At the University of Tokyo, a lab device is trading electric current for electron spin, flipping states in about 40 picoseconds and pointing to computers that run up to 1,000 times faster while shedding heat and power draw. Built from tantalum and manganese-tin, it turns electrical signals into tiny magnetic orientations that can be written with negligible energy and have already survived billions of test cycles. The work, published in Science, hints at data processing that is roughly 100 times more energy efficient. A prototype is targeted by 2030, though turning this quantum switch into commercial hardware remains the hard part.
A computing breakthrough from Tokyo
We have been told for years that faster chips mean hotter servers and bigger power bills. A team at the University of Tokyo is challenging that trade-off. Their researchers built a device that, in lab tests, switches information up to 1,000 times faster than today’s standard approaches, while using dramatically less energy and shedding less heat. If it holds up, the ripple effects could reach cloud, AI, and edge devices alike.
A leap with quantum switching
The core idea pivots from moving electrons to steering their spin, a domain often called spintronics. Instead of relying on current flowing through transistors, the device flips magnetic orientations to represent bits. The team reports bit operations in 40 picoseconds, compared with nanoseconds for mainstream chips. Faster switching and fewer joules lost to heat is a compelling combo for data centers straining under AI workloads.
The materials making it possible
The prototype stacks two materials with complementary roles: tantalum handles the electrical-to-spin conversion, while manganese-tin captures those spins as stable magnetic states. This tandem translates a fleeting electrical nudge into a durable orientation that stores or processes data. The group says the structure stayed stable across billions of cycles in stress tests, a promising sign for reliability at scale.
From the lab to the real world
The findings were published in Science, and the researchers are targeting a functional chip prototype by 2030. That timeline reflects tough engineering hurdles: patterning at nanoscales, integrating with CMOS processes, and proving yields that would satisfy major fabs. Still, durability results are notable. Traditional designs often falter from heat after hundreds of thousands to a million cycles, yet this approach reportedly kept going without failure.
Shaping the next era of computing
Lead investigator Professor Satoshi Nakatsuji says the method could slash information-processing energy by roughly 100 times. That kind of reduction would matter for US operators chasing lower power usage effectiveness, and for chipmakers like Intel, AMD, and Nvidia that face cooling limits as models grow. Could spin-based switching shoulder some of the load that transistors cannot? The research does not replace the CPU or GPU overnight, but it sketches a practical path to specialized accelerators. If the integration challenges give way, we could see new classes of storage-logic hybrids that crunch data with far less heat, opening headroom for speed without the familiar thermal penalty.
Source : https://www.nikkei.com/article/DGXZQOSG132XK0T10C26A5000000/
