Exploring two-dimensional materials for memristors: Modeling and characterization

Abstract

The canonical von Neumann architecture, which distinctly separates the central processing unit (CPU) from the memory unit, faces significant challenges in computational speed and energy efficiency, particularly in data-intensive applications, such as artificial intelligence (AI) and the Internet of Things (IoT). To address these limitations, in-memory computing has emerged as a promising approach by enabling logic operations directly within the memory array, which reduces latency and lowers energy consumption. Among various memory technologies, resistive random access memory (RRAM) or memristor stands out as viable candidates for in-memory computing hardware that offers advantages, such as non-volatile data storage, fast switching speed, long retention time, and high endurance. In particular, two-dimensional materials (2DMs), such as graphene, MoS,, WS, etc., have gained considerable attention due to their atomic-scale thickness, superior electrical and thermal properties with easy and large-scale synthesis. However, implementing crossbar memory and synaptic array with 2DM-based memristor presents several significant challenges, including sneak path currents (ie., a low ON/OFF resistance ratio), limited dynamic range in conductance, and limited endurance. The thesis aims to address these limitations through a combination of physics-based modeling and experimental studies. A comprehensive numerical simulation model is first developed to describe the resistive switching behavior of 2D material-based memristors. This model incorporates electrical and thermal conductivity variations with oxygen ion concentration, which is feedback into a self-consistent solution of the continuity equation, Fourier equation for Joule heating, and Poisson’s equation to accurately capture the switching dynamics. Furthermore, the study explores memristor with graphene oxide (GO) as the resistive layer and compares them with TaOg-based memristor through numerical simulation. The results reveal that GO memristor exhibits superior switching performance, including lower sneak current, a higher ON/OFF resistance ratio, an enhanced read window, and improved non-linearity compared to TaOy-based devices. Additionally, graphene electrode (GE) memristor is investigated to highlight their advantages over conventional metal electrodes, such as Pt and TiN. The findings demonstrate that GE significantly reduces switching voltage, improves switching speed, and enhances analog switching behavior. To further optimize memristor performance, the study evaluates four different metal oxides, such as HfOy, NiOy, TaOy, and TiOy, as resistive layers for GE memristor. Next, the thesis explores defect engineering strategies to enhance the resistive switching characteristics of chemical vapor deposition (CVD)-grown MoS; memristor. Oxygen plasma irradiation is employed as a defect modulation technique, which leads to increased ON/OFF current ratios, reduced OFF-state leakage, and improved multilevel resistance states. The near linear conductance modulation observed during long-term potentiation (LTP) and long-term depression (LTD) in MoS, memristor enables high pattern recognition accuracy (90%) in a 784 x 100 x 10 neural network trained on the MNIST dataset. Moreover, the performance characteristics of the oxygen plasma-treated memristor are compared with the argon plasma treated memristor. The findings reveal that oxygen plasma treated MoS, memristor exhibits a higher ON/OFF ratio, lower SET/RESET voltages, a lower RESET current, more linear and symmetric conductance modulation, and higher recognition accuracy. These studies provide valuable insights into controlled defect engineering in memristive devices, which paves the way for scalable and energy-efficient neuromorphic computing architectures. By integrating rigorous modeling with experimental investigations, this thesis advances the understanding and design of 2D materials for memristor application, bridging the gap between material innovation and neuromorphic system requirements.