Wafer Electrical Test Prediction Using Shot-Region Sampling

The patent performs electrical tests on a subset of shot regions (e.g., 4 out of 16 regions) as a preliminary action, then uses machine learning to predict the electrical test characteristics for the remaining unmeasured regions. This preliminary sampling combined with predictive modeling achieves high prediction accuracy without requiring extensive physical testing of all regions, thus resolving the contradiction between measurement precision and productivity.

The patent introduces machine learning algorithms as an intermediary between the measured electrical test data and the unmeasured regions. The ML model acts as a mediator that infers characteristics of untested regions based on patterns from tested regions, enabling accurate predictions without direct physical measurement of all areas, thereby improving both accuracy and efficiency.

The patent creates a virtual copy of the physical testing process through machine learning models. Instead of physically testing all regions, the system trains ML models on a subset of measured data to replicate and predict test outcomes for unmeasured regions. This copying approach enables efficient prediction with reduced physical testing while managing computational complexity through standardized ML frameworks.

The patent develops a universal machine learning framework that can predict multiple electrical test characteristics simultaneously across different shot regions. The same ML infrastructure handles various test parameters (current, voltage, resistance) and multiple spatial regions, reducing overall system complexity compared to separate specialized systems for each measurement type.

The patent performs electrical tests on a preliminary subset of shot regions (e.g., 4 regions) and uses machine learning to predict results for remaining regions. This preliminary action combined with computational prediction achieves comprehensive test coverage information without requiring sequential physical testing of all 16 regions, significantly reducing testing time while maintaining measurement precision.

The patent replaces the mechanical/physical process of sequentially testing all shot regions with a computational substitution. After measuring a subset of regions, the system uses machine learning algorithms to substitute for the remaining physical measurements, inferring test results through computational modeling rather than continued physical testing, thus reducing time loss while maintaining comprehensive data coverage.