic region and therefore no significant perturbation o f the internal electric field
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
profile. W e can now rewrite the linearized continuity equation as follows (in the iregion):
ty /d t =g0 -vp cp/dc..............................................................................(1.28a)
= go ~vn
............................................................................... (1.28b)
hi linear photodetectors the position o f th e pn-jvnction is particularly critical because
the intrinsic InG aAs layer should be completely depleted at the operating voltage.
Otherwise the photocurrent will contain diffusion currents from the non-depleted portion
o f the n InGaAs absorption layer which deteriorate the pulse response o f the
photodetectors. The p -i-n should be optimized such that the neutral n-region thickness is
made small compared to 1/a (where a is the absorption coefficient (Humphreys, D.A., et
al; 1985]) so that very little portion o f the incident light is absorbed in that region. The iregion thickness is made o f the order o f 1 /a so that most o f the absorption will take place
in this region. Indeed to avoid slow tail diffusion current, the i-region doping level should
be reduced. When such a diode is reverse-biased, the transport o f carriers through the iregion is by drift in the electric field established by the junction bias, and can therefore be
more rapid than diffusion transport. The transit time in the p-i-n photodiode has been
estimated as being o f the order of the thickness o f the i-region divided by the drift
velocity o f the slower moving carrier-type assuming double-carrier-injection.
1.4 Outline of the dissertation
High-speed detectors are in demand for telecommunications and
communications systems for high capacity lightwave networks. A hybrid o f different
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
detector structures and materials are required to cover the range o f applications.
Germanium photodetectors have been routinely used at near infra-red (NIR) windows
covering the telecommunications usage. Germanium however has a relatively small
bandwidth-efficiency product Therefore for high-speed applications direct bandgap
semiconductors such as III-V material are more important. These compound
semiconductor material structures are the main focus o f our work. For example
foo.s3 Gao.4 7 As with a cutoff wavelength o f about 1.7pm is especially useful for
telecommunication photodetectors. Researchers are now also actively exploring practical
devices for long wavelength windows such as 2 to 4pm and beyond - these devices will
be vital to next-generation lightwave communications and laser range finders and Infra­
red imaging. InGaAsSb with appropriate compliant substrate is the promising material o f
choice for all o f these applications.
In chapter 2, we introduce an electrically isolated guard rin g Avalanche PhotoDetector
- APD. We formulate an analytical model o f the device to obtain the electric field
distribution within an APD. A two-dimensional finite element difference method (2-D
FED) is used to demonstrate the effectiveness o f the use o f guard rings in reducing
surface electric field (by 15%) along the semiconductor/insulator interface in planar
avalanche photodetectors (APDs). We demonstrate the advantage o f using the electrically
isolated guard ring and doubly-diflused devices over conventional APD structures with
electrically engaged guard rings. After a discussion o f the results, we conclude that the
introduction o f guard ring indeed provides an extra degree-of-freedom in the design o f
APDs and enhance device characteristics against curvature-effect initiated edge
premature breakdown without incurring additional penalties. We follow this with a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
chapter on practical Separate Absorption and Mutiplication Avalanche photodetectors
and in chapter 4, we present our experimental study o f the superlattice SAM-APD.
In chapter 5 we look at our high performance guard ring p-i-n photodetectors suitable
for use in high speed microwave subcarrier WDM networks and Passive Optical
Networks (PON). We show that diffusion currents originating from the bulk and the
periphery o f the active region induce slaw -tail in conventional p-i-n detectors. We then
consider the problem o f optimizing the device for both analog and digital applications.
The optimal device is also appropriately scaled to control other nonlinear mechanisms as
detailed in section 1.2. Our solution demonstrates the use o f diffused guard ring around
the active area. The electrically isolated guard ring is similar to that introduced in chapter
but now used to control diffusion and other nonlinear effects instead o f curvature edge
electric field premature breakdown. We quantify the slow-tail as an Extinction Ratio. The
extinction ratio is defined as the ratio o f the pulse height to the residual tail.
presents our suggestions for future work - the InGaAsSb material system
for Mid Infra-Red {MIR) applications.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
Electric Field Profiles of Electrically Isolated Guard Ring Avalanche
2.1 Introduction
Researchers have studied high-voltage devices such as the avalanche Photodetector
(APD) extensively because o f their dual function o f serving as a detector of incident
photon, and as an amplifier for photo-generated carriers produced by the absorbed light.
An APD is very attractive to the telecommunication system designers because it can
enhance both the bandwidth and span length between repeaters, compared to that
achievable by other detectors. Therefore APD is favored for repeaterless submarine
applications such as coastal stitching and island hopping where eliminating the number o f
underwater active components is critical to overall cost and reliability, because o f its
potentially large gain-bandwidth product But making APDs requires a proper Electric
field profile for adequate and reliable operation. It is therefore essential to compute the Efield to the required specifications. This will eventually allow for high yield, improve
accelerated aging tests data, and provide longer lifetime. This chapter presents the results
o f field calculations o f APD planar structures (guarded and unguarded). And the intended
purpose is to show that the use o f guard ring and double diffusion in high voltage
junction planar APDs with adequate power dissipation help to reduce maximum edge and
surface fields. This has a direct impact on reliability because surface field strength is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
linked to the breakdown voltage. During surface-initiated breakdown, transient energy
that can be dissipated is considerably less than the amount that can be absorbed by the
central active junction, and cannot be determined apriori. It is therefore essential to
design the device such that the central junction breaks down before the surface fieldinitiated breakdown sets in prematurely. This alleviates the electric field crowding at the
periphery o f the planar junction. It is well known that the breakdown voltage o f a planar
p +r r junction is limited by electric field enhancement at the curvature o f the junction
plane. For example, in planar APD, there is the necessity to prevent premature low
voltage breakdown at the junction periphery, which curves in from the surface o f the
semiconductor device. This is known as the “curvature effect” [Armstrong, H et al 1956;
Sze, S.M et al 1966; Lee, T.P et al 1967]. Premature surface breakdown is to be avoided
in all practical devices, since it causes a concentration o f current at the edge o f the
junction. Furthermore, the breakdown voltage is now a sensitive function o f surface
conditions and not o f the bulk resistivity. It is a known feature in diffused or ionimplanted junction devices. Hence, it is essential to reinforce these weak edges against
uncontrolled surge in E-field. In the discussion section, we present sample potential and
electric field profiles o f guarded and unguarded structures to demonstrate the
effectiveness o f employing an electrically isolated guard ring (since it is completely
electrically isolated from the central planar junction its potentials are allowed to float at a
value somewhat less than that established by the externally applied reverse bias voltage).
We then conclude with a brief summary.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.2 Formulation of the Discrete Poisson Equation
The field profile at the metallurgical p +r r junction and the corresponding space-charge
contour are both subject to Gauss’ law (from Maxwell’s equation), or Poisson’s equation.
This is the vital link between the field shape and strength [jE(x,y)|], o f real and practical
device parameters. The potential distribution is given by
S 2V (x ,y )/d z + d 2V (x,y) / d y 2 = (q/x)[\N j(x,y) - Na(x,y)\ + [p(x,y) - n(x,y)\] ...(2.1)
where we now have V(x,y) as the potential distribution, in 2-d; and we must add that the
RHS o f equation I reduces to 0, m a n oxide or a dielectric layer (th is involves solving the
Laplace’s equation), assuming the breakdown voltage im plies infinitely large gain.
Equation 2.1, implies uniform charge distribution fo r a ll o f the epitaxial layers.
Our approach is to numerically compute the E-field from the discrete form o f the above
equation. And since APDs typically work under very high reverse bias we can ignore the
free carriers and band discontinuities between the InP and InGaAs layers. Uniform
charge distribution is also assumed in the epitaxial layers depending on their dopant
concentrations. To solve the above equation, it is essential to linearize the second order
differential Poisson equation. This requires approximating the equation by its discrete
form on a network o f mesh grid nodes within the simulated edge segment (as illustrated
in figure 2-1). Then a mesh net o f points is defined over the device simulated segment
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and the grid density is chosen depending on the doping level in each layer. The
combination o f the above steps is known as the fin ite difference method [Selberherr S.
1984; Franz A J et al 1983; Adachi T., et al 1979], The solution o f the finite difference
equation is formulated by approximating the continuous electrostatic potential by its
value on a network o f nodes within the device boundary while simultaneously
constricting the solution to be between zero and the applied voltage. The grid structure is
rectangular with non-uniform spacing in both spatial x and y directions. The grid format
is desirable because it can be employed to accurately represent the device features and
can be easily generated automatically. The finite difference equations provides a discrete
approximation o f the continuous Poisson equation that relates the potential value at a
node to the values at neighboring nodes. For our calculations, we used the nearest
neighbors. The perpendicular bisectors o f the lines joining the center node to the outside
nodes define a rectangle (with appropriate rectangular sub-regions).The finite difference
approximation at the center node is obtained by integrating the above two-dimensional
Poisson equation over the triangle. The outer boundaries are defined by the following
boundary conditions:
p ' Central Junction 5
n ln P
n‘ InP
n* InGaAs
n* InP
n 'In P
Figure 2.1 An illustration o f the simulated segment
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
(a) W e require that the electric field along the line 1-4 be define only in the y-direction.
(b) The substrate is assumed to be grounded, therefore the potential on the lines 2-3 and
3-4 are both effectively forced to zero.
(c) Along 5-2, the electric field is effectively forced to be parallel to the surface, making
surface charge insignificant.
(d) And finally, we required that the potential along the chord 1-5 be fixed at the reverse
bias voltage.
In order to obtain the electric field distribution in the device under reverse biased state,
we must solve the Poisson equation as enumerated above. Differential equations are first
converted to finite difference equations as discussed earlier and they can be numerically
solved by using Newton iterative method. The technique we have successfully employed
requires the use o f the combination o f Newton iteration which deals with coupling
between equations. Many authors [Carre B.A., 1961; Ortega J.M. and Rheinboldt W.C.,
1970] have discussed this method in details. An essential portion o f this iteration is the
application o f the depletion region logic. This is a self-consistent means o f determining
which portions o f the device have space charge. Our objective is to formulate the
depletion region logic based on the criteria that the depletion layer approximation is valid
for this device and that the electric field vanishes at the depletion boundary. The
implication o f the first criterion is that complete depletion o f the carrier occurs sharply
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
across the depletion boundary and the second criterion implies that electric field only
exist in the depletion region.
We first attempt to solve the Poisson difference equations at every grid location within
the defined segment in figure 2-1 on page 25, (our simulation is based on a practical APD
wafer), using a charge density equal to the net impurity charge. Thus, the potential is
calculated at each mesh node but it is restricted to positive values only (negative values
are discarded) and limited to be less than or equal to the applied reverse biased voltage.
This argument is valid because lower or higher voltages will lead to the creations o f
regions o f local potential minima (potential wells) for holes o r electrons. Free carriers are
then expected to migrate to these potential wells (until the potential readjusts itself back
to zero o r the applied reverse bias).
One unique feature o f our calculation is that it is not necessary to track the depletion
layer boundary or to vary the depletion region logic with local node position within the
device. This feature makes our approach suitable for modeling devices with multiple
electrically isolated field limiting rings, because it is less complex and consumes less
computational time resources. The real challenge that remains is that while the potentials
o f the neutral portion o f the electrodes are fixed by their contacts , the potential o f the
field limiting electrically isolated guard ring is unknown. This can be seen from the
following discussion. If one applied the simple “one-dimensional theory” to the problem
o f designing field ring devices, very poor design would be obtained since the field ring
would always be placed too far from the main junction. The first problem is that the field
ring potential at breakdown for an optimum design is unknown. Now if the breakdown
always occurred at the surface, the optimal field ring would be midway between the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
anode and cathode potentials, since the field ring will act like a voltage divider for
surface fields. However, the breakdown rarely occurs at the surface but will usually occur
at the maximum curvature. The field ring does not efficiently divide the voltage at this
point since the main junction always has the full applied voltage directly below i t As a
result, field ring must be closer to the potential o f the main junction (anode) than the
cathode in optimal design. As will be seen in figure 2-3, the difference in potential
between the field ring and the anode ranges from about 40% o f the applied voltage to
virtually zero. However, the second problem is that even if the field ring potential is at
breakdown, it would still be grossly incorrect to place the field ring such that punchthrough can be reached at this final voltage difference since it has been demonstrated that
the potential difference between the anode and the field ring changes substantially with
increase in the applied voltage. The technique that we have used to determine this
potential is based on the fret that once the fields have attained steady-state value, there
must be zero net current flowing into each o f the electrically isolated field lim iting guard
ring. To satisfy this criterion the field ring junction cannot be completely reverse biased
or heavily forward biased. Thus the potential o f the ring must be equal to the lowest
potential along the metallurgical ring-cap layer junction less the built-in potential. The
separation between junction and the field ring junction was optimized so that the values
o f multiplication factor through the field ring junction and the anode junction were
simultaneously at breakdown values. This criterion results in the highest breakdown
voltage, since moving the field ring from this position will result in one o f the
multiplication factors increasing beyond an avalanche condition with the other
decreasing. A slightly higher potential implies that the ring-cap junction is heavily
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
forward biased and a potential lower than this by even a fraction o f a volt would suggest
a reverse biased junction. This means a net current is flowing into the field limiting
electrically isolated guard ring. This violates the criterion that the field ring remains at a
constant potential (at breakdown). The condition is valid at above or below punchthrough voltage. This criterion is continuously applied during the iteration together with
the depletion region logic. This is the extra loop used in solving the problem o f the field
limiting guard ring.
2.3 Results and Discussion
Field calculations are performed for the planar structures. The results o f our simulations
are a summary o f numerical calculations investigating the role o f some o f the critical
design parameters such as the ring distance from the central p +r r junction, ring width
and separations. The diffused guard ring can be used to force the curvature o f the
depletion layer into a desirable shape such that the breakdown voltage at the central
junction is less than that at the edge. This implies that the breakdown will occur over the
central junction and uniform avalanching can be achieved. The scope o f our work
includes punch-through condition (and beyond) in which the portions o f the device both
under the central P+r r metallurgical junction and the ring are folly depleted (such that
the guard rings can be engaged prior to the breakdown o f the guarded central junction).
For the planar device, we examined double diffusion for the formation o f the
metallurgical p-n junction independently from the composite use o f guard rings. From
our simulation we observed that each mechanism plays a different role. Specifically edge
breakdown can be effectively controlled by double-diffusion. The double-diffusion
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
creates a graded p ~ rr junction which results in a graded p_ junction edge at the periphery.
For this structure the computed maximum electric field under the graded junction remains
constant under the central junction. In addition we observed from the field profile that the
edge and surface fields changes w ith the first diffusion depth primarily due to the
curvature effect. Overall the edge field is reduced by up to 20% as compared to the p r r
junction (abrupt step junction). The surface field was similarly affected (with a reduction
o f about 15%). The conclusion we can draw from this is that graded p +r r junction may
help reduce surging curvature effect and surface electric fields. To ascertain the role o f
electrically isolated guard rings we simulated a device with a pair o f guard rings in
addition to double diffusion. Notably, we observe no additional edge electric field
reduction but there is a
reduction o f the surface field compared to the unguarded
device (as illustrated in figure 2-2). The reduced surface electric field has direct
implication on device reliability in that the penalty due to the surface leakage portion o f
the dark current will become less. This is evident in the value o f the potential o f -21V at
the surface. The corresponding potential without a field limiting ring at the
semiconductor insulator interface is found to be -35V. Figure 2-3 represents the
distribution o f the device structure with a pair o f
field limiting rings, and the
corresponding structures are shown in figure 2-4. The electric field distribution shows
high field at the junction peripheries o f the first and second diffusion for both the guarded
and unguarded cases. The results o f our simulations seem to suggest that the field
limiting ring is primarily effective in reducing surface fields. One additional constraint
(aside from the obvious one requiring all surface electric field distributions to satisfy the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
breakdown condition) is that our simulations met the requirement o f a peak electric field
o f 650kV/cm. This upper boundary condition ensures that InP which serves as the
multiplication layer does not breakdown and that the electric field threshold (o f 450
a. 2
• 39.1
Simulated Segm ent (micron)
Simulated Segment (micron)
Siniuiu(C({ -
8,,,c,,t Oiii'cron)
Depth (mieren)
Figure 2-2 Potential (a) and E-field (b) Profiles o f a planar device without guard rings
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Simulated segment (micron)
Sim ulated Segment Cm,'cron)
Figure 2-3 Potential (a) and E-Field (b) Profiles o f a planar device with a pair o f guard
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Simulated Segment (micron)
Simulated Segment (micron)
Figure 2-4 Device structures (a) unguarded (b) with one guard ring
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
kV/cm) required for Impact Ionization process is achieved within the device for adequate
gain. From this an optimal device structure can be obtained. O ur calculation suggests that
both the rings width and spacing influence the field distribution. We also observe that a
narrow width ring tends to inhibit field enhancement in its immediate sub-region,
whereas wide field limiting ring appears to aid in reducing peak electric field in the
preceding portion o f the device. We therefore suspect th at in devices requiring multi-ring
limiting capacity narrow-width will probably be effective for exterior application and that
wide ring may be more suited for interior use to prevent excessive field surge. The
numerical technique accounts for lateral impurity diffusion penalty in modeling the
central junction and the guard rings. It can also be used to model fields in devices without
guard rings such as the ‘beveled’ and the mesa structures.
Another source o f catastrophic failure is uncontrolled heat generation during avalanche
breakdown. As stated in section 2.2, we have assumed uniform and defect-free epitaxial
layers (that is no microplasmas). The heat generated should be proportional to the electric
field. It is suspected that heat generation should not be significant at low currents. To
account for this problem at large and excessive current it is quite possible to generate the
contour profile within the device as we have done for the electric field. The equations
describing the temperature distribution in the device can be coupled to the continuity
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.4 Conclusion
In this chapter we have presented a concise result o f a technique for calculating electric
field distribution in a high-voltage device such as the avalanche photodetector, by solving
a set o f difference equations together with an algorithm that in a self-consistent manner
determines which regions o f the device have residual space charge continuously until
breakdown conditions are satisfied. Our results suggest the superiority o f ring-guarded
planar structures for decreasing surface electric field and the doubly diffused central p +r r
junctions for the reduction o f edge electric field. Our simulation allows the prediction o f
the static and dynamic characteristics o f high-voltage devices such as the APD. It can be
employed as a first approximation tool to design high performance avalanche
photodetector. This is the subject o f the next chapter.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Separate Absorption and Multiplication Avalanche
3.1 Introduction
It is generally acknowledged that lightwave communication systems can benefit from
the use o f Iuo.5 3 Gao.4 7 As/InP Separate Absorption and Multiplication Avalanche
Photodetector (SAM-APD) [Nishida K, et al; 1979]. This geometry provides a means of
limiting the dark current due to tunneling in narrow bandgap semiconductor materials.
For these devices photon absorption occurs in the narrow bandgap, with the
photogenerated carriers subsequently swept into the wide bandgap where impact
ionization occurs. An APD can enhance both the bandwidth and span length between
repeaters over that which can be achieved using other detectors such as the p-i-n
photodiodes. The purpose o f this chapter is the application o f the design tools acquired in
chapter 2 for fabrication and micro-machining o f a high performance planar SAM-APD.
The simplicity o f this device can lead to high-yield manufacturing process. The device is
intended for use in the 1.3 and 1.55pm spectral windows required for lightwave
The most important properties o f a photodetector are efficiency, speed, noise and
physical compatibility. foo.5 3 Gao.4 7 As/InP SAM-APD has demonstrated superior
performance for communications applications due to its relatively low bulk leakage
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
current, high quantum efficiency over & wide spectral window, and an attending high
gain-bandwidth product [Campbell J.C et al; 1988]. Making a practical SAM-APD
however, demands the combination o f simple and viable design, good quality material,
and a high degree o f process control. An avalanche photodiode requires the avalanche
multiplication to be spatially uniform over the entire light-sensitive area o f the diode.
Microplasmas that is small areas in which the breakdown voltage is less than that o f the
junction as a whole must be eliminated. The probability o f microplasmas occuring in the
active area is minimized by using low dislocation materials and by designing the active
area to be no larger than necessary to accommodate the incident light (may not be
monochromatic and variation in the penetration depth must also be considered
consequently APDs are) generally from a few microns to 100pm in diameter. The
primary design criterion o f SAM-APD is that the integrated charge density o g in the
multiplication layer must lie in the range 2.2 to 2.7 x 10
cm . This boundary condition
ensures that the maximum field in the InP multiplication layer is high enough to support
impact ionization when the device is biased near o r at breakdown. Also it assures that the
electric field at the InGaAs/InP heterointerface is kept low enough (-200 kV/cm) to
prevent tunneling leakage current, and higher than 100 kV/cm for high speed and good
quantum efficiency (Machida H, et al; 1988].
To achieve a high performance APD, the key characteristics that one must consider
include, low noise, process reproducibility and high reliability. As for the geometry we
have two choices. The mesa and beveled structures are easier to fabricate and to package
in high frequency applications as their geometry can facilitate impedance matching. They
are however prone to edge breakdown due to high surface electric fields. This can lead to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
low reliability. I f the mesa geometry must be used then the issue o f field confinement
must be addressed. One solution is to use a reverse p-n junction as illustrated in figure 31.
InP (p)
InP (p)
Substrate (n*)
Figure 3-1 A reverse p-n junction for better electric field confinement
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For high device reliability it is desirable that the APD be o f a planar architecture. The
planar architecture also provides better power dissipation tolerance before catastrophic
failure. The planar APD structure is however complicated by the necessity to prevent
premature low voltage breakdown at the junction periphery, which curves in from the
bulk o f the device to the surface o f the semiconductor. This is the “curvature effect” cited
in section 2.1. It is a known feature in diffused o r ion-implanted junction devices. As a
rule in semi-cylindrical (to a lesser extent in the semi-spherical) abrupt one-sided
junctions, the breakdown voltage (Vbd ) decreases as the radius o f curvature (and the
junction depth), because o f the associated high field intensity at the edges. Therefore it is
essential to reinforce these weak edges against uncontrolled surge in E-field. A
commonly prescribed solution is large diffused radius. This is not a viable practical
solution. It can offset the delicate boundary conditions required for integrated charge
density as stated earlier. To adequately address this anomaly, there are two good
approaches. One is the introduction o f guard rings at the junction periphery to reduce the
field and therefore help suppress edge breakdown [Chi G. C. et al; 1987]. The second
approach is the enhancement o f the E-field in the central active region o f the junction by
selectively increasing the charge density under this layer [Webb P. et al; 1988]. Some
combination of various methods have been applied in the past requiring additional
fabrication and material constraints, thus leading to complex structures and difficulty in
material processing. There is therefore a need for trade-offs in design.
For this device however, the combination o f a pair o f electrically isolated guard rings
(EIGRs) and double diffusion will suffice. For ease o f material handling however we
I ____________________________________________________________________________________________________________
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
have recommended a single diffusion step as a compromise. Simulation and analytical
results supports this as a viable solution. In this chapter we summarize the design and
fabrication steps o f planar SAM-APDs w ith electrically isolated guard rings. In section
3.2, the results o f the design analysis and a two-dimensional computer simulation are
reviewed. This will be followed by a general discussion and then a conclusion.
3.2 Two-dimensional design and analysis
The structure is based on a proprietary wafer (figure 2) designed solely for the
realization o f production grade APDs. The epitaxial layers are grown on a (100) S-doped
InP substrate (with an Etch Pitch density o f ~ 5.0 x 10 cm ). The first layer grown
consists o f a 2.23-pm-thick InP buffer o r filter layer (n+ type S-doped to ~ 2.8 x 1016 cm"
) for preventing
Thickness (pm)
|Nd -N .|
Lattice Mismatch
Figure 3-2 An Avalanche Photodiode Wafer Spread
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
thermal migration o f defects from the substrate into the photo-absorption layer. This is
followed by an approximately 2.43-pm thick n~ Ino.5 3 Gao.4 7 As photo-absorption layer
with free carrier concentration o f 1.5 x 1015 cm'3. Next, the n-InP multiplication layer is
grown with a thickness o f 0.64 pm, and a sulphur doping o f 4.G x 1016 cm'3 giving an
integrated charge density o f CTb - 2.56 x 10
cm'2 (diffusion depth in the cap layer may
be adjusted for additional charge density as desired),as defined by the boundary condition
in section 3.1. For the cap a 4.03 pm thick unintentionally doped InP layer with a
background concentration o f -
1 .2
1 0 15
cm'3 is grown at the top. The general design is
presented in this section. For analytical and modeling purposes the wafer is divided into
two regions. Namely: (a) the edge region which covers the periphery o f the device
including two curved diffusion fronts and EIGRs, and (b) the central active region where
only a flat second diffusion front exists in the n - cap layer. The problem now is to be
able to predict the conditions under which avalanche breakdown only occurs in the
central region o f the device. The roadmap is as follows. First, the breakdown conditions
in the central active region is determined. The unintentionally doped cap layer allows the
depletion layer to extend laterally far enough along the device surface so that the guard
rings are engaged prior to breakdown. It is also expected that the guard rings will help in
reducing the surging electric field in both the curved junction edge region and along the
device surface. The use o f the cap layer serves to separate the junction diffusion depth p+
and the n-InP multiplication layer thickness into two random variables that can be
independently optimized. This provides us with additional degree o f freedom. The center­
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
piece o f the design is the field profile within the device itself The field is the key linkage
between the epitaxial parameters and the device performance characteristics. The
procedure for extracting the field from Poisson equation has been detailed in chapter 2. It
is very trivial to demonstrate that the Poisson’s relation above collapses into a one­
dimensional expression by using appropriate transformation and symmetry along the yaxis. h i this special case the potential and therefore the electric field along the y-axis can
be solved analytically (or computed as shown in figure 2 - 2 in chapter 2 ) along the crosssection o f the device, h i addition to field computation the next important operation
parameter is the breakdown voltage. For this device avalanche multiplication is the most
dominant (in addition to avalanche multiplication, there are basically tw o other
breakdown mechanisms: thermal instability, and tunneling effect [Capasso F.; 1985])
since the avalanche breakdown voltage imposes an upper limit on the reverse bias o f the
diode. The breakdown voltage can be determined by computing the multiplication gain
which is controlled by the impact ionization coefficients [Cook L.W. et aL, 1982; and
Tagushi K. et al; 1986] for holes (P) and electrons (a). Optimum noise figure can be
attained if only one species ionizes and in a completely deterministic fashion. Therefore
low noise operation is obtained by using a w-type multiplication layer and InGaAs
photoabsorption layer to provide pure hole single injection [McIntyre R. J., 1966]. This
can be readily demonstrated by using first principle statistical analysis o f branching
process, to show that the most critical feature in low-noise APD operation is the absence
of residual electron ionization factor.
An abrupt p +-v one sided junction and uniform doping profiles have been assumed for
convenience. The wafer we introduced above lacks a quaternary InGaAsP speed-up layer
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
that reduces hole pile-up [Campbell J.C. et al; 1988]. We therefore need to use the E-field
to enhance the response speed and at the same time the dark current must be controlled.
To achieve this we have to lim it the maximum electric field at the heterojunction
interface (that is Efai < 200 kV/cm). This constraint also help to eliminate the possibility
o f photo-carrier multiplication in the InGaAs layer (this can lead to additional noise if left
uncontrolled). For our structure it is safe to design for a heterointerface field o f at least
150kV/cm since it lacks a speed-up layer. The InGaAs photoabsorption layer thickness o f
2.43 pm essentially ensures a high quantum efficiency. An external quantum efficiency
(Hext) o f better than 80% is expected at 1.3 pm. The doping level has been deliberately
kept at 1.5 x
1 0 15
cm'3 to facilitate full depletion o f the absorption layer and therefore
assures fast response. The carrier transit time is expected to be significantly reduced.
Using the above device parameters the calculated breakdown voltage is 74.8V. Analysis
showed that the maximum field required to achieve avalanche breakdown is strongly
dependent on the thickness o f the InP layer. Our calculation suggests that the thickness of
InP should be kept below 1.0pm.
3.3 Device Fabrication
The computer simulations detailed in chapter two allowed us to demonstrate the effect
o f guard rings. Different permutations were performed with or without guard rings. A
close study o f the electric field profile revealed that the EIGR has a limited effect on the
curved contour around the junction. The plots however clearly suggest that the guard
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
rings plays a major role in reducing the surface field. This is a very useful role because
there is evidence that the surface fields are clearly responsible for surface leakage (a
reservoir o f dark current). A low surface field is desirable since it provide better
reliability. It has been suggested that failure o f planar APDs with conventional guard
rings may be due to excessively high surface field which is enhanced by semiconductor
and insulator interface charge [Sudo H. and Suzuki M ; 1988]. Low surface field is
therefore crucial for achieving high reliability.
From the computer simulations it suffice that a pair o f electrically isolated guard rings
will be adequate for this device. For our wafer the proposed diffusion depth is l.S3pm.
The spacing between the central junction and the first ring is estimated to be 6 pm. The
guard rings will be 2 to 3pm wide (while the central junction is expected to be 60pm with
some tolerance.
Most structures in past literatures employ one mechanism - the guard rings- to
eliminate edge breakdown. A more effective approach is to use two mechanisms to
eliminate edge breakdown - the electrically isolated guard rings and a stepped junction
edge formed by double Zn diffusion as suggested in section 3.1. Each o f these
mechanisms plays a different role and can therefore be optimized independently. The
mode o f operation o f the EIGRs is in principle similar to the floating rings in some
silicon power devices [ Ghandhi S.K, 1977], During operation when the reverse bias
voltage applied to the central junction increases, the depletion layer extends vertically
and laterally into the v-InP cap layer. However, since the guard rings are heavily doped
they naturally behave like metals and therefore form equipotential surfaces. When the
depletion regions now reaches the inner edges o f the equipotential EIGRs it jumps to the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
outer edges o f the EIGRs. This expansion o f the depletion region acts to increase the
separation o f the equipotential lines at the curved junction edge as well as along the
semiconductor-insulator interface. As a result it simultaneously reduces both the edge and
surface fields. O ur calculations from chapter two also reveal that the reduction in edge
field due to the existence o f the EIGRs is not sufficient by itself to entirely eliminate
surface o r edge breakdown. Computer simulations proves the efficacy o f using a doubly
diffused junction in that the electric field is more enhanced in the central junction where
it is needed for initiating impact ionization collisions (the second diffusion front is seen to
be closer to the n-InP multiplication layer than the junction sub-region). We also observe
that the junction periphery becomes graded during the second diffusion during the drivein process under the SiNx passivation layer. The compositional grading effectively
reduces the maximum electric field around the periphery thus neutralizing the curvature
To ascertain the effectiveness o f the EIGRs, a spatial profile o f the photocurrent
response was determined by scanning a focused 1.3|im wavelength light spot from a
single mode fiber (10pm core) across the device. Figure 3-3 shows a series o f such scans
performed at different biases (near 90% o f the breakdown voltage - 70 to 75V).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3-3 Scans across the active layer o f the device at biases near Vbd
As regards the expected response speed o f the device, measurements shows that the
junction capacitance including the package is 0 .6 pF at 90% o f V b d - The EIGR does not
incur additional capacitive penalty. Since the expected additional capacitance is added in
series to the normal junction edge capacitance. Therefore the total capacitance o f the
diode with the EIGRs should not significantly differ from an identical diode without
guard rings. The central junction is not effectively connected to the EIGRs. Therefore
the guard rings capacitance is nom inally insignificant.
3.4 Device structure and architecture
The EIGRs eliminates the need for large radius guard rings to prevent edgebreakdown. Instead the guard rings are replaced by a series o f heavily doped pairs o f
R ep ro du ced with permission o f the copyright owner. Further reproduction prohibited without permission.
rings. They were formed during the same diffusion (i.e a single photolithographic step)
used to form the central p-n junction. This process involves plasma deposition o f a thin
SiNx layer onto the v-epitaxial surface and etching the central p-n junction window
along with the EIGRs. By careful processing adequate results can be obtained using Zn
as the dopant species. It would also appear that Be ion-implantation technology may be
applicable for accuracy. The p-n junction and the EIGRs were formed by opening up a
ring pattern in a SiNx
mask deposited onto the wafer surface. The precision
characteristics o f ion-implantation promises to limit the spread o f the p-n junction and
guard rings beyond the edges o f the windows after rapid thermal annealing. The
resulting broadened rings from the diffusion process can be reduced to approximately
half this width by using ion-implantation (with some lattice defects being incurred). The
precision offered by the ion-implantation process will be valuable for packing additional
rings in the periphery o f the central junction as required. The “winging” problem
associated with the p-n junction beneath defects in the SiNx mask that features in
diffused ring structures can be reduced. Such “winging” tends to short out adjacent rings
rendering them ineffective. The major disadvantage o f using ion-implantation stems
from the fact that it is characterized by inherent lattice damage as cited above. This may
affect the breakdown conditions o f the device. Since these damages are usually localized
sites in the relatively high-field p-n junction. Rapid thermal annealing o f damages may
help but there is no reliable data on its effect on the performance o f APDs as o f now. If
property done it is expected that ion implantation may provide a higher yield device with
better breakdown conditions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
As a final step, a second nitride layer was plasma-deposited. This step is followed by
deposition and alloying o f top and back surface contacts. Since no contacts are necessary
for the ru g s therefore they will remain electrically isolated. The top metal contact is
only deposited in the central portion o f the p+- region. A window opening was made for
back entrance via the substrate for a back illuminated device. The alternative is for front
illumination device. The wafer can be processed for front illumination. Infact the front
illuminated device at least in theory will appear to be foster (in response) than the backilluminated version. This is to be expected because with a front-illuminated device the
photogenerated carriers are located in part in the depleted portion o f the ternary InGaAs
photoabsorption layer. Since the movement o f the depleted portion is in the same
direction as the approaching illumination from the top o f the device. Therefore diffusion
o f photogenerated carriers is drastically minimized. Whereas in the case o f the backilluminated device the depletion layer and the approaching illumination are counterpropagating. As a result a good portion o f the photogenerated carriers will probably have
to diffuse into the approaching depletion layer. This will slow them down considerably
since the mechanics o f kinetics dictates that carrier drift is always faster than the carrier
diffusion. One means o f alleviating this problem is to incorporate a graded or speed-up
quaternary layer (of InGaAsP) in-between the n InGaAs photoabsorption layer and nInP multiplication layer. This will help reduce some o f the hole pile-up and therefore
speed up their transit through the device thus enhancing efficient photogenerated carrier
collection at the contacts.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.5 Conclusion
This chapter has addressed the main issues that are relevant to the design and
fabrication o f the a highly reliable SAM-APD. Employing a structure consisting o f a
narrow-bandgap InGaAs layer for absorption and a wide-bandgap InP for the
multiplication o f photogenerated carriers solved the problem o f dark current.
Anticipated problems with regards to excessive field build-up due to curvature effect
and associated premature voltage breakdown have been solved by the application o f
guard ring without incurring additional capacitive penalty.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4
Experimental Study of Quarternary Well Superlattice SAM-APD
4.1 Introduction
Superlattice Separate Absorption and Multiplication Avalanche Photodiode (SL>
SAM-APDs) have been studied intensively, first because o f their dual function of
serving as a detector o f incident photons, and as an amplifier for photogenerated carriers
produced by the incident photons. SL-SAM-APD promises to be a high-speed highly
sensitive optoelectric transducer having composite functions o f a photodetector and a
current amplifier. Secondly they have the potential for a large gain-bandwidth (GB)
product and low-noise characteristics due to an enhanced Impact Ionization (II) rate
ratio. A comparison o f various solid state photodetectors reveals that the intrinsic photo­
conductor can provide the highest internal gain (~104), while Schottky barrier
photodiode has the shortest response tim e (10
sec) and the largest bandwidth. On the
other hand the avalanche photodiode has the highest gain-bandwidth product The low
power consumption property o f superlattice SAM-APD is also attractive for practical
optical receivers.
SAM-APDs can be fabricated from a wide variety of semiconductor materials with
different structures. Besides conventional germanium and silicon APDs several APDs
fabricated from m -V compound alloyed semiconductors have been reported widely in
literatures. However, there have been few reports on SL-APDs exhibiting a gainbandwidth product larger than
1 0 0 GHz
and a submicro-Ampere dark current This dark
current specification is required to realize receiver sensitivity improvement in the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
lOGb/s range by increasing the gain-bandwidth product the dark current should be less
than 20nA (the boundary condition for this transmission rate). The low leakage and dark
current requirement ensures that the highest sensitivity is multiplication noise limited
rather than by the dark current. The origin o f this dark current can be traced to tunneling
[Mendez, E.E., Wang, W .I., et al; 198S] o f thermally excited carriers through the well
layers with a small band-gap because the tunneling probability strongly depends on the
bandgap. It is therefore essential to increase the bandgap in order to effectively suppress
the dark current [Kagawa T. et al, 1982]. This can be achieved by introducing a
quartemary alloy such as AlxGaInyAs (1.03eV) into the well. In this chapter we discuss
the experiments performed to study the use o f superlattice for the multiplication layer in
4.2 Device Structure
We investigated two different wafer designs each consisting o f 9 epitaxial layers (as
illustrated in figures 4-1 and 4-2). The processed chips were mounted in a flip-chip backilluminated configuration. This provides a means o f reducing the bonding pad and
junction capacitances (junction diameter can be reduced as needed). This design also
improves the quantum efficiency because o f the p-contact-electrode reflection. With this
configuration the photoabsorption layer can be made thin to reduce the carrier transit time
without any significant penalty. The inductance o f the bonding wire can also be
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The separate Absorption and Multiplication layer was grown on a low resistivity
sulphur-doped n+-InP substrate to reduce RC time constant. The n+ substrate provides for
low optical attenuation and improves the quantum efficiency. The SL-SAM structure
consists mainly o f an undoped AIxGaIhyAs - AlxInAs Superlattice multiplication region
and the lightly Zn-doped InGaAs photoabsorption layer. The superlattice is a thin 0.34pm
thick layer. Its primary function is to enhance the gain-bandwidth product The SL period
is 15. The thickness and the impurity concentration o f the photoabsorption layer are
0.9pm and 8 x 10
cm respectively (structure I).
The wide bandgap o f the (AlGalnAs) SL was employed for reducing tunneling dark
current component as discussed above and for transparency in the 1.3 to 1.55 pm spectral
window [Watanabe I, et al; 1993] so that light can reach all the way down to the p-n
junction (at the field buffer and SL interface). A high resistivity Zn-doped InP field
buffer layer with a thickness o f 0.052pm and a doping concentration o f 7 x 10
was introduced to control the difference between the high electric field strength o f the SL
multiplication region and the photoabsorption layer. It is sandwiched in between the SL
and the absorption layers. The primary function o f this layer is to provide a relatively low
maximum hetero-interface electric field (17.4 kV/cm) in the absorption layer thereby
preventing avalanche multiplication and suppressing tunneling dark current generation
(and to aid in regulating the electric field that is required for carrier drift within the
absorption layer). This layer because o f its relatively high resistivity is used for one o f the
junction where it plays a role in increasing the breakdown voltage (Vbd). The measured
Vbd f°r our devices is typically between 22 and 24Volts (operating voltage : 12 < V <
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2 0 ).
The p -InP layer (with a doping concentration o f 7 x 10 cm and a thickness o f
1 .0 pm)
is grown on top o f the photoabsorption layer to act as a window and a field-
termination layer. Finally a p+- InGaAs ohmic contact layer was grown on the field
termination layer. The epitaxial layers were optimized to minimize series resistance
which tend to aid Joule’s thermal effects heating O^Rseries) [Sheng S. L., et al; 1993]. As
illustrated in figures 4-1 and 4-2 there are two versions o f the design (I & II - 60pm mesa
diameter) that we fabricated and tested extensively - for their electrical and optical
properties. Both were o f mesa geometry. The mesa was formed by chemical wet-etching
technique and a plasma chemical vapor-deposited SiC>2 film was applied for anti­
reflection coating and surface passivation. Ohmic contacts to p+ and n+ layers were made
by alloying evaporated AuZn and AuGeNi respectively (for design I). Design U has an
additional metallization mask for the Ti-Pt-Au alloy that goes on top o f the Si02 layer o f
design I.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Carrier Cone, (cm*3)
1 .0
0 .1 0 0
Thickness (pm)
Figure 4-1 Epitaxial W afer Structure Design #1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Carrier Concentration (cm*3)
Thickness (pm)
1 .0
0 .0 1 2
0 .0 1 2
0 .1 0 0
Figure 4-2 Epitaxial Wafer Structure Design #n
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3 Energy Band Calculation
The energy band diagram at the hetero-interface o f the AlxInAs (1.45eV) and
AlxGaInyAs (1.03eV) is shown in figure 4-3. The InAlAs - AIGalnAs is thought to be
superior for the enhancement o f the Impact Ionization rate ratio [Capasso F., et al; 1981]
- the fundamental agent o f gain mechanism) for the following reasons:
The conduction band discontinuity (AEc) - the origin o f enhanced a (also differences in
scattering rates and the confinements o f electrons and holes) in the SL (the primary
means o f energy transfer to the ionizing species) is not markedly decreased and the
valence band discontinuity can be eliminated.
Although band discontinuity has not been measured directly but computation based on
a simple but realistic computer-generated band structure model for both the conduction
and the valence bands was realized by band alignment. The nominal band discontinuity
can be determined from the temperature dependence o f thermal emission current across a
single barrier structure o f AIGalnAs - InAlAs - AIGalnAs. From the figure it is obvious
that AEC = 0.39eV (correspondingly AEy as small as 0.03eV) was obtained which is
negligibly small. The electrons subsequently gain AEc in kinetic energy upon transferring
into the narrow-bandgap material (well) and lose AEc
kinetic energy at the other end
upon transferring into the barrier layer. The overall effect is equivalent to the
superposition o f a spatially varying electric field within the device. The E-Field profile o f
the heterostructure can be computed as described in chapter 2. The structure is biased
such that the action of the dc field heats up the electrons to energies near the impact
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ionization threshold within the narrow-gap semiconductor. The resulting superposition o f
the dc bias field and the spatially varying “ac-like” field arising from the spatial
periodicity o f the superlattice locally heats the electron distribution to energies at which
impact ionization can occur. Meanwhile the holes on the other hand can transit the SL
relatively unperturbed because AEy is practically negligible. We know from previous
work that InGaAs-InAlAs has a type I heterostructure. The transition to type II
heterostructure (InAlGaAs-InAlAs) must take place when the Al content varies while
maintaining the lattice matching to InP substrate. For the aluminum content at which the
transition from type I to II occurs, AEy vanishes while AEc *s less sensitive to the Al
content. The ionization occurs predominantly near where the spatially varying field
peaks. Our calculations show that the impact ionization collision length is between 50
from the interface.
The device promises low-noise performance. The electron and hole ionization
coefficients are vastly different in the material and the carrier with the largest ionization
coefficient initiates the avalanche process. Furthermore the wafer is designed such that
only one species ionizes (electrons in this case) in a completely deterministic way. This
guarantees that the optimum noise figure can be attained. To achieve this there should be
no residual hole ionization because even a small amount o f hole ionization can generate
excess noise. These properties should combine to make the device suitable for
applications requiring low noise specifications.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4-3 Energy band diagram
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.4 Conclusion
The preliminary data obtained from design #I suggests the realization o f practical devices
from this wafer. Our 60pm diameter flip-chip device shows signs o f meeting some o f the
key specifications required in a practical SAM-APD. We found out that the use o f
InAlGaAs has some advantages over InGaAs for application to superlattice SAM-APD
because the bandgap o f well layers can be made significantly larger while maintaining
almost the same value of AEc. Therefore the inherently larger dark current in the
narrower bandgap InGaAs can be significantly suppressed in the larger bandgap
InAlGaAs. The electron impact ionization is enhanced in the InAlGaAs-InAlAs SL
because the electrons receive kinetic energy from AEc band off-set at the heterointerfaces
and the reliability will not be degraded due to the moderate use o f the aluminum content
[Morris, N. A., 1995],
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5
High Performance Guard Ring P-I-N Photodetector for Digital Fiber
Optics Communications
5.1 Introduction
In digital fiber communication applications ranging from microwave subcarrier WDM
and passive optical nework to high-speed digital satellite antenna remoting and beam
steering, the p-i-n photodiode promises an efficient mechanism for converting optical
signals from the transmitter plane to the electrical domain in the receiver plane. As a
result o f improvement in design and fabrication, many new applications are possible at
microwave frequency. However, the speed o f a p-i-n photodiode is limited by the
intrinsic capacitance as well as the transit time required by the photon-generated carriers
to transverse the eintrinsic’ region. High-speed p-i-n photodiodes have received a great
deal o f attention {Bowers, J.E et al 1985; Gimlett J.L 1989; Wake D. 1992 and Lucovsky
G. et al 1964], and there has been considerable effort in developing these ultra-fast
devices. The requirement o f faster response times dictates the reduction o f the /-layer
thickness to minimize transit time delay. The reduction o f the /-layer however ultimately
leads to a larger intrinsic capacitance, thus increasing the device response time. Another
potential source o f speed limitation in p-i-n detectors is diffusion current problem. The
time response of conventional p -i-n photodiode is plagued by a wavelength-dependent
slow-tail arising from carriers photogenerated in the neutral regions within the bulk and
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the marginal area around the photodiode junction, reaching the depletion layer by
diffusion. It is a serious drawback in applications where weak signals must be detected
just after strong peaks that is Rayleigh scattering after a Fresnel reflection in optical
fibers. The diffusion current can introduce significant delay in the device performance
with the long tail making the device hard to turn o ff (that is, poor Extinction Ratio - ER).
The slow-tail device is particularly unsuitable for digital applications, as successful
is established by distinguishing the presence o f a pulse “on” and its
absence “off”. For example, in PON application the p-i-n photodiode has to have the
capability to resolve weak pulses juxtaposed in a string o f varying and relatively larger
pulses. In principle, reconvolution analysis could be used to recover the data, hi practice,
this is seldom obtainable with good accuracy. The ultimate resolving power o f the
detector is when the incoming signal’s power equals the total background noise power,
for a signal-to-noise ratio o f one. Therefore, the power o f the individual pulses are
relatively insignificant so far the presence o f a pulse can be ascertained. The timing o f the
sequence o f the pulses rather than the strength o f the individual is very crucial to signal
The slow-tail is practically due to photon-generated carriers in the undepleted regions
within the bulk o r near the surface o f the device. For slowly diffusing bulk-generated
carriers, this problem can be alleviated by employing heterojunction devices, with
additional penalties such as lower quantum efficiency, charge trapping and accumulation
at heterojunctions, and processing complexity [Bowers J. E., et al; 1985]. An approach is
to employ an optimized heterojunction device designed for short minority carrier
lifetimes and small width over diffusion length ratios for heavily doped regions [George
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
G. and krusius J. P.; 1994], Decreasing the minority carrier lifetime should improve the
performance o f the detector. However, in doing so the device must be optimized such that
the diffusion length is not decreased proportionately, since decreasing the minority carrier
lifetim e also leads to an increased reverse saturation current In this chapter, we propose
the use o f diffused guard ring around the active area to further minimize the deleterious
effects o f slow-tail originating from near the surface o f a heterojunction device outside o f
the active region.
The experimental data o f the guard ring p-i-n photodiode for a 20nsec pulse will be
presented in this paper. It will also be shown that the series resistance Rseries ^
noticeable effect on the E R penalty with increase in Rseries- The additional parasitic
capacitive penalty due to the guard ring is practically negligible, h i section S.2 , we
identify the source o f the diffusion slow-tail problem in a conventional phenomenological
p -i-n photodiode model. This is followed by a brief description o f the guard ring p-i-n
photodiode device structure. The results and discussion are presented in section 5.3.
Finally a summary and conclusion are presented in section 5.4.
71 —
i -
r e g io n
P reg w n
m m —m m
F ig u r e 5 -1
Schematic o f a typical p -i-n photodiode under illumination
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2 Analysis
The typical p -i-n photodiode (as illustrated in Fig.5-1) employs the photo-voltaic effect
for converting photons into electron-hole pairs, which then produce a current through an
external circuit. This photo-current generates a signal voltage across a load resistor. A
typical p-i-n photodetector is made up o f an /-intrinsic layer sandwiched in between a p
and n-layer. Usually, extra-heavily doped n+ and p + are used to improve ohmic contacts
(not shown in Figure 5-1). Electron-hole pairs can be produced in all the three regions by
photon absorption. The width o f the depletion region in both n and p layers is normally
significantly smaller than the width o f the n or p region itself. In the n layer the photon­
generated holes (within about one diffusion length L„) can diffuse to the metallurgical
junction, and then drift rapidly across the p-n junction under the junction field to produce
an external current. Similarly, the photon-generated electrons in the / 7-layer can diffuse to
the junction where they are eventually swept out into the external circuit.
The response o f the photon-generated holes in the p-layer, and those o f the electrons in
the n-layer is that o f the majority carrier. Holes (electrons) are therefore swept out o f p~
region (n-region) on a time scale that is o f the order o f the dielectric response tim e which
is very small. The speed limitation is virtually imposed by the diffusion o f minority
carriers. This diffusion current will contribute a slow-tail to the detector impulse
response. In addition the diffusion current could last as long as the carrier lifetime and the
charge content in the tail can be as large as the drift component due to the slow diffusion
times. This is the main problem with the conventional p-i-n device. Since the drift in the
junction field is relatively faster than the diffusion process, ultimately the best time
i ___________________________________________________________________________________________________
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
response and higher current is produced in devices where photons can be absorbed
preferably solely in the active depleted region o f the device. But for most applications
electrons-hole pairs are inevitably generated in the margin (bulk and surface) outside o f
the active area. That is the undepleted regions within the bulk and close to the surface o f
the device. There is therefore need for corrective mechanisms to check this effect. For the
bulk diffusion current an optimized heterojunction p-i-n has been proven to be effective
[George G., and Krusius J.P.; 1994].
Several ways can be used to alleviate a slow-tail in a typical p-i-n photodetector pulse
response: (a) Use o f a large area detector to facilitate easier coupling to the center o f the
detector and avoid light foiling outside o f the active area. However, there is the need to
use a smaller-area device, because o f capacitance constraints, (b) Employment o f single
mode fiber to facilitate easier coupling to the center o f the detector and prevent light from
falling outside o f the active area. But there is bias for multi-mode fiber because o f the
need for adequate power budget essential for link or network transparency, (c) Active
coupling o f the detector while monitoring the slow-tail current. This is not easy with
receptacle design as there might be variation from connector to connector and
manufacturing cost may be prohibitively high. We now propose the use o f a diffused p +
guard ring for collecting electron-hole pairs generated in the margin and near the surface
area to further reduce the diffusion slow-tail current. The field limiting capability o f
guard rings has been demonstrated in high-voltage semiconductor devices [Brieger K.-P
et al, 1983], hi these devices, guard rings were typically used for the prevention o f edge
breakdown. In this chapter, w e use diffused guard ring around the active area as a viable
solution for the slow-tail diffusion current problem. The cross section o f a guard ring p -i-
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
n device that we have fabricated is shown in Fig. 5-2. From the top, it consists o f SiNx
anti-reflection coating, another layer o f SiNx for passivation. Following this is the n r InP
cap o r window layer through which the p+ impurities (Zn) are diffused. The p+ diffused
central junction is punched through slightly into the photodetector absorption layer to
increase quantum efficiency by eliminating surface recombination. Its large bandgap
(1.35eV) serves to passivate and reduce surface leakage current in the narrow bandgap
(0.74eV) InGaAs. The n_-InGaAs is the photo-absorption layer. This layer is followed by
the filter or buffer layer, and then the substrate. The diffused p+ guard ring is on the
perimeter o f the device enclosing the active area completely, and slightly punching into
the photo-absorption layer. This guard ring serves as an effective means for collecting
electron-hole pairs generated in the margin and near the surface, and also for decreasing
the nominal diffusion lengths. Such an optimized structure also helps to reduce dark
current contribution from the margin o f the device and facilitate shorter minority carrier
lifetimes in the neutral regions. We have achieved this by improving on the quality o f the
material, tuning the doping level and adjusting the thickness o f the layer.
The shunt resistance Rsh o f this device (400M£2) is very high and can be neglected. It
may be included in complete high frequency circuit model o f the device to account for a
possible leakage current path (that is low shunt). The device is designed such that the
diffused guard ring is electrically isolated from the central junction, as no contact is made
to the floating guard ring. Therefore, any capacitance penalty coming from the guard ring
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
is negligible. The capacitance versus voltage performance o f the device with and without
guard ring will be presented in next section.
C//V A a m a tin ff
SiN passivation.
Au/Zn Contact-
«?i7V A fl m a tin g
■CWpa -n iva tin n
InP cap
rtu a rri PinQ ___
~ G u ard P in g
n InP buffer
InP substrata
A u S n sofrlpr m n ta rt
Figure 5-2 Cross-section o f the (75jun active area) guard ring p -i-n photodiode
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
S.3 Results and Discussion
The response waveform and speed o f the light output against the input current pulse
waveform is one o f the main concerns o f the choice o f the laser for the measurement o f
the ER o f a photodetector. The speed depends on the oscillation delay time and foil times.
We have employed a dc-coupled laser with a very fast recovery time in our test set-up. In
order to shorten the oscillation delay time, it is essential to bias the laser close to the
threshold. The idea is to bias the laser such that the characteristic nominal delay
accompanying the on-set o f
stimulated emission is effectively checked, and most
importantly to avoid additional extinction ratio penalty coming from the laser. We also
demand that the laser not be driven too hard to avoid excessive ringing and over-shoot, a
by-product o f stimulated emission [Tucker R.S. 1984 and Stephens T. et al, 1993]. The
transient tests were carried out in the large signal gain-switching regime. A multi-mode
fiber was used for wider illumination exposure to simulate the worst scenario for the
rigorous slow-tail test performed on the device. For the pulse generator, we used the
HP8116A as the source. The quality o f the electrical pulse was captured and examined on
the high-speed Tektronix Communication Analyzer CSA803A, before being used to
drive the laser. The CSA803A with an excellent resolution allowed us to detect and
estimate the extent o f the diffusion slow-tail current in our device. These tails tend to
increase the Inter-Symbol-interference (ISI) in digital systems by an amount which
depends upon the shape o f the tail and the type o f the line coding being used for
signaling. Tails which are just a few time slots long are the worst since they add the most
uncertainty to the neighboring pulse amplitudes at their sampling times. Tails which are
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
many tim e slots long are less important since they tend to add to and average with other
tails, thus adding dc level to the receiver signal which can be blocked by an ac-coupled
amplifier. Therefore, for a given area in tail, long low tails are much less important than
shorter higher amplitude ones.
The experimental data collected from thousands o f diodes are summarized in Table 5-1,
indicating that the guard ring device can deliver an ER of3IdB . To our knowledge, this is
the first report o f this type o f capability. Furthermore, our rolling-reliability tests confirm
a robust life test performance. We also observed that the accelerated aging test data
improves substantially, with a slight relaxation o f our specification on Rseries (ohmic
drop across the neutral semiconductor regions and contacts). And there was no penalty
on the ER. It is understandable that Rseries will have no effect whatsoever on the ER,
simply because the experiment was performed in the low-injection regime (high-injection
regime provides little additional physical insight into the device behavior). In this regime,
the minority carriers define the respond speed boundary conditions o f the p-i-n diode as
reiterated above because the minority carrier density is roughly equal to the injected
carrier density. The condition o f high-injection during which injected minority-carrier
density is comparable to the majority concentration is not applicable here. At highinjection, we must account for the effect associated with the finite resistivity o f the
quasi-neutral regions o f the junction. The Rseries >s then expected to absorb an
appreciable amount of the voltage drop between the diode terminals. Left unchecked, it
can ultimately lead to catastrophic failure. Fig. 5-3 Illustrates the typical response o f the
device to a 20ns pulse from which we estimated the ER (full pulse amplitude to the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
residual ta il ratio). The residual slow-tail diffusion current is barely visible. And Table 1
presents the data comparing the E R o f ETX75 (standard device) with ETX75G (guard
ring device). The anticipated input optical power penalty required to obtain certain Bit
Error Rate (BER) due to finite E R value is AP = IOlog[(ER + I)] / [(ER - I ) ] dB. This is
the amount o f power compensation required to avoid performance degradation. In the
experimental set-up, we varied the distance (and therefore the received optical power
levels) o f the fiber from the device to simulate a narrow and a wider device exposure to
optical illumination (where the angle o f exposure o r beam angle is a derivative o f NA —
Rseries ( ^ )
- (Standard device)
(Guard Ring device)
R e p ro du ced with permission o f the copyright owner. Further reproduction prohibited without permission.
a 'o
IOm / U *
Figure 5-3 Typical response o f a guard ring photodiode to a 20ns pulse (a) and the
corresponding residual slow tail (b)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Tim e (ns)
Time (ns)
Figure 5-4 Optical response under different power illumination w ith different
i l l u m in a t in g
areas (a) unguarded and (b) guarded
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the numerical aperture o f the fiber. In the limit when the distance between the fiber and
the device is zero, the diameter o f exposure at the device plane is exactly equal to the
core diameter o f the fiber). For the narrow beam cone we expect the cross-section o f the
far field beam pattern at the device plane to be approximately Gaussian and for the wider
beam to be slightly elliptical and therefore illuminate a wider area o f the device. We
should then expect the generation o f electron-hole pairs to be confined to the active highfield regions for narrow beam and low received optical power. Whereas the wider
elliptical (higher level o f received optical power) beam tends to generate additional
electron-hole pairs in the margin hence longer and higher tail (Fig.5-4(a)) in the
unguarded device. By contrast in the guarded device in Fig.5-4(b) under the same
illumination situation as in Fig.5-4(a), wider illumination with higher optical power does
not increase the slow tail. The result demonstrate the effectiveness o f guard ring in
collecting electron-hole pairs generated in the margin. The effect o f the guard ring
compared to an unguarded device at different power levels (-0.2dBm and -!3dBm) with
different illuminating areas is clearly demonstrated in Fig.5-4(b).
The capacitive penalty due to the guard ring has also been measured. This is
demonstrated in the capacitance versus reverse voltage and the corresponding frequency
response shown in Fig.5-5, for both the standard (unguarded) and guarded devices
respectively. The analog 3dB bandwidth for the standard device without guard ring is
2GHz, whereas with the guard ring the bandwidth shrunk to 1.7GHz due to the slight
increase in the capacitance. The slight increase o f the capacitance causes longer RC
response time and therefore slightly narrower bandwidth. The bandwidth penalty is
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
flm m Vollag*
F r e q m n c y (U H s)
fi* 1
K 4
R m n a Voltago
Frequency (MHz}
Figure 5-5 Capacitance versus reverse voltage and Bandwidth (a) unguarded (b)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The differential m inority carrier lifetime is obtained directly from the device
impedance measurement as a function o f frequency using the network analyzer and then
fitted to the total impedance obtained from RC circuit equivalent model o f the device.
From this, we estimate the minority-hole diffusion length w ithin the margin o f the device
to be 20pm. The minority carrier lifetime is effectively decreased by the use o f guard
The device is ideal for use in SCM-WDM and broadband networks. For these
applications the bandgap o f the photoabsorption layer can be tailored to detect
wavelength bands centered around either 1.3 o r 1.55pm thus acting as a bandpass
photodetector. There are two practical approaches for achieving this. The passband
selection can be realized w ith a dielectric coating on the window or lens cap o f the
package. This option is not easy in that there may be a problem with obtaining precision
coating uniformity and therefore loss o f transmission at the required bandpass. The
second option is material optimization and structural re-engineering. This will be
discussed briefly. For 1.55pm spectral passband the filter o r buffer layer in figure 5-2
must be replaced by a quartemary layer (e.g InGaAsP) lattice matched to InP but with a
bandgap o f 0.88eV or 33.85kT @ room temperature and the photodiode must be used in
the back-entrance configuration. That is a back-illuminated version o f the modified
photodiode in figure 5-2. In this mode photons with energy greater than 33.85kT will be
absorbed in the filter layer and those with energy less than 33.85kT will pass through to
be absorbed in the ternary layer (0.74eV). For 1.3pm SCM-WDM applications we must
use a quartemary photoabsorption layer (InGaAsP) [Adachi S., 1982] with a bandgap o f
~ 0 .8 8 eV and lattice matched to the InP (5.8688A; 1.35eV [Adachi S., 1987]) substrate.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To achieve highly linear optical response, a specification required in SCM-WDM
networks the detector structure in figure 5-2 must be optimized to reduce carrier (hole)
pile-up due to band discontinuity at the heterointerface. A speed-up or graded layer may
be used to achieve this. The free carrier levels (doping) and thickness o f the ternary, filter
and the InP cap layers must also be optimized. W ith these modifications a second order
intermodulation distortion as low as -80dB can be achieved at lmW (OdBm) optical
power and 0.7 modulation index in a 80 and 110 channel microwave matrix fiber optic
video transmission system using subcarrier tones at 135MHz and 190MHz respectively.
5.4 Conclusion
We have designed and manufactured a novel guard ring p-i-n photodetector, with an
Extinction R atio o f 31dB (with a 20nsec pulse). This ER meets the stringent requirement
o f the state o f the ait digital fiber communication system. To our knowledge, this is the
first report to document the use o f guard ring in p -i-n photodetector to reduce the slowtail diffusion current, without incurring excessively prohibitive additional penalty. The
ability o f the device to eliminate slow-tail diffusion current should make it ideal for
applications such as fiber-to-the-home with a passive optical network architecture
u tiliz in g
ATM burst dati» packets, where there is large variation from the distribution
point to the home, and there is need for the receiver to recover from a strong signal
quickly in order to recognize the small signal and subcarrier-WDM network which
requires a linear detector.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
Future Work - Mid Infra red (MIR) Photodetectors
Advances in optoelectronic devices usually rely on progress achieved in the optical
fiber fields. Since Mid Infra Red (MIR) photodetectors and semiconductor lasers are o f
great interest for spectroscopy, optical transmission through the atmosphere and
lightwave communications with repeaterless transmission over several hundreds o f
kilometers distance, significant research and development effort is being spent on MIR
fibers whose total losses in the 2-4 pm wavelength region are about 2 to 3 orders o f
magnitude below those o f conventional silica fibers (Tran, D.C, et al; 1984]. If progress
in their fabrication continues ultra-low loss fluoride fibers may find use in long distance
transmission in the 2-4pm wavelength region sometime in the future. Heavy metal oxide
glasses such as GeOz used for fiber Raman lasers and for soliton transmission, halide
crystal glasses - BeF2, Cadmium-based and Zirconium-based - and chalcogenide glasses
with low softening temperature; are some examples o f this material. A loss o f 1x10
dB/km is theoretically possible w ith fluoride glass fibers, even though present total loss
values are still not competitive compared to the 1.3 and l.Spm windows because o f
metallic impurities and excess scattering due to waveguide imperfections [Tran D.C. et
al, 1986; Kanamori et al; 1986]. Presently, short sections o f step- and graded-index fibers
made from KRS-5 TIBr and H I - polycrystalline material are used for thermal radiation
measurements and for the transmission o f CO* laser light.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Compound semiconductor alloys o f InxG ai.xAsySbi_y lattice-matched to GaSb
(6.0969A, 0.72eV [Adachi S., 1987]) (where y/x ~ 0.9) span the wavelength range from
1.7 to 4.3pm, and thus the InGaAsSb/GaSb [York, PJC , et al; 1995; and Lee H., 1995]
material systems appear ideally suited for devices to be used in this spectral region. Until
recently the alloy system GaxIn i.xAsySbi.y has received relatively little attention
because it has a very large miscibility gap with a relatively high critical temperature
(about 1470°C) [DeWinter J.C., et al; 1985 and Astles M ., et al; 1986]. Nevertheless
several research establishments have reported photodetectors based on this alloy system.
At SamofF (RCA Laboratories) for example highly efficient antimonide-based 1.9pm
resonant-cavity photodiodes have been fabricated. The resonant cavity enhances the
quantum efficiency at the resonant wavelength to nearly 100% [Samoff - Internal
Report]. The free-spectral range o f this type o f resonant cavity is defined as the frequency
spacing between two consecutive passbands and the finesse is the ratio o f the freespectral range to fiill-width half-maximum passband width. In this chapter we present a
design for a practical photodiode in the ~ 2 - 4pm spectral window.
We limit our discussion to only mesa structures, for high reliability (as mentioned in
chapter 3) however planar photodiodes are more desirable because mesa structures are
vulnerable to premature edge breakdown. The geometry is that o f a regular p-i-n
structure. The schematic o f the diode is illustrated in figure 6-1. It consists o f a Be
(Beryllium)-doped p+ GaSb window or cap (0.77eV) layer approximately 0.6pm thick,
an undoped n* -Gao.8 4 Ino.l6 Aso.14 Sbo.8 6 absorbing layer about 2.4pm thick (~ lxlO 15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
cm'3) and a 2 |xm thick Te (TeUurium)-doped n+ - GaSb filter o r buffer layer, epitaxially
grown and matched onto a (100) oriented Te-doped n+-GaSb substrate. To define the
mesa geometry, a
1 0 0 pm ( 0
) diameter etch, can be obtained by using photolithographic
techniques and a concentrated solution o f bromine/methanol or potassium/iodide. The
expected (inferential) current-voltage characteristic o f the photodiode at room
temperature is illustrated in figure 6 - 2 . The dark current should increase monotonically
with increasing bias and for a back-bias greater than about 5Volts we actually observed a
soft breakdown o f the type characterized by tunneling in devices with sim ilar structure
and material system. A closer study o f some blind diodes showed that at low voltages the
reverse current is dominated by generation-recombination mechanism in the space-charge
region via deep level traps. The measured ideality factors confirmed our theoretical
expectations. This is typical o f GaSb-based devices. Assuming
n* InGaAsSb
GaSb fn+1
GaSb (ri*)
Figure 6 -l(a ) Schematic o f a InGaAsSb/GaSb-based mesa back-illuminated photodiode
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
re v e rs e b fa s.V
Figure 6-2 A blind-diode dark-current-voltage characteristics
a band-to-band tunneling, our results suggests a tunneling process via deep level traps in
the forbidden band o f the photoabsorption layer.
Relative Spectral quantum efficiency curves were obtained using a monochromator
(with a white lamp as the light source o f known relative output-power spectrum) and
appropriate set o f filters (for these measurements typical dc absolute quantum efficiency
was obtained at 1.523pm by employing an HeNe laser). Typically quantum efficiencies
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f more than 56% without AR coating w ere reproducibly measured. The long wavelength
cutoff is defined by the absorption edge o f the InGaAsSb layer while absorption in the
GaSb cap is believed to be responsible for the short wavelength cutoff As for the dark
current we have routinely measured ~ 5-1 OpA at lOOmV. The primary sources o f this
dark current are surface leakage, generation-recombination, and tunneling. In the domain
o f small bias voltages where applied reverse bias is typically o f the order o f a few
hundreds o f millivolts, the dark current increases rapidly. In the second regime the
variation o f the current with the reverse voltage is relatively small till a bias voltage o f
about 4Volts. This is typical o f generation-recombination current. This current may be
coming from ionic charges located in surface states where they tend to create image
charges within the semiconductor thereby forming the so-called surface channels. Once a
channel is formed it modifies the p -n junction and gives rise to surface leakage current.
To suppress this noise it may be necessary to use extremely pure intrinsic
photoabsorption layer -
2 -6
cm if possible because impurities or, more generally
lattice defects in semiconductors tend to reduce the lifetime o f the electronic states that is
excitons and or free electron-hole pairs thereby changing their electrical and optical
The measured external quantum efficiency without antireflection is uniform at
approximately 56.5% from ~ lpm to ~ 2.4pm. Much improvement in the Q.E can be
obtained by using appropriate antireflection coating, back-illuminated geometry so that
one can use the p+-contact as a reflector to re-circulate the photons. Back-illuminated
devices should result in fast response time, since the structure can be configured for small
active areas and therefore less capacitance. If mounted p-side down it can also help
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
eliminate extraneous parasitic capacitance from a contact pad. A smaller area device in
conjunction with a simple stripline mounting circuit w ill be capable o f detecting light
pulses at a rate greater than 4Gbits/sec. The finished chip may be mounted on a 50Q
stripline circuit with the bias circuit designed to achieve minimum ringing. The stripline
circuit despite poor low frequency response and DC bias return has proved to be useful
for high-speed measurements [Auston D.H., 1983].
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S. Adachi, “Refractive indices o flll-V compounds: key properties
o f InGaAsP relevant to device design,” J. Appl. Phys. 53, pp.
S. Adachi, “Band gaps and refractive indices o f AIGaAsSb,
GalnAsSb and InPAsSb: key properties for a variety o f the 2-4-pm
Optoelectronic device applications,” J. Appl. Phys. 61(10), pp.
4869-4876, May 1986.
T. Adachi, A. Yoshi, and T. Sudo, ‘Two-dimensional
semiconductor analysis using fimte-element method,” IEEE Trans.
Electron D evices, vol. ED-26, pp. 1026-1031, July 1979.
H. L. Armstrong, E. D. Metz and L Weiman, “Design theory and
experiment for abrupt hemispherical p-n junction diodes,” IRE
Trans., voL ED-3, pp. 86-92, April, 1956.
[Astles8 6 ]
M. Astles, H. Hill, A. J. Williams, P. J. Wright and M. L. Young,
“Studies o f the Gai.xIn)cAsi.ySby Quarteraary alloy system I.
Liquid Phase Epitaxial Growth and Assessment,” J. electron.
M ater. 15, pp. 41-49,1986.
D. H. Auston, “Impulse response o f photoconductors in
transmission lines,” /. Quant. Electron., vol. QE-19, no.4, April
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
J. E. Bowers, C. A. Bum is, and R. J. McCoy, “InGaAs p-i-n
photodetectors with modulation response to millimeter
wavelengths,” Electron. L ett., vol. 21, pp. 812-814,1985.
[Brackett8 8 ]
C. A. Brackett, "Dense WDM Networks, “Proc. ECOC’88, pp.
533-540, Sep. 1988.
K.-P. Brieger, W. Gerlach and J. Pelka, “Blocking capability o f
planar devices w ith field limiting rings,” Solid State Electron.
vol. 26, no. 8 , pp. 739-745,1983.
[Campbell8 8 ]
J. C. Campbell, W. T. Tsang, G. J. Qua and B. C. Johnson, “High­
speed InP/InGaAsP/InGaAs avalanche photodiodes grown by
chemical beam epitaxy,” IE E E J. Quantum Electron., vol. 24, pp.
F. Capasso, W. T. Tsang, A. L. Hutchinson, and G. F. Williams,
“Enhancement o f electron impact ionization in a superlattice: A
new avalanche photodiode with a large ionization rate ratio,” Appl.
Phys. Lett., vol. 14, pp. 1188-1191,1981.
F. Capasso, “Physics o f avalanche photodiodes,” in Semiconductor
and Semimetals, vol. 22D: Lightwave Communications
Technology Orlando Florida: Academic Press,
F. Capasso and G. Margaritondo, Eds., “Heterojunction band
discontinuities,” North H olland Physics Publishing, Amsterdam,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
B. A. Carre, “The determination o f optimum accelerating factor for
successive over-relaxation,” Compu. J ., vol. 4, pp. 73-78,1961.
K. Cheung, M. M. Choy, and H. Kobrinski, “Electronic
wavelength tuning using acousto-optic tunable filter with broad
continuous tuning range and narrow channel spacing,” IEEE
Photon. Tech. L e tt, voL 1, no. 2, pp. 38-40, Feb. 1989.
G. C. Chi, e t al “Planar InP/InGaAsP 3-D graded-junction
avalanche photodiode,” IEEE Trans, on Electron Devices, vol.
ED-34, pp. 2265-2269,1987.
M. M. Choy, F. F. Tong, and T. Odubanjo, “A 200Mb/s FSK
subcarrier optical link potentially extendable to 640 channel
WDM-SCM network,” Proc. C LEO '9I, pp. 468-469,1991.
L. W. Cook, G. E. Bulman, and G. E. Stillman, “Electron and hole
impact ionization coefficients in InP by photomultiplication
measurements,” Appl. Phys. Lett., voL 40, pp. 589-591, 1982.
T. E. Darcie, “Subcarrier multiplexing for multiple-access
lightwave networks,” J. Lightwave Technol., vol. LT-S, no. 8 , pp.
1103-1110, Aug. 1987.
W. B. Davenport and W. L. Root, An introduction to the theory o f
Raman signals and noise. McGraw-Hill, 1958.
N. G. de Bruijn and T. Van Arrdenne-Ehrenfest, “Circuits and
trees in oriented graphs,” Simon Stevin, vol. 28, pp. 203-217,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
M . Dentan and B. de Cremoux, “Numerical simulation o f the
nonlinear response o f a p-i-n photodiode under high-illumination,”
J. Lightwave Technol., voL 8, pp. 1137,1990.
[Desem8 8 ]
C. Desem, “Optical interference in subcarrier multiplexed systems
with multiple optical carriers,” J. SAC., vol. 8 , no.7, pp. 12901295,1990. Electron. L ett., voL 24, pp. 50-52,1988.
J. C. DeWinter, M. A. Pollack, A. K. Srivastava, J. L. Zyskind,
“Liquid phase epitaxial Gai.xInxAsySbi-y lattice-matched to (100)
GaSb over the 1.71 to 2.23-pm wavelength range,” J. Electron.
M ater. 14, pp. 729-747,1985
A. F. Franz, G. A. Franz, S. Selberherr, C. Ringhofer, and
P. Markowich “Finite boxes - A generalization o f the finitedifference method suitable for semiconductor device simulation,”
IEEE Trans. Electron. D evices, vol. ED-26, pp. 1026-1031 July
G. George and J. P. Krusius, ‘Transient response o f p-i-n diodes
including diffusion effects,” State Electron., vol. 37, no. 11, pp.
S. K. Ghandhi, Semiconductor Power Devices. New York: W iley,
G. Ghartrand, and L. Lesniak, Graphs and Digraphs. Wadsworth
and Brooks/Cole, 1979.
J. L. Gimlett, “Low-noise p-i-n/HEMT optical receivers with
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
bandwidth greater than 16 GHz,” ./, cfLightw ave Technol., pp.
1432-1437, O c t 1989.
A. S. Grove, Physics and Technology o f semiconductor devices,
New York: W iley, 1967.
R. R. Hayes and D. L. Persechini, “Nonlinearity o f p-i-n
photodetectors,” IKKK Photon. Teclmol. Lett., vol. 5, pp.70,
D. A. Humphreys, R. J. King, D. Jenkins, and A. J. Moseley,
“Measurement o f absorption coefficients o f Gao.47Ino.53As over the
wavelength range o f 1.0-1.7 pm,” Electron. L ett., vol. 21, pp.
[Humphreys8 8 ]
D. A. Humphreys and R. A. Lobbett, ‘Investigations o f an
optoelectronic nonlinear effect in a InGaAs photodiode, and its
application in a coherent optical communication system,” Proc.
IE E , vol. 135, no. 1, pp. Feb., 1988.
[Kaminow 8 8 ]
L P. Kaminow, “Photonic local networks,” in O ptical Fiber
Telecommunications n , S. E M iller and I. P. Kaminow, Eds.
New York, p. 933,1988.
[Kanamori8 6 ]
T. Kanamori and S. Sakagushi, “Preparation o f elevated NA
fluoride optical fibers,” Japan J. Appl. Phys., 25, pp. L468-L470,
L. V. Keldysh, “Behavior o f non-metallic crystals in strong
electric fields,” Soviet Phys. JE W ., vol. 6(33), no. 4, April 1958.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
L. V. Keldysh, “The effect o f a strong electric field on the optical
properties o f insulating crystals,” Soviet Phys. JETP., vol. 34(7),
no. 5, Nov. 1958.
H. Lee, P. K. York, R. J. Menna, R. U. Martinelli, D. Z.
Garbuzov, S. Y. Narayan, and J. C. Connolly,” Room-temperature
2.78-pm AlGaAsSb/InGaAsSb quantum-well lasers,” Appl. Phys.
L ett., 66,1942-1944,1995.
T. P. Lee and S. M. Sze, “Depletion layer capacitance of
cylindrical and spherical p-n junctions,” Solid State Electron., 10,
p. 1105,1967.
G. Lucovsky, R. F. Schwartz, and R. B. Emmons, ‘Transit-time
considerations in p-i-n diodes,” J. o f Appl. Phys., vol. 35, no. 3,
March 1964.
[Machida8 8 ]
H. Machida, M. Kobayashi, T. Shirai, Y. Kishi, and T. Kaneda,
“Reach-through avalanche photodiode for gigabit-rate systems,”
OFC ’88 Tech. Digest, paper T uC 4,1988.
E. E. Mendez, W. I. Wang, B. Ricco, and L. Esaki, “Resonant
tunneling o f holes in AlAs-GaAs-AlAs heterostructures, “Appl..
Phys. L ett., vol. 47(4), pp. 415-417,1985.
[Mclntyre6 6 ]
R. J. McIntyre, “Multiplication noise in uniform avalanche diodes,”
IEEE Trans. Electron Devices, vol. ED-13, pp. 164-168,1966.
N. A. Morris, J. C. Connolly, R. U. Martinelli, J. H. Abeles, D. E.
Cooper, C. B. Carlisle, H. Riris, and A. L. Cook, “Single-mode
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
distributed-feedback 761nm GaAs-AlGaAs quantum-well laser,”
IEEE Photon. Technol., 5, pp. 455-457,1995.
M Nazarathy, W. V. Sorin, D. h i Baney, and S. A. Newton,
“Spectral analysis o f optical mixing measurements,” J. Lightwave
Technol., vol. 7, no. 7, pp. 1083-1096,1989.
K. Nishida, K. Taguchi, and Y. Matsumoto, “InGaAsP heterostructure avalanche photodiodes with high avalanche gain,” Appl.
Phys. Lett., vol. 35, pp. 251-252,1979.
B. Odubanjo, C.S. Wang, & W. L Wang, “An electrically isolated
guard ring avalanche photodetector,” Proc. SPIE, vol. 2 9 9 4 ,1997.
B. Odubanjo, C.S. Wang, & W. I. Wang, “High performance guard
ring p-i-n photodiode for digital fiber optic commmunications,”
Proc. SPIE, voL 2999, pp. 414-423,1997.
T. Okuda, “DFB laser intermodulation distortion analysis taking
longitudinal electric field distribution into account,” IEEE Photon.
Technol, Lett., vol. 6 , p .2 7 ,1994.
J. M. Ortega, W. C. Rheinboldt, Iterative solution o f non-linear
equations in several variables. New York: Academic Press, pp. 214230,1970.
A. Papoulis, Probability, random variables, and stochastic
processes, M cGraw-Hill, second edition 1984.
S. O. Rice, “Mathematical analysis o f random noise,” B ell Syst.
Tech. J., vol. 23, pp. 232-332,1944.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
J. H. Schafiner and W. B. Bridges, “Intermodulation distortion in
high dynamic range microwave fiber-optic links with linearized
modulators,” J. o f Lightwave TechnoL L ett., vol. 6 , p. 27,1994.
S. Selberherr, Analysis and simulation o f semiconductor devcies,
Wein. Springer 1984.
T. Stephens and K. Hinton, “Specifying adiabatic lasers for
2.5 Gbits/sec, high dispersion, IM/DD optical systems,” Electron.
L ett., vol. 129, pp. 1479-1480,1993.
[Sudo8 8 ]
H. Sudo and Suzuki, “Surface degradation mechanism o f InP/
InGaAs APDs,” J. o f Lightwave Tech., vol. 6 , pp. 14961501, 1988.
[Sze6 6 ]
S. M. Sze and G. Gibbons, “Effect o f junction curvature on
breakdown voltages in semiconductors,” Solid State Electron., 9,
p. 831,1966.
S. M. Sze, Physics o f semiconductor devices, New York: Wiley,
[Taguchi8 6 ]
K. Taguchi, T. Torikai, T. Sugimoto, K. Makita, and Ishihara,
“Temperature dependence o f impact ionization coefficients in
InP,” J. Appl. Phys., voL 59, pp. 476-481, 1986.
D. C. Tran, G. H. Sigel and B. Bendow: “Heavy metal fluoride
glasses and fibres: A review,” J. Lightwave Technol., LT-2, pp.
[Tran8 6 ]
D. C. Tran, K. H. Levin, M. J. Burk, C. F. Fisher and D. Bower,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
“Preparation o f properties o f high-quality IR transmitting glasses
and fibers based on metal flourides,” Proc. SPIE 618,48,1986.
K. Vahala and A. Yariv, “Semiclassical theory o f noise in
semiconductor lasers - Part I,” IEEE J. Quantum Electron., vol.
QE-19, pp. 1096-1101, June, 1983.
K. Vahala and A. Yariv, “Semiclassical theory o f noise in
semiconductor lasers - Part n,” IEEE J Quantum Electron., vol.
QE-19, pp. 1102-1109, June, 1983.
D. Wake, “1550-nm millimeter photodetector with bandwidthefficiency product o f 2.4 THz, comprising monolithically
integrated near-travelling wave optical pre-amplifier and
waveguide p-i-n photodiode,” J. Technol., pp. 908-912,
July 1992.
D. Wake, N. G. Walker, and I. C. Smith, “Zero-bias edge coupled
InGaAs photodiodes in millimeter wave radio fiber systems,”
Electron. Lett., vol. 29, no. 21, pp. 1879-1881,1993.
[Webb8 8 ]
P. Webb, R. J. McIntyre, J. Scheibling, and M. Holunga, “Planar
InGaAs/InP avalanche photodiode fabrication using vapor-phase
epitaxy and silicon implantation techniques,” Proc. Opt, Fiber
Comm. Conf. Tech. Series, vol. 1, pp. 129-130,1988.
T. H. Windhom, L. W. cook, and G. E. Stillman, “Temperature
dependent electron velocity-field characteristics for InGaAs at
high electric fields,” J. o f Electron. M ater., vol. 11, no. 6 , pp.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
P. K. York, R. J. Menna, D Z Garbuzov, H. Lee, R. U. Maitmelli,
and S. Y. Narayan, 2-(im GaSb-based distributed Bragg reflector
lasers,” CLEO post deadline paper 47,25 May 1995.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Без категории
Размер файла
3 868 Кб
Пожаловаться на содержимое документа