How tightly does the return current hug the trace?

Tightly, and the distribution has a closed form. For a trace at height h above a reference plane, the return current density falls off as one over one plus x over h squared, where x is the sideways distance from the trace centreline. Integrating that gives the fraction flowing within a given width: 50 percent lies within plus or minus one h, 79.5 percent within three h, and 93.7 percent within ten h. For a trace 0.2 mm above its plane, almost all the return is inside a strip about 4 mm wide. This is why a plane needs continuity only where the traces are, and why a small break in exactly the wrong place matters so much.

What actually happens when a trace crosses a plane split?

The return current cannot follow beneath the trace any more, so it diverts to the nearest way across, around the end of the slot or through whatever decoupling capacitor bridges the gap. The signal still arrives, which is what makes the fault so easy to miss in functional testing. What changes is the loop: it grows from a thin strip a fraction of a millimetre tall to a loop enclosing the whole detour, often by two or three orders of magnitude in area. Since radiated field is proportional to loop area, that area increase appears directly as emissions, and the same loop is equally efficient at receiving, so immunity degrades too.

Does skin effect change where the return flows?

It changes the depth rather than the lateral position. Skin depth in copper is 66 micrometres at 1 MHz, 21 at 10 MHz and 6.6 at 100 MHz, so above a few megahertz the current rides in a thin sheet on the surface of the conductor facing the signal. One ounce copper is 34.8 micrometres thick, which equals the skin depth at 3.6 MHz: above that, adding copper thickness does nothing for high frequency return impedance, while adding width still does. The lateral spreading law is unaffected, because it is set by the geometry of the field between trace and plane.

What is the single most useful layout habit that follows from this?

Trace the return path with a finger for every fast net before the board is released. If the finger has to leave the area directly under the trace, whether because of a split, a slot, a connector cut-out or a change of reference layer, that is a loop, and it will show up in a chamber. This one check catches more radiated emission problems than any component choice, and costs nothing at the point in the project where fixing it is still free.

Return current paths: where current actually goes

Author Hugues Orgitello Last updated 30 August 2026 ~5 min read

Guide, EMC fundamentals

Current flows in loops. Every signal that leaves a driver comes back, and the route it takes home is chosen by physics rather than by the layout engineer. Most radiated emission failures are a return path that went somewhere unintended, and most of those were visible on the layout before the board was made. This page derives where the return actually goes, and what that implies for reading a design.

Lowest impedance, not shortest distance

The return does not seek the shortest route or the least copper. It distributes itself to minimise total impedance, and impedance has two terms whose balance shifts with frequency.

At low frequency, resistance dominates. The return spreads widely through the plane, taking the broadest cross-section available, which is usually nothing like the path under the trace.

As frequency rises, inductive reactance grows with frequency while resistance does not. Above roughly 100 kHz inductance is in charge, and the minimum-inductance arrangement is the one enclosing the smallest loop area. That is the return flowing in the reference plane directly beneath the signal trace.

The transition is gradual, but in any digital or RF design one assumes the high-frequency behaviour and treats the reference plane as sacred.

How tightly it hugs the trace

The lateral distribution has a closed form. For a trace at height h above a plane, return current density falls off as:

J(x) proportional to 1 / (1 + (x/h) squared)

with x the sideways distance from the centreline. Integrating gives the fraction of return current flowing within a strip of half-width x:

fraction = (2/pi) x arctan(x/h)

Half-width	Fraction of return current
1h	50.0 %
2h	70.5 %
3h	79.5 %
5h	87.4 %
10h	93.7 %
20h	96.8 %

The consequence is concrete. A trace 0.2 mm above its reference plane carries almost all its return inside a strip roughly 4 mm wide. A plane therefore needs to be continuous where the traces are, not everywhere, which is the useful half of the result. The other half is that a break in exactly the wrong place, directly under a fast net, does damage out of all proportion to its size.

Depth: skin effect

Skin effect sets how deep the current runs, not where it runs laterally. In copper:

Frequency	Skin depth
1 kHz	2.09 mm
1 MHz	66.1 um
10 MHz	20.9 um
100 MHz	6.6 um
1 GHz	2.1 um

One ounce copper is 34.8 um thick, which is exactly the skin depth at 3.6 MHz. Above that frequency the current rides in a sheet thinner than the copper, so adding copper thickness stops helping while adding width still does. This is why heavy copper improves power distribution and does almost nothing for high-frequency return impedance.

The plane split, and why it is so expensive

Take a fast net crossing a slot in its reference plane. The return current arrives at the slot and cannot continue beneath the trace. It diverts, around the end of the slot or through whatever decoupling capacitor happens to bridge the gap.

The signal still arrives at the far end. Functional testing passes. What changed is the loop.

Before the split, the loop is a thin strip the height of the dielectric, a fraction of a millimetre tall. After it, the loop encloses the whole detour, commonly two or three orders of magnitude more area. From coupling mechanisms, radiated field is proportional to loop area, so that increase appears directly in the emission result.

The same enlarged loop is equally efficient in reverse, so immunity degrades at the same time. A split under a clock net typically costs on both scans at once, which is a useful diagnostic signature.

Connector cut-outs, mounting holes in a row, and via anti-pad chains all form slots without anyone deciding to create one.

Changing reference layer

A via that moves a signal from one layer to another also moves its return, and the return cannot follow through the via. It must find a path between the two reference planes, through a stitching via if one is nearby, through the plane pair capacitance if not.

The rule that follows: a signal changing layers should change between layers referenced to the same plane where possible, and where it cannot, a ground stitching via should sit next to the signal via. A stitching via placed a centimetre away is not the same fix, because the return has to travel there and back.

Reading a layout for return paths

One habit catches most of it. For every fast net, trace the return with a finger, on the layer directly beneath the trace, from load back to source. If the finger has to leave the region under the trace for any reason, that is a loop.

What to look for	Why it matters
Slots, splits or moats under fast nets	Forces a detour; large loop area
Rows of mounting holes or vias forming a line	An unintentional slot
Layer changes without an adjacent stitching via	Return has no path between planes
Connector cut-outs in the reference plane	The return crowds around them
Signals referenced to a power plane cut by domain splits	Return path broken at every split

Key takeaways

Return current takes the lowest-impedance path, which above about 100 kHz means directly under the trace.
The spreading law is (2/pi) arctan(x/h): half the return within one dielectric height either side, 94 percent within ten.
Skin depth sets depth, not lateral position. Above 3.6 MHz, one ounce copper is already thicker than the skin depth, so width helps and thickness does not.
A plane split turns a thin strip into a large loop, and radiated field is proportional to loop area, so emissions and immunity degrade together.
Trace the return with a finger before release. It is free at that point and expensive afterwards.

Frequently asked questions

Why does return current follow the trace instead of the shortest path?: Because it takes the path of lowest total impedance, and which term dominates depends on frequency. At low frequency resistance dominates, so the return spreads out to take the widest, lowest resistance route through the plane, which is rarely under the trace. As frequency rises, inductive reactance grows with frequency while resistance does not, and above roughly 100 kHz inductance dominates. The lowest inductance configuration is the one that minimises enclosed loop area, and that means the return flowing in the plane directly beneath the signal trace. Nobody routes it there; it goes there because that is where the impedance is lowest.
How tightly does the return current hug the trace?: Tightly, and the distribution has a closed form. For a trace at height h above a reference plane, the return current density falls off as one over one plus x over h squared, where x is the sideways distance from the trace centreline. Integrating that gives the fraction flowing within a given width: 50 percent lies within plus or minus one h, 79.5 percent within three h, and 93.7 percent within ten h. For a trace 0.2 mm above its plane, almost all the return is inside a strip about 4 mm wide. This is why a plane needs continuity only where the traces are, and why a small break in exactly the wrong place matters so much.
What actually happens when a trace crosses a plane split?: The return current cannot follow beneath the trace any more, so it diverts to the nearest way across, around the end of the slot or through whatever decoupling capacitor bridges the gap. The signal still arrives, which is what makes the fault so easy to miss in functional testing. What changes is the loop: it grows from a thin strip a fraction of a millimetre tall to a loop enclosing the whole detour, often by two or three orders of magnitude in area. Since radiated field is proportional to loop area, that area increase appears directly as emissions, and the same loop is equally efficient at receiving, so immunity degrades too.
Does skin effect change where the return flows?: It changes the depth rather than the lateral position. Skin depth in copper is 66 micrometres at 1 MHz, 21 at 10 MHz and 6.6 at 100 MHz, so above a few megahertz the current rides in a thin sheet on the surface of the conductor facing the signal. One ounce copper is 34.8 micrometres thick, which equals the skin depth at 3.6 MHz: above that, adding copper thickness does nothing for high frequency return impedance, while adding width still does. The lateral spreading law is unaffected, because it is set by the geometry of the field between trace and plane.
What is the single most useful layout habit that follows from this?: Trace the return path with a finger for every fast net before the board is released. If the finger has to leave the area directly under the trace, whether because of a split, a slot, a connector cut-out or a change of reference layer, that is a loop, and it will show up in a chamber. This one check catches more radiated emission problems than any component choice, and costs nothing at the point in the project where fixing it is still free.