Contract Manufacturing PCB: Who Owns the Defect When Fabrication and Assembly Are Split

The four lines you are actually splitting

When people say "responsibility boundary" they usually mean one line. There are four, and each can sit in a different place.

Line 1 — Procurement. Who buys the components, who owns them in transit, who carries the cost when a part goes obsolete mid-build, who absorbs a price move between quotation and kitting.

Line 2 — Incoming inspection of the bare board. Someone receives fabricated panels and decides they are fit to populate. Whoever accepts them has, in practice, absorbed a share of any fabrication defect that inspection could reasonably have caught.

Line 3 — Defect attribution. When a populated board fails, someone must decide whether the cause is fabrication, assembly, or component. This is the line buyers most often leave blank, and it is the expensive one.

Line 4 — Rework, replacement and scrap liability. Once attribution lands, who performs the rework, who pays for it, and what happens to boards that cannot be reworked.

A contract can place these in several valid combinations. What it cannot do is leave them implicit. An unassigned line does not stay neutral — it defaults to the buyer, because the buyer is the only party with a commercial reason to keep the project moving.

Three structures, and what each one really costs you

Turnkey with one supplier

One party fabricates, buys parts, assembles and delivers tested boards. Attribution becomes internal to them, and that is the main thing you are buying: you raise one ticket against one contract.

The cost is visibility. Internal attribution means you may never learn which stage failed, only that it was fixed. For a product entering volume that opacity has a price, because you cannot target process improvement at a stage you cannot see.

Split fabrication and assembly, buyer as integrator

You contract the fab, you contract the assembler, and you sit between them. You gain supplier choice at each stage and full visibility into both. You also inherit line 2 and line 3 by default, plus the arbitration seat when the two disagree.

That is real work, not a formality. It needs someone on your side who can read a cross-section report and a solder joint image and form an independent opinion. If nobody on your team can do that, you have chosen a structure you cannot operate.

Split, with the assembler taking the bare board

The assembler procures the bare board — from its own fab or from one you nominate — and delivers populated boards. Attribution is again internal, and you keep some influence over who fabricates.

One detail decides whether this works. If you nominate the fab, the assembler will usually decline liability for fab-attributable defects, and reasonably so: it did not select the supplier. A nominated source combined with full assembler liability is not something you get for free, and a contract silent on the point will be read the way that favours whoever drafted it.

Line 1: procurement, and the three words that redistribute risk

Consignment, partial turnkey, full turnkey. These are not billing labels. They move risk.

Full turnkey — the assembler buys everything. One invoice, one party accountable for availability. You pay a handling margin on parts and you lose direct sight of which distributor supplied what, unless you ask for that in the deliverables.

Partial turnkey — you supply the hard, long-lead or controlled items and the assembler buys commodities. Common and sensible. Also the structure that generates the most disputes, because a failure on a buyer-supplied part sits in a grey zone: the assembler did not choose it, but did receive, store and place it.

Consignment — you supply everything. Lowest part cost, highest coordination load. Shortages are yours, including the ones discovered at kitting.

Whichever you choose, three sub-questions decide whether it holds up under stress.

• Attrition. Extra parts for setup, first-article placements and loss. Who supplies them, at whose cost, and who owns the unused remainder. Reel remnants have value and an owner; if the contract names no owner, the party holding them keeps them.
• Buyer-supplied part defects. If a part you supplied is defective out of the bag, who pays for the placement labour already spent and the removal labour still to come. Naming an inspection level for buyer-supplied parts on arrival is what makes that answerable rather than arguable.
• Substitutions. Whether the assembler may substitute without written approval, and what happens to boards already built on an unapproved substitution. The mechanics of running that approval are covered in turnkey PCBA BOM risk and alternate part approval. Here only one thing matters: the contract must name who may decide.

Moisture-sensitive devices deserve a line of their own. Floor life and bake requirements come from the manufacturer's classification, and mishandling them produces failures that appear at reflow and look exactly like a process problem. If you supply MSD parts, the contract should say who tracks exposure and who owns the consequence of a broken dry pack.

Line 2: the handover nobody schedules

Bare boards arrive. Someone unpacks them, and they go to the line. That unremarkable moment is a liability transfer, and in a split supply chain it is usually undocumented.

The question is not whether inspection happens. It is what was inspected, against what, and what acceptance means afterwards. Four points settle it:

1. Who inspects. The assembler on receipt, you before forwarding, or a third party at the fab. Each is workable; the unworkable case is everyone assuming someone else did it.
2. Against which document. A fabrication drawing, a specification sheet, an agreed acceptance plan — and which one governs when they conflict. If you have not set that order of precedence, do it before boards ship, not after a rejection. The sampling and criteria side of this is laid out in the incoming inspection acceptance plan for PCB delivery.
3. What acceptance does and does not cover. Visual and dimensional acceptance cannot clear internal-layer registration, plating continuity through the hole barrel, or laminate integrity. Those are found later, often at assembly or in the field. A contract should state that acceptance is limited to the declared inspection scope and does not waive claims for latent fabrication defects.
4. How long the claim window stays open. A latent defect found in week six is a different commercial conversation from one found on day two. Put a number on it, in writing, and make it the same number in both contracts.

Missing point 3 is the common and costly error. An assembler who accepted panels without qualification has a fair argument that you agreed the boards were good — and you will be funding the rework while the two suppliers correspond politely with each other.

Line 3: attribution, where split supply chains actually break

A board fails. Three candidate causes, two contracts, no named referee. This is the state you must design out before you need it.

Attribution needs four things agreed in advance.

A named decider. Somebody must be able to close the question. Options: your engineer, a mutually accepted third-party lab, or an escalation clause naming one party's judgement as final for a defined class of defect. Any of these beats silence.

A method, not an opinion. Cross-section, X-ray, dye-and-pry, electrical isolation — the technique depends on the defect. What matters contractually is that the method is agreed and its cost is allocated before the analysis starts, because analysis cost is itself a dispute.

A published criterion. IPC-A-610 describes acceptability for electronic assemblies and IPC J-STD-001 describes the requirements for soldered assemblies; both use class levels, so "meets IPC" without a class is not a criterion. For fabricated boards the equivalent conversation is about which specification the drawing invokes. Name the document and the class in both contracts, and make them consistent — an assembler working to one class while your drawing implies another is a dispute you built yourself.

A default when analysis is inconclusive. Some failures cannot be attributed with confidence. Left unaddressed, inconclusive means the buyer pays. If you want a different outcome — shared cost, escalation to a second lab, a threshold below which nobody argues — write it down while everyone is still friendly.

One practical asymmetry is worth planning around. Fabrication defects are often discoverable only after assembly has already added cost to the board. A pad that lifts, a barrel that cracks, a laminate that delaminates under reflow — each destroys a board that now carries the value of its components. Your contract with the fab should address the assembled value of a board scrapped by a fab-attributable defect, not just the bare board price. Most fab terms cap liability at the fabrication value by default. That default is the single largest uncovered exposure in a split structure, and it is negotiable only before you order.

Line 4: rework, and the thing worse than paying for it

Once attribution lands, three questions remain: who performs the rework, who pays, and what happens to boards that cannot be reworked.

Who performs it is not automatic. An assembler may be unwilling to rework a defect attributed to a fab it did not select. A fab cannot rework a populated board at all. The realistic options are the assembler under a separate rework order, a third-party rework house, or scrap. Pick before you need to.

Rework itself has limits. IPC-7711/7721 covers rework, modification and repair procedures, and the relevant constraint is that thermal cycles on a joint and a pad are finite. Some components cannot be reliably reworked. Some cannot be reworked at all without re-qualifying the board. A contract that promises unlimited rework at supplier cost is promising something the physics will not support — better to define a rework limit and what happens when it is reached.

And the part buyers forget: a reworked board may need to be disclosed. If you ship to customers with traceability requirements, rework is a record, not just a repair. Decide who keeps that record and whether reworked units are identifiable in the shipment. Retrofitting that answer after delivery is unpleasant.

What to fix before the purchase order

None of this needs a bespoke legal document. It needs a scope section that both suppliers see and neither can reinterpret later.

• Name the supply model in words — full turnkey, partial turnkey or consignment — and list which BOM lines each party buys.
• Name the attrition owner and the owner of unused material and reel remnants.
• State the bare board inspection scope, the governing document, the order of precedence when documents conflict, and that acceptance does not waive latent defect claims.
• State the latent defect claim window as a duration, identically in both contracts.
• Name the attribution decider, the analysis method and who funds the analysis, including the inconclusive case.
• Name the acceptance standard and its class level, consistently across the fabrication drawing and the assembly contract.
• Address assembled-value liability for boards scrapped by a fab-attributable defect, rather than accepting the default bare board cap.
• Define the rework path, its limit and the record, including whether reworked units are identifiable.

Eight lines. They fit on one page, they are all answerable before you order, and each one you skip becomes a negotiation conducted at the worst possible moment — with a failed build on the bench and a ship date behind you.

If you are still deciding how to divide the work, it helps to have the fabrication and assembly conversations against the same document set. You can review scope on the PCB assembly page, check what a fabrication package needs on the multi-layer PCB page, and send both sides of the job through a single quote request so the boundary is on the record from the first reply.

Related reading on PCB168

• PCBA quote comparison: aligning scope boundaries and hidden charges — normalising assembly quotes once the boundary is set.
• Turnkey PCBA BOM risk and alternate part approval — the approval mechanics behind line 1.
• Defining inspection and test requirements before you get quoted — turning acceptance criteria into quotable lines.
• Incoming inspection acceptance plan for PCB delivery — sampling and rejection criteria for line 2.

FAQ

Is one turnkey supplier always safer than splitting?

Not always — it is simpler, which is different. Turnkey concentrates attribution in one party, so you raise one claim instead of refereeing two. What you give up is visibility into which stage failed and the ability to shop each stage. Splitting is the better choice when you have someone able to arbitrate technical disputes and a reason to control each supplier. It is the worse choice when nobody on your side can read a failure analysis report.

If I nominate the bare board supplier, does the assembler still own fabrication defects?

Usually not, unless the contract says so explicitly. Most assemblers accept liability for suppliers they selected and qualified, and decline it for sources imposed on them. If you need a nominated fab and full assembler liability, that is a negotiated term with a cost, and it has to be written before the order.

We accepted the bare boards on a visual check. Are we now liable for everything?

It depends on what your contract says acceptance covers. That is exactly why the inspection scope should be stated and paired with a clause saying acceptance does not waive latent defect claims. Visual and dimensional checks cannot detect internal registration, barrel plating or laminate problems, so acceptance limited to a declared scope is defensible. Unqualified acceptance is much harder to argue against later.

Who pays for failure analysis when nobody admits fault?

Whoever your contract says — and if it says nothing, you do, because you are the party that needs the answer. Allocate analysis cost up front, with a rule for the inconclusive result. A common structure is that the party found responsible funds the analysis, with cost shared if the result is inconclusive.

Does "meets IPC" settle the criterion question?

No. The relevant standards are class-based, so a reference without a class level leaves the threshold undefined. Name both the document and the class, and make sure the fabrication drawing and the assembly contract invoke the same one.

Can I change the split after production starts?

You can, but treat it as a new qualification rather than an administrative update. A new assembler inherits none of the process knowledge built on the previous line, and the first builds should be treated as first builds. If you are moving from prototype volumes to production, plan the transfer as a defined step with its own acceptance criteria.