Published Nov 12, 2026 · Last updated Nov 16, 2026 · 8 min · IndieRF

Skin Depth at 10 GHz: 0.66 µm Copper, 0.72 µm Nickel

Foil roughness and ENIG nickel more than double a 50 Ω RO4350B line's conductor loss at 10 GHz.

At 10 GHz the skin depth is 0.66 µm in copper and 0.72 µm in ENIG nickel, so the 4.5 µm nickel layer is six skin depths thick. On a 50 Ω microstrip on 6.6 mil RO4350B with Rogers' ½ oz ED foil, conductor loss goes from 0.17 dB/in smooth to 0.28 with roughness and 0.45 with ENIG: about 1.1 dB more per 4 inches.

IndieRF Transmission Line Calculator

50 Ω microstrip, 6.6 mil RO4350B, ½ oz Rogers ED foil (Sq 2.8 µm), no mask, 10 GHz, bare copper

Open in calculator
Length units

Pick a JLCPCB, PCBWay or OSH Park stackup to fill height, εr, and copper. Editing those fields, or the mask, switches this back to Custom.

(mil)
(mil)
(mil)

Lengths are shown to 0.01 mil. That is the drawing resolution, not a fabrication tolerance. Ask your fabricator for the etch allowance.

Hammerstad ran on this foil's Sq of 2.8 µm under the strip and on the ground. The strip's top and sides stay smooth: Rogers measured untreated foil faces adding little loss (Horn et al., DesignCon 2016), so the 0.4 µm top side is not run. The maker prints no Huray values for it.

(GHz)
W = 13.75 milt = 0.69 milh = 6.60 milGround planeDielectric: Rogers RO4350B · εr 3.66 · tanδ 0.00375 mil
TraceDielectricGroundDrawn to scale.

Impedance Z0

50.01Ω

Insertion loss

0.412dB/in

0.0162 dB/mm

Closed form, field solve pending

Effective permittivity ε_eff

2.868

Static ε_eff 2.857

Where the loss comes from

42% of the loss is the copper under the finish (0.174 dB/in of 0.412 dB/in).

  • Copper42.2%
  • Plating0.0%
  • Roughness26.2%
  • Dielectric31.5%
  • Mask0.0%

Skin depth and plating

At 10.00 GHz, 0.0% of the current is in the plating and 100.0% is in the core.

Foil: Rogers ED ½ oz (RO4000). Sq = 2.8 µm. A larger value is a rougher surface and more conductor loss.

  • Copper coret 17.50 µm · δ 0.66 µm · 100.0%

    Conductivity 5.80e7 S/m

Advanced roughness parameters

Huray treats the foil teeth as snowballs. a is the sphere radius and As/Af is the sphere area divided by the flat area. A rougher foil has a larger As/Af and more conductor loss. These are the numbers stored for this foil type, not a measurement of your board.

No snowball values for this foil. The maker prints only the RMS roughness, so the loss uses Hammerstad on Sq = 2.8 µm.

Switch Surface finish from Bare copper to ENIG. Plating goes from 0% to 29% of the loss (0.165 dB/in), and the total goes from 0.408 to 0.573 dB/in.

Six skin depths of nickel at 10 GHz

Skin depth is how far the current gets into a metal before it has fallen to 1/e:

δ=ρπfμ0μr\delta = \sqrt{\frac{\rho}{\pi f \mu_0 \mu_r}}

Copper skin depth is 66/f66/\sqrt{f} µm with ff in MHz: 2.1 µm at 1 GHz, 0.66 µm at 10 GHz. Anything plated thicker than a few skin depths is what the current sees. The rule holds for copper only; nickel needs its own σ and μr. With Transmission Line Calculator's copper (σ=5.8×107\sigma = 5.8\times10^7 S/m) the formula gives 2.090 µm at 1 GHz and 0.661 µm at 10 GHz.

Nickel conducts a quarter as well as copper and is magnetic, and both shrink the skin depth. The library uses σ=1.4×107\sigma = 1.4\times10^7 S/m and μr=3.5\mu_r = 3.5, which puts nickel's skin depth at 2.27 µm at 1 GHz and 0.72 µm at 10 GHz, close to copper's. The ENIG stack in the library is 0.05 µm of gold on 4.5 µm of nickel. That is the IPC-4552B nominal the library cites, not a value from a fab. IPC-4552 allows 3 to 6 µm of nickel (as quoted in the I-Connect007 column below), so the same finish can be two skin depths thicker or thinner at 10 GHz depending on the shop.

The nickel μr\mu_r is the softest number here. Plated nickel's permeability varies with frequency and with the deposit, and electroless nickel-phosphorus is less magnetic than bulk nickel. Rautio and Coonrod fitted measured ENIG lines on RO4003C and RT/duroid 6002 with a nickel-phosphorus conductivity of 5×1055\times10^5 S/m. Treated as non-magnetic, that conductivity gives a skin depth of 7.1 µm at 10 GHz, thicker than the whole layer. Read the nickel rows as one model of nickel, good to the size of the effect, and measure a coupon if 0.05 dB/in matters to you.

The loss slope from 1 to 10 GHz

I had a microstrip filter on Rogers come back with more insertion loss, and a steeper loss slope against frequency, than I expected. I had not accounted for the nickel in the finish, the copper foil or its surface roughness.

The table puts numbers on those three effects for this note's RO4350B line. The width is held at the 10 GHz 50 Ω solve, 13.70 mil (0.348 mm), and each column adds one effect to the one before it.

Scroll sideways for more columns

50 Ω microstrip on 6.6 mil RO4350B, 13.70 mil wide (field-solved at 10 GHz, then held), ½ oz copper, no mask. Conductor loss in dB/in: smooth copper, plus Rogers ED ½ oz roughness (Hammerstad on Sq 2.8 µm under the strip and on the ground; its top and sides smooth, though Rogers prints 0.4 µm there), plus ENIG (4.5 µm nickel on the strip's top and sides, 36% of the loss, which leaves roughness on the rest). Each frequency links to the ENIG line.
Frequencyδ copperδ nickelSmooth copper+ foil roughness+ ENIG nickelTotal, bareTotal, ENIG
1 GHz2.090 µm2.274 µm0.0550.0820.1390.0940.152
2.4 GHz1.349 µm1.468 µm0.0850.1340.2190.1640.249
5.8 GHz0.868 µm0.944 µm0.1330.2130.3430.2860.415
10 GHz0.661 µm0.719 µm0.1740.2830.4480.4080.573

Between 1 and 10 GHz the conductor loss climbs 0.12 dB/in on smooth copper, 0.20 dB/in with the foil's roughness and 0.31 dB/in with ENIG as well. Smooth copper follows f\sqrt{f}. The roughness term grows faster than that, from 0.027 to 0.108 dB/in, because the skin depth shrinks under the foil's teeth: Hammerstad's factor rises from 1.76 to 1.97 on the 2.8 µm bonded side. On the 0.4 µm top side it would rise from 1.03 to 1.30, and the next section says why the calculator leaves that side smooth. The nickel term grows from 0.058 to 0.165 dB/in as its share of the plated-face current goes from 84% to 94%. A model with smooth copper and no finish gets the 1 GHz loss wrong by 0.08 dB/in and the 10 GHz loss wrong by 0.27, so it gets the slope wrong too, and a filter's upper band edge is where you find out.

Roughness and nickel sit on different faces

ENIG coats the top and the sidewalls of the strip. The treated side of the foil, the rough one, is bonded to the laminate on the bottom face. A 2-D boundary-element solve of this line puts 64% of the loss on the two faces bonded to the laminate, the strip's bottom and the ground plane, and 36% on the strip's top and sides. Transmission Line Calculator gets the same 36% from the geometry, by Wheeler's rule applied one face at a time, and puts the nickel on that share and the roughness on the rest. The split follows the width: on a 20 mil core the same solve puts 62% of the loss on the top and sides of a 6 mil strip and 22% on a 98 mil one, so the same nickel adds almost three times the fraction of copper loss to the narrow line.

Bare copper roughens the same two faces and leaves the top and sides smooth. Rogers prints this foil's top side at Sq 0.4 µm, and Hammerstad on that would add 30% to those faces at 10 GHz. Rogers' own measurements say the untreated side adds little. Horn and coauthors (DesignCon 2016) laminated reverse-treated foils both ways up and put the untreated side on the laminate, where the current is densest. At 50 GHz the loss stayed near 1.1 dB/in up to Rq 1.3 µm on that side, while lines with the treated side down rose from 1 to 1.75 dB/in. Hammerstad gives both sides of their ½ oz foil, 0.8 µm each, the same factor, so it cannot tell them apart. The sidewalls are etched copper with none of the bonding treatment, and Rogers prints no sidewall value, so smooth sidewalls are our assumption. At 10 GHz that gives 1.62 times the smooth loss, the same roughness as the ENIG line carries. The smooth-copper loss both columns scale is within 1% of the same solve.

At 10 GHz on the ENIG line the nickel is the larger effect: 0.165 dB/in of nickel against 0.108 dB/in of roughness. Each has its own fix. A smoother foil attacks the bottom face. A finish without nickel, or keeping the long runs on an inner layer, attacks the top. Swap in Rogers' ½ oz LoPro foil (Sq 0.9 µm) on the ENIG line and Hammerstad drops the roughness term from 0.108 to 0.085 dB/in, for 0.425 dB/in of conductor loss. Hammerstad saturates near twice the smooth loss, and Rogers' measured 10 mil lines in the Coonrod article below separate by more than 0.03 dB/in at 10 GHz. The next step is a coupon.

Rogers' calculator puts this line higher

John Coonrod (Rogers, 2012) modeled this same construction, a 50 Ω microstrip on 6.6 mil RO4350B with ½ oz ED copper, in Rogers' MWI-2010 calculator and reported about 0.42 dB/in of conductor loss and 0.13 dB/in of dielectric loss at 10 GHz on bare copper. Transmission Line Calculator gives 0.28 and 0.12 for that line. The dielectric agrees. Two choices in MWI-2010 account for the 0.14 dB/in conductor gap. It multiplies Hammerstad's factor into all of the current, the smooth top side included, which takes 0.28 to 0.34 dB/in. It also appears to start from Hammerstad and Jensen's 1980 conductor-loss formula, which reads 26% above the boundary-element solve on this line (0.217 against 0.173 dB/in smooth) and takes it to 0.43. Horn and coauthors document the first choice. The second is our inference, because that formula with the same factor reproduces Coonrod's 0.42, 0.27 and 0.14 dB/in on 6.6, 10 and 20 mil cores to within 0.01 dB/in. Both tools agree on Hammerstad's factor itself, 1.97 at 2.8 µm and 10 GHz. That factor stops at twice the smooth loss, and Rogers' own measurements (Horn and coauthors, 2010) on 3 µm foil climb past it above about 10 GHz, so budget the total from a measured coupon.

The laminate values come from Rogers' RO4000 data sheet. The library's RO4350B uses the design Dk, 3.66 (8 to 40 GHz, differential phase length), over the process Dk, 3.48 ± 0.05 (10 GHz clamped stripline), because the design value is the one Rogers measures on microstrip and labels for design. Rogers also says the design Dk falls by about 0.1 from 20 mil cores to 4 mil ones. That moves the 50 Ω width. The dissipation factor is 0.0037, Rogers' 10 GHz value, at every frequency; Rogers prints 0.0031 at 2.5 GHz, so the dielectric loss in the 1 and 2.4 GHz rows reads about 19% high.

Same stackup, other finishes (OSP, immersion silver, ENEPIG) and a relieved mask are in ENIG vs bare copper microstrip loss. That note uses the calculator's generic reverse-treated foil with Huray roughness, so its 10 GHz numbers sit a few hundredths of a dB/in from these. To see the LoPro line, open it on ENIG and read the roughness row.

References

  • Rogers Corp., RO4000 Series High Frequency Circuit Materials data sheet, PUB# 92-004, Revised 1592 080322: RO4350B process Dk 3.48 ± 0.05, design Dk 3.66, Df 0.0037 at 10 GHz and 0.0031 at 2.5 GHz, 0.0066" standard thickness, ½ oz (18 µm) ED cladding.
  • Rogers Corp., Copper Foils for High Frequency Materials, Publication #92-243, Revised 1530 081826, Table 1: ED ½ oz on RO4350B, Sq 2.8 µm dielectric side and 0.4 µm top side; LoPro ½ oz, 0.9 µm dielectric side and 0.8 µm top side.
  • J. Coonrod, "Circuit Materials and High-Frequency Losses of PCBs," The PCB Magazine, February 2012, pp. 24 to 29.
  • A. F. Horn III, J. W. Reynolds, P. A. LaFrance and J. C. Rautio, "Effect of Conductor Profile on the Insertion Loss, Phase Constant and Dispersion in Thin High Frequency Transmission Lines," DesignCon 2010; Printed Circuit Design & Fab, May 2010: MWI multiplies the smooth loss by Hammerstad and Jensen's roughness factor on the RMS roughness.
  • A. F. Horn III, P. A. LaFrance, C. J. Caisse, J. P. Coonrod and B. B. Fitts, "Effect of Conductor Profile Structure on Propagation in Transmission Lines," DesignCon 2016: untreated sides of reverse-treated foils, Rq up to 1.3 µm, facing the laminate; insertion loss at 50 GHz flat near 1.1 dB/in against Rq, against 1 to 1.75 dB/in with the treated side down.
  • B. Rautio and J. Coonrod, "An Efficient EM Simulation Model for ENIG Plated Metal Finishes Including Conductor Side-Wall Plating Verified With Physical Measurement from 1 to 50 GHz," Sonnet Software and Rogers Corp.
  • M. Carano, "Trouble in Your Tank: At 40 GHz, Everything Matters, Part 1," I-Connect007, July 2026, for the IPC-4552 nickel window of 3 to 6 µm.
  • IPC-4552B, Performance Specification for Electroless Nickel/Immersion Gold (ENIG) Plating for Printed Circuit Boards.
  • E. Hammerstad and O. Jensen, "Accurate Models for Microstrip Computer-Aided Design," IEEE MTT-S Digest, 1980.

Questions? Contact

Discussion

No comments yet. Start the thread with a measurement, a correction, or a worked example.