PS-BloTAM: Pre-sampling based architecture-level temperature analysis methodology

Abstract

Efficient Thermal Analysis (TA) plays a crucial role in temperature-aware floorplanning design and Dynamic Power and Temperature Management (DPTM) for Multi-Processor System-on-Chip(MPSoC). To accelerate temperature-aware floorplanning with fixed die area and unchanged heat dissipation system, this work proposes a novel architecture-level TA method PS-BloTAM. It first uses HotSpot to build the pre-sampling thermal resistance (TR) matrix S with a sampling block array. Then, according to size and position of modules in the given floorplanning solution, PS-BloTAM analytically computes this given floorplan's TR matrix R with S. Finally temperatures of modules in different working modes can be easily estimated with previous R. Adopting traditional idea of design library, PS-BloTAM is able to efficiently compute temperature distributions of numerous floor-plans. Experiments show that comparing with HotSpot, PS-BloTAM achieves 43X speedup with average and maximum errors less than 1.65% and 6.64% respectively.