Predictive Maintenance
Sensor Data & Feature Engineering
Equipment traces, FDC data, time-series features, and health indicators
Understanding FDC Data
Understanding FDC Data
FDC (Fault Detection and Classification) data is the foundation of PdM in fabs. Understanding its structure is essential:
- Trace data: Time-series sensor readings during each process run. Example: chamber pressure sampled at 10 Hz during a 60-second etch step = 600 data points per parameter per run.
- Summary statistics: Aggregated values per run — mean, std dev, max, min, slope of each parameter during each recipe step. This is often the starting point for ML models.
- Context data: Which chamber, which recipe, lot ID, wafer slot, timestamp, PM history, consumable age.
Key Concept: Feature Engineering for Equipment Data
Raw sensor time series must be transformed into meaningful features. Common approaches:
- Step-level statistics (mean, std, trend, range per recipe step)
- Deviation from golden trace (DTW distance, residual analysis)
- Rolling statistics over recent runs (moving average, EWMA)
- Time since last PM, cumulative RF hours, wafer count
- Rate of change features (how fast is a parameter drifting?)
Building Health Indicators
Building Health Indicators
A health indicator (HI) is a derived metric that tracks equipment degradation over time. Good HIs should:
- Monotonically degrade: Consistently trend in one direction as the component ages
- Be prognostic: Start changing well before failure occurs, giving time to plan maintenance
- Be interpretable: Engineers should understand what physical degradation the HI reflects
Example HIs for a PVD chamber:
- Target life remaining (based on deposition rate trend and voltage drift)
- Chamber matching score (statistical distance from reference chamber)
- Particle adder trend (from periodic blank wafer inspections)
Analogy: Blood Pressure for Machines
Health indicators are like vital signs for equipment. Just as a doctor monitors blood pressure, heart rate, and cholesterol to predict cardiovascular risk, a PdM system monitors equipment vitals to predict maintenance needs.
Knowledge Check
Knowledge Check
1 / 2What is FDC data in semiconductor manufacturing?