Standard and Custom Designs Forge Fruitful Marriage

By Steve Weller Chief Technologist, SBS Technologies The design process road is fraught with obstacles including budget constraints, time to market pressures, complex technologies, integration issues and long product lifetimes. These factors can make for a bumpy and unpredictable ride for even the most skilled and experienced engineers. For designers who wish to overcome these obstacles, standards-based form factors are a practical tool. That?s the case even for those planning on designing and building everything themselves. Standards-based form factors such as PMC, P-PMC, PC?MIP, CompactPCI, and VME define electrical, mechanical, and protocol specifications enabling full interoperability between boards and chassis from different vendors. P-PMC Eases Design Process Designers can use a processor PMC (P-PMC) mezzanine, for instance, can be used to provide a powerful CPU to a single large custom main board. The particular PPMC (or even multiple PPMCs) used can be chosen to suit the processing needs of the application and can be upgraded as memory requirements change or as parts become obsolete. Use of a PPMC effectively removes a great deal of effort for the design team. It also permits use of a simpler and more cost-effective main board due to the lower speed requirements and smaller number of signal layers needed. Moreover, it's still a fully custom product. Use of a standard form factor board doesn?t commoditize the end user product, it just makes it simpler and easier to design. Standards-based form factors are also very effective tools if used as purely internal design specifications. The advantages of splitting the CPU horsepower from the rest of the system in the example above are still present even if the P-PMC itself is designed in-house. Here, however, only two specialized teams can work on the design in parallel, each able to fully test and validate their part using standard off the shelf tools and equipment. A single monolithic design wouldn?t enjoy that kind of access to standard test equipment. If a problem is found when using a monolithic design the whole system would need to be redesigned, rather than just a small part of it being re-spun. Using standards-based form factors aids the debugging process, letting you take advantage of many reference platforms. If the CPU intended for the final design is unavailable you can swap out the P-PMC mezzanine, replace it with one purchased from a third party and use it to validate the motherboard. The board doesn?t even have to have the same CPU. For instance, if it is easier from a motherboard validation standpoint to use an x86 CPU and run Windows, then you can do so. Later you can mate the system with a PowerPC-based P-PMC when the production code is fully written and the motherboard is debugged. Without the standards-based form factors, none of this is possible. If you instead used a PCI board with a different form factor, the system may still work electrically. But interoperability and flexibility are lost unless the form factor and all the other mechanical specifications are followed. The previous example illustrates a benefit to using standards-based form factors as part of a custom design. Such an approach offers a good middle-ground between all-custom designs and a completely standard form factor solution. Sometimes an all-custom approach takes too long to design and carries too much risk. At the other extreme, a completely standard form factor design could be mechanically unsuitable or lack the right boards to achieve the design specification. The small premium paid up front to understand and implement the standards specifications pays off handsomely over the lifetime of the product. For other designs simple market realities dictate a degree of flexibility in the I/O portion of the system. If the product needs to connect to a customer's WAN for instance, what interface should it have? Ethernet? ATM? T1/E1? V.35? They each cover a part of the market, but to build them all into every box would be prohibitively expensive and would eat up most of the space allocated for the WAN interface circuitry. So either several different system designs must be created, multiplying much of the effort, or there must be some sort of pluggable interface just for the I/O. While a custom interface design will certainly work, it will not solve all of the issues. As with the CPU example, off-the-shelf parts can be purchased and used immediately if standards-based form factors are used. Graphics chips are particularly onerous devices to use in an embedded design. To achieve their performance, these sophisticated, expensive machines require an intimate association with the system bus. Unfortunately such chips are often end-of-lifed by their manufacturers very quickly. Unless the graphics capability of the equipment can be upgraded separately from the rest of the system, redesigns and board spins will consume much of the planned profit. To avoid that situation the best approach is to use a standards-based form factor to achieve the mechanical and electrical interoperability. That, combined with graphics standards support such as OpenGL in the system software, ensures useful available parts and minimal disruption to long-term customers. Take What You Need Another approach is to use ?only? the form factors and employ a custom approach to the electrical scheme. If the mechanical characteristics of VME are right for the application, and the connectors, cooling, power supplies and the like are up to scratch, you can buy an off-the-shelf VME rack and design your own cards and backplane. It's probably best to add some sort of keying mechanism, just to make sure a real VME card is never plugged in, of course. An alternate way to go is to use the electrical specifications and the standard connector locations, but ignore the form factor. Height envelopes can be ignored if there is nothing in your system for the parts to interfere with. Board outlines can be extended to allow for more memory, bigger or more connectors. If four ports on a PMC are insufficient, just keep adding board space until you have enough. Board outlines can also be reduced to allow for large components on other boards. While these designs will not fit into most equipment, that does not matter, so long as it fits into your equipment. None of this takes away from the ability to take advantage of the experience that engineers already have with the standards-based form factors used and the silicon available to drive the protocols. Predicting Power Standards-based form factors have other, less obvious advantages as prototyping tools. Let?s assume that you need to know the total power dissipation of your yet-to-be-designed system and to identify hot components that may need special attention. Using pure estimation and data sheets it is easy to be off by a factor of two either way. This will affect a whole slew of other parameters such as the airflow design and the power supply specifications, and hence total system cost. If the product can be made out of or even approximated with standard form factor parts such as CompactPCI or VME boards, then the situation is improved. Board vendors have power information readily available, so once it is clear which boards will be used and what they will be housed in, it is easy to ascertain the total power. A system can quickly be built up and real-life currents determined. Adjustments can be made to measurements to account for the differences between what will be in the eventual system and what was put together for testing. In the test system fans can be turned off and on and temperatures can be measured. Sometimes what appears fully custom on the outside is actually very standard on the inside, and for good reasons. 3U and 6U CompactPCI, for instance, find their way into military equipment. Such applications make use of CompactPCI?s standard features, wealth of rear I/O connections, and its ability to fit into existing equipment form factors. This notion of combining standard and custom designs can also be found in the consumer desktop computer realms. An example is the new flat-screen iMac. Its 10-inch diameter hemispherical base includes a full-size DVD/CD reader/burner. A first glance that seems an odd choice, if space and power dissipation are major design concerns. They could have embedded a much slimmer, more power-efficient laptop drive, but this would have cost more and tied a desktop computer to the whims of the laptop market. In their approach the computer can be upgraded and stay competitive with technology changes available to the rest of the desktop PC world, a healthy trade-off that directly affects sales. The custom parts of the iMac internal design: heat sinks and the circular PCB in the base, have no need to follow standard form factors since the product has no internal expansion capabilities. Such trends will extend into the embedded world, as backplane buses make way for serial fabrics such as InfiniBand. There will still be backplanes of course ? to distribute power and reduce the complexity of fabric interconnections ? but system designers will be much less constrained to choose a single standard form factor for an entire design. Mixing low-cost standards-based form factor boards and devices with expensive, custom-designed pieces will become a more commonplace, and possibly the norm for embedded computer designs.