ABaT-FS: Towards adjustable bandwidth and temperature via frequency scaling in scalable memory systems

Abstract

Traditional memory design aims to improve bandwidth and reduce power by trading off memory width and frequency scaling (FS). In this context, we propose ABaT−FS, a hardware scheduling mechanism that, for the first time, performs FS on ranks in scalable memory systems which employ Double Data Rate (DDR) synchronous dynamic random access memories (SDRAM). ABaT−FS is able to utilize different rank frequencies via controlling FS intensity - defined as the ratio between the amount of time FS is applied and the total selected scheduled cycle. We propose a design space exploration of ABaT−FS with different FS intensities aiming to determine the behavior of system implications such as bandwidth, rank temperature, and power utilization. Our findings show that for 100% of FS intensity, bandwidth increases proportionally while rank temperature is increased of about +23.7°C%, and energy-per-bit magnitude is decreased in up to 67%.