⚡ 本页包含 AI 生成的分析内容，仅供参考
本文提出一种采用16nm工艺的72kb增益单元存算一体宏，支持多种数据格式（MX、LNS、FP等）和可重构的二进制补码与符号数值双表示方案，实现了120.5TFLOPS/W的高能效，解决了通用AI任务中数据表示灵活性需求。
Yao-Kai Yeh1, De-Qi You1, Ashwin Sanjay Lele3, Brian Crafton3, Bo Zhang3, Ping-Sheng Wu2, Ya-Tang Yang1, Chung-Chuan Lo1, Ren-Shuo Liu1, Chih-Cheng Hsieh1, Kea-Tiong Tang1, Meng-Fan Chang1,2 National Tsing Hua University, Hsinchu, Taiwan, 2TSMC Corporate Research, Hsinchu, Taiwan, 3TSMC Corporate Research, San Jose, CA *Equally Credited Authors (ECAs) 1 Abstract Diverse AI workloads require flexible data representations and numerical formats. In this work, we present a reconfigurable 2’s-complement and sign-magnitude scheme integrated within a versatile-format CIM macro supporting MX, LNS, FP, and INT for MAC operations. Implemented in a 16nm 72kb gain-cell array, the macro achieves record energy efficiency (120.5TFLOPS/W) and throughput density (3.18 TOPS/mm2) in MXINT8 mode. Diverse AI workloads require flexible compute-in-memory (CIM) architectures that support multiple numerical formats (e.g., floating-point (FP) and integer (INT)) and data
Jen-Chun Tien*1, Win-San Khwa*2, Le-Jung Hsieh1, Tsung-Han Lou1, Jyun-Cheng Bai1, Yu-Sheng Kao1, Ting-Hao Hsu1, Mai Tseng1, Hung-Hsi Hsu1,