As artificial intelligence becomes more powerful and widely used, the amount of data being created every day is growing at an extraordinary pace.
Every AI chatbot conversation, online search, streamed video, scientific simulation and recommendation system relies on enormous amounts of information being stored, processed and transferred.
This growing demand is placing increasing pressure on data centers, which already consume huge amounts of electricity.
Researchers at the University of Edinburgh have now developed a new theoretical approach that could dramatically reduce the energy needed to store and manipulate digital information.
Their work, published in Advanced Materials, could help pave the way for future computer memory that is far more energy efficient than today’s technology.
Modern computers store information as tiny units called bits, which represent either a 0 or a 1. In many types of memory, changing a bit from one state to another requires switching the direction of magnetization inside microscopic magnetic materials.
Although this process happens incredibly quickly, it still consumes energy. When multiplied across billions or even trillions of operations every day, that energy use becomes significant.
As artificial intelligence continues to expand, experts expect global electricity demand from information and communication technologies to rise sharply.
Without major improvements in efficiency, computer systems and data centers could eventually account for a much larger share of the world’s electricity consumption and carbon emissions.
Instead of using conventional methods to switch magnetic memory, the Edinburgh researchers turned to a branch of mathematics known as optimal control theory. This mathematical technique is designed to find the most efficient way to achieve a particular goal while working within real-world limitations.
Using this approach, the team developed a framework that calculates the ideal sequence of ultrafast magnetic field pulses needed to switch magnetic memory while using the smallest possible amount of energy. Rather than simply applying a magnetic field in the usual way, the method carefully adjusts how the field changes over time to achieve maximum efficiency.
Computer simulations produced striking results.
The researchers found that their optimized switching method could reduce energy consumption by several orders of magnitude compared with some of today’s most advanced memory technologies, including DRAM, spin-transfer torque magnetic random-access memory (STT-MRAM) and the newer spin-orbit torque MRAM (SOT-MRAM).
Perhaps even more importantly, the predicted energy requirements approach what physicists call the Landauer limit. This is the fundamental theoretical minimum amount of energy required to process a single bit of information according to the laws of thermodynamics. While no practical technology can completely eliminate energy use, getting closer to this limit represents an important milestone for computing efficiency.
The researchers also designed the framework with real-world applications in mind. Their study provides practical guidance for future device designs and methods of delivering magnetic field pulses, making it easier for other scientists to test the concept experimentally.
Although the research initially focused on magnetic fields, the mathematical framework is much more flexible. The same principles could also be applied to electrical currents or even ultrafast laser pulses, two technologies already being explored for next-generation data storage.
While the work remains theoretical, it offers an exciting glimpse of how future computer memory could become dramatically more energy efficient.
As artificial intelligence and digital technologies continue to expand, innovations like this may help reduce electricity consumption, lower operating costs and lessen the environmental impact of the world’s rapidly growing demand for computing power.
