Photonic synaptic system for MAC operations by interconnected vertical cavity surface emitting lasers

We report experimentally on high-speed, tuneable photonic synaptic architectures realized with vertical cavity surface emitting lasers (VCSELs) connected in series and in parallel configurations. These are able to perform the controlled weighting of fast (150 ps long) and low energy ( µ W peak power...

Full description

Saved in:
Bibliographic Details
Published in:Optical materials express 2022-04, Vol.12 (4), p.1417
Main Authors: Robertson, Joshua, Alanis, Juan Arturo, Hejda, Matěj, Hurtado, Antonio
Format: Article
Language:English
Subjects:
Computer architecture
Data centers
High speed
Neural networks
Optical communication
Optical pulses
Photonics
Vertical cavity surface emission lasers
Weighting
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We report experimentally on high-speed, tuneable photonic synaptic architectures realized with vertical cavity surface emitting lasers (VCSELs) connected in series and in parallel configurations. These are able to perform the controlled weighting of fast (150 ps long) and low energy ( µ W peak power) optical pulses (or spikes), and permit high-speed (0.5 GHz) dynamic weight tunability, for the implementation of important spike processing functionalities. These include, for the in-series VCSEL synaptic architecture, the performance of accumulative weighting and, due to amplification, the compensation of losses in sequential neural network layers. Additionally, for the in-parallel VCSEL synaptic architecture, we show the system’s ability to perform key multiply and accumulate operations using fast, low-power optical spiking signals as inputs. Moreover, this work uses off-the-shelf VCSELs operating at key telecom wavelengths (1300 and 1550 nm) thus making our technique fully compatible with optical telecommunication networks and data centre technologies. These results therefore highlight the suitability of our approach for hardware-friendly, low power, high-speed and fast tuning VCSEL-based photonic synaptic architectures with excellent scalability prospects for use in future neuromorphic photonic computing systems.
ISSN:2159-3930
2159-3930
DOI:10.1364/OME.450923