Optical transceiver packaging for Gigabit Ethernet and Fibre Channel

Optical transceiver packaging for Gigabit Ethernet and Fibre Channel

Transceiver designers have several options when choosing a packaging design. Here`s what to expect from the most common approaches.

Bryan R. Gregory Molex Fiber Optics Inc.

As the Gigabit Ethernet specification nears completion and optical Fibre Channel links become increasingly common, networking hardware design engineers are under much pressure to incorporate optical links into their products. Generally, all of the rapid design cycle demands, marketing hype, vendor claims, standards committee announcements, and customer expectations seem to be taking their toll on the engineering and purchasing departments. Changes are so rapid that it`s difficult for engineers to spend time analyzing all their design options. This is especially true of all of the optical packaging options available for fiber-optic data links. This article will outline the five most common optical package standards: 1 ¥ 9, gigabit interface card (gbic), gigabit link module (glm), media interface adapter (mia), and the emerging mini-form factor devices.

The 1 ¥ 9 package style is the most common form factor in the market today (see Fig. 1). The package was originally developed for slow-speed light-emitting diode links. Recently, many companies have taken the obvious step of putting their gigabit-speed links into the same package.

The name "1 ¥ 9" originates from the single row of nine electrical pins that sit at the back of the package (two for ground, two for signal input, two for signal output, two for supply voltage, and one for signal detect). This device also has two mechanical mounting pins in the front that help secure the device to the board. This form factor is being considered in almost every new design because of its availability, small size, and low cost.

The format has several advantages. For example, when the market heats up, the 1 ¥ 9 will be the most readily available and cost-effective solution. At present, almost all component suppliers have put gigabit optical trans-ceivers on allocation. Lead times will probably remain long until suppliers increase capacity. Therefore, it is best to design in a package that is being built by most companies and in the highest volume. This format also is the least costly solution. The simple design is intrinsically less expensive than any of the other designs mentioned in this article--a situation unlikely to change in 1998.

The 1 ¥ 9 also is a simple design that offers a very small package. Several formats are available, such as DC-coupled inputs/outputs, AC-coupled inputs/outputs, and short- and long-wavelength lasers. The pin definitions are standard among all vendors, as is the footprint.

However, other factors also must be considered when evaluating this approach. The main concern with the 1 ¥ 9 package is compatibility between vendors. Even though these optical transceivers appear fairly simple, there are a couple of places where variation among optical component suppliers can require different support circuits. These compatibility issues include:

Terminations (AC-coupled, DC-coupled,

and mixed)--The term "AC-coupled" means that the optical transceiver includes a 50-ohm termination for the transmitter and receiver pins inside the device. The term "DC-coupled" means that these terminations are not included in the device and must be placed on the printed circuit board (pcb) near the pins. Most manufacturers offer either a DC transmitter and receiver, or an AC transmitter and receiver. It is also possible to offer a mixed device, where the transmitter and receiver have different termination schemes.

Signal detect trigger point--The ieee

802.3z standard allows a range of -30 to -17 dBm.

Electromagnetic interference (emi) emis-

sions--Designers will probably see significant differences in emi between devices of different manufacturers. Don`t assume that each manufacturer`s device requires special additions such as an external emi shield to meet minimum Class A or Class B requirements. There are some very good designs in the market.

Mechanical mounting pins--Grounded or

not grounded.

Package height--Slight variation among

vendors. Package length can also vary slightly.

Signal detect--Most vendors only offer

an ecl pin. There is some movement toward making this a ttl pin.

Of all of the compatibility issues, the only serious problem is the AC versus DC coupling. The other issues are not too significant, but must be taken into account.

gbic

The gbic is a module that is used in a manner similar to a pcmcia card (see Fig. 2). The optical modules contain an electrical input that is basically an sca2 connector and a duplex SC optical output. The electrical modules have a DB9 or high-speed serial data connector (hssdc) output.

During production, the pcb card has an unpopulated gbic cage mounted on the edge of the board. After production, a gbic card can be inserted into this cage according to the end-user`s specifications. Because these modules tend to be large and expensive, designers usually populate a pcb card or hub with a combination of electrical outputs, 1 ¥ 9s, and gbics. This allows a system to meet the optical short-wavelength or electrical DB9 requirements of the majority of their customers and still offer a couple of configurable ports for long-wavelength, electrical, or additional short-wavelength links. gbics are available in short-wavelength optical, long-wavelength optical, passive electrical, and active electrical formats.

The flexibility provided by this approach is one of the gbic`s advantages. Manufacturers can mount the card housing on the board and then install the correct module just before shipping. The gbic also enables a variety of features, such as laser shutdown control, additional grounding, and good emi performance. The gbic also benefits from the fact that there are few compatibility problems among vendors. With the gbic approach, components that fail in the field can be replaced efficiently.

These advantages must be weighed against two principal drawbacks: large size and relatively greater expense. Expense will likely remain a problem in the short term, particularly in comparison with the 1 ¥ 9 approach. Because the gbic module is intrinsically more complicated than a 1 ¥ 9 and because suppliers produce fewer of them than the 1 ¥ 9s, it is very unlikely that this price difference will change in 1998.

mia

The mia is a simple optical transceiver that has an electrical DB9 input and an optical duplex SC output (see Fig. 3). The device is basically an after-market adapter that is attached to an electrical DB9 output. This adapter protrudes 3 to 6 inches off the edge of the pcb card. It is also possible that, in the future, vendors may offer this with an hssdc input.

One advantage of the mia is that it does not require a dedicated optical card. If the majority of customers demand an electrical interface, the mia might be a good add-on solution for the occasional optical-link request. This is a device that, in theory, end-users could purchase and install, allowing them to adapt an existing system to meet a longer link requirement.

However, the mia suffers several disadvantages. For example, size represents a potential problem. These devices protrude from the end of a card. It can be difficult to design an enclosure that will accommodate the additional 6 inches required by an mia. Aside from the space required, special care must be taken to allow the fiber connector to mate with the mia and maintain a minimum bend radius. The lack of a controlled environment also can prove problematic, due to the delicate nature of optical transceivers. If the transceivers are not mounted on the card, it is difficult to guarantee their safety. Side pull and mechanical shock are the most serious concerns.

Occasionally, there are problems attaching an mia to the output of an electrical glm. It is important to test all of the possible vendor combinations of these two devices. Finally, the mia will usually be more expensive than a 1 ¥ 9.

glm

The glm is basically a daughter card with more-sophisticated functions than other devices (see Fig. 4). In addition to the optical trans-ceiver, the glm includes a serialization and deserialization chipset. These devices ease the design process, but they are large and somewhat expensive. They are available in open fiber control (ofc) and non-ofc versions. ofc is a built-in function that shuts down the laser if no optical signal is being received by the adjacent receiver. This function was popular among the very first generation of gigabit-speed devices, but has generally been eliminated from most vendors` products. Some vendors also offer an electrical glm card that has a DB9 output.

The glm offers parallel input, which eliminates the need to deal with very-high-frequency signals on the main pcb. The inclusion of serialization and deserialization on the same card also can be an advantage.

However, the glm is expensive. The component vendor receives a large portion of the value-added profit. The daughter card also is relatively large, and ofc versions are difficult to work with and test. Finally, the number of suppliers is limited. Most vendors are concentrating on newer packages such as the 1 x 9 and the gbic.

Mini-form factor devices

There has been a lot of discussion about the Mini-MT (mtrj) and Galaxy (Volition) packages (see Fig. 5). The purpose of this article is not to add to the overall debate (or "connector war" as its sometimes been called) but to make several brief comments:

Gigabit-speed versions of these devices

will probably not be available until late 1998.

When the devices are available, they

will probably be expensive. Despite the claims of both sides, the Volition connector is not being manufactured by any company other than 3M Telecom Systems Div., Austin, TX. The Mini-mt/mtrj is based on an MT ferrule that is very expensive and is currently being supplied by a very limited number of vendors. There are currently a number of U.S. companies selling MT-style connectors1, but almost all of these incorporate third-party ferrules. Until the Volition becomes a true multi-source product and the MT ferrules are manufactured by a number of North American and European companies, neither of these packages will be truly viable.

Because most of these gigabit-speed optical transceivers are new products, it is very important to carefully qualify each vendor`s component against the basic Fibre Channel or Gigabit Ethernet specification. Several other recommendations are:

Optical budget--Require a transmitter

output power as close to -5 dBm as possible and a receiver sensitivity of at least -18 or -19 dBm. If you don`t have margin on both the transmitter and receiver ends, interoperability may be a problem.

emi--Open-air transmission of an opti-

cal transceiver should be at least Class A. When these devices are placed inside a metal enclosure or behind a grounded faceplate, this can allow a well-designed system to achieve Class A or Class B.

Eye safety--Ensure that your vendors

have a cdrh (Center for Devices and Radiological Health) accession number and proof of iec 825-1 compliance.

Extinction ratio and jitter--Even though

both of these measurements are difficult to produce, try to verify that each vendor has good control over both these aspects of performance.

In conclusion, following the gigabit market requires constant effort. A number of component trends seem to make sense, but other trends will probably fade away. The mia and glm will probably remain niche products or will gradually phase out. Because of customer feedback and general volume trends, Molex has chosen to focus on the 1 ¥ 9 and gbic markets. These two products seem to be the most robust, with a large, stable supply base. It is also important to follow the developments of the mini-form factor devices. By the first half of 1999, these will probably become commonplace products. u

Bryan R. Gregory is product marketing manager at Molex Fiber Optics Inc., Downers Grove, IL.