Once an optical engine sits on a package, the last few centimetres to the board edge become the expensive part of the link. Two designs compete for those centimetres: a lithographically defined waveguide printed into the board, or a bundle of short fibers terminated at the package. The first removes connectors and manual assembly; the second keeps a replaceable, well-characterised path. Neither is obviously cheaper once yield is counted.
With embedded polymer waveguides on the board, light is guided in a polymer structure fabricated alongside the copper layers, so the route is fixed by the board artwork and the connection to the engine is a passive alignment at assembly. With fiber pigtails to the package edge, each channel is a discrete fiber that must be stripped, aligned, attached and routed with a bend radius, and every one of those operations is a yield event. The attraction of the printed route is that one alignment step replaces dozens of individual ones, which is exactly the kind of arithmetic that improves as lane counts climb.
One passive alignment replaces dozens of individual terminations, but it also turns a bad lane from a replaceable part into a scrapped board.
Part One: What is actually being compared
The two options are often described as competing implementations of the same thing, which is misleading. embedded polymer waveguides on the board and fiber pigtails to the package edge make different promises about where the variability sits: one puts it in the capital plan, the other in the operating tail. A comparison that only measures the mean therefore misses the entire reason a team would choose one over the other in board-level optical interconnect.
There is a quiet accounting asymmetry here. embedded polymer waveguides on the board moves spending into capital and spreads it over the life of the installation; fiber pigtails to the package edge keeps spending in operations where it is easy to approve and hard to cap. Finance teams prefer the second, engineering teams prefer the first, and the argument is usually settled by the budgeting process rather than by measurement in board-level optical interconnect.
Part Two: The arithmetic that decides it
The loss budget tells the other half of the story. A polymer guide might cost on the order of a fraction of a decibel per centimetre plus coupling loss at each interface, while a short fiber path is far lower per unit length but pays a connector or splice penalty at both ends. At a handful of centimetres the two are comparable; at twenty centimetres the polymer route can consume several extra decibels and leave no margin for temperature, ageing or a slightly misaligned engine. Because the whole link budget is usually only single-digit decibels, a difference of two or three is not a detail, it is the difference between a working product and a respin.
Deployment also changes the measurement. After embedded polymer waveguides on the board goes in, the metric people optimise shifts from throughput to predictability, and after fiber pigtails to the package edge it shifts from predictability to unit price. That shift is why two teams can look at the same production data and reach opposite conclusions about which option won.
Part Three: Where the simple model breaks
Repairability is the underrated term. A failed fiber pigtail can often be re-terminated or replaced; a failed embedded guide is inside a laminate that also carries power and high-speed copper, so the board is scrap. That changes the cost of a defect from a component to an assembly, and it changes where in the manufacturing flow the test has to happen. It also changes the thermal story: polymer index and dimensions move with temperature, so a guide that passes at room temperature can drift in a hot rack, while silica fiber is comparatively indifferent.
In a link budget of single-digit decibels, two or three extra decibels of polymer loss across temperature is not a detail but a respin.
Part Four: The cost that arrives later
None of this makes the printed route a bad idea. It makes it a scale idea. The tooling, the alignment fixtures and the process characterisation are fixed costs that only amortise over high volume and stable board designs. Products with short runs, frequent revisions, or a wide mix of lane counts are better served by a discrete, reconfigurable path. The crossover is not a technology verdict; it is a volume and revision-rate verdict, which is why two companies looking at the same optics can rationally choose opposite sides.
Adoption will not be a single flip. Early movers tend to be the cases with the most extreme value of the deciding variable, then the middle of the market follows once tooling and reference designs exist, and the long tail stays with fiber pigtails to the package edge until the cost of switching falls below the cost of an exception. A reasonable reading of the next few years is coexistence with a slowly moving boundary rather than a replacement event.
Part Five: What this changes outside the engineering team
For the technician in a repair depot the difference is stark. One design arrives as a sealed board where an optical fault means a full replacement, the other as a board with a visible, reseatable harness. That affects spare-parts inventory, mean time to repair and the training needed on site. Those costs land on the operator, not on the designer, and they are usually discovered a year after deployment when the first field failures come back.
A fair evaluation totals connector count, worst-case loss across temperature, assembly yield, board scrap cost and expected design revisions over the product life. Lane count alone will point to the printed route; the full bill often points back to discrete fiber for anything but a stable high-volume platform.
A market sizing sanity check
If co-packaged optics moves from a few flagship switches into mainstream AI clusters, the board-level optical routing and assembly segment could plausibly reach $1B to $3.5B annually by 2031, including waveguide materials, alignment tooling and pigtail assembly. The lower bound assumes discrete fiber remains dominant and printed guides stay niche; the upper bound assumes lane counts per package rise enough that manual termination becomes the binding constraint. The swing factor is assembly yield, not raw optical performance, because a single bad lane scraps an expensive package.
All figures here are an author's synthesis of published ranges and a stated model, not a measurement of a shipped system. The point of the arithmetic is to locate the crossover and expose which assumption drives it; a vendor quote will differ, and the direction of that difference is more informative than the final number.
The comparison that matters
The comparison that matters is between connectors eliminated and margin surrendered. Printed waveguides win at high lane count, high volume and stable artwork; fiber pigtails win when the loss budget is tight, the design churns, or a fault must be repairable in the field. The package boundary is where the decision should be made, not the link diagram.
A change in regulation or in the supply of a single component could invalidate the ranking overnight. The ranking is a snapshot of prices and constraints in board-level optical interconnect, not a property of the technologies, and anyone reusing it in a different year should re-run the threshold rather than reuse the conclusion.
The crossover is a volume and revision-rate verdict, which is why the same optical engine can rationally ship with two different last-centimetre designs.
SOL: HCqPmPJM3NVtS6YrmVaTZYnuP8XZ2PFgQZKT94VK1LkS
Creating this takes real effort — thank you for the support.
Top comments (0)