← 返回 JSSC 论文列表
📄 下载 JSSC 原文 PDF
JSSC 2023第3期Memory28nmSRAMCIM

A 8-b-Precision 6T SRAM Computing-in-Memory Macro Using Segmented-Bitline Charge

提出一种新型SRAM-CIM结构,提升能效并降低计算延迟。
28nm CMOS, 7.2ns计算延迟, 22.75 TOPS/W能效
SRAM-CIM分段位线位线组合源注入能效优化
创新点1:分段位线电荷共享方案(SBCS)是一种电路创新,通过将位线分段并利用电荷共享原理实现低功耗的乘加运算(MAC),同时保持高信号容限,解决了传统SRAM-CIM中信号衰减和能耗高的问题,实测能效达22.75 TOPS/W。
创新点2:位线组合方案(BL-CMB)属于系统级创新,通过动态合并位线减少ADC数量,在能效(EF)和推理精度之间提供可配置的权衡选项,显著降低了模拟读出电路的面积和功耗开销。
创新点3:源注入局部乘法单元(SILMC)结合两种全局位线开关的电路设计创新,支持SBCS和BL-CMB方案,并在晶体管工艺波动下保持稳定的信号容限,确保8位输入/权重下20位近全精度输出。
创新点4:优先混合ADC(方法创新)通过优化模拟读出架构,在7.2 ns计算延迟内实现高精度量化,同时抑制了传统逐次逼近ADC的面积和功耗代价。
Abstract
Advances in static random access memory (SRAM)- CIM devices are meant to increase capacity while improving energy efficiency (EF) and reducing computing latency ( T AC). This work presents a novel SRAM-CIM structure using: 1) a segmented-bitline charge-sharing (SBCS) scheme for multiply-and-accumulate (MAC) operations with low energy consumption and a consistently high signal margin across MAC values; 2) a bitline-combining (BL-CMB) scheme to reduce the number of analog-to-digital converters (ADC