An effective state-based predictive approach for leakage energy management on embedded systems

Energy optimization is very important for portable and battery-driven embedded systems. With the shrinking of transistor sizes, reducing leakage power becomes a significant issue. In this paper, we propose a novel prediction approach to predict idleness of functional units for leakage energy management. Using a state-based predictor, historical utilization information of functional units (FUs) is exploited to adjust the state of the predictor so as to enhance the accuracy of prediction; based on it, the idleness of the FUs are predicted and utilized for leakage reduction by applying power gating. We design two prediction algorithms, the prediction with fixed threshold (PFT) and the prediction with dynamic threshold (PDT), respectively. We implement our algorithms based on SimpleScalar and conduct experiments with a suite of fourteen benchmarks from Trimaran. The experimental results show that our algorithms achieve better results compared with the previous work.