RTC Magazine

RTC: By and large, the embedded computer business has been going gangbusters over the past two or three years with compound growth in the communications, commercial, medical, security and industrial sectors growing at better than double-digit rates. Only certain areas of the military have seen programs delayed and business flat if not down. Here at RTC, we happen to be bullish about the industry and believe that the industry will continue to enjoy this kind of growth. Do you believe the industry is going to keep going in the right direction? What market segments do you envision growing at the fastest rate? Why?

Milrod: At BittWare, we’ve also experienced tremendous growth the last few years. But unfortunately, I predict that the overall rate of growth for the COTS board market will slow considerably in 2007, and maybe even through 2008.

My belief is based on the turmoil in the industry regarding next-gen systems standards. Without a clear successor to VME or CompactPCI, I think many COTS customers have adopted a wait and see approach—I know I would in their shoes.

Until one system architecture or spec emerges with the critical mass to attract multiple vendors, healthy competition and longevity, why bet the farm on any of these “better mouse traps?” Of course, most customers will continue to play with new systems concepts and technologies in their R&D labs; but they will be, and should be, very tentative with respect to actually developing new systems and/or products.

Countering this negative development is the fact that “the show must go on.” For many customers, this means continuing to buy and build the older VME, CompactPCI, or PCI based systems. This scenario should provide board vendors with some safe, decent margin continuing business from established design-wins. Other customers, such as the Military and commercial communications, need to get new technologies deployed whether or not the standards are safe. These early adopters might provide significant rewards to board vendors who have taken great risks and find themselves in the right place at the right time with the right stuff, but the success may still be short-lived if a different standard emerges victorious in the long run.

Ultimately, either of these scenarios means that we should see some really serious catch-up growth once one or two standards cross the chasm, enabling customers to safely release their pent-up demand for next-gen systems. Until then, I think both vendors and customers will have to proceed cautiously.

RTC: I noticed that your product listing includes AMC cards. Does BittWare see a current market in this area or do you believe that market is still out in the future? How far? Have you been seeing AMC applications outside of the communications sector? Where?

Milrod: Well, I never said I was very good at taking my own advice! Seriously though, this is a great example of BittWare proceeding with caution—while still attempting to lead. Thanks to some large lead customers who really wanted our high-end signal processing solutions on AMC, we’ve been able to blaze new trails without taking undue financial risks. As a result, I think we’ve been able to have our cake and eat it too. Without compromising our core competencies, customers, or existing design-wins, we’ve still been able to invest in some new, high-risk technology, and have given ourselves a chance to be in the right place at the right time with the right stuff.

As an AdvancedTCA module, AdvancedMC is not very compelling; but as a new, independent form-factor for use in MicroTCA systems, I’m thrilled with this format and excited about its potential. I strongly believe that switched fabric-based, small form-factor boards are the future of the COTS board market, and right now AMC in MicroTCA is the only game in town. We’ve done well with this format, and are forecasting significant growth for 2007. While it’s true that the bulk of our current AMC customers are commercial communications, we are seeing a lot of interest from the military and industrial markets as well.

With a well-established market in the behemoth commercial communications industry, the emergence of the MicroTCA, and the promise of at least a semi-rugged version, AMC has great promise. Clearly, this is the most significant new format in our industry since the introduction of PMC. By the second half of 2007, we’ll know whether or not the MicroTCA platform has really gained acceptance; if the ruggedization spec happens, then we’ll know for sure if it will take off or not.

RTC: Many have been critical of the AdvancedMC and MicroTCA specifications as being inadequate for any but the most benign communications applications. In addition, it has been criticized for other limitations such as the maximum number of layers, frailty of the connector and lack of power management. Do you believe such perceived limitations will limit the useful applications for the spec?

Milrod: No spec is perfect, and no matter what trade-offs you make, there will be people who disagree. We have certainly had our share of problems with AMC. We’ve blown up boards due to connector malfunctions, had to scrap finished designs because we couldn’t build the PCBs thin enough, had to re-spin working boards to accommodate spec changes, and continue to have PCB yield issues because our complex boards require so many layers.

There is no doubt that designing and building AMC boards is challenging, but it is a real spec that supports small format switched fabrics, and it really works. In other words: it might not be the best spec, but it’s the only one we’ve got! We’ve been shipping AMC products for over a year now, and our customers have successfully built and deployed systems. To me, that’s the definition of a good spec.

Also, one of my “truisms” is that you should never underestimate the ability of good engineers to overcome problems. In the early days of VME and CompactPCI, there were legions of problems; even now there are still issues with those mature systems, but everyone knows how to work around them. When there is the possibility of compelling solutions, clever engineers can, and will, overcome a multitude of limitations, as they have with VME and CompactPCI.

My feeling is that AMC has gained a fair bit of momentum, and a lot of clever engineers are starting to think AMC is looking pretty compelling. At BittWare, we’ve already overcome several limitations, and are working at overcoming others. Also, don’t forget that there will be additional spec tweaks, and the environmental extensions are coming soon.

RTC: There seems to be a trend in the embedded computer industry to attempt to supply more and more complete systems to OEMs thus climbing up the food chain in hopes of increasing revenue with the additional parts of the systems supplied. Many companies have done this through acquisitions that allow the company to provide additional parts of systems. Do you believe the industry will continue to consolidate in this way? At what point do you think embedded computer suppliers will begin competing with their customers? Will a time ever come when there are only a small handful of system makers and the traditional merchant board market will go away?

Milrod: I think the main reason that many COTS board vendors have had to climb up the food chain is because the fragmentation of specs has made it extremely difficult to integrate boards together into a system, and because no one spec fragment has broad enough vendor support. BittWare got into systems integration to lower these barriers to entry for our customers.

However, we’ve been careful to provide this value to our customers without competing with them. Typically, this means providing a system and framework upon which our customers can build their applications, while stopping short of developing the actual application itself.

Another important point is that we try to never force the customer to do it our way or to limit them to using only our proprietary pieces. If they want our stuff, great, we’ve got a bunch of stuff; but if they want to put some of their own secret sauce in, or add some other third-party component to differentiate their system, that’s fine with us.

Once a new standard develops critical mass with broad support and customers gain experience with it, the barriers to integration will recede. Then I think we’ll see this trend reverse, and board vendors will, by and large, return to selling boards.

RTC: RoHS has been the bane of just about everyone in electronic manufacturing in recent months. And, it is expected to have wide-reaching impact in the embedded computer industry where there is not enough volume to drive semiconductor makers to update ICs to RoHS compliance. It’s possible that many ICs such as PCI chips may well quickly go end-of-life. Do you think this will happen and cause a major disruption in technologies such as CompactPCI? PMC? What work-arounds do you envision? Can programmable logic take up the slack? Will there be other advantages if that’s the case?

Milrod: Honestly, I don’t know what’s going to happen. Our customers are not asking us to move to RoHS, in fact quite the opposite, but clearly the high-volume consumer products are quickly moving over. I’m not a semiconductor business analyst, but it surely seems like the business case for continuing to make older, non-RoHS parts might be hard to make; justifying the investment to move these older parts to RoHS is even harder. Hopefully, these older parts will still be profitable enough for them to keep making a few lots every once in a while. I can’t think of anything in the past to compare this to, so there’s just no way to tell how this will play out.

The work-arounds that we’ve put in place are twofold. First, we now support mixed process assemblies so that we can use RoHS parts on a non-RoHS board if that’s the only way we can get a given part. Secondly, we are increasing inventories of at-risk parts, and monitoring their status very closely.

For us, I don’t think the programmable logic option makes sense. Since these parts would not be pin-compatible, they would necessitate a PCB re-spin which, in addition to being costly, might introduce latent bugs that could bite our customers in the butt. If we ever need to obsolete a board that our customers still want, I think it would be better for everyone involved if we invested in helping them upgrade to newer products rather than keeping the older stuff alive.

RTC: IBM’s announcement of the Cell processor was initially hailed as a breakthrough in processing power. However, few in the embedded computer community have taken advantage of it outside of Mercury Computers. Do you believe the Cell processor has potential in embedded-computer systems going forward? What about some of the other dual or multicore devices based on established instruction sets?

Milrod: Apparently, the Cell is a pretty awesome processor, but all that processing doesn’t do anyone any good unless it’s useable. Right now, using this processor is very, very hard. My guess is that the total available market for the embedded space is less than one week’s worth of PlayStation 3 production, so IBM doesn’t have a lot of incentive to invest in developing general-purpose tools, training, or support. This puts a tremendous burden on an embedded board vendor and their customers. The vendor must either create very sophisticated tools, or basically program the processor for the customer, or the customer must climb a very steep learning curve—all options that seem untenable.

Another major concern I have about this processor is the thermal problem. As I’ve said previously, I think the future lies in small formats with switched fabrics. It’s hard to see how a Cell processor could support that type of embedding. Not only are other multicore processors with established instruction sets easier to use, many of them are sensitive to the requirements of portable devices, such as laptops, so they often have lower power versions available. I think they are still hard to use in embedded applications, but the amount of existing code and expertise is compelling.

If you’ll indulge a brief stroll down memory lane, this question reminds me of similar questions I was asked a decade ago about the new C80 chip from TI. Remember that one? Most people don’t. It too was a video-oriented, multicore processor that had tremendous processing power and was touted as a major breakthrough, but it was a little hot and difficult to program. TI invested heavily in marketing, tools and training, and several companies built boards based on it. Ultimately, people found the potential simply wasn’t worth the hassles, and its use in the embedded computer space died quickly.

RTC: What is going on in DSP? Are a lot of high-end DSP applications migrating to more general-purpose processors such as the Pentium and Altivec? Both of these are beginning to include instructions for DSP operations. Do you envision traditional DSP circuits such as those from ADI and TI eventually migrating to general-purpose processors? Why, or why not? What is the role of FPGAs in DSP? They seem to be taking hold in numerous applications that have been the domain of traditional DSP chips. Where do you think FPGAs fit in the continuum from general-purpose processors to DSP-exclusive devices?

Milrod: Wow, that’s quite a question! I’ll try to hit the highlights. First off, it’s important to note that the DSP market as a whole is huge and very diverse. Secondly, I think the use and role of DSPs in the embedded computer and COTS board markets is changing rapidly.

DSPs are used to solve a broad range of applications in our community, most of which are complex and involve multiple trade-offs for processing capability, power consumption, scalability, interprocessor communication, real-time issues such as low latencies and determinism, I/O, and of course ease-of-use and cost. I think it’s fair to say that no one DSP can adequately address all of these issues for all applications, and the semiconductor companies seem to have stopped trying.

Historically, most COTS DSP boards have featured floating-point DSPs. These are generally faster and more capable than their fixed point counterparts, and are significantly easier to program as well. Unfortunately, they also cost as much as an order of magnitude more, so volume applications quickly move their solutions to fixed-point implementations. From the semiconductor manufacturer’s point of view, this has left a limited market for floating-point DSPs, mostly prototype, development and lunatic fringe applications—not a compelling business case. While there are plenty of compelling business cases for semiconductor vendors to invest in fixed point DSPs, the need to optimize them for power and cost tends to cause these DSPs to be narrowly targeted at specific applications, and not well suited to general-purpose, high-end COTS solutions.

Despite this, we haven’t seen application migration to Pentium or Opteron. These processors provide plenty of raw number crunching power, but they just don’t have the processing agility or sustained real-time capabilities that DSPs provide. We have seen some applications migrate to the AltiVec floating-point accelerator in the past, but recently, however, we’ve seen some backward migration as the PowerPCs with AltiVec are getting hotter and hotter. Given the fundamentally different nature of the processing and I/O requirements, I don’t see how GPPs can ever replace or incorporate high-end DSPs.

Enter the FPGA. With the addition of specialized DSP blocks, these devices can do some serious DSP. It’s not easy, but it’s possible. While floating-point and complex algorithms are still a challenge, they are quite good at the hard real-time data movement, I/O and low-latency, deterministic number crunching problems, and they continue to improve at the power and cost issues. That being said, I don’t see FPGAs obviating the need for specialized DSP chips.

I think that both of these approaches to high-end signal processing, that is FPGAs and DSPs, are more complementary than competitive, and that even GPPs can add value for certain applications. I now believe that intelligently combining these technologies is the optimal way to approach signal processing. Therefore, we’re investing a lot of time, money and energy creating hybrid signal processing architectures and boards that provide FPGAs in combination with DSPs and GPPs.

RTC: BittWare has been a leader in providing DSP technology on a variety of platforms with a variety of different processor approaches. What can our readers look forward to from BittWare in coming months?

Milrod: We will continue to provide essential building blocks that enable our customers to create innovative signal processing systems. Based on what I’ve already said here, these will be primarily focused on switched fabrics, small form-factors and hybrid signal processing, along with the software and frameworks to make them more useable.

I’m particularly excited about the integration of Altera Stratix II FPGAs on our latest VME/VXS, 3U cPCI and AMC offerings. This has allowed us to implement switched fabrics such as Serial RapidIO and PCI Express, in addition to leveraging these powerful FPGAs in our new hybrid signal processing architectures. We started out adding FPGAs to our DSP boards, and now it seems that we have moved toward FPGA boards that have DSPs on them. In fact, we will soon be releasing some FPGA-only boards. We are also already working with Altera on Stratix III designs.

Recently, we announced our Trident real-time multiprocessing operating environment for the TigerSHARC DSP, and we hope to extend this to the FPGAs in order to create a hybrid operating environment that should greatly reduce development time. Our ATLANTiS FPGA framework is being completely overhauled to take advantage of Altera’s SOPC Builder tools and IP cores, and we plan to significantly expand our own IP offerings and integrated FPGA solutions. Finally, you’ll see more ruggedized and application-targeted boards and systems from us. Our customers are pulling us hard into these arenas, and who are we to deny them?

Thanks for giving me the opportunity to share my thoughts in this great publication; it has really been an honor.

BittWare
Concord, NH.
(603) 226-0404.
[www.bittware.com].