A Deep Learning-Based Approach to Design Metasurfaces From Desired Far-Field Specifications
Chen Niu, Mario Phaneuf, Tianke Qiu, Puyan Mojabi
Abstract
A deep learning neural network model in conjunction with a method to incorporate auxiliary surface waves is developed for the macroscopic design of transmitting metasurfaces. The main input to the neural network model is the user-defined desired far-field specifications. This network is used to calculate the required tangential electromagnetic fields on the metasurface. These fields will then be augmented by incorporating auxiliary surface waves along the metasurface for power redistribution to satisfy the requirement for having lossless and passive metasurfaces. The designs will then be evaluated using full-wave simulations of metasurfaces with three-layer unit cell topology in both 2D and 3D scenarios.