Combinatorial Techniques for Memory Power State Scheduling in Energy-Constrained Systems

Abstract

Energy has emerged as a critical constraint for a large number of portable, wireless devices. For data intensive applications, a significant amount of energy is dissipated in the memory. Advanced memory architectures support multiple power states of memory banks, which can be exploited to reduce energy dissipation in the system. We present a general methodology using combinatorial graph scheduling techniques, which can be used for obtaining efficient memory power management schedules for algorithms. Additional techniques like tiling further improve the efficiency of our approach. Our simulation results show that we can obtain over 98% energy reduction in the memory energy for the Transitive Closure using our methodology.