Nanofabrication Technologies for Making Neuromorphic Devices Based on Two-Dimensional MoS2

Abstract

Neuromorphic devices hold great potentials for next-generation computing and storage applications due to their attractive performance features such as low energy consumption and high execution speed compared to conventional CMOS based electronic devices, which are complicated, high cost, and high energy-intensive. For next-generation neuromorphic device applications, emerging layered semiconductor materials, such as layered transition metal dichalcogenides (TMDCs), have attracted tremendous attention due to their excellent electronic, spintronic and optoelectronic properties. However, to ultimately realize working TMDC-based neuromorphic devices and systems for practical applications, we need to (i) develop scalable nanofabrication methods for producing high-quality 2D-layered devices, (ii) advance the device physics to utilize the unique electronic properties of 2D-layered materials for practical implementations, and (iii) provide scientific insights for controlling the electronic characteristics of 2D-materials and technical guidelines for manufacturing practical neural networks based on 2D materials. The research works presented in this dissertation seek to address part of the aforementioned challenges and include the following topics: (1) Development of a top-down nanofabrication approach capable of producing multilayer TMDC heterostructures; (2) Systematical investigation of the switching characteristics and mechanisms of mechanically-printed few-layer MoS2 memristors for potential neuromorphic applications; (3) Development of a plasma-treating technology for controlling the stoichiometry composition as well as device characteristics of 2D-layered memristors for neuromorphic computing; (4) Fabrication and characterization of 2D-materials based multilevel transistor memory cells with a combination of excellent retention and endurance properties. The first part of the thesis presents a new nanofabrication technique, which is capable of producing WSe2/MoS2 heterostructure arrays with a high uniformity of feature thicknesses. The demonstrated WSe2/MoS2 heterostructure photo-response devices exhibit an enhanced degree of current rectification (forward/backward current ratio ~ 400) and prominent photovoltaic responses in comparison with pristine WSe2 control devices. The second part of the thesis presents a systematical study on the switching characteristics and mechanisms of mechanically-printed few-layer MoS2 memristors. The observed switching characteristics (i.e., rectification-mediated and conductance-mediated modes) can be explained by the modulation of the MoS2/metal Schottky barriers and re-distribution of ionic vacancies in the MoS2 channels. We further found that the switching mode transition (i.e., transition from the analogue mode to a quasi-binary mode under modulation of accumulative electrical stresses) is attributed to the field-induced agglomeration of the ionic vacancies at MoS2/Ti interfaces. The third part of the thesis presents a novel plasma-assisted treatment technology for controlling the stoichiometry composition as well as device characteristics of MoS2 memristors to further enhance analogue pulse-programmed switching characteristics with a better linear weight update scheme and enhanced dynamic range. Simulation results imply that the neural network consisting of Ar-plasma-treated MoS2 memristors could potentially achieve 94.3% learning accuracy. The fourth part of the thesis presents an experiment/simulation-integrated investigation on the validity of the implementation of plasma-treated MoS2-based multilevel transistor cells for neuromorphic computing applications. The plasma-treated multilevel cells can be programmed into at least 3-bit resolvable states for hour-scale computing applications or 4.7-bit states for minute-scale computing. These presented works provide nanofabrication approaches for producing multilayer TMDC-based nanoelectronic device arrays, which could be generally utilized for making other layered materials. Additionally, the obtained device physics knowledge associated with memristors and multilevel cells based on MoS2 is anticipated to leverage the unique electronic properties of layered semiconductors for neuromorphic computing and data storage applications.