It's kinda like a deli, with a lot of choices
We virtually assembled executives from many of the leading firms in embedded small form factor computers and I/O products and asked them to comment on this question: Which approach is better when building small form factor embedded systems: stacking boards, placing modules on carriers, or cabling modules together? Their response to this query, looking at it from many angles, is both insightful and a bit surprising. Read their remarks, and you’ll see why we likened this to a deli.
Our executive panel for this forum comprises some of the best and brightest minds in the industry. Not surprisingly, their comments reflect both a diversity of thinking and a range of experience in hundreds of real-world applications for small form factor computing and I/O products.
What was surprising, though, was how these executives’ answers aligned. We gave them some guidance on how to respond but didn’t ask specific questions beyond what they thought users should know to make a choice. Reading and compiling their responses in a roundtable format they haven’t seen until now resulted in a very interesting discussion.
The first point that became obvious to us was that the executives thought the best approach – stacking, module on carrier, or cabling – mostly depends on how and where it will be used. Rather than immediately calling out a single approach, several of our panel members weighed in with this viewpoint.
Lehrbaum: “As asked, this question fails to acknowledge the fact that embedded systems have little in common other than their need for internal, preprogrammed microprocessors. Beyond that, the requirements of every embedded system – both functional and physical – are dictated by the specific application, its environment, and how many units must be built.”
Blazer: “Any discussion that starts with ‘Which approach is best?’ is by its very nature subjective. For example, it is hard to beat a desktop PC for pure processing power per dollar, but if the computer has to be launched into space, pushed out of an airplane, bounced around in a city bus, or buried in the ice at the South Pole, its limitations become very apparent. This does not make it a bad form factor; it is just not suited to these applications. The key is that it is impossible to compare form factors without considering the application.”
Winfield: “As with any engineering design challenge, the best approach is a function of many variables. These include functionality, size, processing power, human/machine interfaces, cost, packaging, networking, project deadlines, availability of qualified engineering personnel, documentation, and long-term product availability plus the environmental issues of temperature, shock and vibration, RoHS, EMI/RFI, and others. Therefore, the simple answer to the question is: ‘That depends on the application.’”
Persidok: “Choosing a small form factor for building an embedded system is a complex task with many options available to the customer. Any given application may have specific requirements that can be met by multiple form factors. An end user’s precise needs and desires for his project or industry will then refine his choice for a particular approach. There is no superior approach or embedded form factor that can intrinsically meet the needs for every application or customer in the embedded world.”
Narrow it down
Before deciding on functionality, our panel suggests that available space is one of the first things that comes into play. Size constraints prompted our panelists to mention specific approaches.
Wooley: “StackableUSB allows engineers to maximize on the space they have to implement a system. By introducing a one-fourth-size form factor measuring a mere 1.85" x 1.78", an entire system consisting of a 32-bit CPU running at 80 MHz, a GPS receiver, and a ZigBee wireless communication module, for example, can now be implemented in a space smaller than your coffee mug.”
Finstel: “The module plus carrier board method is the top choice … when the solution has to be thin with just two parts stacked (there is 13 mm from the top of the COM Express carrier board to the top surface of a COM Express heat spreader plate). This is an approach we often see in mobile devices where space is critical and the surface of the carrier board has to be adapted to the housing, such as in portable ultrasound devices. This approach is also used when system depth is critical, such as in POS/POI (Point-Of-Service/Point-Of-Information) or digital signage applications, where the computer is mounted behind a flat screen and special I/O and connector placement is needed. There are also considerations of how many units need to be built and how much customization is desired for the particular application.”
Lehrbaum: “When annual build rates exceed about 10,000 units, a product’s embedded electronics tend to be designed with an emphasis on minimizing cost and maximizing manufacturing efficiency. For this class of embedded system, full custom rules. At lower volumes, particularly when 32-bit performance (or more) and standard operating systems are needed, small form factor embedded computers tend to be constructed as a stack of two or more boards. When size, power, or cost constraints are severe, the stacks typically coalesce into two-board sandwiches.”
Persidok: “The [Computer-On-Module or COM] I/O baseboard can be the shape and size desired by the customer, and all connectors can be defined for exactly what the application demands…. It is surprising that some people see these modules as competition to the stacking board standards because they appear to serve quite different market niches. Whereas PC/104 and most other form factors have standard COTS-type I/O boards for the particular form factor, the I/O on COM modules is predominately customer specific. The average COM customer requires more than 500 units a year, while the average PC/104 user requires far below this. It takes an initial investment to make a custom baseboard or carrier I/O board; therefore, it is not practical for most small volume applications unless an off-the-shelf baseboard is considered.”
Winfield: “Stacking boards, as its name implies, allows both an SBC and multiple I/O boards to be plugged together ‘piggyback’ style, one on top of another. This allows a system designer to select various boards from different manufacturers or do a custom design. Comparing a stackable system to a system using a COM module on a carrier is similar [based on I/O requirements]. The main difference is that each carrier module is typically unique to its application due its target volumes being greater than a few thousand per year.”
When it has to be rugged
Our panel also considered the notion of how rugged a solution needs to be and reflected on how standardizing things like mounting holes, connector types, and locations can help increase ruggedness.
Blazer: “PC/104’s small size (3.6" x 3.8") and four mounting holes minimize its susceptibility to vibration, and its proven rugged stacking connectors make it a natural in applications that experience high shock and vibration…. When I think of PC/104, I think of small and rugged.” [Editor’s note: See Figure 1 for an example of a PC/104-Plus board in a rugged slice enclosure.]
Persidok: “Because of [PC/104’s] small size, low-power systems do not have to dissipate as much heat. This attracts more applications with greater temperature range tolerance. More and more vendors are now offering extended temperature components. The secure and rigid four-hole standoff mounting and solid pin contacts allow PC/104 to be used in higher shock and vibration applications.”
Lehrbaum: “The embedded stack or sandwich approach works well for fixed, mobile, indoor, and outdoor requirements. Many of these applications require robust board-to-board connectors, well-distributed mounting holes, and passive cooling mechanisms. These needs can be met by COM standards such as COM Express and by stackable SBC standards, including EBX, EPIC, and several PC/104-sized module variants.”
When it has to be fast
There seems to be a healthy debate about performance, and it’s here where our panel diverged a bit on how to achieve speed. You be the judge on which of these positions is best.
Finstel: “This [COM] approach provides a reliable connection between the processor module and the carrier board via reliable, proven, high-bandwidth connectors for PCI Express, GbE, HD graphics, USB 2.0, and other state-of-the-art serial interfaces defined by PICMG for the COM Express standard. To assure proper signal integrity, a COM and carrier board approach is the most favorable due to the fact that stacking or cabling [can] violate the design rules of high-speed signals.”
Winfield: “The PCI and ISA legacy parallel bus architectures are yielding to higher-speed serial buses such as USB and PCI Express. That is why we have chosen the Stackable Unified Module Interconnect Technology (SUMIT) architecture for future designs. It supports the new lower-power mobile chipsets from Intel and VIA and their serial I/O signals. SUMIT can be used on industry-standard 90 mm x 96 mm modules as well as for I/O expansion on EPIC, EBX, and Pico-ITXe SBCs. [Editor’s note: See Figure 2 for an example of an EPIC board with both SUMIT and PC/104 I/O.] It spans the range from slower-speed Serial Peripheral Interface (SPI) and Low Pin Count (LPC) buses to USB and PCI Express on the same connector because its bandwidth supports up to 5 Gbps. This performance level is measured with up to four modules stacked together and offers excellent performance backed by testing from Samtec’s Signal Integrity division.”
Persidok: “The wired module concept has been growing due to the popularity of the USB interface. The advantages are obvious because I/O is not limited by the CPU board’s form factor. I/O can now be installed in close proximity to the sensors or devices being controlled. In the analog world, this is a critical advantage given the potential of increasing interference and noise commonly seen with longer cables. It is also less burdensome to protect and route a single serial interface rather than a pile of individual sensor points. Also, unlike a bus, there is no bandwidth sharing…. USB 3.0 will be full duplex and capable of 10 times the bandwidth of USB 2.0.”
Wooley: “Technically, the best signal integrity and throughput is achieved by the shortest communication path.... With stacking connectors being able to support speeds of more than 5 Gbps, high-bandwidth buses such as PCI Express and the much-anticipated USB 3.0, which boasts 4.8 Gbps performance, are easily used for interboard communications without speed and bandwidth limitations…. Having a faster bus speed does not default to a higher throughput. The more data passed or multiplexed through a single channel, the slower the performance, resulting in smaller overall system bandwidth…. [With] 10 root USB channels, the concern of bandwidth hogging is eliminated. Unlike an interface that provides a single channel to be multiplexed and shared among many peripherals, StackableUSB allows each individual peripheral connected to a root channel to take full advantage of the 480 Mbps USB 2.0 transfer rate, if required.”
Blazer: “The addition of PCI Express to an embedded standard brings additional performance. The PCI/104-Express specification, which has four x1 links and a x16 link, gives a total throughput of 76 PCI buses. This can be tapped many ways. Everyone thinks of video cards, but FPGAs, DSPs, frame grabbers, data acquisition, SATA, and 1 and 10 GbE can all take advantage of the extra bandwidth.”
When it has to be expandable
If there is more space to work with and there aren’t any dramatic requirements for ruggedness, expansion becomes important. Although stacking modules provides the needed flexibility while offering ruggedness, there are some trade-offs.
Wooley: “By eliminating cables and avoiding the routing complexities that may be associated with carriers, one not only maximizes system performance by utilizing a stacking scheme, but also ensures that all the components are held together in the most rugged fashion.” [Editor’s note: See Figure 3 for an example of a StackableUSB stacking scheme.]
Lehrbaum: “One key disadvantage of stacks, particularly when the stack is more than a two-board sandwich, is that it’s hard to access boards inside the stack for maintenance, repair, or upgrade.”
Blazer: “When building a system, PC/104’s self-stacking ability and modularity are key benefits. You only pay for the boards you need, safe in the knowledge that you can always add another board at will. You will never run out of backplane slots, and you don’t have to oversize the system to allow for possible expansion. PC/104’s standard ISA, PCI, and PCI Express buses make it a very good fit for mezzanine applications, where it can be used as a macrocomponent on a large motherboard to provide an easily upgradable CPU or expansion capability for that extra Ethernet, video controller, or data acquisition card that your boss just told you must be included in the system.”
Finstel: “Stacking SBCs and I/O boards is a good fit when height isn’t a problem and there are appropriate off-the-shelf I/O boards to suit the application’s needs. COM can be thought of as the simple two-board stack, and it offers a different look at customization where the computer is standard and the I/O is dedicated to the application.”
Persidok: “The COM processor module is composed of all the electrical features except that the I/O bus and computer I/O are routed out of a mounting connector or connectors. The baseboard or carrier provides the I/O connectors and expansion. The customer has the option to bring out and combine only the computer I/O connectors desired and application I/O components and connectors needed for the customer’s solution. Multiple vendors can be sourced and qualified. Because the customer I/O is preset, CPU board obsolescence is less painful.”
Lehrbaum: “In the sandwich approach, one board – either a stackable SBC or a pluggable COM – implements the core embedded computer subsystem (CPU, RAM, mass storage, user interface, networking). The second board integrates the system’s application-specific functions, I/O, switches and push buttons, power supply, and others.
Cabling comes to mind when there is more space and distance, the connections are serial in nature, or there are somewhat unusual requirements.”
Winfield: “I prefer not to have high-voltage AC within a stack of boards. An optical isolated coupler should be mounted away for the safety of any maintenance personnel. In this instance, cabling remote modules together makes sense.”
Finstel: “Cabling together boards still works well and is an economical solution if the application is utilizing a larger chassis such as a full-size 3U rack-mount system.”
Persidok: “I/O [modules] can be powered over the USB cable…. Because a parallel bus and connector are not required, more I/O can be put on a smaller package.”
A whole menu to choose from
These are only some of the comments we received, and we certainly thank our panel members for participating in this forum. They raised many issues and choices that need to be considered when looking at both an application and the various approaches that can address it.
As a parting thought, Rick Lehrbaum left us with this gem, which brings us back to the deli metaphor: “Make mine a swiss, provolone, avocado, lettuce, tomato, and mustard on sourdough. Hold the mayo!” Your order will vary, and that’s OK; it’s a small form factor world, after all.
Jim Blazer is vice chairman and CTO of RTD Embedded Technologies, based in State College, Pennsylvania, where he is responsible for managing intelligent data acquisition systems and embedded PC designs. Jim currently serves as president of the PC/104 Consortium and holds a BSEE from Penn State. He can be reached at email@example.com.
Dirk Finstel is CTO of Kontron AG, based in Munich, Germany. Dirk has worked in the industry for more than 18 years and holds a BS in Computer Engineering and Science from the University of North Carolina at Charlotte. He can be reached at firstname.lastname@example.org.
Rick Lehrbaum is Executive VP of strategic development at Diamond Systems Corporation, based in Mountain View, California. Rick has worked in the industry for more than 26 years and holds a BS in Physics from New York University as well as an MS in Physics from the University of Louisiana at Monroe. He can be reached at email@example.com.
John Persidok is CEO, engineering director, and founder of ACCES I/O Products, based in San Diego, California. John has more than 35 years of experience in the electronics industry and holds an MSEE. He can be reached at firstname.lastname@example.org.
Jerry Winfield is president and founder of WinSystems, based in Arlington, Texas. Jerry has been active in the industry for more than 35 years and holds a Bachelor’s degree in Electrical Engineering from the University of Houston. He can be reached at email@example.com.
Susan Wooley is CEO of Micro/sys, based in Montrose, California, and serves as chairman of the Inter-Stackable Standards Group. Susan has more than 30 years of experience in the embedded world and holds a BA from Cal Poly Pomona as well as a Master’s degree in Public Administration from the University of Southern California. She can be reached at firstname.lastname@example.org.