Semtech has outlined power, reach and loss targets for 200G linear pluggable optics, setting out its current assumptions for data centre links.
According to the chip supplier, 200G LPO modules are targeting power consumption of about 10 watts per module, compared with 23 to 25 watts for retimed digital signal processor modules and about 16 watts for retimer-based linear receive optics. Next-generation retimed modules could fall to about 20 watts, while RTLR designs could decline to 14 to 15 watts.
Those numbers matter at switch level because cooling demand tends to track module power. A 512-port switch based on the Tomahawk-6 generation would save about 500W at module level by moving from retimed modules to LPO, rising to nearly 1kW when cooling is included.
At 1,024 ports, that saving would double. The figures underline energy use as a central factor in optical interconnect decisions as operators build larger clusters for AI and other data-intensive workloads.
Power focus
Linear pluggable optics remove the digital signal processor used in traditional retimed optical modules and instead rely on a linear electrical path with tighter constraints on the host system. That shifts part of the engineering burden from the module to the switch platform and board design.
Semtech framed the trade-off plainly: lower module power comes with narrower operating margins. For operators weighing deployment, the issue is not just the wattage of a single pluggable, but whether the full switching platform can stay within the electrical and optical limits LPO requires.
The current target for 200G LPO is 500 metres, corresponding to a 3 dB optical loss budget. In Semtech's view, that would cover most practical data centre link lengths.
Longer distances are possible, but only under specific conditions. Reach could extend to about 2 kilometres in a low-dispersion DR context if receiver sensitivity and transmit power are sufficient, giving an optical budget of roughly 4 dB.
That means the constraint is not simply fibre length. Beyond the 500-metre target, the limiting factor becomes optical power budget, which depends on transmitter output and receiver performance rather than distance alone.
Board limits
Semtech also set out an electrical loss budget of 22dB die-to-die for LPO ports. This refers to the electrical loss between the switch chip and the module die across both transmit and receive sides.
The company is working to extend that tolerance to potentially 26dB once measurement methods confirm that level is practical. For now, however, the design target remains 22dB.
That compares with up to 32dB for fully retimed modules. For switch makers, the gap matters because it leaves less headroom in printed circuit board layouts, connector choices and channel design.
In practice, platforms designed with generous margins for retimed optics may not automatically qualify for LPO modules. Engineers would need to assess whether the electrical channel between the switching silicon and the pluggable can stay within the tighter budget.
Calibration step
Host port calibration is required to make LPO work reliably, according to Semtech. The aim is to ensure the host port presents the correct electrical eye at the module input, with equalisation matched to the continuous-time linear equaliser in the module driver.
The calibration process starts with inserting a host compliance board into the switch and setting the required CTLE target for the module in use. Engineers then apply an equaliser with constrained taps and no precursors, measure output swing, verify eye shape and confirm that channel loss is reflected correctly in the equalisation settings.
Once that work is completed for a given switch platform and module type, the settings can be applied in software when a module is inserted. That would allow plug-and-play operation without manual tuning for each port or module.
The broader message for hardware designers is that LPO changes where complexity sits. Instead of absorbing signal impairments inside a retimed optical module, the host platform must be designed and calibrated with greater precision.
That trade-off has become more prominent as data centre operators seek to cut energy consumption while maintaining high port density. With optical links accounting for a growing share of switch power in dense AI clusters, lower-power pluggables have become more attractive even if deployment requires stricter board and channel engineering.
For designers considering 200G LPO, the current headline figures are a 10W power target, 500-metre reach within a 3 dB optical budget and a 22dB die-to-die electrical loss ceiling.