Stockholm university logo, link to start page

Using AI guides to find new materials for electronics and more

Sidestepping hit-and-miss experiments and the need for luck in the lab, artificial intelligence can help direct chemists and companies to likely new organic materials for use in electronics – and more, including unlocking secrets of the Universe.

The organic crystal shown here is composed of three different molecules, arranged in a periodic pattern. It can transfer an electron charge, which may be ideal for use in electronics.
The organic crystal shown here is composed of three different molecules, arranged in a periodic pattern. It can transfer an electron charge, which may be ideal for use in electronics.
 

A database built over the past five years could provide the next material for everyday use, in mobile phones, flat screens or other electronic devices. And instead of taking years of painstaking lab work to discover that material, researchers can use artificial intelligence (AI) tools to identify likely candidates in the database in five minutes.

Researchers from Nordita, the Nordics physics collaboration hosted by Stockholm University with KTH, have built the Organic Materials Database, based on nearly 41,000 real organic crystalline materials. Led by Alexander Balatsky of Stockholm University, the Nordita team trained AI algorithms to comb through the digital library, teaching the algorithms what properties might be intrinsic to the various arrangements of atoms peculiar to each crystal.

Algorithms predict what organic crystals can do

With that “artificial knowledge”, these algorithms can then predict what unknown organic crystals can do. Can they efficiently absorb sunlight, and store electricity for a new type of solar panel? Would they conduct electrons without resistance at high temperatures? Or maybe they are semiconductors that would be useful for making transistors or electronic circuits, or perhaps they can emit light in specific colors, making them ideal for screens.

By modelling unknown crystals, the AI algorithms can expand the library from the 40,000-plus known materials to 200,000 organic crystals with previously unknown properties. Their characteristics can be predicted by computer modelling.

Saving time and costs for researchers and companies

Matthias Geilhufe
Matthias Geilhufe

That modeling, or mapping of possibilities, will save a lot of time and costs for researchers, companies and their R&D divisions, says Matthias Geilhufe, a lead researcher on the OMDB project, and first author of an opinion published in Nature Physics in January 2021. Instead of painstakingly creating organic molecules in a lab and testing if they have the properties they want – a process that once took decades – researchers can test just a few of the likeliest candidates, selected by AI. If they don’t find the right target material on the first try but are heading in the right direction, they can change course as needed.

“Machine learning is useful to fill in the blanks,” says Matthias Geilhufe. “It can say, ‘Look there, do the experiments around that spot,’” like asking a mapping app to find a restaurant that serves sushi near your office.

Useful in quantum computers

More than 700 users have signed up to explore the OMDB. Matthias Geilhufe himself is using it to find materials that would work for quantum computing. The goal is to find organic molecules with a metal atom core that can be used to make stable Qubits, the components of a quantum computer. The OMDB could hold the key to that material, among many others – beyond screens and electronics, to finding detectors for Dark Matter, the so-far unseen matter that makes up 85 percent of the Universe. Matthias Geilhufe says that now it’s up to the users of OMDB to find these new materials, with the help of their AI guides.

Text: Naomi Lubick
 

Founding sources
The research is founded by ERC Synergy Grant ‘HERO’, VILLUM FONDEN via Centre of Excellence for Dirac Materials, Knut and Alice Wallenberg Foundation, Vetenskapsrådet and Swedish National Infrastructure for Computing (SNIC).