Joint Power Management and Adaptive Modulation and Coding for Wireless Communications Systems With Unreliable Buffering Memories

To guard against process variability in advanced semiconductor nodes, especially for high-density memories, designers resort to overdesigning policies resulting in increased power consumption. A promising approach to save power is to utilize Voltage over-Scaling (VoS). However VoS results into unreliable buffering memories where a predictable statistically amount of errors are introduced to memories. The goal is to trade off channel dependent SNR slack versus hardware induced errors, to achieve predetermined quality metrics, at reduced power consumption. By design, modern communication systems attempt to minimize channel-dependent SNR slack via adaptive modulation and coding (AMC) schemes, thus reducing the gains of on-chip power management. This paper investigates the interaction between on-chip power management via VoS on embedded memories versus network based AMC techniques. A novel mathematical approach that analytically describes the system packet error rate (PER) performance under the VoS induced noise is presented. Based on this model, different AMC and power management algorithms are presented that utilize the received SNR estimates to find the best AMC mode and memory voltage that achieves performance goals at reduced power consumption. Simulation results show that the proposed algorithms can achieve up to 58% energy efficiency for the memory-subsystems compared to conventional AMC algorithm with perfect memories.