Silicon Based Metal-Semiconductor-Metal

Photodetectors

A thesis submitted to the Faculty of Graduate Studies and Research In partial fulfillment of the requirements

For the degree of Master of Engineering In Electrical Engineering

Ottawa-Carleton Institute of Electrical Engineering Department of Electronics

carleton*~n.iversity Faculty of Engineering -

Ottawa, Ontario, Canada

May 2001 O Amanda DeVnes, 2001

uisitions and Acquisitions et Bi mgraphii Services selvices bibibgraphiques 9.

The author has granted a non- exclusive licence aüowing the National Li'brary of Canada to reproduce, loan, d i s û i i e or sel copies of this thesis in microform, paper or electronic formats.

The author retains ownership of the copyright in this thesis. Neither the thesis nor substantid extracts fiom it may be printed or otherwise reproduced without the author's permissicm.

L'auteur a accordé une licence non exclusive permettant à la Bibliothèque nationale du Canada de reprodiiire, prêter, distribuer ou vendre des copies de cette thèse sous la fonne de microfiche/fûm, de reproduction sur papier ou sur format électronique.

L'auteur conserve la propnbté du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de ceîie-ci ne doivent être imprimds ou autrement reproduits sans son autorisation.

With the possibility of Mly optical telecommunication networks now within industry's

grasp, optical designers face numemus challenges as they attempt to intepte optical and

electronic bct ions togethet while continually increasing capacity and speed

This thesis is comprised of two novel projects. An attempt was made to fàbricate, test,

and chanicterize a polysilicon-germanium metal-semiconductor-metal (MSM) 155011x1

detector for long-haul optical communications. It was eventuaîly determineci that the

high-quaiity germanium film required could not be fabricated given the limited resources

at the Carleton University laboratory. Subsequently, a polysilicon MSM detector was

fabricated, tested, and characterized at 8 S h light for use in short-haul fiber

communications and high bandwidth optical interconnects. Electron beam Iithography,

used for the 6rst thne in the Carleton Fabrication Lab, allowed the possibility of

submicron dimensions, resulting in very small, fast devices. The detectors exhibited fast

responses (1 .SGHz), but large dark cunents (40nA over 750 puZ).

1 would like to express my thanks to Siegfried Janz of NRC, whose lab 1 used at the

Institute for Microstructural Sciences to perfom ail of my photonic measurements. As

weli, the assistance of Dan-Xia Xu and Peter Grant was invaluable.

There are a few people who gave me tips and hints dong the way, and 1 greatly

appreciate their help and advice. They are: Chris Pawlowicz (now at Norte1 Networks),

N i d Tait (Carleton University), and Yan Feng (NRC).

1 am very thaniâul to Technical officer Lyall Bemdt and technician Carol Adams, who

helped and supported al1 my work at the fabrication lab.

For the people who were always ready to offer non-technical heip and guidance: my

parents and Chris, my husband.

Final thanks goes to rny supervisor Gany Tarr and bis optimism in spite of my doubts.

................. 1 . h a u c f l o n ..n.""."Uu".....u"..- ".3rn""*.UrnU.......".U*".U.* a--... .- 1

Modem Optical Fiber Communications ...................................................................... 2

..................................... The Role of MSM àetcctors in Modern Fiber Optic Systems 3

...................................................................................................... Pmblern Definition 4

........................................................................................... 1.55 pm Detection 4

Optuelectronic integration ............................................................................. 6 Thesis Contributions ................................................................................................. 8

PolyGermanium Photodiode .......................................................................... 8

........................................ ûptimizatioa of an Existing PolySilicon Photodi& 9

Recent Work in Silicon-Based 1.55 fhotodetection ................. ., ............... 9

2 . Det~fOt Delgn M.., "m." ...c.*w..n-a-.mw--nrrmwm- m m r i i u 13

Wavelength ............................................................................................................... 13

....................................................................................................... Attenuation 13

........................................................................................................ Dispersion 14

StnicnKe ..................................................................................................................... 15

PN Junction .................................................................................................. 16

PIN Diode ......................................................................................................... 16

Avalanche Diode .......................................................................................... 17

................................... Schottky Banier/Hetmju~~ction Diode ..................... ... 17

MSM Diode ...................................................................................................... 17

Material ...................................................................................................................... 19

Germanium Content ......................................................................................... 19

.....*................... .............*...........*....... Straincd versus Relaxed ................ ,, ; 20

Monocrystalline vcrsus Polycrystalline Matenal ............................................ 21

Conclusionç ................................................................................................................ 23

3 . SC~O~~~C@ISM T h e ~ v - . . - ~ - ~ . . - . - . . m n u . u u u . r n . u ~ m . - . - . . ~ - - ~ . - . - 24

Schottky Barrier Diodes ............................................................................................ 24

image Force Effect ........................................................................................... 26

Schottky Barrier Diodes as Photodetectors ...................................................... 27

..................................................................................................... MSM Operation 2 8

Dark C u m t. .................................................................................................... 32

Noise ....................................................................................................... 3 3

.................................................................................... Quantum Efficiency 3 3 . . Responsimty ..................................................................................................... 35

Bandwidth ....................................................................................................... 35

Transit Time ..................................................................................................... 35

............................................................................................ RC Time Constant 35

4 . Experiment: PolyCermrnium MSM Photodetector ...., .. ......... .. ...... .. .... 3 7

. 5 Experiment: P o l y S i n MSM Photodetector ..., ....... ... ........... .......... ....... ....-. . 44

Performance Evaluation of the Polydicon Detector ................................................. 54

B a n k Height ................................................................................................... 54

Dark Current .................................................................................................. 55

Responsivity/External Quantum Efficiency ..................................................... 57

.................................................................................................. Capacitance 6 0

Frequency Response ......................................................................................... 62

6. Conclusions and Rec~rnmendrtions.,.~ . . n . . ~ ~ ~ . . o ~ ~ . e ~ a o l l e l L ~ ~ l ~ e . e . ~ ~ ~ . e . . . e . . 68

. 7 Appeadix A: Summrry of Attempted SiGe Experiments ................. .. ..... .. .... ...... 70

8 . Appendir B: Ut of Instruments used for McrsurementJ . ...,,.L.LWUe-- .. 71

............................................................................................ DC Meamrements 71

High Speed Measurements ............................................................................... 71

. 9 Citition~......-....~--.-...-~~..~ ..... .."..... ................... ............ .... - ..... .... ..... ....... ... . 72

Figure 1: Absorption Coefficients ...................................................................................... 5

Figure 2: Loss Spectnun of Fiber Qtic Glas ............................................................ 14

Figure 3: Dispersive Behaviour of Fiber Optic Glass ....................................................... 15

Figure 4: An MSM Photodiode .............. ........................................................................ 18

Figure 5: Energy Bandgap versus Germanium content ................................................. 19

Figure 6: Criticai film thiclcness vernis Germanium content .......................................... 20

..................... Figure 7: Absorption Coefficient of Amorphous versus Crystalline Ge. Si 22

............................... Figure 8: Energy Band Diagram ofan M-S junction in Equiiiirium 24

............................ Figure 9: Energy Band Diagram of an MSM Detector in Equiliirium 28

Figure 10: ~nergy Bands and Electric Field Distribution at (a) V=Vw and (b) V=VFe .. 3 1

Figure 11: Typical Metal Liftoff Process .......................................................................... 39

...... Figue 12: Images of (a) PolySilicon and (ô) PolyGennanium sarnples ( b e l pm) 41

Figure 13: Schottky characteristic using RTCVD polysilicon ........................................ 41

Figure 14: Surface Roughness versus Substrate Temperature ...................................... 42

Figure 15: 0.5 jm MSM pattern ....................................................................................... 44

Figure 16: Enhanced Metal Lifioff Process ................................................................. 4 6

........................... Figure 17: Noise Effects h m EM interference (6500~ maguification) 49

................... Figure 18: Step Coverage for Angied versus Perpendicular Evaporant Flux 52

......................................................................................... Figure 19: MSM Dark Current 56

................................................................. Figure 20: Photoresponse Measu~ement Setup 58

. . Figure 21: MSM Responsimy ............................................................ ........................... 58

vii

Figure 22: Impulse Response Measurement Setup ........................................................... 63

Figure 23: Temporai Respbnse ......................................................................................... 63

Figure 24: Temporal Response at 4 V .............................................................................. 64

Figure 25: Exponmtial Fitting to Impulse Response Data ............................................... 65

Figure 26: FFT of Temporal Response ............................................................................ 67

Table 1: Schottky barrier height of metaVmonocrystalline siiicon contacts ..................... 23

The use of light for telecommunications is a tapidly changing, very profitable industry.

With more &ta now sent dong communication networks than telephone conversations,

the telecommunications industry has bew hard pressed to meet consumtrs' bandwïdth

and processing speed demands. It is expected tbat the 'holy graii of the

telecommunications industry - the dl-optical global network - will be a reality by the

year 2015, with speeds in the terahertz range and transmission distances over 10,000 km.'

in order to realize this exponential growth and satisfy the increasing demand, focus must

be turned to the system components that comprise a communications network. The most

promising advances to &te involve the impressive evolution of opticai ampiifiers, optical

cross-connects, and integrated opticai components.

integrating the various stages of optical ûansmitters and teceivers on electronic chips

eliminates many speed limitations inherent in metal intercomect networks. Monolithic

integration also promises higher sensitivity, compact devices and reiiable performance2.

New functions can be realized and the cost per compent drops dramatically when

several discrete devices are brought together on a single mbstrate. The ultimate goal of

optical integration is expected to achieve the kind of success thatelectronic VLSI

technology has enjoyed over the last 20 yearsBtS'

htegration dso introduces new design challenges since the opticai layers typicaliy

require very high quaiity, conmileci composition nIms while the electronic devices and

- - - - - - - - - - -

interconnects introduce impedance matching issues. The problem is to reconcile these

concms into a solution that remaius cost effective.

This thesis attempts to reconcile one of the many issues faced by the telecommunications

industry by introducing a photodetector that is highiy integrable and economical to

produce.

Modem Opticai Fiber Communications

Current data rates transmitted over f i e r optic systems are in the range of OC-3 (1 55.52

Mbps), OC-12 (622.08 Mbps), and OC-48 (2488.32 Mbps) data rates. nie next

generation of optical network components at OC-192 (9953 -28 Mbps) is already

cornmercially available, with OC-768 in the foreseeable future4'. Thus the lightwave

telecommunications industry is gaining momentum at breakneck speed.

The digital modulation schemes used for data transpon cmently involve a conversion

hm optical to electrîcal - or microwave - formats (and vice versa) at the transmitter and

receiver ends. The M y optical networlcs envisioned for the near future must perfonn the

modulation, guiding, amplification, and demodulation at the optical level. Optical

networicing technology has been pushed even further with the birth of WDM (or

wavelength division multiplexing). By sending multiple channels 1i.e. wavelengths)

simultaneously over the same fiber, the total mer capacity has dramaticaiiy increased

over a very short period of rime.

In WDM, each access node in the network is assigneci a specifk wavelength value, In

order to esîablish a connetion to a node, the transmitter seiects the appropriate receiver

node wavelength to send its opticd signai. Several optical s i a s , eachtransmitting

using a different wavelength of light, can be multiplexed together and sent over the

opticai network. Thus dynamic WDM systems require wavelmgch-tunable transmitters

and receivers, which are currentiy in the early stages of development.

Using WDM with wavelengths in the band of 13-1.6 pm (that is, the band of lowest loss

in fiber optic glas) and assuming the selected node can support gigabit per second

electrical processing, the tiieoretical agpgate date rate of an optical communication link

could thus appmach the tesahertz range6.

The Rofe of MSM detecton in Modern Fiber Optic Systems

The inkgration of opticd devices with electronic structures has presented many

challenges to the telecommunications industry. Opticai devices are nonnaliy vertical

stnicnires, whiie field effect transistors, often used in the amplification stage of optical

receiven, are planar. Many techniques have been surveyed in order to successfùîiy

integrate the two inherently incompatiile devices, including bridge interconnection and

d a c e poüshing for planarization'.

Lateral MSM detectors have been proven to be fast, reliable detectors that are extcemely

easy to fabncate and are intrinsicaiiy compatiile with planar FE%. For these reasons,

there have been many reports of MSM detectors integrated into photoreceivec systerns.

For example, Harder et al pubiished the performance of a monolithic MSM, a

transimpedance amplifier, and an output buffer using a GaAs MESFET process8. ïhe

Eibrication process used the same b& Iayer of the FET for the detector's active layer

and the electrodes h m its gate metal.

The greatest disadvantage in employing MSMs in photoreceivers is their large dark

current. As discussed in this report, the magnitude of dark curent has a large

dependence on the Schottky barrier height and it is often necessary to take steps to reduce

this effecP O.

Problem Definition

This sîudy begins by identifjing the trends in data communications and attempting to

determine the needs of fiiture communication networks. Once these needs have been

recognized, it will be demwstrated that the two optical detectors presented in this paper

are highly feasble, economical solutions.

1.55 pm Detecrion

Fiber optic cable, normally constnieted fiom fused germano-silicate glass, is

dernonstrated to have its lowest dispersion when propagaîing Light at 1.3 pm and lowest

losses at 1.55 pm, as wiil be discussed further. For this reason, 1.55 pm is the wavelength

of choice in long distance fiber optic commuaication systems9.

An ideal, fully opticai network would nm entitely at this wavelength, however, cunent

technology presents some limitations. At the receiver end, semiconductor detectors use

the principle of light absorption and dsequent gcneration of electron-hole pairs to

convert the optical energy signal into an electrical one that c m be amplineci and

intetpreted as datai0. The emgy bandgap of the materiai is inversely proportional to the

minimum wavelength that c m be detected. Silicon, because of its relatively large

bandgap, can at most detect light at amund 1 pmlO, rendering it unsuitable for the

detection of long haul sigaals. The absorption coefficient of silicon and other cornmon

semiconductors is shown in Figure 1. As a resuit, present &y detectors that have the

correct bandgap are made h m alioys of Indium Gallium Arsenide (InGaAs). On an

integrated receiver, the detectors interface with the pre-amplifier and intercomect stages,

which are made with silicon technologies almost 9% of the theL0. Thus there are

compatlibility issues between the two materiai systems when integrating complex

circuitry. A h , processes that use these materials are not as mature as silicon

technologies, do not have a high qualitty native oxiàe like silicon, and are thus costlyLL.

Figure 1: Absorpîion ~ocfficienb"

%me of the work in this am bas focussed on devices made with vaious strained alloys

of silicon and germanium (SiGe). By vaxying the hction of germanium in a silicon layer,

one can tune the range of wavelength detection fiom tbat of pure silicon (1 pm) to pure

germanium (>2 p)". Thus with the appropriate hction of germanium, it is possible td

target the desired 1.55 prn mark. Germanium bas been shown to be integrable with

existing silicon processes, and to introduce dramatic improvement to the performance of

semiconductor devices. For example, its incorporation into the base of bipolar transistors

increased the device's speedI2. More recentîy, SiGe detectors have been demonstrated in

integrated optical msmitter and receivers for 1550 nm transmission operating at 20

G ~ J s ' ~ .

Given these issues and cowtraints, the initial aim of this thesis was to design, fabricate,

and test a novel, cost-effective, VLSI compati'ble SiGe device with good detection

behaviour at 1550 m.

Optoelec~onic Integrution

Another problem faced by tbe optical communications industry is that of integrauon. As

outIined eariier, there are numemus benefits in combiaiag botb electncal anà optical

functioas onto a smgle chip. Until communication networks are M y optical at ail Ievels,

there wül atways ex& the drive to integrate the optical and electronic functions as much

as possible. Speciticaily in the case of photoreceiver circuits, inkgrathg the detector with

the amplifier's eiecttonics both teduces packaging costs and eIiminates the parasitic

inductance of the bond wire. Accompanied by these bene& are severai dficuities that

7

anse when attempting to inkgrate very different material systems and structures. O h

when âeveloping a satishctory process for inkgmihg optical waveguides, filters and

detectors, for example, the individual performance of each component is compromised.

Thus the search for simple structures and common materials mains in the foregmmd in

the task of optoelectronic integration.

An immediate application that wouid benefit fiom optical-elemonic integration is optical

interconnection at chip-to-chip or board-to-board levels. The potentia! baudwidth of

optical interconnects fat exceeds that of current electrical connections. With the rapid

improvements made to processor speeds, computing networks now h d their bottleneck

at the interco~cct levet, where the metallic connections made between VLSI chips suffer

îiom power consumption delays, crosstalk, and aie ümited in density by the total

capacitance which affects propagation speed. Optical interconnects eliminate these

firadamentai problems, but introduce other factors including alignment issues, low

conversion efficiency of transmitters and receivers, and Iower switching speeds.

However, rhese problems are not fundamentai limitations and can eventuaily be

overcome as optical technologies matureL4.

The concept of opticai interconnects as applied to both paraIIel-processing in digital

compuring systems and optical backpianes in telecommunication switches are currently

areas of active resea~ch*~*'~.

At the optical interco~ect levei, the distances involved are very short and thus issues

such as loss and dispersion are not as signiscant as in the case of long haul

telecommunications. Instead, cost and case of fhbrication become important measutes

with which to determine the suitability of a particular detector. An economicai, diable

choice is to use wavelengths correspondhg to the strong absorption region of silicon. It

is logical to employ silicon as the photoactive material in an integrated photoreceiver

since the receiver elecüonics will aiready be silicon based.

Given the issues introduced by inkgration, the secondary aim of this thesis was to design,

fabricate, and test a silicon device with good detection behaviour at 850 nm to be used in

optical interconnect applications. It was expected that the photodetector would operate at

higher speeds when compared to earlier devices made with similar structural and material

charactenstics.

Tbesis Contributions

PoiyGennanium Photodiode

This section documents the attempt to fabricate a SiGe-baseci photodiode with good

responsivity at 1550 mu that can be easily introduced into current processing

technologies. It begins with aa analysis iato appropriate detector structure and material

composition and is followed by a description of the fabrication process aud test

procedures. Conclusions about the feasibility ofproducing such devices are drawn.

To the author's knowledge, this is the 6rst use of polygerrnanium as the material in the

active layer of an MSM photodetector. 'The ability of silicon-germanium alloys to absorb

1550 nm light holds promise for highly integrable, manufacturable devices. As will be

- - -- - - - - - -

discussed in detail later in this report, using a polycrystalline material offers a high

absorption coefficient, a thin active layer, and therefore a fast device.

ûptimùation of an Exlating PoiySiIicon Photodiode

This section reports the optimization of an existing polysilicon detector whose successful

performance at larger dimensions bas already been rep~rted'~. The original device

demonstrated good performance at 800-860 nm wavelengths for use in short haul

communication links and high bandwidtb optical interconnects.

The optimization of the device involves attempts to reduce its dimensions, thereby

increasing response speed (smaller distance for carriers to travel between electrodes). The

process to redesign, fabricate, and measure the detector's performance is presented. A

mbmicron lithographie process novel to Carleton University's Fabrication Lab is also

outlined. Final results include recommendations for M e r optimimtion.

Recent Work in Silicon-Bmed 1.55 pm Photodetection

To the author's knowledge, an MSM photodeteetor made fiom relaxed polygermanium

has not beea reporteci to date. However, there have been a number of simiiar structures

made and the results of these endeavours are summarized here.

Jager and Kosak performed the earliest experiments with Schottky barrier contacts to

germanium. Their work included metal evaporation on germanium heated to 300 O C .

Those who continueci their work (Thanailakis and Northrop; Trwer and Rhodenck)

tended to have confiicting data with regards to finai barrier heighd8. niis observation

has been confnmed in more recent snidies20. (ne obsemtion made by Tumer and

Rhoderick suggested that contacts to alimwuum age swner than other metals, possibly

because aluminium oxidizes easily with the G a interfacial film.

In April of 1993, Xiao et al reported the fabrication of Schottky banier diodes for

infrared detection using Pt and Pd silicides and strained SiGe laYersL9. in order to combat

problems with misfit dislocations created by growing SiGe fih on Si substrates, the

SiGe was graded. Thus, the Ge fiaction was gradually increased over a region before

depositing the desired Ge content. The entire film was grown in an RTCVD system

between 600-700 O C . From earlier work, Xiao and his colleagues had considered the fact

that platinum reacts preferentially to the Silicon in a SiGe film, leading to Ge segregation.

The isolated Ge tends to form three-dimensional islands, leading to surface roughness. To

combat this, a silicon-capping layer (-40-100 A) was deposited on the surface of the film

that is ideaily completely consumed during the silicidation step. The desired effect is to

lave the SiGe film intact. The Schottky barrier detector was able to detect wavelengths

greater than 8 Pm, with maximum responsivity of 0.1 A/W at 2.5 m. In May 1994, the

same group reported successfui work on Schottky b d e r diodes using SiGe *s for the

purpose of decreasïng bsrrier heights and extending the wavslength detection rangeto.

Work by Jiang et al in Febniary of 1996 c o h e d the need for a silicon-capping layer

for alrrminidSiGc ~chottlcy diodes2'. The fiim was deposited using RTCVD at 600 O C.

Fermi level pinning was also observed, whereby the b d e r height is not determined by

the work fimction difference of the two materials, but rather by interface traps present at

the junction d o t by any interfacial oxide. This effect generaiiy leads to mcreased

barrier heightdO.

- - - - - - - -

Chu et al hvestigated the effe&of P sa&cial silic&apping layer that does not

completely react with the metal to fonn a silicide? The unconsumed silicon tends to

inhi'bit the banier height's reduction, however it was concluded that the effect is minimal

because carriers will tunnel îbrough the layer, wbich is extremely thin ((-0.1-10 nm).

In March 1997 Nur et al concluded b t it was difncult to accurately determine the barrier

heights of diodes made by thermaUy reacted Pt on SiGe due to the Fermi level pinni#.

To overcome this, a novel technique was devised whereby an SiGe strained layer was

deposited using molecular beam epitaxy (MBE) at 550 O C and was followed by a co-

sputtering process of Pt and Si together to fonn the silicide. Fmn this work, it was found

that the value of the barrier height followed the expected pattern with increasing

germanium content, unüke devices made with thexmaily reacted silicide films.

In order to inhibit the Ge segregation, Colace et al fabricated a Schottky banier SiGe

diode using a two step deposition proces?. Using LPCVD, an initialiy relaxed film -50

nm thick was grown at 330 O C. The islandùig of the Ge was inhibited at the low

temperature, and thus a second film (at 500 O C) was p w n without dislocations. The

detectors exhibited a maximum responsivity of 0.12 A N at f 330 nm, and a dark cment

density of 1 iuvW2.

In 1999, othernovel structures using SiGe f b s for 1550 nm phatodetection were studied

by Janz et uL2' A quantum-well device was fabncated uring a combination of ultra high

vacuum CVD (UHV-CVD) and MBE to get a straincd superlattice without dislocations.

A responsivity of 0.16 A/W at 1550 nm was achieved with this structure.

In ~ummary, îhe deposition temperatures surveyed in the teporteci research literature

range h m 50-700 O C and depend on the method u s d Titanium appears to be the

Schottky b h e r metal of choice for several groups. Many reports include efforts to

eiiminate F+& level pinning, and a silicon-cappïng iayerwas favomd as the manna to

do as.

Upon consideration of these papers, it was found that no progres bas been made in the

use of fiilly relaxed SiGe films for 1.55 pm pbotodetection. This may be due to the

possibility that the dark curent of a device made with this mamial my be large due to

the nmber of dislocations, but it has not yet been shown to be beyond acceptable limits.

It was also inferesthg to note that whik much of the work rwiewed here was based on

Schottky barriers for 1.55 jm photodeteetion, very little research has been conducted in

the area of MSMs, which hoId much promise in integrated optoelectronic circuits. ïhus

this study will provide much insight into the possibility of süicon-based MSMs for future

opticai detector schemes.

This section deals with the various critéria that require consideration whw designiag a

photodetector for fiber optic communications. These factors include the optical

wavelength of interest, detector structure, semiconductor material composition, and

eiectrode m a l composition*

Wavelength

As mentioned earlier, fiber optic cable is made h m fused silica glass. Glasses other than

hued rilica tend to exhibit bigher signal attenuation9. The outer layer, or cladding, is

doped to have a lower refractive index than the inner layer, or core. There are wo

mechanisms which cause the propagiâting sipal to deviate fiom its original power level

and original optical fiequency band respectively, amuation and dispersion.

A f temation

Attenuation is caused by two factors: intrinsic propetties of the glas, such as W or IR

absorption, and extrMsic qualities such as imperféct glass structure. Attenuation in fiber

optic glas is at a minimum around 1.55 pm, as shown by Figure 2''.

Figure 2: Los Spcctnim of Fiber @tic ~1ass'O

Dispersion

Because individual fiequencies of ligbt have different propagation constants, they mvel

at H i t velocities in the fiber. This ultirnately affects tbe maximum bit rate of the

optical signai. Two types of dispersion are found in fiber optic cable: chromatic and

intermodaI. Intermodal dispersion is oniy found in mdti-mode nber, and is due to the

inherent differences in travelling vchcities berwêen modes. Chromatic dispersion is

material related and due to the dependence of the nfractive index of nber on the

waveIength of propagtting light. In the latter case, a signiiicant portion of light travels in

the cladding layer (which has different propagation characteristics thaa the cote), giving

l e m dispersion. Opticai 6kr bas a dispersion minimum around 1.27 p, as show in

Figure 36.

Figure 3: Dispersive Behaviour of Fiber Optic Glass

In consideration of these two minima, it would be desirable to operate long-distance

Iightwave communication systems at these particular wavelengths. WDM systems have

thus been designed such that the wavelength bands allocated for communication channels

f d in the 1.3-1.6 pm range. The challenge presented to optoelectronic device designers

is to design retiable, hi@-powered lasers and detectors that operate in this waveband.

One motivation of this thesis is to meet this challenge and constnict a photodetector that

targets this wavelength range, specincally the 1.55 pm mark.

Structure

There are many common photodetector structures, each with their relative merits and

disadvantages. These will be discussed below, b e g b h g with the simplest PN

photodiode.

PN Junction

A large reverse bias is applied to a PN junction diode, heavily doped on one side. Light

incident on the detector with energy equal to or greater than the semiconductor's bandgap

wüi generate electron-hole pairs within the space charge layer. Any excess energy is

dissipated as heat. If the carriers are generated outside the depletion region, they slowly

diffuse either to the electrodes or to the depletion region. Those carriers either generated

in or diffiised to the depletion layer are swept by the high electric field there toward the

electrodes, conm3uting to the reverse c~nent '~ .

The PN junction is easy to fabricate and analyze. Its detected wavelengtb, operating bias,

capacitance, and bandwidth are al1 highiy tuneable by simply varying fabrication

parameters. However, due to its small depletion width (hence narrow high-field ma), the

PNphotodiode has a lot of carrier generation outside this region, limiting its fiequency

response to the kilohertz range.

PaV Diode

An enhancement to the basic Phrsûucture is the introduction of an intrinsic layer in

between the n andp layers, creating a PNdiode. Under sufficient bias the entire iatrinsic

region is depleted, thereby widening the high field tegion and dramatically increasing the

device's response speed.

The PNs process parameters are easily tunable, as in the PNjunction diode, and it is

rehtivity robust in terms of sensitivity to temperature, shock and vl'bration, The PLV

diode exhiiits a high kquency tesponse and is a common detecmr mcnue in

telecommunication receivers.

Avalanche Diode

The Avalanche photodiode ( A D ) uses carier multiplication and impact ionization to

create a current gain under high reverse bias. Its stmcture is si& to a PiNdiode,

however instead of an iotrùisic region, the middle layer is commonly lightly doped as

either n orp type material".

The multiplication effect produces a higher responsivity in APDs versus other detectors.

However, it also has higher noise due to the same avalanche process, and is highly

temperature sensitive, too much so for diable optical received.

Schottky Bamer/Hetero~nction Diode

Instead of complementary n ami p layes bmught together as in the PNjunction, one layer

of îhe structure is either a metal (Schottky barrier) or another semiconductor material

aitogether (heterojunction). The resulting energy-band diagram is similar for both cases,

and both operate using electron-hole pair geacration under strong reverse bias as in the

PlNdiode. The physical mechanism behind the Scho* b h e r is of particdar

importance to this thesis thus a detailed description of its o p t i o n wilI be treated in

Chapter 3.

MMDiade

The MSM Diode is essentiaiiy two Schotkky bader diodes baçk-to-back. The entire

semicductor layer between two metal electrodes becornes M y depleted mder

d c i e n t bias. In order to maximize responsivity and speed, the eleettodes of the diode

are typicaiiy patterned as interdigitated fingers, 'This pattern is iilustrated in Figure 4.

Figure 4: An MSM photodiode3

Cunmt lightwave receiven do not use MSM detectors, but they are an appealing solution

for optoelectronic integrated circuit (OEIC) receivers because they integrate easily wim

the MESFET a d HEMT circuiay that normally make up the pre-amplifier stage of

optical receivers2. MSM structures make for a very simple, con-effective photodetector

design. Tbey e x b i t low capacitance and cm have very high bandwidths. The smaU

capacitance pennits large area devices, which enable impmved coupling to optical Las.

Their main limitations mclude relatively high dark cwrent (thermioic emission current

over the Schottky banier) and relatively poor extemai quantum efficiency (due to

electrode sbadowiug).

In keeping with the objectives of the thesis as outhed earlier, the goal of this project

was to manufacture and test a aovel MSM photodetector. Detailed theory behind the

MSM is given in Chapter 3.

Material

The initial thrust of this study was to hbricate a good working photodetector able to

absorb iight at 1.55 pm and be compatiile with cunent CMOS processes. As rnentioned

earlier, a siiicon-germanium ailoy is suited for this purpose. However, there were several

factors to consider before using this materid.

Germanium Content

Figure 5 illustrates the bandgap of SixGer, aiioys with hcreasing Ge content for strained

and utraineci films. nie required energy bandgap to absorb 1.55 pm light is -0.8 eV,

correspondhg to a Ge fraction of 0.5 in strained layers and XI.9 in reiaxed layers.

Figure 5: EPOW hndgap venus Cennaniwm conted2

20

Stroined versus Relaxed

Depositing a high quaiity SixGer, layer presents some challenges. The large lattice

mismatch (about 4.17% at room temperature) between silicon and germanium results in

sûained (or pseudomo~hic) pwth up to a certain criticai film thicknessl2. Beyond mis

point, tbe misfit between lattice constants induces &locations ramer than straint6. This

minimum thickness demases as the fraction of germanium is increased, as shown by

Figue 6. Layers grown beyond this critical film thickuess are considered 'relaxed' or

Figure 4 Critical film thichas venus Germanium content"

It can be seen fiom Figure 6 that the critical film thickness of a strained film with a large .

Ge content is a very smail value, too srnail to be an adequate active tegion in a

photodeiector. Given these considerations, a SiGe photodetector can be made in two

ways:

1) Increase the critical 6lm thickness by introducing a graded germanium region

between the silicon substnte and the active region, as per Sturm et alL9 and Colace et

aP4, or

2) Grow relaxed layers of whatever thiclrness is required to provide the optimum

responsivity and speed in a detector.

Successfhl detectors have been demonstrated using the first concept?, which aliows a

large bandgap reduction with relatively little Ge. However only very thin layers can be

grown in order to remain dislocation free. In the second case, the relaxed layer can be as

thick as necessary but this solution requires a larger hction of Ge to shrink the bandgap.

Monocrystalline versus Polycrystalline Mateniil

Another consideration is the choice of depositing crystalline or polycrystalline material.

Polycrystdline material was chosen for three reasons:

1) So tfiat device processing could take place in house. The Carleton Univers@

Fabrication Laboratory has a Rapid Thermal Chemical Vapour Deposition (RTCVD)

machine, cu~feptly capable of depositing doped or mirinsic polycrystalline silicon,

germanium materiai, or doys of the two. To take advmtage of this asset, the material

used for the photocteteetor will be polycrystalihe. In industry;depositing a

polycrystall.ine layer is a standard, mature process used for gates, resistors, and

capaciors m weli-estabfished CMOS processesLO.

2) As shown by Figure 7, the absorption coefficient of amorphous germanium is greater

than that of the conespondhg mmocrystalline material. Thus a srnalier fîim thickness

. . - -

is reqykd in the polycrystalline case. The thinner the f&, the faster the device will

respond since the cariers have a shorter distance to travel to tbe ebctrodes.

3) A succéssfid MSM photodetector was fabricated at the Carleton lab using LPCM)

Oow-pressure chernical vapour deposition)" polysilicon, so the extension to

poIygermanium seemed logical.

Figiire I: Absorption Coeüïdent of Amorphous versus Crystflliae Ce, siU

Metai Electrodes

Two factors were considered when selecting the Schonky banier'metal:

1) The redtant Schottky b d e r height, and

2) The adhesion propecties of the metai

Table 1 lists reported Schottky banier heights of common met& used in device

processing. Titanium, because of its good adhesion and Schottky banier height to silicou,

was chosen as the electrode metal.

Table 1: Schottky barder beight of metaiimonucryst.USne silicon c~ntacts'~

The photolithographie mask made in a previous study was reused here", whereby the

widths and spacings of the e l e d e s were both set to 2.5 p.

Conclusions

InP or InGaAs based detectors, dthough proven material for absorbing 1 .SS pm light, are

expensive and difficult to fabncate on a large scale. A preferable detector would be able

to detect light at this fiequency and yet stiil be readily integrable with conventional

silicon processes. As such, an MSM diode made from relaxed polycrystalline SiGe and

titanium was designed, fabricated and tested.

Schottky Barrier Diodes

A Schottky diode is formed wtien a semiconductor is brou* into contact with a metal to

fom a meial-semiconductor (M-S) juuction. A potential b k e r aises due to the

mismatch between work bctions of the two materials. If the Fermi level of the metd is

-ter than that of the semiconductor (in the case of an n-type semiconductor, for

example), then electrons in the semiconductor, having greater wergy, will ieave the

semiconductor and flow towards the metal until the Fermi levels are aligned This state of

equilibrium is illustrated in Figure 8.

Figure 8: Energy Band Diagram of an M-S junction in ~~uiübrium'"

Here, & is the work fuaction of the metai, 6 is the work function of the semiconductor,

and xs is the electron afnaity of the semiconductor. The electrons that migrate across the

interface to the metai leave behind 6xed positive ions in the semiconductor, creating a

depletion layer with a buiit-in potentiai amss it. The energy band diagram illustrates tùii

behaviour with band bending near the interface. The Mt-in voltage is the d i f f i c e

between the original Fermi levels, Le.

Thus gives the value of the energy required for electmns to move h m the

semiconductor to tbe metal. Carriers in the metal see a barrier given by the Schottky

barrier height #b:

Both & and X, are illustrateci in Figure 8.

A positive voltage VA applied at the metal lowers the barrier from the semiconductor to

the metai by -VA. In this forward bias condition the barrier encountered by eiectrons in

the metal is unaffécted. The Richardson-Dushman equation describes this emission

curent under an appüed bias V:

EQ. 3 J = AIT exp( -mi - V)) (IO) kT

Wheic Vis the appüed voltage, k is Boltzmann's constant ( 1 . 3 8 ~ 1 ~ ~ JK), and T is the

temperature in Kelvin. A** is the effective Richardson constant using the effective mass

of the electron, and is given by:

Where m. is the effective m a s of an electron, q is the electronic charge (1.602~10-'~ J),

and h is Planck's constant (6.625~10'~~ Js).

-- -- --

Applying a reverse b i s (or negative voltage on the metal -VA) inmases the barrier

height on the semiconductor side but again the barrier height seen by electrons on the

metal side is constant. Thus in fonvard bias, electrons in the semiconductor may flow

fieely towards the metal, but in reverse bias they are blocked by a large barrier. The

reverse saturation c m n t of a Schottky diode is the mal1 l h g e carrier fîow irom the

metal to the semiconductor and is given by:

EQ. 5 -99b (16) JO = A" T' exp(-) kT

The cunent-voltage relationship of a Schottky diode in both forward and reverse bias can

be expressed as:

EQ. 6 I = Io (exp(V I n& ) - 1) (lm

Where n is the ideality factor used to account for nonlinear effects and & is the thermal

voltage kT/q.

Image Force Effect

in reality, equation 6 does not accurately describe the cunent-voltage relationship

observed experimentally. The non-ideal beiaviour can be accounted for by considering

the image force of an electron. The electric field hes ema~ating 6om an electron with a

charge -q at a distance x h m the interface act as if an image charge +q were situated on

the other side of the interface at the same distance x. A new relationship can be derived

for the energy vernis distance into the semiconductor using Coulomb's force of attraction

F=qE. Thus the band bending at the M-S interface is not as abrupt as was shown

previously in Figure 8, and the work function of the metal is lowered by an amount A$.

The M-S cment-voltage relationship becomes:

EQ. 7 J = A"T' exp( -q(h -A4 - Y)) (a@ kT

where:

EQ. 8

where E,, is the extemally applied elecmc fie14 and E, is the semiconductor permïttivity.

Equation 7 is a more accurate description of the 1-V characteristics of a Schottky banier

diode.

Schottky Barrier Diodes as Photodetectors

A Schotîky barrier diode made for photodeteetion is typically constructed with a mbstrate

layer, an absorbing semiconductor layer, and a thin semi-transparent metal layer

deposited on top. If light energy is incident on the diode, one of two processes occius. If

the photon energy Eph is less than the energy gap E, of the semiconductor but greater than

the Schottky banier height &,, electrons excited in the metal m y overcome the barrier

height, giving rise to current flow. In the second proces, ibe photon mergy E& is greater

than the bandgap E, and the incident photons generate elecûon-hole pairs in the

semiconductor, which are swept away by the elecmc field force in the depletion region

toward their respective contacts. if the absorption region is fdiy depleted, the device is

very fast, as both carriers move at their saturation velocities.

MSM Operation

Under sufficinit applied bias, the two depletion regions at each interface of the MSM

electrodes become wide enough such that they wiiî meet and the semiconductor layer will

be fully depleted. The voltage at which the meeting occurs is called the reach-through

voltage, or VRT. Applying a bias across the two electrodes in this way creates a reverse-

biased M-S junction at one electrode and a forward-biased junction at the other. If the

appIied voltage is increased further, the fiat part of the energy band and the zero electric

field point will shift to the point m. This is called the MSM flat-band condition.

Figure 9 illustrates the energy-band diagram of a symmetrical n-type MSM with no

appiied bias.

Figure 9: Energy Band Diagram of an MSM Detector in ~~nilibrium"

Figure 10 shows the energy bands and electric fields in (a) reach though and (b) tlat-band

conditions. Here, w is the width of the depletion region, hl is the Schottky barrier height

on the reverse-biased side (where electron injection takes place) and governs the electmn

current there. &.2 is the Schottky barrier height on the fornard-biased side (where hole

for contacts 1 and 2 respectively.

Figure 10: Energy Bands and EIecÉric Field Distribution nt (a) V=Vm and @)

v=v",

For good responsivity, the MSM is normally biased at a voltage p a t e r than the flat-band

voltage (VFB). The value of VFB c m be derived fiom Gauss' Law, and is given by:

where ND is the effective carrier concentration, and s is the electrode spacing.

Dark Current

The dark current of a photodetector is the current present even when there is no light

incident on the device.

In the case of an MSM, the fuadamental dark cwrent mechanism is thermionic emission,

and results h m carriers able to overcome the Schottky barrier at the contacts when an

intermediate bias (and no incident light) is applicd. Thus the dark current is significantly

iduenced by the bamier height and the quality of the metai-semiconductor intdace,

which implies a need for tighter tolerances in device fabrication.

As shown by equation 10, the electron and hole injection associateci with the dark cunent

are expoaentially related to the Schottky barrier heighrs &, and @=Eg-q&, respectivel$.

Where A**,,, A**, are the effective Richardson constants for elecuons and holes

respectively, T is the opetaang temperature, and the image force effect has bew taken

into account. At low bias the dark current is negligrile, whereas at high bias, tunnelhg

33

cumnt becomes significant and the thermionic emission mode1 breaks dom3. It has

been suggested that a minimum dark current value exists if the Schottky bar& height is

chosen to be one haif of the energy gap of the semiconductor E, .~ Norrnally, the noise

generated by dark cunent is ignored due to the much larger magnitude of the noise in the

amplifier stages of optical receivers.

In one study, placing the electrode tips and pads on top of a silicon nitride layer

successfully reduced the dark current in GaAs MSM detectors. The insulating layer

eliminated parasitic leakage paths and the high electric field regions at the electrode

tips.30

Noise

The noise of an MSM is similar to that of a PlNdiode. Thermal noise due to the series

resistance can be neglected, however shot noise due to the dark current and the random

arriva1 of incident photons must be considered. It is also possible that MSMs exhiiit l/f

noise because the current flow is lateral, similar to field effect devices3'. It is thought that

the origin of the I/fnaise is due to traps3, and thus may be a significant factor in the

polycrystalline devices fabricated in this study.

Qumtum Eficiency

The quantum efficiency of a detector is the ratio of the number of generated elemon-hole

pairs to the number of photons required. The interna1 quantum efficiency T\I in equation

11 is solely dependent on the absorption process aud the collection of photo-generated

elecmn-bole pairs.

where a is the absorption coefficient of the material and / is the thickness of the

absorptim region.

External qUaaNm efficiency q, takes inm account the dependency of efficiency on the

overail detector stnu:ture. Thus it varies widely between the various deteetor schemes.

MSMs in partidar suffa fiom low extemal quantum efficiency due to the shadowing of

the electrodes on the s& of the device. The quantum efficiency cm be increased

dramatically if very thin electrodes and wide aaive regions are used, however this would

decrease the speed of the device and increase electrode resistance. ïhis relation is given

by equation 12.

where R is the reflectivity, and SIA is the fiaction of device surface tbat is exposed to

light. la the case of the MSM detector:

where s is ttie spacing between the electrodes of widtii W.

Transparent elecaodes have been used to increase quantum efficiency but have yet to be

demonmted to work reiiabl$',

Responsivity

Instead of using the quantum efficiency as a measuse of device performance, it is more

common to use the responsivity R of the detector. The responsivity is given as the ratio

of the generated photocunent Gh to the incident optical power Pk, with units of amps per

watt.

where v is the fiequency of light in Hertz.

Bandwidth

The frequency response of the MSM detector is detennined by tbe transit time of the

carriers and its RC tirne constant,

Transit Time

Both the electrode spacing and the appiied bias determine carrier transit time. The

electrode spacing determines the distance carriers must travel, and the correct applied

bias ensures complete depletion of the active region. Under f i c i e n t bias, aii carrier

generation takes place within the high field region and the carriers travel at saturation

velociîy.

RC Time Constant

The total resistance of the device is made up of the contact and load resistances. The

capacitance of an interdigitated MSM device is Qven by:

EQ. 15 C = (N - l)CoI

Here, N is the number of fïngen with width w, spacing s, and length 1. Co is the gap

capacitance of the contacts per unit le@. The value of Co is given as follows2:

where 6 and E, are the dielectric constants of fiee space and the semiconductor

respectively, and K(k), K1=K@Y are amplete elliptic integrals of the first kind7:

and k k ' are given by :

EQ. 18 k = tan2( m 4(s + w)

),km= 4 3 "1

The capacitance of an MSM diode is nonnally much smaller than that of an equal am

PIN diode and correspondingly the MSM bas a higher fiequency responsez. For the same

reason, the RC time constant is Eiirly mail and therefore carrier transit time is fiequently

the speed limitation in MSM stnictures3'. The low capacitance is-particularly appealing

because it allows the device area to be very large for coupling to fiber optic cabie. As

weii, its contribution to the tord hnt-end of the receiver capacitance is negligible.

-

4. EXPERIMIENT: POLYGERMANIUM MSM PHOTODETECTOR

To simpli6 the processing and to inmre adequate detection of 1550 nrn Iight, initial

experiments used pure germanium material instead of a SiGe alloy. According to 2-Q Li

et al, a silicon b u f k layer m u t fïrst be deposited in order to provide nucleation sites on

the oxide when using S a and Ge& as source gases.32 A silicon capping layer between

the germanium and metal electrode may also be necessary to reduce the dark cunent

inherent over a small Schottky barrier but may also limit the device speed.

The wafer substrate was p-type, with resistivities typicdy between 4-18 52-cm. A 1000 A

silicon dioxide layer was thennally grown before the active layer so that the active layer

was isolated h m the bulk silicon. The deposited films were annealeci with hydrogen to

passivate the dangling bonds between grain boundaries. In polysilicon films, this has

shown to be effective in reducing the potential banier that exists at the boudaries by

reducing the number of trap States there, thereby increasing carrier rn~bility.~~ The

electrodes were patterned using a metal lift-off procedure, since the ümited metal etching

faciIities available in the fabrication lab are not capable of providing g o 4 dennition at

these dimensions. The liftsff process consists of the following steps:

1. Deposit polysilicodpolygermanium film on 1000 A SiOz in a coldwaü AET Rx

Series RTCVD chamber. For details on process conditions, please see Appendix A.

2. Hyàrogen anneal for 100 minutes at 420 O C and atmospheric pressure.

3. Spin coat wafers at 4000 rpm for 30 seconds with HPR504 photoresix

4. Expose and develop windows into the photoresist using a Karl Suss alignment

system.

5. Plasma etch (Techuics PE-II A) in an Ot plasma for 30 seconds at 100 Watts and

0.3 Ton any residual photoresist h m eIecîroàe windows.

6. Blanket sputtet metal for 15 minutes at 100 Watts, wiîh 60 sccm argon using a

Materials Research Corp. sputtering system.

7. Dissolve remaining photoresist using an acetone bath in an Ultrasonic agitator for

-20 minutes.

8. Rime with isopropanol alcohol and blow dry with a nitrogen gun.

Figure 1 1 outlines the metal lifloff process.

Spun on photaresisr

+ Deposiicdfifm

. Dissolve PR

Figure 11: Typicai Metal LiftoEProcess

Initiai deposition attempts induded a -100 A dicon buffér liyer k pemit nuclcation,

followed in the same cyck by the polygennanium iayer. Three metais (dumirilum,

plathum, and titaaium) were used for the elecaodes for the sake of compa"s0~ The

resuit for each metai was an ohmic conta instead of a Schottky banier. The next

experiment mcluded the -100 A silicou capping layer, which was expected to inmase

the Schottky barrier height. It was thought that perhaps the metal would adhere to

silicon better than to germanium, however the resuit was again an ohmic contact. The

device stiii exhi'bited iightdetecting behaviour, but a detector with otimic contacts does

not satisfy the purpose of this project.

The sarnples were examined through a JEOL JSM-840 scanning electron microscope

(SEM) and were show to have very rough surfaces. The germanium appeared to have

deposited in clumps (or 'islands') across the wafer instead of as a smooth, uniforrn layer.

Although the processing conditions were different, other groups made the same

observation about their germanium films'g. When compared to the polysilicon

depositions made in an LPCVD chamber in the siune laboratory", the difference in

surface quaîity was quite apparent. Figure 12 compares the two samples. To c o n f h that

it was the use of germanium, and not the equipment that was the problem, a polysilicon

film was grown in the RTCVD chamber. As shown in Figure 13, a Schonky banier was

formed between the polysilicon and metal contacts.

Figure 12: Images of (a) PolySiiicon and @) PolyGerdum samples (bar4 p)

1 I

-1 -0.5 O 0.5 1

Applimd Bi- (V)

Figure 13: Schottky ciuracteristic wing RTCVD pdysüicon

Fwther research into germanium deposition processes revealed that most successfûi

high quality germanium films are deposited using lower temperatures tban that used in

the present studqm. This holds mie even for films deposited in RTCVD ch am ber^^^.

For example, Kobayashi et al perfomed studies to compare the grain size and surface

roughness of several germanium tilms deposited at different te~n~erahires~~. As shown by

Figure 14, theü resuits indicate that the temperature used in this thesis study was too high

if minimum surface roughness is required. The value of Ta in the figure is the

temperature of a pst-deposition heat treatment which was perfomed on some of the

films.

Figure 14: Surface Roughness versus Subsbrite ~ e r n ~ e n t u r e ~

Steps were taken to reduce the deposition temperature as much as possible. Kowever, the

M m e 4 pyrometer used by the RTCVD was limited in its operation to temperames

above -640 OC. Attempts to operate the RTCVD beIow the rated temperature of the

pyrometer led to Mie or no deposition on the wafer d e .

At d i s point the original idea of a Siûe alloy film seemed as if it could resolve the

surface roughness problem, or at least minimize its effects. The deposited film wouid stiU

inciude the silicon b a e r layer to encourage nucleation. Various compositions of gas

fiows in the RTCVD chamber were tried. Each time however, the surface was too rough

and the contacts were ohmic.

The resuits of these endeavours are summarized in Appendix A.

A thicker silicon capping layer may have alleviated the problem, however the resulting

structure would likely have had a slow response due to carriers piling up at the silicoa-

germanium interface.

It was fïnally concluded that given the limited resources at the Carleton University Lab

and in the iaterests of time, a successfiil working polygermanium or polySiGe detector

couid not be made. Certainly with more flexible equipment and a better understanding of

germanium processing, a polygemanium MSM detector would be possible, but in lieu of

exploring this idea further, the focus of the thesis shified to a related topic.

5. EXPERIMENT: POLYSILICON MSM PHOTODETECTOR

A successfûi polysilicon detector made for short-haul applications such as optical LANs,

or optical interconnects between boards andfor VLSI chips has already been made at the

Carleton University fabrication lab, Therefore, a seemingly logical step would be to take

advantage of the advanced lithography equipment available in the lab and optimize the

detector's performance.

The JEOL JSM-840 scanning electron microscope can be adapted to create minimum 0.1

pm patterns using polpethyl methacrylate (PMMA) as a positive resist and the electron

beam as the exposure source. The SEM uses the infornation contained in a digital

schematic (as opposed to a glas or chrome plate in a W lithographie pmess) to direct

the electron beam and expose the PMMA. The sample is subsequently developed in a 3: 1

methyl isobutyl ketone:isopropanol alcohol (MiBK:IPA) solution to reveal the desùed

pattern. A picture of the final detector is illustrated in Figure 15.

Figure 15: 0 5 pm MSM pattern

Ebeam iithography is not expected to replace photolithography in high-voIume

manufacnuing facilities, however it is an inexpensive way to do submicroa lithography in

a research environment. The fabrication of a 0.5 pm MSM device in actuai

manufacturing conditions would likely use state-of-the-art submicron photolithographie

equipment.

It is expected that the speed of the 0.5 pm detector wiii be dmmaticaliy improved over

the 2.5 pn device previously made; specifically, with thinnet electrode widths and

spacings, the carriers have a smailer distance to mvel. Many groups have attempted to

increase the speed of silicon detectors in other ways. For instance, Chen et al used a very

thin active layer to reduce the distance travelled by the caniers. The corresponding loss

of efficiency was overcome using a scattering r e k t o r to trap light in the active laye?'.

Using a costiy SOI substrate to block photocarriers generated deep in the bulk siiicon

allowed Honkanen et al to achieve speeds of 5 GHz h m doped silicon MSMS~*.

Some groups have also achieved fast speeds in the same manuer employed in this report.

Using electron beam lithography, Alexandrou et al reported success in fabricating

crystalline silicon MSMs in 1993 that worked at 38 GHz under 725 nm light. The diode

used 0.3 finger widths and ~prpacin~s~~. Simiiarly, Chou and Liu fabncated aystalline

silicon MSMs with 0.1 pm widths and spacings, and achieved speeds of 41 G H ~ . ~

Employing a new lithographie pmeas involves some parameter &ing and optimization.

Specifically, the metal iiftoEprocess d e s m i earlier does not work well at submicron

dimensions. A more complex process was use& whereby a two-level stack of PMMA

with different molecuiar weights (Le. 495K and 950K) creates a 'lip' in the resist stack

that allows for easier liftoff. Without this lip, the metal coverage would likely prevent the

solvent fiom getting to the resist in order to dissolve it. This process is shown in Figure

16.

For the most part the rernaining process parameters were the same as for the o@id

polysilicon detector. An initial 1000 A oxide layer was grown to isolate the devices fiom

the bulk siiicon. The polysilicon film was deposited in the LPCVD chamber (rate: -150 A

per minute), with its thickness scaled d o m to be the same order of magnitude as the

electrode dimensions. For cornparison studies, devices were also made with crystallized

amorphous silicon. In this case, amorphous silicon was deposited at 560°C for 175 min.,

followed by a hydrogen ameal at 650°C for 1 hour to partially crystallize the material, The

expected result is a polysilicon material with larger grain sizes, hence a lower density of

trap m e s at the grain boundaries.

For the electrodes, a metal stack of TüAl was used since Ti provides the optimum

Schottky bmier, while Al adheres well to gold bond wires. Ali of the process steps used

employ standard, mature techniques used in typical CMOS processes. For instance, the

deposition of the active layer could coincide with the definition of the polysilicon

MESFET gate in a typical photoreceiver chip.

The enhanceci liftoff process was performed in the fabrication lab at Carleton University

for the first time for this study. Because of this, numecous problems presented themselves

during the course of the work.

The thickness of the resist stack was optimized based on the minimum lithographie

dimensions (Le. 0.5 pm) used. The resist must be thick enough to provide an adequate

banier (with a 'lip' profile) to the substrate. However, it cannat be thicker than the

electrode dimensions, otherwise it may be diicult to dissolve in the liftsff sep. Thus

the resist thickness was tuncd by diluting the commercial standard PMMA solutions

using chiorobenzene. Attempts were also made to tune the spin speed, but it was found

that increasing the spin speed to thin the resist led to probiems with thickuess unifomity

across the wafer. This effect is acknowledged in other works4'.

It was found that PMMA, although a stable compounà, presents some difficulties in its

handling and use. Over long storage times, the casting solvent (in this case

chlorobenzene) tends to evaporate, leading to thicker resists when spua. Thus periodic

measuring of the thickness of a particuiar resist solution is recommended. As well,

caution must be taken when dispensing the resist onto the wa f i in the spimer. Over a

period of tirne, the PMMA tends to dry up around the rim of the bonle, and upon constant

opening and closing of the lid the dried PMMA falls into the solution and appears as tiny

spots on the wafer after spïnning. Usually if care is taken to obtain the PMMA h m the

middle of the solution (that is, not at the bottom or on the sides), the amount of spottiag

cm be kept to a minimum.

There exisrs a tradesff during e-beam exposure between a lengthy write rime with a

smaU beam current, and shorter expomre times with tw large a current. The optimized

exposue settings required tuning of the electron beam cunent, the charge dose (Le.

Coulombs per unit area), and various other settings on the scanning electron mrmrcroscope.

It was also found that the exposed electron beam in the SEM is highiy susceptible to

outside eiectromagnetic effects. It is Mknowu what the parhcular source of interference

was in this case, but it appeared that the recent construction on the flaors above the

- -

fabrication lab included the installation of new EM sources. As a r d t , the manner in

which the beam sweeps the sample had to be adjusted, because the resuiting noise caused

by the EM interference manifestai itself quite sigPificantiy. As shown in Figure 17, this

leads to considerable noise problems.

Figure 17: Noise Eff- fkom EM inttrfereace (6J00r magnifïcation)

For the purposes of this pmject, a few simple changes in parameters to the design file

averted the noise problem, however it would be usefid in the future to monitor the EM

field in the fabrication lab and identify changes to the EM pattern and associated SEM

behaviour. An associate at NRC dso suggested that the potentiometers in some of the

SEM functiom (for example, WY shift dials, rotation dials, etc) may get noisy after years

of use and require replacing. As well, field cancetlation cîrcuitry may be necessary if the

problems intduced by noise become signifiant.

Files which requise long write times cm often be a problern because the electron beam

current tends to drift over the. Currently, a SEM operator must monitor the beam

current duhg the mite tirne and make adjustments as necessary. However, the

lithography system has an optional package which may be a good solution to this

problern, An automated beam current reader takes m e n t readings at regular intervals

during the write t h e aad automatically adjusts the dose to account for current drift.

A planar plasma etcher is used to etch residual resist in the electrode windows, however,

this was found to be a difficult process to control. The PMMA is very soft and readily

etches in the chamber, at a rate ofapproximarely 50 A per second in an oxygen pIasrna at

0.3 Torr and 100 Watts of power. The plasma etch was found to be necessary in order to

completely open windows in the resist and enable good metal adhesion. However, the

process presented a danger in that tw heavy an etch would remove the resist's Iip profile

required for successfiil lift-off. As well, the plasma etcher in the Carleton fabrication Iab

is found to etch unevenly across even small areas, thus to maintain a consistent etch

across the sarnples, the same spot in the chamber was used each tirne. Afier various

experiments, it was found that approximately 200 A of PMMA (most ükely sufficient to

remove any residual resist) could be etched using conditions of 50 W at 0 3 Torr for 15

seconds*

To maintain the clean surface of the open windows in the resist, the samples mut be

immediately transfened to the evaporatim chamber for the metai deposit foiiowing the

plasma etch. Tfie evaporation of 20 nm Ti 80 nm Al is a straightforward procedure,

whereby the samples were positioned directly above the metai target so that teSulhg step

coverage was perpendicuiar to the wafer surface. This type of coverage is desirable in a

la-offprocess, since it is more difficult for the acetone to get under the metai if it is

deposited at an angle. This concept is iiiustrated in Figure 18.

For this same reason, the evaporation system was chosen over the sputtering system

because the sputtering platfonn is typicaiiy quite close to the metal target, creating a flux

of metal ions that do not al1 travel perpendicular to the sample. In contrast, the

evaporation chamber has a large distance between the cmcibie and sample holder, and it

is a reasonable approximation to assume that the majority of the metal travels in a straight

line toward the sample, producing a step coverage that facilitates easy lifisff of field

metal.

Deteminkg the required thickness of the metal stack introduced a tradesff between two

factors. The metal needed to be as thin as possible to allow the lift-off process to work

easily, however a very thin electrode may not be able to carry the generated photocurrent

and may introduce a large resistance, thereby causing the device to become RC time-

limited Chou and Liu estimated that the actual resistance of nanoscale e l e d e s are at

least 3.7 times p a t e r than that predicted by bulk resistivities, due to the fact that

collisions between electrons and the metal boundary become sigaificant at tbis scale4'.

a EVAPORATION

SOURCE

A- PolySi layer

Hete, the metal

barria to the acctone for

EVAPORATION SOURCE

Figure 18: Step Coverage for Augieâ venus Perpendiruiar Evaporant

Finally, the acetone Iift9ff process introduced a Emde4ff between too little agitation

(whereby the metal did not aiways completely Mt) and too much agitation (whereby the

rough handling caused too much metal to lifte@. The technique eventuaily used began

with 20 minutes in the ultrasonic followed by 10 minutes standing in clean acetone. The

devices were then rinsed with alcohol and blown dry with a nitrogen gun.

The buk of the work in this project involved addressing the various issues outlined

above. To mmmanZe, the fabrication process of the polysilicon detector was as follows:

1. Deposit polysilicon 0.5 pm film in at LPCVD chamber at 627 O C and 0.25 Torr.

2. Hydrogen anneal for 100 minutes at 420 OC and atmospheric pressure.

3. Spin stacked 495W950K PMMA resist layer.

4. Expose electrode pattern using SEM.

5. Develop in MIBKJPA for 3 minutes.

6. Light plasma etch in oxygen at 50 W for 20 s at 0.3 Ton to ensure clean windows

for good metal adhesion.

7. Blanket thennal evaporation 20 m Ti/ 80 nm Al.

8. Dissolve temaining PMMA using acetone in ultrasonic agitator for 20 minutes.

Continue dissolution outside of the ultrasonic for another 10 minutes.

9. Rime with isopropanol aicohol and blow dry with a nitrogai gun-

Performance Evaiurtion of the Polysüicon Detector

A series of mea~urcmcnts were made to characterizt the detector's performance. These

included 1-V measurements to extract the actuai barrier height of titanium to silicon, dark

~~llfent rneasurements, DC photoresponse rneasucements using a continuous wave light

source, and high kquency measurements using a pulsed source. h both photoresponse

tesr, an optical fiber was positioned directly above the detector with the incident light

pqendicular to the &tector plane. The fiber core was centred ditectly on the detector to

maximize the amount of incident light, and was as physicaliy close as possible to the

sampies.

BaMr Height

Using 1-V data taken h m a sample Schottky barrier, the barrier height can be extracted

by plotting the forward bias m e on a logarithmic scale- Extrapolating the straight ihe

to the zero b i s point gives a value of the saturation current density, la. Frorn JO the

- @* banier heigbt can be found using JO = A''T' expl-) : kT

This method asmmes aear-ideal Schottky behaviour at the metal-semiconductor juncîion.

In this case, an average of severai 1-V c w e s was taken and the banier height was found

to be 0.48 eV, very close to the expected valw of 0.5 eV. However, the ideality factor of

the contact was fornid to be -3.3, indicating that the barrier iikely does not foiiow

Schottky theary very closely. ïherefore there may be other uuknown current mechanisms

at work in the detector, perhaps due to an interfacial oxide layer at the metai-

semiconductor junction.

DarR Current

Because of the high trap density in polysilicon, it is expected that the dark cunent of the

device will be high due to c d e r recombinatiodgeneration at the t ~ a p centers. Assuming

the cment behaves according to a thermionic emission mechanism, we can estimate the

expected dark current value using the calculated barrier height of 0.48 eV,

Dark current measurements were made with a Hewiett-Packard 41 55 Semiconductor

Parameter Analyser at Carleton University. To illustrate the dependence of grain size on

the magnitude of dark cunent, the dark cunent of crystallized amorphous silicon

detectors is shown with both the measured and theoretical dark current of the as-

deposited polysilicon samples.

1 Ciys ta l lbsd Amciphcus Silicon

j-

Figure 19: MSM Dark Current

The as-deposited polysilicon samples have a larger dark curent value than expected; in

addition, they appear to have a larger bias dependence than predicted by thermionic

emission theory. As mggested eariier, it is possible that the metai-semiconductor

contacts do not foiiow Schottky behaviour, and that the grain boundaries of the

polysiiicon materiai introduce other curent mechanisms. For instance, c h e r

recombiiiationlgeneration at trap States could contribute to the dark cunent. In addition,

emission at the grain bomdaries induced by the applied electric field may account for the

large dependence on bias.

As expected, the crystaliized amorphous silicon devices have a lower dark m e n t due to

the larger grains (hence lower density of traps) present in the materiai. This result may

CO* the presence of field-induceà emission at the grain boundaries, since the bias

dependence is l e s apparent in this materiai. However, depositing amorphous siiicon is a

conventid CMOS processing.

ResponrM'ry/ExtemI Quantum Eflciency

A quick calcularion into the expected ùetector quantum efficiency and responsivity

requires knowledge of the absorption coefficient of the active iayer. According to an

earlier study that used the same photoactive material, the absorption coefficient of the

polysilicw is approximately 2 . 5 1 0 ~ cm" at 980 nm." Using the equations detailed in

chapter 3, and assuming a reflectivity R of 0.3 (air-silicon):

The actual plysilicon thickness was measured after deposition and hund to be -0.42

pm. ThiS reduces the expected quantum efficiency to 3.5% and the responsivity to 27.6

M.

Responsivity measrirements were taken at the Institue for Mimsmictural Sciences at

NRC in ihe mamer iIlustrated in Figure 20. A 980 nm laser was chopped at 400 Hz and

58

the reference signal was split off and fed into a lock-in amplifier. A pre-amplifier

supplieci the bias and tead the photodetector's signai, converthg the output cunent into a

voltage signal that was also fed into the lock-in. Figure 21 illustrates the gencnited

photocurrent as a function of applied bias for three different laser input power levels.

Lock-in Amp 1 7 - OUTPUT BIAS

u MSM Detecîor

Figure 20: Photoresponse Measunment Setup

O 1 2 3 4 5

Applied Bias (V)

Figure 21: MSM Responsivity

From the above curve, the device responsivity was measured to be 3 mAW at 980 nm

and 4 V applied bias. This value is much lower thaa the expected value of -28 mAN.

There are a number of minor practical ciifferences between the calculated and actual

responsivity. n ie meamred value of incident light power was taken at a point that was

not the same as the output facet which illuminated the device; therefore, there is some

lors in the splice and connectors (typical SMA connecter losses are -1 dB, while mort

worst case splice losses are -0.3 d ~ ) ~ .

The spot size of the multimode fiber is about 50 pm, but the device's area is on the order

of 50x15 pm; therefore only about 30% of the ligbt actually illumiaated the detector. If

the illumination power is to be consichci constant over the fiber spot size (in actuaiity,

the modes follow a more gaussian profile), then a recalcuiation of the responsivity given

a smailer incident power level improves the value to about 8 mA/W.

It is suspected that these issues parîially conmibute to a much lower responsivity than

initially calculated. Other theorics, which may require a more detailed study, conceni the

resuitant structure of the polysilicon iayer. The thin layer deposited may not have the

columnar profile of the matenal tûat was deposited earlier and was reported to have an

absorption coetncient o f a = 2 .5~10~ cm-' at 980 nm". Thus this value of a may be

incorrect for this device. The potentiaily lesssrdered structure in this case may also have

a significant density of recombination centers, leading to a marked decrease in carrier

collection at the elecaodes. Expcriments that could determine the amount of

photogeneration taking place and what fraction of the photogenerated cariien are

actually coiiected may provide some information on this topic.

The polysilicon detector may be better suited for applications that operate at shorter

wavelengths, where the response would likely be better since the absoIption coefficient

of silicon increases rapidly below 1 pm wavelength.

Responsivity measurements were attempted on the crystaliized amorphous devices but

they appeared to have no detectable photoresponse. It is thought that due to the smaller

density of grain boundaries, the corresponding absorption coefficient is smaller than that

of the as-deposited polysilicon and closer to that of monocrystaliine (whose absorption

length at 980 nm is -100 pm). Therefore the 0.5 pm active iayer is too thin to provide

any measunible respanse.

Capacitance

Using the fomiulae given in chapter 3 and the vahies specific to this design, the foiiowing

calculations present the expected value of capacitance at 0.5 pm finger widths and

spacings:

EQ. 22 k = tan2( ) =0.1716,k1= d s = 0.9852 4(0.5p +O.Sp)

Computing the complete ellipticals of the first kind is a standard function in MATLAB.

The remlts are presented in the next equation.

Finally, these values are used to compute the total capacitance fiom the gap capacitance

per unit length:

EQ. 25 C = (N - 1)CJ = (20 - 1)(567xi0'" f X50x104m) = S3.8jF cm

In contrast, a PaVdioàe with a 0.5 pm inaiasic layer and with equivalent area would

have a capacitance oE

As stated earlier, the capacitance of MSM photodetecton is smaller thaa that of

comparable PIN diodes. Thecefore, capacitance does not play much of a role in the

fiequency response of the detector since an MSM detector is noxmaliy Iimited by the

transit rime of photogenerated carriers, and not the RC time constant.33'

Frequency Response

As a first order approximation, carrier transit times can be found by dividing the distance

between electrodes by the saturation velocity (assuming full depletion of the active

iayer).

It is assumed that the saturation velocity of polycrystalline silicon is the sarne as that of

monocrystalline silicon, because its value only depends on c h e r effective mass and is

independent of r n o b i ~ i t ~ ~ ~ .

Where w is the distance travelled by carriers between electrodes. Thus an ideal detector

bandwidth can be found:

1 EQ. 28 B =-= 1 = 3 1.8GHz 2Z? 2~-5~10''~

The actual fiequency response of the detector was measured using a 850 nm Tksapphire

laser. The laser, with a FWHM pulse width of 150 fk, was pulsed at a rate of 76 MHz. A

bias-tee circuit was used to both provide the bias to the detector and carry the RF signal

to a 50 GHz digital sampling oscilloscope. This setup is shown in' Figure 22. The

detector response at the osciiioscope at bias points of 1,2, and 4 V is plotted in Figure 23.

Figure 22: Impulse Response Measurement Sehip

F'igre 23: Temporal Respanse

- - - . .

From this data, it can be seen that the response at the fàlling edge can be broken into

two parts: a fast and a slow response. Thus there are multiple hquency components

which wiii appear in the spectral response. This concept in shown in further detail in tbe

following figure using the data coiiected at 4 V:

0.0 ! 1 0.4 0.6 0.8 1 .O 1.2 1 -4 t

Time (ns) I 1 1

Figure 24: Temporal Response at 4 V

In order to sîudy this result M e r , the data was plotted on a Iogarithmic scale and the

two t h e constants were found by fitting the two sections of the graph to appropriate üne

equations. The slope of each line gives the tirne constant associated with that particuiar

area of the graph. This result is plotted in Figure 25.

1 0.4 0.6 0.8 1 .O 1.2 1.4 Tim (nr)

i

Figure 25: ExponenoW Fitting to Impulse Respanse Data

The time constants seen hem are much larger than originally expected in the first order

transit t h e approximation, for several possible reasons. The possibly large number of

grain bouaddes may serve as trap centers for photogenerated carriers, which hold the

caxriers for an indefinite time and then release them. Fuaher study into the actual

structure of the polysiiicon material, for example, using deep-level transient spectroscopy

(DLTS), may subsiantiate this iheory. DLTS measurements can detexmine whether or not

the tirne constants have a eXpcE~kn temperature dependence (where EA is the trap

energy), perhaps leading to the conclusion that trap States piay a signiiicant d e in the

carrier transport within polysiiicon.

The slow response may aiso be due to carriers generated in the outside edges of the

detector, beyond the influence of the high electnc field These cmïers have to rely on the

slow process of diffusion to travel to the electrodes. As mentioned earlier, the spot size

of the fiber illuminating the detector was much Iarger thau the total device area and

indeed a significant hction of the carniers may have been generated this way. A mesa

etch of the active regioa around the detecior would likely eliminate this problem.

The very end of the tail was found to be made up of severai ripples, most likely

refiections within the cable from the bis-tee circuit to the osciiloscope. This idea was

confirmed when a shorier cable was inserted and caused the refiections to shift away

fiom the pulse. Reflections mch as these are achowledged as a typical source of

measurement e r ~ o r ~ ~ , as may be the case in this study.

The fiequency response was found by feeding the impulse data into a Fast Fourier

Transfonn (FFT) algorithm. The result is plotted as follows:

Figure 26: FFï of Temporai Response

ïhe two different thne constants are not obvious within the spectral response because the

FFT idgorithm flnds an overall average system response. T'us the detector bandwidth

was found to have a 3dB point at 1.5 GHz. It was expected that the kquency response of

the 0.5 pm detector, as a fht-order approximation, would scale linearly with the speed of

750 MHz reporteci h m the 2.5 pm MSM made eariier, assurning that the both devices

are transit-time limite& Thus in contrast to the measured response stated above, the

expected bamiwidth was about 3.75 GHz, This discrepancy is psibly due to the issues

presented earlier in this discussion (iicreased trap âensity, wasted illumination outside

detector area, etc. ..)

The fabrication of a polygermaaiuxn MSM for 1550 nm detection using a metai lift-off

process was attempted. ï he resultant germanium fiIm deposited by the RTCVD system

was too mugh to produce the necessary rectifying Schottky contact. A lower temperature

process may pexmit the growîh of a smooth germanium surface, but this option is not

available in the Carleton Fabrication Lab.

A fast but leaky polysiiicon MSM detector was fabricated, tested, and characterized. The

namw widths and spacings allowed photogenerated carriers to be coüected quickly,

however the large trap densicy inherwt in polycrystahe material caused the response to

be much slower than expected. The detector has a responsivity of 3 mA/W at 980 nm, a

dark current of 39 nA at 4 V, and a 3 dB baudwidth of t .5 GHz.

There are several ways to optimize the device fabricated in this project. A number of

published works recornmend various methods to reduce dark current. As mentioned

earlier, placing the elecoode tips and contact pads on top of silicon nitride has been

shown to eliminate leakage psth and high elecmc field regionp. Reducing the da&

current would cerrainly strengthen the signal to noise ratio and hence improve device

performance.

The device total area can be increased in order KI maximize the responsiviîy of the device

by allowing al1 incident light fiom the fiber to be collected- This may further complicate

the liftsff procm and would d t in increased &vice capacitance, but îhis trade-off

. . ...

would likely be acceptable. Another way to boost responsivity is to deposit an anti-

refiection coating such as silicon nitride.

The detector response can be greatly improved by tinding methods to eliminate the slow

component of the fiequency response. A mesa etch (or even a light block) around the

&tector would eliminate the carriers which slowly d i e to the elecrrodes. hcreasing

the total detector area would also suppress this effect, since the electxic field would be

present everywhere that the light is incident, ailowing carriers to drift quickiy to the

elecîrodes.

Certainly there is much work involved in examining the polysilicon structure and the

current mechanisms it uses. For instance, using the same materiai with different h g e r

spacings may c o n f b whether or not the device is in fact transit-time limited. As

mentioued eariier, deep-level transient spectroscopy may be able to provide more insight

into the effect of ûap m e s on detector speed.

This study has demonstrateci the ability to manufacture submicron photodetectors with

satisfactory performance, but has also opened many questions about carrier transport in

polysilicon and its effects on photoresponse and speed.

Experiment Number

1 2

3

Film Composition*

Si BI Ge AL Si BI Ge AU Si C

4

Si

5

H2 to d u c e surface roughness

-se

Initial anempts Detennine if silicon cap

Si BI Ge AU Si C

6

(table continued below)

R d t s

Ohmic Ohmic

will give Schottky contact Determine if Schottky contact is possible in

Si BI Ge AU Si C

1 Experiment 1 Deposition 1 Pressure 1 Gas Flows 1

Schottky

RTCVD Decrease temperature to

Si BI SiGe AU Si C

No film deposited reduce surface roughness Decrease pressure to

' Notes: The substrate was always an n-ye wafer with 1000 A gare oxide grow on top. Base pressure was usually 5x10 Torr or better. Ab breviations: B: buffer layer (-100 A) AL: active layer (-2500-3000 A) C: capping Iayer (-100 A)

Ohmic kduce surfiice roughness Change to SiGe alloy; Add

Number

v

1 2 3 4 5 6

Ohmic

~emperature ("Cl

750 750 680 630-680 640 640

(Torr)

5 5 0.5 5 1.6 1.6

N2

750 750 750 -

750 750 750

S a

2.5 2.5 5 2.5 2.5 2.5

Ge&

500 500 O 500 500 500

H2

O O O O O 100

8. APPFXDIX B: LIST OF INSTRUMENTS USED FOR

DC Measurements

ILX LightWave LDX-3412 Precision Current Source

Semiconductor GaAs Laser Diode

Newport 835 Optical Power Meter

Stanford Research Systems SR540 Chopper Controller

Stanford Research Systems SR530 Lock-In Amplifier

Sbnford Research Systems SR570 Low Noise Cunent Re-Amplifier

High S'ed Measurements

Coherent Systems Miro 900 TiSapphire Laser

50 GHz Hewlett-Packard Bias-Tee

Stanford Research Systems SR570 Low Noise Current Pre-Amplifier

Tektronix CSA 803C 50 GHz Communications Signal Anaiym

' Hirahara et al. 'Opticai Communications Technology Roadmap', IEICE Transactions on Electronics, Vol. E8 1-C, August 1998 pp. 1328-1341 Paliab Bhattacharya, Semiconductor Optoelecîronic Devices, 2nd ed. Prentice Hall 1997 Mario Dagenais et al, Integrated Optoelectronics, Academic Press, 1995 'SONET 101: An Introduction to Basic Synchronous Optical Networks', Intemal Document, Norte1 Networks, December 1995

G.P. Agrawal, Fiber-ûptic Communication Systems, John Wiley & Sons, Toronto 1992 ' M Ito et al, 'Low Dark Current Ga& Metal-Semiconductor-Metal (MSM) Photodiodes

Using WSi, Contacts', IEEE Journal of Quantum Electronics, Vol. QE-22, Jdy 1986 pp. 1073-1077

* C. Harder et al. '5.2 GHz Bandwidth Monolithic GaAs Optoelectronic Receiver', IEEE Electron Device Letters, Vol. 9 No. 4, April1988 pp. 171-173 Ed. Casimer DeCusatis et al. Handbook of Fiber Optic Communications, Academic Press, Toronto, 1998

'O Edward S. Yang, Microelectronic Devices, McGraw-Hill, Toronto 1988 I 1 Richard A. Soref, 'Silicon-Based Optoelectronics', Proceediigs of the IEEE, Vol. 8 1, No. 12, December 1993 pp. 1687-1703

IZ Subramanian S. Iyer et al, Heterojunction Bipolar Transistors Using Si-Ge Alloys, IEEE Transactions on Eleceon Devices, Vol. 36, No. 10, Oct 1989 pp. 2043-2062

l3 H. Teaika '20 Gbls Optical Transmitter and Receiver Using SiGe Integrated-Circuits', Electronics and Communications in Japan Pt. 1 - Communications, Vol. 79 No. L2, December 1996 pp. 25-35

l4 C. Tocci and H.J. Caulfield, Optical Interconnection: Foundations and Applications, Anech House, Boston, 1994.

'' R Lytel, 'Optical interconnections within Modern High-Perfornlance Computing Systems', Proceeedings of the EEE, Vol. 88 No. 6 June 2000 pp. 758-763

l6 D.J. Goodwiil, 'An ATM-Based Intelligent Optical Backplan using CMOS-SEED Smart Pixel Arrays and Free-Space Opticai Interconnect Madules', EEE Journal of Selected Topics in Quantum Electronics, Vol. 2 No.1 April 1996, pp. 85-96. " RP. MacDonald et al, 'MSM Photodetector Fabricated on Polycrystalline Silicon', EEE Photonics Technology Letten, Vol, 11 No. 1 Jan 1999 pp- 108-1 10

'' EH. Rhoderick and RH. Williams, Metal-Semiconduetor Contacts, 2d ed., Oxford University Ras, Oxfard 1988.

l9 X Xiao et al. 'SiliciddStrained Sii&& Schot tky-Be Mared Detectors', IEEE Electroa Device Letters, Vol. 14, No. 4, April 1993 pp.199-201 '' J.R. Jimenez et al. 'SchotîIq barrier heights of Pt and k silicides formed on SilSiGe

measured by interna1 photoemission, Journal of Applied Physics, Vol. 75 No. 10,15 May 1994 pp. 5 160-51 64 " R.L. Jiang et ai. "Roperties of Schottky contacts of Al on SiGe alloys', Applied

Physics Letters Vol. 68 No. 8,19 Febni2iry 1996 pp. 1123-1 125 D.P. Chu et ai. 'Theoretical investigation of CoSiZ1 Sii.,Gex detectors: Influence of a

Si tunnelling barrier on the electro-optical charactenstics', Journal of Applied Physics, VOL 79 (2) 15 January 1996 pp. 1151-1 154 0. Nur et ai. 'Electrical and structural characterization of PtS@ Sii,Ge% low Schottky

h i e r junctions prepared by CO-sputtering', Journal of Vacuum Science Tecbnoiogy, B 15 (2), MadApr 1997 pp. 241-246

24 L. Colace et ai. 'Metal-Ge-Si heterostnictures for near-infiared light deteaion', Jomai of Vacuum Science Technology, B 17 (2) March/April1999 pp. 465-467 S. Janr et al. 'Si/ Sil.& using three-dimensional growth modes to enhance

photoresponse at 1=15SO nm', Journal of Vacuum Science Technology. A 18 (2), MarchlApril2000 pp. 588-592 " LW. Matthews and A I . Blakeslee, 'Defects in Epitaxial Multilayers', Journal of

Crystai Growth, Vol. 27 1974 pp. 118-125 *' X. Xiao et al. 'S ilicidelStrained Si&% Schottly-Barria Infrared Detecton', 'IEEE

Elecaon Device Leîîers, Vol. 14 No. 4, April1993 pp. 199-201 D. Beagiehole and M. Zavetova, 'Fundamental absorption of Amorphous Ge, Si, and

GeSi Alloys', Journal of CrystaUine Solids, Vol. 4 1970 pp. 272-278 29 SM. Sze, et d 'Curreat Transport in MetalSemiconductor-Metai (MSM) Stmctures',

Solid-State Electronics, Vol. 14, pp.1209-1218 W.A. Wohlmuth et al, 'Dark Current Suppression in GaAs MetalScmiconductor-

Metal Photodetectors', EEE Photonics Technoiogy Letters, 8, August 19% pp. 1061- 1 O63

3' J.BD Soole and H. Schumacher, 'TnûaAs Metal-Semiconductor-metal Photodetectors for Long Wavelength Optical Communications', IEEE Journal of Quantum Electronics, Vol. 27, k h 1991 W. 737-752

32 V. Z-Q. Li et al, 'Structure and properties of rapid thermal chemical vapor deposited polycrystaliine silicon-germanium films on Si& using Si2&, Ge&, and B& gasesy, Journal of Applied Physics, Vol. 83 No. 10, May 15 1998 pp. 5469-5476 . - -.

33 C.H. Seager and D.S. Ginley, 'Passivation of grain boudaries in polycrystalline silicon', Appiied Physics Letters, Vol. 34, No. 5, March 1 1979 pp. 337-340 Akira Sakai et al, 'Growth of main-relaxeci Ge films on Si(Oû1) swtkes', Applied

Physics Letters, Vol. 71 No. 24, December 15 1997 pp. 35 10-3512 35 SCO-Kyu Lee et al, 'Low Temperature (1550°C) Fabrication of CMOS TFT's on

Rapid-Thermal CVD Polycrystaliine Silicon-Germanium Films', Japanese Journal of Appüed Physics Vol. 36 1997 pp. 1389-1393

36 H. Koyabashi et al, 'Electrical properties of evaporated polycrystalline Ge t h i . h', Thin Solid Films, 300, 1997, pp. 138-143

37 E. Chen et al. 'High-efficiency and high-speed silicon metal-semiconductor-metal photodetecton operating in the inhed' , Appiied Physics Letters, Vol. 70, Febniary IO 1997 pp. 753-755

38 K Honkanen et al, 'High-speed metal-semiconductor-metal photodetectors fabricated on SOI-substrates', Physica Scripta, Vol. T79 1999 pp. 127-130

39 Alexa~drou et al, ' A 75 GHz siiicon metal-semiconductor-metal Schottky photodiode', Applied Physics Leners, Vol. 62, May 17 1993 pp. 2507-2509 S.Y. Chou and M.Y. Liu, 'Nanoscale Tera-Hertz Metal-Semiconductor-Meta1

Photodetecton', EEE Journal of Quantum Electronics, Vol. 28, October 1992 pp. 2358-2368

4' J-C. Nabity, Nanometer Pattern Generation System User's Manual, Version 7.5, Bozeman, USA January 1997

42 SM. Sze, Physics of Semiconductor Devices, 2* ed. John Wiley & Sons Toronto 198 1 " KT. Hawkins II et al, 'Cornparison of Fast Photodetector Response Measurements by

Optical Hetedyne and Pulse Response Techniques', Journal of Lightwave Technology, Vol. 9, October 1991 pp. 1289-1294