Machine Learning–Driven Optimization of Microcrystalline Silicon Deposition for Display Applications via Raman Spectroscopy and Random Forest Regression

ABSTRACT This study addresses the critical challenges in developing microcrystalline silicon (μc‐Si) processes for display applications, where limited industrial expertise in crystallization characterization and the absence of standardized experimental protocols hinder optimal process optimization. We present a machine‐learning‐driven framework employing Random Forest regression to quantitatively model the relationships between key deposition parameters—chamber pressure, plasma power ( P ), substrate temperature ( T ), and gas flow ratio ( R )—and crystallinity metrics. Through systematic feature engineering and hyperparameter optimization, the proposed model achieved state‐of‐the‐art predictive performance, yielding a mean squared error (MSE) of 20.1172 and a coefficient of determination ( R 2 ) of 0.94 on validation datasets. Feature selection was conducted in two sequential phases: (1) preliminary screening via the Pearson correlation coefficient to remove numerical noise and weakly correlated variables, and (2) identification of second‐order interaction terms, followed by feature ranking based on Random Forest importance scores.