University of Tokyo develops magnetic switch 1,000 times faster than today's AI chips
Researchers at the University of Tokyo have built a magnetic switching device that operates at picosecond speeds-up to 1,000 times faster than the fastest AI accelerators available today-while using a fraction of the energy and generating minimal heat.
The work, published in Science this week, demonstrates a new method for flipping a binary magnetic state at picosecond speeds. That represents a massive jump from the nanosecond-scale switching that defines modern silicon-based processors. One picosecond equals one trillionth of a second.
The heat problem
Processor speed and heat generation are directly linked. Faster chips produce more heat, which drives up power consumption and strains data center infrastructure. The Tokyo team says their approach solves this fundamental constraint.
The researchers built a spintronic device using Mn3Sn, an antiferromagnetic compound made of manganese and tin. Unlike traditional semiconductors that rely solely on electron charge, spintronic devices use both charge and spin to process, store, and transmit data.
In their proof of concept, a 40-picosecond electrical pulse flipped the device's magnetic state while generating minimal resistive heat. Energy consumption was a fraction of what modern AI accelerators require.
What this means in practice
The speed improvement doesn't translate directly to overall computing speed. A computer is more than a switch-it depends on many hardware and software components working together to read, process, and move data. A faster switch alone has limits.
But the efficiency gains could matter for specific applications. Professor Tomo Nakatsuji of the University of Tokyo said the technology could eventually enable cloud-based quantum services and optical quantum computing for general users. He also suggested that data downloads currently taking an hour could be processed in one second.
The work builds on earlier research published in Nature in January 2025. The main challenge now is moving the technology from research labs to commercial manufacturing at scale and cost.
For professionals working in AI for Science & Research, understanding spintronic approaches and their efficiency advantages is increasingly relevant as hardware constraints shape what's possible in the field.
Your membership also unlocks:
