近似乘法器研究综述  

Survey of Approximate Multipliers Research

作　　者：李鹏程[1] 黄立波[1] 谭弘兵 杨乾明[1] 刘威 马胜[1] 邓全[1] 王永文[1] LI Pengcheng;HUANG Libo;TAN Hongbing;YANG Qianming;LIU Wei;MA Sheng;DENG Quan;WANG Yongwen(School of Computer,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]国防科技大学计算机学院,长沙410073

出　　处：《小型微型计算机系统》2025年第2期482-492,共11页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62272475,62172430)资助;湖南省自然科学基金项目(2022JJ10064,2021JJ10052)资助;湖南省科技创新计划项目(2022RC3065)资助;国防科技大学先进微处理器芯片与系统重点实验室支持。

摘　　要：人工智能(AI)和物联网(IoT)的迅速发展,对能效提出了更高的要求.在后摩尔时代,通过牺牲精度以换取高能效的近似计算成为解决资源开销问题的重要方法,其中近似乘法器在容错应用中发挥了关键作用.为促进近似乘法器的相关研究,本文对相关文献进行了归纳、分析和总结.首先回顾了传统乘法器的发展、近似乘法器常用评估指标和应用测试方法,然后从近似乘法器的架构设计和通用优化两个方面详细阐述乘法器中近似技术的研究和发展.架构设计侧重于在乘法运算的不同阶段进行结构设计,通用优化强调可以在多种架构设计中发挥重要作用的近似优化技术.最后,对未来的研究方向进行展望和总结.The rapid development of artificial intelligence(AI)and the Internet of Things(IoT)puts higher demands on energy efficiency.In the post-Moore era,sacrificing precision in exchange for efficient approximation became an important way to solve the cost problem,in which approximate multipliers play a key role in fault-tolerant applications.In order to promote the related research of approximate multiplier,this paper summarizes,analyzes and summarizes the relevant literature.Firstly,the development of traditional multiplier,common evaluation indexes and application test methods of approximate multiplier are reviewed,and then the research and development of approximate technology in approximate multiplier are described in detail from two aspects of architecture design and universal optimization.Architectural design focuses on structural design at different stages of multiplication,and universal optimization emphasizes approximate optimization techniques that can play an important role in a variety of architectural designs.Finally,the future research direction is prospected and summarized.

关 键 词：近似乘法器 精度 开销 架构设计 通用优化 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]