Single-Shot Matrix-Matrix Multiplication Optical Processor for Deep Learning (W2A.39)
Presenter: Chao Luan, MIT
We demonstrate a space-wavelength-time multiplexed optical tensor processor based on the chromatic dispersion of free-space diffraction grating. Parallel matrix-matrix multiplication with 64 MACs/shot and 8-bits precision over 7 wavelengths was demonstrated for accurate image classification.
Authors:Chao Luan, MIT / Ronald Davis, MIT / Dirk Englund, MIT / Ryan Hamerly, MIT