A dynamic temperature control simulation system for FPGAs

Abstract

Rapid increases in transistor density, clock speeds and competition with custom ICs have escalated the demand for aggressive solutions to battle rising operating temperatures in programmable fabrics. In this work, we make several key contributions to temperature management in FPGAs. We develop a novel and robust simulation framework exploring adaptive techniques to reduce on chip temperatures in the reconfigurable core. We implement a thermal driven voltage scaling algorithm based on temperature and performance feedback. Our performance estimation model is an accurate empirical relation between delay, supply voltage and temperature with an average error of 9%. Our final results show significant temperature reductions of up to 13.37degC accompanied by the added benefit of power savings averaging 13.48%. Overheads are limited to an average reduction in worst case operating frequency of 10.78% and a voltage swing of 0.61V.