Prediction of toroidal dipole resonance in dielectric metasurface by deep learning

Abstract

Toroidal dipole (TD) resonance is a promising method for enhancing light–matter interactions, offering significant potential in photonic device design. While numerical simulations are commonly used to study TD resonances, they are computationally expensive and time consuming. In this study, we propose deep learning strategies to predict TD resonances induced by Brillouin zone folding. A fully connected neural network is developed to predict transmission mapping, transmission spectra, multipole scattering, and TD components. Comparison with numerical simulations shows that the neural network predicts TD resonance efficiently and accurately. Experimental validation through fabricated samples further confirms the strong TD response. Our work presents an effective tool for quickly and precisely exploring nanophotonic properties and offers a promising approach for predicting high-quality factor TD resonators.