Edge-Coupled Differential Pairs: Spacing Matters

What Is Edge-Coupled Embedded Microstrip, Anyway?

Edge-coupled embedded microstrip is a way to route differential pairs on a PCB where the traces sit between a dielectric substrate and some kind of cover material — soldermask, conformal coating, prepreg, or just air. The "edge-coupled" part means the traces run side-by-side, with their outer edges separated by a specific gap. This is different from broadside-coupled (one trace above the other) or microstrip routed over an open ground plane.

Why does this matter? Because the cover material above your traces changes everything. The dielectric constant of that cover, combined with the substrate below, creates an effective dielectric environment that's different from what you'd get with a simple microstrip on FR4. Get this wrong, and your impedance calculations are garbage, your signal integrity suffers, and you end up with reflections that kill your high-speed link.

Most engineers either ignore the cover material entirely or treat it as an afterthought. That's a mistake. The cover can shift your impedance by 5–15% depending on what it is and how thick it is.

Understanding the Geometry

Let's set up the mental model. You have a substrate — typically FR4 with a dielectric constant around 4.2, though Rogers RO4350B (3.48) is common in RF work. Your differential traces sit in this substrate at a certain depth from the top surface. Above them is a cover layer: maybe 25 μm of soldermask, 75 μm of conformal coating, 100 μm of prepreg, or nothing at all (just air).

The trace width, trace spacing (measured edge-to-edge, not center-to-center), substrate height, and cover thickness all feed into the impedance calculation. The cover's dielectric constant matters too — soldermask is typically 3.5, conformal coating 3.0, prepreg FR4 about 4.2, and air is 1.0.

Here's the key insight: the field distribution around your traces isn't uniform. Some of the electric field penetrates the substrate below, some goes into the cover above, and some radiates into the air if there's no cover. The calculator accounts for this by computing odd-mode and even-mode effective dielectric constants separately, then deriving the differential impedance (what you care about for your SerDes link) and common-mode impedance (which matters for EMI).

The Math Behind It

The calculator uses closed-form equations (Wadell's formulas and variants) to compute the transmission line parameters. For edge-coupled microstrip, you get two propagation modes: odd and even.

In odd mode, the two traces carry equal and opposite currents. The electric field between them is strong, and they look more tightly coupled.

In even mode, the traces carry equal and in-phase currents. The field is weaker between them, and the coupling is looser.

The differential impedance is derived from the odd mode:

The common-mode impedance comes from the even mode:

Each of these impedances depends on the effective dielectric constant seen in that mode, which is a weighted average of the substrate and cover dielectrics, weighted by how much field energy sits in each region.

The calculator also spits out the single-trace impedance — what one trace would look like without its partner. This is useful for sanity-checking: if your single-trace impedance is way off from what you'd expect for a standalone microstrip, something's wrong with your geometry.

A Real-World Example: 100 Ω Differential USB 3.0 Pair

Let's design a differential pair for a USB 3.0 high-speed signal. The spec calls for 100 Ω differential impedance (±10%). We're routing on a standard 4-layer board with 1.6 mm total thickness, 0.2 mm substrate under our signal layer, FR4 (εr = 4.2), and 25 μm soldermask on top.

Our copper is 1 oz (35 μm). We'll start with a guess: 5 mil trace width, 8 mil edge-to-edge spacing.

Open the Edge-Coupled Embedded Microstrip Calculator and plug in:

• Trace Width: 5 mil (0.127 mm)
• Trace Spacing: 8 mil (0.203 mm)
• Substrate Height: 0.2 mm (200 μm)
• Cover Thickness: 25 μm
• Cover Type: Soldermask (3.5)
• Copper Thickness: 1 oz (35 μm)
• Substrate Dielectric: FR4 (4.2)

The calculator returns something like:

• Differential Impedance: 104.2 Ω
• Odd-Mode Impedance: 52.1 Ω
• Even-Mode Impedance: 198.5 Ω
• Single Trace Without Partner: 95.3 Ω
• Odd-Mode Effective εr: 3.18
• Even-Mode Effective εr: 3.95
• Propagation Delay (odd): 161.8 ps/inch
• Propagation Delay (even): 165.2 ps/inch

We're at 104 Ω, which is just barely outside the 90–110 Ω window. Let's tighten the spacing to 7 mil:

• Trace Spacing: 7 mil (0.178 mm)

Now we get roughly 102 Ω. Better. Try 6.5 mil spacing and we're at about 100 Ω. Perfect.

Notice the odd and even mode effective dielectric constants are different — 3.18 vs. 3.95. The odd mode "sees" less of the substrate and more of the cover (soldermask), so its effective εr is lower. The even mode sees more substrate, so it's higher. This is exactly what we'd expect.

Also note the propagation delays: they're almost the same (161.8 vs. 165.2 ps/inch), which is good for signal integrity. If they diverged significantly, you'd get skew between the differential signals.

Common Mistakes and Gotchas

Forgetting the cover material exists. Engineers often calculate impedance assuming microstrip over an infinite ground plane, ignoring that 25 μm of soldermask sitting on top. That soldermask has εr = 3.5, which is close to FR4, but it still matters. Conformal coating (εr = 3.0) matters less; prepreg (εr = 4.2) matters just as much as the substrate. Confusing edge-to-edge spacing with center-to-center spacing. The calculator takes edge-to-edge spacing. If you have 5 mil traces with 8 mil edge-to-edge gap, the center-to-center distance is 5 + 8 + 5 = 18 mil. Don't accidentally plug in 18 mil as the spacing — that'll give you a completely wrong answer. Assuming single-trace impedance should match differential impedance. It shouldn't. A single trace without its partner is a different beast. For our USB example, the single trace came back at 95.3 Ω, not 100 Ω. That's normal. The presence of the coupled trace shifts the impedance. Not checking if you're in the validated range. The calculator flags whether your geometry is within the range where the closed-form equations are accurate. If you push the spacing to 20 mil or the substrate height to 5 mm, the formulas start to break down. If the "Within Validated Range" flag comes back false, don't trust the numbers — you need a field solver like HFSS or CST. Ignoring propagation delay skew. If your odd and even mode delays differ by more than a few percent, you'll get skew between the differential signals. This isn't always a killer, but it's worth knowing about. For high-speed serial links (PCIe, USB 3.1, etc.), keep it under 5% if you can. Treating "no cover" as truly air. If you specify "no cover," the calculator assumes εr = 1.0 above your traces. In reality, there's always something — even if it's not soldermask, there's the PCB solder resist process, or dust, or humidity. For a rough estimate, "no cover" is fine, but don't bet your design on it. Soldermask will always go down during manufacturing.

When to Use This Calculator vs. Others

If your differential pair is routed on the surface of the PCB with nothing above it, use the microstrip impedance calculator instead — it's simpler and won't confuse you with cover materials.

If your traces are between two ground planes (stripline or dual stripline), use the stripline calculator — the field distribution is completely different.

If your traces are broadside-coupled (one above the other), use the broadside-coupled pair calculator.

Use this one when your traces are edge-coupled and there's definitely a dielectric cover above them. This is the common case for standard PCB manufacturing: your signals are in an inner layer or the surface layer, and they're covered by soldermask or prepreg.

Practical Design Tips

Start conservative. If you need 100 Ω differential, aim for 98–102 Ω in the calculator. Real PCB manufacturing has tolerances: trace width can vary ±10%, spacing can drift, dielectric constant varies with temperature and frequency. A 2 Ω margin gives you breathing room.

Always verify your single-trace impedance makes sense. For a 5 mil trace at 0.2 mm height on FR4, you'd expect something in the 80–100 Ω range. If the calculator says 200 Ω, you've made a unit error somewhere.

If you're designing for a spec that allows a range (like 85–115 Ω for some differential standards), use the calculator to find the spacing that puts you in the middle of that range. Don't design for the edge of the spec — manufacturing variation will push you out of it.

For high-speed work (above 5 GHz or very fast rise times), run your final geometry through a field solver. The closed-form equations are accurate enough for most purposes, but they're still approximations. A few hours in HFSS can save you from a respun board.

Try It Out

Grab a design you're working on — or dream one up — and open the Edge-Coupled Embedded Microstrip Calculator. Pick your substrate (FR4 or Rogers), your cover material (soldermask is the safe bet), and start tweaking trace width and spacing until you hit your target impedance. Pay attention to the effective dielectric constants and propagation delays — they tell you a lot about whether your design is balanced. If the odd and even mode delays diverge wildly, you know you need to adjust spacing or height. And always check that "Within Validated Range" flag before you commit to a layout.