基于最大误差的热传感器数量分配方法

Sensor Allocation Based on the Maximum Error in Thermal Monitoring for Microprocessors

为了更好地监控微处理器的运行温度,提出了一种考虑奇异点的改进聚类求解策略和新颖的热传感器数量分配方案.该方案分析了在给定最大热点监控误差下所需要的热传感器数量,并且在满足工程上所需要求的情况下,使热传感器数量尽可能地少.实验结果表明,所提出的方法能够在广泛应用的微处理器架构上计算出给定误差要求下所需的热传感器数量,并且与现有的k-means聚类算法相比,所使用的传感器数量明显减少.

A new singular value-awared method of solving and sensor allocation strategy were proposed for the temperature sensors embedded in microprocessors as to monitor the thermal behavior. The strategy discussed how many temperature sensors we need at least to satisfy the max error to be allowed by using of improved dual clustering algorithm. The experimental results indicate the superiority of the technique and confirm that it can calculate the minimum number of sensors with the max error allowed in widely used microprocessor architecture. Much less sensors are used compared with the k-means cluster algorithm.