Neural Processing Letters
https://doi.org/10.1007/s11063-022-10814-9
A Parallel Reconfigurable Architecture for Scalable LVQ
Neural Networks
Marwa Gam
1,2
· Mohamed Boubaker
1
· Khaled Ben Khalifa
1,3
·
Mohamed Hedi Bedoui
1
Accepted: 4 April 2022
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022
Abstract
In this work, a new architecture for implementing LVQ neural networks is presented. The pro-
posed solution exploits the advantages of local parallelization in elementary neuro-processors
and partial dynamic reconfiguration to define the number of elementary processors used
according to context and requirements. The RP-LVQ architecture is generic, scalable and
partially and dynamically reconfigurable to ensure the adaptability of the LVQ network to
the application context while respecting the constraints by matching parallelism with the
architectures. As a validation of the RP-LVQ, we report the speed and power performance
of several LVQ network architectures’ implementation on a Xilinx zynq-7000 FPGA. For
a 1,024 neuron-topology in the hidden layer and a parallelism rate of β = 5, we achieved
10,070 million connections per second at a 100 MHz clock frequency with a 16.58 mw power
consumption. These promising results open up ways for the implementation of applications
where the flexibility and adaptability to the context or user are needed.
Keywords ANN · RP-LVQ · Scalable topology · Parallel architecture · Partial dynamic
reconfiguration · Adaptability
1 Introduction
Artificial Neural Networks (ANNs) are processing models that are inspired from the structure
and physiology of the human brain, aiming at mimicking its natural learning abilities. They are
typically made of computing units that are non-linear, parallel, distributed and simply layer-
interconnected. Parallelism, modularity and dynamic adaptation are generally associated with
B Mohamed Boubaker
boubaker.mohamed@gmail.com
1
Technology and Medical Imaging Laboratory, Faculty of Medicine Monastir, University of
Monastir, 5019 Monastir, Tunisia
2
National Engineering School of Sousse, University of Sousse, Erriyadh, BP 264, 4023 Sousse,
Tunisia
3
Institut Supérieur des Sciences Appliquées et de Technologie de Sousse, Université de Sousse,
4003 Sousse, Tunisia
123