High-Speed PCB Engineering | Via Signal Integrity
A controlled-impedance coupon can pass while the via transition used by the actual signal pair remains uncharacterized. The missing question is not simply whether the impedance is nominal. It is how the complete transition reflects energy over the frequency range that matters to the channel.
Engineering answer: A TDR impedance result is evidence for a specific structure, stimulus, reference plane, and measurement setup. It does not automatically establish the broadband differential return loss of an HDI via transition. When the transition is critical to channel margin, review its actual geometry and return path, evaluate mixed-mode S-parameters over a project-defined band, and correlate the model with a representative physical measurement.
1. The Coupon Passed. What Has Actually Been Verified?
Consider a high-speed board with a tightly controlled stripline differential pair. Its fabrication drawing calls for an impedance target, and a companion test coupon meets the agreed tolerance. This is useful manufacturing evidence: it indicates that the tested line structure, on its stated layer and under its specified test conditions, meets the impedance requirement.
Now follow that same pair to a device escape or a layer change. The signals pass through pads, microvias, a longer inner-core via, anti-pad openings and nearby reference planes. Those features are not necessarily present in the straight-line coupon. Passing the coupon therefore says little about the frequency response of that particular vertical transition.
There are three distinct test objects that should not be conflated:
- Controlled-impedance line coupon: a representative trace structure used to check an agreed impedance target.
- Board-level TDR path: a route that may expose discontinuities along the actual interconnect, depending on accessibility, time resolution and fixture effects.
- Via-transition network: the complete electromagnetic structure at the layer change, evaluated with appropriate port definitions and broadband reflection and transmission data.
This distinction is especially important when a design changes PCB suppliers. The same nominal impedance target can be met through different combinations of finished trace geometry and material properties, while the transition itself depends on a separate set of dimensions and reference-plane relationships.
2. Why a High-Speed Via Is More Than a Vertical Conductor
A high-speed via is an electromagnetic transition, not simply a plated connection between two copper layers. Its response is shaped by the full surrounding structure. A signal barrel contributes inductive behavior; the barrel and pads couple capacitively to nearby copper; anti-pad dimensions change that coupling; and the placement of return-path vias influences the current loop and field distribution.
Unused via segments can add stub-related resonance. Where the differential pair changes layers, imperfect geometric symmetry can also create differential-to-common-mode conversion. These effects can change with frequency, so a result that looks acceptable at one test point does not necessarily predict behavior across an entire band.
| Design feature | Electrical concern | Question for review |
|---|---|---|
| Signal-via diameter and length | Distributed inductance, capacitance and transition behavior | Does the model use finished dimensions and the actual layer span? |
| Pad and anti-pad geometry | Local capacitive loading and field redistribution | Are all relevant reference-plane openings included? |
| Ground or return-path vias | Return-current continuity and loop geometry | Is the return path represented at every layer transition? |
| Unused barrel or stub | Frequency-dependent reflection and resonant behavior | Is the residual stub modeled or controlled by the drawing? |
| Differential symmetry | Imbalance and mode conversion | Are both conductors and nearby grounds modeled together? |
None of these parameters should be optimized in isolation. For example, enlarging an anti-pad may reduce one capacitive discontinuity yet change return-path coupling or affect the neighboring transition. It is the assembled structure—not any individual diameter—that determines the response.

3. TDR and Return Loss Are Complementary—Not Competing—Measurements
TDR helps locate and interpret changes in a transmission path by observing reflections from a time-domain stimulus. A faster effective edge and adequate instrument bandwidth can reveal smaller features, but fixture behavior, noise, loss and finite time resolution still matter.
Frequency-domain S-parameters describe how the defined network reflects and transmits energy as frequency changes. For a differential via treated as a four-port network, the relevant quantities may include Sdd11 (differential input reflection), Sdd21 (differential forward transmission), and mode-conversion terms such as Scd21. Whether these are useful depends on the port mapping, reference impedance, calibration and de-embedding.
These are related descriptions of the same physical network. If the complete, properly calibrated broadband network response is available, time-domain and frequency-domain views can be transformed into one another under appropriate assumptions. The limitation here is coverage: a straight-line coupon, a bandwidth-limited time-domain measurement and a via-specific four-port model do not represent identical evidence.

The practical review question is therefore not “TDR or VNA?” It is: Which evidence represents the actual via, at what reference planes and over what bandwidth, against which project requirement?
4. Core Thickness Can Change the Optimization Strategy
An IMPACT 2026 conference contribution by Mengen Zan and colleagues examined a long via architecture connecting outer stacked microvias to an inner plated-through-hole core in a seven-stage HDI construction. The work used modeling and simulation to study how return loss changes with core thickness and surrounding geometry.
The authors did not apply one anti-pad rule to every construction. Instead, they described a progressive approach: for the thinnest modeled cores, tune dielectric properties within a hybrid lamination; for intermediate cores, additionally adjust signal and ground via arrangement; for thicker cores, consider splitting the core and introducing further reference-plane controls. The published abstract uses the study-specific ranges of below 10 mil, 10–20 mil, and above 20 mil.
The same abstract reports a simulation target at 53.125 GHz and an improved simulated reflection response after optimization. These are results for the authors' modeled geometry and boundary conditions, not universal 224 Gb/s compliance thresholds, manufacturing rules, or proof of successful physical board qualification. At the time described in the abstract, a validation board was still being fabricated to assess measurement-to-model correlation.
The engineering lesson is broader than those three thickness bands. When a layer stack changes, review the via as a complete field structure. A solution that works for a thin core may not scale to a thicker inner core; conversely, a process change that reduces total thickness may change the available ground/reference arrangement or manufacturing sequence.
5. From an EM Model to Physical Evidence
For a transition that is important to the channel budget, a defensible validation plan needs both electrical and manufacturing definitions. A practical sequence is:
- Define the test object. Identify the exact layer transition, differential pair, launch geometry, reference planes and boundary of the structure to be evaluated.
- Freeze a manufacturable geometry. Include finished drill and plating assumptions, pad/anti-pad geometry, registration allowances, dielectric thicknesses and relevant material properties.
- Run the appropriate multiport analysis. Examine differential reflection and insertion loss, and check mode conversion when relevant, over a project-defined frequency band. Define port impedance and de-embedding boundaries.
- Design representative physical validation. Use a dedicated via test structure or an accessible board path; specify fixture and calibration approach before fabrication.
- Correlate and release. Compare model and measurement under matched reference planes, investigate meaningful differences, and document the approved configuration and acceptance criteria.
A physical test plan may not be practical for every prototype. In those cases, record the evidence gap rather than treating a nominal impedance result as a substitute. The owner of the SI requirement should decide whether model-only release is acceptable for the project stage, and what additional correlation is needed before qualification.
6. What to Check Before the Stack-Up and Via Design Are Released
The most useful review happens before a fabrication drawing becomes a frozen build instruction. At that point, engineers can still adjust the layer transition without creating uncontrolled changes between fabrication, SI modeling and NPI.
| Release item | Minimum evidence or decision | Accountable function |
|---|---|---|
| Electrical requirements | Interface assumptions; relevant frequency band; S-parameter convention and acceptance criterion | System / SI owner |
| Via topology | Layer span, microvia/PTH structure, stub strategy, signal and ground via positions | PCB design / SI |
| Manufacturing baseline | Finished dimensions, stack-up, materials, drill/plating and registration assumptions | Fabricator / NPI |
| Model definition | Port map, reference impedance, dielectric and conductor parameters, solver boundary | SI owner |
| Validation boundary | Coupon or board test structure, fixture, calibration, de-embedding, correlation plan | Test / SI / NPI |
| Change control | Which geometry or material changes trigger renewed electrical review | Design authority / SQE |
Do not add a generic “224G via compliant” stamp based on a simulation study, a single coupon or a quoted fabrication tolerance. The release record should identify what was evaluated, which limits were applied, and what remains unverified.
The Decision Is About Evidence, Not Just a Number
Controlled impedance remains necessary for high-speed PCB manufacturing. It is simply insufficient as the only evidence for a complex vertical interconnect. A straight trace, a board-level interconnect and a differential HDI via can all be called “impedance-controlled,” but they are not interchangeable test objects.
For channel-critical transitions, start with the actual stack-up and via geometry; assess the frequency-dependent response of the defined network; then establish whether the resulting simulation has been correlated with a representative physical structure. That sequence turns an abstract impedance target into a reviewable electrical requirement.
Review Your High-Speed PCB Stack-Up and Via Transitions
Share your stack-up, via drawing, design files and the electrical requirements that matter to your channel. UltroNiu can review the PCB manufacturing scope and identify items that require joint validation with your design and SI team. Specialized measurement and qualification deliverables must be agreed for each project.
Submit an Engineering Review


