Artificial intelligence: algorithms and chips developed together

Chatbots and virtual assistants, smart home applications and navigation systems - we are increasingly encountering artificial intelligence in our everyday lives. Many industries can benefit from AI, such as energy and medical technology, automotive and mechanical engineering, marketing and logistics. The development of suitable hardware, i.e. powerful microchips, is crucial for the successful use of AI. In addition to high computing power, low costs and energy efficiency, security and reliability are particularly important. "The development of hardware requires special expertise, as the design processes are much more complex than for software," says Mehdi B. Tahoori, Professor of Dependable Nano Computing at the Institute of Computer Engineering at KIT. "But there is a global shortage of hardware experts." In addition, licenses for state-of-the-art hardware design tools are extremely expensive. These challenges present major hurdles, especially for small and medium-sized enterprises (SMEs).

Special attention is paid to security and safety
To enable the efficient design of AI chips and AI systems, the "German Open-Source Tools for AI Algorithm-Hardware Co-Design" project, EDAI for short, is pursuing a new approach. The researchers are coupling the design of AI algorithms and AI chips. "We are developing algorithm-hardware co-design tools that combine the optimization of AI algorithms with the exploration of the hardware design space," explains Tahoori, who is the spokesperson for the project. "We take the hardware requirements into account throughout the entire process." Particular attention is paid to the criteria of security and functional safety.
The EDAI project is based on open source, which means that the software developed is open source. On the one hand, this is intended to expand the community for the design of AI chips and counteract the global shortage of skilled workers; on the other hand, it is intended to facilitate access to AI-based solutions, especially for small and medium-sized companies. This is because AI is an essential component of digitalization and offers SMEs in particular the opportunity to use scientific findings for their value creation and gain competitive advantages.

AI hardware architectures are optimized for applications
EDAI uses an automated design flow to bridge the gap between the optimization of AI models at software level and their implementation in hardware. The researchers use an algorithm/hardware co-design approach that optimizes AI algorithms by taking the hardware into account from the outset. They develop new templates for AI hardware architectures that are automatically selected and optimized to meet the specifications for the intended application.

About EDAI
KIT coordinates EDAI, partners are TU Munich, FAU Erlangen-Nuremberg and RPTU Kaiserslautern-Landau. EDAI started on May 1, 2024 and is scheduled to run for three years. The Federal Ministry of Education and Research is funding the project on the basis of the guideline "Design instruments for sovereign chip development with open source" (DE:Sign). The project has a total budget of around three million euros.
The project consortium brings together internationally renowned AI and hardware experts from leading German universities and is committed to the successful transfer of technology to industry, especially SMEs. The aim is to create a powerful, adaptable AI platform for German industry and for specific application areas such as automotive and industrial automation. This will promote innovation, competitiveness and technological sovereignty. "KIT as coordinator makes Karlsruhe the location for energy-efficient and trustworthy AI chips in Germany and Europe," says project spokesperson Tahoori.