Abstract:

With the growing popularity of bandwidth consuming applications such as cloud computing or online gaming, computation has moved from the distributed devices towards data centers. This puts pressure on the capacity of short-range data center optical interconnects. An interconnect includes a driver circuit, VCSEL, MMF, and a photodiode (PD). High bandwidth multimode 850 nm GaAs VCSELs are typically used in the interconnects to satisfy the high-speed requirement. The capacity of such 850 nm VCSEL-MMF links, defined as bandwidth–distance product, is limited by modal and chromatic dispersion effects. Increasing the speed lowers the achievable transmission distance; conversely, longer links can only operate at lower bitrates. Currently, 10 Gb/s VCSELs are used in data centers and they support the optical links up to 300 m. However, with the 25 Gb/s optical interconnects becoming available the transmission distance becomes limited to 100 m. The majority of the existing data center connections are below 100 m. Yet, a solution for an improved link capacity is required to support future interconnects speeds (40 Gb/s), increasing data center sizes, and already implemented longer links (100 m – 300 m). Installing parallel lanes is a popular approach. Although viable, this solution is short-term because the optical packaging of components with multiple lanes becomes increasingly complex and there is no sufficient space for the interconnect ports. Moreover, the rewiring is costly and the space in the data centers is limited.

Keywords: Problems and trends of supporting hyper-datacenters