800G光模块白皮介绍_第1页
800G光模块白皮介绍_第2页
800G光模块白皮介绍_第3页
800G光模块白皮介绍_第4页
800G光模块白皮介绍_第5页
已阅读5页,还剩25页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

ENABLINGTHENEXTGENERATIION

OFCLOUD&AIIUSING800GB/S

OPTIICALMODULES

PROMOTERS:

Contents

1.CloudExpansionSetsPaceforOpticalModules01

2.DataCenterArchitectures03

3.8x100GSolutionforSRScenario05

3.1800GSRscenariorequirementanalysis05

3.2TechnicalFeasibilityof8x100Gsolutions05

44x200GSolutionforFRScenario

07

4.1800GFRscenariorequirementanalysis

07

4.2Technicalfeasibilityof4x200Gsolutions

08

4.3Packagingfor4x200Gsolution

09

4.4Forwarderrorcorrection(FEC)codefor200Gperlane

10

5.Possiblesolutionsfor800GDRscenario

11

6.SummaryandOutlook

12

01

www.800G

Sales($M)

2020

20212022

100G200G400G

2023

LIGHTCOUNTING

EnablingTheNextGenerationOf

Cloud&AiUsing800Gb/sOpticalModules

1.CloudExpansionSetsPacefor

OpticalModules

Cloudcomputingandstoragehavetakenoverasthetechnologicalbackbonetoamajorityofourmodernbusinessapplicationsprovidinginfrastructure,platform,softwareorvirtuallyanythingasaservice,andtopersonalappliancescoveringphones,laptopsandvarioussmartdevices.UnlikewirelessinfrastructureandstandardslikeLTEand5G,wherethestandardizationandtechnologyareaheadoftheactualapplications,providinga“builditandtheywillcome”businessmodel,therapidandall-encompassingexpansionofcloudapplicationsandservicesvigorouslypushesthedevelopmentofhigh-techelectronicsandoptics,whichoftenseemtorunbehindthepacesetbytheendusers.Theexponentialresourcegrowthofartificialintelligenceapplicationsandtheinherentneedforhighbandwidthforthetransportofbigdataputsafurtherstrainondatacenterarchitecturesandtheunderlyinginterconnects.Thus,thedeploymentsoftheAIcloud,aregainingmomentum.

Cloudapplications,AR/VR,AI,and5Gapplicationgeneratemoreandmoretraffic.Theexplosivegrowthoftrafficrequireshigherbandwidth.AsshowninFigure1,globalinterconnectionbandwidthcapacitywillgrowata48%CAGRin2017-2021.

WORLDWIDEGROWTH

10,000(Tbps)

CAGR:+48%

6,000

4,000

2,000

0

2020

US.EUAPLATAM

8,000

2018

2019

2017

2021

Figure1–GlobalInterconnectionIndex(Source:Equinix)

AsshowninFigure2,marketanalystsareprojectingafirstadoptionof400GDatacommodulesin2020withalargeradoptionof2x400G/800Gmodulesin2022-23.

$7,000$6,000$5,000$4,000$3,000$2,000$1,000$-

2024

2x400G

Figure2–Projectionofthemarketrevenuefordatacommodules(Source:LightCounting)

Ethernetswitchingchipcapacity

EnablingTheNextGenerationOfCloud&AiUsing800Gb/sOpticalModules

“OurLightCountingForecastmodelindicatesthatoperatorsofClouddatacenterswillneedtodeploy800Gopticsby2023-2024 tokeepupwiththegrowthofdatatraffic,”statedfounderandCEOofLightCountingMarketResearch,VladimirKozlov,PhD.“Mostof800Gwillbestillpluggabletransceivers,butweexpecttoseesomeimplementationofco-packagedopticsaswell.”

DatacentercloudarchitecturesarebeingpacedbythecapacityscalingofswitchingASICs,whichisdoublingapproximatelyeverytwoyears,unfazedbythetalkabouttheendofMoore’sLaw.Today,12.8Tb/sEthernetswitchingchipsarebeingcommerciallydeployedwithfirstchipdesignfirmsalreadyprototyping25.6Tb/ssiliconfordeploymentnextyear,asshowninFigure3.Thisputsfurtherpressureontothedensificationofopticalinterconnects,whichdonotscaleatthespeedofCMOSduetothelackofacommondesignmethodologyacrossthevariouscomponentsandacommonlargescaleprocess.

Inthepastfewyears,therapidexpansionofcloudserviceswasfueledbytherapidadoptionandpriceerosionof100Gshort

reachopticalmodulesbasedondirectdetectiontechnologyandnon-returntozero

(NRZ)modules.Afterthebeginningofthe400GbEBandwidthAssessmentactivity

inIEEEinMarch2011,initialdeploymentof400Gopticalmodulesisfinally

startingin2020withastrongerrampprojectedfor2021,asshowninFigure2.

Infact,intheinitialusecases,400Gmoduleswillbemainlyusedtotransport

4x100Gover500minDR4applicationand2x200GFR4opticsover2km,not

makinguseofthe400GbEMAC.Atthesametime,itseemsunlikelythat

IEEEwouldsoonstandardizethenextgenerationofoptics,suchas800GbE,

meaningthatthestandardizationofhigherdensityopticsforthetransportof

8x100GbEor2x400GbEforthe25.6Tb/sand51.2Tb/sswitchinggenerations

wouldbewellbehindactualmarkettimelinerequirementsof2021-22.This

raisestheneedfor800Gindustryinteroperabilityoutsideoftheestablished

standardbodies.

800G

QSFP112-DD&OSFP32x@1U/64x@2U

25.6T/51.2T

256/512Lanes

100GQSFP28

32x@1U

10GSerdes

25GSerdes

50GSedes

128Lanes

100GSerdes

128Lanes

20132015201720192021-22

400G

QSFP56-DD&OSFP

32x@1U

100GQSFP2864x@2U

40GQSFP+32x@1U

256Lanes

256Lanes

1.28T

12.8T

6.4T

3.2T

Figure3–Historicalevolutionofdatacenterswitchingchipcapacity

02

www.800G

03

www.800G

EnablingTheNextGenerationOf

Cloud&AiUsing800Gb/sOpticalModules

2.DataCenterArchitectures

Thehyperscaledatacentermarketisquitefragmentedwithrespecttotheuseddatacenterarchitecturesorthedemandforpluggableoptics.DatacentersforoperatorswithalargerexternalcustomerbaseofferingXaaSaremorelikelydominatedbynorth-southclient-to-servertrafficandcouldhavemoresmallergeographicalclusters.Ontheotherhand,operatorswithalargeinternaldemandforcloudcomputingandstorageseemoreeast-westtrafficbetweenserversandcouldoperatetheirdatacentersashugeclusterswithahigherradix.Evenincaseofsimilarusecases,theoperatorscandeployindividualflavorsofnetworkarchitectureorhaveasubjectivepreferencetoacertaininterconnectssolutionsuchasPSM4orCWDM4orothercost-downvariantsofthereof,suchas100GCWDM4-OCP.

Onecanderiveatleasttwomaintypesoftypicaldatacentersarchitectures.Figure4showsthecommonabstractionofahyper-scaledatacenteranditsopticalinterconnectroadmap.Ingeneral,thesearchitecturesarelarger,haveacertainconvergencefromlayertolayer,e.g.3:1,andrelyoncoherentZRinterconnectsattheSpinelayer.Animportantboundaryconstraintfor800Gnetworkinginthiscaseisthat200Ginterconnects,albeitnotserial,areusedattheservertoTORlayer,whereastheTOR-leaf/spinelayerwouldtypicallyrelyonPSM44x200Ginafan-outconfiguration.

DC

Scenario4

.....

Spine

Scenario3

.....

Leaf

Scenario2

TOR

.....

Scenario1

Server

Scenario4

(DCI)

TypicalOpticalmoduleevolution

800G

ZR

800G

PSM4/FR4

800G

PSM8/4

200G

AOC

100GQSFP28

DWDM

400GQ-DD

ZR

40GQSFP+

eSR4/LR4

100GQSFP28

CWDM4/PSM4

400GQ-DD

DR4/FR4

Scenario3(Spine-Leaf)

40GQSFP+

SR4

100GQSFP28

SR4/PSM4

400GQ-DD

SR8/DR4

Scenario2

(Leaf-TOR)

10GSFP+

AOC/DAC

25GSFP28

AOC/DAC

100G

AOC/DAC

Scenario1(TOR-Server)

2012

2016

2019

2022

Figure4–Typicalhyperscaledatacenterinterconnectroadmap

Forthetypicalhyperscaledatacenternetwork(DCN),deploying200Gserverswillrequirean800Gfabric.It’satrafficconvergencenetwork,whichdependsonthebalancebetweenservicerequirementsandCapexoptimization.Table1showsthedetailedreachrequirementsdependingontheDCNlayer.

Spine

.....

EnablingTheNextGenerationOfCloud&AiUsing800Gb/sOpticalModules

Table1–DetailedrequirementsofthetypicalhyperscaleDCN

Scenario

ServertoTOR

TORtoLeaf

LeaftoSpine

DCI

Bandwidth

200G

800G

800G

800G

Distance

4mwithinrack;

20mcross-rack

≥70m

100mispreferred

500m/2km

80km-120km

Modulesize

QSFP-DD/OSFP

QSFP-DD/OSFP

QSFP-DD/OSFP

QSFP-DD/OSFP

Figure5showsthedatacenternetworkarchitectureofanAIcluster,withlesslayersthanthehyperscalenetworkduetothefactthatitlacksanyconvergencebetweenthelayers.ThedesignofanAIcloudimpliesdifferenttrafficflowswithmuchlargerbigdataflowsandlessfrequentswitching.

Opticalmodulerateevolution(AI/HPCCluster)

Scenario2

400G

PSM4

800GPSM8

Scenario2

(Spine-Leaf)

.....

Leaf

Scenario1

2*200GE

2*400GE

Scenario1

(Leaf-Server)

Server

2019

2021

Figure5–AI/HPCopticalinterconnectroadmap

FortheAI/HPCclusterDCN,deploying400Gserverswillrequirean800Gfabric.ThisDCNdoesn’thaveanytrafficconvergence,withfasterdeploymentthaninthecaseofFigure4.Table2showsthedetailedrequirements.

Table2–DetailedrequirementsoftheAI/HPCclusterDCN

Scenario

ServertoLeaf

LeaftoSpine

Bandwidth

400G

800G

Distance

4mwithinrack;

20mcross-rack

500m

Modulesize

QSFP-DD/OSFP

QSFP-DD/OSFP

Latency

92ns(IEEEPMAlayer)

92ns(IEEEPMAlayer)

Notexplicitlyshown,butalsorelevant,areDCnetworksforsmallercloudorenterprises,wherethedownstreamtotheserverisdecoupledfromthefan-outratesoftheLeaf-Spinelayerandtypicallyhasslowerserverinterconnectspeeds.

04

www.800G

05

www.800G

MACANDHIGHERLAYERS

RECONCILIATION

400GMII400GMII

400GBASE-RPCS

400GBASE-RPCS

PMA

PMA

400GAUI-4C2M400GAUI-4C2M

PMA

PMA

800GMSA

PMD

PMD

MEDIUM

MEDIUM

EnablingTheNextGenerationOf

Cloud&AiUsing800Gb/sOpticalModules

3.8x100GSolutionforSRScenario

3.1800GSRscenariorequirementanalysis

Fortheclassof100m,theindustryisfacingthebasiclimitationsofVCSELsignalingatspeedsof100G/lane.Here,multi-modetechnologywilllikelyallowforreachesof30-50m,thusonlypartiallycoveringtheSRclass,whichisprimarilyemployedbyChinesehyperscaledatacenteroperators.TheMSAtargetsthedevelopmentofalow-cost8x100GmoduleforSRapplications,coveringthesweetspotof60-100m,asshowninFigure6.Particularly,theMSAisintendedtospecifyalowercosttransmittertechnologywiththepotentialtoleveragesub-linearcostscalingwithahighdegreeofintegration.Suchamodulewouldallowforanearlytime-to-marketdense800Gsolution.Alowcost800GSR8couldalsosupportthepotentialtrendsofanincreasingswitchradixanddecreasingservercount-per-rack,whichmaycombinetofavormiddle-of-the-rack(MoR)andend-of-the-rack(EoR)ortop-of-the-rack(ToR)architectures,byprovidingalowcostserial100Gserverinterconnect.AsshowninFigure6,theMSAwilldefinealowercostPMDforsinglemodefiberinterconnectsbasedon100GPAM4.DuetothelowlatencyrequirementsofSRapplications,KP4forwarderrorcorrection(FEC)willbeusedend-to-endwithasimpleclockrecoveryanddataequalizationunitinthemodule.Finally,theMSAwillspecifyaconnectorforthePSM8moduleswhichenablesafan-outto8x100G.

8x100G

DSP

8x100G

PSM8

Figure6–800GSR8blockdiagrams

3.2TechnicalFeasibilityof8x100Gsolutions

Asmentionedabove,signalingrateupto100Gperlanemaylimittheevolvementofmulti-modefiber(MMF)basedsolutionfrom400G-SR8to800G-SR8.BasedonthetheoreticalmodelusedinIEEE,wecanreckonthatthetransmissiondistancethatMMFcansupportisnomorethan50masthebaudrateupto50G(SeeTable3).ThelimitationfactorsarefromthelimitedbandwidthofVCSELandthemodaldispersionofMMF.Withtheoptimizationindevices,fibermediumaswellasenhancedDSPalgorithms,100mMMFtransmissionmayberealizedatthecostofhigherexpense,higherlatency,andlargerpowerconsumption.Hence,in800GPluggableMSA,werecommendthatthe800G-SR8scenarioistakenoverbySMFbasedsolution.

BER

BER

BER

FEC:KP

4

EMLon

linetest

result

EMLBERvs.OMA

10-8-6-4-202

OMA(dBm)

(a)

1.00E-021.00E-031.00E-041.00E-051.00E-061.00E-071.00E-08

1.00E-09

-

SiPh.BERvs.OMA

-10-8-6-4-202

OMA(dBm)

(b)

FEC:KP

4

SiPh.o

nlinetes

tresult

1.00E-021.00E-031.00E-041.00E-051.00E-061.00E-071.00E-08

1.00E-09

-10-8-6-4-202

EnablingTheNextGenerationOfCloud&AiUsing800Gb/sOpticalModules

Table3–FiberchannelbandwidthandtransmissiondistanceofMMFreckonedbythetheoreticalmodelusedinIEEE

Bitrate

50Gbps

50Gbps

100Gbps

100Gbps

SignalType

PAM4

PAM4

PAM4

PAM4

FiberType

OM4

OM3

OM4/OM5

OM3

Fiberchannelbandwidth(GHz•km)

2.301

1.541

2.301/2.377

1.541

Transmission

Distance(m)

100m

70m

50m

35m

IEEEstandards

50G-SR,100G-SR2

200G-SR4,400G-

SR8

Definednow

-

InordertoguaranteetheadvantagesonthecostandpowerconsumptionoftheSMFbasedsolution,reasonablePMDstandardrequirementsareindispensablein800G-SR8.ThePMDrequirementstobedefinedshouldensurethat1)diversetransmittertechniques,suchasDML,EML,andsiliconphotonics(SiPh),canbeappliedinsuchscenario;2)allthepotentialofthecomponentscanbereleasedadequatelytoachievethetargetinglinkperformance;3)keyparametersinPMDlayersshouldberelaxedasmuchaspossible,inthecontextofmaintainingareliablelinkperformance.Accordingtothesethreeprinciples,wewillconductsomebriefinvestigationsanddiscussionsasfollows.

ThepowerbudgetoftheSMFbased800G-SR8solutionwouldbequitesimilarwiththatdefinedinIEEE400G-SR8.TheonlyissuetobedeterminedistheinsertionlossofnewdefinedPSM8SMFconnectors.ItmeansthatthepowerbudgetinSRscenariocanbeachievedwithoutahitchbasedoncurrentlymatureopticalandelectroniccomponentsandDSPASICsusedin400GEopticalinterconnection.Therefore,apartfromspecifyingtheconnectorforthePSM8modules,thekeyissueforthedefinitionofPMDparametersin800SR8scenarioistofindoutthesuitableopticalmodulationamplitude(OMA),extinctionratio(ER),transmitterdispersioneyeclosurequaternary(TDECQ)ofthetransmitterandsensitivityofreceiver.Inordertosettheseparametersintothesuitableposition,thebiterrorration(BER)performanceofthediversetransmittersisinvestigatedandassessed.

DMLBERvs.OMA

FEC:KP

4

DMLo

nlinetes

tresult

OMA(dBm)

(c)

1.00E-021.00E-031.00E-041.00E-051.00E-061.00E-071.00E-08

1.00E-09

Figure7–(a)EMLBERvs.OMAresultsbasedoncommercialavailable400GDSPASICs;(b)SiliconPhotonicsBERvs.OMAresultsbasedoncommercialavailable400GDSPASICs,(c)DMLBERvs.OMAresultsbasedoncommercialavailable400GDSPASICs

Figure7showsthreeBERvsOMAcurvesof100GbpsPAM4signal,whichcorrespondtodifferenttransmittertechnologiesrespectively,asonlineresultsandobtainedusingcommercial400GDSPASICs.Actually,theBERperformancesofEMLandSiPhfor100GperlaneillustratedinFigure7(a)and(b)arewell-knownresultssincethesetwosolutionshavebeenextensivelydiscussedinthepastfewyears.ConsideringrelativelylowlaunchingopticalpowerofSiPhtransmitterandgoodenoughsensitivityofallthreesolutions,theminimumOMArequirementin800G-SR8isrecommendedtoberelaxedappropriately.

06

www.800G

07

www.800G

400GBASE-R

PCS

PMA

8x112G

DSP

4x224G

PSM4

MACANDHIGHERLAYERS

RECONCILIATION

400GMII

400GAUI-4C2M

400GBASE-R

PCS

PMA

PMA

PMA

PMD

400GMII

400GAUI-4C2M

800GMSA

MEDIUM

NewPMD

EnablingTheNextGenerationOf

Cloud&AiUsing800Gb/sOpticalModules

TheBERperformanceoftheDMLinFigure7(c)showsthattheOMAsensitivityinthiscaseiscomparablewiththatinthecaseofEMLorSiPh,eventhoughthecommercialDMLusedinherehasrelativelylowerbandwidththanEMLandSiPh.ThisresultimpliesthatthecommercialDSPASICsusedinpracticehavemuchstrongerequalizationabilitythanthereferencereceiverIEEEdefinedin400GE,andthusitcansupportthetransmitterwithcomparativelylowbandwidthtoachievethetargetingpowerbudget800G-SRrequired.InordertoreleasethepotentialoftheDSPunitadequatelyfor800GSR8PMD,referencereceiverforcompliancetest(i.e.TDECQ)requirestobere-definedtomatchthepracticalequalizationabilityofcommercialDSPs,i.e.moretapsnumbersthancurrentlydefined5tapsaredesired.Meanwhile,consideringtherelativelylowsensitivityrequirementinSRscenarioandrestrictionofthepowerconsumptionofthe800Gmodule,alowcomplexityDSPmodeisrecommendedinfuturemodules.AnotherkeyparameterisERthatisrelatedtothepowerconsumptiondirectly.AlowerERisfavoredaslongasitdoesnotimpactthereliabilityofthelink.Basedontheaboveanalysis,webelievethatalowcostandpowerconsumptionSMF-basedsolutionisfeasibleandpromisingin800G-SR8scenario.

4.4x200GSolutionforFRScenario

4.1800GFRscenariorequirementanalysis

200GperlanePAM4technologyisthenextmajortechnologicalstepforopticalintensitymodulated,directdetectioninterconnectsandwillbethefoundationfora4-lane800Gconnectivity,aswellasanessentialbuildingblockforfuture1.6Tb/sinterconnects.AsshowninFigure8,theMSAwilldefinethefullPMDandpartialPMAlayersincludinganewlowpower,lowlatencyFECasawrapperontopoftheKP4FECofthe112Gelectricalinputsignals,inordertoimprovethenetcodinggain(NCG)ofthemodem.Oneofthekeygoalsofthisindustryalliancewillbethedevelopmentofnewwidebandwidthelectricalandopticalanalogcomponentsforthetransmitterandreceiverassembliesincludingdigital-to-analogandanalog-to-digital(AD/DA)converters.Inordertoachievetheaggressivepowerenveloptargetsofpluggablemodules,theDSPchipswillbedesignedinCMOSprocesswithlowernmnodeandemploylowpowersignalprocessingalgorithmstoachieveequalizationofthechannel.

8x112G

DSP

1234

Mux

4x224G

CWDM4

Figure8–800GFR4/PSM4blockdiagrams

EnablingTheNextGenerationOfCloud&AiUsing800Gb/sOpticalModules

4.2Technicalfeasibilityof4x200Gsolutions

Consideringthatatemperaturecontroller(TEC)isrequiredinLAN-WDM,whichisnotdesiredin200G/lanescenarios,thepowerbudgetwillbeanalyzedbasedonCWDM4.Linkinsertionloss,multipathinterference(MPI),differentialgroupdelay(DGD),andtransmitterdispersionpenalty(TDP)arethecontributionstothelinkpowerbudget.AccordingtothemodelreleasedinIEEEstandards,MPIandDGDpenaltyiscalculatedaslistedinTable4.Inviewoftheincreasedbaudrateof200Gperlane,thedispersionpenaltyisexpectedtobelargerthanthatin100Gperlane.Areasonablesuggestionfortransmitterdispersionpenalty(TDP)is3.9dB.Hence,takingintoaccountthemarginforreceiveragingandcouplingloss,aswellasthetypicallaunchingopticalpowervalueofthetransmitter,wethinkthereceiversensitivityrequiredshouldbearound-5dBm.

Table4–Powerbudgetanalysisof800G-FR4

Description

LinkinsertionlossMPIpenaltyDGDpenalty

TDP

Simulationvalue

4dB

0.4dB

0.4dB

3.9dB

SinceSNRdeterioratesabout3dBcomparedwith100G/laneasthebaudratedoubles,itisexpectablethatastrongerFECisnecessarytomaintainthereasonablereceiversensitivity(~-5dBm)andmarginoferrorfloor.Therefore,asmentionedabove,onthetopoftheKP4,anadditionallowpower,lowlatencyFECasawrapperwillbecarriedoutintheopticalmodule.ThethresholdvalueofthenewFECisdeterminedaccordingtothelinkperformanceandpowerbudgetrequirement.

Linkperformanceof200G/paneispresentedusingsimulationandexperiment.TheparametersofthedevicesadoptedinthelinkarelistedinTable5.TheexperimentalresultshowsthatthereceiversensitivitycanreachthetargetvaluewhilethenewFEC’sthresholdissetto2E-3asdepictedinFigure9(a).However,inthisexperiment,maximumlikelihoodsequenceestimation(MLSE)wasrequiredtocompensatetheexcessiveinter-symbol-interferenceinducedbychannelbandwidthlimitation.ThedashlineinFigure9(a)showsthesimulationbasedonthemodelinwhichthemeasuredparametersofthedevicesusedintheexperimentareadopted.Togetherwithexperimentalresults,simulationsshowthatthesystemislimitedbythebandwidthofcomponents,suchasAD/DA,driverandE/Omodulators.Consideringthathighbandwidthcomponentsareexpectedtobeavailableintheyearstocome,simulationresultsbyusingthesamesystemmodelbutwithexpandedbandwidthisillustratedinFigure9(b).Itshowsthereceiversensitivityof@2E-3canmeettheabove-mentionedrequirementwithonlyFFEequalizationintheDSPunit,whichisinaccordancewiththetheoreticalexpectation.

Table5–Parameterscomparisonbetweensimulationandexperiment

Description

3dBBandwidthofTOSA

3dBBandwidthofROSA

3dBBandwidthofAD/DAEffectiveNumberofBits(ENOB)

RIN-OMA

ChromaticDispersion

TIANoise

PDResponsivity

ValueinExperimentalSetup

42GHz

56GHz*

37GHz

4.5

~-137dB/Hz

7ps/nm

15pA/sqrt(Hz)0.5A/W

EnhancedvalueinSimulation

56GHz

56GHz

50GHz

4.5

-137dB/Hz

7ps/nm

15pA/sqrt(Hz)0.5A/W

*ROSAinexperimentsetupisahigh-bandwidthinstrument.

08

www.800G

09

www.800G

SolutionB

SolutionA

Experimentvs.Simulation

Exp-FFEExp-FFE+M

LSE

Simu-FFESimu-FFE+FEC:2E-3

MLSE

FEC:KP4

-8-6-4-2024

OMA(dBm)

(a)

1.00E-05

1.00E-01

1.00E-02

1.00E-03

1.00E-04

1.00E+00

EnablingTheNextGenerationOf

Cloud&AiUsing800Gb/sOpticalModules

Basedontheaboveanalysis,TDECQisstillsuggestedtobefollowedincompliancetestinginthe800G-FR4scenario.However,FFEtapnumbersofthereferencereceiveradoptedinTDECQmeasurementisanticipatedtobeincreasedtoareasonablevalueandneedstobefurtherdiscussed.Additionally,itshouldbenotedthatiftheabilityoffuturedevicetargeting100Gbaudunderperformsourexpectation,morecomplicatedalgorithms(e.g.MLSE)maybeusedinFR4scenarios,whichimpliesthatanewcompliancemetrologymustbedeveloped.

Simulation@ExpectedParameters

8-6-4-2024

OMA(dBm)

(b)

Simu-FFEFEC:2E-3

1.00E-05

1.00E-01

1.00E-02

1.00E-03

1.00E-04

1.00E+00

-

Figure9–(a)200G/laneexperimentandsimulationresultsmatchwellwitheachother;(b)200G/lanesimulationresult:FFE

equalizationcanmeettherequirementofpowerbudgetwhencomponentbandwidthinlinkisimproved.

4.3Packagingfor4x200Gsolution

Forthe4x200Gmodule,thepackagingforboththetransmitterandreceiverneedstobereconsideredtoensuresignalintegritywithintherangeundertheNyquistfrequencypoint(56GHz).TwopossiblesolutionsforthetransmitterareillustratedinFigure10.SolutionAisatraditionalapproachwherethemodulatordriver(DRV)isclosetothemodulator.Incontrast,inSolutionB,DRVinflip-chipdesignisco-packagedwiththeDSPunittooptimizethesignalintegrityontheRFtransmissionline.Bothofthesetwosolutionscanberealizedbythestate-of-arttechnology.PreliminarysimulationsshowthatSolutionBcanachievegoodresultsandcanensureabandwidthlargerthan56GHz.TheripplesontheS21curveofSolutionAareduetothereflectiononDRVinputandcanbeoptimizedthroughthematchingdesignoftheDRV.Eventually,itisexpectedthattheoverallperformanceofSolutionAcanbefurtherimproved.

-24

-26

-28

-30

-32

-34

01020304050607080

freq.GHz

Figure10–Twopackagingsolutionsforthetransmitter.TheS21simulationputstheRFline(markedinred),thewire-bondingand

modulatorintoconsideration,andthebandwidth@-3dBoftheEMLCOCis60GHz.

KP4

KP4

800G

800G

Legacy

C2M

C2C

C2M

Concatenated

EnablingTheNextGenerationOfCloud&AiUsing800Gb/sOpticalModules

Atthereceiveside,thehighbandwidthphotodiode(PD)withlessparasiticcapacitanceandthehighbandwidthtrans-impedanceamplifier(TIA)areneededtoensurethebandwidthperformanceofthereceiver.Thereisnoobstacletorealizingthesecomponentsbythestate-of-artsemiconductortechnology.Asfarasweknow,somestakeholdersinindustryalreadyputmucheffortindevelopingthesecomponentsthataredesiredtobeavailablein1~2years.Ontheotherhand,theconnectionbetweenPDandTIAisalsocritical.Theparasiticeffectintheconnectionalwaysdegradestheperformanceandthusshouldbecarefullyanalyzedandoptimized.

4.4Forwarderrorcorrection(FEC)codefor200Gperlane

AstrongerFECwithathresholdperformanceof2E-3isrequiredtoachievethesensitivityrequirementof200GPAMreceiver.Figure11illustratesacomparisonbetweenterminatedschemeandconcatenatedscheme.Inthefirstoption,KP4isterminatedandreplacedwithanewFECwithlargeroverhead.TerminationhasadvantagesonNCGandoverhead.Inthesecondoption,aconcatenatedschemekeepsKP4astheoutercodeandcombinesitwithanewinnercode.Concatenationhasadvantagesonlatencyandpowerconsumptionandismoresuitablein800G-FR4applicationscenario.

New100G/laneAUIC2M

8

100G/laneAUI

Legacy

BER<1E-5

BER<1E-5

BER<2E-3

Terminated

KP4

NewFEC

KP4

Concatenated

KP4&newFEC

Figure11–800GFEC:TerminatedFECschemevsConcatenatedFECscheme

SerialconcatenationofKP4andanalgebraiccodeshowninFigure12isastraightforwardsolutiontoachieve2E-3BERthresholdperformance,aswellastominimizethepowerconsumptionandend-to-endlatency,sinceKP4isnotterminated.Noisewithbiterrorratepe<1E-5introducedinC2MelectricalinterfaceistransparenttoPMA.TheoverallperformanceoftheconcatenatedschemewillnotbedeterioratedbypesincepeismuchlowerthanthedecodingthresholdofKP4.HammingcodeswithsingleerrorcorrectingcapabilityandBCHcodeswithdoubleerrorcorrectingcapabilityaregoodcandidatesforthealgebraic

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论