跳到主要导航 跳到搜索 跳到主要内容

MixMixQ: Quantization with Mixed Bit-Sparsity and Mixed Bit-Width for CIM Accelerators

  • Beihang University
  • Beijing Jinghanyu Electronic Engineering Technology Co. Ltd.

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

Quantization is vital for deploying neural networks on Computing-In-Memory (CIM) based accelerators due to inherent limitations in memory devices and data interfaces' representational capacities. However, traditional quantization algorithms often overlook CIM's unique computing paradigm, leading to suboptimal performance. To address this, we introduce MixMixQ, a novel quantization algorithm specifically designed for CIM accelerators that strategically integrates mixed bit-sparsity and mixed bit-width, enhancing overall hardware efficiency while preserving high accuracy. Notably, our method can enhance hardware efficiency by up to 294% compared to traditional quantization methods, with only a minimal 0.13% decrease in accuracy compared to a full-precision network.

源语言英语
主期刊名GLSVLSI 2024 - Proceedings of the Great Lakes Symposium on VLSI 2024
出版商Association for Computing Machinery
537-540
页数4
ISBN(电子版)9798400706059
DOI
出版状态已出版 - 12 6月 2024
活动34th Great Lakes Symposium on VLSI 2024, GLSVLSI 2024 - Clearwater, 美国
期限: 12 6月 202414 6月 2024

出版系列

姓名Proceedings of the ACM Great Lakes Symposium on VLSI, GLSVLSI

会议

会议34th Great Lakes Symposium on VLSI 2024, GLSVLSI 2024
国家/地区美国
Clearwater
时期12/06/2414/06/24

指纹

探究 'MixMixQ: Quantization with Mixed Bit-Sparsity and Mixed Bit-Width for CIM Accelerators' 的科研主题。它们共同构成独一无二的指纹。

引用此