Architectural Exploration to Enable Sufficient MTJ Device Write Margin for STT-RAM Based Cache

As a promising nonvolatile memory technology, magnetic tunnel junction (MTJ) based spin-torque transfer RAM (STT-RAM) has recently attracted much attention. However, recent device research suggested that, in order to maintain sufficient MTJ write margin to prevent device breakdown, MTJs may be subject to unconventionally high random write error rates (e.g., 10^-3 and above) as memory cell size is being scaled down. In this paper, we aim to develop a STT-RAM cache design solutions that can effectively tolerate high MTJ write error rates at small performance and implementation cost, which makes it much easier to maintain sufficient MTJ write margin and hence push the STT-RAM scalability envelope. Using the full system simulator PTLsim and a variety of benchmarks, we show that the proposed architecture design can readily accommodate MTJ write error rate upto 2% at the penalty of less than 3% processor performance degradation, less than 10% silicon area overhead, and negligible energy consumption overhead.