Different Dielectric Constants in High-Frequency PCBs: What Dk Does and Does Not Do

A dielectric constant is a ratio: how much more charge a material stores than vacuum does in the same field. In a high-frequency PCB it does exactly two things — it sets the speed at which a signal travels and, together with the geometry, the impedance the signal sees — and it does not set the loss, which is the job of the dissipation factor. That distinction is worth a whole article, because most of the advice written about different dielectric constants gets it wrong in the same three ways. Dk 2.20 gives 4.56 ps/mm and a 21.9 mm guided wavelength at 10 GHz; Dk 10.2 gives 9.08 ps/mm and 11.0 mm — a factor of two in delay and in the length of every resonator, and no change at all in loss if the dissipation factors match.

Different Dielectric Constants in High-Frequency PCBs

A Ratio, Not a Quality

The dielectric constant, more precisely the relative permittivity εr, is the ratio of the permittivity of a material to that of free space. In a capacitor it is the factor by which the material multiplies the stored charge; in a transmission line it is the factor by which the material slows the wave. Both statements describe the same physics, and only one of them is useful when choosing a laminate.

What a designer needs from Dk is three numbers: the delay per millimetre, the guided wavelength at the frequency of interest, and the line width that produces the target impedance on a given dielectric thickness. Everything else that is written about Dk — that low Dk means low loss, that high Dk means thick dielectrics, that low Dk is required for high impedance — is either a correlation or a misconception, and the three are dealt with in section 10 with numbers.

One further point of vocabulary. The Dk that appears in a field solver is not quite the Dk of the material: in a microstrip, part of the field is in air, so the effective permittivity εeff lies between 1 and Dk and depends on the geometry. On Dk 2.20 laminate a 50 Ω microstrip has εeff of about 1.87; on Dk 11.20 it is about 7.41. The gap between Dk and εeff is why a stripline and a microstrip on the same laminate travel at different speeds.

Speed, Wavelength and Delay: The Arithmetic

Propagation velocity is v = c / √εeff, so delay is 3.336 × √εeff picoseconds per millimetre, and guided wavelength is λ0 / √εeff. The table below is the whole of what Dk does, in three columns.

Dk (50 Ω microstrip)DelayGuided wavelength at 10 GHz50 Ω width at 0.254 mmQuarter wave at 2.4 GHz
2.20 → 1.874.56 ps/mm21.9 mm0.79 mm28.6 mm
3.00 → 2.425.19 ps/mm19.3 mm0.65 mm25.2 mm
3.38 → 2.675.45 ps/mm18.4 mm0.59 mm23.9 mm
3.66 → 2.855.63 ps/mm17.8 mm0.56 mm23.2 mm
11.20 → 7.419.08 ps/mm11.0 mm0.22 mm14.3 mm

The last two columns are why high-Dk materials exist. A quarter-wave resonator at 2.4 GHz is 22.9 mm long on a PTFE laminate and 11.5 mm on a ceramic-filled one, and since a filter or an antenna patch scales with area rather than length, the difference in footprint is closer to four to one. On a small module where the antenna has to sit inside a 20 mm envelope, that is not a refinement but the reason for the material choice.

The delay column matters on the digital side of the same board. 4.56 against 9.08 ps/mm is a factor of two in electrical length, so a length-matched interface that was tuned on one laminate is 100 % out of match on the other. Whenever a design moves between material families, every matched length, every stub and every resonant element has to be recomputed — which is a mechanical consequence of a material parameter that is usually presented as a single number in a selection table.

How Dk Is Measured, and Why Two Values Exist

Dk is not a constant of nature, it is a measurement, and different fixtures give different numbers. Two methods dominate in laminate data sheets. The split-post dielectric resonator measures a thin sample at a fixed frequency, usually 10 GHz, with the sample clamped between two dielectric posts; it is accurate and it is the usual source of the process value. The clamped stripline resonator measures a sample in a resonant line at a lower frequency, often around 1 GHz, and is closer to how the material behaves in a board.

Neither is wrong, and both are in the same table on some data sheets, which is where the confusion starts. RO3010 is published with a process value of 10.2 ± 0.30 and a design value of 11.20; RO4003C with 3.38 ± 0.05 and 3.55; RO4350B with 3.48 ± 0.05 and 3.66. The design value is the one a field solver expects, because it comes from the same kind of broadband characterisation the tool assumes. Using the process value in simulation produces a line that is wider than it needs to be, and on RO3010 the error reaches 5 % of impedance.

The practical rule is to state which value is meant, on the drawing, next to the impedance target. It is also worth remembering that all polymers show dispersion: Dk falls slowly as frequency rises, so a value quoted at 1 MHz is not the value at 10 GHz, and a data sheet that names neither frequency cannot be compared with one that does.

A third source of disagreement is anisotropy. A glass-reinforced laminate has a different Dk along the glass bundles than across them, and a different value again through the thickness, so a stripline that sees the z-direction value and a microstrip that sees mostly the in-plane value can measure differently on the same sheet. The ceramic-filled PTFE grades are filled rather than woven and are close to isotropic; the glass-reinforced versions of the same Dk, such as RO3210 against RO3010, are not, and their z-axis expansion of 34 ppm/°C against 16 shows the glass. Where a design depends on Dk to three digits, ask which direction the number refers to.

The Three Families in Practice

Commercial high-frequency laminates cluster into three Dk bands, and the boundaries are useful because they correspond to different design intentions rather than to different chemistry.

BandTypical materialsWhat the Dk buysWhat it costs
2.0–2.3RT/duroid 5880 (2.20), TLY-5 (2.20), 5880LZ (1.96)Longest wavelength, widest lines, lowest dielectric loss with matched DfArea; poor thermal conductivity (0.20 W/m·K); weak bonding to foil
2.4–3.0Taconic TLX-0 to TLX-6 (2.45–2.65), RF-30 (3.00), RO3003 (3.00)Low loss with better mechanical behaviour than pure PTFEStill low k, still limited by foil bonding
3.3–3.7RO4003C (3.38), RO4350B (3.48), RO4835 (3.48), RO3035 (3.50), RO6035HTC (3.50)FR-4 process compatibility; the mainstream RF choiceHigher loss than PTFE at equal thickness
6–11RO3006 (6.15), RO3010 (10.2), RO3210 (10.2), RO6010.2LM (10.2)Small features: filters, coupled lines, antenna elementsTight dimensional tolerance; narrow lines; large TCDk on some grades

The bands are not a quality scale. A 2.20 laminate and a 10.2 laminate can have the same dissipation factor and therefore the same loss per wavelength, and the choice between them is a decision about size, tolerance and thermal path rather than about electrical quality. Comparing Dk values as though a lower one were better is the single most common error in laminate selection.

Dk Sets Width, Which Sets Tolerance

The 50 Ω solution on a given dielectric thickness is a fixed ratio of width to thickness, and that ratio depends on Dk. On 0.254 mm laminate the width falls from 0.79 mm at Dk 2.20 to 0.22 mm at Dk 11.20. The consequence that is rarely mentioned in selection tables is what happens to tolerance.

Etch tolerance is roughly a fixed number of micrometres, so its relative effect grows as the line narrows. On a 0.79 mm line, ±25 µm is ±3.2 %; on a 0.22 mm line it is ±11.4 %, and at W/h ≈ 2 that is nearly 5 % of impedance, most of a ±10 % window spent on etch alone. A high-Dk material therefore demands tighter etch control, tighter dielectric thickness control, or a wider acceptance window, and a supplier who can hold ±10 % on a Dk 3.38 stack may only hold ±15 % on a Dk 10.2 one. That is a property of the geometry, not of the supplier.

The same arithmetic explains why high-Dk designs often use a thinner dielectric: it restores the width. Going from 0.254 mm to 0.508 mm on Dk 10.2 doubles the 50 Ω width to 0.43 mm and halves the relative etch error, at the cost of a thicker board and longer vias. The design is a triangle of Dk, thickness and tolerance, and no corner can be chosen alone. The impedance side of the same trade is in impedance control in high-frequency PCB manufacturing.

Loss per Inch and Loss per Wavelength Are Different Questions

Dielectric loss per inch follows αd ≈ 2.6 f √Dk Df dB/in, which contains Dk under a square root. Read carelessly, that suggests high Dk loses more. Read with the wavelength in mind, it does not, because a shorter guided wavelength means fewer inches for the same number of wavelengths.

LaminateDk / DfLoss at 10 GHzLoss per wavelength50 Ω width at 0.254 mm
RT/duroid 58802.20 / 0.00090.035 dB/in0.020 dB0.79 mm
RO4003C3.38 / 0.00270.13 dB/in0.061 dB0.59 mm
RO4350B3.48 / 0.00370.18 dB/in0.080 dB0.60 mm
RO301010.2 / 0.00220.18 dB/in0.050 dB0.22 mm

The 3010 row is the argument: it loses the same 0.18 dB/in as 4350B, because the larger √Dk is offset by the smaller Df, and it loses less per wavelength than 4003C despite a Dk three times higher. Loss is a Df story with a Dk footnote, and any comparison that ranks materials by Dk alone will rank them wrongly.

TCDk: The Third Dimension of the Constant

Dk also changes with temperature, and the coefficient is not small in either sign. Over a 100 °C excursion the Dk of RT/duroid 5880 falls by about 1.25 %, that of RO4350B rises by about 0.5 %, and on a high-Dk ceramic-filled laminate the change can reach 4 %. Since delay scales with √Dk, half of that appears as electrical length, and electrical length that changes is a phase error or a frequency shift.

LaminateTCDkDk change over 100 °CResonator drift at 2.4 GHz
RO4003C+40 ppm/°C+0.4 %≈ 5 MHz
RO4350B+50 ppm/°C+0.5 %≈ 6 MHz
RO6035HTC−66 ppm/°C−0.66 %≈ 8 MHz
RT/duroid 5880−125 ppm/°C−1.25 %≈ 15 MHz
RO6010.2LM−425 ppm/°C−4.25 %≈ 51 MHz

Fifteen megahertz on a 2.4 GHz resonator is a channel-width problem, and 51 MHz is a redesign. The sign matters as much as the magnitude: a positive and a negative coefficient cannot be substituted for one another in a stack that must hold frequency, even when the room-temperature Dk is identical, and a mixed stack of two materials with opposite signs will drift differently at every temperature. The measurement side of the same property, including how it is verified, is in electrical characteristics in high-frequency PCB manufacturing.

Moisture is the other variable with the same effect and the opposite sign of some coefficients: absorbed water has a Dk near 80, so a laminate that takes up 0.02 % by weight shifts its effective Dk upward, and a board that is dried before measurement and humid in service will not behave identically. PTFE is the least affected of the families because it absorbs almost nothing; polyimide and some hydrocarbon systems absorb more, and the direction of the shift is the opposite of a negative TCDk material, which is how a design can appear to be stable over temperature in the lab and drift in the field.

Where High Dk Is the Whole Point

Three classes of structure are designed around a high dielectric constant rather than in spite of it. Coupled-line filters. Coupling depends on the ratio of gap to dielectric thickness, so a high Dk allows a tight gap at a manufacturable dimension, and the resonator length falls with √Dk — a 10.2 material makes a 2.4 GHz filter roughly a third the length of a 2.20 one. Patch antennas. The patch dimension scales as λ0/√Dk, so a Dk 10.2 patch at 2.4 GHz is about 40 % of the linear size of a Dk 2.2 patch, and the ground plane shrinks with it. Delay lines and phase shifters. A fixed electrical length in a smaller physical length, with the caveat that TCDk then becomes a phase-stability question.

The counter-case is equally concrete. A long antenna feed on a 10.2 laminate has a 0.22 mm line carrying the full transmit power, with 11 % width tolerance from etch alone and a conductor loss per inch higher than a wider line would have. High Dk belongs in the small, tolerance-critical, low-power features; the feed usually belongs on a lower-Dk, wider-line material, which is why mixed-Dk stacks exist at all. Stack design for that case is covered in high-frequency PCB stack-up design.

Choosing a Dk in Six Steps

1. Fix the size constraint. If the antenna or filter must fit an envelope, that sets a minimum Dk: L = λ0 / (4 √εeff) for a quarter-wave resonator. 2. Fix the loss budget in dB per wavelength, not per inch, so that materials of different Dk compare fairly. 3. Check the width the chosen Dk produces at the available dielectric thickness, and the etch tolerance as a percentage of it. 4. Check TCDk against the operating range if the structure is resonant. 5. Check the thermal path, since Dk and thermal conductivity are unrelated and a low-loss material can be a poor conductor of heat — the arithmetic is in what thermal management refers to. 6. Confirm the value you are designing to — design Dk for the solver, process Dk for acceptance.

Steps 1 and 3 are the ones that get skipped, and they are the two that produce a design that cannot be manufactured: a Dk chosen for loss that gives an un-etchable line width, or one chosen for size that gives a resonator with 15 MHz of thermal drift.

Three Statements About Dk That Are Backwards

StatementWhat is actually true
A low Dk means low lossLoss follows Df. RO3010 at Dk 10.2 and Df 0.0022 loses the same 0.18 dB/in at 10 GHz as RO4350B at Dk 3.48 and Df 0.0037, and less per wavelength than RO4003C. Low-Dk materials are often low-loss because the PTFE family happens to have low Df, which is a correlation, not a rule
A high Dk requires larger layer spacingThe opposite. For a fixed impedance, higher Dk allows a thinner dielectric and a narrower line, which is exactly why ceramic-filled laminates are used to shrink filters and antennas. What a high Dk does require is tighter dimensional control, because the line is narrower
Only a low Dk can give high impedanceImpedance is set by W/h and Dk together. 100 Ω is available on Dk 10.2 at a narrower width or a thicker dielectric; the constraint is manufacturability, not physics. What is true is that high impedance on a high Dk needs a line so narrow that etch tolerance dominates

All three errors have the same shape: a real correlation turned into a rule, and then used to choose a material for a reason that belongs to a different parameter. The fix is to ask, for each material decision, which of the three quantities — delay, loss, size — is actually binding.

Conclusion

Different dielectric constants do two things in a high-frequency PCB: they set the delay, 4.56 ps/mm at Dk 2.20 against 9.08 ps/mm at 11.20, and they set the line width that produces a given impedance, 0.79 mm against 0.22 mm on 0.254 mm laminate. Everything else that is attributed to Dk is either a correlation with Df or a consequence of geometry. The guided wavelength at 10 GHz falls from 21.9 mm to 11.0 mm across the same range, which is why high-Dk materials exist: every resonator, patch and coupled section shrinks with the square root of Dk.

Two numbers deserve to travel with the Dk on every drawing. The first is which value is meant, design or process, since the gap reaches 10 % on RO3010 and moves impedance by 5 %. The second is TCDk, from −425 ppm/°C on a high-Dk laminate to +50 on a hydrocarbon one, because a resonator that drifts 51 MHz over 100 °C is a different product from one that drifts 6 MHz even though both were specified at the same room-temperature Dk.

XCE PCB stocks and processes all three Dk bands — PTFE from 2.20, hydrocarbon from 3.38, ceramic-filled to 10.2 — with 2 to 72 layers, impedance held to ±10 %, panels to 600×1200 mm, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. The geometry that turns a Dk into a width is in microstrip and stripline design, and the width-to-spacing trade in trace width and spacing.

Frequently Asked Questions

What does the dielectric constant control in a PCB?
Two things: how fast a signal travels and, together with the geometry, the impedance it sees. Delay is 3.336 × √εeff ps/mm, so Dk 2.20 gives about 4.56 ps/mm and Dk 11.20 about 9.08 ps/mm. Loss is not controlled by Dk — that is the dissipation factor — and size is a consequence, because guided wavelength falls as √Dk.
Does a higher Dk mean higher loss?
No. Dielectric loss is 2.6 f √Dk Df dB/in, so loss is linear in Df and only square-root in Dk. RO3010 (Dk 10.2, Df 0.0022) loses the same 0.18 dB/in at 10 GHz as RO4350B (Dk 3.48, Df 0.0037), and less per wavelength than RO4003C. Ranking laminates by Dk alone ranks them wrongly.
Why does RO3010 have two Dk values, 10.2 and 11.20?
10.2 ± 0.30 is the process value, measured on the laminate and used for acceptance; 11.20 is the design value derived from broadband characterisation, which is what a field solver expects. The 10 % gap moves a 50 Ω line by about 5 % if the wrong one is used. Simulate with the design value, accept on the process value, and say which one the drawing means.
Is a high Dk better for small antennas?
It is the only way to make them smaller. Patch and resonator dimensions scale as λ0/√Dk, so a 2.4 GHz quarter-wave resonator is 22.9 mm on Dk 2.2 and 11.5 mm on Dk 10.2, and the patch area difference is closer to four to one. The costs are a narrower line for the same impedance, tighter etch tolerance, and often a larger TCDk.
How much does Dk drift with temperature?
Between −425 and +125 ppm/°C depending on the family: about +40 on RO4003C, +50 on RO4350B, −66 on RO6035HTC, −125 on 5880 and −425 on RO6010.2LM. Over 100 °C that is a Dk change of 0.4 % to 4.25 %, which appears as a 5 to 51 MHz shift on a 2.4 GHz resonator. The sign matters — opposite coefficients cannot be substituted in a stack that has to hold frequency.
Why does the same Dk give a different delay on stripline and microstrip?
Because the microstrip has part of its field in air, so its effective permittivity is lower than the material Dk. On Dk 2.20 laminate a 50 Ω microstrip has εeff near 1.87 while a stripline is fully embedded and sees close to the material value. The same board therefore has two different propagation velocities, which matters whenever lengths are matched across layers.
RF & Microwave PCB Fabrication

Need the Dk You Simulated, Not the Dk on the Datasheet?

Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.

2–72 layers ±10% impedance tolerance Manufacturing since 2009