Ultrafast magnetic-pulse memory could cut AI data center energy use by 100x, study finds

Magnetic memory stores data by switching its magnetic state, but the standard switching techniques can be energy-intensive. The Edinburgh researchers instead apply ultrafast magnetic-field pulses to achieve the switching, which the paper says can cut energy usage by "up to two orders of magnitude" compared with conventional methods.

The findings are described in a paper in the journal Advanced Materials. They come as data center power demand grows rapidly with the expansion of artificial intelligence. The paper's authors warn that without significant improvements in efficiency, information and communication technologies could eventually represent a sizable share of worldwide electricity consumption and carbon emissions.

In the paper, study author Dr. Elton J.G. Santos said: "The same framework can be adapted to electrical currents and even ultrafast laser pulses, which are among the most cutting-edge technologies for future data storage. That means the ideas developed here could have applications far beyond the systems we studied."

The research is still theoretical, and the authors have outlined practical steps that could help others build prototypes and run experiments. TechRadar cautions that it may be a long time before the ideas are put into practice, but it notes they offer hope that data centers of the future could be far less energy-intensive than today's.