Intel's Haswell Tablet Breakthrough

by Mark W. Hibben

Compelling Improvements

Intel's new 4th generation Core processors (Haswell) included low power Systems on Chip (SOCs) that are being used in thin tablets without fans, a first for the Core series. These Haswell based tablets have the potential to devastate the market for ARM tablet processors. Without a swift response from ARM's Common Platform partners, IBM, Samsung and GlobalFoundries, the ARM processor ecosystem could lie in ruins by this time next year.

Intel made a number of significant improvements that enabled it to fit a Core processor into the thin tablet form factor, but the main ones were improved power management and the development of single chip SOCs, eliminating the Platform Controller Hub, an accessory chip needed by the previous Core generation. These Core SOCs still consume roughly 5 times the power (5-10 Watts) compared to a typical ARM processor, but they are usually more than 5 times faster. Getting into the 5-10 Watt range allows these SOCs to be used in a tablet, and it can be thin since no fan is required.

These processors are full Intel 64 bit architecture machines capable of running Windows 8 Pro, or Chrome OS, or Android. At Computex, the industry trade show where the chips and devices were unveiled, Intel made it clear it was targeting all three OS markets.

The introduction of 64 bit Core computing into the tablet form factor is nothing short of revolutionary. The Microsoft Surface Pro hinted at what was to come, but it was bulky, heavy and needed a fan for cooling its 3rd Gen Core processor. Now Microsoft will be able to update the Pro with a 4th Gen Haswell and get the size and weight down to something closer to the Surface RT.

But not quite to the Surface RT. Here, the main problem is the power requirement of the Haswell Y series processor. Five times the power means five times the battery capacity for equivalent battery life. Haswell tablets won't be able to provide the battery life of ARM based tablets, but this is the price to be paid for having a much more powerful computer.

For this reason, all of the Haswell tablets shown at Computex were what Intel calls 2-in-1 with detachable keyboards, and Intel really wanted to push the utility of the keyboard. In fact, the keyboard is mainly there to hold the extra battery pack that the Haswell tablet needs.

Window of Opportunity

The necessity of carrying around the extra battery pack will tend to reduce somewhat the desirability of Haswell compared to current and future Atom or ARM based tablets, and this goes far to explain Samsung's decision to use the soon to be obsolete Clover Trail Atom in its new Galaxy Tab 3. Combined with the fact that the Bay Trail Atoms, based on the same 22nm process as Haswell, won't be in tablets until Q4, the ARM ecosystem has a very limited window of opportunity to respond to the Intel tablet threat. The ARM-based ecosystems of Android and iOS can carry forward on some momentum due to their size, but this will not sustain them for long in the face of the Intel onslaught. Can ARM and its partners respond?

Yes, I think they can and they will. This may come as a surprise to Intel bulls, many of whom regard the largest ARM foundry, TSMC, with derision. I often hear it claimed that "TSMC will never get their 20 nm process off the ground", despite the fact that TSMC has formally announced that it is ready to enter production. One of the constants of technological change is the rapidity with which technical advantage leaks away to competitors. Intel is able to produce x86 chips to compete with ARM chips because it uses a superior 22 nm process while the rest of the ARM world still uses 32 nm. This situation will probably not continue indefinitely.

In addition to TSMC's new 20 nm process, the Common Platform alliance, composed of ARM, IBM, Samsung, and GlobalFoundries, is working on their own 20 nm process and a 14 nm FinFET technology superior to Intel's current 22 nm. IBM holds many patents related to the FinFET technology which is also used in the Intel "tri-gate" 22 nm process, so Intel has no exclusive lock on the technology. GlobalFoundries is currently working to bring the 14 nm process on line at its Saratoga County, New York facility. The problem, of course, is time to implementation. Currently, only their 28 nm process is in production.

The Common Platform alliance has been in existence for 10 years. In effect, it has served as a licensing pathway for IBM research developments (such as the FinFET) to find a pathway into production facilities. The ARM ecosystem has benefited from this and been able to more or less keep pace with Intel, although a step behind in process technology. This will be the Common Platform's most difficult test, since remaining a step behind Intel may no longer be an option.

How much time does the Common Platform and the rest of the ARM ecosystems have? Here, the fact that the ARM ecosystems are very developed with over half a billion users each for Android and iOS may help. Consumers are basing buying decisions on the ecosystem of services for which the device is a portal as much as on the device itself. So immediate inroads by Windows and Chrome based tablets may be modest and small compared to the overall growth of the tablet market. In this scenario, ARM foundries might have until late 2014.

But Samsung's decision on the Galaxy Tab 3 could be the beginning of a wave of defections from Android, especially among Android manufacturers who have had a tough time competing with Samsung and are looking for an advantage.

I believe the 20 nm processes in development at TMSC and the Common Platform partners will be in production in early 2014, if not sooner. Given the onslaught of Intel Haswell and Bay Trail processors in Q4, I expect ARM processor shipments to stagnate if not actually decline in the quarter. This will bring on a long overdue correction in ARM's share price, as I have been expecting since I wrote "Does ARM Still Have Momentum". However, I expect ARM processor shipment growth to resume in Q1 2014 as ARM foundries bring their 20 nm processes on line, and a new generation of ARM processors hits the market.


Intel's Unlikely Ally, Samsung

by Mark W. Hibben

An Endorsement and a Rebuff

Intel announced today that an Intel Atom processor would be in the forthcoming Samsung Galaxy Tab 3, a 10.1 inch Android tablet, a move that has implications for ARM Holdings, Microsoft and Apple. Intel's long awaited assault on mobile computing is finally gathering steam.

The announcement was made even more jolting by the fact that the Intel processor is not from Intel's latest and greatest Haswell processor family, which is also being rolled out this week (I'll have more on Haswell in my next post). The processor selected for the Tab was a fairly humble dual core Clover Trail + Atom using Intel's older 32 nm process and clocked at 1.6 GHz.

Various Clover Trails have been finding their way into Windows 8 tablets since the beginning of the year, and these tablets from companies such as HP (Elitepad 900) were reasonably thin and light. Moreover, independent testing by AnandTech showed that they could outperform multicore ARM processors such as the nVidia Tegra 3. In a seperate AnandTech article, a Clover Trail Atom (Z2760) was shown to be more power efficient than competing ARM processors, the Tegra 3 and the Qualcomm APQ8060A.

But the Clover Trail tablets and ultrabooks hadn't made much of a splash as PC sales continued to decline through Q1, and attention shifted to the arrival of Haswell, which was assumed would usher in a new era of high performance Intel tablets.

Samsung's decision to use the Atom is both an endorsement of Intel architecture and a rebuff to Haswell. As I'll discuss in my next post, Haswell may not be able to meet the very high expectations for energy efficiency that I and many others had.

Samsung makes ARM based SOCs that go into its phones, and provides foundry services to other manufacturers such as Apple. Samsung has a huge stake in the ARM ecosystem, so the decision to go with Intel for a flagship product probably didn't come easily.

But, as I pointed out, the Clover Trail Atoms were already outperforming ARM processors, and Samsung knew that even more efficient 22 nm Bay Trail Atoms would be available in Q4. Intel could pitch the Clover Trail adoption as a way to ensure access to Bay Trail when it became available.

Samsung has not abandoned ARM, of course, but they may believe it best to disengage gradually from the ARM ecosystem as Intel processors become dominant.

ARM processor deliveries will not be immediately impacted by the Samsung decision, so we'll probably not see any change in the roughly 20% per year growth in ARM processor deliveries and licensing revenue when ARM reports their results for the second quarter. However, ARM processor deliveries could slow down beginning in Q3 if other manufacturers adopt Intel for their Android devices.

Does Microsoft Get the Message?

Microsoft started the disintegration of the Wintel marriage with its Windows on ARM inititiative announced at CES in 2011. Since then, the partners have drifted apart as Windows 8 was released to mixed reviews and accusations that it had contributed to the steep decline in PC sales in Q1. Until now, Clover Trail processors had been used exclusively in Windows 8 tablets and ultrabooks. Intel in an Android tablet takes a direct swipe at Windows RT and Windows 8, in effect saying that Intel won't wait any longer for Windows 8 to gain traction in Mobile. Other manufacturers with Android tablets (and which of the PC makers doesn't have an Android tablet these days?) may follow Samsung's lead, especially those with considerable Intel experience may follow Samsung's lead.

Given that Windows RT and Surface RT have not been successful, it seems certain that Redmond regrets the whole endeavor, although Microsoft management continue to put on a brave face. Earlier this year, the sensible thing for Microsoft to do would have been to propose reuniting with Intel. Microsoft would give up on ARM and in return Intel would give up on Android. Now, it appears to late for that. Microsoft still has issues with Windows 8 that will hopefully be addressed in 8.1, so it doesn't have much leverage. Meanwhile, Intel has come to believe that the Android market could be its market.

Windows 8 was supposed to bring the mobile market to Intel, but Intel has decided not to wait for it to arrive. Intel Inside would have been the key advantage of Wintel tablets, but now that's gone for Microsoft. The immediate impact of Samsung's decision will be negligible, but as other Android manufacturers join Samsung in adopting Intel, Microsoft's Windows 8.x sales could be hurt.

Whither Apple?

As an iOS developer, I really don't care what processor platform Apple adopts for their mobile devices, I just don't want them to become disadvantaged. Apple has almost as much invested in the ARM ecosystem as Samsung. It's not clear exactly how much of Apple's capital budget goes to support the manufacture of its custom SOCs, but its probably a significant portion of the approximately $10 billion Apple plans to spend this year.

Given the momentum in favor of ARM, I don't expect Apple to jump to Intel any time soon. In fact, I wonder if the delays in some products such as the next generation iPad might be due to a delay in getting the next Apple designed SOC into production. (I'll have a better feel for this after WWDC next week.) To compete with mainstream Android tablets, Apple needs to have a quad core processor, but this would be a first for Apple and they may have hit a technical snag.

The proliferation of Haswell ultrabooks and Bay Trail tablets in the Fall (whether Windows 8 or Android) will probably eat further into Apple's tablet market share without a very compelling iPad refresh. Apple certainly doesn't need to lose more tablet market share. According to IDC's Tablet Market share data, Apple's share went from 58.1% in 2012 Q1 to 39.6% in 2013 Q1, while Samsung went from 11.3% to 17.9%.

Apple's declining market shares for both smartphones and tablets underscores the need to increase R&D spending at Apple, which as I pointed out in "The Case for Lower Apple Margins", is much lower as a percentage of revenue than Google or Microsoft. Higher R&D spending would allow Apple to introduce new products and refresh existing products faster.

If Apple doesn't respond adequately to the challenge posed by Intel's processors in Q4, this will set the stage for Apple to abandon ARM in 2014, further damaging, if not killing, the ARM ecosystem.