光学矩阵器件.docx

光学矩阵器件.docx

《光学矩阵器件.docx》由会员分享，可在线阅读，更多相关《光学矩阵器件.docx（18页珍藏版）》请在冰豆网上搜索。

光学矩阵器件

ResearchStatusandHotspotsofOpticalMatrixCalculationDevices

Abstracts:

Whenwesayaboutmatrix,wearemostlytalkingaboutmultipliers.Vectormatrixmultipliers（VMMs）canbeusedtoperformFourier

transformandsolvelinearequations.Theyhavewideapplicationsinimage

processing,beamforming,radardetection,wirelesscommunication,etc.

Withthefastdevelopmentofmicro-electronictechnologiesduringthepast

decades,theVMMcomputationspeedhasbeenimprovedsignificantly.

HoweverthecontinuousimprovementinperformancesofCMOSchips

encountersbottlenecksinbothphysicalsizeandmanufacturingprocess.As

wellknown,photonicshasadvantagesofbroadband,highspeed,andlow

powerconsumption.Itiseasiertorealizeparallelprocessing.Hence,

photonictechnologiesmayprovideasolutiontosolvethebottlenecks.

Nowadays,researchesonopticalvectormatrixmultipliers（OVMMs）have

becomeahottopicandattractedmoreandmoreattentioninrecentyears.

BecausethemultipliersarethemostimportantdevicesofOpticalMatrixCalculationDevices,

OurfocusismainlyonOVMMs.Wewillintroducethetheoryofmatrixmultiplicationandwecan

seesomehotpotsofOpticalMatrixCalculationDevices.

Introduction:

Traditionalelectroniccomputingtechnologyshows

moreandmoreshortcomingswiththeincreaseof

requirementsforthecomputingcapacity.Optical

computingisafieldwithabrightfuturebecause

ofthenaturalcharacteristicsofphotons,suchasparallelism,highspeed,andhighbandwidth[1,2].Many

achievementsoftheoryandengineeringtechnology

havebeenmadeonthissubjectduringthepastdecades.Theopticalvector–matrixmultiplier（OVMM）

isaveryimportantdeviceintheopticalcomputing

domainasadedicatedprocessor[3].Manystudies

[4,5]focusingonthealgorithmsbasedonvectorand

matrixmultiplicationhavebeenreported,whichimpliestheOVMMcanplayimportantrolesinseveral

fields.

TheOVMMiscomprisedofanopticalsystemand

electricalcircuitsthatprovidethedrivingsignaland

detecttheoutputsignal.Thecoreprocessorofthe

OVMMistheopticalsystem,whichhasbeendemonstratedinavarietyofwayssuchasopticalengineering[6–8],integratedoptics[9],andnonlinearoptics

[10,11].Inthispaper,anopticalmoduleofthe

OVMMbasedonopticalengineeringisproposed.

Itcantransformthewavefrontshapeandcontrolthe

propagationpathofthelightcarryinginformation

withoutcrosstalktoimplementacertainarithmetic

operation.Inordertotransformthebeamshape

precisely,thestructureandtheparametersofthe

systemmustbedeterminedandoptimizedelaborately.Theyarealsoadjustedinconsiderationof

theparametersoftheexistingcommercialoptical

devicesandelectronicdevicesusedinthissystem.

Finally,thesystemshouldbefabricatedaccurately

toensuretheimageryisformedatthecorrectlocationwithacceptableaberrations.

BasicArchitectureandPrinciple

OVMMismainlyusedformatrixmultiplication,vectormatrixmultiplication,matrixmultiplicationoperations.ItsoperationmodeisTypeofStanfordmultiplierstructure.

[34]

Vectormatrixmultiplicationstructureisfunctionallydividedinto:

inputvectors,amatrixlightmodulator,outputvectors.Themainroleoftheinputunitofthelightsourceismodulatedtocarryauniformlightbeamtoinputtotheinformationsignalprocessingunitforeachcolumnofthearray.Themainroleofthesignalprocessingunitisinputopticalsignalstoachievethemultiplicationofthecontrolsignals.Themainroleoftheoutputunitistoconvergethelightofeachrowofthearraytothereceivingunitandoutputstheoperationresultsignals.

Traditionalspaceopticalvectormatrixmultiplierinthestructureconsistsofthefollowingparts:

1）NNon-coherentlightsourcesmakesasourcearray.Itisusedforinputvectors.Usingasphericallensandacylindricalmultiplelenstogetopticalspectralandspatialaggregation.

2）Thematrixthatistobemultipliedisalightmodulatedmatrix.

3）Aftersignalmodulationmatrixoutputsthroughmultiplesetsofsphericallensandacylindricallens,puteverylineoflighttogethertooutputvectorportthatiscomposedofnlightdectectors.Finally,wecanoutputthecalculationresultsthoughanalogtodigitalconverter（ADC）.

Theaboveopticalprocessmodelcanbemathematicallyexpressedasanm×nmatrixAandann-dimensionalvectorobtainedbymultiplyingavectorBtogetam-dimensionalvectorC.Take2×2matrixforexample,inputvectorA（a1,a2）multiplythemodulationmatrixBandyouwillgettheoutputvectorC（c1,c2）.Asshownintable2.1:

Therearesomeothertheoriesandprinciplesthatareveryusefulandeffective.Duetospacelimitations,theyarenotpresentedoryoumayseetheminthefollowingpages.

Hotpotsoftheoryanddevi