深度CNN模型在嵌入式存算一体架构中的应用  

Application of deep CNN model in computing-in-memory architecture

作　　者：谢宾铭 郑磊 汪林[1,4] XIE Bin-ming;ZHENG Lei;WANG Lin(NARI Group Corporation Information System Integration Company,Nanjing 210000,China;School of Management,China University of Mining and Technology(Beijing),Beijing 100083,China;School of Mechanical Engineering,Southeast University,Nanjing 211189,China;Department of Computer Science,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南瑞集团有限公司信息系统集成分公司,南京210000 [2]中国矿业大学(北京)管理学院,北京100083 [3]东南大学机械工程学院,南京211189 [4]南京航天航空大学计算机系,南京211106

出　　处：《信息技术》2025年第4期73-82,共10页Information Technology

基　　金：南瑞集团有限公司课题项目(524608210271)。

摘　　要：为提升存储和运算性能,将深度CNN模型应用于嵌入式存算一体架构设计之中。改装存储器、运算器和嵌入式处理器的内部结构,加设加速器设备,利用调整电路实现硬件设施的连接。以深度CNN模型作为架构运算业务的执行逻辑,并通过存储模块与运算模块的协同工作,实现存算一体软件功能。结果表明:优化设计嵌入式存算一体架构的存储完整度提高了4.4%,架构的运算速度和吞吐量均得到明显提升,即深度CNN网络模型在嵌入式存算一体架构设计中具有较高的应用价值。In order to improve the performance of storage and computing,the deep CNN model is applied to the design of computing-in-memory architecture.The internal structure of memory,arithmetic and embedded processor is modified,accelerator equipment is added,and hardware facilities are connected by adjusting circuit.The deep CNN model is used as the execution logic of the architecture computing business,and through the cooperation of the storage module and the computing module,the computing-in-memory software function is realized.The results show that the storage integrity of the optimized embedded computing-in-memory architecture is increased by 4.4%,and the computing speed and throughput of the architecture are significantly improved,showing the deep CNN network model has high application value in the embedded computing-in-memory architecture design.

关 键 词：深度CNN网络 嵌入式 存算一体架构 协同工作 吞吐量 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]