Attojoule Hexagonal Boron Nitride-Based Memristor for High-Performance Neuromorphic Computing

Abstract

In next-generation neuromorphic computing applications, the primary challenge lies in achieving energy-efficient and reliable memristors while minimizing their energy consumption to a level comparable to that of biological synapses. In this work, hexagonal boron nitride (h-BN)-based metal-insulator-semiconductor (MIS) memristors operating is presented at the attojoule-level tailored for high-performance artificial neural networks. The memristors benefit from a wafer-scale uniform h-BN resistive switching medium grown directly on a highly doped Si wafer using metal-organic chemical vapor deposition (MOCVD), resulting in outstanding reliability and low variability. Notably, the h-BN-based memristors exhibit exceptionally low energy consumption of attojoule levels, coupled with fast switching speed. The switching mechanisms are systematically substantiated by electrical and nano-structural analysis, confirming that the h-BN layer facilitates the resistive switching with extremely low high resistance states (HRS) and the native SiOx on Si contributes to suppressing excessive current, enabling attojoule-level energy consumption. Furthermore, the formation of atomic-scale conductive filaments leads to remarkably fast response times within the nanosecond range, and allows for the attainment of multi-resistance states, making these memristors well-suited for next-generation neuromorphic applications. The h-BN-based MIS memristors hold the potential to revolutionize energy consumption limitations in neuromorphic devices, bridging the gap between artificial and biological synapses. This article presents wafer-scale hexagonal boron nitride (h-BN)-based memristors with metal-insulator-semiconductor (MIS) configuration, operating at the attojoule level. These h-BN-based memristors are the first to demonstrate multi-states in response to nanosecond stimuli among existing 2D materials-based memristors. The h-BN-based memristors have the potential to revolutionize the current challenges in neuromorphic applications, bridging the energy efficiency gap between artificial and biological synapses. image