Wafer Defect Detection Training Using Image-Adjusted ML Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems for semiconductor wafer defect detection face challenges in accuracy due to high heterogeneity in input data such as defect size, noise level, and resolution, often missing false negatives and requiring labor-intensive manual corrections.
Innovation Solution
A method and system for training a second ML system by generating datasets with image adjustments, identifying defects, assigning ground truths, and training on these datasets to improve defect detection, using predefined image adjustment parameters like noise, resolution, and blurriness adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a first ML system is used for defect detection, then detection speed is maintained, but detection accuracy deteriorates due to false negatives in heterogeneous input data
Solution Approach 1:
The system performs preliminary actions by generating multiple adjusted versions of input images (noise-adjusted, resolution-adjusted, blur-adjusted) before defect detection. This preprocessing step ensures that defects are more visible and consistent across variations, enabling the ML system to detect defects more accurately and reduce false negatives.
Solution Approach 2:
The system changes image parameters (noise levels, resolution, blur) to create multiple versions of the same image. By detecting defects across these parameter variations, the system achieves more reliable and accurate defect detection, as defects remain consistent while background noise varies.
2Measurement precision
If manual corrections are applied to improve detection accuracy, then measurement precision improves, but ease of operation deteriorates due to labor-intensive processes
Solution Approach 1:
The system performs self-service by automatically generating multiple image adjustments and performing defect detection on all variations without human intervention. The ML system independently processes the dataset, compares results across variations, and identifies defects, eliminating the need for manual corrections while maintaining high accuracy.
Solution Approach 2:
The system uses feedback by comparing defect detection results across multiple image adjustments. Defects that consistently appear across variations are identified with high confidence, while false positives are filtered out. This automated feedback mechanism replaces manual verification and improves both accuracy and operational ease.
3Manufacturing precision
If multiple image adjustments are generated for training, then manufacturing precision improves, but device complexity increases due to additional processing steps
Solution Approach 1:
The system achieves universality by using a single ML model that processes multiple types of image adjustments (noise, resolution, blur). Rather than requiring separate processing pipelines for each adjustment type, the model is trained to handle all variations uniformly, simplifying the overall system architecture while maintaining high detection consistency.
Solution Approach 2:
The system merges multiple defect detection results from different image adjustments into a single consolidated output. By combining the detection results and identifying consistent defects across variations, the system achieves high manufacturing precision without requiring complex separate processing systems for each adjustment type.
Data Source
AI summary
A method for training a second Machine Learning (ML) system to determine semiconductor wafer defects includes: receiving a first set of semiconductor wafer images for which defect detection failed in a first ML system; generating a first dataset based on the received first set of images and corresponding prediction results of the first ML system; modifying the images in the first dataset using predefined image adjustment parameters to generate a second set of images; identifying, using the first ML system, defects in the second set of images; assigning ground truths to the second set of images based on the identified defects; generating a second dataset based on the first dataset, the second set of images, and the ground truths associated therewith; and training, based on the generated second dataset, the second ML system to determine semiconductor wafer defects.