Efficient design of structural parameters and materials of plasmonic fano-resonant metasurfaces by a tandem neural network

Abstract

We address inverse design of plasmonic Fano-resonant metasurfaces by using a tandem neural network (TNN) which can correctly predict both materials and structural parameters of target spectra. To train this TNN, 19530 groups of data from asymmetric double bar (ADB) nanostructures of varied dimensional parameters and different materials (Ag, Cu, and Al) respectively were collected. Our approach successfully addresses a non-uniqueness problem that commonly exists in nanophotonic inverse design. Besides, we choose target spectra generated outside the collected dataset in order to test the applicability and robustness of the TNN, which proves that the developed TNN is able to retrieve the nanoparticles of appropriate sizes and compositing material matching well Fano-profiled unknown target spectra within the spectral window of study.