LONDON – The leading customer of processor technology licensor ARM Holdings plc (Cambridge, England) by revenue contribution in 2010, according to Nomura Equities Research, comes as a bit of a surprise. The financial brokerage and analysis house says it was Intel Corp. (Santa Clara, Calif.).
That was my reaction too and it begs the question: what ARM-supplied services, licenses or royalties was Intel paying millions of dollars for in 2010? ARM's revenue in 2010 was $631.3 million so a 7 percent contribution means that about $44 million flowed from Intel to ARM – according to Nomura.
I am scratching my head to make sense of this. Could it be some hard disk drive controller that Intel makes by the bucket-load includes an ARM core? Or is there some ARM core that has made its way into some dusty, otherwise-forgotten corner of an Intel memory controller or image processing block that has ended up in an Intel microprocessor?
And guess what? AMD is on the list ahead of the likes of Infineon, Apple and Qualcomm.
I make no defense of the list which is actually 15 entries long, but even if it is just Nomura's opinion it is an interesting talking point.
And it is what Nomura is telling potential investors in ARM Holdings plc in a 53-page note to clients that re-assesses Nomura's long-term view of ARM as a potential investment. Nomura attributes the customer ranking to "Bloomberg data as of Nov. 3, 2011," without much further discussion, except to say that ARM's customer base is broad and the company is not dependent on just a few customers.
Here's that Nomura/Bloomberg list in full, ranked by percentage contribution to ARM's revenue in 2010.
Intel 7.0% TSMC 5.7% Samsung 5.7% TI 4.6% NEC 3.5% ST 3.5% ZTE 2.8% Broadcom 2.6% AMD 2.5% Infineon 2.4% Apple 2.1% Qualcomm 2.0% Fujitsu 1.9% UMC 1.9% Lenovo 1.8%
It would, of course, be very interesting to see an equivalent list for 2011.
It continues to surprise me that given the ubiquitous presence of ARM cores in all things mobile that their business model does not generate more revenue. Given their out sized importance to many companies you would think that they could charge more.
Their margins are very low indeed. This has advantages of course (low cost provider) but it also means they are quite tiny compared to the big beasts (Intel, Apple etc.) I have argued that ARM should move into making their own solutions and move up the value chain, but that's a different business model altogether.
I think ARM's policy has always been not to take too much out of their partners, so as to be the much lower cost option and to be non-threatening.
Of course after 20 years of following that policy they are now in a good position to start cranking up license fees and royalty rates.
I think initial license fees have gone from about $1 million per core to $5 million but ARM would, I am sure, argue that they are providing much greater performance on ARM15 than on ARM7 so paying more is appropriate. And it is very hard to compare like with like as license varieties have increased.
With regard to royalty rates even if ARM does push them up, their customers go to higher and higher volumes thereby earning discounts so that the average ARM royalty per shipped unit is actually going down.
According to ARM supplied data it was 6.1 cents per unit in 2007, 5.7 cents in 2008, 5.4 cents in 2009 and 4.8 cents in 2010. But over the same period units shipped per year had gone from 2.9 billion to 6.1 billion.
These numbers do not seem stupid but interesting because knowing customers of a company (more, ranking) is always a bit taboo.
I posted this discussion in ARM Based Group community because and added ARM's 2010 Annual Report (http://www.linkedin.com/groups/ARM-Inside-in-proportion-7-85447.S.88812149?qid=680c98d6-b2bc-41ab-8f32-dddd59e137cd&trk=group_most_popular-0-b-ttl&goback=%2Egmp_85447%2Egde_85447_member_88812149%2Egmp_85447).
The Nomura list, if correct, suggests that ARM gets more of its royalties from companies MAKING chips with its cores as opposed to DESIGNING OR SELLING those chips.
That would explain TSMC and Samsung being ahead of say Qualcomm and Apple. And it would imply Intel may be acting as foundry to some ARM-based chips.
But this is all guesswork on my part.
Intel as ARM's biggest customer is 2010 is odd - but foundry manufacture is good call Rick.
Intel maybe making stuff for Apple that includes ARM cores. I wonder who AMD was making stuff for..they were almost (but not quite?) out of manufacturing by 2010 with the creation of Globalfoundries.
I think people forget that ARM has a physical libraries, IO cells libraries and memory compilers and other assorted physical IP as part of the Virage acquisition. Given Intels scale of revenues I am sure thats where the money is coming from. Not ARM CPU licences.
Since Nvidia is mostly fabless, its Tegra family would probably account for much of the TSMC number. With the resolve of the legal dispute between Intel and Nvidia, it is possible that Nvidia has also used Intel for some manufacturing.
It is not appropriate for me to discuss the specific top 15 ranking listed by Nomura. But I thought I could add a couple of clarifications using public data that might be of interest
1) The per-device royalty is typically a % of the device's ASP. One of the main drivers we discussed that caused this to drop in that time period was the rapid growth in MCU shipments where the ASPs are substantially lower (NXP as one example has devices priced at less than $1) than devices in smartphones
2) Of the $631.3M in FY10, ARM reported that the contribution of Physical IP (ARM aquired Artisan as opposed to Virage, the latter being purchased by Synopsys) royalty was $43.8M. Processor royalty for that year was $291.5M. So the majority of our revenue in that year came from processors
Intel biggest customer of ARM?!
Ah a conspiracy theory.
Some of the new low power Intel chips are really ARM cores emulating the X86 instruction set.
This way AMD can be kept in the dark of how Intel has achieved such a low power in their devices.
But with the the Transitive processor emulation technology developed by Alisdair Rawsthorne at Univ of Manchester that would not be an impossibility. The Transitive technology was used by both Intel and Apple in the mid-2000s, and the technology was eventually bought by IBM as I remember.
However, ever I think it was mainly applied to PowerPC-x86 emulations.
I asked Bloomberg to check the Intel figure and they said it was an error and they removed it from their database. Pity Nomura didn't check before publishing. I checked with ARM, and Intel (post Infineon wireless acquisition) is a top 10 customer. But ARM have never given a revenue breakdown by customer so I'm not sure where the rest of the Bloomberg data comes from ...
Apart from being of questionable provenece, this list is from 2010, which was that bit earlier in the rise of ARM microcontrollers.
The main thing would be that microcontrollers sell for much lower ASPs than microprocessors but the royalty rate tends to be the same percentage rate. So NXP would have to sell a a lot of MCUs to start to climb up any ranking. But perhaps they did in 2011?
Not sure how useful this retrospective is, since ARM is just a few weeks away from announcing 2011 (and 4Q11) numbers. However as Peter notes, since this is revenue, you have to look at units multiplied by per chip royalty (which is a % of ASP). More public info - we shared that for 4Q10 (which is based on actual shipments from our silicon partners during 3Q10) that Cortex (A, R and M) was 13% of the 1.8BU shipments for that quarter. We also showed publically that Cortex-A shipments for that quarter was ~55M. So that leaves 234-55 =179M split between Cortex-R and M. MCU shipments for 2010 (which includes some legacy ARM7 and ARM9 devices) was pretty significant overall. But the significantly higher ASPs of apps processors wins out.
Intel's presence in the list may be discredited (see earlier comments in this forum).
TSMC is involved because it makes the chips for many ARM licensees ..and is therefore intimately involved in being licensed itself to integrate ARM cores using ARM physical IP, holding ARM cores and physical IP in its libraries.
Both the foundry and the fabless are held accountable by ARM although the balance of payments depends on the nature and the detail of the licenses.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.