FR-4 stops being a transmission-line material somewhere in the GHz range. The limit is not the dielectric constant — it is the loss tangent. At low frequency a Df of 0.02 is a rounding error; at 5 GHz it is the difference between a link that closes and one that does not.
The two numbers that matter
Dk (dielectric constant) sets trace impedance and velocity. What matters just as much as its value is its stability over temperature and across the panel — RF designs need Dk variation to be small and specified. Df (loss tangent) sets how much signal is converted to heat per wavelength. FR-4 sits near Df 0.02; the laminates below sit an order of magnitude lower.
The grades we run
| Material | Dk | Use it when |
|---|---|---|
| Rogers RO4000 series | 3.3–3.6 | RF, antenna, anything above 1 GHz — low loss with stable Dk over temperature |
| Taconic RF grades | 2.5–3.5 | Microwave, high-Q filters |
| Panasonic Megtron 6 / 7 | 3.3–3.6 | High-speed digital channels in the 25–56 Gbps class |
| Al₂O₃ / AlN ceramic | 9–10 / 8–9 | High-power RF and power modules where no organic outgassing is allowed |
Hybrid stackups: the cost answer
You rarely need a full panel of expensive laminate. The standard approach is a hybrid stack: one or two RF layers on the low-loss material, bonded to conventional FR-4 layers that carry power and digital logic. The RF traces get the electrical environment they need; the rest of the board keeps FR-4 economics. The cost of this approach is process — different cure profiles and bonding cycles in the same panel — so it needs a fab that runs hybrids routinely.
Design notes that save iterations
- Spec the controlled impedance coupons on the RF material itself, not on an FR-4 test panel — Dk tolerances differ.
- Mind the glass weave: on very fine traces, weave skew matters for high-speed serial links.
- Copper roughness adds loss at high frequency; low-profile foils are worth asking about on long channels.