by Meghan Fuller Hanna
In its recent report, "New Markets for Telecom & Datacom Lasers: 2007 to 2012," CIR analysts note that parallel optics will dominate short-reach computing center interconnects and eventually LX4-type transponders for the 100-Gigabit Ethernet (100GbE) market. As early as 2012, they say, more than 3.4 million lasers will be used to support connections at 100 Gbits/sec and higher, with most of this bandwidth being supplied through parallel schemes.
In his presentation, "On the Road to 100-Gigabit Ethernet Transmission," delivered last month during BICSI's Winter Conference in Orlando, FL, Andrew Oliviero, senior product manager for optical fiber products at OFS, noted that vertical-cavity surface-emitting laser (VCSEL) arrays likely would be used as the light source in such applications, and those VCSEL arrays would be connected to MPO/MTP connectors.
So what does this burgeoning parallel optics market mean for the makers of parallel optical connectors like the MPO/MTP? As it turns out, not a lot; the connectors themselves already are sufficient for use in 100-Gbit/sec applications. Ditto for the emerging quad small-form-factor pluggable (QSFP) module, whose multisource agreement (MSA) specifies its use with the MPO.
Developed in the late 1980s by NTT, the multifiber push-on or MPO connector (an enhanced version is sold under the acronym "MTP" by US Conec), can be fitted with ribbons of 12 to 72 fibers. The connector was first deployed to aggregate fibers and alleviate fiber congestion.
"People would take LC or SC connectors on one side and aggregate those into a single MPO," explains Dave Carter, product manager of fiber-optic cable assemblies at Tyco Electronics. "You could have a fan-out cable that would plug into, say, six SFP ports on one side, and then you could come out your face plate with an MPO connector and save a lot of space on your front ports, on your I/O panels."
Russ Granger, a product manager at US Conec, says using MPO/MTP connectors with parallel optics modules like the newly developed QSFP transceiver is "just a natural progression of thought. The QSFP comes along, and what connector can we plug in?" he muses. "The MTP/MPO is a natural selection because it's mature."
And it is well understood and field-proven, already having been deployed with POP4 and SNAP12 parallel optic modules. As Tom Schiltz, global product manager of MT-based products at Molex, is quick to note, the QSFP transceiver is only a third as dense as the SNAP12 module, requiring only four transmit and four receive channels. The QSFP will likely inhabit the space of traditional SFP modules in data center applications, including InfiniBand. "You can save footprint space on the front panel by going with one QSFP versus four SFP modules," he says.
Perhaps the most interesting application for the MTP/MPO is 40/100GbE transmission, which the IEEE's Higher Speed Study Group (HSSG) hopes to standardize by mid-2010. Says Doug Coleman, Corning Cable Systems' manager of technology and standards for private networks, "We probably won't see actual parallel optics being used in Ethernet, I don't believe, until the 40-gig work is done. I think the first use of MTP actually will be with 40 gig, and I think it will probably end up being something like a QSFP."
Even 100GbE transmission, "from a connector point of view, [is] going to be a piece of cake," declares Tyco's Carter.
According to Coleman, Corning Cable Systems presented an economic analysis at an IEEE HSSG, and even though the HSSG has announced 100 m as its objective for OM3 fiber, Corning believes that distance could and should be extended out to 150 to 200 m. "We believe that for 100 gig on OM3 fiber, there is clearly an economical value proposition to be able to support those distances by utilizing two ribbons, using one MTP to transmit and the other MTP ribbon to receive on each ribbon," he explains.
If the HSSG standardizes on the short-reach OM3 parallel optics architecture for 100GbE, Coleman believes electronics manufacturers will use two 12-fiber MTP connectors. "Some people were speculating that they might use a 24-fiber MTP, but I think there are electronic design and packaging issues that are going to make it more feasible to use two MTP connectors at each end versus one 24-fiber MTP at each end," he says. "I would send multiple 10-Gbit wavelengths over 10 fibers in one direction, and do the same thing over 10 fibers in the opposite direction. Ultimately, it's going to take, we believe, 20 fibers."
In other words, this implementation would use four MTPs total, two at each end, with 12 fibers per MTP ferrule. Two slots would be allocated on the switch card, one for transmit and one for receive.
While the connectors themselves may be robust enough to handle an aggregate bandwidth of 100GbE or greater, a parallel optical architecture does present some technical challenges. Alan Ugolini, manager of data center market development at Corning Cable Systems, cites two major issues: the total connector loss budget allocated for 40/100 Gbits/sec and the skew rate. To mitigate the effects of the former, he says, Corning is particularly conscious of its polishing process. "We want to make sure that our connectors are very precise and the alignment is such that we can reduce losses and meet any kind of feature performance requirement."
Tyco's Carter agrees. "The biggest challenge that the cable assembly manufacturers have in terminating these connectors is to make sure that they have a proven, consistent polish process such that when you randomly mate any two connectors together, the endface geometries are good enough to guarantee you physical contact so you get the best optical performance over all environmental and mechanical conditions," he explains. Ideally, those fibers need to be polished "absolutely perfectly," he says, "and that's challenging enough to do when you have a single-fiber LC, but when you're doing that across 12 fibers in an array, like in the MPO connector, it is a challenge."
For single-fiber connectors like the SC and LC, there are industry standards that dictate endface geometries, fiber protrusion, and other characteristics, but no such specifications exist in the MPO/MTP world. For now, it's up to each individual vendor to put its polished connectors through various environmental and mechanical tests to prove their reliability and robustness. As a result, any two cable assembly manufacturers could have radically different endface geometries. "A cable assembly [from supplier A] mated with supplier B may not give you the same results because their endface geometry characteristics -- the fiber protrusion, the delta between the fibers, the X angles, Y angles -- some of those things may be very different," says Carter, "but they are all critical to how that thing performs."
Thanks to the increasing popularity of parallel connectors, standards bodies like Telcordia and the IEC in Europe have begun to look more closely at the issue. "They have been reluctant to dictate what that endface geometry has to look like, but they are starting to describe more and more what the performance characteristics have to be," Carter reports.
Corning's Coleman believes monitoring the skew rate also may be critical for distances beyond 100 m. Parallel optics relies on spatial division multiplexing, in which the circuit is spatially divided among multiple fibers and simultaneously transmitted across those fibers. "The difference between the fastest pulse and the slowest pulse is an elementary definition of skew," Coleman explains.
He says the laser-optimized 50-Âµm multimode (OM3) fiber specified by the HSSG will be of such high quality, it likely will not have a significant impact on the overall skew. "It's how you cable it and how you terminate it that's going to impact the skew," he maintains. "We can't necessarily take a cable that has a whole bunch of loose fibers in there, ribbonize them and terminate them with an MTP connector, and be assured that they are going to meet a skew requirement. If you have fibers of different lengths, those differences in lengths can result in [the inability] to conform to the skew objectives."
At press time, those skew objectives had not yet been defined for 40/100GbE applications, but, says Coleman, "This may be a big deal for cable and connector manufacturers."
Despite the technical hurdles ahead, the connector suppliers are bullish on the existing and emerging parallel applications. "We have seen an increase in usage in the parallel optics, mainly the SNAP12 modules right now," says Schiltz. "And we've seen people evaluate and start to test the QSFP modules, but I haven't seen anyone do a full release on the QSFP yet. I think they are still in the evaluation stage, but we expect it to be relatively strong in the years to come."
Granger says the market, by and large, is becoming more comfortable with the MTP/MPO connector. It's taken some time, he says, but now the MTP/MPO is in volume use and people are gaining confidence in it. "Now when designers are creating new systems, it's a choice for them to use it," he notes, "and that choice is made more easily because of their confidence in the connector's capability."
Meghan Fuller Hanna is senior editor at Lightwave.