Range-Based SEM Non-Visual Defect Binning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for classifying wafer defects in semiconductor manufacturing are time-consuming, prone to human error, and require significant resources, especially for non-visual defects that are difficult to identify manually and require extensive training for each layer.
Innovation Solution
A system and method using a defect review tool, such as a scanning electron microscope, with a classifier that filters topographical, intensity, and energy attributes to identify non-visual defects across multiple layers, reducing the need for layer-specific classifiers by defining common attribute ranges for non-visual defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual classification of defects is performed, then defect identification can be done with human judgment, but it requires significant time and effort and introduces inaccuracies and inconsistencies
Solution Approach 1:
The patent replaces the mechanical human judgment system with an automated classification system that uses machine learning algorithms and image processing techniques. The system automatically analyzes defect images, extracts features, and classifies defects without human intervention, thereby eliminating time consumption and human errors while maintaining or improving classification accuracy.
Solution Approach 2:
The classification system is designed to be self-sufficient by automatically training classifiers using collected defect data. The system performs self-learning and adaptation without requiring external human expertise for each classification task, enabling it to handle diverse defect types autonomously and reduce dependency on manual classification efforts.
2Productivity
If layer-based automatic classification with custom classifiers is implemented, then classification can be automated, but it requires extensive resources including training data, human resources, and time for creating and maintaining classifiers
Solution Approach 1:
The patent creates a universal classification framework that can handle multiple defect types and layers using a single integrated system. Instead of requiring separate custom classifiers for each layer, the system uses a unified approach with configurable parameters that can adapt to different layers and defect types, significantly reducing the complexity of classifier creation and maintenance while maintaining high productivity.
Solution Approach 2:
The system enables flexible adaptation to different layers and defect types by changing parameters rather than creating entirely new classifiers. By adjusting classification parameters, thresholds, and feature weights, the same base classifier can be optimized for different contexts, reducing the need for extensive retraining and resource investment.
3Measurement precision
If training of classifiers for each layer is performed, then accurate classification for each layer can be achieved, but it involves huge repetitions in classifier creation for same defect types across layers
Solution Approach 1:
The patent segments the classification process into modular components: a base classifier framework that handles common defect types across all layers, and layer-specific parameter configurations that adapt to individual layer characteristics. This segmentation allows the system to reuse the core classification logic while making targeted adjustments for each layer, eliminating redundant classifier creation and reducing training time.
Solution Approach 2:
The system performs preliminary training on a comprehensive dataset that encompasses multiple layers and defect types before deployment. This preliminary training creates a robust base classifier that can be quickly adapted to specific layers through parameter adjustment rather than complete retraining, significantly reducing the time required for layer-specific classification setup.
Data Source
AI summary
A technique to identify non-visual defects, such as SEM non-visual defects (SNVs), includes generating an image of a layer of a wafer, evaluating at least one attribute of the image using a classifier, and identifying the non-visual defects on the layer of the wafer. A controller can be configured to identify the non-visual defects using the classifier. This controller can communicate with a defect review tool, such as a scanning electron microscope (SEM).