Artificial neural network inference on fpgas using open-source tools

Artificial neural networks are widely spread in the modern world. Various hardware is used for neural network inference: from CPUs and GPUs to FPGAs and ASICs. An important research area is inference acceleration. Many open-source tools have been proposed in this area. This article contains a review of a range of open-source tools for neural network inference, acceleration and hardware synthesis. Some of the tools have been selected for evaluation on an FPGA. Five neural network examples have been used as test models. Intel CPU, NVIDIA GPU and Cyclone V FPGA have been used as evaluation platforms. Results show that TVM/VTA and LeFlow tools can successfully process neural network models and run them on the FPGA. However, execution results are controversial.