Title: Trends in the High-Speed Embedded Market
Subtitle: Linley Gwennap explains the future of multi-core embedded.
By Geoffrey James
Linley Gwennap, founder of The Linley Group and one of the most respected analysts in the microprocessor industry, recently co-authored the fourth edition of “A Guide to High-Speed Embedded Processors.” We asked him to characterize the current state of the market, its trends, and the vendors that dominate it.
Embedded Intel Solutions: Who is the intended audience for your report?
Linley Gwennnap: Our primary audience consists of people who design a piece of equipment that requires a general-purpose processor or network chip. The reports cover companies developing their own CPUs to deliver extra performance, which is why we excluded processors below 400 MHz—a speed range easily achievable by a synthesizable core. The report thus provides an in-depth look at the top products and top vendors in the high-end embedded processor space, specifically AMCC, AMD, Broadcom, Cavium, Freescale, IBM, Intel, Marvell, PMC-Sierra, RMI, and Via Technologies. People using CPUs from these firms have diverse requirements for performance, power dissipation, peripheral integration, and price. This guide is intended to help them make the right selection.
EIS: How does it accomplish that?
LG: The number of markets for high-speed processors continues to grow. In networking alone, these speedy chips are needed for complex functions, such as intrusion detection and other security functions, storage management, router control plane, and networking services. Consumer devices, such as set-top boxes, HDTV receivers, and automobile navigation systems, also need high-performance CPUs, as do high-speed printers, thin clients, kiosks, industrial control, medical imaging, and a host of other devices. System designers prefer a chip that integrates easily into their designs. We help them identify that chip.
EIS: What’s your research methodology?
LG: We go directly to the vendors and conduct in-depth interviews of the product managers, architects, and executives to make sure we have a thorough understanding of feature sets, market strategy, and business strategy. Because we have a strong background in the semiconductor industry, we take these opportunities to dig deep and drill down well beneath the brochure level to uncover the real differences between vendors and products. We boil all that down in our report and offer informed opinions about what applications will work well with each product. Essentially, we’re trying to provide all the information and perspective needed for system designers to make an intelligent decision.
EIS: What long-term trend is driving the market for high-end embedded processors?
LG: I’d have to say that it’s the movement toward multi-core architectures. Just a few years ago, you’d be hard pressed to find any multi-core CPU in this market segment. Now you’d be hard pressed to find one that isn’t. This trend is important because, unlike previous CPU innovations, multi-core forces system manufacturers to change their software. The traditional ways to improve processor speed, like raising the clock speed, simply caused the software to run faster automatically. By contrast, you must make software multi-threaded in order to take full advantage of multi-core. In addition, multi-core adds complexity to the chip design, mandating changes to interconnects, memory, and I/O to ensure that all those processors are being constantly fed the right data. It’s not an easy task and some multi-core CPU designs work better and more efficiently than others.
EIS: Where do you see multi-core as a key requirement for an embedded system?
LG: Multi-core, by definition, is at the high end of the embedded market. A system that requires, for example, an 8-bit controller is obviously inappropriate. While embedded systems are beginning to permeate nearly every kind of electronic product, we see multi-core playing a major role in broadcast video processing, surveillance applications that compress multiple streams, and low-level networking where multiple packet streams must be analyzed and processed. In all of these applications, you can assign each task to a single CPU on the chip. Where multi-core will prove much less useful is any complicated application where processing must take place a step at a time or where there’s a single data stream that must be number-crunched in real time.
EIS: How are the major vendors positioned in this market?
LG: Intel dominates the market at the dollar level because their CPUs tend to dominate the high end of the embedded market--especially systems, like airport kiosks and ATMs, which have a strong resemblance to the personal computer. Freescale, on the other hand, has been able to optimize their CPUs to meet the needs of networking--by integrating an Ethernet interface, for instance. They’ve also used a RISC core that consumes less power than a comparably powerful Intel core and included circuitry that otherwise would have to be provided on a separate chip. As a result, Freescale has been particularly successful in base stations, DSL, and enterprise routers.
EIS: What do you see in the future for multi-core embedded?
LG: Any time there’s a big technology shift, it creates an opportunity for new entrants. This is no exception and there are a number of smaller companies participating in interesting market niches. For example, there’s a company called Cavium that came right out of the gate with a chip with 16 processors running 1 GHz each based upon the MIPS core. That’s a lot of processors compared with Freescale’s multi-core units (which have two processors) and Intel’s multi-core units (which have up to four). However, they’ll eventually have some competition in this space because the other vendors, most notably Intel, have announced plans to gradually increase the number of CPUs on a chip. For Intel, this would probably mean using the Atom CPU core rather than the x86 family, simply because the power requirements are fairly demanding.
EIS: Do you see IBM’s CELL chip making a play in the embedded space?
LG: You can’t rule it out because IBM has been making noise for years about expanding the market for that chip family. However, IBM appears to be focusing on markets where there’s a big dollar value, like computer gaming, where they’ve been successful with the Playstation 3. I’m not sure that they’re all that interested in adapting the CELL to put it into a networking application or signal-processing application. The entire market for high-end embedded CPUs is only around $1 billion a year and that’s apparently not big enough to attract IBM’s attention.
Worldwide revenue market share for high-end/mid-range embedded CPUs.
Source: The Linley Group
++++++++++++++
Geoffrey James is a contributing author for Embedded Intel Solutions magazine. He is both an author and journalist who writes about business, technology, public policy, strategy, and sales/marketing. Geoffrey has written over a hundred feature stories for national publications.