The failure signature: a relay that passes, then shorts
A relay leaves the line with contact resistance in spec, dielectric withstand passing, and a functional test that cycled it a few hundred times. Months later it is a short across adjacent terminals. The instinct is to look for a test escape — a marginal hip check, a bent contact, contamination on the armature. Those are worth ruling out, but they are not the mechanism that fits the timeline. The mechanism that fits is a conductive filament that grew after the test, from a tin-rich surface, until it touched something it should not have.
Tin whiskers are conductive crystalline structures of tin growing outward from tin-rich surfaces, formed by addition of atoms at the base rather than the tip (Panashchenko and Osterman, Examination of Nickel Underlayer as a Tin Whisker Mitigator, ECTC 2009). Lengths run from a few micrometers to millimeters; thicknesses are typically 0.5–10 µm; densities range from a few whiskers to thousands per component; and the process may take hours, days, or years. That last clause is the whole problem. A relay that passed every test is not exculpated by the test — it is simply earlier in the growth curve.
NASA’s tin whisker reference page includes an image captioned “Tin Whisker Growing from Pure Tin Plated Relay Terminal” (NASA NEPP Tin Whisker Homepage). That is the specimen class to keep in mind: not the solder joint, but the terminal finish. The relevant surface is the tin plating on the relay terminal, and the relevant failure is a bridge between adjacent conductors that were within a whisker length of each other.
What the whisker does electrically
The 2009 Panashchenko/Osterman presentation lists three failure modes induced by whiskers: electrical short circuit; permanent short if the current exceeds the whisker’s melting current; and metal vapor arc, where high current and voltage vaporize the whisker into a conductive plasma of metal ions capable of sustaining hundreds of amps. For a signal-level relay, the first mode is the one that shows up as an intermittent or hard short. For a relay switching higher energy, the arc mode is the one that turns a filament into a welded fault.
This is why the post-mortem often finds no whisker at all. A whisker that carried enough current to melt is gone; what remains is a small crater or a discolored contact face, and the failure looks like a mystery weld. If the relay failed at low energy, the filament may still be present, sometimes folded against the contact or broken loose inside the housing.
Why “it passed every test” is not exculpatory
The most useful sentence in the 2009 NASA/UMD work for anyone writing a test plan is this: “Environmental tests provided no acceleration as compared to room-ambient growth, but instead – induced growth.” That is a direct challenge to the assumption that a temperature-cycling or humidity qualification screen has exercised the whisker mechanism on behalf of the field. It has not. It has grown whiskers on its own schedule, and the field will grow more.
So the bench question after a delayed short is not “what did the test miss?” It is “what grew after the test, on which surface, and how far was it from the nearest conductor?” That reframing changes what you inspect and what you record.
Bench inspection: a measurement discipline, not a glance
Hewlett-Packard’s 2005 tin whisker inspection training material (HP experience with tin whisker inspection training, V. Schroeder) is the most concrete public account of what happens when capable people try to inspect for whiskers. The recommended optical setup was a microscope with up to 300× magnification, a stage with 360° twist and 180° rotation, and digital image acquisition with calibrated length measurement. Initial inspection was recommended at 160×; whisker determination at 300×.
The training results are the part to read before writing your own inspection procedure. Six trainees inspected four samples. On the same part, the average maximum whisker length reported ranged from about 52 µm to about 85 µm depending on the operator, with standard deviations of 9–17 µm. When operators measured what was ostensibly the same whisker, the standard deviation was 7–48% of the length measured, and 5–30 µm in absolute terms. One whisker was measured as 108 µm and 70 µm depending on rotation. With a 40 µm pass/fail criterion in play, that error band is larger than the criterion.
The practical consequences for a failure investigation:
- Record the viewing angle for every measurement. A length without an angle is not a measurement; it is an observation.
- Use multiple views. The HP training found that operators did not agree on which lead carried the longest whisker, and the same set of six longest whiskers was rarely identified by more than one operator. A single top-down look will miss filaments lying along the lead edge.
- Control the lighting. The HP guidance was to turn down or off the microscope’s built-in light and use two fiber lights, with at least one illuminating from below. Whiskers are thin and specular; coaxial illumination can make them disappear.
- Do not touch the plating. Handle the relay by the body, not the terminals. A whisker is a surface feature and handling destroys the evidence.
The 2009 Panashchenko/Osterman presentation also describes a tilt-offset measurement method — two images separated by a known tilt angle — as a more consistent way to estimate whisker length than the JEDEC-recommended single-view method, particularly when there are too many whiskers to tilt each one individually. If your lab does whisker measurement routinely, that method is worth adopting; if you are doing a one-off failure analysis, the minimum is to document the angle and take more than one view.
What the micrograph has to show
A whisker is a single crystal, often with a faceted or striated surface, growing out of a grain boundary or a scratch in the tin finish. It is not a dendrite (which grows in the presence of moisture and ionic contamination, usually with a branched or fern-like morphology) and it is not solder wicking or a sliver of flashing. Under SEM, the base attachment and the absence of a molten or resolidified zone distinguish it from a broken weld or a melted filament. Under optical microscopy at 160–300×, the practical discriminators are: it emerges from the plating surface, it is metallic and reflective, it is straight or kinked rather than branched, and it is rooted — not lying loose on the surface.
If the relay failed at low energy and the filament is still bridging, photograph it in place before any cleaning or handling. If the relay failed at higher energy, look for the crater and the discoloration pattern on both contact faces; the arc leaves a signature on the side that received the plasma, and the whisker’s origin may be identifiable as a small pit on the plating near the contact gap.
Mitigation: what the evidence actually supports
The 2009 NASA/UMD test is a useful corrective to mitigation folklore. In that work, a 2 µm nickel underlayer under an 8 µm tin electroplate was not effective in preventing tin whisker growth for the tested finish. The authors’ summary line is blunt: “Mitigation ≠ Elimination.” Nickel underlayers are widely cited as a mitigation; the cited test says they are not a guarantee, at least not for that finish and that test.
That does not mean nickel underlayers are useless. It means an engineer should not treat “Ni underlayer” as a checked box without finish-specific evidence — a supplier’s whisker test data on the actual plating system, at the actual thickness, under the actual environmental conditions. The same discipline applies to conformal coating: it is a layer of defense that can slow growth and reduce bridging risk, but it is not a barrier that can be assumed to hold for the service life of the relay.
The defensible posture at the bench is to assume some whiskers will grow and to design the inspection and the geometry to catch bridging before it becomes a field short. That means:
- Identify every pure-tin finished terminal in the assembly. The finish is on the component datasheet, but it is worth confirming by XRF on a sample, because supplier finishes change and because “matte tin” is not a standard-defined category (the 2009 presentation notes there is no standard definition of matte versus bright finish).
- Document the inspection angle set for each terminal geometry. A relay with terminals on 2.5 mm centers and a relay with terminals on 5 mm centers do not have the same bridging risk, and the inspection angles that reveal a whisker on one may not reveal it on the other.
- Treat any filament that bridges or nearly bridges as a failure mechanism, not a cosmetic finding. A whisker that is 80% of the gap is a field short waiting for a vibration event or a thermal cycle to close the last 20%.
- Do not rely on the qualification test as evidence of field margin. The 2009 result — environmental tests induced growth rather than accelerating it — means the qualification test tells you the part can grow whiskers, not that it has stopped.
What to write in the failure report
A report that says “tin whisker found” is not actionable. A report that says “a 62 µm filament, measured at 45° tilt with two fiber lights, rooted at a grain boundary on the pure-tin plated terminal of the coil-side contact, bridging to the adjacent terminal at a gap of 1.8 mm” is actionable — it tells the next engineer which finish, which geometry, and which inspection angle to look at on the next build.
The evidence-first version of this failure is not a story about lead-free solder. It is a story about a specific surface finish, a specific gap, and a specific time after test. The relay passed every test because the whisker had not grown yet. The next one will pass every test too, unless the inspection and the mitigation are aimed at the mechanism rather than at the test.
FAQ
Is the solder alloy the problem, or the terminal plating?
The cited evidence concerns whiskers growing from tin-rich surfaces. On a relay, the relevant surface is typically the terminal plating, not the solder joint. The failure investigation should identify the finish on the terminals that bridged, not assume the solder alloy is the cause.
Can a whisker short be intermittent?
Yes. A whisker that is close to bridging but not touching can close the gap under vibration or thermal expansion, then open again. That is why a relay that “works now” after a field return is not evidence that the whisker was not the cause.
Does a conformal coating eliminate the risk?
No. The 2009 NASA/UMD work states the general principle directly: mitigation is not elimination. Coating is a layer of defense; it should be combined with finish selection, geometry, and inspection rather than treated as a substitute for them.
How long after assembly can a whisker grow?
The 2009 presentation states the process may take hours, days, or years. There is no service-life cutoff after which a pure-tin finished relay is safe from whisker growth.