A Data-Aware Fourier Neural Operator for Modeling Spatiotemporal Electromagnetic Fields

Neural operators have recently emerged as a powerful tool for solving partial differential equations (PDEs), including Maxwell's equations in computational electromagnetics. Prior machine learning models often fail to capture both temporal field evolution and generalization to irregular geometries. Here, we introduce a Data-Aware Fourier Neural Operator (DA-FNO) as a surrogate solver. Applied autoregressively, it predicts the temporal evolution of all field components while monitoring energy dynamics, terminating automatically once energy converges. The model generalizes to complex geometries and the optical C-band without retraining, achieving a 7.5* speedup with nearly 92.5% accuracy. This approach offers a potentially efficient and accurate alternative to conventional iterative solvers for C-band photonic simulations.