Microstrip vs GCPW vs Stripline

On 1.6 mm FR-4 at 1 GHz, 1 oz copper, no solder mask, a 50 Ω microstrip is 3.008 mm wide. A grounded coplanar waveguide with a 6 mil gap and a 100 mil pour is 1.072 mm. A coplanar waveguide with the same gap and no backside ground is 1.321 mm. A symmetric stripline with 0.8 mm of FR-4 on each side of the strip is 0.678 mm, and it is slower: 177.72 ps/in against the microstrip’s 154.96 ps/in. On a 0.2104 mm four-layer prepreg the microstrip is already 0.374 mm, and GCPW with a 5 mil gap only gets to 0.344 mm. Use the coplanar gap when the microstrip will not fit. Differential 90 Ω and 100 Ω pairs are not calculated.

Which line is narrow enough on 1.6 mm FR-4?

The wide microstrip is the whole complaint. A 3 mm trace does not fit between the pads of a small radio, and it does not look like the picture in a module datasheet that was drawn on a thin core. The return path is the entire backside plane, so the only knobs are width, height, and εr. Height is the board. εr is the laminate. Width is what is left, and it comes out at 118.43 mil.

GCPW puts a slot on each side of the strip and a ground behind it. The slot capacitance lets the strip stay at 42.21 mil, 1.072 mm, for the same 50 Ω. You pay a via fence, because the model assumes the coplanar grounds and the backside plane are the same potential. How much the gap and the pour move Z0, and what the fence does not do, is does a via fence change GCPW impedance?.

CPW without the backside ground, Wadell’s form, lands at 52.00 mil, 1.321 mm, for the same 6 mil gap. $W/h$ stays inside 0.1 to 10. It is a real option on a thick core when you do not have a plane, or when you do not want to stitch one. The return current is in the coplanar pours. Break those pours and the impedance is no longer this number.

Stripline at 26.69 mil is the narrowest of the four, because both sides are dielectric and εeff is εr, 4.400. A standard four-layer board does not give you this layer. Layer 2 and layer 3 are the planes, and there is no signal between them. The 0.8 mm figure is a strip centered in a 1.6 mm dielectric, the construction you build if you mean to, not the inner layers of a 7628 stackup. Parts do not sit on a buried strip without a via.

At 1 GHz this model’s loss is 0.098 dB/in on the microstrip and 0.160 dB/in on the GCPW. The GCPW number is higher because more of the current rides the slot edges. Radiation is not in either number. At millimeter-wave frequencies a full-wave ranking can flip. Do not use the 1 GHz column to settle a 40 GHz argument.

Does GCPW still help on a 4-layer prepreg?

Less. Over 0.2104 mm of FR-4 the microstrip is 14.72 mil, 0.374 mm. GCPW with a 5 mil gap and a 40 mil pour is 13.54 mil, 0.344 mm. You still owe the fence, and the etch of a 5 mil gap moves Z0 as much as the small width saving. I would stay with microstrip on that prepreg unless I need the ground next to the trace for a shunt part.

The stripline row on that board is 5.83 mil, 0.148 mm, with 0.2104 mm of dielectric on each side. That is a symmetric pair of grounds, not “the inner layer of a four-layer board.” Loss at 1 GHz is 0.439 dB/in, mostly the FR-4. εeff stays 4.400. If the delay budget cares, read effective permittivity and trace delay before you bury the run.

CPW with a 5 mil gap on this thin core solves to a width with $W/h$ above 10. The table marks it. I would not build that width from this closed form. The substrate is too thin for an unbacked coplanar line to reach 50 Ω inside the model’s range.

What changes on 10 mil RO4350B?

This is the microwave core, εr 3.66 in this library (the design Dk, not every process Dk on a Rogers sheet), 0.5 oz, 10 mil.

Microstrip is 21.18 mil. GCPW with a 6 mil gap and a 40 mil pour is 18.93 mil. The gap is not buying you a routing channel. It buys a ground next to the strip and a launch that already looks like a probe. Stripline with 10 mil on each side is 9.83 mil, εeff 3.660, delay 162.09 ps/in against the microstrip’s 143.25 ps/in.

Unbacked CPW on this core, same 6 mil gap, also leaves $W/h > 10$. Same warning as the thin FR-4 row. The number is in the table so you can see the solver did not refuse. Do not treat it as a tape-out width.

Losses at 1 GHz on RO4350B are a few tenths of a dB per inch or less in this model. The finish, if the line is on the outer layer, is a separate question. Nickel is ENIG versus bare copper loss. It does not pick the topology.

What about CPW with no backside ground?

Use it when there is no plane and the solved $W/h$ stays between 0.1 and 10. On the 1.6 mm FR-4 row, with a 6 mil gap, it does. The impedance is Wadell’s elliptic form: $k = W/(W+2S)$, a finite-height correction $k_1$ from the sinh mapping, and $Z_0 = 30\pi / \sqrt{\varepsilon_{\mathrm{eff}}} \cdot K'(k)/K(k)$. There is no backside term. Adding a plane without stitching it is not a small edit. That structure is GCPW, and an unstitched plane launches a parallel-plate mode the CPW formula does not describe.

What this comparison is not

It is not a 100 Ω differential pair, and it is not a 90 Ω USB pair. Those need an odd-mode solve this calculator does not have. It is not etch compensation. It is not the loss at 28 GHz. The 1 GHz loss column is conductor plus dielectric only.

GCPW Z0 here is the zero-thickness Ghione–Naldi map. Copper weight changes the microstrip width and the loss. It does not change the GCPW width. Stripline uses Cohn’s zero-thickness Z0 plus the Wheeler / IPC-2141 thickness correction, and the mask is forced off because the strip is buried.

If the line has to be 50 Ω and the microstrip fits, build the microstrip. Fewer vias, less gap etch, and the current is mostly on the copper face against the laminate.

References

• E. Hammerstad and Ø. Jensen, “Accurate Models for Microstrip Computer-Aided Design,” IEEE MTT-S International Microwave Symposium Digest, 1980.
• B. C. Wadell, Transmission Line Design Handbook. The unbacked CPW elliptic integrals.
• G. Ghione and C. U. Naldi, “Coplanar Waveguides for MMIC Applications: Effect of Upper Shielding, Conductor Backing, Finite-Extent Ground Planes, and Line-to-Line Coupling,” IEEE Trans. Microwave Theory Tech., vol. 35, no. 3, pp. 260–267, Mar. 1987.
• S. B. Cohn, “Characteristic Impedance of the Shielded-Strip Transmission Line,” IRE Trans. Microwave Theory Tech., vol. 2, no. 2, pp. 52–57, 1954. Zero-thickness symmetric stripline. The thickness term used here is the Wheeler / IPC-2141 effective-width correction.

Related

• 50 Ω microstrip width on 1.6 mm FR-4
• Does a via fence change GCPW impedance?
• Effective permittivity and trace delay
• GCPW calculator

FAQ

Should I use microstrip, GCPW, or stripline?

On 1.6 mm FR-4 at 1 GHz, 1 oz, no mask, a 50 Ω microstrip is 3.008 mm wide. GCPW with a 6 mil gap and a 100 mil pour is 1.072 mm. CPW with the same gap and no backside ground is 1.321 mm. A symmetric stripline with 0.8 mm of FR-4 on each side is 0.678 mm and is slower, 177.72 ps/in against 154.96 ps/in. Use GCPW when the microstrip is too wide. On a 0.2104 mm prepreg the microstrip is already 0.374 mm and a 5 mil-gap GCPW is 0.344 mm, so the gap buys little.

When is coplanar waveguide without a ground plane a bad solve?

When the 50 Ω width leaves 0.1 < W/h < 10. On 10 mil RO4350B with a 6 mil gap, and on 0.2104 mm FR-4 with a 5 mil gap, the Wadell CPW width is outside that window. Those rows are labeled in the table. Do not build them from this closed form.

Does this calculator do 90 ohm or 100 ohm differential pairs?

No. It solves single-ended microstrip, stripline, CPW, and grounded coplanar waveguide. A 100 Ω pair is not a 50 Ω line with a second trace next to it in this model.

Which loss number should I trust?

The loss column is conductor plus dielectric loss in the quasi-TEM model at 1 GHz. On 1.6 mm FR-4 that is 0.098 dB/in for the microstrip and 0.160 dB/in for the GCPW. Radiation, via fences, and etch are not in it. At millimeter-wave frequencies a full-wave model can rank GCPW and microstrip differently.