2D memristor reliability and modeling for neuromorphic computing

Two-dimensional (2D) materials have been reported to exhibit non-volatile resistive switching (NVRS) phenomenon and applied on memristors or resistive random-access memory (RRAM) devices over the past few decades. Recent research further demonstrates the potential for 2D RRAM devices to be implemented in neuromorphic applications. However, reliability is a major challenge for practical application and industrialization. This dissertation presents the improvement of reliability in 2D material-based memristors and in-depth research in electrical characteristics, failure mechanisms, modeling, and neuromorphic applications. Chapter 2 discusses the reliability improvement of 2D monolayer MoS₂-based memristors by electron irradiation treatment. Sulfur vacancies, the density of which is modified by irradiation dosage, are revealed to be significant in the improvement of yield and endurance. Finite element analysis and Monte Carlo modeling are applied to help understand the role of sulfur vacancies in resistive switching and reliability improvement. In Chapter 3, further optimizations in reliability have been done by introducing sulfurization method in the preparation of MoS₂ films and the fine tune of metal deposition, extending the defect engineering from monolayer to multilayer MoS₂. Intriguing convergence of resistive switching metrics from the statistical measurements is highlighted along with the largely improved endurance performance. An effective switching layer model has been proposed to illustrate the underlying physics of endurance improvement and switching metrics convergence. In Chapter 4, a Monte Carlo modeling tool, which helps visualize the physical process and provides additional insight into the effective switching layer model, is discussed in detail. Transmission electron microscope (TEM) measurements provides experimental support to the model. In Chapter 5, pulse measurements of 2D ReSe₂-based memristors are discussed, demonstrating long-term potentiation and depression (LTP and LTD) behaviors in long-term plasticity programming. A Verilog-A model is proposed based on the multiple-step resistive switching behavior. Further, an artificial neural network (ANN) is trained based on the LTP/LTD parameters from experiments, showing the potential application of 2D memristors. In Chapter 6, the application of the constructed ANN model is discussed by investigating the temperature effect in TaOₓ-based memristors. The simulation results provide additional insights for the designs of potential hardware-based neuromorphic computing applications.