Artificial intelligence is transforming nearly every corner of modern life, but behind every generated image, internet search, recommendation and scientific simulation lies a rapidly expanding demand for energy. Researchers at the University of Edinburgh have developed a theoretical framework that could help reduce the electricity required to store and manipulate digital information in future magnetic memory technologies. Their approach uses mathematics to design magnetic switching processes that operate with dramatically less energy than conventional methods.
The scale of the challenge is enormous. Data centres already consume vast quantities of electricity to run servers, maintain cooling systems and move information across global networks. As artificial intelligence becomes embedded in healthcare, finance, science, manufacturing and everyday digital services, the amount of data being created and processed is expected to surge. Memory devices, which store the binary states underlying digital information, are a major part of this energy demand because billions or trillions of bits may be switched repeatedly during routine computing operations.
The Edinburgh team focused on magnetic memory, a technology in which information is encoded in the orientation of magnetisation. A magnetic element can represent a binary “0” or “1” depending on whether its magnetic state points in one direction or another. Writing data requires that state to be reversed, a process known as magnetic switching. In current devices, switching is often driven by electrical currents or magnetic fields that are not perfectly tailored to the material’s response, meaning that much of the supplied energy can be dissipated as heat.
To address this inefficiency, the researchers applied Optimal Control Theory, a mathematical method used to determine the most effective way to guide a system from one state to another. Instead of treating a magnetic pulse as a simple on-or-off signal, their framework calculates how the field should vary over time to produce the desired reversal while using as little energy as possible. The model can also incorporate realistic constraints, including limits on pulse strength, switching speed and the physical behaviour of the magnetic material.
In computer simulations, the optimised pulses produced striking results. The calculations suggested that switching energies could be reduced by several orders of magnitude compared with established memory technologies such as dynamic random-access memory, spin-transfer torque magnetic random-access memory and emerging spin-orbit torque magnetic random-access memory. These results do not represent a finished commercial device, but they indicate that the fundamental energy cost of magnetic information processing may be far lower than the cost associated with many existing engineering strategies.
The predicted performance is particularly significant because it approaches the Landauer limit, a fundamental thermodynamic boundary associated with irreversible information processing. The limit states that erasing one bit of information requires a minimum amount of energy proportional to temperature. At room temperature, that minimum is extremely small, but real-world devices operate far above it because of material imperfections, electrical resistance, unwanted heating and control inefficiencies. Moving closer to this limit could make future computing systems substantially more energy efficient, especially when scaled across vast data-centre infrastructures.
The researchers emphasise that the framework is not restricted to magnetic-field pulses. The same mathematical principles could be adapted to switching driven by electrical currents, including the mechanisms used in spintronic devices. It may also be useful for controlling ultrafast laser pulses, which are being investigated for advanced data-storage technologies capable of operating at extraordinary speeds. By changing the control signal while retaining the optimisation strategy, scientists could potentially apply the method to a range of emerging memory architectures rather than a single device design.
Dr Elton Santos of the University of Edinburgh, who led the research, said that every digital operation carries an energy cost and that this cost is becoming increasingly important as artificial intelligence and data-intensive technologies expand. He explained that carefully controlling how a magnetic field changes over time can allow magnetisation to switch much more efficiently than it does under conventional conditions. The work, he added, could represent “the next best thing” for energy-conscious information technology because its underlying mathematics can be transferred to multiple physical systems.
The study, published in Advanced Materials, also offers guidance for turning the theoretical concept into an experimentally testable technology. Future work will need to determine which magnetic materials, device geometries and field-delivery systems can reproduce the simulated performance under laboratory conditions. Researchers will also have to examine how thermal fluctuations, manufacturing imperfections and the need for reliable high-speed operation affect the energy savings. If those obstacles can be overcome, optimised magnetic switching could become an important component of lower-energy computing, helping data infrastructure keep pace with artificial intelligence without allowing its electricity demand to grow unchecked.
Subject of Research: Magnetic memory technologies and energy-efficient information processing
References: Advanced Materials
Image Credits: Dr Elton Santos, University of Edinburgh
Keywords
Artificial intelligence, magnetic memory, magnetic switching, Optimal Control Theory, energy-efficient computing, spintronics, Landauer limit, data centres, information technology, computational modelling
