Computational Study of Switching Mechanism and Data Retention in Dielectric Thin Film memristor Using Phase-Field Methodology

The possibility of neuro-inspired computing with eNVMs has increased drastically within the last decade as these devices proved to have the required characteristics such as linearity and scalability to be used as synapses in order to bring together memory and computational process in the network. Memristors with metal oxide stack are demonstrated to have increased number of multi-level states, with long-term stability, making them strong candidates to be used as synaptic devices in STDP.Since the conductive path formation in a metal oxide memristor devices plays a major role in training process in Spiking Neural Network, this thesis focuses in using a self-consistent computational phase field method to study conducting channel morphology of resistive switching thin film structures. This approach successfully predicts the formation and annihilation of conducting channels in typical dielectric thin film structures, comparable to a range of resistive switches, offering an alternative computational formulation based on metastable states treated at the atomic scale, as the system is biased by electric field potential, and as the external temperature of the system changes. In contrast to previous resistive switching thin film models, our formulation makes no a priori assumptions on conducting channel morphology and its fundamental transport mechanisms. This study, also, suggests that the generation and growth of nuclei sites in the system due to the influence of external electric field to be one possible root cause of retention failures of ON and OFF states, and eventual reliability degradation of the memristor device