Penn Researchers Demonstrate 4-Quadrillionth-Joule All-Light AI Switch
Updated
Updated · ScienceDaily · May 22
Penn Researchers Demonstrate 4-Quadrillionth-Joule All-Light AI Switch
4 articles · Updated · ScienceDaily · May 22
University of Pennsylvania researchers built an all-light switching device for AI computing using exciton-polaritons, cutting the energy per operation to about 4 quadrillionths of a joule.
The switch works by coupling photons with electrons in an atomically thin semiconductor, creating a hybrid light-matter particle that can perform the nonlinear signal switching ordinary light struggles to handle.
That matters because current photonic AI chips often send light signals back into electronics for activation and decision steps, adding delay and eroding the energy advantage of optical computing.
If scaled, the approach could enable photonic chips to process camera data directly with fewer light-electricity conversions, lowering AI power demand and potentially supporting basic quantum-computing functions.
As computing moves from electrons to light, what new abilities beyond speed will be unlocked?
Can AI’s energy crisis be solved by a particle that only works near absolute zero?
Record-Breaking 4-Femtojoule All-Light AI Switch: A Leap Toward Ultra-Efficient, Sustainable Computing
Overview
Physicists at the University of Pennsylvania have achieved a major breakthrough by developing an all-light AI switch that operates at a record-low energy consumption of just 4 femtojoules. This switch uses light instead of electrons, allowing it to process information much faster and with significantly less heat. These improvements are critical for advanced AI applications, as they address the main challenge of high energy demand in scaling AI systems. By leveraging the exciton-polariton system, this innovation lays the groundwork for ultra-efficient, high-speed AI hardware, marking a crucial step toward the future of sustainable and powerful artificial intelligence.