Microstrip vs. grounded coplanar waveguide: choosing a 50 Ω reference line
When you evaluate an RF substrate, characterize a probe station, or build a fixture for a switch or amplifier, the first thing you need is a known-good transmission line. Two planar geometries dominate thin-film and PCB work: microstrip and grounded coplanar waveguide (GCPW). They both aim for a controlled 50 Ω, but they behave very differently as frequency, mounting, and component density change.
The two geometries
A microstrip line is a single signal trace on the top of a dielectric with a continuous ground plane on the back. Fields live partly in the substrate, partly in air, so the effective dielectric constant is below the bulk value and dispersion is real above a few gigahertz.
A grounded coplanar waveguide puts the signal trace in the centre of the top face with coplanar ground pours on both sides, a controlled gap between signal and ground, and the backside ground plane retained. The signal is therefore "sandwiched" laterally and vertically.
Insertion and return loss
A well-made thin-film microstrip on 200 µm alumina (εr 9.8) supports 50 Ω lines around 170–180 µm wide and stays usable past 100 GHz; measured S11 better than −16 dB across 1–110 GHz is typical for short straight standards, with the worst match usually at the launch, not the line. Bends degrade the match: a same-side double bend may show −17 dB worst case, while a carefully mitred single bend can stay around −19 dB.
GCPW ties the side grounds to the backside plane with vias, suppressing the parallel-plate mode that otherwise ruins high-frequency behaviour. On fused silica (εr 3.78), a 200 µm-wide line with a 25 µm gap over 400 µm substrate delivers measured return loss better than −29 dB across DC–40 GHz. The wider effective geometry and enclosed field make GCPW intrinsically easier to match over a broad band.
Five practical decision factors
1. Component mounting
GCPW wins when parts mount on the top surface. The side grounds provide a nearby return for shunt elements, DC blocks, and bias tees without dropping a via right next to the pad. Microstrip needs plated or wrapped grounds beside every shunt component.
2. Density and isolation
GCPW side grounds isolate adjacent lines from each other, so two GCPW feeds can run close together with modest crosstalk. Microstrip relies entirely on spacing; dense arrays of microstrip lines couple, especially on thick substrates.
3. Dispersion and modelling
Microstrip is quasi-TEM: effective εr and impedance drift with frequency, and simple calculators stop being accurate above roughly 10 GHz. GCPW is closer to pure TEM when the grounds are properly stitched, so broadband models and fixture de-embedding are more forgiving.
4. Fabrication tolerance
A 50 Ω microstrip on high-εr ceramic is a narrow line where a few micrometres of etch tolerance move impedance noticeably. On low-εr glass or fused silica, GCPW lines are wider and impedance is shared between trace width and gap, giving designers two knobs to land on 50 Ω with the available process.
5. Launch design
Microstrip launches from coax or probe are well understood, but the transition concentrates field at the edge of the ground plane. GCPW launches naturally from ground-signal-ground probes and edge-mount connectors; this is why V-band and W-band probe standards and calibration substrates are overwhelmingly coplanar.
When to pick which
| Situation | Preferred |
|---|---|
| Probe station calibration, TRL/TRL standards | GCPW |
| Simple, lowest-cost interconnect on ceramic above 40 GHz | Microstrip |
| Dense array with shunt-mounted components | GCPW |
| Maximum bandwidth for a 50 Ω reference / test fixture | GCPW |
| Lowest conductor loss for a fixed impedance and substrate thickness | Microstrip (wider effective trace) |
| Co-packaged or glass-interposer feed with vias nearby | GCPW |
| Quick filter/resonator prototyping on alumina | Microstrip |
Bends matter more than people expect
If a microstrip has to turn, the bend is a real discontinuity. Right-angle corners add capacitance and excite the first mismatch in the return-loss curve; unmitred bends on 176 µm lines can cost several dB of S11 at the low end. Best practice is to chamfer (mitre) the outside corner, keep bend count minimal, and never place a bend closer than a substrate thickness to a launch. Our thin-film line family reflects this directly: the straight standard (TL-A) is the reference, the single mitred bend (TL-D) is the cleanest turn, and the same-side double bend (TL-B) is the least forgiving.
Reference standards on thin film
For fixture and calibration work, two productised standards are useful:
- Thin-film alumina microstrip (1–110 GHz) — straight, single-bend, and double-bend configurations in 50 Ω, Au metalization on 200 µm alumina, for band-edge and launch characterisation into W-band.
- Fused-silica GCPW (DC–40 GHz) — 200 µm signal, 25 µm gap on 400 µm fused silica with continuous backside Au ground, for broadband 50 Ω reference use and fixture embedding.
Fused silica is particularly attractive for the GCPW role: tanδ near 0.00005 at 30 GHz, near-zero CTE, and dimensional stability that keep the reference line repeatable across temperature.
Bottom line
Choose microstrip for the simplest, lowest-loss straight run on high-frequency ceramic, especially above 40 GHz where its geometry is proven. Choose grounded coplanar waveguide whenever the board is dense, the launch is a GSG probe or edge connector, the substrate is low-εr glass, or you need a broadband 50 Ω reference that stays matched without heroic modelling. And whichever geometry you use, treat the launch and the first bend — not the long straight middle — as the parts that decide your measured S11.