Tunable grating surfaces with high diffractive efficiency optimized by deep neural networks

Abstract

High diffractive efficiency gratings, as a core component in optics, can engineer light transport and separation. This Letter predicts a grating surface with high diffractive efficiency within the visible light wave band with the aid of deep neural networks (DNNs). The predicted grating surface can have more than 99% diffractive efficiency for the −1th order within the bandwidth of ∼100 nm in the visible wave band, outperforming previously reported structures. Accordingly, the strategy of the DNN-aided design is an efficient and feasible method for optical devices. Moreover, changing the period of the predicted grating surfaces can shift the workable wave band, not only exhibiting the tunability but also bringing about the predicted gratings with more than 90% diffractive efficiency within the whole visible light wave band.