Low-Energy Heterogeneous Non-Volatile Memory Systems for Mobile Systems

Memory systems consume significant energy in hand-held embedded systems. Existing techniques for reducing memory energy requirements in low-power systems have addressed energy consumption when the system is turned on; here we also consider data retention energy during the power off period. Semiconductor non-volatile memory is indispensable for hand-held devices that cannot afford magnetic disks due to excessive space, weight, cost and energy consumption. Current hand-held systems are generally equipped with more than one type of non-volatile storage device, such as battery-backed SDRAM, NOR Flash memory or NAND Flash memory, because each technology has its distinct and complementary features. In this paper, we introduce an energy-aware memory allocation in heterogeneous non-volatile memory systems to maximize the battery life. For this purpose, we first characterize cycle-accurate active mode energy and the data retention energy of non-volatile memory systems. Next, we present an energy-aware memory allocation for a given task set, taking into account arrival rate, execution time, code size, user data size and the number of memory transactions; we do this using trace-driven simulation. Experiments demonstrate that an optimized allocation can save up to 26% of the memory system energy compared with traditional allocation schemes.