Design and Development of a Machine Learning Model for Thin Film Thickness Prediction

Abstract

Thin film technology plays a vital role in numerous applications, including semiconductors, photovoltaics, and optical devices. Accurate prediction of thin film thickness is critical for process optimization and quality control. In this study, a machine learning (ML) model was developed to predict the thickness of chemically deposited thin films, including CdS, CdSe, and MnO₂, based on material type, molar concentration, deposition time, and optical interference fringes. The model utilizes four key input features: material type, molar concentration, deposition time, and number of optical interference fringes. A Lasso Regressor was selected in this study. The model was trained using experimental data, validated on an independent dataset, and tested to assess generalization performance. The developed model demonstrated high predictive accuracy, with mean absolute percentage errors (MAPE) under 1% across all phases, showcasing its potential as a reliable tool for in-situ thickness estimation and process tuning. The model achieved high accuracy, with Root Mean Squared Error (RMSE) values below 0.07 µm across all materials and R² scores above 0.95, indicating strong generalization. Validation results show minimal error between predicted and actual thickness values, with an average prediction error below 2% for most cases.