SuperNeuroMAT: An Efficient Matrix-based Simulator for Spiking Neural Networks

SuperNeuroMAT is an open-source, Python-based simulator designed to address the performance and accessibility gaps in current spiking neural network (SNN) frameworks. By utilizing a novel matrix-based approach to model leaky integrate-and-fire (LIF) neuron dynamics, the simulator provides a scalable solution for researchers using standard laptops and desktops. It natively supports both dense and sparse execution modes, allowing it to handle a wide range of network sizes and connection probabilities without requiring specialized hardware.

The authors developed SuperNeuroMAT to bridge the gap between neuroscience-focused simulators (which are often slow) and machine learning-focused simulators (which often lack biological parameters). By leveraging efficient libraries like NumPy and SciPy, the simulator achieves significant performance gains. Benchmarking results indicate that SuperNeuroMAT consistently outperforms established tools—including NEST, Brian2, BindsNET, and snnTorch—in both execution speed and peak resident memory usage across various network configurations.

Beyond standard machine learning benchmarks like the Digits and citation network datasets, SuperNeuroMAT is designed for general-purpose neuromorphic computing. The authors demonstrate its utility in diverse tasks, including event-based vision (N-CARS and ASL-DVS) and algorithmic primitives such as the neuromorphic shortest path algorithm and basic arithmetic operations. Its availability via the Python Package Index (PyPI) is intended to lower the barrier to entry for students and researchers, facilitating broader adoption of neuromorphic algorithms.