Bitmap Cluster Analysis for Semiconductor Defect Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional defect analysis systems in semiconductor manufacturing struggle to effectively classify random and systematic bitmap failures, often misclassifying nearby failed bits and failing to identify systematic patterns in a timely manner, which hinders yield improvement and quality control.
Innovation Solution
A bit clustering and aggregation system that automatically identifies and classifies random and systematic bitmap failed patterns by allowing users to set radius, minimum, and maximum counts for bit clusters, generating a bit cluster analysis report for easy interpretation, and providing flexible and powerful defect bitmap analysis tools.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If rigid predefined patterns are used to classify bit failures, then classification speed is improved, but classification accuracy deteriorates because random and systematic patterns cannot be distinguished
Solution Approach 1:
The system dynamically adapts the classification approach based on the characteristics of the bit failures being analyzed. For systematic patterns, it uses cluster analysis with radius and count parameters to identify regular geometries. For random patterns, it falls back to predefined pattern matching. This dynamic selection resolves the contradiction by applying the appropriate method for each case rather than using a single static approach.
Solution Approach 2:
The system changes classification parameters based on the detected pattern type. When systematic patterns are detected through cluster analysis, it uses geometric parameters (radius, minimum count, maximum count) to classify them. When random patterns are detected, it switches to using predefined pattern parameters. This parameter adaptation allows accurate classification of both pattern types without sacrificing speed.
2Ease of operation
If predefined patterns are used for classification, then ease of operation is improved, but the ability to identify systematic patterns deteriorates
Solution Approach 1:
The system performs preliminary cluster analysis on the bit failures before final classification. This preliminary action identifies potential systematic patterns by grouping nearby failures within a specified radius and checking against minimum and maximum count criteria. This preliminary step enables the system to detect systematic patterns that would otherwise be missed by predefined pattern matching alone.
Solution Approach 2:
The system uses feedback from the cluster analysis results to determine the appropriate classification approach. If clusters meet the systematic pattern criteria (appropriate radius, minimum count, maximum count), the system feedbacks this information to switch to systematic pattern classification. This feedback mechanism ensures that systematic patterns are reliably identified while maintaining ease of operation through automated detection.
3Measurement precision
If cluster analysis with multiple parameters is implemented, then pattern identification accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the pattern identification process into distinct phases: cluster formation based on radius and count criteria, systematic pattern detection, and random pattern classification. This segmentation allows each phase to focus on specific tasks using appropriate parameters, improving overall accuracy while managing complexity through modular organization of the analysis steps.
Solution Approach 2:
The system introduces cluster analysis as an intermediary step between raw bit failure data and final pattern classification. This intermediary process groups failures using radius and count parameters to identify systematic patterns before applying final classification. The intermediary cluster analysis acts as a bridge that enhances accuracy without requiring the final classification system to directly handle all the complexity of raw failure data.
Data Source
AI summary
A system and method for defect analysis are disclosed wherein a defect data set is input into the system. A radius value is selected by a user, which is the maximum number of bits that bit failures can be separated from one another to be considered a bit cluster. When a defect data set is received, the system and method start with a fail bit and search for neighboring fail bits. The specified radius is used to qualify the found fail bits to be part of the bit cluster or not. If a minimum count of fail bits is not met, the system and method will stop searching and move to the next fail bit. If a minimum count of fail bits is met, the search continues for the next fail bit until the maximum fail bit count specified by the user is reached. Aggregation is provided such that once bit clusters have been classified, the number of clusters that have the exact match or partial match to each other is counted. The user may set the partial match as a threshold count to establish a match.