Leakage power reduction of functional units in processors having zero-overhead loop counter

Hanmin Park, Jong Kyung Paek, Jinho Lee, Kiyoung Choi

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

Abstract

As the semiconductor process advances below 90nm technology node, leakage power has been an ever more serious concern. One of the most effective ways of reducing leakage power is power gating the circuits. Our work tries to find a microarchitectural way of applying power gating technique to functional units in a processor, while avoiding processor stalls, separate power control instructions, and much hardware overhead. We focus on loops, which typically cause long idle period for some functional units that are not used in the loops. Assuming that the target processor has zero-overhead loop feature, we exploit the existing hardware loop counter to determine when to wake up the functional units that have been turned off with the start of the loop execution. We use our own processor, ODALRISC, synthesized with 45nm process library and a C compiler to execute several benchmarks and realistic applications and obtain power report. The experimental results show that our approach achieves about 30% leakage power reduction on average in functional units for JPEG applications, with no performance degradation.

Original languageEnglish
Title of host publication2009 International SoC Design Conference, ISOCC 2009
Pages492-495
Number of pages4
DOIs
Publication statusPublished - 2009
Event2009 International SoC Design Conference, ISOCC 2009 - Busan, Korea, Republic of
Duration: 2009 Nov 222009 Nov 24

Publication series

Name2009 International SoC Design Conference, ISOCC 2009

Other

Other2009 International SoC Design Conference, ISOCC 2009
Country/TerritoryKorea, Republic of
CityBusan
Period09/11/2209/11/24

All Science Journal Classification (ASJC) codes

  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Leakage power reduction of functional units in processors having zero-overhead loop counter'. Together they form a unique fingerprint.

Cite this