Example Project for STANN

This repository contains the LeNet and TenNet examples for the STANN library.

Requirements

The example project in this repo is implemented for the Xilinx Alveo U50.

Xilinx Vitis_HLS and Vivado version 2020.2 are needed for the hardware design, and the Alveo U50 platform files need to be installed in /opt/xilinx/platforms. For the software application, you also need to have Xilinx XRT installed.

How to build the example project

First clone the repository:

git clone --recurse-submodules [email protected]:ce-uos/stann-example.git

After sourcing the settings files for Vitis_HLS 2020.2 and Vivado 2020.2, build the hardware design:

source /tools/Xilinx/Vitis_HLS/2020.2/settings64.sh
source /tools/Xilinx/Vivado/2020.2/settings64.sh
cd hw_src
make all

The make target "all" will build the LeNet example design. After sourcing the setup files for Xilinx XRT, build the software application:

source /opt/xilinx/xrt/setup.sh
cd ../sw_src
mkdir build 
cd build
cmake ../..
make

C Simulation

If you want to run the C testbench for debugging, use the following commands:

cd hw_src
make tb

Citation

If you use this repository, please cite this paper:

@InProceedings{stann2023_mrothmann,
    author="Rothmann, Marc
    and Porrmann, Mario",
    editor="Rojas, Ignacio
    and Joya, Gonzalo
    and Catala, Andreu",
    title="STANN -- Synthesis Templates for Artificial Neural Network Inference and Training",
    booktitle="Advances in Computational Intelligence",
    year="2023",
    publisher="Springer Nature Switzerland",
    address="Cham",
    pages="394--405",
    abstract="While Deep Learning accelerators have been a research area of high interest, the focus was usually on monolithic accelerators for the inference of large CNNs. Only recently have accelerators for neural network training started to gain more attention. STANN is a template library that enables quick and efficient FPGA-based implementations of neural networks via high-level synthesis. It supports both inference and training to be applicable to domains such as deep reinforcement learning. Its templates are highly configurable and can be composed in different ways to create different hardware architectures.",
    isbn="978-3-031-43085-5"
}