Brown University engineers say they have built ultrathin films of a multiferroic material that keep both their electrical and magnetic behavior, removing a barrier that has kept promising materials out of chips. The work, led by Lucas Caretta, an assistant professor of engineering, and published in Nature Materials, targets one of the costliest problems in AI hardware: energy. It is early-stage research, not a product that will appear in data centers soon.

The Dead-Layer Problem

Multiferroics combine electrical and magnetic properties that can be controlled together, using very small voltages. That combination appeals to chip designers because it could allow memory and logic that need far less energy than today's electronics. The catch is thickness. When a multiferroic is made thin enough to fit into a chip, it often loses its behavior, a failure researchers call a dead layer.

According to Brown's account, the material's atomic structure loses its needed shape and chemistry at that scale, and internal electric fields overpower its polarization. The result is a film that no longer works as either an electrical or a magnetic component, which has made thin multiferroics hard to use in real devices.

How the Brown Team Approached It

The team grew bismuth ferrite, a widely studied multiferroic, on a substrate of dysprosium scandate. That substrate has a narrower crystal structure than the film, so it compresses the bismuth ferrite and helps preserve its lattice. The researchers also added a metallic layer to absorb the stray electric fields that would otherwise disrupt polarization.

Caretta's group reports that the film can be made "down to a few angstroms" thick while remaining both ferroelectric and magnetic. The paper also describes altermagnetism, a magnetic state that combines traits of ferromagnets and antiferromagnets. According to the Brown release, the stacked structure makes the bismuth layer behave as an altermagnet, a finding that adds a second line of interest beyond the thickness result.

Why It Matters

AI data centers are power-hungry, and efficiency per operation is becoming as important as raw speed. A material that stays functional at very small thicknesses and switches with small voltages would address two limits at once: how much energy each operation takes and how small devices can get. The article frames the work as a step toward multiferroics in low-power computing, not as a finished component.

The caveats are significant. The Brown release gives no commercial timeline, and the paper itself says the technology still needs further development before it can be incorporated into commercial chips. Lab results on thin films often take years to become manufacturable devices, and there are many steps between a working film and a chip that runs in production.

The release also does not include a specific energy-savings figure, and this article does not supply one. Claims that multiferroics will cut AI chip power should be treated as prospective until devices built from these films are tested in working circuits.

Why Thin Films Are the Hard Part

Thin films are central to chip manufacturing, because the performance of modern transistors and memory cells depends on layers only a few atoms or unit cells thick. A material that works in a bulk sample but fails at chip-scale thickness is of limited use to engineers, however promising its properties look in the lab. That is why the dead-layer problem has been treated as a gatekeeper for multiferroic devices rather than a minor technical detail.

The Brown approach is notable because it tackles the problem from the substrate and the surrounding structure, not from the bismuth ferrite alone. Growing the film on a substrate that compresses its lattice, and adding a metallic layer to absorb stray electric fields, are engineering choices that could in principle be adapted to other thin-film stacks. Whether they transfer to other materials is an open question that the current paper does not settle.

What Comes Next

Several questions will decide whether the work moves beyond the lab:

  • Whether prototype devices can switch and read reliably inside a working circuit.
  • Whether the growth process can scale beyond specialized research substrates.
  • Whether other research groups can reproduce the altermagnetic behavior in ultrathin films.

The Brown team's contribution is a clearer path to thin multiferroics that keep working, which has been one of the field's central obstacles. Whether that becomes a route to lower-power AI hardware will depend on the next stage of engineering rather than on the first demonstration alone.

Sources