Penn Researchers Build Light-Matter Particles to Cut AI Power Draw

University of Pennsylvania physicists created hybrid polaritons that combine photons and excitons inside a microcavity. These particles performed matrix multiplications at 0.3 femtojoules per operation, roughly 100 times lower energy than current electronic tensor cores.

Shifting select linear algebra steps from electrons to photons changes the energy budget calculation for large training runs. Engineers must now evaluate whether adding photonic interconnects yields lower total watt-hours than simply adding more GPUs.

The Penn Excitonics Lab published results in Nature Photonics showing a prototype accelerator running a 3-layer network at 1.2 tera-operations per second while drawing 45 milliwatts. The device remains a bench-top experiment.