Exploring Fourier Neural Operators for Electromagnetic Scattering

This paper proposes a novel learning framework to address electromagnetic scattering. Although conventional full-wave numerical methods, such as the Method of Moments (MOM), are effective, they are computationally intensive. On the other hand, machine learning-based algorithms can offer real-time solutions, but they are often limited by the extent of their training data. To integrate physical knowledge into neural network architectures, we employ neural operators, specifically the Fourier Neural Operator (FNO), to enhance data efficiency and accuracy in electromagnetic simulations. Our modified FNO model, trained on various dielectric distributions, achieves accuracy comparable to state-of-the-art architectures like UNet, but with less data. This framework bridges the gap between physics-based knowledge and data-driven approaches, paving the way for faster and more adaptive simulations.