Light Powered Artificial Intelligence Chips: Experimentation and Optimization of the First Photochromic, Photonic Memristor

Abstract

Memristors, first hypothesized by Leon Chua as the fourth fundamental circuit element, offer notable applications for in memory computing due to their analog switching properties and non-volatile behavior. Photonic memristors, in particular, are a breakthrough technology that offer ultrafast switching and superior energy efficiency. Yet, to date, these devices have predominantly relied on thermally driven phase change materials (PCMs), limiting integration density and resulting in degradation. This thesis pioneers the simulation and design of the first photochromic-actuated photonic memristor, extending the foundational work in ‘Dynamics of A Photochromic-Actuated Slot Microring Photonic Memristor. By leveraging reversible isomerization in diarylethene molecules confined within a slot microring resonator, the device achieves switching governed purely by photon exposure, effectively eliminating thermal bottlenecks.

A full stack simulation framework was developed using Tidy3D and Python to model photoinduced state dynamics. This work verifies the three canonical fingerprints of memristive behavior, including hysteresis loop formation, area decay above a critical threshold observed at 5 Hz, and eventual collapse into a quasi-linear regime. This algorithm also performs radius tuning to optimize memory, yielding an optimal length of 104.3 microns, and verifies that hysteresis can arise from pure phase modulation without reliance on lossy mechanisms. Beyond design, this thesis creates several original optimization techniques: a dual-wavelength waveguide enabling co-confinement of UV and visible light to eliminate bulky flood illumination, and the construction of a novel photochromic neurosynaptic network. Novel integration strategies–such as quantum-dot enhanced light delivery–and molecular engineering via a custom ternary phase diagram of chemical substituents, further extend the design space. Translating theoretical equations into physically simulated architectures shows that the memristor is a viable method of high-performance, on-chip photonic computing.