Most comparisons between high-frequency and standard PCBs stop at a material list: FR-4 here, PTFE there. That list is true and almost useless, because the difference is not that one board is “better” — it is that a different term dominates the error budget. On FR-4 the dielectric owns the loss and the impedance tolerance. On a 0.0027 laminate it owns almost neither, and copper roughness, dielectric height and surface finish take over. This article follows the numbers from the datasheet to the drawing to the fabrication floor.
The Short Answer: the Budget Changes Owner
Every PCB loss and impedance budget has the same terms — dielectric loss, conductor loss, dielectric height tolerance, Dk tolerance, etch tolerance. What separates a standard board from a high-frequency board is which of those terms is large enough to be worth managing.
| Budget term | Standard FR-4 board | High-frequency board |
|---|---|---|
| Insertion loss | Dielectric dominates: roughly 90–95% of the total at 10 GHz | Conductor and dielectric are comparable; copper roughness becomes a specification |
| Impedance tolerance | Dk spread (±0.2 or worse lot to lot) consumes most of the window | Dk tolerance ±0.05 tightens the window; dielectric height and etch take the margin |
| Phase / delay stability | Not usually specified; Dk drifts with temperature and moisture | TCDk and moisture absorption are datasheet columns and design constraints |
| Mechanical | X/Y CTE 14–17 ppm/°C, close to copper; Z expansion above Tg is the concern | X/Y often below copper (good for registration); PTFE Z-CTE can be 12–14× copper |
That single shift explains almost everything else in this article: why the material changes, why five process steps change with it, and why high-frequency boards cost what they cost.
The Material Table, Side by Side
Four representative materials, one column each. Figures are manufacturer datasheet values at 10 GHz unless noted; “relative cost” is the market range for laminate per square metre, not a quotation.
| Property | FR-4 (standard) | RO4350B | RO4003C | RT/duroid 5880 |
|---|---|---|---|---|
| Chemistry | Glass-reinforced epoxy | Hydrocarbon / ceramic thermoset | Hydrocarbon / ceramic thermoset | PTFE + microglass fibre |
| Dk | 4.2–4.6 | 3.48 ±0.05 (design 3.66) | 3.38 ±0.05 (design 3.55) | 2.20 ±0.02 |
| Df | 0.018–0.025 | 0.0037 | 0.0027 | 0.0009 |
| TCDk | Not specified on most sheets | +50 ppm/°C | +40 ppm/°C | −125 ppm/°C |
| CTE X / Y / Z | 14–17 / 14–17 / 45–70 below Tg, 200–300 above | 10 / 12 / 32 | 11 / 14 / 46 | 31 / 48 / 237 |
| Thermal conductivity | 0.25–0.40 W/m·K | 0.69 | 0.71 | 0.20 |
| Moisture absorption | 0.10–0.20% | 0.05% | 0.06% | 0.02% |
| UL 94 | V-0 | V-0 | No rating | V-0 |
| Relative cost per m² | 1× | 3–5× | 3–5× | 10–20× |
Two columns deserve attention because they are the ones buyers miss. RO4003C carries no UL 94 rating at all — a compliance fact, not a performance one, and a reason the 4350B/4003C choice sometimes has nothing to do with electricals. And PTFE conducts heat worse than FR-4 (0.20 against 0.25–0.40), which is why “high-frequency laminate solves thermal problems” is backwards.
The Number That Decides: Loss per Inch
Dielectric loss follows a simple scaling, accurate enough for budgeting: attenuation in dB per inch is roughly 2.6 × fGHz × √Dk × Df. Conductor loss at 10 GHz on a 0.56 mm wide 50 Ω microstrip is about 0.20 dB/inch with smooth foil and 0.30–0.40 with standard foil, because roughness adds 50–100% on top of the smooth-copper figure.
| Material | Dielectric loss @10 GHz | Conductor (50 Ω, 1 oz) | Total, smooth foil | Total, standard foil |
|---|---|---|---|---|
| FR-4 | 1.08 dB/in | 0.20 dB/in | 1.28 dB/in | 1.40–1.48 dB/in |
| RO4350B | 0.18 dB/in | 0.20 dB/in | 0.38 dB/in | 0.48–0.58 dB/in |
| RO4003C | 0.13 dB/in | 0.19 dB/in | 0.32 dB/in | 0.42–0.51 dB/in |
| RT/duroid 5880 | 0.035 dB/in | 0.18 dB/in | 0.22 dB/in | 0.30–0.36 dB/in |
The third and fourth columns are the point of the whole exercise. Moving from FR-4 to RO4350B cuts dielectric loss by 6×, but total loss only by 2.5–3.5×, because the copper term did not move. Moving again to 5880 cuts dielectric loss by another 4× and total loss by less than 2×. Past about 0.0027, further Df reduction buys progressively less unless the copper is treated as a design variable — smoother foil, thicker plating where the skin depth allows, and a surface finish that does not put nickel in the current path.
That is also the honest answer to “which laminate do I need”: on a 30 mm run at 6 GHz, FR-4 loses 0.9 dB and 4350B loses 0.3 dB — a difference a single amplifier stage can absorb. On a 300 mm feed network the same comparison is 9 dB against 3 dB, and the material stops being a preference.
The Ceiling of FR-4, in Decibels
FR-4 does not fail at a frequency; it fails when the accumulated loss or the accumulated phase error crosses a threshold. Dielectric loss alone, per 100 mm of 50 Ω line:
| Frequency | FR-4 (Df 0.02) | RO4350B (0.0037) | RT/duroid 5880 (0.0009) |
|---|---|---|---|
| 1 GHz | 0.43 dB | 0.07 dB | 0.014 dB |
| 2.4 GHz | 1.02 dB | 0.17 dB | 0.034 dB |
| 5 GHz | 2.13 dB | 0.36 dB | 0.070 dB |
| 10 GHz | 4.25 dB | 0.72 dB | 0.14 dB |
| 24 GHz | 10.2 dB | 1.72 dB | 0.34 dB |
| 77 GHz | 32.7 dB | 5.5 dB | 1.08 dB |
The usual working rule — FR-4 is acceptable to about 3 GHz — comes from this table read at 100 mm: at 3 GHz you lose about 1.3 dB of dielectric loss, which most link budgets can absorb. Shorten the run and the ceiling rises: a 20 mm stub at 10 GHz loses 0.9 dB on FR-4, which is often tolerable. Lengthen it and the ceiling falls.
Two non-electrical ceilings arrive earlier than most people expect. Dk tolerance on FR-4 is ±0.2 lot to lot, which is ±2.4% on impedance at fixed geometry — a quarter of a ±10% window gone before any process error. And above Tg the Z-axis expansion of FR-4 runs 200–300 ppm/°C, so a lead-free reflow plus a thick board is a barrel-stress problem, not an electrical one.
What Changes on the Drawing
Five items move from “default” to “specified” when the board becomes high-frequency:
- Impedance becomes a tolerance, not a target. On a controlled-impedance RF board, ±10% is the outer limit and ±5% is common on critical lines. That forces a conversation about which terms are inside the window: dielectric height (±10% thickness alone eats about ±6% of it), etch (±25 µm on a 0.254 mm layer is about ±2.8%), and Dk (±0.05 is about ±0.6%). The impedance control guide walks through that budget line by line.
- Trace width is an output, not an input. Width follows from the impedance target and the dielectric height. On RO4350B at 0.254 mm, 50 Ω is a 0.56 mm line; on 5880 it is 0.79 mm; on a Dk 10.2 material it is about 0.22 mm. “Use wider traces to reduce loss” is only true if you also thicken the dielectric — at fixed height, widening the trace lowers impedance.
- Spacing is a crosstalk rule, not a loss rule. The working convention is s ≥ 3h between parallel runs, because coupling is set by the s/h ratio. Reducing h and keeping s improves isolation, which is the opposite of the intuition that tighter spacing to ground is dangerous.
- Reference planes and via stubs become explicit. Every signal layer needs an adjacent continuous plane, no crossing a split. Through-via stubs resonate at c/(4L√εeff): about 23 GHz on a 1.6 mm board, 12 GHz on 3.2 mm. Either back-drill, or keep the board thin, or accept a notch.
- Copper and finish become electrical specifications. At 10 GHz the skin depth in copper is 0.65 µm and in nickel 1.33 µm, so a 3–6 µm ENIG nickel layer carries essentially all the current on the surface. Below roughly 10 GHz that is acceptable; above it, immersion silver or OSP is the lower-loss choice. The surface treatment guide covers the trade against durability.
What Changes on the Floor
Five process steps genuinely differ; the rest of the flow is the same flow. The full sequence is mapped in the high-frequency manufacturing process guide; the differences are these:
| Step | Standard board | High-frequency board | Why |
|---|---|---|---|
| Lamination | 170–190 °C epoxy cure | PTFE: FEP or modified-PTFE bondply at 260–300 °C; RO4000: 190–200 °C | PTFE has no melt flow, so there is no resin to fill gaps; the bondply is the adhesive |
| Drilling | Standard carbide, high hit counts | Sharp carbide, lower hit counts, entry/exit sacrifice sheet; aspect ratio ≤6:1 for PTFE | PTFE smears rather than cuts, and its Z-CTE (217–237 ppm/°C) strains the barrel far more |
| Desmear / activation | Permanganate | Sodium naphthalenide or plasma (O2/CF4) | Permanganate does not attack PTFE; without activation the plated copper has nothing to bond to |
| PTH plating | 20–25 µm, standard ductility | Same thickness, high-ductility chemistry (≥20% elongation) | The barrel must stretch with the substrate, not resist it |
| Machining | Router, laser scoring allowed | Router only; no laser cutting of PTFE | PTFE decomposes above ~400 °C, releasing HF and carbonising the edge |
The practical consequence is scheduling, not skill: PTFE cannot be co-pressed with FR-4 in one cycle, so a hybrid board is a sub-assembly first and a lamination second. That is the same constraint that makes the stack-up a manufacturing decision and not only an electrical one.
Where the Cost Actually Goes
A 10–20× material multiplier sounds fatal until it is weighted by area. A typical mixed board puts high-frequency laminate on the RF section only — often 20–40% of the panel. At 30% area and a 12× multiplier, the material contribution rises to about 4.3×, and material is rarely more than half of a finished board price.
| Cost driver | Effect of going high-frequency |
|---|---|
| Laminate | 3–5× for hydrocarbon/ceramic, 10–20× for PTFE, weighted by area share |
| Panel utilisation | Some RF materials ship in smaller standard panels; nesting loss can add 10–20% |
| Process time | Extra lamination cycles for hybrids, slower drilling, plasma in place of permanganate |
| Test | Coupon TDR and S-parameter measurement instead of bare continuity test only |
| Yield | The hidden term: PTFE delamination and registration errors cost more than the laminate when they appear at rework |
The corollary is worth stating plainly: choosing a low-loss laminate where the link budget does not need it is a cost with no return. A 2.4 GHz sensor node with a 25 mm antenna feed does not become better on 5880 — it becomes 10× more expensive and slightly worse thermally.
“Standard PCB” Is Not One Thing Either
The line is blurrier than the marketing suggests. Three intermediates matter:
- Mid-loss FR-4. Modified epoxies reach Df 0.008–0.012 — better than standard FR-4 by 2×, still 3× worse than RO4350B. They are a real option at 2–5 GHz where the whole board must stay in one cheap process.
- Hydrocarbon/ceramic thermosets that process like FR-4. RO4003C and RO4350B cure at 190–200 °C with standard prepreg, drill with standard carbide, and desmear with permanganate. They cost 3–5× FR-4 and remove most of the process disruption — which is why they dominate commercial RF. The 4350B vs 4003C comparison covers the choice inside that family.
- PTFE. The genuine process break: bondply lamination, plasma or sodium-naphthalene holes, no laser, high-ductility plating. Df 0.0009 is only available here, and so is most of the processing risk.
So the question is not “high-frequency or standard” but “which of these three tiers does the budget require”. Material limitations that are not obvious from the datasheet table — moisture, drill smear, CTE, panel formats — are collected in the laminate limitations article.
A Decision Table
Read across: the frequency and the longest critical run decide the tier; everything else is a refinement.
| Frequency / longest run | Realistic tier | Watch out for |
|---|---|---|
| ≤1 GHz, any length | Standard FR-4 | Nothing electrical; keep the stack symmetrical for assembly |
| 1–3 GHz, ≤100 mm | FR-4 or mid-loss FR-4 | Impedance tolerance on Dk spread; use coupon TDR to confirm |
| 3–10 GHz, ≤50 mm | RO4003C / RO4350B | Copper roughness; finish choice; via stub if the board is thick |
| 3–10 GHz, >100 mm | RO4003C or 5880 | Total loss budget; phase drift if the run is a feed network |
| 10–40 GHz | 5880 / TLY-5 class | PTFE process route; Z-CTE 217–237; back-drilling |
| >40 GHz, high layer count | Low-Df PTFE, often with LZ variants | Glass weave, drill registration, measurement uncertainty |
| High power in the substrate | High-conductivity laminate (6035HTC class) | Thermal resistance per unit area, not conductivity alone |
Claims in Circulation That Do Not Hold
| Common claim | What the data says |
|---|---|
| “High-frequency signals generate more heat, so these boards need special thermal design” | Backwards on both halves. Lower Df means less heat generated. The real thermal problem is conductivity: PTFE runs 0.20–0.25 W/m·K against FR-4’s 0.25–0.40, so the same heat is harder to remove. Hydrocarbon ceramics (0.66–0.71) are the exception |
| “Use wider traces and larger spacing to reduce loss” | Width is set by the impedance target and dielectric height; widening a trace at fixed height lowers impedance. Spacing is a crosstalk rule (s ≥ 3h), not a loss rule |
| “There are specialised high-frequency FR-4 variants” | Mid-loss FR-4 exists at Df 0.008–0.012, but there is no FR-4 formulation at 0.0037 or below. Above ~3 GHz on a 100 mm run it is a different material family |
| Material named as “Teflon” or “ROGERS 4350B” | PTFE is the polymer; Teflon is one brand of it. And 4350B is RO4350B, an RO4000-series hydrocarbon — not an RT/duroid product |
| “High frequency means above 1 GHz” | A frequency label is unreliable. What matters is electrical length: whether the run is a meaningful fraction of a wavelength, and whether the loss and phase terms are measurable. A 5 mm trace at 10 GHz is not a difficult board |
Conclusion
The difference between a high-frequency PCB and a standard PCB is not a material grade, it is the migration of the binding constraint: from dielectric loss and Dk spread to conductor finish, dielectric height and process control. That is why the same board can be built on FR-4 at 2.4 GHz and needs PTFE at 24 GHz, and why the material table alone cannot answer the question — the length of the run and the size of the loss budget have to be in the same sentence.
Practically: quantify loss before choosing a laminate, treat copper roughness and surface finish as part of the RF design rather than as purchasing details, and verify impedance on coupons rather than on the drawing. The design steps guide and the copper thickness article cover the two terms that most often get missed.
XCE PCB builds both sides of this comparison — standard FR-4 and the RF laminate range — with 2–72 layers, impedance held to ±10%, panels up to 600 × 1200 mm, 5,000 m² of production space and 80+ engineers, and samples in as fast as 24 hours.
Frequently Asked Questions
At what frequency do I need a high-frequency laminate instead of FR-4?
Is a high-frequency PCB just FR-4 with a better resin?
Do high-frequency boards run hotter than standard boards?
What actually changes in fabrication?
Can I mix FR-4 and high-frequency laminate on one board?
How much more does a high-frequency PCB cost?
Need a Second Opinion on FR-4 vs High-Frequency Laminate?
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657