AI agents found two magnetic semiconductor candidates. Experiments still come next.
spintronics
Technology that stores and reads information using electron spin.
Luttinger-compensated magnet
A magnet whose opposing effects cancel while its electron spins still separate by energy.
spin window
An energy range where available electron states have the same spin.
What happened
On October 4, 2026, Vals AI published a research post about two possible memory materials. Claude Opus 5.5 agents helped search for them. The post describes candidates, not a finished device. Both materials are predicted to have almost no total magnetism. Yet their electrons may still separate by spin. Vals AI’s report
The story also became a discussion on Hacker News. That attention shows interest. It does not prove the report is correct. Hacker News discussion
Why the materials are unusual
A semiconductor controls how electricity moves. Spintronics uses an electron’s spin to store and read information. A ferromagnet sorts spins well, but its magnetic field can leak outside. That field can disturb nearby parts. Ordinary antiferromagnets have little outside field. They can sit closer together and switch much faster. However, their electron spins are usually mixed. That makes them harder to read.
The report focuses on Luttinger-compensated magnets. Opposing magnetic effects cancel, so the total spin moment becomes zero. The atoms still sit in different environments. Their electron spins can therefore separate by energy. The useful range is called a spin window. The larger it is compared with room-temperature thermal motion, the better.
The two candidates
The first candidate is YBaMnFeO₅. It is a new five-element design. The authors found no record of it being made as this type of magnet. Simulations predict a semiconductor with a 2.35 eV band gap. The predicted spin windows measure 1.0 eV for holes and 1.4 eV for electrons. Magnetism remains in the simulation near 420 K. A calibrated estimate reaches about 490 K.
There is a serious manufacturing problem. Manganese and iron must form a precise checkerboard pattern. Simulations show that pattern breaking down around 950 K. The oxide may need 900–1300°C to make. At lower temperatures, atoms barely move. A normal synthesis could produce the wrong arrangement.
The second candidate is KV[Cr(CN)₆]. Chemists first made it in 1999. It belongs to the same family as Prussian blue. Earlier work showed related spin-resolved electron states. It did not identify this material as a Luttinger-compensated semiconductor. New calculations predict a 2.1 eV band gap. A 1999 sample stayed magnetically ordered up to 376 K, or 103°C.
Why it matters
If confirmed, one material could combine two useful traits. It could avoid a large outside magnetic field while still sorting electron spins. That combination might help future spin-based memory. It could allow denser placement without losing readable spin information. These are possibilities, not product results.
What is confirmed, and what is not
The report used two quantum-mechanical calculation methods. It shares input files, raw outputs, analysis code, reruns, and known caveats in a public repository. The old sample’s magnetic ordering was measured. However, neither candidate has measured band-gap and spin-sorting results. The existing KV[Cr(CN)₆] sample is a water-containing powder. It has a small leftover magnetic moment of 0.125 Bohr magnetons per formula unit. HSE06 predicts spin sorting survives water. Faster PBE+U predicts the hole window shrinks by more than half.
What to watch next
Researchers need to make KV[Cr(CN)₆] again and measure its spin sorting directly. They also need to test whether ordered YBaMnFeO₅ can be made. Independent reruns and experiments will decide whether these calculations describe useful materials.
AI found two possible room-temperature magnet materials
📰 Full story: AI agents found two magnetic semiconductor candidates. Experiments still come next.
The materials might help future computer memory. People still need to test them.
spin
A small up-or-down feature of an electron.
spintronics
Technology that uses electron spin to store and read information.
band gap
An energy gap that helps define how a semiconductor handles electrons.
💡 The gist
- AI studied two possible magnetic semiconductors.
- One is new on paper. The other was made in 1999.
- The important properties still lack direct experiments.
Vals AI published the report on October 4, 2026. Claude Opus 5.5 agents helped study the materials. Read the report
A semiconductor controls electric flow. These materials also involve electron spin. Spin is a small up-or-down feature of an electron. Spintronics uses spin to store information.
Normal magnets can send magnetic force into nearby parts. That force can cause problems. Some materials cancel their magnetic forces. They are called antiferromagnets. They can work close together. But their electron spins often mix together.
The new candidates may solve both problems. Their total magnetism may cancel. Their electron spins may still separate. This kind of material is called a Luttinger-compensated magnet.
The first candidate is YBaMnFeO₅. It is a new design. Nobody has made it yet. Its atoms must form a very exact checkerboard pattern. Computer tests say the pattern may break apart at high temperatures.
The second candidate is KV[Cr(CN)₆]. Chemists made it in 1999. That old sample kept magnetic order up to 376 K. This equals 103°C. The material may be easier to make than the first candidate.
The computer results are not enough. Scientists have not directly measured the band gap. They have not measured spin sorting either. The old sample also contains water. Different computer methods predict different effects from that water.
The next step is clear. Scientists need to make KV[Cr(CN)₆] again. Then they need to measure its spin sorting.
The report became a popular Hacker News discussion. Popularity shows attention, not truth. See the discussion
💬 AI found computational candidates, not a verified material
The thread clarified that this is about magnetic semiconductors, not superconductors. The candidates may be useful, but no experimental proof is reported.
- Some readers thought room-temperature meant a new superconductor like LK-99. It is actually about magnetic semiconductors; room-temperature semiconductors already exist, so the magnetic behavior is the point.
- The agents used Quantum ESPRESSO and DFT, with PBE+U and HSE06, to search computationally. They proposed candidates; they did not make them in a lab.
- This is close to an AI-guided search where ordinary simulations score ideas. People can inspect the code and logs, but DFT is approximate, Quantum ESPRESSO is not automatically state of the art, and high-temperature magnetism is hard to predict.
- The thread has no report of lab synthesis, measurements, or independent replication. Commenters claim one candidate is very hard to make because heat could destroy its checkerboard atom pattern, while the other was made only once 27 years ago.
- Supporters see useful testable ideas. Critics say the computational work is within reach of materials-science graduate students and that synthesis and measurement are the real challenge.
- It is therefore a promising, possibly previously overlooked lead—not a verified new material. Experiments, provenance checks, and outside confirmation are still needed.
initial digest at 154 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
A computer helper found two special magnet materials
📰 Full story: AI agents found two magnetic semiconductor candidates. Experiments still come next.
They might help computers remember things someday. People must test them first.
semiconductor
A material that can control how easily electricity moves.
spin
A tiny up-or-down feature found in an electron.
YBaMnFeO₅
A new material design that people have not made yet.
Vals AI is a company that studies materials with AI.
Claude Opus 5.5 is a computer helper. It studied two materials with people.
Both materials may keep magnet behavior in a normal room.
YBaMnFeO₅ is a new material design. People have not made it yet.
KV[Cr(CN)₆] was made in 1999.
Tiny parts called electrons have a small feature. That feature is called spin. Computers might use spin to remember things.
The first material needs a very neat atom pattern. The pattern may fall apart when it gets hot.
The older material has water inside its powder. Computer guesses disagree about what water does.
People have not measured the key features yet. They need to make the older material again. Then they need to measure its spin.
Hacker News talked about the report. Attention does not prove the report is true. Report
💬 The AI's material idea still needs testing
A computer found an interesting possibility, but people have not proved that the material is real and useful.
- This is not a room-temperature superconductor. AI used computer calculations to suggest semiconductors with interesting magnetic behavior.
- The AI tried many ideas with an existing tool such as Quantum ESPRESSO and two approximations. That can find useful clues, but calculations can be wrong, especially when guessing high-temperature magnetism.
- Nobody in the thread reports making and measuring the new candidates. Commenter reports say one may be very hard to build without ruining its atom pattern, and the other was made once 27 years ago.
- So some people see a useful research hint, while others say the hard work is still making and testing it. It is not a confirmed discovery, and it may even be an old idea found again.
initial digest at 154 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
💬 AI-proposed room-temperature magnetic semiconductors remain unverified
The HN discussion mixed a correction of the superconductor misunderstanding with concerns about DFT, synthesis, and how much credit the agents deserve. The common bottom line was that computational candidates are not experimentally discovered materials.
initial digest at 154 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.