ST Edge AI Core Documentation

ST Edge AI Core logo

ST Edge AI Core is a command-line tool ( stedgeai) that compiles pre-trained deep learning (DL) and machine learning (ML) models into optimized libraries ready to run on ST devices — enabling edge AI on microcontrollers (MCUs), microprocessors (MPUs), and smart sensors.

Key capabilities:

  • Analyze a model to estimate memory footprint (ROM/RAM) and computational complexity (MACC)

  • Validate the deployed C-model against the original model using random or custom datasets, either on the host or directly on target hardware

  • Generate specialized C-files optimized for the selected ST target

Supported targets include the STM32 family (Cortex-M series, including STM32N6 with ST Neural-ART NPU), Stellar automotive MCUs (SR5/SR6 families), ISPU intelligent sensor processing units, and MLC machine-learning cores.

Getting started

  1. Install the required components using the modular installer

  2. Set up the environment (PATH configuration)

  3. Use the Command Line Interface to analyze, validate, and generate code from your model

For detailed metrics and validation methodology, see the Evaluation report and metrics article.

Explore the documentation

ST Edge AI Core

Core basics: install the components, set up the environment, and drive the stedgeai command line to analyze, validate, and generate code.

ST Edge AI Core
Embedded Client API

The ST Edge AI ( stai) API to integrate the generated model into your embedded application, with migration guidance and examples.

Embedded Client API
Supported Operators and Layers

Operator and layer coverage per framework toolbox — TensorFlow Lite, Keras, and ONNX — including DQNN and ONNX-ML support.

Supported Operators and Layers
Advanced Features

Power-user options: the AiRunner package, I/O data type and layout (NHWC/NCHW), stateful Keras LSTM/GRU, and custom layers.

Advanced Features
STM32 Target

Deploy on the STM32 series: CLI usage, relocatable models, running a c-model locally, and USB-CDC validation.

STM32 Target
Neural-ART Accelerator Target

Deploy and profile models on the STM32N6 Neural-ART NPU: concepts, compiler, example projects, profiling, and encryption.

Neural-ART accelerator™ target
STELLAR Target

Generate and validate models for Stellar automotive MCUs, including the on-target validation flow.

STELLAR target
ISPU Target

Generate, validate, and integrate models for ISPU intelligent sensor processing units.

ISPU target
MLC Target

Configure and deploy decision-tree models on the Machine Learning Core (MLC) of ST MEMS sensors.

MLC target

Resources