decision trees for root cause

**Decision Trees for Root Cause Analysis** is the **application of decision tree algorithms to identify which process conditions split wafers into good and bad groups** — producing human-readable, interpretable rules that manufacturing engineers can directly use for root cause investigation. **How Are Decision Trees Used?** - **Features**: Process parameters from equipment data (chamber, recipe, gas flows, temperatures). - **Labels**: Pass/fail or yield categories from downstream measurements. - **Tree Structure**: Each node splits on the most discriminating variable — the path from root to leaf is an if-then rule. - **Pruning**: Control tree depth to prevent overfitting while maintaining interpretability. **Why It Matters** - **Interpretability**: Unlike black-box models, decision trees produce human-readable rules (e.g., "IF chamber B AND temperature > 405°C THEN yield < 90%"). - **Variable Ranking**: Variables appearing near the root are the most important discriminators. - **Fast Investigation**: Engineers can immediately test the identified conditions and verify the root cause. **Decision Trees** are **the automated detective for fab problems** — finding the simplest set of process conditions that separate good wafers from bad.