Penn team traps light and matter to accelerate matrix operations at lower power.

Engineers at the University of Pennsylvania coupled photons with excitons inside a micro-ring resonator, forming polaritons whose spin precession performs 4-by-4 matrix multiplies in 50 femtoseconds. A prototype chip executed a BERT layer at 2.3 picojoules per MAC, two orders of magnitude below an equivalent electronic systolic array. The device is fabricated in a standard silicon-photonics foundry process.

You begin to treat photons as first-class compute elements rather than mere data carriers, which alters how you partition workloads between electronic and optical stages in your inference pipeline.

The Penn Quantum Photonics Lab taped out a 64-polariton array in AIM Photonics MPW runs; they report 8 times throughput gains on transformer feed-forward layers while drawing 180 milliwatts versus 1.4 watts for a comparable GPU block.