What counts as a research collection?
A research collection is a dated, structured set of records that shows how a technical question was investigated and answered. The most useful collections include:
- Process recipes linked to run results
- Characterization and metrology data
- Yield, defect and failure analyses with root causes
- Lab notebooks and experiment reports
- Simulation inputs and outputs compared with measurements
Why AI companies want it
Scientific and engineering models need examples that link a cause to a measured effect, and that evidence is rarely public. Buyers use it to train and test models for materials discovery, process optimisation and failure analysis. See what qualifies for semiconductor companies and research labs.
How to license a collection, step by step
- Inventory your systems — MES, LIMS, electronic lab notebooks and file shares — and the years they cover.
- Pick a scope, such as one process family or product line.
- Clear ownership. Exclude customer-owned recipes, joint-development data and work with restricted funding terms.
- Check export controls; see export controls and AI data licensing.
- Plan de-identification of customer names, people and, if needed, tool identifiers.
- Describe the collection — counts, years, formats — so buyers can assess it without seeing files.
- Agree the licence; see what the agreement should include.
Who needs to sign off?
| Role | What they check |
|---|---|
| Engineering or R&D lead | Scope, data quality and what counts as a trade secret |
| Legal | Ownership, customer contracts and licence terms |
| Export compliance | Classification of the technical data |
| IT and privacy | Personal data and secure extraction |
| Leadership | Price, exclusivity and strategic risk |
Protecting trade secrets
License only what you are comfortable sharing. Restrict onward use, require deletion at the end of the term and prohibit attempts to identify customers. For the most sensitive data, consider an evaluation-only licence, which lets a buyer test models without training on the data.
Frequently asked questions
Can we license data from government-funded research?
Sometimes. Grant and contract terms can restrict data rights, so check them before including that work.
Is old research data still valuable?
Often. Ten years of linked experiments can be worth more than recent data alone, as long as results were recorded with their settings.
Do we have to share raw files?
Not at first. Buyers assess a description and a small sample; the full collection is delivered only after a licence is signed.