DarkMem: Fine-grained power management of local memories for accelerators in embedded systems

Abstract

SRAM consumes a growing fraction of the static power in heterogeneous SoCs, as embedded memories take 70% to 90% of the area of specialized accelerators. We present DARK-MEM as a comprehensive solution for fine-grained power management of accelerator local memories. The DARKMEM methodology optimizes at design time the bank configuration for each given accelerator to maximize power-gating opportunities. The DARK-MEM microarchitecture dynamically varies the operating mode of each memory bank according to the accelerator workload. In our experiments, DARKMEM reduces the SRAM static power by more than 40% on average, which translates into a reduction of the total power by almost 18% on average with less than 1% overhead.

A Survey on Domain-Specific Memory Architectures

“…Reducing the number of physical memories saves resources, but also saves the static and dynamic energy required to access those memories. Fine-grained voltage regulation of the memory banks can help reduce the power consumption of the memory banks, especially when stalling for system congestion [24]. This optimization, however, requires dual-rail memory IPs and integrated voltage regulators, making the physical implementation more complex.…”