Related Experiment Video
Updated: Apr 27, 2026
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.8K
Evolutionary fuzzy ARTMAP neural networks for classification of semiconductor defects
Summary
This study presents novel evolutionary artificial neural networks (EANNs) for wafer defect detection in semiconductor manufacturing. These models effectively address imbalanced datasets and overlapping samples, improving quality control.
Area of Science:
- Semiconductor Manufacturing
- Artificial Intelligence
- Machine Learning
Background:
- Wafer defect detection is crucial for semiconductor quality improvement, aiming to enhance process stability, production capacity, and yields.
- Imbalanced datasets, where defective units are a minority, pose significant challenges for traditional machine learning techniques.
- Overlapping samples within datasets further complicate accurate classification in manufacturing environments.
Purpose of the Study:
- To introduce novel evolutionary fuzzy ARTMAP (FAM) neural network models for addressing imbalanced datasets in semiconductor manufacturing.
- To develop effective machine learning solutions for classifying defective units despite data scarcity and class overlap.
- To enhance the robustness and accuracy of intelligent systems for wafer defect detection.
Main Methods:
- Integration of fuzzy ARTMAP (FAM) neural networks with hybrid genetic algorithms to create evolutionary artificial neural networks (EANNs).
- Development of two distinct EANN models, with one specifically designed to handle overlapping samples.
- Classification of imbalanced datasets using the proposed EANN models.
Main Results:
- The proposed evolutionary FAM neural networks demonstrated effectiveness in classifying imbalanced datasets.
- One EANN model successfully learned from overlapping samples within the imbalanced data environment.
- Comparative analysis using various classification metrics confirmed the positive outcomes of the developed networks.
Conclusions:
- The developed evolutionary FAM neural networks are effective in handling classification problems with imbalanced data in semiconductor manufacturing.
- The proposed models offer a viable solution for improving quality control through intelligent wafer defect detection.
- The integration of evolutionary algorithms and FAM networks provides a powerful approach for complex manufacturing data challenges.
