New AI Function on the Horizon Thanks to Electrically Programmable Spintronic Device

New AI Function on the Horizon Thanks to Electrically Programmable Spintronic Device
by Riko Seibo
Tokyo, Japan (SPX) Feb 18, 2025

The rapid evolution of artificial intelligence (AI) continues to reshape industries, but with its growing capabilities comes increased power consumption. Addressing this challenge requires innovative AI hardware solutions that enhance computational efficiency while reducing energy demands.

Researchers from Tohoku University, the National Institute for Materials Science, and the Japan Atomic Energy Agency have developed a novel spintronic device that facilitates electrical mutual control of non-collinear antiferromagnets and ferromagnets. This advancement allows for highly efficient magnetic state switching, optimizing data storage and processing with minimal energy consumption-akin to brain-like AI chips.

The findings were published in Nature Communications on February 5, 2025.

"While spintronic research has advanced significantly in electrically controlling magnetic order, most spintronic devices separate the functions of the controlled magnetic material and the driving material," explained Shunsuke Fukami of Tohoku University, who led the research.

Traditional spintronic devices operate with a fixed binary switching mechanism, but this new innovation enables electrically programmable switching across multiple magnetic states, a major leap forward in AI chip technology.

The research team utilized the non-collinear antiferromagnet Mn3Sn as the central magnetic material. When an electrical current is applied, Mn3Sn generates a spin current that drives the switching of an adjacent ferromagnet, CoFeB, through the magnetic spin Hall effect. This interaction not only influences the ferromagnet but also reciprocally affects the magnetic state of Mn3Sn, enabling electrical mutual switching between the materials.

In their proof-of-concept experiment, the researchers demonstrated that data written to the ferromagnet could be controlled electrically via Mn3Sn's magnetic state. By fine-tuning the applied current, they achieved multiple magnetization states in CoFeB, facilitating analog switching. This mechanism mirrors how synaptic weights function in neural networks, a crucial operation in AI processing.

"This breakthrough is a critical step toward more energy-efficient AI chips. By achieving electrical mutual switching between a non-collinear antiferromagnet and a ferromagnet, we open new opportunities for programmable neural networks," stated Fukami. "Our next focus is on reducing operating currents and enhancing readout signals, which are vital for AI chip applications."

The team's innovation paves the way for AI hardware that is both highly efficient and environmentally sustainable.

Research Report:Electrical mutual switching in a noncollinear-antiferromagnetic-ferromagnetic heterostructure