Power to the Pocket
With PC sales flatter than an interstate in Iowa, chip vendors are turning to handhelds for a hand up. Intel Corp. has supplemented its Pentium, Celeron, Xeon, and Itanium with a processor product line for PDAs and smart phones, dubbed XScale, to carry its dominance into this emerging market. AMD has stepped into the arena with last year’s purchase of Alchemy Semiconductor. Of course, the established suppliers of handheld CPUs, Motorola and Texas Instruments, have no plans to cede the market to these invaders from the desktop space.
The appeal for chip vendors is obvious: Even though PDA sales slipped last year, users who formerly owned a single, bulky PC are relying on a greater variety of digital devices, and expecting more from them. When you consider the growing appeal of organizer-enabled wireless phones (or is it cell-phone-enabled PDAs?), the stakes are high: Market researchers at IDC predict the global PDA market will grow around 50 percent annually through 2004.
The technical challenges for handheld CPUs are obvious, too — dwarfing, so to speak, even those of moving from the desktop to notebook environment. A typical laptop CPU draws just 25 or 30 watts of power, compared to recent desktop chips’ 60 or 70 watts? Fine; cut that to less than one watt, and we’ll talk. AMD boasts that the Athlon XP is a compact 80 square millimeters versus the Pentium 4’s 146? One of Intel’s cell-phone chips squeezes both a CPU and flash memory into a teensy 13 square millimeters.
We’ll address the handheld CPU market in two articles. This one will check out the chips that power today’s handheld gadgets and see how they compare; next week, we’ll discuss the new handheld processors and applications on the horizon. And right now, the big trend in handheld silicon is the migration to ARM-based designs.
The Chipmaker Behind the Chipmakers
ARM Ltd. is a U.K. company formerly known as Advanced RISC Machines, spun off from the British PC builder Acorn. Rather than selling silicon, ARM sells intellectual property — partners like Intel and TI pay licensing fees for the right to build chips based on ARM’s architecture. Typically, vendors customize their chips to tweak performance, but the core logic — in, the company boasts, almost three-quarters of the 32-bit embedded RISC processors in the world — is all ARM’s. And while the company’s designs have been used in mobile phones and Pocket PCs for years, it’s never had the wide industry support it enjoys now.
The ARM architecture offers vendors a compelling balance of processing power and low energy consumption. “Power [requirement] is generally the most significant aspect for handheld designs as it impacts directly on features perceived by the user, [such as] talk time and standby time,” says David Cormie, product manager for CPU products at ARM. “Size is also very important, as it directly impacts manufacturing cost.”
Successive generations of the ARM architecture have walked a tightrope between adding performance and adding bulk or battery drain. For instance, while desktop and server fans are eager to move from 32- to 64-bit processing, ARM insists a complete 64-bit CPU would be too big and expensive for a PDA or smart phone.
Its latest ARM11 architecture compromises by providing 64-bit data buses between the integer unit and the instruction and data caches, while its Thumb instruction set heads in the other direction by letting programmers mix 32- and short 16-bit instructions. The latter, ARM says, provide memory savings of up to 35 percent while still giving access to a full 32-bit address space.
Other features of ARM11 — introduced last October and not yet fully implemented by any of ARM’s licensees — range from special Java acceleration, designed to provide faster execution than a software-based Java virtual machine, to the ARMv6 instruction set’s single-instruction-multiple-data (SIMD) multimedia extensions for up to quadruple performance with audio and video codecs (see CPU Planet’s October 23, 2002 article on multimedia extensions — Ed.) And while ARM11 and its eight-stage integer pipeline are expected to debut at 350MHz in 0.13-micron-process products, the company’s roadmap leads to 1GHz at 0.10 microns.
Though ARM-based chips are rapidly becoming dominant in the handheld market, no one silicon vendor can claim the same status. So far, PDA and CPU vendors have paired off in relatively neat teams, such as Garmin’s choice of Motorola, Sony’s of Intel, and Palm’s of Texas Instruments. Let’s look at some of today’s variety of offerings, while sorting out chip names you may have heard, such as DragonBall, StrongARM, and XScale.
From the premiere PalmPilot to recent times, Motorola dominated the handheld market with its DragonBall processors — variations on the venerable 68000 CISC architecture used in pre-PowerPC Apple Mac systems. First released in 1995, according to IDC, the DragonBall is found in more than 75 percent of PDAs worldwide, thanks to its use in all Palms (and Handspring Visor and Sony Clie compatibles) until just last year.
Motorola is currently shipping its fourth- and fifth-generation PDA processors, the DragonBall Super VZ and DragonBall MXL/MX1, respectively. The former doubles the speed — from 33MHz to 66MHz — and adds a host of system-on-chip functionality — ranging from DMA and USB to MultiMedia Card/Secure Digital and Sony Memory Stick flash-card controllers — to its 68K-based predecessor, the DragonBall VZ. Sony chose the DragonBall Super VZ for the Clie PEG-NR70V, first of its digital camera- and keyboard-equipped Palm OS handhelds.
Motorola’s fifth generation, however, jumps from CISC to RISC and joins the ARM camp. The DragonBall MX (Media Extensions) chips combine an ARM9 core with highly integrated system functions, running at 140MHz to 200MHz. Performance is comparable to that of Intel’s StrongARM processors (more on them in a moment).
Texas Instruments’ OMAP
When developing its Palm OS 5 response to Microsoft’s surging Pocket PC platform, Palm chose to optimize the operating system for ARM-based processors. When the company released its first OS 5 handheld, the Tungsten T that debuted last October, it put the Texas Instruments OMAP1510 under the hood.
Part of TI’s Open Mobile Application Processor initiative, the OMAP1510 is a dual-core processor that combines a 175MHz, TI-enhanced ARM925 CPU for general OS tasks with a 200MHz TI DSP (digital signal processor) core for demanding multimedia and speech applications. The chip can run business applications, such as displaying Word, Excel, and PowerPoint files, faster and with higher resolution than the DragonBalls that powered previous Palms.
There are still plenty of non-ARM Palm devices available; the forthcoming Tungsten W, for example, is a keyboard-equipped smart phone that runs the older Palm OS 4.1.1 on a 33MHz DragonBall VZ. But most future Palm devices, at least on the high end, will be built around the ARM architecture.
If backers like European smart-phone leader Symbian and TI weren’t enough to make a company confident, ARM must have popped a few buttons when mighty Intel signed up. StrongARM — developed jointly by ARM and Digital Equipment Corp., the latter later acquired by Intel — and XScale are Intel’s brands for ARM-architecture processors, used in what the chip giant calls PCA or Personal Internet Client Architecture products for PDA, wireless, or networking applications.
In terms of architectural generations, says ARM’s Cormie, “StrongARM performance is approximately equivalent to ARM9, while XScale performance is approximately equivalent to ARM11” (although XScale has some differences, such as a seven- rather than eight-stage integer pipeline and what Intel calls Media Processing Technologies rather than ARM’s SIMD instructions).
The newer XScale, Cormie adds, is gradually replacing the ARMv4-based StrongARM in Intel’s product portfolio. One reason is clock speed that blows away the older design: Intel’s fastest StrongARM processor, the SA-1110, runs at 206MHz, which is still on the high side of the handheld spectrum, but the XScale PXA250 is available at up to 400MHz.
As users of the first XScale Pocket PC handhelds have discovered, however, you shouldn’t expect the newer chip to deliver double the performance. Not only is clock speed no more a perfect predictor of real-world results in the handheld space than it is on the desktop, but most PDA buses can’t move data fast enough to keep the XScale satisfied. It would help if Microsoft would write a version of Pocket PC 2002 specially optimized for the XScale instead of the ARMv4 code base, but that doesn’t seem likely — Microsoft doesn’t want to orphan the many StrongARM iPaq owners.
Intel is working hard to get application developers to write fresh code for XScale, optimizing applications by writing for Integrated Performance Primitives (IPPs) and Graphics Performance Primitives (GPPs) that bunch data into libraries which can be quickly read and processed by the XScale CPU. This kind of batch processing could greatly improve bandwidth-intensive operations such as playing video clips or displaying 3D animations.
That begs the question: Why do you need to support video and 3D graphics on a handheld? Intel is clearly betting that PDAs and smart phones will do more than just carry names and numbers in the future, and it’s not alone in making that bet. Next week, we’ll examine how soaring subminiature CPU speeds will be put to use — and how AMD plans to use a little Alchemy and a custom-built MIPS core to start a new chip war in the handheld space.