SRAM存算一体芯片研究:发展与挑战
SRAM-based compute-in-memory:status and challenges
叶乐 1贾天宇 1陈沛毓 1武蒙 1黄如1
作者信息
- 1. 北京大学集成电路学院,北京 100871
- 折叠
摘要
人工智能时代对计算芯片的算力和能效都提出了极高要求.存算一体芯片技术被认为是有望解决处理器芯片"存储墙"瓶颈,大幅提升人工智能算力能效和算力密度的关键技术和重要解决方案.SRAM存算一体芯片技术由于其在兼容性、鲁棒性、灵活性等方面的优势,已经得到多个旗舰公司的认可和相关领域的产业布局.本文基于国家自然科学基金委员会第347期"双清论坛(青年)"的讨论内容,回顾SRAM存算一体芯片领域近年来的研究现状和发展趋势,分析并总结了该领域未来的研究需求,凝练关键科学问题并进一步探讨前沿研究方向和科学基金资助战略.
Abstract
The era of artificial intelligence(AI)has placed extremely high demands on the throughput and energy efficiency of AI hardware.Compute-in-memory technology is considered promising to address the"memory wall"bottleneck faced by conventional processors.It is a key technology and an important solution for significantly improving the effective throughput and energy efficiency of AI.Compute-in-memory technology has gained common recognition in the industry,with several industry giants starting research and industry in this field.Based on the 347th Shuang Qing Forum(Youth)of the National Natural Science Foundation of China,this paper summarizes the significant demands in the research of SRAM-based compute-in-memory chips.This study reviews the research status and development trends in recent years,distills the major key scientific problems and challenges,and further explores frontier research directions and funding strategies.
关键词
人工智能/存算一体/SRAM存算/科学问题Key words
artificial intelligence/compute-in-memory/SRAM-based CIM/scientific topics引用本文复制引用
基金项目
国家自然科学基金(92164301)
国家自然科学基金(62225401)
国家自然科学基金(61927901)
出版年
2024
NETL
NSTL
万方数据