Catalog of SIP Cores
System on Chip design resources
System on Chip design resources
July 30, 2026 -
Erlangen/Nuremberg -- AI has already found its way into many areas of work and everyday life. However, training modern AI models requires enormous computing resources, resulting in high energy consumption and significant CO₂ emissions. Reducing this demand through intelligent methods has therefore become a pressing priority. The QC-Train project team, consisting of infoteam Software AG, the Fraunhofer Institute for Integrated Circuits IIS, OptWare GmbH, and the associated partners DATEV and Schaeffler Technologies, is developing a quantum computing-based solution to significantly reduce emissions while simultaneously increasing the performance of training large foundation models. The project is funded under the Bavarian Collaborative Research Program “Digitalization” (BayVFP) of the Free State of Bavaria.
The quantum computing specialists from Fraunhofer IIS, infoteam Software AG, and OptWare GmbH in the QC-Train project team. From left to right: Detlef Wong (Optware GmbH), Ulrich Schwenk (OptWare), Dr. Michael Schlotter (Schaeffler), David Weiß (infoteam), Tim S. Tabrizi (infoteam), Dr. Daniel Scherer (Fraunhofer IIS), Daniel Blümel (infoteam), Marc Maußner (infoteam), Rahul Banerjee (Fraunhofer IIS)
The increasing adoption of Large Language Models (LLMs) such as ChatGPT, Gemini, and others, as well as specialized surrogate networks that approximate complex physical, technical, or mathematical systems based on data, is driving a growing demand for computing power, energy, and storage resources. Training these models is pushing data centers to their technical and economic limits. Against this backdrop, the EU AI Act also calls for greater transparency regarding the energy consumption of AI systems and supports the development of standards aimed at improving resource and energy efficiency.
How can this continuously increasing demand for computing power be brought under control, which grows with every prompt and every additional training cycle? Conventional approaches are no longer sufficient. To address this challenge, the quantum computing experts at Fraunhofer IIS, together with their partners, are developing a solution that bridges the untapped potential of quantum computing and specialized AI algorithms.
As part of the QC-Train project, training methods for large AI models are therefore being fundamentally re-thought and re-designed. “The goal is to transfer the most resource-intensive mathematical operations, which are currently executed on conventional hardware, to quantum processors,” explains Dr. Daniel Scherer, Head of the Quantum Compilation Research Group at Fraunhofer IIS, outlining the project's core approach. “Initial technical forecasts indicate that quantum-accelerated training workflows could reduce resource consumption by up to a factor of 100, paving the way for the AI systems of the future.”
The project will first analyze a wide range of strategies for implementing training methods and adapt them for execution on quantum computers. In parallel, these quantum systems will be evaluated regarding robustness, scalability, and performance.
The software tools developed in this process will then be used to identify precisely those application scenarios in which quantum-based acceleration can deliver significant economic benefits and improved resource efficiency from an ecological perspective, even if today's quantum computers are still limited to only a relatively small number of qubits.
Together with the associated project partners, machine learning models based on their respective training data will be developed and evaluated. The resulting findings will directly contribute to the further advancement of the software tools.
In collaboration with infoteam and OptWare, the quantum computing experts at Fraunhofer IIS will then subject the developed model solutions to a practice-oriented evaluation and benchmarking phase. The objective is to demonstrate that, despite the expected continued growth in AI usage and associated computing demands, quantum-enhanced training approaches can make a substantial contribution to climate-friendly AI development while simultaneously increasing computational performance.
The project was launched in July under the Bavarian Collaborative Research Program “Digitalization” (BayVFP) of the Free State of Bavaria and is scheduled to run for three years.