800G ZR/ZR+ Coherent Pluggables: IPoDWDM Design Guide

Coherent optics used to live in a transponder shelf between the router and the line system. ZR and ZR+ pluggables collapsed that layer into a QSFP-DD or OSFP cage on the router itself, and the 800G generation is what made the change economically interesting: one pluggable now covers metro, regional and, in ZR+ modes, long-haul distances. Here is what the module actually delivers, what changes between ZR and ZR+, and the design constraints that decide whether the link works.

Why 800G ZR Is the Inflection Point

400ZR did the proof of concept: a QSFP-DD module, 80 km reach over a single fibre pair, deployed at scale through 2024 and 2025 as the standard campus and metro DCI building block. It proved that routers could own wavelengths directly without a transponder in the path. What it could not do was reach far enough, or carry enough, for the current generation of distributed AI clusters.

The 800ZR generation closes both gaps. The OIF implementation agreement for 800ZR/800ZR+ landed in November 2024 and moved quickly into production, with hyperscalers among the first large-volume buyers — Meta is a notable reference customer for 800ZR+ specifically, which is a useful signal that the reach numbers are real and not datasheet-only. Two things drive the uptick:

• Per-port economics. An 800G wavelength on a router port replaces two 400G wavelengths plus a transponder, a patch panel and rack unit. The saving compounds per port, per site, per year.
• Distributed AI clusters. Training clusters now span multiple buildings and campuses, so the inter-building links need several hundred gigabits each and they need them without a DWDM engineering team provisioning every wavelength by hand.

It is worth being clear about what 800G ZR is not. It is not a replacement for a full long-haul coherent line card in a chassis. Its value is that the vast middle of the market — metro, regional, short-haul — no longer requires one.

What the 800G Module Contains

A digital coherent optics (DCO) module integrates the coherent DSP, the optics and a tunable laser in the pluggable form factor. The current generation is built around a 140 GBaud integrated coherent transmitter-receiver optical subassembly (IC-TROSA) using indium phosphide photonics, which is how vendors reach high transmit power without an external micro-EDFA. The transmitter output power is what makes the difference in deployment:

• −7 dBm — the OIF 800ZR amplifier-free interface.
• 0 dBm — the higher-output mode intended for ROADM-based line systems, so the pluggable can drive the mux directly.

Putting the DSP inside the module rather than on the host line card is a deliberate trade. It makes the pluggable self-contained and vendor-interoperable — the router does not need to know how to run a coherent modem — but it also means the DSP wattage, the laser thermals and the tuning logic all sit inside a cage roughly the size of a matchbox. That is the origin of most of the deployment constraints discussed later: these modules are thermally dense and they need airflow or a cold plate.

The tunable laser covers the C-band on the ITU grid, which is what lets one part number drop onto any 50 GHz or 100 GHz channel without stocking channel-specific variants.

ZR versus ZR+: reach is a mode, not a product

The same pluggable runs several line modes. In ZR mode the module is interoperable and short/medium reach — roughly 500 km class in 800G operation. ZR+ modes trade rate for reach using probabilistic shaping and higher-performance DSP settings: beyond 1000 km at high-performance modes at 800G, and over 2000 km at lower rates such as 400G. Multi-rate operation (800G/600G/400G) via 16QAM, 8QAM and QPSK is the norm, which means one spare part number can serve several link budgets.

The mechanism behind the reach is worth understanding because it determines what you can promise. Probabilistic shaping changes the constellation so that inner points are used more often than outer ones, which raises the effective SNR margin at a small cost in raw bits per symbol. The DSP then spends that margin on either reach or capacity. In practice this means the mode selection is a continuum, and vendors expose several presets rather than two. A conversation that asks “what is the reach?” is unanswerable without specifying the rate, the mode, the span loss and the fibre type.

The related distinction is between high-performance modes and interoperable modes. Interoperable modes are the ones that comply with the OIF specification and will link up against a different vendor’s module. High-performance and proprietary modes reach further but generally require the same vendor at both ends. Design your core links in interoperable mode if you want a competitive supply chain, and reserve proprietary modes for the links where reach genuinely cannot be achieved otherwise.

Interop, Form Factor and the MSA Landscape

Three things must align before two 800G ZR modules will pass traffic:

1. Specification version. The OIF 800ZR and 800ZR+ implementation agreements define the interoperable modes and the amplifier-free interface. Modules claiming compliance should link in those modes across vendors; anything outside them is a vendor-specific extension.
2. Form factor and cage. QSFP-DD and OSFP are both available. They are not interchangeable, they have different thermal envelopes, and OSFP generally offers more heat dissipation room. Check what your router line card actually accepts before ordering.
3. Host support. The router must be able to configure the module’s line mode, wavelength and output power, and must surface the coherent telemetry. A module that links but reports nothing is an operational liability.

Where vendors differentiate is mostly in the DSP modem performance, the laser and the thermal design — the areas where a specification says what must work, not how well. That is why the interoperability tests that matter are contractual: insist on a lab trial against your actual line system and your actual second-source vendor before a volume order.

Link Budgeting: Where Deployments Actually Fail

Reach figures are marketing maxima measured on clean, low-loss fibre with no intermediate nodes. Real spans have splices, connectors, patch panels and ROADM pass-through loss. Budget the following explicitly:

• OSNR at the receiver, computed from the number and spacing of amplifiers, not from the fibre length. A 400 km span with six amplifiers is a harder OSNR problem than an 800 km span with three.
• Chromatic dispersion, which coherent DSP compensates electronically but only up to a limit. Long legacy spans on non-dispersion-shifted fibre can exceed it.
• Polarisation mode dispersion on older aerial plant, which the DSP equaliser also compensates, with its own ceiling.
• ROADM passband and filtering if the wavelength traverses intermediate nodes.
• Launch power per channel against the fibre’s nonlinear threshold: pushing the 0 dBm ROADM mode into a high-power amplifier is a recipe for self-phase modulation penalties.

Our optical power budget guide walks through the arithmetic, and the ITU grid implications of 50 GHz versus 100 GHz spacing are covered in our channel spacing article. If the path passes through ROADMs, also read the ROADM power equalisation guide — automatic power control interacts directly with the per-channel launch power you just budgeted.

IPoDWDM: what you gain and what you give up

Putting coherent optics on router ports removes an entire transponder layer: fewer devices, fewer patch panels, less rack space, lower power per bit, and port-by-port economics — you upgrade one wavelength at a time instead of a whole shelf.

What you give up is operational flexibility. The router port is now a wavelength, so the line system must accept the pluggable’s output power and OSNR profile, and every optical engineering task (power balancing, dispersion compensation, amplifier gain) moves to the IP team’s remit. Practical checks before you order:

• Confirm the line system supports the module class (amplified vs ROADM vs passive mux) and the required output power mode.
• Budget OSNR and dispersion for the real span, not the marketing maximum. Reach numbers assume a clean, low-loss plant.
• Check the router’s power and cooling budget per cage: coherent pluggables run hot, and a full faceplate of them changes the thermal profile of the chassis.
• Check management: these modules expose rich DDM/DOM values, but the router must surface the coherent parameters (pre-FEC BER, OSNR, chromatic dispersion, laser bias) for the NOC to act on them.
• Confirm the failure model. A pluggable failure is a router port failure, not a transponder failure, so your sparing and RMA process has to be built around optics SKUs rather than line cards.
• Agree who owns the fibre. In IPoDWDM the IP team is now responsible for span loss, connector cleanliness and patch panel hygiene — capabilities that often lived with the transport team previously.

The organisational point is the one that derails projects. Moving the optics onto the router also moves the accountability. Teams that plan for that transition explicitly — shared runbooks, a joint change process, the transport team retained as an advisory function — deploy successfully. Teams that do not end up with a fibre problem that nobody owns.

Deployment pattern that works

Start with DCI links where two routers in different buildings need N×400G and there is a fibre pair available. Terminate the pluggable directly into the router, put the mux/amplifier at the node, and monitor pre-FEC BER as your primary health metric — it degrades long before the link drops. Keep one ZR+ spare per site rather than one per module type, since a single part number covers multiple line modes.

A useful extension of that pattern for larger deployments:

# Sketch: bring up an 800G ZR+ link and capture the baseline
router# show interface FourHundredGigE0/0/0/0 transceiver detail
router# show interface FourHundredGigE0/0/0/0 optics
  Pre-FEC BER      : 1.2e-03
  Post-FEC BER     : 0.0e+00
  Rx OSNR (0.1nm)  : 24.8 dB
  Chromatic Disp.  : 412 ps/nm
  Laser Bias       : 38.5 mA
router# show interface FourHundredGigE0/0/0/0 controller

Record those numbers at commissioning. Pre-FEC BER is the leading indicator: a healthy link sits comfortably inside the FEC threshold, and a link climbing toward it will fail weeks before it actually drops. Post-FEC BER should be zero or effectively zero — if it is not, the link is running without margin and will become intermittent. Our coherent optics pre-FEC BER and OSNR checks guide covers which counters to trend and what each failure signature means.

Troubleshooting an 800G Coherent Link

The failure modes are reasonably distinct, and reading the telemetry usually identifies the culprit without touching the fibre:

• Link does not come up at all, no light received. Wrong channel, wrong direction on a duplex pair, an unseated module, or a laser that has not been enabled in the host configuration.
• Link comes up, then flaps. Usually thermal or output-power related. Check module temperature and whether the cage is in an air-cooled area that has lost airflow; also re-check launch power against the amplifier’s input range.
• Link is up but pre-FEC BER is marginal. OSNR is the first suspect, then dispersion, then the transmitter at the far end. Compare OSNR measured at both ends — asymmetric values point at an amplifier or a dirty connector on one side.
• High post-FEC BER with good OSNR. Points at a DSP mode mismatch or a module running a proprietary mode against an interoperable one.
• Wavelength conflict. Two modules on the same ITU channel will not report an obvious error; they will simply not converge. Confirm the grid assignment from the line system’s side, not the module’s.
• Clean connectors. A dirty connector is the single most common preventable cause of a marginal link. Inspect and clean before escalating, every time.

For the physical layer, our optical transceiver power thresholds guide gives the Rx/Tx dBm windows to check against, and the DWDM node itself is described in our DWDM system components guide if you need to verify the mux, EDFA and demux path the wavelength traverses.

Spares, Inventory and the Lifecycle Trap

IPoDWDM changes the shape of your spares holding. A transponder shelf failure was a single high-value replacement; the equivalent now is potentially dozens of optics SKUs across a fabric. Two rules keep this manageable:

1. Standardise ruthlessly. Where the physics allows, use one ZR+ part number across multiple link classes and use its multi-rate modes to cover the shorter links too. Fewer SKUs means smaller spares holding and simpler firmware management.
2. Track module firmware as an asset. Coherent pluggables ship with upgradeable DSP firmware, and version drift across a link is a real interoperability risk. Record the version at commissioning alongside the BER baseline.

Finally, plan the refresh. Coherent DSP generations move quickly, and a module bought today may be superseded within the life of the router chassis. Because these are pluggables, that is an advantage — you upgrade one wavelength at a time — but only if you have kept the channel plan and the port inventory clean enough to do so.

Related reading: DWDM system components explained, 400G spine / 100G leaf link consistency and optical transceiver power thresholds.

原文链接：https://www.coherent.com/news/press-releases/general-availability-of-800g-zr-zrplus-in-qsfp-dd-form-factor