A full year of rather positive growth for embedded microprocessors (MPUs), microcontrollers (MCUs), and digital signal processors (DSPs) is now under our belt, and the general economic picture has improved as well. Gartner Dataquest is forecasting 20.1% revenue growth for all semiconductors in 2004, following a forecasted 11.7% growth in 2003. This is good expansion of the market, though not the extreme swing that we're accustomed to in the semiconductor industry. A lot of the unknown remains for 2004.
Nearly all end-equipment categories are stable and increasing, but no one application is really booming. There's a diverse demand rather than one segment pulling all others. Cell phones are churning slightly, and forms of 3G are teasing. But the features that catch on seem to be unique to geographic areas. More PCs are shipping, but there are no critical upgrades forcing the purchase of new PCs. The Internet continues on its merry way, cluttered with spam, pop-ups, and quickly circulating humor in large files.
Consumer electronics is chugging along, confusing the unenlightened with ever-changing standards. "Multi" seems to be the dominant part of the word multimedia. Clear winners are hard to identify. This past Christmas didn't really bring forth any "must-have" electronics present. It will be another year or two before the new generation of video-game consoles appear. In this arena, China looms large as a near-vacuum of a market, but foreign businesses find it difficult to become established or claim any real dominance. Anyway, China has its own idea of where it wants to go.
Semiconductor prices are low, and essentially all end-equipment has great price pressure. Volumes are up, but pricing is not. Some markets have boomed, yet they were either relatively small or so quickly overcrowded that revenues never really got to enjoy their prime. The number of offerings in home networking, WiFi, Bluetooth, and MP3 quickly commoditized the markets. High prices are limiting the market for exciting new display technologies like LCD, plasma, and Texas Instruments' digital light processing (DLP).
Fab utilization is at a critical point right now. The fabs are essentially at capacity, with availability only in certain fabs or certain technologies. Yet there's little excess capacity, and equipment orders are increasing slowly. Any real surge in chip demand will force troublesome shortages and could spur price rises. How well chip price increases can be absorbed by a moderate demand in end-equipment may test the elasticity of the market. This year could be rocky.
Global economics have stabilized and are expected to improve. So far, though, few jobs have opened up, and in their second- or third-year hiatus, opportunities for the returning jobs moved to a different geography. Without significant employment increases in higher-level jobs, the consumer's disposable income and the demands for information technology equipment won't return to levels that cause the boom that we like so much in semiconductors. However, programmable processors dominate the future of electronics.
MPUs, MCUs, and DSPs will grow 18.0% in 2004. Additional processor growth is hiding in the ASICs and ASSPs (application-specific standard products) that contain cores in ever-increasing quantities. The distinction of MPUs and MCUs from DSP and standard products-general-purpose and application-specific-blur as Moore's Law advances through time.
The flexibility afforded a system that's designed around a programmable MPU, MCU, or DSP lets designers quickly adjust the end product for an uncertain and rapidly changing market, even after leaving the factory. Coupled with flash memory, a network connection, or maybe a service call, a piece of equipment can be given new life with a programming change to correct errors, adapt to new standards or use patterns.
Processors themselves have become more flexible. As the most widely licensed processor architecture, ARM seems to be available in a vast array of varieties. A year ago, Philips and Sharp announced ROM or flash-based MCUs running ARM code. These MCUs bring that universal instruction set to a new category of applications, whether they're self-sufficient (as MCUs are the original system-on-a-chip) or they utilize off-the-shelf configurations. This also opens up the lower end of the processor market to more use of ARM. A handful of chips emerged in 2003.
Configurable processors like ARC and Tensilica have gained ground, often with multiple cores on one chip. Although it's not apparent that such designs will overwhelm the general processor market, in certain niches, designers can devote the resources to crafting instructions that get the needed performance. MIPS Technology Inc. added a Pro Series to the MIPS architecture that provides similar configurability, with the huge benefit of a well-established base instruction set. Adding special instructions to program code that was refined over the last 10 years sounds much better than starting from scratch.
Vendors of DSPs were challenged last year. BOPS and 3DSP slipped away, but StarCore is pulling itself up by the bootstraps (again). A more independent company has made a synthesized version of the scalable DSP, offering it for licensing by others besides Motorola, Agere, and Infineon.
Primarily focusing on cellular handsets, StarCore may start getting real traction in the next year. Meanwhile, LSI Logic embraced the ZSP, promoting it as a licensable DSP core in LSI and IBM ASICs. ParthusCEVA, now known as just CEVA, is revitalizing the old DSP Group cores with CEVA-X. The company expects to extend its licensing program with much higher performance and greater varieties of DSP cores. Motorola moved closer to a concept it calls Reconfigurable Compute Fabric (RCF), which adds the instruction-set level of programmability to an array of very fast processor cores. The cores can be reconfigured on-the-fly to handle a vast diversity of application demands.
Intel is making a real nuisance of itself in cellular handsets. The ARM processor Intel got from DEC looks really good as XScale. Along with the DSP architecture it built with Analog Devices, known as MicroSignal Architecture (MSA) at Intel and Blackfin at Analog Devices, XScale can be creatively coupled with Flash in multichip packages and on silicon. The full-court press Intel is putting on XScale and MSA has hit paydirt in handheld PCs and PDAs, and it may be making its way into smart phones and 3G handsets. Bold predictions of capabilities, flexibility, and performance have created plenty of buzz among handset vendors, even if users are still coming to grips with the possibilities.
TI, the mainstay of DSP, is suffering from its own success with so many new challengers. It's holding up well against the onslaught, but it's getting hit from a number of fronts. Nonetheless, the attraction of a DSP core in an ASIC steers users away unless they work completely with TI, since the 320C architectures aren't licensable.
Renesas will emerge as the top vendor of MCUs in 2004 because the Hitachi and Mitsubishi MCUs all come together under that one roof now. But grabbing market share from competitors is the true sign of success. The last year has been tough for many MCU vendors, with prices dropping on 16-bit MCUs to attract 8-bit MCUs up the chain. (Similar cost-cutting is occurring with 32-bit chips.) The market is growing for 16- and 32-bit MCUs, but at lower-than-expected ASPs (average selling prices). More peripherals, including exotic flavors such as Ethernet, are showing greater sophistication in the highly integrated MCU and their markets.