AI Memory Energy Reduction via Optimal Control Magnetic Switching

Researchers at the University of Edinburgh have developed a novel mathematical framework utilizing optimal control theory to dramatically reduce the energy required to write digital information in magnetic memory. Published on August 21, 2026, in Advanced Materials, the study demonstrates that carefully shaping magnetic field pulses can reverse magnetization in 1 to 10 picoseconds while consuming as little as 0.94 nanojoules compared to 91.2 nanojoules for conventional pulses. The author highlights that as artificial intelligence and large-scale data processing continue to expand exponentially, memory energy consumption has become a critical bottleneck. Traditional field pulses are inefficient and difficult to scale, driving industry interest toward spin-transfer torque and spin-orbit torque magnetic random access memory. The new optimal control method steers magnetization smoothly by leveraging natural material dynamics rather than fighting them. Through computer simulations of three ultrathin van der Waals magnets namely Fe3GaTe2, Fe3GeTe2, and CrSBr, the research team proved that optimized pulses achieve ultrafast switching with magnetic fields over 10 times weaker than standard methods. Furthermore, by tuning properties like magnetic damping and anisotropy, energy consumption could eventually enter the femtojoule range, closely approaching the fundamental thermodynamic Landauer limit. This versatile mathematical framework is not limited to magnetic fields and can be adapted to electrical currents and ultrafast laser pulses used in future spintronics. Engineers can potentially combine multiple switching mechanisms to create highly energy-efficient hardware architectures, offering a vital pathway to curb the massive power footprint of future artificial intelligence systems and data centers.