Data-driven investigation of thickness variations in multilayer thin film coatings

Abstract Design and fabrication of multilayer thin film coatings for photonics applications require careful consideration of various parameters such as layer thickness, refractive indices and number of stacks. A growing trend uses machine learning for efficient navigation in the complex parameter space of photonics applications to efficiently extract valuable insights from the extensive datasets and to predict the optical performance. We developed an approach that combines Monte-Carlo and Finite-Difference Time-Domain simulations to model multilayer thin films. After conducting 95 200 runs, the data were analyzed using Neural Network fitting to explore how thickness variations influence the optical performance. An experiment validation on magnetron sputtered coated samples demonstrates the high accuracy of our method in predicting the optical performance of the thin film stacks ( R 2 > 0.99), contributing to the understanding and enhancement of photonics stack properties for diverse optical applications using machine learning approaches.