Related Experiment Video
Updated: Sep 28, 2025
A Photonic System for Generating Unconditional Polarization-Entangled Photons Based on Multiple Quantum Interference
Published on: September 5, 2019
Programmable photonic neural networks combining WDM with coherent linear optics
Angelina Totovic1, George Giamougiannis2, Apostolos Tsakyridis2
1Department of Informatics, Center for Interdisciplinary Research and Innovation - CIRI, Aristotle University of Thessaloniki, Balkan Center - Building A, 10th Km Thessalonikis-Thermis Av, 57001, Thessaloníki, Greece. angelina@auth.gr.
This study introduces a novel neuromorphic photonics neuron architecture combining coherent optics and Wavelength-Division-Multiplexing (WDM) for multifunctional programmable neural networks. The platform supports four operational modes, demonstrating reliable performance for advanced AI applications.
Area of Science:
- Photonics
- Artificial Intelligence
- Computer Engineering
Background:
- Neuromorphic photonics traditionally uses coherent or Wavelength-Division-Multiplexing (WDM) for computations.
- Existing architectures have limitations in versatility and multifunctionality.
Purpose of the Study:
- To present a novel neuron architecture combining coherent optics and WDM.
- To develop a multifunctional, programmable neural network platform.
- To enable parallelization across fan-in and weighting stages.
Main Methods:
- Integration of coherent optics with WDM for enhanced parallelization.
- Development of a reconfigurable photonic hardware platform.
- Mathematical validation of performance across four operational modes.
Main Results:
- The proposed architecture supports four distinct operational modes: multi-layer, convolutional, fully-connected, and power-saving layers.
- Mathematical analysis confirms reliable operation considering crosstalk and spectral dependencies.
- Achieved a Matrix-Multiply Accumulate (MAC) relative error of [Formula: see text].
Conclusions:
- The combined coherent optics and WDM approach offers a versatile and programmable neuromorphic photonic platform.
- This architecture significantly advances the capabilities of photonic neural networks.
- The platform demonstrates potential for efficient and high-performance AI computation.
