Researchers Pitch a Ready-to-Run SOT-MRAM Prototype as a High-Efficiency Memory for Better Computing
AI boom could be fed by cache memory powered by magnetism, rather than electrical charge.
Researchers at the Johannes Gutenberg University Mainz and the Norwegian University of Science and Technology have made a breakthrough that they say, could help reduce the energy needs of future computers: a working spin-orbit torque magnetic random-access memory (SOT-MRAM) device.
"This prototype is one of a kind and could revolutionize data storage and processing. It aligns with global goals to reduce energy consumption and paves the way for faster, more efficient memory solutions," claims lead author Rahul Gupta of the team's creation. "Our approach harnesses a novel fundamental phenomenon by utilizing orbital currents derived from charge currents through the orbital Hall effect, eliminating the dependency on costly and rare materials."
The desire for increasing computing performance has not been matched by gains in efficiency, a problem that has only become more pressing as the current artificial intelligence bubble grows ever-larger. The issue is the focus of a wide range of research projects, looking at everything from more efficient semiconductor materials to a move away from electronics altogether — with the team's SOT-MRAM prototype delivering the promise of dramatic gains in efficiency and overall energy consumption.
Using MRAM, which is based on magnetism rather than an electric charge, as its base, the team's SOT-MRAM prototype is claimed to deliver a 50 percent reduction in overall energy usage over current equivalent devices, a 30 percent boost in efficiency, and a 20 percent reduction in the current required to store data by flipping its magnetic bits — alongside a claimed data storage longevity of at least a decade in the absence of power. Better still, the team says it's ready for integration into commercial devices — and doesn't require the rare, expensive materials of earlier MRAM devices.
The team's work has been published in the journal Nature Communications under open-access terms; no timescale for commercialization has been disclosed, but the technology may be delayed by the closure of project partner Antaios, a French company specializing in SOT-MRAM technology, in 2023.