Semiconductor Defect Classification Using Density Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection methods in semiconductor manufacturing fail to accurately classify new types of defects and excursion defects, leading to misclassification, inaccurate process evaluation, and potential severe issues due to underrepresentation of minority defect classes.
Innovation Solution
A system and method utilizing a processing and memory circuitry to select a subset of attributes, determine density estimation functions, and apply thresholds to classify defects into new classes, identifying outliers and excursions by analyzing attribute distributions and scoring defects based on these functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional defect classification methods are used, then known defect types can be identified, but new defect types and excursions are misclassified
Solution Approach 1:
The system performs preliminary density estimation and threshold determination using training data before actual defect classification. By pre-calculating density functions and thresholds for known defect classes, the system prepares classification criteria in advance, enabling accurate identification of both known defects and novel excursions that deviate from established patterns.
Solution Approach 2:
The patent introduces density estimation functions as an intermediary mechanism between raw defect attributes and classification decisions. These functions serve as mediators that transform attribute data into probabilistic density scores, which are then compared against thresholds to determine whether defects belong to known classes or represent new excursions, thereby improving classification accuracy for both familiar and novel defect types.
2Reliability
If all attributes are used for classification, then comprehensive defect characterization is achieved, but computational complexity increases
Solution Approach 1:
The system extracts and utilizes only the most relevant attributes for density estimation and classification by selecting attributes that provide meaningful discrimination between defect types. This extraction approach maintains comprehensive defect characterization while reducing computational complexity by focusing on the most informative features rather than processing all available attributes equally.
Solution Approach 2:
The patent transforms the classification problem by changing parameters from direct attribute comparison to density-based scoring. By converting raw attribute values into density estimates and comparing these against predetermined thresholds, the system achieves reliable classification with reduced computational burden, as threshold comparison is more efficient than complex multi-attribute analysis.
3Productivity
If minority defect classes are underrepresented in training data, then training speed is maintained, but classification accuracy for these defects deteriorates
Solution Approach 1:
The system incorporates feedback mechanisms where classification results and density estimation performance are continuously monitored. When minority defect classes are misclassified or show poor density separation, the system can adjust threshold values or reweight density estimates for those specific classes, thereby improving minority class detection accuracy without requiring extensive retraining and maintaining training efficiency.
Data Source
AI summary
A system, method and computer readable medium for classifying defects, the method comprising: receiving classified first defects, and potential defects, each first and potential defect having values for attributes; processing the first and potential defects to select a subset of the attributes that differentiates the first defects from the potential defects; obtaining first and second functions based on the first defects and potential defects, respectively; obtaining a first threshold for the first function, and a second threshold for a combination of the first and second functions; applying the first function and the second function to each potential defect to obtain first and second scores, respectively; and determining a combined score of the first and second scores; and indicating as a defect of a potentially new type a potential defect when the first score is lower than the first threshold or the combined score exceeds the second threshold.