A VCSEL BASED PHOTONIC NEUROMORPHIC PROCESSOR FOR EVENT-BASED IMAGING FLOW CYTOMETRY APPLICATIONS
UNIVERSITY OF THE AEGEAN & UNIVERSITY OF WEST ATTICA
M. Skontranis, G. Moustakas, A. Bogris, C. Mesaritakis
ABSTRACT
Τhis work introduces a novel scheme that merges a photonic neuromorphic spiking computing scheme with a bio-inspired event-based image sensor, aiming to combine real time processing of sparse image data with increased accuracy and lightweight back-end digital processing. The neuromorphic computing scheme consists of a time-delayed spiking extreme learning machine realized through a two-section laser, biased in an excitable regime. The application targeted consists of experimental data acquired through a high-flow imaging cytometry system, aiming to classify artificial particles of different size. Results achieved demonstrate that the proposed scheme can offer comparable performance to a lightweight digital neural network (accuracy of 97.1%) but with a reduced number of parameters by a factor of 6.25. These results highlight the merits of combining neuromorphic computing with event-based sensing; paving the way for fast, low-power technologies in fields like environmental monitoring, biomedical analysis, and smart sensing.
