PCB Edge and Outer Layer Inspection: What You Can Actually Judge On

The dispute is never really about the defect

A board edge or outer layer complaint almost never arrives as a technical question. It arrives as a photograph.

Someone in receiving inspection holds a board up, sees a nick along one edge or a patch where the surface looks uneven, and sends an image with three words attached: *is this acceptable?* And at that moment nobody on either side of the transaction can answer, because the question as asked has no criterion behind it.

The supplier will say it is within normal variation. You will say you can see it from across the room. Both statements are true and neither is a judgement. What follows is usually a week of photographs going back and forth, then a commercial settlement that has nothing to do with the board and everything to do with how much either side wants the next order.

This happens at the board edge and on the outer layer more than anywhere else on a bare PCB, and for the same underlying reason: both locations are visible to the naked eye. You do not need equipment to find a finding there. Hole wall quality, internal registration, dielectric thickness — those require instruments, which means by the time anyone is looking, a method has already been agreed. Edge and surface findings get discovered by someone who was not looking for them, with no method in hand, and that is why they become arguments.

The way out is not a better camera. It is recognising that the findings in these two locations fall into two groups with completely different rules:

• Functional findings affect whether the board can do its job. The criterion is performance, the evidence is measurement or test, and the burden sits with the party claiming the board still works.
• Cosmetic findings affect how the board looks. There is no performance criterion to appeal to, so the only thing that can settle it is a limit agreed in advance — and if none was agreed, the burden sits with the buyer who is now trying to introduce one.

Sort your finding into the right group and most disputes resolve themselves. Fail to sort it, and you will argue about a cosmetic issue as though it were a functional one, which is an argument you cannot win because you are asking the supplier to disprove a standard that does not exist.

Board edge: the four findings worth separating

The edge is a cut surface. Nothing about it was imaged, plated or printed — it was produced by removing material, and every removal method leaves a signature. Four findings dominate, and they do not belong in the same category.

1. Edge dimension and profile deviation — functional. The board does not fit, the cutout is in the wrong place, the outline does not match the drawing, the tabs that should have broken flush are standing proud enough to interfere with a housing. This is the easy class, because it is measurable against a drawing and the drawing is the criterion. If the outline is out of tolerance against what you supplied, there is nothing to argue about. Note the condition: *against what you supplied.* An edge that deviates from what you assumed but matches what your file says is your finding, not the supplier's.

2. Edge material condition — nicks, burrs, delamination and fibre exposure. This class splits, and the split is the single most useful distinction on the whole page. A raised burr that interferes with assembly, or an edge where the laminate has opened up, has a functional consequence you can articulate. A small nick in a region with no copper nearby, on an edge that goes into open air, does not. The same visual vocabulary covers both, which is exactly why "there is a nick on the edge" is not a complaint anyone can act on. The complaint that can be acted on names the consequence: it interferes with the rail, it is within the keep-out, it has exposed fibre where the edge is a sealing surface.

3. Copper-to-edge relationship — functional, and the one most often discovered late. Where a conductor, plane or pad runs close to the cut line, the edge stops being a cosmetic surface. The cut can expose copper, smear it across the edge, or leave it close enough to the edge that handling damage reaches it. This is a real functional class and it is also frequently a design-side finding rather than a manufacturing one: if the artwork placed copper at the edge, the cut did what it was told. Which is why the edge keep-out on your drawing matters more to this dispute than any inspection method does.

4. Edge appearance — roughness, colour variation, visible cut texture — cosmetic. How the cut surface looks. Different removal methods produce visibly different textures, and a board that looks rougher at the edge than the last batch may simply have been separated differently. There is no performance criterion here at all. If this matters to you — because the edge is visible in the finished product, or because it touches a user — it has to be declared as a requirement before manufacture, not raised as a defect after.

What to take from the list: only one of the four classes is purely cosmetic, but it is the one that generates most of the photographs. Before you raise an edge finding, decide which of the four you are actually holding.

Outer layer: the four findings worth separating

The outer layer is the opposite kind of surface. It was imaged, etched, coated and printed, so every finding on it points at a specific thing that happened to it. Again, four classes.

1. Conductor geometry — functional. Trace width, spacing, and whether a conductor is continuous. This is the class that gets tested rather than looked at, and it is also the class where the test already exists: electrical test finds opens and shorts, and conductor width is measurable. If you have a geometry finding, you are not in a visual dispute at all — you are in a measurement, and measurements settle.

2. Surface finish coverage and condition — functional, with a cosmetic shadow. Whether the finish is present where it should be, whether it is continuous across a pad, whether the pad will accept solder. The functional question is solderability, and it has a test. The cosmetic shadow is colour and uniformity: a finish that is intact and solderable can still look patchy, mottled or inconsistent between panels. Those two things get reported in the same sentence by receiving inspection, and separating them is the entire job. *Will it solder* is answerable. *Does it look even* is only answerable against a declared limit.

3. Soldermask condition — the class that splits hardest. Mask misregistration that encroaches on a pad is functional and consequential. Mask that is thin over a conductor, or that has a pinhole in an area with no adjacent conductor, sits somewhere in between and the answer depends on what the mask is being relied on to do in that location. Mask colour variation, gloss difference, or minor surface texture is cosmetic. One vocabulary, three outcomes. This is why a mask complaint has to name the location and what the mask is doing there, not just the appearance.

4. Legend and marking — mostly cosmetic, occasionally functional. Legend that is offset, broken, low contrast or illegible. Cosmetic by default — but it becomes functional the moment something downstream depends on reading it. If a legend carries a part reference an operator uses during assembly, a date code, a serial mark, or an orientation indicator, then legibility is a requirement and not a preference. Decide which of those two you are in before you raise it, because the two have different criteria and different burdens.

The parallel with the edge list is the point: in both locations, the findings that look most alarming are often cosmetic, and the ones with real consequences are often the ones nobody photographed. Copper at the edge and mask encroachment on a pad are both quiet findings. Rough edge texture and uneven finish colour are both loud ones.

Why the functional/cosmetic split decides who has to prove what

This is the part that changes how a dispute goes, and it is not obvious.

For a functional finding, the criterion already exists — in your drawing, in your file set, or in the test the board is supposed to pass. You do not have to establish that a short circuit is unacceptable. You point at the net list and the matter is closed. The practical burden lands on whoever is asserting the board is still fit: if the supplier wants to ship a board with a finding you have characterised as functional, the supplier has to show it performs. Measurement, continuity, solderability, fit against the drawing. Evidence exists and it is producible.

For a cosmetic finding, no criterion exists unless someone created one. And that has a consequence buyers routinely walk into: if you did not declare an appearance limit, you are not enforcing a standard, you are proposing one after delivery. The burden is yours, and it is a burden you cannot discharge, because there is nothing to appeal to except your own reaction to the photograph.

That asymmetry explains the shape of every stuck dispute in these two locations. The buyer has a cosmetic finding and is arguing it with functional energy. The supplier has nothing to disprove, so it does not engage with the technical content at all and moves straight to a commercial offer. Both sides then negotiate about money while believing they are discussing quality.

Two consequences worth acting on:

• If your finding is functional, say what it affects and stop describing how it looks. "Copper is exposed at the cut edge inside the keep-out" obliges a response. "The edge looks bad" does not.
• If your finding is cosmetic and you have no declared limit, you have a commercial conversation, not a quality claim. That is not the same as having no case — appearance can matter and suppliers do accommodate it — but recognise which conversation you are in. Dressing it up as a technical rejection costs you credibility on the next finding, which may be a functional one.

The four declarations that have to exist before anything can be judged

A judgement needs something to be judged against. In these two locations, four declarations carry almost all the weight, and all four are cheap to make before manufacture and impossible to retrofit after.

1. The acceptance class the board is built to. IPC-A-600 is the common framework for bare-board acceptability and it is written around classes with different criteria. Which class applies to your board is a decision, not a default, and it is the difference between two boards that are physically identical and legitimately judged differently. Naming the class in the RFQ is the single highest-value line in this entire article, because it converts a large number of appearance questions from arguable into looked-up. Not naming it means every finding starts from zero.

2. The inspection method and condition. A finding discovered under magnification and a finding discovered at normal viewing distance are not the same finding, and a surface judged under bench lighting can look different under daylight. If the method is not stated, each side inspects the way that supports its position and both are inspecting honestly. State the viewing condition and magnification for visual acceptance, and the dispute stops being about whose eyes are right.

3. The drawing-side inputs that make edge findings decidable. Edge keep-out for copper, which features are dimensioned and to what tolerance, whether tab or break locations are specified or left to the supplier, and whether any edge is a functional surface — a sealing face, a sliding face, a user-touchable face. Most unresolvable edge disputes trace back to one of these being absent, which means the supplier made a reasonable choice and you are now unhappy with a choice you delegated.

4. Any appearance requirement that goes beyond the class. If the board is visible in the finished product, if a legend must be readable by an operator, if finish uniformity matters for a reason the class does not cover, that is a requirement and it belongs in the RFQ. Declared in advance it is a specification the supplier can price and build to. Raised after delivery it is a preference, and preferences do not travel backwards through a purchase order.

Notice none of these four requires either party to publish a number. They require the framework, the method, the drawing inputs and the extras to be named. The limits then come from the class you named, and both sides read the same document.

What order to put evidence on the table

When a finding is already in dispute, the sequence matters more than the content. Run it in this order, because each step can end it without the next one being needed.

Step one: which of the eight classes is it? Four edge classes, four outer layer classes, from the two lists above. This costs you nothing and resolves a surprising share of cases, because it frequently reveals that the thing being argued about is cosmetic and that no limit was ever declared. Better to know that internally before you send the email.

Step two: name the consequence, not the appearance. For a functional finding, state what it prevents: the board does not fit, the pad will not accept solder, copper is exposed where the drawing says it must not be, this net is open. A consequence is answerable. An adjective is not.

Step three: produce the declaration you are judging against. The acceptance class, the drawing dimension, the keep-out, the stated appearance requirement. If you can point at a document, you are enforcing an agreement. If you cannot, skip ahead to step six, because steps four and five will not go anywhere.

Step four: establish the inspection condition. Say how the finding was observed — viewing distance, magnification, lighting. This matters because it lets the supplier reproduce the observation. A finding the supplier cannot reproduce is a finding the supplier will dispute, and reasonably so.

Step five: ask for the performance evidence, not an opinion. For functional findings, the question is not "do you agree this is a defect" but "show that the affected boards meet the requirement". Solderability on the affected pads. Continuity on the affected nets. Dimensional check against the drawing. This is where a functional dispute resolves, and it resolves on data rather than on who is more insistent.

Step six: if it is cosmetic and undeclared, move it to the commercial track openly. Say that you recognise no limit was specified, that the appearance matters to you for a stated reason, and ask what can be done on this batch and declared for the next. This closes faster than the alternative and it costs you nothing you actually had. The thing you gain is a declared limit going forward, which is the only durable fix.

The ordering has one more benefit worth naming: it keeps your functional findings credible. A buyer who sends every visual observation as a rejection trains the supplier to treat all of them as negotiating positions. A buyer who sorts first and sends consequences gets a different response on the finding that genuinely matters.

The findings you will not be able to judge, no matter what you write

Some things cannot be settled at incoming inspection on a bare board, and recognising them early saves the week you would otherwise spend trying.

Subsurface conditions you can see but cannot bound. Something visible through the mask or at a cut edge that suggests an internal condition. You can see it; you cannot characterise it without destructive work on a sample. Destructive evaluation consumes a board and needs to be agreed as a step with an owner, which is a different conversation from a visual acceptance.

Anything whose criterion is "different from the last batch". Batch-to-batch appearance variation within an acceptance class is not a defect. It is extremely annoying, and it is the single most common cosmetic complaint, and it is also the one with the weakest footing — because the previous batch was never the specification. If consistency between batches matters to you, that is an appearance requirement to declare, which puts it in declaration four above rather than in a rejection.

Findings that depend on an undeclared downstream process. Whether a surface condition will cause a problem during your assembly depends on your assembly, which the bare-board supplier did not specify and cannot predict. If a surface or edge condition has to survive a specific downstream step, that step is an input you owe the supplier, not a consequence you can assert afterwards.

Anything on a board that has already been through your process. Once a board has been handled, baked, assembled or reworked, an edge or surface finding can no longer be cleanly attributed to the bare-board supplier. This is the time-sensitivity that makes incoming inspection worth doing at all: the window in which attribution is clean closes quickly, and it closes quietly.

Each of these is best handled the same way — as an input or a declared requirement before manufacture, rather than a judgement after delivery.

How to put this into the RFQ

Six lines cover both locations, and none of them asks the supplier for a number it has not seen your files for.

1. The acceptance class the board is to be built and inspected to.
2. The visual inspection condition for appearance acceptance — viewing distance, magnification, lighting.
3. Edge keep-out for copper, stated on the drawing, plus which edge dimensions are toleranced.
4. Which edges, if any, are functional surfaces, and what they have to do — fit, seal, slide, be touched.
5. Any appearance requirement beyond the class, with the reason, so it can be built to rather than inspected for.
6. Whether legend legibility is a requirement, and if so which markings and for whom.

Send these with your Gerber, drill, drawing and acceptance requirements when you request a quote. A supplier that answers all six in writing has given you a criterion set you can judge against, which is the thing you actually needed when you held that board up to the light.

For the manufacturing scope and verification these sit inside, see the capabilities overview and the product range.

Related reading on PCB168

• Board edge and depaneling acceptance criteria — how the separation method itself sets what the edge can look like
• Outer layer etching defect review — identifying what a surface finding actually is
• Cosmetic defect repair disclosure and acceptance — the obligation to disclose boards that were repaired
• Incoming inspection and acceptance plan for PCB delivery — building the plan these criteria sit inside
• Incoming sampling plan, retest and dispute ownership — how many boards to check and who owns a retest
• PCB quality inspection methods — the method inventory behind the criteria
• Request a quote

FAQ

Can I reject a bare PCB because the board edge looks rough?

Only if roughness was declared as a limit, or you can state a functional consequence. Edge texture is a product of how the boards were separated, and different methods produce visibly different surfaces within the same acceptance class. If the edge is visible in your finished product or touches a user, that is a legitimate requirement — but it has to be an RFQ input, because after delivery you are proposing a standard rather than enforcing one. If instead the edge interferes with a rail, breaches a keep-out, or has exposed fibre on a sealing face, say that, and you are on functional ground with evidence available.

What is the difference between a functional and a cosmetic finding in outer layer inspection?

A functional finding has a performance consequence you can name and test — the net is open, the pad will not solder, mask has encroached on a pad that needs to be wetted. A cosmetic finding affects appearance only — colour variation, gloss difference, uneven finish tone, offset legend that nobody has to read. The distinction matters because it decides who must produce evidence. Functional findings are settled by measurement against a criterion that already exists. Cosmetic findings can only be settled against a limit declared before manufacture.

Our supplier says the finding is within IPC. How do we check that?

First confirm which acceptance class your order was placed against, because the criteria differ by class and a board built to one class is legitimately judged by that class. If no class was named in the RFQ or the quotation, there is no agreed basis in either direction, and that is the actual finding to fix. Then establish the inspection condition — viewing distance and magnification — so both sides are observing the same way. Most "within IPC" disagreements turn out to be a class question or a viewing condition question rather than a disagreement about the board.

Copper is exposed at the cut edge. Is that the fabricator's problem or ours?

It depends on what the artwork and drawing specified. If your files placed copper closer to the cut line than the stated edge keep-out allows, the cut did what the data told it to and the finding originates on the design side. If copper was within the declared keep-out and is still exposed or smeared at the edge, that is a manufacturing finding with a clear functional consequence and it should be raised as one. This is why the keep-out on the drawing does more work in this dispute than any inspection method — it decides in advance which answer applies.

This batch looks different from the last one but I cannot point to a defect. Do I have a case?

Not as a rejection, if both batches are inside the acceptance class. Appearance variation between batches within a class is not a defect, and the previous batch was never the specification. What you do have is a reason to declare a consistency requirement going forward, which turns it from a recurring argument into a stated input. If the variation is large enough that you suspect something changed in how the boards were made, ask that question directly rather than framing it as an appearance rejection — a process change question gets a factual answer, an appearance claim without a limit does not.

Does legend quality count as a defect if it is hard to read?

By default legend is cosmetic, so an offset or low-contrast marking is not a rejection on its own. It becomes functional the moment something downstream depends on reading it — an operator placing parts by reference designator, a date or serial marking used for traceability, a polarity or orientation indicator. If that is your situation, legibility is a requirement and belongs in the RFQ with the specific markings named, so it is built to rather than argued about.

How long do we have before an edge or surface finding stops being attributable to the supplier?

Shorter than most buyers assume. Once boards have been handled, stored, baked, assembled or reworked, an edge or surface condition can no longer be cleanly attributed to the bare-board stage, because your own process is now a plausible source. That is the practical case for inspecting these two locations early and documenting what you see on arrival with a dated record — not because early inspection catches more, but because it is the only window in which attribution is clean.

Related links

• A flat green rectangular bare circuit board photographed against a white background at a slight angle, with a rectangular cutout window near its centre, a concentric rectangular trace pattern across the surface, two small protruding tabs along the bottom edge each carrying a group of gold dot-grid pads, one further small tab at the upper right with a group of gold dot-grid pads, and small printed text near the centre.