Interactive Anomaly Classification for Semiconductor Wafer Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for detecting and classifying defects in semiconductor wafer imaging datasets are labor-intensive and prone to incomplete or incorrect annotations, leading to inefficiencies and high false positive rates, especially with large datasets and varying defect classes.
Innovation Solution
An iterative refinement method using a machine-learned classification algorithm that re-trains based on user annotations, allowing for interactive classification and flexible class adjustments, including unsupervised clustering and autoencoder neural networks for anomaly detection, to minimize human effort and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation of sample tiles is used to train classification algorithms, then the algorithm can be trained on imaging data, but the process becomes very laborious and time-consuming on large imaging datasets
Solution Approach 1:
The system performs preliminary automated anomaly detection and classification before manual annotation, pre-processing the large imaging dataset to identify only the most uncertain or ambiguous cases that require human review. This reduces the annotation workload from the entire dataset to only a small subset of challenging cases.
Solution Approach 2:
The system implements an iterative feedback loop where manual annotations are used to retrain and refine the classification algorithm, which then improves its performance on subsequent batches of data. This progressive refinement reduces the need for extensive initial manual annotation while continuously improving accuracy.
2Quantity of substance
If manual annotation is performed on large imaging datasets, then more training data is available, but the risk of incomplete representation, missed defects, misclassification, and false positives increases
Solution Approach 1:
The system replaces manual mechanical annotation with an automated machine-learning-based classification system that uses unsupervised learning and anomaly detection algorithms. This automated approach processes the entire dataset consistently without human fatigue, reducing misclassification and false positives while maintaining high reliability.
Solution Approach 2:
The classification algorithm performs self-training and self-refinement by automatically identifying patterns and anomalies in the data, then using these insights to improve its own performance. The system serves itself by continuously learning from the data without requiring extensive manual curation, thereby maintaining high reliability across large datasets.
3Measurement precision
If conventional inspection methods are used on multi-resolution imaging datasets with multiple magnification scales, then comprehensive defect detection is possible, but the detection process becomes significantly slower
Solution Approach 1:
The system segments the multi-resolution imaging dataset by processing different magnification scales separately and independently. The classification algorithm analyzes each resolution level autonomously, identifying defects at appropriate scales without requiring sequential human review of all magnifications, thereby maintaining detection completeness while improving processing speed.
Solution Approach 2:
The system performs partial processing by focusing computational resources on detecting and classifying only the most significant anomalies at each resolution level, rather than exhaustively analyzing every pixel at every magnification. This selective approach maintains high detection completeness for critical defects while significantly reducing overall processing time.
Data Source
AI summary
A method includes detecting a plurality of anomalies in an imaging dataset of a wafer. The wafer includes a plurality of semiconductor structures. The method also includes executing multiple iterations. At least some of the iterations include determining a current classification of the plurality of anomalies using a machine-learned classification algorithm and tiles of the imaging dataset associated with the plurality of anomalies. The current classification includes a current set of classes into which the anomalies of the plurality of anomalies are binned. The method further includes, based on at least one decision criterion, selecting at least one anomaly of the plurality of anomalies for a presentation to a user. In addition, the method includes, based on an annotation of the at least one anomaly provided by the user with respect to the current classification, re-training the classification algorithm.