Photonic spiking reservoir computing based on a single passive silicon microring and time-wavelength multiplexing

Tools

Donati, Giovanni and Biasi, Stefano and Lugnan, Alessio and Pavesi, Lorenzo and Hurtado, Antonio; (2025) Photonic spiking reservoir computing based on a single passive silicon microring and time-wavelength multiplexing. In: 2025 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC). 2025 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC) . IEEE, DEU. ISBN 979-8-3315-1252-1 (https://doi.org/10.1109/cleo/europe-eqec65582.2025...)

[thumbnail of Donati-etal-2025-Photonic spiking-reservoir-computing]
Preview
 Text. Filename: Donati-etal-2025-Photonic_spiking-reservoir-computing.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (405kB)| Preview

Abstract

Spiking artificial neural networks leverage both analog (spike firing time) and digital (spike or quiescent state) processing modalities to efficiently process information through spike sparsity and are widely explored in electronic hardware [1]. In this work, we introduce a novel framework for all-optical and sparse spiking processing utilizing a passive chip-integrated add-drop silicon microring resonator (MRR). We demonstrate how optical pulse perturbations, when properly calibrated in temporal duration and wavelength detuning relative to the MRR's resonance, can elicit deterministic spiking events followed by a refractory period, without requiring a pump-and-probe approach. Leveraging this finding, we show multi-wavelength neuromorphic applications, such as coincidence detection and a novel framework for sparse spiking reservoir computing with MRRs, inspired from [2]. The spiking MRR reservoir is benchmarked on the Iris flower classification task, encoding the flower datapoint's features onto two input wavelength carriers (at 1545.37 nm and 1558.782 nm, respectively) to double the processing speed. A classification accuracy of 92% is achieved, using only 48 reservoir nodes and an average of just three optical spikes fired per flower datapoint. The simplicity of the proposed photonic spiking reservoir computing approach shows promise for future studies involving on-chip coupled MRR structures, where more complex spiking dynamics can be harnessed [3]. Furthermore, it is compatible with the implementation of binary output weights, simplifying the hardware required to train the reservoir's output layer; thus, offering exciting prospects for simplified learning protocols directly implemented in optical hardware.

ORCID iDs

Donati, Giovanni ORCID logoORCID: https://orcid.org/0000-0001-6156-8394, Biasi, Stefano, Lugnan, Alessio, Pavesi, Lorenzo and Hurtado, Antonio ORCID logoORCID: https://orcid.org/0000-0002-4448-9034;
  • Item type:Book Section
    ID code:93910
    Dates:
    Date
    Event
    15 August 2025
    Published
    Subjects:Science > Physics
    Department:Faculty of Science > Physics
    Depositing user: Pure Administrator
    Date deposited:25 Aug 2025 10:20
    Last modified:18 Nov 2025 18:06
    URI:https://strathprints.strath.ac.uk/id/eprint/93910
CORE (COnnecting REpositories)