A Complete Pipeline for deploying SNNs with Synaptic Delays on Loihi 2

2025-10-15

AI Summary

To address the absence of synaptic delay modeling, significant performance degradation during GPU-to-chip migration, and constrained edge energy efficiency and real-time capability when deploying Spiking Neural Networks (SNNs) on Intel Loihi 2, this paper proposes an end-to-end compilation and training framework for delay-augmented SNNs. Methodologically: (i) it explicitly incorporates learnable synaptic delays into GPU-based training to enhance spatiotemporal representation; (ii) it introduces a lightweight mapping strategy coupled with an event-driven calibration mechanism to ensure high-fidelity deployment on Loihi 2. Experiments on speech keyword spotting demonstrate that Loihi 2 achieves 18× speedup and 250× energy reduction over GPU inference, with <0.5% accuracy loss. This work constitutes the first empirical validation of efficient, delay-augmented SNN migration onto real neuromorphic hardware.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

Spiking Neural Networks are attracting increased attention as a more energy-efficient alternative to traditional Artificial Neural Networks for edge computing. Neuromorphic computing can significantly reduce energy requirements. Here, we present a complete pipeline: efficient event-based training of SNNs with synaptic delays on GPUs and deployment on Intel's Loihi 2 neuromorphic chip. We evaluate our approach on keyword recognition tasks using the Spiking Heidelberg Digits and Spiking Speech Commands datasets, demonstrating that our algorithm can enhance classification accuracy compared to architectures without delays. Our benchmarking indicates almost no accuracy loss between GPU and Loihi 2 implementations, while classification on Loihi 2 is up to 18x faster and uses 250x less energy than on an NVIDIA Jetson Orin Nano.

Problem

• Deploying energy-efficient SNNs with synaptic delays on neuromorphic hardware
• Developing complete training-to-deployment pipeline for spiking neural networks
• Improving classification accuracy while reducing computational energy consumption

Innovation

• Efficient event-based SNN training on GPUs
• Deploying SNNs with synaptic delays on Loihi 2
• Achieving higher accuracy with delay-enhanced architectures