Discussion ARM vs Qualcomm: The Lawsuit Begins

[LightningDust]

A lot of OEMs are going to be unhappy if ARM gets what they claim to want and Qualcomm is actually prohibited from selling Oryon-based ICs.

 

[lakedude]

A lot of OEMs are going to be unhappy if ARM gets what they claim to want and Qualcomm is actually prohibited from selling Oryon-based ICs.

ARM doesn't really want what they claim to want, they just want to be compensated.


"So tell me what you want, what you really really want"

 

[Raqia]

I don’t think there would be concern about current licenses. However, most innovative startups are bought by the major players, who all seem to have ARM licenses, and even if they don’t, they are going to want to make sure there are no potential roadblocks to running with the companies they buy and the IP that they want to use.

That is going to make it a lot harder for future Nuvias to get venture capital to see through new concepts. Those financiers will want the startup to be sellable, and the threat of extortion makes ARM designs less marketable. If they don’t think RISC-V is yet ready to do what they want and Power or MIPS or just that unpalatable, they might be just assed out of their vision.

The new owners of ARM have clearly wanted a short-term and profitable exit in trying to sell to nVidia; now they might be longer term owners so their discounted cashflow guys decided they need to extract more royalties to justify valuation. Agree that they're still quite short sighted in essentially driving existing and potential future licensees to alternatives in making this move at all on weak grounds: smaller licensee won't be able to afford the legal bills and big and small, they all know what terrible terms await them once their current contract expires so long as Softbank owns ARM.

 

[deb57]

I don't understand the semiconductor industry as much but I have been following these companies, this lawsuit, their CEOs interviews for a while now. The only reasonable idea I can come up with on why ARM would continue down the lawsuit is pressure from Softbank to turn it into (not more but) much more profitable company.

The 2021 Nvidia acquisition of ARM was posed at $40B valuation vs today's ARM market cap suggest its $160B (4 times in 3 years) though income or revenues have barely moved. Now the ambitions of Softbank (with 90% stake in AMR) have grown 4x at least and they are willing to change the strategy they had, and even accept playing long term game as ARM currently can't offload the remainder of their stake in the market at the current stock price levels, valuing ARM at $160B.

These current levels are artificially inflated due to short supply as only 10% of ARM stocks are publicly traded, with 2-3% held by companies like Nvidia, Apple for strategic purposes and hence not really publicly traded. They launched v9 with significantly higher contract rates and want all customers to move to v9, they also prefer TLAs, for Apple they had to license ALA till 2040 to get Apple's long-term commitment to the architecture, which is too big company and helps keep ARM architecture as the state of art technology for years to come.

Qualcomm's contract with ARM is till 2026 with option to extend till 2033 (for $1M/yr, which is nothing) and because Oryon designs are based on v8 ALA, ARM will not see much increase in revenues from Qualcomm for years to come, not even v9 ALA level revenues.

My understanding in semiconductor design/manufacturing is limited so please help me understand what constraints ARM will face in adopting either of the below strategies, why they might or might not work. These are just assumed strategies, I am trying to brainstorm are these realistic alternatives for ARM.

1. Softbank still wants to sell ARM: instead of selling today they want to put the revenues and income on growth trajectory, thus attain much higher multiples and sell their stake. This requires new contracts, or different players who might not have a contract with ARM in place already. In which case ARM prefers to either negotiate better contracts or let Qualcomm sink as other companies can pick up the reigns, like Samsung and MediaTek, with AMD and Nvidia already planned to enter ARM PC market in early to mid 2025, similar trend can follow for phones and other IoT. The manufacturing is mostly outsourced to Fabs like TSMC which will not be affected, just instead of producing Qualcomm designs they would product MediaTek, Samsung, AMD, Nvidia, and new player designs.

2. ARM wants to compete and sell their own chips: The chips market has evolved like anything in last 2-4 years, with demand and future potential skyrocketing. ARM's upcoming new cores performance is comparable to Pheonix cores and they can even if it upsets the rest of the market, sell own chips to at least make 3-10x more money in near future and maybe continue. The RISC-V architecture is not mature enough today and will likely take 5-7 years to be usable in mainstream phones and devices with Operating System, App Developers, Chip Designs support - which is too much to ask from other players just to support Qualcomm at this point.

 

[poke01]

RISC-V is rapidly developing. I wouldn’t be surprised if Apple and Qualcomm have RISC-V cores in their labs. A true ISA free of any one company be it Intel and ARM is needed.

This case shows it’s never good to centralised power to one company to dictate ALAs and contracts and future developments.

x86 and ARM are dead man walking ISAs.
Apple is fine for 40 years but they might change earlier. Qualcomm depends on this case.

 

[deb57]

There are certain advantages of ARM ISA over x86, hence moving to ARM helped the industry achieve much greater power efficiency in chips for mobile purposes (including IoT, laptops). Are there benchmarks or otherwise reasons to believe RISC-V can perform at par with ARM ISA, or just better than x86?

 

[gdansk]

There are certain advantages of ARM ISA over x86, hence moving to ARM helped the industry achieve much greater power efficiency in chips for mobile purposes (including IoT, laptops). Are there benchmarks or otherwise reasons to believe RISC-V can perform at par with ARM ISA, or just better than x86?

The primary theory why ARM would be more energy efficient as an ISA is fixed width instructions which every RISC also has (almost by definition now).

Not that RISCV would work for Qualcomm since they have existing customer software to run.

 

[poke01]

There are certain advantages of ARM ISA over x86, hence moving to ARM helped the industry achieve much greater power efficiency in chips for mobile purposes (including IoT, laptops). Are there benchmarks or otherwise reasons to believe RISC-V can perform at par with ARM ISA, or just better than x86?

ARM isn't going anywhere for 10-15 years... but the path to RISC-V is slowly being laid. New papers are on RISC-V and start ups are RISC-V. Its all got to lead somewhere.

 

[Doug S]

RISC-V is rapidly developing. I wouldn’t be surprised if Apple and Qualcomm have RISC-V cores in their labs. A true ISA free of any one company be it Intel and ARM is needed.

Why would Apple care about being "free of any one company"? They have an ALA good for 20 years, so their pricing is locked in. They are able to add their own extensions like they did with AMX (before it became part of the ISA as SME) if they want something ARM won't give them. They have all the advantages of having their own ISA or having an ISA free of any one company, with the advantage of not having to take on 100% of the responsibility for developing the compiler for that ISA and a massive ecosystem of developers familiar with it.

They have as much chance of switching their CPU cores to RISC-V in the next 15 years as Intel or AMD does.

 

[del42sa]

https://www.tomshardware.com/pc-com...pdragon-x-pc-chips-are-almost-entirely-custom

Qualcomm says its Oryon CPU cores have 1% or less of Arm's original technology — cores in Snapdragon X PC chips are almost entirely custom​

 

[FlameTail]

Day 3 of Qualcomm vs ARM trial

Qualcomm vs. Arm trial, Day 3 – Testimonies end, dispute funneled down to two key questions – Tantra Analyst

www.tantraanalyst.com

Arm Squares Off Against Qualcomm: Day 3

Apple CPUs were increasing in performance by 20% per generation while the Arm CPUs were increasing by only 10%.

www.forbes.com

Throughout expert testimony, Arm has been asserting that all Arm-compliant CPUs are derivatives of the Arm instruction set architecture (ISA). Dr. Annavaran provided a tutorial on designing a CPU and SoC to demonstrate that the ISA does not drive a CPU design. According to Dr. Annavaran, you cannot develop a CPU from an ISA.

The conundrum of ISAs and microarchitectures.

Thursday morning, both sides will make their closing arguments before the jury is given its instructions.

Seems like the trial will be over by tommorow.

 

[Tuna-Fish]

Seems like the trial will be over by tommorow.

Jury could deliberate for weeks if they cannot come to agreement, but given the date I suspect they all want to go home sooner rather than later.

 

[poke01]

Why would Apple care about being "free of any one company"? They have an ALA good for 20 years, so their pricing is locked in. They are able to add their own extensions like they did with AMX (before it became part of the ISA as SME) if they want something ARM won't give them. They have all the advantages of having their own ISA or having an ISA free of any one company, with the advantage of not having to take on 100% of the responsibility for developing the compiler for that ISA and a massive ecosystem of developers familiar with it.

They have as much chance of switching their CPU cores to RISC-V in the next 15 years as Intel or AMD does.

Apple won't move on from the ARM ISA anytime soon. Apple Sillicon is barely close to stock ARM cores. Like you say Apple has full control of their designs so it wouldn't matter to Apple for the foreseeable future.

 

[Nothingness]

According to Dr. Annavaran, you cannot develop a CPU from an ISA.

That's such a misleading generic statement, I'm speechless. He's not 100% lying, but that's so approximate he would not pass the recruiting barrage where I work if he insisted on being 100% correct.

An ISA assuming VLIW or some variants of it (e.g. TI c6x, Intel Itanium) or some memory ordering variant or the existence of SIMD instructions, etc. has dramatic impacts on how a CPU is designed. Of course the ISA alone doesn't drive all decisions on the design of a CPU, but it has enough impacts all around the place that disconnecting one from the other leads to an underperforming uarch. People thinking changing an Arm front-end with an x86 one is just plug and play are delusional; I don't mean to say you can't have a uarch doing well on both ISA, it's just that the extra work beyond decoding is massive and has roots in the supported ISA's.

 

[FlameTail]

He's not 100% lying, but that's so approximate he would not pass the recruiting barrage where I work if he insisted on being 100% correct.

The man holds a PhD and was a professor at USC.

https://twitter.com/x/status/1869611388571971830

Meanwhile live update from Day 4 of the trial:

Arm just completed it's closing statement. As mentioned in my summary yesterday, there are 3 (really 2 questions for the jury to answer:
1. Did Nuvia breach the contract (Nuvia #ALA)?
2. Did Qualcomm
breach Nuvia ALA
3. Are Qualcomm products with Nuvia code covered under Qualcomm ALA?

Arm reiterated their same arguments and addressed some of the issues Qualcomm had raised
- Nuvia/Qualcomm needed Arm consent for acquisition
- Breached confidentialality requirements of ALA by sharing it with QC
- Bought Nuvia because of cost savings not performance
- Arm Technology includes Architecture Compliant Core (Remember my comments about it)
- Many things Qualcomm mentioned are irrelevant (e.g. SoftBank ownership, CEO salaries etc .)

 

[Nothingness]

The man holds a PhD and was a professor at USC

That doesn't mean much in particular in the context of a trial (and even less if he works at Qualcomm).

 

[Doug S]

Jury could deliberate for weeks if they cannot come to agreement, but given the date I suspect they all want to go home sooner rather than later.

I have to think the judge already discussed scheduling with them (i.e. who has travel plans and when) and they already know what dates they would be in session during the following two holiday weeks if they aren't able to reach a quick verdict.

I can't imagine a judge would want to start a trial the week before Christmas without making such allowances to prevent the jury from having an incentive to reach a verdict more quickly. The last thing the court wants is for 10 or 11 jurors to reach a decision and the last couple to give in even if they disagree just so no one's kid misses a Disney vacation.

 

[Thibsie]

Apple won't move on from the ARM ISA anytime soon. Apple Sillicon is barely close to stock ARM cores. Like you say Apple has full control of their designs so it wouldn't matter to Apple for the foreseeable future.

Yeah, but Apple changes their arch every 20 years. Clock is ticking...

 

[Tuna-Fish]

Yeah, but Apple changes their arch every 20 years. Clock is ticking...

Every time they have done that they have had a very good reason to. I understand that their ALA with ARM is the broadest and longest-lasting of the bunch, negotiated from a position of strength before Apple jumped fully in on ARM.

 

[jdubs03]

Any predictions?
Based on what I’ve seen I think Qualcomm is at an advantage.

 

[Tuna-Fish]

My prediction is that whichever way the jury decision goes, the loser will appeal and if the loser is Qualcomm they also get a stay on the judgement until appeal is done.

Which means that the decision will just be the end of the opening act of a legal battle that will still last for many years.

 

[digitaldreamer]

- Cristiano explained that Arm's performance lag behind Apple was the main reason for buying Nuvia, and he would be happy to get TLA from Arm if it provided high-performance cores even now.

This appears to be a misleading statement. Qualcomm had a ALA license. It was Qualcomm's performance lag behind Apple that prompted the company to purchase Nuvia. Of course, the TLA license of ARM's cores didn't meet the performance Qualcomm desired and that's one reason they got a ALA license. The other reason was to save money, in the long run, hopefully.

 

[Thibsie]

Every time they have done that they have had a very good reason to. I understand that their ALA with ARM is the broadest and longest-lasting of the bunch, negotiated from a position of strength before Apple jumped fully in on ARM.

Sure but if Arm try to StrongArm (pun intended) Apple at any time, Apple will not hesitate a second.

 

[Doug S]

Yeah, but Apple changes their arch every 20 years. Clock is ticking...

They have because there were good reasons. 68K was a dead end, they moved to RISC with PPC. Then PPC was getting underinvested while x86 was leaving it in the dust. Then they had their own chips that offered not only better efficiency but better raw performance than x86.

For them to ever leave ARM there has to be something better to switch to. They aren't going to arbitrarily switch in 2040 just because someone flipped to a new year on the calendar.

 

[lakedude]

My early impression was that Qualcomm knew what they were doing was not completely according to Hoyle but close enough they thought they could get away with it.

Later after reading the sources posted here it started to seem more like an ARM cash grab.

Reading what both companies say they want (not the stupid stuff, just the money) it seems like there should be room to settle.

Qualcomm wants to avoid spending over a billion extra dollars a year and ARM wants to prevent the loss of only fifty million per year which is peanuts. Lot of room for compromise between those figures.

IMHO they should settle for say a one time 100 million dollar payment to get current and the Nuvia license transferred to Qualcomm plus $50M a year iff Qualcomm only sells server chips or say $100M a year to broaden the license to a larger market. This is way cheaper than the billions Qualcomm was worried about and the possibility of more than ARM would have received had Nuvia remained independent.