Edge AI

Hello, I'm trying to create a project with custom model via STM32 Cube AI studio, which uses STedge Core AI v. 3.0.0. Validation is performed on STM32N6570-DK. I'm getting following error:E200(ValidationError): TARGET: Unable to bind the ST.AI runtime with "network" c-model: [] E801(HwIOError): Invalid firmware - COM7:115200 When running same command from command line I see that progress goes to 74,8 % and then stucks until error is displayed.See attached failed report for details. 

  I'm looking at the test model and the optimized model in the Model Visualization view.However, the menu option to export as SVG is disabled—I can't select it.Do I need to change any other settings?

Hi team, Reposting my earlier query as it appears to have been missed.How do I measure the peak RAM and peak Flash usage of an AI model?I need to calculate peak RAM and peak Flash from the IDE, but the model generated by ST Edge AI is in HEX format and is not included in the IDE's reported size. So, I'm adding the HEX file size to the IDE's peak RAM and peak Flash figures — is this approach correct? Details are shared below. KWS cpu+npuIDE  Flash = text + data = 205,404 + 1,616 = 207,020 bytes ≈ 202.17 KBRAM  = data + bss = 1,616 + 102,040 = 103,656 bytes ≈ 101.23 KBFor cpu+npu we flash the model in  hex format, so we need to add the peak ram and peak flash of hex file with IDE peak ram and peak flash. model size - Total Peak Ram and Peak FlashPeak RAM = (103,656 + 22,065 =  122.77 KB) Peak Flash = (207,020 + 22,065 =  223.72 KB)

Hello, I am trying to convert my Onnx model with the ST Edge AI Core tool to the .nb format for the STM32MP25-DK target.Since the results were not matching I stripped down the model into the single operations and tried to debug them It looks like the problem comes from the MatMul operation. My model looks like this: When I run my test script were I compare the Onnx model with the NB model, then I get the following results:================================================================================ ONNX OUTPUTS ================================================================================ [ONNX] matmul_result shape=(1, 16, 8) dtype=float32 min=-10.684526 max=7.463792 mean=-0.168686 values=[3.1317086219787598, -0.7650761604309082, 0.822991132736206, -6.243682384490967, -1.3690400123596191, 0.5091162323951721, -3.2105088233947754, 3.2010514736175537, 1.773209571838379, -2.338696002960205] ========================================================================

I try to use my own model which build form python and keras, however when I import the model to the tool it have error as follow:>>> C:\ST\STEdgeAI\4.0\Utilities\windows\stedgeai.exe validate --model C:\Users\user\Documents\battery_model_MLP.keras --batch-size 10 --mode host --name network --verbosity 1 --c-api st-ai --optimization balanced --target stm32g4 --memory-pool C:\Users\user\.stm32cubeaistudio\workspace\EISv3_SOCSOHpredictor\.ai\run\run-3\.ai\mempools.json --save-csv --workspace C:\Users\user\.stm32cubeaistudio\workspace\EISv3_SOCSOHpredictor\.ai\run\run-3\.ai\st_ai_ws --output C:\Users\user\.stm32cubeaistudio\workspace\EISv3_SOCSOHpredictor\.ai\run\run-3\.ai\st_ai_outputST Edge AI Core v4.0.1-20581 7ed50de05INTERNAL ERROR: File is not a zip fileRun failed: Error: validate error: INTERNAL ERROR: File is not a zip file. Please check the report for more details. Output: ST Edge AI Core v4.0.1-20581 7ed50de05PASS:   0%|          | 0/187 [00:00<?, ?it/s]              

hiI installed version 5.1.1 today and reactivated the license key, but I got the following error. nanoedge ai studio license has reached its activation limit Is there a solution?   I installed version 5.1.1 today and reactivated the license key, but I got the following error. Is there a solution?

Hi all,I came across an older thread here ("Anomaly Detection on ISPU", ISM330ISN) where the conclusion was that NanoEdge AI's learned knowledge cannot be preloaded onto the ISPU — the "Include knowledge from benchmark" option isn't available when targeting the ISPU, because the model has to relearn after every power cycle. One of you confirmed this is expected behavior given the ISPU sits directly on the sensor with no on-board non-volatile memory for it.I'm doing research on vibration-based bearing fault detection (a manuscript is currently under review at Mechanical Systems and Signal Processing, with validation across four public datasets — CWRU, Huang–Baddour, XJTU-SY, Paderborn), and this relearning constraint is directly relevant to what I'm working on: detectors whose healthy-state reference is small enough to be committed to flash and restored instantly, with no relearning window after a restart.Two questions for the community / ST team:1. Does this same constraint apply to th

Dear ST Community,I’m currently working with an ALIENTEK STM32MP257 development board and ran into an issue when deploying a model converted via the ST Edge AI online tool.The original model is a plant disease recognition model from the ST model zoo (in .tflite format). I used the ST Edge AI webpage to convert it to an .nb model for deployment on my STM32MP257 board. However, when I test the converted .nb model, the inference output is unexpected: it only predicts two classes — “Background_without_leaves” and “Corn__Northern_Leaf_Blight” — both with relatively high confidence. This happens regardless of the input image.In contrast, when I run inference directly using the original .tflite file on the same test images, the results are correct and show proper class diversity.I double-checked my test code and also consulted AI assistance, which suggested that the converted .nb model might be problematic.I’ve attached a compressed file conta

HIST EDGE AI can generate projects for various development boards based on the STM32U0 (e.g., NUCLEO-U083RC). However, the STM32U0 does not appear in the list of supported MCUs (only U3 and U5 are supported). Am I doing something wrong? BRGilberto

Hi,I am trying to install ST Edge AI Core 2.0.0 for an STM32N6 project. During the installation process, the installer asks me to specify the ST Neural-Art package archive.However, I cannot find the ST Neural-Art archive on this page:https://www.st.com/en/development-tools/stm32n6-ai.html#overviewI only found the ST Edge AI Core download, but not the separate Neural-Art archive required by the installer.Could you please let me know where I can download the correct ST Neural-Art archive compatible with ST Edge AI Core 2.0.0?Thank you.

 Hi Team,I am currently working on a Driver Monitoring System (DMS) project using the STM32MP257F-DK board, and I have successfully integrated the OV5647 camera sensor.At this stage, I am facing an issue related to ISP calibration. The camera is currently running with auto exposure enabled, due to which the frame contrast is not constant for any detection. To stabilize and properly tune the image, a fixed YAML calibration file is required. However, a suitable YAML file for the OV5647 sensor is not available.From my understanding, the YAML calibration file needs to be generated using the ISP tuning/calibration tool. However, I tried following the complete process mentioned in the Wiki page  documentation, but the ISP tool is not detecting the board, which is blocking further progress.Could anyone please guide me on the following:How can I generate a custom YAML calibration file for the OV5647 sensor using the ISP tool?Is there any alternative or manual method to create the YAM

Hi ,I am currently working with the NUCLEO-N657X0-Q board and the STM32N6570-DK board. I am deploying and testing my machine learning models using STM32Cube AI Studio.I have noticed something unusual. The model runs successfully on all target boards except the NUCLEO-N657X0-Q when the NPU is enabled. I am not sure why this is happening and would appreciate any insights or suggestions from the community.Another issue is that the NUCLEO-N657X0-Q shows significantly higher inference/execution times compared to the other boards. Surprisingly, even a Cortex-M7-based board appears to perform better in some cases. This is unexpected, especially considering the capabilities of the Cortex-M55 and NPU.Has anyone experienced similar behavior with the NUCLEO-N657X0-Q? Are there any known configuration issues, performance settings, memory considerations, or NPU-related limitations that I should check?Any help or guidance would be greatly appreciated.Thank you!

I am unable to generate project [ consisting of .onnx format - model is using logistic regression] with the help of STM32 cube AI studio . I have attached the error in screenshot below. Can we not implement simple models like logistic regression while doing real time deployment in STM32N6570-DK? Kindly help.RegardsChanchal  

what is edge ai

Hello ST Community,I am using the NanoEdge AI Studio to generate an anomaly detection library for an STM32L4 device (fixed installation, vibration monitoring). The library works fine, but I need to persist the knowledge learned during runtime so that after a power cycle or reboot, the device does not have to learn again from scratch.I followed the official documentation and community advice (including the solution marked as accepted in [this thread](https://community.st.com/t5/edge-ai/is-there-any-way-to-save-or-export-knowledge-which/td-p/123456)):- Added the `.neai` section in my linker script with `KEEP(*(.neai))`- Declared `extern char _sneai_knowledge[];` and `extern char _eneai_knowledge[];`- Computed `NEAI_KNOWLEDGE_SIZE = (uint32_t)_eneai_knowledge - (uint32_t)_sneai_knowledge;`- Implemented functions to save this RAM section to internal Flash and restore it on boot.However, `NEAI_KNOWLEDGE_SIZE` is **0** and the `.map` file shows that the `.neai` section is empty:```.neai 0x20

I am currently exploring STM32Cube AI Studio / Edge AI Cloud, but I am not very familiar with them yet. I encountered an issue during the model conversion process and would like to ask for your assistance. I am trying to analyze and convert an ONNX model exported from PyTorch using STM32Cube AI Studio or Edge AI Cloud.Development board: STM32N6570-DKWhen running Analyze / Generate Code, I encountered the following error:  I would like to ask:Is this error usually caused by the ONNX model format, opset version, or unsupported operators? Are there any recommended ONNX export settings, such as opset version, dynamic shape settings, input/output naming, etc.? Could you help identify the possible cause of this error or provide further debugging suggestions? Thank you for your assistance.

Hello, I am trying to convert my Onnx model with the ST Edge AI Core tool to the .nb format for the STM32MP25 target.While converting the model i get the following error:ST Edge AI Core v2.2.0-20266 INTERNAL ERROR: unsupported operand type(s) for *: 'int' and 'shape_struct'I could track down the error to a resize operation in the Onnx model. Here is a short snippet how the model looks like:When i remove the resize operations and downscale the 28 x 28 patches to 14 x 14 patches. Then the error is gone.So i think that the resize operation is the problem. This happens for quantized and non quantized models. Like stated here (STM32MP2 NPU description - stm32mpu), there should be some resize operations implemented at least on the GPU. Did anyone had similar issues and know how to fix them?Thank you in advance for your help.Best regardsNils

win11  install st edge ai core(version4.0.1) ，everything is ok，but doesn't have a shortcut(try 3 times),  and it doesn’t work, what's wrong? st edge ai core has bugs? Help me!

 Good morning, I try to compile a CNN model for the STM32N6 board but I get the following logs :Generate ProjectCopying previous project files from C:\Users\maick.huguenin\.stm32cubeaistudio\workspace\test-N6\.ai\run\run-9 to C:\Users\maick.huguenin\.stm32cubeaistudio\workspace\test-N6\.ai\run\run-9\.ai\generate\backupCalculating memory footprint...Run failed: Error: analyze error: E103(CliRuntimeError): Error calling the Neural Art compiler. Please check the report for more details. Output: ST Edge AI Core v4.0.0-20500 359356bb0PASS: 0%| | 0/103 [00:00<?, ?it/s] >>>> EXECUTING NEURAL ART COMPILER C:/ST/STEdgeAI/4.0/Utilities/windows/atonn.exe -i "C:/Users/maick.huguenin/.stm32cubeaistudio/workspace/test-N6/.ai/run/run-9/.ai/st_ai_output/model_int8_OE_3_3_1.onnx" --json-quant-file "C:/Users/maick.huguenin/.stm32cubeaistudio/workspace/test-N6/.ai/run/run-9/.ai/st_ai_output/model_int8_OE_3_3_1_Q.json" -g "network.c" --load-mdesc "C:/ST/STEdgeAI/4.0/Utilities/c

I am trying to debug AI application via FSBL (generated itself in STM32cube AI studio) , I have attached the debugger settings and Problem hereby. I had added some printf statements also in between , whose Teraterm output I have attached.Kindly help. RegardsChanchal 

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: 2.1.1CubeMX: 6.16.1CubeProgrammer: 2.22.0ST Edge AI: 4.0ST-LINK: V3J17M10I have attached one of the project files I’ve been testing for your reference. Thank you always for your s

 Hello,I’m working on a project generated by STM32Cube AI Studio targeting the NUCLEO-N657X0-Q, using an XIP configuration with external XSPI flash ( It was said in another post that the generated projects from STM cube AI Studio follow XIP configuration).SetupBoard: STM32N657X0-QToolchain: STM32CubeIDE(2.1.1) + STM32CubeProgrammer(2.22.0)Project generated with STM32Cube AI StudioExecution mode: XIP (application in external flash)Runtime: Bare metal (LL_ATON_OSAL_BARE_METAL)Memory mappingModel location (from generated network.c):.addr_base = (unsigned char *)(0x71000000UL)Programming stepsI programmed manually using STM32CubeProgrammer:second_Appli.elf network_atonbuf.xSPI2.bin → placed at 0x71000000Debug configuration (FSBL project)I start debugging from the FSBL project.In Debug Configuration → Startup → Load Image and Symbols, I configured:second_Appli.elfDownload = OFF (already flashed manually)Load symbols = ONSo FSBL is loaded/debugged, while App and model are already p

Hi! I’m working with the STM32N6570-DK board using an object detection AI model, and I would like to manually modify the camera parameters. I believe settings such as exposure, white balance, and similar controls are currently configured automatically, but I want to use fixed/manual values instead.I already tried modifying the cmw_camera.c file, but nothing changed. Which files do I actually need to modify in order to set specific camera parameter values? Do I need to use a dedicated tool for this?I’m using the IMX335 camera module.  

Hello,I am currently experiencing difficulties using the STM32N6 NUCLEO board.My goal is to test NPU execution using X-CUBE-AI within CubeMX. However, I am stuck in an infinite loop inside the MX_X_CUBE_AI_Process function. Specifically, LL_ATON_RT_RunEpochBlock() continuously returns only the LL_ATON_RT_DONE status, preventing the loop from terminating as expected. do {ret = LL_ATON_RT_RunEpochBlock(&NN_Instance_Default);if (ret == LL_ATON_RT_WFE) LL_ATON_OSAL_WFE();} while (ret != LL_ATON_RT_DONE); I Can't get the LL_ATON_RT_DONE, only LL_ATON_RT_WFE.This issue persists even when using files generated via Cube AI Studio.I have configured the AXI and Clock settings in CubeMX (CPU: 600MHz, NPU: 800MHz, AXI: 800MHz). Since the model is simple, I am using internal memory only and have not enabled any security (TrustZone) or external memory settings.Environment Versions:CubeIDE: 2.1.1CubeMX: 6.16.1CubeAI Studio: 1.1.0CubeProgrammer: 2.22.0ST Edge AI: 4.0ST-LINK: V3

Hi, I am trying to deploy a custom object detection model on the STM32N6-Nucleo board.I started with this:https://github.com/STMicroelectronics/stm32ai-modelzoo-services/tree/main/object_detectionThe training gets done, I get the performance metrics,python stm32ai_main.pyThe deployment code works fine, and I get the deployment complete message. But when I change from development mode to flash boot mode. It doesn't detect the camera.When I program the demo camera.hex filehttps://github.com/STMicroelectronics/x-cube-n6-camera-capture/tree/main/Binary/NUCLEO-N657X0-Qor object detection .hex file,https://github.com/STMicroelectronics/STM32N6-GettingStarted-ObjectDetection/tree/main/Binary/NUCLEO-N657X0-Q/USB-UVC-Displaythe camera works perfectly.Below are the training and deployment configuration files. These are the only files I changed. I would really appreciate any help in this regard. Thank you,# training configure.yaml file:general:project_name: custom_food_yoloxn_256logs_dir: logssav