Semiconductor Training Set Augmentation via Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor fabrication processes face challenges in accurately detecting and classifying defects in high-resolution images due to variations in physical processes such as noise, tool calibration, and environmental factors, which limits the effectiveness of automated examination and defect review.
Innovation Solution
A method is developed to generate a training set for Deep Neural Networks by simulating the effects of physical processes on fabrication process images, using augmented images that account for different parameter values, thereby enhancing the robustness of automated defect classification and review systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated examination systems use standard training sets, then examination speed is maintained, but detection accuracy deteriorates due to variations in physical processes
Solution Approach 1:
The system performs preliminary actions by generating augmented training images that simulate various physical process variations (noise, tool calibration differences, environmental factors) before the actual examination process. This preprocessing step creates a more robust training set that prepares the examination system to handle real-world variations, thereby improving detection accuracy without increasing operational complexity
Solution Approach 2:
The system applies parameter changes by systematically varying physical process parameters (noise levels, calibration values, environmental conditions) in the generation of augmented training images. This allows the training set to encompass a broader range of possible examination conditions, improving the system's ability to accurately detect defects across different physical scenarios
2Reliability
If examination systems are trained with diverse physical process variations, then robustness improves, but training data requirements increase
Solution Approach 1:
The system uses copying by generating synthetic augmented images that replicate real examination scenarios with varied physical process parameters. Instead of collecting large volumes of actual diverse examination data, the system creates copies of existing training images with simulated variations, achieving robustness while maintaining manageable training data volumes
Solution Approach 2:
The system extracts the essential physical process variations (noise patterns, calibration offsets, environmental effects) from real examination data and isolates them as separate simulatable parameters. This allows the generation of augmented training data by applying these extracted variation patterns, reducing the need for large volumes of diverse actual training data
3Measurement precision
If automated defect classification uses simulation-augmented training, then classification accuracy improves, but processing time increases
Solution Approach 1:
The system performs the computationally intensive image augmentation and simulation processes as preliminary actions during the training phase, before actual examination operations. This allows classification accuracy to be improved through comprehensive training on varied data, while the actual examination processing time remains unaffected since the augmented training set is prepared in advance
Data Source
AI summary
There is provided a system and method of generating a training set usable for examination of a semiconductor specimen. The method comprises: obtaining a simulation model capable of simulating effect of a physical process on fabrication process (FP) images depending on the values of parameters of the physical process; applying the simulation model to an image to be augmented for the training set and thereby generating one or more augmented images corresponding to one or more different values of the parameters of the physical process; and including the generated one or more augmented images into the training set. The training set can be usable for examination of the specimen using a trained Deep Neural Network, automated defect review, automated defect classification, automated navigation during the examination, automated segmentation of FP images, automated metrology based on FP images and other examination processes that include machine learning.