A board comes off the line in Shenzhen, passes functional test, is packed in an anti-static bag and ships. The design is yours. The firmware is yours. The bill of materials is yours. What is not obviously yours is the bed-of-nails fixture that tested it, the program that ran on the test machine, the programming jig that loaded the firmware, the golden sample the factory compares against, and the record of which component lot went into which serial number.
Those assets are the difference between owning a product and owning a supplier. A company that holds the drawings but not the means of test and programming cannot move production, cannot qualify a second source, and cannot prove what it shipped. It can only ask the factory for help, at the moment when the factory has least incentive to give it.
The legal position is usually worse than the commercial one. Where the contract says nothing, the factory is in possession of the fixture, owns the test program it wrote, holds the programming image it was given, and has no obligation to hand any of it over until the last invoice is paid. Chinese law will not fill that gap in the buyer’s favour by default. The gap has to be closed in drafting, and it has to be closed before the first production run, because the leverage moves to the factory the moment the tooling is installed and the line is running.
What the buyer owns, and what it only appears to own
The asset map below is the one to take into contract negotiation. Each row is a thing that has to exist for the product to be built, and each row has a default custodian, a default owner and a failure mode when the programme moves.
| Asset | Who usually creates it | Who usually holds it | What happens at a transfer | What the contract should require |
|---|---|---|---|---|
| Gerber and fabrication data, stack-up, impedance specification | Buyer, or the buyer’s PCB designer | Factory and PCB house | Easily transferred, but the stack-up may not be documented | Delivery of the full fabrication package, including stack-up and acceptance criteria |
| Stencil and SMT tooling | Factory, from the buyer’s data | Factory | Usually scrapped or reused; rarely returned | Ownership, a stencil register, replacement at cost if lost |
| Pick-and-place program and feeder set-up | Factory | Factory | Rebuilt from scratch at a new site | Delivery in a defined format, or a documented rebuild obligation |
| AOI program and component library | Factory | Factory | Rebuilt, with a new library and new false-call rates | Delivery, or acceptance that ramp yield will be worse |
| SPI program and print parameters | Factory | Factory | Rebuilt | Delivery with the print recipe |
| ICT fixture and ICT program | Factory, or a fixture house engaged by the factory | Factory | The pair cannot be separated; the fixture is physical and stays | Buyer ownership, asset tag, delivery of the program, right to remove |
| Flying probe program | Factory | Factory | Rebuilt | Delivery |
| Functional test rig, harness, test software | Buyer’s engineers, often built at the factory | Factory | The rig is physical; the software may be on a factory PC | Ownership, documentation, delivery of source or a runnable image |
| Programming jig and programming station | Factory | Factory | Stays | Ownership, delivery, calibration records |
| Firmware image and the keys used to load it | Buyer | Factory, once handed over | The factory keeps a copy unless deletion is required and evidenced | Licence terms, deletion on termination, key custody model |
| Test limits and guardband specification | Buyer’s engineering, often negotiated with the factory | Factory | Redefined at the new site, usually more loosely | Written limits, change control, no widening without approval |
| Golden sample or reference board | Factory, sometimes signed by both | Factory | “Lost” during a dispute | Two sealed samples, one held by each party, with a signed record |
| Reflow profile and process window | Factory | Factory | Re-profiled at the new site | Delivery with the process documentation |
| X-ray and AOI images for critical features | Factory | Factory | Not retained beyond the run | Retention for a defined period, delivery on request |
| Component lot and date code records | Factory | Factory | Often incomplete | Per-lot traceability pack, delivered with the shipment |
| Serial number and programming count records | Factory | Factory | The best available proxy for units built | Monthly reconciliation against shipped units |
| Failure analysis and RMA data | Factory | Factory | The new site starts with no failure history | Ownership of the data, monthly delivery in a defined format |
Three observations follow from the table. The first is that almost everything in the right-hand columns is intangible, and intangible assets are the ones companies forget to schedule. The second is that the physical assets, fixtures and jigs, are the ones that cannot be moved quickly, which is why they carry the negotiating leverage. The third is that the test-related assets are a matched set: a fixture without its program is scrap metal, and a program without its fixture is a file.
The default position when the contract is silent
Chinese manufacturing agreements are usually drafted around price, quantity, delivery and quality, with a short intellectual property clause that addresses designs and drawings. The tooling, fixtures, test programs and programming files are dealt with, if at all, in an annex that lists “tooling” as a single line item with a single cost.
That drafting habit creates four problems.
Ownership is unclear, and unclear ownership favours the party in possession. Where the buyer pays for a fixture, the better view is that the fixture belongs to the buyer, because the factory was engaged to make a thing for the buyer with the buyer’s money. Where the fixture is embedded in the unit price and never separately invoiced, there is no document that says so, and the factory can plausibly treat it as its own production equipment. The argument is winnable and expensive, and it is being had while the buyer’s line is stopped.
Possession is a form of security, whether or not a lien exists. A factory holding a fixture, a mould or a set of finished boards has leverage that does not depend on a legal right of retention. In a work contract, the contractor is entitled to retain the completed work until it is paid, and a buyer who wants the fixture released while disputing a batch of boards will find the two issues bundled together in practice even where the law would separate them. The practical consequence is that fixture costs should be invoiced and paid on their own purchase order, with their own acceptance criteria, so that a dispute about assembly charges does not contaminate the fixture.
Title is a weaker instrument than control. A retention of title clause is effective between the parties, and Chinese law recognises retention of title in sale contracts, but an unregistered arrangement may not bind a third party who acquires the asset in good faith. Title clauses are worth having. They are not a substitute for physical marking, a register, and a contractual right to enter and remove.
Silence covers the software, and software is where the value sits. Nobody thinks of the AOI program, the ICT program or the programming script as an asset until a second factory quotes six to ten weeks of engineering to reproduce them. At that point the buyer discovers that the factory regards them as its own process know-how, which, absent an express term, is a defensible position.
Fixtures, programs and golden samples
A test fixture is a custom mechanical assembly with spring-loaded probes, a vacuum or clamping mechanism and an interface to a specific tester. It is built for one board, at one factory, against one test program. It is not portable in any meaningful sense: moving it to another site requires the same tester model, the same interface, a re-validation and, in most cases, mechanical modification.
The economics are therefore asymmetric. For the buyer, the fixture is a capital asset that should last the product’s life. For the factory, it is production equipment that occupies a shelf. When the relationship ends, the factory has no reason to protect it and every reason to keep it.
Four mechanisms make the position enforceable.
A fixture register. List every fixture, jig, stencil and carrier by serial number, with a photograph, the board it serves, the tester it interfaces with, the date of manufacture, the replacement cost and the physical location. Update it at each engineering change. A register that is maintained is evidence; a register that is created during a dispute is a wish.
Physical identification. Every fixture should carry a durable plate or label reading “Property of [buyer], not for use on other customers’ product”. The plate is not a legal instrument, but it changes the conversation on the shop floor, and it defeats the argument that the factory did not know the fixture was not its own.
A bailment clause. The contract should state that the fixture is the buyer’s property, that it is held at the factory at the factory’s risk, that the factory insures it or bears the cost of replacement, that it may be used only for the buyer’s product, and that it must be maintained and calibrated on a defined schedule. A factory that treats the fixture as a borrowed asset behaves differently from one that treats it as equipment.
A release right. The buyer needs a right, on notice, to enter the premises and remove the fixtures, and an obligation on the factory to pack them to a defined standard and make them available. The clause should survive termination, should not be conditional on payment of disputed sums, and should be backed by a power of attorney so that a third party can collect if the factory does not cooperate.
The golden sample deserves separate treatment because it is the cheapest and most neglected control. A golden sample is a board, or a set of boards, that has been verified against the specification and sealed as the reference. It resolves arguments about what “good” means, it anchors AOI tuning, and it is the first thing that disappears when a relationship sours. Keep two: one at the factory, sealed and signed by both parties with a date and a serial number; one with the buyer, stored under the same conditions. Photograph both, and record the firmware version and the test result on the label.
Firmware, programming files and keys
The firmware is the product’s brain, and in a turnkey arrangement the factory holds a copy of it, a method for loading it, and often the keys used to authenticate it. That combination is enough to build a working unit without the buyer’s involvement, which is the definition of an over-production risk.
Three assets need to be separated in the contract, because they carry different risks.
The image. The compiled firmware that is loaded onto the device. If the factory holds the image, it can load it onto boards that were never ordered. If the image is unprotected, it can also be extracted from a shipped device and reused.
The programming procedure. The jig, the station, the sequence and the parameters. Losing this means the new factory has to re-engineer the programming step, which is usually a few weeks and a re-qualification.
The keys. The cryptographic material that authenticates the firmware, unlocks the device or derives a per-device identity. Keys are the asset that companies most often hand over without noticing, usually because the factory asked for them to make the line work.
The control model that works at small and medium volume does not require a hardware security module, although it benefits from one. Build the image in the buyer’s own environment, sign it there, and give the factory only the signed production image. Derive per-device keys from a master held by the buyer, so that the factory can inject an identity without learning the master. If the device supports secure boot, enable it, and keep the signing key off any machine the factory can reach. Where the silicon does not support any of this, the fallback is procedural: an encrypted image unlocked per batch, a programming station that logs every unit, and a reconciliation between units programmed and units shipped.
Reconciliation is the control that catches over-production. The programming log records a count; the shipping documents record a count; the scrap record accounts for the difference. A gap is either a process error or unauthorised production, and either way it needs an answer. Ask for the reconciliation monthly, in a fixed format, and treat an unexplained gap as a quality incident rather than an administrative one.
The contract should describe the firmware as licensed rather than sold, prohibit reverse engineering and extraction, prohibit programming of any unit not covered by a purchase order, require deletion of all copies on termination with a written certificate, and give the buyer an audit right over the programming logs. Where the factory needs the source to support production, that should be an escrow arrangement with defined release conditions, not a delivery.
From bare board to tested assembly
The manufacturing content matters because the ownership questions only arise in relation to specific process steps. A short account of the process is enough to locate them.
A PCB is the bare board: fibreglass or another substrate with copper traces, pads and vias, and no components. It may be single-sided, double-sided, multilayer at four, six or eight layers and beyond, high-density interconnect with finer pitch and blind or buried vias, flexible, or rigid-flex. Fabrication runs from design and CAM preparation through imaging, etching, drilling, plating, solder mask, legend printing, surface finish and electrical test.
A PCBA is the assembled board: the PCB plus components, solder joints, inspection and test, and optionally conformal coating, programming and functional test. Assembly is surface mount, through-hole, or the mixed technology that most products use.
| SMT step | What happens | The control that matters |
|---|---|---|
| Solder paste printing | Paste is forced through a stencil onto the pads | Stencil thickness, paste viscosity, print pressure and alignment |
| Solder paste inspection | An optical system measures the deposits | Catches insufficient, excessive or misaligned paste before placement |
| Pick and place | Machines place components from reels and trays at accuracies of roughly 25 to 50 microns | Feeder set-up and component verification, which is where wrong-part errors originate |
| Reflow | The board passes through a multi-zone oven and the paste melts | The profile, because too hot damages parts and too cold leaves joints unreliable |
| Automated optical inspection | Cameras check placement and visible joints | Program coverage and library accuracy, not merely the presence of a machine |
| Through-hole assembly | Wave, selective or hand soldering for leaded parts | Hand-soldering skill, which varies more than any machine parameter |
| Test | Electrical and functional verification | Test coverage and limit discipline |
| Post-processing | Cleaning, coating, programming, packing | Residue, coating coverage, programming records |
Testing is a ladder, and each rung answers a different question. Solder paste inspection answers whether the print was right. AOI answers whether the placement and the visible joints are right, but cannot see under a package. X-ray answers what AOI cannot, at the cost of speed. In-circuit test answers whether the right components are present with the right values and whether the nets are connected, using a bed-of-nails fixture built for that board. Flying probe answers the same questions without a fixture, more slowly, which suits low volume. Functional test answers whether the board works as a system, and it is the only rung that tests the product rather than the process. Boundary scan tests interconnect through compliant devices without physical probes.
| Test | What it catches | What it misses | When it is worth the cost |
|---|---|---|---|
| Solder paste inspection | Print volume, alignment, bridging | Everything after printing | Any volume where fine-pitch parts are used |
| AOI | Missing, wrong, misaligned parts; visible joint defects | Hidden joints; anything electrical | Always |
| X-ray or automated X-ray | BGA and QFN joints, voids, some counterfeit indicators | Electrical function; cost and cycle time | Boards with area-array packages |
| In-circuit test | Wrong values, shorts, opens, missing parts | Function; parts that fail only under load | Medium and high volume, where fixture cost amortises |
| Flying probe | The same electrical checks without a fixture | Speed | Prototypes and low volume |
| Functional test | Whether the board works | Marginal units that fail later | Always, at least on a sample |
| Boundary scan | Interconnect between compliant devices | Analogue and non-compliant parts | Complex digital boards |
The choice of ladder is a commercial decision, and it should be written into the contract as a test specification rather than left to the factory’s standard flow. A factory will test to the level it is paid for. Where the specification is silent, the default is AOI plus a power-on check, which is adequate for a toy and inadequate for a device with a warranty.
The supply base
China is the largest centre of PCBA production in the world. The industry runs from workshops with a single SMT line to electronics manufacturing services companies with hundreds of lines, and the density is concentrated in two regions: the Pearl River Delta around Shenzhen, Dongguan and Guangzhou, and the Yangtze River Delta around Suzhou, Shanghai and Hangzhou. Both have deep local supply of laminates, solder, stencils, fixtures, components and the engineering labour that supports them.
The commercial consequence is a cost advantage that is real but often overstated. PCBA in China is typically thirty to sixty per cent cheaper than in the United States or Europe, and the gap comes from labour cost, equipment utilisation, component pricing at volume and lower overhead rather than from any single factor. The gap has narrowed as automation has reduced the labour share and as energy and logistics costs have risen. A buyer who assumes a fixed discount will be disappointed; a buyer who benchmarks total landed cost, including freight, duty, inventory and the cost of quality, will get a clearer picture.
Component supply runs through four channels. Franchised distributors, including the large global houses with China operations, sell authorised product with traceability and, usually, technical support. Local franchised distributors serve the same function at smaller scale. Independent distributors and brokers trade in whatever is available, which is where counterfeit risk concentrates. Component markets, of which Huaqiangbei in Shenzhen is the best known, serve prototyping and repair. Commodity parts are also bought direct from the manufacturer at volume.
Most factories buy from a mix of all four, and the mix shifts with the market. During allocation, when lead times stretch from weeks to many months and prices rise, the broker share rises, small customers receive less than large ones, and the counterfeit risk rises with both. The controls that work in a shortage are unglamorous: plan further ahead, qualify a second source before it is needed, place orders early, hold buffer stock on the critical lines, and buy through a distributor with an allocation relationship rather than through whoever answers the phone.
Component sourcing: who buys, who owns, who answers
Three sourcing models are common, and they allocate risk very differently.
| Model | Who buys the components | Who owns them before assembly | Counterfeit exposure sits with | Control lever |
|---|---|---|---|---|
| Turnkey | Factory | Factory, until the finished board is delivered | Factory in contract, buyer in practice | Approved manufacturer list, traceability pack, audit right |
| Consigned | Buyer | Buyer, throughout | Buyer, unless the factory mishandles the material | Incoming inspection at the factory, insurance, scrap accounting |
| Kitting | Buyer supplies a complete kit | Buyer | Buyer | Full control, full administrative burden |
| Buyer-nominated distributor | Distributor appointed by the buyer, ordered by the factory | Depends on the terms | Shared, which is the problem | Three-way agreement between buyer, factory and distributor |
The phrase “buyer in practice” in the turnkey row is the one that matters. In a turnkey arrangement the factory warrants that components are genuine and from authorised sources, and the contract can make it liable for replacement and rework. What the contract cannot do is undo a field failure. When a substituted memory device fails after thermal cycling in a customer’s hands, the buyer absorbs the warranty cost, the freight, the reputational damage and the engineering time, and then pursues the factory for a sum that is almost always smaller than the loss. The warranty allocation is therefore a recovery mechanism, not a risk control. The risk control is the specification, the sourcing restriction, the inspection and the traceability record.
Ownership during the process deserves a line in the contract even in a turnkey deal. Under a Chinese sale contract, property in goods that are not yet identified passes when the goods are ascertained and appropriated to the contract, which in a factory setting means the point at which specific boards are set aside for a specific order. In a consignment, the buyer’s components remain the buyer’s property throughout, but proving which reels on the line are the buyer’s after they have been opened and loaded is impractical without lot records and physical segregation. Consigned material should be segregated, labelled and reconciled, and the contract should say what happens to surplus material, scrap and obsolete stock at the end of the programme.
The specification is the strongest lever, and it is usually the weakest document. A bill of materials that says “10k resistor 0603” or “or equivalent” gives the factory discretion that the buyer did not intend to give. Critical lines should name the manufacturer, the part number, the grade and the package, and should state that substitution requires written approval supported by a sample and a test result. For commodity parts, an approved manufacturer list with two or three named sources is enough. For memory, power management, radio and microcontroller parts, the approved list should be one manufacturer, and any change should be treated as an engineering change with a qualification run.
Counterfeit components and the traceability pack
Counterfeit parts reach Chinese assembly lines through the broker market, and the broker market grows when allocation is tight. The common categories are remarked parts, where a cheaper device is relabelled as an expensive one; pulled parts, recovered from scrapped equipment and cleaned and re-marked; cloned parts, which function but not to specification; and wrong-grade parts, where a commercial-temperature device is sold as industrial. Each fails differently. Remarked and cloned parts fail early or intermittently; pulled parts fail after thermal cycling; wrong-grade parts fail at temperature extremes, which is exactly where a customer will use them.
The failure signature is expensive because it is diffuse. A bad batch of boards produces field returns that pass on the bench, intermittent faults that cannot be reproduced, and a rising warranty rate that nobody can attribute. Companies often replace the factory before they identify the component, and the problem follows the design to the new supplier.
Screening is a matter of matching the test to the risk. Visual inspection under magnification catches re-marking, inconsistent date codes and packaging that does not match the manufacturer’s style. Marking permanence tests catch parts that have been cleaned and re-lasered. X-ray reveals die and lead-frame differences inside a package that looks identical outside. X-ray fluorescence checks plating and lead finish. Decapsulation and die inspection settle the question for high-value parts. Curve tracing and parametric electrical test compare the part against the datasheet. The industry has codified much of this in standards for counterfeit avoidance and for the acceptability of components bought in the open market, and independent test houses apply them for a fee.
The economic logic is risk-based. Testing every line is uneconomic; testing nothing is negligent. Screen where the loss is asymmetric: obsolete and end-of-life parts, long-lead parts bought during allocation, parts bought from anyone other than a franchised distributor, parts whose failure would create a safety issue, and parts whose unit cost is high enough that counterfeiting is profitable. For everything else, the control is the channel rather than the test.
The traceability pack is what makes a finding bounded. For each lot, the factory should deliver the distributor invoice or a certified copy, the packing list, the reel label photographs, the date code and lot code, the country of origin and, where the manufacturer provides one, a certificate of conformity. That pack turns a field failure into a batch, and a batch into a recall of hundreds rather than hundreds of thousands. It also changes behaviour at the factory, because a purchasing team that knows it must produce the invoice buys differently.
Contract terms that decide the outcome
The clauses that matter are not exotic. They are ordinary clauses that are usually omitted because nobody thought about the asset.
The ownership schedule should list every fixture, jig, stencil, carrier, test rig and programming station, with serial numbers, and state that each is the buyer’s property. The bailment clause should govern storage, insurance, use and maintenance. The release clause should give a right to remove on notice, surviving termination and not conditional on disputed sums.
The test clause should attach the test specification, define the limits, require written approval for any change to a limit, and require the factory to retain test data for a defined period and deliver it on request. The golden sample clause should require two sealed samples and a signed record.
The firmware clause should license rather than sell, prohibit programming outside a purchase order, require deletion on termination with a certificate, and grant an audit right over programming logs. The escrow clause should define what is deposited, who may release it and on what conditions.
The sourcing clause should attach the approved manufacturer list, prohibit substitution without written approval, require sourcing from franchised distributors for named parts, and require the traceability pack with each lot.
The counterfeit clause should warrant authenticity, require containment and replacement at the factory’s cost, allocate field failure and recall costs, and grant an audit right over purchasing records. The audit clause should permit entry, inspection of fixtures and records, and interviews with the production team.
The termination clause should require a transition period, delivery of the fabrication package, test programs, programming files and process documentation, and cooperation with a successor manufacturer. Transition assistance is the clause most often omitted and the one that determines whether a move takes six weeks or six months.
Choosing the factory
Factory types map to volume, and a mismatch is expensive in both directions. Prototype and low-volume shops run one or two SMT lines with quick changeover and high per-unit cost. Mid-volume factories with one to five lines serve programmes in the range of a thousand to fifty thousand units a year. High-volume electronics manufacturing services companies offer very low unit cost and very high minimum order quantities. Full-service providers add design, engineering, final assembly and supply chain management. Specialised houses serve medical, automotive and aerospace customers with the corresponding quality systems.
The criteria that predict a good outcome are capability, system, experience and candour. Capability means the equipment can place the smallest part on the board, inspect what it places, and test what it builds: 01005 and 0201 placement, fine-pitch and area-array packages, SPI, AOI, X-ray, in-circuit or flying probe, and functional test. System means certification with a scope that covers the product, typically ISO 9001 as a floor, IPC-A-610 for acceptability, and IPC Class 2 or Class 3 depending on the product, with ISO 13485 for medical, IATF 16949 for automotive and AS9100 for aerospace. Experience means they have built something like the product at something like the volume. Candour means they answer questions about component sourcing, test coverage and failure rates with specifics rather than assurances.
The red flags are consistent across the industry: pricing well below the market, reluctance to disclose component sources or traceability, no functional test capability, visible ESD indiscipline, refusal of audits, slow or vague technical responses, and a certificate whose scope does not cover the work.
Where foreign companies get this wrong
The fixture that was never bought. The fixture cost is bundled into the unit price and described in the quotation as “tooling included”. No invoice, no register, no serial number. Two years later the buyer wants to dual source and discovers there is nothing to move and nothing to prove.
The test program that belongs to the factory. The buyer assumes the program is part of the service. The factory regards it as process know-how, which is a reasonable position where the contract is silent. The second source then quotes a programme of work to rebuild test coverage, and the schedule slips by a quarter.
The keys handed over to make the line work. An engineer sends the signing key so that the factory can provision devices, because otherwise the line stops. Nothing in the contract records the transfer, and nothing requires the key to be rotated afterwards.
“Or equivalent” in the bill of materials. The phrase is inserted to keep sourcing flexible during a shortage. It becomes the factory’s authority to substitute, and the substitution is discovered in the field rather than in the incoming inspection record.
No lot records, so the recall is total. A counterfeit finding arrives from a customer. The factory can identify the month but not the lot, so every unit ever shipped is in scope. The cost of the recall is then set by the absence of a document rather than by the size of the defect.
Cost and commercial impact
The cost structure explains where negotiating effort pays. Components dominate, and assembly labour, which is what buyers usually negotiate, is a small share.
| Element | Share of total cost | What moves it |
|---|---|---|
| Components, the bill of materials | 60 to 85 per cent | Part selection, volume, allocation, brand and grade |
| Bare PCB | 5 to 15 per cent | Layer count, size, surface finish, panel utilisation |
| Assembly labour | 5 to 15 per cent | Component count, mixed technology, line speed, volume |
| Test | 2 to 10 per cent | Coverage, fixture amortisation, test time |
| Tooling and set-up | 1 to 5 per cent | Stencils, fixtures, programming, first-article work |
| Overhead and margin | 5 to 15 per cent | Factory utilisation, order size, payment terms |
Two conclusions follow. The first is that a buyer who spends a week negotiating assembly labour is optimising a tenth of the cost. The second is that the same buyer can usually save more by changing three lines of the bill of materials, provided the change does not require re-certification of the product.
Cost reduction therefore runs through the bill of materials, the design and the process. On the bill of materials, choose parts that are available from more than one source, use standard packages and common values, and reduce component count where the design allows, since every part removed saves placement, inspection and a failure mode. On the design, follow surface-mount rules with standard footprints and adequate spacing, minimise mixed technology because a board with through-hole parts costs more to assemble than an all-surface-mount board, panelise to standard sizes, and include test points so that in-circuit test remains possible. On the process, improve first-pass yield, which reduces rework more than any price negotiation, shorten test time, and increase boards per panel.
The cost of getting the ownership questions wrong sits outside this table, which is why it is easy to miss.
The cost of reproducing the means of production. A new factory needs a new stencil set, a new fixture, a new test program, a new programming jig and a new process qualification. The fixture is a custom mechanical build whose cost scales with probe count and mechanical complexity; the program is engineering weeks; the qualification run consumes material and line time. None of this appears in a unit price comparison, and all of it lands in the first quarter of a transfer.
The cost of the ramp. Yield at a new site is worse than at a mature site for the first several runs, because the AOI library is new, the operators are new and the process window has not been tuned. The scrap and rework during ramp is a real cost, and it is larger where the buyer cannot supply the process documentation.
The cost of an unbounded quality event. A counterfeit batch or a design change made without approval produces field failures whose cost is a multiple of the unit price. Ho Qianggao, a senior lawyer at the firm with more than ten years of electronics manufacturing work (licence 14403201310092337), describes a client whose product began failing in the field at an alarming rate. The factory had substituted cheaper unbranded memory for the specified branded part, saving about $1.50 per board; the chips failed after thermal cycling, and the warranty cost exceeded $20 per unit. The saving was visible on the purchase order. The cost was invisible until the returns arrived.
Against those numbers, the cost of the controls is small. A fixture register takes an afternoon. A test specification is a document the engineering team already has in draft. A traceability requirement adds a filing step per lot. A deletion certificate costs a signature. The comparison is not between a control and no control, but between a control and the loss it prevents.
Practical checklist
- Build an asset schedule that lists every fixture, jig, stencil, carrier, test rig and programming station, with serial numbers and replacement values.
- State in the contract that each item is the buyer’s property, held at the factory’s risk, insured and used only for the buyer’s product.
- Invoice fixture and tooling costs separately, on their own purchase order with their own acceptance criteria.
- Obtain a right to enter and remove the fixtures on notice, surviving termination and not conditional on disputed sums.
- Require delivery of the fabrication package, the test programs, the programming procedure and the process documentation, on a schedule rather than on request.
- Attach a test specification that names the tests, the coverage and the limits, and prohibit limit changes without written approval.
- Keep two sealed golden samples with a signed record of serial number, firmware version and test result.
- Licence rather than sell the firmware, and separate the image, the programming procedure and the keys in the drafting.
- Reconcile units programmed against units shipped every month, and treat an unexplained gap as a quality incident.
- Attach an approved manufacturer list, prohibit substitution without written approval, and require the traceability pack with each lot.
Geason Law Offices advises foreign electronics companies on PCBA contracts, tooling ownership, test and programming arrangements and counterfeit risk allocation. The relevant service lines are Commercial Contracts & Manufacturing Agreements for the drafting, and Manufacturing Legal Support in China for factory qualification and dispute support. Two related guides cover the adjacent ground: PCB Manufacturing in China: A Guide for Foreign Hardware Companies on the bare board, and Quality Control in China Hardware: Inspection, Testing and Defect Prevention on inspection practice. To discuss a specific board, fixture or programming arrangement, contact the firm; the first consultation is free and the form prepares an email in your own mail application.
Frequently Asked Questions↑ Back to article
What is the difference between PCB and PCBA manufacturing?
PCB (Printed Circuit Board) is the bare board — the fiberglass/copper substrate with traces and pads, no components. PCBA (Printed Circuit Board Assembly) is the fully assembled board with all components soldered on. In China, you can get PCB fabrication from a PCB factory, assembly from an EMS/PCBA factory, or turnkey PCBA from a factory that does both (or sources the PCB and assembles it). Many companies offer turnkey PCBA — you send them the design files and BOM, and they deliver fully assembled, tested boards.
What is the typical PCBA manufacturing process in China?
PCBA manufacturing typically follows these steps: (1) PCB fabrication — making the bare boards, (2) Solder paste printing — applying solder paste to PCB pads using a stencil, (3) SMT placement — pick-and-place machines put components on the board, (4) Reflow soldering — the board goes through an oven to melt the solder, (5) AOI (Automated Optical Inspection) — checking for placement and soldering defects, (6) Through-hole assembly — for components that go through holes (wave soldering or hand soldering), (7) Testing — functional test, in-circuit test (ICT), flying probe test, (8) Cleaning and conformal coating (if required), (9) Final inspection and packaging. The exact process depends on the board complexity and components used.
How much does PCBA cost in China compared to Western countries?
PCBA in China is typically 30-60% cheaper than in the US or Europe. The savings come from: lower labor costs, lower equipment operating costs, better component pricing (due to volume and local supply chain), and lower overhead. However, the gap has been narrowing in recent years. The total cost depends on: board complexity, number of components, component types (BOM cost is usually 60-80% of total PCBA cost), assembly type (SMT vs. through-hole), volume, and testing requirements. For simple boards, the assembly cost can be just a few dollars per board; for complex boards with hundreds of components, it can be $50-100+ per board.
What are common quality issues with Chinese PCBA?
Common PCBA quality issues include: (1) Solder defects — cold joints, solder bridges, tombstoning, insufficient solder, (2) Component issues — wrong components, counterfeit components, reversed polarity, missing components, (3) PCB issues — wrong trace width, poor plating, delamination, (4) Placement errors — misaligned components, wrong orientation, (5) Testing failures — functional failures, short circuits, open circuits, (6) ESD damage — components damaged by electrostatic discharge during handling, (7) Cleanliness issues — flux residue, contamination. Good PCBA factories use AOI, SPI, and functional testing to catch these issues.
How do I ensure component authenticity in Chinese PCBA?
Counterfeit components are a real concern with Chinese PCBA. To protect yourself: (1) Specify approved component brands and part numbers clearly in your BOM, (2) Require the factory to source from authorized distributors (not the gray market), (3) Ask for component certificates of authenticity or batch traceability, (4) Consider consigning critical components yourself — buy them from a trusted distributor and ship them to the factory, (5) Do incoming inspection of key components — visual inspection, part number verification, (6) Use X-ray or decapping testing for high-value or high-risk ICs, (7) Include strong provisions in your contract about counterfeit components — with penalties, (8) Work with reputable PCBA factories that have a track record and certifications.
What is the difference between turnkey PCBA and consigned PCBA?
Turnkey PCBA means the factory sources the PCB, all components, and handles full assembly — you provide design files and receive finished boards. Consigned (or partial turnkey) PCBA means you provide some or all components, and the factory supplies the PCB and assembly labor. Turnkey is most convenient but requires trust in the factory's component sourcing; consigned gives you more control over component quality and cost but requires more work on your side. For most production, turnkey is standard, but critical components may be consigned for authenticity assurance.
Which PCBA tests do I need for my product volume?
The required tests depend on product complexity and production volume. For prototypes and very low volume, you need AOI (Automated Optical Inspection) for visual defects, flying probe test for basic electrical verification, and functional testing to confirm the board works. For medium volume production, add SPI (solder paste inspection), ICT or flying probe, and X-ray for BGA/QFN components. For high volume or high reliability products, add SPI + AOI as standard, X-ray for BGAs, ICT for fast thorough testing, functional test, burn-in test, and environmental stress screening. Always specify test requirements in your contract.