Work in Progress

This repository shows how to build Edge AI applications with Cortex-M/Ethos-U based microcontrollers. It uses ModelNova for the AI model development and Keil MDK for the embedded development.

Learn more at Embedded World 2026

Arm: Hall 4-504

Application development, data capturing, and workload analysis of an Edge AI system based on Alif E8.

ModelNova: Hall 4-600

AI model selection, data labeling, and model creation for an Edge AI system based on Alif E8.

Exhibitor forum: Hall 5-210 on Wed, 03/11/2026, 09:30 - 10:00

From model to market: accelerated AI development on Arm. This talk explains the overall AI model development workflow.

AI model development

The diagram below shows the overall development of an AI model that is integrated into an embedded system. During AI model development, the SDS framework is used as a workbench for data capturing and system analysis. Once the optimized model delivers the expected performance it can be integrated into the final application that may be based on FreeRTOS, Keil RTX, or Zephyr.

Initial development steps

1. Create the input interface with signal conditioning and start capturing data for AI model training.
2. Select an AI model, use the captured data for training, analysis, and creation of the optimized AI model.
3. Integrate the AI model into the SDS framework and analyze performance.

Refine AI model development

1. Capture new data where the AI model does not deliver the expected results.
2. Retrain the AI model using additional training data to optimize performance.
3. Add regression testing before integrating a new AI model into the embedded system.

Quick Start

The RockPaperScissors project implements an AI model that detects three hand gestures (RPS_cls_dataset provides test data). The AppKit-E8_USB/SDS.csolution.yml project uses the SDS framework for testing the AI model on the Alif AppKit-E8 hardware or Arm FVP simulation model.

Keil MDK

1. Install Keil Studio for VS Code from the VS Code marketplace.
2. Clone this repository (for example using Git in VS Code) or download the ZIP file. Then open the base folder in VS Code.
3. Open the CMSIS View in VS Code and use the ... menu to choose an example via Select Active Solution from workspace.
4. The related tools and software packs are downloaded and installed. Review progress with View - Output - CMSIS Solution.
5. In the CMSIS view, use the Action buttons to build, load and debug the example on the hardware.

ModelNova Fusion Studio

1. Download and install ModelNova Fusion Studio.
2. ...

Issues or Questions

Use the Issues tab to raise questions or issues.