How can I use a custom board with Deep Learning HDL Toolbox?

As of R2022b, Deep Learning HDL Toolbox supports the following boards out-of-the-box:

• Xilinx Zynq ZC706
• Xilinx Zynq Ultrascale ZCU102
• Intel Arria10 SoC development board
• Intel Arria10 GX FPGA board

How can I use my own FPGA or SoC board with Deep Learning HDL Toolbox?

Answer 1

[MathWorks Support Team]

In general, Deep Learning HDL Coder has the ability to generate portable, synthesizable Verilog and VHDL code from your deep learning network. You can manually integrate these IP Cores into your project and deploy them on any Xilinx or Intel FPGA. For more information, see the following documentation:

https://www.mathworks.com/help/deep-learning-hdl/dlarchitecture.html

In addition to this, Deep Learning HDL Coder offers the following board-specific features:

• Improved network performance estimation using calibration bitstream.
• Board resource utilization estimation.
• Generate a custom deep learning processor and bitstream for automatic deployment and prediction.

To leverage these integrated board-specific workflows with MATLAB and Deep Learning HDL Coder for your board, the following options are available:

JTAG Support for custom FPGA & SoC boards:

The following example shows how you can define a custom board and reference design for Deep Learning HDL Toolbox. This will allow you to connect to the board using the MATLAB JTAG interface:

https://www.mathworks.com/help/deep-learning-hdl/ug/define-custom-board-and-reference-design-for-dl-ip-core-workflow.html 

Ethernet Support for custom SoC Boards:

If your custom board is an SoC Board (e.g. Zynq Ultrascale), there is a possibility to enable a faster Ethernet connection from MATLAB to the board. The Ethernet connection leverages the ARM processor running Linux with LIBIIO drivers to transfer data between MATLAB and the SoC board:

https://www.mathworks.com/help/deep-learning-hdl/ug/libiioethernet-connection.html

You will need to build a custom Linux image from the MathWorks Buildroot GitHub repository, as explained in the following MATLAB Answers post:

https://www.mathworks.com/matlabcentral/answers/880703-how-can-i-create-a-custom-linux-image-for-my-custom-board-to-use-the-zynq-workflow-with-matlab-simul

Since this task is not trivial, the recommended workflow is to first enable JTAG support and explore this option first.

Further Assistance:

MathWorks Consulting Service may be able to help with any customization tasks:

https://www.mathworks.com/services/consulting.html