Oxford Light-Driven AI Revolution: Partial Coherence Boosts Photonic Computing Performance by 100x

University of Oxford researchers achieved a paradigm-shifting discovery in August 2024, demonstrating that "poorer" light sources can actually enhance performance in photonic AI applications, challenging decades of conventional optical engineering wisdom.

Published in Nature, the research shows that partially coherent light sources can outperform expensive, high-spec lasers in specific applications such as photonic AI accelerators - emerging technology where photons replace electrons for AI computations.

The team used an electrically-pumped erbium-doped fiber amplifier, distributing its partially coherent light across multiple input channels for parallel AI computation, achieving unprecedented scaling benefits.

Testing demonstrated remarkable improvements:

"You can run your AI models 100 times faster compared to a laser system, if the photonic accelerator scales to 100 input channels," explained first author Dr. Bowei Dong.

This breakthrough could prove transformational by removing the need for multiple expensive coherent lasers, dramatically reducing costs while boosting computational power. The discovery opens pathways for cheaper, less energy-intensive light sources in applications traditionally requiring high-spec lasers.

Professor Harish Bhaskaran, who led the work, noted potential applications extending beyond photonic computing to optical communications and emerging optical interconnect technologies, representing a rapidly moving field with exciting science and engineering opportunities.