FrostRune: An Asymmetric Translational Framework for Spiking Neural Networks from High-Level Models to FPGA Deployment

Shane Harrigan; Sonya Coleman; Dermot Kerr; JP Quinn; Kyle Madden; Shuo Liu

FrostRune: An Asymmetric Translational Framework for Spiking Neural Networks from High-Level Models to FPGA Deployment

Shane Harrigan, Sonya Coleman, Dermot Kerr, JP Quinn, Kyle Madden, Shuo Liu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

6 Downloads (Pure)

Abstract

Spiking Neural Networks (SNNs) provide energy-efficient neuromorphic computing, but their deployment on FPGAs requires specialised hardware knowledge. We introduce FrostRune, a framework for automatically converting Python-based SNN descriptions into FPGA-ready VHDL code using the YAML format. Our experiments on the MNIST-DVS dataset demonstrate that FrostRune achieves high accuracy and behavioural equivalence, while significantly reducing latency, power consumption, and resource usage compared to traditional HDL coding approaches. This work lowers the barrier to FPGA-based neuromorphic computing and enables efficient SNN deployment.

Original language	English
Title of host publication	Symposium Series in Computational Intelligence (SSCI)
Publisher	IEEE
Publication status	Published (in print/issue) - 17 Mar 2025
Event	2025 IEEE Symposium Series on Computational Intelligence - Trondheim, Norway., Trondheim, Norway Duration: 17 Mar 2025 → 20 Mar 2025 https://ieee-ssci.org/?ui=home

Conference

Conference	2025 IEEE Symposium Series on Computational Intelligence
Abbreviated title	SSCI 2025
Country/Territory	Norway
City	Trondheim
Period	17/03/25 → 20/03/25
Internet address	https://ieee-ssci.org/?ui=home

Keywords

neuromorphic
FPGA
spiking neural networks

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

Accepted versionAuthor Accepted Manuscript, 219 KB

Efficient level set method for novelty detection
Liu, S. (Author), Ding, X. (Supervisor), Liu, J. (Supervisor) & Coyle, D. (Supervisor), Mar 2024
Student thesis: Doctoral Thesis

Cite this

@inproceedings{fa76ce68a7b9488eb8a5e5941068b4cf,

title = "FrostRune: An Asymmetric Translational Framework for Spiking Neural Networks from High-Level Models to FPGA Deployment",

abstract = "Spiking Neural Networks (SNNs) provide energy-efficient neuromorphic computing, but their deployment on FPGAs requires specialised hardware knowledge. We introduce FrostRune, a framework for automatically converting Python-based SNN descriptions into FPGA-ready VHDL code using the YAML format. Our experiments on the MNIST-DVS dataset demonstrate that FrostRune achieves high accuracy and behavioural equivalence, while significantly reducing latency, power consumption, and resource usage compared to traditional HDL coding approaches. This work lowers the barrier to FPGA-based neuromorphic computing and enables efficient SNN deployment.",

keywords = "neuromorphic, FPGA, spiking neural networks",

author = "Shane Harrigan and Sonya Coleman and Dermot Kerr and JP Quinn and Kyle Madden and Shuo Liu",

year = "2025",

month = mar,

day = "17",

language = "English",

booktitle = "Symposium Series in Computational Intelligence (SSCI)",

publisher = "IEEE",

address = "United States",

note = "2025 IEEE Symposium Series on Computational Intelligence, SSCI 2025 ; Conference date: 17-03-2025 Through 20-03-2025",

url = "https://ieee-ssci.org/?ui=home",

}

TY - GEN

T1 - FrostRune: An Asymmetric Translational Framework for Spiking Neural Networks from High-Level Models to FPGA Deployment

AU - Harrigan, Shane

AU - Coleman, Sonya

AU - Kerr, Dermot

AU - Quinn, JP

AU - Madden, Kyle

AU - Liu, Shuo

PY - 2025/3/17

Y1 - 2025/3/17

N2 - Spiking Neural Networks (SNNs) provide energy-efficient neuromorphic computing, but their deployment on FPGAs requires specialised hardware knowledge. We introduce FrostRune, a framework for automatically converting Python-based SNN descriptions into FPGA-ready VHDL code using the YAML format. Our experiments on the MNIST-DVS dataset demonstrate that FrostRune achieves high accuracy and behavioural equivalence, while significantly reducing latency, power consumption, and resource usage compared to traditional HDL coding approaches. This work lowers the barrier to FPGA-based neuromorphic computing and enables efficient SNN deployment.

AB - Spiking Neural Networks (SNNs) provide energy-efficient neuromorphic computing, but their deployment on FPGAs requires specialised hardware knowledge. We introduce FrostRune, a framework for automatically converting Python-based SNN descriptions into FPGA-ready VHDL code using the YAML format. Our experiments on the MNIST-DVS dataset demonstrate that FrostRune achieves high accuracy and behavioural equivalence, while significantly reducing latency, power consumption, and resource usage compared to traditional HDL coding approaches. This work lowers the barrier to FPGA-based neuromorphic computing and enables efficient SNN deployment.

KW - neuromorphic

KW - FPGA

KW - spiking neural networks

M3 - Conference contribution

BT - Symposium Series in Computational Intelligence (SSCI)

PB - IEEE

T2 - 2025 IEEE Symposium Series on Computational Intelligence

Y2 - 17 March 2025 through 20 March 2025

ER -

FrostRune: An Asymmetric Translational Framework for Spiking Neural Networks from High-Level Models to FPGA Deployment

Abstract

Conference

Keywords

UN SDGs

Access to Document

Fingerprint

Student theses

Efficient level set method for novelty detection

Cite this