Collaborative fuzzy‐based partially‐throttling dynamic thermal management scheme for three‐dimensional networks‐on‐chip. Issue 1 (30th September 2016)

• Record Type: Journal Article
• Authors: Yan, Gaizhen; Wu, Ning; Ge, Fen; Xiao, Hao; Zhou, Fang
• Is Part Of: IET computers & digital techniques. Volume 11:Issue 1(2017)
• Page Start: 24
• Page End: 32
• Publication Date: 2016-09-30
• Subjects: network‐on‐chip -- fuzzy set theory -- thermal management (packaging) -- network routing; three‐dimensional networks‐on‐chip -- collaborative fuzzy‐based partially‐throttling -- dynamic thermal management scheme -- performance improvement -- network availability -- CFP‐DTM -- clock gating scheme -- throttling ratio -- throttled nodes -- adaptive throttling‐aware routing scheme -- fully‐throttling based vertical throttling scheme; Computers -- Periodicals; Digital electronics -- Periodicals; Computer engineering -- Periodicals; Computer architecture -- Periodicals; Computer organization -- Periodicals; 621.39
• DOI: 10.1049/iet-cdt.2015.0198
• Languages: English
• ISSNs: 1751-8601

Abstract

Abstract : Three‐dimensional networks‐on‐chip are beneficial for performance improvement, but suffer from severe thermal issues. Dynamic thermal management (DTM) schemes have been proposed to keep the temperature below the thermal limit while improve the system performance. However, existing fully‐throttling DTM schemes degrade the network availability and thus decrease the system performance. In this study, a novel collaborative fuzzy‐based partially‐throttling DTM (CFP‐DTM) scheme is developed. Two main components are involved in the CFP‐DTM: (i) a fuzzy‐based clock gating scheme dynamically adjusting the throttling ratio and throttled nodes (ii) a highly adaptive throttling‐aware routing scheme for packets to detour the easily congested channels. Experiments show that, compared with the fully‐throttling based vertical throttling scheme, the proposed CFP‐DTM can improve the throughput by 27.5% and reduce the thermal control oscillation by 3°C under the maximum system workload.