Thin-film neural networks for optical inverse problem

Abstract

The thin-film optical inverse problem has attracted a great deal of attention in science and industry, and is widely applied to optical coatings. However, as the number of layers increases, the time it takes to extract the parameters of thin films drastically increases. Here, we introduce the idea of exploiting the structural similarity of all-optical neural networks and applied it to the optical inverse problem. We propose thin-film neural networks (TFNNs) to efficiently adjust all the parameters of multilayer thin films. To test the performance of TFNNs, we implemented a TFNN algorithm, and a reflectometer at normal incidence was built. Operating on multilayer thin films with 232 layers, it is shown that TFNNs can reduce the time consumed by parameter extraction, which barely increased with the number of layers compared with the conventional method. TFNNs were also used to design multilayer thin films to mimic the optical response of three types of cone cells in the human retina. The light passing through these multilayer thin films was then recorded as a colored photo.