Wafer Anomaly Classification With Active Learning for Cold Start
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in high-precision defect detection and classification in wafer imaging datasets, particularly due to high nuisance rates and the need for extensive prior knowledge and annotation effort, making cold-starting scenarios infeasible.
Innovation Solution
A computer-implemented method integrating anomaly detection, classification, and active learning within a single workflow to minimize prior knowledge and annotation effort, allowing for the selection of informative anomalies based on similarity measures and unsupervised clustering, enabling efficient defect classification with reduced user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If anomaly detection algorithms based on die-to-die or die-to-database principles are used, then defects can be detected by comparing wafer portions, but the methods are sensitive to underlying SEM simulation and hard to generalize to new sample types
Solution Approach 1:
The system performs self-training by automatically selecting informative anomalies and generating training data without requiring external annotation. The anomaly detection algorithm iteratively improves itself by identifying anomalies, determining their informativeness based on similarity measures, and using selected anomalies to refine the detection parameters, making the system adaptable to new sample types without re-annotation
Solution Approach 2:
The system dynamically adjusts detection parameters and similarity thresholds based on the selected informative anomalies. By changing parameters iteratively through the selection process and re-training on selected anomalies, the system adapts to different wafer types and manufacturing conditions while maintaining reliable defect detection
2Productivity
If multibeam scanning electron microscopy (mSEM) is used to image large regions of wafer surface with high resolution, then inspection speed increases, but the resulting datasets become huge and cannot be analyzed manually
Solution Approach 1:
The system extracts only the most informative anomalies from the huge mSEM datasets by calculating similarity measures and selecting anomalies that provide maximum information for training. Instead of analyzing all data points, the system identifies and extracts a small subset of representative anomalies that capture the essential defect patterns
Solution Approach 2:
The system introduces an automated selection mechanism as an intermediary between data acquisition and analysis. This mediator automatically identifies informative anomalies based on similarity criteria and prepares training data, bridging the gap between huge raw datasets and the need for efficient analysis without manual intervention
3Reliability
If conventional quality control using low resolution optical tools followed by high-resolution SEM review is used, then defects can be detected, but the process is resource intensive and has long turnaround times
Solution Approach 1:
The system merges anomaly detection and classification into a single integrated workflow that operates directly on high-resolution imaging data. By combining detection and classification steps and using automated selection of informative anomalies, the system eliminates the sequential two-stage process while reducing turnaround time through efficient automated processing
4Measurement precision
If extensive annotation effort and prior knowledge are required for defect classification, then classification accuracy can be improved, but cold-starting scenarios become infeasible
Solution Approach 1:
The system performs preliminary automated selection of informative anomalies that will serve as training data before classification begins. By pre-identifying representative anomalies based on similarity measures and automatically selecting training samples, the system prepares high-quality training data without requiring prior manual annotation or expert knowledge
Solution Approach 2:
The system is self-sufficient in generating its own training data by automatically identifying informative anomalies and selecting representative samples for training. The anomaly selection mechanism autonomously determines which anomalies provide maximum information, enabling the system to start from scratch without external annotation efforts
Data Source
AI summary
A computer implemented method detects and classifies anomalies in an imaging dataset of a wafer comprising a plurality of semiconductor structures. The method comprises determining a current detection of a plurality of anomalies in the imaging dataset, and obtaining an unsupervised or semi-supervised clustering of the current detection of the plurality of anomalies. Based on at least one decision criterion at least one cluster of the clustering is selected for presentation and annotation to a user via a user interface. An anomaly classification algorithm is re-trained based on the annotated anomalies. A system for controlling the quality of wafers and a system for controlling the production of wafers are also disclosed.