STM32CubeAI : External memories

[BKOSI.1]

Hi,

We are working to an ANN development on TensorFlow, this ANN works on computer and not on MCU.

During implementation throught CubeMX, an estimation of RAM and Flash permit to select an appropriate MCU.

My question is, ANN implementation is only available if we use internal RAM and Flash or we can run ANN on external RAM ?

BR,

[fauvarque.daniel · Accepted solution]

On the network view you have an additional popup window (with the button) where you can ask to use external memories.

On STM32 known boards, if there is external memory, it will automatically insert the BSP code to use that memory in the generated code.

If you use external memory you can ignore the message that the amount of RAM / Flash required is bigger than the one availbale

When using external memory, the weights are located in external memory and a network_data.bin file is generated so you can flash it on the external memory

The same applies for external RAM, you can put the activation buffer in external RAM

Regards

Daniel

[BKOSI.1]

Hi,

Thank you for this precision,

Regards,

Benjamin