摘要
冯·诺依曼计算架构本质上面临“存储墙”和“功耗墙”瓶颈,近年来摩尔定律的放缓进一步加剧了上述瓶颈.新型存算一体芯片技术通过器件–架构–电路–工艺的协同创新,将数据存储与计算融合一体化,大幅降低数据搬运及其开销,被视为后摩尔时代突破冯·诺依曼架构瓶颈的重要技术方向之一.同时,存算一体芯片通过大规模并行运算方式实现高算力,在一定程度上可以缓解工艺微缩压力,对我国在新一轮人工智能革命中突破算力困境,具有重要的战略意义与应用价值.基于国家自然科学基金委员会第347期“双清论坛(青年)”,本文介绍存算一体芯片领域当前的研究现状,分析了存算一体芯片技术对我国的重要战略意义,展望了未来亟待解决的关键科学问题与技术挑战,并进一步给出了该领域的发展趋势和建议.
von Neumann architecture inherently faces“memory wall”and“power wall”bottlenecks,while the slowdown of Moore’s Law recently has further exacerbated the above bottlenecks.The emerging memory computing technology integrates data storage and computing through a collaborative innovation of the device–architecture–circuit–process,greatly reducing data movement and its related overheads.It is one of the most important techniques for breaking through the bottleneck of von Neumann architecture in the post-Moore era.Meanwhile,the in-memory computing chip achieves increased computability through a highly parallel computing paradigm,which can alleviate the pressure of process shrinkage.It is also an important strategic technology for our country in the artificial intelligence era.Following the discussions and proposals from the 347th“Shuangqing Forum(Youth)”,this paper addresses the current research status in the field of in-memory computing technology,summarizes its strategic significance for our country,looks forward to the key scientific problems and technical challenges that need to be addressed in the future,and provides further development trends and suggestions in this field.
作者
康旺
寇竞
赵巍胜
Wang KANG;Jing KOU;Weisheng ZHAO(School of Integrated Circuit Science and Engineering,Beihang University,Beijing 100191,China)
出处
《中国科学：信息科学》
CSCD
北大核心
2024年第1期16-24,共9页
Scientia Sinica(Informationis)
基金
国家自然科学基金(批准号:62274008)
北京市自然科学基金(批准号:L223004)资助项目。
关键词
存内计算
近存计算
异构集成
异构架构
in-memory computing
near-memory computing
heterogeneous integration
heterogeneous architecture
