Tiny AI models for embedded systems

Unlock the potential of measurement data, create intelligent ECU functions

Experience the revolutionary potential of AI in your embedded systems. Traditionally, deploying advanced machine learning models to microcontrollers and microprocessors demands extensive skills in data science, AI algorithms, hardware, and safety. Our platform simplifies this entire journey, making AI integration intuitive and efficient.

The Challenge

The Challenge

For over two decades, engineers have leveraged Machine Learning (ML) to optimize complex automotive systems like engines, brakes, and actuators, producing advanced models, virtual sensors, and precise controls with tools like ETAS ASCMO. Integrating these models directly into ECUs can cut costs, replace physical sensors, accelerate development and calibration, and enable real-time estimation, prediction, and intelligent functions—turning years of expertise into on-device AI for smarter, more efficient vehicles.

Key challenges for scaling ML models into production

Finding the best AI model

Developing AI for ECUs is complex, requiring expertise in model types, coding frameworks, pipelines, hyperparameter tuning, and data handling. Generic tools often clash with automotive workflows, creating steep learning curves and inefficiencies.

Deployment on embedded device

Turning AI models into production-ready ECU code is challenging, requiring ECU compatibility, strict safety standards, real-time performance, and integration into automotive toolchains. AI expertise rarely overlaps with embedded engineering skills, creating a major gap.

Graphic of Tiny Ai models in embedded systems

Bridging the gap

Automatic AI Model Creation and Calibration

The ASCMO tool family enables automatic creation of AI models from data, delivering high-precision modeling, analysis, and optimization of large, complex systems. It performs automatic model search to identify the best candidates and selects the optimal model on the Pareto front using metrics like RMSE, model size, inference time, RAM, and ROM.

Embedded AI Coder for High-Performance Deployment

The Embedded AI Coder converts trained neural networks into optimized embedded C code for microcontrollers and microprocessors, without requiring special hardware or libraries. It delivers best-in-class performance, supports functional safety up to ASIL-D (ISO 26262), and integrates easily into existing toolchains and software architectures like AUTOSAR.

Woman reading Tiny AI white paper on tablet

White paper: How to integrate advanced AI into automotive ECUs

This ETAS white paper introduces engineers to an innovative approach for developing, optimizing, and successfully deploying powerful machine learning (ML) models to production ECUs without requiring in-depth AI expertise. Serving as a hands-on guide, the paper presents a toolchain that abstracts the complexity of ML methods through user-friendly workflows and converts these models into efficient, safe embedded code suitable for microcontrollers and microprocessors.

Learn how to generate highly accurate models with remarkable ease and provide them as highly optimized C-code for embedded devices.

Get access to white paper