Metamaterial Waveguide Modelling by an Artificial Neural Network with Genetic Algorithm

The need to predict data with more consistency provided studies that use Artificial Neural Networks (ANN) in applications with a satisfactory error rate. Traditional methods to optimize an ANN have shown good results, however, to achieve a greater degree of efficiency, a Genetic Algorithm (GA) was used in the parameterization, that is, a hybrid model was built that contributes to a significant increase in the performance of the Neural method. Here, we consider waveguides based on metamaterials made of thin-layer metallic and dielectric coatings surrounding a dielectric core. The main parameters to be considered are the length propagation (L-p) as the output of the ANN, as a function of the excitation wavelength of light (lambda), the metal filling rate (r) of the composite coatings as well as the indices of metal refraction (n(m)) and dielectric layers (n(d)), waveguide core width (d) and core refraction index (n(c)) which are the ANN inputs to the model. For the inverse design, the propagation length as input will generate new geometric parameters, which feed back the hybrid system until it reaches the stopping criteria.