Abstract The coating process is critical in determining the performance and yield of finished semiconductors. However, conventional parameter adjustment methods rely on previous experiences and do not ensure high precision and efficiency, especially when new materials and processes are introduced. Therefore, we developed a model to improve the coating process using chemical vapor deposition (CVD) and machine learning to enhance process performance, energy efficiency, and processing time. We identified parameters that have the most significant impact on coating quality through feature selection using analysis of variance, principal component analysis, Lasso regression, and Mutual Information (MI). As a result, 91 parameters were selected, including pressure, temperature, gas flow rate, processing time, and the coating quality index of photoluminescent tungsten disulfide (WS₂), which was the most important optimization target. A model was developed using machine learning algorithms such as support vector machine, random forest, linear regression, and extreme gradient boosting. The optimal combination of models for the optimization of parameters was selected based on evaluation metrics. The optimized model significantly improved the stability, yield, and quality of coating, while enhancing process accuracy and reproducibility. The model also reduced energy consumption and processing time, which led to lowered production costs and environmental impact, thereby increasing manufacturing efficiency. Due to its high adaptability and scalability, the model can apply to various processes with different materials in manufacturing high-end electronic products. The developed data-driven optimization model ensures the automation and intelligent transformation of semiconductor manufacturing.
Tu et al. (Mon,) studied this question.