Linking inclusion characterisics to stainless steel corrosion in semiconductor HCl environments via machine leaning approaches

Abstract

Micro-scale inclusions play a critical role in determining the corrosion performance of semiconductor-grade 316 L stainless steel, yet their quantitative influence remains insufficiently understood. In this work, an integrated experimental-data-driven framework was established to elucidate the corrosion-promoting effects of sub-2 μm inclusions under electronic specialty gas environments. Inclusion characteristics were experimentally characterised and correlated with corrosion responses, forming the basis for model development and validation. To address data scarcity, statistically verified Gaussian-noise and k-nearest-neighbor-based augmentation strategies were employed. Among eight machine-learning algorithms, a weighted ensemble model exhibited the highest predictive accuracy (R² = 0.95). Model predictions were further validated against independent corrosion experiment datasets rather than internal cross-validation alone. SHAP and partial dependence analyses revealed the importance features that small inclusions (0–2 μm), particularly silicates and spinels, significantly promote corrosion once their total area fraction and number density exceed critical thresholds. Based on these findings, a corrosion-recommendation map focusing on sub-2 μm inclusions was constructed for semiconductor-grade 316 L stainless steel. The results highlight inclusion engineering as an effective pathway to improving corrosion resistance in high-purity electronic specialty gas applications.