An international team of researchers, in an effort to produce cheap, defect-free graphene in larger quantities, have used GCS HPC resources to develop more efficient methods for producing graphene at an industrial scale.

Real-Time Multiscale Monitoring and Tailoring of Graphene Growth on Liquid Copper image

Recent research indicates that using a liquid copper catalyst may be a fast, efficient way for producing graphene, but researchers only have a limited understanding of molecular interactions happening during these brief, chaotic moments that lead to graphene formation, meaning they cannot yet use the method to reliably produce flawless graphene sheets.

In order to address existing challenges and help develop methods for quicker graphene production, a team of researchers at the Technical University of Munich (TUM) has been using the JUWELS and SuperMUC-NG high-performance computing (HPC) systems at the Jülich Supercomputing Centre (JSC) and Leibniz Supercomputing Centre (LRZ) to run high-resolution simulations of graphene formation on liquid copper.

Graphene’s appeal primarily stems from the material’s perfectly uniform crystal structure, meaning that producing graphene without impurities is the main goal. To produce large amounts of graphene, methods that could reliably produce high-quality graphene cheaper and faster are required. One of the more promising methods under investigation involves using a liquid metal catalyst to facilitate the self-assembly of carbon atoms from molecular precursors into a single graphene sheet growing on top of the liquid metal. While the liquid offers the ability to scale up graphene production efficiently, it also introduces a host of complications, such as the high temperatures required to melt the typical metals used, such as copper.

While technological advances have opened up new ways for gaining insight into atomic-scale behavior even under extreme conditions such as very high temperatures, experimental techniques do not always allow researchers to observe the ultra-fast reactions that facilitate the correct changes to a material’s atomic structure (or what aspects of the reaction may have introduced impurities). This is where computer simulations can be of help, however, simulating the behavior of a dynamic system such as a liquid is not without its own set of complications.

“The problem describing anything like this is you need to apply molecular dynamics (MD) simulations to get the right sampling,” Mie Andersen from TUM said. “Then, of course, there is the system size—you need to have a large enough system to accurately simulate the behavior of the liquid.” Unlike experiments, molecular dynamics simulations offer researchers the ability to look at events happening on the atomic scale from a variety of different angles or pause the simulation to focus on different aspects.

While MD simulations offer researchers insights into the movement of individual atoms and chemical reactions that could not be observed during experiments, they do have their own challenges. Chief among them is the compromise between accuracy and cost—when relying on accurate ab initio methods to drive the MD simulations, it is extremely computationally expensive to get simulations that are large enough and last long enough to accurately model these reactions in a meaningful way.

Andersen and her colleagues used about 2,500 cores on JUWELS in periods stretching over more than one month for the recent simulations. Despite the massive computational effort, the team could still only simulate around 1,500 atoms over picoseconds of time. While these may sound like modest numbers, these simulations were among the largest done of ab initio MD simulations of graphene on liquid copper. The team uses these highly accurate simulations to help develop cheaper methods to drive the MD simulations so that it becomes possible to simulate larger systems and longer timescales without compromising the accuracy.

The team published its record-breaking simulation work in the Journal of Chemical Physics, then used those simulations to compare with experimental data obtained in their most recent paper, which appeared in ACS Nano.

Andersen indicated that current-generation supercomputers, such as JUWELS and SuperMUC-NG, enabled the team to run its simulation. Next generation machines, however, would open up even more possibilities, as researchers could more rapidly simulate larger numbers or systems over longer periods of time.

Despite the need for more research into using liquid catalysts to produce graphene, Andersen indicated that the two-pronged approach of using both HPC and experiment would be essential to further graphene’s development and, in turn, use in commercial and industrial applications. “In this research, there is a great interplay between theory and experiment, and I have been on both sides of this research,” she said.