Scientists at Aalto University (Finland) have developed a revolutionary technique that uses a single beam of light to perform AI calculations at supercomputer speeds, introducing a new class of ultra-fast, energy-efficient hardware.
The innovation encodes data into light properties — such as amplitude and phase — enabling complex tensor operations to occur passively as light travels, rather than step-by-step like in traditional systems.
Instant Processing, No Electronics 🔬⚡
Unlike electronic systems, the method is fully passive: the physics of light does the work all at once, with no need for stepwise control. The researchers successfully demonstrated convolution and attention operations — the building blocks of neural networks — using multiple wavelengths of light interacting in parallel.
Dr. Yufeng Zhang, lead author, explained: “Our method performs the same operations as today’s GPUs, but all at once and at the speed of light.”
The system can be implemented across various photonic platforms. Aalto's team is now working to integrate it into on-chip photonic processors, enabling AI acceleration with radically reduced power consumption — potentially deployable in 3–5 years.
📅 Date: November 16, 2025
🔗 Source: ScienceDaily
In short, this photonic computing method offers a path to instant AI computation with minimal energy use, revolutionizing how neural networks process data. It could power everything from real-time image recognition to large language models — at light speed.