Fraunhofer Institute for Integrated Circuits IIS
Project Overview
The Challange: The Carbon Footprint of Frontier AI
The rapid evolution of Large Language Models (LLMs) and Foundation Models - such as ChatGPT, Claude, and specialized industrial surrogate neural networks - is creating a major computational bottleneck. Training these cutting-edge, frontier AI systems demands a massive infrastructure of GPU clusters, consuming vast amounts of electrical energy and pushing modern data centers to their ecological and economics limits. According to Neural Scaling Laws, achieving better model performance typically forces an exponential increase in training time and compute resources. This continuous need for larger datasets and bigger hardware setups represents one of the greatest barriers to the sustainable advancement of modern AI.
Our Vision: Revolutionizing Foundation Model Training via Quantum Computing
The QC-Train project bridges two defining future technologies: Quantum Computing (QC) and Artificial Intelligence (AI). Our core focus is to fundamentally redesign how massive AI Foundation Models are trained by moving the most rescource-intensiv mathematical operations from traditional hardware to quantum processors. Early technical projections show that quantum-accelerated training workflows could enable a significant reduction in resource consumption compared to traditional classical methods - paying the way for the sustainable development of next-generation frontier AI.
Core Operational Approach
To achieve this, the project is developing software tools that adapt both classical an quantum algorithms for the creation of efficient training workflows of such foundational models. First, standard model-trainig strategies from classical computing will bre re-formulated into routines that can be offloaded to a quantum computer. Following this, these routines will be formulated as quantum algorithms, and analyzed for robustness, scalability, and resource requirements. Finally, advantage regimes will be identified, to pinpoint the exact scenarios where quantum acceleration delivers a clear economic and environmental advantage, once fault-tolerant quantum computers (FTQC) are available.
The Role of Fraunhofer IIS
As the technical lead for quantum algorithms and quantum software design, Fraunhofer IIS (represented by the Quantum Compilation research group) is the fundamental link - that aims to bridge the gap between applications, algorithm, and quantum hardware. We intend to achieve this by designing automated software tools to translate classical equations that govern the training process in large foundational models, into quantum circuits that implement quantum algorithms that are expected to speed up the training pipeline. In particular, we develop techniques and tools for the resource-efficient compilation of block encoding, the basic building block underlying the targeted quantum algorithm.
