Design of a dynamically reconfigurable arithmetic unit for matrix algorithms

Weijiang Wang, Yingtao Ding, Shan Cao, Xianli Zhao

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Matrix operations, such as multiplication, inversion operations, are widely used in high-perforamce industrial control, scientific computing, and media processing applications. To meet the increasingly intensive timing and power demands of matrix opertions, dynamically reconfigurable structures are introduced as a new design paradigm to make a good balance between performance and flexibility. In this paper, a dynamically reconfigurable multi-operation arithmetic unit is proposed basing on multiply-add-fused unit. The proposed arithmetic unit consists of three pipeline stage, each of which can be dynamically configured by control signals. The proposed arithmetic unit can be used for addition, subtraction, multiplication and multiply-add-fused operations for different data types. Compared with the traditional arithmetic units, the chip area is reduced by the proposed unit since the most functional units are reused for different operations.

Original languageEnglish
Title of host publicationProceedings - 2015 IEEE 11th International Conference on ASIC, ASICON 2015
EditorsJunyan Ren, Ting-Ao Tang, Fan Ye, Huihua Yu
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781479984831
DOIs
Publication statusPublished - 21 Jul 2016
Event11th IEEE International Conference on Advanced Semiconductor Integrated Circuits (ASIC), ASICON 2015 - Chengdu, China
Duration: 3 Nov 20156 Nov 2015

Publication series

NameProceedings - 2015 IEEE 11th International Conference on ASIC, ASICON 2015

Conference

Conference11th IEEE International Conference on Advanced Semiconductor Integrated Circuits (ASIC), ASICON 2015
Country/TerritoryChina
CityChengdu
Period3/11/156/11/15

Fingerprint

Dive into the research topics of 'Design of a dynamically reconfigurable arithmetic unit for matrix algorithms'. Together they form a unique fingerprint.

Cite this