当前位置: X-MOL 学术IEEE Trans. Very Larg. Scale Integr. Syst. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Low-Power, Highly Reliable Dynamic Thermal Management by Exploiting Approximate Computing
IEEE Transactions on Very Large Scale Integration (VLSI) Systems ( IF 2.8 ) Pub Date : 2020-10-01 , DOI: 10.1109/tvlsi.2020.3012626
Somayeh Rahimipour , Wameedh Nazar Flayyih , Noor Ain Kamsani , Shaiful Jahari Hashim , Mircea R. Stan , Fakhrul Zaman B. Rokhani

With the continuous downscaling of semiconductor processes, the growing power density and thermal issues in multicore processors become more and more challenging, thus reliable dynamic thermal management (DTM) is required to prevent severe challenges in system performance. The accuracy of the thermal profile, delivered to the DTM manager, plays a critical role in the efficiency and reliability of DTM, different sources of noise and variations in deep submicron (DSM) technologies severely affecting the thermal data that can lead to significant degradation of DTM performance. In this article, we propose a novel fault-tolerance scheme exploiting approximate computing to mitigate the DSM effects on DTM efficiency. Approximate computing in hardware design can lead to significant gains in energy efficiency, area, and performance. To exploit this opportunity, there is a need for design abstractions that can systematically incorporate approximation in hardware design which is the main contribution of our work. Our proposed scheme achieves 11.20% lower power consumption, 6.59% smaller area, and 12% reduction in the number of wires, while increasing DTM efficiency by 5.24%.

中文翻译:

通过利用近似计算实现低功耗、高度可靠的动态热管理

随着半导体工艺的不断缩小,多核处理器中不断增长的功率密度和热问题变得越来越具有挑战性,因此需要可靠的动态热管理 (DTM) 来防止系统性能面临严峻挑战。传递给 DTM 管理器的热分布的准确性在 DTM 的效率和可靠性、不同噪声源和深亚微米 (DSM) 技术的变化方面起着至关重要的作用,这严重影响了热数据,这可能导致显着降低DTM 性能。在本文中,我们提出了一种新的容错方案,利用近似计算来减轻 DSM 对 DTM 效率的影响。硬件设计中的近似计算可以显着提高能效、面积和性能。为了利用这个机会,需要能够在硬件设计中系统地结合近似的设计抽象,这是我们工作的主要贡献。我们提出的方案实现了 11.20% 的低功耗、6.59% 的面积和 12% 的线数减少,同时将 DTM 效率提高了 5.24%。
更新日期:2020-10-01
down
wechat
bug