High-Frequency Microwave PCB Materials and Their Properties

High frequency microwave PCB materials are usually presented as a table of dielectric constants and dissipation factors, as if both were constants. They are not. Dk falls slowly and Df rises roughly linearly with frequency, and the number on a datasheet depends on how it was measured: a clamped stripline at 1 MHz, a split-post resonator at 10 GHz and a split-cylinder resonator at 40 GHz can differ by several percent on Dk and by a factor of two on Df. The practical consequences are large. RO3010 is quoted at Dk 10.2 but designed at 11.20 — a 9 % gap that changes a 50 Ω line by more than 40 %. This article reads the numbers properly, gives the families by frequency band, and corrects four claims that the conventional version of this material keeps repeating.

The dielectric loss of a laminate per unit length follows a closed form that every selection argument should start from: attenuation in dB per inch is approximately 2.6 × f_GHz × √Dk × Df. Two things follow immediately. Loss rises linearly with frequency, and for a fixed electrical length the total dielectric loss depends only on Df — Dk cancels, because a higher Dk means a shorter physical line for the same phase. A high-Dk material saves area; it does not save dielectric loss.

The last row explains the whole material market in one line. At 1 GHz FR-4 costs 0.11 dB/in, which is survivable; at 40 GHz it costs 4.3 dB/in, which means a 100 mm run loses 17 dB before the copper is counted. Microwave material selection is therefore not about a “better” dielectric but about where the loss curve becomes unacceptable for the link budget.

The number a designer needs and the number a fabricator measures are not the same number, and the industry solved the problem by publishing both. Process Dk is what a coupon measures — the material as built, at the frequency and with the method stated on the datasheet. Design Dk is a slightly higher or lower figure used in simulation to account for how the etched geometry and the measurement setup differ from the ideal model.

Using the process number in a field solver therefore produces a line that is too wide and an impedance that is too low, and the error is largest exactly where high-Dk materials are used — in filters and antennas where a 10 % dimensional error is a tuning problem rather than a tolerance. The second source of disagreement is the test method itself: a clamped stripline measurement per IPC-TM-650 2.5.5.5 is a low-frequency, pressure-dependent measurement; a split-post resonator is a 10 GHz measurement on a bare coupon; a split-cylinder resonator measures a different mode entirely. Comparing a 1 MHz Df with a 10 GHz Df is not a comparison.

Polytetrafluoroethylene is the lowest-loss solid dielectric in routine production, and the two materials that matter most are 5880 and TLY-5. Both are Dk 2.20 ±0.02 with Df 0.0009, but they are not interchangeable: 5880 uses micro-glass reinforcement with a Z-CTE of 237 ppm/°C, TLY-5 uses woven glass with 217 ppm/°C and a lower peel strength of 3.04 N/mm against 5.5 N/mm.

The property that decides whether PTFE can be used is usually not electrical. Its thermal conductivity is 0.20 W/m·K, so a 0.254 mm dielectric has the same thermal resistance as 500 mm of copper; it cold-flows under lamination pressure, thinning the dielectric by up to 4 % and taking a 50 Ω line to 48.8 Ω; and it cannot be desmeared with permanganate, only with sodium naphthalene or plasma. Those three constraints are processing considerations in their own right, covered in processing considerations for high-frequency PCBs.

One further number is routinely missed: the temperature coefficient of Dk. 5880 is −125 ppm/°C, so a 100 °C excursion changes Dk by 1.25 % and moves a 10 GHz resonant frequency by about 63 MHz. On a wideband link that is invisible; on a narrowband filter it is the difference between in-band and out-of-band. The woven-glass alternative is compared grade by grade in TLY-5 parameters and production notes.

Adding ceramic powder to PTFE raises Dk while keeping the loss floor intact, which is how a single chemistry covers 3.00 to 10.2. The trade is mechanical: the ceramic raises CTE in X and Y towards copper, and it raises flexural modulus, so the material stops behaving like a soft plastic. The table below is the full RO3000 set with the parameters that actually get specified.

The 3010 row carries the argument for high-Dk material in a filter. Its in-plane CTE of 13 and 11 ppm/°C is below copper’s 17, so a 40 mm resonator changes length less than the copper it is bonded to over temperature, and the resonant frequency moves less than the housing. That is worth more than the area saving on a narrowband design, and it is the reason 3010 appears in cavity and filter work rather than in general routing.

Ceramic-filled PTFE also holds the Td above 500 °C across the series, which is a processing fact rather than a marketing one: the material survives reflow without bulk degradation, so the constraint moves to the copper adhesion and the desmear route instead. The Taconic equivalents sit in Taconic high-frequency laminates if the selection has to stay within one supplier’s stack.

The RO4000 series is a different chemistry from everything above, and the original version of this article got that wrong twice: it listed 4000-series material as a “ceramic-enhanced composite” alongside Rogers, and it repeated the claim that the series has a Df of 0.002–0.004. The series is a ceramic-filled hydrocarbon thermoset — not PTFE — with a Df of 0.0027 to 0.0037, and it laminates at 190–200 °C like an FR-4 rather than at 260–300 °C like PTFE.

Two rows deserve attention. 4835 has the same electrical parameters as 4350B and differs in oxidation resistance, which matters on a high-power amplifier that runs hot for years rather than on a link budget. 6035HTC has a third of the loss of 4350B and twice the thermal conductivity, and its 0.254 mm thermal resistance of 1.76 K·cm²/W produces a 17.6 K rise at 10 W/cm² against 36.8 K for 4350B — the grade exists because heat, not loss, was the limit.

A footnote that catches designers regularly: at 0.101 mm core thickness the Dk of 4350B is 3.36 and 4835 is 3.33, below the 3.48 quoted for standard thickness. A thin-core design carried over from a 0.508 mm stack therefore calculates the wrong width. The head-to-head comparison with 4003C is in 4350B versus 4003C.

Standard FR-4 is a legitimate microwave material up to about 3 GHz, and the boundary is arithmetic rather than prejudice. At Df 0.020 the dielectric loss is 0.108 dB/in at 1 GHz, 1.08 dB/in at 10 GHz and 4.3 dB/in at 40 GHz, so a 50 mm feed at 10 GHz loses 2.1 dB in the laminate alone. Modified epoxies and high-Tg blends move Df to 0.008–0.015, which extends the useful range but does not reach the 0.002–0.004 band of the hydrocarbon family.

The other limit is dimensional. FR-4 has a thickness tolerance of roughly ±10 % on a laminate and a Dk that varies with glass weave, so a 50 Ω line with a 0.24 mm width sits inside a ±3 % impedance window only by luck. On a 0.254 mm stack the same tolerance is worth ±6 %, which is why impedance-controlled FR-4 is quoted at ±10 % rather than ±5 %.

The honest summary is that FR-4 is not excluded from microwave work by its dielectric constant — 4.3 is not far from 3.5 — but by its loss tangent and its tolerance stack. Where the run is short, the band is below 3 GHz and the impedance window is loose, FR-4 wins on cost by a factor of three to five. The full comparison, step by step, is in high-frequency PCB versus standard PCB.

Three resin families are routinely mis-described in summaries of this subject, and the corrections matter because each one leads to a wrong selection.

The pattern is that each of these families is chosen for a non-electrical reason — temperature, moisture, layer count, process compatibility — and then justified on electrical grounds that do not hold at the top of the band. Polyimide is the clearest case: it survives 260 °C for extended periods and is the standard for rigid-flex and high-reliability multilayer, and it is a poor choice for a 20 GHz feed because 0.012 of Df costs 0.6 dB/in before any copper loss.

Electrical parameters select the material; mechanical and thermal parameters decide whether it can be processed on the line that exists. Four of them account for most of the surprises.

The peel strength row is the one that decides between two electrically identical materials on a fine-line board. 4350B at 0.88 N/mm is adequate for 0.15 mm lines and marginal for 0.075 mm, while 5880 at 5.5 N/mm tolerates the handling that a thin PTFE panel receives. Conversely 4350B can be processed on an FR-4 line, which 5880 cannot.

Selection becomes simple once the band is fixed, because the loss curve does most of the work. The table below is the version to keep on a wall.

Two entries need a note. In the 20–40 GHz row the dielectric term has stopped being the problem: at 10 GHz a 0.38 dB/in total on 4350B splits into 0.18 dielectric and 0.19 copper plus roughness, and by 40 GHz the copper term dominates because it grows as the square root of frequency while roughness scattering grows faster. In the 40–77 GHz row the specification that decides yield is Dk uniformity across the panel, because a phase array with a 2 % Dk gradient steers wrongly regardless of how low the average loss is.

The 5880 multilayer build is the one that concentrates all of these constraints, and its process route is set out in multilayer Rogers 5880 processing challenges. Where a lower Dk is needed than 2.20, 5880LZ gives 1.96 with Df 0.0019 and a Z-CTE of about 41 ppm, at the cost of a much larger CTE mismatch to copper in X and Y.

Laminate price is the least useful comparison in this subject, because the material is a small fraction of the finished board cost on a thin RF design and a large fraction on a thick one. The comparison that survives a purchasing review is cost per decibel avoided over the actual run length.

Take a 150 mm feed at 10 GHz on a 0.254 mm stack. FR-4 costs 1.08 dB/in of dielectric loss, or 6.4 dB over the run, and no amount of design work recovers that. 4350B costs 0.18 dB/in, or 1.07 dB, and processes on an FR-4 line. 5880 costs 0.035 dB/in, or 0.21 dB, and requires a separate press cycle, sodium-naphthalene desmear and a 6:1 aspect ratio. The difference between 4350B and 5880 is 0.86 dB for a step change in process complexity; the difference between FR-4 and 4350B is 5.3 dB for a material change alone. Most programmes should buy the second difference and question the third.

Five items make a laminate order unambiguous, and each one is a place where a wrong assumption becomes an impedance error that nobody can trace later.

1. Grade and thickness, with the Dk basis stated — process or design, and at which frequency, because 4350B at 0.101 mm is 3.36 rather than 3.48. 2. Copper weight per layer and the finished outer copper, since plating adds 20–25 µm. 3. The impedance target per layer with coupon geometry, so width and impedance cannot drift apart. 4. The desmear route and press cycle, because PTFE and thermoset cannot share either. 5. A flammability and peel requirement, since 4003C has no UL94 listing while 4350B is V-0, and the two are otherwise interchangeable in a design review.

With those five, material selection stops being a conversation about which laminate is better and becomes a check on whether the datasheet number being used was measured the same way as the coupon that will verify it. The station-by-station route those materials travel is in the high-frequency PCB manufacturing process.

The material families in this article differ less in dielectric constant than the usual summary suggests — 2.20 to 10.2 covers the whole commercial range — and far more in loss, in tolerance and in what the shop has to do to build them. Dk is a design variable that buys area; Df is the variable that decides whether the link closes; and the tolerance on both, together with Z-CTE, thermal conductivity and peel, decides whether the board can be made at all.

Read every datasheet number with its method and frequency attached. A Df of 0.0009 measured by split-post resonator at 10 GHz and a Df of 0.002 measured by clamped stripline at 1 MHz are not on the same scale, and a design Dk of 11.20 against a process Dk of 10.2 is not a rounding difference — it is a 20 % impedance error waiting for a prototype.

XCE PCB builds across the range: PTFE and ceramic-filled PTFE with sodium-naphthalene or plasma desmear and 6:1 aspect ratio, 4000-series thermoset on FR-4-compatible cycles, 2 to 72 layers, panels to 600×1200 mm, impedance ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. The RO3000 series in more detail is in the RO3000 series introduction.