What this calculator does
A differential pair is two identical traces of width separated by a gap , driven with equal and opposite signals. Because the traces are close together, each one's field couples to the other. A coupled pair therefore has two characteristic impedances, one for each of its two natural modes:
- Odd mode (equal and opposite currents, i.e. a differential signal): the field concentrates in the gap, so the capacitance to ground rises and falls below the single-line value.
- Even mode (equal currents, i.e. common-mode noise): the traces shield each other, so rises.
The quantities engineers specify follow directly from these:
USB, PCIe, Ethernet and HDMI all specify (typically 85–100 Ω). Common-mode impedance matters for termination and EMI, and the coupling coefficient tells you how tightly coupled the pair is.
The models
Edge-coupled microstrip. The calculator uses the Kirschning–Jansen (1984) equations for the static even- and odd-mode and . They are corrections to the Hammerstad–Jensen single-line model, which the calculator reuses from the microstrip calculator. Because part of the field is in air, the modes travel at different speeds. The odd mode has more of its field in the air above the gap, so and the odd mode is faster. That difference is the main source of far-end crosstalk in microstrip.
For copper thickness, Jansen's (1978) split of the width increment (an extra sidewall term for the odd mode) is combined with Hammerstad–Jensen's single-line approach: one width for and an air width for the reduction of , applied mode by mode. This combination is our own construction rather than a single published model, so the accuracy table below reports how it compares with a field solver.
Edge-coupled stripline. For zero thickness, Cohn's (1955) conformal mapping is exact:
Finite thickness uses Cohn's fringing-capacitance correction, built on the single thick strip from the stripline calculator. The odd mode switches to a separate formula for , which adds the parallel-plate field between the strip sidewalls. In a homogeneous stripline both modes have , so there is no far-end crosstalk from mode-velocity mismatch.
How to use solve-for
Most designs start from a target with the stack-up fixed. Click ⇄ Solve next to Trace spacing and enter 100 to find the gap that gives 100 Ω at the current width. You can also solve for the width at a given spacing. Solving for is usually the more practical choice, since fabricators prefer a minimum trace width. You can also target , or , or back out from a measured .
If no solution exists, the target is out of reach at that width. can never exceed twice the single-line impedance (the limit ), so the width has to come down first.
Worked example
An outer-layer pair on a 0.1 mm (4 mil) prepreg with , mm, mm and 1 oz copper (m), as microstrip (the defaults):
- Ω, Ω, Ω and Ω, with .
- and , so the differential (odd) delay is 131.21 ps/in and the common-mode delay is 147.77 ps/in.
- For exactly 100 Ω, solve for the gap: mm. Alternatively, keep and widen the traces to mm.
The same traces as a centred stripline with mm give Ω and Ω, with both modes at 169.45 ps/in. Here 100 Ω needs mm or mm. The stripline pair couples more tightly (). Each plane is about 0.2 mm away, twice the microstrip's 0.1 mm, so the neighbouring trace takes a larger share of the field.
Accuracy and validity
We compared the models with an independent 2-D finite-volume field solver (tooling/validation/laplace2d.py), which
reproduces Cohn's exact coupled stripline to better than 0.05 % and Hammerstad–Jensen microstrip to about 0.15 %.
| Geometry | Range | Error vs. field solver |
|---|---|---|
| Microstrip, | , , | within 0.4 %, within 0.4 % |
| Microstrip, | , | within ≈ 3 %, within ≈ 2.6 % |
| Microstrip, , narrow, high | , | +7 %, −13 % (warned) |
| Stripline, | any | exact (Cohn) |
| Stripline, , | −0.8 % … +1.3 % | |
| Stripline, , | up to +3 % | |
| Stripline, , | up to +5 % (warned) |
Kirschning–Jansen's published range is , and , with a stated accuracy of about 1 % for zero thickness. As the microstrip model reduces to the isolated line, which we checked against scikit-rf's independent Hammerstad–Jensen implementation to 0.02 %. Results are quasi-static: they do not include dispersion, which slightly raises both values above a few GHz.
Practical notes
- Solder mask lowers . On outer layers a 15–25 µm mask fills the gap with and typically lowers by 2–6 Ω. Fabricators' impedance tables include this, and this calculator does not.
- Tight versus loose coupling. Tightly coupled pairs () keep stable through neck-downs and reject common-mode noise better, but they are more sensitive to etching of the gap. Loosely coupled pairs () behave like two independent 50 Ω lines.
- Odd and even delays differ on microstrip. Here the difference is about 16 ps/in. Over long buses this produces far-end crosstalk and mode conversion. Stripline does not have this problem.
- Intra-pair skew comes from length mismatch and glass-weave effects, not from the impedance. Match the lengths of the two traces to a few mils at multi-Gb/s rates.
- Thick copper pulls down, because the field between the sidewalls is in air. On outer layers with plated copper (), confirm the final geometry with a field solver.