Intel, Arm, 3D FinFETs & mobile computing’s future

Bernard Cole

Intel Corp.’s move to a 22 nm Trigate vertical 3D FinFET (Fin-shaped field effect transistor) structure for use in nextgen CPUs means that for the first time the company may be a jump ahead of ARM Ltd. and its licensees in many embedded and mobile apps, and may also be able to cut off a move by ARM into PCs and servers.

Although Intel is hoping that the new FinFET structure will give it a big lead – maybe as much as three years - on its competitors, the move into this new and still largely untried transistor structure is definitely a high stakes gamble on Intel’s part. For one thing, ARM and its partners may not be far behind, with several ARM test chips implemented with FinFETs, with semiconductor foundries used by Intel’s competitors already working on the necessary tooling. One, Taiwan Semiconductor Manufacturing Company (TSMC), is planning the rollout of an initial FinFET process at the 14-nm node level in the near future.

The impression from coverage in the N.Y. Times, television network channels, and even some electronics industry news sources is that Intel has caught the industry flatfooted and that its success is assured with its “revolutionary” 3D structure. But not so fast. Something that has been under investigation for at least a decade by many companies is hardly revolutionary.

While Intel is breaking new ground in being the first to use this transistor structure in 22nm integrated circuits, it is not unique in its expertise in FinFET CMOS. At least half a dozen companies have been actively involved in FinFET development.

Since the year 2000, according to the U.S. Patent Office, there have been about 500 patents applied for or issued on vertical trigate FinFET transistor structures. Of these, Intel Corp. is listed as assignee on only 40 or so. Same results came from a search of databases with links to technical papers submitted to journals and conferences: about 450, with about the same ratio, based on authorship. And a rough count of the FinFET assignees other than Intel included about 10 to 15 companies that are licensees of the ARM architecture or one of several other CPU architectures, including MIPS and Power CPUs.

One argument in favor of Intel’s gamble succeeding is that it has an unmatched in-house nanometer-scale semiconductor process capability, now primed to start building CPUs and SoCs using the FinFET structure within a few months.

While Arm’s business is the licensing of its architecture and so does not have any in-house fabrication capabilities, many of its partners do.

Those that don’t have in-house capabilities use one of several independent fab houses to build their devices. Two of those fab houses - IBM Microelectronics and TSMC - have written FinFET technical papers and submitted patents. No significant activity was found from a third one – the recently formed Global Foundries. But it has entered into a strategic, long-term partnership on sub-22 nm CMOS scaling with IMEC, the European research institute, which has had significant FinFET patent and technical paper activity over the last ten years..

Another factor which would seem to favor Intel is the size of the financial commitment it can throw at making FinFETs work, based on the huge sales and income muscle it generates: an estimated $40 billion in sales each year, compared to Arm Ltd., with estimated annual revenue of just under $500 million from licensing its processor architecture. But Arm Ltd. does not have to depend on its own resources alone; it also has numerous licensees, some of whom can match Intel in annual sales and income.

The very fact that there are numerous licensees each with their own corporate strategy for taking on Intel has also been cited as an argument in favor of Intel’s eventual success: the “too many cooks spoil the broth” charge. Microsoft pulled out this old chestnut when Linux started threatening its dominance. But Linux, with thousands of developers participating in its development, has not done too shabbily. Nor has open-source Android.

And in spite of the number of corporate cooks under the ARM licensee umbrella, that architecture has done well compared to the Intel CPUs, especially in the balance between low power and high performance that current Atom X86 offerings have still not been able to achieve. Indeed, the success the ARM licensees have had in keeping the power down without sacrificing performance is one reason Intel has had to move to its bleeding edge FinFET strategy.

What makes me particularly optimistic about the Arm architecture in the face of Intel’s massive FinFET assault has to do with what I call the “first to be second” rule. While being first to market with a product or technology would be a best-of-all-possible-worlds outcome, being second is almost as good because that company can learn from the mistakes of the first company and avoid them. The interesting competition to watch is who among the Arm licensees will claim the title of “first to be second” at the 22 nm node.

For all these reasons, I don’t think Intel is likely to do better than continue to be runner-up in building the CPUs that will power the next generation of mobile and embedded computing devices, given how successful the Arm licensees been so far against Intel and their well researched understanding of what it will take to get to the 22 nm node and below.