基于先验信息的压缩感知重建温度场技术
Technology of Temperature Field Reconstruction Using Prior Information Based Compressed Sensing
黄俊瑜 1李文昌 2刘剑 3张天一 4朱旻琦5
作者信息
- 1. 济南大学信息科学与工程学院;山东产业技术研究院研究生院
- 2. 中国科学院半导体研究所;中国科学院大学集成电路学院
- 3. 中国科学院半导体研究所;中国科学院大学材料与光电研究中心
- 4. 中国科学院半导体研究所
- 5. 中国电子科技集团公司第五十八研究所
- 折叠
摘要
针对利用压缩感知方法重建微处理器芯片温度场存在错失真实热点、从而影响重建精度的问题,提出了一种基于先验信息的压缩感知重建温度场方法.通过对温度场先验信息的处理,结合主成分分析和模拟退火算法有针对性地设计观测矩阵,优化温度传感器的布置,提高重建精度.实验结果表明:与基于特征图的重建方法、基于模拟退火算法布置传感器的方法和基于随机采样的压缩感知重建温度场方法相比,该方法在平均温度误差、最大温度误差和均方误差方面至少分别提升了 29.9%、46.6%和53.7%,有更优越的温度场重建性能.
Abstract
To solve the problem that the randomly placed sensors miss the real hot spots of the processor and affect the re-construction accuracy in the temperature field reconstruction of microprocessor chips using compressed sensing methods,a prior information based compressed sensing method for temperature field reconstruction was proposed.By processing the prior informa-tion of the temperature field,the observation matrix was designed targeted by combining principal component analysis and simula-ted annealing algorithm,and the temperature sensor layout was optimized to improve the reconstruction accuracy.The experimental results show that,compared with EigenMaps-based reconstruction method,the method based on simulated annealing algorithm for sensor layout and the method based on random sampling for compressive sensing temperature field reconstruction,the method im-proved the reconstruction accuracy by at least 29.9%,46.6%and 53.7%in terms of average temperature error,maximum temper-ature error and mean square error,respectively,and has superior temperature field reconstruction performance.
关键词
温度场重建/压缩感知/观测矩阵/温度传感器/先验信息Key words
temperature field reconstruction/compressed sensing/measurement matrix/temperature sensor/prior information引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据