FINN
FINN is an experimental open-source compiler from AMD Research that generates dataflow-style FPGA accelerators for quantized neural networks, aiming for high inference throughput and low latency.
FINN is an experimental framework from AMD Research's Integrated Communications and AI Lab for exploring deep neural network inference on FPGAs. It specifically targets quantized neural networks, generating dataflow-style hardware architectures customized for each network so the resulting FPGA accelerators can reach high throughput and low latency.
The compiler is fully open source so users can adapt it across the software and hardware abstraction layers involved in FPGA inference, and the project maintains a separate finn-examples repository with pre-built examples for several neural networks. Only Docker-based execution of the FINN compiler is currently supported, and documentation includes a getting-started guide plus Jupyter notebook tutorials best run from inside Docker.