Sunking PCB

IPC-A-600 Class 2 vs Class 3: What Changing the Class Actually Changes

Design Guide2026-10-14Sunking Circuits Technical Team
IPC-A-600 Class 2 vs Class 3: What Changing the Class Actually Changes

That one line on the drawing — "inspect to IPC-A-600 Class 3" — is the easiest line in the whole technical package to write, and the easiest one to render meaningless.

It is easy to write because you only change a 2 into a 3. It is easy to render meaningless because the person writing it usually believes they have turned a dial one notch tighter, while the shop floor receives something else entirely: a class is not a dial you can turn up in isolation. It is the name of a list of acceptance criteria. Change the name and you change the list — and the most consequential kind of change in that list is not "the same characteristic, judged to a tighter limit." It is that certain characteristics are not on the list at one class at all, and only start being inspected, judged, and recorded at the other.

Most of the quality material already on this site answers how to inspect and how to dispose of what you find: PCB quality inspection methods covers the techniques, outer layer pattern inspection checklist covers what to look for on outer layers, and defect classification and the non-reworkable points covers where a defect lands and which process steps are one-way doors. This article answers the upstream question none of them touch: what the class field itself switches.

1. "A strictness scale" is the wrong model. The class switches the object of judgement

Picture Class 2 and Class 3 as two marks on one ruler, and two false conclusions follow.

The first is that "building to Class 3 just means tightening every tolerance; the process stays put." But items appearing on or disappearing from the criteria list, and limits being tightened, affect production on completely different scales. A tighter limit means the same process route has to deliver a narrower distribution — tune parameters, add control points, raise process capability. A characteristic moving from "not judged" to "judged" means there was no such inspection, no such record, and quite possibly no such test specimen. What you have to add is the step itself, not a parameter. The second kind does not fit into an existing flow. It changes the flow.

The second false conclusion is that "a Class 2 board built a bit more carefully can ship as Class 3." But if a characteristic is not a criterion under Class 2 acceptance, then a lot that passed Class 2 has no data on that characteristic at all. No data is not a pass and not a fail. It is the absence of a conclusion. Re-judging after the fact to Class 3 means answering a question nobody asked at the time, about finished boards whose test specimens may no longer exist. That is the real cost of a late class change, and it has nothing to do with care.

In one sentence: a class does not answer "how strict." It answers "which characteristics, judged to what, with what evidence retained."

2. The three things a class change actually alters

Lay the differences out and they fall into three groups. Their engineering weight escalates — from "tune a parameter" to "change the flow."

Group one: where the acceptance limit sits. The same characteristic is judged at both classes, but the accept/reject boundary differs. This is the group everyone already understands and the easiest to absorb: it mostly tests process capability and rarely adds steps. It covers dimensional and visual characteristics — the same family of fields as our published manufacturing tolerances. The tolerance table states the manufacturing window we can hold consistently; the class states how the result is judged at acceptance. Both must be present to be complete. Omit one and you get either "we can build it but it cannot pass" or "it passes but we cannot build it."

Group two: whether the criterion exists at all. Some characteristics are not criteria at one class and enter the scope of judgement at the other. This is the focus of this article, because it changes the length of the inspection list rather than a number on it. The consequences:

  • new inspection operations or stations;
  • new test specimens — typically a coupon carried with the panel, because destructive conclusions can only come from a coupon, never from the deliverable board;
  • new record fields, because an inspection you cannot evidence is an inspection you cannot claim.

Group three: sampling and record requirements. The same inspection, but how much is judged, by whom, and what survives afterwards. High-reliability acceptance usually demands more than "the verdict is pass." It demands traceability — which lot, which board, which coupon, measured when. This group does not change the physical process. It changes the data chain, and a gap in the data chain is as fatal at delivery as a defect. Our quality assurance page describes incoming verification as a data-matrix scan performed twice, at incoming and again at lamination — that is a data-chain design decision, not an inspection-technique one.

Group one can be absorbed by adjustment. Groups two and three can only be absorbed by changing the flow. Quoting and scheduling a class change as if it were only group one is the main reason "build to Class 3" fails to land on the floor.

3. Which process steps get rerouted

A longer criteria list does not only affect the inspection bench.

CAM and front-end engineering. The criteria list defines the review scope. If a characteristic will be judged at final acceptance, its manufacturability has to be confirmed during data review — otherwise you are releasing a known risk into production. The class also drives specimen design: whether a coupon is needed, where it sits in the array, and which process steps it must accompany. That can only be settled at CAM. After material is cut, adding a coupon means re-laying out the panel. A coupon footprint also feeds back into outline and depanel layout; for how those are dimensioned, see outline routing and V-cut tolerances.

Lamination and drilling. These are the classic one-way doors. Once past them, the interlayer structure and hole geometry are fixed; a characteristic that fails afterwards is not a rework question. A higher class brings more interlayer and hole characteristics into the criteria list, which is equivalent to moving more "non-reworkable verdicts" to just after these two steps. That in turn demands more conservative upstream control — leave margin in the process window instead of relying on inspection to catch it. See multilayer lamination registration control and drilled hole quality inspection.

Plating. When plating-related criteria change, what changes is the release condition, not the bath parameters. Which items must be measured, measured before release, and what the release record retains — these are flow fields. Our electroplating quality inspection and release covers that gate.

Electrical test and final inspection. A class may change test coverage requirements, and the coverage requirement in turn determines which test method is even feasible. The method trade-off is in flying probe vs fixture electrical test. What matters here is sequence: fix the class and the coverage requirement first, then choose the test method. Do it the other way round and you get a method already committed that cannot reach the required coverage.

The coupon chain. Characteristics needing a destructive conclusion can only be answered by a coupon carried with the panel. The coupon has to be placed in the array at CAM, travel through every relevant step, and be destroyed at the end. Break any link and the conclusion becomes permanently unavailable — boards shipped, question unanswerable. The same mechanism on the impedance side is in impedance test coupon verification.

4. Why "build to Class 3" is not executable, and what to add

The phrase is not executable because it names the list without supplying the parameters needed to use it. To make it executable, the drawing or technical agreement has to add at least four things.

1) Standard and revision. There is more than one acceptance standard. IPC-A-600 addresses acceptability of the finished bare board's visual and structural features; IPC-6012 is the performance specification for rigid boards; on the assembly side it is IPC-A-610 — our registered assembly standard is IPC-A-610 Class 3, which is an independent field from the bare-board IPC-A-600 class and cannot be inferred from it. Writing only "Class 3" without naming the standard asks the factory to choose your criteria list for you.

2) Scope and zoning. A class need not, and usually should not, apply uniformly to the whole board. What genuinely needs the higher class is often local — one connection area, one hole type, one functional block. Imposing the highest class board-wide spends the cost on areas unrelated to reliability and buys no reliability back. Zoning only works if the drawing can state it: zone boundaries, the class per zone, and which governs when they conflict.

3) Evidence requirements. Which conclusions are required, from whose specimen, which records are retained, and which documents ship with the goods. Skip this and groups two and three surface at the last moment before delivery — by which point the specimens are gone.

4) Who may approve a deviation. Borderline cases will occur. Who raises a deviation request, who approves it, and where the approval is recorded — that path has to exist beforehand. Without it, a borderline case has only two exits: shipped quietly, or scrapped wholesale.

With those four in place, "build to Class 3" turns from a wish into a set of executable instructions. Our published inspection standard is IPC-A-600 Class 2 (see quality assurance), and capabilities lists both Class 2 and Class 3; on the heavy-copper side, acceptance can follow IPC-6012 Class 2 or Class 3 as you specify. In other words, the class is a field to settle in writing at order confirmation — not an adjective you can append mid-production.

5. Zoning: why the highest class board-wide is the wrong default

The instinct that "stricter is never wrong" fails on cost and lead time, and often fails on reliability too.

The reason is that the constraints a higher class brings do not land only on the critical area. They land on the whole board. A longer criteria list applies to every characteristic of that type everywhere, including regions with no influence on product life. Two kinds of waste follow. One: the process window gets narrowed for the sake of irrelevant regions, squeezing the adjustment room available where it actually matters. Two: reject exposure is spread across the whole board, shrinking the room for selection and rework.

The more effective move is to answer "where does a failure cause the product to fail" first, apply the higher class there, run the rest at the normal class, and state that clearly on the drawing. This requires a judgement from the design side — and only the design side can make it. What the factory can supply is which process constraints each class implies in each zone, which is exactly why the class question belongs in the RFQ stage rather than at the inspection bench.

6. Four paths after a reject, and the conditions each requires

Even with the class stated properly, rejects happen. A reject is not a scrap, but each of the four exits has conditions, and the conditions live in the order documents.

Path one: rework and re-judge. Applies only to characteristics whose process step can be reworked without disturbing already-formed conductors or interlayer structure. The cost is labour and lead time. Requires that the order permits rework and that the re-judgement standard matches the first.

Path two: selection within the array. When the defect affects only some units on the production panel, remove the failing units and deliver the good ones. Requires the order to explicitly permit in-array selection — a field that has to be written beforehand, because negotiating it after the fact is usually too late.

Path three: deviation release. The characteristic falls outside the criteria but assessment shows no impact on this product's use; an authorised party approves, releases, and records it. Requires item four above to already exist. Without that path, these cases are forced onto path four.

Path four: scrap and rebuild. Structural problems no downstream step can repair — wrong laminate grade, interlayer registration out of limits, lamination delamination. Which is also why incoming verification is worth two separate scans.

The fork between these four is not on the inspection bench. It is at order confirmation: whether rework is allowed, whether in-array selection is allowed, and who may approve a deviation determine how many exits you have when something is rejected. For where a given defect lands and the process timeline of non-reworkable points, see defect classification and non-reworkable points.

Summary

A class is not a strictness scale. It is the name of a criteria list, and the differences between lists come in three layers: where the limit sits, whether the criterion exists, and sampling and record requirements. The first layer is absorbed by adjustment; the second and third can only be absorbed by changing the flow — new inspections, new specimens, new records. So for "build to Class 3" to be executable, the drawing must simultaneously state the standard and revision, the scope and zoning, the evidence requirements, and who handles exceptions. Putting those four into the RFQ and order confirmation is far more useful than arguing at the inspection bench over whether one feature counts as a defect.

FAQ

Q1: What is the fundamental difference between IPC-A-600 Class 2 and Class 3?

Not "the same metric, held tighter," but a different object of judgement. The differences come in three layers: where the acceptance limit sits, whether the criterion exists at all, and sampling and record requirements. The first layer is the same characteristic against a different boundary and can be absorbed by process capability. The second is characteristics that are not criteria at one class and enter the scope of judgement at the other, which adds inspections, specimens and records. The third changes the data chain. Reading a class change as only the first layer is what leads to underestimating its effect on the flow.

Q2: If the drawing says only "build to Class 3," can the factory execute that?

Not completely — four pieces of required information are missing: which standard and revision applies (IPC-A-600, IPC-6012, or IPC-A-610 on the assembly side), whether it applies board-wide or by zone, which evidence and records must be retained and shipped, and who may approve a deviation. Omitting the first asks the factory to choose your criteria list. Omitting the third is the most damaging, because characteristics needing a destructive conclusion can only be answered by a coupon, and the coupon has to be placed in the array back at CAM.

Q3: Can a lot that passed Class 2 acceptance be re-judged to Class 3 afterwards?

Usually not to any valid conclusion. If a characteristic is not a criterion under Class 2, the lot has no data on it — not a pass, not a fail, simply no conclusion. Items requiring destructive verification are worse: if the specimen was never designed in and never travelled with the panel, it no longer exists. This is precisely why the class has to be settled in writing at order confirmation.

Q4: Isn't applying the highest class board-wide the safest option?

Usually not. The constraints land on the entire board, including regions with no influence on product life. The result is a process window narrowed for irrelevant regions, reject exposure spread board-wide, and less room for selection and rework — without buying real reliability. The more effective approach is to determine where a failure causes the product to fail, apply the higher class there, run the rest at the normal class, and state zone boundaries, the class per zone, and precedence on conflict in the drawing. Only the design side can make that call.

Q5: Is the IPC-A-600 class the same as the IPC-A-610 class?

They are two independent fields and neither can be inferred from the other. IPC-A-600 addresses acceptability of the bare board's visual and structural features; IPC-A-610 addresses assembly acceptability (solder joints and components). Our registered assembly standard is IPC-A-610 Class 3; on the bare-board side our published inspection standard is IPC-A-600 Class 2, with capabilities listing both Class 2 and Class 3. A delivered assembly needs both fields stated separately.

Q6: Does a reject always mean scrap?

No — there are four paths: rework and re-judge (only for characteristics that can be reworked without disturbing formed conductors or interlayer structure), selection within the array (when only some units on the panel are affected), deviation release (assessed and approved by an authorised party, and recorded), and scrap and rebuild (structural problems such as wrong laminate grade, interlayer registration out of limits, or lamination delamination). The conditions for the first three all live in the order documents: whether rework is allowed, whether in-array selection is allowed, and who may approve a deviation. Without those three fields confirmed, only the fourth exit remains.

Need Professional PCB Manufacturing?

Founded in 2009, we provide one-stop PCB solutions from prototype to mass production.