Wafer Defect Classification via SEM and Layout Image Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated defect classification techniques for integrated circuits require significant manual calibration and are limited in detecting various defects, making them time-consuming and inefficient for identifying manufacturing flaws in increasingly complex and miniaturized circuit designs.
Innovation Solution
A machine learning-based defect classification system that uses a scanning electron microscope to capture images of manufactured devices and generates predicted images from layout designs, allowing for the identification and classification of defects by comparing captured and predicted images, thereby reducing manual effort and improving defect detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated defect classification techniques are used, then defect detection speed is improved, but manual calibration effort and time increase significantly
Solution Approach 1:
The system performs preliminary actions by pre-processing SEM images to enhance defect visibility and pre-aligning layout designs before comparison. This preliminary preparation reduces the need for extensive manual calibration during the actual defect classification process, allowing automated techniques to operate more efficiently with minimal manual intervention.
Solution Approach 2:
The system introduces an intermediary comparison mechanism that automatically matches SEM image features with corresponding layout design elements. This intermediary process serves as a bridge between raw imaging data and defect classification, eliminating the need for manual calibration by automatically establishing the relationship between physical device features and their design specifications.
2Adaptability or versatility
If traditional automated defect classification is used, then some defects can be detected, but the system is limited in detecting various defect types
Solution Approach 1:
The system achieves universality by implementing a multi-functional defect classification framework that can detect and classify multiple defect types including bridging, opening, misalignment, and dimensional variations. The comparison mechanism between SEM images and layout designs is designed to be versatile, handling different defect categories through a unified approach that maintains high accuracy across all defect types.
Solution Approach 2:
The system applies dynamics by adapting its analysis parameters and comparison thresholds based on the specific defect type being detected. The classification process dynamically adjusts its sensitivity and methodology according to the nature of the defect, allowing the system to maintain high reliability across diverse defect types rather than using a static, one-size-fits-all approach.
3Measurement precision
If contour extraction and filtering processes are used, then defect identification can be performed, but the process becomes extensive and time-consuming
Solution Approach 1:
The system extracts only the essential features needed for defect identification by directly comparing key geometric elements between SEM images and layout designs. Instead of performing extensive contour extraction and filtering on entire images, the system selectively extracts and compares only the relevant structural features, maintaining high measurement precision while significantly reducing processing time and increasing throughput.
Data Source
AI summary
This application discloses a scanning electron microscope system to capture an image of an electronic device manufactured according to a layout design describing the electronic device, and a computing system to generate a predicted image of the electronic device using the layout design. The predicted image corresponds to an expected image of the electronic design system captured by the scanning electron microscope system. The computing system identifies manufacturing defects present in the electronic device based on differences between the predicted image of the electronic device and the captured image of the electronic device, and utilizes the captured image of the electronic device to classify the manufacturing defects identified based on the predicted image of the electronic device from the layout design. The computing system can generate a manufacturing defect report identifying the manufacturing defects used to perform repair of the electronic device or modification of the layout design.