PCIe clock distribution in embedded systems.docx
- 文档编号:9699327
- 上传时间:2023-02-05
- 格式:DOCX
- 页数:10
- 大小:850.68KB
PCIe clock distribution in embedded systems.docx
《PCIe clock distribution in embedded systems.docx》由会员分享,可在线阅读,更多相关《PCIe clock distribution in embedded systems.docx(10页珍藏版)》请在冰豆网上搜索。
PCIeclockdistributioninembeddedsystems
PCIeclockdistributioninembeddedsystems
ByIanDobsonandJimHolbrook,IntegratedDeviceTechnologyInc-February19,2009
ThePCIe(peripheral-component-interconnectexpress)protocolishighlydesirableforcommunicationacrossbackplanesinembeddedandothersystemtypes.However,foranembedded-systemenvironmentinwhichbackplaneconnectorpinsareoftenatapremium,PCIe’spreferredclock-distributionscheme—usingastarconfigurationofpoint-to-pointconnections—islessthanideal.YoucandistributeaPCIe-compatibleclockusingasinglemultidropsignalandstillmeetthetightjitterrequirementsofthePCIeGeneration2specification.
ClockinginPCIe
PCIeBaseSpecifications1.1and2.0definethreeclock-distributionmodelsforthe2.5-and5-Gbpssignalingrates(figure1,figure 2, andfigure3).Thecommon-clockarchitectureisthecommonmethodforavarietyofreasons.First,mostofthecommerciallyavailablechipssupportingPCIeinterfacesuseonlythisarchitecture.Second,thisarchitectureistheonlyonethatdirectlysupportsspread-spectrumclocking,whichcanbeimportantinreducingEMI(electromagnetic-interference)peakingand,hence,simplifiesthetaskofmeetingelectromagneticemissionslimitsforthesystem(Figure4).Finally,thisarchitectureisthesimplesttoconceptualizeanddesign.
Themostsignificantdisadvantageofthecommon-clockarchitectureistheneedtodistributethereferenceclocktoeachPCIeendpointinthesystem.Theclock’s100-or125-MHzfrequencyandthePCIeprotocol’stightjitterrequirementsfurthercomplicatethistask.For2.5-Gbpsoperation,thelimitis86-psecp-pphasejitterforasamplesetof106samples.The5-Gbpsoperationallimitis3.1-psec-rmsjitter.However,tooperateat5Gbps,atransceiverfirstnegotiatesat2.5Gbpsandthenmovesuptothehigherrateifbothendscandoso.Thatis,ifthesystemsupportsany5-Gbpslinks,thenthereferenceclockmustmeetbothjitterspecifications.
Theseparate-anddata-clockarchitecturesavoidtheselimitationsbutsubstantiallyincreasethecomplexityoftheclock-systemdesignanddon’tsupportspread-spectrumclockingwithouttheuseofsidebandsignaling.Thegoverningspecificationsforreference-clockjitterarePCIeBaseSpecifications1.1and2.0,andPCIeJitter-ModelingRevision1.0DandPCIeJitterandBER(bit-error-rate)Revision1.0detailthemethodforverifyingjittercompliance.Theelectromechanicalspecificationsprovidemechanical-form-factorinformation,electrical-signaldefinition,andfunctionaldefinitions.Someofthesespecifications,suchasCard-ElectromechanicalSpecifications1.1and2.0,alsoprovidejitterbudgetingamongthereferenceclock,transmittingPLL(phase-lockedloop),receivingPLL,andmedia.Strictlyspeaking,theCard-ElectromechanicalSpecificationappliesonlytoPC-,serverATX-(advanced-technology-extended),andATX-basedformfactors.Industrygroupshavepublishedadditionalelectromechanicalspecificationstocoverotherformfactors,suchasMini-Card-ElectromechanicalSpecification1.2formobile-computingplatforms.
Formostembeddedsystems,thesespecificationsprovideguidelinesthatdesignerscanuseinwholeorinparttospecifytheembeddedsystem’sPCIeclock-distributionscheme.Forexample,manyoftheCard-ElectromechanicaldocumentsspecifytheuseoftheHCSL(host-clock-signal-level)protocolfordistributingthereferenceclock.However,manyembeddedsystemsuseLVPECL(low-voltage-positive-emitter-coupled-logic)signalingorM-LVDS(multipoint-low-voltage-differentialsignaling)toachieveagreaterreach,noisemargin,orbothontheirclock-distributionnetwork.
Manyembeddedsystemsdistributealargenumberofhigh-speedsignals,includingclocks,acrosstheirbackplanes.Todealwiththeoften-heavyelectricalloadingonthosebackplanes,thesesignalstendtohavepowerfuldriversand,hence,highedgerates.Thissituationpresentsthedangerofcrosstalkandothersignal-integrityproblems,especiallywhenthebackplanehasalighterloadthantheworst-casedesign.AnotherrelateddesignchallengeisthatPCIespecifiesreferenceclocksof100or125MHz,whicharedifficulttodistributecleanlyoveralong,heavilyloadedbackplane.
InadditiontothePCIespecifications’tightjitterlimitsandneedforalongersignalreach,thenumberofsignalsthatcantransitthebackplaneconnectorsandthebackplaneitselfalsoconstrainembeddedsystems.Definingtheconnectorpinoutsisoneofthemorecriticaltaskswhenspecifyingthesystem.
Common-clock-distributionscheme
Duetotheclockfrequencyandjitterconstraints,mostcommon-clock-architecturedesignsdistributetheirreferenceclocksusingpoint-to-pointdifferential-signalingpairs,oneofwhichgoestoeveryPCIeendpointinthesystem.IfyourdesignhasmultiplePCIeendpointsonasinglecard,youcantakeinareference-clockinputfromthebackplaneandprovideaclock-distributionnetworkonthecardusingzero-delaybuffers.Eventhistaskcanbedifficulttodesign,however,giventhejitterconstraintsof5-GbpsPCIeoperation.
Assumingthatyoucoulddesignsuchon-carddistributionschemes,theystillrequireapoint-to-pointconnectionfromthePCIeroottoeverycardinthesystem.Inembeddedsystems,thisrequirementaddsalotofconnectorpinstotheroot-cardslotsandalotoftraceswithspecialroutingrequirementstothebackplane.Italsomeansthattheslotthattherootcardplugsintohasadifferentpinoutfromthatoftheotherslots.
OneapproachtosolvingtheseproblemsistodividethePCIereferenceclockontherootcardanddistributeitacrossthebackplaneusingamultidropM-LVDSandthentomultiplyittothedesiredfrequencyorfrequenciesonthedestinationcards.Althoughconceptuallysimple,thisapproachistrickytoachievewithinthejitterconstraintsofPCIe(Figure5).
ThisapproachallowsyoutouseanM-LVDSpairtodriveorreceiveaPCIe-compliantreferenceclock.Inmanyembeddedsystems,thecardsoperateasrootsorendpointsdependingontheapplication,theslotassignment,orboth.AcardthatoperatesinonlyoneofthosemodeswouldbesimplerthantheoneinFigure5.Onecardinthesystemwouldactastheroot,generatingareferenceclockmeetingthePCIeconstraintsfromitsonboardcrystal.ThisclockwoulddriveanyonboardPCIedevicesfromaninternalclock-distributionnetwork.Theclockwouldalsogotoanon-PLLdividercircuitthatwoulddivideitfrom100or125MHztothebackplanefrequencyof25MHz.Itwouldthendrivethedivided-downreferenceclocktotherestofthecardsinthesystem.Alltheothercardsinthesystemwoulddisabletheuseoftheironboardclockgenerators,tristatetheirdriversforthereference-clocktraces,andreceivethereferenceclockfromthebackplane.ThisclockwouldmultiplyusingaPLL-basedzero-delaybuffertotherequiredonboardreference-clockfrequencyandthentraveltotheothercards.Thecircuitrythatreceivesandmultipliesthereferenceclockfromthebackplanewouldusuallyresideontherootcardandcouldgeneratethesecondreference-clockfrequency,ifnecessary.ToachievethelowjitterthatPCIerequires,youcanincorporatejitterattenuatorsfortheclocksynthesizerandthezero-delaybuffer.
OneofthemainchallengesofadesignsuchasthisisthatPLLsfilterhigh-frequencyjitterhigherthantheirloopbandwidthbutaddjitteratmodulationfrequencieslowerthantheirloopbandwidth.PLLsalsoinducetrackingskewbecausetheydonotperfectlytrackphaseandfrequencyvariationsofthereference-clockinput.Forabackplane-PCIeimplementationsuchasthisone,whichinvolvestwoormorecascadedPLLsforfrequencygenerationandtranslation,youmusttakegreatcaretominimizephasejitterandPLL-trackingskew.
PCIe-jittermeasurement
Beforedivingintoananalysisoftheperformanceofthisdesign,youmustunderstandtheprocessbywhichPCIeanalyzesjitterperformance.OneoftheoverarchingconcernsofthePCIeJitterWorkingGroupwastoneitheroverspecifynorunderspecifythereferenceclock.Tothatend,thegroupaccountedforthefilteringeffectofthetransmittingandreceivingPLLsandphaseinterpolatoronthereferenceclockandthepeakingeffectsofthesePLLs.
Althoughthegrouphasyettodetailmanyportions,theprocessnowhasfourhigh-levelsteps.First,determinetheaccumulatedphaseerrorforeachcycle.Forserial-datatransfer,theaccumulatedphaseerrorismoreimportantthancycle-to-cyclejitterorperiodjitter,whichareimportantcharacteristicsofparallelbuses.Second,applytheDFT(discreteFouriertransform)totheaccumulatedphase-errordatatochangefromtime-domaintofrequency-domainanalysis.Then,applythesystem-transferfunctiontotheDFToftheaccumulatedphase-errordataandperformaninverseDFTtotransferthefilteredaccumulatedphase-errordatabackintothetimedomain.
YouperformthefilteringanalysisofthePLLsysteminthecomplexfrequencydomainbysettings=jωinthesystem-transferfunctions.Thisequationworkswellforcontinuoussystems,butmostmodernPLLimplementationsarenotpure-analogsystemsbecausetheyhavedigitalcomponents,suchasthephasedetectorandfeedbackdivider;thus,Z-domaindigitalanalysisismoreaccurate.However,briefstudiesbythePCIeJitterWorkingGroupshowedthatS-domainanalysisimposesminimalerror,sothegroupusedS-domainanalysisformodeling.TheS-domainapproximationdeviatessignificantlyfromrealitywhenthereferencefrequencyislessthan10timesthePLLbandwidth,anddesignersmustkeepthatfactinmindwhenselectingaPLL(Reference1).
Jitter-measurementtips
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- PCIe clock distribution in embedded systems