Gold-plated Rogers RO4350B PCBs are specified when the laminate has to hold its electrical values and the copper has to survive repeated assembly, test and storage. The laminate supplies the stability; the finish supplies solderability, coplanarity and shelf life — and above roughly 10 GHz the finish also adds a measurable loss penalty that most supplier pages never mention. XCE PCB builds RO4350B boards from 2 to 72 layers, holds impedance to ±10% and ships samples in as fast as 24 hours. For the wider material picture, see our Rogers PCB materials guide.
What the RO4350B Laminate Brings to a Gold-Plated Board
Dielectric performance you can put in a fabrication drawing
RO4350B is a woven-glass reinforced hydrocarbon/ceramic laminate — not PTFE. That is why it runs on FR-4 lines, and it is also why the numbers below are quoted as process values: Dk 3.48 ±0.05 and Df 0.0037 at 10 GHz, with a design Dk of 3.66 used for 8–40 GHz modelling. Mixing the two is the most common cause of a first-article impedance miss.
| Property | Value | Why it matters on a gold-plated board |
|---|---|---|
| Dk (process / design) | 3.48 ±0.05 / 3.66 | Line width is calculated from 3.66; the coupon is measured against 3.48 |
| Df @ 10 GHz | 0.0037 | About 0.18 dB/in of dielectric loss — roughly one sixth of FR-4 |
| TCDk (−50 to 150 °C) | +50 ppm/°C | 125 K excursion shifts Dk about 0.6% → ~5° of phase per 100 mm at 10 GHz |
| CTE X / Y / Z | 10 / 12 / 32 ppm/°C | X/Y below copper: a registration advantage. Z is 1.9× copper: a barrel-strain cost |
| Thermal conductivity | 0.69 W/m·K | 0.254 mm ≡ 3.68 K·cm²/W, i.e. 37 K at 10 W/cm² |
| Td / Tg | 390 °C / >280 °C | Td, not Tg, sets the reflow-cycle budget for a thermoset |
| Water absorption | 0.05% | One fifth of standard FR-4: less Dk drift in humid sites |
| Peel strength | 0.88 N/mm (5 lb/in) | Lower than PTFE laminates: fine-line adhesion is a process question |
| Flammability | UL 94 V-0 | The one place 4350B beats 4003C outright, which has no rating |
What the gold layer actually does
Two different claims get merged in supplier copy, and only one of them is true. Gold gives a flat, oxidation-free, solderable surface with a long shelf life — that part is real. Gold does not improve high-frequency transmission at the thicknesses ENIG uses.
The arithmetic is short. At 10 GHz the skin depth is 0.65 µm in copper, 0.78 µm in gold and 1.33 µm in nickel. An ENIG gold layer of 0.05–0.10 µm is therefore 8–15× thinner than the gold skin depth: it carries almost no current. The 3–6 µm nickel beneath it is 2–5× its own skin depth and has about 4.8× the resistivity of copper, so the top layer behaves like a nickel conductor. Expect roughly +0.1 to +0.15 dB/in at 10 GHz and +0.2 to +0.3 dB/in at 28 GHz against a nickel-free finish. Below 3 GHz the same difference sits inside coupon-to-coupon scatter.
| Finish | Typical thickness | Added loss vs bare copper | Wire bonding | Choose it when |
|---|---|---|---|---|
| ENIG | Ni 3–6 µm, Au 0.05–0.10 µm | +0.1–0.15 dB/in @10 GHz | No | Solderability, ICT probing, long storage |
| EPIG / DIG (nickel-free) | Pd 0.05–0.15, Au 0.03–0.07 µm | Near zero | Yes | RF path above 20 GHz, or bonding required |
| Immersion silver | 0.12–0.40 µm | Near zero | No | Lowest loss with a simple process; needs sulfur control |
| OSP | 0.2–0.5 µm | Near zero | No | Cost-first RF, short shelf life accepted |
| Hard gold (selective) | ≥0.76 µm | Adds loss; keep it off RF traces | Yes | Edge connectors and wipe contacts only |
The working rule: ENIG for assembly and test, a nickel-free or silver finish when the trace itself is the RF path above 20 GHz. The full thickness windows and their failure modes are in our PCB surface treatment guide, and the mixed-dielectric version of the same question is covered in RO4350B + FR-4 hybrid with immersion gold.
Thermal: the dielectric sets the resistance, not the gold
0.254 mm of RO4350B at 0.69 W/m·K gives 3.68 K·cm²/W vertically. At 10 W/cm² that is a 37 K rise through the laminate alone — against about 85 K for FR-4 and 127 K for RT/duroid 5880. A 0.05–0.10 µm gold film changes none of it: heat leaves through copper area, thermal vias and the baseplate, which is why a 4350B power amplifier design is a stacking problem, not a finish problem.
Expansion is the other half of the thermal story. Z-CTE of 32 ppm/°C is 1.9× copper, so the plated barrel absorbs the difference: 1.6 mm across a 215 K reflow excursion moves about 11 µm. That is manageable — but only with ≥20% elongation plating and a drilled aspect ratio kept inside the process window, not with a thicker gold layer.
Reliability statements worth putting in a purchase order
- Humidity: 1,000 h at 85 °C/85% RH is a customer qualification, not an IPC-6012 requirement — ask for the test method and the coupon design, not the sentence.
- Reflow: six lead-free cycles are realistic against a 390 °C decomposition temperature; what actually fails first is the barrel, not the laminate.
- Nickel barrier: the Ni layer does stop copper–tin intermetallic growth. Its known failure mode is black pad, controlled by keeping phosphorous at 7–10 wt%, capping gold at 0.10 µm and writing the IPC-4552B corrosion rating into the PO.
Critical Applications
- 5G NR massive-MIMO antenna boards — phase consistency matters more than absolute loss; see the RO4350B performance and cost analysis.
- 77/79 GHz automotive radar — the design rules are covered in RO4350B in automotive radar.
- Satellite communication phased arrays — where the finish has to survive thermal vacuum as well as reflow.
- High-power aerospace RF systems — thermal vias and baseplate contact, not surface finish, set the junction temperature.
Where a Gold-Plated RO4350B Board Stops Making Sense
- Long RF runs above 20 GHz: the nickel penalty (0.2–0.3 dB/in at 28 GHz) starts to rival the dielectric loss of the material itself.
- Bonding without a nickel-free finish: ENIG is not a wire-bondable surface; specify ENEPIG or hard gold.
- Ultra-low-loss front ends: at Df 0.0009 the PTFE laminates (RT/duroid 5880, Taconic TLY-5) are a different class from 0.0037 — compare on loss versus cost before choosing.
Manufacturing Capability You Should Verify
For gold-plated RO4350B, XCE PCB quotes and builds against these numbers, and every line below can be confirmed on a first article:
- Layer count: 2–72 layers, with hybrid RO4350B + FR-4 stackups (the process rules live in hybrid lamination).
- Impedance: ±10% on controlled-impedance nets, verified by TDR on coupons — not ±5%, and not quoted “up to 110 GHz”.
- Dk tolerance: whatever the laminate is certified to (±0.05 for 4350B). It is a material property; no laser compensation can tighten it.
- Finish: ENIG to IPC-4552B (Ni 3–6 µm, Au 0.05–0.10 µm, P 7–10 wt%), immersion silver, OSP and selective hard gold.
- Sample lead time: as fast as 24 hours; board size up to 600×1200 mm in a 5,000 m² plant with 80+ engineers.
“On a gold-plated 4350B board the finish is an electrical layer, not a cosmetic one. We publish the nickel thickness because it is the number that shows up on the VNA.” — XCE PCB engineering team
Related reading
- Rogers PCB materials guide: RO4350B, RO4003C, RO3003, RT5880
- RO4350B high frequency PCB: RF performance versus manufacturing cost
- RO4350B vs RO4835: what actually changes
- Choosing between Rogers 4350B and 4003C
- RO4350B + FR-4 hybrid lamination process rules
- PCB surface treatment: eight finishes and their thickness windows
Frequently Asked Questions
Is RO4350B a PTFE material?
Does immersion gold improve high-frequency performance?
What nickel and gold thickness should I specify?
Can a gold-plated RO4350B board be wire bonded?
What impedance tolerance can be held on RO4350B?
Which finish should I use for a 28 GHz antenna?
Conclusion
A gold-plated Rogers RO4350B board is a laminate decision and a finish decision, and they fail in different ways. The laminate gives 3.48 Dk, 0.0037 Df, V-0 flammability and FR-4 processing; the finish gives solderability and shelf life — and, above 10 GHz, a nickel loss penalty that has to be spent deliberately. Specify the design Dk, the Ni/Au window and the impedance tolerance, and the board will do what the simulation promised.