Deep Learning-based Inverse Design of the Complete Photonic Band Gap in Two-Dimensional Photonic Crystals

Abstract

Background: With the continuous development of computer science, data-driven computing methods have shown their advantages in various fields. In the field of photonics, deep learning (DL) can be used to inversely design the structure of optical devices. Objective: The two-dimensional (2D) photonic crystal (PCs) with adjustable structural parameters and a large complete photonic band gap (CPBG) are inversely designed in terms of DL neural network (NN) tagged to obtain a specified width of CPBG. Methods: The new PCs structure is designed by combining multiple factors that produce a CPBG. Tandem networks are used to speed up the training of the NN and tackle the problem of nonuniqueness that arises in inverse design. Results: After various attempts and improvements, the ideal PCs structure was obtained. It is found that the connecting channel between the primitives in the PCs unit cell has a dominate effect on the CPBG. The use of a tandem network enables better convergence of the network. Finally, suitable NN can be obtained, which can realize the forward prediction of the CPBG and the inverse design of the structure. Conclusion: DL can realize forward prediction and inverse design of 2D PCs targeting the width of the CPBG, which broadens the application scope of DL in the field of PCs.