Embedded Microprocessor Watch

Issue #158 -- 01/27/2003

Editor: Tom R. Halfhill

In this issue:

A Few Suggestions
Peter Glaskowsky - Editor-in-Chief  {01/27/2003}

I wasn’t entirely forthcoming in my editorial last month (see MPR 12/30/02-02, “Toward a Brighter Tomorrow”) when I declined to name specific new ways for semiconductors to add value to our lives in areas such as clothing, food, housing, and furniture. In fact, I can point to a few applications that could double the total world market for semiconductors—from tiny sensors to complex microprocessors.

Some of the smallest chips we’ll ever use may have the highest-volume opportunity. Radio-frequency identification (RFID) tags could eventually replace bar codes on all kinds of consumer goods. RFID tags can be read at a distance, eliminating much of the labor associated with inventory taking and with both wholesale and retail transactions. They can also be overprinted with packaging graphics; don’t underestimate the value of improved aesthetics.

Even smaller chips could eventually be manufactured by the trillions each year: so-called “smart dust.” Chips just 5–10 microns on a side would be individually invisible to the naked eye.

Smart dust could be mixed into building materials, such as concrete and plywood, to report temperatures, pressures, and mechanical stresses during construction, ensuring that homes and office buildings meet all structural codes. These tiny chips could also be vulcanized into tires to detect improper inflation and give advance warning of tire failures. Smart dust could even be ingested or inhaled to monitor our health.
I’m not speaking here of nanotechnology, the future dream of machines built from individual atoms. This is current technology, devices we can start building today. A chip 10 microns square, made with 0.13-micron technology, can have hundreds of integrated components. With the 32nm technology scheduled to arrive by 2009, the component count will be in the thousands. That’s more than sufficient to implement a fairly complicated sensor with local storage and processing, along with a radio transceiver—albeit one that would have to operate at a frequency of about 2THz, midway between the common definitions of radio and light.

Smart houses have been “just a few years away” for almost my whole life. It’s easy to imagine the potential benefits of computer-controlled housing. If my house knows where I am and whether I’m cold or hungry, it can turn off the lights in other rooms, adjust the furnace, and even begin preparing a preselected meal. The old notion of the smart house was based on the concept of a central computer running all the systems, but the smarter approach is distributed computing—with intelligent appliances throughout the house and no single point of failure. With every new generation of microprocessor-controlled ovens, thermostats, and digital video recorders, we get closer to that future.

Our furniture can be smart, too. I don’t want my couch to beep at me just because I’ve opened a second bag of potato chips in one day, but I might not mind being awakened if I fall asleep in front of the TV. I would certainly pay extra for a bed that could detect a stroke or heart attack and call for medical assistance.

Smart clothing is already here. Indeed, it’s old news. The first smart clothing was created by students at the University of California, Santa Cruz in the 1970s to facilitate cheating at roulette. Shoes with MOS Technology 6502 microprocessors, radios, and vibrating elements used a nonlinear dynamics simulation to give the players an advantage over the house.

Today, some 25 years later, most smart clothing still isn’t that smart, mostly because the clothing industry is smart enough to let the consumer-electronics industry provide the smarts. The Scott eVest incorporates no microprocessors, but it can carry several electronic gizmos connected via internal wiring channels. The Burton Amp jacket, co-designed with Apple Computer, has buttons to control Apple’s iPod music player sewn directly into its sleeve—and it’s rugged enough for use by snowboarders.

On a more serious note, the VivoMetrics LifeShirt may help save the lives of people with heart and lung problems. This machine-washable T-shirt incorporates sensors that monitor dozens of cardiopulmonary parameters. The collected data are stored in a Handspring Visor PDA for later evaluation, providing valuable insight into a patient’s health during her/his daily activities.
I don’t want to overlook the many new applications for semiconductors in computer systems. Despite the growth of the Internet and wireless networking, the world is full of computerized devices that don’t talk to each other. Most offices include a digital private branch exchange (PBX) system, and many PBXs today are built around commodity personal computer technology—yet PBXs are almost never connected to the company’s computer network.

There’s no need to maintain what amounts to two independent data networks in each office building. The smarter approach would be to fold the functions of the local-area network (LAN) into the PBX system, using voice over Internet Protocol (VOIP) technology for traditional telephone functions. The phone then becomes the office PC’s network adapter. Costs are reduced by eliminating redundant cabling, and productivity is improved by integrating email and voice mail. Cisco and other companies already offer such systems, but they’ve been slow to catch on.
Perhaps they need to deliver even more value. An integrated voice and data network can be extended to support videoconferencing that works as easily as making an interoffice phone call. Once you have videoconferencing, why not video messaging and document sharing? If these features come with the flexibility of the PC and the reliability of the phone system (not vice versa), they may attract a more sizable market.

I actually like the idea of allowing these LAN-attached telephones to drive the computer monitor directly. This might be the best way to make videoconferencing as reliable as phone calls are today. Monitors would have to become slightly smarter to handle multiple simultaneous inputs, but that should be easy enough.

Monitors need to be smarter for other reasons. Only a tiny fraction of all computer monitors today are ever calibrated to ensure they are displaying the correct colors. Even the best of these calibrated monitors is unable to display more than about half of the colors the eye can see, and changing ambient light conditions—or even the user’s clothing—can alter appearances despite whatever calibration may be in place. Instead, monitors make do with approximations based on simple predictive models.

Digital photography is becoming commonplace, and photo-quality printers are widely available. Random variations from screen to paper should be considered unacceptable, but most computer users take them for granted. Users know that printed photographs look different than they did on the monitor, but they don’t know why.

The truth is that it takes a lot of processing power in the computer, and more in the monitor, and still more in the printer, to correct all these problems. Only recently has it become practical to close the loop with sensors and signal processing to guarantee that what you get is exactly what you want.

Audio devices also use an open-loop control scheme. Audio CDs are encoded with the exact music we want to hear, but no stereo system is designed to make sure we hear it. CD players create a reasonably correct analog signal from the digital bitstream, but that’s the end of the feedback loop. Amplifiers drive speakers with unknown characteristics; cables and physical acoustics add even more uncertainty.

The answer is to deliver the digital bitstream to speakers that amplify the signal and monitor the result, using wireless microphones at the intended listening position. The commercial success of the Bose Wave radio, which offers minor improvements over the quality of previous products, proves there’s a big market for good sound at a good price. All-digital sound systems will require a lot of signal processing, but the going cost of 1GFLOPS is around $20 today and falling rapidly.

I’m running out of room here, and I haven’t even gotten around to explaining the huge markets waiting for portable video devices and secure digital two-way radio systems for police, fire, and air-traffic control. Also, I have also discovered a truly marvelous new computer system architecture that this editorial is too small to contain…

To find out more about Microprocessor Report, please visit: www.mdronline.com.

For Your Reading Pleasure
Peter Glaskowsky - Editor-in-Chief  {01/27/2003}

We rarely do book reviews at Microprocessor Report, but within the past few weeks four new books worth mentioning came into our offices.

Designing Embedded Hardware
John Catsoulis
ISBN 0-596-00362-5, trade paperback, 298 pp.
O’Reilly & Associates, $39.95

A hardware book from the noted publisher of software titles? As far as we know, this is O’Reilly’s first and only book devoted to hardware design. Like O’Reilly’s other titles, Designing Embedded Hardware has an old-style illustration of an animal on the cover. In this case, it’s the shy and elusive porcelain crab, found in tidal pools along the coast of the Pacific Ocean.

This isn’t even a hardware book for software engineers. It’s about the low-level design of microprocessor-based embedded systems. The book contains introductory text, starting with definitions of voltage and current and continuing through the principles of system design and implementation. There are chapters on PIC and AVR microcontrollers, Motorola’s 68000, and TI’s DSP56800. About a third of the book covers peripherals and interfacing. Sample designs are included for SPI, I2C, RS-232 and RS-422 serial links, three types of local-area networks, and many useful analog peripherals.

The author runs his own embedded-systems design firm and has a conversational writing style. The book is heavily illustrated with schematics, waveform diagrams, and sample printed-circuit board layouts. System designers will need additional information to create complete designs, but manufacturers’ data sheets and application notes will usually bridge this gap.

This is the first venture into hardware-design book publishing for both O’Reilly and Catsoulis, and it’s a strong debut. We hope O’Reilly will follow this title with further books about hardware design, especially ones that leverage the company’s extensive experience with software development.

Digital Video and HDTV
Algorithms and Interfaces
Charles Poynton
ISBN 1-55860-729-7, hardcover, 736 pp.
Morgan Kaufman Publishers, $59.94

Is it a coincidence that the price of this volume matches the refresh rate of NTSC video, 59.94 fields per second? I don’t think so. The author makes good use of humor in this book and demonstrates great attention to detail, so I’m sure this is another inside reference.

Poynton is one of the graphics industry’s leading experts on video and television technology. This book builds on Poynton’s earlier Technical Introduction to Digital Video, previously my favorite work on the subject, but it has more than twice as many pages and goes into greater depth on more subjects.

Digital Video and HDTV opens with a comprehensive introduction to the foundations of video—how moving images are captured, processed, and displayed on computer monitors and TVs. Part 2 of the book goes into detail on specific implementations of these underlying technologies, including filtering methods, color science, color coding in standard- and high-definition video, and the operation of video-tape recorders.

Part 3 is devoted to digital video-compression schemes, including motion JPEG, the DV standard used in digital camcorders, and the MPEG-2 standard used in DVD players and some digital theater systems. Unfortunately, Poynton gives short shrift to MPEG-4 as not “relevant to handling studio- or distribution-quality video signals.” In fact, MPEG-4 can be used for high-quality video coding and may become part of the future high-definition DVD standard. MPEG-4 is a complex standard (see MPR 3/29/99-05, “MPEG-4: Way Beyond Video”), but some discussion of it here would have been useful.

Part 4 describes important standards for studio video-production work. These include standard-definition component and composite video for both NTSC and PAL, standard-definition test signals, and the two HDTV resolutions that will likely be widely used in studios: 1,280 ´ 720 pixels and 1,920 ´ 1,080 pixels. The final section provides similar detail on broadcast and consumer standards, principally the analog standards used around the world. High-definition broadcast standards are summarized briefly. The book concludes with some background information on imaging science and a comprehensive glossary.
We recommend this book to anyone with a professional interest in digital video or television standards. No other book available provides such comprehensive coverage of this complex topic.

The Essential Guide to Semiconductors
Jim Turley
ISBN 0-13-046404-X, trade paperback, 240 pp.
Prentice Hall, $34.99

Former Microprocessor Report analyst and current editorial board member Jim Turley has written a solid introduction to the business and technology of semiconductors. The book is easy to read but still contains enough technical depth to provide a coherent explanation of complex subjects. A summary of the business issues related to semiconductor companies is the largest portion of the book, but there are also chapters on chip design and manufacturing, microprocessors, memories, and custom chips. Appendices list important industry standards bodies, conferences, and other resources.

This book is not aimed at experts, but even an expert can learn something from Turley’s easy style. This would also be the ideal book to lend to executives, investors, and coworkers who need to understand what you do for a living.

Network Processor Design
Issues and Practices, Volume 1
Edited by Patrick Crowley, Mark A. Franklin, Haldun Hadimioglu, and Peter Z. Onufryk
ISBN 1-55860-875-3, trade paperback, 352 pp.
Morgan Kaufman Publishers, $59.95

As its preface explains, this book grew out of the first Workshop on Network Processors, held in February 2002. The book consists of two sections, the first covering network processor design principles such as benchmarking, modeling, and compiler technology. The second half of the book describes specific implementations, including Agere’s PayloadPlus, Cisco’s Toaster 2, IBM’s PowerNP, Intel’s IXP2400, Motorola’s C-5e, PMC-Sierra’s ClassiPI, and TranSwitch’s Aspen.

The book contains a roughly equal mix of theory and practice, but even the theoretical portions are of substantial practical value. The section on design principles makes frequent reference to commercial network processors and real-world applications, conveying information that is directly relevant to the many other network processors on the market.

The “Volume 1” reference in the title seems to imply that Morgan Kaufman is planning to publish a series of these books. If so, it’s off to a good start.

To find out more about Microprocessor Report, please visit: www.mdronline.com.

Analog and CPU Wizards Reduce Digital Power
Max Baron - Principal Analyst  {01/21/2003}

In November 2002, National Semiconductor Corp. and ARM announced a strategic business relationship to jointly develop and market power-efficient systems that, they claim, will increase the battery life of handheld portable devices in several stages—from 25% to as much as 400%. The two companies’ joint effort will leverage ARM’s penetration in the mobile-phone market and National Semiconductor’s expertise in analog design and power management.

NSC and ARM’s joint project aims to create circuits, software, and tools that address three energy-consumption tasks: frequency reduction, minimal voltage levels to support it, and reduction of leakage current.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0121/170301.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

TI DSPs Play the Limbo Game
Max Baron - Principal Analyst  {01/21/2003}

In December 2002, Texas Instruments introduced two DSPs—the dual-MAC TMS320VC5501 and the TMS320VC5502—that it claims can deliver high performance, reduced power consumption, and low price. The less-expensive one, the C5501, clocks at 300MHz and sells for $5.00 in 10,000-unit quantities.

The C55x DSP core can control the clocking of its individual functional units. The core monitors which parts of the chip are in use, powering them off when they are not needed. Programmers receive an additional tool for power reduction: user-configurable idle domains they can use to customize power consumption for a specific application.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0121/170302.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Transmeta Charges the Embedded Market
Tom R. Halfhill - Senior Editor  {01/13/2003}

With the debut of Intel’s new Pentium M mobile-PC processor (code-named Banias) only months away, Transmeta is trying to expand the potential market for its competing x86-compatible chips. The company has announced a new line of Crusoe SE (Special Embedded) processors aimed at embedded systems that need to run x86 software with high performance and relatively low power consumption.

For now, Crusoe SE processors are identical to the TM5800 Crusoe chips Transmeta sells for mobile PCs and blade servers; architectural differentiation will come later. However, Transmeta does promise five-year availability for Crusoe SE parts and certifies them for 24/7 operation without a cooling fan at junction temperatures up to 100°C and case temperatures up to 85°C. In addition, before shipping, each Crusoe SE must pass a 24-hour burn-in test while running a real-time operating system (RTOS).

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0113/170201.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Intrinsity Delivers On Its Promise
Max Baron - Principal Analyst  {01/06/2003}

Intrinsity, a company founded in 1997 and headquartered in Austin, Texas, has received its first samples of FastMath, a chip based on a MIPS32-compliant core, and has been able to operate the chips at 2GHz. FastMath is the company’s most important contribution to computer architecture. It uses the FastMIPS core as part operating-system processor, part controller of a 16-SIMD coprocessor. At 2GHz, the SIMD engine can compete with massively parallel arrays and with high-performance dedicated DSPs.

Intrinsity has benchmarked FastMath on simple DSP workloads.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0106/170101.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Transmeta Shows New TM8000 Astro
Tom R. Halfhill - Senior Editor  {01/06/2003}

Transmeta is showing customers first samples of its next-generation mobile PC processor, promising to deliver production quantities by 3Q03. Although Transmeta is withholding most architectural details about the new TM8000 Astro until it’s closer to release, the company says the processor will offer higher performance and additional power-management features.

The biggest difference between the TM8000 and current TM5800 Crusoe processors is a much wider VLIW core. The TM8000 has 256-bit-wide instruction words that can issue up to eight 32-bit operations in parallel, compared with the TM5800’s 128-bit-wide instruction words containing four 32-bit operations. Although the common wisdom is that typical PC applications yield little instruction-level parallelism beyond four-way execution, it’s possible that the TM8000 can do better. Its “code-morphing software” dynamically translates x86 instructions into native VLIW instructions, often reducing complex x86 instructions into two or more simpler operations suitable for parallel execution. (See MPR 2/14/00-01, “Transmeta Breaks x86 Low-Power Barrier.”)

In addition to wider instruction words, the TM8000 has many new instructions, instruction types, addressing modes, and registers; it’s virtually a new architecture. Transmeta has the freedom to overhaul the processor’s architecture so extensively because the VLIW core is invisible to x86 software, which runs atop the emulation layer. The native VLIW architecture is visible to only Transmeta’s system programmers, who wrote a new version of the code-morphing software for the TM8000.

Transmeta won’t publicly discuss whether the TM8000 will support the latest x86 media-processing extensions, such as Intel’s SSE and SSE2. (The TM5800 doesn’t.) However, Transmeta CTO David Ditzel says the TM8000 will have “significantly enhanced” floating-point performance, which is a notable feature of SSE2. Because the TM8000 will hit the market in 3Q03, about the same time MPR expects Intel to roll out SSE3, it will be unfortunate if Transmeta doesn’t support at least the x86 extensions introduced to date. State-of-the-art media processing isn’t critical for the TM8000, which is designed primarily for mobile computers and business applications, but that situation may change as more users follow the trend of replacing desktop PCs with laptops as their primary system.
Ditzel says Transmeta will announce the TM8000’s clock speeds when the chip is closer to introduction. For now, he says only that the TM8000 will exceed 1GHz, a clock frequency already attained by the TM5800. Transmeta’s foundry, TSMC, will manufacture the first TM8000 chips in a 0.13-micron process—the same process TSMC uses for the TM5800.

Although mobile PC users are generally less obsessed with clock frequencies than desktop users are, mobile processors that run at gigahertz speeds command higher prices and compete more effectively with Intel’s fast mobile Pentium 4 and Pentium III. It will be interesting to see how Transmeta and Intel tune their marketing strategy after Intel’s 1H03 introduction of Banias, a new mobile-PC processor. Banias will run at lower clock frequencies than the mobile Pentium 4 does, yet it will deliver strong performance. Intel, like Transmeta, may downplay clock speeds in favor of promoting low power consumption and longer battery life. (See MPR 11/25/02-01, “Intel Spills the Beans About Banias.”) There will almost certainly be a war of words and benchmarks as the two companies try to establish which mobile processor delivers the better balance of power consumption and performance.

Blade servers are a potential growth market for Transmeta’s cool-running chips. In rack-mounted blades, high processing density and low heat dissipation are more important than maximum performance. Transmeta is already branching out in this direction. Earlier this month, RLX Technologies introduced its new ServerBlade 1000t, which uses a 1GHz TM5800 to replace an older 667MHz version of the chip. Los Alamos National Laboratory is using a 480-processor cluster of ServerBlade 1000t boards.

Transmeta is also trying to enter the embedded market, where low power is definitely a virtue. The relatively large memory requirements of Transmeta’s code-morphing software and translation cache will rule out many embedded applications, but lucrative opportunities still exist for an x86-compatible processor that runs at 1GHz and sips only a few watts of power. Examples might be set-top boxes, personal video recorders, and other consumer appliances in which fanless operation is a plus.

To find out more about Microprocessor Report, please visit: www.mdronline.com.