Reproduce “Spiking Neural Networks for Inference and Learning: A Memristor-based Design Perspective” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

M. E. Fouda, F. Kurdahi, A. Eltawil, E. Neftci

The authors review co-designing task-modulated learning rules with memristor behavior for neuromorphic computing. Three-factor rules combine pre- and postsynaptic activity with an error, surprise, or reward signal; proposed pulse adaptation accounts for device nonlinearity. One RRAM example reports total update asymmetry of 0.77 at ±3 V. An ICA simulation shows weights oscillating after about 10⁴ samples, not a completed hardware system.

Key insights

  • The Leaky Integrate & Fire (LI&F) neuron model, in its discrete form, is equivalent to a recurrent neural network (RNN), enabling the application of advanced machine learning theory.
  • Memristor crossbar arrays offer an efficient architecture for parallel vector-matrix multiplication, crucial for neuromorphic systems.
  • RRAM non-idealities, such as asymmetric non-linear conductance updates and stochasticity, can be mathematically modeled and potentially exploited as computational resources rather than solely as limitations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.