VLSI implementation of deep neural networks using integral stochastic computing

Arash Ardakani, Francois Leduc-Primeau, Naoya Onizawa, Takahiro Hanyu, Warren J. Gross

Research output: Chapter in Book/Report/Conference proceedingConference contribution

15 Citations (Scopus)

Abstract

The hardware implementation of deep neural networks (DNNs) has recently received tremendous attention since many applications require high-speed operations. However, numerous processing elements and complex interconnections are usually required, leading to a large area occupation and a high power consumption. Stochastic computing has shown promising results for area-efficient hardware implementations, even though existing stochastic algorithms require long streams that exhibit long latency. In this paper, we propose an integer form of stochastic computation and introduce some elementary circuits. We then propose an efficient implementation of a DNN based on integral stochastic computing. The proposed architecture uses integer stochastic streams and a modified Finite State Machine-based tanh function to improve the performance and reduce the latency compared to existing stochastic architectures for DNN. The simulation results show the negligible performance loss of the proposed integer stochastic DNN for different network sizes compared to their floating point versions.

Original languageEnglish
Title of host publication2016 9th International Symposium on Turbo Codes and Iterative Information Processing
Subtitle of host publicationPaths to 5G and Beyond, ISTC 2016
PublisherIEEE Computer Society
Pages216-220
Number of pages5
ISBN (Electronic)9781509034017
DOIs
Publication statusPublished - 2016 Oct 17
Event9th International Symposium on Turbo Codes and Iterative Information Processing, ISTC 2016 - Brest, France
Duration: 2016 Sep 52016 Sep 9

Publication series

NameInternational Symposium on Turbo Codes and Iterative Information Processing, ISTC
Volume2016-October
ISSN (Print)2165-4700
ISSN (Electronic)2165-4719

Other

Other9th International Symposium on Turbo Codes and Iterative Information Processing, ISTC 2016
CountryFrance
CityBrest
Period16/9/516/9/9

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Information Systems
  • Theoretical Computer Science

Fingerprint Dive into the research topics of 'VLSI implementation of deep neural networks using integral stochastic computing'. Together they form a unique fingerprint.

Cite this