摘要
在处理依托于大数据的计算任务时,冯·诺依曼计算架构的硬件平台存在的内存瓶颈使得计算系统能耗高、速度慢。而利用新型的忆阻器可以实现原位存储和计算,这样的架构减小了能耗和延时,最大化地消除了传统平台存在的内存瓶颈问题。首先阐述基于忆阻器的存内计算的相关技术;然后介绍存内计算的实现方法,包括布尔逻辑和模拟计算两个方面;接着介绍存内计算的实验研究进展;最后讨论基于忆阻器的存内计算研究中现存的挑战和将来的研究方向。
The memory bottleneck of the hardware platform based on von-Neumann computing architecture limits the energy efficiency and processing speed of the computing system when dealing with the computing task centering on large amount of data.The emerging memristor devices can be used for in-situ storage and computation.This architecture reduces energy consumption and delay,and eliminates the memory bottleneck problem of traditional platform.In this paper,first of all,the relevant technology of memristor based in-memory computing is described.Then,the principle of in-memory computing in Boolean logic and analogue computing is introduced.After that,the state-of-the-art of in-memory computing is discussed.And finally,the existing challenges and future research directions of memristor based in-memory computing are discussed.
作者
林钰登
高滨
王小虎
钱鹤
吴华强
LIN Yudeng;GAO Bin;WANG Xiaohu;QIAN He;WU Huaqiang(Institute of Microelectronics,Tsinghua University,Beijing 100084,China;School of Microelectronics,Tianjin University,Tianjin 300072,China)
出处
《微纳电子与智能制造》
2019年第2期35-46,共12页
Micro/nano Electronics and Intelligent Manufacturing
基金
国家自然科学基金应急管理(61851404)项目资助.
关键词
存内计算
忆阻器
非挥发逻辑
模拟计算
in-memory computing
memristor
nonvolatile Boolean logic
analogue computing
