Energy-efficient implementation of spiking recurrent cells on FPGA

energy efficiency; FPGA; hardware acceleration; neuromorphic computing; spiking recurrent cell; Cell model; Energy; Field programmables; Field-programmable gate array; Hardware acceleration; Neural-networks; Neuromorphic computing; Neuronal dynamics; Programmable gate array; Spiking recurrent cell; Electronic, Optical and Magnetic Materials; Hardware and Architecture; Artificial Intelligence; Electrical and Electronic Engineering

Abstract :

[en] Spiking neural networks (SNNs) can significantly reduce energy consumption compared to conventional artificial neural networks when spiking activity is sparse and the neuron model is hardware-friendly. However, biologically faithful models are often too costly for hardware implementations like field-programmable gate arrays (FPGAs), whereas highly simplified models, such as the leaky integrate-and-fire (LIF) neuron, sacrifice essential neuronal dynamics. In this work, we present an FPGA accelerator for an SNN utilizing the spiking recurrent cell (SRC) model, which offers an intermediate level of biological plausibility and hardware efficiency between simple LIF and complex conductance-based models. The SRC model features continuous spikes and critical dynamical properties like the refractory period, yet remains mathematically simple enough for efficient FPGA deployment. To optimize SRC computation, we propose a set of mathematical simplifications such as piecewise-defined approximations that eliminate costly non-linear functions ((Formula presented) (Formula presented), (Formula presented) (Formula presented)) and using scaling to avoid floating-point arithmetic. We then integrate these units into a cohesive hardware architecture using a dedicated binding layer to permit modularity and scalability of the network VHDL architecture. To demonstrate this modularity, two complete networks, with respectively 1 and 4 SRC layers, were implemented and validated using Spiking Traces (SpT) derived from, respectively, the MNIST and Fashion-MNIST datasets. Weight matrices computed offline are stored directly in LUT-registers without retraining or hardware-specific adaptation to strictly evaluate the robustness of SRC. The reference implementation achieves (Formula presented) (Formula presented) accuracy on MNIST with a (Formula presented) (Formula presented) -image SpT and a processing time of (Formula presented) (Formula presented) per digit. We further investigate accuracy-energy trade-offs by reducing the SpT length and quantizing synaptic weights down to (Formula presented) (Formula presented), achieving (Formula presented) (Formula presented) accuracy at (Formula presented) (Formula presented) per digit ((Formula presented) (Formula presented), (Formula presented) (Formula presented) weights) and (Formula presented) (Formula presented) at (Formula presented) (Formula presented) ((Formula presented) (Formula presented), (Formula presented) (Formula presented) weights). These results demonstrate that SRC-based SNNs deliver competitive performance and low energy consumption while preserving richer neuronal dynamics than standard LIF models.

Disciplines :

Electrical & electronics engineering

Author, co-author :

Harmeling, Pascal ; Université de Liège - ULiège > Département d'électricité, électronique et informatique (Institut Montefiore) > Capteurs et systèmes de mesures électriques

De Geeter, Florent ; Université de Liège - ULiège > Département d'électricité, électronique et informatique (Institut Montefiore) > Systèmes et modélisation

Drion, Guillaume ; Université de Liège - ULiège > Département d'électricité, électronique et informatique (Institut Montefiore) > Systèmes et modélisation

Language :

English

Title :

Energy-efficient implementation of spiking recurrent cells on FPGA

Publication date :

June 2026

Journal title :

Neuromorphic Computing and Engineering

eISSN :

2634-4386

Publisher :

Institute of Physics

Volume :

Issue :

Pages :

024022

Peer reviewed :

Peer Reviewed verified by ORBi

Additional URL :

https://iopscience.iop.org/article/10.1088/2634-4386/ae759c

Funders :

SPF BOSA - Service Public Fédéral Stratégie et Appui

Available on ORBi :

since 05 July 2026

Statistics

Number of views

69 (2 by ULiège)

Number of downloads

59 (2 by ULiège)

More statistics

Scopus citations^®

Scopus citations^®
without self-citations

OpenAlex citations

Bibliography

Strubell E, Ganesh A, McCallum A 2019 Energy and policy considerations for deep learning in NLP Proc. 57th Annual Meeting of the Association for Computational Linguistics pp 3645 50 pp 3645–50
Dampfhoffer M, Mesquida T, Valentian A, Anghel L 2022 Are SNNs really more energy-efficient than ANNs? An in-depth hardware-aware study IEEE Trans. Emerg. Top. Comput. Intell. 7 731 41 731–41 10.1109/TETCI.2022.3214509
Yan Z, Bai Z, Wong W-F2024Reconsidering the energy efficiency of spiking neural networks(arXiv:2409.08290)
Aliyev I, Lopez J, Adegbija T 2025 Exploring the sparsity-quantization interplay on a novel hybrid SNN event-driven architecture 2025 Design, Automation & Test in Europe Conf. (DATE) IEEE pp 1 7 pp 1–7
Abbott L F 1999 Lapicque’s introduction of the integrate-and-fire model neuron (1907) Brain Res. Bull. 50 303 4 303–4 10.1016/S0361-9230(99)00161-6
Feng J 2001 Is the integrate-and-fire model good enough?—A review Neural Netw. 14 955 75 955–75 10.1016/S0893-6080(01)00074-0
Hodgkin A L, Huxley A F 1952 A quantitative description of membrane current and its application to conduction and excitation in nerve J. Physiol. 117 500 10.1113/jphysiol.1952.sp004764
Skinner F K 2006 Conductance-based models Scholarpedia 1 1408 10.4249/scholarpedia.1408
Wang Q, Youjie Li, Shao B, Dey S, Peng Li 2017 Energy efficient parallel neuromorphic architectures with approximate arithmetic on fpga Neurocomputing 221 146 58 146–58 10.1016/j.neucom.2016.09.071
Modaresi F, Guthaus M, Eshraghian J K 2023 Openspike: an openram SNN accelerator 2023 IEEE Int. Symp. on Circuits and Systems (ISCAS) IEEE pp 1 5 pp 1–5
Li S, Zhang Z, Mao R, Xiao J, Chang L, Zhou J 2021 A fast and energy-efficient SNN processor with adaptive clock/event-driven computation scheme and online learning IEEE Trans. Circuits Syst. I 68 1543 52 1543–52 10.1109/TCSI.2021.3052885
Carpegna A, Savino A, Di Carlo S 2022 Spiker: an FPGA-optimized hardware accelerator for spiking neural networks 2022 IEEE Computer Society Annual Symp. on VLSI (ISVLSI) IEEE pp 14 19 pp 14–19
Carpegna A, Savino A, Di Carlo S 2024 Spiker+: a framework for the generation of efficient spiking neural networks FPGA accelerators for inference at the edge IEEE Trans. Emerg. Top. Comput. 13 784 98 784–98 10.1109/TETC.2024.3511676
Gupta S, Vyas A, Trivedi G 2020 FPGA implementation of simplified spiking neural network 2020 27th IEEE Int. Conf. on Electronics, Circuits and Systems (ICECS) IEEE pp 1 4 pp 1–4
Han J, Zhaolin Li, Zheng W, Zhang Y 2020 Hardware implementation of spiking neural networks on FPGA Tsinghua Sci. Technol. 25 479 86 479–86 10.26599/TST.2019.9010019
He Z, Shi C, Wang T, Wang Y, Tian M, Zhou X, Ping Li, Liu L, Nanjian W, Luo G 2021 A low-cost FPGA implementation of spiking extreme learning machine with on-chip reward-modulated STDP learning IEEE Trans. Circuits Syst. II: Express Briefs 69 1657 61 1657–61 10.1109/TCSII.2021.3117699
Li J, Shen G, Zhao D, Zhang Q, Zeng Y 2023 Firefly: a high-throughput hardware accelerator for spiking neural networks with efficient DSP and memory optimization IEEE Trans. Very Large Scale Integr. (VLSI) Syst. 31 1178 91 1178–91 10.1109/TVLSI.2023.3279349
De Geeter F, Ernst D, Drion G 2024 Spike-based computation using classical recurrent neural networks Neuromorph. Comput. Eng. 4 024007 10.1088/2634-4386/ad473b
Sepulchre R, Drion G, Franci A 2019 Control across scales by positive and negative feedback Ann. Rev. Control, Robot. Auton. Syst. 2 89 113 89–113 10.1146/annurev-control-053018-023708
Mendolia L, Wen C, Chicca E, Indiveri G, Sepulchre R, Redouté J-M, Franci A2025A neuromodulable current-mode silicon neuron for robust and adaptive neuromorphic systems(arXiv:2512.01133)
Panchapakesan S, Fang Z, Jian Li 2022 SyncNN: evaluating and accelerating spiking neural networks on FPGAs ACM Trans. Reconfigurable Technol. Syst. 15 1 27 1–27 10.1145/3514253
Roy A, Venkataramani S, Gala N, Sen S, Veezhinathan K, Raghunathan A 2017 A programmable event-driven architecture for evaluating spiking neural networks 2017 IEEE/ACM Int. Symp. on Low Power Electronics and Design (ISLPED) IEEE pp 1 6 pp 1–6
Jiadong W, Lun L, Wang Y, Zhiwei Li, Chen C, Qingjiang Li, Chen K 2025 Efficient spiking convolutional neural networks accelerator with multi-structure compatibility Front. Neurosci. 19 1662886 10.3389/fnins.2025.1662886
Koravuna S, Sanaullah S, Jungeblut T, Rückert U 2024 Spiking neural network models analysis on field programmable gate arrays Int. Conf. on Intelligent and Innovative Computing Applications pp 259 70 pp 259–70
Indiveri G 2025 Neuromorphic is dead. Long live neuromorphic Neuron 113 3311 4 3311–4 10.1016/j.neuron.2025.09.020
Xiao H, Rasul K, Vollgraf R 2017 Fashion-MNIST: a novel Image Dataset for Benchmarking Machine Learning Algorithms (arXiv: 1708.07747 [cs, stat])
Mirsadeghi M, Mirbagheri M, Kheradpisheh S R2026A multiplication-free spike-time learning algorithm and its efficient FPGA implementation for on-chip SNN training(arXiv:2604.23218)
Liang Z, Zhang X, Vousden M, Thomas D B, Bragg G M 2025 Quantitative simulations of spiking neural networks on an event-driven fpga cluster Integration 106 102590 10.1016/j.vlsi.2025.102590
Harmeling P2026Energy-Efficient Implementation of Spiking Recurrent Cells on FPGA Github(available at: https://github.com/Harmel-lab/SRC—SNN—01.git)