An Efficient SNN Model Generation Method for IoT Edge Computing

Neuromorphic hardware-based IoT edge services allow intelligent processing on end devices, which makes them suitable for IoT edge computing. However, Comprehending the complex operating processes of Spiking Neural Networks (SNNs) used in neuromorphic hardware can be challenging for IoT developers. In this paper, we propose an efficient SNN generation method to simplify the process for IoT developers. Our proposed method generates SNN models considering the constraints of FPGA devices and neuromorphic hardware, while meeting user performance requirements. We utilize trained model by extracting the set of effective cost data through the pre-processing. Additionally, we focus on minimizing the size of the network model using the set of effective cost data to enhance efficiency.