ML Defect Classification for Semiconductor Optical Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical inspection systems struggle to efficiently classify defects in semiconductor manufacturing due to the difficulty in distinguishing between types of defects, especially when they are small or not clearly defined by linear dimensions.
Innovation Solution
The use of multiple machine learning models trained with optical inspection data and ground truth data from high-resolution inspections allows for the efficient classification of defects, with selected models deployed for inference to classify defects in semiconductor manufacturing samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If optical inspection is used for defect detection, then inspection speed and productivity are improved, but defect classification accuracy deteriorates
Solution Approach 1:
The patent introduces machine learning models as an intermediary between optical inspection data and defect classification. The ML models process optical inspection features (reflectivity, scattering, absorption characteristics) and translate them into defect type classifications, enabling accurate classification without requiring high-resolution imaging for all defects
Solution Approach 2:
The patent replaces manual visual inspection and high-resolution imaging with automated machine learning classification. Instead of relying on human analysts or complex imaging systems to classify defects, the system uses trained ML algorithms that process optical inspection data to automatically identify defect types, significantly improving throughput while maintaining accuracy
2Measurement precision
If high-resolution imaging is used for defect classification, then classification accuracy is improved, but inspection time and productivity deteriorate
Solution Approach 1:
The patent segments the inspection process into two stages: (1) optical inspection for rapid defect detection and characterization, and (2) machine learning classification for defect type identification. This segmentation allows the system to use fast optical inspection for all defects and reserve high-resolution imaging only for ambiguous cases, reducing overall inspection time while maintaining accuracy
Solution Approach 2:
The patent applies partial action by using high-resolution imaging only for a subset of defects that require it, rather than applying it to all defects. The machine learning models first triage defects and identify those that need detailed imaging, allowing the system to use high-resolution imaging partially and only when necessary, thereby reducing total inspection time
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
This approach significantly reduces the time required for defect classification, allowing for the majority of defects to be classified using optical inspection alone, while reserving high-resolution imaging for only a small portion of defects, thereby enhancing manufacturing efficiency.
Implementation Method 1
Optical inspections can include reflectometry techniques, spectrometry techniques, ellipsometry techniques, etc. Optical inspections can be performed using specularly reflected light, diffusely reflected (scattered) light, transmitted light, or various combinations thereof.
Implementation Method 2
Optical inspections can be performed using specularly reflected light, diffusely reflected (scattered) light, transmitted light, or various combinations thereof.
Data Source
AI summary
Implementations disclosed describe, among other things, a systems and techniques for perform efficient inspection of a semiconductor manufacturing sample. The techniques include collecting optical inspection data for training sample(s) that have a plurality of defects. The techniques further include generating, using the optical inspection data, a training data set that includes descriptions, images, and ground truth classifications for the defects. The techniques further include using the training data set to train a plurality of machine learning (ML) classifiers to generate predicted classifications for the defects in the training sample(s). The techniques further include selecting, using the predicted classifications and the ground truth classifications, one or more ML classifiers that meet one or more accuracy criteria, and using the selected ML classifier(s) to classify defects in the semiconductor manufacturing sample.