Low Loss Materials for Millimetre Wave PCB Stack-Ups
Above roughly thirty gigahertz the wavelength on a printed circuit board is measured in millimetres, and low loss material stops being an optimisation and becomes the starting point of the design. Every stub, pad and connector launch that was invisible at lower frequencies is now a significant fraction of a wavelength, and the dielectric itself absorbs a measurable share of the signal on any line of useful length.
What Changes at Millimetre Wave
At millimetre wave frequencies the geometry and the material interact, so they cannot be chosen one after the other. A via that behaved as a short circuit at five gigahertz has inductance that matters, a pad behaves as a capacitor, and a plane split that used to be harmless radiates from the edge as if it were an antenna.
Loss also rises for two different reasons. Conductor loss grows with the square root of frequency while dielectric loss grows almost linearly, so the dielectric term quickly dominates on a long line. That is why the choice of laminate decides the link budget long before the layout details are settled, and why a single decibel of attenuation is often the difference between a link that closes and one that does not.
Dielectric Constant and Dissipation Factor
The dielectric constant sets the impedance and the physical width of a controlled trace, and its tolerance usually matters more than its nominal value. A material quoted at 3.0 with a range of plus or minus 0.04 gives repeatable impedance, while an average figure taken from a batch that varies by ten percent does not.
The dissipation factor describes how much energy the dielectric absorbs per cycle, and it is the number that decides attenuation over distance. Both figures change with frequency and with temperature, so the supplier curve should be used in the model rather than a single room temperature value quoted on a data sheet.
Where the Loss Comes From
Total insertion loss is the sum of dielectric loss, conductor loss and radiation. The dielectric term depends on the material, the conductor term depends on the foil and on copper thickness and roughness, and the radiation term depends on the geometry and on how well the transition into the line has been designed. Radiation is the term most often ignored, and it is the one that a poorly designed launch turns into a dominant contributor; our high speed design rules cover the layout habits that keep it small.
That split matters because it shows where effort is worth spending. On a short line inside a module the conductor and launch terms dominate, while on a long antenna feed the dielectric term decides everything, so the same expensive material can be essential in one place and completely wasted in another.
Glass Weave and the Fibre Effect
Woven glass fabric consists of fibre bundles separated by resin rich regions, and the two have different dielectric constants. When a differential pair runs along the weave, one trace may sit over a bundle while its partner sits over resin, and the resulting skew appears as a phase error that length matching cannot remove.
The usual remedies are to rotate the routing relative to the weave, to specify a spread or flat glass fabric, or to accept a material in which the effect is smaller. Glass weave effects are visible well below millimetre wave, and they are one reason a laminate that measures well on a coupon can behave badly on a real board. A supplier that can quote the weave style and the resin content gives the designer something to work with.
Copper Foil and Conductor Loss
At high frequency the current travels in a thin skin at the surface of the conductor, and rough copper forces it along a longer path. Low profile and rolled foils reduce that penalty, and the difference between a standard electrodeposited foil and a smooth rolled product is measurable at ten gigahertz and obvious beyond thirty.
Skin depth also decides how much copper is actually useful. Plating a trace far thicker than a few skin depths adds cost and capacitance without reducing loss, because the current is not flowing in the extra copper. Our high frequency laminate guide covers the foil side of the loss budget in more detail.
Material Families for Millimetre Wave
The realistic candidates are PTFE based laminates, hydrocarbon and thermoset materials with low filler content, and in some designs liquid crystal polymer. PTFE offers the lowest loss and the most stable dielectric constant, at the cost of a soft material that needs careful drilling, plating and handling throughout fabrication.
Hydrocarbon systems are stiffer and easier to process with slightly higher loss, and they suit mixed stack-ups where only one or two layers carry millimetre wave signals. Our notes on PTFE PCB properties and on laminate material properties describe the trade-offs between them.
Stack-Up and Layer Thickness
A millimetre wave stack is normally thin, because thin dielectric widens a fifty ohm line and reduces radiation from the trace edges. It is also usually simple, with one or two signal layers sandwiched between ground planes so that the field stays confined and the reference under every trace is continuous.
Those ground planes must be solid beneath the signal, which means no plane splits and no slots crossing the trace. A split that crosses a trace forces the return current to detour around it, and the inductance of that detour appears as an impedance bump at exactly the frequency where the design has least margin. Layer transitions should use a via with its ground return placed immediately alongside, or a launch structure that has been simulated as a complete transition rather than as a via plus a pad.
Fabrication Tolerances That Matter
At millimetre wave the tolerances that matter are dielectric thickness, conductor width and the registration between layers. A ten percent change in dielectric thickness moves impedance by roughly five percent, which is enough to spoil a return loss specification that looked comfortable during the design phase.
Plating thickness, surface finish and solder mask shift the electrical result as well. Mask is normally removed from a millimetre wave trace, the finish is chosen for low loss rather than for solderability alone, and the fabrication drawing names the tolerance class that the design genuinely needs instead of a default. Saying so on the drawing avoids a difficult discussion after the panels have already been etched.
Qualification and Measurement
Material is qualified with a resonant cavity or transmission line measurement that reports dielectric constant and loss tangent across frequency, not with a single point figure taken at one gigahertz. The frequency range matters as much as the value, because a curve that is flat to ten gigahertz can turn upward sharply beyond it. Coupons travelling with the panel then confirm that the fabricated board matches the model the design was built on.
A first article should include a time domain reflectometry check of impedance and a network measurement of a known line length. Comparing measured insertion loss with the simulated value tells the designer whether the model, the material or the process is responsible for the difference, and that answer is worth more than the number itself.
FAQ
Is PTFE always the right choice at millimetre wave? It gives the lowest loss, but a hydrocarbon material may be adequate for a short run and much easier to process. The decision follows from the link budget, not from the data sheet alone.
Can standard FR-4 be used above thirty gigahertz? It can carry a signal over a very short distance, but the loss is high and the dielectric constant varies with frequency. For anything longer than a few millimetres the margin normally disappears.
How does gopcb handle millimetre wave work? We build on the laminate named in the design, hold dielectric thickness and line width to the agreed class, keep the ground reference continuous, and supply coupon measurements so the customer can compare the board with the simulation.
