Machine-Learning-Enabled Fast Optical Identification and Characterization of 2D Materials - Paper Evaluation

Aug 3, 2026

Upload by Oliver

Two-dimensional materials are atomically thin systems with diverse electronic and quantum properties, where determining layer thickness—especially for monolayers—is essential yet commonly slow and resource intensive. Accurate characterization often relies on expert-driven workflows and costly instrumentation such as atomic force microscopy. This work accelerates identification by applying machine learning to optical microscopy images to infer layer thickness. Three models—SegNet, 1D U-Net, and 2D U-Net—are evaluated for monolayer recognition, alongside experiments on labeling and image-processing strategies to determine an effective approach.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

AccessFree to download

Abstract

Introduction

Background of 2D materials and thickness control

Existing characterization methods and limitations

Optical microscopy contrast mapping and challenges

Methods

Machine learning approach and supervised learning rationale

Model comparison and dataset labeling need

arXiv :2406 . 16211v1 [ cond-mat .mtrl-sci ] 23 Jun 2024 Machine-Learning-Enabled Fast Optical Identification and Characterization of 2D Materials Polina A. Leger, 1 Aditya Ramesh, 1 Talianna Ulloa, 1 and Yingying Wu 1 1 Department of Electrical and Computer Engineering, University of Florida, Gainesville, Florida 32611, USA. Abstract Two-dimensional materials are a class of atomically thin materials with assorted electronic and quantum properties. Accurate identification of layer thickness, especially for a single monolayer, is crucial for their characterization. This characterization process, however, is often time-consuming, requiring highly skilled researchers and expensive equipment like atomic force microscopy. This project aims to streamline the identification process by using machine learning to analyze optical images and quickly determine layer thickness. In this paper, we evaluate the performance of three machine learning models-SegNet, 1D U-Net, and 2D U-Net-in accurately identifying monolayers in microscopic images. Additionally, we explore labeling and image processing techniques to determine the most effective method for identifying layer thickness in this class of materials. I. INTRODUCTION Two-dimensional (2D) van der Waals (vdW) materials are a class of materials that can be atomically thin, down to a thickness of ∼0.8 nm. These materials exhibit weak interlayer vdW interactions and strong intralayer covalent bonding. Due to their unique interlayer interaction, they provide access to a broader class of materials with tunable properties such as energy dispersion relations, bandgap control and carrier mobility [1–8] . This tunability allows for greater design control over device properties, making them suitable for applications ranging from insulators in gate dielectrics to groundbreaking new electronics, such as superconductors for quantum computers [1, 4, 9–11] . These materials were first discovered in 2004 when graphene was fabricated using a piece of scotch tape to exfoliate an atoms-thick layer from graphite [12, 13] . Today, large-scale manufacturing processes, such as vapor-phase epitaxy[14–16] and chemical vapor deposition[17–19], are employed alongside the exfoliation method to grow these materials. Precisely controlling variables such as layer thickness during fabrication is challenging. As a result, characterization techniques are essential to properly identifying the grown material based on its physical and electrical properties. Optical spectroscopy[20, 21], Raman spectroscopy[3, 7], photoluminescence[10, 22], and atomic force microscopy[8, 23] are some of the techniques used to determine the thickness of 2D flakes. However, these methods can be time-consuming and inefficient. In addition, the delicate nature of 2D materials[24–26] requires complex set-up like a closed environment with inert gas, making certain characterization techniques expensive and not easy to operate. Among these methods, optical microscopy stands out as an easy and cost-effective method. It works by measuring the light path’s reflection at different intensities, depending on material absorption, with the light travelling an additional distance of material thickness when incident on the underlying substrate. The resulting interference between all the wavelengths of incident light can be calculated using contrast equation in Equation 1, where R mat and Rsub denote the reflection spectra of the 2D flake and the substrate [27], respectively. C (λ) = Rsub~~ ~~(λR)s−ub(Rλmat)(λ) (1) The RGB values of each pixel are the averaged contrast spectra at the corresponding range of wavelengths. This allows a functional mapping that associates the thickness of a 2D sheet with its apparent color, like in the case of graphene [27] . However, the contrast spectra vary depending on illumination, substrate thicknesses, substrate type, and materials being studied, leading to a tedious process of determining the correct fun

FAQs

Why is identifying monolayer thickness in 2D materials important?

Accurate layer thickness identification is crucial for correctly characterizing the diverse electronic and quantum properties of two-dimensional materials, especially when the target is a single monolayer.

What existing techniques are used to measure 2D material thickness, and what are their drawbacks?

Techniques such as optical, Raman, photoluminescence, and atomic force microscopy can be time-consuming and may require complex, expensive setups, making them inefficient and difficult to operate for routine thickness determination.

How does the proposed machine learning workflow improve thickness identification using optical images?

Machine learning analyzes optical microscopy images to learn the relationship between pixel color and layer thickness from labeled data, enabling automated classification of layers and reducing the tedious mapping process required by conventional contrast calculations.

Aug 3, 2026

Upload by Oliver