SNN / DVS Projects
A PyTorch-based framework for developing spiking neural networks with deployment-aware support for Intel Loihi neuromorphic hardware.
Hardware-aware SNN framework overview
Demo panel showing the framework diagram. Blue: implemented. Gray: partially implemented. White: planned.
This project is a PyTorch-based research platform for developing spiking neural networks with deployment-aware support for Intel Loihi hardware. It is designed to bridge algorithm development and neuromorphic deployment by keeping hardware constraints visible throughout the modelling and training workflow.
The project focuses on temporal learning, delay-based memory mechanisms, and practical neuromorphic deployment workflows. It supports experimentation with architectures that can be trained with hardware constraints in mind.
The framework currently implements multilayer feedforward and recurrent SNN architectures, explicit synaptic delay mechanisms, quantization-aware training for Loihi-compatible representations, and export of weights and delays to NxSDK-compatible formats. Development is continuing toward convolutional SNNs and membrane-state quantization.
Current work is extending the framework toward convolutional SNNs and membrane-state quantization, while keeping the export and deployment pipeline compatible with Intel Loihi workflows.
Representative fidelity comparisons are shown below, highlighting how model variations affect deployment-oriented behavior under neuromorphic constraints.
