Semiconductor Defect Classification Using Synthetic Minority Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect classification systems in semiconductor manufacturing face challenges with accurately identifying minority class defects due to insufficient training data, leading to misclassification, inaccurate process evaluation, and potential severe defects going unnoticed.
Innovation Solution
A system and method that utilize a processing and memory circuitry to select a subset of attributes, generate a temporary training set, and train an engine to improve classification accuracy by differentiating between majority and minority class defects, with the ability to generate synthetic defects and update the classifier based on confidence levels, thereby correcting misclassifications and optimizing classification results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a classifier is trained on available defect data, then classification can be performed, but minority class defects are misclassified due to insufficient training data
Solution Approach 1:
The patent creates synthetic copies of minority class defects by generating artificial defect images that mimic the characteristics of rare defect types. These synthetic defect images are generated using various transformation techniques (rotation, scaling, adding noise) to existing minority class defect samples, thereby creating an expanded training dataset that enables the classifier to learn and accurately identify minority class defects without requiring additional real defect samples
Solution Approach 2:
The patent performs preliminary data preparation by identifying and separating minority class defects from the training dataset before actual model training. This preliminary action involves detecting underrepresented defect classes, extracting their characteristics, and preparing synthetic defect images in advance to augment the training data, ensuring that the classifier receives balanced and sufficient training examples for all defect classes before deployment
2Measurement precision
If classification is performed with limited training data, then processing speed is maintained, but classification accuracy deteriorates
Solution Approach 1:
The patent generates synthetic defect images through digital copying and transformation of existing minority class defect samples. This approach creates additional training data without requiring physical collection or manual annotation of real defects, thereby improving classification accuracy for minority classes while avoiding the time-consuming processes of real defect collection, preparation, and annotation
Solution Approach 2:
The patent transforms existing defect images by modifying parameters such as rotation angles, scale factors, brightness levels, and noise levels to generate diverse synthetic defect variations. These parameter changes create a richer training dataset that improves model generalization and accuracy without requiring additional training time, as the transformations are applied computationally to existing data
3Productivity
If all defects are classified into majority class, then processing simplicity is maintained, but minority class defects are overlooked
Solution Approach 1:
The patent implements preliminary analysis to identify minority class defects in the training data before model training. This preliminary step involves calculating class distribution, detecting underrepresented defect types, and preparing synthetic defect images specifically for minority classes. By performing this preparation in advance, the system ensures comprehensive defect detection without requiring complex real-time adjustments during classification
Solution Approach 2:
The patent generates synthetic defect images for minority classes to balance the training dataset. This copying approach ensures that minority class defects receive adequate attention during training, enabling the classifier to distinguish them from majority class defects. The synthetic images are created through systematic transformations of existing minority class samples, providing sufficient training examples without increasing system complexity
Data Source
AI summary
There are provided a classifier and method of classifying defects in a semiconductor specimen. The method comprises receiving defects classified into a majority class, each having values for plurality of attributes, some defects belonging to a minority class, and some to the majority; selecting an attribute subset and defining differentiators for attributes wherein a second classifier using the subset and differentiators classifies correctly to minority and majority classes at least part of the defects; generating a training set comprising: defects of the majority and minority classes, and additional defects which the second classifier classifies as minority; training, upon the training set, subset, and differentiators, an engine obtaining a confidence level that a defect belongs to the majority class; applying the engine to second defects classified to the majority class, to obtain a confidence level of classifying each defect to the majority class; and outputting defects having a low confidence level.