NOMAD (Novel Materials Discovery)

NOMAD, the Novel Materials Discovery laboratory, is a large open European repository for computational materials science, run by a German-led consortium with roots at the Fritz Haber Institute and Humboldt University of Berlin and now part of the FAIRmat and NFDI research-data infrastructure. It stores both the raw input and output files from electronic-structure calculations and a normalised, machine-processed version of the same results, drawn from around 40 community simulation codes. The NFDI consortium reports roughly 19 million entries covering more than 4 million materials, built from well over 100 million individual calculations.

You browse and search it through a web portal that needs no login, and every record carries FAIR metadata so that entries are findable and comparable across the different codes that produced them. Programmatic access runs through a REST API, bulk downloads handle larger pulls, and uploads can be given a persistent DOI for citation. Because the normalised Archive exposes a consistent schema, you can pull out structures, total energies, band structures and other computed properties as structured data rather than scraping pages.

For RAG and AI work the data is numeric and relational, so the practical step is templating each calculation into readable text (composition, method and key results), embedding that, and keeping the raw numbers for exact lookup. It suits materials-science assistants, provenance-aware property lookups and training sets for machine-learning interatomic potentials, where the raw DFT outputs are especially valuable.

On licensing, the headline is Creative Commons Attribution, so credit the uploader, but licences are set per upload and some data sits under an embargo of up to three years before it opens, so confirm the status of each record before reuse and treat commercial use as something to check per upload rather than assume. It complements the sources we already list: Materials Project and OMat24 offer curated, ready-to-use computed datasets, and the Crystallography Open Database holds measured structures, while NOMAD is the broad, raw, multi-code archive underneath them.