Runtime network-on-chip thermal and power balancing
In Network-on-Chip (NoC), most thermal and peak power balancing methods are monitored by centralized power/thermal managers. This increases inter-core communication latency and imbalanced thermal distribution. These factors directly affect the hot spot formation caused by high power densities develo...
Saved in:
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
ARQII Publication, Malaysian Simulation Society
2017
|
Subjects: | |
Online Access: | http://eprints.utm.my/id/eprint/80736/1/MohdShahrizalRusli2017_RuntimeNetwork-on-ChipThermal.pdf http://eprints.utm.my/id/eprint/80736/ http://arqiipubl.com/ojs/index.php/AMS_Journal/article/view/6/6 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | In Network-on-Chip (NoC), most thermal and peak power balancing methods are monitored by centralized power/thermal managers. This increases inter-core communication latency and imbalanced thermal distribution. These factors directly affect the hot spot formation caused by high power densities developed with increasing per-core transistor number. As a result, reliability decreases along with static power dissipation. This proposal aims to introduce hierarchical agents for balancing power and thermal distribution by manipulating system’s parameters such as power, thermal, voltage below the optimal values. As some level of control is applied, this proposal also targets to achieve network scalability by implementing some level of independencies; self-organize and self-optimize distributed agent in overcoming core-level homogeneous processing element (PE) thermal and power variations at runtime. The aims of this work are to significantly contribute to achieving runtime thermal and power balancing, power and thermal management and reducing thermal hot spot formation in NoC. |
---|