New OSFP-XD form factor could challenge co-packaged optics
posted on
Jun 12, 2021 05:37PM
Such modules could supply sufficient connectivity to support the requirements of a 51.2T switch chip in a 1RU box via 32 modules. The creation of 3.2-Tbps versions of the OSFP-XD would accommodate 102.4T switches in the same form factor.
During an OFC 2021 workshop Sunday, June 6, Arista Networks Founder, Chief Development Officer, and Chairman Andy Bechtolsheim introduced the OSFP-XD (for “extra dense”) form factor. The form factor will leverage twice the number of electrical channels of the current 800G OSFP while maintaining the same module size. The increase in the number of channels would enable the OSFP-XD to support 1.6 Tbps with current 100G-per-lane SerDes technology and a power efficiency commensurate with expectations expected to arise with 51.2T switch chips, as well as 3.2 Tbps for the needs of 104.4T switching devices with the advent of 200G SerDes. And it will do so while offering a faceplate density similar to that which co-packaged optics promises.
The OSFP-XD specifications will feature a power envelope of 33 W and a (2x8) 16-lane electrical interface, with optical interfaces that vary by module scheme. Bechtolsheim described five 1.6T variants during his talk as part of the “Which Device Technologies Will Get Us Beyond 400G?” workshop. Two, the 1.6T-ZR2 (for 100 km at 30 W via fixed-wavelength lasers) and 1.6T-LR2 (for 10 km at 28 W), will leverage dual-channel 16QAM coherent transmission technology with dual LC connectors. The other three will be based on 200G PAM4 and offer 1 to 2 km of reach. The 1.6T-CWDM8 will feature an LC connector, the 1.6T-2FR4 will feature dual LCs, and the 1.6T-DR8 an octal SN/MDC connector. The power consumption for this last module is targeted at 21 W, while the other two have a 22-W target.
Such modules could supply sufficient connectivity to support the requirements of a 51.2T switch chip in a 1RU box via 32 modules. The creation of 3.2-Tbps versions of the OSFP-XD would either supply further faceplate density for 51.2T switches or handle the needs of 102.4T switches in the same 1RU form factor. Such pluggable modules would thus offer an alternative to co-packaged optics, Bechtolsheim argued at this workshop and on other panels during the week, while obviating some of the concerns and uncertainty about the creation of an entirely new technology. Speakers from Google during the week and in other forums (including a speaker who immediately followed Bechtolsheim on Sunday) have expressed doubts about whether co-packaged optics technology can be perfected in time for the arrival of 51.2T switches, so at least one hyperscaler may agree with Bechtolsheim’s viewpoint.