Review · 2026 · Independent Researcher
A Reproducibility Analysis of Memristive Spiking Neural Networks with Spatio-Temporal Backpropagation
PythonLaTeXSNNMemristor ModelingSTBP
§ I
Abstract
This work reproduces and extends recent results on memristive SNNs trained via Spatio-Temporal Backpropagation (STBP). We model memristor conductance dynamics analytically and integrate them into a custom PyTorch layer that supports surrogate gradients over discrete spike events.
§ II
System Architecture
Custom PyTorch modules for LIF neurons and memristor synapses; STBP training loop with truncated BPTT across T=20 timesteps; LaTeX manuscript with reproducible figure pipeline.
§ III
Results
Convergence within 5% of baseline reported values on N-MNIST; identified sensitivity of accuracy to memristor drift parameters σ_G and non-linearity coefficient α.