A Flash-SRAM-ADC-Fused Plastic Computing-in-Memory Macro for Learning in Neural Networks in a Standard 14nm FinFET Process

📋 论文概要

该论文提出了一种融合Flash、SRAM和ADC的塑性计算存储宏（CIM），用于神经网络学习，解决了边缘AI设备在低功耗、高实时性推理的同时需要具备在线学习能力的问题。该宏在标准14nm FinFET工艺中实现，支持塑性权重更新以适应动态环境。

💡 主要创新点

• 首次将Flash、SRAM和ADC融合在单个计算存储宏中，实现塑性计算存储架构，支持神经网络在线学习。
• 采用14nm FinFET工艺，在保持高能效的同时实现了可塑性权重更新机制，适用于边缘AI设备。

🏷 关键词

📄 原文摘要

Wang Ye1,2, Hongyang Hu1, Jing Liu1, Jinshan Yue1, Jianguo Yang1, Qing Luo1, Chunmeng Dou1,2, Qi Liu1,3, Ming Liu1,3 Institute of Microelectronics of the Chinese Academy of Sciences, Beijing, China University of Chinese Academy of Sciences, Beijing, China 3 Fudan University, Shanghai, China 1 2 AI edge devices are not only required to perform inference tasks with low power and high real-time performance but are also expected to have the capability to learn and adapt to dynamic and unpredictable environments, without heavily relying on cloud-based training. The recent rise of computing-in-memory (CIM) has offered a competent solution by minimizing the power and latency associated with data movement. While many existing CIM designs [1-6] have primarily focused on improving the performance of AI inference, those with learning abilities have, so far, been relatively less studied. As shown in Fig. 34.9.1, synaptic plasticity (SP) is one key enabler for learnable CIM

👥 作者与机构

Linfang Wang1,2, Weizeng Li1,2, Zhidao Zhou1,2, Hanghang Gao1,2, Zhi Li1,2,