Catenaa, Monday, December 1, 2025- Researchers at Aalto University have demonstrated single-shot tensor computing using light, enabling AI operations at the speed of light and bypassing traditional electronic limitations.
The breakthrough could accelerate future artificial intelligence hardware and reduce energy consumption.
Tensor operations, critical for AI tasks such as image recognition and language processing, traditionally require step-by-step computation on GPUs.
The Aalto team led by Dr. Yufeng Zhang developed a method that encodes digital information into the amplitude and phase of light waves.
As these structured light fields propagate, they perform matrix and tensor multiplications in a single pass, allowing multiple calculations to occur simultaneously.
By introducing multiple wavelengths, the team expanded the approach to support higher-order tensor operations.
The process is passive, requiring no active electronic switching, and can be adapted to existing optical platforms.
Professor Zhipei Sun said the method could be integrated onto photonic chips, producing low-power AI processors capable of complex calculations.
Zhang likened the method to inspecting and sorting multiple parcels simultaneously rather than one by one, emphasizing its parallel processing advantage.
The researchers believe this optical computing framework could be incorporated into major AI platforms within three to five years, creating a new generation of high-speed, energy-efficient computing systems.
