← 返回 JSSC 论文列表JSSC 2023第10期Memory28nmEmerging MemoryNeural Network Accelerator
A 28-nm RRAM Computing-in-Memory Macro Using Weighted Hybrid 2T1R Cell Array and
28nm工艺的2T1R RRAM存内计算宏单元,提升边缘智能设备能效。
28nm HKMG, 1-bit IN/3-bit W/4-bit O, 30.34–154.04 TOPS/W
存内计算RRAM能效优化模拟计算边缘智能
▸创新点1:解耦存储与计算数据路径的宏结构(系统创新)。通过分离存储和计算的数据路径,显著减少了寄生效应的影响,提高了输入并行度,从而提升了整体能效比(30.34–154.04 TOPS/W)。
▸创新点2:加权混合2T1R单元阵列(电路创新)。采用2T1R结构结合加权设计,有效降低了器件非理想性导致的信号衰减,同时优化了模拟读出的硬件成本,提升了计算精度。
▸创新点3:MSB冗余子阵列映射方案(方法创新)。通过冗余子阵列映射最高有效位(MSB),显著提升了推理精度(CIFAR-10和CIFAR-100分别提升0.96%和2.83%),同时增强了系统的容错能力。
▸创新点4:参考减法电流感测放大器(RS-CSA)(电路创新)。通过引入参考减法机制,进一步提高了信号检测的精度和稳定性,优化了模拟乘加运算(MAC)的性能。
Abstract
Non-volatile computing-in-memory (nvCIM) can
potentially meet the ever-increasing demands on improving the
energy efficiency (EF) for intelligent edge devices. However, it still
suffers from limited input parallelism due to the parasitic effects,
signal margin degradation due to device non-idealities, and large
hardware cost for analog readout. In this work, we present
a two-transistor-one-resistor (2T1R) resistive memory (RRAM)
nvCIM macro featuring: 1) a macro structure with decoupled
memory a