TEdgeAI

AI chips for edge devices have to compute efficiently and at the same time be trustworthy and resistant to attacks. That is hard: design flow, tools and AI model have to fit together, and at the edge of the network computing power and energy are scarce. TEdgeAI (design concepts and methods for trustworthy and attack-resistant AI chips at the edge) develops methods and open-source tools for this.

• An open-source tool that generates optimised hardware descriptions directly from AI models
• Extensions of existing open-source design tools with security analysis and hardening of neural networks
• Testing the tools and methods by designing an AI accelerator on an FPGA

The project also strengthens German–Taiwanese cooperation in microelectronics and the training of chip designers.

Our contribution

The University of Lübeck coordinates the project. The ITI analyses hardware and software together for weaknesses (HW/SW co-analysis) and develops solutions for the trustworthiness of AI. To this end, a toolchain and a design flow for dedicated AI hardware are being built, aimed above all at the edge.

Partners and funding

Our partners are National Taiwan University (Taipei) and Chung Yuan Christian University (Taoyuan). Funded by the German Federal Ministry of Research, Technology and Space (then BMBF) under the DE:Sign call of the microelectronics design initiative.