Microchip aims to help developers use its PolarFire FPGAs for creating overlay-based neural network applications for edge use, without learning an FPGA tool flow, with a product called VectorBlox Accelerator software development kit (SDK).

“FPGAs are ideal for edge AI applications, such as inferencing in power-constrained compute environments, because they can perform more Gop/s with greater power efficiency than a CPU or GPU, but they require specialised hardware design skills,” according to the company. “VectorBlox Accelerator SDK is designed to enable developers to code in C/C++ and program neural networks without prior FPGA design experience.”

The tool kit  works on Windows and Linux operating systems, and can execute models in TensorFlow and open neural network exchange (ONNX), with Microchip contending that ONNX supports many frameworks, such as Caffe2, MXNet, PyTorch, and MATLAB.

It also includes bit-accurate simulation to model the FPGA implementation while still on the PC or server. Neural network intellectual property (IP) included with the kit also supports the ability to load different network models at run time.

PolarFire FPGAs, said Microchip, have maths blocks that can deliver up to 1.5Top/s, and packages as small as 11 x 11 mm.

This is part of Microchip’s ‘Smart Embedded Vision initiative’, launched last July, to promote the use of PolarFire in edge applications by providing hardware and software developers with tools, IP cores, and boards.

You will have to wait until the third quarter for VectorBlox availability, although an early-access programme is scheduled for June.