Improving energy and performance with spintronics caches in multicore systems

Spintronic memory (STT-MRAM) is an attractive alternative technology to CMOS since it offers higher density and virtually no leakage current. Spintronic memory continues to require higher write energy, however, presenting a challenge to memory hierarchy design when energy consumption is a concern. Various techniques for reducing write energy have been studied in the past for a single processor, typically focusing on the last-level caches while keeping the first level caches in CMOS to avoid the write latency. In this work, use of STT-MRAM for the first level caches of a multicore processor is motivated by showing that the impact on throughput due to increased write latency is offset in many cases by increased cache size due to higher density. The Parsec benchmark suite is run on a modern multicore platform simulator, comparing performance and energy consumption of the spintronic cache system to a CMOS design. A small, fully-associative level-0 cache is then introduced (on the order of 8-64 cache lines), and shown to effectively hide the STT-MRAM write latency. Performance degradation due to write latency is restored or slightly improved, while cache energy consumption is reduced by 30-50% for 12 of the 13 benchmarks.