Machine learning for layer number identification of black phosphorus based on Raman spectra.

Area of Science:

• Materials Science
• Nanotechnology
• Artificial Intelligence

Background:

• Black phosphorus (BP) is a 2D material with tunable properties, but determining its thickness is challenging.
• Conventional methods for BP thickness determination are inefficient and complex.
• Accurate thickness measurement is vital for understanding BP's electronic, optical, and thermal characteristics.

Purpose of the Study:

• To develop an efficient and accurate machine learning (ML) method for determining the layer number of black phosphorus (BP).
• To compare multiple ML algorithms for BP layer number identification.
• To identify key Raman spectral features for accurate BP thickness prediction.

Main Methods:

• Feature extraction from Raman spectra of BP, including peak position, intensity, and width.
• Analysis of feature importance to identify crucial predictors, such as substrate peak to Raman mode intensity ratio.
• Application and comparative analysis of various ML algorithms for layer number prediction.
• Dataset augmentation and model architecture refinement to overcome data limitations.

Main Results:

• The ML model achieved high accuracy, with R-squared values not less than 0.9 across all tested algorithms.
• The intensity ratio of the substrate (Si) peak to the Raman mode was identified as a critical feature.
• The study successfully identified discriminative Raman spectral features for BP layer number prediction.
• The developed ML approach demonstrated superior efficiency and accuracy compared to conventional methods.

Conclusions:

• Machine learning models can accurately and efficiently predict the layer number of black phosphorus.
• The proposed method reduces the analytical burden on researchers and promotes AI applications in 2D material characterization.
• This work provides a robust framework for automated characterization of 2D materials using ML and Raman spectroscopy.