AI's Energy Use Hacked by New Physics

Artificial intelligence and other digital technologies are creating and processing unprecedented amounts of data, leading to a significant increase in computing power and data storage needs. This expansion presents a major challenge, as data centers already consume substantial electricity, with projected increases in the coming decades. Failing to improve efficiency could make information and communication technologies a large contributor to global energy consumption and carbon emissions.

To address this, scientists at the University of Edinburgh have devised a novel theoretical framework. This approach focuses on optimizing the energy used to store and manipulate digital information in future magnetic memory systems. By employing Optimal Control Theory, a mathematical method for finding the most efficient processes, the research team has designed ultrafast magnetic-field pulses.

These optimized pulses can switch magnetic states with minimal energy expenditure. Computer simulations indicate this new method could reduce switching energy by several orders of magnitude compared to current leading memory technologies like DRAM and MRAM. The predicted energy requirements bring future magnetic memory significantly closer to the Landauer limit, the theoretical minimum energy needed for processing a single bit of information.

The framework extends beyond pure theory, offering practical implementation guidance including optimized device designs and magnetic field delivery methods. Dr. Elton Santos, who led the research, highlighted the versatility of the mathematical approach, suggesting it could be adapted for electrical currents and ultrafast laser pulses, potentially revolutionizing energy efficiency across various cutting-edge data storage technologies.