Cube AI doesnt generate all files needed

[LukasOst]

I am working with the Nucleo-N6 Board and i am trying to get a neural network to run on the ART accelerator. The following steps are done to create the project:

• Make new project
  • Select NUCLEO-N657X0-Q in the Board Selector
  • In Project Structure select "Secure domain only" and "Appli"
  • Then "Finish" and "Generate demonstration code"
• Configure .ioc
  • Middleware and Software Packs: X-Cube-AI 10.0.0 for context "Application" and check box under core and select "System performance" under Application
  • Configure LPUART1 Mode: Asynchronous
  • Then again select X-Cube-AI under Middleware and Software Packs and add own network (TFLite) and use LPUART1 under "Platform Settings" (b.t.w. some weird windows pop up which say "Please wait, configuring platform settings" which cant be closed)
  • Resolve Errors in RCC and Clock Configuration
• Save .ioc and generate Code

While building -> fatal error in "app_x-cube-ai.c" where the line #include "aiSystemPerformance.h" cant be resolved.
Are my steps to create a project with AI correct?

STM32CubeIDE Version: 1.17.0
OS: Ubuntu 22.04.5 LTS
X-Cube-ai: 10.0.0
N6 Firmware: 1.0.0

[Julian E.]

Hello @LukasOst ,

Sorry for the late answer.

We have some bug when regenerating code with CubeMX and X Cube AI and we think your issue may come from that.

I've ask the dev team for help. I'll update you when I have more information.

Have a good day,

Julian

[stmhxc]

Hello Julian.

Will a simple guidance of NPU be published and in the next update? I use NUCLEO-N6 Board too and can't wait to run my own model on N6 chips.

Really looking forward for your reply.

Best wishes,

[fauvarque.daniel · Accepted solution]

You can find the online documentation here 

https://stedgeai-dc.st.com/assets/embedded-docs/index.html

with a chapter on how to perform inferences on the target.

AI Code generation with MX for N6 has been fixed in the upcoming April release. At this point the systemPerformance application is not supported (it should be removed from the possible choices), you can use the Template Application for a simple template.

You'll still have to manage the loading of the secure and non secure programs from the FSBL code. This part is not directly generated by X-CUBE-AI but can be generated with MX.

I recommend using the latest STM32CubeMX 6.14 as some bug fix have been done also on N6 support

Regards

Daniel

[LukasOst]

Hey @fauvarque.daniel thanks for your answer.

My problem at the moment is that the number of choices in CubeMX when creating a new project is a bit overwhelming. Can you maybe explain when do i need to use the FSBL? From the information i found i understood it that way that i need FSBL if i want to store the Network in extern memory. Is this correct?

Also i am kind of confused by the architecture of the project after creating a new project in CubeMX/CubeIDE. Why is it setup like a "nested" project?

[LukasOst]

Hey @fauvarque.daniel
do you know the date when the April release is planned to publish?

Regards

Lukas

[Julian E.]

Hello @LukasOst ,

I think it is end of april

Have a good day,

Julian