CMSIS-MLEK Reference Application Package
This documentation explains the usage of the CMSIS-MLEK Reference Application Package for developing Machine Learning (ML) and Edge AI applications on Cortex-M processors with Ethos-U NPUs.
Target Audience
This user's guide assumes basic knowledge about Cortex-M software development and CMSIS-Toolbox workflows. It is written for embedded software developers who work with C/C++ compiler toolchains and develop machine learning applications for microcontroller devices with Cortex-M processors and Ethos-U NPUs.
Package Overview
The CMSIS-MLEK package provides ready-to-use templates and examples for machine learning applications. It includes three main template categories:
- Audio Reference Applications - Keyword spotting and audio processing examples
- Video Reference Applications - Object detection and video processing examples
- Generic Reference Applications - General-purpose inference runner for custom ML models (available from version 1.0+)
Manual Chapters
- Overview explains the features and capabilities of the MLEK templates package.
- MLEK Reference Applications - Audio describes the available ML templates for audio use-cases.
- MLEK Reference Applications - Video describes the available ML templates for video and image use-cases.
- Target Configuration Reference Application Hardware gives step-by-step guidance for deploying the reference applications to a hardware platform in Visual Studion Code.
- Target Configuration Arm Virtual Hardware shows how to select a virtual target in Visual Studio code.
Available Templates
The package includes the following example templates:
| Template | Description | Key Features |
|---|---|---|
| Audio/KWS | Keyword spotting application | Real-time audio processing, wake word detection |
| Audio/User Algorithm | Customizable audio processing template | Template for implementing custom audio ML algorithms |
| Video/Object Detection | Object detection using camera input | Real-time object detection, bounding box visualization |
| Video/User Algorithm | Customizable video processing template | Template for implementing custom video ML algorithms |
| Generic/Inference Runner | General-purpose ML inference | Support for custom TensorFlow Lite models |
Supported Platforms
The templates support multiple Arm Cortex-M platforms:
- Corstone-300 (Cortex-M55 + Ethos-U55/U65)
- Corstone-310 (Cortex-M85 + Ethos-U55/U65)
- Corstone-315 (Cortex-M85 + Ethos-U65)
- Corstone-320 (Cortex-M85 + Ethos-U85)
Each template can be built for both Arm Virtual Hardware (AVH) simulation and physical evaluation boards.
Revision History
| Version | Description |
|---|---|
| 0.1.0 | Initial draft release of CMSIS-MLEK templates package |
| 0.5.0 | First published release. |