Mobile processors are rapidly evolving. The demands of Web browsing are driving higher-performing smart-phone processors. At the same time, cost pressures demand greater integration from chip vendors. Yet the desired integration is different for smart phones, PNDs, e-book readers, or netbooks.
Intel continues to develop its Atom platform, hoping to break into the high-volume smart-phone market. Demand for Bluetooth, GPS, Wi-Fi, and other connectivity functions continues to grow. To drive down cost, vendors are developing new combinations of these functions.
CPU PERFORMANCE SOARS
Performance is approaching a quantum leap. Next-generation processors will deliver as much as 10 times the peak performance of the original Apple iPhone CPU. Just as with the PC market a decade ago, 1 GHz has become the magic number for CPU speed in smart-phone processors. A further boost comes from the move from the one-instruction-at-a-time ARM11 to sophisticated dual-issue CPUs such as ARM’s Cortex-A9, Qualcomm’s Scorpion, and Marvell’s Sheeva. New dual-CPU designs promise to redouble performance.
End users, however, are likely to be disappointed with the performance of these new processors. As we have seen in other markets, it takes years for software to adapt to the dual-CPU model. Until it does, there will be about a 30% gain in performance after putting background tasks on the second CPU. Dual-issue CPUs also fail to meet lofty expectations: on CoreMark testing, Cortex-A9 delivers a 40% advantage over ARM11 in per-megahertz performance, much less than the 100% gain that ARM claims.
INTEGRATION DOs AND DON’Ts
Another way to reduce cost is through integration. The most effective integration in smart phones combines the application processor and the cellular baseband. Qualcomm is the biggest proponent of this strategy, but other processor vendors are following suit. The Linley Group expects 40% of all smart phones in 2010 to use an integrated processor.
This trend reduces the opportunity for mobile processor vendors, such as Texas Instruments and Samsung, that lack cellular technology. Although TI processors appeared in about 37% of all smart phones in 2009, its share has dropped from 87% three years ago. TI will probably lose the smart-phone lead in 2012.
Because new smart phones require a full set of connectivity functions—Bluetooth, FM, GPS, and Wi-Fi—it is difficult to integrate just one or two of these functions on the processor. Broadcom added Bluetooth and FM to its BCM21551 Zeus processor. However, smart-phone customers still had to add an external GPS chip and an external Wi-Fi chip. After customer feedback, Broadcom redefined the Zeus chip and removed the Bluetooth and FM functions.
Other vendors are developing specialized mobile processors. For e-readers, Marvell recently released the first processor with an integrated e-ink display controller. Vendors targeting in-dash navigation are adding automotive interfaces such as CAN and MOST. Processors for netbooks or smartbooks must support PC-style memory and I/O interfaces. As demand grows for these new applications, application-specific mobile processors will displace general-purpose application processors.
Rather than integrate connectivity into the processor, the primary trend has been to combine multiple connectivity functions into “combo” chips. The most popular combo for feature phones in 2010 will be Bluetooth+FM+GPS. The Linley Group estimates the attach rate for each of these three technologies will exceed 33% in 2010, meaning that they will appear in most feature phones and nearly all smart phones. Thus, it makes sense to combine them into one chip.
Although Wi-Fi has not gotten traction in feature phones, it appears in nearly all smart phones. Today, smart phones can use a Bluetooth+FM+GPS combo along with a separate Wi-Fi chip. Another choice is to use a Bluetooth+FM+Wi-Fi combo plus a GPS chip. The holy grail for smart phones is a single chip combining all four connectivity functions, which are likely to sample in 2010.
Once these four-function combos appear, they will also drive the adoption of Wi-Fi in feature phones. Bluetooth 3.0 depends on Wi-Fi for its physical layer.Essentially, it allows Bluetooth software to take advantage of the higher speed of Wi-Fi, so feature phone designers wishing to implement high-speed Bluetooth today must add a separate Wi-Fi chip, which is relatively expensive. Four-way combo chips will bring down the cost of adding Wi-Fi to feature phones, making Bluetooth 3.0 more prevalent.
What’s next? Integration of cellular RF into the combo connectivity chip makes sense. Like cellular RF, connectivity functions consist largely of analog and RF circuits. A chip combining both can be built in an IC process optimized for RF circuitry. In contrast, integrating cellular RF in the baseband processor requires making the RF circuitry work in a process optimized for digital logic. In early 2010, Qualcomm will sample its QTR chip, the industry’s first to combine cellular RF, Bluetooth, FM, and GPS.
INTEL MAKES PROGRESS
In 2008, Intel rolled out its first Atom processor, known as Menlow, with plans of going after ARM in the mobile market. So far, Menlow has been a big hit in netbooks but a dud in smaller devices. Menlow’s active power is about 2.8 W, versus less than half a watt for most smart-phone processors. Worse, Menlow burns 1 W even when idle.
In 2010, the second-generation Atom known as Moorestown will debut. This time around, Intel focused on reducing idle power, and the new chip set is just as good as an ARM-compatible processor at doing nothing.
Unfortunately, Moorestown needs about 1.4 W to actually run applications. The new Atom also requires two or even three chips to match the feature set of today’s smart-phone processors. It is too hot and bulky for most smart-phone designs, but it will be used in larger devices such as mobile Internet devices (MIDs) and tablet computers.
Intel’s progress shows that the x86 instruction set is less of a problem than previously thought. Moorestown’s power is high, but its performance is higher than that of most mobile processors. As other vendors strain to match Atom using dual-CPU ARM designs running at 1.0 GHz or faster, their designs are exceeding 1.0 W, approaching the power level of Moorestown. Thus, at this extreme level of performance, ARM-based processors and Atom seem to require similar power.
Mobile chip design has always been an extreme technical challenge. Now, chip vendors must also understand the specific and evolving needs of their target markets and integrate the right set of functions. To succeed, these vendors must fine-tune their crystal balls.