Enhancing Lifetime of PCM-Based Main Memory with Efficient Recovery of Stuck-at Faults

Abstract

Among several nonvolatile memory (NVM) candidates, PCM is selected as an attractive replacement to DRAM and comes with both challenges and opportunities. It has beautiful characteristics like non-volatility, better scalability, and lower leakage power. However, limited write endurance is the main burden toward its adoption in practice. It means that after a certain number of writes, some memory cells permanently stuck at either '0' or '1'. To solve this problem and provide strong fault tolerant system, some recovery techniques with minimal storage overhead are required. In this work, we propose a recovery mechanism that relies on static partitioning of a data block into some small number of groups and spreading out faults across the groups uniformly. We then exploit inversion mechanism along with shifting mechanism to continue the use of the failed cell with stuck-at value. So, our proposed method can recover multi bit stuck at faults per partition. Compare to the existing mechanisms, our experimental results for multi-threaded workloads reveal considerable improvement in lifetime and the number of recoverable failures per data block