QSFP-DD, OSFP, and CFP8: Which Is the Best for 400G?

There are three criteria for a successful form-factor: small size, low power consumption, and interoperability between all systems vendors. As we all know, the SFP/SFP+ and QSFP+/QSFP28 are successful form-factors for 1G/10G and 40G/100G networks. In fact, for 100G networks, there are 4 different form-factors: CFP, CFP2, CFP4, and QSFP28.

100G form-factors: CFP vs. CFP2 vs. CFP4 vs. QSFP28

100G form-factors: CFP vs. CFP2 vs. CFP4 vs. QSFP28

100G

The transmission departments in telecommunication networks need a pluggable transceiver able to cover long reach also using some dedicated technologies such as Coherent detection, while data centers need a small form-factor with the lowest power consumption and the lowest cost per unit due to their application is for short reach only (max 2km generally).

During the first instances of the 100G transceivers, the CFP form-factor was preferred because it was impossible to make a transceiver less than 12W power consumption, even for intermediate reach. Once the technology and components availability were better, it was then feasible to CFP2, and then CFP4. Still today, the Coherent technology for 100G and 200G is only available on CFP and CFP2 form-factors.

In parallel, the GAFA (Google, Apple, Facebook, and Amazon) with their phenomenal need for additional data center capacity, have pushed the QSFP28 form-factor for various short reach applications such as DAC, AOC, SR4, PSM4, and CWDM4.

Today, with the technology maturity and QSFP28 wide-adoption, most of the 100G applications are available on QSFP28 form-factor, with some exceptions for reach more than 40km, including Coherent detection.

400G

For 400G bit-rate, some essentials interrogations must be raised before going deeper into the subject:

  1. Who need the 400G transceivers?
  2. For which application?
  3. How about technology maturity?
  4. Any interoperability with former form-factors?

Following the market situation, 400G is a priority for the intra-connections in large data centers and at a lower scale for the transmission department in telecommunication networks. Because the 400G bit-rate requires PAM4 modulation, the reach is becoming more and more challenging and is limited to a few kilometers only. Longer reach will require Coherent detection technology and/or amplification, dispersion compensation, etc.

Again, we will observe a similar scenario: for 400G networks, a dedicated form-factor for the data center intra-connection (Intra-DC) and another one for transmission. However, thankfully it seems that “intermediate” form-factors won’t happen for 400G.

Intra-DC

400G is coming with 2 form-factors for Intra-DC: QSFP56-DD (QSFP-DD for QSFP Double Density) and OSFP (Octal SFP). Both form-factors are running 8 lanes of 50G PAM4 on the electrical side while the optical side can be either 8 lasers of 50G PAM4 or 4 lasers of 100G PAM4. In the 4-laser design, a “gearbox” is added to convert the PAM4 electrical signal from 8x50G to 4x100G.

The QSFP-DD is defined by the QSFP-DD MSA while the OSFP is defined by the OSFP MSA. They are similar but have three key differences:

  1. OSFP allows more power (15W) than QSFP-DD (12W) so that the OSFP allows an early adoption because it’s easier to release a technology designed for 15W than 12W.
  2. QSFP-DD port is backward compatible with QSFP including 40G QSFP+, 100G QSFP28, and 200G QSFP56, while OSFP port requires a QSFP to OSFP adapter.
  3. OSFP integrates thermal management directly into the form factor, but QSFP-DD does not.

400G form-factors: QSFP56-DD vs. OSFP

400G form-factors: QSFP56-DD vs. OSFP

Both QSFP-DD and OSFP are designed for intra-DC applications including DAC, AOC and optical connection up to 2km. Additional variants will come for other applications such as Data Center Interconnect (DCI) with longer reach and other technology like DWDM super channel.

Transmission

The CFP8 form-factor, defined by the CFP MSA, is radically different compared to QSFP-DD and OSFP as:

  1. It allows up 24W power consumption.
  2. It has 16x25G NRZ on the electrical side instead of 8x50G PAM4 for QSFP-DD and OSFP.
  3. It has an MDIO management interface instead of I2C for QSFP-DD and OSFP.

With its large space and max 24W power consumption, the CFP8 is intended for transmission application. Available in an initial version of 10km, it has 16 electrical lanes of 25G NRZ which are converted to 8 lanes of 50G PAM4.

However, other variants are coming for longer reach, including Coherent detection technology. A version called CFP8 ZR (80km) will come at a later stage but it also opens the door for a CFP8 800G! By using the 16 electrical lanes and apply a 50Gbps PAM4 signal, it is feasible to reach 800G; then adding a DSP, Coherent detection and multiplexing lasers will enable the optical transmission. Clearly, this is not for today yet.

400G form-factors: QSFP56-DD vs. OSFP vs. CFP8

400G form-factors: QSFP56-DD vs. OSFP vs. CFP8

For 400G applications, others form-factors than the one listed above are also available, but for dedicated applications. We can list the COBO (Consortium for On-Board Optics) and the CDFP for cable application enabling 16 electrical lanes of 25Gbps.

Originally published at QSFP-DD, OSFP, and CFP8: Which Is the Best for 400G?.

The Technologies of Next-Generation Optical Transceivers — PAM4 and 64QAM

PAM4 and 64QAM Technologies

PAM4 and 64QAM Technologies

The shift to cloud services and virtualized networks has put the data center in the middle of our world and meant that connectivity within data centers and between data centers has a huge impact on the delivery of business and personal services. Hyperscale data centers are being installed across the world and these all need connecting. To meet this demand, optical transceiver suppliers are delivering new solutions based on PAM4 and 64QAM, providing coherent modulation that will drive down the cost of connectivity and increase the bandwidth of each connection.

Connections to many servers are already 25G and links between switches in large data centers are already 100G. The introduction of SFP28 and QSFP28 transceivers integrating new technologies and built using efficient manufacturing techniques has driven down the cost of these connections and allowed massive growth in the market. The next stage is the introduction of 100G single lambda solutions and cost-effective 400G transceivers for links between switches. The PHY devices needed for this next step are already becoming available, 12.8T switch devices are in production, and the first 400G QSFP-DD and OSFP optical transceivers are sampling.

QSFP-DD

The rise of the hyperscale data center operator has dramatically changed the market. The switch to 25G and 100G from 10G and 40G has happened very quickly. The sheer scale and numbers of data centers being installed or upgraded means that the new technologies can be shipped in volume as soon as the price is right, the components have been qualified, and the production lines are operational. We are now seeing the first 400G PHY devices and optical transceivers for data centers becoming available and companies are vying for market position as we wait for the leading hyperscale operators to commit to large deployments.

Many of those companies that have benefited from 25G and 100G are putting their investments into single lambda PAM4 100G and 400G solutions for the data center. This has required new PAM4 PHY devices designed to meet the power constraints of 400G OSFP and QSFP-DD transceivers. A few companies have also invested in 50G and 200G PAM4 PHYs, enabling a cost-effective upgrade from 25G and 100G. 50G SFP56 and 200G QSFP56 transceivers are expected to be interim solutions, but it is unclear how widespread their use will be or for how long. 40G was an interim solution that lasted for many years.

Coherent technology, originally developed for 100G long-haul networks, is now widely used for long-haul connections, including subsea, metro networks, and Data Center Interconnect (DCI) between data centers. The market for DCI has grown rapidly, with many systems vendors offering solutions with 80km to 500km reach. For long-haul and metro applications, several leading equipment manufacturers continue to use in-house coherent Digital Signal Processor (DSP) designs. Coherent DSP solution is now available to optical transceiver vendors such as Gigalight that is going to ship 400G transceivers based on this design. The latest DSP ASICs are enabling 600G (64Gbaud 64QAM) solutions and CFP2-DCO transceivers. The next step is the introduction of the 7nm DSPs that will enable the cost-effective 400G ZR transceivers planned for 400G links up to 100km starting in 2020.

This continues to be a market in flux. Lumentum has completed the acquisition of Oclaro, Cisco has completed the acquisition of Luxtera, and several Chinese optical transceiver vendors have joined the charge to 400G in the data center. The PAM4 PHY devices required for 100G single lambda and 400G in the data center are proving to be very challenging to deliver. PAM4 PHY solutions in 28nm and 14/16nm technology have been sampling for more than six months and these are now being joined by 7nm solutions.

Related articles: PAM4 — The High-Speed Signal Interconnection Technology of Next-Generation Data Center