Related Experiment Video
Updated: Jul 29, 2025
07:56
A Photonic System for Generating Unconditional Polarization-Entangled Photons Based on Multiple Quantum Interference
Published on: September 5, 2019
8.5K
Experimental demonstration of a photonic convolutional accelerator based on a monolithically integrated
Optics Letters
|May 23, 2023
Summary
We developed an energy-efficient photonic convolutional accelerator using a novel semiconductor laser. This device achieves high-speed real-time image recognition for handwritten digits with impressive accuracy.
Area of Science:
- Optoelectronics
- Artificial Intelligence Hardware
- Integrated Photonics
Background:
- Convolutional neural networks (CNNs) are crucial for image recognition but computationally intensive.
- Photonic approaches offer potential for energy-efficient and high-speed computation.
Purpose of the Study:
- To propose and demonstrate a simple, energy-efficient photonic convolutional accelerator.
- To leverage integrated semiconductor lasers for neural network acceleration.
Main Methods:
- Utilized a monolithically integrated multi-wavelength distributed feedback semiconductor laser.
- Employed a superimposed sampled Bragg grating structure for laser design.
- Experimentally validated the accelerator's performance on real-time recognition tasks.
Main Results:
- Achieved an operational speed of 44.48 GOPS for a 2x2 kernel.
- Successfully generated 100 images for real-time recognition.
- Demonstrated 84% prediction accuracy on the MNIST handwritten digit database.
Conclusions:
- The proposed photonic convolutional accelerator is compact and low-cost.
- This technology enables efficient realization of photonic convolutional neural networks.
- Offers a promising pathway for high-performance, energy-saving AI hardware.
