Semiconductor Defect Review via Cluster-Seed Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in accurately detecting and classifying defects in semiconductor wafers due to errors from raw materials, mechanical, electrical, or optical issues, leading to false positives and inefficiencies in the examination process.
Innovation Solution
An examination system utilizing a processing and memory circuitry to assign cluster-seed scores to defects, enabling the selection of defects for review based on these scores, and employing a set of sampling utilities to calculate review scores, thereby optimizing the defect review process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If comprehensive defect inspection is performed on all defects in semiconductor wafers, then defect detection completeness is improved, but examination time and resources increase significantly
Solution Approach 1:
The system performs preliminary actions by examining a first subset of defects before the main inspection process. This preliminary examination allows the system to identify defect patterns and characteristics in advance, which are then used to guide the selection and inspection of a second subset of defects, reducing the overall time required for comprehensive defect detection.
Solution Approach 2:
The defect inspection process is divided into multiple segments: examining a first subset of defects, examining a second subset of defects, and selecting defects for review based on patterns identified in previous subsets. This segmentation allows the system to process defects in manageable portions while maintaining comprehensive coverage, balancing detection completeness with examination efficiency.
2Measurement precision
If all defects are reviewed in detail, then classification accuracy is improved, but review budget and resources are exceeded
Solution Approach 1:
The system applies partial action by reviewing only certain defects in detail rather than all defects. Defects are selected for detailed review based on patterns and characteristics identified during preliminary examination of defect subsets. This approach maintains classification accuracy for critical defects while reducing the overall review budget required.
Solution Approach 2:
The system uses the information gathered from examining defect subsets to automatically guide the selection and review process. The examination results from first and second defect subsets inform the automated selection of defects requiring detailed review, reducing the need for external expert intervention and optimizing resource allocation within the review budget.
3Productivity
If defect examination process is simplified, then processing speed is improved, but false positive findings increase
Solution Approach 1:
The examination process is segmented into multiple stages with increasing detail. The system first examines defect subsets to identify patterns, then uses these patterns to guide more detailed examination of specific defects. This multi-stage approach maintains processing speed by avoiding uniform detailed examination of all defects while reducing false positives through progressive verification.
Solution Approach 2:
The system implements feedback mechanisms where examination results from defect subsets are used to inform and adjust the examination of subsequent defect subsets. Patterns and characteristics identified in earlier stages provide feedback that improves the accuracy of defect classification in later stages, reducing false positives while maintaining processing efficiency.
Data Source
AI summary
A first defect map representing defects in a first semiconductor specimen in an attribute hyperspace may be received. Scores may be assigned to classified defects in the first defect map where an assigned score of a given defect of the classified defects in the first defect map is indicative of a number of defects within a threshold distance in the attribute hyperspace to the given defect in the first defect map that are classified to a same defect class as the given defect. A second defect map representing defects in a second semiconductor specimen in the attribute hyperspace may be received. Defects in the second defect map may be selected for review based on the scores assigned to the classified defects in the first defect map. The selected defects in the second defect map may be selected for classification.