Generative Foundation Model for Multi-Scale Materials Discovery, Design and Deployment

GENMAT will develop and validate Europe’s first integrated multi-scale, multi-modal materials-science foundation model (M3FM), shortening the 10–20-year lab-to-market cycle. A unified AI infrastructure will connect atomic chemistry, microstructure evolution, manufacturing optimisation and in-service performance monitoring, providing physics-aware insights from design to use so European industry delivers sustainable materials faster and at lower cost. M3FM combines physics-aware learning with EMMO semantics, multi-modal fusion (chemical structures, process parameters, imaging and sensors), and a hierarchical architecture that bridges scales, under open-science practices for reproducibility and trust. We address three high-risk, high-impact challenges: (1) a FAIR data-governance ecosystem integrating existing resources (e.g., NOMAD) into an interoperable platform; (2) a generalisable backbone that learns across materials classes, replacing fragmented single-task models; and (3) trustworthiness via physics-informed constraints, uncertainty quantification and transparent decision-making. The toolkit will provide encoders for chemistry/imaging/sensors, physics-informed neural operators, cross-scale transfer, generative models for molecules and microstructures, UQ and human-centred XAI. EuroHPC will support pre-training; parameter-efficient fine-tuning will enable partner-side adaptation, with secure options to combine proprietary and public datasets. Assets will be shared via EOSC/AI-on-Demand. Impact will be demonstrated through three use cases: recyclable vitrimer-based polymers; PFAS-free durable coatings; and AI-driven structural health monitoring for composite structures. Together these advance circularity, durability and safety, shorten development cycles and de-risk industrial uptake. By delivering open, sovereign AI infrastructure aligned with European values, GENMAT will place Europe at the forefront of AI-enabled materials discovery and deployment.