Towards Automated FPGA Compilation of Spiking Neural Networks

Edge computing and the Internet of Things (IoT) have created a growing demand for low-cost and low-power field programmable gate arrays (FPGAs) to facilitate rapid and efficient processing. At the same time, spiking neural networks (SNNs) have gained popularity for their capacity to replicate biological processes and efficiently handle temporal data, making them a favorable choice in various IoT applications. However, designers must possess a comprehensive knowledge of FPGA architecture and programming methodologies to fully exploit the potential of FPGAs for SNN implementations. In this work, we propose SpiComp framework aiming to automate the implementation of SNNs on FPGAs, leveraging their reconfigurable nature and high efficiency without requiring domain expertise. Results show that the SNN implementations produced by the framework achieve high throughput and have a small energy footprint compared to other implementations in the literature while providing design automation.