Photo credit: phys.org
Groundbreaking Advances in Photonic Computing Using Partial Coherence
Recent research from the University of Oxford, alongside its partners from the Universities of Muenster, Heidelberg, and Ghent, has unveiled a significant advancement in the field of photonic computing. Their study, titled “Partial coherence enhances parallelized photonic computing,” published in Nature, challenges the traditional reliance on high-quality lasers, suggesting that simpler light sources can enhance performance in applications including light-driven artificial intelligence (AI).
The revelation that low-coherence light sources, such as those from the sun and incandescent bulbs, may outperform lasers in specific contexts opens new avenues for more affordable and less energy-intensive technologies. Typically, light sources are evaluated based on their coherence, which describes the uniformity of light waves over time and space. While high-coherence sources, like lasers, provide precise and narrow wavelength ranges suitable for applications in optical communications and medical imaging, the new findings suggest that low-coherence sources could also be viable for high-order tasks in photonic computing.
Particularly, this study indicates that partially coherent light can serve remarkably well in photonic AI accelerators, where computations are executed using photons instead of electrons. The researchers utilized a specific configuration involving a narrow portion of incoherent light produced by an electrically pumped erbium-doped fiber amplifier, traditionally used in optical communications. This device allowed them to distribute partially coherent light across multiple input channels in a way that enhanced AI computational parallelism significantly.
In their experiments, the team successfully employed this innovative lighting system to identify individuals with Parkinson’s disease based on their walking patterns, achieving a remarkable classification accuracy exceeding 92%. Furthermore, the capability of a setup using a single partially coherent light source to perform high-speed AI computations at rates reaching 100 billion operations per second stands out as particularly impressive. This level of performance, which would typically require several discrete lasers, exemplifies how a singular, simpler light source can match or even surpass existing technologies.
Lead researcher Dr. Bowei Dong notes the transformative potential of this approach, stating, “The benefit of using ‘poorer’ light sources has a scaling effect. You can run your AI models 100 times faster compared to a laser system, if the photonic accelerator scales to 100 input channels.”
Professor Harish Bhaskaran, who oversaw the project and is also a co-founder of Salience Labs, emphasized the broader implications of their work, stating, “While this work showcases the use of such partially coherent light in some emerging areas of photonic computing, we will in future also investigate whether this insight might apply to optical communications, particularly in the emerging optical interconnect technology space. This is an area of research that is moving rapidly with a lot of interesting science and engineering to explore.”
More information: Bowei Dong et al, Partial coherence enhances parallelized photonic computing, Nature (2024). DOI: 10.1038/s41586-024-07590-y
Citation: New ‘game-changing’ discovery for light-driven artificial intelligence (2024, August 2) retrieved 2 August 2024 from https://phys.org/news/2024-08-game-discovery-driven-artificial-intelligence.html
This document is provided for informational purposes only and is subject to copyright. Reproduction requires written permission unless for fair dealing related to private study or research.
Source
phys.org