loading
loading
The Arm-Qualcomm verdict made headlines for its $2 award. The finding that matters more is about who controls the compliance tests a custom chip must pass before it can ship. The verdict nobody read closely On Oct. 13, a federal jury in Delaware delivered the second verdict in two years between Arm and Qualcomm. The number that drew the most attention was the damages award: $2, one for each claim Qualcomm won. Both companies declared victory, and IMHO both had some reason to. Arm prevailed on four of the six claims. On the other hand, Qualcomm won a finding that Arm interfered with its business with Meta, plus an additional finding under its Architecture License Agreement that I think deserves far more attention than it received. The jury found that Arm breached Qualcomm’s Architecture License Agreement by withholding two deliverables, ACKs and OOBs. To most people, these acronyms mean very little, which may explain why most coverage moved quickly on to royalties and to the $2. But they describe something we may know very well: the tests a custom processor must pass before anyone can call it an Arm processor. What a compliance kit actually does Arm licenses its technology in two ways. Most customers take Arm-designed cores “as is” and integrate them into their chips, under a Technology License Agreement (TLA). A smaller group, including Apple, Qualcomm, and Nvidia, signed up for an Architecture License Agreement (ALA) that lets them design their own cores executing the Arm instruction set. That freedom comes with an obligation. A custom core has to behave exactly as the architecture specifies, down to obscure corner cases, or software written for Arm will eventually break on it. A licensee demonstrates this with the Architecture Compliance Kit, or ACK. Qualcomm’s filing says its engineers run roughly 48,000 ACK tests. According to Tirias Research, an industry analyst firm that reported daily from the courtroom and drew on Qualcomm’s pre-trial statement of proof, out-of-band (OOB) packages tell engineers which of those tests apply to a specific processor design, and ACK patches fix problems in the tests themselves. Qualcomm alleged that Arm withheld these materials from late 2022 into 2025. Anyone who has run a large verification team will recognize why this matters. When a test fails, the first question is whether the bug lives in the design or in the test. With tens of thousands of tests, a custom microarchitecture and a development effort measured in many person-years, a test bug without a patch can cost weeks of engineering time chasing a problem that does not exist in the silicon. Multiply that across a program, and the delay becomes strategic, especially since, as Tirias observed, licensing decisions made today shape products that reach customers several years later. On the other hand, Arm argued that the underlying ACK remained available, that quarterly releases continued, and that the additional OOBs, patches, and engineering support were not mandatory deliverables under the contract. The jury disagreed on the contract question, but the underlying point is a fair one to debate. Tirias framed it well, asking when withholding support effectively becomes withholding part of the product. The clause worth billions The $2 award made the verdict look surprisingly small. However, Section 8.3 of Qualcomm’s architecture license, as Tirias describes it, provides that if Arm breaches certain obligations and fails to cure within 30 days of written notice, royalty rates on qualifying products can fall to zero for five years. Qualcomm says it gave notice of these breaches in November 2022, repeated it in December, and received a denial from Arm. This means the largest potential remedy in the entire dispute, which Arm itself values in the billions of dollars, rests exactly on a breach about test deliverables. The remedy depends on neither the Meta interference finding, nor access to Arm’s v10 architecture, nor any pricing dispute. Its financial consequences could be major, but only if the court finds that it applies to Qualcomm’s products and that it is enforceable, and both questions remain open. Arm argues that the provision is unenforceable, and whether it applies is for Judge Maryellen Noreika to decide. She heard closing arguments in the parallel bench trial on Oct. 20 and, according to Bloomberg, showed reluctance to get involved in setting licensing terms, calling Qualcomm’s request that the court order Arm to negotiate v10 terms in good faith “very, very drastic.” That request concerns a separate question from the royalty clause, namely whether Arm negotiated v10 terms in good faith, and a decision on one does not necessarily settle the other. Both are now with the judge, and a written ruling will follow. IMHO the outcome is genuinely open, and I would not try to predict it. What is already clear is that a verification issue sits at the very center of the money question. When the referee also plays There is a structural reason why this matters. In the Arm model, one company writes the architecture specification, writes the compliance tests, decides which tests apply to which design, fixes the tests when they are wrong, and stands as the authority on whether a core is compliant. Arm’s own 2022 complaint illustrates the last point, describing Qualcomm seeking Arm’s verification that a new Qualcomm core complied with the architecture so it could be incorporated into a product. From the timing, that core was most likely the Nuvia-derived design that later became Oryon. Other parts of the industry assign these roles differently. PCI Express and USB compliance run through industry bodies, PCI-SIG and USB-IF, where competing vendors share a common test process that no single competitor owns. Arm’s arrangement worked for decades because Arm was the industry’s “Switzerland,” a word that came up in court. A neutral supplier holding the referee’s whistle raises few concerns. But that neutrality is changing by Arm’s own account. Asked at trial whether roughly two-thirds of Arm’s revenue would come from chips by 2030, Rene Haas said yes, adding that Arm is “adding to our business a chip business.” He was equally direct about who should set the price of Arm’s technology in licensing negotiations: “It’s our product. Of course, we should decide.” He said this about pricing, not compliance, but the same logic of ownership applies to both. Competing with your customers is often complicated, but it is not misconduct. Still, once the architecture owner is also a chip vendor, the compliance toolchain also becomes a control point over how quickly a competitor’s custom core can reach the market. I suspect this is why such a narrow contract claim ended up carrying so much weight, and why the FTC may find it relevant too. The FTC is not a party to this case, but it reportedly opened an antitrust investigation into whether Arm is illegally trying to dominate parts of the market by refusing or degrading licenses. The RISC-V contrast RISC-V is often pitched as the answer to dependence on a single architecture owner. RISC-V removes the need for a proprietary architecture license from a single owner, although individual RISC-V cores and implementations can still carry their own intellectual property and licensing constraints. The compliance picture, however, is less mature than the licensing picture. RISC-V International’s own documentation describes its Architectural Compatibility Test suite as a minimal filter, and states that passing it does not mean a design complies with the architecture. Cores must re-pass the tests yearly when the suite changes in order to keep their compatibility listing. The organization has formed a Certification Steering Committee to build a formal certification program, but that work is still in progress. Meanwhile, the RVA23 profile is becoming the practical target: Ubuntu now requires it, and Linux is only now gaining a simple way to report whether a system is RVA23 compliant. A chip designer choosing an architecture today therefore faces a tradeoff. Arm offers mature, rigorous compliance under one company’s control, while RISC-V offers neutral, community-governed compliance that is still being built. Qualcomm’s acquisition of Ventana, and the recently reported addition of the Akeana team, reads to me as a bet that this gap will close, or at least an insurance policy against further disputes with Arm. IMHO, the decisive question for RISC-V in high-performance computing is less about the instruction set and more about who builds trusted compliance and certification at scale, and who can match the broad software ecosystem Arm has built around its architecture. What to watch Three things will tell us how this plays out. The first is Judge Noreika’s ruling on Section 8.3 and on whether Arm negotiated v10 in good faith. The second is v10 itself, since access to a new architecture generation includes access to its compliance kit, so the eventual terms will cover tests as well as royalties. The third is RISC-V certification, where a credible and independent program would change the calculus for every company weighing a custom core. Only the first is a court decision, and its timing is not yet known. The other two will play out over a longer period. The $2 headline will be forgotten quickly, but the finding behind it deserves a longer memory. Whoever writes the test suite controls the path from a custom design to a shipping product, and this trial gave the industry a rare look at what that control is worth. Read also: Arm, Qualcomm Case Goes to Court Over Arm Architecture Licenses Qualcomm Wins Key Claims Against Arm, but Damages Total Just $2 Qualcomm Forecasts Billions in Additional Revenue from New Data Center Solutions The post Two Dollars and a Test Suite appeared first on EE Times.
document fetched 2026-10-24 02:54 · canonical https://eetimes.com/two-dollars-and-a-test-suite