BGC汽车计算的未来-2024-11-智能网联

June2023

WHITEPAPER

TheFutureofAutomotiveCompute

AreChipletSystemsapromisingtechnologysteponthepathtowardacentralizedstack?

AtAGlance

Ashifttosoftware-definedcarsisdrivingdemandforautomotivecompute

Withsoftware-enabledusecasessignificantlyincreasingthedemandforautomotivecom-

putechips,theautoindustryisundergoingashiftinfocusfromhardwaretosoftware.Asaresult,theindustryisevolvingtowardsinglecentralizedcomputestackswithintheelectricalandelectronic(E/E)architecturesofcars.

Currentstate-of-the-artsolutionscomewithtrade-offs

Tofulfilltheseincreasingcomputerequirements,highperformancecompute(HPC)chipsareneededas,nowmorethanever,automakersdesirearesilientsupplychainfortailoredand

modularSystem-on-Chips(SoCs),whilechipsuppliersneedhighvolumestorecoupup-frontresearchanddevelopment(R&D)efforts.

Multipleapproachesarebeingconsideredonthepathtowardasinglecomputestack

Toaddressthesetrade-offsonthepathtowardasinglecomputestack,theindustryhas

multipleoptionstoconsider.TheseincludeagradualmigrationtowardacentralizedE/E

architecturewithstrongdomaincontrollers,customizedSoCs,andChipletSystems,amongothers.

ChipletSystemsareoneapproachbeingusedtocustomizehardwaredesignsforspecificrequirements

DisaggregatingSoCsintofunctionalChipletstoformChipletSystemsallowsforbothcus-

tomization—forspecificsegmentsormodelsviahighermodularityandscale—andahigherdegreeofintegrationbycombiningdifferentnodesizesononechip,whichisusableas

bridgingtechoreveninfuturecentralizedcomputestacks.

AnewvaluechaindynamiccanbeenabledbyanopenecosystemofChipletSystems

AChipletSystem’svaluechaincontainsnewvaluepoolswithsizeableopportunities,astheChipletSystemleadstoadisaggregationofvaluechainstepswithintheautomotivecomputemarket—whichisexpectedtogrowto~$20billionto$22billionin2030.Ifthevaluechainiscoveredinanopenecosystem,multiplenewplaysareenabled,includingcombiningChipletsfromdifferentsuppliers.

Overall,theautomotivecomputeindustrymayprofitfromnewdynamics

Automakers,Tier-1suppliers,andsmallerchipsupplierswithoutcomputepresencemay

profitfromopenChipletSystems,enablingthemtobeinvolvedinchipdevelopment—po-tentiallycustomizingautomotivecomputechipsinanopenecosystemforincreasedsupplychainresilienceandareducedriskoflock-ineffects.

2THEFUTUREOFAUTOMOTIVECOMPUTE

TheFutureofAutomotiveCompute

ShifttoSoftware-DefinedCarsDrivesDemandforAutomotiveCompute

Currently,theautomotiveindustryisundergoingtremendouschange.Alongwithelectrifica-tioneffortsbyallautomakersglobally,theautomationofdrivingfunctions—whichwilleven-tuallyleadtofullyautonomousvehicles—isoneofthekeyglobaltrends.Weexpectthe

shareoflightvehiclesaleswithAdvancedDriverAssistanceSystems(ADAS)functionsof

Level2orhighertoreachabout50%by2030,meaningthatshareislikelytodoublerelativeto2022.Furtheradvancementsfocusprimarilyonthecockpit,withtheintroductionofmorevirtualanddigitalusecasessuchasaugmentedrealityandgaming.

Thekeytothistrendissoftware.Asoutlinedin

ChasingtheSoftwareDefinedDreamCar,

softwarewillcontinuetoplayacentralroleinautomobiles,asitenablesnotonlyinnovationbutalsodifferentiationamongOriginalEquipmentManufacturers(OEMs)—andalsohasanoutsizedinfluenceoncustomers’purchasedecisions.

Toenablethesecomplexsoftwarefeatures,moreandmorepowerfulchipsarerequired,whichwillleadtoastronggrowth—withanexpectedCompoundAnnualGrowthRate

(CAGR)of22%between2023and2030—intheautomotivecomputesemiconductormarket,asshownin

Exhibit1.

Exhibit1-AutomotiveComputeSemiconductorMarketGrowingto~$20Billionin2030

1

Automotivecomputesemiconductormarket(in$B)

Viewonsemiconductorcontent,notincludingelectricalboards,integration,etc.

>

19.7

16.9

2.1

+22%14.5

3.7

12.2

3.2

9.2

7.4

6.0

7.5

5.0

1.41.8

20232024202520262027202820292030

AD(L4/L5)ADAS(L2+/L3)ADAS(L1/L2)IVI(Basic+Enhanced)IVI(Premium)Other

Source:IHSLightVehicleForecast,BCGanalysis

Note:IncludesSoC,CPU,GPU,Chiplet-integrablememory,MCUs,etc.

1.BasedonIHSautomotivesemiconductormarkettracker2022-11filteredforautomotivecompute,whichincludesmainlySoCsbutalsoGPUs,memory,etc.

BOSTONCONSULTINGGROUP3

Theincreasedneedforautomotivecomputepowerdrivenbyafocusonsoftwareisalso

leadingtoanevolutiontowardmorecentralizedE/Earchitectures.Comingfromanetworkof50–100distributedElectricalControlUnits(ECUs),thenextstepisarchitectureswithdo-

maincontrollersforeachfunctionalcluster(e.g.,autonomousdriving,in-vehicleinfotain-

ment).Hybridarchitecturesarecombiningthisdomain-basedapproachwithzonehubs

clusteringinformationlocallyandsendingittoavehicleserver.Inpurelycentralizedarchi-

tectures,onlyzonehubsareusedtocollectedge-devicedataandsendittoacentralvehicleserver.

2

Inthelongrun,weexpecttheindustrytoshifttowardasinglecomputestackina

centralizedE/Earchitecture—includingamixed-criticalitysafetystacktoaccountforauto-motivespecificrequirements—andwilllikelyapplyahorizontallayeredapproachinordertoreachsufficientsalesvolumes.

CurrentState-of-the-ArtSolutionsComewithTrade-Offs

Sincetheautomotiveindustryis,atanygiventime,evolvingtowardacentralizedE/Earchi-tectureatdifferentspeeds,therequirementsfromautomakersregardingautomotivecom-

putesemiconductors—inparticular,chipsforvehiclecomputers—arediverse.Requirementsmayincludealowtimetomarkettoadaptarchitectureandfeatureinnovationsquickly.Orautomakersmightpreferabroadofferingofchipsupplierswithamplechoicestoavoid

lock-ineffectsandenforcearesilientsupplychain.Further,updatability,exchangeability,andtheavailabilityofasoftwarestackcanreduceadaptionanddevelopmentefforts.Plus,auto-makersmightexpressawishforcustomizationandenvisionhavingchipstailoredaccordingtotheirarchitecturechoicesandcomputepowerneeds.Thus,amodularandscalablesolu-tionisdesiredinordertorealizedifferentimplementations(e.g.,fordifferentplatformsor

regions).Inthisway,onlythenecessaryamountofcomputepowerisintegrated,possibly

improvingthecostper(necessary)performance.Also,customizedcomputechipsmayallowautomakerstobetterdifferentiatethemselvesfromtheirpeers.

Chipsuppliers,ontheotherhand,needtolookattheautomotivemarketfromanotherpointofview.ThedesignofnewchipscomeswithlargeR&DeffortsintherangeofseveraltenstohundredsofmillionsofUSdollars.Sincethepricesofchipsarelimitedbyautomakers’

willingnesstopay,inordertoreachsufficientmarginstorecouptheup-frontR&Defforts,

largesalesvolumesareneeded.Automotivesales,especiallywhenlookingatspecificcar

models,havemuchsmallervolumesthandoothermarkets,suchasconsumerelectronics.Forexample,in2022,smartphonesalesweremorethan10timeshigherthancarsales,

amountingtoroughly66millionunits.Therefore,chipsuppliersaimtomaximizetheirsalesvolumesintheautomotiveindustrytomakethisanattractivebusinesscase.

MultipleApproachesAreBeingConsideredonthePathTowardaSingleComputeStack

Atthispoint,theindustryhastheopportunitytoshapethepathforautomotivecompute.

Eventually,thisdecisionwillalsobeinfluencedbytheE/Earchitectureimplementedbythe

respectiveautomakers,ascertainarchitecturesarebettersuitedtospecificcomputesolutions.Inordertoaddressthetrade-offbetweencost,customization,scalability,

modularity,availabilityofsoftwarestack,resilientsupplychains,andtimetomarketconsid-erations,therearemultipleapproachesautomakersandchipsuppliersmightconsider,

including:

2.Thevehicleserverdoesnothavetobeasingle(physical)deviceandcanbesplitupintoseveralcomputeunits.

4THEFUTUREOFAUTOMOTIVECOMPUTE

1.Gradualmigrationwithstrongdomaincontrollers

Someautomakersarechoosingtoimplementagradualmigrationtowardacentralized

computestackwithstrongdomaincontrollers.Compute-intensivefunctionalclustersare

formedinsuchamanner,meaningthat,forexample,allADASfunctionsareexecutedon

onedomaincontrollerwithadedicatedSoC.Differentupdatetimesandcycles(perdomaincontroller)canbeacknowledgedlikethis,andtheadaptioneffortfromatraditionalE/E

architecturewithdistributedECUislowerthanforthefollowingoptions.

Takingthispath,automakersmaypreferastandardizedcomputesolutionthatisinlinewithchipsuppliersdrivingtowardanindustrysetupofhorizontalplatformsandstandardized

SoCs.Thesechipsuppliersalsoofferpartsofasoftwarestack(intheformofSoftwareDevel-opmentKits(SDKs),etc.)tosimplifydevelopmentforautomakersbyreducingtheirR&D

efforts.

2.CustomizedSoC

InacentralizedE/Earchitecture(hybridorpurelycentralized),automakersmayprefera

morecustomizedcomputesolutionandthereforeinvolvethemselvesinthedesignofSoCsforautomotivecompute.EventhoughtheseSoCsaretypicallybasedonexistingdesigns,

thereisasignificanteffectonthevaluechain,sinceSoCdesignisahigh-marginbusiness(intherangeof~25%–35%).OneexampleofthisapproachisTeslatakinganexistingchipfromSamsungandusingitasabasistoreplaceanSoCfromNVIDIAinitsAutopilot

hardwarefortheTeslaModel3.Inthisway,Teslacannotonlycustomizethedesignfor

strategicpurposes,butalsoactasafablessplayeranddirectlyinteractwithafoundry,whichcutsoutthemarginoftheoriginalchipsupplier.

Despitethissizeableopportunity,chipdesignisacomplexendeavorthatrequireslarge

investmentsinR&D.Tosomedegree,thiscanbemitigatedby“simply”modifyingexisting

designs,butmanyOEMsstillhesitatetoinvolvethemselvesinchipdesignduetothelackofcapabilitiesandresultingbusinessrisk.

3.ChipletSystems

AnotherapproachtodesigningcustomizedchipsisdisaggregatingSoCsintosingleChipletsandpackagingthemwithadvancedtechnologies.Atfirstglance,thismayappeartobea

technologicalmethodforchipdesign,whichcanbeleveragedbythetraditionalsuppliers.

However,thisdisaggregationmaychangethevaluechain,astheindividualChipletscouldbedevelopedandsuppliedbydifferentplayers—therebybringinginnewplaysandopportuni-ties,especiallywhenanopenChipletecosystemevolves.Whileinthelongrunthismay

comewithmanybenefitsforautomakers,suchashighermodularityandscale,therearealsosignificantshortcomingsregardingthistechnologyifnosolutionisfound(e.g.,thepo-tentiallyhighcostsof(advanced)packaging,theneedforaChipletarchitecture,andcom-monagreed-uponstandardsacrosstheindustry).

Whilenoneofthesethreeapproacheshasyetemergedastheclearestpathforward,this

studytakesadeeperlookintothethirdapproach,asChipletSystemsareanoptionthathasgainedincreasedtractionintheautomotiveindustryoflate.

BOSTONCONSULTINGGROUP5

ChipletSystemsasanApproachtoCustomizingHardwareDesignsforSpecificRequirements

ThebasisforChipletSystemsisatechnologytrendfromthesemiconductorindustrythatisnowslowlyenteringtheautomotivedomain.Mercedes-BenzCTOMarkusSchäfer,for

example,recentlyannouncedthat“theindustryurgentlyneedshigh-performance,energy-

efficiency,andcost-effectiveautomotive-gradeChiplets.”3RobertBoschCEOStefan

HartungrecentlyalsointroducedadaptivechipletsystemsforautomotiveattheITF2023.4

InsteadofhavingallpartitionsofanSoConasingledie,inChipletSystems,theSoCissplitintosmallerunits—calledChiplets—andcombinedwithemerginghigh-performancepack-agingtechnologiessuchas2.5Dpackaging

(seeExhibit2)

.Thispracticeisalreadybeing

usedinapplicationssuchasforhigh-performanceserverchipsandinconsumerelectronics(e.g.,inAMD’sEpyc7002(“Rome”)orApple’sM1Ultra).

Exhibit2-ChipletSystemsasanAlternativeAlongsideMonolithicSoC

Illustrative

SensorI/0

7–14nm

CPU&GPUCluster7–14nm

Connectivity

7–14nm

Memory

sub-system

7–14nm

Safety&Security7–14nm

AI/ML

accelerator

7–14nm

MonolithicSystem-on-Chip

AllSoCpartitionsfullyintegratedonsame(low)nodesize

7–14nm

Nodesize

#1#2#3#4#5#6

oIscurrentstandardinSoCdevelopment

Optimalindividualpartitionnodesize

CollectivenodesizeforintegratedSoC

Partition

oEnsureshighperformanceforusecasessuchasAD

oHighdevelopmentandfabricationcost,especiallywithlargerdiesizesandlowernodesizes

Sensor1/10

16–28nm

Chiplet-basedSystems

OptimalChipletnodesizes,integratedviapackagingtech

Connectivity

22–28nm

Safety&Security28–45nm

AI/ML

accelerator

7–14nm

CPU&GPUCluster7–14nm

Memory

sub-system

14–16nm

Packaging

Nodesize

#1#2#3#4#5#6

Optimalindividualpartitionnodesize

=

NodesizeofindividualChiplets

Chiplet

Allowsforsmallerindividualdiesizes

andcombinationofvaryingnodesizeso

Easierupgradeabilityandre-usabilityof

individualchipletswithinsystemderivates

Costefficiencyofpackagingtechunclear

FurtherS/C

contentmaybeintegratedintoChiplet-basedsystem

Source:BCGanalysis

Note:Nodesizesareillustrative,architecturedesignmaydifferdependingonapplication.

3.LinkedInpostbyMarkusSchäfer:

/posts/markus-sch%C3%A4fer_chiplets-

futureofmobility-leadincarsoftware-activity-7011253968311865346-0wlw/

4.

/2023/world/program/itf-world-day-1/program-part-3/stefan-hartung

6THEFUTUREOFAUTOMOTIVECOMPUTE

Therearefivekeyreasonswhythistechnologytrendmightbeaninterestingopportunityalongsidemonolithicdesignsintheautomotiveindustry:

1.Possibilitytomitigateyieldlimitationsbycombiningseveralnodesizes

Inrecentyears,afterdecadesofsuccessfuldevelopmenttowardsmallernodesizes,semi-

conductorplayersappeartohavereachedphysicallimitations,especiallyinensuringeco-

nomicyields.SincemonolithicSoCsaredesignedasfullyintegratedchips,allpartsneedtobeonthesamesmallnodesize.Forchipdesigners,thisleadstoatrade-off,assmallnode

sizesarebeneficialforHPCbutnotidealforcertainotherparts,suchasanalogfunctions.BydisaggregatingthemonolithicchipintoChiplets,thisneedforonenodesizeisremoved,as

eachChiplet’snodesizecanbechosenindividually.Fromayieldperspective,thisisvery

interesting:OnesmalldefectonalargefullyintegratedmonolithicSoCleadstothewasteoftheentirechip.WhereasinChipletSystems,thedefectwillonlyimpactthesingleChiplet

wherethedefectislocated,whileotherChipletsarenotwasted.Overall,ChipletSystemscomewithyieldandcostbenefitsduetothedisaggregationintodifferentnodesizes.

2.Highscalabilityandmodularity

FormonolithicSoCs,changestopartsofachipleadtoare-design(atleastpartly)ofthe

wholeSoC,whichresultsinhighR&Dcosts.InChipletSystems,singleChipletscanbeinter-changedandreplaced.Thisstepispossiblewithoutare-designofthearchitectureaslongasthearchitecturewasdesignedcorrectlyandthemodifications/changesstaywithincertain

limits.ThedevelopmenteffortformodularsolutionsaddressingscalabilityorcustomizationrequestsfromOEMsmightbesignificantlyreducedinthisway.ChipletSystemscanthere-foreserveasbridgingtechnologyinhybridE/Earchitecturesandalsoasanalternative

alongsidemonolithicSoCsinfuturecentralizedcomputearchitectures.

3.Chancetoincreasesupplychainresilienceandlowerrisksforlock-ineffects

BydisaggregatingSoCsintosingleChiplets,thevaluechaincouldbeopenedupbynew

playersbeingabletoengage,asisdiscussedinmoredetailinthefollowingchapter.This

couldleadtoabroaderofferingontheautomotivecomputemarketenrichingOEMs’choic-es.Withmorealternatives,theriskforlock-ineffectscanbemitigatedandsupplychain

resilienceincreased.

4.Longterm:LowerR&Deffortandacceleratedtimetomarket

ChipletSystemswillneedacertaintimetodeveloptheirfullpotential,astheinitialarchitec-turesanddie-to-dieinterfaceswillfirstneedtobedeveloped.When(andif)theseanda

sufficientofferingofChipletsareavailableonthemarketforautomotivecompute,new

ChipletSystemscanbedesignedmoreeasilybasedonexistingmodules.Then,themodular-itywillhelptoreduceR&Deffortsandlowertheexpectedtimetomarket.

5.Longterm:VisionofopenChipletecosystem

ShouldChipletSystemsevolveinthelongrunwithopenarchitecturedesignsandenough

playersarewillingtoparticipateinanopensystemtocoverthewholevaluechain,Chiplet

SystemsmayleadtoanopenChipletecosystem.Thiswouldallowautomakerstomixand

matchChipletsfromdifferentsuppliers,therebyincreasingscalability,modularity,andcus-tomization(contributingtokeyreason#2)aswellastheirsupplychainoptions(contributingtokeyreason#3).

BOSTONCONSULTINGGROUP7

Atthispointinitsdevelopment,therearealsofivekeyshortcomingsandchallengestothistechnology:

1.Specialarchitecturedesignnecessary

ChipletSystemsrequiresystemarchitecturetobecompatiblewithseveralChiplets,possiblyfromdifferentsuppliers.ThiscreatesanadditionalR&Deffortwhosepayoffmaybeunclear,dependingontheapplication.

2.Needend-to-endChipletvaluechaincoverage

InordertoproduceChipletSystems,theentireChipletvaluechainneedstobecovered

end-to-end.SincetheChipletSystemvaluechaincontainsentirelynewsteps(e.g.,theinte-grationofindividualChipletstoChipletSystems),playersforthesestepsneedtobeidenti-fiedtoreachthementionedcompletevaluechaincoverage.

3.Highcostofadvancedpackaging

ChipletSystemsareenabledbyinnovativepackagingmethods,whichensurehigh-speed

low-latencycommunicationbetweenthesingleChiplets.Typically,advancedpackagingmeth-odssuchas2.5Dor3Dpackagingareused.Comparedwithtraditionalpackaging,themanu-facturingcostsoftheseissignificantlyhigher(byafactorintherangeof10orgreater)whichcouldleadtonon-competitivepricescomparedwithmonolithicSoCs(aspackagingcosts

wouldthenaccountforamajorshareofthechipcosts).

4.Needforstandardizationtoreachopenecosystem

TocombineChipletsfromdifferentsuppliers,thestandardizationofinterfacesisnecessary.Chipletsuppliersneedtoagreeonstandardsforthedie-to-dieinterconnects,includingthe

physicallayerandthecommunicationprotocol.Sincetheinterests/incentivesofdifferent

playersintheindustrymaynotbealigned,itispossiblethatnocommonstandardorseveralcompetingstandardsevolve,whichwouldreducemix-and-matchoptions.Theprocessof

standardizationalsoslowsdowndevelopment,increasingbothcostsandtimetomarket.

5.Longterm:Needforresponsibleorchestratorinopenecosystem

InanidealopenecosystemofChiplets,automakerscandesignchipsbymixingandmatch-ing.Asaprerequisite,anorchestratingplayerisneededtobothberesponsibleforthewholeChipletSystemandensurethefulfilmentofitsrequirements.Thisincludescoordination

withtheChipletsuppliers,assurancethatstandardsaremet,validationoftheChipletSys-temasawhole,etc.,whichmayleadtohighercoststhanformonolithicSoCs.

NewValueChainDynamicCanBeEnabledbyOpenEcosystemofChipletSystems

ThemodularityofChipletSystemsmayopenupthevaluechain,asdepictedin

Exhibit3

.

Generallyspeaking,therearefourmajorstepsintheautomotivecomputevaluechainbasedonmonolithicchips:design,manufacturing,assemblyinvehiclecomputeronaprinted

circuitboard(PCB),andapplicationsoftware.Typically,fablessplayerssuchasNVIDIA,

Qualcomm,andMobileyefocusonSoCDesign,whileplayerssuchasTSMC,GlobalFound-ries,andUMCcoverSoCmanufacturing.AutomotiveTier-1ssuchasBosch,ZF,andConti-

nentaltraditionallytakeovertheintegrationstepinwhichthevehiclecomputerisbuiltusingtheSoC.Theapplicationsoftwareisusuallyprovidedbythefablessplayers,Tier-1s,orOEMs(e.g.,TeslaorVolkswagen/Cariad).

8THEFUTUREOFAUTOMOTIVECOMPUTE

Exhibit3-ChipletTrendWillOpenUpValueChainforNewPlays

SoCDesignSoCManufacturingVehicleComputer(PCB)Software

SoCValueChain

FromonemonolithicSoCthatneedsbundledcompetenciesatfewcompaniestowardsdisintegratedchipletsandintegratorsthatenablescompetenciestobespreadouttomultipleparties

ChipletSysValueChain

ChipletChipletSysVehicle

ManufacturingIntegrationComputer(PCB)

ChipletSysDesign

ChipletDesign

Software

Source:BCGanalysis

InaChipletworld,thefirsttwostepsarenowdisaggregatedupintofoursteps:ThefirststepinvolvesthedesignandverificationofaChipletSystemarchitecture,includingthenecessarymiddlewaresoftware.Inthesecondstep,theindividualChipletsaredesignedsuchthattheyconformwiththedefinedarchitecture,followedbymanufacturingandintegrating(includingautomotivegradevalidation)ontoonechipusinginnovativepackagingmethods—usually

2.5or3Dpackaging—toensuredie-to-dieinterconnectswithhighperformance.The

remain-ingtwosteps(assemblyonPCBandsoftware)staylargelythesameasbefore.5

Whilethedisaggregatedstepscouldbecoveredbythesamefablessplayersasbeforeina

closedvaluechain,ChipletSystemsalsoprovidetheopportunitytoopenupthevaluechain.Theyofferthepossibilitytospreadoutcompetenciestomultipleparties,whichenablesnewplayerstoenterthisvaluechain.Forexample,specialistsforHPC,ML/AI,communications,sensorpreprocessing,security,automotivesafety,andmanymorecaneachdelivertheir

functionalcompetencyasaChiplet,aslongasitconformstotheChipletSystemarchitec-ture.Inthisway,playerswithouttheknow-howtodesignawholeautomotivecomputeSoCin-housecanjointlydevelopachipandprofitfromscalabilityandmodularitybenefitsof

ChipletSystems.Thesenewplayersmightincludesemiconductorplayersthataremore

specializedorhavenopresenceinautomotivecompute,aswellasautomotiveTier-1sandautomakers.

However,suchanopenecosystemofChipletplayerswouldrequiresignificantorchestrationefforts,clearlydefinedstandards,andfullcoverageofthenewdisaggregatedvaluechain.

Regardingthedie-to-dieinterconnectandorchestration,aconsortiuminitiatedbyIntel

calledUniversalChipletInterconnectExpress(UCIe)iscurrentlythemostprominentexam-ple.UCIehasgainedtractionwithmanystrongsemiconductorplayers,andrecentlyalso

madeaconnectiontotheautomotiveindustrywithMercedes-Benz.AnotherinitiativeisthestandardBunchofWires(BoW),whichwasinitiatedbytheOpenComputeProject,anindus-trytradegroupfortheserverHPCmarket.Bothinitiativesaimforanopenecosystemof

Chiplets.Itremainstobeseenwhetherthisendeavorwillbesuccessfulfortheautomotiveindustryoraclosed-systemapproachwillultimatelyprevail.

5.WiththeshifttoChipletSystems,theintegrationofcomponentsandtheirautomotivevalidationmaybeshiftedfromthePCBtotheChipletSystemintegration,sincemoresemiconductorcomponentsmaybeintegratedintoaChipletSystemthanamonolithicSoCDesign.

BOSTONCONSULTINGGROUP9

TheChipletSystem’svaluechaincontainssizeableopportunities,asouranalysisofthevalueaddperprocessingstepin

Exhibit4

shows.ThemarketsizeforChipletSystemsinautomo-tivecomputewillamounttoabout$21billionto$23billionin2030,ofwhicheachofthe

fourprocessingstepswillcontributeabillion-dollaramountinthemid-singledigits.NewplaysarelikelytoevolveinChipletSystemDesignandChipletDesign,whichbothofferattractivemarginsintherangeof25%–35%.

Theintegrationinavehiclecomputer—meaning,onaPCB—addsthelargestamountofvalue(~$14billion–$16billion)butoffersonl