Researchers have successfully demonstrated a method to execute complex artificial intelligence computations in a single shot using light waves, potentially overcoming the speed and energy limitations of conventional electronic hardware.
An international collaboration led by Aalto University has developed “single-shot tensor computing” to replace the sequential operations of standard digital platforms. By encoding digital data into the amplitude and phase of light, the system performs the matrix multiplications that underpin deep learning algorithms passively as the light propagates.
“Our method performs the same kinds of operations that today’s GPUs handle, like convolutions and attention layers, but does them all at the speed of light,” says Dr Yufeng Zhang from the Photonics Group at Aalto University. “Instead of relying on electronic circuits, we use the physical properties of light to perform many computations simultaneously.”
Passive computation
The technology avoids the need for active electronic switching, allowing for massive parallelism where inputs are connected to outputs instantly via “optical hooks”. The researchers extended the approach to handle higher-order operations by introducing multiple wavelengths of light.
“In the future, we plan to integrate this computational framework directly onto photonic chips, enabling light-based processors to perform complex AI tasks with extremely low power consumption,” says Professor Zhipei Sun, leader of the Photonics Group.
The team estimates the method could be integrated into existing major industry platforms within three to five years.
