Upload
ngokhanh
View
235
Download
1
Embed Size (px)
Citation preview
Ronald A. Nordin
Anthony F. J. Levi
A System Design Perspective onOptical Interconnection Technology
Demands for increased interconnection density and higher bandwidth, coupled with constraints on the costofadvanced wide-bandwidth telecommunication switching and high-throughput computer architectures, areexhausting the capabilities ofconventional electrical interconnection. Advances inintegrated circuit technologies and enhanced digital services have created a"bottleneck" at the circuit-board-to-eircuit-board level ofthe interconnectionhierarchy. Toalleviate thisproblem, AT&T is developing new techniques inparallel optical interconnection. Inserting parallel optical interconnectiontechnology into advanced electronic processing systems notonly meets projected performance requirements, butalso potentially offers them ata competitive cost. This paper describes how future high-performance paralleloptical interconnect technology might be packaged and used.Introduction
Technological advances in integratedcircuits (Ics) and demand forenhanced digital telecommunication services andadvancedcomputer services are creating a needforincreased interconnection density andhigherbandwidths, while maintaining stringentreliability specifications. Enhanced digital telecommunications services, which includeBroadband Integrated Services Digital Network (BISDN), high-definition television(HDTV) , video conferencing, cellular services,and microcell interconnection, havespurredresearchand design ingraphics, real-timeimage processing, patternrecognition, modeling,speechprocessing, and parallel processing. (See Panel 1fordefinitions ofabbreviations, acronyms, and terms.)
Increased interconnection density andbandwidth at alllevels ofthe interconnectionhierarchy are shapingsystemdesign advancesin traditional electronic packaging andopticaltechniques.1-4 End-to-end optical datalink(ODL) products (e.g., ODL® 40 and ODL50) havebeen used successfully in high-performancecomputer and telecommunication switchingsystems, such as the AT&T Network Systems5ESS® switch. To meetfuture systemneeds,AT&T is developing newtechniques in paralleloptical interconnection.
Parallel optical interconnection willmostlikely be achieved usinga generic componentdesigned to communicate overdistancesbetween 1meter (m) and 10kilometers (km). These devices are known as onedimensional onts (lD-ODL).
Figure 1ashows the conventionalelectronic packaging technology forlargedigital switching andcomputer systems. Thelengthofthe corresponding interconnectionhierarchy is roughly bounded byits packagingtechnology, as shown in Figure lb. Electrical interconnections are efficient andsupport the density and bandwidth needed tointerconnect lengthsofless than 1m. DiscreteODLs are efficient forinterconnection lengthsofmorethan 25m, but theycannot supportthe density or cost constraints imposed byhigh-performance system design in the 1- to100m range. In this range, weexpect the 10ODL to be a competitive solution.
Tofulfill the needsofmany smallvolume users,which in total createa largedemand for 1O-0DLs, the industry needsastandard, such as the IEEE 1596-1992 ScalableCoherent Interface (SCI), that will notneed to be customized fordifferent applications. Although it supplies computer-bus-likeservices, it uses a collection offastpoint-topoint links (ofsize16N +2), instead ofa
AT&TTECHNICAL JOURNAl. 0 SEPTEMBER/OCTOBER 1993 37
Panel 1. Abbreviations, Acronyms, and Tenns
1O-oDL - one-dimensional optical datalinkAlGaAs - aluminum gallium arsenideAu-goldBER - biterror rateBH-GRINSCH - buried-heterostructure graded-index
separate-confinement heterostructureBISDN - broadband Integrated Services Digital Net-
workcw - continuous wavedBm - decibels referenced to a milliwattDFB - distributed feedbackECL-emittedcoupkrwgicEEL - edge-emitting laserEMI - electromagnetic interferenceFP - Fabry PerotGaAs - gallium arsenideGaInAsP - gallium indium arsenide phosphideGbit/s - gigabit(s) per secondIC - integrated circuitInP- indium phosphidekm- kilometerLED -light-emitting diode
physical bus, to reachdatarates as highas 1gigabit persecond (Gbit/s) per link. This standard provides a convenientframework fordeveloping the 1O-GDL
The keytechnological developments neededbythe 1O-ODL are laserdiode arrays, detector diode arrays,solderbumpbonding, electronic laser driver arrays, electronic receiver arrays, optical submount technologies(e.g., micro-machined silicon with waveguide capability),optical fiber ribbon cables, andoptical fiber connectorization. Figure 2ashows a schematic diagram ofAT&Ts1O-ODL technology. System design advantages ofthistechnology include low-eost optical datalinks (packageand manufacturing costare sharedovermany channels);highbandwidth; high reliability; piece parts ruggedenough formanufacturing environments; high opticalinterconnection density; laser-to-Iaser optical delay differences, or skew, low enough to be used in synchronousdesigns; a small module footprint (approximately 6.5crrr'): andlow electromagnetic interference (EM!). This
38 AT&TTECHNICALJOURNAL. SEPTEMBER/OcroBER1993
LGA -laser gatearrayUNb0 3 -lithium niobatem-meterMAC - multi-fiber arrayconnectorMo- molybdenumNRZ - nonretumto zeroNPN - transmitteroc- optical carrierODL - optical data linkOEIC - opto-electronic integrated circuitPiN-diodePN- diodePOLYFIC - thin-film polymer film integrated circuitPOLYHIC - thin-film polymer hybrid integrated circuitPQFP - plastic-quad flat packps- picosecondR1V - room-temperature vulcanizationSCI - scalable coherentinterfaceSEED - self-electro-optic effect deviceS.SEED - symmetric SEEDSEL - surface-emitting laserSn-tinW-watt
represents a significant cost/performance advantageovera serialODL scheme, which uses discrete devices, oran all-electronic interconnection scheme.
Figure 2bshows how1O-ODLs may be incorporatedin a systemdesign. Theycanbe mounted to thesurface ofa board, andthen,at the appropriate point inthe manufacturing process, the multi-fiber arrayconnectors (MAC)-IIs canbe mounted andconnected to the 10ODL Figures 2candd shows 181O-ODL modules (18 wide)mounted ona circuit board. Each module has 18fibers,andeachfibertransmits at 1Gbit/s, collectively yieldinga bandwidth of324 Gbits/s. Forhigh-bandwidth connections, this represents a significant improvement over"traditional" connections between a circuit board andbackplane. The slim profile andlow power dissipation of1O-ODLs allow themto supportmoremodules percircuitboard andmorecircuit boardsper equipment shelf.
Critical to the development ofthe ID-ODL programis the adoption ofan appropriate packaging
Switching/computer system
-1 234Multiple frames ~
Multl-chlp PCB
-"~Q(MCM~ Q 7Figure 1. Conventional large system(a) packaging and(b) Interconnectionhierarchies and theirtypical lengths.
Shelf
0123456789
MCM
000Integrated ---t~O 0 0 0
circuit chips 0 0 0
(a)
QCM-to-MCM::> ~me-tO-fra~
@I~to-c~ dhelf-to-sh~~-8te--to-g-~ CfCB-to-PCj) C LAN =:::>
1[).()OL
Electrical OOL
0.1 mm 1 mm 1 cm 10 cm 1 m 10 m 100 m 1 kin
Interconnection length
(b)
scheme, in which the transmitter and receiver modulesarecompatible with an existing fiberribbon connector.One such fiberribbon connector isAT&Ts MAC-II,which consistsof 12or 18multimode (orsingle-mode)glassfibers placed inaccurately machined silicon vgrooves on 25G-micrometer {J..Lm) centers.Guide pinsensure that the polished ends ofthe two connectorhalves canbe accurately attached with less than a O.!HI.Bcoupling loss.For the MAC-II, the standard deviation for
fibermisalignment is less than 2 urn. Bonding the laserarray (ordetectorarray) on a silicon submount thatincorporates v-grooves andguidepinsmakesit compatiblewith the MAC-II connector, and usable ina manufacturable, self-aligned packaging scheme.
Backplane Interconnection RequirementsPanel 2 listsgeneralsystemspecifications foran
advanced synchronous high-performance system based
AT&TTECHNICALJOURNAL. SEPTEMBER/OCTOBER 1993 39
Figure 2. 1D-ODLtechnology. (a) Schematicdiagram of a transmitter connected to areceiver by 18-wldeoptical fiber ribbon.The physical design ofthe system Is shownIn (a) a side view,(b) an end view, and(c) and (d) a topview.
Transmitter module
Driver array
(a)
Receiver module
Decisioncircuit array
Transimpedanceamplifier array
Rberribbonconnectorized pigtail
(18 wide) 1D-ODLPQFP
~=n1=!I!="=!!!=!!!="=!\~==========:::JMAe-II
(c)
o12
1617
(b)
MAe-II 1D-ODL
(d)
Printed circuit board
EIeetronIcptocie8$Ing
board space<e.a., MOMs,power supply,memory, etc.)
on environments that contain genericcentralofficeswitching equipment. To attain the requiredperformancewithin acceptable constraints, the electronic driverandreceiver mustbe integrated intoan arrayform. Althoughintegrating veryhighgain/bandwidth amplifiers is diffi-
40 AT&TTECHNICAL JOURNAL • SEPTEMBER/OCTOBER 1993
cult, because ofcrosstalk and power/ground "noise,"integrating low-gain transimpedance amplifiers (e.g.,300n) is muchmorefeasible. The evendistribution ofpower dissipation between the transmitter and receiveris also desirable, from a systemperspective. Key to the
Panel 2. System Specifications for an Advanced Synchronous HIghPerformance System
The generalsystemspecifications foran advanced synchronoushigh-performance systembasedon environments that contain genericcentraloffice switching equipment include:- Cost/interconnection - Less than$50, which includes driver,
receiver, connectors, and backplane media- Module reliability - 105 hours < (J
- Zero systeminterconnect error - Appropriate digital coding(e.g., parity) to ensure systemintegrity
- Ambient temperature - O°C < T < 70°C- Absolute humidity - Lessthan 0.026 (5percent< relative humi-
dity< 95percent)- Interconnect length (L) - 1< L < 100m- Compatible with standardEeL input/outputlevels- Module electrical differential inputsandoutputs- Power supplies - +5, ground, -2, and-4.8- Bitrate (BR) - 0 < BR< 1Gbit/s- Average groupinterconnect delay variation - Lessthan 200
picoseconds (ps)- Skew (delay variation within a group) - Total interconnect skew
less than 400 ps (toavoid clock recovery circuitry or anyjitteraccumulation)
- Small package size- 1" xl" (e.g., integrated 18channels)- External to module dc coupling- Low module power dissipation ofless than 2W.
successofthe lD-ODL is an appropriate lightsource,which weaddress in the nextsection.
Optical Transmitter Source &electionThe three competitive optical transmitter tech
nologies that provide an optical backplane strategy! forbothone- or two-dimensional arraysare:- Light-emitting diode (LED) arrays- Optical modulator arrays- Laser diode arrays.Thesystemperformance and costofeach is discussed inthe sections that follow.
LED Amlys. LEO arrays" have beenregardedasattractive, economical optical transmitters becausetheyare inexpensive to manufacture; highly reliable; easilyextendible to large, two-dimensional (i.e., surfaceemitting) arrays; and morestablein temperature thanotheroptical methods. Unfortunately, LEOs have a fewdisadvantages. Their low external power efficiency (a
coupled power efficiency ofabout0.2 percent, with anintegral lens) manifests itselfin unwanted chippowerdissipation, therebylimiting the width ofthe LEO array.LEOS also havea longcarrier recombination lifetime,which controls their modulation bandwidth (neglectingparasitic capacitance). Resonant cavity ("super") LEOsrepresenta possible upgrade in LEO performance. Theyimprove spectral width byan order ofmagnitude,therebyreducing the chromatic dispersion infiber.Quantum wells alsoimprove LED bandwidth. Despite allthis, LEOs havea fundamentally longercarrierlifetimethan injection lasers,and,moresignificantly, cannot supply the lightpower necessary for the lD-OOL.
OptIcal Moduletor Amlys. Another ofthe opticaltransmitter technologies is the classofoptical devicescalled modulators. These devices havean input lightpower sourcedirected to each modulator andan outputconnection that directs lightto the receiver. The lightismodulated either with an electrical signal (e.g., IiNb0 3
AT&T TECHNICAL JOURNAL • SEPTEMBER/OCTOBER 1993 41
signal, or theycan operate transparently. Examples ofdata-transparent optical modulators include deformablemirrordevices," LiNbO 38 liquid crystal devices (electrooptic effect),9 and magneto-optic devices. 10 The selfelectro-optic effect device (SEED) 11 is an example ofthelatter type oflatchable modulator.
Although a modulator-based lD-ODL isunlikelyto be available soon, researchon modulators andadditional related topics is underway, and may offer futureadvantages.
LaHr Arrays. In recentyears, laserarrays havebeen considered foruse as optical transmitters. Onedimensional arraysofedge-emitting lasers (EELs) 12.13
and two-dimensional arraysofsurface-emitting lasers(SELs) ,14.15 in both GaAs and InPmaterial systems, havestimulated the searchfornewnonlinear devices withbeneficial characteristics for lD-ODL applications. Lasersare preferred over LEDs becausetheyhave:- Higherexternal power efficiency- Anarrow output radiation conethat allows simple butt
coupling offibers- Modulation capabilities that extend into the giga
hertz16 range (reduction ofcarrier lifetime with stimulatedemission).
There are three fundamental ways ofmodulatinga laser: directcurrent drive (with or without pre-bias),external cavity modulation (e.g., LiNbO 3 or other electrooptic devices), or intra-eavity modulation. Different systemspecifications (orperformance specifications) anddevicelimitations dictate the type ofmodulation that canbe used.Ifdesigners use pre-bias modulation, or an external modulation schemein which the contrastis low, theymust"code" the transported data to set the properthresholdadjustment forthe receiver. This resultsinan ac-eoupledsystem. Foran ac-eoupled systemto work, the thresholdlevel at the receiver mustbe adjusted automatically usingautomatic gaincontrol. Aback-facet monitor at the transmittercontrols the sourcepower. To reducethe complexityand associated cost,a moredigital approach (i.e., decoupled systems) is desirable. Direct modulation producesa dc-eoupled system (i.e., logic 0 equals notransmitted photons, which results ina highcontrast).
Currentresearchon digital applications oflaserarraysfocuses on direct-drive lasers," driven from alogic family such as emitted coupler logic (ECL) , withoutcurrent pre-bias. The philosophy behind the transmittermodule electronic design centerson the directdigital
1
Input currentpulse shaped
Pulse Stepshaped
0.5Time (nanoseconds)
o
3.2
4
i.s.8l;::: 24~ ..S:
fp:0 1.6GlC.::;,
80.81---+---+-------4--+.----..
or InPdirectional coupler), an optical signal, a magneticsignal, a mechanical signal, or a combination ofallofthese. Usually, the lightpower sourceis carriedinafiber-optic cable. Onceinside the lD-ODL, the lightispower divided and routed to each modulator. This allowsthe continuous wave (cw) high-power lightsourceto bephysically separate, unconstrained by sizeor power, andcapable ofproviding reliability through redundancy. Theincident cwlightis directed on the optical device, whichcan then reflect or transmitlight. The modulated outputlightis carriedoutofthe lD-ODL in waveguides. Thesemodulators can either interactwith, or latch, the data
Figure 3. A simulated "optical eye diagram" using two different current pulse shapes to support a 1-Gbltjs datarate. The bright red line represents a step function currentpulse with t r =t f =250 ps and a 2D-mA magnitude. Theother line Is a shaped current pulse with a 4D-mA Initialspike settling back down to the 2D-mA level. Pulse shapingrelaxes oscillation peaks and reduces settling time. Thisdelay, together with relaxation oscillations, dictates thelaser bandwidth.
42 AT&T TECHNICAL JOURNAL • SEPTEMBER/OCTOBER 1993
drive ofthe laser.The electronic current driverdeliversa stepofcurrent to the laser, in which the magnitude ofthe current pulse in the "on"state is usually twoto tentimes greater than the lasingthreshold.This is donewithout feedback (i.e., back-facet monitors).
Twodesignconstraintsimmediately arisewhenconsidering this strategy.First, the magnitude ofthe current pulsemust be large enough to deliver the necessaryamount ofoptical powerin the worst-case conditions(i.e., high temperatureand near-end of life), and largeenoughto minimize the turn-on delay. Second, becausethe laser is not pre-biased above threshold, the turn-ondelay and jitter can be excessive. This obstaclemaybeaverted usingappropriately shaped current pulsesandengineering the laser propertiesto achieve the optimalcoating design,cavity length, active area design, etc.Maintaining a lowlaser threshold current (while stillmaintaining high outputpowerlevels) maximizes signalfidelity and minimizes transmitterpowerdissipation.
Figure3 showsa simulated "optical eye diagram" using twodifferent current pulseshapes to supporta 1-Gbit/s data rate. Pulseshapingreduces oscillationpeaksand settlingtime.This delay, together withrelaxation oscillations, dictatesthe laser's usefulbandwidth. The uniformity ofturn-on delayacross the array(i.e., skew) alsolimitsthe usablebandwidth in synchronousapplications. Decreasing the laser's cavitylength and improving facetreflectivities decrease theturn-on delay. Ahigh-bandwidth, direct-drive laser arraydeveloped byAT&T Bell Laboratories consistsofa 12laser InP/GaInAsP array18 in which the lasingthresholdis controlled to 9.9±0.9 rnA, the optical poweris 8.6±0.4mW per facet, the bandwidth is greater than 4 gigahertz(GHz) (at3 mWofoptical power), and the crosstalkisless than -26 dB. Similar reports on InP-based laserarrayshavebeen madeby IBM 19 and ORTEL in GaAs,andby NTI20 and lincoln LabS.21The conclusion drawnfrom this simulation is that direct-current modulation,with pulseshaping, can result in an excellent eye diagram at 1 Gbit/s, in which a digital logiczero correspondsto no emittedphotons.
Atemperature-stable laser - one that does notrequirea thermal electriccooler- costs less and is morereliable. To increasethe temperaturestability, designerscan lowerthe threshold currents and/or change the materialsystemto one witha widerbandgap. GaAs lasers aremore temperaturestablethan InP lasers, owing to the
improved To constant, that is Ithoc exp(TITo), and typicalvaluesfor InGaAsP and AlGaAs are 50°C and 120°C,respectively. In addition, the gain ofa GaAs laser islarger than that ofan InP laser, indicating that GaAslaser arrays shouldbe superiorto InP laser arrays. However, InP lasers are currentlymore reliable than GaAslasers,22 because they sufferless facetdeterioration withtime,fewer dark linedefects, and less nonradiativerecombination in the cavity. As GaAs laser technologyimproves, the reliability difference shoulddisappear.Future switching systems, which will havelower-costoptical connectorsand plastic fibers, will use evenshorter-wavelength optical sources.Although the opticalinterconnection applications being described inthispaperare for distancesofless than 100m, and for modestbit rates (1 Gbit/s), the wavelength used need notbe dictated by minimum loss or dispersion in the fiber. Thefundamental concernsfor interconnect designersincludelowcost and minimum skew (matched delay) in the fiberribbon. Eventhough minimum loss and dispersion in thefiberdo not controlthe wavelength, wavelength doesbecomean issue at bit rates greater than 1 Gbit/s,and/or at distancesgreater than 1km.Here,InGaAsP/lnP lasers operating at a minimum l.3-mmfiberdispersion havea distinct advantage.
Device and circuitdesignersare currentlydebating the use ofn-substrate versus p-substratelasers.Sincethere are no differences between them infactors such as device processing costs,yields, or characteristics, p-substrate lasers would be preferable, froma circuitdesignperspective, becausethe electronicdriver:- Uses NPN transistors, which havea higherIt- Has a less complex circuitdesign- Has a lowerpowerdissipation.
Recent research in SELs for optical interconnectionapplications alsosuggest that they are an attractivecandidate for the lD-GDL lightsource.First, they canbetested at the wafer level, which lowers the device costand provides a naturalway to evaluate defectmaps.Contrary to popular opinion, however, this does not represent a significant advantage forhigh-volume arrayproduction.
The secondadvantage ofSELs is their potentially high coupling efficiency to fiber, the result ofasingle-mode gaussianoutputprofile. The couplingefficiency can be increasedbecauseofthe single-mode
AT&T TECHNICAL JOURNAL. SEPTEMBER/OCroBER 1993 43
FIgure 4. A schematic and a photomicrograph of a lasergate array.
gate (absorber section) is voltage modulated. The highinputimpedance gate is terminated at 500. Achange inthe gate voltage affects the magnitude ofthe loss (orgain) in the absorbersection, which modulates the output lightintensity. The lightoutputis either in the spontaneous state,wherethe gate voltage is low, or the stimulated emission state,wherethe gate voltage is high.Whenthe anode is de-biased (e.g., I, = 50rnA), there is alargenonlinear regime. Figure 4 shows howsmallchangesin the gate voltage cancausea largechange inlightoutputintensity, producing a largecontrastratio.The absorbersection becomes electrically saturated, andalmost lOO-percent modulation canoccur. Thisenablesthe electrical or optical (with integrated PiN) input pulseto be reconditioned, provide de coupling, andsupport awider temperature change.
These devices alsoexhibitpower gain, and,assuch,may be considered to have transistor-likeamplification characteristics. Forexample, a 3O-~Wchangeofelectrical input power candeliver 7 mWofoptical outputpower. The laser described is anInGaAs/InP buried-heterostructure graded-indexseparate-eonfinement heterostructure (BH-GRINSCH) laserdiode with fourquantum wells.
Conventional lasersoutperform LED andmodulator devices and cost less (if the costofthe receiver functionis included). Compared with direct-drive, lowthresholdlaserarrays, lasergate arrays (LGAs) have:- Wider bandwidth, because ofthe high input
impedance ofthe gate- Lower power dissipation, owing to the lower effective
drive currentand lackofpower dissipation from an ICdriver
- Direct ECL compatibility- Reconditioned pulse shape,resulting inan improved
bit error rate (BER)- Less intersymbol interference- Wider temperature margins (nothermal crosstalk)- Lower delay times, to minimize skew.
Currently, the only disadvantage that LGAs havecompared with direct-drive low-threshold lasers is a marginally increased processcomplexity. Isolating absorberregions and defining gate contact padsare the only additional processing required.
Receiver Design. The receiver module design philosophy contradicts that oftraditional telecommunicationreceiver design strategies. Here, high-incident optical
1000 11m
r! I " I ,''''• I .'~
6
Anode
G~5:qCathode
OL-_"""-'L__~.-LL-_....I.----J
o 50 100Anodecurrent, 'a (rnA)
natureofthe outputpower. SEL technology has progressed enormously to date,especially incertainperformance specifications. ButforSELs to play an active rolein lD-ODLs, they mustdemonstrate:- Direct, dc-eoupled modulation- Highersingle-mode outputpower- Higher"wall plugefficiency"- Proven reliability performance- High performance overa wider temperature rangethan that ofa comparable edge emitter. Althoughresearch on SELs and related topics continues, a SELbased lD-ODL is technologically premature.
Figure 4 showsanother type ofedge emitter,one that uses an intra-eavity loss rnodulator-" to controllightoutput. This FP lasergate is a three-terminal device.The anode (gain section) is de biasedabove its lasingthreshold. The cathode is at groundpotential, and the
44 AT&TTECHNICALJOURNAL. SEPTEMBER/OCTOBER 1993
102 103
Current(~)
-35
Cil -30
:8~:; -25!Ul
~() -20E::JE";
::::E -15
-1070 10
10
Receiver Transmitter
Figure 5. A comparison of lIghtsource technologiesand receiver characteristics for anoptimal data-lInkdesign. The receiverperformance Indicesshown on the left aretranslmpedance gainrequirements to produce a so-mv outputsignal, the translmpedance amplIfierpower dissipation,and the minimumpower/ground supplycrosstalk allowed toprevent oscillation.The receiver powerdissipation was deter·mined by simplyassuming 20 mW pergain stage (at a gainof 10 per stage).
power on the detectorarrayproduces a largecurrent signal to the transimpedance amplifier array IC.Sucha largesignal significantly relaxes the gainand dynamic rangerequirements on the transimpedance amplifier. The transimpedance gaincan be relaxed (e.g., to 3oon), to integratemultiple amplifiers ontoa single tc, This, in tum,will reducesignal delay variations and minimize cost.Otherwise, interchannel crosstalk resulting from the useofhigher-gain amplifiers would prevent this level ofintegration. Forexample, 200 fJWoflightpower absorbed inthedetectorat an 85-percent quantum efficiency will produce an outputcurrent of170 J,!A Atransimpedance gainof300n will produce an outputvoltage of51 millivolts(mY). Generally, the outputofa transimpedanceamplifier is connected to a decision circuit, wheretheinput signal is "regenerated." The 51-mV signal in thisexample is enough to drive the decision circuit. Weestimate a per-ehannel power dissipation of 100 mW on the
module, which generatesa total power dissipation of1.8W on the receiver module.
Figure 5compares plotsofdifferent light sourcetechnologies. The receiver power dissipation was determined bysimply assuming 20mW pergainstage (atatransimpedance gainof 10per stage).The crosstalk wasscaled linearly with gain, anda maximum crosstalk of-30 dBwasused as a reference at a transimpedance gainof2000n (asper private discussions with Dr. R. N. Nottenburg on transimpedance amplifier arraywork at theUniversity ofSouthern California).
The EEL power dissipation wascalculated with85-percent detectorefficiency, 3O-percent quantumefficiency, 3-dB coupling efficiency at the detector andtransmitter, 5-mA threshold current,twice calculated current pulseheight (tominimize turn-on delay), andcurrent derived from a 5V power supply. For the calculatedSEL power dissipation, the assumptions varied from the
AT&T TECHNICAL JOURNAL. SEPTEMBER/OCTOBER 1993 45
Figure 6. 1D-ODL
transmitter. (a) Optical silicon subassembly connecting toMAC-II connectorplug. (b) Modulepackaging scheme.
Wirebonds
MAe-II plugretaining screw
(a)
Laser array chip
MAC-II plug(fiber array)
Copperspacer t 'PQFP heat I •spreader j---0.5--+I
Plug-socketinterface
(b)
EELonly in two areas: a la-percentquantum efficiencyanda 6-dB coupling loss,to account forspacial holeburningeffects. These parameters wereplotted againstthe output current exiting a detectorarray. With a detector current closeto 500~ the receiver and transmitter(EEL) power dissipation balance to approximately 90mW.
Clearly, it is desirable to lower the amount ofdetectorcurrent,eventhoughthis results ina slightimpairment ofthe power balance with respect to the EEL.The evidence suggeststhat EELsshould be the predominantlightsourcetechnology for lD-GDL applications.They provide a lower power dissipation and an overwhelmingly widertolerance ofpower/groundcrosstalkthan do other technologies at this time.
PlICkage De..... for Perellel Optics. To successfullyintroduce newconcepts intoproduction-grade products,it is essential to design and develop an appropriate packagingscheme. It is advantageous to blendcomponentscurrently in production (orextensions thereof) with aminimum oftechnological development, so products canbe developed fasterand with less risk.The lD-ODL package designis described in the next two sections. Thefirst section details the subassembly, wherethe optical
46 AT&TTECHNICALJOURNAL. SEPTEMBER/OcroBER 1993
components are housed, and the second sectiondescribes how the optical subassembly is packaged.
Silicon Submount De...... The lD-ODL package wasdesigned-' as a connectorized transmitter (orreceiver)module. The package wasassembled within the tolerancesneededto couple lightfrom a laserarrayinto amultimode fiberarray. Astrictly passive alignmentscheme25•26 wasused,one inwhich the module wasdesigned to fit inside a standard plastic-quad flat pack(PQFP). To makethe design compatible with existing ribbonconnectors, designers choseAT&Ts MAC-II,27 an18-wide fiberribbon connector. Consistent with the fiberarrayconnector, the laser (and PiN) subassemblies usedsilicon. Silicon also provides a good thermal pathforheatdissipation andcanbe manufactured to lithographicaccuracies.P Figures6aand b, respectively, show thealignment methodology andthe lD-GDL transmitter modulepackaging scheme. The alignment between the multimode fiberarrayand the laserwasobtained by springloading the alignment pinsin the fiberarray (MAC-IIplug) intothe alignment v-grooves in the transmittermodule. The alignment v-grooves, located along theedges ofthe transmitter module, were formed by
Rgure 7. (a) Assembled transmittermodule without Its RTV
or lid attachment.(b) Eye diagram of apseudo-random pattem(2 15 - 1) NRZ Inputsignal transmittedthrough the 1D-ODLat622 Mblts/s.
precisely etchingthe silicon submount and the cap. Thebonding padson the submount (corresponding to the pcontacts ofthe laser array) are located within lithographic accuracies ofthe etchedv-grooves.
The laserarraychipconsistsof 18lasers locatedon25O-Ilm centers. It is bondedontothe silicon submount usingAu/Sn solder, with the epitaxial sidefacingthesubmount. The solderbonding padson the siliconsubmount in Figure6a,located at the end ofelectricalinterconnection lines, are used fordriving the lasers.Thesolder bumpheightwasoptimized to provide a goodmechanical joint, to dissipate the heat generatedby thelasers, and to allow the laserarrayto self-align duringthesolder reflow process. Aresidue-free active atmospheresolder reflow process.F' which does not requireanypost-eleaning, wasused to keep the laser's surface free ofcontamination. Figure7ashows a photograph ofthecompletely assembled transmitter module.
HybrId De....... The package forthe lD-ODL isbased on a family ofmultilayer hybrid integrated circuits,called POLYHICS,30,31 developed byAT&T. Precisionthin-film resistors, fabricated on the ceramic substrate,
(8)
(b)
are placed closeto the ends ofsignal traces to provideexcellent termination ofhigh-data-rate lines. The leadframe supports data ratesgreater than 1 Gbit/s. Much ofthe POLYHIC package wasdesigned to minimize developmentcostand interval, which wascontrolled byincorporating pre-existing, premolded leadframe anda quadflat-pack geometry. For an initial predevelopment product, the package configuration wassatisfactory.
lD-ODL Performance. How well anyinterconnectiontechnology performs ina largedigital systemdesignis a function of:- Signal density (on/offofthe PCB and its footprint)- Signal bandwidth- Power dissipation- Interconnection lengthlimitations- Costper interconnection- Skew- Signal groupdelay variations- Biterror rate- Reliability.The next two sections describe the performance ofthetransmitter and receiver modules in the lD-ODL.
AT&TTECHNICALJOURNAL.SEPTEMBER/OCfOBER 1993 47
Transmitter. The 18-wide laser diode arraysusedin the lD-ODL wereN-substrate, FP, InGaAs/InP (A= 1.3urn), bulk-active arrays, with PN junction-blocking layers.Laserfacets, on 250-J..lm centers (tomatch the fiber-ribbonconnector), havea cavity lengthof250 urn, The -3 dBbandwidth ofthese lasers is greater than 2 GHz. Since thearraysare "hightemperature" lasers,there is no needforthermal electric coolers. These lasers,whose thresholdcurrent is in the 10- to 15-rnA range, with reasonable uniformity, can supply several milliwatts oflightpower at85°C. "Butt" coupling ofthe lasers to the MAC-II connectorreducesthe amount ofpower to the fiberby no morethan3 dB. About 1mWis the the optimal amount ofopticalpower that should be coupled to the fiber.
Receiver. The 18-wide detectorarraysused in thelD-ODL are N-substrate, bottom-illuminated, InGaAs PiNdiodes, whoseoptically active region is 75urn in diameter on 250-J..lm centers.Detectorarraysare mounted ontopofa turningmirror (anisotropically etched on the siliconsubmount), to "up-reflect" the lightfrom the buttcoupled MAC-II connector. Loss ofoptical power from theMAC-II to the detectorarrayis less than 3 dB.
Figure 7bshowsan eye diagram fora channelusing lD-ODL modules. The receiver function wasimplemented with discretedevices (Hewlett Packard transimpedance amplifiers and decision circuits), usinganinput synchronous optical network datarate of622Mbits/s (OC-12). The patterngeneratorproduces apseudo-random patternlengthof215 - 1 ofnon-return tozero (NRZ) data.
ConclusionsAdvanced switching (and computer) system
architecture designsrequirehigher performance andlower-eost packaging technologies to produce newproductsandenhanced services. Electronic interconnectiontechniques - usingtechnologies such as controlled collapsechipconnect (C4) and multichip modules (MCMs)- offer performance superiorto other methodsat a costthat is lower for short-length interconnections (e.g.,chip-to-chip packaging). Optical interconnection strategies are useful at higher levels ofpackaging (i.e., boardto-board, shelf-to-shelf, andframe-to-frame level ofinterconnections). Furthermore, parallel optical datalinkswith high datarates now offer greater flexibility andlower cost than time-multiplexed serialdatalinks.
Ofthe numerous techniques used to implement
48 AT&TTECHNICALJOURNAL. SEPTEMBER/OCTOBER 1993
a parallel datalink, the lD-ODL may be the first to provide high performance at low cost. Designers agreethat the keyconcepts of lD-ODL technology include anECL-eompatible dc-eoupled system, highoptical outputpower andcontrastthat vastly simplify the receiver circuitdesign, improved skew, lower total power dissipation, small PCB footprint, and low thermal impedance athigh temperatures.
AcknowledgmentsWeespecially thankthese people from AT&T
Bell Laboratories: M.Cappuzzo andJ. Shmulovich(Murray Hill), forpreparing the silicon submounts;J. Zilko (Solid StateTechnology Center), forprovidinglaserarraychips; R. Frahm (Murray Hill), forprovidingdetectorarraychips; D. Buchholz and R. Huisman(Indian Hill), forhybrid design andexperimental measurements; N. Basavanhally, M.Brady, H. Nguyen, andR. Roll (Engineering Research Center), fordesigningand performing the silicon submount bonding andassembly operations; andother members ofthe lD-ODLteam, toonumerous to name, whose efforts weregreatly appreciated.
References1. R. A Nordin, etal.,"A Systems Perspective on Digital lnterconnec
tion Technology," IEEE]. Lightwave Technology, Vol. 10,No.6,June 1992, pp. 811-827.
2. Y. Sakurai, etal, "Integrated Experimental System for BISDN," ISS'92, Proceedings ofthe XlVInternational Switching Symposium,Yokohama, Japan, October 25-30, 1992, Vol. I, pp. 209-213.
3. J. F. Ewen, etaI., "GaAs Fiber-Optic Modules for Optical Data Processing Networks," IEEE]. Lightwave Technology, Vol. 9, No.. 5,December 1991, pp. 1755-1764.
4. J. W. Parker, "Optical Interconnection for Advanced Processor Systems: AReview of the ESPRIT II OLIVES Program," IEEE]. Lightwave Technology, Vol. 9, No.5, December 1991, pp. 1765-1773.
5. M. K Kilcoyne, etal., "Optoelectronic Integrated Circuits for HighSpeed Signal Processing," Optical Engineering, Vol. 25, No. 10,October 1986, pp. 1119-1126.
6. KKaede, etaI., "12-Channel Parallel Optical Fiber TransmissionUsing aLow Drive Current 1.3 mm LED Array and aPiN PD Array,"IEEE]. Lightwave Technology, Vol. 8, No. 6,June 1990, pp. 883-887.
7. D. R. Pape and L.J. Hornbech, "Characteristics of the DeformableMirror Device for Optical Information Processing," Optical Engineering, Vol. 22, 1983, pp. 675-681.
8. W. A Payne and H. S. Hinton, "System Considerations for thelithium Niobate Photonic Switching Technology," Proceedings ofthe First Topical Meeting ofPhotonic Switching, Incline Village,Nevada, March 18-20, 1987, pp. 196-199.
9. A R. Tanguay, "Materials Requirements for Optical Processing andComputer Devices," Optical Engineering, JanuaryIFebruary 1985,pp.2-18.
10. W.E. Ross, et al., 'Two-Dimensional Magneto-Optic Spacial LightModulator forSignal Processing," Society ofPhoto-Optical lnstrumentation Engineers, Vol. 341, 1982, pp. 191-198.
11. R. A Novotny, et al., 'Two-Dimensional Optical DataLines," Proceedings ofthe43rd Electronic Components and Technology Conference, Orlando, Florida, June 1-4, 1993, pp.790-794.
12. S.Yamashita, et aI., "Low-Threshold (3.2 rnAper element) 1.3 mmInGaAsP MQW LaserArray on a p-type Substrate," IEEE PhotonicTechnology Letters, Vol. 4, No.9, September1992, pp.954-957.
13. N.Yamanaka, et al.,"AGigabit-Rate Five Highway GaAs OE-LSIchipset forHigh-Speed Optical Interconnections Between Modulesor VLSIs," IEEE Journal on Selected Areas in Communications,Vol. 9, No.5,June 1991, pp.689-697.
14. M. W.Maeda, et al.,"Multi-Giga Bit/s Operation of 16-WavelengthVertical Cavity Surface Emitting LaserArray," IEEE Photonics Technology Letters, Vol. 3, No. 10, October1991, pp.863-865.
15. M. Orenstein, et aI., "Matrix Addressable Vertical Cavity SurfaceEmitting LaserArray," Electronics Letters, Vol. 27,No.5,January1991, pp.437-438.
16. H.Z.Chen,et aI., "HighFrequency Modulation ofAlGaAs/GaAsLasers Grown on SiSubstrateby Molecular Beam Epitaxy," AppliedPhysics Letters, Vol. 52,No.8, February1988, pp.605-606.
17. K.Y. Lau, et al.,"HighSpeedDigital Modulation of UltralowThreshold « 1 rnA) GaAs Single Quantum Well Lasers WithoutBias," Applied Physics Letters, Vol. 51,No.2, July 13, 1987,pp.69-71.
18. L. A. Koszi, et aI., "1.5 mm InP/GaInAsP Linear LaserArray withTwelve Individually Addressable Elements," Electronic Letters,Vol. 24, No.4, February1988, pp.217-219.
19. E. Marc1ay, et aI., "Scaling ofGaAslAlGaAs LaserDiodes forSubmilliampere ThresholdCurrent,"Electronic Letters, Vol. 25,No. 14,July 1989, pp.892-894.
20. M.Nakao, et aI., "1.55 mmDFB LaserArray with 1/4Shifted FirstOrderGratings Fabricated by X-ray Lithography," Electronic Letters, Vol. 25,No.2,January 1989, pp. 148-149.
21. D.Yap, et aI., "HighPerformance InGaAsP/InP Buried Heterostructure Lasers and Arrays Defined by IonBeam AssistedEtching," Applied Physics Letters, Vol. 52,No. 18, May 1988,pp.1464-1466.
22. F. R. Nashand R. L. Hartman, "Accelerated FacetErosion Formationand Degradation of (AI,Ga)As Double-Heterostructure Lasers,"IEEE]. Quantum Electronics, Vol. QE-16, No. 10, October1980,pp.1022-33.
23. J O'Gorman, et aI., "Dynamic and Static Response of Multielectrode Lasers," Applied Physics Letters, Vol. 57,No. 10, September3,1990, p.968.
24. N. R. Basavanhally, et al.,"LaserOptical FiberConnector,"U.S. PatentNo. 104892, 1992.
25. M.S.Cohen, et aI., "Passive LaserFiberAlignment by IndexMethod," IEEE Transactions in Photonic Technology Letters, 1991,p.985.
26. C.A. Armrento, et al.,"Passive coupling of InGaAsP/InP LaserArray and Single ModeFibersUsing Silicon Wafer Board," Electronic Letters, Vol. 27, No. 12, June 6, 1991, p. 1109.
27. R. Bonanni, "Optical Connector," U.S. PatentNo. 4818058,1989.28. P. E. Petersen,"Silicon as Mechanical Material," Proceedings ofthe
IEEE, Vol. 70, No.5, 1982, pp.422-457.29. R. D.Deshmukh, et aI., "Active Atmosphere SolderSelfAlignment
and Bonding ofOptical Components," Proceedings ofthe International Electronic Packaging Society, 1992, pp.1037-1051.
30. C. Edge,et aI., "Flip-Chip SolderBond Mounting of LaserDiodes,"Electronic Letters, Vol. 27, No.6, March 4, 1991, pp.499-501.
31. C.C.Shiflett, et a\.,"HighDensity Multilayer Hybrid CircuitsMadewith Polymer Insulating Layers (POLYHICs)," Proceedings ofthe International Symposium onMicroelectronics, 1986, pp.481-486.
32. J L. Brandner, et a\.,"Electrical Characterization of POLYHIC, aHighDensity, HighFrequency, Interconnection and PackagingMedium forDigital Circuits," Proceedings ofthe 39th ElectronicComponents Conference, 1989, pp.759-764.
(Manuscript approved October 1993)
RonaldA. Nordin is a distinguished member of technical staffin the Broadband and Photonic Switching Department at AT&TBell Laboratories in Indian Hill Park (Naperville), Illinois. Hedevelops optical interconnection equipment within the switchand to the subscriber for broadband systems. Mr. Nordinreceived a B.S.E.E. from Purdue University, West Lafayette,Indiana, and an M.S.E.E. and Ph.D. from Northwestern University, Evanston, Illinois, all in electrical engineering. He holdsfour U.S. patents, in the areas of optics and high-data-ratepackaging. Mr. Nordin joined AT&T in 1977.AnthonyF. J. Levi was appointed Professor of ElectricalEngineering at the University of Southern California in LosAngeles in 1993. From January 1984 to mid-1993, Mr. Leviworked in the Physics Research Division of AT&T Bell Laboratories in Murray Hill, New Jersey, first as a member of technical staff, and then, in 1988, as a distinguished member oftechnical staff. Mr. Levi's research interests include nonequilibrium phenomena in semiconductors, the scaling ofelectronic and photonic devices to submicron dimensions,and integration of electronic and photonic devices. He holdsnine U.S. patents in these research areas. Mr. Levi receiveda B.Sc. from the University of Sussex, England, and a Ph.D.from Cambridge University, England, both in physics.
AT&T TECHNICAl. JOURNAL. SEPTEMBER/OCTOBER 1993 49