Jhon OrdoñezGuillermo Sahonero-Alvarez2026-03-222026-03-22201910.1109/iestec46403.2019.00126https://doi.org/10.1109/iestec46403.2019.00126https://andeanlibrary.org/handle/123456789/52479Citaciones: 3Convolutional Neural Networks (CNNs) have become a popular and useful deep learning algorithm. However, the requirements of computation have also increased. As implementation of CNNs in Embedded Systems or Cyber-Physical Systems (CPS) is required, the need of efficient technologies of computation like FPGAs arise substantially. In this paper, we present a 3D convolution accelerator implemented on a Xilinx ZCU102 FPGA board. It achieves 32.08 GOP/s of performance and an efficiency of 3.58 GOP/s per Watt. This accelerator has been developed in Xilinx SDSoC environment.enField-programmable gate arrayComputer scienceConvolution (computer science)ComputationConvolutional neural networkEmbedded systemHardware accelerationDeep learningComputer hardwareArtificial neural networkarticle