Machine Learning-Enabled Inverse Design of Subwavelength Waveguide Gratings using Environment-Aware Effective Medium Approximations

Accurate design of complex grating structures using 3D simulations is computationally intensive. Our ultimate goal is to create a 2D effective index model of a 3D subwavelength waveguide structure to carry out inverse design of the device with accuracies comparable to 3D simulations. Here we present a deep neural network surrogate model to predict the optical response of the 2D effective structure as a first step.