Third Time's A ChARM? Intel To Manufacture ARM Chips After Atom Failures In Mobile Devices

After a failure to gain significant market share in the mobile market with its x86-based Atom line of chips, Intel is ready to switch strategies and start building ARM chips for other companies. Intel announced at the Intel Developers Forum (IDF) that it has licensed the ARM architecture so it can build 10nm ARM chips for fabless semiconductor companies.

Intel’s (Tragic) History With ARM

Intel has a history with ARM. In fact, the company acquired the StrongARM division from Digital Equipment Corporation (DEC) in 1997, and later renamed it XScale. However, because of various internal conflicts of interest between the ARM and x86 divisions and bad management decisions, Intel failed to turn Xscale into a success. It ended up selling Xscale to Marvell in 2006 for $600 million after having invested billions of dollars. That was only one year before Apple launched the ARM-based iPhone.

Intel had decided to focus on its x86 chips, and if the company was going to go lower-end than even its budget Celeron chip line, it was going to use an x86-chip to do it: Atom. However, Atom was initially targeted at Mobile Internet Devices (MIDs), which were some sort of mini-tablets with higher power consumption than smartphones. MIDs weren’t the success Intel thought they would be.

Fortunately for the company, netbooks started growing fast around the same time. The netbooks kept the Atom line alive and even thriving for several years. At the same time, Intel realized that touchscreen smartphones were an even bigger opportunity, but it didn’t have an ARM division anymore, and Atom's power usage was too high for such a device. That’s how Intel began the multi-year project to reduce Atom’s power consumption with each new generation.

When the 1.6 GHz Atom came out, it had vastly higher performance than the typical 300-400MHz ARM11 chips that were being put inside touchscreen smartphones such as the iPhone at the time. Not only did it have four times more clock speed, it could also execute significantly more instructions per clock as well.

On the performance side, Atom would easily beat the highest-end ARM chip. However, by the time Intel managed to bring Atom’s power consumption in line with the high-end ARM chips, the ARM chips became highly competitive in performance. They were also already too entrenched in the market for Intel to move the needle in its favor.

Not only that, but Intel also had to spend billions of dollars every year subsidizing its Atom chips so they could compete in price, due to the higher cost of building x86 chips on a more advanced process node. In comparison, multiple ARM chip makers could offer low-priced ARM chips for similar performance and power consumption levels, while still turning a profit. Intel’s strategy was simply unsustainable.

Intel tried to switch strategies by starting to license out Atom designs to other chip makers in the same way that ARM licenses out its designs to other companies. Intel licensed Atom to Rockchip and Spreadtrum, but as Intel saw before, Atom was only competitive if it also benefited from Intel’s more advanced process technology. Rockchip and Spreadtrum Atoms would have been built on TSMC’s 28nm planar process, so they wouldn’t have been as competitive in performance and power consumption as similarly priced ARM chips.

Becoming An ARM Chip Manufacturer (And Its Perils)

Although companies like Samsung and TSMC have started eating away at Intel’s process technology advantage since they switched to FinFET technology and 14/16nm process nodes, Intel still possesses an advantage in manufacturing capabilities.

The company is now hoping that by becoming a manufacturer of chips for other companies, it can leverage that advantage to finally become a significant player in the smartphone market, even if not with its own chip designs. This way, Intel becomes not a competitor to Qualcomm, but a competitor to Samsung and TSMC’s foundry businesses.

On the face of it, this seems like a good strategy for Intel. The company has a leading manufacturing process and it’s the kind of business that makes billions of dollars every year for Samsung and TSMC. From that point of view, everything may just work out great for Intel, especially if it can capture Apple as a customer (so far it has gained LG as a customer, but LG doesn’t make too many chips of its own).

However, there are some downsides and pitfalls to this strategy, too.

First off, Moore’s Law seems to be fast approaching its death, as in the point beyond which transistors can’t be shrunk anymore. Intel will have to find some other way in which to significantly improve the manufacturing process every two years or so, or it won’t have much of a competitive advantage compared to other foundries. IBM, for instance, already seems to be ahead of Intel with its 7nm process technology, which uses Silicon-Germanium transistors and EUV lithography. That means it's not a given that Intel will continue to lead in process technology five years from now. Also, if Intel can’t keep its manufacturing prices low, it may find that customers will still flock to Samsung and TSMC.

Second, becoming an ARM chip manufacturer also represents a high risk for Intel’s x86 chip business. Intel has an x86 lock-in advantage in the Windows PC market right now, because Microsoft failed to make Windows RT a success. However, as Microsoft keeps promoting the Universal Windows Platform, and as the architecture-agnostic Chromebooks gain more market share, that lock-in may also disappear. Then, ARM chips can compete head-to-head against Intel in the notebook market, which wouldn’t be good news for Intel’s margins, as it would finally have some real competition in the PC market again.

Becoming an ARM chip manufacturer is probably not a bad strategy for Intel to try, especially since those foundries are already built for its own chips. However, much of the success of this business will depend on how attractive Intel’s offer to customers compares to those of competitors. For instance, just having a 10nm process six months or even one year early may not be enough if it costs customers twice as much. It will also depend on how well Intel can continue to protect its x86 business while making ARM chips that much more competitive.

Lucian Armasu
Lucian Armasu is a Contributing Writer for Tom's Hardware US. He covers software news and the issues surrounding privacy and security.
  • wifiburger
    seriously Atom CPU were and still are pure garbage, any ARM chips runs circle around it,
    can't do anything on that stupid atom cpu
  • rwinches
    To be correct, after the 1.6 Atom chips which I have in netbook, Intel designed a Pentium chip that sips power at around 7 watts max. Binning netted 2 and 4 core Celeron and 4 core Atom chips that has shown up in many laptops running Windows or Chrome . That series continues with smaller nm scale Plenty of stick computers have been sold with Atom chips.

    So Intel has done well with the low power series, the rep of the slow Atom chips that came before the 1.6 and the gap before the newer faster low power series made it hard to get things going on the Atom front. To get the sales moving MS cooperated with 'Windows 8 with Bing' which was Free on under $200 systems which has helped a lot and is continued with Win 10 as you see the with the compute sticks.

    So the great sales of Windows Laptops, ChromeBooks, Mimi PCs, Stick PCs and 2 in 1s these lower binned Pentiums have done well overall.

    My ASUS X205TA 11.6" laptop gives 8 hours runtime and has sold like hotcakes as has Atom models from other makers.

    I turned off most of the whizzbang UI effects and quick start (not needed) stopped or de-installed most live tiles installed Classic Shell. This laptop street price $140-199.

    This Updated Atom Laptop $250 is a great deal.

    Would I like to have a 4 core dual mem channel faster Celeron or new Z8300 Atom 64 bit Win version? Yes.

    So Intel is still in the game and improving it's lower power mobile market.

    I guess if it wont play AAA 3D games it's garbage.
  • bit_user
    I knew it! I was sure they had this in their back pocket, as a hedge.
  • bit_user
    18453201 said:
    seriously Atom CPU were and still are pure garbage, any ARM chips runs circle around it, can't do anything on that stupid atom cpu
    This is going too far. First, newer Atoms are much better than the originals. Secondly, when did you compare the exact same software on both?

    Yes, ARMs have maintained a strong lead in perf/W, but not so much that I'd call recent Atoms pure garbage.
  • RedJaron
    18453201 said:
    seriously Atom CPU were and still are pure garbage, any ARM chips runs circle around it,
    can't do anything on that stupid atom cpu
    Circles in what tasks?

    Older Atoms had problems. Newer ones, since Silvermont, have been quite good. Sadly, they still have the stigma of the older chips and people think Atom is still a dirty word in the CPU game. No, they don't offer performance levels of the i3, but neither does an ARM chip in the same pricing bracket. Newer Atoms in small tablets and 2-in-1 devices are more than adequate for basic use.
  • artk2219
    Ah atoms and contra revenue. What gets me is how beema and mullins were direct and very good competition, yet we almost never saw any small pc's, sticks, or tablets thanks to the rebates intel was giving on atom.
  • jasonelmore
    this news is HUGE! although i'm not to keen on having apple use all of intel's 10nm capacity and delay desktop cpu's which is what happened with GPU's with TSMC. Apple bought all the capacity for the first 6 months, and delayed everyone else from getting any silicon baked.
  • bit_user
    18455746 said:
    ... which is what happened with GPU's with TSMC. Apple bought all the capacity for the first 6 months, and delayed everyone else from getting any silicon baked.
    I'm actually not so sure about that. See, desktop GPUs are necessarily very big, whereas mobile SoCs are comparatively rather small. Big = expensive, especially on a new process, and doubly-so, if yields aren't great. So, I think it wouldn't have been economically viable to fab GPUs at 14/16 nm much sooner than they did. And even now, Nvidia is still struggling to meet demand (due, in part, to yield problems... or so suggests semiaccurate, on the free side of their paywall).

    Now, regarding Intel's 10 nm... let's keep one thing in perspective, here. As far as I'm aware, Intel is still in full control of its fabs. I think it's safe to assume that no one is going to prevent them from using their own fabs as and when they deem it strategically valuable to do so.

    And, they've been fabbing others' chips, for a while. So, it's not even like this is an entirely new situation.
  • jabliese
    Used a couple brands of Atom based handhelds, they are junk.
  • alextheblue
    That's one of the hidden future benefits of UWP. Once you get enough UWP apps, you can release a second ARM variant of Windows (they already have W10M) and this time you'll have an instant library - the UWP apps are architecture agnostic and are cloud compiled. Granted, they don't have any motivation to do that at this time, even for tablets. But they're looking ahead and not letting themselves be locked to Intel.

    This is a wise move if you look at what Intel is doing with Atom. On the low-power tablet front there's no direct replacement for Cherry Trail, hence the lack of a non-Pro Surface 4. Yes Intel has other low-power chips but the Core series is rather expensive compared to Cherry Trail, hence being limited to Surface Pro in their lineup.

    Then again they may not have to jump back to ARM if Zen scales down low enough, a 2 core 4 thread low-power Zen might be able to take the place of the retreating Atom 4C/4T low-cost lineup.