Predicting Fundamental Transverse Electric Mode of Slab Waveguide Based on Physics-Informed Neural Networks

Abstract

Over the past few years, deep learning has proven to be an effective and fast tool in many areas, especially in the field of photonics. The design of integrated optical devices relies on optical waveguides which requires a reliable and fast methods for determining the waveguide’s characteristics before the fabrication process. In this work, the newly emerging paradigm of physics-informed neural networks (PINNs) is employed for analyzing and predicting the fundamental transverse electric (TE) mode and effective refractive index ( ) of a slab waveguide. PINNs is particularly useful as it is a data and mesh-free method, which solve the most critical problems of computational modeling which are the speed and computational hardware cost. The suggested model has a prediction accuracy of up to 99% and effective refractive index relative error ranges between 10-5 and 10-6. Model results are validated against finite difference time domain method using Lumerical software and variational method.