Trace impedance
A trace-impedance study computes the transmission-line parameters of a printed circuit board trace from its cross-section: the characteristic impedance , the effective permittivity , the propagation delay per length, and the per-length capacitance and inductance matrices. For a differential pair it also gives the differential and common-mode impedances, the odd- and even-mode impedances and the near- and far-end coupling coefficients. The cross-section comes from the board stackup: layer thicknesses and Dk, copper thickness, trace width and etch angle, solder mask and reference planes.
The study is an electromagnetic line_parameters analysis on a cross_section geometry. It
publishes these outputs:
| Output | Symbol | Unit |
|---|---|---|
impedance | per signal conductor | Ω |
effective_permittivity | per signal conductor | 1 |
delay | per length | s/m |
capacitance, vacuum_capacitance | , | F/m |
inductance | H/m | |
differential_impedance, common_mode_impedance, odd_mode_impedance, even_mode_impedance | , , , | Ω |
odd_mode_effective_permittivity, even_mode_effective_permittivity | modal of a pair | 1 |
odd_mode_delay, even_mode_delay | modal of a pair | s/m |
near_end_coupling, far_end_coupling | , | 1 |
This version is lossless. Conductor loss from skin depth and dielectric loss from the loss tangent (resistance and conductance per length, and attenuation) are a separate output Not yet releasedStatus.
A 50 Ω microstrip and a 100 Ω pair
The study below sizes a single-ended microstrip and an edge-coupled differential pair on the
top layer of a four-layer board. The top copper is 35 µm (1 oz) under 20 µm of solder mask,
on 0.2 mm of prepreg (Dk 4.1) over the ground plane on In1.Cu. The etch leaves 60° sidewalls.
geometry single_ended {
cross_section {
layer "F.Mask" : solder_mask { thickness = 20 um; permittivity = 3.8 }
layer "F.Cu" : signal { thickness = 35 um }
layer "dielectric 1" : dielectric { thickness = 0.2 mm; permittivity = 4.1 }
layer "In1.Cu" : plane { thickness = 35 um }
trace line : signal { layer = "F.Cu"; width = 0.36 mm; etch_angle = 60 deg }
}
}
geometry pair {
cross_section {
layer "F.Mask" : solder_mask { thickness = 20 um; permittivity = 3.8 }
layer "F.Cu" : signal { thickness = 35 um }
layer "dielectric 1" : dielectric { thickness = 0.2 mm; permittivity = 4.1 }
layer "In1.Cu" : plane { thickness = 35 um }
trace p : signal { layer = "F.Cu"; center = -0.2 mm; width = 0.23 mm; etch_angle = 60 deg }
trace n : signal { layer = "F.Cu"; center = 0.2 mm; width = 0.23 mm; etch_angle = 60 deg }
}
}
mesh single_ended on geometry.single_ended { triangles {} }
mesh pair on geometry.pair { triangles {} }
study microstrip {
geometry = geometry.single_ended
mesh = mesh.single_ended
analysis line { electromagnetic line_parameters {} }
output z0 = analysis.line.impedance.line
output eps_eff = analysis.line.effective_permittivity.line
output delay = analysis.line.delay.line
}
study differential_pair {
geometry = geometry.pair
mesh = mesh.pair
analysis lines { electromagnetic line_parameters {} }
output z_diff = analysis.lines.differential_impedance
output z_common = analysis.lines.common_mode_impedance
output next = analysis.lines.near_end_coupling
output fext = analysis.lines.far_end_coupling
output c = analysis.lines.capacitance
output l = analysis.lines.inductance
}runmat run trace-impedance.fea --study microstrip reports close to 50 Ω and
--study differential_pair reports close to 100 Ω. To hit a target, change
the width (and, for the pair, the gap) and run again: each solve takes about a second.
Describing the cross-section
A cross_section block lists the stackup from the top of the board down, then the traces.
| Layer role | Fields | Meaning |
|---|---|---|
dielectric | thickness, permittivity | core or prepreg with its Dk at the frequency of interest |
solder_mask | thickness, permittivity | conformal mask over the outer copper next to it: thickness deep on the laminate between traces and over each trace's top and sidewalls |
signal | thickness, optional fill_permittivity | the routed copper layer; between traces it is filled with mask or air on an outer layer and with the dielectric directly above on an inner layer, unless fill_permittivity says otherwise |
plane | thickness | a full reference plane at 0 V |
| Trace field | Meaning |
|---|---|
role signal or ground | a line whose parameters are reported, or coplanar ground copper held at 0 V |
layer | the copper layer it is etched on; every trace of one section is on the same layer |
center | lateral position of the trace centre (default 0) |
width | width of the base face, the face bonded to the laminate |
etch_angle | sidewall angle from the base face (default 90°); the far face is narrower by |
base = top or bottom | which face is the base; by default the laminate side (bottom on the top layer, top on the bottom layer, bottom on an inner layer) |
The domain lies between the reference planes nearest the signal layer. Layers beyond those
planes do not affect the result, so a section may list the stackup only down to (or up to)
the plane. An outer layer without a plane on its open side is surrounded by air: the section
extends 60 dielectric heights to the sides and above, with insulating walls, which changes
by less than 0.05 %. enclosure { half_width = …; air = … } sets other distances. A line
needs a return conductor: a reference plane or ground traces.
Taking the stackup from the board
stackup = "./board.circuit.layout" reads the layer list (names, thicknesses, Dk and copper
weights) from the board's layout file, so the section and the board cannot disagree. The
traces then name the layout's copper layers, for example layer = "F.Cu". A layout with
several stackups (rigid-flex) names one with stackup_name = rigid.
Broadside-coupled lines
Conductors on two copper layers of one section, such as a broadside-coupled pair Not yet releasedStatus, solve as one multiconductor line.
How the parameters are computed
The line is quasi-TEM: its cross-section is small compared with the wavelength, so the transverse field at each instant satisfies the 2D electrostatic equation on the section,
with prescribed on the conductors and zero normal flux on the enclosure walls.
Capacitance. Holding signal conductor at 1 V and every other conductor at 0 V gives the potential . The Maxwell capacitance matrix per length is the energy form
symmetric, with positive diagonal and negative off-diagonal terms. A second solve with every gives .
Inductance. Copper and laminate are non-magnetic, so the inductance matrix follows from the vacuum capacitance: .
Single line. , and per metre.
Several lines. The characteristic impedance matrix of the lossless multiconductor line is
.
impedance is its diagonal, the impedance of each line with the others held at 0 V. For a
pair driven differentially, , and in common, :
and the modal effective permittivity is . On a microstrip the odd mode travels faster than the even mode, because more of its field is in air.
Coupling. With the mutual terms and against and :
A step of with rise time on one line of a coupled length gives a near-end voltage on the quiet line (once ) and a far-end pulse of about . In a homogeneous dielectric (stripline) ; on microstrip is negative because inductive coupling dominates.
Discretization. The mesher builds a graded triangle mesh whose lines follow every layer interface, the mask surfaces and the trace edges, including sloped sidewalls. Elements are smallest at trace edges and corners, where the field is singular, and grow geometrically away from them. On linear triangles the element matrix is
with the area and , the corner coordinate differences. Conductor nodes are prescribed; one sparse Cholesky factor of the free block serves every signal conductor as a right-hand side, on the admitted CPU threads.
| Mesh field | Default | Meaning |
|---|---|---|
edge_size | smallest feature / 1000 | element size at trace edges and corners |
growth | 1.1 | size ratio between neighbouring elements |
max_size | twice the dielectric height | largest element, reached in the far field |
The smallest feature is the least of the copper thickness, trace widths, gaps, dielectric and mask thicknesses. A section that would not fit the run's memory budget is rejected with its size and the budget.
Accuracy
Against Hammerstad and Jensen's microstrip forms the default mesh gives and within 0.5 %. Against Cohn's exact stripline and coupled-stripline solutions it gives each impedance within 0.3 %. Refining the edge size converges to better than 0.1 %. The electromagnetic validation page lists these checks with their problems, tolerances and results, including agreement with the IPC-2141A forms inside their stated ranges.
The parameters are quasi-static. They hold while the cross-section is small compared with the wavelength in the dielectric, which covers board traces well into the gigahertz range; dispersion of at higher frequencies and loss are not included.