Software developer Mipsology announced today the availability of its Zebra FPGA IP, a solution for deploying machine learning systems on FPGAs and SoCs. It is expected to ship early Q3.
The IP takes TensorFlow, PyTorch, or ONNX convolutional neural network (CNN) models and maps them to the FPGA. It uses only part of the FPGA, however, enabling designers to integrate their own functions and fully utilize ARM CPUS/GPUs.
Zebra is designed with edge applications in mind, such as industrial automation, security, autonomous vehicles, and smart cities, as well as for image super resolution enhancement.
For more information on Zebra, see our story “FPGA-based neural network software gives GPUs competition for raw inference speed,” from our May/June 2021 issue.
