SuperNeuroMAT is a Python package for simulating and analyzing spiking neural networks.

Documentation available: https://ORNL.github.io/superneuromat/

Unlike its sister package, SuperNeuroABM, SuperNeuroMAT uses a matrix-based representation of the network, which allows for more efficient simulation and GPU acceleration.

SuperNeuroMAT focuses on super-fast computation of Leaky Integrate and Fire (LIF) spiking neuron models with STDP.

It provides:

1. Support for leaky integrate and fire neuron model with the following parameters:

• neuron threshold
• neuron leak
• neuron refractory period

2. Support for Spiking-Time-Dependent Plasticity (STDP) on synapses with:

• weight
• delay
• per-synapse disabling of learning

3. Support for all-to-all connections as well as self connections
4. A turing-complete model of neuromorphic computing
5. Optional GPU acceleration or Optional Sparse computation

• Note that long delays may impact performance. Consider using an agent-based simulator such as SuperNeuroABM for longer delays.

1. Install using pip install superneuromat
2. Update/upgrade using pip install superneuromat --upgrade

The installation guide covers virtual environments, faster installation with uv, installing support for CUDA GPU acceleration, and more.

Import the spiking neural network class:

from superneuromat import SNN

See the tutorial for more.

Additionally, the migration guide may be of use to those coming from older versions of SuperNeuroMAT.

1. Please cite SuperNeuroMAT using:

   @inproceedings{date2023superneuro,
     title={SuperNeuro: A fast and scalable simulator for neuromorphic computing},
     author={Date, Prasanna and Gunaratne, Chathika and R. Kulkarni, Shruti and Patton, Robert and Coletti, Mark and Potok, Thomas},
     booktitle={Proceedings of the 2023 International Conference on Neuromorphic Systems},
     pages={1--4},
     year={2023}
   }
   

2. References for SuperNeuroMAT:
   • SuperNeuro: A Fast and Scalable Simulator for Neuromorphic Computing
   • Neuromorphic Computing is Turing-Complete
   • Computational Complexity of Neuromorphic Algorithms