DynRBLA: A High-Performance and Energy-Efficient Row Buffer Locality-Aware Caching Policy for Hybrid Memories

Phase change memory (PCM) is a promising memory technology that can offer higher memory capacity than DRAM. Unfortunately, PCM’s access latencies and energies are higher than DRAM and its endurance is lower. DRAM-PCM hybrid memory systems use DRAM as a cache to PCM, to achieve the low access latencies and energies, and high endurance of DRAM, while taking advantage of the large PCM capacity. A key question is what data to cache in DRAM to best exploit the advantages of each technology while avoiding their disadvantages as much as possible. We propose DynRBLA, a fundamentally new caching policy that improves hybrid memory performance and energy efficiency. Our observation is that both DRAM and PCM contain row buffers which cache the most recently accessed row. Row buffer hits incur the same latency in DRAM and PCM, whereas row buffer misses incur longer latencies in PCM. To exploit this, we devise a policy which tracks the access and row buffer misses of a subset of recently used rows in PCM, and caches in DRAM only rows which are likely to miss in the row buffer and be reused. Compared to a cache management technique that only takes into account the frequency of accesses to data, our row buffer locality-aware scheme improves performance by 15% and energy efficiency by 10%. DynRBLA improves system performance by 17% over an all-PCM memory, and comes to within 21% of the performance of an unlimitedsize all-DRAM memory system.