Will Arm make and sell its own processors? We're gonna go with no
Not for a long while at least, anyway
Analysis Arm has designed its own example of a high-end processor, and is getting samples of the chip made for select customers.
There was a report in The Financial Times the other day with the headline, "Arm makes its own advanced prototype semiconductor." Although the newspaper did its best to cover the nuances of this tale, some might jump to the conclusion that Arm is getting into the chipmaking business. Having spoken to our own industry sources, what's in fact happening here is that Arm is stepping up efforts to showcase its work and make a point to some customers.
For those who don't know, Arm, which is headquartered in the UK and today owned by Japan's Softbank, creates blueprints of chips and licenses those designs to clients. Those customers take the blueprints and typically combine them with custom circuitry to produce complete, finished chip designs, which can be sent off to be fabricated by places like TSMC, Intel's foundry unit, and Samsung. Arm collects royalties as well as licensing fees; it doesn't make the chips itself.
That said, Arm does at times design complete chips and have these made for it; these are just prototypes for customers to try out rather than final products for mass production.
These samples include things like a multicore 64-bit 10nm mobile-friendly test chip made with TSMC; a 28nm Samsung-fabbed part for embedded systems; and a datacenter-grade processor mainly for the likes of Google and Microsoft.
The key point being that these are limited samples of completed chips just for engineers so that they can test their software works as expected on Arm's latest architectures and technologies. Senior staff at Arm we've spoken to previously have openly detailed the availability of this test silicon, and said Arm simply isn't interested getting into making and selling chips by itself. It just doesn't make a lot of sense to them, business wise and strategically.
What's happening now is that Arm has spent the past six months or so designing a complete high-end processor to, as the FT put it, “demonstrate the power and capabilities of its designs to the wider market.” That's a polite way of saying that some at Arm feel that some particular customers aren't taking full advantage of its designs, and so it's making some top-end silicon to show off what the architecture is capable of. Arm may well have also been spurred into such action by the resurgence of AMD, which is in the way of Arm's desire to bag more of the datacenter and personal computing market, and perhaps by RISC-V, an upstart architecture snapping at Arm's heels.
We've heard of one prominent Arm licensee that selects a specific cache hierarchy to save money, as it reduces the processor die size, which shaves off performance, leaving some at Arm frustrated that their technology isn't getting a full and proper opportunity to perform at its best. The high-end Arm-designed processor sample reported on this month is hoped to make a point to current and future licensees that Arm's blueprints can perform well, if done right.
That's done right in Arm's eyes, anyway.
Did we mention the IPO?
This is a time of major change for the Brit chip operation as it prepares to be floated on the stock market by SoftBank in an initial public offering. The Brit biz is said to be exploring new ways of bringing in revenue as part of that move, in order to attract shareholders and keep investors happy.
We've learned that Arm has put together a “solutions engineering” team to produce limited-run chip samples, which are said to be targeting mobile devices and laptops initially.
It's quite obvious to us that Arm is working on silicon that's more akin to a reference design than a product it expects to sell by itself to device makers, otherwise the company is putting at risk its long-standing position as a neutral provider of technology, or the “Switzerland of the processor industry” as it has been dubbed.
Indeed, there are, we're assured, no plans to sell or license the samples. Arm declined to comment on the matter. Its representatives said the business is in a quiet period.
Experts The Register spoke to believe this work is likely linked with the IPO and efforts to generate more revenue, and they too do not believe Arm intends to start selling its own processor silicon into the market.
“It might be some Arm strategy to derive more value from its IP,” said IDC’s senior research director in Europe Andrew Buss. “If you look at the amount of value that is generated throughout the entire supply chain, the licensing fees that Arm gets are really just a tiny fraction of it."
The licensing fees that Arm gets are really just a tiny fraction of it
Arm reported total revenue of $746 million for the quarter up to the end of 2022, for example, compared to the many billions its licensees are together making each quarter from Arm's licensed technology.
If Arm did make its own line of completed chips, the strategy might be compared to Microsoft’s decision to start building and selling its own Surface devices, effectively competing with the OEM system builders that licence Windows for their hardware.
However, Buss noted that Surface was always really intended to shift perceptions, and push vendors into building the kind of systems that Microsoft wants to see brought to market.
“A key issue is, what market would Arm be trying to hit?” Buss said of the comparison. He pointed out that whatever market Arm-compatible system-on-chip (SoC) silicon ends up in, “Arm cores are typically a small part of the overall solution,” with other major component blocks such as cellular modems making up the rest. Arm would have to come up with these in order to deliver a competitive end product; we note that Arm's IP library is pretty extensive, as it includes GPUs, image processors, neural network accelerators, and so on.
Manoj Sukumaran, principal analyst at Omdia, agreed, telling us: “Designing a market ready CPU or SoC is no small effort. There is a lot of testing and validation effort involved in launching a market-ready chip when compared to designing a test chip or validation platform.”
At this point in time, Sukumaran said he does not think that Arm is moving in that direction.
“However, by doing a part of the testing, validation and software porting in-house, Arm might be able to command a higher premium as royalty from its IP customers as it would enable them to lower the time to market,” he opined.
- QEMU 8.0 hatches more support for Arm and RISC-V
- Europe finalizes €43B Chips Act it hopes will help free it from foreign fabs
- Arm liable for $8.5B SoftBank loan if IPO is a no-show
- Stratus ships latest batch of fault-tolerant Xeon servers
“Eventually, this might prompt even the device makers to have their own CPU/SoC, like the way Apple does its own silicon. Device makers like Xiaomi, Lenovo and others could have their own SoC rather than buying from Qualcomm or Mediatek,” he added.
This may or may not be linked with industry talk that Arm wishes to change the way it licences its technology in a bid to drive up revenue.
According to reports, this would still see Arm license its processor designs to chipmakers, but those companies would only be allowed to sell completed processors to device makers that have inked agreements with Arm to pay a royalty based on the overall price tag of their end products. That would allow Arm to take a larger slice of Arm-powered phone, laptop, and server sales, potentially.
Arm is said to have been under pressure from owner SoftBank to raise more revenue ahead of the IPO, now expected to fall sometime before the end of this year. ®