An Adaptive ECC Scheme for Runtime Write Failure Suppression of STT-RAM Cache

Spin-transfer torque random access memory (STT-RAM) features many attractive charac- teristics, including near-zero standby power, nanosecond access time, small footprint, etc. These properties make STT-RAM perfectly suitable for the applications that are subject to limited power and area budgets, i.e., on-chip cache. Write reliability is one of the major challenges in design of STT-RAM caches. To ensure design quality, error correction code (ECC) scheme is usually adopted in STT-RAM caches. However, it incurs significant hard- ware overhead. In observance of the dynamic error correcting requirements, in this work, we propose an adaptive ECC scheme to suppress the runtime write failures of STT-RAM cache with minimized hardware cost, in which the cache is partitioned into regions protected by different ECCs. The error rate of a data is speculated on-the-fly and the data is allocated to a partition that provides the needed error correcting capability. Moreover, to accom- modate the time-varying error correcting requirements of runtime data, the thresholds that determine data’s destination cache partition will be adaptively adjusted. Our experimental results show that compared to conventional ECC schemes, our scheme can save up to 80.2% ECC bit overhead with slightly degraded write reliability of the STT-RAM cache. Moreover, the detailed analysis shows that through simultaneous optimization in cache access patterns and reducing STT cell programming workload, our method outperforms conventional ECC design in power and energy consumptions