Annotation labeling technique for machine learning-enhanced optical characterization of hexagonal boron nitride thickness on 300-nm oxide substrate

We propose an annotation technique that may enhance characterization of hexagonal boron nitride flake thickness on 300-nm silicon oxide substrate via machine learning-enhanced machine vision of optical microscopy data. Manually interpreting optical microscopy data is labor intensive, time-consuming, and faulted by human subjectivity due to periodic behavior in light reflectance as thickness increases, resulting in colors that are visually difficult to disambiguate. To address this issue, we propose a labeling technique incorporating thickness-related shadowing around flake borders to differentiate thicknesses when colors are similar. By focusing on annotation labeling, we aim to solve the limitations of conventional human characterization and a tool for machine vision model training. Our technique promises a precise means of flake thickness identification by optical microscopy alone, bypassing labor-intensive scanning probe characterization. Our approach promises accelerated characterization of hBN thickness when implemented with machine vision, with potential applications for enhancing efficiency in 2D materials research and development.

Depiction of annotation labeling technique incorporating thickness-related shadowing around flake borders to differentiate thicknesses when colors are indistinguishable for two flakes (110-nm (left) and 225-nm (right)) in a trilayer stack of hBN/SiO₂/Si on 300-nm oxide.