Georgia Tech uses supercomputers and machine learning to analyze electronic materials to create better capacitors

According to foreign media reports, capacitors will play an important role in powering future machines such as electric vehicles and mobile phones due to their high energy output and fast charging speed, but a major obstacle for capacitors to become energy storage devices is that they store much less energy than batteries of the same size . Researchers at the Georgia Institute of Technology have found a novel way to solve the above problems. Using supercomputers and machine learning technology, the researchers finally succeeded in creating more powerful capacitors. The research involved teaching computers to analyze the atoms of two materials used to produce capacitors – aluminum and polyethylene.

(Image credit: University of California, San Diego)

The researchers are focused on finding ways to more quickly analyze the electronic structure of capacitor materials to identify properties that affect capacitor performance. "The electronics industry wants to understand the structure and properties of all materials used to make electronic devices, including capacitors," said Rampi Ramprasad, a professor in Georgia Tech's School of Materials Science and Engineering. "For example, polyethylene is a very good insulator with a large band gap, an energy range that charge carriers cannot reach, but it has a disadvantage that excess charge carriers can enter the band gap, reducing efficiency."

"To understand where the defects are and what role they play, we need to calculate the entire atomic structure of the material, which has been very difficult until now," Ramprasad said. "Currently, analyzing such materials using quantum mechanics is too slow, limiting the amount of analysis that can be done in a given time."

Ramprasad and his colleagues used machine learning to develop new materials. They used data samples generated by quantum mechanical analysis of aluminum and polyethylene to teach a powerful computer how to simulate such analysis. Quantum mechanical analysis of the electronic structure of materials involves solving the Cohen-Sham equation of density functional theory, which produces wave functions and energy level data that can be used to calculate the total potential energy and atomic forces of the system.

The researchers used the Comet supercomputer at the San Diego Supercomputer Center, an organized research unit at UC San Diego, for early calculations, and the Stampede2 supercomputer at the Texas Advanced Computing Center at the University of Texas at Austin for later stages of this research.

The new machine learning method produced several orders of magnitude more similar results than conventional techniques based on quantum mechanics. "The increase in computing power will allow us to design electronic materials that are better than what is currently available," Ramprasad said.

Although this study focused on aluminum and polyethylene, machine learning methods can be used to analyze the electronic structure of a wider variety of materials. Ramprasad said that in addition to being able to analyze electronic structure, other aspects of material structure that are now analyzed by quantum mechanics can also be accelerated by machine learning methods.

The faster processing speeds enabled by machine learning methods will allow researchers to more quickly simulate how changes to a material will affect its electronic structure, thereby finding new ways to improve its efficiency. "Supercomputer systems enable high-throughput computing, which allows us to create a massive database of knowledge about various material systems, and this knowledge can then help us find the best material for a particular application," Kamal said.