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

Area of Science:

• Materials Science
• Chemical Engineering
• Computer Science

Background:

• Semiconductor manufacturing complexity and tighter process tolerances necessitate advanced control strategies.
• Traditional experience-driven methods are insufficient for modern plasma-based processes.
• Machine learning (ML) offers a powerful approach to interpret complex plasma diagnostics.

Purpose of the Study:

• To review recent advancements in ML for intelligent diagnostics in plasma-based semiconductor manufacturing.
• To analyze ML applications from both equipment and ML-focused perspectives.
• To identify challenges and propose future directions for AI in plasma processing.

Main Methods:

• Review of recent studies on ML applications in plasma-enhanced chemical vapor deposition, reactive ion etching, and sputtering.
• Classification of ML applications including anomaly detection, plasma diagnostics, and predictive maintenance.
• Examination of emerging strategies like physics-informed learning and explainable AI.

Main Results:

• ML models successfully capture nonlinear plasma behavior and enable virtual metrology.
• ML facilitates early fault detection, process optimization, and noninvasive monitoring.
• Significant progress has been made in real-time data analysis and in situ diagnostics.

Conclusions:

• ML is transforming semiconductor plasma processing, enabling intelligent diagnostics and real-time decision-making.
• Barriers such as data availability and model interpretability need addressing.
• Emerging strategies and a technological roadmap are crucial for industrial readiness and autonomous manufacturing.