Xilinx/finn

Health v2 · maintenance only

Adoption not part of the score

1046 stars · 306 forks observed · 2026-09-01

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

FINN is an open-source dataflow compiler from AMD/Xilinx that generates highly efficient FPGA accelerators for quantized neural network (QNN) inference. It transforms trained QNNs into customized streaming dataflow architectures with sub-microsecond latency and high throughput.

Use cases

• compile quantized neural networks to FPGA accelerators
• build ultra-low-latency DNN inference on FPGAs
• generate dataflow architectures for QNNs
• explore quantization and parallelization design space for hardware
• deploy neural networks on Xilinx FPGAs
• research hardware/software co-design for neural network inference

When to choose

• you need sub-microsecond latency or very high throughput inference on FPGAs
• your network is a quantized neural network (e.g., BNN, QNN from Brevitas)
• you want an open-source, customizable FPGA DNN compiler
• you're doing research across the hardware/software abstraction stack

When to avoid

• you need generic DNN acceleration for arbitrary float models
• you don't have access to Xilinx FPGA hardware or Vitis toolchain
• you want a plug-and-play deployment without design space exploration
• you can't use Docker, as the compiler only supports Docker-based execution

Facets

framework · maturity active

compiler machine-learning llm-inference deep-learning machine-learning deep-learning embedded-systems hardware gpu-computing python fpga quantized-neural-networks dataflow-architecture hls vitis hardware-acceleration qnn amd-xilinx linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Xilinx/finn")