基于纳米散射结构的可集成光学神经网络及其逆向设计（英文）-豆柴文库

您所在位置：网站首页 / 基于纳米散射结构的可集成光学神经网络及其逆向设计（英文）.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

文本预览

基于纳米散射结构的可集成光学神经网络及其逆向设计（英文）
Abstract:
Withtherapiddevelopmentofartificialintelligenceanddeeplearning,opticalneuralnetworkshavereceivedincreasingattentionasanalternativetotraditionalelectronicneuralnetworksduetotheirhighenergyefficiencyandhigh-speedparallelprocessingcapabilities.Inthispaper,weproposeanovelintegratedopticalneuralnetworkbasedonanano-scatteringstructureanddemonstrateitssuperiorityininformationprocessingandcomputationalefficiencycomparedtotraditionalelectronicneuralnetworks.Furthermore,weexploretheinversedesignmethodsfortheproposedopticalneuralnetwork,whichcanoptimizethedatatransmissionqualityandnetworkperformance,providingapromisingapplicationprospectforfutureopticalneuralnetworks.
Introduction:
Artificialneuralnetworkshavebeenwidelyusedinvariousfieldssuchasimagerecognition,naturallanguageprocessing,andmachinelearning.Thetraditionalelectronicneuralnetworks,whichrequirelarge-scalecomputationalresourcesandhighpowerconsumption,limittheirapplicationsduetothelimitedprocessingcapabilityandhighenergyconsumption.Incontrast,opticalneuralnetworks(ONNs)haveshownremarkablepotentialforenergy-efficient,high-speedparallelprocessingcapabilitywithultra-fastlightpropagationspeedandtheabilitytoprocessmultiplechannelsinparallel.
Recently,therehasbeengrowingresearchindevelopingintegratedONNswithhighintegrationlevelandlowpowerconsumption,whichcanmeetthedemandsofportableandmobiledevices.Inthispaper,weproposeanintegratedopticalneuralnetworkbasedonanano-scatteringstructureanddemonstrateitsexcellentperformanceininformationprocessingandcomputationalefficiency.
Nano-scatteringstructure-basedOpticalNeuralNetwork:
Theproposedintegratedopticalneuralnetworkconsistsofanano-scatteringstructureandaback-propagationalgorithm.Thenano-scatteringstructurecanrealizetheparallelandnonlinearcomputationoftheinputopticalsignalsbymanipulatingthelight-matterinteractioninthenanoscalerange.Moreover,thestructurecanbeeasilyintegratedwithCMOStechnology,facilitatingthemassproductionandlow-costimplementati