Electronics are shrinking faster than Las Vegas housing prices, the market for 2D kiddie flicks, and Lindsey Lohan's showbiz career. But for electronics, at least, shrinkage is good. Today's cell phones, tablets, and other trendy mobile gadgets would be impossible without the miniaturized electronics of system-on-chip (SoC) processors.
Indeed, SoCs are all the rage in today's gadget-loving tech circles. So why is Intel nowhere to be found?
SoCs: Your Basic PrimerSoCs cram essential system functions into a single chip. Ideally, the SoC would be the only chip, but few SoCs are so well integrated. Usually, a few smaller chips must be added for additional functions. This adds bulk, and in a mobile gadget, size matters.
The other crucial factor is power. The less electricity a device consumes, the longer it can run on small batteries, and the less heat it dissipates. Engineers must juggle these factors to find the ideal balance for their hardware designs.
To complicate matters further, users keep wanting more features. When cell phones added still cameras, they needed image processors and additional memory. Now those cameras are shooting videos, too, so they need video processors and even more storage. Two-way video for Skype? Add a second camera and more horsepower. HD video? Ditto. It never ends.
Two of the most powerful families of smartphone SoCs are Snapdragon from Qualcomm and OMAP from Texas Instruments (TI). Known as application processors, these chips are awesomely complex, as the OMAP4 block diagram below shows. Although their CPUs aren't as powerful as today's PC processors, they are as fast as the PC procs of a few years ago and typically sip less than one watt-a stunning combination of performance and power efficiency.
Texas Instruments' latest SoC, the OMAP44x, is a dual-core chip that uses ARM's most advanced architecture, Cortex-A9. As you can see from this diagram, the SoC handles pretty much every function that, in a traditional PC, would be executed by separate hardware components. The video subsystem uses the same 3D acceleration architecture as Apple's A4, and supports resolutions up to 1680x1050.
Equally impressive is their large-scale integration. The TI and Qualcomm SoCs are packed with features not yet found in PC processors, such as special hardware for audio, video, graphics, image capture, data compression, cryptography, networking, security, GPS navigation, and wireless communications. They also have built-in interfaces for keypads, keyboards, I/O ports, microphones, speakers, and touch screens. Power management is especially sophisticated. They can throttle each part of the chip at different speeds and shut off circuits when they're not needed.
X86: Powerful-and Power-HungryIntel is trying hard to break into the SoC market but is hampered by an x86 architecture that's historically geared toward high performance, not low power. Virtually all application processors use ARM, which is optimized for power efficiency and low cost.
ARM has another advantage: open licensing. Unlike Intel, ARM doesn't sell chips. Instead, ARM sells licenses to its CPU architecture and processor cores. ARM's army of licensees (like Freescale, Marvell, Samsung, Qualcomm, and TI, to name just a few) design and sell the actual chips while ARM sits back and collects royalties. Intel won't openly license the x86, so Intel's customers are limited to the SoCs that Intel offers. Intel launched a custom design program last year, but it appears to have fizzled.
So far, Intel's application processors aren't as integrated as the best Snapdragon and OMAP chips. Case in point: Intel's latest product, the Atom-based Moorestown chipset. Although it's much better than the previous Menlow chipset, Moorestown still requires two chips to duplicate the application-processor functions that TI's OMAP4430 delivers in one chip. The Intel chips are also more than twice the size of the OMAP4430 and consume about 50 percent more active power than the OMAP4430, by some estimates.
The latest trend is to integrate more wireless functions into application processors. For years, mobile phones have used a separate baseband processor that converts analog radio-frequency signals into digital data and vice versa. Adding these functions to the application processor eliminates the baseband processor, but it's difficult to integrate analog and digital functions without causing interference and manufacturing problems.
Rising to the challenge, Qualcomm's Snapdragon QSD8650 combines a cellular baseband processor with a full-featured application processor. It even incorporates a power-management chip that would normally be separate. Another Qualcomm SoC, the QTR8600, integrates cellular radio functions with Bluetooth, FM radio, and GPS. Even Intel's next-generation Medfield SoC won't go that far.
Intel is trying to catch up through acquisitions. The company recently acquired the assets of Comsys, a small wireless company, and will probably buy the wireless business unit of Infineon, a major supplier to Apple, Nokia, and Samsung. In addition, Intel has licensed technology from Nokia.
Traditionally, Intel's advantages have been world-class chip-fabrication technology and enormous production capacity. However, Intel fabricates Moorestown in an older 45nm process instead of the smaller 32nm process introduced for PC processors this year. Qualcomm and TI are still using 45nm for Snapdragon and OMAP4, so Intel could be exploiting its technology advantage more aggressively. But PC processors are Intel's cash cow, so Intel won't manufacture its slow-selling SoCs in 32nm until Medfield debuts next year.
Another problem for Intel is system software. Since the first IBM PC in 1981, Intel has enjoyed a symbiotic relationship with Microsoft's popular DOS and Windows operating systems. But Microsoft's supremacy on PCs isn't translating into similar success on mobile phones and tablets.
Last April, Microsoft introduced its own smartphone, the Kin. Never heard of it? Kin was killed only six weeks after launch. And in the wake of Apple's hot-selling iPad, Hewlett-Packard withdrew a Windows-based tablet even before introduction. Intel is hedging its bets with two GNU/Linux-based operating systems (Moblin and MeeGo), but they're looking like also-rans against Apple's slick iOS and Google's fast-rising Android.
The simple truth is that Intel needs the mobile market more than the mobile market needs Intel. ARM's lower-power processors already dominate the market, they're growing more powerful each year, and they have a larger software base. Marvell, Qualcomm, and probably Apple have ARM architectural licenses, so they can design their own ARM-compatible processors-something Intel won't allow with the x86. To out-wrestle ARM, Intel needs something better, not something almost as good.
By the end of this decade, we may see the first single-chip smartphone or tablet. This super-SoC will probably be a stacked chip that combines the memory and processor in one package-technically, not a single chip, but close enough. On the other hand, new features yet to be imagined may still require some auxiliary chips. But it's a certainty that SoCs will keep getting smaller, faster, and better.
source : http://gizmodo.com/5638574/can-intel-compete-in-a-mobile-arms-race