Grating waveguides by machine learning for augmented reality
Xi Chen, Dongfeng Lin, Tao Zhang, Yiming Zhao, Hongwei Liu, Yiping Cui, Chenyang Hou, Jingwen He, Sheng Liang
Abstract
We propose a machine-learning-based method for grating waveguides and augmented reality, significantly reducing the computation time compared with existing finite-element-based numerical simulation methods. Among the slanted, coated, interlayer, twin-pillar, U-shaped, and hybrid structure gratings, we exploit structural parameters such as grating slanted angle, grating depth, duty cycle, coating ratio, and interlayer thickness to construct the gratings. The multi-layer perceptron algorithm based on the Keras framework was used with a dataset comprised of 3000-14,000 samples. The training accuracy reached a coefficient of determination of more than 99.9% and an average absolute percentage error of 0.5%-2%. At the same time, the hybrid structure grating we built achieved a diffraction efficiency of 94.21% and a uniformity of 93.99%. This hybrid structure grating also achieved the best results in tolerance analysis. The high-efficiency artificial intelligence waveguide method proposed in this paper realizes the optimal design of a high-efficiency grating waveguide structure. It can provide theoretical guidance and technical reference for optical design based on artificial intelligence.