A breakthrough in magnetism could make AI chips use dramatically less energy

A new method developed by researchers at Brown University could allow future computer chips to use significantly less energy, potentially revolutionizing artificial intelligence technology. The breakthrough centers around a special class of materials called multiferroics, which possess both electrical and magnetic properties that can be controlled simultaneously with small voltages.

While traditional computer chips require multiferroics to be extremely thin to function properly, a problem known as the "dead layer" effect has prevented their widespread use.

The Brown University team overcame this limitation by creating extremely thin films of bismuth ferrite, a widely studied multiferroic material, without losing its essential properties. By doing so, they demonstrated that the material can be made as thin as just a few angstroms while still retaining its ferroelectric and magnetic characteristics.

What makes multiferroics particularly attractive for computing is the unique connection between their electrical and magnetic properties. By applying a small voltage, the material's atomic structure can be switched between binary values (ones and zeros), allowing for the storage of digital information using minimal electricity. Moreover, the material's magnetic behavior is linked to its electrical polarization, enabling the creation of memory and logic components that require significantly less energy than existing technologies.

The key to the Brown University team's success was their innovative approach to producing thin films of bismuth ferrite. They deposited the material onto a foundation made of dysprosium scandate, a compound derived from rare earth elements. This underlying material creates a mismatch in atomic arrangements, which compresses the bismuth ferrite and helps preserve its atomic structure and properties.

Additionally, the team incorporated a second metallic layer to counteract unwanted electric fields, preventing them from disrupting the material's electrical polarization.

The result is a bismuth ferrite film that maintains its multiferroic properties even when reduced to extremely thin dimensions. Surprisingly, the engineered layers also improved the material's magnetic behavior, causing it to exhibit altermagnetism. This unusual form of magnetism combines the desirable characteristics of both ferromagnetism and antiferromagnetism, offering a potentially valuable combination of speed, compactness, and readable signals for future computer components.

The discovery of this enhanced multiferroic material could be a game-changer for AI computing, as it brings the technology closer to practical use in low-power memory and computing devices. While further development is still needed, the researchers believe that this breakthrough could pave the way for more energy-efficient AI hardware, ultimately reducing the environmental impact of rapidly growing artificial intelligence demands.