Low Loss Laminate Selection for High Speed PCB Designs

Material selection for a high speed board used to be a simple choice between ordinary FR4 and an expensive low loss laminate. With serial links now running at tens of gigabits per second per lane, the decision has become a graded one, and the right answer is the cheapest material that still meets the loss budget at the required distance.

What Determines the Loss

Loss in a transmission line comes from two sources that behave differently. Conductor loss arises from the resistance of the copper and from the roughness of its surface, and it rises with the square root of frequency. Dielectric loss arises from the polarisation of the material and rises linearly with frequency.

At lower frequencies the conductor dominates, and at higher frequencies the dielectric takes over. The crossover is why a material change matters more for a long link at high frequency than for a short one, and why the same stackup can be adequate for one channel and inadequate for another on the same board.

Reading the Dissipation Factor

The dissipation factor is the material property that sets the dielectric loss, and it is the number to compare when selecting a laminate. Ordinary FR4 sits around zero point zero two, mid loss materials are around zero point zero one, and the low loss and ultra low loss grades fall to zero point zero zero five or below.

The relationship is roughly linear: halving the dissipation factor halves the dielectric part of the loss for the same geometry. When a link budget is tight, that is a direct gain, and it is the main reason for paying more for the material. The same property affects a filter or a resonant structure, as described in the general treatment of a microstrip and stripline line.

Insertion Loss Budget

The useful way to choose is to build a budget. Start with the total loss the link can tolerate, taken from the interface specification, then subtract the connector, the package, the via transitions and the crosstalk allowance. What remains is available to the trace.

Dividing the remaining loss by the length gives a loss per unit length, and that figure can be compared against the published loss curves for candidate materials at the relevant frequency. Doing that arithmetic before selecting a material usually shows that one grade is sufficient for the longest channel and a cheaper one covers everything else. Where the trace length varies across the board, a mixed stackup with different materials on different layers is possible, though it adds process complexity. The routing decisions that shorten the longest paths are described in the material on high frequency traces and data bus routing.

Dielectric Constant and Impedance

The dielectric constant sets the impedance of a line for a given geometry, and it also sets the propagation velocity. Higher grades of material often have a lower dielectric constant, which means a wider trace for the same impedance and, at the same thickness, slightly lower loss.

Stability matters as much as the value. A material whose dielectric constant varies with frequency or temperature changes the impedance along the channel and introduces reflections that no equaliser can fully correct. The specification should state the tolerance over the operating range, not only a nominal figure at one frequency.

Copper Foil and Surface Roughness

Conductor loss depends on the copper as well as on the geometry. A rough foil lengthens the path the current follows at high frequency, and a standard foil can add more loss than the difference between two dielectric grades.

Low profile and smooth foils reduce that contribution, and they are usually specified together with a low loss laminate for the fastest channels. The trade is adhesion, since smooth copper bonds less strongly, so the combination should be one the fabricator has experience with rather than one assembled from two data sheets.

Material Grades in Practice

Standard FR4 remains adequate for most digital designs below a few gigabits per second per lane and for short links at higher rates. Mid loss materials cover a band above that and are the common choice for server and networking boards where the traces are of moderate length.

Low loss materials are used for long backplane channels and for links that run close to the interface limit, and ultra low loss materials for the highest rates and the longest distances. The price rises with each grade, and the increase is not linear, so specifying the highest grade as insurance is an expensive habit.

Stackup and Layer Assignment

Where a board carries both high speed channels and ordinary digital signals, the stackup can assign the low loss material only to the layers that need it. That approach keeps most of the board on a cheaper material and concentrates the cost where the loss matters.

Hybrid stacks have their own requirements: the materials must be compatible in the lamination cycle, and the expansion mismatch between them affects warpage. The general principles of building a stack that behaves predictably are set out in the material on a multilayer PCB for high speed.

Process and Cost Considerations

Low loss materials are not difficult to process in the way that PTFE is, but they still differ from FR4 in their drilling, their lamination cycle and their dimensional behaviour. A fabricator who works with them routinely will hold the registration and the impedance tolerance more easily than one who does not.

Cost has three components: the raw material, the process premium and the test cost. The material dominates, so the decision should be revisited whenever the channel length or the data rate changes, because a grade that was necessary at one revision may be unnecessary at the next, and the reverse is also true.

Design Practices That Reduce the Requirement

Before paying for a better material, it is worth reducing the distance the signal has to travel. Placing the devices that must be connected close together, avoiding unnecessary layer transitions and keeping the reference plane continuous all reduce the loss the material has to survive.

Via stubs are another source of loss at high frequency. Back drilling a long through via removes the unused barrel and its resonance, which can be more cost effective than moving to a lower loss laminate for the whole board. The two measures should be compared on the same budget rather than considered separately.

Verification

The result is verified with a test coupon on the panel and with a measurement of a representative channel. Impedance coupons confirm the geometry, and a loss measurement on a known length of line confirms the material behaviour against the published curve.

Where the measured loss exceeds the prediction, the cause is usually the copper roughness or the actual dielectric constant of the delivered material rather than the design. Identifying which of the two is responsible is what makes the next revision cheaper, and it requires the coupon data to be kept with the board records.

FAQ

Is the material the biggest factor in high speed loss? Usually, but not always. A rough foil or a long via stub can contribute as much as a material change, and both are cheaper to fix.

Can FR4 be used for a ten gigabit link? For short distances, yes. Beyond a few inches the loss usually forces a mid loss or low loss material, depending on the interface specification and the equalisation available.

Does a lower dielectric constant always mean lower loss? Not directly. The two are related but independent, and a material should be chosen on its loss and its stability rather than on its dielectric constant alone.