The X-Modules

10-Gbps fiber-optic ports are shrinking, but the hype meisters aren’t. A trio of front-panel, hotpluggable, transponder MSAs bring new opportunities for I/O designers who don’t accept vague claims that try to take the place of documentation and performance measurements.

Little pays tribute to the success of distributed computing so much as the evolutionary path of I/O technologies. With the notable exception of the portable sector, in which power users’ expectations ri se as dependably as—and at a somewhat faster rate than—cases shrink, the availability of computational resources is rarely the problem. Communication throughput is more often the limiting resource. The situation isn’t static, either. The business models for many of the new services you hear about from the press or on trade-show floors these days depend on the availability of inexpensive broadband communications, from large data stores to large numbers of arbitrarily located end users. The confluence of requirements for datacom and telecom and the development of physical-layer technologies that can address multiple protocols have formed a sweet spot at the 10-Gbps node (Reference 1). It appeared only a year ago that the market might converge on one or two I /O architectures and form factors. But the physicallayer market just a few quarters later offers those and three new module types, along with a host of questions you have to ask yourself before deciding which is best for your application. This flurry of activity in a sector as beset as any during the period, ought to make you wonder about not only the technical issues that distinguish the various physical-layer structures, but also the long-term prospects for whichever one you pick.

OBJECTS IN THE MIRROR ...

Modular I/O structures bring a host of benefits to the development, manufacture, and maintenance of datacom and telecom ports:

• Modularizing the port largely isolates the design, materials, manufacturing, and test issues linked to the line card's highest frequency analog circuits from those involving the rest of the product.


• Vendors can populate line cards and adapter boards with the most expensive functions as a post-assembly process, minimizing the yielded cost of their products and reducing the inventory-carrying costs.


• End users can populate dense cards as their needs evolve rather than tie up capital in unused capacity.


• Interoperability issues are isolated to the modular level, allowing system vendors, their board suppliers, and their customers to agree on and specify proven, stock I/Os.


• New suppliers can enter the market with a predefined, market-accepted qualification regime, fostering competition and securing the supply chain to a greater extent than can strictly proprietary designs.

Early modules that manufacturers developed under MSAs (multisource agreements)—such as the SFF (small-form-factor), SFP (small-form-factor-pluggable), 200PIN, and 300PIN MSAs—succeeded in demonstrating these points and reducing the barri ers to market acceptance of photonic technologies and the time to market for vendors providing them. The MSAs also provided a model for multiple module makers and component vendors to converge on electrical-, optical-, and mechanicalimplementation specifications beyond the scope of standards organizations.

The release of the Xenpak MSA a couple of years ago re p resented an important step in transponder evolution by providing the means for module replacement without disturbing the line card or any of its other ports (Figure 1). A front-panel, hot-pluggable module requires several electrical, electro mechanical, and mechanical features that board-mounted modules need not possess.


LOWERCASE X

As at tractive as the Xenpak form factor is compared with modules that are not front-panel hotswappable, the MSA does not meet the dimensional requirements of some important market segments. Though many services have converged on data rates at or near 10 Gbps, other parameters, such as the link length and its corresponding power dissipation, cannot converge.

The price the Xenpak form factor pays for a small-footprint module that can dissipate 10W is a tall package that straddles the pc board and, as a result , requires a board cutout. The cutout has been the source of much criticism of the Xenpak form factor because it adds to the pc-board manufacturing cost and eliminates precious routing space. Additionally, though the functional segmentation that Xenpak off e r s has proved successful, it is by no means the only possible arrangement.

Last year brought definitions of two new Xenpak-like transponder MSAs—Xpak and X2—that boast smaller package dimensions for high-density line-cards, PCI adapters, and other small-form at interface cards. Infineon, Intel, and Picol ight founded the Xpak MSA, which some 25 companies have since joined. This module definition is strongly similar to its older brother, Xenpak, in that it segments the modularized functions of the I/O channel at the physical-layer/media-access-controller interface through a four-lane (10-Gbps attachment - unit - interface) connection.

The Xpak MSA uses the Xenpak pinout, except that it includes a pair of clock signals dropped from later versions of the Xenpak standard. The addi tional clock pins provide for operation at 10GFC (10-Gbps Fibre Channel), 10GbE (10-Gbps Ethernet ) , and OIF SFI4-P2 (Optical Internet - working Forum SERDES /Framer Interface). The MSA provides a second pinout for the SFI4-P2 connection. The Xpak module’s dimensions demand les s of the I /O card’ s air rights than does Xenpak, eliminating the board cutout and allowing a standard top-side mount on a PCI card (Figure 2). Two versions of the Xpak module exist (Reference 2). Both occupy a 1.4232.98- in. maximum footprint—the length measured from the faceplate.

The low-profile version rises to a maximum of 0.391 in. above the board surface to fit within the standard PCI-card-component-height limit. This low-profile form factor can dissipate 4W wi th s tandard mounting in PCI NIC (network-interface card) or InfiniBand HBA (host-bus-adapter) and TCA (target - channel-adapter) applications. For longer reach applications outside a small-card environment, a tall version of the Xpak module provides for additional cooling fins to a maximum overall height of 0.881 in. The module footprint and its associated moduleholder design address board-edge and midboard mounting for elecommunications applications (Reference 3). Installations that require many channels, as is the case in the SAN (storage-area-network) and switching-center segments, can take advantage of double-sided “belly-to-belly” mounting arrangements. You can mount as many as 20 devices on a 17-in. pc board using such a layout.

The Xpak retaining mechanism comprises a rotating wire bale, similar to the SFP’s mechanism, that interlocks with the module holder. When you connect fiber-optic cables to the Xpak module, the cable’s connectors prevent the wirebale from moving to the unlatched position. Disconnect the cables, and this simple latching mechanism allows a module swap in less time than the Xenpak retainer that uses two captive thumbscrews. The Xpak latch provides a specified amount of compression to a conductive gasket implementing one part of the b reakaway EMI shield. A second module-guide mounting option provides an EMI seal for systems with painted faceplates.

With the variety of available fiber-optic module outlines, mounting options, packing densities, and cooling strategies, OEM engineers need to determ i n e the thermal perf o rmance of their I/O designs. You may simulate the thermal behavior, but, at some point in the design cycle, you need to reconcile your results with the fact that, from your customers’ perspectives, your models and their experiences are not interchangeable: Your simulations ought to correlate to physical measurements. Appendix A1 of the Xpak MSA contains a useful introductory guide to making controlled thermal - performance measurements. It provides dimensions and mounting details for a simple test chamber, recommended temperature- and air flow-measurement locations, and module operating conditions. As the document’s authors point out, this appendix can also guide discussions between module makers and OEMs about thermal performance and requirements.

X TOO

On the heels of the Xpak MSA announcement, a separate consort ium, the X2 MSA group , announced its own module agreement (Reference 4) . Eight companies—Agere Systems , Agil ent Technologies, JDS Uniphase, Mitsubishi Electric, NEC, Opnext, Optillion, and Tyco — formed the X2 g roup. (Since signing the X2 MSA, Agere has sold its optoelectronic business to TriQuint.)

X2 also uses the Xenpak electrical interface, though with a few exceptions. X2 provides a 4-bit port - address space-one bit narrower than Xenpak and Xpak. X2 also reduces the number of power pins and makes common the chassis and the electrical ground. The MSA also reserves Xenpak’s four vendor-specific pins. On the optical end, the X2 MSA supports 10GbE, OC192 SONET, 10GFC, and others (Reference 5).

As dimensionally similar as X2 and Xpak modules may be, they are mechanically incompatible, primarily due to different approaches to the guide mechanism (Table 1). Xpak uses different guides for its two module heights. Agilent claims that X2 has an advantage by using a single guide design for its three defined module heights (Reference 6). Beyond the obvious reduction of guide-part inventory, the claimed advantage may be real, but only in applications that meet two criteria: The assembly must use a midboard mounting arrangement; otherwise, the module position i s height-specific due to the panel cutout. The application must benefit from mixing module types, ostensibly as a group from instance to instance of the product. Perhaps an under-appreciated benefit of Xpak’s approach is that it prevents end users from mixing module heights within a single instance of a multimodule product. Such a mix would have airflow implications, the thermal consequence of which the end customer’s system operators may not find obvious.

Agilent also claims that X2 has better EMI and thermal performance. X2 uses the Xenpak EMI gasket-and-flange design and thus theoretically inherits the larger module’s EMI-seal performance. The Xpak design uses a more modest seal geometry, the adequacy or lack of which the MSA groups should demonstrate, not infer. EMI performance can be challenging to assess. Small differences in the evaluation method, which includes setup geometries, operating conditions, and evaluation criteria, can lead to significant diff e rences in the results. EMI requirements are largely a function of the end application, and many applications have adopted EMI standards that specify the evaluation method. EMI-performance claims and comparisons need to either reference a standard or document the evaluation method. This situation is true in the same sense that comparisons between track-and-field athletes need to specify an event.

Though differences in the two module’s case features are clearly visible and suggest dissimilarities in thermal performance, the extent of those performance dissimilarities is not apparent to the unaided eye. Here again, claims of superiority by a vendor may well be valid under application constraints, but these constraints remain unidentified, unspecified, and uncharacterized. Until the vendors identify the operating conditions and evaluation criteria and verify measurement results, consider these excellent topics for you to evaluate as you consider any high-speed, modular I/O.

In the meantime, the fact that the X2 document contains little information beyond mechanical drawings does not aid comparisons between the two form factors. Six months after its initial release, the X2 group has not updated the MSA with so much as a list of pin definitions.

UN XEN

A third MSA group, XFP (10-Gbit small-form-factor-pluggable module), offers a departure fro m the Xenpak architecture and its four-lane interface . XFP is a full-speed, single-lane serial alternative to Xenpak and its derivatives using an XFI (10-Gbit ser ial inter face) connect ion. Wi th no SERDES (serializer /deserial izer ) in the module, XFP is smaller and cheaper, and it uses less power. But it also does less of the job and leaves you with pcboard traces operating at four times the speed of the other MSAs. If port-packing density is an issue in the applications you design for, XFP is hard to beat. The dimensions provide for 16 modules on one side of a 17-in. -wide pc board. The footprint also p rovides for belly-to-belly mounting of a second course to the same board.

The trade-offs between XFP and either of the smaller-than-Xenpak alternatives are numerous and need careful consideration. Remember that you need to compare the total channel requirement for a given parameter to compare XFP with either Xpak or X2. You need to asses your product-development costs, bill of materials, power budget, manufacturing costs, and sustaining engineering costs for a channel from end to end, not just for the module, because the XFP segments the channel differently from the other MSAs. You cannot calculate the values one form factor provides outside the context of your project and your organization’s resources.

If you go back to the benefits of modularizing the I/O, the XFP form factor does force you and your customers to pay for the SERDES function a board at a time, not with single-channel granularity. On the other side of that issue, with XFP, you can take advantage of some of the multichannel SERDES chips and technologies available from companies specializing in high-speed communication on FR4. This flexibility mitigates some of the costs and may provide an overall channel-density win that none of the other current form factors can match.

Single-lane XFI is more compact to lay out than four-lane XAUI, and maintaining equal lengths and corner counts in your layout is less complicated if you’ re dealing with differential trace pairs and not clusters of eight. Layouts on FR4 are more challenging at 10 Gbps than they are at 2.5, so you should be sure that your team has the necessary engineering resources to generate a high-quality, high-speed board. The XFP-module vendors and the SERDES providers will no doubt have a good deal of application-support information. The end result, however, will be your team’s responsibility.

The XFI interface is protocol-agnostic and operates at 9.95 to 10.7 Gbps. XFP implements a two-wire I2Cserial monitor-and-control capability similar to that of Fibre Channel and OC-x SFP modules. Additional control lines provide reset, module detection, interrupt, module-not-ready, and similar signals.

APPEARING SOON ON A CARD NEAR YOU

Interestingly, virtually all of the Xpak and X2 MSA group members are also members of XFP. The perception may be that the overall 10-Gbps fiberoptic I/O market will segment into little-Xen XAUI and the littler XFP/XFI form factors. Alternatively, Xpak and X2 may be poised to re-enact the infamous VHS/Betamax battle. A market-adoption delay for Xpak and X2 could give XFP a footing despite the additional OEM-engineering load that form factor brings by allowing module makers and SERDES vendors the time to show I/O designers that 10 Gbps on FR4 is manageable.

The first models that vendors based on these MSAs are now available for sampling, and several manufacturers are readying others for this quarter and the second quarter of this year. Most vendors have not determined volume prices for the various form factors, so you should check with the vendors as they set prices during the next few months. A recent i-Suppli report suggests that the multi-thousand-dollar average selling price of 10-Gbps transponders is due to fall by as much as 50% in the next year (Reference 7). This trio of SAs gives credence to that prediction if the vendors can early on gain market momentum. On the one hand, end users tend to make do with their infrastructure during difficult times. On the other, Intel, for example, quotes International Data Corp statistics that Ethernet forms the basis of more than 85% of all installed network connections (Reference 8) . Between that number and the ubiquity of the PCI bus in enterprise environments, which Ethernet dominates, only a small fraction of the available nodes would need to become early adopters to make a market for new Xpak and X2 transponder modules.



References
1. Israelsohn, Joshua, “Fiber lights the short haul,” EDN, March 21, 2002, pg 61.
2. “A cooperation agreement for a small form factor pluggable 10 -Gbit/s transceiver package,” Xpak MSA Group, Revision 2.2, December 2002.
3. “Xpak MSA group announces build-to specification availability, enhancements to small-form-factor 10 -Gigabit Ethernet standard,” Xpak MSA Group press release, Sept 3, 2002.
4. “A cooperation agreement for a small versatile 10 -Gigabit transceiver package,” X2 MSA group, issue 0.9, July 31, 2002.
5. “Industry leaders announce ‘X2’ multi-source agreement for 10 -Gigabit pluggable optical transceivers,” X2 MSA group, July 22, 2002.
6. “X2 MSA Q&A document,” Agilent Technologies, July 22, 2002.
7. Rebello, Jagdish, “10 -Gigabit Ethernet in LANs, MANs, and WANs: real or just plain hype?” i-Suppli, April 2002.
8. “Evolution of Gigabit technology: from the backbone to the desktop,” Intel, 20 01.

Acknowledges

Thanks to Ed Rodrigues and Peter Bradshaw from BitBlitz and Steve Skiest from Molex for their contributions to this article.

You can contact Technical Editor Joshua Israelsohn at
(1) 617-558-4427, Fax (1) 617-558-4470
E-mail jisraelsohn@edn.com


Sidebar article: XEN AND THE ART OF FIBER-SYSTEM MAINTENANCE

Though conceptually simple and attractive, a front-panel, hot-swappable transponder module that operates at 10 Gbps requires design inputs from several engineering disciplines. Solid mechanical, thermal, photonic, RF, and power - management engineering must all combine efficiently to address module attachment, EMI, dissipation, and interconnection-integrity requirements.

The electromechanical elements of the design include an EMI shield, its conductive seal, and a dense, multiservice connector that carries power, control, and RF signals. The EMI shield’s breakable connection to ground must exhibit a low impedance at the signaling frequency; if it does not, the connection impairs the shield’s effectiveness. A gasket compressed between the module’s flange and the front panel during module installation establishes the Xenpak front-shield ground. The module uses a 70-pin connector, 16 pins of which form four differential signal paths in each direction that typically carry signals as fast as 3.125 Gbps.

Key mechanical attributes of the Xenpak MSA (multisource agreement) include the module guide, the retaining mechanism, and case designs. The guide and the retaining mechanism allow end users to quickly swap modules and establish the appropriate contact positioning and force; they also preclude the possibility of damage to either the module or the line card at the hands of an over exuberant technician. The case design sets the limits on module density and thermal load-critical issues that are always in tension, because, at the board level, they are the physical and thermodynamic reflections of channel count, channel speed, and link length. Implementation technologies link together the physical, thermodynamic, and photonic parameters.

Unlike the 300PIN (300-pin) MSA, which implements the 16-bit, parallel OIF (optical internetworking forum) SFI-4 interface standard, the Xenpak architecture suits the IEEE 802.3ae standard with its systemside, 4-bit XA UI (10 -Gbps attachment-unit interface). IEEE 802.3ae defines physical-layer interfaces for 10 -Gbps LAN and OC - 192 WAN payloads and seven PMD (physical-media-dependent) interfaces (Table A, Reference A).

Beyond the raw systemside-bus interface, comprising power, control, status, and data lines, the electrical interface also provides hot-swap protection across its width. Simple management schemes for power - line protection are well-established for a range of supply voltages and currents. Control-line, status-line, and data-line protection methods, however, are more application-specific and need to address signal bandwidths, amplitudes, and polarities.

REFERENCE:
A. Thatcher, Jonathan, “10 Gigabit Ethernet: brief introduction to the IEEE 802.3ae project,” Networld + Interoppresentation, May 2000."



Sidebar article: DON’T LOSE YOUR FOOTING ON THE BRANCHES OF THE DECISION TREE

Not long after the X2 MSA (multi source-agreement) Group announced its modular form factor, industry watchers began to speculate about how long it would be before X2 and Xpak converged. The similarities between the two MSAs were so striking, and their potentially significant differences were so undocumented that those outside the MSA groups found it challenging to find a discussion on that topic based more on measurements than on conjecture. Six months later, X2 has yet to fill in the blanks in its publicly accessible document. Several companies with interests in X2 have lodged vague claims and criticisms at the more complete Xpak MSA, mystifying some industry observers and leading to a bit of conjecture among others.

More than one participant in this market says off the record that the greatest attraction to X2 is that it is not Xpak. That statement seems cryptic until they add the fact that few people like it when their suppliers become their competitors.

Intel’s prominent position as a cofounder of the Xpak MSA group and as a manufacturer of some of the first modules in that form factor are perhaps the logical result of that company’s focus on the communications market as a chip supplier. But its role as a module maker puts it in a potentially awkward position with respect to an admittedly small portion of its ow n customer base. Keep in mind that no law exists against participating in your customers’—or your suppliers’—markets . The history of integration is the story of functional consolidation.

The lack of criticism of XFP for Intel’s membership is as mystifying as the original criticism of Xpak. In either case, it is unlikely that the world’s largest provider of semiconductors is going to ask permission before it enters markets that interest it.

Although the carping (and potential cross-carping) is of little engineering interest, the way it manifests itself can be. As you evaluate products that comply with the various MSAs, be aware that some marketing messages are more puff than substance, and some are just misleading. Choose the MSA footprint that best suits your application on its merits. Consider parametric performance, time to market, module cost, and channel oper ating budgets. Within an MSA group, choose module suppliers on their merits for design experience, ability to del iver, quality, and price. Climb the decision tree with care, and be aware that a few folks have been trying to undercut some of the branches.