Hasty Briefsbeta

A robust, open-source framework for Spiking Neural Networks on low-end FPGAs

19 days ago
https://arxiv.org/abs/2507.07284
Copy Link
  • #Spiking Neural Networks
  • #Neuromorphic Computing
  • #FPGA
  • Spiking Neural Networks (SNNs) offer a power-efficient alternative to traditional neural networks by using 0/1 spikes instead of multiply-and-accumulate operations.
  • Existing SNN acceleration architectures like Loihi, TrueNorth, and SpiNNaker are largely inaccessible to the wider community.
  • This paper introduces a robust, open-source framework for SNNs targeting low-end FPGAs, featuring a synaptic array for spike propagation.
  • The framework includes a PyTorch-based SNN model compiler and requires minimal resources (6358 LUT, 40.5 BRAM).
  • Tested on a Xilinx Artix-7 FPGA at 100 MHz, the framework achieves competitive performance (0.52 ms/img for MNIST digit recognition).
  • The framework accurately simulates hand-coded any-to-any SNNs on toy problems and is available as open-source.
AboutLogin