Penn Researchers Achieve 4-Femtojoule Optical Switching With Hybrid Particles for AI Computing
Updated
Updated · Interesting Engineering · May 16
Penn Researchers Achieve 4-Femtojoule Optical Switching With Hybrid Particles for AI Computing
4 articles · Updated · Interesting Engineering · May 16
University of Pennsylvania researchers built a nanocavity device that creates exciton-polaritons—hybrid light-matter quasiparticles—and used it to switch one optical signal with another at about 4 femtojoules.
That matters because photons move fast and waste little energy but normally do not interact enough for computing; the hybrid particles add the nonlinear response needed for logic without converting signals back into electricity.
The proof-of-concept could cut power use and heat in AI hardware, where electronic chips consume heavy electricity and require increasingly aggressive cooling in dense data centers.
The result, published in Physical Review Letters, sets a benchmark for 2D exciton-polariton switching, though scaling the lab device into reliable, large-scale optical computers remains an open engineering challenge.
As AI's energy demand soars, can a new light-based chip prevent a global power crisis before 2030?
How will computers that 'think' at the speed of light reshape society beyond just saving energy?
4-Femtojoule Exciton-Polariton Switches: Transforming AI Hardware and Quantum Photonics
Overview
In May 2026, researchers at the University of Pennsylvania, led by Bo Zhen, announced a major breakthrough in ultra-efficient optical switching using exciton-polaritons. These unique hybrid particles combine the speed of light with the interactive power of matter, blending the low dephasing of photonic systems with strong particle interactions. By precisely coupling light into nanoscale cavities and interacting with atomically thin materials, the team established strong light-matter coupling. This innovation promises to move beyond the limits of traditional electronics, paving the way for faster, more energy-efficient data processing and a new era in computing technology.