Friday, March 12, 2010

Mobile Wi-Fi Growth Set To Accelerate

The market for mobile Wi-Fi is about to move into a new phase of growth driven by user demand for web browsing and peer-to-peer communications. Traditionally, Wi-Fi connectivity is associated with smartphones but not feature phones. Although overall handset shipments in 2009 declined compared to a year earlier, smartphone shipments increased by 17%. This growth was driven by the popularity of smartphones such as RIM’s BlackBerry and Apple’s iPhone. The number of smartphones with Wi-Fi grew an astounding 75% in 2009 as the attach rate of this feature soared.

Wi-Fi is now poised for adoption even in feature phones. Feature phones do not use an open operating system, but many still offer a web browser. The LG Arena is one of the first feature phones to support Wi-Fi for accelerated web browsing, but we expect others to follow suit. Wi-Fi is also required to enable the High Speed mode of Bluetooth 3.0. The new Wi-Fi Direct feature will enable simple peer-to-peer connection with other Wi-Fi devices, further increasing the popularity of that standard.

We believe that these trends will drive the Wi-Fi attach rate to 40% of all handsets in 2014. This attach rate translates to mobile Wi-Fi chip shipments (including combos) of 750 million units in that year, a 28% annual growth rate from 2009. These chips will generate more than $2.4 billion in revenue.

In handsets, the Wi-Fi functionality is initially delivered as a standalone chip and is gradually moving to a combo connectivity chip. Eventually the Wi-Fi functionality could be integrated with the cellular baseband chip set. We project the standalone Wi-Fi market in handsets to peak in 2012 and than start declining as this technology gets integrated into combo chips.

Currently, the leading supplier of Wi-Fi chips is Broadcom, followed by TI, Marvell, and Atheros. Broadcom and TI are also leading suppliers of combo chips and have all of the four major (Bluetooth, FM, Wi-Fi, and GPS) mobile-connectivity technologies in production. TI was the first vendor to sample a four-way combo chip, and we expect Broadcom to follow suit by the end of 2010. Consequently, both of these vendors should maintain leadership positions for mobile Wi-Fi. Atheros and Marvell have significant ongoing Wi-Fi revenue from the Nintendo DSi and Sony PSP, respectively.

Other vendors that could increase their Wi-Fi market share include ST-Ericsson, Qualcomm, and MediaTek. All of these vendors share a common trait of offering cellular baseband processors as well as all four mobile connectivity technologies. Each of these four vendors can create a handset bundle to effectively block other suppliers of Wi-Fi chips. This strategy represents the biggest threat to the current leaders of mobile Wi-Fi chips. --Jag

Jag Bolaria, senior analyst

Additional coverage of this market appears in our recent report A Guide to Mobile Connectivity Chips.

For analysis and news on semiconductors for mobile and wireless, subscribe to our free newsletter, Linley on Mobile

Thursday, February 18, 2010

LSI Announces Multicore Network Processor

Last week, in advance of Mobile World Congress, LSI announced its Axxia Communication Processor (ACP) family. Combining multiple PowerPC CPUs with data-plane technology from LSI's APP network processors, the ACP chips represent a hybrid of a multicore processor and an NPU. The chips also include content-inspection technology from LSI's Tarari line.

The top-of-the-line ACP3448 includes four PowerPC 476 CPUs that operate at up to 1.8GHz and each have a 512KB level 2 cache. An IBM/LSI joint development, the new 476 core can issue up to five instructions per cycle and supports out-of-order execution. The 476 does not, however, support multithreading. Using a coherent interconnect, the ACP supports symmetric multiprocessing (SMP).

The ACP data plane uses blocks similar to those in LSI's shipping APP3300/APP650 NPUs. But whereas the APP chips use a fixed pipeline, the new ACP uses what LSI calls a virtual pipeline to enable flexible data flow between these blocks as well as to the PowerPC CPUs. The ACP3448 is rated at 20Gbps of throughput, 10Gbps of security processing (e.g., IPSec), and 3Gbps of regular-expression processing. The data plane integrates eight GbE MACs and a pair of 10GbE MACs. The ACP also includes PCI Express and RapidIO ports to connect external hosts, DSPs, or peripherals.

Built in a 45nm SOI process, the ACP34xx family will begin sampling this month. LSI is offering two other ACP34xx family members: the ACP3442 with four CPUs and a 10Gbps data plane, and the ACP3421 with two CPUs and a 5Gbps data plane. All three ACP34xx family members share a common pinout.

What sets LSI's ACP apart from multicore competitors, such as Cavium's Octeon and NetLogic's XLP, is its autonomous data plane. For many designs, the ACP can perform complete packet processing without any CPU intervention or load. Compared with network processors, such as EZchip's NPA or Wintegra's WinPath, the ACP is first to integrate powerful multicore control-plane processing. LSI must now show this hybrid design can simultaneously deliver deterministic throughput. Depending on pricing and power dissipation, which were not announced, the ACP34xx should be attractive for a variety of designs in wireless infrastructure, network security, and various Layer 4-7 applications. --Bob

Bob Wheeler, senior analyst

For analysis and news on processors for networking and communications, subscribe to our free newsletter, Linley Wire

Wednesday, February 3, 2010

What's In the Apple A4?

For chipheads, the big surprise in Apple's iPad announcement was the disclosure that the device uses an "internally designed" processor called the Apple A4. This new 1GHz chip is also likely to appear in the fourth-generation iPhone later this year, although it may be underclocked to 800MHz to save power in the phone.

A popular theory is that the iPad processor includes a custom CPU designed by the former PA Semi team, which Apple acquired in April 2008. For this to be true, the team would have had to design a new ARM CPU from scratch (its previous CPU used PowerPC) and put it into production in about 20 months. A new CPU design typically takes about three years of design, verification, testing, and qualification; the PA Semi team was exceptional, but completing such a task in less than two years would be highly improbable.

Another theory is that the A4 instead uses an ARM-designed CPU, such as the Cortex-A9 used in Nvidia's new Tegra 2 and other forthcoming mobile chips. The Cortex-A9 is capable of reaching 1GHz in state-of-the-art 45nm manufacturing technology. If indeed the iPad has a Cortex-A9 CPU, it would be the first mobile product to use that design, beating the Tegra 2 to market by a few months. It would be strange, however, for ARM to work with Apple as the lead customer for its new CPU, knowing that the company would keep the processor to itself. Furthermore, the Cortex-A9 is designed for dual-CPU implementations, and one would expect Apple to brag about this capability if its processor supported two CPUs.

A third idea is that the A4 uses the 1GHz Cortex-A8 CPU known as Hummingbird, which is designed by Intrinsity and manufactured by Samsung. This choice would allow Apple to continue working with Samsung, a long-time Apple supplier that makes the Cortex-A8 processor for the iPhone 3GS. Staying with the Cortex-A8 would also simplify software development. Samsung announced that the Hummingbird CPU had already been validated in silicon last July, putting it on track to be production-ready in time for the iPad launch.

Apple touts the low power consumption of the A4 as enabling the iPad's long battery life. The battery life of the iPad is about the same as that of the iPhone 3GS, but the iPad's large battery has five times the capacity of the iPhone's battery. Most of this extra juice is needed for the big display; the power consumption of the processor has little to do with it.

For 3D graphics, some reports indicate that the A4 uses ARM's Mali engine, but Samsung typically uses Imagination's PowerVR cores, and Apple is an investor in Imagination. From the iPad specifications, we also assume that the A4 includes a video engine capable of decoding H.264 main-profile video at 720p resolution and 30fps. MPEG-4 video is decoded at VGA resolution, a poor fit for the iPad's XGA screen.

Outside of the processor, we expect that the iPad leverages many of the same chips as the iPhone. However, it probably replaces Broadcom's BCM4325 combo chip with the newer BCM4329, which adds 802.11n capability. The iPad has the same cellular specifications as the iPhone 3GS and probably uses the same chips from Infineon, which has supplied all of Apple's iPhone models.

These mysteries should be resolved once the iPad is available. A quick check of the A4's CPU ID and performance characteristics should identify it as a licensed ARM design or a wholly new CPU. If it is the former, Apple is not getting much benefit from designing its own processor chip. --Linley

Linley Gwennap, principal analyst

Additional coverage of this market appears in our recent report "A Guide to Mobile Processors."

For analysis and news on semiconductors for mobile and wireless, subscribe to our free newsletter, Linley on Mobile


Wednesday, January 27, 2010

Tegra 2 Sets Mobile Speed Record

Nvidia's Tegra 250 will be the first mobile processor to reach production using ARM's Cortex-A9 CPU, giving it a performance advantage for smartphones, tablet computers, and other mobile devices. The second-generation Tegra processor has been sampling since 3Q09 and is expected to reach production in 2Q10. ST-Ericsson's U8500 and TI's OMAP4, which also use Cortex-A9, are due to enter production later in 2010.

Tegra 250 contains two Cortex-A9 CPUs, each running at up to 1.0GHz. Extrapolating from initial tests on the ARM CPU, this configuration should score 5760 on the Coremark test (using two threads). This represents more than twice the speed of the iPhone 3GS, Nexus One, or any other smartphone shipping today. It is also 40% faster than the Coremark score of the 1.6GHz Atom CPU (also using two threads with hyperthreading).

Nvidia would not specify the power consumed when both CPUs are firing at full speed, but we believe it exceeds 1.0W. The chip will typically operate at this level for only brief bursts, however, as the second CPU will rarely be needed. In fact, when playing audio or video, neither CPU is required, because the chip provides dedicated hardware engines for these functions. The company says that power consumption in these modes is similar to that of the original Tegra, which uses 150mW for HD (720p) video playback and just 15mW for audio playback.

Tegra 250 can decode video at 30fps and full HD (1080p) resolution for H.264 base and main profiles, but it is restricted to 720p for the ultrachallenging H.264 high profile (Blu-ray). The chip can also encode video at 1080p for H.264 base profile. Few other mobile processors can handle 1080p video at even the base profile.

On the 3D side, Nvidia says it has doubled the performance of the initial Tegra, resulting in a peak speed of 90 million triangles per second. This level is well beyond the performance of any mobile processor shipping or even sampling today.

Tegra 250 will appear initially in tablet computers and similar-sized devices that will debut by mid-2010. The chip will also be used in smartphones, but due to longer design and qualification cycles, these products will not ship until late 2010. Although new processors from ST, TI, and Intel will approach the performance of Tegra 250, Nvidia is already working on its third-generation processor, which it expects to sample before the end of this year. Now that it has the performance crown, Nvidia does not plan to give it up. --Linley

Linley Gwennap, principal analyst

Complete coverage of the U8500, OMAP4, and Atom appears in our recent report "A Guide to Mobile Processors."

For analysis and news on semiconductors for mobile and wireless, subscribe to our free newsletter, Linley on Mobile




Monday, January 18, 2010

Smart TVs Lure Intel Into CE

Although 3D TVs received the most publicity at CES this past month, the emergence of the “smart” or Internet-connected TV is more significant. TV-maker Vizio exemplifies the trend. The company displayed TVs running widgets to connect to Facebook, Yahoo Widgets, Netflix, and YouTube. Most of these applications previously had been in the domain of PCs.

The emergence of PC-like applications in TVs has set up a clash between long-time suppliers of consumer-electronics (CE) chips and Intel, paralleling the battle for smartphone technology. Intel asserts that as CE systems connect to the Internet and gain the smarts to run a variety of software that it has an advantage because of the performance of its CPU technology and the surrounding infrastructure.

Suppliers of CE chips counter that the market requires integrated chips that combine CPUs with traditional CE functions such as video processing and analog conversion plus new functions for home-networking connectivity such as Wi-Fi and new technologies such as 60GHz radios. CPU designs from companies such as MIPS and ARM deliver sufficient computing performance and better power dissipation than Intel’s CPUs. Moreover, the software most important to the PC—Windows and Office—is irrelevant.

Arguments from both camps have validity but on balance Intel is the one that must play catch-up. Integration and cost-effectiveness are clearly important product attributes. Intel has developed a few CE-specific chips, including the Atom-based CE4100, but the company has had little uptake. Competitors such as Broadcom and STMicroelectronics have wider product lines and chips that integrate functions such as demodulators for cable, satellite, or broadcast TV and digital-analog converters.

Meanwhile, smartphones have shown that software developers are excited to target systems based on processors other than x86. Crucially, Adobe is enabling Flash to run on various architectures, further opening the door for hardware-neutral software. JavaScript—used by Yahoo Widgets—and Android also facilitate cross-platform software development.

While the battle heats up among chip suppliers, another group of technology companies stands to benefit from the trend toward smart, connected TVs: semiconductor IP suppliers. Beyond demand for ever-faster CPUs, companies need 3D-graphics processors for glitzy user interfaces, better audio processors, multi-standard video encoders and decoders, and I/O controllers ranging from SATA to Wi-Fi. Ultimately it is the consumer that wins as the TV morphs from Idiot Box to something a bit smarter. --Joe

Joseph Byrne, senior analyst

Friday, December 18, 2009

Broadcom Sneaks Out New Ethernet Products

At Broadcom's Analyst Day, Enterprise Group VP Nariman Yousefi spent much of his time discussing market trends and forecasts. The presentation also embedded, without much fanfare, a handful of new products that implement Energy Efficient Ethernet (EEE), which is being finalized in IEEEE 802.1az work group. EEE reduces energy use by reducing link speed during periods of low link utilization. Broadcom products that support EEE include GbE PHYs, 10GbE PHYs, and controller. The company claims its EEE PHYs will use 70% less power during periods of low traffic utilization compared with a PHY without EEE.

Hidden in the EEE press release is a new 10GBase-T product, the BCM84823. This new dual-port PHY appears to be replacing the short-lived BCM84812 dual PHY that was introduced at the April 2009 Interop. We believe most of the EEE operations are incorporated through firmware changes to the existing silicon and thus represent a low risk. Like its predecessor, the BCM84823 is built in 65nm and consumes about the same power (5.5W) when active. Compared with its predecessor, the BCM84823 adds XFI and uses a smaller size package. XFI is popular at Cisco, a Broadcom customer that may have pushed for such a change.

Separately, Broadcom's VP of manufacturing, Neil Kim, stated that the Enterprise Group is sampling a 40nm product. Although he provided no further details on this product, we suspect it to be a high-density (>48 port) 10GbE switch. This switch would compete with Marvell's 65nm Lion silicon, which is also a 48x10GbE switch but has already been sampling for more than nine months, with OEM products expected in 1H10.

Broadcom's goal is to own the Ethernet market by proliferating products into all niches in order to prevent competitors from gaining a foothold. To achieve this goal, Broadcom develops proliferations such as these EEE products, CE products for emerging markets, and FCoE devices for future growth segments as well as acquiring companies such as Dune to plug product-line holes. In Ethernet as a whole, Broadcom has tremendous momentum and an insurmountably large infrastructure. Apart from Marvell, few other companies will be able to challenge Broadcom's Ethernet position, and thus any risks for future downsides to its business are more from poor internal decisions than external factors. --Jag

Jag Bolaria, senior analyst

For analysis and news on processors for networking and communications, subscribe to our free newsletter, Linley Wire



Thursday, December 17, 2009

Broadcom's Not-So-Smartphone Strategy

At its Analyst Day this week, Broadcom discussed its progress in the cell-phone market. Broadcom's connectivity (Bluetooth, FM, GPS, and Wi-Fi) chips have been quite successful. According to Bob Rango, VP of the connectivity group, these products have won designs with all of the Big Five handset makers except Motorola and all of the top four smartphone makers except RIM.

Scott Bibaud, VP of the cellular group, reported that his products have made impressive headway with Samsung, which has adopted Broadcom's EDGE and UMTS baseband processors into a number of phones. For example, the EDGE chip appears in the popular Samsung Star, which has already shipped more than 10 million units. The group's other major customer, Nokia, has been less enthusiastic, so far putting into production only one model that uses a Broadcom baseband.

Product execution continues to be a problem for the cellular group, however, which has announced no new baseband or application processors in more than two years. In fact, the company admitted that its ballyhooed "Zeus" processor, the BCM21551, has been terminated. Zeus, as you may recall, was a technological tour de force combining an application processor, HSPA baseband, 2G/3G RF circuitry, and Bluetooth/FM. We originally heralded the chip's announcement as a sign of Broadcom's technology leadership in smartphones. In the words of the immortal Emily Litella: never mind.

CEO Scott McGregor said he pulled the plug on Zeus due to a "lack of customer interest." Of course, when a product hasn't entered production two years after its announcement, customer interest may wane. Because of this delay, the processor's ARM11 CPU did not match up well against the more powerful Cortex-A8 used in many popular smartphones shipping today. Another trouble spot may have been the integrated nine-band RF circuitry, which would have been the industry's first; Broadcom said that the chip was fully functional but admitted that its customer "preferred" to use its own RF chips.

As a result, the company's only current offering for smartphones is the BCM2153, which combines a 312MHz ARM11 application CPU with an HSDPA baseband. As this CPU is slower than a two-year-old iPhone, it is not well suited to new smartphone designs. Broadcom offers no standalone application processors for smartphones and, given that the market is shifting toward integrating the application and baseband processors, downplayed its interest in developing one.

To draw attention away from this situation, Bibaud attempted to define a new product category that he called a "smarter phone" or a "smart-feature phone." This type of phone can provide a widget-based interface and strong multimedia capabilities, but it does not run a smartphone OS and therefore cannot download applications. Although we agree that feature phones are improving and can mimic some smartphone capabilities, Apple's dramatic success with its App Store demonstrates that the future is all about downloadable apps. Every major market analyst agrees that smartphones will continue to eat into the feature-phone market over the next several years.

Given that Broadcom's current baseband share is tiny, focusing on the feature-phone segment will be enough to greatly improve its market share in the near term. To achieve longer-term success, the company must restart its product pipeline, and it must deliver processors suitable for the lucrative and growing smartphone segment. --Linley

Linley Gwennap, president and principal analyst