Neural network modeling for electromagnetic structures

This paper presents an overview of emerging neural network (NN) modeling techniques for electromagnetic (EM) structures. Techniques including NN modeling for frequency and time domain simulations, NN inverse modeling, and NN modeling for EM-based simulations are discussed. NN models for EM structures are developed by training the NNs with EM data generated from either frequency or time domain EM simulators. After training, NNs become fast and accurate models of EM structures, which can be incorporated into various simulation methods to realize the analysis of different EM systems. Numerical examples show that simulations using NN models are much faster than conventional EM simulations, while maintaining high accuracy.