RF Trace Bend Design | Miter and Curve Rules

Why RF Trace Bends Matter

Every bend in a microstrip or stripline trace introduces a localized impedance discontinuity. A sharp 90° corner adds excess capacitance roughly proportional to the square of the trace width, producing a reflection that worsens with frequency. At 2.4 GHz on a 10 mil (0.254 mm) wide 50 Ω microstrip, the extra return loss from an un-mitered right-angle bend is typically only −35 to −40 dB—often negligible. But at 28 GHz or 77 GHz the same geometry can degrade return loss to −15 dB or worse, radiating energy and corrupting amplitude and phase balance in antenna feed networks.

Two proven techniques exist to compensate: mitering (chamfering the outer corner) and curving (replacing the corner with a circular or spline arc). Both aim to keep the cross-sectional width—and therefore the characteristic impedance—constant through the bend. This article gives the formulas, trade-offs, and verification steps for each approach.

The Physics of a Right-Angle Bend

At a 90° corner the outer edge of the trace extends farther than the inner edge, widening the effective conductor area over the ground plane. The excess patch of copper acts as a shunt capacitance to ground. For a microstrip of width W on a substrate of height h, the added capacitance is approximately:

C_bend ≈ (W / h) × √ε_eff × 0.001 pF (per GHz of operating frequency, per empirical Douville–James model)

This capacitance lowers the local impedance, reflects part of the incident wave, and couples to surface-wave modes at millimeter-wave frequencies. The resulting mismatch ripple degrades filter passband flatness, mixer conversion loss, and S-parameter performance of the entire signal chain.

When Can You Ignore It?

A common rule of thumb: if the trace width W is less than λ/20 at the highest frequency of interest, the un-mitered bend reflection is below −30 dB and usually acceptable. On FR-4 (ε_eff ≈ 3.3) at 1 GHz, λ_eff ≈ 165 mm, so λ/20 ≈ 8.3 mm—far wider than any practical 50 Ω trace. At 24 GHz on Rogers RO4350B (ε_eff ≈ 2.9), λ_eff ≈ 7.3 mm, and λ/20 ≈ 0.37 mm (14.5 mil). A typical 50 Ω microstrip on 5 mil RO4350B is about 10.8 mil wide—within the limit, but barely. Above 40 GHz, mitering or curving is almost always required.

Miter (Chamfer) Compensation

Mitering removes a triangular or trapezoidal wedge from the outer corner. The miter percentage M is defined as the ratio of the cut depth d to the diagonal of the square corner patch (d_max = W√2):

M (%) = (d / (W × √2)) × 100

Optimal Miter Percentage

The Douville–James empirical formula, widely adopted in microwave design, gives the optimal miter as a function of W/h:

M_opt (%) = 52 + 65 × e^{−1.35 × (W/h)}

For a 50 Ω microstrip where W/h ≈ 2 on a typical low-loss laminate, this yields M_opt ≈ 57–60 %. Thinner substrates (W/h closer to 1) push the optimum above 65 %; thicker substrates bring it toward 52 %. An under-miter leaves residual capacitance; an over-miter creates excess inductance. Both produce reflections, so hitting the right value matters.

Most EDA tools (Altium, KiCad, Cadence Allegro) can apply a chamfer automatically. Verify the computed miter depth against the formula above before releasing to fabrication, especially when the tool defaults to a generic 50 % cut.

Fabrication Limits on Mitering

The sharp inner tip of a mitered corner creates an acute copper feature. Standard etching processes reliably resolve features down to about 3–4 mil (75–100 µm). At millimeter-wave frequencies where traces may be only 6–8 mil wide, the miter tip can fall below this limit, causing etch rounding that shifts the effective miter percentage. Discuss minimum feature sizes with your fabricator, and consider a truncated miter (a trapezoidal cut with a flat tip) to keep the smallest feature above 3 mil. The performance penalty of truncation is typically less than 1 dB of return loss below 80 GHz.

Curved (Radial) Bends

A curved bend replaces the corner with a circular arc. The key parameter is the bend radius R, measured to the trace centerline.

Minimum Radius Guidelines

• R ≥ 3 × W: adequate for most applications below 6 GHz. Return loss stays below −30 dB.
• R ≥ 5 × W: recommended for designs above 20 GHz and for phase-matched antenna feed networks.
• R → ∞ (45°-45° double bend): two 45° bends separated by a short straight segment approximate a curve with minimal CAD complexity and excellent performance.

Curved bends generally outperform mitered bends above 40 GHz because they avoid the sharp tip fabrication issue entirely. The trade-off is routing area: a 5W-radius curve on a 10 mil trace consumes roughly 50 mil of extra board real estate per bend compared to a miter, which matters in dense RF isolation and shielding zones.

The 45°-45° Double-Bend Alternative

Two 45° mitered bends separated by a straight segment of length ≥ 1.5 × W behave like a smooth curve in practice. This is the go-to compromise in many KiCad RF layout workflows because the geometry is easy to parameterize and DRC-check. EM simulations consistently show this topology matching or beating a single optimally mitered 90° bend up to at least 60 GHz.

EM Verification of Bends

Analytical formulas give a starting point, but the only way to confirm bend performance in a real stackup is full-wave electromagnetic simulation. Tools such as Keysight ADS Momentum, Ansys HFSS, and Sonnet handle this well.

Simulation Setup Tips

1. Model the bend plus at least 3 × W of straight trace on each side to let higher-order modes decay.
2. Use wave ports (not lumped ports) for frequencies above 20 GHz to capture the correct field distribution.
3. Set mesh density so the smallest cell is ≤ λ_eff/30 at the highest simulation frequency.
4. Include the ground plane via fence if your design uses coplanar waveguide with ground (CPWG).
5. Compare S₁₁ and S₂₁ against a straight through-line of equal electrical length to isolate the bend’s contribution.

For a detailed walkthrough, see the EM simulation setup guide covering port definitions, boundary conditions, and meshing strategies. After simulation, export the Touchstone file and integrate it into your system-level budget.

Design Change ROI: Is Re-routing Worth It?

Mitering or curving every bend in a complex RF board adds layout time. Before committing, estimate whether the electrical improvement justifies the effort. Enter your current measured or simulated return loss, the target spec, and the number of bends to rework below.

For example, a 77 GHz automotive radar front-end with 14 un-mitered bends showing −18 dB aggregate return loss may need only the 6 bends closest to the antenna feed to be curved to reach a −25 dB system target. Reworking all 14 adds layout time with diminishing returns—the tool helps you quantify that trade-off so you spend engineering hours where they matter most.

Bend Design Rules Summary

RF Trace Bend Design Checklist

Use the interactive checklist below to make sure every bend in your layout has been addressed before tapeout. It covers miter percentage verification, minimum radius checks, EM simulation sign-off, and DFM compliance.

After generating the checklist, cross-reference each item against your full RF PCB design review checklist to catch interactions between bend geometry and other discontinuities like via transitions, pad entries, and connector launches.

Quick Pre-Tapeout Bend Audit

• Confirm miter percentage matches the Douville–James optimum for your W/h ratio, not a default 50 %.
• Verify that the smallest copper feature at any miter tip exceeds your fab’s minimum etch resolution (typically 3 mil for standard processes).
• For curves, confirm R ≥ 3W (sub-6 GHz) or R ≥ 5W (above 20 GHz).
• Run a full-wave EM sim on any bend carrying a signal above 10 GHz or any bend in a phase-critical path (antenna feed, LO distribution).
• Check that the ground plane beneath each bend is continuous—no split planes, no via clearance holes encroaching within 2W of the trace edge.
• Review Dk and Df frequency dependence of your laminate at the operating frequency, since effective dielectric constant shifts the optimal miter.

Frequently Asked Questions

Does a 90° bend really cause signal loss?

A 90° bend primarily causes reflection (return loss), not absorption loss. The reflected energy reduces the power delivered to the load and creates standing waves. At low frequencies the reflection is negligible, but above roughly 10 GHz it becomes measurable and above 40 GHz it can dominate system performance.

What miter percentage should I use for 50 Ω microstrip?

For most 50 Ω microstrip geometries the optimal miter is between 55 % and 65 %, depending on the W/h ratio. Use the Douville–James formula: M(%) = 52 + 65 × e^−1.35(W/h). A generic 50 % default is almost always an under-miter.

Are curved bends always better than mitered bends?

Not always. Below about 20 GHz, a properly mitered bend performs comparably to a curved bend and uses less routing area. Above 40 GHz, curved bends (R ≥ 5W) are generally superior because they avoid the fabrication-sensitive sharp miter tip. Between 20 and 40 GHz either technique works when executed correctly.

How do I handle bends in differential pairs?

Keep both traces at equal electrical length through the bend. The inner trace of a curved differential bend is shorter, so add a small meander or use the “accordion” technique on the inner trace. Maintain the differential spacing through the bend to avoid common-mode conversion. EM simulation is strongly recommended for differential bends above 10 Gbps or 10 GHz.

Can I use the same bend rules for stripline?

Yes, the same miter formulas apply. Stripline bends are slightly less problematic than microstrip bends at the same frequency because the homogeneous dielectric eliminates dispersion effects. The Douville–James miter percentage still applies; use the stripline W/h ratio (where h is the distance to the nearest reference plane).