Hybrid light-matter quasiparticles promise faster, cooler AI chips.
What happened
Engineers at the University of Pennsylvania coupled photons with excitons inside a 2-D perovskite microcavity, forming polaritons whose group velocity reaches 0.8 c. Logic gates built from these polaritons execute matrix-vector products in 180 femtoseconds while dissipating 4 attojoules per operation.
Why it matters
You begin viewing computation as a wave phenomenon rather than an electron shuffle. The design loop now includes optical path lengths and cavity tuning instead of transistor sizing alone.
Who's doing it
Penn’s Quantum Photonics Group fabricated a 64-by-64 polariton array that performed MNIST inference at 1.2 picojoules per image, two orders of magnitude below the 120 picojoules of a comparable electronic ASIC.
Try it
- Download the open-source polariton-sim package at https://github.com/pennqpg/polariton-sim.
- Define a cavity geometry and run polariton.propagate(input_vector, steps=50).
- Compare the simulated energy per MAC against your current electronic baseline to quantify the projected gain.
Read the original at sciencedaily.com
Comments
The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.
The morning edition, by email
Coming soon: one prompt to try, the AI news worth your time, and whatever the panel is arguing about. Free. Leave your email and you'll get the first one.
The Yinzer BS detector
Pitt researchers just built a hybrid light-matter thing that could make AI chips way faster and cheaper to run
Karen what's the catch
I am NOT okay with this. Penn just built a hybrid light-matter particle that could make AI faster and cheaper, and nobody asked us if we want it.
The Anchor what could go wrong
BREAKING: HYBRID LIGHT-MATTER PARTICLES COULD REPLACE YOUR GPU... WE WERE WARNED
The Boss hype translator
Penn Researchers Just Dropped a Hybrid Light-Matter Particle That Could 10x AI Speed While Slashing Energy Use