How Copper and Solder Mask Shift Z0

ENIG does not move Z0 in this model. On a 0.2104 mm FR-4 microstrip at 2.4 GHz, the 1 oz, unmasked 50 Ω width is 14.71 mil. Bare copper, OSP, and ENIG all read 50.00 Ω on that width. ENIG does change the loss, from 0.475 dB/in to 0.625 dB/in. Copper weight and solder mask are what move the impedance. Two-ounce copper on the same width reads 48.60 Ω. A flooded mask reads 49.36 Ω. On a 1.6 mm board those corrections shrink, because the foil and the mask are a smaller fraction of the dielectric.

The loss, and why the nickel is the loss, is ENIG versus bare copper. This note is only the impedance.

Does ENIG change the impedance?

The plating stack is a surface impedance. IPC-4552B in this library is 0.05 µm gold on 4.5 µm nickel, applied to the top and the sidewalls. It is not added to the copper thickness that Hammerstad–Jensen uses for the effective width. Microstrip puts most of its current on the bottom face, against the laminate, which is still copper. The model assigns the plated surface to 25% of the current, and that fraction changes conductor loss. It does not change εeff or Z0.

OSP is electrically bare copper here. Same Z0, same loss. If a quote says “ENIG shifted our 50 Ω coupon,” look for a different finished width, a different mask, or a different Dk before you blame the gold flash. The gold is thinner than a skin depth. It is not a geometry change.

The chart is the thin prepreg, where $t/h$ is large enough to see. The zero-length ENIG bar is the point.

How much does copper weight move Z0?

Thicker copper looks wider electrically. The Hammerstad–Jensen correction adds a $\Delta u$ that grows with $t/h$, so Z0 falls if you keep the drawn width. Hold 14.71 mil on the 0.2104 mm board:

Half-ounce, 17.5 µm, reads 50.91 Ω. One ounce, 35 µm, is the 50.00 Ω solve. Two ounce, 70 µm, reads 48.60 Ω. Re-solve 50 Ω and the widths are 15.19 mil, 14.71 mil, and 13.95 mil. A two-ounce power pour next to a one-ounce radio, with the same drawn width, is not the same impedance. Spec the weight the solver used.

On 1.6 mm FR-4 the same two-ounce step, width held at that board’s 50 Ω solve, reads 49.68 Ω. The foil is the same 70 µm. The dielectric is almost eight times thicker, so $t/h$ barely moves $\Delta u$. Copper weight is a first-order width correction on a thin prepreg and a small one on a two-layer 1.6 mm board.

How much does solder mask move Z0?

The mask is a capacitance filling factor, not a conformal coat in a field solver. This library uses εr 3.3 and tanδ 0.02, 20 µm thick, which is a Taiyo-class LPI figure, not your fab’s measured mask. A flooded mask covers the fringe. A relieved mask pulls back. The default relief is 8 mil.

On the 0.2104 mm line, relieved reads 49.89 Ω and flooded reads 49.36 Ω. Re-solving 50 Ω, flooded wants 14.38 mil instead of 14.71 mil. That is 0.33 mil, which is less than a typical etch tolerance and more than nothing if you are stacking it on top of a two-ounce correction. On 1.6 mm FR-4 the flooded mask reads 49.91 Ω. I would not retune a 3 mm trace for the mask. I would retune a 15 mil trace if the mask is flooded and the target is actually 50 Ω rather than 50 Ω ± a few ohms.

The mask also adds a little loss. On this thin line it is the difference between 0.475 dB/in with the mask off and 0.491 dB/in flooded, at 2.4 GHz. The nickel is the larger loss term. The mask is the larger impedance term.

Do you retune the width?

Lock 50 Ω with the copper weight and the mask that are on the traveler. Leave the finish out of the width. Put the finish in the loss comparison. Then look at the fab’s Dk. A move from εr 4.4 to the process Dk they press will dwarf 0.33 mil of mask if their Dk is not 4.4. The heights in the 1.6 mm FR-4 width note are published prepreg thicknesses used with this library’s εr, on purpose, so you can see the height effect without pretending to be their impedance coupon.

If the line is GCPW, copper thickness is not in Z0 at all. The map is zero-thickness. Changing ounces changes the loss geometry factor and not the width the solver prints. Do not “compensate” a GCPW gap for two-ounce copper inside this model. Compensate the microstrip.

What this shift is not

It is not a stackup field solver, and the mask is not 3D. Extreme $t/h$, or a mask much thicker than 20 µm, is outside the story these filling factors were shaped for. ENIG thickness here is the IPC-4552B window used by the loss model, not a coupon cross-section. Nickel permeability is a loss parameter, μr 3.5 in that model, and it never enters Z0.

Roughness does not enter Z0 either. Huray or Hammerstad changes the conductor loss and leaves the impedance table alone. If 0.5 Ω is your margin, measure the coupon. Use the table to decide which variable is worth putting on that coupon. On this board, at this frequency, it is copper weight and a flooded mask. It is not the gold.

References

• E. Hammerstad and Ø. Jensen, “Accurate Models for Microstrip Computer-Aided Design,” IEEE MTT-S International Microwave Symposium Digest, 1980. The thickness correction $\Delta u$ is what moves Z0 with copper weight.
• IPC-4552B. The ENIG stack used for loss: 0.05 µm Au on 4.5 µm Ni. It is not a Z0 term.
• IPC-2141. Controlled-impedance geometry, including copper thickness and mask as stackup inputs. This calculator’s mask is a filling-factor overlay, not the IPC field solution.
• The solder-mask constants are a typical LPI, εr 3.3, tanδ 0.02, 20 µm. They are not a fab measurement.

Related

• ENIG versus bare copper loss
• 50 Ω microstrip width on 1.6 mm FR-4
• Microstrip calculator, 0.2104 mm, 1 oz, no mask, 2.4 GHz

FAQ

Does ENIG change microstrip impedance?

Not in this model. On a 0.2104 mm FR-4 microstrip at 2.4 GHz, the 1 oz unmasked 50 Ω width is 14.71 mil. Bare copper, OSP, and ENIG all read 50.00 Ω. ENIG changes the loss, from 0.475 dB/in to 0.625 dB/in, because the nickel is in the surface impedance. The plating stack is IPC-4552B, 0.05 µm gold on 4.5 µm nickel, and it is not added to the trace thickness.

How much does copper thickness change impedance?

Holding that 14.71 mil width, 0.5 oz reads 50.91 Ω and 2 oz reads 48.60 Ω. The Hammerstad–Jensen thickness correction widens the effective strip. Re-solving 50 Ω, 2 oz wants 13.95 mil and 0.5 oz wants 15.19 mil. On 1.6 mm FR-4 the same 2 oz step only reaches 49.68 Ω, because t/h is smaller.

How much does solder mask change impedance?

The mask model is a capacitance overlay, εr 3.3, 20 µm, not a 3D conformal solve. On the 0.2104 mm line, a relieved 8 mil opening reads 49.89 Ω and a flooded mask reads 49.36 Ω. Re-solving 50 Ω, the flooded mask wants 14.38 mil instead of 14.71 mil. On 1.6 mm FR-4 the flooded mask reads 49.91 Ω. The correction shrinks as the dielectric gets thick.

Should I include mask and copper when I lock 50 ohms?

Yes, if the traveler has them. Lock the width with the copper weight and the mask you will actually print. Leave ENIG out of the width solve. Put ENIG in the loss comparison instead. If the remaining error is a few tenths of an ohm and the fab’s Dk is not 4.4, the stackup Dk dominates this correction.