Semiconductor Defect Classification Using Design Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect classification methods for semiconductor wafers rely on comparing defect images to reference images, lacking knowledge of the actual device design, leading to potential misclassification of defects based on location rather than type, and require extensive scanning processes, which are time-consuming and inefficient.
Innovation Solution
A method utilizing a defect classification engine that combines defect images with design information to characterize defects, allowing for classification based on both image and design data, including size, shape, topological, and luminance comparisons, and enabling the creation of synthetic reference images for faster processing without the need for separate reference images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional defect classification methods are used, then defect images can be classified based on comparison with reference images, but the system lacks knowledge of actual device design leading to potential misclassification
Solution Approach 1:
The patent merges defect image information with design information from the semiconductor layout to create a comprehensive classification system. The classification engine combines features from both the defect image (morphology, texture, color) and design information (layer structure, pattern recognition) to accurately identify defect types and locations, resolving the contradiction by integrating multiple information sources rather than relying on reference images alone.
Solution Approach 2:
The patent introduces design information as an intermediary element that bridges the gap between defect detection and accurate classification. This intermediary provides contextual knowledge about the intended device structure, allowing the system to distinguish between actual defects and design variations, thereby improving classification accuracy without requiring direct comparison to reference defect images.
2Measurement precision
If extensive scanning processes are used to classify defects, then more comprehensive defect information can be obtained, but the process becomes time-consuming and inefficient
Solution Approach 1:
The patent performs preliminary action by pre-processing the defect images to extract key features (morphology, texture, color) and pre-aligning them with the design information before classification. This preliminary feature extraction and registration reduces the computational burden during actual classification, allowing accurate results to be obtained faster without requiring extensive scanning processes.
Solution Approach 2:
The patent replaces the mechanical scanning process with a computational approach. Instead of physically scanning the entire wafer or using multiple reference images for comparison, the system uses image processing algorithms to extract features and compares them directly with design information through computational matching, significantly reducing processing time while maintaining accuracy.
3Loss of time
If a single defect image from a single perspective is used, then processing time is reduced, but classification accuracy may be compromised
Solution Approach 1:
The patent adds another dimension to the classification process by incorporating design information as a fourth dimension of analysis. Instead of relying solely on the two-dimensional defect image, the system integrates the defect characteristics with the three-dimensional structural context from design information, enabling accurate classification from a single image perspective while compensating for the limited viewpoint through contextual knowledge.
Data Source
AI summary
Defects observed by imaging tools may be classified by automatic comparison of features observed in a defect image with design information relating to corresponding portions of the image. Defect information may be generated from a defect image from a defect imaging tool. Design information relating to one or more structures to be formed on the substrate in a vicinity of the defect may be retrieved. The defect may be classified based on a combination of the defect information from the defect image and design information relating to one or more structures to be formed on the substrate in the vicinity of the defect.