Comment by fc417fc802
1 year ago
This is really confusing me. Is Arm seriously claiming that all design work that makes use of their ISA is derivative work? I feel like I have to be misunderstanding something.
Wouldn't that be similar to the Google v Oracle Java API case except the claim would be even stronger - that all programs making use of the Java API were derivative works of the Java API and thus subject to licensing arrangements with Oracle?
Or similarly, a hypothetical claim by Intel that a compiler such as LLVM is derivative work of the x86 ISA.
That can't possibly be right. What have I misunderstood about this situation?
> Or similarly, a hypothetical claim by Intel that a compiler such as LLVM is derivative work of the x86 ISA.
Intel has been lenient toward compiler implementers, but their stance is that emulation of x86 instructions still under patent (e.g., later SSE, AVX512) is infringing if not done under a license agreement. This has had negative implications for, for example, Microsoft's x86 emulation on ARM Windows devices.
(I'm guessing Apple probably did the right thing and ponied up the license fees.)
Yeah, I did a double-take when I read that too - but that does seem to be the case. From a different article [^1]:
> "Throughout expert testimony, Arm has been asserting that all Arm-compliant CPUs are derivatives of the Arm instruction set architecture (ISA)."
> "Arm countered with an examination of the similarities in the register-transfer language (RTL) code, which is used in the design of integrated circuits, of the latest Qualcomm Snapdragon Elite processors, the pre-acquisition Nuvia Phoenix processor, and the Arm ISA (commonly referred to as the Arm Arm)."
Were they trying to argue that the RTL is too similar to the pseudocode in the ARM ARM or something?? That is absolutely crazy. (Of course, [when we have a license agreement and] you publish a public specification for the interface, I am going to use it to implement the interface. What do you expect me to do, implement the ARM ISA without looking at the spec?)
edit: Wow, I guess this really is what they were arguing?? Look at the points from Gerard's testimony [^2]. That is absolutely crazy.
[^1]: https://www.forbes.com/sites/tiriasresearch/2024/12/19/arm-s...
[^2]: https://www.tantraanalyst.com/ta/qualcomm-vs-arm-trial-day-2...
I would assume (but don't actually know) that compiler authors make extensive use of the (publicly available) ARM as well. But claiming that any associated llvm backends are a derivative work seems absurd to me.
I really feel like I must have misunderstood something here.
Well they did lose the case. Whatever they were contending was clearly incorrect.
Bloomberg article indicates the jury is still out on the question of whether or not Nuvia breached the license. They only agreed that Qualcomm's own ALA covers use of the tech in the event that they happen to possess it.
In other words, Nuvia failing to destroy the designs might or might not have been a breach of contract. At least if I understand all of this correctly. But I feel like I must be missing some key details.
Actually Sun v Microsoft in the late 90s.
Is Arm seriously claiming that all design work that makes use of their ISA is derivative work?
I assume Arm has some patents on the ISA [1] and the only way to get a license to them is to sign something that effectively says all your work exists at Arm's sufferance. After that we're just negotiating the price.
[1] You and I hate this but it's probably valid in the US.
Qualcomm already had a license for ARMs patents. An older licence with much better terms.
ARM has unilaterally cancelled both the Nuvia architecture license agreement (ALA) and the Qualcomm ALA.
Because all Arm ALAs are secret, we do not know if Arm has any right to do such a unilateral cancellation.
It is likely that the ALAs cannot be cancelled without a good reason, like breech of contract, so the cancellation of the Qualcomm ALA must be invalid now, after the trial.
The conflict between Arm and Qualcomm has started because the Qualcomm ALA, which Qualcomm says that it is applicable for any Qualcomm products, specifies much lower royalties than the Nuvia ALA.
This is absolutely normal, because Qualcomm sells a huge number of CPUs for which it has to pay royalties, while Nuvia would have sold a negligible number of CPUs, if any.
Arm receives a much greater revenue based on the Qualcomm ALA than what they would have received from Nuvia.
Therefore the real reason of the conflict is that Qualcomm has stopped using CPU cores designed by Arm, so Arm no longer receives any royalty from Qualcomm from licensing cores, and those royalties would have been higher than the royalties specified by the ALA for Qualcomm-designed cores.
When Arm has given an architectural license to Nuvia, they did not expect that the cores designed by Nuvia could compete with Arm-designed cores. Nuvia being bought by Quacomm has changed that, and Arm attempts now to crush any competition for its own cores.