A latest article in Scientific Studies detailed the characterization of disordered buildings in graphene and monolayer hexagonal boron nitride (hBN) induced by low-energy argon ion irradiation. The research employed automated scanning transmission electron microscopy (STEM) imaging coupled with convolutional neural network-based evaluation to look at these structural adjustments.
Picture Credit score: ogwen/Shutterstock.com
Background
Vacancies, topological defects, or grain boundaries considerably affect the properties of two-dimensional (2D) supplies. For example, level defects in hBN can perform as quantum emitters with properties similar to excessive stability throughout a large temperature vary, vibrant zero-phonon line emission, broad spectral protection, and potential lifetime-limited emission at room temperature.
Equally, graphene amorphized by means of electron irradiation permits the manufacturing of free-standing, steady, and steady monolayer amorphous carbon. Nevertheless, precisely characterizing the construction of amorphous or extremely faulty supplies stays difficult.
Two sensible approaches for acquiring atomic-resolution data are scanning probe microscopy and STEM. STEM, particularly, is extra scalable for analyzing giant areas of samples however requires freestanding 2D materials specimens. Automated picture evaluation strategies will help deal with a few of these limitations, enhancing the effectivity and accuracy of structural characterization.
Strategies
Graphene and hBN samples had been ready on Au Quantifoil STEM grids through chemical vapor deposition (CVD) utilizing commercially out there supplies from Graphenea© (graphene) and Graphene Grocery store (hBN).
For cleansing, graphene samples had been handled with a diode laser (445 nm), whereas one hBN pattern underwent laser pulse cleansing inside the STEM imaging column. One other hBN pattern with larger defect density was annealed in an ultra-high vacuum (UHV) heating stage at 500 °C for one hour. Following these steps, the samples had been uncovered to low-energy ion irradiation utilizing a microwave plasma supply.
After irradiation, the samples had been imaged through STEM beneath UHV circumstances. Graphene was imaged at 40 kV, and hBN at 60 kV. Automated picture acquisition enabled the seize of atomic-resolution photos throughout giant areas (tons of of nanometers per lateral dimension), with extra semi-automatic imaging carried out for comparability.
Every dataset, comprising tons of of photos, was analyzed utilizing an automatic pipeline consisting of three steps: neural community preprocessing, core neural community processing, and postprocessing of the outcomes. The preprocessing mannequin was skilled on 170 randomly chosen experimental photos of graphene and hBN, primarily from low- and medium-defective regimes.
The neural community produced atomic density segmentation maps for every enter picture. These maps had been then analyzed within the postprocessing step to find out atomic positions, offering detailed insights into the structural traits of the samples.
Outcomes and Dialogue
The graphene and hBN buildings displayed notable variations. On the highest defect density, graphene exhibited disordered buildings with quite a few non-hexagonal carbon rings (amorphization). In distinction, faulty hBN largely retained its unique crystal construction with minimal non-hexagonal rings.
Postprocessing of graphene utilizing the steady Delaunay graph offered knowledge on defect areas, lacking atoms, and ring sizes. This methodology carried out successfully for low to medium defect densities, requiring no guide enter. Nevertheless, at excessive defect densities, most defects prolonged past the imaged subject of view as a consequence of in depth lattice amorphization. On this regime, lattice polygon ring sizes had been indicators of structural defectiveness.
For hBN, postprocessing concerned a number of steps, together with the removing of invalid atomic positions, similar to these at picture boundaries or inside contamination. This course of decided the variety of related lacking atoms and their species. Like graphene, the core mannequin precisely recognized atomic positions throughout various defect ranges however struggled to guage pore buildings. For extremely faulty hBN, native lattice injury was estimated primarily based on the imaged space and porous areas.
The preprocessing neural community struggled with porous hBN and extremely faulty graphene, necessitating guide threshold choice to calculate porous or defect areas. In distinction, the postprocessing neural community fashions for graphene and hBN carried out properly for low to medium defect densities. Nevertheless, these fashions had been restricted to options inside the coaching dataset and demonstrated minimal capability for extrapolation to extra complicated buildings.
Conclusion
The researchers examined defect regimes in freestanding monolayer graphene and hBN, starting from pristine materials to excessive defect concentrations and full amorphization. Defects had been launched utilizing low-energy ion irradiation at various ranges.
1000’s of STEM photos had been semi-automatically acquired to offer detailed statistical representations of defects. The evaluation revealed distinct variations between the 2 supplies. In graphene, bond rotations facilitated lattice amorphization, whereas in hBN, growing dysfunction primarily resulted in empty lattice websites with out important adjustments to the general crystal construction.
The research demonstrated the effectiveness of mixing automated atomic-resolution picture acquisition with neural network-based evaluation for characterizing 2D supplies throughout totally different ranges of dysfunction. This method minimizes human bias in picture choice and atomic construction identification, providing a speedy and constant methodology for analyzing such supplies. Nevertheless, a completely automated methodology relevant to any experimental dataset doesn’t but exist.
Journal Reference
Propst, D. et al. (2024). Automated picture acquisition and evaluation of graphene and hexagonal boron nitride from pristine to extremely faulty and amorphous buildings. Scientific Studies, 14(1). DOI: 10.1038/s41598-024-77740-9, https://www.nature.com/articles/s41598-024-77740-9
