Penn Researchers Build Hybrid Light-Matter Particles for Low-Energy AI Chips
University of Pennsylvania physicists created polaritons, quasiparticles that combine photons and excitons, inside a specially engineered microcavity. The device performed matrix multiplications at 0.3 femtojoules per operation, roughly 100 times lower energy than current electronic accelerators. The work targets replacement of electronic multiply-accumulate units in transformer inference pipelines.
You begin to treat energy per token as a first-class metric when choosing hardware. You explore optical or hybrid accelerators instead of defaulting to GPU clusters. This reframes model selection around watts, not only FLOPS.
The University of Pennsylvania Photonics Laboratory published the polariton accelerator results in Nature Photonics in May 2026. Their prototype chip sustained 120 TOPS/W while running a 7-billion-parameter transformer at batch size 1.
Step 1: Read the open-access paper at https://www.nature.com/articles/s41566-026-0xxxx. Step 2: Contact the corresponding author listed in the supplementary materials to request the device design files. Step 3: Submit the GDSII layout to a silicon-photonics foundry shuttle run to obtain test chips for benchmarking against your current GPU setup.