Joe Brindley, Oisin Boyle, Marcus Law, Benoit Daniel, Gencoa Ltd, Liverpool, United Kingdom
Residual gas analysis using Remote Plasma Emission Monitoring (RPEM) has been developed over the past decade as a diagnostic method for monitoring and control of vacuum deposition processes. While RPEM provides rich process data, translating this information into actionable outcomes remains a key challenge—data alone delivers limited value without clear operational insight.
Previous work demonstrated the feasibility of applying Artificial Intelligence (AI) to RPEM data to detect air leaks in vacuum systems. Building on this foundation, the present work introduces a multi-mode fault detection framework capable of distinguishing between air leaks, water leaks, and contamination events. This expanded capability enables more precise root-cause identification and faster corrective action.
Critically, fault detection alone does not address the central question: do these identified issues materially affect process success? To answer this, the AI framework has been further extended to predict overall process success directly from RPEM data. By linking fault signatures to measurable process outcomes, the approach moves beyond anomaly detection toward predictive process monitoring.
Initial case studies from sputter deposition processes are presented, demonstrating the system’s ability to classify fault modes and forecast process performance. The results highlight the potential of AI-enhanced RPEM to transition from passive monitoring to proactive process control and yield assurance in vacuum deposition environments.