Resistive Random-Access Memory With Enhanced Reliability by 2D Material Engineering

The modern era is being encompassed by machine learning and artificial intelligence. As developments in these areas continue to expand, the hardware used to implement them has remained largely stagnant. This is where new technologies such as emerging non-volatile memories and, specifically, resistive random-access memory (RRAM) have gained momentum. RRAM utilizes low-power, high-speed resistive switching of a metal oxide thin film, and offers a cross-point memory architecture that facilitates the matrix-vector multiplication commonly used in machine learning. However, cycle-to-cycle variability is a substantial reliability concern that may significantly impact the model accuracy. Different methods have been explored to reduce this variability, including different material stacks and oxygen exchange layers (OEL). 2D materials have certain properties (e.g., atomic layer thickness and tunable defects) that are desirable for an OEL. Therefore, through the implementation of reduced graphene oxide as an OEL a more consistent and stable RRAM device with less variability was fabricated. Additionally, properties such as low power switching and multi-level conductance, which are useful in neuromorphic computing applications, were demonstrated.