Title:Machine Learning for Mie-Tronics

Abstract:Electromagnetic multipole expansion theory underpins nanoscale light-matter interactions, particularly within subwavelength meta-atoms, paving the way for diverse and captivating optical phenomena. While conventionally brute force optimization methods, relying on the iterative exploration of various geometries and materials, are employed to obtain the desired multipolar moments, these approaches are computationally demanding and less effective for intricate designs. In this study, we unveil the potential of machine learning for designing dielectric meta-atoms with desired multipolar moments up to the octupole terms. Specifically, we develop forward prediction models to unravel the intricate relationship between the scattering response and the topological attributes of individual meta-atoms, and an inverse design model to reconstruct scatterers with the targeted multipolar moments. Utilizing a tandem network trained to tailor dielectric meta-atoms for generating intended multipolar moments across a broad spectral range, we further demonstrate the generation of uniquely shaped meta-atoms for exciting exclusive higher order magnetic response and establishing super-scattering regime of light-matter interaction. We also illustrate the accurate prediction of electric field distributions within the given scatterer. Our versatile methodology can be readily applied to existing datasets and seamlessly integrated with various network architectures and problem domains, making it a valuable tool for the design of different platforms at nanoscale.