Inverse design of dispersion for photonic devices based on LSTM and gradient-free optimization algorithms hybridization

Abstract

As an inherent property of optical devices, dispersion plays an important role in the areas of optical communication and nonlinear optics. Traditional dispersion optimization approaches are time-consuming and power-hungry. In this paper, to accelerate the design of dispersive optical devices, an indirect inverse design method based on the long short-term memory forward model combined with gradient-free optimization algorithms is proposed. In the case of photonic crystal fiber, the results show that the forward model can predict the group velocity dispersion (GVD) with an accuracy of up to 99.62%, and the calculation speed is more than one thousand times faster than the conventional numerical simulations. The prediction accuracy of the inverse model is higher than 93%, with a calculation time of less than 20 s. In the case of slot waveguide, the results show that the forward model can predict the GVD with a prediction accuracy of 96.99% and the inverse design accuracy goes to 99%. The proposed machine learning model offers an efficient tool for dispersion optimization in both fiber and waveguide platforms.