Parallel Implementation of CNN on Multi-FPGA Cluster

We developed a PYNQ cluster that consists of economical Zynq boards, called M-KUBOS, are interconnected through low-cost high-performance GTH serial links. For the software environment, we employed open-source platform. The is anticipated to be multi-access edge computing (MEC) server for 5G mobile networks. implemented ResNet-50 inference accelerator on image recognition MEC applications. By estimating execution time each layer, layers were divided into multiple boards so board would as equal possible efficient pipeline processing. Owing in which FPGAs directly connected by high-speed links, stream processing without network bottlenecks and between readily realized. implementation 4 achieved 292 GOPS performance, 75.1 FPS throughput, 7.81 GOPS/W power efficiency. It 17 times faster speed 130 more efficiency compared CPU, 5.8 GPU.