Investigating the impact of bidispersity on spin-coated polystyrene thin films using machine learning

The thickness of spin-coated polystyrene (PS) thin films largely influences their physical properties. While industrial applications primarily utilize polydisperse PS, existing models address only monodisperse systems. We created bidisperse PS films across varying concentrations, molecular weights, and blend ratios, evaluated monodisperse models with bidisperse data, and tested other machine learning models. We discovered that bidisperse films were systematically thinner than monodisperse films of equivalent weight average molecular weight, with the overlap parameter (c/c*) emerging as a key predictor. Our Gaussian Process Regression achieved MAPE = 3.82% (63% improvement over monodisperse models), R² = 0.9919, and RMSE = 75.2 Å.