Design of Macroscale Optical Systems with Metaoptics Using Transformer-Based Neural Networks

Abstract

Metaoptics are thin, planar surfaces consisting of many subwavelength optical resonators that can be designed to simultaneously control the amplitude, phase, and polarization to arbitrarily shape an optical wavefront much in the same manner as a traditional lens but with a much smaller form factor. The incorporation of metaoptics into a conventional optical system spans multiple length scales between that of the individual metaoptic elements (< {\lambda}) and that of the entire size of the optic (>> {\lambda}), making computational techniques that accurately simulate the optical response of metaoptics computationally intractable, while more efficient techniques utilizing various approximations suffer from inaccuracies in their prediction of the optical response. To overcome the trade between speed and accuracy, we implement a transformer-based neural network solver to calculate the optical response of metaoptics and combine it with commercial ray optics software incorporating Fourier propagation methods to simulate an entire optical system. We demonstrate that this neural net method is more than 3 orders of magnitude faster than a traditional finite-difference time domain method, with only a 0.47 % deviation in total irradiance when compared with a full wave simulation, which is nearly 2 orders of magnitude more accurate than standard approximation methods for metaoptics. The ability to accurately and efficiently predict the optical response of a metaoptic could enable their optimization, further accelerating and facilitating their application.