Researchers at Japan's National Institutes for Quantum Science and Technology (QST) have demonstrated a magnetic memory material that can be rewritten using ultrashort laser pulses instead of electric current. The team says the approach could help reduce the energy consumed by future AI systems and data center infrastructure.

Published in Applied Physics Letters, the research describes a magnetic memory material that can be repeatedly written and rewritten using ultrashort laser pulses instead of electrical current. The researchers argue the approach could enable much faster memory switching while reducing energy losses associated with electrically driven memory technologies.

The material is based on a CoFeB artificial ferrimagnet, a structure the researchers say combines optical switching with compatibility with existing magnetic memory technologies. The researchers engineered layers of cobalt, gadolinium, and CoFeB to work together, creating a material in which magnetic states can be switched using a single femtosecond laser pulse.

According to the paper, the system differs from previous all-optical switching demonstrations because the CoFeB layer is directly exchange-coupled within the stack rather than indirectly coupled through a nonmagnetic layer. The researchers argue that this enables ultrafast optical writing while maintaining compatibility with magnetic tunnel junction architectures commonly used in magnetic memory technologies.

Previous demonstrations of all-optical switching have largely been limited to ferrimagnetic materials that are difficult to integrate into practical memory devices. The researchers argue that achieving the effect in a CoFeB-based system is significant because CoFeB is already widely used in spintronic memory technologies due to its strong magnetic readout characteristics and compatibility with magnetic tunnel junctions.

The work comes as data center operators and chipmakers grapple with the growing energy demands of AI workloads, prompting interest in technologies that could reduce power consumption in memory, interconnects, and other supporting infrastructure. As AI workloads increase, attention has increasingly focused on the power consumed not only by processors but also by memory and data movement.

Researchers say optical switching could potentially reduce heat generation and energy losses associated with electrically driven memory while enabling much faster write speeds. In accompanying materials, QST said the approach could eventually support future AI hardware, edge computing systems, and optoelectronic platforms that combine photonic and electronic components.

The team used Japan's NanoTerasu synchrotron facility to study spin arrangements and interlayer interactions within the material, helping guide the design of the multilayer structure.

The research remains in the materials demonstration stage and does not represent a deployable memory technology. However, the authors argue that compatibility with existing magnetic tunnel junction architectures could improve its prospects for future device integration. While the researchers suggest that practical optoelectronic applications could emerge within the next decade, significant engineering work remains before the technology could be incorporated into commercial memory products.