Proposal for an Artificial Neural Network Tool for the Process of Generating Metasurface Unit Cell Geometries

Abstract

This paper focuses on presenting an intelligent model that can generate the desired geometry of a unit cell metasurface for a given resonant frequency at which we expect the metasurface structure to work. The model consists of the use of a multilayer perceptron and filters, which represent the output geometry of the unit cell as a 6 × 6 matrix stored in a binary state. The value 0 in the matrix denotes the dielectric substrate on which the geometry of the unit cell is built, and the value 1 denotes the blocks as the conducting parts of the unit cell metasurface. The proposed model was tested using simulation data from the Comsol Multiphysics environment. The test confirmed the effectiveness of the model, and it is possible to develop and apply it to larger and other datasets.