Geometric analysis algorithm based on a neural network with localized simulation data for nano-grating structure...

Area of Science:

• Materials Science
• Optical Physics
• Computational Science

Background:

• Mueller matrix spectroscopic ellipsometry (MMSE) is a key nondestructive technique for analyzing nanostructures.
• Advancements in computational power and neural networks are improving the analysis of complex nanostructures.
• Accurate characterization of nanostructure dimensions is crucial for device performance.

Purpose of the Study:

• To develop a fast and accurate deep learning-based method for nanostructure parameter prediction using MMSE data.
• To predict the dimensions (width and height) of 1D grating structures.
• To validate the deep learning approach through experimental measurements.

Main Methods:

• Utilized a two-step deep learning algorithm combining neural networks with limited simulation data.
• Trained the neural network to predict nanostructure parameters from Mueller matrices.
• Performed experimental validation using a silicon dioxide (SiO2) grating.

Main Results:

• Achieved high accuracy in predicting grating width and height, with a Mean Absolute Error (MAE) below 0.1 nm.
• The deep learning model successfully predicted dimensions for a SiO2 grating.
• Experimental results showed good agreement with scanning electron microscopy (SEM) measurements.

Conclusions:

• The proposed deep learning method offers a fast and accurate approach for nanostructure analysis via MMSE.
• This technique effectively determines critical dimensions of 1D grating structures.
• The study demonstrates the potential of AI-driven MMSE for advanced nanometrology.