Chip companies can license design and process records they own, and few domains are scarcer: there is no large public corpus of chip design or fab data. Buyers want RTL with verification logs and bug history, process recipes tied to measured results, and yield and failure analysis. Three limits decide what can move: export controls under the EAR, foundry and customer NDAs, and the trade-secret value you would give away.
- Public chip data is scarce, so owned design and process history is rare supply.
- The valuable record links settings or code to a measured result: a test outcome, a yield, a failure mode.
- Releasing controlled technology to a foreign person, even in the US, can be an export.
- Foundry process kits and customer designs usually belong to someone else.
Can a chip company license semiconductor process data for AI?
Yes, for records it owns and is allowed to share. The reason buyers ask is scarcity. ChipAgents, which builds AI for chip design, writes that "Semiconductor design has no equivalent public corpus." and that "Most valuable engineering knowledge in this industry resides within individual companies' proprietary design files and internal logs, data that never reaches the public internet." (ChipAgents). A developer building an open data set describes the RTL-to-placement data as the "corpus that ML-for-EDA has been missing" (Hugging Face).
This page covers design, fab and yield records. For lab notebooks and materials experiments across chemistry and biotech, see R&D experiment records, which also lists published buyer ranges for semiconductor and materials run records.
Which semiconductor records are useful for AI
ChipAgents' own agent reasons "across specifications, RTL, verification plans, assertions, logs, waveforms, and EDA tool outputs" (ChipAgents), which is a fair list of what buyers look for.
| Record set | What it holds | Why it is valuable |
|---|---|---|
| RTL with revision history | Design code and every change | Shows how designs evolve and bugs are fixed |
| Verification plans, testbenches and regression logs | Tests, coverage, failures and fixes | Links code to a pass or fail result |
| Bug tracker and review threads | Bugs found, root cause, fix and review | Multi-step debugging reasoning |
| Timing, power and physical-design reports | Tool outputs across iterations | Ties design choices to measured results |
| Process recipes with metrology | Settings for each step and the measured outcome | A known ground truth for each run |
| Yield, test and failure-analysis records | Wafer and die yield, failure modes, corrective actions | Decisions linked to outcomes |
| Equipment logs and maintenance | Tool states, alarms, fixes | Fault detection and recovery examples |
Export controls under the EAR
Much semiconductor technology is controlled under the US Export Administration Regulations, and the rules reach sharing inside the US. An export includes releasing "technology" "or source code (but not object code) to a foreign person in the United States (a "deemed export")" (15 CFR 734.13). Release includes "Oral or written exchanges with a foreign person of "technology" or source code in the United States or abroad." (15 CFR 734.15). BIS summarizes: "A deemed export refers to the sharing or release of controlled technology or source code to a foreign person within the United States." (BIS).
For a data license, that means classifying the records before anything moves, and knowing who at the buyer, and at its cloud and labeling vendors, will see them. Other countries run their own dual-use regimes. Your export compliance team and counsel make this call.
Trade secrets and customer or foundry NDAs
Process recipes are often trade secrets. US law protects information only if "the owner thereof has taken reasonable measures to keep such information secret" and it "derives independent economic value, actual or potential, from not being generally known" (18 U.S.C. 1839). Licensing a recipe without strong confidentiality and use limits can weaken that protection. ChipAgents notes that design files and debugging logs "are, for many semiconductor companies, too sensitive to transmit to third-party cloud environments" (ChipAgents).
- Foundry process design kits and models are licensed to you under NDA; they stay out.
- Customer designs, including build-to-print and design-service work, belong to the customer.
- Third-party IP blocks come with their own licenses.
- Consider older process nodes or retired products first, where the competitive and control questions are smaller.
- Ask about buyer-side controls: where the data is processed, who can access it, and whether it can be resold.
What semiconductor data is worth
General company data programs publish $100K to over $3M (micro1) and $100K to $4M+ (Handshake AI); semiconductor-specific buyer ranges are summarized in the R&D records guide. These are published ranges, not offers. Note that micro1 says it licenses data only to buyers "based in the U.S. and allied countries", which matters for controlled technology (micro1).
| Raises value | Lowers value |
|---|---|
| Code or settings linked to measured results | Final designs with no history |
| Failed runs, bugs and their fixes kept | Only clean, released versions |
| Several product generations | One project |
| Records you own outright | Data mixed with foundry or customer IP |
| Uncontrolled or properly classified technology | Records whose control status is unknown |
How to prepare
Related: R&D experiment records, CAD and engineering data, data licensing for manufacturers, source code and which training data holds its value.
- Inventory records by system: version control, bug tracker, regression farm, MES, SPC and yield databases.
- Mark which records include foundry, customer or third-party IP.
- Get an export classification for each record set.
- Pick a retired product or older node as a pilot set.
- Agree with counsel on the confidentiality and use limits a license must carry.
Next step
This page states what each rule says and where to read it. It is not legal advice: your counsel decides how the rules apply to your records.
Does your company hold records like these? See what your data is worth.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
Why is semiconductor data valuable for AI?
Because there is little of it in public. ChipAgents writes that semiconductor design has no equivalent public corpus and that the knowledge sits in companies' design files and internal logs.
Can sharing chip data with an AI company be an export?
Yes. Under the EAR, releasing controlled technology or source code to a foreign person in the US is a deemed export. Classify records and know who at the buyer will see them; your export team decides.
Can a fab license process recipes?
Only recipes it owns and can share without breaching NDAs or losing trade-secret protection. Foundry kits and customer designs stay out.
How is this different from licensing R&D lab notebooks?
The R&D guide covers experiment records across chemistry, materials and biotech. This page covers chip design, verification, fab process and yield records and their export and NDA limits.
How much is semiconductor data worth?
General programs publish $100K to over $3M (micro1) and $100K to $4M+ (Handshake AI). Value rises when code or settings are linked to measured results across several product generations.
Sources and further reading
- ChipAgents: collaboration with NVIDIA on agentic AI for chip design
- Hugging Face blog: an open, generated dataset of SoC RTL
- 15 CFR 734.13: export (eCFR)
- 15 CFR 734.15: release (eCFR)
- BIS: Deemed exports
- 18 U.S.C. 1839: definitions, including trade secret
- micro1: Company Data Partnerships
- Handshake AI: data partnerships