April 7 - 11, 2025
Seattle, Washington
Symposium Supporters
2025 MRS Spring Meeting & Exhibit
QT01.03.03

Rapid Automated 2D Material Characterization with Optical Contrast Heuristics

When and Where

Apr 10, 2025
5:00pm - 7:00pm
Summit, Level 2, Flex Hall C

Presenter(s)

Co-Author(s)

Glenn Foster1,Matthew Carbone2,Suji Park2,Houk Jang2,Patrick Vora1

George Mason University1,Brookhaven National Laboratory2

Abstract

Glenn Foster1,Matthew Carbone2,Suji Park2,Houk Jang2,Patrick Vora1

George Mason University1,Brookhaven National Laboratory2
Two-dimensional (2D) flakes are of exceptional interest to the materials research community for their unique optical, thermal, and electronic characteristics. These properties are strongly correlated with their thickness, i.e., the number of stacked layers the flakes are composed of. Mechanical tape exfoliation is a common method for producing 2D flakes, but the process results in flakes with a range of thicknesses. Manually determining their thicknesses is a tedious process that is heavily dependent on human recognition,1 leading to the community developing a suite of automated flake characterization techniques. These systems rely on machine learning methods that are often overly sensitive to imaging conditions or require large volumes of training data1, 2 which may not exist for novel materials. Here, we detail an automated flake characterization system that makes rapid, accurate identifications without machine learning techniques. Our method works by using optical contrast as a heuristic. A flake’s optical contrast against a background substrate varies as a function of the flake’s thickness.3 Flakes in microscope images are found by comparing each pixel’s optical contrast with the expected contrast for a given flake and substrate combination. We demonstrate the invariance of contrast with respect to camera parameters such as light intensity, exposure time, and gain for monolayer graphene on SiO2. Because optical contrast does not vary under such conditions, discrete ranges for the optical contrast of desired flake thicknesses can be set with just a few pre-identified images. This heuristic approach, therefore, eliminates the need for a large training data set, while still returning accurate thickness and size identifications of 2D materials. Our system had a precision of 88.7% when tested on a 1,517 image tilescan of exfoliated graphene flakes, with an average per-image processing time of one second. Additionally, our system’s reliance on fundamental optical laws makes it generalizable to any 2D material. We anticipate that our identification system will greatly improve the efficiency of manufacturing 2D heterostructures, especially in the context of fully autonomous 2D materials fabrication systems like Brookhaven National Laboratory’s Quantum Material Press (QPress).4

References
[1] Lu, B., Xia, Y., Ren, Y., Xie, M., Zhou, L., Vinai, G., Morton, S. A., Wee, A. T. S., van der Wiel, W. G., Zhang, W., & Wong, P. K. J. (2024). When Machine Learning Meets 2D Materials: A Review. Advanced Science, 11(13), 2305277. https://doi.org/10.1002/advs.202305277
[2] Mao, Y., Wang, L., Chen, C., Yang, Z., & Wang, J. (2023). Thickness Determination of Ultrathin 2D Materials Empowered by Machine Learning Algorithms. Laser & Photonics Reviews, 17(4), 2200357. https://doi.org/10.1002/lpor.202200357
[3] Blake, P., Hill, E. W., Castro Neto, A. H., Novoselov, K. S., Jiang, D., Yang, R., Booth, T. J., & Geim, A. K. (2007). Making graphene visible. Applied Physics Letters, 91(6), 063124. https://doi.org/10.1063/1.2768624
[4] BNL | Quantum Material Press (QPress). (n.d.). Retrieved October 17, 2024, from https://www.bnl.gov/qpress/

Keywords

2D materials

Symposium Organizers

Andrew Mannix, Stanford University
Suji Park, Brookhaven National Laboratory
Dharmraj Kotekar Patil, University of Arkansas
Amirhossein Hasani, Montana State University

Symposium Support

Bronze
MonArk NSF Quantum Foundry - Montana State University
MonArk NSF Quantum Foundry- University of Arkansas
QUANTUM DESIGN

Session Chairs

Amirhossein Hasani
Andrew Mannix

In this Session