Predicting the eigenstructures of metamaterials with QR-code meta-atoms by deep learning

Deep neural networks (DNNs) facilitate the reverse design of metamaterial perfect absorbers (MPAs), usually by predicting the MPA structure from the input absorptivity. However, this suffers from the difficulty that the spectrum that actually exists is unknown before the structure is known. We propose an MPA structure with quick response (QR)-code meta-atoms and construct a novel DNN to predict and reverse design the eigenstructures by inputting designated eigenfrequencies. In addition, the meta-atom has a tremendous number of degrees of freedom, providing rich properties such as multiple absorption peaks. This work paves the way for the study of eigenproblems of complicated metamaterials and metasurfaces.