SVM (Support Vector Machine) for semiconductor classification is the application of maximum-margin classifiers to separate process conditions or wafer types — finding the hyperplane that maximally separates classes in feature space, with kernel functions handling non-linear boundaries.

How Does SVM Work?

• Margin: Find the hyperplane that maximizes the distance to the nearest data points (support vectors).
• Kernel Trick: Map data to higher-dimensional space (RBF, polynomial kernels) for non-linear boundaries.
• Soft Margin: Allow some misclassifications (controlled by parameter $C$) for noisy data.
• Multi-Class: One-vs-one or one-vs-all strategies for multi-class problems.

Why It Matters

• Small Datasets: SVMs excel when training data is limited — common early in a new process development.
• Feature Space: Kernel SVMs can model complex, non-linear decision boundaries efficiently.
• Defect Classification: Effective for wafer map pattern classification and defect type identification.

SVM is the maximum-margin classifier — finding the widest possible gap between classes for robust classification of semiconductor data.