Layer Change or Detour: Choosing Between a Via and a Bend

The question comes up on almost every high speed layout: is it better to change layer through a via, or to stay on the same layer and route around the obstacle? The two options trade one discontinuity for another, and which is preferable depends on the kind of circuit, on the frequency, and on what the trace is carrying.

The answer is not the same for a radio frequency board and a high speed digital one, because the two are analysed differently and because they care about different quantities.

What the Two Options Cost

A layer change adds a via to the path. The via has a small capacitance and a small inductance, it interrupts the impedance of the trace, and on a thick board it leaves a stub below the layer where the signal exits, which resonates at a frequency set by its length. It also requires the return current to change layer, so a ground via must be provided beside it, otherwise the return path has to find its own way across.

Routing around, by contrast, adds length. A longer trace has more loss, more coupling opportunity and more delay, and the detour itself may involve two bends. The bends are minor discontinuities if they are made at forty five degrees or as arcs, but a detour around a large obstacle can add several centimetres where a via would have added a fraction of a nanosecond of delay and a small discontinuity, and the added length is unconditional.

<img src="https://www.gopcba.com/wp-content/uploads/2026/09/270-2-scaled.jpg" alt="signal via with a ground return via” />

Why the Digital Case Is the Simpler One

On a multilayer digital board with solid reference planes, the analysis is comparatively well defined. The trace is a transmission line over a plane, the return current flows directly beneath it, and a via is treated as a lumped element in the path. Whether a via is preferable to added length then becomes a calculation: compare the effect of the via discontinuity with the effect of the extra loss and delay, at the data rate and the rise time of the interface.

For many high speed digital traces the answer is that a signal via with a short stub is acceptable and a long detour is not, because the detour consumes timing margin that the interface needs. This is why escape routing from a dense package is done with vias, and why the effort goes into controlling the stub rather than avoiding the via.

Why the RF Case Needs More Care

A radio frequency circuit is analysed differently. The board often has fewer layers, sometimes only two, which means the return current is distributed among whatever ground and supply copper is nearby rather than flowing in a well defined plane beneath the trace. Both the voltage and the current in the circuit matter, because in an analogue circuit the current path affects the behaviour of the network, not only the voltage at the receiver.

In that situation RF routing is less forgiving: a bend and a via both need to be examined rather than assumed. A bend changes the local geometry and therefore the local impedance and the field distribution; a via introduces a shunt capacitance and an inductance in series, both of which shift the match. At frequencies where a fraction of a millimetre of extra trace is a significant phase shift, either change can move a tuned network off its target, and the analysis has to include the physical structure rather than a lumped approximation.

Practical Guidance

On a multilayer board with continuous reference planes, as described in the stack-up plan, prefer the via over a long detour for high speed digital nets, and control the stub: use a blind or buried via where the fabrication process allows, or a back drill to remove the unused barrel, and place a ground via next to the signal via so the return current has a defined transition. Keep the number of layer changes in a critical net to the minimum the routing allows, because the cost is cumulative.

On a two layer or RF board, treat both options as changes to the circuit rather than as routing conveniences. Keep the trace geometry consistent along the path, place ground copper adjacent to the trace and connect it with vias at intervals so the return has a defined route, and verify the result with a measurement rather than with a rule of thumb. That measurement belongs in the test plan for the board, since it is the only confirmation that the chosen geometry matches the model. Where the trace is part of a tuned network, the effect of a via or a bend is a question for the field solver or the network analyser.

The Return Path Is the Real Question

Underneath both cases is the same issue. A layer change is only cheap if the return current can change layer with the signal, which requires a ground via nearby. A detour is only cheap if the return current can follow the trace around the obstacle, which requires the reference plane to be present along the whole detour. If neither condition is met, both options are expensive, and the correct answer is to change the layout so that the signal does not have to leave its reference in the first place.

That last point resolves most of the debate, and it is the question worth asking before any geometry is chosen. The choice between a via and a bend matters much less than whether the return path under the trace is continuous, both before and after the change. A layout that keeps a solid reference beneath every critical net can afford a via, and the copper and dielectric behind that reference come from the fabrication choices made at the start; a layout that routes a fast trace across a plane split cannot be rescued by choosing a different geometry.

FAQ

Is one via always worse than one bend? Not necessarily. The bend adds length and delay, while the via adds a localized discontinuity. Which is worse depends on the interface and the geometry.

How many vias are acceptable in a high speed net? As few as the routing allows, and each additional one should be justified by a reduction in length.

Does the ground via beside a signal via have to be adjacent? It should be as close as the layout permits, so that the loop formed by the transition is small.

What about a via that does not change layer? A via used purely as a test point or a connection to an inner plane is still a discontinuity, and it should be evaluated the same way.

Summary

A layer change and a detour both cost something: the via adds a discontinuity and a stub, the detour adds loss and delay. On a multilayer digital board with solid planes, the via is usually the better trade and the effort belongs in controlling the stub and providing a return via. On an RF or two layer board, both are changes to the circuit and should be analysed as such. In every case, the continuity of the return path decides whether either option is acceptable at all.