一种非均匀采样重构芯片热信号的方法

摘要

实时片上热信号是集成电路芯片动态热管理(DTM)的重要信息.提出了一种基于Voronoi图的非均匀采样重构芯片热信号的方法,该方法在16核的处理器上每核9个传感器的情况下给出了最小2.25％的平均误差,并且通过与距离倒数加权平均法的平均运行时间的对比验证了该种方法的优越性.%The runtime on-chip thermal signal is essential information for dynamic thermal management ( DTM). This paper proposes a method based on Voronoi diagram to reconstruct thermal signals with non-uniformly spaced samples, by applying this method on a 16-core processor with 9 thermal sensors on each core, it can reconstruct the thermal signal of the processor with average error as small as 2. 25% , and proves the superiority of our proposed method to inverse distance weight by comparing the average runtime of each other.