Generally speaking, the current focus of the domestic optical communications industry is focused on the following directions:
1. Full implementation of 400G
2. Accelerated deployment of G.654.E optical fiber
3. The rise of LPO
4. FTTR and 50G PON
5. High-performance computing cluster network
1. Full implementation of 400G
After years of preparations upstream and downstream of the industry chain, this year, the domestic optical communications backbone network will finally usher in the full implementation of 400G.
According to the information provided by experts at the meeting, driven by the “digital data in the east and computing in the west” strategy and the vision of building a “computing power network”, operators are actively deploying all-optical transport capacity construction and carrying out 400G construction and trial operation:
China Telecom has built the first 400G all-optical transport network in the Greater Bay Area, and the ChinaNet backbone network has completed the 400GE IP+ optical long-distance transmission live network pilot.
China Mobile has built a 400G all-optical test network across the four provinces of Zhejiang, Jiangxi, Hunan, and Guizhou, and is preparing to launch relevant deployment and implementation by the end of 2023. (It is reported that the centralized procurement of 400G in the province will start in October.)
China Unicom has built 400G trial networks in Shandong, Zhejiang, Shanghai and other places.
400G high-speed interconnection is another upgrade of all-optical transport capacity. It is the result of all-optical forwarding, low latency, high-speed optical modules and other technologies. Its goal is to provide deterministic carrying and the ability to account for quality.
The current reality is that with the massive construction of data centers, the demand for backbone network bandwidth continues to increase. Overall, the inter-provincial export bandwidth will reach the level of over 100 Terabits.
In terms of latency, the basic requirements put forward by our country’s east-to-west computing strategy are: 1 millisecond within the city, 5 milliseconds from the city to the hub node, and 20 milliseconds between hub nodes.
So these all mean that it is imminent to upgrade the backbone network to 400G.
After years of exploration, the single-wavelength 400Gb/s system based on 130GBaud baud rate and QPSK modulation method has become the first choice for domestic long-distance trunk line construction.
CCSA has now completed the release of standards for metropolitan 400G and long-haul 400G, and standards for metropolitan 800G and 400G ultra-longhaul are also in the process of being compiled.
In terms of band expansion, the C6T+L6T band (a total of 12 T) has also become a consensus.
It is worth mentioning that in addition to 400G, technical research and standard construction of 800G and 1.6T are also progressing steadily. Some manufacturers have launched samples and are conducting trials.
Optical modules for 800G and above are continuously being developed in multiple standards organizations. Some standards, such as IPEC, 800G Portal and CCSA, have already been released. Most standards may be released one after another in 2024-2025.
Speed upgrading seems simple, but it involves spectrum expansion, optical device upgrades, module power consumption and volume control, integration requirements, and industrial chain reuse. It is really not as simple as imagined.
The road ahead is long and full of challenges.
2. Accelerated deployment of G.654.E optical fiber
I believe you have also recently seen China Mobile’s centralized procurement bidding announcement for G.654E optical fiber and cable products.
This purchase has accumulated 8,463 skin-length kilometers, equivalent to 1.2279 million core kilometers. Compared with the first 654E optical fiber centralized procurement in 2022 (2,134 skin-length kilometers, equivalent to 332,400 core kilometers), the scale of this centralized procurement has increased nearly 4 times!
The increase in G.654E optical fiber also pave the way for the comprehensive upgrade of the backbone network to 400G.
G.654.E optical fiber has the characteristics of ultra-low loss and low non-linearity, and has demonstrated very good performance in ultra-long-distance optical transmission. It has been unanimously recognized by the three major operators and will be used to build a comprehensive computing power network. Network backbone.
In terms of industry, G.654.E optical fiber has already achieved large-scale production capabilities and has entered the stage of engineering application.
At present, the total length of G.654.E optical fiber in China is only about 30,000 kilometers, accounting for less than 3% of the entire trunk network. In the next few years, the construction scale of G.654.E optical fiber has great potential.
In terms of performance, the loss of G.654.E optical fiber is expected to be optimized to 0.15dB/km in the future, and the transmission flatness of the entire C+L band may also be further improved. This will also be helpful for C+L band applications.
According to statistics, as of June this year, the total length of domestic optical cable networks has reached 61.96 million kilometers, and long-distance optical cable lines have exceeded 1.11 million kilometers.
As the construction of computing power networks further accelerates, 130 trunk optical cables need to be built around the computing power hub nodes.
These newly constructed new optical cable networks will further improve data transmission bandwidth and performance, which is conducive to the upgrade of network architecture.
In terms of fiber optic cables, there are two important technical directions worthy of attention.
First of all, the first one is spatial division multiplexing of multi-core few-mode optical fiber.
Spatial division multiplexing of multi-core few-mode optical fiber has become a feasible path to break through Pbps capacity.
This year, the National Key Laboratory of Optical Communication Technology and Network of China Information Technology Group achieved a single-mode 19-core optical fiber transmission system experiment with a total transmission capacity of 4.1Pb/s and a net transmission capacity of 3.61P/s.
The super optical network constructed in the Guangdong-Hong Kong-Macao Greater Bay Area has a total length of more than 160 kilometers, connecting Guangzhou and Shenzhen. It uses FiberHome’s independent space division multiplexing optical fiber and cable technology to create the longest distance and largest capacity space division multiplexing in the world. Optical communication systems.
Standardization around spatial division multiplexing is also being gradually advanced.
In TC6 of China Communications Standards Association, three research topics related to space division multiplexing have been established. In September last year, the ITU-T SG15 meeting released the “Technical Report on Space Division Multiplexing Transmission”.
In general, domestic and international standards organizations are very concerned about this area.
Another important direction is hollow-core optical fiber.
Hollow core fiber, as the name suggests, has an air or vacuum core in the center of the fiber instead of glass or other materials. It is considered a disruptive technology with the characteristics of large bandwidth, low latency and low loss, and is widely favored.
Because the entire medium changes and is transmitted in the air, the delay per kilometer is reduced by 1.54 microseconds.
In terms of ultra-low loss, the theoretical minimum loss can be less than 0.1dB/km. Currently, the one disclosed by the University of Southampton is 0.174dB/km.
Another very important feature of hollow-core optical fiber is that it has ultra-low nonlinearity.
At present, hollow-core optical fiber has attracted a lot of industry attention. Its standardization of optical cable structures and collaborative innovation with transmission systems are still in the early stages, and many institutions are participating in pre-research.
A bottleneck worthy of attention in hollow-core optical fiber is the drawing length.
Currently, solid optical fiber can stretch 10,000 kilometers. However, the limit of hollow-core optical fiber is only 10 kilometers, which is a difference of 3 orders of magnitude. This directly brings huge cost differences and affects large-scale production.
3. The rise of LPO
Last year and the beginning of this year, we were still discussing CPO/NPO. Now, LPO is here again.
As we mentioned earlier, driven by the demand for data bandwidth, optical modules have evolved from 400G to 800G and further to 1.6T.
As the speed increases, problems such as integration and power consumption of traditional pluggable optical modules will become very difficult to solve.
Previously, the industry proposed CPO and NPO. Now, LPO (Linear Pluggable Optics, linear pluggable optical module) is proposed.
LPO replaces the traditional DSP with linear-drive technology and transfers the corresponding overall compensation function to the module’s analog electrical chip and the corresponding ACK Serdes functional unit, achieving low loss, low power consumption, and low latency. , low cost and hot-swappability, etc., have relatively large advantages.
LPO maintains the module’s pluggable form. According to industry data, the power consumption of LPO is 50% lower than that of traditional pluggable optical modules, which is close to that of CPO.
After adopting the linear direct drive solution, the power consumption of silicon photonics, VCSEL, and thin film lithium niobate can be reduced by about 50%.
Low power consumption not only saves power, but also reduces the heat generated by components within the module.
After removing the DSP chip, the system reduces the time for signal recovery and the delay is greatly reduced.
DSP is relatively expensive, and the BOM cost of DSP accounts for about 20-40% of the 400G optical module. LPO’s driver and TIA integrate EQ functions, so the cost will be slightly higher than that of DSP, but the LPO solution can still reduce the cost of optical modules a lot.
Compared with CPO, LPO does not significantly change the packaging form of the optical module. It uses pluggable modules for easy maintenance and can make full use of existing mature technologies.
According to predictions, LPO will achieve mass production by the end of 2024.
Experts at the meeting also expressed different opinions on whether LPO is the optimal solution, believing that in-depth demonstration is needed through design and experiments.
LPO not only has advantages, but also disadvantages.
Because the DSP is removed, stronger SerDes is needed to compensate. And stronger SerDes means that the cost will become higher.
Previously, the most widely used optical module was based on 50G SerDes. Currently, 400G and 800G optical modules are all based on 100G SerDes, and in the future it will be 200G SerDes.
SerDes refers to the speed of the electrical part, and the speed of the optical part has also evolved accordingly. The impact of this evolution on the optical module is that the speed is constantly increasing.
LPO will also bring about interconnection issues. Not only the interconnection between switches, but also the interconnection and interoperability of traditional optical modules. This limits the application scenarios of LPO.
The technical details of LPO are still relatively complicated. Later, Xiao Zaojun will write a special article to introduce it.
By the way, encapsulation.
Traditional optical modules come in various packaging forms. With 400G, 800G, and 1.6T, this situation will change. Packaging formats are constantly converging, such as being reduced to QSFP-DD and OSFP, and related modules may be reduced to OSFP and CFP8.
The convergence of packaging formats is a good thing for industry development.
4. FTTR and 50G PON
At this meeting, another focus is on FTTR and 50G PON at the access network level.
Operators have been actively promoting FTTR in the past two years. At present, various operators have millions of users, and it is said that the number will exceed 10 million by the end of the year.
Operators also implicitly expressed that there is a certain lack of demand for FTTR for home users. Therefore, the focus of FTTR promotion has now begun to shift to a certain extent from FTTR-H (for families) to FTTR-B (for enterprises), including big B and small B (small and micro enterprises).
In terms of PON technology, the current trend is from 10G PON to 50G PON.
China will start promoting the construction of 10G PON around 2021. In less than 3 years, the entire Gigabit optical network has covered more than 500 million households and there are more than 100 million Gigabit users.
Now, what operators are actively carrying out technical verification and reserve is 50G PON. According to predictions, 2024-2025 will be the time for the launch of 50G PON. From 2027 to 2030, 50G PON will reach a certain scale.
At present, the standard formulation work of 50G PON has been basically mature. There are already many prototypes of related products, and operators have also organized trials.
From a technical perspective, 50G uplink is the most difficult and challenging. It is unrealistic for ONU to remain unchanged as before. Either integrating SOA or using high-power lasers remains to be further verified.
In addition to home scenarios, operators have begun to introduce PON technology into industry scenarios, such as industrial PON.
Industry scenarios have higher requirements for latency, so 50G PON needs to focus on improving latency capabilities. Industrial PON has certain requirements in terms of compatibility with multiple protocols in factories, remote power supply capabilities, and anti-interference capabilities. Its challenges are much more complex than those in home broadband scenarios.
In addition, it needs to be mentioned that the sinking of OTN is still progressing.
OTN point-to-multipoint quality dedicated lines can support OTN to further extend to users and further integrate OTN technology with existing ODN, transmission network, and access network. The access side passes through fixed allocation, and the transmission side passes through the hard pipes of OICO and ODO to achieve end-to-end hard isolated transmission.
5. High-performance computing cluster network
AIGC is the hottest topic this year. The optical communications industry has also been driven by the rapid development of AIGC large models and has achieved good performance.
I have written many articles about high-performance networks this year, introducing the network support technology behind the AIGC large model.
AIGC large models require a large number of GPUs to support calculations. The scale of the cluster is getting larger and larger, and the performance requirements of the cluster network are extremely high.
The bandwidth, delay, stability and reliability of the network directly affect the computing time of the GPU cluster and also determine the cost of the entire computing.
Currently, the mainstream technical routes are InfiniBand (IB) and RoCE solutions.
IB is NVIDIA’s private protocol, and the cost is too high, basically 3-5 times that of the latter. Therefore, more and more manufacturers are choosing the new Ethernet RoCE that is transformed from traditional Ethernet combined with RDMA technology.
RoCE is open source, and various manufacturers have related solutions. There are many choices and high cost performance.
Currently, the main GPU used in China is Nvidia A800 (A100 is not available). The interconnection bandwidth of A800 is 400Gbps and A100 is 600Gbps.
The interconnection bandwidth of H100 is as high as 900Gbps (H800 is 450Gbps).
Therefore, foreign countries are stepping up efforts to develop intelligent computing clusters based on 800G optical modules. We still focus on 400G, and the demand for 800G is not too strong. But continued pursuit is still necessary. In the next few years, we will just find ways to move from 400G to 800G.
From a macro perspective, RoCE provides a good opportunity for domestic manufacturers to catch up, and also provides domestic companies with options to develop AIGC large models.
Well, the above is the current progress in the key areas of focus in the domestic optical communications industry.