Related Experiment Video
Updated: Jan 2, 2026
11:34
Subsurface Defect Localization by Structured Heating Using Laser Projected Photothermal Thermography
Published on: May 15, 2017
11.5K
Data-driven approaches to optical patterned defect detection
Mark-Alexander Henn1, Hui Zhou1, Bryan M Barnes1
1Nanoscale Device Characterization Division, Physical Measurement Laboratory, National Institute of Standards and Technology, 100 Bureau Drive, Gaithersburg, MD 20899, USA.
OSA Continuum
|December 12, 2019
Summary
Machine learning, specifically convolutional neural networks (CNNs), enhances defect detection in semiconductor manufacturing. These advanced computer vision methods improve the identification of minuscule defects, even those smaller than the inspection wavelength.
Area of Science:
- Semiconductor Manufacturing
- Computer Vision
- Machine Learning
Background:
- Growing computational power fuels advancements in computer vision and classification methods.
- Traditional defect detection methods in semiconductor manufacturing struggle with increasingly smaller defects.
- Line edge roughness (LER) further complicates defect inspection due to scattering effects.
Purpose of the Study:
- To investigate the application of data-driven methods for improving defect detection in semiconductor manufacturing.
- To address the limitations of intensity threshold-based approaches for small defect identification.
- To enhance the detectability and classification of defects, including those affected by LER.
Main Methods:
- Utilized machine learning techniques, including convolutional neural networks (CNNs).
- Employed simulated scattering of realistic geometries with and without defects to generate image data.
- Incorporated line edge roughness (LER) into simulations to mimic real-world fabrication challenges.
Main Results:
- CNNs demonstrated improved detectability and classification of defects in simulated semiconductor images.
- The proposed method successfully identified defects significantly smaller than the inspection wavelength (over 20x smaller).
- Effectively addressed challenges posed by LER in defect inspection.
Conclusions:
- Machine learning, particularly CNNs, offers a powerful solution for advanced defect detection in semiconductor fabrication.
- This data-driven approach overcomes limitations of traditional methods for sub-wavelength defect identification.
- The methodology shows promise for improving quality control and yield in semiconductor manufacturing.
