Sunking PCB

Inner-Layer AOI: The Last Moment Inner Layers Are Visible

Design Guide2026-10-16Sunking Circuits Technical Team
Inner-Layer AOI: The Last Moment Inner Layers Are Visible

The short version: the inner layers of a multilayer board are visible only before lamination. After lamination they cannot be seen, cannot be reworked, and cannot be sampled. Inner-layer inspection is therefore not an ordinary stage in the quality chain — it is a one-way gate. Whatever it releases is what gets delivered.

1. The irreversibility is structural, not attitudinal

Most inspection stages have a second chance behind them. If outer-layer imaging goes wrong, the panel is still open: it can be re-evaluated, treated locally, re-imaged if necessary. If solder mask prints off-position, remedies exist. Release decisions at those stages carry a cushion — being wrong is recoverable at the next step.

Inner layers have no cushion, and the reason has nothing to do with how strict anyone is. It is that the panel has physically become one body. A multilayer board is built by imaging each inner layer separately, then stacking those layers with dielectric and bonding the whole assembly under heat and pressure. From the moment lamination completes:

  • It cannot be seen. The inner layer sits between dielectric and outer copper. No surface inspection method reaches that plane.
  • It cannot be reworked. No process opens a bonded panel, corrects one inner trace, and closes it again.
  • It cannot be sampled. Confirming the real state of an inner layer requires destructive cross-sectioning, and the sectioned coupon is no longer product.

Those three facts combine into one consequence: a defect released at inner-layer inspection will not be found and corrected downstream — it will be laminated into the finished board as-is. That is a statement about process order, not an exhortation to take inspection seriously.

Bond and fill failures created *by* lamination itself are a different class of problem from imaging defects that already exist *before* lamination; see preventing lamination voids and delamination.

2. Why downstream inspection cannot cover for it

A common assumption is that even if inner-layer inspection misses something, electrical test, visual inspection and final inspection will eventually screen it out. That reasoning collapses "can detect anomalies" into "can detect this anomaly."

Electrical test verifies connectivity, not geometry. Continuity and isolation testing answers whether what should be connected is connected and what should be separated is separated. An inner trace narrowed by over-etch but still conducting will pass, because under test conditions it genuinely is conducting. This class of defect is "works now, margin consumed," and margin is outside what the test evaluates. For differences between test methods, see flying probe versus fixture test.

Outer-layer inspection looks at outer layers. Outer-layer AOI compares outer geometry against its data. Inner layers are not in its field of view, no matter how carefully it is run. How that inspection should be organised is a separate topic: outer-layer pattern inspection.

Visual and dimensional inspection look at reachable surfaces and edges. They have no view of inner layers either.

So every downstream stage is doing exactly its own job; none is negligent. The issue is that none of their observation targets includes an already-encapsulated inner layer. Relying on downstream stages to catch inner-layer defects assigns a task to a set of processes that cannot see it.

3. What this gate does catch

Inner-layer inspection happens while the panel is still open and the pattern can still be magnified and compared against its data. Per common practice, the categories it addresses include:

  • Connectivity: whether the pattern is continuous where it should be and separated where it should be — the pre-forms of opens and shorts.
  • Line geometry: whether traces and spaces deviate in form, including deformation from over-etch and under-etch, and edge irregularity.
  • Residue and foreign material: whether copper that should have been removed is still present, or whether anything is attached to the surface.
  • Feature correspondence: whether pads, annular features and similar elements correspond to the data or have shifted.

Whether those readings are usable downstream depends on one easily overlooked prerequisite: the comparison baseline must be locked first. Inspection compares an artifact against a set of data. If that data is not the released final revision, then no amount of inspection rigour changes the fact that it was compared against an outdated reference.

4. What this gate does not catch

The boundary matters just as much. Inner-layer inspection observes geometry; it does not predict assembly behaviour and does not substitute for layer-to-layer control:

  • It sees single-layer geometry, not inter-layer relationships. Every layer can pass individually and the relative position between layers can still shift once stacked. That belongs to lamination and registration control; see accumulation and control of multilayer registration deviation.
  • It sees the present form, not changes introduced later. Material flow and the action of heat and pressure during lamination all occur after inspection.
  • It reads observable surface features. Internal material states that are not pattern geometry are not among its observation targets.

Writing this down has practical value. It explains why "inner-layer inspection passed" is not the same guarantee as "this board's multilayer structure is sound." The converse also holds: not every problem found after lamination should be attributed to an inner-layer escape.

5. The nature of the release decision: who carries the irreversibility

Because a released defect enters the finished product, inner-layer release is a decision that transfers risk onto the product. That settles several questions of ownership.

First, disposition of suspect material cannot be deferred to the next step. The next step is lamination, after which the information is gone. "Laminate it and decide later" is not a valid option here, because no mechanism for deciding later exists.

Second, the record must be sufficient to support traceability. If a lot turns out to have a problem, the only route back to its pre-lamination state is the record made at the time. A missing record is not merely a documentation gap — it means the evidence of that lot's inner-layer condition is permanently gone.

Third, some judgements require the designer to answer, not the fabricator to decide alone. When a deviation raises the question "is the margin on this trace still sufficient" or "is this feature shift acceptable in your application," the fabricator holds the geometric facts but not the design intent. Such questions are only meaningful while the panel is still open. For how the quality system organises these releases, see quality assurance and inspection.

6. What designers must declare before lamination

Irreversibility imposes requirements upstream. If the following are not stated before lamination, the fabricator can only infer from common practice — and the consequence of that inference is laminated in permanently:

  • Which data revision is the release baseline. Name the final revision so the comparison reference is unambiguous. For stackup and drawing release, see multilayer stackup drawing release.
  • Which inner-layer features are functionally critical. Not all traces on a layer are equally sensitive to margin loss. Marking the critical regions tells the fabricator where nothing may be relaxed when judging borderline material.
  • Layer order and layer correspondence. Which layer is which, without guessing.
  • The disposition path and contact for suspect material. Who answers, and within what time, when a question arises before lamination. Without this, schedule pressure pushes the floor toward a default choice.
  • Known boundaries tied to stackup and layer count. For how layer count and the press window constrain each other, see multilayer press process and layer-count limits.

None of this asks the designer to understand process detail. It asks only that what matters and who decides be made explicit. For overall multilayer manufacturing boundaries see multilayer PCB manufacturing; for process capability scope see manufacturing capabilities.

7. Putting the step back into the process chain

Read together, the position of inner-layer inspection in the chain looks like this:

  1. Inner-layer patterns are complete. At this point inner layers are visible, measurable and dispositionable.
  2. Inner-layer inspection performs comparison and judgement. Suspect material must reach a conclusion here.
  3. Lay-up and lamination execute — this is the end of visibility.
  4. Subsequent processes act on outer layers and reachable surfaces. Their observation targets do not include inner layers.
  5. Electrical test verifies connectivity; visual and dimensional inspection verify reachable features.

Step 3 is the only point on this chain where information disappears in one direction. Engineering has a single standard way of handling such a point: move the judgement forward to while the information still exists, rather than searching for substitute evidence after it is gone.

8. Summary

Inner-layer inspection is special not because it is stricter than other inspections, but because nothing follows it. Lamination seals the inner layers into the board permanently, which makes this step's release conclusion the delivery conclusion. Once that is understood, its requirements stop looking like procedural rules and become the direct consequence of process order: lock the baseline, conclude suspect material here, keep records that support traceability, and ask the designer before lamination about anything only the designer can decide.

Related technical articles

FAQ

Q: If inner-layer AOI passed, does that mean the inner layers of the multilayer board are sound?

A: It means the inner-layer pattern matched its comparison data at inspection and its geometry was judged acceptable. Relative position shift between layers after stacking belongs to lamination and registration control, which is outside the guarantee of single-layer pattern inspection. The two are different guarantees and do not substitute for each other.

Q: Can electrical test cover a defect that escaped inner-layer inspection?

A: It depends on the defect type. Electrical test evaluates connectivity, so a defect that already causes an open or a short will be found. A trace that has narrowed but still conducts will pass under its criteria. The consumed margin is not within what the test observes.

Q: Can inner layers be re-inspected after lamination?

A: Not non-destructively. Confirming the actual inner-layer state requires destructive cross-sectioning, and a sectioned panel cannot be delivered as product. This is precisely why the pre-lamination record is the only traceable evidence.

Q: We supplied design data but did not mark critical regions. Does that matter?

A: Yes. The fabricator can judge geometric deviation but does not know which trace carries the tightest margin in your application. Without marking, borderline material can only be judged uniformly per common practice, and the release decision loses the support of design intent.

Q: When suspect material is found, why not laminate first and decide afterwards?

A: Because after lamination the information needed to decide no longer exists. Before lamination, suspect material is something that can be discussed, sampled and dispositioned; after lamination it is an invisible, unchangeable fact. Process order does not permit deferring that decision.

Q: Is this gate held to the same requirements as outer-layer pattern inspection?

A: The methods overlap; the nature does not. After outer-layer inspection the panel can still be observed and dispositioned, so a wrong call has room for correction. After inner-layer inspection comes lamination, with no room for correction. The release threshold must therefore be set for irreversibility, not for "there is another stage after this."

Want this stackup and HDI process checked?

Layer count, blind/buried vias and impedance control drive both yield and cost. Send us the stackup and we will return DFM feedback with optimisation options.