Cryogenic Evaluation of Resistive Random Access Memory With Enhanced Endurance at 14 K

Abstract

The nonvolatile cryogenic memories can play an important role in realizing energy-efficient and scalable low-temperature electronics for quantum computing and future high-performance computing systems. In this article, we evaluate the cryogenic performance of HfOx-based resistive random access memory (RRAM) and demonstrate that the addition of extremely thin ~0.5-nm AlOx barrier layers enables a high endurance of $\gt 10^{{7}}$ cycles, which represents a $20\times $ improvement compared to operation at room temperature (RT). We also show that by leveraging the analog behavior of the RESET at cryogenic temperatures in contrast to the abrupt RESET at RT, multiple resistance levels beneficial for multibit memory and weight tuning in deep neural networks (DNNs) can be realized. The multibit capability coupled with high endurance and low operational voltages at 14 K presents promising opportunities for incorporating RRAMs into memory and machine learning applications within cryogenic computing environments.