Physicists find a way to dramatically reduce computer memory power consumption — the solution was on the surface

Researchers from the University of Edinburgh have demonstrated that the energy needed for writing data in electronic memory can be greatly reduced, a pressing issue especially relevant after the rise of generative AI platforms. They proposed that current methods of changing a bit in magnetic memory cells are far from optimal, resembling brute-force approaches that use strong pulses against the initial magnetization. Instead, they suggested using magnetic fields with specially calculated time-varying shapes that switch a magnetic element from '0' to '1' along the most energy-efficient path. The pulses were computed using optimal control theory, and computer simulations were performed for single-layer van der Waals magnets Fe3GaTe2, Fe3GeTe2, and CrSBr. The spins were coherently reversed within 1–10 picoseconds, with the optimized magnetic field amplitude more than ten times lower than in traditional switching. The energy per switching event was about 0.94–9.7 nJ, compared to 42.8–91.2 nJ for ordinary magnetic field pulses. Further optimization of device size, magnetic anisotropy, and damping coefficient could theoretically bring costs down to femtojoules. According to the authors' calculations, this method could be more energy-efficient than STT-MRAM and SOT-MRAM technologies under certain parameters. Moreover, the proposed method may also be applied to control currents and lasers, suggesting a broader optimization of data handling.