Microstrip vs stripline: how to choose and how to calculate impedance
Every high-speed net on a board, from a USB pair to a DDR bus to an RF feed, lives or dies on one number: characteristic impedance. Pick the wrong trace structure or size it with the wrong formula and the line reflects its own signal back, which shows up as ringing, overshoot and, on fast links, a closed eye and intermittent failures that are miserable to debug. Almost all controlled-impedance routing uses one of two structures, the microstrip or the stripline. This guide explains how they differ, when to use each, the one input people get wrong, and why the formula behind your calculator matters more than most people realise. Where it helps, it ties back to the calculator so you can put real numbers to it.
The two structures at a glance
A microstrip is a trace on an outer layer with a single reference plane beneath it. A stripline is a trace on an inner layer, sandwiched between two planes. That one structural difference drives everything else: how the fields behave, how much the line radiates, how easy it is to route and probe, and how wide the trace has to be for a given impedance.
| Property | Microstrip | Stripline |
|---|---|---|
| Layer | Outer | Inner |
| Reference planes | One, below | Two, above and below |
| Field path | Part in air, part in board | Entirely in the dielectric |
| Effective permittivity | Between air and the laminate | The full dielectric constant |
| Radiation and EMI | Higher | Lower, shielded by planes |
| Crosstalk | Higher | Lower |
| Routing and probing | Easy, on the surface | Needs inner layers, hard to probe |
| Width for the same Z0 | Wider | Narrower |
| Best for | RF feeds, tight layer counts, surface routing | The fastest, most sensitive nets |
When a trace stops being a wire
The trigger is the signal's rise time, not its clock frequency. When the time it takes a signal to travel down a trace becomes an appreciable fraction of its rise time, the trace behaves as a transmission line and impedance matching matters. Because a square edge carries energy well above the fundamental, a modest clock with sharp edges acts like a much faster signal. In practice that means every multi-gigabit serial link, every DDR bus and most RF routing.
Characteristic impedance is fixed by the cross-section: trace width, the height to the nearest plane, the copper thickness and the dielectric constant. It does not depend on length. Length changes delay and loss, not the impedance.
The number people get wrong: H is to the plane, not the board
The single most common reason a calculated impedance does not match the finished board is the dielectric height. The height that sets impedance is the distance from the trace to its nearest reference plane, read from the layer stackup. It is not the overall board thickness. On a typical four-layer board the signal-to-plane height is a small fraction of the total thickness, so using board thickness can be off by a factor of several and hand you an impedance that looks plausible and is completely wrong. Pull the height from the stackup every time.
Why the formula behind the calculator matters
Here is the part most articles skip. Plenty of online calculators use the IPC-2141 microstrip equation because it is short and easy to code. It is also known to be inaccurate, and it gets worse the wider the trace. The Hammerstad-Jensen equations, which our calculator uses, stay within about one percent across the practical range. The table below is computed for FR-4 (er 4.3, thin copper) and shows the two side by side as the width-to-height ratio grows.
| W / H | Hammerstad-Jensen | IPC-2141 | Error |
|---|---|---|---|
| 0.5 | 96.4 Ω | 98.5 Ω | +2% |
| 1.0 | 71.8 Ω | 73.2 Ω | +2% |
| 2.0 | 49.4 Ω | 48.0 Ω | -3% |
| 3.0 | 38.0 Ω | 33.2 Ω | -13% |
| 4.0 | 31.0 Ω | 22.8 Ω | -27% |
| 5.0 | 26.2 Ω | 14.6 Ω | -44% |
| 7.0 | 20.1 Ω | 2.4 Ω | -88% |
| 8.0 | 18.0 Ω | -2.5 Ω | negative |
| 10.0 | 14.9 Ω | -10.6 Ω | negative |
For stripline the IPC-2141 closed form is accurate and still in common use, including in our tool. The takeaway is not that IPC-2141 is bad, it is that the right formula depends on the structure, and a calculator that uses one equation for everything will quietly mislead you on wide microstrips.
Common impedance targets
Most controlled-impedance work comes down to a handful of recurring targets. These are the ones worth memorising.
| Interface | Target | Type |
|---|---|---|
| RF, many digital lines | 50 Ω | Single-ended |
| Video, some RF | 75 Ω | Single-ended |
| USB 2.0 / 3.0, HDMI | 90 Ω | Differential |
| Ethernet, PCIe, SATA | 100 Ω | Differential |
| DDR (varies) | 40 / 80 Ω | Single / differential |
Differential pairs, and the limit of formulas
A differential pair carries equal and opposite signals on two coupled traces. The differential impedance is the single-ended impedance of one trace adjusted for the coupling between the pair, which the spacing controls: tighter spacing means more coupling and a lower differential impedance. This is where closed-form models reach their limit. Published differential equations disagree with each other by ten to twenty-five percent for the same geometry, a wider spread than single-ended formulas show. Use an analytical differential number to get close, then confirm the final width and spacing against your fabricator's stackup and field solver, which account for the real-world effects below and are the authoritative answer. Our calculator reports the differential, odd, even and common-mode values and says this plainly in the result.
When to reach for coplanar waveguide
At higher RF and microwave frequencies, a grounded coplanar waveguide (CPWG) is often the better choice. It places the signal trace between two coplanar ground pours, with a plane underneath, so the fields are tightly confined and the gap to the adjacent ground becomes a tuning knob alongside width and height. It also lets you route without via transitions that would otherwise break the impedance. The calculator includes a CPWG mode that uses the conductor-backed coplanar equations with exact elliptic integrals, rather than a rough approximation.
How the board house verifies it
Fabricators that offer impedance control build a test coupon on the panel and measure it with a time-domain reflectometer (TDR), which sends a fast edge down the line and reads impedance from the reflection along its length. The methods for characterising and reporting those measurements are standardised in IEEE 370. The practical consequence is that you should specify a target impedance and a tolerance, not a fixed trace width: the fab adjusts the width to hit your target on their actual materials and reports the measured result. Closed-form calculators ignore solder mask, copper surface roughness, trapezoidal etching and frequency dispersion, so expect a few percent difference on single-ended lines and more on differential pairs.
Putting it to work
Choose microstrip or stripline from the routing and shielding needs, take the trace-to-plane height from the stackup, size the width with an accurate model, and treat differential and final production numbers as something the fabricator confirms. For the sizing step, the PCB impedance calculator handles microstrip, stripline, edge-coupled differential pairs and CPWG, and can solve directly for the width at a target impedance. A high-speed trace usually has to satisfy current and heat limits as well, which is what the PCB trace width calculator covers.