virtual metrology

**Virtual Metrology (VM)** is a **prediction technique that estimates wafer quality metrics from process tool sensor data without making a physical measurement** — using machine learning models trained on historical process-metrology correlations to predict CD, thickness, and other parameters. **How Does Virtual Metrology Work?** - **Sensor Data**: Collect process parameters (temperature, pressure, gas flows, RF power, time, etc.) from the tool. - **Model Training**: Train ML models (regression, neural networks, random forests) on sensor data → metrology measurement pairs. - **Prediction**: For new wafers, predict metrology values from sensor data alone. - **Validation**: Periodically validate against actual measurements to detect model drift. **Why It Matters** - **100% Prediction**: Every wafer gets a predicted measurement, even without physical metrology. - **Excursion Detection**: Detects process excursions in real time from sensor signature anomalies. - **Cost Reduction**: Reduces the number of physical measurements needed (expensive, slow). **Virtual Metrology** is **predicting measurements without measuring** — using process sensor data and ML to estimate wafer quality for every wafer.