Home

Welcome to FINN+ 🚀

FINN+ is a powerful framework for deploying quantized neural networks on FPGA hardware. It specifically targets quantized neural networks (QNNs), with emphasis on generating dataflow-style architectures customized for each network.

🎯 What is FINN+?

FINN+ takes your quantized neural networks (in QONNX format) and automatically generates optimized FPGA implementations. Whether you're targeting Zynq SoCs or Alveo data center cards, FINN+ handles the complex hardware generation process for you.

FINN+ is a fork of AMD's FINN framework and is developed as part of the EKI research project. For additional information on the original FINN framework, we refer to the project website: https://xilinx.github.io/finn/

✨ Key Features

• 🔄 Automated Compilation - From ONNX model to FPGA bitstream
• ⚡ High Performance - Optimized dataflow architectures
• 🎛️ Flexible Configuration - Fine-tune performance vs. resource trade-offs
• 📊 Resource Estimation - Predict FPGA utilization before synthesis
• 🔧 Multiple Targets - Support for Zynq and Alveo platforms
• 🐍 PYNQ Integration - Python driver for rapid prototyping
• ⚡ High-Performance C++ - Optimized C++ driver for production deployment

🚀 Quick Start

1. Install FINN+

# Using pip (recommended)
pip install finn-plus

# Or using Poetry for development
git clone https://github.com/eki-project/finn-plus.git
cd finn-plus
poetry install

2. Prepare Your Model

Train and export your quantized model using Brevitas:

# Export your Brevitas model to ONNX
from brevitas.export import export_qonnx
export_qonnx(model, dummy_input, "model.onnx")

3. Create Build Configuration

# config.yaml
board: U55C
generate_outputs:
  - estimate_reports
  - bitfile
  - pynq_driver     # Python driver for prototyping
  - cpp_driver      # C++ driver for production
shell_flow_type: vitis_alveo
synth_clk_period_ns: 10.0
target_fps: 1000

4. Build Your Accelerator

finn build config.yaml model.onnx

Check out our Quick-start guide page for detailed setup instructions!

📚 Documentation Overview

Section	Description
Building an Accelerator	How to configure your accelerator builds
DataflowBuildConfig	Complete reference for all configuration options
Development Guide	Contributing and development IDE setup
API Documentation	The FINN+ API documentation

🎯 Use Cases

🖼️ Computer Vision

• Image classification on edge devices
• Real-time object detection
• Video processing pipelines

🔊 Signal Processing

• Audio classification and enhancement
• Sensor data processing
• Time-series analysis

🏭 Industrial Applications

• Predictive maintenance
• Quality control systems
• Process optimization

🛠️ CLI Commands Overview

Command	Description
finn build <config> <model>	Build FPGA accelerator from ONNX model
finn run <flow_file>	Run custom build flow (legacy)
finn config create <path>	Create configuration template
finn config list	Show current configuration
finn deps update	Update FINN+ dependencies
finn test	Run test suite

🤝 Community & Support

• 📖 Documentation: This wiki contains comprehensive guides
• 🐛 Issues: Report bugs on GitHub
• 💬 Discussions: Join the community discussions
• 🔧 Contributing: See our Development guide

🔗 Related Projects

• Brevitas - Quantization-aware training library
• QONNX - Quantized ONNX representation
• PYNQ - Python framework for FPGA acceleration

---

Ready to get started? 👉 Begin with Installation or jump to Quick Start Guide