Sophisticated electromagnetic scattering solver based on deep learning

In this paper, a deep learning (DL) framework is proposed to predict the scattering field, emerging superior efficiency without sacrificing accuracy. 2D and 3D scatterers in the scheme can be either lossless medium or metal. To achieve precise approximation, medium-scale data sets are sufficient in training the proposed model. As a result, the fully trained framework can realize three orders of magnitude faster than the conventional FDFD solver. Furthermore, our model also exhibits robust generalization ability in forecasting the field scattered by utterly distant from the training data set. We believe this work offers great potential for applications pertaining to EM forward problems.