Via & Backdrill Design for High-Speed Channels
Why Vias Fail Before Traces
Engineering Premise
In high-speed PCB channels, vias are usually the first structures to distort the signal.
Not the long trace.
Not the connector.
The via.
This is because a via is:
- a vertical impedance discontinuity
- a geometry transition
- a reference plane transition
- a manufacturing tolerance amplifier
If vias are not engineered explicitly, no amount of trace optimization can save the channel.
1. Via stubs: why “unused length” becomes a resonator
1.1 What a via stub really is
A via stub is the unused conductive length of a plated hole that extends beyond the signal layer connection.
Electrically, a stub behaves as:
- an open-ended transmission line
- a frequency-dependent resonator
At certain frequencies, the stub reflects energy back into the channel.
1.2 Stub length vs frequency: the non-linear effect
Stub impact is not linear.
As data rate increases:
- rise time shortens
- higher harmonics become relevant
- stub resonance enters the signal spectrum
This causes:
- reflection-induced jitter
- eye height reduction
- deterministic jitter increase
A stub that is “acceptable” at 5 Gbps can collapse the eye at 16–25 Gbps.
1.3 Why eye diagrams close before specs are violated
Eye closure often appears:
- before insertion loss limits
- before impedance specs are exceeded
Because reflections from stubs:
- distort the waveform shape
- increase timing uncertainty
- interact with equalization in non-intuitive ways
This is why via problems are often misdiagnosed as “SerDes tuning issues.”
1.4 Executable checklist (stub control)
- Minimize via depth wherever possible (layer assignment matters).
- Avoid routing high-speed nets on layers that force long unused via sections.
- Treat stub length as a budgeted parameter, not an afterthought.

2. Backdrilling: what it really fixes—and what it does not
2.1 What backdrill actually removes
Backdrilling removes:
- the unused portion of the via barrel
- the dominant stub resonator
It does not remove:
- via pad capacitance
- antipad geometry
- reference plane discontinuity
Backdrill is necessary, but not sufficient.
2.2 Residual stub: the hidden variable
Backdrilling always leaves a residual stub due to:
- drill depth tolerance
- registration accuracy
- layer-to-layer thickness variation
This residual stub still:
- reflects
- varies across panels
- widens channel-to-channel variation
Backdrill improves the mean, but the distribution width matters at volume.
2.3 When backdrill provides real benefit
Backdrill yields diminishing returns below certain stub lengths.
Real benefit depends on:
- data rate and rise time
- channel length
- equalization strategy
At lower speeds, backdrill may show little improvement.
At higher speeds, it may be mandatory to avoid eye collapse.
2.4 Executable checklist (backdrill strategy)
- Define maximum allowed residual stub, not just “backdrilled or not.”
- Align backdrill spec with data-rate class (DDR vs PCIe Gen4/5/6).
- Ensure all lanes in a channel use identical backdrill geometry.
3. Via types: through, blind, buried, and microvias
3.1 Through vias: simple but dangerous at high speed
Through vias:
- are easy to manufacture
- have long barrels
- create long stubs
They are acceptable only when:
- speed is low
- backdrill is applied
- geometry is tightly controlled
Without backdrill, they are a primary SI risk.
3.2 Blind and buried vias: shorter, but not free
Blind/buried vias:
- reduce stub length
- reduce vertical discontinuity
But introduce:
- additional lamination cycles
- registration tolerance stacking
- yield and cost penalties
They improve SI but narrow the manufacturing window.
3.3 Microvias: excellent electrically, fragile mechanically
Microvias offer:
- minimal stub
- excellent high-speed performance
But are sensitive to:
- aspect ratio limits
- thermal cycling fatigue
- stacking reliability
They solve SI, but introduce reliability engineering requirements.
3.4 Executable checklist (via type selection)
- Choose via type based on speed class + volume reliability, not SI alone.
- Avoid mixing via types within the same high-speed channel.
- Standardize via architecture per channel class.
4. Reference plane continuity through vias: the invisible SI killer
4.1 Why vias break return paths
When a signal changes layers through a via:
• the return current must also transition
If reference planes are:
- not adjacent
- not stitched
- perforated
the return current detours, causing:
- added inductance
- mode conversion
- timing distortion
This effect is often invisible in schematic review.
4.2 Ground stitching is not optional
Without stitching vias or capacitors:
- the return path becomes large and unstable
- EMI and jitter increase
- channel behavior becomes temperature and lot dependent
Return path control is as important as signal path control.
4.3 Executable checklist (reference continuity)
- Ensure every signal via has a nearby return path transition.
- Use ground stitching vias adjacent to high-speed signal vias.
- Avoid reference plane swaps without explicit return bridges.
5. Why via-related SI issues explode at volume
5.1 Prototype hides via variability
Prototypes:
- use fewer panels
- tighter operator control
- less drill wear
Via geometry is more consistent.
5.2 Volume exposes tolerance stacking
At volume:
- drill wear varies
- plating thickness varies
- backdrill depth distribution widens
Small via differences become:
- lane-to-lane skew
- channel-to-channel inconsistency
Via-induced issues are distribution-driven failures.
For high-speed channels:
- traces carry the signal
- vias define whether it survives
Robust via design requires:
- Stub length budgeting
- Backdrill tolerance awareness
- Via type standardization
- Explicit return path engineering
- Manufacturability-first thinking
A via that “works once” is not a via that scales.
What ULTRONIU Does
ULTRONIU treats vias as high-speed channel components, not passive holes.
Capabilities include:
- via stub budgeting tied to data-rate class
- backdrill strategy with residual-stub control
- via/reference co-design for return path integrity
- channel-class standardization for volume builds
- SI risk evaluation under manufacturing variation
The objective is not a clean eye on one board.
It is clean eyes across thousands of boards.
Via-Induced SI Reality: Transition Discontinuities, Return Paths & Early Architecture Lock-In
Q1: Why do vias distort signals more than long traces?
Because vias introduce abrupt impedance and reference changes in a short distance.
Q2: Does backdrill eliminate all via-related problems?
No. It removes the main stub, but residual stubs and reference issues remain.
Q3: Are blind vias always better for high speed?
Electrically yes, but they introduce yield and reliability trade-offs.
Q4: What is the most common via-related SI mistake?
Ignoring return path continuity during layer transitions.
Q5: When must via architecture be finalized?
With stackup definition—after routing, options are limited.
Tags:
Related Products
Related Products/Solutions
Quick links


