Phd in MATHEMATICAL MODELS FOR ENGINEERING, ELECTROMAGNETICS AND
NANOSCIENCES

Thesis title: Reconfigurable Plasmonic Networks Via Hybrid Plasmonic-Solitonic Interconnections

This thesis documents the culmination of an extensive research and development endeavor focused on the design, implementation, and optimization of a highly efficient and compact photonic system inspired by biological neural networks. The primary objective was to create a system capable of emulating the intricate functionalities of biological neurons and their interconnections while addressing inherent limitations through material adjustments and geometric refinements. The journey commenced with a fundamental goal of miniaturization and increased density of neurons and interconnections, guided by the aspiration to closely mimic biological neural networks. Central to our approach was the integration of memory and processing functions, achieved passively through the utilization of photorefractive materials. This passive integration, devoid of active control mechanisms, ensured seamless operation while enabling nuanced and adaptive responses to incoming signals. Key milestones in the development process included the introduction of solitonic waveguides within a simple geometry, demonstrating the feasibility of generating solitons from diffracted light of Surface Plasmon Polaritons (SPP). This milestone required addressing challenges related to diffraction losses and divergence angles, culminating in efficient soliton generation and distribution. Further refinements were achieved through innovative beam splitter designs, enabling balanced signal distribution over short distances and enhancing system compactness. Transitioning to Metal-Insulator-Metal (MIM) waveguides optimized signal propagation, reducing losses while facilitating angle-insensitive coupling and backpropagation procedures vital for photonic neural networks. The system's evolution culminated in the realization of increased interconnection density, enabling 4-digit configurations with supervised and unsupervised training capabilities. These advancements underscore the robustness and versatility of the developed system, which can accommodate a wide range of machine learning algorithms, demonstrating its potential in various applications requiring complex signal processing and decision-making tasks. This thesis contributes to the growing body of research in photonic systems and their application in computational and cognitive domains. The iterative process of innovation and refinement outlined herein serves as a blueprint for future endeavors aimed at pushing the boundaries of photonic technologies in advancing computational and cognitive systems.