multivariate analysis

**Multivariate Analysis (MVA)** in semiconductor manufacturing is the **statistical analysis of high-dimensional process and metrology data** — using techniques like PCA, PLS, and clustering to extract patterns, detect anomalies, and identify root causes from hundreds of correlated process variables. **Key MVA Techniques** - **PCA (Principal Component Analysis)**: Reduces dimensionality, identifies dominant variation patterns. - **PLS (Partial Least Squares)**: Relates process variables to quality outcomes. - **MSPC (Multivariate SPC)**: Hotelling T² and Q-statistic for multivariate process monitoring. - **Contribution Plots**: When MSPC detects an anomaly, contribution plots identify which variables caused it. **Why It Matters** - **Hundreds of Variables**: Modern process tools generate 100-1000+ sensor readings — univariate SPC cannot handle this. - **Correlated Variables**: MVA naturally handles correlations between variables (temperature, pressure, flow are interdependent). - **Root Cause**: Contribution analysis identifies which specific variables are responsible for detected anomalies. **MVA** is **seeing the big picture in process data** — extracting meaningful patterns from the overwhelming dimensionality of modern fab data.