Multicore Trends Continue to Drive the Embedded Market in 2007

An interview with Senior Analysts at Venture Development Corporation and In-Stat

By Cheryl Ajluni

 

Last year, a great deal of focus centered on the trend toward embedded multicore platforms. More and more engineers were projected to adopt multi-core processor architectures in the coming years. Is this still the case today? And what exactly are the current “big picture” trends in embedded hardware and software systems? To find the answer to these questions and others, I recently spoke with Matt Volckmann and Tom Halfhill about emerging trends in the embeddedsystems market. Our e-mail exchange focused on various key market trends including the use of multi-core platforms and electronic-system-level (ESL) tools as well as power consumption and custom ASIC alternatives. Highlights of those discussions are as follows:

Question: Are multi-core platforms still a hot trend this year?

[Matt] When we last talked with Embedded Intel Solutions about this topic, there was limited penetration of multicore and early indications of anticipated growth in adoption on future projects. Since that time, we have conducted several additional end-user surveys on the issue and the level of interest and expected adoption of multi-core has further increased. What is most surprising is not just the change in the current level of adoption, but the portion of engineers expected to use multi-core processors on future projects.

The ESL survey we conducted through February and March of this year showed that 12.5% of respondents surveyed— working across various hardware, system, and software engineering roles—were using multi-core on their current projects. Most astounding was that over 36% of these respondents believed they would be using some form of multi-core processing technology within the next two years. Clearly, attitudes have shifted since 2006, when our data showed this metric to be less than 15%. Of course, these figures suggest a growing degree of awareness and planning around multi-core. Whether the level of actual adoption follows this expectation is yet to be seen.

Certainly, the vision of multi-core in the embedded market promoted by Intel and AMD (which is an extension of the move within the personal and enterprise computing markets) has had an impact on raising awareness. However, this perspective is only part of the larger vision of multi-core that includes the idea of enabling the design of either homogeneous and/or heterogeneous cores. In a broad analysis, it is important to recognize the impact of the adoption of various potential architectures for multi-core systems.

Should the penetration of multi-core design continue, this will also raise the importance of having more advanced tools for designing and verifying both hardware and software. The tool vendors in the embedded industry are monitoring this trend with interest. A significant number of these suppliers are augmenting their solutions in preparation for change in the area of multi-core design.

[Tim] Power consumption and heat dissipation continue to limit the increase of throughput by traditional clock-frequency scaling. Therefore, at this time, multi-core processors are the only path to significantly higher performance. Some people haven’t yet grasped this reality. Others are resisting the trend because it’s more difficult to design multi-core processors and harder to write software for them. Nevertheless, applications that can’t take advantage of multiple cores—for whatever reason—will not see much improvement in performance after they hit the wall on a single core. Multi-core is the future.

Many people don’t realize that embedded processors are leading the way in multi-core integration. Whereas PC microprocessors grab attention for integrating two or four processor cores on a chip, Microprocessor Report has written about embedded processors that integrate hundreds or even thousands of cores on a chip. And these massively parallel chips aren’t mere lab experiments like the 80- core prototype chip that Intel showed earlier this year. Embedded processors with dozens or hundreds of cores are in production and are in shipping products right now. Anyone who doubts the trend toward multi-core integration simply isn’t keeping in touch with the market.

 

Question: What are the “big picture” trends in 2007 and looking forward to 2008? What’s being done in the industry to try to address these trends?

[Matt] Across the various engineering domains (e.g., embedded hardware and software) in the embedded-systems space, there are a number of interesting changes occurring. Within the systems-engineering space, for example, one of the most significant factors is a continued focus on enabling early software development and verification/test as project teams look to build more sophisticated devices within short time frames, while at the same time controlling development costs.

Of course, an increased focus on software does not necessarily mean that hardware design becomes a less important component of the embedded-systems engineering process. On the contrary, it is reasonable to conclude that more sophisticated methods of approaching hardware design and description are critical to advancing software engineering capabilities. Certainly, the ESL is a response to this objective within embedded-systems development. It aims to unify the many diverse and complex hardware and software engineering tasks through a more complete model of the total system (at higher levels of abstraction) in order to bring added design and verification efficiency.

[Tim] Power consumption is still a major concern and is driving other things, such as multi-core integration and the quest for greater efficiency in processing. Another continuing trend is the exploration of alternatives to custom ASICs, which are becoming prohibitively expensive for all but the highest-volume applications. As FPGA prices fall, they become more popular as alternatives to ASICs. It’s significant that ARM, for the first time, is licensing a synthesizable embedded-processor core specifically for FPGAs. At Microprocessor Report, we are also seeing new alternatives, such as Atmel’s Customizable Atmel Processors (CAPs), which mix gate-level programmability with microcontroller functions on a single chip. Embedded-systems designers need more choices halfway between ASICs and FPGAs.

 

Figure 1. Shown here is a graphical representation of ESL segmentation, as defined by Venture Development Corporation.

Figure 2. This graphic, courtesy of Venture Development Corporation, indicates an expected increase in the use of virtual prototyping methodologies over the new two years.

Question: What is ES L and what does it mean to the industry?

[Matt] Over the last year, VDC conducted substantial research on ESL, looking to develop a consensus definition for this space with the help of key EDA, ESL, and embedded software firms. In defining ESL, we felt it was important to distinguish between ESL design tools and system-level design tools. In our view, ESL solutions are a central sub-segment of a broader range of diverse technologies that ultimately enable system-level design. One of the other challenges was that ESL tools, in practice, can potentially be defined on at least three levels:

    • Design/verification of IC/SoC hardware at higher levels of abstraction

    • Design/verification of both hardware/ software elements within an IC/SoC, independent of the level of abstraction

    • Design/verification of more complete systems encompassing not just ICs/ SoCs but also multiple chips, boards, and other system elements associated with a complete computing device, network, or computing environment

     

Based on our research, elements of these concepts were combined to create a working definition of an ESL tool: a tool capable of assisting in the creation, assembly, simulation, and testing/verification of hardware and system designs modeled in software at an abstraction level above the register transfer level (RTL) of description—or in a manner that more comprehensively addresses system-level design/verification and software interaction (see Figure 1).

Regardless of how the space is defined, the success of ESL tools will ultimately depend on customer adoption of more advanced methods. There continue to be signs that adoption and interest in ESL methodologies is continuing—especially within the semiconductor segment, where companies are looking to ensure the success of their latest technologies. Here, ESL is viewed as a method for enabling customers to be successful in the use of the semiconductor company’s physical silicon, IP, and other products.

Also, there is evidence of growth in the virtual-system-prototyping and simulation segment of the ESL market. This space is defined as tools that enable the creation, assembly, and simulation of hardware/system designs modeled at a high level of abstraction and offer simulation speeds fast enough to enable efficient software development and debugging. There are certainly many associated challenges. However, the growth in the market for these types of tools is being driven by several key market factors including an increasing interest in using virtual hardware prototyping in conjunction with more traditional physical methods (see Figure 2).

 

Question: Do you see any crossover between the trend toward multi-core platforms and the other trends you mentioned?

[Matt] Yes. A rapid migration to multi-core technologies clearly raises the level of difficulty in both hardware and software design and verification. The associated challenges are exactly the issues that many ESL tools are attempting to address. The pace of adoption of multi-core architectures and ESL tools are therefore very much connected

Ultimately, rapid growth in the ESL market will depend on real demand from customers for new solutions and a widespread realization that more traditional methodologies alone are not sufficient. Therefore, any additional complexity added to the system engineering process is a potential driver of ESL growth.

[Tim] Yes, it’s all interrelated. The general trend in personal computing and consumer electronics is toward greater mobility, which requires lower power consumption for longer battery life. At the same time, these mobile products (e.g., cell phones, notebook PCs, ultramobile PCs, and digital cameras) need more processing power to handle their additional functionality. Simply cranking up the clock speed won’t work beyond a certain point, so designers must resort to multi-core chips and more efficient processing. And those decisions, in turn, force designers to rethink their software. It’s not an easy road to follow. But for now, it appears to be the only road.

Matt Volckmann is a senior analyst/ program manager, Embedded Software Practice, Venture Development Corporation (VDC), www.vdc-corp.com.

Tom R. Halfhill is a senior analyst, Microprocessor Report, In-Stat (Reed Business Information U.S.), www.instat.com.


Cheryl Ajluni is a contributing editor and freelance technical writer. She is the former Editor in Chief of Wireless Systems Design (WSD) Magazine