Rice University researchers used a microstructure model of radiation-damaged hexagonal boron nitride to help them study the benefits of deep learning techniques for simulating two-dimensional materials to understand their characteristics. Image: Prabhas Hundi/Rice University.Scientists are discovering new two-dimensional (2D) materials at a rapid pace, but they don't always immediately know what those materials can do.
Researchers at Rice University's Brown School of Engineering say they can now quickly deduce this information by feeding basic details of the materials’ structures into ‘deep learning’ agents that have the power to map their properties. Better yet, the agents can quickly model materials that scientists are thinking about making to facilitate the ‘bottom-up’ design of novel 2D materials.
Rouzbeh Shahsavari, an assistant professor of civil and environmental engineering, and graduate student Prabhas Hundi were exploring the capabilities of neural networks and multilayer perceptrons. These can take minimal data from the simulated structures of 2D materials and make ‘reasonably accurate’ predictions of their physical characteristics, like strength, even after the materials have been damaged by radiation and high temperatures.
Once trained, Shahsavari said, these agents could be adapted to analyze new 2D materials using as little as 10% of their structural data. That would return an analysis of the material's strengths with about 95% accuracy, he said.
"This suggests that transfer learning (in which a deep-learning algorithm trained on one material can be applied to another) is a potential game-changer in material discovery and characterization approaches," the researchers suggested. They report the results of their extensive tests on graphene and hexagonal boron nitride in a paper in Small.
Since the discovery of graphene in 2004, atom-thick materials have been touted for their strength and range of electronic properties for use in composites and electronics. Because their atomic arrangements have a significant impact on their properties, researchers often use molecular dynamics simulations to analyze the structures of new 2D materials even before trying to make them.
According to Shahsavari, deep learning offers a significant speed boost over such traditional simulations of 2D materials and their characteristics, allowing calculations that now take days of supercomputer time to run in hours.
"Because we can build our structure-property maps with only a fraction of the data from graphene or boron nitride molecular dynamics simulations, we see an order of magnitude less computational time to get a full behavior of the material," he said.
Shahsavari and Hundi decided to study graphene and hexagonal boron nitride because of their high tolerance to deterioration under high temperatures and in radiation-rich environments, important properties for materials in spacecraft and nuclear power plants. The Shahsavari group had already carried out more than 11,000 molecular dynamics simulations on radiation cascade damage for another paper on 2D materials, and so they decided to see if they could reproduce their results with the much faster ‘deep learning’ method.
They ran thousands of ‘deep learning’ simulations on 80 combinations of radiation and temperature for hexagonal boron nitride and 48 combinations for graphene, hitting each combination with 31 random doses of simulated radiation. For some, the researchers trained the deep learning agent with a maximum of 45% of the data from their molecular dynamics study, but found that the agent achieved up to 97% accuracy in predicting defects and their effects on the material's characteristics.
Adapting trained agents to different materials, they found, required only about 10% of the simulated data, greatly speeding up the process while retaining good accuracy.
"We tried to figure out the corresponding residual strengths of the materials after exposure to extreme conditions, along with all the defects," Shahsavari explained. "As expected, when the mean temperature or the radiation were too high, the residual strength became pretty low. But that trend wasn't always obvious."
In some cases, he said, the combined higher radiation and higher temperatures made a material more robust instead of less, and it would help researchers to know that before making a physical product.
"Our deep learning method on the development of structure-property maps could open up a new framework to understand the behavior of 2D materials, discover their non-intuitive commonalities and anomalies, and eventually better design them for tailored applications," Shahsavari said.
This story is adapted from material from Rice University, with editorial changes made by Materials Today. The views expressed in this article do not necessarily represent those of Elsevier. Link to original source.