Hybrid light-matter quasiparticles promise faster, cooler AI chips
Engineers at the University of Pennsylvania coupled photons with excitons inside a 2-D perovskite layer to create polaritons. These quasiparticles performed matrix operations at 0.3 femtojoules per MAC, beating electronic CMOS by a factor of 40 while maintaining coherence over 50 micrometers.
You begin viewing data movement, not transistor count, as the dominant energy cost. The technique encourages you to co-design algorithms with photonic hardware rather than treating accelerators as black boxes. This reframes your hardware roadmap from silicon-only to hybrid electro-optic pipelines.
The Photonics Foundry at Penn fabricated a 64-by-64 polariton array and demonstrated real-time image classification at 240 frames per second while drawing 12 milliwatts total system power.
Step 1: Download the simulation notebook from https://github.com/UPennPhotonics/polariton-sim. Step 2: Define a 32-by-32 weight matrix and run the included polariton propagation script on Google Colab with a T4 GPU. Step 3: Compare the reported energy per operation with a standard PyTorch matmul on the same matrix to quantify the efficiency gain.