학술논문

Densely Parallelized Photonic Tensor Processor on Hybrid Waveguide/Free-Space-Optics
Document Type
Conference
Source
2023 International Conference on Photonics in Switching and Computing (PSC) Photonics in Switching and Computing (PSC), 2023 International Conference on. :1-3 Sep, 2023
Subject
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Photonics and Electrooptics
Performance evaluation
Transmission line matrix methods
Tensors
Optical computing
Benchmark testing
Optical fiber communication
Optical waveguides
optics
machine learning
neuromorphic device
optical waveguide
free space optics
Language
ISSN
2166-8892
Abstract
We demonstrate photonic tensor processor based on hybrid waveguide and free-space-optics. Our device can realize 32x100λ matrix operation thanks to the dense space/wavelength parallelization. By constructing the optical benchtop, we confirmed all matrix value can be controlled with good accuracy (>7 bit). Benchmark experiments for the standard neural network models showed good performance compatible to the simulated ones, suggesting the successful operation of our device. The device is considered to be able to perform 188 tera operations per second, which is one magnitude faster than previous approach. We also demonstrate the computation on the optical link using our device.