Efficient electro-thermal co-analysis on CPU+GPU heterogeneous architecture

Abstract

Since supply voltage and temperature (V&T) directly influence IC performance and reliability, electro-thermal (ET) analysis including power/ground (P/G) analysis and thermal analysis is very important in IC design. On the observation that temperature's influence on leakage current (ET coupling effect) and supply voltage's influence on power consumption, this work proposes a novel iteration-based ET co-analysis method that simultaneously solves V&T and then with them to refresh power consumption for the next round of V&T solving. Different from present methods that regard the P/G analysis and thermal analysis as independent processes without interaction, the ET co-analysis method takes their interaction into the consideration, which leads to more practical results. Since both P/G analysis and thermal analysis are very time-consuming, this work further employs multi-thread and GPU parallel computing techniques to speed up the ET co-analysis based on a parallel computing system of CPU+GPU heterogeneous architecture (PCS_CGHA). Experimental results show that compared with our method, present ET analysis methods will give too pessimistic or optimistic results W/O considering ET coupling effect. And our efficient method on PCS_CGHA provides 44 times speedup over the naive analysis method with no acceleration.