摘要
多核处理器等超大规模集成电路芯片通常采用动态热管理技术处理热问题。高精度重构芯片温度场是动态热管理正确运行、保证被测芯片运行性能及工作可靠性的前提条件。基于频域分析的温度场重构技术损失了一部分高频信息,重构精度较低。为了提升温度场重构精度,该文提出了一种基于稀疏字典学习的温度场重构方法。该方法通过字典学习将温度场先验信息稀疏表示,同时设计温度传感器位置分配方案,实现了温度场的重构。实验结果证明了相比基于频域分析的温度场重构方法,所提方案具有更优越的温度场重构性能。
Dynamic thermal management is used to handle the thermal problem of very large scale integrated circuits(VLSI),such us multicore processors.Accurate monitoring of the temperature field can insure dynamic thermal management working correctly,guarantee the chip working performance and reliability further.The temperature field reconstruction techniques based on analysis in frequency domain ignore the information in high frequency zone,which leads to thermal field recovery inaccurate.In order to improve the precision of thermal field reconstruction,a new thermal field reconstruction method based on sparse dictionary learning technology is proposed.In this method,the prior information of temperature field is sparse represented by dictionary learning,and the location assignment scheme of temperature sensor is designed to realize the reconstruction of temperature field.The experiments prove that the proposed strategy have better performance than the methods based on analysis in frequency domain.
作者
张天一
李文昌
肖金玉
刘剑
ZHANG Tian-yi;LI Wen-chang;XIAO Jin-yu;LIU Jian(Institute of Semiconductors,Chinese Academy of Sciences,Haidian Beijing,100083;School of Microelectronics,University of Chinese Academy of Sciences,Sijingshan Beijing,100049;CSMC Technologies Corporation,Wuxi Jiangsu,214028;College of Materials Science and Opto-Electronic Technology,University of Chinese Academy of Sciences,Sijingshan Beijing,100049)
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2021年第4期502-507,共6页
Journal of University of Electronic Science and Technology of China
关键词
字典学习
动态热管理
稀疏表示
温度传感器
温度场重构
dictionary learning
dynamic thermal management
sparse representation
temperature sensors
thermal field reconstruction
作者简介
张天一(1990-),男,博士生,主要从事数模混合集成电路设计及热分析方面的研究;通信作者:李文昌,E-mail:liwc@semi.ac.cn。
