Industry news — On 2 June 2026 GlobalFoundries completed its acquisition of the ARC processor IP business from Synopsys, folding it into MIPS to form a portfolio of more than 150 patents serving over 300 IP customers, while Synopsys retained its interface and foundation IP — a reminder that who owns the core you licensed, and under what terms it will be supported, can change in the middle of a programme (MIPS, a GlobalFoundries company, 2 June 2026).
In short: An FPGA IP licence is defined by what you receive (encrypted netlist, obfuscated RTL or VHDL source), how widely you may use it, whether you pay per unit or once, what maintenance covers, which certification artefacts come with it and what happens if the vendor changes hands — and the licensing models differ far more on those axes than on headline cost.
Supplier continuity is a licence term, not a footnote. When an IP portfolio changes hands, the programmes that carry on unaffected are the ones whose contract already answered three dull questions: what did we receive, what may we do with it, and what do we still owe. Most engineers reach that point late — the core has been chosen on merit, the integration is scoped, and then procurement asks what exactly is being bought. This article maps the axes of an FPGA soft-IP licence so that conversation is short. It is deliberately even-handed: there is no best model, only one that fits your volume, your certification burden and your appetite for supplier risk.
- The short answer
- What you actually receive
- What the licence lets you do, and where
- Royalty or royalty-free
- The licensing models side by side
- What a maintenance year actually buys
- Certification, documentation and traceability
- Escrow, continuity and supplier risk
- Evaluation, and the questions to ask
- Where this stops applying
- Frequently asked questions
The short answer
An FPGA soft-IP licence is six independent decisions, and cost is only the visible one. What you receive — encrypted netlist, obfuscated RTL, or plain VHDL or Verilog source — decides whether you can fix a timing failure yourself. Scope decides how many projects, sites and FPGA families the fee covers, and whether a re-spin onto a new device costs again. Royalty or royalty-free decides whether the core is a permanent line in your bill of materials or a single entry in this year’s budget; on a low-volume, high-value programme that is the axis that most changes the answer. Maintenance decides who pays when the standard revs. Certification artefacts decide whether a DO-254 programme can use the core at all. Escrow and change-of-control terms decide what happens if the vendor is acquired. Get those six answered in writing and the comparison becomes arithmetic.
What an FPGA IP licence actually delivers
The delivery format is the first thing to settle, because it silently defines every engineering option you will have for the next five years. Three formats dominate, and they are not interchangeable.
- An encrypted netlist is a post-synthesis representation, locked to a vendor’s toolchain and usually to one device family. It builds and it simulates through a supported flow, and it is opaque: you cannot read it, restructure it, or port it anywhere the encryption keys are not honoured.
- Obfuscated RTL is real HDL with identifiers stripped and structure flattened. It compiles in any tool that reads the language, so it survives a toolchain migration better than a netlist, but debugging it in simulation is close to reading a disassembly listing.
- Source delivery is the unmodified VHDL or Verilog with its constraints, testbench and, where one exists, the C driver source. You can read it, instrument it, re-time it and re-target it.
For an FPGA project the case for source is practical rather than ideological. Timing closure is device-specific: a core that closes comfortably in a fast speed grade may need a pipeline stage inserted or a clock-domain crossing re-balanced to close in the part your board already carries. With a netlist that fix is a support ticket and a schedule risk; with source it is an afternoon. Interfaces rarely match on the first attempt either — a core exposing AXI4-Lite where your fabric speaks Avalon or APB needs an adapter written against real behaviour, not a datasheet. And a certification authority may simply require that the code be readable and traceable.
Logic Design Solutions states its position openly. Across the catalogue — the I2C and SPI macros, the UART, the ARINC 429 controllers — the IP is described as “available in VHDL source code format for ease of customization” and as something that “can be customised by Logic Design Solutions”. Those two clauses are a delivery model: you get the code, and the vendor will make the modification itself if that is faster. The published figures show why it matters — the I2C master macro is specified with a system clock up to 180 MHz and the SPI slave up to 220 MHz, both explicitly speed-grade dependent, which is exactly the kind of number that moves when you change part.
What the licence lets you do, and where
Scope is three separate questions that vendors routinely bundle into one word, and that word means different things at different companies. Ask them separately.
Project scope. A single-project licence covers one named product. A variant for a second customer, a derivative board or a follow-on contract is a new licence — the cheapest entry and the most expensive way to run a product line. A multi-project or company-wide licence prices the portfolio once.
Site scope. A single-site licence covers engineers at one address. That matters more than it sounds: if verification is done by a subcontractor, layout by a partner, or the team is split across two countries, a single-site clause can make ordinary working practice a breach. Ask whether named subcontractors may hold the code and whether it may cross a border — a hard constraint on export-controlled defence work.
Device scope. This is the one that surprises people. Many soft-IP licences are granted per FPGA family, because the vendor’s integration work genuinely is per family: transceiver primitives, hard PCI Express blocks and memory interfaces differ between an AMD (Xilinx-heritage) UltraScale+ part, an Altera Agilex device and a Microchip PolarFire. LDS reflects that in its own catalogue, publishing separately validated entries per family and per evaluation board rather than one universal part number. The question is therefore concrete: if this design re-spins onto another family in three years, does the licence follow it, and at what fraction of the original fee?
Two clauses belong in the same conversation. Derivative rights — if you modify the source, who owns the modification and may you ship it? Sub-licensing — may your customer receive the core inside a bitstream, and as source if they will maintain the board after you? Both are routinely left undefined until they become a dispute.
Royalty or royalty-free
A royalty is a per-unit payment that outlives the project. It is not merely a price structure but an obligation with administrative weight: a reporting cadence, a duty to keep production records, an audit clause letting the vendor inspect them, and often a minimum annual commitment payable whether you ship or not. Royalty-free means one payment, no ongoing count, no reporting. Neither is better; volume and programme length decide.
Low-volume, high-value programmes — avionics, test instrumentation, defence electronics — usually favour royalty-free, and not because a per-unit fee on a few hundred units is punishing. You bid a fixed-price contract years before the first article ships, so a known one-time cost is a number you can put in the bid, while a royalty is a liability to forecast across a production tail that may run fifteen years. You will also be asked to report unit counts on a programme where the customer treats those counts as sensitive. On a small build the administration can genuinely cost more than the royalty.
High-volume programmes often favour the royalty. The up-front fee is lower, which matters when the product may not survive its first year; cost scales with success rather than with optimism; and a cancelled programme has not been paid for in advance. A royalty is the vendor sharing your volume risk, and the up-front fee is priced accordingly.
Three details decide whether a royalty is workable. Is there a cap after which the licence becomes fully paid? Is it charged per device, per board or per end system, and what happens when two instances of the core sit in one FPGA? And does the obligation survive a change of ownership of the vendor, which it almost always does?
The licensing models side by side
Five arrangements are what you will usually be offered. The comparison is qualitative by design — the point is the shape of each model, because the figures move with the core, the family and the volume.
| Model | Up-front cost | Per-unit cost | Modification rights | Re-target cost | Certification fit | Supplier-risk exposure |
|---|---|---|---|---|---|---|
| Evaluation licence | None or nominal | Not for production | None; time- or hardware-locked | n/a | Poor — no artefacts | n/a |
| Per-project, encrypted netlist | Low to moderate | None | None beyond exposed parameters | Usually a new licence per family | Weak unless artefacts supplied separately | High — you cannot maintain the core |
| Per-project, VHDL or Verilog source | Moderate | None | Full within the project; ownership of derivatives set by contract | Often a new project licence, but you hold the code | Strong — readable, reviewable, traceable | Moderate — you keep the last release |
| Company-wide source licence | Highest | None | Full, across every programme in scope | Usually none — any family, any project | Strong | Lowest |
| Royalty-bearing (any format) | Lowest | Per shipped unit, with reporting and audit | Set by the delivery format, not the royalty | Negotiated; royalties may restart | Set by artefacts, not by the royalty | High — survives an acquisition |
Read the table by column, not by row. If certification is the binding constraint, the last two columns decide and the first two are noise; if you are bidding fixed price, the first three decide. Very few teams are optimising the same thing.
What a maintenance year actually buys
Maintenance is a subscription, and it is worth asking exactly what falls inside it, because vendors draw the line in different places. A well-defined maintenance year covers four things: correction of defects against the delivered specification, support for new releases of the FPGA vendor’s toolchain, ports to new device families, and updates when the underlying standard is revised. The first two are near-universal; the last two are where schedule and money live. Tool-version support is unglamorous and indispensable — a synthesis release can change how an inference is handled or reject a constraint syntax that worked last year, so ask which versions are supported today and how quickly a new one is qualified.
The distinction nobody writes down is that a specification revision is not a bug fix. When NVM Express publishes a new revision of its command set, when PCI-SIG defines the next PCI Express generation with a different PHY, or when a new SATA generation arrives — the Serial ATA generations run at 1.5 Gbps, 3 Gbps and 6 Gbps — the resulting work is new development, not remediation. Most agreements say so and most engineers assume the opposite. Settle it: does maintenance entitle you to the next revision, to a discounted upgrade, or to nothing?
Two mechanical questions close the section. If you let maintenance lapse and want to resume, is there a reinstatement fee or back maintenance payable to the date of lapse? And is the renewal price capped, or free to move once you are committed to the core? On what a defect report should contain before it reaches a vendor, see our note on verification services for storage and bus IP.
Certification, documentation and traceability
DO-254 is the guidance an airborne electronic hardware programme is built around — RTCA’s document, published in Europe as EUROCAE ED-80 and titled “Design Assurance Guidance for Airborne Electronic Hardware”, recognised by the FAA through advisory material as an acceptable means of compliance. Commercially it means one thing: a certification programme cannot treat a licensed core as a black box that happens to work. It must show how the core’s requirements were captured, how they were verified, and that the item under review is the item that was verified.
An IP supplier is therefore asked for some combination of: a requirements specification the core traces to, design and verification artefacts including the testbench and its coverage, bidirectional traceability from requirement to implementation to test, configuration management and a firm version identity for the delivered code, a problem-report history, and identification of the tools that produced the deliverables. Design assurance is an evidence exercise, and evidence is a deliverable that has to be named in the contract. If the vendor cannot supply it, certification does not become impossible — the work transfers to you, as verification of a component you did not write and, with an encrypted netlist, cannot read. That is a budget line, and it belongs in the commercial comparison rather than in a schedule discovered later.
Logic Design Solutions publishes DO254 variants of several cores — among them the SPI master and slave macros, the UART and the ARINC 429 controller — and states on those pages that “DO254 design, verification and traceability documentation available on request”. That is a documentation commitment, not a certification credential: no supplier’s paperwork certifies your equipment, it only makes your certification tractable. Programme-level context sits in our piece on designing avionics data recorders on FPGA with ARINC 429 IP, where that controller is specified for all standard rates and up to 2.5 Mbit/s.
Escrow, continuity and supplier risk
Supplier continuity is a contract term you either bought or did not. The June 2026 transfer of the ARC processor IP business to MIPS illustrates it cleanly: the cores did not change, but the counterparty did, and every support commitment, roadmap assumption and renewal conversation attached to them now sits with a different company. A programme with a fifteen-year production life should assume at least one such event.
Source-code escrow is the classical answer: a third party holds the source, build scripts and documentation, released to you on defined triggers — insolvency, discontinuation, or sustained failure to meet support obligations. It works, with three caveats that are routinely skipped. The deposit must be verified, meaning somebody has confirmed that what is in the vault actually builds; it must be updated at every release, or you inherit a version two years behind your own; and the triggers must include the case you actually fear, which is usually quiet abandonment rather than bankruptcy.
A source licence removes most of the need for escrow, one of its underrated commercial arguments: if you already hold buildable code and its testbench, the vendor’s corporate future is a support question rather than an existential one. Three clauses do the rest. Assignment and change-of-control states that your licence and its terms survive an acquisition unchanged. A support commitment period states for how many years the core will be maintained, and an end-of-life notice period gives you time for a last purchase. A right to the last release guarantees you may keep and use the final delivered version whatever happens. None is expensive at signature; all are impossible afterwards.
Evaluation, and the questions to ask
An evaluation licence is a time-boxed right to build and run the core, and it comes in three shapes: a time-limited or hardware-locked bitstream, a feature-reduced version of the core, or a complete reference design for a named evaluation board. The third is the most informative, and it is what a serious vendor tends to offer — LDS publishes its 10 GbE MAC as separately validated designs on specific boards including the KC705, VC707, ZC706, KCU105, ZCU102 and ZCU106, and on the Microchip PolarFire evaluation kit.
Be clear-eyed about what that proves. A reference design proves the core builds and runs on that board, with that tool version, in isolation, at low utilisation. It does not prove it will close timing inside your design, at your speed grade, sharing your clock network with your other cores. The evaluation worth the fortnight is the one you run yourself: instantiate the core in your project with your constraints and read the implementation reports rather than the demonstration. If what you are evaluating is a netlist, run that experiment before signing, because afterwards the fix is no longer yours to make. Vendor selection more broadly — track record, engineering depth, support responsiveness, references — is covered in our buyer’s checklist for choosing an FPGA IP vendor; this page covers only the commercial terms once the vendor is chosen.
Questions to put to any IP vendor before signing
- What exactly is delivered — encrypted netlist, obfuscated RTL or plain VHDL or Verilog source — and are the testbench, constraints and driver source included?
- Is the licence per project, per site or company-wide, and may named subcontractors hold the deliverables?
- Does it cover one FPGA family or any family, and what does a re-target to another vendor’s device cost?
- Is there a per-unit royalty? What triggers it, is it capped, what must be reported and how often, and what does the audit clause allow?
- What does a maintenance year include, and is a revision of the underlying standard a maintenance item or a new purchase?
- Which certification artefacts are available, on what terms, and when in the schedule?
- If I modify the source, who owns the modification, may I ship it, and is the core still supported?
- What happens on acquisition, discontinuation or insolvency — is there escrow, is the deposit verified, and do I retain a right to the last release?
Eight questions, all answerable in writing, and the answers are the licence. If a vendor is uncomfortable putting any of them on paper, that is itself information. The same discipline applies to the engineering relationship, set out in what an IP-focused design house delivers, and to the category basics in soft IP versus hard IP.
If you are scoping a core and need the commercial terms to fit a real programme, describe the protocol, the target device family, the production volume and the certification context, and we will tell you plainly what is possible.
Where this stops applying
This is a map of the axes an FPGA soft-IP licence is negotiated on. It is not legal advice, and it has clear edges:
- The contract wording governs, not the model name. “Site licence”, “project licence” and “company-wide” are not standardised terms. Read the definitions clause before comparing two quotations.
- It covers soft IP delivered as RTL or netlist for FPGA. ASIC IP, hard macros, foundry-tied physical IP and processor architecture licences are commercially different animals with their own norms.
- Open-source cores are out of scope. Permissive and copyleft licences raise separate questions — provenance, patent grants, obligations on redistribution — that this comparison does not address.
- No vendor’s commercial terms are described here, including ours. The only Logic Design Solutions facts stated above are those published on our product pages: VHDL source delivery, customisation by us, DO254 documentation available on request, and the named validation boards.
- Certification requirements are set by your authority and your assurance level, not by an article. The artefact list above is the usual shape of a request, not a compliance checklist, and it does not substitute for your own plan for hardware aspects of certification.
In summary
An FPGA IP licence is six decisions wearing one price tag. Delivery format decides whether a timing failure is your problem or a support ticket, and for FPGA work source is what keeps re-timing, re-targeting and certification review in your own hands. Scope decides whether the fee covers the next project, the next site and the next device family. Royalty against royalty-free is settled by volume and programme length rather than preference: low-volume, long-life programmes want a known number they can bid, high-volume ones want cost that scales with success. Maintenance is worth exactly what its definition of a defect says, and a revision of a standard is not a defect.
The two clauses most often omitted are the ones that have become expensive: certification artefacts, which are a deliverable rather than a favour, and continuity — escrow that is verified and updated, an assignment clause that survives a change of ownership, and a stated right to the last release. Ask the eight questions in writing, and what began as a procurement obstacle becomes a comparison you can actually make.
Frequently asked questions
What is the difference between a royalty-free and a royalty-bearing FPGA IP licence?
A royalty-free licence is paid once and carries no per-unit obligation afterwards. A royalty-bearing licence has a lower up-front fee and adds a payment for every unit shipped, plus a duty to report unit counts, keep production records and accept an audit clause, and often a minimum annual commitment. Neither is inherently better: royalty-free suits low-volume programmes that must bid a fixed price years in advance, a royalty suits high-volume products where cost should scale with success rather than with a forecast.
Do I need source code, or is an encrypted netlist enough?
It depends how much risk you are willing to hand to the vendor. A netlist is enough when the core drops into a supported family at modest utilisation and nothing about the standard or the device will change. Source becomes necessary when you may have to insert a pipeline stage to close timing, adapt an unusual interface, port to another FPGA family, or let a certification authority read the implementation. With a netlist, each of those is a support request with a schedule attached.
Why does re-targeting a core to a new FPGA family often cost again?
Because the integration work genuinely is per family. Transceiver primitives, hard PCI Express blocks, clocking resources and memory interfaces differ between AMD, Altera and Microchip devices, and between generations of one vendor’s parts, so a validated core is validated against a specific set of them. Many licences are therefore granted per family. Ask before signing what a re-target costs, whether it is a fresh licence or an upgrade, and who performs the port.
What does a DO-254 programme need from an IP supplier?
Evidence, in a form that can be traced: a requirements specification the core maps to, design and verification artefacts including the testbench and its coverage, bidirectional traceability from requirement to implementation to test, configuration management with a firm version identity, a problem-report history, and identification of the tools used. If the supplier cannot provide it, the verification work transfers to your programme — a budget line rather than an impossibility, but one to price before the contract, not after.
Is source-code escrow still worth it if I already have the source?
Much less so, which is one of the quieter arguments for a source licence. Escrow exists to give you buildable code if the vendor disappears; if you already hold the code, its testbench and its build scripts, the remaining exposure is support rather than survival. What still matters is an assignment clause that keeps your terms intact through a change of ownership, a stated support commitment period, an end-of-life notice period, and an explicit right to keep and use the last delivered release.
Can an evaluation licence prove that a core will close timing in my design?
No, and treating it as proof is a common and costly mistake. A reference design on a named evaluation board proves the core builds and runs on that board, with that tool version, in isolation and at low utilisation. Your design has different constraints, a different speed grade, a shared clock network and other cores competing for routing. The evaluation worth running is your own: instantiate the core in your project and read the implementation reports rather than the demonstration.
- MIPS, a GlobalFoundries company — GlobalFoundries completes acquisition of the ARC processor IP business from Synopsys, 2 June 2026: the transaction cited above; a worked example of an IP portfolio changing owner mid-programme.
- RTCA — DO-254 / EUROCAE ED-80, Design Assurance Guidance for Airborne Electronic Hardware: the guidance an avionics hardware programme is structured around, and the origin of the artefact and traceability expectations placed on an IP supplier.
- Federal Aviation Administration — Advisory Circulars library (AC 20-152A, airborne electronic hardware): the FAA advisory material recognising DO-254 as an acceptable means of compliance.
- NVM Express, Inc. — NVM Express specifications: the published revision history behind the maintenance question: a new command-set revision is new development, not a defect fix.
- PCI-SIG — PCI Express specifications: the PCI Express generations; a PHY generation change in a host core is a port, not a bug fix.
- SATA-IO — Serial ATA International Organization: the body defining the Serial ATA generations at 1.5 Gbps, 3 Gbps and 6 Gbps referenced above.
- IEEE 802.3 Ethernet Working Group — IEEE 802.3 standards development: IEEE 802.3ae, the 10 Gbit/s Ethernet amendment the 10 GbE MAC cores cited in the evaluation section are built to.
- Logic Design Solutions — ARINC 429 IP (M429GEN) product page: source for the statements quoted above: VHDL source code delivery, customisation by Logic Design Solutions, DO254 design, verification and traceability documentation available on request, and the 2.5 Mbit/s ARINC 429 data rate.
- Logic Design Solutions — Ethernet MAC 10G SFP — validated board variants: source for the named evaluation boards, each published as a separately validated variant of the same core.
