Stronger than steel and lighter than aluminum, carbon fiber is a staple in aerospace and high-performance vehicles. Now, scientists at the U.S. Department of Energy’s Oak Ridge National Laboratory have found a way to make it even stronger.

ORNL researchers simulated 5 million atoms to study a novel process for making carbon-fiber composites stronger and more cost-efficient by incorporating a reinforced layer of polyacrylonitrile nanofibers, or PAN nanofibers. Led by ORNL’s carbon and composites group, the team combined fundamental science with molecular dynamics simulations using the Frontier supercomputer to better understand how the reinforcement process works at the atomic scale. Their findings, published in the journal Advanced Functional Materials, could lead to new, ultradurable materials for airplanes, vehicles, and a wide range of manufacturing applications that require stronger, more lightweight materials.

Carbon-fiber composites are made by embedding thin strands of carbon—each thinner than a human hair—into a polymer matrix. Carbon fiber strands are incredibly strong; however, their bond with the surrounding polymer is somewhat weak by comparison.

“When a carbon-fiber composite fails, it typically begins at the interface between the carbon-fiber strands and the polymer matrix,” says Tanvir Sohail, a postdoctoral researcher at ORNL’s National Center for Computational Sciences. “By incorporating a layer of PAN nanofibers at the interface, we redirect stress from the carbon fibers into the surrounding polymer, which improves load distribution and enhances the composite’s overall strength.”

The process of reinforcing the carbon-fiber composites involves a technique called electrospinning, which uses an electric field and a spinning drum to create a spool of PAN nanofibers no larger than 10 nanometers. For comparison, a single sheet of paper is about 100,000 nanometers thick.

Carbon fiber is expensive to manufacture and test, and running extensive physical experiments can quickly become time-consuming and cost-prohibitive. That’s where supercomputing can drastically accelerate the search for materials with the desired characteristics and help guide the experimental development. But simulating the synthesis process is also expensive—computationally, that is.

“Carbon fiber is extremely dense, and modeling it with molecular dynamics requires tracking the behavior of millions, if not billions, of atoms,” says ORNL computational scientist Swarnava Ghosh, who led the Frontier simulations along with Sohail. 

The material is assembled and tested for strength. Strands of PAN fibers with varying diameters are bundled and tightly bonded to a polymer coating for reinforcement. A graphene sheet is used to simulate pulling on the material to identify weak points. Credit: ORNL, U.S. Department of Energy

Most high-performance computing clusters used by industry and universities can simulate upward of a few thousand atoms. But the calculations use approximation techniques that group atoms together instead of calculating them individually. The technique is faster and far less computationally demanding, but it’s also less accurate and can introduce errors.

The research team was given a small allocation of time on Frontier via a director’s discretionary award to demonstrate the benefits of using a leadership-class supercomputer. Frontier is the world’s most powerful supercomputer for open science, with a peak performance of 2 exaflops per second, meaning it can perform more than a billion-billion calculations per second.

Using only a fraction of Frontier’s power, Sohail and Ghosh built a 5-million-atom model of carbon-fiber composites reinforced by PAN nanofibers. This model provided unprecedented molecular-level insights into the fundamental forces that bind the materials together.

To find the right fit, the team modeled PAN nanofibers with diameters that ranged 6–10 nanometers by using the Large-Scale Atomic/Molecular Massively Parallel Simulator, or LAMMPS, a powerful open-source code used for molecular dynamics simulations on supercomputers. It’s widely used to model the behavior of atoms in materials such as metals, polymers, and biomolecules, and to investigate how the materials respond to conditions such as stress, temperature, and chemical changes over time. 

The simulations showed that PAN nanofibers with a diameter of approximately 6 nanometers offered the best performance. The thinner fibers aligned more uniformly at the interface, thereby improving both mechanical strength and stress transfer from the carbon fiber to the polymer. 

“Simulating materials with 5 million atoms would not have been possible without the power of Frontier,” Sohail says. “To our knowledge, this work is the first hierarchical, fully atomistic simulation of a complete bulk PAN nanofiber integrated within a polymer matrix without relying on any assumptions or simplified calculations.”

Building on the success of their simulations, Sohail and Ghosh plan to submit a proposal for additional time on Frontier through the Innovative and Novel Computational Impact on Theory and Experiment program. The allocation would allow them to harness more of the machine’s power to expand their research. They also plan to integrate artificial intelligence technology into their simulations to characterize a wider range of advanced composites with multifunctional properties that can be used to develop energy-efficient technologies.

“Ultimately, we’re trying to make materials not just stronger but smarter and more efficient,” Ghosh says. “Now, with a proven method for strengthening carbon fiber, we can apply it to a wide range of other materials that are vital for industry and manufacturing.”

In addition to Sohail and Ghosh, the ORNL research team includes Sumit Gupta, Marti Checa, Michael Toomey, Logan Kearney, Rajni Chahal, Sargun Singh Rohewal, Nihal Kanbargi, Liam Collins, David McConnell, Ilia N. Ivanov, Amit K. Naskar, and Christopher Bowland.

ORNL houses Frontier at the Oak Ridge Leadership Computing Facility, a U.S. Department of Energy Office of Science user facility.

Published June 19, 2025, by Oak Ridge National Laboratory.