Device quantization policy in variation-aware in-memory computing design

Chih Cheng Chang, Shao Tzu Li, Tong Lin Pan, Chia Ming Tsai, I. Ting Wang, Tian Sheuan Chang, Tuo Hung Hou*

*此作品的通信作者

研究成果: Article同行評審

摘要

Device quantization of in-memory computing (IMC) that considers the non-negligible variation and finite dynamic range of practical memory technology is investigated, aiming for quantitatively co-optimizing system performance on accuracy, power, and area. Architecture- and algorithm-level solutions are taken into consideration. Weight-separate mapping, VGG-like algorithm, multiple cells per weight, and fine-tuning of the classifier layer are effective for suppressing inference accuracy loss due to variation and allow for the lowest possible weight precision to improve area and energy efficiency. Higher priority should be given to developing low-conductance and low-variability memory devices that are essential for energy and area-efficiency IMC whereas low bit precision (< 3b) and memory window (< 10) are less concerned.

原文English
文章編號112
期刊Scientific reports
12
發行號1
DOIs
出版狀態Published - 12月 2022

指紋

深入研究「Device quantization policy in variation-aware in-memory computing design」主題。共同形成了獨特的指紋。

引用此