impedance_field_solvervia APIThe yield and the stack-up search run on Pro and API plans.
Added around the outer-layer copper; 0 for none.
εr and tanδ are the values at this frequency; the solver follows the material model to each listed frequency.
0 for smooth copper.
Up to 8 frequencies in GHz, separated by commas.
Where solve-for, the tolerance spread and the 0.1 % accuracy rule act.
Run history
Your runs appear here, kept in this browser.
Run the simulation to see its results here.
About this tool
Solve a PCB cross-section with a 2-D field solver: microstrip, stripline, coplanar waveguide and differential pairs, with trapezoidal traces, plating, solder mask and mixed dielectrics. Impedance or the full mode set of a pair, loss with copper roughness, and the width or spacing for a target impedance, each with its stated uncertainty.
How it works
The impedance calculators on this site use closed-form approximations, each fitted to one idealised geometry and accurate within the range its authors state. This tool solves the cross-section itself, so it handles what those formulas leave out: trapezoidal etched traces, plating, conformal solder mask, a layered dielectric stack, coplanar grounds beside a trace or a pair, and two traces on different layers.
The method. A transmission line on a PCB carries a quasi-TEM wave, so its impedance follows from the two-dimensional electrostatics of its cross-section. The solver places unknown charge on every conductor surface and on every boundary between two dielectrics, and solves for the charge that holds each conductor at its potential: a boundary-element method. Ground planes enter through image terms, so they need no panels. Solving once with every dielectric replaced by vacuum gives the inductance, and solving with the materials gives the capacitance. A single line's Z₀ and εeff follow from those two. For a pair, the inductance and capacitance matrices give the differential, common-mode, odd-mode and even-mode impedances, each mode's delay, and the impedance of one trace with its partner removed.
Accuracy. The mesh is refined until the stated uncertainty is no more than 0.1 %: the estimated distance of the reported figure from the converged value, judged from how the last refinements moved it. Every structure family is checked on every change against an exact solution, a published formula within its stated accuracy, or an independent solver. The Accuracy section lists the worst deviation found for each.
Materials. Each dielectric's εr and tanδ are entered at a reference frequency. By default they follow the Djordjevic-Sarkar wideband model to every listed frequency, so impedance, εeff and loss change with frequency as real laminates do. A material can also be marked constant.
Loss. Conductor loss comes from the solved field by Wheeler's incremental-inductance rule: how much the inductance changes when every metal surface recedes by a skin depth. It is multiplied by a roughness factor, from Hammerstad-Jensen (RMS roughness) or Huray (nodule radius and surface ratio). Dielectric loss comes from each material's loss tangent, weighted by the share of the field it holds. A frequency where the skin depth is more than a third of the copper thickness has its conductor loss flagged. So does any frequency where a dimension exceeds a tenth of a wavelength, where the quasi-static solve stops being valid.
Design aids. Solve-for finds the trace width, or a pair's spacing, that meets a target impedance at the design frequency within 0.1 %, or states the range that can be reached. The tolerance spread moves each fabrication parameter by its tolerance and reports how far the impedance moves: a first-order estimate, not a yield.
Accuracy
Each figure is the largest difference the reference suite found between this solver and an independent result for that structure: an exact solution, a published formula within its stated accuracy, or an independent solver. The suite runs on every change and fails if any figure is exceeded.
| Structure | Within | Worst measured | Against | Cases |
|---|---|---|---|---|
| Surface microstrip | 0.075 % | 0.059 % | Hammerstad & Jensen 1980, stated to 0.2 % | 15 |
| Embedded microstrip | 0.015 % | 0.011 % | method of moments (test-only mom2d) | 9 |
| Stripline | 0.0041 % | 0.0033 % | Cohn 1954, exact | 12 |
| 0.42 % | 0.33 % | Wheeler 1978, stated to 0.5 % | 9 | |
| Coplanar waveguide | 0.0029 % | 0.0022 % | Ghione & Naldi 1984 conformal map, exact | 3 |
| Grounded coplanar waveguide | 0.084 % | 0.067 % | finite elements, closed by a distant cover | 1 |
| Edge-coupled surface pair | 0.0054 % | 0.0043 % | method of moments (test-only mom2d) | 8 |
| Edge-coupled embedded pair | 0.0068 % | 0.0054 % | method of moments (test-only mom2d) | 8 |
| Edge-coupled stripline pair | 0.0025 % | 0.002 % | Cohn 1955, exact | 8 |
| Edge-coupled pair with coplanar ground | 0.046 % | 0.036 % | finite elements, closed by a distant cover | 2 |
| Edge-coupled pair in coplanar waveguide | 0.069 % | 0.055 % | finite elements, closed by distant planes | 2 |
| Broadside-coupled pair | 0.06 % | 0.048 % | finite elements | 4 |
| Conductor loss | 0.013 % | 0.01 % | Wheeler's rule by finite elements | 3 |
| Dielectric loss | 0.001 % | 0 % | exact, one material | 1 |
Impedance is compared for single lines and for each mode of a pair. A comparison against a published formula includes that formula’s own error, which its authors state.
References
Characteristic impedance of the shielded-strip transmission line · Shielded coupled-strip transmission line · Analytical formulas for coplanar lines in hybrid and monolithic MICs
- Characteristic impedance of the shielded-strip transmission line — S. B. Cohn, IRE Trans. MTT-2 (1954): the exact stripline the solver is held to
- Shielded coupled-strip transmission line — S. B. Cohn, IRE Trans. MTT-3 (1955): the exact edge-coupled stripline, both modes
- Analytical formulas for coplanar lines in hybrid and monolithic MICs — G. Ghione and C. Naldi, Electronics Letters 20(4) (1984): the exact coplanar waveguide with finite grounds
- Accurate models for microstrip computer-aided design — E. Hammerstad and Ø. Jensen, IEEE MTT-S Symposium Digest (1980): microstrip within its stated accuracy, and the roughness factor
- Transmission-line properties of a strip line between parallel planes — H. A. Wheeler, IEEE Trans. MTT-26 (1978): thick stripline within its stated 0.5 %
- Formulas for the skin effect — H. A. Wheeler, Proc. IRE 30 (1942): the incremental-inductance rule for conductor loss
- Wideband frequency-domain characterization of FR-4 and time-domain causality — A. R. Djordjevic, R. M. Biljić, V. D. Likar-Smiljanić and T. K. Sarkar, IEEE Trans. EMC 43(4) (2001): the material model
- Impact of copper surface texture on loss: a model that works — P. G. Huray et al., DesignCon (2010): the nodule roughness model
Every structure family is checked against an exact solution, a published formula within its stated accuracy, or an independent solver (method of moments, or finite elements in a closed box for structures with neither), on every change. The accuracy section lists the worst deviation found for each.
Frequently asked questions
6 questions
How is this different from the microstrip and differential pair calculators?
The calculators evaluate closed-form approximations, which are instant but fitted to idealised shapes: rectangular traces, one dielectric, no mask, no coplanar grounds. This tool solves the actual cross-section numerically, so etched trapezoids, plating, conformal solder mask, mixed dielectric stacks and coplanar grounds are all in the answer, and every result states its own uncertainty.
How accurate is it?
Every result is refined until its stated uncertainty is no more than 0.1 %, unless the free tier's size limit is reached first, in which case the result says so. Against exact solutions the solver agrees to a few thousandths of a percent; the Accuracy section lists the worst deviation measured for each structure and what it was measured against.
Why is the differential impedance not twice the single-ended impedance?
Because the two traces are coupled. The odd mode's field crowds into the gap between them, which lowers the differential impedance below twice the impedance of one trace alone. The result shows both, together with the uncoupled impedance. In stripline the coupling fades within a few plane spacings; a surface pair stays coupled much further apart.
Which roughness model should I use?
Hammerstad-Jensen needs only the RMS roughness and is the common default, but it saturates at a factor of two. Huray models the copper nodules themselves and keeps rising with frequency, which matches measurements on rough foils above a few GHz. Copper suppliers publish the nodule radius and surface ratio for their foils.
Why does the impedance change with frequency?
Laminates are dispersive: their permittivity falls slowly as frequency rises. With the Djordjevic-Sarkar model on, each listed frequency is solved with the materials as they are at that frequency, so impedance and εeff drift a little across a wide band. Mark a material constant to remove the effect.
What do I need to run it?
Nothing: a solve runs on the free tier without an account, after a short ad, up to 1,200 unknowns and 8 frequencies. Pro and API plans skip the ad, run on faster workers, and solve up to 4,000 unknowns and 64 frequencies. The same solve is available through the API, the Python SDK and the MCP server.