Scientists build AI chip that computes with light in trillionths

Researchers at the University of Sydney have developed a nanophotonic chip prototype that performs artificial intelligence (AI) computations using light (photons) instead of electricity (electrons). This innovative chip processes information at the picosecond scale—trillionths of a second—by guiding light through nanoscale structures that act as artificial neurons arranged in a neural network architecture. Unlike traditional electronic processors, which generate heat and require substantial energy for cooling due to electrical resistance, this photonic approach significantly reduces energy consumption and heat production, addressing a major challenge in scaling AI infrastructure. The team tested the prototype by training it to classify over 10,000 biomedical images, including MRI scans, achieving accuracy rates between 90% and 99%. These results demonstrate that neural network models can be physically embedded in nanoscale photonic structures rather than relying solely on software running on conventional silicon chips. The researchers aim to scale up the design to handle more complex datasets, potentially enabling photonic chips to complement or replace traditional