Investigating Device Degradation and Revival in Resistive Random Access Memory

Understanding and correcting the degradation of Resistive Random Access Memory (ReRAM) devices is critical for hardware performance and accuracy, especially in the context of Neural Networks (NN) and hardware-based accelerator platforms. Device degradation can be mitigated by using switching conditions that prolong longevity at the expense of memory window performance, however, when devices do degrade or fail, the common approach is to bypass the affected device within the NN or memory array. This paper explores the potential for device revival, examining both short-term switching characteristics and long-term reliability.