Heavy Copper PCB Failure Attribution: Evidence Retention Terms to Fix Before the PO | PCB168
The argument nobody can win
A heavy copper board ships, passes incoming inspection, goes through assembly, gets potted or conformal coated, and goes into a cabinet that draws real current. Eight weeks later a unit comes back from the field with a burned trace, a lifted pad, a cracked barrel, or an intermittent that only shows up warm.
At that point the buyer's engineering team looks at the board and sees a material or construction problem. The supplier looks at the same board and sees thermal abuse: a reflow profile that was never agreed, a bolted terminal torqued beyond what the pad could take, a current loading nobody declared at quote time, an enclosure with no airflow.
Both positions are reasonable. Neither can be proven. That is the actual problem, and it is not a technical problem — it is an evidentiary one. The board's as-delivered condition stopped existing the moment it entered assembly, and nobody wrote down what it looked like before that happened.
What follows is not a failure analysis guide. It is the set of clauses that decide, before anything fails, whether an attribution argument will have evidence to work with or will come down to who has more leverage on the next order.
Why attribution collapses: the as-delivered state is gone
Attribution on a heavy copper board is harder than on a standard board for a reason that has nothing to do with the board being worse. It is that heavy copper failures concentrate in exactly the attributes that assembly and service most aggressively modify.
Thick copper layers behave as heat sinks. They change how the board absorbs and releases thermal energy during reflow, how long solder stays molten locally, how much stress accumulates at the interface between copper and resin, and how the board sits after it cools. Then in service, high current loads those same structures continuously. Every one of those processes leaves marks on the features that attribution depends on.
So when a failed heavy copper board arrives on a desk, the features you would want to examine have been written over at least three times: once by reflow, once by any rework, once by hours or months of thermal cycling under load. Distinguishing an incoming defect from an accumulated one requires a reference point for the as-delivered condition. If no such reference was created and preserved, there is nothing to compare against and the question is formally unanswerable.
This is why evidence retention is a purchasing clause rather than a quality department procedure. The reference point has to be created at a moment when both parties still agree on what the board is — and that moment is before shipment, not after failure.
What gets destroyed, and at which step
It helps to be specific about where evidence disappears, because most of it is destroyed by people acting entirely reasonably.
At assembly. The as-received surface finish condition, the original solder mask edges around large copper features, the original flatness, and any pre-existing surface marks are all covered or altered. Nothing malicious happens here; this is just what assembly does. If no one photographed or sampled the lot as received, the as-delivered baseline is now permanently inferred rather than observed.
At first rework. This is the big one. The instinctive response to a bad joint on a heavy copper board is to reheat and fix it. That single action destroys the joint geometry, the local thermal history, and often the pad interface that would have shown whether the pad was compromised on arrival or pulled during assembly. A reworked failure site is usually unusable for attribution.
At field triage. A technician swaps the board, sometimes cuts a wire, sometimes desolders a terminal to get the unit running again. The failed board goes into a box without a label, a serial reference, or a record of what the load condition was when it quit. The physical artifact survives; the context that made it interpretable does not.
At shipping the sample back. A board returned loose in a bag, without the mating hardware, without the serial or lot linkage, and without the operating record is a board that can only support an opinion. The most common reason attribution fails is not that the board was examined badly. It is that the sample arrived incomplete.
At the record horizon. Both sides keep process and test records for some period, then dispose of them. If a failure surfaces after the supplier's records for that lot are gone, and after the buyer's assembly logs for that build are gone, the paper trail terminates on both ends simultaneously. Nobody chose this outcome; a default retention period chose it.
Evidence custody: who holds what
The first clause family answers a question that is almost never asked at quote time: for this part number, who is responsible for holding which evidence, and for how long?
The structure worth writing is a simple division into three custodies.
Supplier-side custody. The records that only the supplier can produce: the material and copper foil identification for the lot as actually built, the lot traceability linkage from shipped serial or lot number back to the production records, the electrical test record for the lot, the inspection records against the agreed specification, and any first article documentation referenced by the order. The clause should name which of these are held, that they are held per shipped lot rather than per month, and that they remain retrievable by lot identifier for an agreed period. The period itself is a negotiated figure — write the basis for it, which is usually the longer of the product's warranty term and the buyer's own field-return window, and let the PO carry the number.
Buyer-side custody. The records that only the buyer or their assembly house can produce: the as-received condition record for the lot, the assembly process record for the build including the reflow profile actually run, any rework performed and on which positions, the handling and storage conditions between receipt and assembly, and the operating conditions under which the unit was running at failure. Heavy copper attribution depends heavily on this half, and buyers routinely assume the supplier owes evidence while assuming nothing of themselves. A clause that only binds the supplier produces a one-sided record set, which in practice means the supplier can decline attribution on grounds of missing counterpart data.
Joint custody. The small set of artifacts both sides should be able to point to: the agreed specification revision that governed the lot, the agreed acceptance criteria, the reflow profile ownership statement if one exists, and the approved deviation or change records for that part number. These are the documents that define what "conforming" meant for this lot. Where thermal profile ownership has already been settled for the part — in its own clause, written at a different time, for a different purpose — attribution inherits a ready-made piece of evidence. Where it has not been settled, it becomes one more thing to argue about at the worst possible moment.
Two practical notes. Custody should be stated per part number, not per supplier relationship, because retention needs differ between a prototype run and a board going into a ten-year industrial product. And custody should specify retrievability by lot identifier, not merely existence; records that exist but cannot be tied to the shipped lot do not function as evidence.
The disposition order when a board fails
The second clause family is the one that saves the most cases, and it is almost absurdly cheap to write: when a board is suspected of a material or construction defect, what happens first?
The default behavior, absent a clause, is repair first. Production needs the unit running. Somebody reheats the joint. Attribution dies there.
The clause should invert that by naming a sequence, and by naming it in a document the people on the floor will actually see:
- Quarantine before touching. The suspected board is removed from service and set aside physically, not reworked, not cleaned, not desoldered, not cut. The point is to freeze the artifact.
- Record before moving. Identification is captured while the board is still in context: lot or serial linkage, position in the assembly, visible condition, and the operating state at failure. A record created at the bench is worth more than a reconstruction created a week later.
- Notify within the agreed window. The supplier is informed within a period the PO names, with the identification already attached. Late notification is the second most common reason attribution fails, because by the time the supplier hears about it the lot records may be closed and the remaining units already consumed.
- Hold the sibling units. The other boards from the same lot are the control group. If they are allowed to flow into production while the investigation runs, the chance to compare a failed unit against its untouched siblings is lost — and that comparison is frequently the only thing that separates a lot-wide incoming issue from a one-off assembly event.
- Only then decide on repair or scrap. Once the artifact and its context are preserved, repairing the unit costs nothing evidentially. Repairing it first costs everything.
Write this as an explicit order of operations with the words "before" in them. A clause that says both parties "shall cooperate in investigation" produces nothing. A clause that says the board shall not be reworked prior to identification and notification changes what happens at the bench.
One thing to settle alongside the sequence: who bears the cost of the hold. Quarantining sibling units has a real schedule cost, and if the PO is silent on who absorbs it, the hold will quietly not happen.
Minimum completeness of a returned sample
The third clause family exists because incomplete samples are the single largest cause of inconclusive attribution, and completeness is entirely specifiable in advance.
Define, for this part number, what a returned failure sample must include to be accepted for joint evaluation:
- The board itself, unreworked, with the failure site intact. If rework already happened, that fact is disclosed rather than omitted — a disclosed rework narrows what can be concluded; an undisclosed one invalidates the whole exercise.
- Lot and serial linkage sufficient to retrieve the supplier's production and test records for that specific lot. A board with no retrievable identity cannot be matched to any record set on either side.
- Mating hardware and interface parts where the failure involves a terminal, a bolted joint, a busbar connection, or a connector. On heavy copper designs the failure is frequently at an interface, and an interface cannot be evaluated from one side of it.
- The assembly record for that unit, including the thermal profile actually run and any rework history on the board.
- The operating condition at failure as observed: load state, how long it had been running, ambient and enclosure conditions, and whether the failure was sudden or progressive. Qualitative is fine and far better than nothing.
- Packaging appropriate to preserving the evidence, so that transport damage is not added to the list of candidate causes.
Then write the consequence, because a list without one is a wish: a sample that does not meet the minimum completeness is handled as an undetermined case under whatever commercial route the PO specifies, rather than defaulting to either party's account. Both sides need to feel that incompleteness has a cost, or the list will be treated as optional.
It is also worth agreeing on what happens to the sample afterward — whether it is returned, retained by the supplier, or held by a third party — because a sample that disappears into an investigation and never comes back cannot be re-examined if the conclusion is contested.
When both sides agree to go outside
The fourth clause family is the escape valve. Some cases will not resolve bilaterally: both parties have complete records, both interpretations remain consistent with the evidence, and the commercial exposure is large enough that neither can concede.
Settle in advance, when nobody is angry:
- The trigger. What conditions send a case to an independent party — typically a combination of unresolved status after an agreed interval, exposure above a threshold the PO names, or a suspected lot-wide rather than unit-level issue.
- Who qualifies. The characteristics of an acceptable laboratory or adjudicator, and ideally a short pre-agreed list. Picking the lab during a dispute turns lab selection into a second dispute.
- The scope of the question. What the third party is actually asked to determine. A referral that asks "whose fault is this" tends to come back unhelpfully. A referral that asks specific, bounded questions about the physical condition of the artifact comes back usable.
- What they get. Which evidence from both custodies is handed over. Incomplete disclosure to a third party reproduces the original problem at higher cost.
- Who pays, and how that shifts. The initial split and whether it changes based on the outcome.
- The binding status of the result. Whether the finding is binding, advisory, or binding only below a stated exposure. This is a commercial decision, not a technical one, and it should be made deliberately rather than discovered later.
The purpose of this clause family is not to use it. It is that its existence changes bilateral behavior: when both sides know there is a defined path to an outside determination, both tend to preserve evidence more carefully and argue more narrowly.
Writing the four clause families into the PO
Where these live matters, because a term buried in a document nobody reads at the moment of failure is not operative.
The custody and retention terms belong in the quality agreement or supply agreement, since they govern the relationship over time and need to survive individual orders. Reference them from the PO rather than restating them, so revisions do not fragment across orders.
The disposition order needs to reach the people who will touch the board. That means it belongs not only in the agreement but in the buyer's own internal handling instruction for the part, and ideally on the return or nonconformance form itself. A clause that exists only in a contract file will be discovered after the rework, which is to say too late.
The returned-sample completeness list should be attached to the RMA or return request form as a checklist, for the same reason: it has to be visible to whoever packs the box.
The third-party route belongs in the agreement with its thresholds carried on the PO, since exposure scales with order value.
A few things to get right in the drafting itself. Every period, threshold and quantity in these clauses is a figure the two parties must negotiate — write the basis for each figure into the clause and let the commercial documents carry the actual numbers, so the basis survives even when the numbers get renegotiated. Bind both sides symmetrically; a one-sided evidence obligation reliably produces a one-sided refusal to conclude. State the specification revision that governs each lot, because attribution against an unstated revision is attribution against a moving target. And align the retention period with the product's real field exposure rather than a generic default, since a failure that surfaces after both parties' records have lapsed is unattributable no matter how good the clauses were.
What these terms deliberately do not do
Three limits worth stating plainly, so the clauses are not oversold internally.
They do not make failures less likely. Nothing in this document improves a board. Evidence terms change what happens after a failure, not whether one occurs. The work of reducing failures sits in specification, design inputs and process control, which are different conversations.
They do not determine cause. A complete evidence set makes attribution possible; it does not make it automatic. Some cases will remain genuinely undetermined even with every record intact, which is precisely why the undetermined-case commercial route needs to be written rather than left to be invented.
They do not substitute for acceptance criteria. If the specification never said what conforming meant for this part, having perfect records of a lot's production tells you what was built but not whether it was right. Evidence terms sit on top of a specification; they do not replace one.
Used with that framing, these four clause families are among the cheapest terms in a heavy copper order. They cost a negotiation session and some form redesign, and they convert the most expensive category of supplier dispute from a contest of assertions into a comparison of records.
To settle these terms against a specific heavy copper part, the construction and acceptance details belong on the heavy copper PCB page, and the commercial figures — retention periods, notification windows, referral thresholds — are best fixed during quoting via the quote request rather than after the first failure.
FAQ
We already have a warranty clause. Isn't that enough?
A warranty says the supplier is liable for defects attributable to them within a period. It does not say how attribution is performed, who keeps which records, or what happens to a failed unit before anyone examines it. In practice a warranty without evidence terms is a clause that activates exactly when attribution has already become impossible. The warranty defines the remedy; these clauses define whether you can ever reach it.
Why do heavy copper boards need this more than standard boards?
Because of when they fail and what modifies the evidence. Heavy copper failures concentrate around thermal and high-current loading, which means they surface after assembly and often after time in service, not at incoming inspection. Standard-board defects more often show up at receipt, while the board is still in its as-delivered state and attribution is straightforward. On heavy copper the board is typically several process steps away from as-delivered by the time anyone looks at it.
Our assembly is outsourced. Does that change the custody split?
It changes who physically holds buyer-side evidence, not who owes it. The assembly house holds the process record, the actual reflow profile, and the rework history — so the buyer needs those obligations flowed down into the assembly agreement. If they are not, the buyer has contracted away the half of the evidence set they are responsible for producing, and will be unable to meet their own side of the attribution.
What retention period should we specify?
That figure is yours to negotiate; this article does not supply one. The basis most buyers use is the longer of the product's warranty term and the realistic field-return window, since a record that lapses before failures can surface is a record that will not be there when needed. Get the basis agreed first, then the number follows from it and can be defended in renegotiation.
Is photographing incoming lots actually worth the effort?
It is the cheapest single item on the list and the one most often skipped. An as-received condition record does not need to be exhaustive — enough to document the lot's visible state, with the lot identifier legible. Its value only appears later, when it becomes the one piece of evidence showing what the board looked like before assembly touched it.
The failed board was already reworked. Is attribution over?
Not necessarily over, but narrowed. Disclose the rework and what was done. Sibling units from the same lot often remain available and untouched, and the lot-level question — was there a systematic incoming issue — can sometimes still be answered from them even when the specific unit cannot. This is the main reason the disposition order includes holding sibling units.
Should the third-party finding be binding?
That is a commercial judgement about risk appetite, not a technical one. Binding findings resolve disputes but remove the option to contest a determination you consider wrong. Advisory findings preserve that option but may not end the dispute. Many buyers choose binding below a stated exposure and advisory above it. The important part is deciding deliberately at PO stage rather than discovering the ambiguity mid-dispute.
Won't suppliers refuse these terms?
Capable suppliers generally do not, because most of the supplier-side custody items are records they already generate and keep for their own quality system. What suppliers reasonably push back on is asymmetry — obligations on them with none on the buyer — and open-ended retention with no stated basis. A symmetric clause with a justified period is usually an easy conversation. Resistance to retaining lot traceability at all is itself informative.
Related links
Need PCBs for Your Project?
Get an instant quote — prototypes from $2, delivered worldwide.
Get Instant Quote