论文部分内容阅读
高性能处理器普遍集成热传感器,采用动态热管理技术对芯片实施连续热监控。然而,由于实际芯片中的模拟或者数字热传感器不可避免伴随噪声,使动态热管理的可靠性受到很大影响。因此,为了提高热监控的精确性,本文运用主成分分析(PCA)技术对原始热图像样本矩阵进行降维近似处理,并结合矩阵扰动分析提出基于模拟退火算法的热传感器位置分布优化方法。实验结果表明:该方法比现有的贪婪算法在热重构误差、信噪比(SNR)和误警率等性能方面有了一定提高,能够有效运用在动态热管理中实现精确的热监控。
Thermal sensors are commonly integrated into high-performance processors and thermal continuous thermal monitoring of the chip is implemented using dynamic thermal management techniques. However, the reliability of dynamic thermal management is greatly affected by the unavoidable accompanying noise of analog or digital thermal sensors in real chips. Therefore, in order to improve the accuracy of thermal monitoring, principal component analysis (PCA) is used to reduce the dimensionality of the original thermal image sample matrix. In combination with the matrix perturbation analysis, an optimization method of thermal sensor position distribution based on simulated annealing algorithm is proposed. The experimental results show that the proposed method has some improvements over the existing greedy algorithms in terms of thermal reconfiguration error, signal-to-noise ratio (SNR) and false alarm rate, and can effectively utilize thermal monitoring in dynamic thermal management.