上海交通大学学报(英文版) ›› 2017, Vol. 22 ›› Issue (4): 481-492.doi: 10.1007/s12204-017-1861-1

• • 上一篇    下一篇

Rigid Sensor Allocation and Placement Technique for Reducing the Number of Sensors in Thermal Monitoring

LI Xin* (李鑫), ZHOU Wei (周巍), JIANG Wen (蒋雯)   

  1. (School of Electronics and Information, Northwestern Polytechnical University, Xi’an 710072, China)
  • 出版日期:2017-08-03 发布日期:2017-08-03
  • 通讯作者: LI Xin (李鑫) E-mail:xinli@nwpu.edu.cn

Rigid Sensor Allocation and Placement Technique for Reducing the Number of Sensors in Thermal Monitoring

LI Xin* (李鑫), ZHOU Wei (周巍), JIANG Wen (蒋雯)   

  1. (School of Electronics and Information, Northwestern Polytechnical University, Xi’an 710072, China)
  • Online:2017-08-03 Published:2017-08-03
  • Contact: LI Xin (李鑫) E-mail:xinli@nwpu.edu.cn

摘要: Abstract: Using embedded thermal sensors, dynamic thermal management (DTM) techniques measure runtime thermal behavior of high-performance microprocessors so as to prevent thermal runaway situations. The number of placed sensors should be minimized, while guaranteeing accurate tracking of hot spots and full thermal characterization. In this paper, we propose a rigid sensor allocation and placement technique for determining the minimal number of thermal sensors and the optimal locations while satisfying an expected accuracy of hot spot temperature error based on dual clustering. We analyze the false alarm rates of hot spots using the proposed methods in noise-free, with noise and sensor calibration scenarios, respectively. Experimental results confirm that our proposed methods are capable of accurately characterizing the temperatures of microprocessors.

关键词: thermal sensors, dynamic thermal management, allocation and placement, dual clustering

Abstract: Abstract: Using embedded thermal sensors, dynamic thermal management (DTM) techniques measure runtime thermal behavior of high-performance microprocessors so as to prevent thermal runaway situations. The number of placed sensors should be minimized, while guaranteeing accurate tracking of hot spots and full thermal characterization. In this paper, we propose a rigid sensor allocation and placement technique for determining the minimal number of thermal sensors and the optimal locations while satisfying an expected accuracy of hot spot temperature error based on dual clustering. We analyze the false alarm rates of hot spots using the proposed methods in noise-free, with noise and sensor calibration scenarios, respectively. Experimental results confirm that our proposed methods are capable of accurately characterizing the temperatures of microprocessors.

Key words: thermal sensors, dynamic thermal management, allocation and placement, dual clustering

中图分类号: