Penn Researchers Demonstrate All-Light AI Switching at 4 Quadrillionths of a Joule
Updated
Updated · ScienceDaily · Jun 2
Penn Researchers Demonstrate All-Light AI Switching at 4 Quadrillionths of a Joule
3 articles · Updated · ScienceDaily · Jun 2
University of Pennsylvania researchers achieved all-light switching with exciton-polaritons, showing an AI-relevant computing step can run without converting optical signals back into electronics.
About 4 quadrillionths of a joule powered each switching event, a tiny energy cost aimed at easing the heat and resistance losses that increasingly limit electron-based AI hardware.
Exciton-polaritons combine photons with electrons in an atomically thin semiconductor, giving light enough interaction strength to perform the nonlinear switching operations ordinary photonic systems struggle to handle.
If scaled, the approach could enable photonic chips that process camera data directly, cut power use in large AI systems, and support some basic quantum-computing functions.
Will this light-particle breakthrough solve AI's energy crisis or just create new, unforeseen technological hurdles?
Is this the ultimate future for AI chips, or will quantum computing make this new technology obsolete before it scales?
If AI becomes nearly cost-free to run, what are the hidden societal dangers of such limitless computational power?
University of Pennsylvania Achieves 4-Femtojoule All-Optical AI Switch, Transforming Energy Efficiency in AI Hardware
Overview
University of Pennsylvania researchers have achieved a major breakthrough by demonstrating all-optical AI switching at a record-low energy level of 4 femtojoules. This innovative platform performs AI switching entirely with light, enabling direct processing of light signals from sources like cameras. By handling complex nonlinear operations within the optical domain and reducing the need to convert signals back to electrical form, the technology streamlines data flow and minimizes conversion losses. If scaled, this advance could lead to more efficient, powerful, and sustainable AI systems, marking a significant step forward in AI hardware development.