From September 11 to 13, the 25th China International Optoelectronic Exposition was held in Shenzhen. During the "New Optical Transport Technologies Forum for the Era of Computing Power" held concurrently, Li Han, Chief Expert at China Mobile and Director of the Basic Network Technology Research Institute at China Mobile Research Institute, stated that compared to 800G/1.2T, 1.6T is poised to become the next major generational technology following 400G.
Li Han pointed out that high-speed optical communication is a crucial foundation for enhancing the computing efficiency of intelligent computing centers. For distributed training among intelligent computing centers, differentiated strategies need to be employed for different parallel algorithms. For intra-data center interconnection, it is worth considering the introduction of Optical Circuit Switching (OCS) and Multi-Tier Network (MTN) to support multi-level low-latency slicing, meeting the hard isolation needs of multiple tenants.
Single-Wavelength 1.6T is the Next Major Competitive Focus in Optical Transport Networks
Li Han introduced that, in response to the new demands of computing power networks on optical networks, China Mobile has constructed a new flexible and efficient all-optical network technology architecture based on 400G high-speed interconnection. This architecture promotes ultra-large bandwidth, flexible scheduling, optical carrier intelligent computing, and cutting-edge technological innovation, aiming to create a highland for technological innovation in the integration of computing and networking.
As of now, China Mobile has completed the 400G interconnection network construction project for the eight major hubs of the national "Eastern Data, Western Computing" initiative, covering 135 cities. The network is built using a dual-plane concept to provide high-quality, end-to-end computing power network services. The inter-hub network meets the latency requirements of 1-5-20 milliseconds, operating stably and with no difference compared to the 100G era.
Li Han pointed out that 400G is expanding further from inter-provincial backbone networks to intra-provincial backbone networks and metropolitan area networks. There are three potential technical routes: QPSK, 16QAM-PCS, and 16QAM, which should be considered in combination with factors such as transmission capacity, frequency efficiency, and deployment costs. Currently, QPSK and 16QAM-PCS coexist in intra-provincial backbone 400G networks. Provinces with large networking scopes can use QPSK technology to enhance transmission capacity, while those with lower transmission distance requirements can use PCS technology for higher spectrum efficiency and lower costs. In metropolitan 400G networks, the main competition is between 16QAM and 16QAM-PCS, which has now mostly converged to the single 16QAM-PCS code type. However, whether to introduce the C+L band still requires further study.
Li Han believes that single-wavelength 1.6T will be the next major generational competitive focus in optical transport networks over the next decade. Compared to 800G/1.2T, 1.6T is expected to become the next major generational technology following 400G. The ITU-T has already initiated the research and development of the B1T electrical layer standard, starting with the G.709.b1t project, which involves frame structure, overhead, rate, mapping, and FEC, among other aspects. The main technical direction of the 1.6T electrical layer includes optimization for mainstream Ethernet service carrying and the use of large time slots, discarding small granularity to simplify multiplexing levels.
"The single-wavelength 1.6T optical layer faces significant technical challenges in transmission distance, transmission capacity, and transmission medium. It requires collaborative efforts from industry, academia, and research institutions to tackle high-speed optical chips and wide-spectrum optical devices, and to clarify the technological evolution route and strategy for optical fiber infrastructure to meet the demand for 1.6T backbone transmission in existing networks," Li Han said.
Optical Communication is the Crucial Foundation for the Scaling of Intelligent Computing Centers
The demand for intelligent computing (AI computing) is rapidly growing, with scale increasing sharply. High-speed optical communication technology is foundational for enhancing computing efficiency within intelligent computing centers and in multi-site interconnection scenarios.
Li Han introduced that long-distance interconnection will lead to latency degradation and increased interconnection bandwidth. For different large model parallel strategies such as data parallelism, model parallelism, and mixture of experts parallelism, differentiated strategies should be adopted.
Meanwhile, with the rapid development of intelligent computing AI, multi-tenant separation within the industry is an important direction for the development of data centers. Data centers serving multiple tenants can improve DC utilization and reduce costs, but at the same time, they must protect the security and privacy of tenant data and ensure that the performance of each tenant's business is not affected by others.
Li Han suggested that to address the differentiated needs of industry multi-tenancy, Optical Circuit Switching (OCS) and Multi-Tenant Network (MTN) technologies could be considered to support multi-level low-latency slicing. OCS can support optical slicing to achieve physical isolation, while MTN can support electrical slicing to achieve fine-grained, flexible dynamic isolation. To meet the flexible configuration and green development needs of intelligent computing, China Mobile is expediting research on the efficient collaboration of OCS in intelligent computing centers, pushing for OCS performance enhancements.
Li Han also pointed out that compared to solid-core optical fibers, which achieve total internal reflection through material doping, hollow-core optical fibers utilize a novel air-guiding mechanism, substantially reducing nonlinear effects and transmission latency by over 30%. This can break through the two major physical bottlenecks of solid-core optical fibers: the "nonlinear Shannon capacity limit" and the "transmission latency limit".
As system rates increase, the transmission distance supported by 850nm VCSEL-based multimode fibers continues to shorten, compressing their application space within data centers. The application scenarios have reduced from 500 meters to as low as 30 meters, making it difficult to meet the interconnection needs of Spine-Leaf switches in the 400G era.
To maximize the technical potential of VCSEL, China Mobile proposed a VCSEL solution based on hollow-core optical fibers. At the 850nm wavelength, hollow-core optical fibers have significant advantages over solid-core multimode fibers in terms of transmission loss, chromatic dispersion, and mode purity. This could enable single-lane 200G signal transmission over distances of ≥2km using VCSEL + hollow-core optical fibers, meeting the inter-building interconnection requirements of intelligent computing centers and significantly broadening the application scenarios of the VCSEL technology route.
In conclusion, Li Han summarized that the combination of VCSEL optical modules and anti-resonant hollow-core fibers could potentially extend the reach of B400G short-distance optical modules from 30 meters to kilometer levels. However, technical challenges remain in VCSEL single-mode characteristics, hollow-core fiber design, and coupling aspects. "In the future, VCSEL + hollow-core fiber is expected to provide higher performance and lower-cost solutions for intelligent computing centers."