Embedded Microprocessor Watch

Issue #159 -- 02/24/2003

Editor: Tom R. Halfhill

In this issue:

Nomadik Roaming in 2003
Max Baron - Principal Analyst  {02/24/2003}

Nomadik, the new chip from STMicroelectronics (ST), is designed to deliver its best performance and longest battery life for users that favor multimedia and Internet communications over lengthy processing of applications, but it’s no slouch in the applications area either. ST expects to use variants of the family it calls Nomadik in mobile handsets, personal digital assistants (PDA), Internet appliances, and automotive entertainment systems.

ST opted for a distributed processing model that uses accelerators for data-intensive functions and a general-purpose processor for applications and control. The company’s architecture decision is not surprising, since ST is one of the pioneers and market leaders in dedicated audio and video codecs.

Nomadik’s peripherals and memory interface resources follow the microarchitecture recipes that have become accepted in cellular telephony. The chip’s microarchitecture, however, has been laid out to define three separate noninterfering domains of intense activity: video, audio, and application processing—each with its own processor, local memory system interrupts, and DMA services.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0224/170801.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Here Today, (Maybe) Gone Tomorrow
Markus Levy - Senior Editor  {02/24/2003}

Since this is Microprocessor Report, I’ll take the position that the microprocessor is the most important component of every system design. This implies that a system designer must perform extensive due diligence on a short list of processors.

There are three fundamental questions every system designer should ask before making the ultimate decision. First, will the processor fulfill the needs of the application? Second, will the processor be available in the future? Third, will the processor vendor be around in the long-term, or even short-term, future to develop successors, ensuring the application's ability to evolve?

The answer to the first question is easy, because it represents a quantitative process that’s performed using a thorough technical analysis, an evaluation of benchmark results, and a reading of user manuals. The other two questions pose more serious challenges for any system designer, because their answers could mean life or death for a product.

Will the processor and the processor vendor be around in the future? The importance of these questions has grown to enormous proportions during the past few years, owing to the volatile economy in the high-tech industry. Many large, stable processor vendors have struggled to maintain profits, while many smaller vendors have literally disintegrated. A good case in point is Lexra. A few months ago, we published an article on Lexra’s new and interesting multithreaded processor technology. (See MPR 11/25/02-03, “Tying Up a MIPS32 Processor With Threads.”) Lexra claimed customers were on the way, but in the next couple of months the company disintegrated so extensively that Lexra’s Internet domain is now up for sale. That's bad news for any customers that may have locked the new Lexra processor into their designs.

BOPS has met a similar fate to Lexra. Likewise, TriMedia recently expired, although that company had been flailing for several years. Fortunately, TriMedia’s parent company, Philips Semiconductors, has taken the technology back in house, presumably to provide continued support to licensees such as National Semiconductor. Another example of a deceased company and technology is Chromatic, which had a decent architecture. Before it failed, Chromatic had approximately 250 people working on software support for two to three years.

Obviously, this is a short list of processor companies that have met their demise; it doesn’t even include the vanished hordes of network processor companies. It’s sad to watch these companies fail, especially for the hopeful employees that are thrown out of work. Their customers are often hurt, too. All the positive attributes of a processor are worthless when the vendor vanishes after its processor has been designed into the customer’s product. There’s not much one can demand from a bankrupt company.

Of course every company has its fluctuations. Even processor giants such as Intel, Motorola, and Texas Instruments (to name but three) sometimes have delivery problems or even decide to cancel a product. Large companies may also be plagued with inflexibility, making it more difficult to change directions on a product definition for the benefit of a single customer. On the other hand, most large and successful processor companies have spent enormous amounts of time and energy establishing and maintaining corporate-level relationships to help make customers feel cared for.

The small processor vendors that are smart and successful are those that recognize, appreciate, and cherish customer relationships. If their entire customer relationship were based solely on technological superiority, it is definitely more at risk than a relationship having the proper balance. True, some companies become successful purely on technology, but the ones that have survived over the years have also evolved into dedicated marketing machines. Long-term differentiation is based not on continued technology delivery alone but also derives from product branding, customer experience, and far-reaching distribution channels.

What can you do to minimize the risk to your company’s future if you select a processor from a small vendor? First, let’s establish the most probable reason for even considering a small vendor. Simply put, unique processor technology may allow the rewards to outweigh the risks: the unique technology may be the necessary ingredient to give your product the competitive advantage. In addition, small vendors can occasionally do things established companies cannot do; there’s an advantage to not having an established infrastructure. Furthermore, if you’re not a top-tier customer, the large processor vendor may be unable to provide the attention and hand-holding you need to bring your product to market.

Although we advise caution, we believe there will be small processor vendors that beat the odds; however, time is needed to tell the whole story. Improv Systems may be an example. The company recently closed another round of funding, obtaining enough cash to support itself for the next 18 months (perhaps enough time to outlast the recession). In general, any startup must publicly announce solid design wins to make it easier to prove their momentum and long-term viability. Although it may appear obvious, take advantage of leading-edge technology by exploring opportunities with small vendors, but watch your step.

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

Embedded Processors Chill in 2002
Markus Levy - Senior Editor  {02/18/2003}

Each year, Microprocessor Report’s analysts and editorial board work together to select the best high-performance embedded processors on the market. The purpose of this article is to announce the winner of the MPR Analysts’ Choice Award for Best High-Performance Embedded Processor in 2002. At the beginning of our decision process, we nominated six outstanding processors and then carefully sorted through all the technical and market-related details to reach our final decision.

Some of the nominated processors compete head-on for the same applications, whereas others have unique positions in the embedded marketplace. Therefore, the MPR analysts judged processors in this category by their level of integration as well as by the I/O subsystem that ties the processor cores and peripherals together. Our judgment was also based on the engineering achievement and innovation used to design and manufacture these device We also analyzed these chips on the basis of their practical value, for factors that include price, power, and development-tool support.

Among the six nominees, one is a returning winner of the Analysts’ Choice Award for Best High-Performance Embedded Processor Chip in 2001—Broadcom’s BCM1250. We had planned to bring back Motorola’s PowerPC MPC7455, another winner from 2001, but found out from Motorola that the company was secretly sampling the new PowerPC MPC7457 (code-named Apollo). We’ve also brought back PMC Sierra’s MIPS-based RM9000x2, a nominee from 2001. The new contenders for 2002 are IBM’s PowerPC 440GX, Intrinsity’s FastMIPS, and NEC’s MIPS-based VR7701.

Broadcom’s BCM1250 can be described as a general-purpose dual-core SoC, although it includes many features that help position it for network and communications infrastructure markets. In addition to the dual 800MHz, 64-bit MIPS processor cores (SB-1), the 0.13-micron BCM1250 includes a HyperTransport(HT) fabric, 10/100/1,000 Ethernet MACs, dual 55Mb/s serial I/O channels, a PCI interface, and integrated system I/O. The BCM1250 ties its cores and peripherals together using Broadcom’s proprietary Z-bus, a scalable split-transaction multiprocessor bus. Similar to Broadcom, PMC-Sierra targets the RM9000x2 for the network and communications markets, but the device is best described as a dual-core general-purpose processor. The central ingredient of the RM9000x2 is its buffer pool, a shared memory fabric connecting the on-chip bus controllers and the 64-bit MIPS-based CPUs (E9000). This buffer pool combines a multiported register file and a crossbar switch, and it supports buffered concurrent data transfers among all on-chip ports. The RM9000x2 is the first dual-core device to sample at 1GHz.

IBM and Motorola are our two PowerPC-based nominees. Motorola’s MPC7457 is built in Motorola’s HiPerMOS7 and uses 0.13-micron lithography and silicon-on-insulator (SOI) technology, along with copper interconnects. The MPC7457 runs 25% faster than the MPC7455 and consume approximately one-half the power (at the same clock frequency). It uses the same seven-stage pipeline as the MPC7455, and, combined with AltiVec vector unit, it is the heavy-lifting champion in single-core embedded processors, according to EEMBC benchmarks. The 440GX is IBM’s highest-performance integrated processor. It contains the superscalar PowerPC 440 core, the 128-bit CoreConnect bus, an I2O messaging unit, 256K of on-chip SRAM, and two 10/100/1,000 Ethernet ports with TCP/IP hardware-acceleration logic, in addition to two 10/100 Ethernet ports. IBM fabricates the 440GX in its 0.13-micron CMOS CU-11 high-volume copper process, giving the processor a target frequency range of 500–600MHz and plenty of performance headroom.

Intrinsity’s FastMIPS is the first embedded processor to reach the 2GHz mark. The FastMIPS is based on a new MIPS32-compliant core that connects to a half-frequency 1MB L2 cache, served by the chip’s integrated DDR-400 SDRAM controller, and a 4GB/s interconnect fed by two 1GB/s full-duplex RapidIO ports. In 2002, the MIPS-based VR5500 returns in a different incarnation, as the processor core inside the VR7701. With this integrated chip, NEC has focused on conservative high-volume production of this integrated processor based on the superscalar out-of-order VR5500 core (also used in the VR5500 standalone chip). In addition to the processor core, the most significant features in the VR7701 are its integrated 256K L2 cache, memory controller, PCI-X 133 interface, and dual 10/100Mb/s Ethernet MACs.

And the winner is…you'll have to read the article to find out.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170703.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

The Long-Distance Runner
Max Baron - Principal Analyst  {02/18/2003}

During the next few years, the search for energy-efficient computing will become more important than the drive for performance. Multiple solutions of varying complexity have appeared, currently leaving most of the power management to the chip designer but soon expected to involve more software know-how.

This year, five low-power embedded microprocessors were nominated for MPR’s Analysts’ Choice Award: AMD’s Au1100, Intel’s PXA250, Motorola’s DragonBall MX1, NEC’s VR4131, and NeoMagic’s MiMagic 5.

The PXA250’s flexibility and advanced power-management approach won it the MPR Analysts’ Choice Award for Best Low-Power Embedded Processor of 2002.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170704.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Extremely High Performance
Max Baron - Principal Analyst  {02/18/2003}

Computing workloads continue to underscore the importance of performing short, relatively simple operations on enormous quantities of fast-changing data. Classic instruction-intensive processors are extending their ISAs to cope with the onslaught of high-speed data from sources such as multimedia and communications. Clock-frequency increases and modest parallelism have produced sufficient performance for the individual user but have been unable to offer a practical solution for systems that must serve tens and hundreds of people.

Two to three years ago, 3G communications were hailed as the principal catalyst that would enhance freedom of movement via portable computing. With this optimism at its apex, many companies tended to underestimate regulatory and commercial hurdles still facing 3G. As in the network-processor business not long ago, companies began to develop architectures that would meet massive data- and signal-processing needs. Quite a few parallel architectures popped up, and some of them—the ones that offer flexibility and performance—will make it to the top.

This year, six extreme-performance processors were nominated for MPR’s Analysts’ Choice Award: CMU/ST Microelectronics’ PipeRench, Intrinsity’s FastMath, Micron’s Yukon, NEC’s DRP, PACT’s XPU128, and Sandbridge’s SandBlaster. Intrinsity’s FastMath offered the simplest, most complete product, and won the MPR Analysts’ Choice Award for Best Extreme Processor of 2002.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170705.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Soft Cores Gain Ground
Tom R. Halfhill - Senior Editor  {02/18/2003}

Although 2002 was a soft market for almost everything high-tech, including embedded intellectual property (IP), soft microprocessor cores made great strides last year. All the major vendors introduced new or enhanced versions of their leading synthesizable products. Microarchitectures grew more sophisticated; configurability became more popular; benchmark scores broke records; and clock frequencies zoomed to levels once considered the exclusive domain of hard cores and high-performance embedded chips.

We nominated six 32-bit processor cores for the Microprocessor Report Analysts’ Choice Award in the IP Core Processor category: ARC International’s ARCtangent-A5; ARM’s ARM1026EJ-S and ARM1136JF-S; Improv Systems’ Crescendo Jazz DSP; MIPS Technologies’ M4K; and Tensilica’s Xtensa V. All are outstanding products.

The customizable ARCtangent-A5 processor core was ARC’s most important new product of 2002. It retains all the basic features of the earlier ARCtangent-A4 and adds the 16/32-bit ARCompact instruction-set architecture (ISA). ARCompact can reduce code size in a typical embedded application by 30%, according to ARC.

We nominated two of ARM’s new processor cores: the ARM1026EJ-S and ARM1136JF-S. Both cores signal a new course for ARM—the company has changed its design flow to produce soft implementations of new microarchitectures first, followed by optimized hard cores.

The ARM1026EJ-S was ARM’s first synthesizable ARM10 core. ARM tuned the instruction pipeline for flexible timings and greater compatibility with caches comprising compiled RAMs. The ARM1020E variant is also the first (and so far, only) ARM core that has certified EEMBC scores.

An even bolder new product is the ARM1136JF-S. As the first ARM11 core, soft or hard, it’s the debutante of ARM’s new design flow. It has an integrated DMA controller for tightly coupled memories, two branch-prediction modes, and surprisingly high clock-frequency targets for a soft core: 400MHz (worst-case) and 500–700MHz (typical) in a 0.13-micron process. What makes this possible is an eight-stage instruction pipeline, the longest of any ARM processor. Among the other improvements in the ARM1136JF-S is a truly integrated FPU for greater floating-point performance.

By far the most unusual contender for an award in this category is the Jazz DSP core and Crescendo “solution kit” from Improv. This is the second generation of Improv’s VLIW media-processor core and development software, which designers use to create a custom processor known as a Jazz DSP. Although we excluded DSPs from the IP-cores category this year, the Jazz DSP isn’t really a digital-signal processor in the traditional sense. It’s more accurately described as a customizable media processor with DSP and parallel-processing capabilities.

At Embedded Processor Forum 2002, MIPS introduced the MIPS32 M4K synthesizable microprocessor core, which has new features for hardware-supported multitasking, lower interrupt latencies, chip multiprocessing (CMP), bit manipulation, and greater configurability. All these improvements reflect the demand for powerful SoCs in communication, networking, and storage applications.

Tensilica was busy in 2002, announcing a new version of its configurable microprocessor core (Xtensa V), new VLIW extensions (FLIX), support for IBM’s CoreConnect bus, record-breaking EEMBC benchmark scores, and four important technology patents. With help from custom instructions, a simulated 260MHz configuration of Xtensa V racked up a higher EEMBC ConsumerMark score (optimized) than any other tested microprocessor.
Our decision: we choose the ARM1136JF-S for the Microprocessor Report Analysts’ Choice Award for Best IP Core Processor of 2002.

We believe the ARM1136JF-S is a significant new processor for both ARM and the embedded-systems industry. For ARM, it’s the first ARM11 core and the most architecturally sophisticated core the company has ever produced. Its features make it suitable for a wide range of modern embedded applications, and it can reach clock frequencies once considered outlandish for a synthesizable processor. We think the ARM1136JF-S raises the bar for soft cores and will be a hit with designers who need high performance with low power consumption.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170706.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

The 10Gb/s Plateau
Peter Glaskowsky - Editor-in-Chief  {02/18/2003}

Just as the air is thinner at high altitudes, the market for network processors is rarified at higher data rates. Several major vendors—including AMCC, IBM, and Motorola—have delayed their plans to offer 10Gb/s network processors until the market develops. Even fewer niches exist for 40Gb/s (OC-768) networks, and there are as yet no shipping 40Gb/s NPUs. We expect to see at least one 40Gb/s part announced in 2003, but very little demand for NPUs at this speed will exist until 2005 or so.

Our Analysts’ Choice Award nominees are a mix of 10Gb/s products (AMCC’s nP7510, Cisco’s Toaster3, EZchip’s NP-1, Intel’s IXP2800, and Silicon Access Networks’ iPP) and slower devices (IBM’s NP4GS3 and Motorola’s C-5e). Intel’s IXP2800, though nominated, did not become available during 2002 and was withdrawn from consideration for the award.

We select EZchip’s NP-1 as the Microprocessor Report Analysts’ Choice for Best Network Processor of 2002. The NP-1 is one of the fastest NPUs on the market and offers a high degree of integration with its on-chip search engine and multiple standards-based interfaces. A follow-on NP-1c device, with higher performance, is due out this quarter. Of all the NPUs available during 2002, we believe the NP-1 offers the best overall package of current capabilities and future potential.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170708.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Security by Design
Peter Glaskowsky - Editor-in-Chief  {02/18/2003}

When a chip makes it onto the motherboard, you know it has proved its value. Security processors have been delivering increasing value to server manufacturers over the past few years. Now, one server-motherboard vendor believes security processing has become a must-have item for some of its customers. Abit’s new SI-1Ns uniprocessor Pentium 4 motherboard includes Cavium’s Nitrox Lite CN1005 security processor, and Abit will have dual-processor boards later this year.

The Abit design win is important to Cavium, but it has broader implications for the industry. We believe that, over time, as the cost of securing high-speed Internet connections diminishes, security will become a mandatory feature for most Internet-connected devices.

Security processors such as Broadcom’s BCM5841, Cavium’s Nitrox Plus CN1340p, Corrent’s CR7120, and Hifn’s 8154 HIPP II define the current state of the art in this market. Each device offers a different combination of features and speeds to meet customer needs. We believe the best security processor available today is Cavium’s Nitrox Plus CN1340p, a chip so fast it may enable entirely new ways of thinking about Web surfing.

A single CN1340p could protect all Web transactions for a major portal or e-commerce site. All Web requests, not just the most sensitive financial transactions, could be protected from prying eyes across the Internet. This revolutionary potential, unmatched by any other security processor, earns the CN1340p our Microprocessor Report Analysts’ Choice Award for Best Security Processor of 2002.

The arrival in 2003 of network processors with integrated security acceleration, such as Intel’s IXP2850, will pose a new challenge to dedicated security processors. These new chips will do much to make security processing ubiquitous across the public Internet.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170709.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

New Tech for a New Year
Peter Glaskowsky - Editor-in-Chief  {02/18/2003}

For major chip makers, the semiconductor industry slowdown of 2001–2002 was not an excuse to curtail research and development. Managers may have put more emphasis on projects with practical potential, but pure research continued to find support at more-forward-looking companies.

Technology research led to many interesting announcements in 2002. Some were eminently practical, such as Intel’s placing an order in April for the first extreme ultraviolet (EUV) beta production tool. Some were purely theoretical, such as IBM's "molecule cascade" demonstration, in which carbon monoxide atoms on a copper surface were used to perform simple computations by a mechanical means that resembles toppling dominoes.

Our award nominees for 2002 are more toward the practical end of the scale. Recent advances in 3D chip packaging provide a good example of research with immediate benefits. Nanowires are sure to find applications once it becomes possible to fabricate them affordably and controllably. Strained silicon merited nomination this year despite a variety of implementations across the industry. Probably the most interesting and controversial of these is our fourth nominee: value speculation, an umbrella term that covers a variety of ways to leverage likely assumptions about data values to accelerate instruction dispatch and execution.

Among these nominees, one stood out as offering the clearest path to commercialization as well as having the greatest potential to add value to the microprocessor industry during the near future. We select strained silicon as the Best Technology of 2002. We have chosen not to associate this award with a specific company; too many companies have been involved in developing strained silicon for us to name some over others. We congratulate all these companies together, and we look forward to enjoying the fruits of their labor.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0218/170710.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Benchmarks Reveal Design Tradeoffs
Markus Levy - Senior Editor  {02/10/2003}

The real decision to select a processor doesn’t usually happen until the processor has been adequately benchmarked. Prior articles in MPR have covered the processors discussed throughout this article, specifically the IBM 405GPr and 440GP, and the MIPS 20Kc, but new data is available to allow us to provide additional insight into the processor’s real-world behavior. In this analysis, we’ll be using data derived from EEMBC’s benchmarks to point out the architectural differences of IBM’s 405GPr and 440GP, as well as highlight the benefits and limitations of the 440GP’s superscalar architecture. Furthermore, we’ll also use examples taken from MIPS Technologies’ recent announcement of certified EEMBC scores for its 20Kc to demonstrate the tradeoffs of a memory subsystem design.

Performing a fair comparison requires keeping things constant. IBM provided an example of this by using the same compiler to benchmark its 405GPr and 440GP. Although there were differences in the processor and its memory subsystem, using the same compiler eliminated one of the major variables. Similarly, MIPS ran its 20Kc at two different operating speeds, which eliminated many variables, including the compiler, but pointed out the difference of using two different core:memory-bus ratios. Within the scope of this article, we also analyze the symbiotic relationship between the IBM processors and compilers from Green Hills Software, Wind River, and GNU.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0210/170601.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Intel Gets Extreme in 2009
Tom R. Halfhill - Senior Editor  {02/10/2003}

Intel claims it will be the first company to mass-produce microprocessors using extreme ultraviolet (EUV) lithography, a revolutionary new photomask technology. Pilot production is scheduled to begin with the 45nm fabrication process in 2007–2008, using tools and techniques now being refined. Mass production is scheduled to debut with the 32nm fabrication process in 2009.

EUV lithography will enable a significant leap forward in the circuit density of chips, because the shorter-wavelength light allows stepper tools to draw features at least 10 times smaller than is possible with today’s deep-ultraviolet lithography. Intel is betting heavily on EUV to keep Moore’s law from expiring for at least another decade. Without EUV, the famous “law,” which observes that circuit density doubles every two years, would soon be unsustainable.

The skyrocketing costs of conventional mask sets are a growing obstacle for developers of ASICs and SoCs, especially for chips that have little chance of attaining high volumes. If the leading-edge technology of EUV lithography costs even more than optical lithography does, many chips could be priced out of their markets. Although the actual cost of EUV masks will almost certainly increase with the rising complexity of microprocessor designs, Intel believes their relative cost, compared with that of conventional masks, may decline over time. Others dispute that hypothesis, citing the higher cost of manufacturing the many-layered EUV mask blanks and other factors.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0210/170602.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

ARM Tools Link With ParthusCeva
Markus Levy - Senior Editor  {02/03/2003}

Few CPU and DSP IP vendors have joined forces to provide the hardware and software hooks necessary to adequately "glue" their cores together. Almost seven years ago, ARM attempted to combine its processor cores with its ill-fated Piccolo core. Beyond Piccolo, ARM has not invested engineering resources in developing and supporting a new DSP core. Instead, ARM has begun work with the DSP Group, the start of a series of partnerships, to integrate the dual-core support into ARM’s RealView development platform, designed to ease the pain of the hardware verification time and costs for such system-on-chip (SoC) designs. RealView provides a common development environment and combines the ARM Integrator development platform, the RealView debugger (RVD), and RealView compiler tools. The development environment can be used for multicore debugging, including multicore cross-triggering and simultaneous single-stepping, to be used in multiprocessor configurations and multi-DSP environments, although we haven't yet seen an independent evaluation to give evidence of the robustness of these tools.

Microprocessor Report readers can access the full story here: www.mdronline.com/mpr/h/2003/0203/170501.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.