A low-computation-complexity, energy-efficient, and high-performance linear program solver using memristor crossbars

Ruizhe Cai, Ao Ren, Yanzhi Wang, Sucheta Soundarajan, Qinru Qiu, Bo Yuan, Paul Bogdan

Research output: Chapter in Book/Entry/PoemConference contribution

4 Scopus citations

Abstract

Linear programming is required in a wide variety of application including routing, scheduling, and various optimization problems. The primal-dual interior point (PDIP) method is state-of-the-art algorithm for solving linear programs, and can be decomposed to matrix-vector multiplication and solving systems of linear equations, both of which can be conducted by the emerging memristor crossbar technique in O(1) time complexity in the analog domain. This work is the first to apply memristor crossbar for linear program solving based on the PDIP method, which has been reformulated for memristor crossbars to compute in the analog domain. The proposed linear program solver can overcome limitations of memristor crossbars such as supporting only non-negative coefficients, and has been extended for higher scalability. The proposed solver is iterative and achieves O(N) computation complexity in each iteration. Experimental results demonstrate that reliable performance with high accuracy can be achieved under process variations.

Original languageEnglish (US)
Title of host publicationProceedings - 29th IEEE International System on Chip Conference, SOCC 2016
EditorsKaran Bhatia, Massimo Alioto, Danella Zhao, Andrew Marshall, Ramalingam Sridhar
PublisherIEEE Computer Society
Pages317-322
Number of pages6
ISBN (Electronic)9781509013661
DOIs
StatePublished - Jul 2 2016
Event29th IEEE International System on Chip Conference, SOCC 2016 - Seattle, United States
Duration: Sep 6 2016Sep 9 2016

Publication series

NameInternational System on Chip Conference
Volume0
ISSN (Print)2164-1676
ISSN (Electronic)2164-1706

Other

Other29th IEEE International System on Chip Conference, SOCC 2016
Country/TerritoryUnited States
CitySeattle
Period9/6/169/9/16

ASJC Scopus subject areas

  • Hardware and Architecture
  • Control and Systems Engineering
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'A low-computation-complexity, energy-efficient, and high-performance linear program solver using memristor crossbars'. Together they form a unique fingerprint.

Cite this