Machine Learning-Enabled In Situ Diagnostics for Intelligent Plasma-Based Semiconductor Manufacturing: A Review

Semiconductor device architectures are becoming more complex, and process tolerances are tightening; therefore, plasma-based manufacturing increasingly requires predictive and adaptive control beyond conventional experience-driven methodologies. Machine learning (ML) is a powerful tool in process intelligence that transforms high-dimensional plasma diagnostics into actionable insights for real-time decision making. This review examines recent progress in ML for intelligent diagnostics in plasma-based semiconductor manufacturing from two perspectives: equipment-based classification (plasma-enhanced chemical vapor deposition, reactive ion etching, and sputtering) and ML application-focused classification (anomaly detection, plasma diagnostics, and predictive maintenance). This dual framework highlights the transformative effects of these technologies on semiconductor plasma processing, including real-time data analysis, process optimization, and in situ diagnostics, which demonstrates the current state of advancement in the era of artificial intelligence (AI). Recent studies have shown that ML models can successfully capture nonlinear plasma behavior, enable virtual metrology (VM), early fault detection, and process optimization, and support noninvasive monitoring compatible with high-volume manufacturing environments. However, despite rapid progress, several barriers remain, including the limited availability of labeled data sets, difficulty of model interpretability, poor domain transferability, and persistent gap between laboratory validation and fab-scale deployment. This review highlights emerging strategies that can improve robustness and industrial readiness, such as physics-informed learning, multimodal sensor fusion, uncertainty-aware modeling, and explainable AI. Moreover, it establishes a technological roadmap for intelligent and autonomous plasma manufacturing.