Fast Steady-State Thermal Analysis

Thermal issues are gaining critical in system-on-a-chip. We propose a two-stage thermal analysis approach using big and little matrix operations with a matrix compression method in general successive over-relaxation (SOR) to speed up the analysis of a steady-state thermal profile, referred to as TSA. Experimental results show that the analysis performance of TSA is faster than the general SOR by 2145× in average and faster than the popular Supernodal LU (SuperLU) at least by 1.55× for large size of sparse matrices.