STAM: System level state-machine-based thermal behavior analysis for multicore processor

Abstract

As the density of transistors is increased, the processor suffers from not only high power consumption but also high temperature. We presented the system level thermal behavior analysis model (STAM) which uses minimum physical characteristics and activities for components of the processor to obtain thermal behaviors. STAM employs the state-machine-based model for a thermal transient response. The results show that STAM is more than three orders of magnitude faster than the existing temperature estimation tool, with about a 6% average error to find hotspots.