JSSC 2022第4期Memory40nmEmerging Memory

CHIMERA A 092-TOPS 22-TOPSW Edge AI Accelerator With 2-MByte On-Chip Foundry Res

CHIMERA是一款基于40nm CMOS工艺的非易失性边缘AI加速器，支持训练和推理，采用RRAM技术。

40nm CMOS, 0.92-TOPS峰值性能, 2.2-TOPS/W能效

边缘AIRRAM非易失性内存DNN加速器低秩训练

▸首个使用RRAM宏的非易失性DNN芯片，支持边缘AI训练和推理

▸低秩训练算法显著减少RRAM权重更新步骤和能耗

▸ENDURER硬件模块提升RRAM的写入耐久性

Abstract

Implementing edge artiﬁcial intelligence (AI) infer- ence and training is challenging with current memory technolo- gies. As deep neural networks (DNNs) grow in size, this problem is only getting worse. This article presents CHIMERA, the ﬁrst non-volatile DNN chip for both edge AI training and inference using foundry on-chip resistive RAM (RRAM) macros and no off-chip memory, fabricated in 40-nm CMOS. CHIMERA’s DNN accelerator is speciﬁcally optimized for RRAM and achieves 0.92-TOPS peak perform