PCB Stackup Design: Principles for Multilayer Boards

PCB stackup design is the least glamorous document in a hardware project and one of the most consequential. It fixes the distance between every pair of layers, decides which copper is a reference plane, and determines whether the impedance numbers in the layout tool survive contact with a real fabrication line.

What a Stackup Must Deliver

A working stackup answers four questions at once. It gives every high-speed signal a continuous signal reference plane, holds the impedance targets within tolerance, keeps the finished thickness and material combination manufacturable, and stays balanced enough that the panel does not bow after lamination.

Get those four right and the rest is detail. Get one wrong and the failure appears late, usually as an impedance deviation on a coupon or a warp measurement at final inspection, when fixing it means a new set of tooling.

Stackup Symmetry Around the Core

Lamination presses the stack at high temperature under pressure, and the materials expand and shrink as they cool. If the construction is asymmetric, with more prepreg on one side of the core than the other, the residual stress is unbalanced and the board curls. Thickness alone is not enough: dielectric type and copper distribution have to be mirrored as well.

The practical test is to fold the stackup drawing at its centre line. Dielectric thicknesses, copper weights and layer types on either side of the fold should match, or the difference should be justified by a specific electrical requirement.

Even Layer Counts and Warpage

Almost every proven stackup uses an even number of layers, and the reason is both mechanical and economic. A core is a double-sided copper-clad sheet, so a construction built from full cores naturally yields an even count. An odd count normally means adding a copper layer onto one side of a core, which requires extra processing and introduces an unbalanced build.

An odd layer requirement is therefore handled by adding a layer rather than removing one. If the signal layers are odd and the planes even, add a signal layer. If the planes are odd, add a ground plane in the middle of the stack, which costs nothing electrically and improves the return path.

Orthogonal Routing Directions

Adjacent signal layers should route in perpendicular directions. A horizontal layer between two vertical layers reduces broadside coupling, because the traces cross at right angles rather than running parallel over any meaningful distance. The benefit appears as reduced crosstalk and cleaner return paths under the traces that matter most.

The rule is a default, not a law. A dense area escape or a bus that must change direction will break it locally, and the correct response is to keep the parallel run short and, where possible, to place a plane between the two layers instead of routing them adjacent to each other.

Reference Planes and Return Paths

Every signal layer should be adjacent to a solid plane. That adjacency is what turns a trace into a controlled-impedance transmission line and gives the return current a defined path. When a signal layer sits between two plane layers, the design gains flexibility because both sides offer a reference, but the designer has to know which plane the return current actually uses.

Crossing a plane split is the failure mode to avoid. If the return path has to detour around a gap, the loop area grows and the trace radiates. A plane that carries a different net than the one the trace nominally references creates the same problem.

Power and Ground Pairing

A power plane placed directly against a ground plane creates distributed capacitance between them, which is useful at high frequency and costs nothing in a stackup where both layers are already required. The thinner the dielectric between them, the higher the capacitance and the more the pair contributes to power integrity.

Some designs use multiple power planes to avoid splitting a single layer into several voltage regions. Whether that is necessary depends on how many supplies the board needs and how sensitive the reference is, and splitting a plane with care is often the cheaper answer.

Mixed Dielectric and Microwave Boards

Boards that mix materials, for example a high-frequency laminate on the outer layers and standard FR-4 inside, need extra care because the two materials have different expansion coefficients and different dielectric constants. Placing the mixed dielectric near the centre of the stack minimizes the imbalance.

On microwave boards the stackup often includes a blank signal layer near the middle purely to keep the build symmetrical. That layer may never carry a net, but it preserves the mechanical and electrical balance of the panel.

Cost and Manufacturability

Cost rises with layer count, but not linearly, and not in the way most people expect. A well-planned six-layer board often costs less than a poorly planned four-layer board, because the four-layer version forced extra components, a larger outline or an expensive assembly step to work around routing congestion.

Material choice matters as much as count. Standard FR-4 covers most digital products, while high Tg grades are worth the premium when lead-free reflow and wide thermal cycling are expected. Thinner prepreg may be cheaper but can squeeze the impedance tolerance, particularly on outer layers where plating thickness is hardest to control.

Reviewing the Stackup Before Release

Review the stackup as a document, not an afterthought. Confirm the layer functions, the reference for every signal layer, the dielectric thicknesses and the material callouts against the fabricator’s standard offering. Check that the impedance targets are achievable with the chosen construction rather than with an idealised one.

A short review against layer stackup examples from one to eight layers catches most mistakes, and the guidance on balanced stackups with an odd layer count explains how to add a layer without inflating cost. Where planes carry multiple voltages, review power plane splitting rules before releasing the design.

Documenting the Stackup for the Fabricator

The stackup drawing is a manufacturing document, so it has to state what the fabricator needs and nothing that contradicts it. Include layer numbers and functions, dielectric thickness and material for each layer, copper weights, the finished thickness with tolerance and the impedance targets with the layer and reference for each.

Two conventions prevent most misunderstandings. Number the layers from the top, and state whether the impedance values apply to the coupon or to the finished board. Ambiguity on that point produces arguments after the panels are built rather than before.

Where the fabricator has a standard construction that meets the electrical requirement, using it is usually faster and cheaper than an exotic sequence, even if it means adding a layer. Standard builds have known yields, and yield is the hidden cost in every custom stackup.

FAQ

Is a four-layer board always cheaper than six layers? Usually yes for the bare board, but not always for the product. If the extra two layers remove a connector, avoid a larger outline or simplify assembly, the six-layer option can win on total cost.

Can I build a stackup with two different prepreg thicknesses? Yes, and it is common. What matters is that thicknesses are mirrored around the centre line and that each dielectric is a standard stock item. Non-standard combinations increase lead time and price.

How much does warp actually matter? More than most teams expect. Warpage affects component placement, coplanarity for large packages and solder joint reliability, and it is one of the few stackup errors that cannot be corrected after fabrication.