It’s likely that your current designs have you pushing the proverbial envelope. If so, then high-speed serial interfaces are the way to go. Their overall bandwidth beats their parallel counterparts. Also, the newer technologies offer plenty of other benefits, such as lower pin counts and hot-swap support. The clear leader is PCI Express (PCIe), followed by Ethernet, Serial RapidIO, and InfiniBand. HyperTransport remains a chip-to-chip link, even though board standards are defined for it. PCI Express has edged out AGP, and PCI is quickly disappearing on the PC motherboard, just as PCI pushed ISA into near oblivion. Still, the compatibility between PCI and PCIe makes PCIe much easier to support. Serial links also have significant advantages outside the box. External PCI Express (ePCIE) is finding homes in a host of applications, from box-to-box links to external peripherals. Likewise, eSATA is an alternative to USB and IEEE 1394 when it comes to storage. In terms of backplane standards, only USB has found a niche at the low end. Storage connections tend to wind up in specialized environments.
GOING ALL SERIAL
Established parallel bus standards remain the mainstay, but serial alternatives exist with the same form factors (see “Serial- Parallel Alternatives” at www.electronicdesign.com, Drill Deeper 18798). Only the connectors are different, giving designers a significant advantage since it’s often possible to mix boards with the proper backplane. This is typical on the PC side, where motherboards have a collection of PCIe, PCI, and, sometimes, ISA slots. These often show up in motherboards targeted at embedded environments where legacy boards abound.
The high-speed serial interfaces each have their niche with minor overlap, typically involving PCIe and Ethernet. Interestingly, from a backplane point of view, the wiring and connector requirements for PCIe, Ethernet, Serial RapidIO, and InfiniBand are essentially identical, as are the serializers/deserializers (SERDES) used to implement them. The SERDES are found in FPGAs, which is why many serial backplane standards support a range of serial interfaces and why a single FPGA board can support any of these standards.
The convergence of board and connector form factors contrasts with the partitioning of products based on the backplane interface. The rooted-tree nature of PCIe is great from a compatibility standpoint, but it means alternatives like Ethernet, Serial RapidIO, and InfiniBand are needed for more networkstyle connectivity. PCIe can support multiple peers, though architecture, overhead, and legacy support tend to get in the way of turning it into a fabric backplane.
The switch to serial has also made a difference in the move from 6U to 3U form factors. Of course, increased integration and higher-performance chips have played a part as well. But to take advantage of these advances, the smaller boards need off-board throughput more than they did than in the past.
Developers continue to innovate, but the new high-speed serial interfaces are up to the challenge. Tom Cox, executive director of the RapidIO Trade Association, notes that members are quite comfortable with the performance of the RapidIO Specification 1.3, even though 2.0 has been approved and 3.0 sits on the drawing board.
Performance remains an issue as evidenced by the x16 PCIe video interface. But the range of requirements in the embedded space often makes even x1 PCIe overkill. This does illuminate another key advantage of serial interfaces—they’re scalable, allowing designers to jump from x1 to x2 to x4 and so on without needing to move up to higher link transfer rates.
The combination of interfaces and board form factors leads to a large number of options, though in practice, only a few areas compete directly. For example, the 3U and 6U CompactPCI/Compact- PCI Express form factors match up with the VME/VPX/VXS standards, but established use of the parallel versions often dictates the choice of the serial versions. Things are getting a little more interesting with serial interfaces when it comes to stacking standards.
STACKING UP STANDARDS
The 2008 Embedded System Conference last month in San Jose was the site of two key announcements from the PC/104 Embedded Consortium and the Small Form Factor SIG. Both look to bring PCI Express to the stacking, small-form-factor arena built by PC/104 (see the table). They both use a high-performance connector like Samtec’s SUMIT (Stackable Unified Module Interconnect Technology) with a 0.6-in. height that matches the PC/104 standard.
The PC/104 Embedded Consortium defines two standards: PCIe/104 and PCI/104-Express. Like PC/104 and PC/104-Plus, the new standards designate a single- and double-connector implementation. PCI/104-Express maintains the PC/104-Plus PCI connector, replacing the ISA connector with a PCI Express connection. As a result, the new boards can be combined with new and existing PCI-104, PC/104 boards that have only the PCI connector.
The PCI/104-Express stacking system (Fig. 1) resembles the system originally defined for EPIC Express, the precursor to this standard (see “EPIC Express Rides The Rails” at www.electronicdesign.com, ED Online 14190). This allows point-to-point PCI Express connections to be routed to boards in a stack.
The PCI-104 and PC/104 boards don’t have this problem because they utilize a bus. With PCI Express, each board uses the first PCI Express connection and shifts the connections so the next board in the stack sees the next unused connection as its first connection.
The shift only works if there are enough connections at the outset. The number depends on the types of connectors that are used. A full-featured system can support an x16 link or a pair of x4 links and four x1 links. The approach essentially puts PC functionality in a PC/104 form factor. This is comparable to what the original PC/104 standard did, followed by PC/104-Plus.
The Small Form Factor SIG Express104 employs a similar stacking approach, but the connections aren’t limited to PCIe. It uses the SUMIT standard connector, even though SUMIT is the connector standard while Express104 is an instance of a standard that uses SUMIT. Expect more standards from the Small Form Factor SIG based around SUMIT in the near future.
Continue to page 2
USB, SPI, SMB, and LPC also must be considered with PCIe. They target lowerend, peripheral complements that may or may not contain PCIe devices. This actually makes PCIe a better replacement for PC/104. Moreover, the PCIe/USB combination permits support for ExpressCard devices.
The standard also allows processor boards to implement a subset of the interfaces, making the standard very interesting from a microcontroller perspective. That’s because many microcontrollers don’t have a PCIe interface, but they do support USB, SPI, SMB, or LPC.
Express104 is an interesting combination that mixes bus interfaces (SMB and LPC) with point-to-point interfaces (PCIe and USB). The signal shifting on a board for USB depends on the interfaces required by the board. Express104 also has different board sizes, though a PC/104-size board is in the mix. The smaller boards enable the stacking architecture to go into places where PC/104 will not fit.
The Small Form Factor SIG and the PC/104 Consortium standards overlap in their target audience, but they also address different arenas. Express104 is I/O-oriented with a nod to the mobile end of the spectrum. It can handle almost any peripheral used for data acquisition except for high-end video peripherals, though it can easily handle video with its PCIe links.
The PC/104 Consortium is looking toward the high end where high-performance video needs the x16 bandwidth. In the not too distant future, don’t be surprised to see a combination of the two in a single system.
Both PCIe standards will require new processor boards. The mix of peripheral boards will change as well. In fact, it will even be possible to build a board that mixes the two standards. Still, PC/104-Plus single-board computers like VersaLogic’s Cheetah will make up the bulk of shipments in this space for the next couple of years (Fig. 2).
Don’t expect to see InfiniBand and RapidIO in this space, but Ethernet, including Gigabit Ethernet, is quite common. The big difference is that Ethernet is used to link an end node to a network versus the backplane fabric found in new rack-mount systems.
Based on Motorola’s VERSAbus back in the 1970s, VME has one of the longest track records in the business. Since then, the system has grown to the 64-bit VME64. The 2eSST protocol found on VME320 systems boosted performance—though the 320-Mbyte/s throughput is high, high-speed serial links deliver better performance.
CompactPCI’s history isn’t as long as VME, but it has a similar 3U and 6U form factor. It showed up in 1995 as PIGMG’s PICMG 2.0 standard, based on the parallel-bus PCI architecture.
Both CompactPCI and VME are now found in rugged and military applications. They’ve also served as the backbone for a range of industrial applications. Both have complements in the test arena with board standards such as VXI and PXI. Likewise, each has moved into the high-speed serial space.
AdvancedTCA and its little brother MicroTCA are more recent entrants in the board space. Designed for carrier-grade communications applications, AdvancedTCA is a larger form factor that’s slowly emerging into other embedded application areas. Elma Bustronic offers an AdvancedTCA backplane that implements a dual-star serial fabric for redundancy not typically found in PCbased systems (Fig. 3).
MicroTCA is based on the Advanced Mezzanine Card (AMC) standard. AMC cards come in a range of sizes and can be found in AdvancedTCA carrier boards as well as MicroTCA racks. PICMG 3.x standards address the AdvancedTCA, MicroTCA, and AMC architectures.
The big difference between AMC and the VME/CompactPCI systems is the backplane. VME and CompactPCI use a parallel bus. Of course, neither of the standards organizations for these platforms has remained idle. For instance, PICMG 2.16 incorporates incorporates Ethernet into the backplane, while the newer CompactPCI Express standard blends PCI Express with the CompactPCI form factor. It’s even possible to use the Ethernet support without the PCI or PCI Express side, depending on the application.
In a similar vein, VXS and VPX build on the VME tradition with a range of serial fabrics. The VME International Trade Association (VITA) defines the VME, VXS, and VPX standards. VPX comes in 3U and 6U form factors and uses only serial interfaces for backplane communication. VXS (VME Switched Serial) is essentially a blend between VME’s parallel interface and VPX’s serial interface.
VPX was adopted for a growing number of applications using boards such as Curtiss Wright Controls Embedded Computing’s CHAMP-AV6 (Fig. 4). Like AdvancedTCA and MicroTCA, the VPX and VXS standards define a range of serial fabric support.
Mezzanine cards for the VME and CompactPCI factions haven’t turned into another backplane standard like AMC and MicroTCA, though they do tend to overlap. PMC (PCI Mezzanine Card) and XMC (Express Mezzanine Card) sockets can be found on a range of VME and CompactPCI families of boards. XMC supports PCIe, but, as with the other serial board standards, it also can handle interfaces such as Serial RapidIO.
Given the SERDES standardization, it’s not surprising that this is true. Of course, moving between standards and off-the-shelf products is another matter since demand usually drives availability. Something like an InfiniBand XMC board is likely to be a rare commodity. On the other hand, Infini- Band boards with a PCIe link through the XMC connection do exist.
KNOW YOUR OPTIONS
Surprisingly few choices exist when it comes to form factors. 3U and 6U boards are the most common in the board space, and CompactPCI, VME, and MicroTCA offer designers a large number of options that’s only exceeded by what’s available in the PC/104 and PC space.
Adding in the serial fabrics quickly raises the number of combinations significantly, but in practice many solutions wind up targeted at specific markets. InfiniBand settled into a supercomputer niche. Meanwhile, Serial RapidIO garnered much of the interest for designers in the high-performance military space as well as in midrange communications. AdvancedTCA remains the carrier-grade communications alternative.
PCI Express and Ethernet persist as the mainstays for low- to mid-range solutions from a backplane perspective. Both have significant limitations, as well as significant advantages. For example, legacy support is key to the success of PCI Express, yet the rooted-tree architecture makes peer-to-peer communication an interesting exercise that isn’t as easily expandable as RapidIO or InfiniBand.
Likewise, Ethernet offers the advantages of compatibility across its range and familiarity to developers because of its extensive use in networks. Ethernet interfaces are also integrated onto most higherend microcontrollers or the interface support chips of high-end microprocessors.
Downsides include power consumption, latency, and overhead on the network and the processor. TCP offload engines (TOEs) help, but they usually aren’t found in the low-cost Ethernet interfaces in most microcontrollers.