How and where are SFP connectors and cabling used?

By Ed Cady, Contributing Editor

Small FormFactor Pluggable (SFP) connectors and cables are long time primary types of high-speed IO interface interconnect systems used to connect server, storage, switch, video and communication systems. Major market segment implementations include cloud datacenters, enterprise datacenters, HPC (High Performance Computing) labs, camera surveillance systems, Internet provider systems and machine vision systems.

The SFP interconnect system evolved from the HSSDC (high-peed serial data connector) back in the late 1990s. HSSDC-1 and HSSDC-2 were breakaways from the traditional use of two-piece metal connector contact systems, as the plug connector used a PCB with gold-plated pads to mate with the edge-style receptacle connector’s metal contacts. The HSSDC connector was selected for use by the Gigabit Ethernet IEEE-802.3z and Gigabit FibreChannel NCITS T11 standards organizations. These much smaller serial IO interface connector and cables were a welcome change from the previous very large, heavy, wide parallel IO interface connectors and cabling. SFP also has been used for 2.5, 5 and 8 Gbps data rate applications including InfiniBand standard single lane links.

The SFP interconnect system includes the use of various data speed rate transceiver modules that plug into the receptacle edge style connector that is inboard from the box bulkhead and under a SFP metal shield cage. The back end of the module is flush with the box bulkhead or back-panel. The small module outboard end has a dual LC receptacle port that provides mating connection with different types of external fiber-optic cables relative to link reach requirements.

SFP copper cables are usually constructed using two individually shielded twin-axial transmission elements within an outer shielding layer for controlling EMI and achieving EMC regulations. The copper wire conductors and differential pair shields and system shield are carefully laser trimmed, prepped, processed and terminated with reflow soldering or laser welded to the PCB pads. Increases of higher speed data rates once caused temporary use of awkwardly large 22-AWG wire size cables for longest length reaches. Over time, copper reach length requirements have been greatly reduced and some short applications now use very small 33-AWG wire size twin-axial elements, making the outer diameter cable size more acceptable. SFP is still popular for 1-, 2.5- and 5-Gbps camera network applications.

SFP+ is the 10-Gbps mainstream and current very high volume version of this interconnect family. Besides being specified in the 10 G Ethernet IEEE-802.3ak, this connector has been chosen for 14 Gbps InfiniBand and 16 Gbps FibreChannel. To meet the increased speed rate for medium and long reaches, active copper chips were added to the cable’s PCB plug connector for different signal conditioning, retiming or re-driving electrical options. Some OEM systems’ house cables have had EPROM chips on board for various system management functionality like identification and closed and secured networked system requirements. Sometimes there have been EPROM handshake inter-compatibility problems between different company products causing headaches for end-users. Although primarily a single lane interconnect, some passive copper SFP+ four lane designs have been used for 4L x 10 G = 40 Gbps short links versus using the larger traditional QSFP+ connector. SFP+ Active Optical Cables have been mostly used for 10+ meter reaches. This is done by having an E/O engine chip on the plug’s PCB and terminating optical fibered cable to the plug.

SFP28 is the new 25-32 Gbps version. The Ethernet Alliance and InfiniBand trade association plug-fest testing events have worked out most of the technical issues. It appears the markets will use both active copper and active optical cables for most reach requirements. See the Integrators List at www.infiniband.org for various qualified suppliers. It appears a back-to-back receptacle connector is being developed that would have internal copper twin-axial flat cables or better substrate quality, and PCB jumpers extending over the primary, but lower quality substrate PCB. These internal SFP+ jumpers connect back to the PCB very close to the switch chip or on top of the chip itself. IEEE802.3bj and IEEE802.3bm Ethernet standards make use of SFP28.

SFP56 is the latest developing 50-56 Gbps version. It appears that this would be used for mostly active optical modules, active optical cables and maybe some very short active copper cable applications which may be more expensive and power consuming.

µSFP is an emerging effort to specify a smaller package size connector that would enable more ports to be on a standard 1U box bulkhead. It may be a reaction to the possibility of USB3.1 Type C connectors and half size RJ connectors being evaluated for various commercial networks applications. It seems targeted for 25 Gbps single lane links being specified by the Cloud Datacenter Consortium and IEEE802.3by committee.

Some SFP specifications include SFF-8402, SFF-8074, SFF-8431 and SFF-8432, which can be found at www.sffcommittee.org. The members of the SFF committee do detailed interconnect specifications for open industry standard groups as well as the originating SFP MSA private consortium. The most popular applications are in ToR to Leaf servers all connected in one rack with a fan-out or hydra four SFP+ multi-legged cables which are connected together on one side with a QSFP+ connector and 12 SFP+ cable legs are connected to one CXP connector. The major threat to continued high volume SFP related interconnect usage is the advent of inside-the-box, mid-board optical engines and external passive optical cables like the new MXC interconnect.