University of Southern California: Discovering New Materials: AI Can Simulate Billions of Atoms Simultaneously

Imagine the concrete in our homes and bridges not only withstanding the ravages of time and natural disasters like the intense heat of wildfires, but actively self-healing or capturing carbon dioxide from the atmosphere. Now, researchers at the USC Viterbi School of Engineering have developed a revolutionary AI model that can simulate the behavior of billions of atoms simultaneously, opening new possibilities for materials design and discovery at unprecedented scales.

The current state of the world's climate is a dire one. Brutal droughts, evaporating glaciers, and more disastrous hurricanes, rainstorms and wildfires devastate us each year. A major contributor to global warming is the constant emission of carbon dioxide into the atmosphere. Aiichiro Nakano, a USC Viterbi professor of computer science, physics and astronomy, and quantitative and computational biology, was contemplating these issues after the January wildfires in Los Angeles. So, he reached out to longtime partner Ken-Ichi Nomura, a USC Viterbi professor of chemical engineering and materials science practice, with whom he's collaborated for over 20 years.

Discussing these issues together helped spark their new project: Allegro-FM, an artificial intelligence-driven simulation model. Allegro-FM has made a startling theoretical discovery: it is possible to recapture carbon dioxide emitted in the process of making concrete and place it back into the concrete that it helped produce. "You can just put the CO2 inside the concrete, and then that makes a carbon-neutral concrete," Nakano said. By simulating billions of atoms simultaneously, Allegro-FM can test different concrete chemistries virtually before expensive real-world experiments. This could accelerate the development of concrete that acts as a carbon sink rather than just a carbon source — concrete production currently accounts for about 8% of global CO2 emissions. The breakthrough lies in the model's scalability. While existing molecular simulation methods are limited to systems with thousands or millions of atoms, Allegro-FM demonstrated 97.5% efficiency when simulating over four billion atoms on the Aurora supercomputer at Argonne National Laboratory.