Digital Implementation of On-Chip Hebbian Learning for Oscillatory Neural Network

Edgar Luhulima, Madeleine Abernot, Federico Corradi, Aida Todri-Sanial

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

2 Downloads (Pure)


This work proposes a digital implementation of an Oscillatory Neural Network (ONN) in a Field-Programmable Gate Array (FPGA), demonstrating excellent associative memory capabilities. This work goes beyond previous implementations by enabling on-chip learning directly in the FPGA. More specifically, we implement on-chip Hebbian learning, and we compare three different design strategies. The first strategy takes advantage of a System-on-Chip (SoC) composed of a Processing System (PS) and Programmable Logic resources (PL) to integrate Hebbian learning in PS. The two other strategies implement the Hebbian learning directly in PL. We compare the three different design strategies on a digit recognition task in terms of accuracy, utilization, execution time, and maximum frequency. We show that implementing Hebbian learning in PL gives more advantages in terms of resource utilization and latency than implementing Hebbian in PS with several orders of magnitude because the weight matrix computation is performed in hardware. Moreover, we develop an application interface to demonstrate the pattern learning and recognition capabilities of our digital ONN implementation.
Originele taal-2Engels
Titel2023 IEEE/ACM International Symposium on Low Power Electronics and Design, ISLPED 2023
UitgeverijInstitute of Electrical and Electronics Engineers
Aantal pagina's6
ISBN van elektronische versie979-8-3503-1175-4
StatusGepubliceerd - 2023


Duik in de onderzoeksthema's van 'Digital Implementation of On-Chip Hebbian Learning for Oscillatory Neural Network'. Samen vormen ze een unieke vingerafdruk.

Citeer dit