摘要
智能化时代,计算任务正从“计算密集型”向“数据密集型”转变,这使得传统冯·诺依曼(von Neumann)架构在算力和能效上面临严峻挑战.阻变器件凭借其高集成密度、高开关速度和低开关能耗等优势,在嵌入式存储、存算融合及类脑计算等领域展现巨大潜力,为解决传统架构挑战提供了创新路径.本文将讨论阻变器件的基本概念、关键电学特性及应用现状,分析其在先进工艺节点下推动嵌入式存储、存算融合和类脑计算发展的重要作用.特别地,本文将讨论阻变器件在赋能存算融合技术及类脑计算核心单元(如神经元和突触等)中的发展现状和挑战,揭示其对智能计算体系发展的支撑作用,并展望未来发展前景.总而言之,阻变器件凭借其独特的优势,为传统计算系统带来变革性创新,成为助力人工智能时代加速发展的基础器件.
In the era of intelligence,computational tasks are shifting from being“computation-intensive”to“data-intensive”,posing significant challenges for the traditional von Neumann architecture in terms of computational power and energy efficiency.Resistive switching devices,with the advantages of high integration density,fast switching speed,and low switching energy,have shown tremendous potential in embedded memory,in-memory computing(IMC),and neuromorphic computing,providing innovative solutions to these challenges.In this work,we review the basic concepts,key electrical characteristics,and applications of resistive switching devices,analyzing their crucial role in advancing embedded memory,IMC,and neuromorphic computing at advanced process nodes.Specifically,we discuss the development status and challenges of resistive switching devices in empowering IMC and the core units of neuromorphic computing,such as neurons and synapses,unveiling their support for intelligent computing system evolution and forecasting future development prospects.In summary,resistive switching devices,with their unique advantages,offer transformative potential for traditional computing systems,contributing to the accelerated development of the AI era.
作者
余杰
李超
张续猛
刘琦
刘明
Jie YU;Chao LI;Xumeng ZHANG;Qi LIU;Ming LIU(State Key Laboratory of Integrated Chips and Systems,Fudan University,Shanghai 200433,China;Frontier Institute of Chip and System,Fudan University,Shanghai 200438,China)
出处
《中国科学:信息科学》
2025年第4期749-765,共17页
Scientia Sinica(Informationis)
基金
中国科学院战略重点研究计划(批准号:XDB44000000)
教育部创新平台项目资助。
关键词
阻变器件
嵌入式存储
存算融合
类脑计算
resistive switching devices
embedded memory
in-memory computing
neuromorphic computing
