Arm offers limited flexibility compared to RISC-V or MIPS. No one wants to spend months negotiating license terms under today's cost and time-to-market pressures.
Arm is under the gun. With the rise of open-source architectures such as RISC-V and now MIPS, winds of change are blowing through in the microprocessor industry.
We hear stories about new licensing practices at Arm since it was acquired by Japan’s SoftBank. Arm’s rivals tell us that they are engaged in many more talks with current Arm licensees who are looking for alternatives.
Product developers no longer have the luxury of two-year product development cycles. And many don’t have the big budgets for licensing fees, often quoted as the huge barrier to entry for system-on-chip (SoC) design.
Arm denies rumors of hikes in its licensing fees. Arm’s latest high-end core historically comes in at a higher price, and that was true with the A76, an Arm spokesman told EE Times.
Nevertheless, sources who are mostly Arm’s past and current rivals say that Arm’s contracts are becoming more complex, resulting in licensees paying more. Arm’s business model is what most in the IP licensing industry now use — an upfront license fee to design with the architecture and then royalties based on the number of chips shipped. Arm never confirms its license fees openly, but they are widely reported to range from $1 million to $10 million.
One of the promises of open architectures like RISC-V and now MIPS is the ability to ‘play’ on top of the instruction set architectures (ISAs) to innovate and develop your own application-specific SoC. Users can customize them without having to pay an upfront fee. Hence, the cost of entry is lower. It’s not completely free because you still have costs of tools, test, and verification, but there is no license fee for the ISA itself. This is what may hurt Arm in the long term.
This is a next-step evolution from the cores that Arc International and Tensilica offered in the late 1990s. Even then, cost of entry and configurability were the arguments that these companies used to pitch. (Full disclosure: I was part of the early team at Arc).
Arc started with a license fee of about $250K for the processor core, as opposed to Arm’s typical $3 million to $5 million at the time. The completely configurable 32-bit RISC core was a huge attraction to customers, as opposed to Arm’s fixed instruction set. Many licensees liked this because they could configure the core to only what they needed, significantly reducing silicon size and power consumption and improving performance. As a result, Arc was successful with large and small companies, including Intel, Fujitsu, Canon, and SanDisk.
When I spoke to Tim Whitfield, Arm’s VP of strategy for embedded and automotive, he argued that beyond processor cores, the company provides many other IP blocks and tools. Arm takes any competition and disruptive technology seriously, and open-source rivals certainly have been a key focus since being acquired by SoftBank. The company has been exploring new business models and markets and brought out new cores to address customer needs.
Overhead and revenue challenges
The corollary is that Arm has a lot of overheads, and it needs to recover its costs. It says that it can take up to eight years from designing an architecture to seeing a product with its customer. And with an engineering team of up to 5,000 across high-salary locations like Cambridge in the U.K. maintaining some 14 architectures, costs can mount. I’m told that developing the Arm 926 core took some 85 man years and the Arm 7 about 27 man years.
Meanwhile, Arm reported that license fees have remained flattish since 2014. In the quarter ending December 2018, it recorded a 34% drop in license revenue compared to the same quarter in 2017, falling from $190 million to $125 million. Clearly, this is partly impacted by its own DesignStart program, which removes the license fee for Cortex-M processors — it signed 38 DesignStart Pro licensees in the same quarter.
According to SoftBank’s financial reports, in the nine months to Dec. 31, 2018, Arm recorded a 27.7% decrease in year-on-year licensing revenue. It blamed the shortfall partly on delays signing new contracts in China in the first half of the year. The value of licenses signed in those nine months was down to $334 million, compared to $462 million in the same period in 2017.
Rising costs and declining revenues could be motivating Arm to be creative with license fees and royalty structures. But that risks confusing customers, pushing them to look at alternatives.
Arm clearly foresaw declining licensing revenues. To counter the trend, it also made some big acquisitions in the IoT space to bring in software-as-a-service–type revenues.
Ultimately, many IP vendors set their pricing based on total cost of ownership and recovery of development costs. Under such a model, one might argue that every customer pays the same — but this is not entirely true.
Those with the deepest pockets can afford an architecture license, giving them freedom to customize cores. Smaller players who use something like the DesignStart program may end up paying significantly less, but they won’t have the same design freedom.
A CPU core is not free — whether it’s from Arm, SiFive, or Andes, all need to make money, one Arm exec argued. At the lower end, risk appetites may be higher and open architectures may be more appropriate. At the higher end, he said that Arm doesn’t see economic models shifting much. To me, that seems to be like saying that those who can afford it won’t get fired if they choose Arm, much like the old adage, “You won’t get fired for choosing IBM.”
It’s a brand-new day
A CEO of a company selling RISC-V IP said that Arm’s days could be numbered. He gives the company no more than five years with its existing business model.
The bottom line is that Arm’s approach offers limited flexibility compared to a more open architecture. No one wants to spend months negotiating license terms under today’s cost and time-to-market pressures.
Arm could choose to compete by giving designers more for their license fee or even take open architectures head-on by eliminating license fees altogether and finding other revenue models, just as Wave Computing hopes to do with MIPS and RISC-V solution providers do as well. But that doesn’t appear to be in the cards right now.
I think that Arm needs to not just evolve but radically rethink how it might compete with the democratization of IP that open architectures are moving us toward. They might have seen off competitors like Arc, Tensilica, and MIPS in the past, but this time, it won’t be so easy.