Implementing AI/ML on KW45B41Z-EVK using eIQ TSS and Ported TFLM Middleware

cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Implementing AI/ML on KW45B41Z-EVK using eIQ TSS and Ported TFLM Middleware

141 Views
Srushti_Mulimani
Contributor I

Hi NXP Community,

I am working on an edge AI application for the KW45B41Z-EVK (Cortex-M33) and would like advice on deploying a machine learning model on this board.

Since the KW45 SDK does not include the eIQ/TFLM middleware by default, my plan is:

  1. Model Training: Train an n-class classification model for time-series data using eIQ Time Series Studio (eIQ TSS). Since the KW45 is not listed as a target in eIQ TSS, I selected the FRDM-MCXN947 (Cortex-M33) as the target platform to export the model.

  2. Runtime Deployment: Port the middleware/eiq/tensorflow-lite folder from the FRDM-MCXN947 SDK into my KW45 SDK project and link it with CMSIS-NN.

My Questions:

  • Is porting the eIQ TFLM middleware from the MCXN947 SDK to the KW45 SDK a valid and supported approach?

  • Is selecting the FRDM-MCXN947 in eIQ TSS suitable for generating models to run on the KW45?

  • Is there any other recommended or native workflow to implement AI/ML on the KW45B41Z-EVK?

Thanks for your help!

Labels (1)
0 Kudos
Reply
2 Replies

14 Views
sofiaurueta
NXP Employee
NXP Employee

Hello @Srushti_Mulimani , hope you are doing well.

 

If you are using eIQ Time Series Studio (TSS), I would recommend porting your project from the FRDM-MCXW71 instead. It is an officially supported target in eIQ TSS and is software compatible with the KW45. Both devices share the same Cortex-M33 core architecture and a similar memory profile, so the generated library should be a much closer match for your hardware.

When using a device with such a different hardware profile as the FRDM-MCXN947 as a target for porting, there is a risk of generating incompatible models, for example, due to the differences in available RAM between both devices, or the generation of NPU-accelerated models specific to the MCXN947.

 

Hope this helps!

 

Best regards,
Sofia.

0 Kudos
Reply

46 Views
Srushti_Mulimani
Contributor I

Hi NXP Support Team & Community,

I am writing to follow up on this inquiry, as finding a resolution is quite urgent for my current project timeline.

Could someone from the NXP team please provide a quick confirmation on whether porting the eIQ/TFLM middleware from the FRDM-MCXN947 SDK over to the KW45 SDK is a valid and supported workaround? Any brief guidance or confirmation on the recommended workflow would be extremely helpful so I can confidently proceed with development.

Thank you in advance for your time and assistance!

KW45B41Z-EVK EIQ-TOOLKIT EIQ-TFLITE-MICRO #AI/ML

0 Kudos
Reply
%3CLINGO-SUB%20id%3D%22lingo-sub-2410174%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3EImplementing%20AI%2FML%20on%20KW45B41Z-EVK%20using%20eIQ%20TSS%20and%20Ported%20TFLM%20Middleware%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-2410174%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3E%3CP%3E%3CSTRONG%3EHi%20NXP%20Community%2C%3C%2FSTRONG%3E%3C%2FP%3E%3CP%3E%3CSPAN%3EI%20am%20working%20on%20an%20edge%20AI%20application%20for%20the%20%3CSTRONG%3EKW45B41Z-EVK%3C%2FSTRONG%3E%20(Cortex-M33)%3C%2FSPAN%3E%3CSPAN%3E%20and%20would%20like%20advice%20on%20deploying%20a%20machine%20learning%20model%20on%20this%20board%3C%2FSPAN%3E%3CSPAN%3E.%3C%2FSPAN%3E%3C%2FP%3E%3CP%3E%3CSPAN%3ESince%20the%20KW45%20SDK%20does%20not%20include%20the%20eIQ%2FTFLM%20middleware%20by%20default%3C%2FSPAN%3E%3CSPAN%3E%2C%20my%20plan%20is%3A%3C%2FSPAN%3E%3C%2FP%3E%3COL%3E%3CLI%3E%3CP%3E%3CSPAN%3E%3CSTRONG%3EModel%20Training%3A%3C%2FSTRONG%3E%20Train%20an%20n-class%20classification%20model%20for%20time-series%20data%20using%20%3CSTRONG%3EeIQ%20Time%20Series%20Studio%20(eIQ%20TSS)%3C%2FSTRONG%3E%3C%2FSPAN%3E%3CSPAN%3E.%20Since%20the%20KW45%20is%20not%20listed%20as%20a%20target%20in%20eIQ%20TSS%2C%20I%20selected%20the%20%3CSTRONG%3EFRDM-MCXN947%3C%2FSTRONG%3E%20(Cortex-M33)%20as%20the%20target%20platform%20to%20export%20the%20model%3C%2FSPAN%3E%3CSPAN%3E.%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLI%3E%3CLI%3E%3CP%3E%3CSPAN%3E%3CSTRONG%3ERuntime%20Deployment%3A%3C%2FSTRONG%3E%20Port%20the%20middleware%2Feiq%2Ftensorflow-lite%20folder%20from%20the%20%3CSTRONG%3EFRDM-MCXN947%20SDK%3C%2FSTRONG%3E%20into%20my%20KW45%20SDK%20project%20and%20link%20it%20with%20CMSIS-NN%3C%2FSPAN%3E%3CSPAN%3E.%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLI%3E%3C%2FOL%3E%3CP%3E%3CSTRONG%3EMy%20Questions%3A%3C%2FSTRONG%3E%3C%2FP%3E%3CUL%3E%3CLI%3E%3CP%3E%3CSPAN%3EIs%20porting%20the%20eIQ%20TFLM%20middleware%20from%20the%20MCXN947%20SDK%20to%20the%20KW45%20SDK%20a%20valid%20and%20supported%20approach%3C%2FSPAN%3E%3CSPAN%3E%3F%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLI%3E%3CLI%3E%3CP%3E%3CSPAN%3EIs%20selecting%20the%20FRDM-MCXN947%20in%20eIQ%20TSS%20suitable%20for%20generating%20models%20to%20run%20on%20the%20KW45%3C%2FSPAN%3E%3CSPAN%3E%3F%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLI%3E%3CLI%3E%3CP%3E%3CSPAN%3EIs%20there%20any%20other%20recommended%20or%20native%20workflow%20to%20implement%20AI%2FML%20on%20the%20KW45B41Z-EVK%3C%2FSPAN%3E%3CSPAN%3E%3F%3C%2FSPAN%3E%3C%2FP%3E%3C%2FLI%3E%3C%2FUL%3E%3CP%3EThanks%20for%20your%20help!%3C%2FP%3E%3C%2FLINGO-BODY%3E%3CLINGO-LABS%20id%3D%22lingo-labs-2410174%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3E%3CLINGO-LABEL%3EKinetis%20W%20Series%20MCUs%3C%2FLINGO-LABEL%3E%3C%2FLINGO-LABS%3E%3CLINGO-SUB%20id%3D%22lingo-sub-2411056%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%20translate%3D%22no%22%3ERe%3A%20Implementing%20AI%2FML%20on%20KW45B41Z-EVK%20using%20eIQ%20TSS%20and%20Ported%20TFLM%20Middleware%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-2411056%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3E%3CP%3E%3CSTRONG%3EHi%20NXP%20Support%20Team%20%26amp%3B%20Community%2C%3C%2FSTRONG%3E%3C%2FP%3E%3CP%3EI%20am%20writing%20to%20follow%20up%20on%20this%20inquiry%2C%20as%20finding%20a%20resolution%20is%20quite%20urgent%20for%20my%20current%20project%20timeline.%3C%2FP%3E%3CP%3ECould%20someone%20from%20the%20NXP%20team%20please%20provide%20a%20quick%20confirmation%20on%20whether%20porting%20the%20eIQ%2FTFLM%20middleware%20from%20the%20FRDM-MCXN947%20SDK%20over%20to%20the%20KW45%20SDK%20is%20a%20valid%20and%20supported%20workaround%3F%20Any%20brief%20guidance%20or%20confirmation%20on%20the%20recommended%20workflow%20would%20be%20extremely%20helpful%20so%20I%20can%20confidently%20proceed%20with%20development.%3C%2FP%3E%3CP%3EThank%20you%20in%20advance%20for%20your%20time%20and%20assistance!%3C%2FP%3E%3CP%3E%3CA%20href%3D%22https%3A%2F%2Fcommunity.nxp.com%2Ft5%2Fc-pwmxy87654%2FKW45B41Z-EVK%2Fpd-p%2FKW45B41Z-EVK%22%20class%3D%22lia-product-mention%22%20data-product%3D%223146-1%22%20target%3D%22_blank%22%3EKW45B41Z-EVK%3C%2FA%3E%26nbsp%3B%3CA%20href%3D%22https%3A%2F%2Fcommunity.nxp.com%2Ft5%2Fc-pwmxy87654%2FEIQ-TOOLKIT%2Fpd-p%2FEIQ-TOOLKIT%22%20class%3D%22lia-product-mention%22%20data-product%3D%222670-1%22%20target%3D%22_blank%22%3EEIQ-TOOLKIT%3C%2FA%3E%26nbsp%3B%3CA%20href%3D%22https%3A%2F%2Fcommunity.nxp.com%2Ft5%2Fc-pwmxy87654%2FEIQ-TFLITE-MICRO%2Fpd-p%2FEIQ-TFLITE-MICRO%22%20class%3D%22lia-product-mention%22%20data-product%3D%222671-1%22%20target%3D%22_blank%22%3EEIQ-TFLITE-MICRO%3C%2FA%3E%26nbsp%3B%23AI%2FML%3C%2FP%3E%3C%2FLINGO-BODY%3E%3CLINGO-SUB%20id%3D%22lingo-sub-2411442%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%20translate%3D%22no%22%3ERe%3A%20Implementing%20AI%2FML%20on%20KW45B41Z-EVK%20using%20eIQ%20TSS%20and%20Ported%20TFLM%20Middleware%3C%2FLINGO-SUB%3E%3CLINGO-BODY%20id%3D%22lingo-body-2411442%22%20slang%3D%22en-US%22%20mode%3D%22CREATE%22%3E%3CP%20lang%3D%22es-419%22%3EHello%20%3CA%20href%3D%22https%3A%2F%2Fcommunity.nxp.com%2Ft5%2Fuser%2Fviewprofilepage%2Fuser-id%2F265218%22%20target%3D%22_blank%22%3E%40Srushti_Mulimani%3C%2FA%3E%26nbsp%3B%2C%20hope%20you%20are%20doing%20well.%3C%2FP%3E%0A%3CP%20lang%3D%22es-419%22%3E%26nbsp%3B%3C%2FP%3E%0A%3CP%3EIf%20you%20are%20using%20eIQ%20Time%20Series%20Studio%20(TSS)%2C%20I%20would%20recommend%20porting%20your%20project%20from%20the%20FRDM-MCXW71%20instead.%20It%20is%20an%20officially%20supported%20target%20in%20eIQ%20TSS%20and%20is%20software%20compatible%20with%20the%20KW45.%20Both%20devices%20share%20the%20same%20Cortex-M33%20core%20architecture%20and%20a%20similar%20memory%20profile%2C%20so%20the%20generated%20library%20should%20be%20a%20much%20closer%20match%20for%20your%20hardware.%3C%2FP%3E%0A%3CP%3EWhen%20using%20a%20device%20with%20such%20a%20different%20hardware%20profile%20as%20the%20FRDM-MCXN947%20as%20a%20target%20for%20porting%2C%20there%20is%20a%20risk%20of%20generating%20incompatible%20models%2C%20for%20example%2C%20due%20to%20the%20differences%20in%20available%20RAM%20between%20both%20devices%2C%20or%20the%20generation%20of%20NPU-accelerated%20models%20specific%20to%20the%20MCXN947.%3C%2FP%3E%0A%3CBR%20%2F%3E%0A%3CP%3EHope%20this%20helps!%3C%2FP%3E%0A%3CBR%20%2F%3E%0A%3CP%3EBest%20regards%2C%3CBR%20%2F%3ESofia.%3C%2FP%3E%3C%2FLINGO-BODY%3E