400G and 800G Optical Module PCB: Loss, Vias and Heat

An optical module is one of the densest mixed signal assemblies in production. A pluggable 400G or 800G module contains a digital signal processor, several analogue front ends, a driver for each optical lane, a microcontroller and a thermal path that has to remove tens of watts through a metal case the size of a matchbox.

The board inside is small, and every millimetre of it is contested. The design rules that follow from that density are the subject of this article: channel loss, via transitions, the host connector interface, power integrity and the thermal path.

What the Module Demands

At 400G the module carries eight lanes at 50 Gb/s, and at 800G the same package carries eight lanes at 100 Gb/s. The electrical channel from the host connector to the optical device has to meet an insertion loss and return loss mask over that bandwidth, and the losses accumulate quickly in a board a few centimetres long.

The density of the routing is the second constraint. The signals arrive in a narrow connector footprint, they have to reach a digital signal processor in the centre of the board and then travel out to the optical components, and all of that happens in a few square centimetres with a power plane and a ground structure in between.

Channel Loss and Material Choice

Standard FR4 is rarely adequate for the host side of a 400G channel, so the module boards use low loss materials with a dissipation factor well below 0.005. The material choice is driven by the total loss budget, which includes the connector, the board, the package and the optical device.

Conductor loss matters as much as dielectric loss at these frequencies. A smooth copper foil with a low profile treatment reduces the attenuation, and the stackup is usually designed so that the host side signals run on the layers closest to the surface, where a thick dielectric and a wide trace minimise loss.

Via Stub and Layer Transitions

Every layer transition is a source of reflection, and the stub left by a through via is the most damaging. A signal at 50 Gb/s sees a stub of a few tens of mils as a resonant element, so the board uses back drilling, blind vias or a thin panel to reduce the residual stub below the target.

The number of transitions matters as much as their quality. Each one adds a small reflection, so the routing is planned to change layers as few times as possible, and where a transition is unavoidable, the pad geometry is tuned with the field solver rather than copied from a previous design.

Connector and Host Interface

The host connector is part of the module’s electrical specification, so the launch from the connector into the board has to be designed with the connector vendor’s model rather than with a generic footprint. Pad size, ground via placement and the layer on which the signal leaves the connector all affect the impedance through the transition.

The connector is also a mechanical element. It is pressed into the board or soldered to it, it takes the insertion force of a plug, and it has to hold its position through the thermal cycles of the module. The mounting features and the pad geometry are therefore designed together with the signal launch.

Power Integrity and Sequencing

The digital signal processor in a 800G module draws tens of amperes at a low voltage, and the current changes quickly when the lanes are enabled. The power distribution network has to hold the voltage within a narrow window across the frequency range from the regulator to the die.

Decoupling is placed under the processor, with the smallest capacitors closest to the die and the bulk capacitance near the regulator. The plane structure in a small board is limited, so the ground plane is kept continuous and the power distribution is designed as a plane pair rather than as routed traces wherever the stack permits.

Thermal Path in a Small Package

Heat leaves the module through the case, and the case is cooled by the host cage and by the airflow through the system. The board therefore has to move heat from the processor and the optical drivers into the case, usually through a thick copper area, thermal vias and a gap filler or a metal block.

The optical components add a constraint, because their temperature affects wavelength and output power. The layout separates the hot digital devices from the optical devices as far as the density allows, and the thermal simulation is used to confirm that the optical section stays within its window across the operating range.

Assembly and Component Constraints

The board is densely populated with small packages, and the assembly process is a mixed one: fine pitch surface mount components on one side, an optical sub assembly on the other, and a metal case that is attached after electrical test. Warpage is a real risk at this density, and the stackup is designed for symmetry to control it.

Stencil design, paste volume and reflow profile all have to be matched to the component mix. A module that works in the laboratory can still lose yield in production because of a paste deposit that is marginal on a 0.35 mm pitch part, so the assembly process is qualified together with the design rather than afterwards.

Test and Verification

Production test on an optical module board is normally functional at the module level, because probing the board alone does not capture the interaction with the connector and the optics. Where the board is tested separately, the test access is designed in as pads beside the processor and the analogue front ends.

Verification includes eye diagrams at the host interface, bit error rate over temperature, and the loss and return loss of the channel against the specification. Those measurements are made with the module assembled, because the case and the connector both change the result.

The routing discipline that follows from all of this is the reason an optical module PCB is treated as a transmission system rather than as a circuit board. Every net on the high speed side is designed from the loss budget backwards: the material is chosen to meet the attenuation, the layer transitions are counted and minimised, and the connector launch is modelled before the footprint is placed. The result is a board where the layout tool is used to confirm a calculation rather than to discover one.

Design Checklist

Confirm the loss budget across the whole channel before choosing the material, plan the layer transitions to be as few as possible and specify the stub target for each, model the connector launch with the vendor data, and design the power distribution and the thermal path as one problem rather than two.

gopcb builds high speed boards with low loss laminates, back drilled vias, controlled impedance coupons and the panel level inspection that optical module assembly requires.

FAQ

Which material is used for 800G module boards? A low loss laminate with a dissipation factor below 0.005, often in a mixed stackup with a mid-loss material on the less critical layers, chosen from the loss budget rather than from a preferred brand.

Is back drilling necessary at these rates? On a thick panel with through vias, yes. On a thin module board the stub may already be short enough that blind vias or a thinner stack meet the target at lower cost.

How is the module kept cool? Through the case, using a copper thermal pad on the board, thermal vias and a filler material between the board and the metal housing, with the optical section thermally separated from the processor.

Related reading: high frequency laminate selection, back drilling design, impedance discontinuity analysis, and power plane design.