A NoC-Based Spatial DNN Inference Accelerator With Memory-Friendly Dataflow

IF 1.9 4区工程技术 Q3 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

IEEE Design & Test Pub Date : 2023-12-01 DOI:10.1109/MDAT.2023.3310199

Lingxiao Zhu, Wenjie Fan, Chenyang Dai, Shize Zhou, Yongqi Xue, Zhonghai Lu, Li Li, Yuxiang Fu

引用次数: 0

Abstract

Editor’s notes: This article addresses the challenges of excessive storage overhead and the absence of sparsity-aware design in Network-on-Chip (NoC)-based spatial deep neural network accelerators. The authors present a prototype chip that outperforms existing accelerators in both energy and area efficiency, demonstrated on TSMC 28-nm process technology. —Mahdi Nikdast, Colorado State University, USA —Miquel Moreto, Barcelona Supercomputing Center, Spain —Masoumeh (Azin) Ebrahimi, KTH Royal Institute of Technology, Sweden —Sujay Deb, IIIT Delhi, India

查看原文本刊更多论文

具有内存友好数据流的基于noc的空间DNN推理加速器

编者注:本文解决了基于片上网络(NoC)的空间深度神经网络加速器中存储开销过大和缺乏稀疏感知设计的挑战。作者提出了一种原型芯片，在能量和面积效率上都优于现有的加速器，并在台积电28纳米工艺上进行了演示。-Mahdi Nikdast，美国科罗拉多州立大学-Miquel Moreto，西班牙巴塞罗那超级计算中心-Masoumeh (Azin) Ebrahimi，瑞典皇家理工学院-Sujay Deb，印度德里IIIT

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

IEEE Design & Test COMPUTER SCIENCE, HARDWARE & ARCHITECTURE-ENGINEERING, ELECTRICAL & ELECTRONIC

CiteScore

3.80

自引率

5.00%

发文量

期刊介绍： IEEE Design & Test offers original works describing the models, methods, and tools used to design and test microelectronic systems from devices and circuits to complete systems-on-chip and embedded software. The magazine focuses on current and near-future practice, and includes tutorials, how-to articles, and real-world case studies. The magazine seeks to bring to its readers not only important technology advances but also technology leaders, their perspectives through its columns, interviews, and roundtable discussions. Topics include semiconductor IC design, semiconductor intellectual property blocks, design, verification and test technology, design for manufacturing and yield, embedded software and systems, low-power and energy-efficient design, electronic design automation tools, practical technology, and standards.