A significant portion of the energy consumed by modern electronics goes not into computation, but into writing and storing data. This is especially noticeable in data centers and artificial intelligence systems, where memory operations become one of the main consumers of electricity. Researchers from the University of Edinburgh have proposed a method that makes it possible to switch magnetic memory cells with much lower energy costs—by several orders of magnitude compared with existing technologies. The work was published in the journal Advanced Materials.

Why memory "guzzles" so much energy

In magnetic memory, a bit of information is written by flipping the magnetization of a tiny cell. This process has long been known, but it still remains quite wasteful: a significant part of the energy is dissipated as heat. For many years, scientists have been trying to approach the so-called Landauer limit—the theoretical minimum amount of energy that, according to the laws of thermodynamics, is required to process one bit of information. Modern technologies are still far from this limit.

How the ideal pulse was calculated

The team led by Elton Santos applied the mathematical framework of optimal control theory. Using it, the researchers calculated exactly what the shape of the magnetic pulse over time should be so that the magnetization switches as economically as possible.

It turned out that if, instead of simply applying a constant or crudely changing field, the law of its variation is carefully defined, energy losses drop sharply. According to the authors’ calculations, the new approach makes it possible to reduce energy consumption by several orders of magnitude compared with common types of memory—standard DRAM, as well as magnetoresistive STT-MRAM and SOT-MRAM.

“By carefully designing how the magnetic field changes over time, magnetization can be switched much more efficiently,” Elton Santos explained.

Versatility of the method

An important advantage of the approach is its versatility. The method works not only with an external magnetic field. The same principles of optimal control can be applied to electric current and ultrafast laser pulses. This means that the idea can be implemented in different types of devices and is not tied to one specific technology.

The authors believe that such solutions open the way to fundamentally more energy-efficient electronics. This is especially relevant against the backdrop of the rapid growth of data centers and artificial intelligence systems, where memory operations are becoming one of the main consumers of energy.