Deep learning enabled inverse design of nanocrystal-based optical diffusers for efficient white LED lighting

Abstract

Angular color uniformity and luminous flux are the most important figures of merit for a white-light-emitting diode (WLED), and simultaneous improvement of both figures of merit is desired. The cellulose-nanocrystal (CNC)-based optical diffuser has been applied on the WLED module to enhance angular color uniformity, but it inevitably causes the reduction of luminous flux. Here we demonstrate a deep-learning-based inverse design approach to design CNC-coated WLED modules. The developed forward neural network successfully predicts two figures of merit with high accuracy, and the inverse predicting model can rapidly design the structural parameters of CNC film. Further explorations taking advantage of both forward and inverse neutral networks can effectively construct the coating layer for WLED modules to reach the best performance.