Design of porous-core photonic crystal fiber based on machine learning approach

Abstract

The traditional numerical analysis in waveguide design can be time-consuming and inefficient. This is even more prominent in the THz region and with complex shapes and materials. As an alternative to overcome these drawbacks, we propose a machine learning (ML) approach to design porous-core photonic crystal fibers (PCFs) for the THz band. This method is based on an artificial neural network (ANN) model trained to predict key parameters such as the effective refractive index, effective area, dispersion, and loss values with accuracy and speed. In that sense, the network was trained to perform multiple-output regression of the above parameters. The training data for this model comes from numerical calculations that use the finite element method (FEM) to simulate and evaluate analytical expressions. Our results demonstrate the ML model's ability to capture the complex and nonlinear relationships between the input and output parameters and accurately predict the behavior of the THz PCF. Moreover, the proposed model has an inference time of ∼0.03584 s for a batch of 32 data sets, which substantially outperforms typical calculation times needed in FEM simulations for THz waveguide design. These results show that this approach is efficient and effective and has the potential to significantly accelerate the design process of PCFs for THz applications.