Machine-learning models for analyzing TSOM images of nanostructures

Area of Science:

• Nanotechnology
• Optical Microscopy
• Machine Learning

Background:

• Through-focus scanning optical microscopy (TSOM) is a cost-effective, non-destructive technique for 3D nanostructure measurement.
• Current TSOM analysis relies on library-matching, which can limit dimensional measurement accuracy.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) method for improving dimensional measurement accuracy in TSOM.
• To extract texture information from TSOM images using feature vectors for enhanced analysis.

Main Methods:

• Feature extraction using Gray-level Co-occurrence Matrix (GLCM), Local Binary Pattern (LBP), and Histogram of Oriented Gradient (HOG).
• Training and testing of three ML regression models: Random Forest, Gradient Boosting Decision Tree (GBDT), and AdaBoost.
• Evaluation of feature vectors used in isolation, in pairs, and in combination.

Main Results:

• The proposed ML method demonstrates considerably higher measurement accuracy compared to the library-matching method.
• The AdaBoost model with combined LBP and HOG features excels in measuring features across a wide size range.
• For narrower size ranges, the AdaBoost model with HOG features shows superior performance.

Conclusions:

• Machine learning, particularly using texture features like LBP and HOG, significantly enhances dimensional measurement accuracy in TSOM.
• The choice of feature extraction method and ML model can be optimized based on the size range of the nanostructure features being measured.