Deep Neural Network Training Incorporating Equivalent Circuit Model for Metasurface Design

In this paper, we propose an efficient method for designing metasurfaces. A deep neural network (DNN) is trained using data from various metasurfaces, including their S-parameters extracted from full-wave (HFSS) simulation results. Notably, the capacitance and inductance parameters from the equivalent circuit model of the metasurfaces are incorporated for the first time in training the model and constructing the network. To verify the performance of the proposed method, we apply the numerical results obtained from a slot-based metasurface structure. Numerical experiments demonstrate that, with a limited number of epochs, this approach significantly enhances the accuracy of the trained DNN model. In other words, this method offers a more accurate and efficient neural network-based solution for metasurface design.