90 Ω USB Differential Pair Width

On JLC04161H-7628, a 90 Ω edge-coupled microstrip at 1 GHz with a 5 mil gap is 10.61 mil wide, 0.269 mm. That row is an outer pair over layer 2: 7.1 mil of 7628, εr 4.4, finished 1 oz copper of 1.6 mil, and solder mask 0.6 mil at εr 3.8. Bare copper. The same stackup with an 8 mil gap is 12.28 mil, 0.312 mm. Turning the mask off at the 5 mil gap is 10.95 mil, 0.278 mm. JLC04161H-3313 at a 5 mil gap, 3.5 mil of 3313 at εr 4.1, is 6.41 mil, 0.163 mm.

The 5 mil and 8 mil gaps are numbers typed into this calculator. JLCPCB’s stackup pages do not set the gap, and this note does not invent a minimum space. Ask the fab for the space they will etch.

Open the 5 mil JLC04161H-7628 pair. The lock solves the width for 90 Ω. Zdiff is 90.00 Ω, so Zodd is 45.00 Ω and Zeven is 67.13 Ω. Zcommon is 33.56 Ω. The coupling coefficient k is 0.197. Odd-mode ε_eff is 3.088 and even-mode ε_eff is 3.559. Odd-mode delay is 5.862 ps/mm. Mode skew is 0.431 ps/mm, the even mode lagging the odd mode because more of the odd-mode field sits in the slot air.

What each row assumes

A smaller gap couples the traces more tightly, so Zdiff falls and the solved width for 90 Ω gets narrower. The 5 mil row is 10.61 mil. The 8 mil row is 12.28 mil, k drops from 0.197 to 0.134, and mode skew drops from 0.431 ps/mm to 0.376 ps/mm. A wide gap lets Zdiff approach twice the impedance of one trace alone, and the two ε_eff values move toward each other.

The mask sits on the odd mode only. Kirschning and Jansen do not include solder mask. This library’s single-strip mask overlay is applied to Zodd, then Zdiff is twice that. Zeven stays the bare-line value. That is why the mask-off width, 10.95 mil, is wider, and why its mode skew, 0.518 ps/mm, is larger: the odd mode is less loaded.

JLC04161H-3313 uses the same copper and the same mask, on 3.5 mil of 3313 at εr 4.1. At a 5 mil gap the width is 6.41 mil. k is 0.123. Odd-mode delay is 5.932 ps/mm. Mode skew is 0.222 ps/mm.

W/h on the 7628 5 mil row is 10.61 / 7.1 = 1.49. S/h is 5 / 7.1 = 0.70. Both sit inside 0.1 to 10.

Intra-pair skew and the neighbor

Mode skew is not a length error. It is the even-mode delay minus the odd-mode delay, per length of pair. On this microstrip it is 0.431 ps/mm at the 5 mil gap. A stripline pair is one dielectric, so both modes share εr and that difference is zero.

Intra-pair skew is a length mismatch between the two traces. Odd-mode delay on the 5 mil row is 5.862 ps/mm, so 1 ps of skew is 0.171 mm of extra length on one trace. This calculator does not take that length difference as an input. Match the two lengths in the layout.

The 2W/3W rule is about the next pair, or an unrelated trace, sitting about two to three trace-widths away. It does not set the gap inside this pair, and it does not enter the impedance model.

What the model is

Edge-coupled microstrip uses Kirschning and Jansen, IEEE Trans. MTT, January 1984, with the March 1985 corrections, on a Hammerstad–Jensen single strip. They state static impedance better than 0.6% for zero thickness, 0.1 ≤ W/h ≤ 10, 0.1 ≤ S/h ≤ 10, and 1 ≤ εr ≤ 18. ε_eff dispersion is about 1.4%. Impedance dispersion is under about 2.5%. Copper thickness and solder mask are outside that claim.

Jansen’s thickness correction is applied here even though the gap is narrower than 20 times the copper. Qucs only applies that correction when S > 20 t. At 5 mil versus 1.6 mil copper, this row is past that gate. The 0.6% figure does not cover it.

Symmetric edge-coupled stripline, if you switch the pair reference, uses Cohn’s 1955 zero-thickness map plus a Wheeler width correction. When the dielectric above and below differ, the number is still the centered-strip formula at the total ground spacing. It is not an offset-pair solution. Edge-coupled grounded CPW is not in this calculator.

These rows are a closed-form check of the published stackup at a gap you type. They are not the fab’s coupon, and they are not a replacement for the fab’s impedance calculator. The 50 Ω microstrip note uses the same JLC04161H-7628 preset for a single trace: 12.06 mil at 1 GHz.

References

• M. Kirschning and R. H. Jansen, “Accurate Wide-Range Design Equations for the Frequency-Dependent Characteristic of Parallel Coupled Microstrip Lines,” IEEE Trans. Microwave Theory and Techniques, vol. 32, no. 1, pp. 83–90, January 1984.
• Corrections, IEEE Trans. Microwave Theory and Techniques, vol. 33, no. 3, p. 288, March 1985.
• E. Hammerstad and Ø. Jensen, “Accurate Models for Microstrip Computer-Aided Design,” IEEE MTT-S International Microwave Symposium Digest, 1980.
• S. B. Cohn, “Shielded Coupled-Strip Transmission Line,” IRE Trans. Microwave Theory and Techniques, vol. 3, no. 5, pp. 29–38, October 1955. Used for the symmetric stripline pair.
• JLCPCB impedance stackup, including JLC04161H-7628 and JLC04161H-3313. Accessed 2026-10-11. https://jlcpcb.com/impedance
• JLCPCB impedance calculator guide, updated 16 September 2026. Finished 1 oz copper 1.6 mil, mask 0.6 mil at εr 3.8. Accessed 2026-10-11. https://jlcpcb.com/help/article/user-guide-to-the-jlcpcb-impedance-calculator

Related

• KiCad vs JLCPCB 50 Ω width
• 5 mil gap on JLC04161H-7628, 90 Ω
• 8 mil gap on the same stackup
• 5 mil gap, mask off
• JLC04161H-3313, 5 mil gap

FAQ

How wide is a 90 ohm USB pair on a JLCPCB 4-layer board?

On JLC04161H-7628, this calculator’s 90 Ω edge-coupled microstrip at 1 GHz with a typed 5 mil gap is 10.61 mil, 0.269 mm. The stackup is 7.1 mil of 7628 at εr 4.4, 1.6 mil finished copper, and 0.6 mil mask at εr 3.8. Zdiff is 90.00 Ω, Zodd is 45.00 Ω, Zeven is 67.13 Ω, and k is 0.197. An 8 mil gap on the same board is 12.28 mil, 0.312 mm. The 5 mil gap is not a JLCPCB minimum.

How much does the gap change the 90 ohm width?

On that JLC04161H-7628 stackup, moving the gap from 5 mil to 8 mil moves the 90 Ω width from 10.61 mil to 12.28 mil. k falls from 0.197 to 0.134. Mode skew falls from 0.431 ps/mm to 0.376 ps/mm. A tighter gap lowers Zdiff, so the width solved for 90 Ω gets narrower.

What does solder mask do to the pair?

The mask overlay is applied to the odd mode only. On the 5 mil JLC04161H-7628 row, mask on is 10.61 mil and mask off is 10.95 mil, 0.278 mm. Odd-mode delay with the mask is 5.862 ps/mm. Kirschning and Jansen do not include a solder mask.

Does this replace the fab coupon?

No. It is a closed-form check of the published stackup at a gap you type. Kirschning and Jansen state static impedance better than 0.6% for zero thickness inside 0.1 to 10 for W/h and S/h. This 5 mil row has 1.6 mil copper and a mask, and the gap is narrower than 20 times the copper, so that 0.6% figure does not cover it. Build from the fab’s coupon. JLC04161H-3313 at a 5 mil gap is 6.41 mil, 0.163 mm.