Tensor operations are a form of advanced mathematics that support many modern technologies, especially artificial intelligence. These operations go far beyond the simple calculations most people encounter. A helpful way to picture them is to imagine manipulating a Rubik's cube in several dimensions at once by rotating, slicing, or rearranging its layers. Humans and traditional computers must break these tasks into sequences, but light can perform all of them at the same time.

Today, tensor operations are essential for AI systems involved in image processing, language understanding, and countless other tasks. As the amount of data continues to grow, conventional digital hardware such as GPUs faces increasing strain in speed, energy use, and scalability.

Researchers Demonstrate Single-Shot Tensor Computing With Light

To address these challenges, an international team led by Dr. Yufeng Zhang from the Photonics Group at Aalto University's Department of Electronics and Nanoengineering has developed a fundamentally new approach. Their method allows complex tensor calculations to be completed within a single movement of light through an optical system. The process, described as single-shot tensor computing, functions at the speed of light.

"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. Zhang. "Instead of relying on electronic circuits, we use the physical properties of light to perform many computations simultaneously."

Encoding Information Into Light for High-Speed Computation

The team accomplished this by embedding digital information into the amplitude and phase of light waves, transforming numerical data into physical variations within the optical field. As these light waves interact, they automatically carry out mathematical procedures such as matrix and tensor multiplication, which form the basis of deep learning. By working with multiple wavelengths of light, the researchers expanded their technique to support even more complex, higher-order tensor operations.