A team of AI agents and researchers has found two candidates for a room-temperature magnetic semiconductor. The discovery, detailed in a new report, combines machine learning design with quantum mechanical simulation to target next-generation computer memory.
The project brought together the agents with the author. The agents ran density functional theory simulations at two levels of approximation: a faster one (PBE+U) and a slower, usually more accurate one (HSE06). The band gaps and spin windows below come from the more accurate HSE06 method.
The Magnet Primer
Magnets come in two familiar varieties. Ferromagnets, like fridge magnets, align their atomic magnets in the same direction. Antiferromagnets (AF) arrange their atomic magnets in alternating directions, cancelling each other out. Both behave very differently when it comes to storing information.
Spintronics relies on the direction an electron spins — up or down — to encode data. Ferromagnets naturally sort electrons by spin, while ordinary antiferromagnets mix spins at each energy level, making reading and writing difficult.
Three Kinds of Magnet
The report describes three categories:
- Ferromagnetic (Ferro): A macroscopic magnetic field that interferes with nearby materials. Switching is slow and power hungry.
- Antiferromagnetic (Antiferro): No macroscopic field, so materials can be packed closer together. Switching is fast, but spins are mixed by energy level.
- Luttinger compensated (LC): Antiferromagnets where spin-up and spin-down atoms have equal magnetic strength, cancelling out to zero net moment. The up and down atoms sit in different environments, allowing spin sorting by energy.
The Design Target
The ideal material would combine a semiconductor’s band gap with LC behavior. It would have a large spin window — the slice of energy where every available electron state shares the same spin — at the edge of the gap. That window should stay open even at room temperature, where thermal agitation adds noise.
The report describes two paths to that goal. One was designed by the agents themselves. The other was found through the search.
Candidate 1: YBaMnFeO₅
The agents designed a new compound made of five elements: yttrium, barium, manganese, iron, and oxygen. As far as the researchers could find, it has never been made or proposed as this kind of magnet.
The compound is predicted to be a semiconductor with a 2.35 eV band gap. Spin sorting occurs at both sides of the band gap: a window of 1.0 eV for holes and 1.4 eV for electrons. Thermal agitation at room temperature causes only around 26 meV of fluctuation, leaving plenty of room for sorting to hold.
What Comes Next
The discovery advances the search for room-temperature semiconductors. The design approach shows what machine learning can do alongside traditional simulation methods.
The report covers the second candidate and the methods behind the search in full. The report itself offers the full technical account of how the materials were found.
See the video the story is built around at vals.ai.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.