Mitigating methodology of hardware non-ideal characteristics for non-volatile memory based neural networks

作     者：Lixia HAN Peng HUANG Yijiao WANG Zheng ZHOU Haozhang YANG Yiyang CHEN Xiaoyan LIU Jinfeng KANG 

作者机构：School of Integrated CircuitsPeking University Beijing Advanced Innovation Center for Integrated Circuits School of Integrated Circuit Science and EngineeringBeihang University 

出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))

年 卷 期：2025年第68卷第2期

页      面：307-321页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 081201[工学-计算机系统结构] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported in part by National Key R&D Program of China (Grant No. 2023YFB4402405) National Natural Science Foundation of China (Grant Nos. 92064001, 62101018) Joint Funds of the National Natural Science Foundation of China (Grant No. U20A20204) 111 Project (Grant No. B18001) 

主　　题：computing in memory neural network accelerators non-volatile memory hardware non-ideal characteristics software-hardware co-design 

摘      要：Non-volatile memory-based computing-in-memory(nvCIM) paradigm has been extensively studied to boost the energy efficiency of neural network accelerators in edge applications. However, the degradation of inference accuracy induced by the non-ideal characteristics across circuits, arrays, and devices is becoming a crucial issue. In this work, we establish a hardware characteristic behavior model to analyze the impact of nvCIM non-ideal characteristics on neural network *** we propose a hardware aware training and weight mapping correction methods to mitigate inference accuracy *** simulation verification, about 95% inference accuracy degradation is recovered by adopting the proposed mitigation method for various non-ideal characteristics and various neural network models. The feasibility of the proposed method is further proved in an experimental example with a flash-based LeNet recognition system.