Retention Characteristic Optimization Based on Combined Forming Scheme for Resistive Random Access Memory Chip

Abstract

The long-time retention issue of resistive random access memory (RRAM) presents a significant challenge in maintaining the performance of large-scale RRAM-based computation-in-memory (CIM) systems. To address the long-term inference accuracy degradation caused by RRAM instability, we proposed a combined forming strategy, which can effectively suppress resistance drift and improve inference accuracy without periodic updates of RRAM cells. With this optimized strategy, the probability of high resistance state (HRS) drift was reduced to 5%, and the inference accuracy could be maintained at 88% even after $10^{{5}}$ s at 125°C. This work provided a valuable strategy for enhancing devices retention and sustaining accuracy in RRAM-based CIM systems.