Embedded Processor Watch

Editor: Cary D. Snyder

Contributors to this issue: Max Baron, Markus Levy, and Cary Snyder

In This Issue:

• Editorial: Waiting for The App
• NEC Processor Goes Out of Order
• Ripping Good Audio Optimization
• Narrowed I/O Options Part II
• MIPS Processor Hits 1GHz

Of Interest:

• Microprocessor Forum 2001 - News Update

Editorial: Waiting for The App
By Max Baron {9/24/01-02}

A few days ago, Intel broke through the 2GHz barrier-a historic achievement, because of both the frequency itself and the availability of chips in volume quantities. Really good athletes make difficult feats look easy. And it did look easy when, one more time, at San Jose's Fairmont Hotel, with red lights reflecting off huge balloons, Intel told the press it had achieved its goal. The professional presentation and the quick demos made it all look very easy. Automotive design applications and music ripping followed medical applications. Rapid creation of videos was demonstrated: montages of background, images, and sound became a real show. You walk out in awe, thinking of the hundreds of man-years that went into this, confident that there are more "gigahertzes" where these came from-and then you remember the killer app.

The killer app, as it was expectantly dubbed a few years ago, was supposed to be a program that required top performance (and memory) and kept coming back for more. Its appeal was to be so great that it would trigger mass desktop purchases by people who had never before used a computer. Additionally, this app would make existing computer owners run to the nearest store to buy a new, more powerful computer. It would bring revenue growth and comfortable margins back to semiconductor and system vendors alike.

An application has arrived, but it's not a program. Instead of channeling more revenue into the PC area, it's taking the money unto itself. I'm talking about the Internet. Internet connectivity has become a sink of money or a source of major revenue, depending on which side of the cash register you occupy. It has also become a sink of time.

High-speed connectivity has become more important than high-speed local processing in a desktop. Web sites use high-quality animation to differentiate themselves from each other; audio and video bandwidth requirements compete with downloads of ever-increasing file sizes. Depending on a subscriber's being inside the magic mile or outside it, the charge for ADSL can run from approximately $40 per month to $80. Add about $20 per month if, for professional reasons, the subscriber needs to keep an email address that was registered with a previous access provider. A 1GHz bare-bones computer sells today for about $700. Assuming a two-year depreciation (or a new impulse to buy a new computer), we compare $700 spent for a computer with $960 for ADSL, or worse, with all the problems tacked on, to $2,400.

Leaving aside the reasons for the high cost of connectivity and server storage, let's concentrate instead on using the new gifts of processor performance. They should help reduce, as much as possible, the need for bandwidth and the time spent in searches. The compression of voice and music, still images, and video has always been dependent on the throughput capabilities of the encoding and decoding processors. The lower the throughput expected at the decoding end, the more bandwidth required for the communication itself, helping connectivity and storage costs soar upward. With more decompression capability, we should expect a reduction in needed bandwidth, using known methods. More may be done to reduce bandwidth: aside from compression that reduces sound frequency or image sharpness, psychological effects that help compress voice and music may be able to make further use of performance and may be extended into video.

The new performance capabilities should also help reduce the time users spend searching for relevant information-a problem largely ignored until now. Search engines, rich in storage of databases, are forcing Internet users to ask them questions in baby talk. Worse, these engines are coming up with answers that most of the time are irrelevant to the real search; some of them are not even willing to prioritize by date or "relevance" the order in which they provide answers or support searches within subsets of searches.

Other search engines will let you "AND" two or three words but then give you (for good measure?) all the text they can find that contains any one of the words. How often do we get a response that reads "24,560 documents found, displaying the first 10?" It seems like a long time ago that there were a few programs a user could buy to help poll several search engines. But the programs were not much more customizable than the search engines on the Web and quickly succumbed to the "ask" and "mamma" type of engines. The world was too busy making quick money by creating complex, animation- and video-loaded, time-consuming, beautiful but hard-to-read pages and Web sites. Many of these beautiful new Web sites couldn't be found easily, because users had only the tools offered by the popular search engines.

The new gifts of processor performance may be able to help provide quality media at lower connectivity demands. Additionally, high processor performance can increase user productivity in text and image searches, thus removing some of the barriers that prevent increases in general-purpose computer use. They may help create virtual private networks (VPN) and distributed storage that will make communications and databases secure, easy to update, and almost impossible to eradicate.

As Intel and other semiconductor manufacturers take performance beyond 2GHz, they may want to encourage development of a new app that will make the Internet experience more efficient and less costly. And in the process, they may shift some of the revenue back to themselves.

NEC Processor Goes Out of Order
By Markus Levy {9/10/01-01}

NEC has typically provided processors that have low to midrange complexity, offering devices that range from tiny 4-bit microcontrollers to high-end, yet conservative, 64- bit microprocessors. This image may have changed, however, with the company's recent Embedded Processor Forum introduction of its newly designed VR5500 architecture. The most notable features of this superscalar processor are its 10-stage pipelines combined with out-of-order (OOO) execution and register renaming. The least notable feature is NEC's releasing of the first VR5500-based device at only 300MHz in its 0.13-micron process, but there's more to good performance than clock rate alone.

The VR5500 is upward compatible with NEC's mature VR5432, allowing customers to continue using tools, operating system, and software based on the MIPS IV ISA. The VR5500 is built from the ground up to be a scalable architecture and is loaded with plenty of headroom to boost the operating frequency over time. Don't be fooled by the company's initial 300MHz device; NEC is taking a conservative approach to guarantee volume production for the digital consumer market. (The full version of this article is available online to Microprocessor Report subscribers at: http://www.mdronline.com/mpr/h/2001/0910/153701.html)

Ripping Good Audio Optimization
Hot Consumer App Leads Recovery Charge
By Cary D. Snyder {9/10/01-02}

There is no denying the popularity of the Internet as a conduit for disseminating digital music; look at how much publicity Napster and the MP3 compressed audio format have received. In-Stat market forecasts predict that between 10 million and 30 million flash-based digital music players will be sold annually by the year 2004. In this market, audio-chip manufacturers Cirrus Logic, Micronas GmbH, and Texas Instruments have taken the lead in fueling the skyrocketing use of digital music by providing silicon for digital music processing. These silicon manufacturers are now developing third- and fourth-generation processors for portable digital audio players. Cirrus has the Maverick EP7409, Micronas has its new Zenon architecture, and TI is busy working to bring the DA250 to the market-indicating digital music is here to stay, with MP3 currently the most popular digital audio compression algorithm.

Integrated MP3 chips will do much to bring the retail price of flash-memory-based players below $100 while doubling battery life-a winning combination. The most important drivers for manufacturers of many portable digital audio players are cost to manufacture and price to the end user. To help meet OEM cost requirements, silicon manufacturers are increasing integration in an attempt to lower costs for player products. The implementation is complicated by the need to have the same chips support full-featured, higher- margin products.

Providers of digital music processor chips offer a prime example of the way intelligent integration leads to success-lower-cost bills of materials, smaller players, longer battery life, and a number of other feature improvements. An In-Stat/MDR study (New Products and Technologies Gradually Gaining Acceptance, www.instat.com/catalog/cat-mm.htm#mm0101mi) estimates the declining average price a person would be willing to pay for an MP3 player. From the information in this study, it is easy to conclude that the market will experience a significant increase in sales as the price of MP3 players goes below $100. Additional information on the increasing popularity of digital music is available in a related In- Stat/MDR report (Portable Digital Music Players Ride the MP3 Wave: http://www.instat.com/abstracts/mm/2000/mm0012da_abs.htm).

This article examines three different approaches to creating silicon for digital music players: Cirrus Logic's third-generation EP7409, which combines a 74MHz ARM7TDMI with a companion 24-bit DSP; TI's use of 55x DSP cores in its newest chip-the DA250; and Micronas GmbH's new Zenon ARM-based processor architecture and audio-processing companion chip. (The full version of this article is available online to Microprocessor Report subscribers at: http://www.mdronline.com/mpr/h/2001/0910/153702.html).

Narrowed I/O Options Part II
The PCI Successor Is 3GIO (Serial PCI)
By Cary D. Snyder {9/4/01-01}

Future PC connection standards are clearer now. According to the PCI Special Interest Group (PCI-SIG), the Intel-born 3GIO is in the catbird seat as the heir-apparent replacement for the ubiquitous PCI. Backers of HyperTransport and RapidIO realize that 3GIO (or whatever moniker PCI-SIG chooses) will be the narrow I/O bus replacement for PCI. The PCI-SIG general membership is expected begin to review new PCI standard for completeness and accuracy early next year.

The real push for and acceptance of 3GIO will come from increased bandwidth at the lowest interface-pin count. The initial 3GIO specification's value of 100MB/s per pin translates into the lowest cost-per-pin implementation option for a variety of bandwidth requirements. The good thing about having the PCI SIG develop and administer this specification is that current members can work together to create a smooth transition. There is no reason to rush this process, as current PCI/PCI-X solutions fulfill today's bandwidth needs.

HyperTransport and RapidIO will find their own niches in network communications applications. What is most interesting about these standards is that FPGA vendors Xilinx and Altera are leading ASIC vendors in production release of physical-layer components for RapidIO and HyperTransport.

Intel's claim that the 3GIO specification has long-lasting attributes is valid; its 10-plus years of expertise in developing and driving successful I/O standards-PCI, AGP, USB, and InfiniBand-are being put to good use. Intel's willingness to share its expertise with the entire digital electronics industry should be applauded-not lambasted as domineering control. This article examines the technical fit, form, and function of narrowed I/O buses. (The full version of this article is available online to Microprocessor Report subscribers at: http://www.mdronline.com/mpr/h/2001/0904/153601.html).

MIPS Processor Hits 1GHz
PMC-Sierra Discusses Details at Embedded Processor Forum
By Markus Levy {8/27/01-02}

PMC-Sierra's MIPS Processor Division, formerly known as QED, has a history of developing high-performance MIPS processors, particularly for high-end networking equipment and printers. At the Embedded Processor Forum 2001, PMC- Sierra demonstrated the continuation of its high- performance trend as it unveiled the details of its double- headed RM9000x2, a dual-core device capable of reaching speeds of 1GHz and having more than the average number of high-speed buses use for moving data around. (The full version of this article is available online to Microprocessor Report subscribers at: http://www.mdronline.com/mpr/h/2001/0827/153502.html).

Upcoming Events of Interest:

Microprocessor Forum 2001 News Flash

!! MPF 2001 News Flash !!
Special Bonus for our Registered Conference Attendees:
Monday Night Welcome Party will feature Federico Faggin, co-founder of Synaptics, who led the design and development of the world's first microprocessor and more than 25 integrated circuits. He was also inducted into the National's Inventor's Hall of Fame for the co-invention of the microprocessor.

Movie Night will be at the San Jose Tech Museum's IMAX theater on Wednesday October 17, starting at 6:15, The movie will be Shackleton's Antarctic Adventure. You won't want to miss all the fun and surprises we have in store for you.

Due to the overwhelming number of first time introductions planned at this year conference, we have decided to begin the conference one half day early. The conference will now begin on Monday October 15th at The Fairmont Hotel in San Jose, CA.

Here is what's happening:

• Tuesday's schedule was so jam packed with 1st time introductions that we had to add an extra half day to the conference.
• Conference now set to begin on Monday October 15th at 1:00 PM.
• Register before August 31st for the early bird discount.

This is one event you will not want to miss, so register now while you're thinking about it!

Complete program details and fast on-line registration are available now at http://www.mdronline.com/mpf, or call us toll-free at 800.527.0288 or (480.483.4441 (AZ) outside North America).