githubofaliyev/SNN-DSE: Surrogates, Spikes, and Sparsity

Artifact Release for ISPASS 2026

Paper: Surrogates, Spikes, and Sparsity: Performance Analysis and Characterization of SNN Hyperparameters on Hardware

Authors: Ilkin Aliyev, Jesus Lopez, Tosiron Adegbija
Venue: IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS 2026)

Overview

This release contains the reproducibility artifacts for our ISPASS 2026 paper, including:

• Training scripts for spiking neural networks with configurable surrogate gradient functions (Fast Sigmoid, Arctangent, Spike Rate Escape, Stochastic Spike Operator) and neuron models (LIF, Lapicque)
• FPGA instrumentation platform (SystemVerilog RTL) for cycle-accurate latency and spike count measurement
• Weight extraction utilities for deploying trained models to hardware

Key Results Reproducible

• Figure 5: Accuracy sensitivity to surrogate gradient functions
• Figure 6: Hardware inference latency characterization
• Figure 7: Pareto analysis of neuron configurations (LIF vs. Lapicque)
• Table II: Benchmarking against prior ASIC/FPGA implementations

Requirements

• Python 3.11, PyTorch 2.2.2, CUDA 12.8
• snnTorch 0.7.0, Brevitas 0.10.2, tonic
• Xilinx Vivado (for hardware simulation)

Datasets

• DVS128-Gesture
• N-MNIST
• DVS-CIFAR10

License

MIT License

Citation

@inproceedings{aliyev2026surrogates,
  title={Surrogates, Spikes, and Sparsity: Performance Analysis and Characterization of SNN Hyperparameters on Hardware},
  author={Aliyev, Ilkin and Lopez, Jesus and Adegbija, Tosiron},
  booktitle={IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)},
  year={2026}
}