Via Stub Effects and Back Drilling at High Frequency

A via that connects two layers also connects to every layer below, whether those layers need the connection or not. The unused portion is a stub, and at high frequency it behaves like a small antenna attached to the signal path. On a fast channel a stub only a few millimetres long is enough to create a notch in the frequency response that no amount of tuning elsewhere will remove.

What a Stub Is and Why It Exists

Vias are plated through the whole thickness of the board because that is how the process works. When a signal enters from an outer layer and leaves at an inner layer, the plating below that inner layer remains as a conductor hanging off the signal path.

A via stub is a transmission line that ends in an open circuit, and an open circuited line reflects energy back towards the source. Its electrical length depends on the physical length of the plating and on the effective permittivity of the material surrounding it. That material is a mixture of laminate and, in some stack designs, of the air inside an unfilled via, so the effective value differs from the figure on the laminate datasheet.

The Quarter Wave Resonance

An open circuited line presents a short circuit at a quarter of a wavelength, and at that frequency the stub destroys the signal rather than merely disturbing it. The result is a deep notch in the insertion loss curve and a corresponding peak in the reflected energy. In network terms the notch is the frequency domain view of an impedance discontinuity, and most of the incident energy is reflected rather than transmitted.

The notch frequency falls as the stub gets longer, which is why the effect appears at the data rates used today for stubs of only a few millimetres. On a thick backplane the stub of an ordinary through via can easily be several millimetres long. A long stub also produces further resonances at odd multiples of the quarter wave frequency, so the damage is not confined to a single band.

How Stub Length Sets the Frequency

The relationship is simple: the resonant frequency is inversely proportional to the electrical length of the stub. That length is the physical distance from the layer where the signal leaves the via to the far end of the plating, multiplied by the square root of the effective permittivity.

The practical consequence is that the board thickness and the layer stack both matter. Moving a signal to a layer closer to the bottom shortens the stub, and reducing the total thickness shortens the stub of every via on the board at the same time. The same reasoning explains why a thin board with a controlled stack performs better than a thick one at the same data rate, even when both use identical materials.

Symptoms on a Real Channel

On a real link the stub shows up as a rise in insertion loss at the top of the band, and it is often accompanied by a change in group delay that distorts a fast edge. The eye diagram opens less and the bit error rate rises at the highest data rate even though the same link works perfectly at a lower one.

Because the effect is narrow in frequency, it can be invisible in a continuity test and obvious in a channel measurement. That is why high speed designs are verified with a network analyser or a pattern generator rather than with a multimeter.

Back Drilling as the Remedy

Back drilling removes the unused portion of the plating with a slightly larger drill after the board has been plated. The result is a via that stops close to the layer where the signal leaves, and the stub is reduced to the short distance between the signal layer and the bottom of the cut.

The process is controlled by the depth of the cut, which has to be deep enough to remove the stub and shallow enough to leave a reliable connection to the layers above. The remaining tolerance is usually expressed as a maximum stub length rather than as a drill depth. A typical requirement might be that the remaining stub is shorter than a defined figure, which the fabricator meets by controlling the drill depth and the board thickness together.

Design Measures Before Back Drilling

Back drilling adds cost and a process step, so the design should be arranged to avoid the need where it can. Placing the signal layer near the bottom of the stack, keeping the board thin within its aspect ratio, or routing the critical net on a different layer all reduce the problem before it is manufactured.

Blind and buried vias are the other approach, because a via that does not pass through the whole board has no stub on the unused layers. They cost more to fabricate, but they solve the problem structurally rather than by removing material after plating. The choice between them is usually made on the total cost of the board, since back drilling is applied to every via of a given type rather than only to the critical ones.

Process Control for Back Drilling

Control depends on the drill depth and on the registration of the drill to the via. Depth is set by the machine and verified on a coupon, while registration depends on the same tooling that positions the holes in the first place.

The back drill is larger than the via, so it also removes some of the surrounding dielectric. That removal leaves a small recess which can trap chemistry, so the sequence of plating, back drilling and surface finishing has to be planned carefully rather than improvised.

Verification: TDR and Loss Measurement

The stub can be measured directly with time domain reflectometry, which shows the reflection from the stub as a feature on the trace. The measurement is quick and can be made on a coupon that includes the same via structure as the product.

Insertion loss measured on a coupon with a known line length shows the effect in the frequency domain, which is closer to the way the link behaves in service. Both measurements should be taken on the production panel or on a coupon built with it. The coupon should replicate the via structure of the product, including the pad sizes and the plating thickness, because both affect the impedance of the transition.

Choosing the Right Fix for a Product

The choice between accepting the stub, back drilling and using blind or buried vias depends on the data rate, the stack, the board thickness and the cost target. A conservative design with a short via and a slow edge may need nothing at all. That assessment should be made with the highest frequency content of the signal in mind rather than with the nominal clock rate.

Where the channel margin is thin, the decision should be made from a measurement rather than from a rule of thumb, and the measurement should be repeated after any change to the stack or to the laminate. The design rules that support those decisions are set out in the guide to high speed design.

FAQ

How long can a stub be before it matters? It depends on the data rate and on the margin in the link. As a rough guide a stub becomes significant when its resonant frequency approaches the highest frequency content of the signal, which for modern links can be a few millimetres.

Does back drilling weaken the via? It removes material below the signal layer, so the current path is unaffected and the mechanical connection remains. The depth has to be controlled so that sufficient plating is left at the layer being connected.

Can a stub be simulated rather than measured? It can, and simulation is useful for comparing options before the board is built. The final check should still be a measurement on a coupon from the actual build.