Board Edge Isolation: Pulling Ground Back From the Edge

Traces that run along the board edge behave differently from traces in the middle, and the difference is not caused by the absence of laminate. It is caused by the ground plane. Every plane has to be pulled back from the routed edge for manufacturing reasons, and that pullback opens a window through which the fields of an edge trace can reach the outside world and couple to other edge traces on the opposite side of the board.

Why Edge Traces Are Exposed

Inside the board the field of a trace is confined between the trace and the reference plane below it. At the edge, the plane stops before the copper does, so a portion of the field is no longer bounded by a conductor. That unbounded portion fringes beyond the laminate, and the trace behaves partly like a radiator and partly like one half of an unintentional coupler.

The result is a board edge that is noisier than the interior, which is why sensitive analogue and radio traces are usually kept away from the perimeter as a matter of course. What is useful is knowing how far away they have to be.

The Geometry That Sets the Coupling

The controlling dimension is not the distance from the trace to the board edge in isolation, but the distance from the trace to the edge of the ground plane. That gap determines how much of the field escapes, and it can be parameterised and simulated in a field solver to get a quantitative answer rather than a rule of thumb.

Consider a stack with microstrip on the outer layers, stripline on the inner layers and solid ground planes between them, all on the same dielectric thickness. Pull the ground planes back from the edge by a gap that varies, and measure the isolation between the traces on different layers.

Microstrip to Microstrip Across the Edge

For two microstrip traces on opposite faces of the board, the isolation improves as the ground pullback increases. There is no sharp threshold because the coupling mechanism is dominated by the fringing field at the plane edge, which falls off gradually with distance.

The practical conclusion is that a critical signal routed as microstrip on the surface has no defined safe distance that applies to every design. The more critical the signal, the more the ground should be pulled back and the further the trace should be moved towards the interior of the board.

Microstrip to Stripline

The behaviour changes when the two traces are on different kinds of layer. Simulation shows that once the ground pullback reaches about twice the dielectric thickness, the isolation between a surface microstrip and an inner stripline stops improving. The curve flattens because the dominant coupling mechanism has changed: beyond that distance the field at the plane edge is no longer what governs the coupling, and the shared return current in the ground plane takes over.

That gives a usable number. If a surface trace and an inner trace are the concern, a pullback of roughly two times the dielectric thickness removes the edge effect from the budget, and improving it further requires a different measure such as separating the traces or adding a via fence.

Stripline to Stripline

For two inner striplines the effect starts even earlier. The isolation improves with the gap until the pullback reaches about one dielectric thickness, at which point the shared return path dominates and further pullback buys nothing.

The reason is that a stripline is already shielded above and below, so the only exposure is at the plane edge, and it takes a much smaller pullback to remove that exposure. An inner layer is therefore a good home for a critical signal even when it has to run near the perimeter.

What the Simulation Does Not Tell You

The results are expressed as multiples of the dielectric thickness, not as absolute dimensions, because the structure scales. A thin dielectric gives a small absolute pullback and a thick one gives a large pullback, and the isolation at a given multiple is similar in both cases.

What the model does not cover is the effect of the trace leaving the board through a connector, the presence of a metal enclosure close to the edge, or the influence of nearby vias and component pads. Those have to be assessed separately, and they often dominate in a real product.

Sensitivity, Scaling and Frequency

Isolation is a frequency-dependent quantity, and the curves show a dip at the frequency where the trace length approaches a half wavelength along the structure. Beyond that length the trace no longer behaves as a short stub, and a longer trace does not necessarily couple more than a shorter one at a specific frequency.

The practical reading is that the worst case is bounded, and the number to design against is the one at the frequency of interest. The rules of thumb above are a starting point for the pullback, not a substitute for checking the band that matters.

Applying the Numbers

Set the ground pullback from the routed edge according to the fabricator’s minimum, then decide how much further to go on the basis of which layer the critical trace is on. An inner stripline needs about one dielectric thickness of additional pullback to be clear of edge coupling; a surface microstrip benefits from as much as the layout allows, and should be moved inboard if the trace is genuinely sensitive.

Where the trace has to run near the edge, a via fence connecting the ground planes along the perimeter restores much of the lost shielding, and it is the standard remedy on boards where the outline leaves no room. Microstrip and stripline routing covers the geometry of the two transmission-line types, and High frequency traces and data bus routing covers the routing rules that go with them.

Where This Fits in the EMC Plan

There is also a practical limit to how much the pullback can be increased. Removing copper from the perimeter reduces the plane area, which raises the plane impedance and weakens the reference for any trace that runs along the edge. Beyond the point where the edge coupling has been removed, additional pullback is a net loss.

Edge coupling is one contributor among several, and it is usually not the largest. Loop area, reference continuity and connector placement matter more on most boards. The reason to control it anyway is that it is cheap: the same layout that keeps traces away from the perimeter also keeps them away from the enclosure seam and the mounting hardware, which are the other structures at the edge of a product. EMI suppression design principles puts edge isolation in that wider context.

FAQ

How far should a trace be kept from the board edge? For an inner stripline, a ground pullback of about one dielectric thickness is enough to remove the edge effect. For a surface microstrip there is no fixed figure; the trace should be moved inboard as far as the layout allows.

Does the dielectric thickness matter? It sets the scale. The isolation depends on the pullback expressed as a multiple of the dielectric thickness, so a thicker dielectric requires a larger absolute distance.

What if the trace must run at the edge? Use an inner layer if possible, add a via fence along the perimeter connecting the ground planes, and keep the trace away from any enclosure seam or metal hardware.