Issue with NPU operation on NUCLEO-N657X0-Q – Output values not updating

[Student_0726]

Hello!

I am having trouble verifying NPU operation on the NUCLEO-N657X0-Q board.

I am trying to test a simple INT8 FNN (Feed-Forward Neural Network) using an ONNX model. The project was generated via X-CUBE-AI within CubeMX. I have configured the AXI and Clock settings (CPU: 600MHz, NPU: 800MHz, NPU AXI: 800MHz) and, given the model's simplicity, I set it to use internal SRAM only.

While the code execution seems to flow correctly without errors, I noticed that the output values are not being updated. The output buffer remains unchanged, or in some cases, still holds values that look like the input data. It seems the NPU's computation results are not being written back to the memory correctly.

I would greatly appreciate any advice or suggestions, even if they aren't complete solutions!

Environment Versions:

• CubeIDE: 2.1.1

• CubeMX: 6.16.1

• CubeProgrammer: 2.22.0

• ST Edge AI: 4.0

• ST-LINK: V3J17M10

I have attached one of the project files I’ve been testing for your reference. Thank you always for your support!

[Julian E.]

Hi @Student_0726,

 

X Cube AI is depricated, could you please use our new tool: STM32Cube AI Studio.

Then please let me know if you still have this or another issue.

 

Have a good day,

Julian

 

[Student_0726]

Hi @Julian E.,

Thank you for your reply. Apologies for the delayed response; I was taking some time to test things out in AI Studio.

 

After switching to AI Studio, I encountered an issue when clicking "Generate project" (please refer to the attached photo). I suspect this might be related to the External Memory loader settings, but it seems I cannot manually modify the .ioc file and regenerate the code to fix it.

Because of this, I tried using "Generate model Code" with different configurations. In the resulting project, the compiler couldn't read app.x-cube-ai.c or network.c when they were placed in the external project folder (outside of Appli). To address this, I moved the App and generated folders inside the Appli folder and tried building again.

However, this resulted in an obscure error that makes it hard to pinpoint what went wrong: (LL_Arithacc_Init): Unknown destination type (ARM/Thumb). For your information, I have already configured the preprocessor and path settings.

I've also attached the .ioc file I used with the 'Generate model Code' option.

 

 

Are there any reference materials or documentation available for using AI Studio? It would be especially helpful if there is a guide specifically related to the N6.

Additionally, if the reason my X-CUBE-AI tests aren't working is due to compatibility issues with the latest software version, which version would you recommend downgrading to? I am really eager to get this issue resolved.

 

Thanks again for your help.

[Julian E.]

Hi @Student_0726,

 

Regarding the recommended versions, you can find them in the setting part of CubeAI Studio:

 

For the issue that you are facing, I am looking for answers internally.

 

Have a good day,

Julian