SMT Stencil Incoming Metrology and Release Criteria

The conclusion first: by the time a stencil reaches your dock, whether it can be used is already a settled fact. Whether the apertures were cut where they belong, whether the foil thickness is uniform, whether tension is where it should be, whether the walls received any post-processing — all of that happened in the supplier's plant. Transport will not improve any of it and, under normal packaging, will not destroy it either. Incoming inspection does not create quality. It does exactly one thing: it reads an already-settled fact at the cheapest possible moment.

What happens in practice is that the step gets skipped and the first printed board stands in for it. That is the most expensive inspection method available: by then you have committed paste, laminate, labour and machine time, and a problem means stopping the line, cleaning, and rework. Worse, the symptom and the cause are rarely one-to-one — the same "insufficient paste" can originate in the aperture, the foil, the tension or the wall. Waiting for a printed board to start the investigation compresses a problem that could have been decomposed into four independent checks on a bench into one tightly coupled puzzle.

This article covers which groups of attributes can actually be measured at incoming, which line failure each group predicts, and where the three dividing lines fall: inspect every unit, sample, or substitute evidence for what cannot be observed at all.

1. Only four groups are measurable at incoming

A stencil is a structurally simple object — a metal foil with apertures in it, mounted in a frame. The number of independently observable attributes is therefore limited, and it comes to roughly four groups:

The table is not meant to be used as "check all four and sign". It is meant to be read backwards: when a specific failure appears on the line, go to the corresponding row first rather than starting from printer parameters. More importantly, the cost of verifying these four groups differs enormously, so they should not all be checked at the same intensity.

2. Aperture geometry and position: the only group worth 100% checking

The apertures are the reason a stencil exists, and they are the only group where being wrong leaves no recovery path. Slightly off foil thickness can be partly compensated with parameters; slightly loose tension can be traded against print speed. But an aperture cut in the wrong place, or an aperture that was never cut at all, cannot be compensated by any parameter on the line.

This is why this group sits on the "inspect every unit" side of the three dividing lines:

• Presence and form of apertures. Check every stencil against the drawing: confirm every aperture that should exist does exist, that the forms match the data, and that nothing was cut extra or omitted. The cost is low; the consequence of missing it is high.
• Relative positional relationships. The positions of apertures relative to each other, and relative to the fiducials, determine whether printing can be aligned at all. What you are looking at here is whether the overall positional relationship is self-consistent, not the absolute coordinate of any single aperture — that requires dedicated metrology conditions and is normally beyond line-side capability.
• Identification of anomalous apertures. Process anomalies during cutting are usually not uniformly distributed; they concentrate in a region or in one class of aperture. Visual inspection with magnification is looking for "this area differs from the rest", not "this aperture is off by X".

The boundary worth stating plainly: line-side work is verification and anomaly detection, not metrology. Absolute aperture dimensional accuracy is the supplier's process capability and should be evidenced by the supplier's own process records, not re-measured by the receiving party at the line — where neither the metrology conditions nor the business case exists. Misplacing that responsibility onto receiving usually produces numbers nobody is willing to act on.

3. Foil thickness and flatness: how wide the parameter window is

Foil thickness is easy to skip on the assumption that the supplier simply built to order, but what it governs sits further upstream than that: thickness sets the paste volume baseline, and thickness consistency sets the width of the parameter window.

Those two need to be separated. Thickness itself is a design-stage choice, decided together with pad size, component types and aperture geometry — that is a design decision and outside the scope of incoming inspection. What incoming looks at is consistency and condition:

• Whether thickness is uniform. If thickness varies across regions of the same foil, the paste volume baseline differs by region. Print parameters are applied board-wide, so such a difference cannot be removed by tuning; it simply shows up as "this region is always heavy or always light".
• Creases, pressure damage, localised deformation. This class of defect mostly comes from transport and storage, and incoming inspection is the only point where transport damage is genuinely intercepted. A creased area cannot conform to the board surface, so skipping is certain — and it is easy to spot visually, so missing it is purely a matter of not having looked.
• The relationship between the mesh plane and the frame plane. Whether the foil sits in the plane it should occupy determines contact behaviour during printing. This couples with tension and is covered in the next section.

Thickness consistency belongs on the sampling side: it is determined by the supplier's material and process, is usually stable within a batch, and gains nothing from 100% measurement. Creases and deformation, however, belong to 100% visual — the cost is near zero and it is the single checkpoint on the transport link.

4. Tension and frame: the only attribute that drifts

The first three groups share a property: once set, they stay set. Apertures do not move on their own, foil does not thin on its own, and wall condition changes slowly in normal use. Tension is different — it is the only one of the four that changes continuously with use time and use count, and that property demands a completely different management approach.

Two practical consequences follow.

First, the incoming tension value is not the in-service value. Confirming tension at incoming confirms only the starting point. As print cycles accumulate, as cleaning cycles accumulate, and as storage time passes, tension departs from its initial state. Tension management is therefore not a one-off release decision but a line that has to be watched across the whole life of the stencil — which is precisely why stencils need cleaning and life management, not just incoming inspection.

Second, tension failure presents as "repeatability got worse", not as "it printed badly". The characteristic symptom of insufficient tension is not that one board fails to print, but that variation between boards grows under the same parameter set. Problems of this class pass first-article inspection easily — the first article may be entirely normal — while producing scattered defects throughout the batch. So judging a tension problem means looking at the trend in within-batch consistency, not the result on a single board.

The frame-to-foil bond belongs to the same class: whether it is sound, whether any area has loosened, governs how the mesh behaves under print pressure. Visual inspection at incoming catches obvious abnormalities; subtle changes can only be inferred from in-service behaviour.

Tension therefore occupies a special position: incoming should confirm its state, but incoming confirmation must not be treated as the whole of release. It needs an in-service tracking line attached, or the attribute is simply unmanaged.

5. Wall condition and cleanliness: the part you cannot measure

Walls are the hardest of the four groups, for a direct reason: they determine release, and they are almost unobservable at the line.

Whether paste releases cleanly from an aperture depends on the friction between wall and paste, and on wall roughness and post-processing condition. These are properties at a microscopic scale requiring dedicated observation methods. Visual inspection with magnification — what a line can actually do — only finds gross anomalies: visible clogging, residue, or a region of apertures whose appearance differs from the rest.

So this group falls on the third dividing line: it cannot be measured, so supplier process evidence substitutes for measurement. Specifically, what should be requested and archived is:

• A statement of cutting method and parameters. Wall condition is a direct result of the cutting process, and the process route sets the basic character of the walls.
• A statement of wall post-processing. What was done after cutting (whether post-processing was applied, and which kind) often matters more to release than the cutting itself.
• Material batch information. This lets units from the same batch be read against each other, so the scope of an anomaly can be traced.
• Cleaning and packaging condition at dispatch. This sets the cleanliness starting point at arrival.

Archiving these is not paperwork for its own sake. The practical use is this: when a release problem appears on the line, there is a path to walk. Without those records, a release problem can only be expressed as "this stencil does not work well" — impossible to localise and impossible to avoid on the next order. That is the actual cost of treating walls as "unmeasurable, therefore unmanaged".

Cleanliness itself is time-varying, unlike the static properties of the wall: cleanliness at arrival is a packaging and transport question, while cleanliness in service is a cleaning-management question. Incoming only needs to confirm the starting point is normal; it cannot and should not predict what follows.

6. The three dividing lines: spend inspection effort where it pays

Putting the previous four sections together, the actual structure of incoming inspection looks like this:

The value of this table is that it rejects two common misallocations.

One is checking everything. Pushing every attribute to maximum intensity means the inspection time budget cannot hold, and the whole step gets skipped in practice. The most common reason an inspection procedure fails is not that the standard was too loose but that it was too strict for anyone to execute.

The other is checking only what is measurable. Because walls cannot be measured they get ignored, and release problems then never yield a root cause. Evidence substitution is not a downgraded fallback — it is the only effective method for unobservable attributes. Abandoning it means voluntarily giving up traceability for an entire class of problem.

One clarification is needed: these three lines define the receiving party's inspection intensity, not the ownership of quality responsibility. Absolute aperture accuracy, wall condition and foil consistency remain the supplier's process responsibility, and verification by the receiver neither replaces nor reduces it. Incoming inspection is risk interception, not a quality endorsement.

7. What the release record should contain, and who reads it later

Release records are the easiest thing to turn into a form that exists only to be signed. There is a practical test for whether yours is useful: when the same product is rebuilt six months from now, will anyone open it?

Records that do get opened usually contain:

• Which data revision this stencil corresponds to. After a product revision, the relationship between an old stencil and new data is the single most frequent source of trouble at rebuild. The record must uniquely identify the data revision, otherwise rebuild comes down to guessing.
• The actual state of the four attribute groups at incoming, not just a "pass". When a later problem requires deciding "was it always like this, or did it change in service", only the starting-point record can answer.
• The process evidence supplied (cutting method, post-processing, material batch). As covered above, this is the only traceability entry point for release-class problems.
• Actual behaviour on first use. Putting incoming state and first-use result side by side forms the baseline for this stencil. Any later deviation is read against it.
• The starting count for cleaning and use. Tension is time-varying; without a starting point there is no way to judge how far it has drifted.

Conversely, what the record should not contain is numbers the receiving party cannot actually measure but filled in anyway. A value with no metrology basis behind it is worse than a blank — it will later be used as a real baseline, turning a measurement that never happened into a fact that misleads decisions.

SKPCB runs multiple inspection gates through production (Q-GATES) and organises acceptance under the IPC framework (QC-IPC); assembly-side detection includes SPI, inline 2D AOI, post-reflow 3D AOI, X-Ray and ICT/FCT (PCBA-TEST). The purpose of a stencil release record is to connect this one link into that chain: a paste volume deviation seen at SPI should be traceable back to the incoming thickness and tension state, and a joint defect found by AOI or X-Ray should be matchable against the wall evidence. An isolated release record has no value; one wired into the chain does.

8. Turning this into a procedure

Finally, collapsing the above into an executable order. On arrival:

1. Verify the data revision — confirm this stencil corresponds to the currently valid revision. If the revision is wrong, nothing downstream matters.
2. 100% visual and magnified aperture verification: presence, form, positional self-consistency, anomalous-region identification.
3. 100% visual mesh condition: creases, pressure damage, localised deformation — the only checkpoint for transport damage.
4. Sample thickness consistency and initial tension state, recorded as a baseline rather than as a pass/fail verdict.
5. Archive supplier process evidence: cutting method, wall post-processing, material batch, dispatch cleaning and packaging state.
6. Record actual first-use behaviour afterwards, merging it with the incoming state into this stencil's baseline.
7. Enter in-service tracking: cleaning count, use count, tension trend — managed as life, not as a one-off release.

No step in this sequence requires a metrology lab at the line. The entire requirement is: look at what can be seen at the cheapest moment, substitute evidence for what cannot be seen, and wire both into somewhere they will actually be used.

FAQ

Q1: Can incoming inspection fully replace first-article inspection?

No — they examine different things. Incoming reads the settled state of one object, the stencil. First-article reads the actual result of an entire combination: stencil, paste, laminate, machine parameters and board support. Normal incoming with abnormal first-article is entirely possible, because the problem may sit anywhere else in that combination. The role of incoming inspection is to remove the stencil from the suspect list when the first article fails — or, conversely, to pin it down.

Q2: Should aperture dimensions be measured at the line?

Not advisable. Absolute aperture accuracy requires dedicated metrology conditions a line normally lacks; a value produced without those conditions can neither support a verdict nor support traceability, and in a record it actively misleads. What the line should do is verification and anomaly detection — are all required apertures present, do the forms match the data, does any region differ visibly from the rest. Absolute accuracy is supplier process capability and belongs in the supplier's process records.

Q3: Why must tension be managed separately instead of being judged once at incoming with everything else?

Because it is the only one of the four groups that changes continuously with use. Apertures, thickness and wall condition stay essentially constant in service, so one incoming confirmation covers them; tension departs from its initial state with print cycles, cleaning cycles and storage time. Judging tension only at incoming means knowing the starting point and not the present value. And since tension failure presents as degraded within-batch consistency rather than one bad board, first-article inspection rarely catches it.

Q4: If wall condition cannot be seen at the line, what is the practical use of requesting supplier records?

The use arrives when something goes wrong. Release-class problems — incomplete paste release, clogging, first-article deviation — look highly similar at the symptom level and cannot be separated by examining finished boards. With cutting method, post-processing statements and material batch records on file, you can judge whether the problem is process-inherent or batch-specific, and therefore whether to change supplier, change process route, or just quarantine this batch. Without them the conclusion stops at "this stencil does not work well", which gives you no way to avoid a repeat on the next order.

Q5: Which release-record item is most often omitted, with the largest consequence?

The data revision. After a product revision, the correspondence between an old stencil and the new data is the highest-frequency problem source at rebuild, and it is completely inconspicuous at incoming time — the stencil looks fine, because it genuinely is fine; it just matches the previous revision. If the record cannot uniquely identify the data revision, rebuild depends on someone's memory or on re-comparing from scratch, and neither is reliable.

Q6: Should sampled thickness data be written as pass/fail?

Record it as a baseline state rather than a verdict. The purpose of line-side sampling is to establish a reference for "roughly what state this stencil left the plant in", for comparison against later deviations. Writing it as a pass/fail verdict loads a sampling observation with more conclusive weight than it can carry. The actual acceptance verdict should rest on the supplier's process records and the acceptance terms agreed between the two parties.

Related links

• SMT stencil manufacturing service
• Quality system and inspection gates
• Products and manufacturing capability
• Aperture area ratio and paste release
• Wall polish and nano-coating selection
• Laser cutting versus electroforming
• Stencil data package: paste layer to cut file
• Defect classification and IPC acceptance class