AI at the speed of light has just become possible

Researchers at Aalto University have demonstrated single-shot tensor computing at the speed of light. This is a remarkable step toward next-generation general purpose artificial intelligence hardware that uses optical computing rather than electronics.

Tensor operations are a type of arithmetic operation that forms the backbone of almost all modern technology, especially artificial intelligence, but they go beyond the simple mathematics we are familiar with. Imagine the mathematics behind rotating, slicing, or rearranging a Rubik’s cube along multiple dimensions. Humans and classical computers have to perform these operations step by step, but light can perform these operations all at once.

Today, every AI task, from image recognition to natural language processing, relies on tensor operations. However, the explosion of data is pushing traditional digital computing platforms such as GPUs to their limits in terms of speed, scalability, and energy consumption.

How light enables instant tensor calculations

Motivated by this pressing problem, an international research collaboration led by Dr. Yufeng Zhang of the Photonics Group in the School of Electronics and Nanoengineering at Aalto University has revealed a new approach to perform complex tensor calculations using the propagation of a single light. The result is single-shot tensor computing achieved at the speed of light itself.

This research was published in Nature Photonics.

“Our method performs the same kinds of operations that today’s GPUs handle, such as convolutions and attention layers, but all at the speed of light,” says Dr. Zhang. “Instead of relying on electronic circuits, we use the physical properties of light to perform many calculations simultaneously.”

To accomplish this, the researchers encoded digital data into the amplitude and phase of light waves, effectively converting the numbers into physical properties of the light field. When these light fields interact and combine, they naturally perform mathematical operations such as matrix and tensor multiplication, which form the core of deep learning algorithms.

By introducing multiple wavelengths of light, the team extended this approach to handle even higher-order tensor operations.

Potential impact and future applications

“Imagine you are a customs official and you have to inspect every parcel with multiple machines with different functions and sort them into the appropriate boxes,” Zhang explains.

“Typically, we process each package one at a time. Our optical computing approach combines every package and every machine. We create multiple ‘optical hooks’ that connect each input to the correct output. With just one operation, one pass of light, all the inspection and sorting happens instantly and in parallel.”

Another important advantage of this method is its simplicity. Optical manipulation occurs passively as the light propagates, so no active control or electronic switching is required during the calculation.

“This approach can be implemented on almost any optical platform,” said Professor Zhipei Sun, leader of the Photonics Group at Aalto University. “In the future, we plan to integrate this computational framework directly into photonic chips, allowing light-based processors to perform complex AI tasks with very low power consumption.”

The ultimate goal is to bring the technique to existing hardware and established platforms from major companies, Zhang said, conservatively estimating that the approach will be integrated into such platforms within three to five years.

“This will usher in a new generation of optical computing systems that will greatly accelerate complex AI tasks across a myriad of fields,” he concludes.