Automatic generation of FPGA-based learning accelerators for the neural network family
DeepBurning is an end-to-end automatic neural network accelerator design tool for specialized learning tasks. It provides a unified deep learning acceleration solution to high-level application designers without dealing with the model training and hardware accelerator tuning. You can refer to DeepBurning homepage for more details.
DeepBurning mainly includes the following four parts:
YOSO:search for the optimized neural network architecture and the NPU configuration
Model-zoo:pre-compiled neural network instructions
Zynq-prj: Pre-built zynq project on ZC706 and MZ7100.
NPU-IP: NPU ip core (netlist) It is a general NPU core that supports almost all the main-stream neural network models. It can be further customized for specific learning tasks and run at higher speed and less resource overhead.