A simple thermal model for multi-core processors and its application to slack allocation

摘要

Power density and heat density of multicore processor system are increasing exponentially with Moore's Law. High temperature on chip greatly affects its reliability, and the cost of packaging and cooling system increases exponentially with power consumption. For a multicore processor, the peak temperature of a block depends on its own power density as well as power density of other blocks on chip. In this paper, we have developed a simple thermal model, called Matrix Model (MM), that can be used to derive temperature profiles for all the cores of a multicore processor. We theoretically demonstrate the correctness and efficiency of MM. Our simulation results show that the model is comparable to the HotSpot Model for predicting the peak temperature. Besides having lower computational cost, the MM is succinct (a single matrix) and can be used to derive algorithms for a variety of scenarios. We use this model to develop a novel slack allocation algorithm for a workflow represented by Directed Acyclic Graph on a multicore processor.