what kind of basic FPGA system is needed for deep learning IP core generation?

I'm tring to deploy my deep learning network on FPGA. I need to create my FPGA evaluation board.

Now, I have two questions.

1. in the "Register Custom Board" , step 2, how many interface should contain? just necessary interfaces or all the interface?
2. in the "Register Custom Reference Design", step 2, a custom design is need? What does the custom design need, or what kind of design should I build?

Waiting for your answer, thank you!

Answer 1

[Kanishk]

Hi @KH,

I understand you want to create your own FPGA evaluation board.

When registering a custom board in MATLAB, include all the necessary interfaces that are required for the specific application. All possible interfaces available on the board do not necessarily need to be included, only those that are essential for the application are fine.

For registering a custom reference design, a custom design is needed. The custom design should be tailored to meet the requirements of the deep learning application. It should define the data flow and processing logic, ensuring that the FPGA resources are utilized efficiently.

Please go through the following official MATLAB documentation of “Register a Custom Board” and “Define Custom Board and Reference Design for Microchip Workflow” to learn more.

• https://www.mathworks.com/help/releases/R2021a/hdlcoder/ug/register-a-custom-board.html
• https://www.mathworks.com/help/releases/R2021a/hdlcoder/ug/register-a-custom-reference-design.html

Hope this helps!

[KH]

Thanks for your answer!

Answer 2

[Wang Chen]

Hi KH,

On top of Kanishk's answer, you can also refer to following example:

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

Thanks,

Wang