Deployment & Operations

Real-time inference, alert systems, maintenance scheduling, and ROI

Deploying PdM in Production

Moving from a Jupyter notebook to a production PdM system involves significant engineering:

Data pipeline: Real-time ingestion of FDC data from 1,000+ tools, cleaning, feature computation, and storage. Must handle missing data, sensor failures, and recipe changes.
Model serving: Low-latency inference after each process run (seconds, not minutes). Models must handle multi-chamber, multi-recipe scenarios.
Alert management: Converting model scores into actionable alerts. Too many false alarms = alert fatigue (engineers ignore them). Too few = missed failures.
Integration with MES: Alerts flow into the Manufacturing Execution System for maintenance scheduling and wafer routing decisions.
Model monitoring: Track model performance over time. Equipment changes (new PMs, recipe updates) can invalidate models — requiring retraining or adaptation.

Key Concept: ROI of PdM

A successful PdM system typically delivers 5–15% reduction in unplanned downtime and 10–20% reduction in maintenance costs. For a large fab, this translates to $10–50M annual savings. The ROI is compelling, but achieving it requires strong data infrastructure and close collaboration between data scientists and equipment engineers.

ML Models for PdM

Defect Types in Fabs

Knowledge Check

1 / 1

What is the biggest operational challenge in deploying PdM?

Balancing alert sensitivity (catching failures) with specificity (avoiding false alarms)Training the ML modelCollecting any sensor data at allGetting management approval