An FPGA-based parallel accelerator for matrix multiplications in the Newton-Raphson method

Xizhen Xu, Sotirios Ziavras, Tae Gyu Chang

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

4 Scopus citations

Abstract

Power flow analysis plays an important role in power grid configurations, operating management and contingency analysis. The Newton-Raphson (NR) iterative method is often enlisted for solving power flow analysis problems. However, it involves computation-expensive matrix multiplications (MMs). In this paper we propose an FPGA-based Hierarchical-SIMD (H-SIMD) machine with its codesign of the Hierarchical Instruction Set Architecture (HISA) to speed up MM within each NR iteration. FPGA stands for Field-Programmable Gate Array. HISA is comprised of medium-grain and coarse-grain instructions. The H-SIMD machine also facilitates better mapping of MM onto recent multimillion-gate FPGAs. At each level, any HISA instruction is classified to be of either the communication or computation type. The former are executed by a controller while the latter are issued to lower levels in the hierarchy. Additionally, by using a memory switching scheme and the high-level HISA set to partition applications, the host-FPGA communication overheads can be hidden. Our test results show sustained high performance.

Original languageEnglish (US)
Title of host publicationEmbedded and Ubiquitous Computing - International Conference EUC 2005, Proceedings
Pages458-468
Number of pages11
StatePublished - Dec 1 2005
EventInternational Conference on Embedded and Ubiquitous Computing, EUC 2005 - Nagasaki, Japan
Duration: Dec 6 2005Dec 9 2005

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3824 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

OtherInternational Conference on Embedded and Ubiquitous Computing, EUC 2005
CountryJapan
CityNagasaki
Period12/6/0512/9/05

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'An FPGA-based parallel accelerator for matrix multiplications in the Newton-Raphson method'. Together they form a unique fingerprint.

Cite this