Diagnosing a Via Plating Defect in Production
An intermittent communication fault on a small percentage of units is one of the harder problems to solve, because the board passes its functional test when it is cold and fails after it has been running for a while. The case below describes how a via plating defect was isolated on a batch of networking equipment, and it illustrates a diagnostic sequence that applies to any fault that appears only under specific conditions.
The Symptom
During pilot production of a new device, units were subjected to extended ageing and then to a system performance test. Two units lost their communication link, a failure rate of about four percent of the units tested. The failure was not reproducible on demand, and the boards had passed the earlier stages of production testing, which meant the fault was either marginal from the start or produced by the ageing process. Both possibilities point to a connection that is present but not sound.
Step One: Confirm the Failure and Clear the Obvious Causes
The first task is to establish that the fault is real and to eliminate the explanations that are easy to test. The supply voltages of every functional block were measured and found to be within specification, and the phase-locked loop of the main logic device was checked and found to be stable. This matters because an unstable supply or a jittering clock produces symptoms that resemble a communication fault, and ruling them out prevents the investigation from following a false trail.
Step Two: Narrow the Fault Using Log Data
With the obvious causes excluded, the next step is to localise the problem rather than to inspect the whole board. Diagnostic output from the failing unit was compared against the data the link partner was transmitting, which identified the specific signal path on which the corruption occurred. This is the point at which diagnosis becomes efficient: a board has thousands of nets, and a fault that reduces to one of them can be investigated with a probe instead of a rework station.
Step Three: Resistance Measurement Along the Net
The suspect path was then measured in sections to find the point of abnormal resistance. Possible causes include a pin that is not soldered, a bridge between adjacent pins, a defective via, or a short or open in the copper. Microscopic inspection showed that the soldering was sound. Comparative measurement found that the resistance across the suspect via was about 500 milliohms, while a comparable via elsewhere on the same board measured around 20 milliohms. That difference of more than an order of magnitude in contact resistance identified the via as the source, and it is the point at which the fault stopped being a puzzle and became a measurable defect: the barrel was not fully plated, so the connection existed but with far too much resistance.
Step Four: Confirm the Mechanism by Heating
A hypothesis is only useful when it can be tested. A via with incomplete copper plating behaves differently from a sound one as it warms, because the laminate expands more than the copper, and the expansion pulls on the plating. The suspect via was heated with hot air to about 100 degrees Celsius and measured again, and its resistance increased with temperature, while a good via in the same area did not change materially. That behaviour is consistent with a cracked or thin barrel and inconsistent with any other explanation, and it also explains why the fault appeared only after the unit had been running: the defect is a thermal effect. The general defect mechanisms behind this behaviour are described in this article on copper plating defects and prevention.
Step Five: Prove It With a Fly Wire
The final step is to remove the suspected cause and see whether the symptom disappears. A fly wire was soldered from the top of the board to the bottom, bypassing the suspect via, and the unit was returned to the system test for 72 hours. All functions remained normal and no communication fault occurred, which confirms the diagnosis to a level that inspection alone cannot reach. Confirmation by substitution is worth the time it costs, because it converts a plausible explanation into a demonstrated one and prevents a corrective action being applied to the wrong thing.
Root Cause and Corrective Action
The root cause was insufficient copper in the plated barrel of the via, which produced a connection whose resistance was high enough to disturb a high-speed link and which worsened as the board warmed. The corrective actions followed three lines. The finding was reported to the board supplier with the measurement data, and improved samples were requested for volume testing. A second fabricator was qualified in parallel, so that the product was not dependent on a single process. Finally, a via inspection requirement was added to the incoming inspection procedure and enforced at goods receipt, so that the defect would be detected before assembly rather than after the product had been built.
Why the Defect Escaped Earlier Screening
Understanding why a defect was not caught is as useful as understanding what caused it. In this case three factors combined. The defect was intermittent, so a single test pass proved nothing. The resistance was high enough to disturb a link but low enough that a continuity check treated the net as connected. And the defect worsened with temperature, so a unit tested at room temperature and then shipped behaved differently from one that had been running in an enclosure for an hour. A screening process that does not apply the stress under which the failure occurs, and that does not measure a parameter rather than a state, will pass this kind of defect every time. Thermal cycling during product validation would have exposed it earlier, because the resistance of a poorly plated barrel changes with temperature in a way that a sound one does not.
What the Design and the Test Should Do Differently
Two lessons extend beyond this case. The first is that a production test which only checks continuity will not find a via with excessive resistance, because the net is connected. Where a link is sensitive, the test should include a resistance measurement across critical vias, or an in-circuit test with a threshold on the measured value rather than a simple pass or fail. The second is that the layout should give the vias the best chance of being plated well: adequate annular ring, a hole size within the fabricator”s demonstrated capability, and enough copper to plate. Those requirements are described in this article on via design rules, and the qualification of a stackup before volume production is described in this article on multilayer PCB prototype requirements.
FAQ
Why did the board pass its original test? Because the via was connected. A partially plated barrel still conducts, so a continuity check passes; the defect only appears as a resistance that is too high and that grows with temperature.
How was the via distinguished from a solder defect? By inspection and by measurement. Microscopic inspection showed sound solder joints, and comparative resistance measurements showed that only the via was abnormal.
How can this be prevented in future production? Specify a hole size and annular ring the fabricator can plate reliably, qualify the process on a prototype, and add a resistance limit for critical vias to the incoming inspection procedure.
