Scientists in Australia have developed a tiny artificial intelligence chip that performs calculations using light instead of electricity.
The experimental device, created by researchers at University of Sydney, could help make future AI systems faster and far more energy efficient.
The prototype chip was built at the university’s Sydney Nano Hub and uses particles of light, known as photons, to process information.
Unlike traditional computer chips that rely on electrical signals, the new chip allows light to pass through specially designed nanostructures.
As the light moves through these structures, the chip automatically performs calculations.
The research was published in the journal Nature Communications.
Traditional computer processors move electrons through tiny circuits to process data. This process generates heat and requires significant energy, especially in large data centers that power modern artificial intelligence systems.
As AI technologies continue to expand, energy consumption has become a growing concern for the technology industry.
The new chip aims to solve part of that problem by replacing electricity with light. Because light can travel through materials with very little resistance, it produces much less heat and can operate extremely quickly.
Professor Xiaoke Yi, director of the Photonics Research Group at the University of Sydney, said the team reimagined how photonics could be used to design a new type of AI processor.
“Artificial intelligence is increasingly limited by energy consumption,” Yi explained. “By using light to perform neural computations, we can create processors that are faster, more energy efficient, and extremely compact.”
The chip’s computing elements are built from nanostructures that are only tens of micrometers wide, roughly the width of a human hair. These structures work together to form a neural network, a type of computing system designed to mimic how the human brain processes information.
When light travels through the network, the nanostructures guide and transform the light waves in ways that represent mathematical operations. This allows the chip to carry out machine-learning calculations almost instantly.
The processing happens on a picosecond timescale, meaning each calculation takes just trillionths of a second—the time it takes light to pass through the chip’s tiny structures.
To test the prototype, the researchers trained the photonic neural network to analyze more than 10,000 biomedical images, including MRI scans of the breast, chest, and abdomen. In both simulations and experiments, the chip achieved classification accuracy ranging from about 90 percent to 99 percent.
The technology highlights a potential path toward more sustainable AI infrastructure. Current data centers often consume large amounts of electricity and water to power and cool their equipment. Photonic computing could help reduce this environmental impact while supporting the growing demand for artificial intelligence.
Photonics, the science of controlling light particles, already plays a key role in many technologies such as fiber-optic communications, lasers, and medical imaging. Only recently have researchers begun exploring how photonics can also perform complex computing tasks.
Following the success of the prototype, the research team is now working to expand the technology into larger and more powerful photonic neural networks.
If the approach continues to improve, future AI hardware may process information not with electricity—but with light traveling at the fastest speed possible.
Source: University of Sydney.
