The past two weeks have been a busy – if not tumultuous – period for Intel. Driven by continued challenges in various semiconductor markets, culminating in weaker-than-desired earnings in the most recent quarter, Intel has set out to change their direction and refocus the company towards what they see as more lucrative, higher growth opportunity markets such as data center/server markets and cellular (5G) connectivity. To get there, the company is making changes to both their product lines and their head count, with the goal in the case of the latter to cut 11% of their workforce by the middle of next year.

Today’s big news out of Intel is along these lines, and with strategy and workforce news behind them, we have our first announcements on product changes that will come from Intel’s new strategy. In a report on Intel’s new strategy published by analyst Patrick Moorhead, Moorhead revealed that Intel would be radically changing their smartphone SoC plans, canceling their forthcoming Broxton and SoFIA products and in practice leaving the smartphone market for at least the time being.

Given the significance of this news we immediately reached out to Intel to get direct confirmation of the cancelation, and we can now confirm that Intel is indeed canceling both Broxton (smartphone and tablet) and SoFIA as part of their new strategy. This is arguably the biggest change in Intel’s mobile strategy since they first formed it last decade, representing a significant scaling back in their mobile SoC efforts. Intel’s struggles are well-published here, so this isn’t entirely unsurprising, but at the same time this comes relatively shortly before Broxton was set to launch. Otherwise as it relates to Atom itself, Intel's efforts with smaller die size and lower power cores have not ended, but there's clearly going to be a need to reevaluate where Atom fits into Intel's plans in the long run if it's not going to be in phones.

For the moment Intel’s announcement leaves some ambiguity in their larger mobile plans – where does the remaining Apollo Lake fit into the picture for tablets, if at all? – but for now we have a very clear picture of the smartphone SoC market, and how Intel will no longer be a part of it.

Intel’s full statement:

Intel is accelerating its transformation from a PC company to one that powers the cloud and billions of smart, connected computing devices. We will intensify our investments to fuel the virtuous cycle of growth in the data center, IoT, memory and FPGA businesses, and to drive more profitable mobile and PC businesses. Intel delivers a broad range of computing and connectivity technologies that are foundational to this strategy and that position us well to lead the end-to-end transition to 5G. Our connectivity strategy includes increased investment in wired and wireless communications technology for connecting all things, devices and people to the cloud, and to power the communications infrastructure behind it. We re-evaluated projects to better align to this strategy.

I can confirm that the changes included canceling the Broxton platform as well as SoFIA 3GX, SoFIA LTE and SoFIA LTE2 commercial platforms to enable us to move resources to products that deliver higher returns and advance our strategy. These changes are effective immediately.

Update 4/30: After publication, Intel sent along a second message clarifying that Broxton is canceled for both "phones and tablets," as the latter was not mentioned in their original message.

Smartphone SoCs: The Path so Far

Anyone following Intel’s exploits in the smartphone space over the last few years has been watching them with interest on product, timeliness and execution.

We’ve interviewed and appeared on video speaking with Aicha Evans, Intel’s current corporate Vice President of the Communication and Devices Group, whose large enthusiasm, energy and mantra of time to market has steered Intel over the past few years into the mobile scene, after bashfully missing an early entry. In that time, Intel has invested many billion dollars in both SoC and modem development to claw a market from the slew of ARM-based solutions in the wild. Aside from having a process node advantage during that time, Intel has had to redevelop its microarchitecture products and radio business into something that could be efficient, performant and price competitive, all the while maintaining the high margins Intel's overall business requires. Particularly in the radio business, the bread and butter of the CVP, Intel acquired and merged several companies to expand its radio portfolio, including the CDMA assets of VIA Telecom announced as recently as Q4 2015, as well as Infineon Wireless (modem/RF) and Silicon Hive (ISP).

As admitted by Intel, the first few generations were rough, either resting on their laurels or not having a complete solution. Earlier this decade Intel used a ‘contra-revenue’ strategy, investing into OEMs that would buy their chips, causing operating losses for the mobile division of $3.1 billion in 2013 and $4.2 billion in 2014 with a much lower revenue stream. Intel subsequently combined the financial reports of their mobile and consumer PC businesses into a new Client Computing Division, bringing all CPU/SoC development under a single roof but also obfuscating the investments and losses behind a high performing, high margin part of the company.


(Image Courtesy Tweakers.net)

Thus Intel’s big wins in the smartphone space have been rather limited: they haven't had a win in any particularly premium devices, and long term partners have been deploying mid-range platforms in geo-focused regions. Perhaps the biggest recipient has been ASUS, with the ever popular ZenFone 2 creating headlines when it was announced at $200 with a quad-core Intel Atom, LTE, 4GB of DRAM and a 5.5-inch 1080p display. Though not quite a premium product, the ZenFone 2 was very aggressively priced and earned a lot of attention for both ASUS and Intel over just how many higher-end features were packed into a relatively cheap phone.

Meanwhile, just under two years ago, in order to address the lower-end of the market and to more directly compete with aggressive and low-margin ARM SoC vendors, Intel announced the SoFIA program. SoFIA would see Intel partner with the Chinese SoC vendors Rockchip and Spreadtrum, working with them to design cost-competitive SoCs using Atom CPU cores and Intel modems, and then fab those SoCs at third party fabs. SoFIA was a very aggressive and unusual move for Intel that acknowledged that the company could not compete in the low-end SoC space in a traditional, high-margin Intel manner, and that as a result the company needed to try something different. The first phones based on the resulting Atom x3 SoCs launched earlier this year, so while SoFIA has made it to the market it looks like that presence will be short-lived.

Overall, Intel’s strategy of ‘Time To Market’ in order to generate revenue in a fast paced market makes sense - if you are late, then you are behind on performance, efficiency, and no-one will buy the chips. However, TTM has drawbacks if the chip comes without the features it needs, and the end result has seen Intel always play catch-up in one form or another, hoping that their strategy would encourage customers. Intel got serious about mobile, but it would appear it hasn't been enough.

Intel's Leaving the Trail: Broxton & SoFIA Cancelled

With Intel announcing the cancelation of their entire suite of smartphone SoCs, this has a significant impact on the company's overall strategy. The next generation of Intel's in-house mobile SoCs, Broxton, was lined up to use Intel’s newest generation 14nm Atom core, Goldmont. Goldmont has already been announced at IDF Shenzhen this year as part of the Apollo Lake netbook/low-cost PC platform, but we have been expecting it to arrive as part of a few handsets this year. Despite the fact that we assume Broxton should be in the final stages of silicon development and less than a few months out, the official word from Intel today is that the Broxton commercial platform has been cancelled for both smartphones and tablets, effective immediately. The resources working on the Broxton platform are being moved to areas within the company that offers better returns on investment and are more aligned with Intel’s connectivity (read: 5G) strategy.

Comparison of Intel's Atom SoC Platforms
  Node Release Year Smartphone Tablet Netbook
Notebook
Saltwell 32 nm 2011 Medfield
Clover Trail+
Clover Trail Cedar Trail
Silvermont 22 nm 2013 Merrifield
Moorefield
Bay Trail-T Bay Trail-M/D
Airmont 14 nm 2015 'Riverton' Cherry Trail-T Braswell
Goldmont 14 nm 2016 Broxton
(cancelled)
Willow Trail
(cancelled)
Apollo Lake
Apollo Lake

The other side of this news is the cancellation of the SoFIA 3GX, LTE and LTE2 commercial platforms as well. SoFIA as a platform had missed its original targets, was delayed (some analysts suggest up to a year), and in the end was developed through agreements made with RockChip and Spreadtrum to manufacture some of the SoFIA SoCs for those markets using a less expensive process node but also using the expertise of these two bulk SoC sales companies. We were expecting SoFIA with Intel’s 2nd generation LTE, as well as the next microarchitecture in SoFIA, to be announced this year. As of today’s email exchange with Intel, these programs are now cancelled, again effective immediately.  At this point details on how the arrangements with RockChip and Spreadtrum are unclear (Intel declined to comment).


One of Intel and Rockchip's current SoFIA SoCs

The Road Ahead for Intel

Intel’s announcements over the past week have included layoffs of 12000 staff, but also a clarification of Intel’s future strategy. Among those five focal points include the Cloud, the Client business, Memory and FPGAs, R&D through Moore’s Law, and 5G Connectivity. These five areas are all high margin, high grossing and high volume market segments. Sometimes an introspective look and an internal refocus on the core strengths is a good thing, depending on how your competitors are doing, but that means shedding parts of the business that don’t meet those expectations.

For the moment at least, Intel is out of the SoC side of the smartphone market. This will allow ARM architecture based SoCs to absorb the remaining market share they didn’t have already.

What's less clear at the moment is whether this will also impact the low-cost/non-premium tablet market, as embodied by products such as the Surface 3. In their updated statement, Intel has told us that Broxton is cancelled for both "phones and tablets." Our current understanding is that Broxton is the SoC at the heart of the Willow Trail platform – the successor to the widely used Cherry Trail-T – but at this time Intel has not explicitly confirmed whether this is in fact Willow Trail, or if Broxton's tablet variation represented another platform altogether. Though regardless of what happens with traditional tablets, we'll continue see Intel in more premium tablet-like devices such as 2-in-1s (e.g. Surface Pro) via Apollo Lake and the Core processor lineup, as Intel has previously identified convertable devices as a growth market for the company.

Update 5/02: In a newer statement, Intel has confirmed that Apollo Lake will be offered to tablet manufacturers. At this point it's not clear what the tradeoffs are for that versus Willow Trail, and whether Apollo Lake is suitable for all types of devices that the current-generation Cherry Trail has been used in. But this does mean we will see tablets using the Goldmont CPU core, while Intel Intel will flesh out the rest of their tablet SoCs with Core-based parts. Intel will also "continue to support" their tablet customers with Bay Trail, Cherry Trail, and SoFIA parts.

Also not discussed in greater detail is Intel's future plans for their overall Atom lineup. With Apollo Lake announced just earlier this month, it's clear that Intel's Atom efforts have not been cancelled entirely. We will still see the new 14nm Goldmont cores appear in low-cost PCs under Apollo Lake, most likely in several 11-to-13 inch high volume devices. However for the moment there is not an Atom core on Intel's roadmap beyond Goldmont.

Finally, despite all of this one key target for Intel will be the rest of the discrete modem market, which is currently Qualcomm’s domain, and the late 2015 acquisition of VIA Telecom’s CMDA assets will help. To put some perspective on this, two things: Intel recently hired Dr. Renduchintala, former Qualcomm VP of Mobile, to head up the client business, as well as Amir Faintuch, also formerly Qualcomm, to co-manage Intel’s Platform Engineering Group. Secondly, at Mobile World Congress 2016 in February, Aicha Evans said that she wanted a big contract in 2016, otherwise we might not see her in 2017.

Source: Intel, tip-off from Patrick Moorhead via Forbes

POST A COMMENT

157 Comments

View All Comments

  • zodiacfml - Saturday, April 30, 2016 - link

    If they are thinking what I'm thinking, I think they are about to buy another company. Reply
  • watzupken - Saturday, April 30, 2016 - link

    I feel Intel has been complacent and joined the party way too late. Reply
  • nemo183 - Saturday, April 30, 2016 - link

    Look, there's much informed debate going on here, so I am reluctant to look stupid in asking a question that will seem barboursly naive at the very best. But since it's causing me a considerable amount of grief, I'll have to get over the social embarrassment, given this opportunity to become better informed. So, here goes!

    Isn't this really the formal announcement of the death of 'Moore's Law of Computing'? Gordon Moore always said it would reach a natural conclusion, an end-point, and this is it. Accepting there are still performance gains to come, both from architecture tweaks and software efficency improvements, hasn't semiconductor technology come to the end of the road?

    Since 1965 it has probably been one of the most important drivers of the economy and even today is constantly referred to as the force that will provide the perfomance gains required to power the future for just about anything from the Fourth Industrial Revolution up to national and global security.

    There is much talk of the development of different forms of computing - quantum, DNA and so on - the list is endless, but seemingly no general consensus on what, when and by how much. There is certainly nothing in the pipeline that would allow for a seamless transition from one technology to another.

    Yet few people mention this or appear concerned. Is this because it's just taken taken as a 'given', accepted by everyone and too depressing to talk about? I can think of many more options, but I won't list them, because I'd like to appeal to this community for their own informed views, for which I'd be very grateful and better informed by.

    There is a very serious reason for me asking this. Over the last few years I've been a very junior member of a research project (basically, I buy, lick the stamps and post the letters of the clever people) that has had access to some of the cleverest people alive, which has resulted in a view of the future that is alarming different from what most people expect. One thing is certain - on this topic, nobody agrees, and most are unwilling to state their position due to insufficient information.

    Actually, that's not strictly true. There is complete consensus on one scenario. Any talk of a technological singularity, in the terms in which it is generally understood, is ludicrous at best. In addition, the promoters of the idea are talked of using language ruder than I've ever heard used before.

    I'd like to ask any member of this community to comment or reply in the knowledge that it serves a serious purpose. I should be clear that the ideas expressed in this post are not ones I hold personally, simply because I don't have the knowledge or experience required to do so. Please be as direct as you wish!
    Reply
  • abufrejoval - Saturday, April 30, 2016 - link

    First of all, Moore's law is a little more complex than compute performance will double every x months. There are some important economical factors underneath, which are currently changing, and one of those is *direct* competition, which is no longer really happening. Instead we see an evolution into niches and each niche guarded by its own gorilla.

    One of the cornerstones of Moore type progress was that the square increase in usable transistors due to a linear process density increase could be translated directly into computing power.

    That computing power gain curve has hit serious trouble with things like the Gigahertz wall and the resulting core number explosion hasn't resulted in linear gains for many consumer workloads.

    It's probably more appropriate to say that Moore's law is running out of steam, especially as the cost of shrinking is rising sharper than the compute power gains (e.g. ELV vs. multi-patterning/multimasking).

    But what's coming to an end before Moore's law is general purpose compute, one architecture for all jobs, which is what x86 stands for most.

    We've seen this a bit with GPUs taking a huge chunk out of general purpose CPUs both in games and HPC and I believe we are going to see it also at the other side with memory intensive workloads, where transporting data to the CPU for processing is starting to use more energy than the processing itself.

    The logical escape for those compute scenarios is to move some of the compute towards the RAM itself and take advantage of the vast parallelism available within those row buffers.

    But you won't see Intel pushing that just as you didn't see Intel pushing non x86 GPUs until it was about hurting AMD and Nvidia (today more than 50% of a desktop "CPUs" silicon real-estate is actually "GPU contra revenue" only targeted at keeping the competition starved from revenue).

    Intel wants an effective monopoly to maintain their margins and they are fighting hard and dirty to achieve that. Their natural escape valve, the one they have used for decades, faster general purpose CPUs, is getting both more difficult to build and less and less successful in creating value, because special purpose architectures made possible by web scale companies like Amazon, Google and Facebook (and the ARM mobile space), byte (intentionally misspelled ;-) off large chunks of their general purpose CPU slice.

    So do not mistake Intel's trouble as a sign that compute will stop to evolve further.

    It will just evolve in places an in ways which are less Intel and there is some inertia as the entire IT industry will have to turn to new architectures.

    The x86 may well become another mainframe, but won't be the one and only architecture Intel wanted it to be, including graphics (remember the Larrabee?) or including the IoT space (remember the Intel Quark?)
    Reply
  • KAlmquist - Saturday, April 30, 2016 - link

    Good point. Most of the increase in the computing power of Intel desktop chips over the past few years has been in the GPU. So Moore's law is still operating, but not in a way that helps people who are going to buy a discrete GPU in any case. Reply
  • TheOtherBubka - Saturday, April 30, 2016 - link

    The materials and technology path forward in computing power to keep the same pace as what "Moore's Law" has yielded over the last 40+ years is a bit difficult even over the next 10 years, let alone almost half a decade. This is simply due to the laws of physics that Moore himself knew all those years ago. Hence, 'the natural conclusion' you refer to.

    So what are the challenges? Well, there are 2 major ones. The first is the light source to enable the lithography. EUV is coming along, but the development has not been difficult. Although the research world has options such as e-beam writers and synchotrons, these techniques are not 'fast' techniques relative to current UV light sources and production. They are also expensive. Hence the reason that companies are pushing current UV light sources to the limit by doing things such as double, triple, quadruple exposure/patterning. But each patterning adds what I am estimating as 10 steps (adhesion promoter HMDS, bake, spin photoresist, bake, expose to light, bake, developer, rinse, dry, quality control check) which adds significant time, cost, and increased difficulty for the final product due to alignment issues. Plus, the extra mask costs and development time. Multiply this out for a few of the bottom layers in device and it's no easy task to make things work. Hence the R&D costs and years of hard work by all of the materials scientists, engineers, and staff behind the scenes before it even goes to production line people and their worry about yields, defects, ...

    The other major challenge is in the materials side of making a transistor. My take is FinFETs were developed as a means of lowering the leakage rate of electrons across the device (hence reducing the 'Off' state power) by making the gate length equal to a larger node process. But silicon itself has some fundamental limits to it and hence the development of III-V materials on large silicon wafers because even 40 years ago, people knew GaAs was faster (higher electron mobility) than silicon. Further, the recent interest in 2-D materials (WS2, WSe2, MoSe2, graphene, ...) as a means of being able to conduct an electron from one place to another quickly and at lower power. But all of these options take time to develop. Also keep in mind the solution must be able to be scaled to manufacturing speeds and be reproducible beyond normal comprehension to make a billion transistor device at a 100 million per year. One thing the semiconductor industry has done correctly is to make standards (e.g., 300 mm wafers as the defined unit) so multiple companies have a direction to make a machine that works interchangeably with others in the overall manufacturing production line. Compare that to the numerous starters, battery sizes, tires, rims, etc of the automotive world.

    So is "Moore's Law" dead? Well, yes if "Moore's Law" is doubling transistors every node jump every 24 months. Node jumps are getting fewer and fewer. Look at any roadmap. But will 'performance' increase with time? Sure. There is a need in terms of computing power per unit energy input. It just won't be as fast and as easy.

    As for the rest, sounds like you work at an interesting place!
    Reply
  • TheOtherBubka - Saturday, April 30, 2016 - link

    Edit: should have been 'EUV is coming along, but the development has not been easy'. Reply
  • zodiacfml - Saturday, April 30, 2016 - link

    End of Moore's law for Intel but not for computing as a whole. Actually, it is Intel or consumers decision to drive Moore's law. As long there is no urgent need for compute power, Moore's law will be violated. Yet, if there's an application that will utilize it, Moore's law might be exceeded.

    If I'm right, Intel is preparing for a big change and focus for that new industry they are going into and they haven't left a clue yet. I think this is the case due to the massive changes that is not related to the revenues.
    Reply
  • Shadowmaster625 - Sunday, May 1, 2016 - link

    Moore's Law ended years ago. It went from 18 months to 19 months. Then it went from 19 months to 20 months. No one can really say exactly when this happened because it has been happening so gradually. Reply
  • way2funni - Saturday, April 30, 2016 - link

    Any thought to the notion that intel gambled on Microsoft's windows mobile OS getting better traction by now?

    It seems to me there has been and remains a concerted effort to make your phone a very miniaturized PC running a real windows kernel. The newest high end Lumia 950 is a peek at the first layer of that onion being peeled with it's clumsy wired plug display setup .

    I always kinda saw that merger happening at a certain convergence in technologies - say, about where 7nm core that does what a Skylake i5 does now along with a 'Thunderbolt III 'AIR" ' level bus where you just walk in , drop your phone on your wireless charge pad and instantly your big screen lights up along with your Kinect and Cortana and whatever their VR setup but you don't have to wear a whole helmet - just an earpiece with a little bar like a mic but at eye level that projects across your eyes.

    By about 2020 -2022 Windows has AI and now you're in Tony Stark's lab with JARVIS
    Reply

Log in

Don't have an account? Sign up now