George Washington University researchers created a new approach for AI computing that uses light instead of electricity. The method uses photons within neural network TPUs to improve the efficiency and speed of ML learning.
- Their paper was recently published in the journal Applied Physics Reviews.
- The photonic tensor core can conduct multiplications of matrices in parallel. It was able to able to perform two to three orders of magnitude higher than an electric TPU.
- Photonic specialized processors could boost response time, reduce data center traffic, and save a great deal of energy, paper co-author Mario Miscuglio said.