Computational modeling of memristive nanowire junctions for neuromorphic computing

The exponential growth of artificial intelligence and data-intensive algorithms has exposed the scalability limitations of the classical von Neumann computing paradigm. In particular, the increasing latency and energy cost associated with data transfer between memory and processor give rise to the architectural bottleneck known as the "Memory Wall". To address these limitations, neuromorphic engineering aims to emulate the efficiency of the biological brain through in-memory computing. Within this landscape, atomic switch networks composed of randomly interconnected silver nanowires have emerged as promising physical substrates for reservoir computing. However, fully exploiting these architectures requires a rigorous computational understanding of how the localized and volatile dynamics of individual memristive junctions determine the collective behavior of the network. The mathematical model of an electrochemical atomic switch junction constitutes the fundamental building block for the numerical simulation of electrical transport in neuromorphic nanowire networks. This thesis investigates mathematical models describing nanowire junctions as memristive elements. These models attribute the evolution of the junction resistance to the growth and dissolution of a conductive filament through the otherwise insulating coating surrounding the nanowires. An in-house numerical program is employed to perform parametric investigations of the junction-resistance model and to evaluate the influence of material properties and electrical loading conditions. Particular emphasis is placed on the role of spontaneous filament dissolution, which governs the device's characteristic volatility and time-dependent conductance decay. As an intermediate step toward the analysis of complete nanowire networks, the mutual interaction among multiple memristive junctions is investigated through simplified network motifs. Two representative configurations are considered, each reducing to an equivalent resistor network with time-dependent junction resistances. This mesoscopic analysis enables the characterization of the spatiotemporal redistribution of electrical conductivity and the dynamic formation of preferential conductive pathways within the network. Overall, the proposed computational framework offers a systematic approach for investigating how the intrinsic volatility and morphological sensitivity of individual junctions give rise to the collective nonlinear dynamics underlying the operation of neuromorphic nanowire networks.