AIBRIEF
Both are still computational predictions for low power magnetic memory chips, the class behind MRAM and hard drive read heads: one was designed from the laws of physics, not empirical tuning, the other is a 1999 material being re checked
A research team running an AI-assisted search across a large chemical space has published two shortlist candidates for a room-temperature magnetic semiconductor, the class of material that underpins faster, lower-power memory chips like MRAM and spintronic read heads.
One candidate, YBaMnFeO₅, was designed from first principles by the workflow. The other, KV[Cr(CN)6]·2H2O, is a 1999 compound being re-evaluated computationally, not a freshly invented material. Both remain predictions on paper; nothing has been synthesized or measured in a lab.
The team's preprint (arXiv 2502.18136) frames the candidates as a "Luttinger compensated bipolarized magnetic semiconductor," a 1950 theoretical class the source blog describes as never experimentally realized, though that novelty claim needs cross-checking against the original Luttinger literature. The team's public ledger on GitHub lists inputs, structures, and the workflow so other groups can test the predictions.
The work is computational only, and the framing drew pushback on a Hacker News discussion. A PhD commenter in magnetic materials noted that magnetism is not a simple ferromagnet-vs-antiferromagnet binary; it also includes diamagnets, paramagnets, non-collinear antiferromagnets, and other orders the vendor copy collapses. The shortlist is now public. The test is whether an experimental group can grow either compound and confirm the predicted room-temperature behavior.