At this year's ECOC, Marvell presented a report discussing the development trend of the next generation 1.6T / 3.2T optical modules. By now, the 800G optical module scheme for data centers has been mostly clarified, and many leading optical module suppliers have already produced 800G optical module prototypes or small quantities of orders. The next step is to consider the next generation of optical modules. There are currently two paths to increasing the total conversion rate of high-speed optical modules: increasing the number of channels or increasing the conversion rate per channel, commonly known as channel generation and speed generation.
Increasing the number of channels is technically feasible without increasing bandwidth or modulation complexity. However, it presents challenges regarding limited space, accommodating more optical devices, carrying out multi-channel optical channel interface and electrical channel connection, and serious issues with density and power consumption. Increasing the single-channel rate involves using optical and electrical devices with larger bandwidth or a more complex modulation format based on the original number of channels. However, this approach is difficult due to the challenges of increasing the number of channels, power consumption, limited space, and high-density packaging. Some manufacturers have done a 1.6T optical module demonstration based on OSFP-XD for hot-swappable optical modules. To increase the single-channel rate, it is necessary to improve the analog bandwidth of optoelectronic devices, strengthen the DSP signal processing ability, employ stronger FEC, and enhance the optical integration.
Marvell believes that the technical route of the 1.6T optical module is becoming increasingly clear, and the industrial chain is gradually maturing. Three technical solutions can be realized on the basis of the 800G optical module, and industrialization capabilities are expected to be achieved in 2024. However, the technical route of 3.2T is not yet clear, and further technical optimization and exploration are needed.
Comments