Optical inspection of nanoscale structures using a novel machine learning based synthetic image generation algorithm.

Area of Science:

• Semiconductor manufacturing
• Optical microscopy
• Machine learning

Background:

• Optical image quality is degraded by diffraction and noise, complicating nanoscale defect detection.
• Acquiring sufficient high-quality training data for rare defects is often prohibitively expensive.
• Existing machine learning methods require large datasets and face challenges with simulations.

Purpose of the Study:

• To present a novel interpretable machine learning technique for classifying nanoscale defects in semiconductor wafer images.
• To address challenges in defect detection, including noise, limited training data, and computational complexity.
• To enable accurate detection, classification, and size estimation of nanoscale defects.

Main Methods:

• Developed an interpretable machine learning technique incorporating physical insights specific to noisy optical images.
• Utilized a small number of training samples for defect classification.
• Applied the method to analyze semiconductor wafers using visible light microscopy.

Main Results:

• Successfully detected both parallel and previously undetectable perpendicular bridge defects in a 9-nm node wafer.
• Accurately classified the shapes of nanoscale defects.
• Provided reliable estimations of defect sizes from optical images.

Conclusions:

• The proposed machine learning technique offers an effective solution for detecting and classifying nanoscale defects in semiconductor manufacturing.
• This approach overcomes the limitations of traditional methods requiring extensive training data and complex simulations.
• The technique enhances defect analysis capabilities, particularly for rare and critical defects.