← 返回 JSSC 论文列表JSSC 2021第4期Memory65nmCIM
Z-PIM A Sparsity-Aware Processing-in-Memory Architecture With Fully V ariable We
Z-PIM是一种支持稀疏性和可变位精度的内存处理架构,用于高效能深度神经网络。
65nm CMOS, 1.0V, 200MHz, 5.294mW
内存处理稀疏性可变位精度能效深度神经网络
▸稀疏性处理:Z-PIM通过零跳过卷积SRAM技术实现高效的稀疏性处理,显著减少无效计算,提升能效比。具体采用通道级数据映射和跳过零权重输入通道的方法,在VGG-16模型中实现95.42%的开关率降低(系统创新)。
▸全可变位精度:采用位串行算术(bit-serial arithmetic)技术,支持1至16比特的完全可变权重精度,通过多周期逐比特乘法降低单周期操作复杂度,兼顾灵活性与能效(方法创新)。
▸层次化位线结构:提出分层位线设计以减少寄生电容,结合电荷复用方案,显著降低卷积层能耗。实测显示该结构在1.0V/200MHz下仅消耗5.294mW,能效达0.31-49.12 TOPS/W(电路创新)。
▸零跳过卷积SRAM:基于定制8T-SRAM单元实现原位AND运算与通道级累加,配合对角线累加SRAM完成空间/比特级累加,最终输出卷积结果。该设计通过读操作流水线化提升66.1%吞吐量(架构创新)。
Abstract
We present an energy-efficient processing-in-
memory (PIM) architecture named Z-PIM that supports both
sparsity handling and fully variable bit-precision in weight data
for energy-efficient deep neural networks. Z-PIM adopts the
bit-serial arithmetic that performs a multiplication bit-by-bit
through multiple cycles to reduce the complexity of the operation
in a single cycle and to provide flexibility in bit-precision.
To this end, it employs a zero-skipping convolution SRAM, which
performs in-memor