The 400G era (as the 400 Gb / s transfer rate is called) within data centers started in 2017 and has not stopped growing in options since then. Although it was commercially available some time before, the beginning of this new era was marked by the approval of the 400GBASE-SR16 interface by the IEEE 802.3 Ethernet group. At that time, the available base channel was 25 Gb / s and multiplying by 16 resulted in the desired 400 Gb / s, over a distance of up to 100 meters, using a 32-fiber MPO connector - something quite complicated to manufacture, test and maintain.

In March 2020, the 400GBASE-SR4.2 interface was approved, with several new features. One is the use of a 50 Gb / s base channel in PAM4 (Pulse-Amplitude Modulation 4 level) and, for the first time in a pattern, more than one wavelength was inserted in multimode optical fiber, taking advantage of the newly -developed OM5 to reach 150 meters. Efforts to expand the range of 400G continue in working groups that develop short-distance and low-cost interfaces for connecting servers, even those for distances of 80 kilometers, which will enable the connection of data centers for synchronization of content and high demand services such as video, games and conferences.

But if 400G interfaces appear to be technically mature and diverse and, at the same time, are expected to experience an increase in market demand in the coming years, what can come next?

There are several initiatives taking place in parallel, on several fronts. The most popular option seems to be 800 Gb / s (800G), but there are considerations about 1.6 Tb / s. This discussion, which was already taking place, was enhanced by the publication of the Bandwidth Assessment Report II (BWA) by the IEEE, in April 2020, which brings together the growth trends in demand in various applications, based on the rate of 400 Gb / s compared to 800 Gb / s and 1.6 Tb / s.

The increase in speed, in general, happens through a multiplication of "dimensions" that are directly impacted by the factors of cost and technical feasibility. Among the challenges to reach speeds above 400G, the following stand out:

Base channel of 100 Gb / s or greater: by mid 2021, the basic electrical channel of 100 Gb / s and its multiplications of 200 Gb / s and 400 Gb / s should be standardized. This may be the first option for an 800 Gb / s interface, by parallelizing the speed in 8 channels. This can be done using more optical fibers - in this case, it would be 16 - or even more than one wavelength per fiber, thus taking more channels - as in 400GBASE-SR4.2. But a very interesting option is the development of a 200 Gb / s base channel. In this way, 800 Gb / s would be achieved by multiplying the channel by 4, reducing complexity at the level of the transmission medium. In this case, the complexity would be taken to signaling (it is possible to continue with PAM4 or larger orders) and error correction with consequences for processing. The challenge is to seek a balance between technical and economic feasibility for a new base channel.

Transmission and connectivity medium: Another balance that must be balanced and that depends directly on the base channel. A 200 Gb / s base channel requires half the medium required by the 100 Gb / s channel. This is the main change between the 400GBASE-SR16 (16x25 Gb / s) and 400GBASE-SR8 (8x50 Gb / s) interfaces. Another example that illustrates how dimensions are used is in the difference between the 400GBASE-SR8 and 400GBASE-SR4.2 interfaces. Both use the 50 Gb / s base channel and multiply by 8 to reach 400 Gb / s. The difference is that the former uses one channel per fiber pair, which requires 16 fibers, while 400GBASE-SR4.2 places 2 channels per fiber pair at different wavelengths, thus requiring only 8 fibers. Connectivity is simplified in the second option, as it uses an 8-fiber MPO connector (the first, 16-fiber). On the other hand, it is more complex to manufacture a transceiver that uses two wavelengths in each fiber. This same consideration can be made to reach 800 Gb / s, for example. Assuming two base channel options and two wavelength multiplexing options, we would eventually have four methods of connectivity and use of the physical medium.

The format of the transceiver: the development of new transceivers has sought to anticipate the needs of the industry, facing the challenge of accommodating the greater power and thermal dissipation, required by processing at higher speeds, and the need to occupy little space, since the density of connections is increasing. The new formats that are natural candidates to accommodate the next interfaces are OSFP (Octal Small Form-Factor Pluggable) and QSFP-DD (Quad Small Form-factor Pluggable Double Density). Both were designed with 8 channels for direct application of 400 Gb / s (8X50 Gb / s), but have already been adopted by MSAs (Multi-Source Agreements) for 800G.
As an alternative to the "pluggable" formats of tranceivers, direct optical connection methods are being developed on the communication boards, as proposed by the COBO - Consortium for On-Board Optics initiative. In this case, the optical connection is made directly on the board, next to the chipset, and should save a lot of space in front of the switches.

Some compatibility and specification agreements between manufacturers (MSAs) and organizations have already taken the lead and intend to invest in the development of new speeds:

In addition to market initiatives, within the normative scope, the NEA (New Ethernet Applications) group of IEEE 802.3 - the same that led the edition of the BWA Report - is considering a new standard proposal with no definition of speed, but which will be the “ beyond 400G ”. Several technical challenges are under discussion, which will require a lot of work from this new study group, which can start in 2021.
As for the applications for the next speed, we can consider two broader perspectives: the use from the core of the network and the intra and inter-data center connections.

Higher speeds are always adopted at the center of the network, naturally depending on the volume of data and the bandwidth required at that level. It is expected that the connections between Spine and Leaf equipment are prepared for a future 800 Gb / s, in the same way that today they are being connected for 400 Gb / s, or even with 100 Gb / s. The more traditional hierarchical architecture must also accommodate, first, the connections between the Core Switches and the Aggregators.

This movement should "push" the speeds considered high, today in the center, further to the edge of the networks, reaching the layers towards the servers. In a short time, the servers will be connected by links of 50 Gb / s and 100 Gb / s, for example. Today it is common for speeds of 10 Gb / s and 25 Gb / s to connect the servers, often still using copper cables.

The inter-data center scenario is rapidly adopting the highest direct speeds, taking advantage of this feature to maintain communications in real time and meet the increased demand for cloud services, with the necessary synchronization. The 400G-ZR implementation, published by the Optical Internetworking Forum (OIF) in March 2020, is today one of the most discussed and considered for 80 to 120 km connections between data centers. In that same forum, what is coming next is being discussed, with the 800G-ZR being one of the options. The path is already being opened, since the OIF itself has started a study on the electrical implementation of 224 Gb / s, which should be a basis for the next leap.

The conclusion is that the speed of the networks will always increase, since the demand for more bandwidth, greater availability and more instantaneous access only grows. If the growth rate will continue to defy Moore's Law, only the future will tell. The search continues to make technology capable of delivering this ever-increasing speed at an acceptable cost. The data center remains the main actor in this scenario. It should concentrate the first investments in the next wave, when, in a few years' time, 400G will be fully operational.