Multi-Angle Scattered Light Defect Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect inspection systems for semiconductor wafers lack a method for accurately training classifiers to differentiate between defects and noise, limiting their ability to classify defects with high precision.
Innovation Solution
A defect inspection method and system that utilizes multiple detectors arranged at various elevation and azimuth angles to collect scattered light data, which is then input to a learning device trained with defect and noise information to classify defects accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple detectors are arranged at multiple elevation angles and azimuths to detect scattered light, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The detection system is segmented into multiple independent detectors positioned at different elevation angles and azimuths around the sample. Each detector captures scattered light from a specific angular direction, and the classification process is segmented into training and application phases. This segmentation allows comprehensive angular coverage while maintaining manageable system complexity through modular detector units.
Solution Approach 2:
The system transitions from single-angle detection to multi-dimensional angular detection by arranging detectors in three-dimensional space around the sample. Detectors are positioned at multiple elevation angles (vertical angle from sample surface) and multiple azimuths (horizontal angle around irradiation point), creating a comprehensive angular sampling network that captures scattered light from all directions.
2Measurement precision
If a learning device is trained using output information from multiple detectors and defect information, then defect classification accuracy is improved, but loss of information increases during training
Solution Approach 1:
The learning device undergoes preliminary training using defect information and output information from multiple detectors before actual defect inspection. This preliminary training phase prepares the classification model in advance, enabling it to accurately classify defects during routine inspection without requiring real-time complex processing that could lead to information loss.
Solution Approach 2:
The system implements feedback mechanisms where defect information (including ground truth labels) is fed back to the learning device during training. The learning device uses this feedback to adjust its classification parameters and improve accuracy. The output information from multiple detectors serves as input feedback, creating a closed-loop training process that continuously refines classification performance.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate classification of defects and noise by leveraging an appropriately trained learning device, enhancing the precision and efficiency of defect inspection processes.
Implementation Method 1
a detector for detecting a scattered light generated by irradiation of the sample with a light
Data Source
AI summary
The present disclosure proposes a method for classifying defects and the like by using a learning device that has been suitably trained, a system, and a computer-readable medium. As one aspect thereof, the present disclosure proposes (see FIG. 1) a defect inspection method, etc., in which one or more computers are used to inspect a defect on a sample on the basis of output information from detectors that detect scattered light produced via the irradiation of the sample with light, wherein defect information is outputted by: receiving output from a plurality of detectors disposed at a plurality of angles of elevation with reference to the sample surface, and at a plurality of sample surface-direction orientations with reference to the irradiation points of the light on the sample; and inputting the output information of the plurality of detectors into a learning device that has been trained using the output information from the plurality of detectors and the defect information.