High Speed I/O Connection with Small Form Factor

Small Form Factor connectors and assemblies, both copper and fiber, have been the evolving solution to the need for high-speed I/O interconnect in telecom and datacom applications. The following is a discussion with Joe Dambach, Molex global new product development manager.

What is the history of Small Form Factor with high-speed I/O connections for telecom/datacom (from first iterations of SFF to QSFP+)? What drove and shaped the standards?

As data rates for I/O have increased, the need for greater flexibility in the way that specific ports are implemented has grown. For most networking/switching equipment, it is difficult to always know the specific run length each port requires. Shorter distances can be connected with passive copper cables, medium distance with active copper or fiber, and long distance needs fiber. In order to optimize the overall economics for an installation, this kind of port flexibility is very desirable, as opposed to dedicating specific ports to specific transmission media types.

How did Molex first become involved with high-speed I/O connections for telecom/datacom?

Molex has been involved in the I/O product space for a number of years, with products evolving as market needs evolved. SCSI was one of the first high-speed (for the time!) areas we became involved with, then PCIe, InfiniBand, SAS, Fibre Channel, and Ethernet. Today, Molex is involved in all of these standards, and supplies products to all the major markets.

SFP/SFP+ can be described as the 1 Gb/s to 10 Gb/s interconnect system. How are these data rates achieved? Are they through single channels or multiple channels, and why?

SFP/SFP+ is intrinsically a single transmit channel and receive channel system. It always implements a point-to-point interconnect. SFP as a form factor pretty much replicates the density of all the existing RJ-45 stacked/ganged products, and supports the same kind of network topology, while offering longer transmission distance than is supported in 10GbaseT interfaces. SFP+ is also very popular for supporting high bandwidth box-to-box links, at 10 Gb/s, while supporting a higher number of user interfaces running at 1 Gb/s speeds, or even Fast Ethernet.

Are SFP and SFP+ intermatable? What are the differences and similarities between the two?

Yes, legacy intermate is always a critical consideration in any standardized I/O. Use of existing cable and module infrastructure is a must, as long as possible. There are some differences, such as an extra degree of EMI shielding on the plug for SFP+, and of course, cable and internal circuitry. SFP+ cables and connectors provide full intermateability and compatibility with SFP, and will dial-down speeds to the lowest common denominator in the system.

QSFP/QSFP+ can be described as the 10 Gb/s to 40 Gb/s interconnect system. How are these data rates achieved? Are they through single channels or multiple channels, and why?

QSFP/QSFP+ basically leverages the pluggable I/O concept of SFP into applications that require a fatter pipe for a point-to-point connection by offering more channels. QSFP started in support of 4 Gb/s Fibre Channel, aggregating 4 channels of data for a 16 Gb/s total. Today, QSFP+, running 10 Gb/s (or now 14 Gb/s to support InfiniBand FDR) on each channel, increases this aggregate bandwidth to 40 Gb/s (54 Gb/s for InfiniBand) to support 40GbE and 4 aggregated 16 GFC links. (Note: 16 GbFC actually runs at around 14 Gb/s).

Are QSFP and QSFP+ intermatable? What are the differences and similarities between the two?

Very much like SFP and SFP+, the QSFP+ supports legacy interconnects. The Molex QSFP and QSFP+ connectors are actually the same. We try to provide as much future-proofing as possible. QSFP and QSFP+ are virtually identical in appearance, and most of the differences lie in the cable and the silicon.

What are the differences and similarities between Molex’s zQSFP+ and QSFP+? What performance differences are there between the two, and why?

The existing QSFP+ and the new zQSFP+ will continue to support the legacy requirements as above. zQSFP+ is targeted to 4 channel, 25 Gb/s+ serial applications, focused initially in 100GbE and Infiniband. From a signal integrity viewpoint, zQSFP+ is a much more capable connector. This is achieved by using a different technology for the basic connector, which we term preferential coupling. Different manufacturing techniques are required, but the performance improvement is dramatic.

The next step on the high-speed I/O ladder seems to be 100 Gb/s. When is this likely to happen, and why? Do you think there will be intermediate steps, like 4-25 Gb/s channels, and why? What will the 100 Gb/s connector likely look like? How will it dissipate the heat from the active optical components? Is it likely to be 100% optical, or a mix of copper and optical, and why?

100 Gb/s is the primary goal in the development of the 4X25 Gb/s products described above. My opinion is that as long as bandwidth demands continue to grow, denser and faster interconnect systems will be required. Indeed, the entire world should be interconnected via glass fiber today, per the prognosticators of 10 years ago. However, as long as copper and electrons are the primary processor paradigm, I think copper interconnects will have a place and continue to scale with bandwidth requirements. Incredible things have been accomplished to take Ethernet from its 10Mb origin to today’s planned 100 Gb/s, and I expect future innovation to be just as incredible.