Optimizing hardware-software co-design based on non-ideality in memristor crossbars for in-memory computing

被引:0
作者
Jiang, Pinfeng [1 ]
Song, Danzhe [1 ]
Huang, Menghua [1 ]
Yang, Fan [1 ]
Wang, Letian [1 ]
Liu, Pan [1 ]
Miao, Xiangshui [1 ,2 ]
Wang, Xingsheng [1 ,2 ]
机构
[1] Huazhong Univ Sci & Technol, Sch Integrated Circuits, Wuhan 430074, Peoples R China
[2] Hubei Yangtze Memory Labs, Wuhan 450205, Peoples R China
基金
中国国家自然科学基金;
关键词
memristor crossbar; IR-drop; neural network; activation function; hardware-software co-design; EFFICIENT;
D O I
10.1007/s11432-024-4240-x
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
The memristor crossbar, with its exceptionally high storage density and parallelism, enables efficient vector matrix multiplication (VMM), significantly improving data throughput and computational efficiency. However, its analog computing is vulnerable to issues like IR-drop, device-to-device (D2D) variation, and stuck-at-fault (SAF), leading to a substantial decrease in the inference accuracy of neural networks deployed on crossbars. This work presents a hardware-software co-design approach tailored to deal with memristor crossbar non-ideality. We introduce an end-to-end functional array simulator (FAST) for precise and ultra fast end-to-end training, mapping, and evaluation of neural networks on the memristor crossbar. Utilizing the sparsity of the memristor crossbar coefficient matrix, it achieves simulation with low storage and computational resource requirements, dynamically selecting the optimal solution to complete the process. It can also precisely simulate the impact of non-ideal effects such as IR-drop, retention, variation, SAF, and AD/DA precision. Using FAST, we assess memristor crossbar matrix operations under non-ideal conditions, identifying the max throughput and the most energy-efficient crossbar configurations. Additionally, we propose a comparator-based activation function modulation (CAFM) scheme and its corresponding hardware architecture with programmable activation function circuits to address the IR-drop issue, enabling low power and area overheads, resulting in the recovery of neural network accuracy by 54% or more. This is validated within FAST, demonstrating the success of our hardware-software optimization co-design.
引用
收藏
页数:16
相关论文
共 45 条
[1]   Xbar-Partitioning: A Practical Way for Parasitics and Noise Tolerance in Analog IMC Circuits [J].
Amin, Md Hasibul ;
Elbtity, Mohammed E. ;
Zand, Ramtin .
IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, 2022, 12 (04) :867-877
[2]   Design memristor-based computing-in-memory for AI accelerators considering the interplay between devices, circuits, and system [J].
An, Junjie ;
Wang, Linfang ;
Ye, Wang ;
Li, Weizeng ;
Gao, Hanghang ;
Li, Zhi ;
Zhou, Zhidao ;
Tian, Jinghui ;
Gao, Jianfeng ;
Dou, Chunmeng ;
Liu, Qi .
SCIENCE CHINA-INFORMATION SCIENCES, 2023, 66 (08)
[3]   Parasitic-Aware Modeling and Neural Network Training Scheme for Energy-Efficient Processing-in-Memory With Resistive Crossbar Array [J].
Cao, Tiancheng ;
Liu, Chen ;
Gao, Yuan ;
Goh, Wang Ling .
IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, 2022, 12 (02) :436-444
[4]   Energy-efficient computing-in-memory architecture for AI processor: device, circuit, architecture perspective [J].
Chang, Liang ;
Li, Chenglong ;
Zhang, Zhaomin ;
Xiao, Jianbiao ;
Liu, Qingsong ;
Zhu, Zhen ;
Li, Weihang ;
Zhu, Zixuan ;
Yang, Siqi ;
Zhou, Jun .
SCIENCE CHINA-INFORMATION SCIENCES, 2021, 64 (06)
[5]  
Chen P Y., 2017, Proceedings of IEEE International Electron Devices Meeting (IEDM)
[6]   In-memory computing with emerging nonvolatile memory devices [J].
Cheng, Caidie ;
Tiw, Pek Jun ;
Cai, Yimao ;
Yan, Xiaoqin ;
Yang, Yuchao ;
Huang, Ru .
SCIENCE CHINA-INFORMATION SCIENCES, 2021, 64 (12)
[7]  
Duff I S., 1983, Direct Methods for Space Matrices
[8]   Overcoming Crossbar Nonidealities in Binary Neural Networks Through Learning [J].
Fouda, Mohammed ;
Lee, Jongeun ;
Eltawil, Ahmed ;
Kurdahi, Fadi .
NANOARCH'18: PROCEEDINGS OF THE 14TH IEEE/ACM INTERNATIONAL SYMPOSIUM ON NANOSCALE ARCHITECTURES, 2018, :31-33
[9]   Novel Weight Mapping Method for Reliable NVM based Neural Network [J].
Han, L. X. ;
Xiang, Y. C. ;
Huang, P. ;
Yu, G. H. ;
Han, R. Z. ;
Liu, X. Y. ;
Kang, J. F. .
2021 IEEE INTERNATIONAL RELIABILITY PHYSICS SYMPOSIUM (IRPS), 2021,
[10]   Efficient evaluation model including interconnect resistance effect for large scale RRAM crossbar array matrix computing [J].
Han, Runze ;
Huang, Peng ;
Zhao, Yudi ;
Cui, Xiaole ;
Liu, Xiaoyan ;
Kang, Jinfeng .
SCIENCE CHINA-INFORMATION SCIENCES, 2019, 62 (02)