Silicon Wafer Weak Point Prediction via UCF Index Sampling

The patent replaces expensive EDA vendor pattern classification licenses with a custom-built, cost-effective pattern classification system using machine learning models trained on historical defect data. This disposable-like approach uses affordable computational resources instead of perpetual expensive licenses, achieving similar classification capability without the high licensing cost burden.

The patent creates a copy of the pattern classification functionality by training machine learning models on harvested patterns from ORC/MRC/DRC markers. Instead of relying on proprietary EDA vendor tools, the system replicates and enhances pattern classification capabilities through custom-trained models, eliminating licensing dependencies while maintaining or improving classification accuracy.

The patent performs preliminary clustering and prioritization of ORC/MRC/DRC flagged locations before full classification. By pre-grouping markers into clusters and identifying high-priority regions based on defect history and pattern severity, the system prepares a structured dataset that enables efficient subsequent processing, allowing comprehensive coverage without excessive processing time.

The patent segments the design into multiple regions and processes ORC/MRC/DRC markers in distributed parallel clusters. Each cluster independently classifies patterns in its assigned region, then results are aggregated. This segmentation enables full design coverage while reducing the processing burden on any single computational resource, achieving both completeness and efficiency.

The patent merges multiple EDA vendor tool decks (ORC, MRC, DRC, and custom DFM decks) into a unified defect prediction system. By combining the analytical capabilities of multiple tools and integrating their markers into a single machine learning framework, the system achieves superior predictive accuracy that leverages the strengths of each individual tool while maintaining manageable overall complexity through centralized coordination.

The patent creates a composite defect prediction system that combines multiple analytical approaches and data sources from different EDA tools. Like composite materials that combine different substances to achieve superior properties, the system integrates diverse marker types and classification methods to achieve predictive accuracy greater than any single tool could provide alone.