Opus 5.5 agents surface two room-temperature magnetic semiconductor candidates

Vals AI has described how a team of AI agents built on Claude Opus 5.5, working alongside researcher Geby Jaff, turned up two candidate magnets for spintronics — the technology that stores data using electron spin.

What they were after

Two familiar magnet types compete in memory. The ferromagnet (your fridge magnet) sorts electrons by spin nicely but leaks a disruptive field and switches slowly. The antiferromagnet throws off almost no field, switches roughly a thousand times faster and can be packed densely — but its spins are mixed, so reading the information is hard.

The coveted middle ground is the Luttinger-compensated magnet: net magnetisation is zero, yet the "up" and "down" atoms sit in inequivalent sites, so spins can still be sorted by energy as in a ferromagnet. The ideal is a semiconductor with a band gap, no stray field, and a "spin window" far wider than room-temperature thermal noise (about 26 meV).

Two candidates

The agents ran quantum-mechanical calculations (density functional theory, at PBE+U and the more accurate HSE06).

YBaMnFeO₅ — a new five-element compound that, per the authors, had never been made. The calculation gives a 2.35 eV band gap, spin windows of 1.0 and 1.4 eV, and magnetism up to roughly 420–490 K. The catch: the scheme needs a perfect checkerboard of Mn and Fe atoms, which falls apart around 950 K. Under standard synthesis the crystal would likely come out scrambled, losing its spin sorting.

KV[Cr(CN)₆] — a relative of the 300-year-old pigment Prussian blue, first made back in 1999. Its zero magnetisation and spin sorting had essentially already been seen (a 2008 study), but nobody had called it a compensated semiconductor. The prediction: a band gap near 2.1 eV, windows of 2.6 and 1.6 eV, and the 1999 sample stayed magnetically ordered up to 376 K — above room temperature. Crucially, its structure locks each metal into its own site, exactly what the first candidate lacked.

Caveats

The numbers are for a perfect, dry crystal. The only sample is a powder with water in its pores and a small residual moment; the two methods disagree on how much the water matters. Neither the band gap nor the spin sorting has been measured yet. The next step is to remake KV[Cr(CN)₆] and test it directly. The authors have posted all calculations, code and a list of known caveats on GitHub.

Source: Hacker News