Toward Low-SWaP Cognitive Agents: Neuromorphic Intelligence and FPGA-Based Deployments of Event Neural Networks

Trevor Bihl, Rajashree Majumder, Zhewei Wang, Avinash Karanth, Jundong Liu; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 4715-4722

Traditional artificial intelligence (AI) solutions are often high-SWaP (Size, Weight, and Power), data hungry, and lack the efficiency, resilience, and autonomy exhibited by biological intelligence. As a consequence, autonomous and cognitive agents suffer from various technical bottlenecks precluding their widespread use. Neuromorphic computing offers the potential to develop learning machines that perceive, adapt, and learn continually. In this paper, we survey the state-of-the-art in neuromorphic intelligence, including advances in spiking neural networks (SNNs), emerging brain-inspired hardware platforms, supporting software stacks, and representative application domains. Key to these advances are deployment pipelines for development, evaluation, and testing of solutions. Thus, in parallel, we further present a preliminary FPGA-based deployment pipeline for both artificial neural networks (ANNs) and SNNs. This approach not only provides a practical low-SWaP alternative for edge AI but also enables head-to-head benchmarking of neuromorphic solutions against conventional neural models on reconfigurable hardware. Together, these perspectives define a roadmap toward robust, low-power cognitive systems capable of real-world autonomy across embedded and constrained environments.