NeuroPlat: An FPGA-Based Neuromorphic Computing Platform with Optimized Toolchain for Efficient SNN Deployment

作者：Zhipeng Liao, Tianyang Li, Ziyang Shen, Chaoming Fang, Jie Yang, Mohamad Sawan · 发表于：Mind · 年份：2025 · DOI：10.1109/mind67540.2025.11351864

As the applications of brain-inspired computing become increasingly complex, the efficient hardware implementation of spiking neural networks faces two critical challenges: the lack of specialized hardware architectures for simulating spiking neuron behaviors and the absence of dedicated toolchains for mapping SNN algorithms to neuromorphic hardware. To address these limitations, we propose a reconfigurable FPGA-based computing platform featuring specialized computational cores optimized for spiking neuron behavior simulation, incorporating multiple hierarchical storage units, a $\mathbf{1 6}$-parallel spiking processing array, and a dedicated instruction set. Additionally, we develop a comprehensive deployment toolchain that enables the combination of the PyTorch model with our dedicated instruction set, facilitating efficient end-to-end deployment of brain-inspired algorithms on neuromorphic hardware. Spiking-YOLO network has been implemented on this platform and achieves a peak processing speed of 20 FPS, achieving an optimal balance between computational efficiency and resource utilization.