Why Dk Changes With Frequency: Data 1–110 GHz

The dielectric constant of every PCB laminate drops as frequency rises, and the rate of that drop varies dramatically between materials. This article explains the physics, compares real Dk-vs-frequency data for common RF laminates from 1 GHz to 110 GHz,…

Cross-section of a multilayer RF PCB showing copper and dielectric layers

Dk vs Frequency: The Short Answer

Every PCB dielectric’s relative permittivity (Dk, or εr) decreases as frequency increases. The drop is caused by molecular dipoles that can no longer follow the alternating electric field fast enough, a mechanism called dielectric relaxation. For a filled PTFE laminate like Rogers RO4350B, the published Dk falls from roughly 3.66 at 1 GHz to about 3.48 at 10 GHz (process Dk). For a pure PTFE material like RT/duroid 5880, the change is far smaller—around 2.20 ± 0.02 across the same span. That difference determines whether your 50 Ω trace is still 50 Ω at the operating frequency.

The Physics: Why Dk Must Fall With Frequency

Microstrip traces on a high-frequency PCB laminate with gold finish

A dielectric’s permittivity comes from polarization mechanisms, each with a characteristic response time. At RF and microwave frequencies, two matter most:

  • Dipolar (orientational) polarization — polar molecules or side-chains in the resin rotate to align with the applied field. Above a few hundred MHz, they start lagging, and their contribution to Dk rolls off.
  • Interfacial (Maxwell-Wagner) polarization — charge accumulates at boundaries between the resin, glass weave, and ceramic fillers. This effect is strongest below ~1 GHz and fades rapidly above it.

The result is a Dk curve that drops steeply between roughly 100 MHz and 2 GHz, then continues to decrease more gently through the millimeter-wave range. The slope of that gentle region is what separates a good RF laminate from a mediocre one. Materials with non-polar backbones (PTFE, hydrocarbon) have very little dipolar contribution to lose, so their Dk is nearly flat. Laminates with polar resin systems (standard FR-4 epoxy, for instance) show a much steeper decline and higher loss.

This frequency-dependent behavior is captured mathematically by the Debye relaxation model and its extensions (Cole-Cole, Havriliak-Negami), which describe the complex permittivity ε*(ω) = ε'(ω) − jε”(ω). The real part ε’ is what we call Dk; the imaginary part ε” drives dielectric loss, quantified by the dissipation factor Df = ε”/ε’.

Real Data: Dk From 1 GHz to 110 GHz

The table below compiles published and commonly referenced Dk values for laminates frequently used in high-frequency PCB designs. All values are “design Dk” (the effective value seen by a transmission line on a finished board) unless noted otherwise. Measurement methods vary—split-post resonator, stripline resonator, clamped broadband—so treat cross-vendor comparisons as approximate.

Laminate 1 GHz 10 GHz 24 GHz 60 GHz 77 GHz 110 GHz
Standard FR-4 (mid-Tg) 4.50 4.25 ~4.1* — — —
Panasonic Megtron 6 (R-5775) 3.71 3.63 3.58 3.50 — —
Isola Astra MT77 3.07 3.00 2.97 ~2.94 — —
Rogers RO4350B 3.66 3.48 3.44 3.40 3.39 —
Rogers RO3003 3.02 3.00 2.99 2.97 2.97 ~2.96
Rogers RT/duroid 5880 2.20 2.20 2.19 2.18 2.18 2.17

*FR-4 above 10 GHz is rarely measured formally because loss makes it impractical; the value is extrapolated.

Key observations:

  • FR-4 drops ~6 % from 1 to 10 GHz; RT/duroid 5880 drops less than 1 % across the entire range to 110 GHz.
  • RO4350B, the workhorse for many designs up to 40 GHz, loses about 5 % of its Dk between 1 and 10 GHz but then flattens. This plateau is why Rogers specifies its “design Dk” at 10 GHz.
  • RO3003, a ceramic-filled PTFE laminate commonly chosen for 24 and 60 GHz radar modules, stays within 2 % of its nominal value all the way to 110 GHz.

Estimating Dk Drift in Your Design

If you know the Dk your laminate vendor specifies at one frequency and need to estimate the effective Dk at your operating frequency, the calculator below applies a first-order logarithmic drift model. Enter the reference Dk, reference frequency, target frequency, and the material’s approximate Dk slope (often available in the vendor datasheet as “Dk vs frequency” curve slope or derivable from two data points).

Loading Dk Drift Estimator…

For example, entering RO4350B’s Dk of 3.48 at 10 GHz with a target of 28 GHz and a moderate drift coefficient yields an estimated Dk around 3.43. That 0.05 drop may seem small, but run it through an impedance calculator and you will find it shifts a nominally 50 Ω microstrip line by roughly 0.6–0.8 Ω—enough to matter in a tightly toleranced impedance-controlled stackup. For designs pushing into the W-band (75–110 GHz), always request measured Dk data at your operating frequency rather than extrapolating from a 10 GHz spec.

Dk Drift and Dielectric Loss: Two Sides of One Coin

Automotive radar module PCB with mmWave components and shielding

The same polarization lag that reduces Dk also generates heat—dielectric loss. In the Debye model, the peak of ε” (maximum loss) occurs at the relaxation frequency, and the steepest drop in ε’ (Dk) happens in the same region. This is why materials with a large Dk slope also tend to have a high and rising Df. Standard FR-4’s Df climbs from about 0.018 at 1 GHz to 0.025 or more at 10 GHz, while RT/duroid 5880 stays below 0.001 across the same range.

Understanding the split between conductor and dielectric loss is essential when choosing a laminate. Below roughly 5 GHz on copper microstrip, conductor loss usually dominates and spending on ultra-low-Df material may not be justified. Above 10 GHz, dielectric loss grows rapidly, and the Dk-vs-frequency slope becomes a reliable proxy for how lossy the material will be at your target band.

How Dk Drift Affects Impedance and Timing

Impedance shift

Characteristic impedance Z₀ of a microstrip is approximately proportional to 1/√εeff. A 3 % drop in Dk from 1 GHz to 28 GHz translates to roughly a 1.5 % rise in Z₀—about 0.75 Ω on a 50 Ω line. That is within IPC-6012 Class 3 tolerance (±5 Ω) but can push a marginal design out of spec when combined with etching and laminate thickness tolerances.

Propagation delay and data rate

Propagation delay is proportional to √εeff, so a falling Dk means signals arrive slightly earlier at higher frequencies—effectively creating frequency-dependent group delay, which distorts wideband pulses. In high-speed serial links running at 28 or 56 Gbaud PAM-4, this dispersion can close the eye diagram. The Nyquist bandwidth of such a link is half the symbol rate; use the calculator below to see the Nyquist frequency for your data rate and judge whether Dk drift across that bandwidth is significant.

Loading Nyquist Data Rate Advisor…

For a 56 Gbaud PAM-4 link, the Nyquist frequency is 28 GHz. If your laminate’s Dk shifts by 0.10 between DC and 28 GHz, the resulting differential group delay across the channel can exceed 1 ps/inch—enough to require equalization. Choosing a material with a flatter Dk curve (Megtron 6 or better) is often cheaper than adding DSP taps.

Practical Guidelines for Managing Dk vs Frequency

1. Use design Dk, not bulk Dk

Laminate vendors report two numbers: “bulk” or “material” Dk (measured on a bare slab) and “design” or “process” Dk (measured on a finished stripline coupon). The design Dk accounts for the copper-roughness effect and the resin/glass ratio of the actual construction. Always use design Dk for impedance calculations. Rogers’ RO4350B datasheet explicitly distinguishes the two.

2. Match the test frequency to your operating frequency

A Dk value measured at 2.5 GHz via clamped stripline is not valid for a 77 GHz radar design. When scaling an RF design to a higher frequency band, request or look up Dk data at the target frequency. Isola’s Astra MT77 data, for example, provides characterization up to 40 GHz.

3. Account for temperature as well

Dk also shifts with temperature, typically +50 to +200 ppm/°C for PTFE-based materials. In automotive radar modules that see −40 °C to +125 °C, the combined frequency-plus-temperature Dk swing can reach 3–4 %, demanding tighter manufacturing tolerances at mmWave.

4. Simulate with a wideband dielectric model

Do not assign a single Dk value in your EM simulator for a broadband design. Tools like Ansys HFSS and Keysight ADS accept Djordjevic-Sarkar or Debye multi-pole models that capture the frequency dependence of both Dk and Df. Feeding a flat Dk into a 77 GHz simulation can produce impedance and insertion-loss errors exceeding 10 %.

5. Request measured data from your supplier

For critical RF builds, ask the fabricator for a test coupon with Dk extraction at your operating frequency. A split-post dielectric resonator (SPDR) test at a single frequency is inexpensive and eliminates guesswork.

Frequently Asked Questions

Does Dk ever increase with frequency?

In standard PCB laminates, no. The monotonic decrease is a fundamental consequence of polarization mechanisms losing their ability to follow the field. Apparent increases in published data usually indicate measurement artifacts, mode conversion, or a resonance in the test fixture. In exotic engineered metamaterials the effective permittivity can behave differently, but those are not conventional PCB substrates.

Can I use FR-4 Dk values for a 5.8 GHz design?

You can, but you must use the Dk at 5.8 GHz (roughly 4.2–4.3 for mid-Tg FR-4), not the 1 MHz value of 4.7 that many generic datasheets list. The bigger concern at 5.8 GHz is FR-4’s high Df, which causes significant insertion loss on anything longer than a short stub.

How flat does Dk need to be for 77 GHz radar?

Automotive radar designers typically target a Dk variation of less than 1 % from 76 to 81 GHz. Materials like RO3003 (Dk ~3.00 ± 0.01 across that band) and RT/duroid 5880 meet this requirement. RO4350B is borderline at 77 GHz and is more commonly used for the baseband or IF sections of the radar module.

What measurement method gives the most accurate Dk at mmWave?

Fabry-Perot open resonator and free-space transmission/reflection methods are preferred above 30 GHz because they avoid the conductor-loss de-embedding errors inherent in stripline methods. The IEEE 1785.1 standard on waveguide measurements provides relevant calibration guidance for mmWave setups.

Is a lower Dk always better for RF?

Not always. A lower Dk widens the trace for a given impedance (good for loss and fabrication tolerance) but also increases the board area needed for distributed elements like quarter-wave stubs and coupled-line filters. At frequencies below 6 GHz, a moderate Dk of 3.0–3.5 often gives the best trade-off between size and performance.

Ready for a quotation?

Send your requirements for a material, stackup and manufacturing review. Price and schedule are confirmed for your project.

Upload Gerber