Funding project on embedded AI

More and more products require integrated AI functions - at the same time, traditional development processes are often complex, time-consuming and require specialized expertise. This is precisely where the 'Edge AI Platform' comes in: It enables companies to develop their own AI models directly for their products and integrate them seamlessly into embedded systems, even without in-depth expert knowledge.

Complete AI pipelines can be created, evaluated and rolled out via a web-based application. The platform automates key steps such as training, optimization and deployment. Users simply define their application and provide data; as a result, they receive optimized software libraries, detailed performance analyses and specific recommendations for suitable hardware.

Technological advancement in version 3.0

In the third project phase, the project team is adding new functions to the platform. Among other things, the focus is on automated data labeling for complex use cases, the expansion to new and related image classes and the retraining of modern vision language models for annotating previously unknown objects. In addition, methods for time series analyses are being integrated, for example through self-supervised learning, and functions for object recognition and tracking are being further developed.

The project also opens up new possibilities at hardware level: Specialized AI accelerators increase energy efficiency and processing speed and enable real-time applications directly on edge devices. This is complemented by approaches such as knowledge distillation to optimize powerful models for resource-limited hardware.

Focus on economic benefits for companies

The economic added value for companies is key: "With the Edge AI platform, we are significantly lowering the barriers to entry for the use of AI in product development. Companies can go from idea to operational model much faster and integrate AI economically into their products," says Dr. Christian Wiede, Head of Embedded AI at Fraunhofer IMS.

Development times and time-to-market are significantly reduced, while prototypes and proofs of concept can be implemented quickly and resource-efficiently. At the same time, new opportunities arise to relieve existing teams and make development processes more efficient. This enables companies to test new use cases at an early stage and make well-founded decisions on the integration of AI into their products.

With a total volume of around 860,000 euros and a duration from December 2023 to November 2026, the project will be implemented in three consecutive sub-projects - from model development to the optimization of AI pipelines and integration into edge devices. The Fraunhofer Institute for Microelectronic Circuits and Systems IMS, the Fraunhofer Institute for Integrated Circuits IIS and the Fraunhofer Institute for Manufacturing Engineering and Automation IPA are involved. Funding is provided by the Fraunhofer Cluster of Excellence Cognitive Internet Technologies.