The CDFP 400 Gig module
- The CDFP will be a 400 Gig short reach module
- Module will enable 4 Terabit line cards
- Specification will be completed in the next year
A CDFP pluggable multi-source agreement (MSA) has been created to develop a 400 Gigabit module for use in the data centre. "It is a pluggable interface, very similar to the QSFP and CXP [modules]," says Scott Sommers, group product manager at Molex, one of the CDFP MSA members.
Scott Sommers, MolexThe CDFP name stands for 400 (CD in Roman numerals) Form factor Pluggable. The MSA will define the module's mechanical properties and its medium dependent interface (MDI) linking the module to the physical medium. The CDFP will support passive and active copper cable, active optical cable and multi-mode fibre.
"The [MSA member] companies realised the need for a low cost, high density 400 Gig solution and they wanted to get that solution out near term," says Sommers. Avago Technologies, Brocade Communications Systems, IBM, JDSU, Juniper Networks, TE Connectivity along with Molex are the founding members of the MSA.
Specification
Samples of the 400 Gig MSA form factor have already been shown at the ECOC 2013 exhibition held in September 2013, as were some mock active optical cable plugs.
"The width of the receptacle - the width of the active optical cable that plugs into it - is slightly larger than a QSFP, and about the same width as the CFP4," says Sommers. This places the width of the CDFP at around 22mm. The CDFP however will use 16, 25 Gigabit electrical lanes instead of the CFP4's four.
"We anticipate a pitch-to-pitch such that we could get 11 [pluggables] on one side of a printed circuit board, and there is nothing to prohibit someone doing belly-to-belly," says Sommers. Belly-to-belly refers to a double-mount PCB design; modules mounted double sidedly. Here, 22 CDFPs would achieve a capacity of 8.8 Terabits.
The MSA group has yet to detail the full dimensions of the form factor nor has it specified the power consumption the form factor will accommodate. "The target applications are switch-to-switch connections so we are not targeting the long reach market that require bigger, hotter modules," says Sommers. This suggests a form factor for distances up to 100m and maybe several hundred meters.
The MSA members are working on a single module design and there is no suggestion of future additional CDFP form factors as this stage.
"The aim is to get this [MSA draft specification] out soon, so that people can take this work and expand upon it, maybe at the IEEE or Infiniband," says Sommers. "Within a year, this specification will be out and in the public domain."
Meanwhile, companies are already active on designs using these building blocks. "In a complex MSA like this, there are pieces such as silicon and connectors that all have to work together," says Sommers.
Luxtera's 100 Gigabit silicon photonics chip
Luxtera has detailed a 4x28 Gigabit optical transceiver chip. The silicon photonics company is aiming the device at embedded applications such as system backplanes and high-performance computing (HPC). The chip is also being used by Molex for 100 Gigabit active optical cables. Molex bought Luxtera's active optical cable business in January 2011.
“Do I want to invest in a copper backplane for a single generation or do I switch over now to optics and have a future-proof three-generation chassis?”
Marek Tlalka, Luxtera
What has been done
To make the optical transceiver, a distributed-feedback (DFB) laser operating at 1490nm is coupled to the silicon photonics CMOS-based chip. One laser only is required to serve the four individually modulated 28Gbps transmit channels, giving the chip a 112Gbps maximum data rate. There are also four receive channels, each using a germanium-based photo-detector that is grown on-chip.
The DFB is the same laser that Luxtera uses for its 4x10Gbps and 4x14Gbps designs. What has been changed is the Mach-Zehnder waveguide-based modulators that must now operate at 28Gbps, and the electronics amplifiers at the receivers. “The chip [at 5mmx6mm] is pretty much the same size as our 4x10 and 4x14 Gig designs,” says Marek Tlalka, director of marketing at Luxtera.
Source: Luxtera
Luxtera is announcing the 100 Gigabit chip which it is sampling to customers. Molex, for example, will package the chip and the laser to make its active optical cable products. Luxtera will package the transceiver chip and laser in a housing as an OptoPHY, a packaged product it already provides at lower speeds. The company will sell the 100Gbps OptoPHY for embedded applications such as system backplanes and HPC.
Applications
The 100GbE transceiver chip is targeted at next-generation backplane applications as well as active optical cables. And it is enterprise vendors that make switches, routers and blade servers that are considering adopting optical backplanes for their next-generation platforms, says Luxtera.
According to Tlalka, system vendors are moving their backplanes from 15Gbps to 28Gbps: “It is pretty obvious that building an electrical backplane at this data rate will be extremely challenging.”
When vendors design a new chassis, they want it to support three generations of line cards. Even if a system vendor develops a 28Gbps copper-based backplane, it will need to go optical when the backplane data rate increases to 40-50Gbps in 2-3 years’ time and 100Gbps when that speed transition occurs. “Do I want to invest in a copper backplane for a single generation or do I switch over now to optics and have a future-proof three-generation chassis?” says Tlalka.
Exascale computers, 1000x more powerful than existing supercomputers planned for the second half of the decade, is another application area. Here there is a need for 25-28Gbps links between chips, says Tlalka.
System platforms and HPC are ideal candidates for the packaged transceiver chip but longer term Luxtera is eyeing the move of optics inside chips such as ASICs. Such system-on-chip optical integration could include Ethernet switch ICs (See example switch ICs from Broadcom and Intel (Fulcrum)) and network interface cards. Another example highlighted by Tlalka is CPU-memory interfaces.
However such applications are at least five years away and there are significant hurdles to be overcome. These include resolving the business model of such designs as well as the technical challenges of coupling the ASIC to the optics and the associated mechanical design.
Standards
Luxtera's 100Gbps transceiver chip supports a variety of standards.
Operating at 25Gbps per channel, the chip supports 100GbE and Enhanced Data Rate (EDR) Infiniband. The ability to go to 28Gbps per channel means that the transceiver can also support the OTN (optical transport network) standard as well as proprietary backplane protocols that add overhead to the basic 25Gbps data rate.
In addition the chip supports the OIF's short reach and very short reach interfaces that define the interface between an ASIC and the optical module.
The chip is also suited for some of the IEEE Next Generation 100Gbps Optical Ethernet Study Group standards now in development. These interfaces will cover a reach of 30m to 2km.
400GbE and HDR Infiniband
Luxtera says that it is working on different channel ’flavours' of 100G. It is also following developments such as Infiniband Hexadecimal Data Rate (HDR) and 400GbE.
HDR will use 40Gbps channels while there is still an industry debate as to whether 400GbE will be implemented using ten channels, each at 40Gbps, or as a 16x25Gbps design.