Automated Calibration Sample Selection for Die-to-Database Photomask Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current manual selection of calibration samples for die-to-database photomask inspection is subjective, incomplete, and prone to failure, leading to high first-time failure rates and reduced throughput due to the complexity of image rendering and the need for experienced operators.
Innovation Solution
An automated method using local binary pattern (LBP) analysis, clustering, and evaluation scores to select effective calibration samples, which reduces operator knowledge requirements and stabilizes the selection process, enabling standardized evaluation and improved first-time success rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual selection of calibration samples is performed by experienced operators, then the selection quality may be improved through expert judgment, but the process becomes subjective, incomplete, and prone to failure leading to high first-time failure rates
Solution Approach 1:
The system performs automatic calibration sample selection without requiring human operators to manually choose samples. The automated method uses LBP analysis and clustering algorithms to independently evaluate and select calibration samples from the reticle design data, eliminating the subjective and unreliable manual selection process while maintaining high selection quality through objective computational metrics
Solution Approach 2:
The manual operator-based selection system is replaced with an automated computational system using LBP (Local Binary Pattern) analysis and clustering algorithms. This substitution transforms the subjective human judgment process into an objective mechanical/computational process that consistently evaluates sample representativeness and diversity through standardized algorithms, thereby improving reliability
2Ease of operation
If manual selection of calibration samples is performed, then operator experience can guide the selection process, but the enormous data volume makes visual inspection of every sample impractical and the selection incomplete
Solution Approach 1:
The system extracts only the most representative and diverse calibration samples from the enormous dataset of reticle design data. Through LBP analysis and clustering, it identifies and extracts a small subset of samples that capture the essential variability of the entire dataset, eliminating the need to manually inspect every sample while maintaining complete and representative calibration coverage
Solution Approach 2:
The system transforms the evaluation of calibration samples from visual inspection to computational parameter analysis. By converting image data into LBP feature vectors and evaluating samples based on quantitative parameters such as cluster representativeness and diversity metrics, the system can comprehensively assess all samples in the enormous dataset without manual intervention, preventing information loss
3Reliability
If repeated inspection procedures are performed after first-time failure, then calibration can be achieved, but the process takes several hours and severely affects product throughput
Solution Approach 1:
The system performs preliminary automatic calibration sample selection before the actual inspection process begins. By pre-selecting optimal calibration samples using LBP analysis and clustering algorithms, the system ensures that calibration is accomplished in the first attempt without requiring repeated inspection procedures, thereby maintaining both high reliability and productivity
4Ease of operation
If automated methods are used for calibration sample selection, then operator knowledge requirements are reduced and selection stability is improved, but advanced algorithms and processing power are required
Solution Approach 1:
The system introduces LBP (Local Binary Pattern) analysis as an intermediary computational method between the raw reticle design data and the calibration sample selection decision. This intermediary transformation converts complex image data into standardized feature vectors that can be systematically clustered and evaluated, providing a structured bridge that automates the selection process while managing algorithmic complexity through established image processing techniques
Data Source
AI summary
A method for selecting samples of reticle design data patterns in order to calibrate the parameters based on which the reference image used in a die-to-database reticle inspection method is rendered, the method comprising the steps of applying local binary pattern (LBP) analysis to a plurality of samples to obtain a p-dimensional vector output for each of the plurality of samples, clustering the q-D data points to M groups, selecting one sample from each clustered group, calculating evaluation scores for the samples selected, and, selecting a portion of the M samples on the representativeness score and the diversity score.