Problems faced by pluggable optical modules after 1.6T?
From the COBO onboard optical module to the CPO alliance, the industry has been looking for low-power, low-cost solutions. North American Internet companies are planning to deploy CPO switches in the second half of 2022. Lot of experts mentioned that the optical module will reach the bottleneck period after 1.6T, so what problems will the pluggable optical module face? For example, power consumption, cost per bit? Does it coexist with CPO?
The most important thing for CPO is to solve the problem of electricity (interference). Light is inherently orthogonal and it is easier to resist interference. Pluggable and CPO coexist for a long time and do not conflict. CPO on-board optical transmission has other application scenarios. It can be used as an inter-board connection between chip and chip, which can increase the extremely high density. In addition to exchanges, AI and supercomputers in particular will become the pioneer market for CPO and on-board optical modules.
From the perspective of equipment, CPO technology is first of all power consumption. With the same single board density, power consumption has to be increased by more than 4 times. From the perspective of network topology, some people have proposed that CPO technology can reduce TOR and reduce system power consumption; at the same time, heat dissipation The problem is that the temperature rise of 400G is already very high, and 800G will be even higher. Secondly, the future trend of switch equipment is inevitably higher and higher port density and bandwidth. Should CPO technology be the direction to ensure high-bandwidth and high-density transmission? Even if it is not used in switches for the first time, integration and high speed are the inevitable direction.
CPO and pluggability must coexist. CPO is more idealized and has a complicated process. From the perspective of North America, the players are Intel, Broadcom, and Cisco. From the perspective of silicon photonics, there are few manufacturers of real energy producing silicon optical modules, and the packaging density is at least 32 channels, 64 channels, and 128 channels. At present, it is difficult to control the cost of 400G DR4 4-channel to be accepted by the industry. It is conservatively estimated that pluggable modules will have a period of at least 5 to 6 years, and CPO should have a part of the market.
The power consumption of the switch chip is already very high. Although the CPO has low power consumption and low cost, placing the silicon optical module near such a large heat source will still be subject to temperature fluctuations, and the challenge is still great. CPO is a very good technology trend and may be the first to be applied in HPC, but in the long run, when it will enter the large-scale market, it will require many industry colleagues to work together to overcome various problems.
Challenges brought by CPO to devices, modules, and switch systems
If CPO is accepted by more Internet companies, the design, testing, and maintenance of connectors, electrical chips, optical chips, light sources, high-power lasers, and switchboards will bring great challenges.
From the perspective of connector manufacturers, different scenarios in CPO require different special connectors, such as the connection of lasers: Intel uses heterogeneous integration; most companies use external lasers (pluggable), so you need Integrate optical connector and electrical connector, consider alignment tolerance, dust is not sensitive. From the perspective of the light engine, silicon optical coupling needs to consider converting high-density optical fibers into ordinary optical fibers, and consider using beam expansion methods. From light engines to switch panels, optical fibers with thousands of cores require high-density connectors.
From the perspective of industrial layout, CPO players such as Nvidia, Intel, and Broadcom have completed many industrial layouts through mergers and acquisitions, and they have optical and electrical capabilities. In the domestic industry chain, giants without light and giants with electricity have the ability to integrate. Centec can do the electrical part of the CPO switch. It also needs optical partners, packaging partners, and a process of complete cooperation in the industry chain. Standardization and mass production are both tedious and long work of mutual cooperation.
CPO uses all the electricity and chip parts for sealing, but the light source is still external. The advantage is that the laser is more resistant to heat than the electric chip, and the Raman solution is a direction. For laser chip manufacturers, it is enough to make corresponding changes in accordance with changes in technical solutions, customer needs, and industry implementation. If you really go to CPO, then ing the core chip is the biggest source, and the core lies in whether it can be recognized by customers in the head market.
To be conclude, if CPO does come, how to embrace it from the perspective of devices, modules, and chips, share three points:
First, based on hypothetical judgments, CPO can only be integrated, considering the cost of silicon photonics integration. Is there a Fab that supports the so-called multi-material system silicon-optical integrated process platform? Such a platform is urgently needed in China.
Second, the best companies in the industry to commercialize silicon-optical integration are all full-capable, with both optical and electrical design capabilities. Therefore, domestic companies can not only focus on optical chips, but also need integrated design with electricity.
Third, there is a lack of domestic manufacturers that focus on switching chips like Centec. Broadcom and Cisco overseas have obvious advantages, from switching chips to silicon photonics to packaging capabilities in the entire industry chain. Since CPO is a photoelectric sealing, to seal the light engine and the switching chip together, it is necessary for the optical switching chip company to lead the device layer to further improve the industrial chain.