How can Machine Learning Advance Graphene Innovation?

How can machine learning drive graphene innovation?

Over the past five years, the discovery and significant interest in twisted bilayer graphene has spawned an entirely new subfield in advanced materials science and quantum physics: twistronics. In a new study, scientists are applying machine learning algorithms to discover more about this remarkable phenomenon.

Image credit: CNStock/

Discovering the Magic Corner: Twisted Double Layer Graphene

The discovery of two-dimensional graphene shocked the scientific world when it was published several decades ago. Since then, graphene innovation has progressed significantly in terms of synthesis methods, understanding of its unique properties and development of practical applications.

In 2018, researchers at the Massachusetts Institute of Technology (MIT) discovered that they could make graphene superconducting by stacking two atom-thick layers of the material on top of each other at a precise angle of 1.1°.

At this orientation, which the researchers called the “magic twist angle,” double-layer graphene transforms from a weakly correlated Fermi liquid into a highly correlated two-dimensional electron system. The properties of the system are exceptionally sensitive to carrier density, proximity to nearby ports and variation in angle of rotation.

Twisted double-layer graphene was found to have several unique properties. It is superconducting, but it also has interaction induced isolation states, electronic nematicity, low temperature linear resistance at low temperatures, quantized anomalous Hall states and magnetism.

Challenges in Twisttronics

The discovery of twisted graphene led to a new field in materials science and quantum physics: twistronics.

Scientists working in this field are now looking for more two-dimensional materials and stacking angles that can produce remarkable properties.

Researchers are focused on finding magic angles for other van der Waals materials (such as graphene) that exhibit a magnetic ground state — whether that be antiferromagnetic or ferromagnetic — when produced in ultra-thin layers.

But discovering magic angles like the 1.1° angle of graphene is not an easy task.

High-resolution microscopy using techniques such as transmission electron microscopy (TEM) or scanning probe microscopy (SPM) can help researchers measure rotational angles accurately to below 0.01° accuracy. However, these techniques are time consuming and require the samples to be either freestanding or supported on a conductive substrate.

Not only this, but these measurements can only provide very local information on areas smaller than a micrometer – angles of rotation can vary significantly within just a few micrometers.

In practice, therefore, this approach is not viable. Practical applications for twisted bilayer materials require characterizations on arbitrary substrates over a much larger area and in a relatively shorter time frame.

Other techniques include transport measurements of devices under magnetic fields and at low temperatures, but this approach is complex and limited to small areas.

Low-energy electron diffraction (LEED) provides information about the superimposed layers and their orientations, covering larger areas than TEM or SPM. However, the method also requires a conductive substrate and high vacuum conditions.

Raman spectroscopy is a promising technique for bulk identification of magic angles because it obtains a large amount of information about the state of the material from the Raman spectrum at once. However, the differences in Raman spectra are often extremely subtle, and so using the technique to manually identify angles of rotation is prohibitively expensive and time-consuming.

Using machine learning to drive graphene innovation

Rapid, accurate, non-destructive methods of angle discovery are needed to continue innovation in this field.

In new research, scientists at Kyushu University, Japan, propose a machine learning analysis technique to automatically classify the Raman spectra of twisted graphene samples into a series of twist angles.

The study was published in the journal Applied nanomaterials in 2022 alongside an open-source codebase for the machine learning algorithm that researchers developed.

This algorithm has low computational requirements; it is fast and has an accuracy of about 99% compared to manual spectra labeling.

The described method takes advantage of the flexible, non-invasive nature of Raman spectroscopy measurements in combination with the speed and predictive accuracy of machine learning. As a result, the authors argued that it could facilitate exploration in the emerging field of twistronics.

The method involves extracting features from the Raman spectrum of twisted graphene, which were used to train the machine learning model to derive the twist angle within predefined ranges.

The flexibility of the Raman spectroscopy technique also allows the method to be extended to determine how much stress and doping is present in graphene samples.

The method could also be applied to almost any other heterostructure, the researchers say. This means that the field of twistronics can use it to find more materials and magic angles that may see significant practical applications in the coming years.

Not only this, but the study also showed how open source machine learning algorithms can enable the easy and effective integration of artificial intelligence into other research techniques and areas.

Read more: What is Twisted Graphene?

References and further reading

Andrei, EY and AH MacDonald (2020). Graphene bilayers with a twist.

Cao, Y., V. Fatemi, S. Fang, et al. (2018). Unconventional superconductivity in graphene superlattices with a magic angle.

Park, J. G. (2016). Opportunities and challenges of 2D magnetic van der Waals materials: magnetic graphene? Journal of Physics: Condensed

Pilkington, B. (2022). What is Twisted Graphene? [Online] AZO Nano. Available at:

Solís-Fernández, P., and H. Ago (2022). Machine learning determination of the twist angle of bilayer graphene by Raman spectroscopy: implications for van der Waals heterostructures. Applied nanomaterials.

Disclaimer: The views expressed here are those of the author expressed in a personal capacity and do not necessarily represent the views of Limited T/A AZoNetwork, the owner and operator of this website. This disclaimer is part of the Requirements from using this website.

#machine #learning #drive #graphene #innovation

Leave a Comment

Your email address will not be published. Required fields are marked *