Machine learning-driven analysis of thickness variations in multilayer thin film coatings for photonics

Multilayer thin film coatings are vital for photonics, yet their optical properties are highly sensitive to thickness variations and material parameters. We developed a hybrid approach combining Monte-Carlo and FDTD simulations, generating a dataset of over 95,000 scenarios to investigate these effects. Neural Network (NN) modeling was applied to extract meaningful relationships between thickness fluctuations and optical performance. Validation experiments using magnetron sputtered coatings confirmed the model's predictions with R2 > 0.99. Benchmarking with multiple linear regression (MLR) demonstrated the NN’s superior performance. Experimental validation with SiO₂/Ta₂O₅ coatings confirmed the model’s reliability, establishing a robust tool for the photonics community. This work demonstrates a powerful synergy between computational simulations and machine learning, offering precise insights for optimizing thin film stacks in diverse photonic technologies.