Ensemble Learning for Automated Semiconductor Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor defect detection and classification methods using scanning electron microscopy (SEM) face challenges in accurately localizing and classifying defects due to noisy microscopy images, especially in high numerical aperture applications, and rely heavily on expert intervention and rule-based systems, which are inefficient and unreliable for multi-defect instances.
Innovation Solution
A computer-implemented training method using an ensemble of learning structures comprising feature extractor, region proposal, detection, and segmentation modules, trained with SEM image datasets to predict defect classes and instance segmentation masks, reducing the need for manual labeling and expert review.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scanning electron microscopy (SEM) is used for defect detection, then resolution and localization capability are improved, but image noise increases making defect detection more difficult
Solution Approach 1:
The patent replaces traditional rule-based mechanical inspection systems with a deep learning-based automated detection system. The neural network model processes noisy SEM images and automatically identifies defects, eliminating the need for manual threshold setting and expert intervention while handling the noise inherent in SEM imaging.
Solution Approach 2:
The patent introduces an intermediary processing layer using deep learning models between the noisy SEM images and the defect detection output. The neural network acts as a mediator that learns to distinguish defect patterns from noise through training on labeled data, enabling reliable detection despite image quality degradation.
2Ease of operation
If rule-based defect detection methods are used, then ease of operation is improved, but reliability and detection accuracy deteriorate
Solution Approach 1:
The patent implements a self-service automated detection system where the deep learning model independently performs defect detection, classification, and localization without requiring expert operator intervention. The system automatically adjusts detection parameters and processes images autonomously, maintaining high reliability while preserving ease of operation through automated decision-making.
Solution Approach 2:
The patent incorporates feedback mechanisms where the neural network is trained on labeled defect data and continuously refines its detection capabilities. The system learns from previously identified defects and adjusts its detection thresholds and criteria, improving reliability over time while maintaining user-friendly operation through automated iterative optimization.
3Measurement precision
If manual defect classification and localization is performed, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent performs preliminary actions by pre-training the deep learning model on extensive labeled defect datasets before actual inspection. The model learns optimal defect patterns, classes, and localization criteria in advance, enabling rapid and accurate defect detection during production without requiring time-consuming manual analysis for each defect identification.
Solution Approach 2:
The patent ensures continuity of useful action through automated parallel processing where multiple images are analyzed simultaneously by the neural network. The system continuously processes defect detection, classification, and localization without interruption or manual intervention, maintaining constant productivity and eliminating the time losses associated with sequential manual review.
4Measurement precision
If expert intervention is required for defect detection, then measurement precision is improved, but productivity deteriorates
Solution Approach 1:
The patent substitutes the mechanical process of expert human review with an automated neural network system. The deep learning model performs defect detection, classification, and localization with precision comparable to expert operators while dramatically improving productivity by processing images automatically at high speed without human intervention requirements.
Solution Approach 2:
The patent creates a universal automated system that performs multiple functions including defect detection, classification into different defect types, localization with precise coordinates, and confidence scoring. This multi-functional neural network replaces the need for multiple specialized expert operations, improving both precision and productivity by consolidating tasks into a single automated workflow.
Data Source
AI summary
The present disclosure related to a computer-implemented training and prediction method for defect detection, classification and segmentation in image data. The training method comprises providing an ensemble of learning structures, each learning structure comprising a feature extractor module, a region proposal module, a detection module, and a segmentation module. Each learning structure is trained individually and validated. Learning structures whose validation prediction score exceeds a predetermined threshold score are selected and their predictions combined, using a parametrized ensemble voting structure.