RTC Interviews Bill Kehret, CEO Themis Computer
Page 1 of 1
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?
Kehret: Vertical integration can add revenue for the supplier, reduce logistics overhead for the procurer and potentially reduce integration risk. If the procurer perceives that the supplier is adding value (“one stop shopping”), the relationship is strengthened and the supplier may be granted a gross margin premium for his “commodity.” The procurer, often a first-tier prime contractor or system integrator or OEM, adds value by using his application domain knowledge and, perhaps, proprietary hardware and software to create a differentiated system solution for the end customer. This mutually beneficial relationship usually means that the supplier is expected to remain weakly differentiated from his competitors (remain a commodity supplier), ensuring that the procurer can continue to enjoy competitive pricing, from alternative suppliers. Problems arise when the supplier attempts to rebalance the value-add stack.
Rebalancing the value-add stack between the embedded computing supplier and his first level customer, may reduce the total cost to the integrator’s customer, but it’s usually not in the integrator’s interest to have his value-add ratio reduced. That’s a nice way to say that he doesn’t welcome competition for his value-add ROI. Usually, this isn’t a problem, because the embedded computer supplier has a difficult time managing, selling and supporting a broad functional range of products, diluting his application domain knowledge. In other words, it is very hard for a broad-range supplier to have the competence to compete in the deep vertical niches that the traditional primes dominate. But, this is a two-way street. The primes have increasingly outsourced much of their deep technology, so they are now obliged to rely on Integrated Product Teams (IPTs) to provide complete solutions for their end customer.
Themis’ strategy is to add incremental value, as an Integrated Product Team leader, for mission-critical embedded computing. Our domain knowledge is kinetic and thermal management and resource management for server and I/O consolidation. The benefits for our customer, and the end user, are reduced life cycle cost of ownership. The key to conflict avoidance with the “prime” is to add value by reducing the cost of maintenance and technology refresh/insertion. We also scrupulously adhere to open standards, supporting and interoperating with all existing hardware and infrastructure software platforms. Themis does this by supporting and driving standards with such organizations as OMG and VITA.
RTC: As we go to press, the European Union’s directive on the Restriction of Hazardous Substances (RoHS) will just be going into effect. It’s been speculated that it may cause problems as many leaded components go to end of life. This is causing a lot of confusion in many areas where the RoHS regulations don’t necessarily apply but the availability of many components—but not all—is disappearing. How will the transition through mixed inventories with no clear distinction in part numbers or even appearance, affect our industry? What work-arounds—if any—do you envision? Can programmable logic take up the slack? Will there be other advantages if that’s the case?
Kehret: Themis’ enjoys market and product exclusions to RoHS. Having said that, we are very focused on driving all new products to RoHS compliance. Where we have extended life cycle programs in place with key customers, we attempt to inventory complete kits, or completed assemblies, to avoid the risk of mixing component technologies. We’re very concerned. Even certification by the best assembly houses in the industry can’t guarantee that improperly marked parts don’t enter the supply stream. Themis believes that a rapid transition to RoHS compliance is the best way to reduce the risk of mixed processing, and when combined with supplier certificates and segregation of kit and WIP inventories for legacy product, can provide the best overall level of risk management.
RTC: VMEbus has been a mainstay in the military and aerospace markets for the past 20-plus years. And while there have been mid-life kickers to upgrade the technology, it would appear to be reaching the end of a long run. Do you believe VME can continue to be a viable approach throughout and beyond this decade? Will other, newer standards such as VITA 46, 51 and others rapidly displace VME? On what timetable do you expect to see this happening? Why?
Kehret: The long life cycles of military platforms pretty much guarantee a long exit ramp for VME64 backplane-compatible technology. The fact that new military programs are still designing-in parallel bus VME backplane ecosystems, tells me that the eventual transition will take more than ten years. VITA and the VME ecosystem vendors have worked hard to adapt best of breed interconnect technology for use in mission-critical environments. So far that has resulted in a divergence of serial protocols, which complicates the life of system integrators. The benefit to integrators and end customers alike is that the shelf form-factor doesn’t have to change. The war-fighter’s 6U shelf or hotel space doesn’t have to be re-architected.
An added architectural benefit is that equipment shelves can be interconnected, locally and remotely, with low latency serial links. VITA 46 (VPX) acknowledges that switched serial protocols, (in our estimation PCIe), will be the dominant LSI building block interconnect system, at the I/O level. Given increasingly high levels of chip-level integration (SOC), PCIe and some, yet other serial bus-du jour, will quickly dominate at the board edge, even for mezzanine modules (VITA 42, XMC). Power density scaling pretty much demands conduction cooling and opens the door for flowthrough liquid cooling and thus VITA 48 VPX-REDI (Ruggedized Enhanced Design Implementation). What this means for VME is that the old Euro Standard 6U form-factor has lots of life left in it. This is important because in both 3U and 6U formats it occupies an important rung on the volumetric modularity ladder. Modularity, thermal and kinetic management are where VME shines, and these new specifications extend the commodity reach for this popular embedded computing space.
RTC: The VME Standards Organization recently started up a committee for VITA 56, which is defined as a front-panel access, live insertable mezzanine card that will be available in versions to handle full military shock and vibration requirements. The dimensions and capabilities are somewhat similar to the AdvancedMC. Do you think there’s room for two similar mezzanine form-factors—one for rugged applications, the other for more benign jobs? For that matter, as for example, the military moves to a more network-centric structure, might systems based on ATCA and perhaps in the not too distant future, MicroTCA, migrate from the commercial world to some military applications—particularly shipboard and land-based? Does Themis have any plans to join the ATCA/MicroTCA marketplace? Why or why not?
Kehret: Well, we’re not particularly pleased to have board edge finger connectors, sans connector shells, migrate into the VME space. This is a giant step backward in connector integrity. Most of the defensive writing we’ve seen focuses on the surface mount compression connection reliability, but that’s not the focus of our concern. Rather, we are concerned with the mechanical stress path, which isn’t mediated by strong mating connector shells. Fortunately, the VPX wafer-style connectors do address this requirement. The problem with using less robust connectors in open standards is that there is no control over the way independent board vendors implement the standards, or control fingerplate tolerances. The relatively lower mass and shorter stress paths of mezzanine cards does move the first resonance mode frequency up, significantly, a mediating benefit for this application. However, lateral stress paths do cause blade-mating surfaces to scrub, reducing the reliability of the mated contact.
Connector reliability is a big deal for Themis and its customers. A major part of our program to mitigate connector reliability concerns is a regimen of stress-to-fail and Highly Accelerated Life Testing (HALT), during the design phase and 100% vibration screens during manufacturing, both at the board and system levels. Further, we do continuous life testing, with sampled Highly Accelerated Stress Screening (HASS), to ensure quality and reliability of our products. Board-level interconnects both add overhead cost per function and reduce reliability. Our test regimes help improve reliability, but there’s little to be done for the increased component cost of mezzanine modules.
Regarding ATCA and related ecosystems, Telecommunications and Data Communications markets have demonstrated an appetite for this large form-factor standard. As implied above, it’s another, now established, rung on the shelf-level volumetric modularity ladder. Themis has products under development that will push computing density in these form-factors. We’re less excited about MicroTCA, for the reasons elaborated above, but we leave the market to sort this out. We do have aggressive plans for the AMC form-factor, and our extensive Symmetric Multiprocessing (SMP) and chip-level multi-threading experience, at the hardware and OS levels, should make us a strong competitor in this market.
RTC: As processors and attendant systems have become increasingly powerful, there has been a need to develop clever cooling techniques. Themis has been a leader in a variety of techniques over the years including mist cooling. What are the challenges today, and how is Themis and other companies addressing the problems?
Kehret: Thanks for asking! We think there’s a lot more life left for air-cooled systems, and conduction cooling is a key to extending the life of air-cooled systems. The question is, where to put the heat sink? We like to use as much of the board surface area as possible for the heat exchanger (heat sink) on our high-power-density boards. Actually, we’ve been doing this for more than ten years. We pioneered four-way SMP servers in the VME form-factor, using discretionary pins and 12V power to get the power into the slot and wall-to-wall heat sinks that slide into the VMEboard card guides. Interestingly, VITA 46/48 takes a page out of this book and adds liquid flowthrough cooling in the bargain. We’ve been doing flowthrough-cooling for several years too, with our Slice modular switched computing, but that’s beyond the scope of this VME discussion.
These “unitary” heat sinks create a whole separate set of problems for the mechanical designers. The naïve design approach runs afoul of thermal, as well as shock and vibration stress on the chip-to-board interfaces, regardless of bonding technology. Themis’ extensive stress screens, continuous life testing and accelerated life testing are part of our “customer commitment” to quality and reliability, an essential gatekeeper and counterpoise to design innovation.
The trouble with bus and board open standards is the freedom that vendors have for interpretation. ATCA especially suffers from all the choices of how to use cooling air channels on the board. Our industry likes to go crazy with modularity. The down side of this is that the air channels get messed up by a proliferation of AMCs. Little if anything is done to control airflow across down stream modules, and the AMCs block much of the remaining surface area available for heat exchangers. At the level of the shelf, or chassis, there is very little effort made to regulate the available slot-to-slot inlet air pressure, so much of the touted power density headroom is fiction. These problems put a special burden on the system integrator and can, in part, be mitigated by the ecosystem vendor, but it means the vendor that builds the high-power-density boards also needs to design the shelf/chassis and its cooling system.
But Themis isn’t only a story about high-performance, high power density computing. We’re hard at work on several computing architectures that push high performance way down the thermal power scale. That should have benefits for Space, Weight and Power (SWAP)-challenged applications.
RTC: VME has been a staple in Themis’ embedded computer strategy for many years, serving the communications, industrial and military markets. However, the business is changing radically. What was once the number one VME supplier (Motorola) has gone end-of-life on many of its designs, and according to rumors, has backed off on most new development. What was the number two maker of VME hardware (Force Computers) abandoned the architecture and was acquired by Motorola—the combined company is now focusing significantly on ATCA and communications strategies. While the VME market remains relatively strong, 1) can it continue to be a viable force in the embedded computer board and subsystem market in light of growing competition and 2) will a critical mass of suppliers continue to offer the variety of products required? 3) Does there continue to be a market for VME outside the military?
Kehret: OK, as you suggest, there are really several ideas in play regarding the longevity of the “VME” market. We believe that VME is more than VME64 or any other particular flavor. In the broadest sense, I think the market thinks of VME as the 6U/3U Euro Standard card and shelf packaging dimensions. Backplane compatibility, of course, makes all the difference in terms of a variant’s survivability. The splintering of 6U/3U, with the emergence of cPCI, was really the industry’s reaction to a lack of responsiveness by the standards groups to extend and refresh backplane protocol and connector-based I/O constraints. In any case, the 3U/6U VME ecosystem is, for all practical purposes, here to stay. The viability of particular specification, and “dot”-level variants, will be worked out by market forces. The important point is that VITA and other standards organizations, through their endorsement, are providing enough flexibility in the ecosystem for it to mutate and survive. Whether it thrives is up to the embedded community, their ingenuity and perseverance.
The question of critical mass in the ecosystem is largely answered above. We think the VME ecosystem will actually make gains at the expense of cPCI and ATCA, in non-military markets. Will MicroTCA steal market share from 3U VME? That’s hard to say. As I’ve discussed above, the mechanical/electrical integrity of the AMC board edge connectors are seriously compromised for high shock and vibration applications. I rather suspect there will be parallel universes for ATCA, its ecosystem and the evolving VME ecosystem. Once in a while there will be some design win crossovers, with an attendant flurry of concern, as expressed in the trade media, but for the most part, the two ecosystems will remain vital standards for their respective, highly differentiated markets.
Where do VME customers go when they leave the bus-and-board market?
While you didn’t ask this question, it’s a logical extension. The answer is, more blade servers and all-in-one boxes. Commercial blade servers are outperforming ATCA at lower price points, and while they are no match for VME—in terms of robust service—they are great for server consolidation. Themis has been working on its Slice Switched Computing initiative for several years. This product family has the advantages of VME, in terms of cost of ownership, thermal and kinetic management, but also the benefits of a larger format board (think rack slice), vastly improved cooling and extensibility, while matching or winning on SWAP. The market has learned that COTS is good. VME-like attributes at commodity prices, should give Themis an “unfair” advantage in the server consolidation markets. Slice and Slice-Lite are our answers for those who would leave the bus-and-board fold.
RTC: Switched fabric technology appears to be something that’s going to happen—it’s just a matter of when. Several variants from ATCA and AMC to other approaches such as VITA 36 and VITA 41 are emerging, as is CompactPCI Express. Do you see any of these emerging as critical technologies over the next several years? Which one (s)? How will the emergence of hybrid and non-VME products impact the backplane and packaging business? Will the proliferation of many different architectures cause the lack of critical mass and deterioration of the market?
Kehret: VITA 41 is an incremental approach to serial fabrics. While its use of a high-performance wafer-style connector is a good idea, the new connector comes at the expense of board footprint and routing channels. It’s a useful evolutionary step for those integrators who need some serial connectivity but need to preserve their legacy, proprietary VME boards. For now, we think it’s a rather narrow, niche market.
VITA 36 is an important complement to the PMC spec for those who need rear connector I/O. Those who need it, usually because the front panel is too crowded, or those who want cable management, really get a benefit. Often the VME baseboard vendor also controls the other end of this link, so it’s not much of a problem for the industry at large.
RTC: Themis goes back a long way in the embedded computer industry, and over the years our editors have always enjoyed your tremendous insight and perception of the industry. Embedded computer technology has continued along a growth curve following Gordon Moore’s Law. Advances such as multicore processors and developments in programmable logic continue to add speed and density. Can you give us any sense of what applications you envision for embedded computers, say, a decade away? Twenty years?
Kehret: I’ll offer the usual caveats about technology forecasting. Technology forecasting, using auction and pseudo markets, is a good way to improve accuracy through community participation, but individual prognosticators are often way off base. Still, in the spirit of your question, I’ll take a cut at it. An architectural trend that is gathering momentum in telecom markets as well as semiconductor design can be called “low and wide.” I believe George Gilder popularized the paradigm, calling it “low and slow” or “wide and weak,” some years back. That certainly makes sense for the all-optical-networks (high channel count and low power), but the paradigm is starting to get legs in computing architectures too. Sun Microsystems has consistently beat the big semiconductor players to market with multicore designs and its recent “Chip Multi-Threading” (CMT) offerings are a good example of what can be done to reduce power per thread, while retaining significant transaction rates.
Optical interconnects, optical switching and storage can break the planar processing bottleneck by going to 3D and back again. Both technologies have a transformational potential that may change computing density far faster than Moore’s Law, and in so doing, change the application landscape for computing (as if it wasn’t changing fast enough already). In any case, these trends are very hopeful because they portent increases in computing density/power (transaction rate per unit volume per watt). I’m not going to trip out about fly-size UAVs, but I do think these trends will enable huge rates of advancement for diagnostic medicine and for personalized pharmacological interventions, for disease control and management and for virology. With the threat of biological terrorism, that’s a comforting thought. I also look forward to talking and maybe just “thinking” about how to interface with high-mobility computing devices. These advances, along with retinal scanning and heads-up displays, ought to take the pain out of our neck for everyday computer users in the next five to ten years.
As you might imagine, we plan to be part of this exciting evolution and there will be plenty of spinout technology to benefit our more traditional, embedded markets.
You will have to ask a real “seer” about the twenty-year horizon!