To read original PDF of the print article, click here.
Embedded ICs:Expanding the Possibilities
As designs start to proliferate in the more complex 16-, 32-, and 64-bit realms, the days of selecting a mild-mannered, off-the-shelf processor may be numbered. That's because semiconductor vendors are unleashing a new crop of heavy-duty cores, beefed up instruction sets that add DSP support, and system-on-chip options. This article will explore what's driving these trends and look at a few of the latest offerings.
The explosion-and several new technical directions-in embedded integrated circuits were in ample evidence at the latest Embedded Systems Conference (Sept. 24 – 28, San Jose, CA).
Perhaps it's heresy to say it, but silicon is no longer the hot hardware area. No one is suggesting that standard controllers and processors will die out anytime soon. Yet much of today's embedded buzz is coming from system-on-chip (SoC) architectures, which enable engineers to pretty much take any core and turn it into a custom processor.
“Far and away the most important trend that's occurring is we're seeing a lot of SoC development,” said Dean McCarron, principal analyst at Mercury Research. “That's not to say that piece-parts aren't important. What's happening now is product development and perhaps initial, low-volume manufacturing is occurring with standalone devices. But as soon as the determination has been made that the product is a winner, it's almost immediately going over to SoC.”
Also taking shape is a kind of shotgun marriage between the heretofore largely separate worlds of the microprocessor and the digitial signal processor (DSP). In practice, this is resulting in embedded chips and cores getting fitted with DSP-like instruction-set additions, which deliver matrix and Fourier transform capabilities, for example, in hardware.
The SoC edge
Domestically, however, things are moving at a slower pace.The other two apps at the top of the list are the Internet appliance and the cable-television set-top box.Interestingly, as a side note, it was impossible at this year's ESC to pin down any single factor as shaking up the embedded arena. That stands in stark contrast to two years ago, when the show was dominated by real-time Java, and last year, when the confab was atwitter with news of Linux.
However, what matters most of all are the chips and cores that will fill up engineers' toolboxes in the coming months. Let's take a look at a few of the offerings.Hyperstone Electronics has entered the embedded world from a hybrid slant. The German-based company makes a combined RISC/DSP processor (see Figure 1). The part has been licensed by Hyundai for use in DVD and arcade-game applications.
Though Hyperstone says they expect to ship more than one million off-the-shelf parts this year, the company is actually taking a three-prong marketing approach. The second leg offers the processor in the form of intellectual property (IP) that can be implemented on an ASIC. Thirdly, said Matthias Steck, vice president of Hyperstone, “We offer our IP also as a standard component, so the customer doesn't need to take all the risk of designing an ASIC right from the start.”
Despite dreams of SoC glory, Steck remains well aware that cost continues to be the leading criterion in device selection. “At the end of the day, it's always a matter of the system costs and the development costs,” said Steck. “If an 8-bit or 16-bit controller is sufficient for your application, then you should stick with it. On the other hand, if your system requires an SoC, then the cost of the intellectual property for the controller is usually the dominant cost of the system.”
However, in today's changing embedded world, where applications range from cell-phone handsets to handheld computers, there are no hard and fast rules. Specificially, there is little to guide the designer as to what a safe volume crossover point is, where they might want to go to SoC design versus an off-the-shelf part. “That's really hard to say,” noted Steck. “It depends on the volume.” For example, he pointed to one of his company's customers with 200,000 units a year who went the SoC route, while noting another customer, using one million units a year, who is still sticking with the piece-part.
Over the moon
Vulcan's unique twist, which it hopes will jumpstart a still-sleepy Java hardware market, is that Moon won't be sold as an off-the-shelf part. “The whole idea is retargetable IP from scratch,” said Mark Goodson, joint managing director of Vulcan, which began operation in 1996 as a design consulting house.
The 32-bit Moon core directly executes Java, in native mode, in hardware. Vulcan officials claim Moon uses silicon real estate that's one-third the size of the layout of the picoJavaII core. picoJava, though never a big market success, was Sun Microsystem's original stab at a Java hardware architecture.Now, after several years in the doldrums, numerous new hardware players appear ready to jump into the Java waters. “The infrastructure for Java chips is now here and design engineers are being pushed to think more creatively in terms of designs and applications,” said Paul Zorfass, principal embedded analyst with First Data.
Indeed, the impetus for the burgeoning activity is the broadening need for low power and mid-level performance devices, at a decent price, which can power Internet appliances such as smart phones and handheld Web browsers.
Vulcan's Goodson sees a ripe market for IP that can translate into quick-turnaround Java silicon. This is a result of several concurrent developments, he believes. “We are seeing the confluence of two trends,” Goodson said. “First is the rapid uptake of Java by developers, and then there's the huge growth in embedded apps.”
He laid the failure of previous industry efforts to catch fire not so much at the chips themselves but rather at the maturity of the embedded community as a whole. “To a certain extent, until now Java processors have been ahead of their time,” Goodson contended.Vulcan built Moon as a member of Altera's Consultants Alliance Program, implementing first silicon on Altera's Apex 20K FPGA technology. Moving forward, Vulcan says they want to assist Altera users who have used low-density devices in the past but are now ready to move to higher-density FPGAs designed using standard HDLs and synthesis.
That architectural angle is predicated on a design that implements most of the Java byte-code instructions directly in hardware. That is, the more complex tasks are run either in microcode, or directly in a predefined area of ROM. Moon is equipped with a single port with an extensive pre-fetch queue to enable the execution of multiple instructions per cycle.Moon also has an instruction-fetcher that looks ahead along the bytecode stream to determine if stack operations can be optimized.
Moon operates at 32 bits internally, but allows a user-programmable input/output bus width.
Moon is retargetable for programmable logic, Goodson added.
As of now, Vulcan has Moon up and running in the lab at 100MHz in a 0.35-micron processor. First silicon of 0.25-micron devices are on the way and should be ready shortly. Goodson would not provide hard performance figures (he claimed the company hasn't finished benchmarks yet) but characterized Moon's performance as “an order of magnitude increase” over current Java chips. More specifically, he said he expects that Moon will be eight to 20 times faster than its competition.
Stalwarts fight back
Over at NEC, the company has repositioned itself from a supplier of piece-parts to a vendor whose offerings encompass cores, SoCs, and the addition of new, DSP-like instructions into their standard embedded parts.
According to Karl Auker, assistant general manager for the VR microprocessor at NEC Electronics, the company is aiming at four major applications markets: networking, office automation and imaging, consumer applications, and the Internet appliance.
Both based on the MIPS architecture, NEC's VR4100 part is aimed at low-power mobile applications; the VR5432 features added instructions for image manipulation.
“Over the past couple of years, the changes in the marketplace have caused us to become more focused,” said Auker. “Namely, our main focus is towards the digital community and the Internet appliance.”He added that this requires better allocation of company resources.
“We are working a lot more with people outside NEC partnerships,” Auker said. “These include people who are making specialized controllers or ASICs for specific market segments like set-top boxes or video decoding.” The primary objective is to get NEC's microprocessor designed onto a reference platform, where a potential customer can see it put through its paces.
Intel's quiet play
Now, however, Intel has signalled its intent to take StrongARM squarely into the embedded markets for wireless telecommunications and for computing. They've even give their home-grown successor to the StrongARM architecture a new name: Xscale.
Aiming at the higher-end telecom infrastructure apps requires more performance, but Intel says they've demonstrated Xscale running at 1.0GHz while dissipating a scant 1.0 watts.Perhaps more important, Intel sees this infrastructure market-which includes basestations and the like-as one that many other companies have not focused on. “It's an open playing field right now,” said Tom Yemington, a manager at Intel. “Pick your analyst to pick your market size!”
The venerable PowerPC processor is also taking on new challenges in network applications (see Figure 3).
Also on the rise is the new category of the reconfigurable embedded processor. While it might seem like such chips aim to be all things to all developers, they do offer a heretofore unavailable level of flexibility for embedded designs. Many of them are aimed at telecom applications-such as routers-where software revisions are considered par for the course.
Two of the most prominent items on the reconfigurable radar screen have been the ARC core, a 32-bit design, and the highly touted upcoming offering from start-up Tensilica. The Tensilica processor archictecture is called Xtensa, and it will be implemented in Altera's programmable-logic devices.
Programming the bear
One big fear that took shape last year and lingers on is concern about tool availability. A year ago, the embedded community began to worry that fewer development tools would be available for each chip on the market. That concern was heightened by Motorola's acquisition of software-tools vendor Metrowerks.
Those concerns have not been borne out. Rather, there may be cause for optimism instead. A case in point is the StarCore DSP architecture, which comes out of the eponymous company formed as an alliance between Lucent Technologies and Motorola. StarCore has actively financed the development of compilers and toolsets from Green Hills Software. The latter firm says the product is about to ship.
Meanwhile, Tasking says they are going to offer their own, robust set of tools for StarCore, even though Motorola hasn't officially taken them under its wing. Tasking points to its 15 years of experience with DSPs as the reason it believes it will come up with a winning toolsuite. So here is an example of an embedded part (albeit a DSP) that is garnering more tools support, not less.
The same can be said on the real-time-operating-system front, where a host of next-generation RTOSes are being aggressively served up by the likes of Wind River, Enea OS, LynuxWorks, and others.
Nevertheless, it behooves the embedded developer to remember that, no matter how cutting edge the application, the oldest rules in the book still apply to the selection of the embedded device and tools chain. Almost to a person, the engineers interviewed for this article agreed on three factors: cost, cost, and cost. esp
Alexander Wolfe is editor-at-large for ESP. He holds a BE in electrical engineering from Cooper Union. He has written assembly language code for embedded systems. He is co-author of From Chips to Systems: An Introduction to Microcomputers, 2nd Edition (Sybex, 1987). He can be reached at