As artificial intelligence continues to expand, the world’s data centers are using enormous amounts of electricity.
Some large AI facilities already consume as much power as small cities, raising concerns about energy costs and environmental impact.
Scientists are now searching for new types of computer memory that can process and store data using far less energy.
Researchers from POSTECH and Chungnam National University may have found an important breakthrough.
They have developed a new memory technology that stores information using tiny temperature changes instead of large electric currents, dramatically reducing energy use.
Their study was published in the journal Advanced Functional Materials.
The new system is based on a field called spintronics. Traditional electronics store and process information using the electrical charge of electrons. Spintronics works differently. It uses a property called “spin,” which can be thought of as the tiny magnetic direction of an electron.
In spintronic devices, different spin directions represent the binary numbers 0 and 1 used in digital computing.
Spintronics has attracted attention because it could allow faster and more energy-efficient computing. However, most existing spintronic memory systems require strong electric currents to switch spin directions. Those currents generate heat and waste a large amount of energy.
The research team wanted to avoid this problem by controlling spin using temperature instead of electricity.
Previous attempts at temperature-based switching faced a major challenge. When the temperature returned to normal, the spin direction often switched back as well, meaning the memory could not reliably store information over time.
The researchers solved this issue using a phenomenon called thermal hysteresis. In simple terms, this means a system can remain in a changed state even after the original trigger is removed.
To create this effect, the team built a structure using two magnetic materials called gadolinium iron garnet and holmium iron garnet. These materials respond differently to temperature changes, creating a kind of magnetic competition between them.
The scientists compare the process to a tug-of-war game. Each material pulls the system’s magnetic direction in a different way. As the temperature changes, one side becomes stronger and eventually pulls the system into a new stable state.
Importantly, the system does not immediately switch back when the temperature changes again. Instead, the magnetic state remains stable, allowing the memory to store information without continuous power. This is known as non-volatile memory.
The researchers successfully switched the memory states using only small temperature changes of about 25 degrees Kelvin and a relatively weak magnetic field.
Compared with current spin–orbit torque memory technologies, the new method reduced energy use by up to 66 times. Under ideal conditions, the reduction could reach as much as 452 times.
Professor Hyungyu Jin said the work shows that memory devices may someday operate using tiny thermal changes rather than large electrical currents.
Although the technology is still in the research stage, it could eventually help create ultra-low-power memory systems for AI computers, mobile devices, and future data centers, reducing both electricity costs and environmental impact.
