Laminate Selection for High-Speed Designs: Performance Against Cost

The material a board is built on stops being a background choice once the data rates rise. Above roughly ten gigabits per second, a conventional laminate absorbs enough of the signal to consume the link budget on its own, and laminate selection becomes a design decision: which material removes that loss, and at what cost.

The Electrical Properties That Matter

Two properties decide whether a material is suitable, and a third decides how predictable the design will be.

The first is the loss tangent, the proportion of the field energy the dielectric absorbs on each cycle. It is the property that most directly determines how far a signal can travel before it becomes unreadable. The practical thresholds are well established: a conventional laminate with a loss tangent of around 0.02 is adequate for moderate rates; materials in the range of roughly 0.008 to 0.01 suit rates around ten gigabits per second; and the fastest links need materials below about 0.002, with the low-loss families generally below 0.005.

The second property is the dielectric constant, which sets the propagation velocity and, together with the stack-up geometry, the impedance of every controlled trace. A lower dielectric constant allows wider traces for the same impedance, which is convenient for manufacture, but the absolute value matters less than its stability.

That stability is the third consideration and it is easily overlooked. A material whose dielectric constant varies with frequency, with temperature or from batch to batch produces traces whose impedance is not what the model assumed, and the resulting variation between boards is more damaging than a consistent offset would be. Materials intended for high-rate designs are specified with a tolerance on the dielectric constant for exactly this reason.

Thermal Behaviour

The thermal requirements come from two directions: the temperatures the board sees during manufacture and the temperatures it sees in service.

Manufacturing applies the most severe conditions, since pressing, plating and assembly all heat the panel, and the lead-free processes used for most products now run hotter than the tin-lead processes they replaced. The glass transition temperature describes the point above which the resin softens and its expansion behaviour changes sharply; the decomposition temperature describes where the chemistry begins to break down. A material chosen for a high-rate application needs both to be comfortably above the process temperatures it will meet, because a laminate that softens during pressing loses dimensional control.

A third thermal property matters for reliability rather than for processing. Certain material systems resist the growth of conductive filaments through the laminate under the combination of moisture and voltage bias, and where a design uses fine spacing at elevated voltage, the rating that describes that resistance is a selection criterion rather than a detail.

Mechanical and Manufacturing Behaviour

Mechanical properties determine whether the material can be built into a board at all.

Thickness control affects the impedance, since the dielectric thickness beneath a trace is one of the parameters that sets it. Dimensional stability affects layer registration, and materials differ in how much they move during pressing; a laminate with poor stability produces boards whose layers do not align as the artwork intended.

Drilling is where the newer materials show their differences. The lowest-loss materials are often filled or based on chemistries that are more brittle than epoxy, and they drill differently: the tool wears faster, the hole walls are more prone to damage, and the drilling parameters have to be adjusted. The desmear route may also change, since the chemistry that removes epoxy smear effectively does not necessarily remove the residue these materials leave behind, and a plasma process may be needed instead.

Availability is the last mechanical consideration in practice. A material that is only stocked in limited constructions, or that has a long lead time, converts a two-week build into a six-week one. Confirming availability before finalising the stack-up is cheaper than discovering it at the quotation stage.

Cost and Manufacturability

Low-loss materials cost substantially more per unit area than conventional laminate, and the difference is not the only cost. They are often processed on more conservative parameters, which reduces yield; they may require different lamination cycles; and they may be supplied in smaller panel formats.

The way that cost is managed is to use the material only where it earns its place. A hybrid stack-up puts the high-rate signals on layers built from the low-loss material and uses the conventional laminate for the power and general routing layers, which is where the volume of the board usually is. Because the layers are pressed together into one board, this requires the two materials to be compatible in their pressing behaviour, which is a question for the fabricator and is covered in the article on high-frequency material lamination. The stack-up arrangements that follow from such a choice are described in the article on layer assignment.

Matching the Material to the Rate

The practical decisions map onto the rate rather than onto the product category.

Below roughly ten gigabits per second, a conventional laminate usually meets the requirement, and the design effort is better spent on the stack-up geometry and the routing than on the material. Around ten gigabits per second, a mid-loss material with a well-controlled dielectric constant is the usual answer. Above that, low-loss materials become necessary, and at millimetre-wave frequencies the choice moves to materials with very low loss and very stable dielectric properties, which bring their own processing requirements and are usually handled by a fabricator with specific experience in them.

What the choice always implies is that the material’s properties have to be used in the design rather than assumed. The dielectric constant and its tolerance are inputs to the impedance calculation, and the loss tangent is an input to the link budget. Selecting a material and then calculating on nominal values is how a design ends up with a channel that behaves differently from the model.

Confirming the Choice

Three questions are worth settling with the fabricator before the stack-up is fixed. What is the actual construction that will be used, including the prepreg types and the resulting dielectric thicknesses? What tolerance can be held on the impedance with that construction? And what is the lead time for the material in the quantities the project needs?

The answers feed directly into the impedance calculation and into the schedule, and they are the difference between a stack-up that describes the board that will be built and one that describes a board that could be built. The process of specifying and verifying controlled impedance is set out in our guide to impedance control, and the choice of what goes on the copper surface is covered in the article on surface finish selection.

FAQ

When is FR-4 no longer suitable? When the loss at the operating rate consumes too much of the link budget. As a rough boundary, designs around ten gigabits per second and above generally need a lower-loss material.

What matters more, the dielectric constant or the loss tangent? The loss tangent decides how much signal survives the route; the dielectric constant decides the impedance and the propagation velocity. For a long high-rate channel the loss matters more, but the dielectric constant tolerance matters more for predictability.

Why not build the whole board from a low-loss material? Because the cost is per unit area across every layer, while only the high-rate signal layers benefit. A hybrid construction puts the expensive material where it is used.