ESC: Connectivity success impacts design traditions

The top challenges designers need to overcome to achieve success in meeting market demands will be examined by all of our In Focus contributors, who will participate in technical sessions at this week's Embedded Systems Conference.

First comes the Internet-centric connectivity challenge, made truly ubiquitous with the rapid adoption of wireless technology. Next is the explosive growth in the type and variety of embedded consumer/home electronics and entertainment devices, driven by buyers who are increasingly sophisticated and demanding. To satisfy their demands, developers are moving to more-sophisticated processors and higher levels of silicon integration in an effort to provide those features at as low a price point as possible. Faced with such seemingly incompatible demands, developers are now finding they have to shift from traditional board-level design methodologies to new system-on-chip (SoC) designs, where the problem of hardware and software integration-always problematic-becomes even greater.

Wireless connectivity has become especially intriguing to embedded developers, said Cliff Bowman, software engineer at Ember Corp. (Boston), who will present a paper on wireless-mesh topologies at ESC. "This is because it reduces considerably the range of issues compared to the costs and development time related to the physical wired connections. But the issues relating to protocols are just as complex, if not more so, especially when real-time and determinism are factored in," he said. These are important not only in traditional embedded control applications, but also in many media-intensive small-footprint computing and entertainment appliances. Excerpts from Bowman's ESC presentation can be found in an exclusive online contribution.

Often forgotten by developers considering wireless in their designs is that unlike wired methodologies, latency and reliability are significantly impacted by the network topology. "In any wireless system, network latency grows additively with the number of hops," said Bowman. "Furthermore, link reliability falls multiplicatively with the number of hops unless the system employs some form of error correction or recovery. As a consequence, shortening message paths by distributing tasks can significantly improve overall network performance."

Jimmy Schaeffler, chairman and CEO of the Carmel Group (Carmel, Calif.), a market research firm, who is on the consumer media-products panel, believes the consumer market has always been a fast-moving one. According to Schaeffler, it has also been challenging for embedded designers simply because of the pace of innovation and competitive pressures. And those factors have considerably shortened time-to-market windows. "But in the past, that change was linear, in one direction with reasonably well-defined market segments," he said. "The difference now is the introduction of nearly ubiquitous connectivity-via satellite, via wireless and via the Internet-as the third dimension. Now you have a market in which the directions a particular consumer device will go are not nearly as predictable."

Doug Rasor, vice president of strategic marketing at Texas Instruments Inc.'s Semiconductor Group (Dallas), who will be participating in the "Embedded Visionaries" panel, noted that this phenomenon requires that companies and embedded designers change the way they design products. Rasor believes they must opt for as much flexibility as possible, to allow for the increased number of permutations a particular design might take.

As the contributions illustrate, developers are turning to the creation of companywide, applicationwide, SoC- specific and architecturally specific "platforms," in which the primary focus is not just on performance, power dissipation and consumption, and cost issues, but on flexibility and reuse as well.

"I would bet you, though, that when push came to shove," said TI's Rasor, "that most companies at this moment in market time would opt for flexibility and reuse first, and give up a little on the other elements to get it. That is how much the dynamics of the embedded consumer market have changed."

"The nature of this new embedded consumer-device market is going to require methodologies that are not just used to get you to market faster than your competition," said Jerry McGuire, general manager of the media platforms and services group at Analog Devices Inc. (Norwood, Mass.). McGuire, who is on the "Embedded Visionaries" panel, noted that "it will also require the use of tools and platforms that will allow you to turn your design and your company on a dime, when a particular market segment takes an unexpected right turn."

What this means to Richard Tobias, vice president of the ASIC and Foundry Business Unit at Toshiba America Electronic Components Inc., is that until some sort of industrywide methodology finally emerges, it is up to embedded developers, their companies and the hardware vendors to take up the slack. Until that occurs, he said, as much as they dislike getting involved in implementation rather than design issues, embedded designers are going to have to pay more attention to the range of SoC design methodologies, how to choose them, how to develop the hardware and software underpinnings of the designs, how to debug and test these new designs, how to integrate multiple processors into a single-chip design, how to adapt tools originally used in board-level designs and how to integrate new tools into the traditional embedded-system design flows.

Integrating embedded and design-automation tools in a way that is beneficial to systems designers has been particularly problematic for Tobias at Toshiba America.

"I thought I was talking with many of the EDA companies about their need to adapt their methodologies and design philosophies to the needs of the embedded developers," he said. "But with some exceptions, I think I was just talking to them; that is, I was talking and but I am not sure they were listening."

Impressive as the range co-design and co-verification tools and the attempts by large EDA vendors have been, Tobias said, EDA developers still do not get the point that embedded designers do not want to learn and adapt to new tools.

They want a design framework that allows them to keep their focus on developing the algorithm and the corresponding code to solve the problem, he said, rather than spend too much time on implementation issues. "I think everyone has forgotten why embedded designers in the '70s rushed to microprocessor design methodologies and away from building their designs from a mix of SSI, MSI and LSI logic blocks and bit-slice ALUs," he said. "It was because software-programmable microprocessors allowed them to keep their focus on solving the problem and coming up with the right algorithms and spend less time on hardware implementation issues."

In terms of advanced system-on-chip designs, Tobias noted that the closest the industry has come to a design framework that relieves embedded developers from delving too deeply into EDA implementation issues is the emergence recently of the concept of semiconductor-based platforms. "Because there are very few, if any, EDA companies providing them," he said, "we have had to invent our own high-level frameworks, taking a lot of time and money away from actual product development."

The concept of using applications platforms to simultaneously achieve the often-conflicting goals of managing complexity, fast time-to-market and product differentiation is not new, according to Peter Santos, vice president of marketing at Barcelona Design Inc. (Newark, Calif.). Santos, a contributor who will present a class on platform design as it relates to analog circuits, said, "They have been in existence for decades in fields such as automobile design and computers."

But, recently, he said, "complexity pressures have brought the platform concept to receive strong attention in [the area of] semiconductor design." That should come as no surprise because all of these areas share a common emphasis and need for real-world interfaces, rapidly increasing operating frequencies, higher functionality and, of course, lower cost," he said.

TI's Rasor noted that, ultimately, some common methodology will have to emerge in which all segments of the computer industry take part: embedded developers and tool vendors, the makers of the semiconductor devices, the EDA companies and the end users. "There are just too many tough issues to solve that require more coordinated, comprehensive industrywide effort," Rasor said.

On the one hand, he said, such semiconductor platforms need to be fixed enough so that the it is possible to bring together all of the elements-processor, chip interconnect, functional IP or memory-so as to accomplish fast, reliable time-to-market. But on the other hand, such platforms must allow the user to apply significant differentiating value, without losing the design time advantages conferred by leaving the other part of the design fixed. "Also, any manufacturing cost penalty of using a platform must be small enough as to be outweighed by reduced time-to-market and design risk," said Santos.

According to Ubicom's Chris Waters, as promising as the development of flexible semiconductor platforms are in helping designers keep up with the twists and turns of this new wireless and ubiquitously connected consumer device market, they are, because of the intellectual and financial capital required, a revolution in which only the largest companies can participate. "As a small company, we found that the best way to maintain the flexibility we need and our customers need is to use advances in semiconductor process technology to push the performance of the silicon," he said. "This allowed us to build the additional functions in software rather than hardware and our customers to adapt quickly to shifts in the marketplace."

Unless such semiconductor-based platforms become more widely available to the smallest of companies, he said, embedded developers are going to keep looking for alternatives.