Substrate Image Classification Refinement Using Triplet Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for substrate image classification in manufacturing processes often result in erroneous metrology results due to the incorrect selection of algorithms, leading to discarded good substrates, incorrect manufacturing parameters, decreased yield, and increased processor overhead, energy consumption, and bandwidth.
Innovation Solution
A method involving training a machine learning model using substrate images sorted into classes and refining it with a triplet loss function based on misclassified images to improve accuracy, followed by using the refined model for substrate processing actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used for substrate image classification, then the classification process is simple, but the accuracy is low leading to erroneous metrology results
Solution Approach 1:
The classification process is segmented into two distinct phases: initial training phase and refinement phase. The refinement phase specifically targets misclassified images from the training phase, applying triplet loss function only to these problematic cases rather than retraining on all data. This segmentation allows the model to improve accuracy where needed without unnecessary complexity elsewhere.
Solution Approach 2:
Instead of applying refinement to all training data, the method applies triplet loss refinement selectively only to misclassified images. This partial action focuses computational resources on the specific subset of data that benefits most from refinement, improving classification accuracy while avoiding the excessive complexity of full-data retraining.
2Productivity
If conventional classification methods are used, then the processing speed is fast, but the yield decreases due to discarded good substrates
Solution Approach 1:
The system implements a feedback mechanism where misclassified images are identified and fed back into the refinement process. The triplet loss function uses these misclassified examples to adjust model parameters, creating a closed-loop system that continuously improves classification accuracy. This feedback-driven approach ensures that good substrates are not erroneously discarded while maintaining high processing throughput.
3Loss of energy
If conventional methods are used, then the energy consumption is low, but the processor overhead increases due to reprocessing
Solution Approach 1:
The system performs preliminary classification using the trained model to identify misclassified images before applying the computationally intensive triplet loss refinement. This preliminary action filters the data so that expensive refinement computations are applied only to necessary cases, reducing overall processor overhead and energy consumption while maintaining high processing efficiency for the majority of correctly classified images.
Data Source
AI summary
A method includes identifying substrate images that have been sorted into classes. The method further includes training a machine learning model using data input including the substrate images and target output including the classes. The method further includes refining the trained machine learning model using a triplet loss function based on one or more substrate images misclassified by the trained machine learning model to provide a refined trained machine learning model associated with performance of an action associated with substrate processing.