Curved Trace
Routing RF Signals Around a Bend Without Reflections
Every transmission line that has to change direction on a circuit board introduces a discontinuity. A square 90° corner is the worst case: the extra copper at the outer apex stores charge and behaves as a small shunt capacitance, usually a few tens of femtofarads up to roughly 0.15 pF for typical 50-ohm microstrip widths (larger on wide, low-permittivity boards). That capacitance produces a local impedance dip and a reflection whose magnitude rises with frequency, so a corner that looks harmless on a DC schematic can drop return loss below 20 dB above 20 GHz. The curved trace solves the problem geometrically. By sweeping the conductor through a constant-radius arc, the layout removes the apex altogether, the surface current follows a continuous path, and the local line width seen by the wave stays close to its straight-section value.
The governing parameter is the ratio of bend radius to trace width. As the centerline radius grows relative to the width, the inner and outer edges of the arc carry more nearly equal path lengths and the field crowding along the inner edge falls off. Once the radius reaches about three times the trace width, the bend is electrically almost invisible: the residual impedance change is on the order of a percent and the added insertion loss is dominated by ordinary conductor and dielectric loss over the arc length, not by the bend itself. Tighter curves are sometimes unavoidable in dense layouts, and even a radius of 1.5 to 2 times the width is far superior to a square corner, but designers verify those cases with a 2.5D electromagnetic solve before committing to fabrication.
Curved routing also matters for phase. The arc adds physical length equal to one quarter of its circumference per 90° turn, and that length must be counted when equalizing matched pairs or balancing the arms of a feed network. A curve and a mitered corner occupying the same board footprint do not have the same conductor length, so substituting one for the other late in a layout can detune a carefully tuned distribution network.
Bend Geometry and Impedance Equations
R ≥ 3 × W (centerline radius, microstrip)
Added length per 90° arc:
ΔL = (π × R) / 2
Added electrical phase:
Δφ = (360° × ΔL × √εeff) / λ0
Square-corner excess capacitance (approx.):
Ccorner ≈ W × (1.6εr + 1.0) × 0.01 pF/mm
Where W = trace width, R = centerline bend radius, εeff = effective permittivity, λ0 = free-space wavelength. Example: W = 0.2 mm → R ≥ 0.6 mm; a 90° arc at R = 0.6 mm adds ΔL ≈ 0.94 mm, about 18.6° at 10 GHz on εeff ≈ 2.7.
Bend Style Comparison
| Bend Style | Return Loss @ 20 GHz | Excess Reactance | Board Area | Length Predictability | Best Use |
|---|---|---|---|---|---|
| Square 90° corner | 15 to 20 dB | High (up to ~0.15 pF) | Smallest | Poor | DC and low-freq only |
| 45° chamfer | 22 to 28 dB | Moderate | Small | Fair | Routine digital and sub-6 GHz |
| Optimized mitered bend | 28 to 34 dB | Low (tuned) | Small | Fair | Narrowband mmWave |
| Curved trace (R ≥ 3W) | > 35 dB | Very low | Larger | Excellent | Wideband feeds, matched pairs |
| Curved trace (R ≈ 1.5W) | 26 to 32 dB | Low | Compact | Good | Space-constrained mmWave |
Frequently Asked Questions
What bend radius should a curved RF trace use to stay impedance-continuous?
A common rule keeps the centerline radius at three or more times the trace width (R ≥ 3W), holding the impedance perturbation below about 1% and return loss above 30 dB through mmWave. For a 50-ohm microstrip 0.2 mm wide on 0.13 mm RO4350B, that is roughly 0.6 mm. Tighter radii crowd current on the inner edge and raise loss; if area is tight, R near 1.5W to 2W is still far better than a square corner, but confirm S11 with a 2.5D solve.
Why does a curved trace outperform a sharp right-angle corner at high frequency?
A square corner leaves excess metal at the outer apex that acts as a shunt capacitance of roughly a few tens of femtofarads up to about 0.15 pF, creating an impedance dip and a reflection that worsens with frequency, dropping return loss to about 20 dB above 20 GHz. A curve removes the apex so current follows a smooth path and local impedance stays near 50 ohms, and it radiates less. A 45° chamfer helps similarly, but a true constant-radius arc gives the lowest reflection and most predictable phase.
How much does a curved trace add to electrical length, and does it matter for matched pairs?
A 90° arc adds πR/2 of centerline length; for R = 0.6 mm that is about 0.94 mm, roughly 18.6° of phase at 10 GHz on εeff near 2.7. In length-matched differential pairs and array feeds you must count this, since a curve and a mitered corner of the same footprint differ in conductor length. Layout tools compute arc length automatically; hand layouts should add πR/2 per quarter turn rather than the diagonal distance.