Cluster Analysis

Cluster Analysis in semiconductor manufacturing is the unsupervised grouping of wafers, lots, or process runs into similar clusters — identifying natural groupings in process data that may correspond to different process states, equipment conditions, or failure modes.

Common Clustering Methods

• K-Means: Partition data into $K$ clusters minimizing within-cluster variance.
• Hierarchical: Build a dendrogram of nested clusters by iterative merging/splitting.
• DBSCAN: Density-based clustering that finds arbitrary-shaped clusters and identifies outliers.
• Gaussian Mixture Models: Probabilistic soft clustering with cluster shape flexibility.

Why It Matters

• Process Grouping: Identifies that wafers naturally fall into distinct groups (good vs. marginal vs. bad).
• Equipment Comparison: Clusters tool-to-tool variation to identify systematic equipment differences.
• Failure Classification: Groups defect signatures into categories for automated root cause analysis.

Cluster Analysis is finding natural groups in fab data — letting the data reveal its own structure for equipment matching, failure classification, and process optimization.