What Linear Pluggable Optics Means for AI Data Centers

Summary

Semtech explains linear pluggable optics (LPO) as an active optical transceiver architecture that removes the DSP retimer from the module and shifts most equalization to the host switch SerDes. A conventional retimed module uses DSPs to retime, reshape, and re-amplify degraded electrical signals, consuming about 23 to 25 watts in current 1.6T DR-8 deployments. LPO instead combines analog components such as a CTLE, driver, modulator, TIA, and potentially a linear equalizer, without DSP retiming or clock recovery, and is intended to support 200G AI data center interconnects. For 200G links, LPO modules target about 10 watts, compared with roughly 16 watts for RTLR or LRO and 23 to 25 watts for current retimed modules; next-generation DSP modules are expected to reach around 20 watts. At 512 ports, Semtech estimates about 500 watts of module-level savings and up to 1 kilowatt at the system level after cooling, with savings doubling at 1,024 ports. LPO is active rather than passive and requires host-SerDes calibration. It may be less suitable for high-loss channels, strong reflections, or systems needing maximum TXFIR flexibility. The OIF CEI 224 specification, IEEE 200G requirements, and the LPO MSA are developing the relevant standards, while calibration and standards work remain active but mature enough for deployment planning.