Inverse design of electromagnetically induced transparency(EIT) metasurface based on deep convolutional generative adversarial network

Abstract With the increasing complexity of electromagnetically induced transparency (EIT) metasurface structure and the limitations of traditional optimization methods, there is an urgent need for an advanced design approach to accelerate the design of complex EIT metasurface. In this study, we propose an improved deep learning model based on deep convolutional generative adversarial network (DCGAN) to simplify the design process of EIT metasurface. The proposed model enables the optimization of metasurface with eight structural parameters, obtaining single-band or dual-band EIT effects for y polarization incidence. The training of the network model achieves convergence with a mean square error (MSE) of 0.2 for the generator and 0.41 for the discriminator. The average errors between the predicted results and the target parameters are within 0.6 μm. The relative spectral error (RSE) is utilized to quantify the deviation between the simulated spectra obtained from the predicted structures and the target spectra, exhibiting a minimum RSE of 6.26%. Furthermore, comparisons between the proposed model and two other convolutional neural networks validate the superior prediction capability and higher accuracy of our proposed model.