Can Chip ‘Waste’ Make AI Faster and Greener? Korean Team Says Yes

Scientists at the Korea Institute of Science and Technology (KIST) have flipped a key assumption in AI chip development by harnessing what was once considered wasted energy.

Electron spin loss, long considered an efficiency drain in magnetic devices, has been shown to drive magnetization switching with up to three times better energy performance.

This breakthrough undermines decades of research that focused on minimizing spin loss in spintronic devices. Instead of fighting against it, researchers led by Dr. Dong-Soo Han found that spin loss can be used as an energy source, creating a new way forward for low-power AI hardware.

How spin loss boosts energy efficiency in AI chips

Spintronics technology uses electron spin to process and store information; however, traditional designs require large electrical currents that waste significant energy due to spin loss before the currents reach their destination.

Han’s team demonstrated experimentally that increasing spin loss paradoxically reduces the power required for magnetization switching.

The phenomenon works similarly to a balloon moving when air is let out, triggering spontaneous magnetization flips within magnetic materials without needing external energy stimuli.

This self-switching effect lets magnetic states switch between digital "1" and "0" while using drastically less power.

Did you know?
Contrary to conventional thought, increasing electron spin loss can reduce power consumption in magnetic switching, a key step in digital data storage.

Market potential and industry applications

The global spintronics market is expected to skyrocket from $2.20 billion in 2025 to $40.26 billion by 2034, led by demands for ultra-efficient AI memory and neuromorphic computing.

The simplicity of the new method, which integrates smoothly with existing semiconductor manufacturing, makes it well-suited for rapid commercial adoption.

Applications include AI semiconductors, edge computing devices, and ultra-low-power memory systems. Unlike previous approaches requiring complex materials, this technique’s compatibility with current chip processes paves the way for major industry disruption.

ALSO READ | A Magnet That Bends Light? Japan’s New Discovery Explained

A paradigm shift in spintronics research

Until now, spintronics focused solely on reducing spin loss. Han explained that using spin loss as an energy driver redefines device design and efficiency goals.

The team’s paper in Nature Communications outlines this shift and its implications for computing technology.

By tapping into spin loss energy, the researchers open avenues to develop ultra-small, low-power AI chips that can surmount the thermal and physical limits of existing AI hardware architectures, helping move AI beyond data centers onto mobile and IoT devices.

Toward a greener AI future

This discovery addresses the critical need to cut power consumption in AI hardware, especially as AI chip demand reaches a $40.79 billion market in 2025. Lower energy use means less heat generation, longer device lifespans, and the ability to deploy intelligent computing in a much wider range of environments.

The research not only challenges conventional approaches but also offers an elegant, practical path to energy-efficient AI computing, promising faster, greener, and more accessible AI technologies worldwide.