The general investing public has been essentially brainwashed to hate Intel (INTC) these days. While the first half of 2012 was filled with analyst praise for its growing revenues, clearly defined mobile strategy (first given at the analyst day in May 2011), and fabrication advantage, the second half has not been so kind.
Media pundits, who were calling the stock a "screaming buy" earlier this year, are now vehemently spreading fear, uncertainty, and doubt. For example, CNBC's Jim Cramer, on a recent edition of his popular show "Mad Money," had this to say about Intel:
It has a good yield but no products being used by mobile products. I'm saying don't buy.
I believe that this statement - likely to be taken to heart by a good majority of the investment public - is misleading. In fact, not only is this misleading, it's patently false, as Intel's latest Atom Z2760 tablet chip will be featured in no less than 20 very thin and light tablet design wins. Further, the company's Atom Z2460 is already powering smartphones that offer better performance and battery life characteristics than many of its ARM (ARMH) based competitors.
In fact, over the last several months - especially after last month's Intel Developer Forum - Intel's mobile strategy has become quite clear. In this article, I hope to explain just how clever its mobile strategy really is, and why it will be a dominant force in the smartphone and tablet spaces going forward.
Attacking From Above - Core
Intel has been relentless over the last few years about pursuing power efficiency on its notebook, desktop, and server chips. Each iteration of its processors has focused on improving performance in an incredibly power efficient manner. In fact, the "rule of thumb" for Intel's micro-processor design teams is that a feature can only be added if, for every 1% increase in power consumption, the design gains a 2% increase in performance.
To this end, we have seen lower power chips with higher performance in each generation. A combination of clever circuit design, strong micro-architectural decisions, tighter platform integration, sensible instruction set extensions, and process technology leadership has allowed Intel to fit significantly greater performance within ever-decreasing power envelopes. In fact, as demonstrated at the Intel Developer Forum this September, an 8W "Haswell" will provide the same performance as a 17W "Ivy Bridge" - even built on the same transistor technology (22nm tri-gate).
So, with the firm's upcoming "Haswell" able to fit into an 8W thermal design power, the implication here is quite clear: Intel intends to use its high-end "Core" processors in thin and light tablet and convertible form factors within a generation or two (14nm + further integration should whack the TDP down even more). The "Core" product line will be the high end, high performing, and high margin offerings for everything from tablets to high end desktops.
Attacking From Below - Current Atom Manages To Compete
Now, the more interesting part of the Intel mobile equation is the "Atom" line of systems-on-chip. While the high end "Core" line will be very power efficient for their performance levels going forward, the "Atom" has traditionally been the company's "low power, low end" solution.
Unfortunately, the emphasis here has been on "low end." Since introducing the first "Atom" processor in 2008, the core has not fundamentally changed, while the higher end cores have been getting refinement after refinement under the "tick-tock" strategy. As a result, the competition (ARM Holdings and its various licensees) was able to produce very competitive designs on the performance/watt front by employing more modern micro-architectural techniques such as out-of-order execution.
In tests, Intel's current "Medfield" and "Clover Trail" system-on-chip designs are actually still quite a bit ahead of the shipping ARM-based designs today such as the Nvidia (NVDA) Tegra 3, the Qualcomm (QCOM) Snapdragon S4, and even the custom designed Apple (AAPL) "Swift" (performance data here). It even does so while being quite power efficient. In fact, according to the aforementioned Anandtech review of the iPhone 5,
At least based on this data, it looks like Intel is the closest to offering a real competitor to Apple's own platform from a power efficiency standpoint. [...] I can't stress enough that the x86 power myth has been busted at this point.
However, the big challenge for Intel in 2013 will not be the mobile SoC designs shipping today, but rather the upcoming designs based on the very powerful ARM Cortex A15. In terms of performance, it is unlikely that the current Atom processors - even in their brand new mobile SoC forms - will be able to compete with SoCs based on the new ARM chip in terms of raw performance, although the latest Anandtech review of the Samsung Chromebook (which runs a dual core Cortex A15) notes that the performance will need to be scaled back in order to fit into tablet/smartphone thermal envelopes.
Accelerating The Atom Roadmap, Unifying Intel
That being said, the laziness here is over. Intel's previous Atom roadmap called for an update every 5 years, but the new Atom roadmap moves to a "tick-tock" strategy in which the firm will release either a new micro-architecture or the previous one on a new process node in alternating years. In 2013, the "Silvermont" chip will be both a "tick" and a "tock" as it combines a brand new design as well as a shrink to 22nm tri-gate.
More importantly, the company underwent a fairly significant reorganization during 2011. Previously, the "Atom" and the "Core" development teams were separate, meaning that they carried out work fairly independently of each other. This made sense - the "Core" design team has traditionally been about making faster, smaller, and more efficient CPUs, while the "Atom" team has been more about building low cost, low performance, but more integrated system-on-chip solutions.
However, the high performance processors are becoming more and more integrated and the lower power processors are becoming faster and faster. It made sense, then, to put the creative and technical juices from both teams under one roof to make sure that all of Intel's products - from "Atom" to "Xeon" - are the very best in every metric:
Can Intel Win?
The most pressing question here is whether this strategy will be successful. While I do not have a crystal ball handy to me, I can tell you that Intel has a habit of dominating every space that it feels threatened in. In the server space, Intel rose from a bit player in the early 90s to eventually singlehandedly overtaking each and every vaunted "RISC" based processor vendor (such as IBM (IBM) and Oracle (ORCL)) by aggressively innovating in process technology, micro-architecture, and the software ecosystem.
Note, however, that since 2010, AMD's (AMD) server market share has shrunken to nearly negligible levels and Intel's share has actually increased as its products have become even clearer winners on a raw performance and performance/watt basis.
Intel's chief competitors in the mobile space are ARM Holdings and its licensees. These licensees include:
The ARM business model is very interesting, and this model is the only reason that the aforementioned companies can even compete in the space. Let me explain.
Instruction Sets, Barriers To Entry, And Software
An important part of a microprocessor is the instruction set architecture that it implements. Loosely speaking, an instruction set architecture defines what operations the processor will support and how those operations are exposed to the programmer and compiler. It is exceptionally difficult to move to a new instruction set in this day and age because software writers (especially the operating system guys like Microsoft (MSFT), Google (GOOG), and Apple) don't want to have to create a brand new OS for each different instruction set architecture. Further, program compatibility across different versions of the same OS on different instruction sets is a challenge (although modern programming paradigms that implement virtual machines tend to mitigate this problem from the user's standpoint).
So, the key to having a commercially viable instruction set is to have not only software support on the application side, but to have a suite of development tools for the actual developers to optimize the code for the instruction set. There cannot be a dozen different instruction sets to support if developers actually want to get any work done, so only a handful survive. One of these is the ARM instruction set.
While ARM itself is fairly small and has a background in mostly very low power, low performance solutions, its popularity in the smartphone world has led to a fairly strong ecosystem that has made continued support for the instruction set architecture quite viable going forward. Google's Android, Apple's iOS, Microsoft's Windows RT and Windows Phone platforms are all now compatible with the ARM instruction set. This opens up the chip and SoC design ecosystem to a number of companies that had previously been unable to compete as they have not been able to acquire an x86 license from Intel.
With software compatibility between x86 and ARM a complete non-issue for mobile devices, the contest now really does come down to having the best chip (defined by performance/watt & peripheral features) at the best price. It's not about ARM or x86; it's about which of the SoC design houses ends up having the advantage going forward from a product and cost structure standpoint.
Intel Will Not Have A Mobile Monopoly
Unlike in the x86 space where Intel, over the last 10 years, had only to compete with a mismanaged and stumbling competitor, Intel has very strong competitors in the mobile SoC space:
- Nvidia is extremely well run and its graphics technology should prove to be an asset as mobile devices require stronger multimedia capabilities.
- Qualcomm is the world leader in baseband and already has its baseband chips built right into its "Snapdragon" solutions (it also designs its own CPU and GPU cores that perform very well).
- Samsung is a "jack of all trades" that is excellent at putting together CPU/GPU/peripheral ingredients from others into a nice product that it manufactures itself with its own fabrication plants
Further, on the manufacturing side, the non-Samsung fabless partners rely primarily on Taiwan Semiconductor (TSM), which has been producing and shipping its 28nm low-power transistor technology for about a year now. Intel, Samsung, and Taiwan Semiconductor are all essentially on the same page in terms of low power manufacturing (32nm from Intel and Samsung is more-or-less equivalent to 28nm from Taiwan Semiconductor).
Intel has the resources, patents, and experience to hold a technological lead on the ARM-based competitors on micro-architecture, instruction set design (it controls its own ISA, so it can add in new performance enhancing features at will, while the ARM folks need to wait for ARM's revisions), and fabrication (should Intel follow through with its plans to accelerate its low power processes). This means:
- Smaller, cheaper to manufacture chips
- Faster, more power-efficient processors
- Quicker time-to-market for semi-custom designs for the phone vendors due to full vertical integration
- Stronger software support due to the control it has over its instruction set as well as its extremely large army of software developers (these folks were responsible for bringing Android to Intel's X86 and making sure everything worked flawlessly)
However, the competition is not incompetent and there will naturally continue to be a strong ARM-based presence in the mobile space. The competitors I listed earlier are all extremely well-run operations that represent the strongest players in the ARM ecosystem. I believe that in spite of Intel's financial and technical advantages, Intel does face a number of headwinds that cannot be ignored:
- Intel still needs to make sure its notebook/desktop "Core" chips are relevant, so it may sandbag innovation in its lower cost "Atom" lineup to prevent cannibalization (although I would sincerely hope that this is not the case)
- Intel will still go through the entirety of 2013 without releasing its next generation "Atom" core, meaning that during this year, designs based on ARM's "Cortex A15" such as the upcoming Nvidia Tegra 4 and the Samsung Exynos 5-series will likely be superior solutions across the board, giving the ARM players another year to further build momentum
- It is not clear how well Intel's low power graphics solutions will perform - currently, its competitors lead it in this area by a significant margin (and Nvidia is a graphics specialist)
- The lower ASPs of mobile chips cannot be ignored and should tablets and smartphones significantly cannibalize PC sales, Intel's revenues might shrink over time.
Intel will be a big player in the mobile space. It has the technological resources, experience, and money to produce some of the best solutions for the space. In 2013, Intel will use its "Clovertrail" and "Medfield" platforms to fight its competitors, but it is unclear how well these parts will be on a performance/watt basis compared to the latest ARM Cortex A15 based designs. However at the end of 2013/early 2014, we will see the upcoming "Silvermont" core in the smartphone SoC codenamed "Merrifield" and the tablet SoC codenamed "Bay Trail." At that point, we will know if Intel's renewed focus will be good enough to validate my thesis.