| Episodes | 121 |
| Mentions | 158 |
| Cited here | 51 |
| First — last | #1 — #724 |
| Top guests | Jeff Keyzer, Chris Denney, Jonathan Hirschman |
| Related | pcb fabrication · pick and place machine · contract manufacturer · component sourcing · pcb |
PCB assembly (PCBA) is the process of populating a bare printed circuit board with electronic components to produce a functional assembly, distinct from PCB fabrication, which produces the unpopulated board.[445] The process matters economically because failures concentrate at the assembly stage—solder that does not flow, joints that do not make contact, or wrong parts—rather than in the silicon itself.[483] Where a design’s bare boards and components are entirely in stock, an assembly house can turn a build in about twenty-four hours, a turnaround that distinguishes board assembly from semiconductor packaging, where prefabricated dies and substrates still take weeks.[650]
Production process
A production assembly line runs as a conveyor sequence: solder paste is applied by machine through a stencil, components are placed by pick-and-place machines, and the board passes through a reflow oven.[121] In a well-run shop, a human inspection station sits between placement and reflow, whose operator looks at the board without touching it to catch catastrophic errors; the oven is followed by a further quality check and automated optical inspection.[121] Through-hole placement, which follows, remains manual in most shops because it has not been automated, and the line finishes with automated functional test fixtures and flying-probe testing.[121]
Solder paste application
Stencil printing is a single-shot deposition of paste across an entire board or panel, which is what makes it repeatable at volume; the alternative of jetting paste pad by pad rasters across the board and takes far longer per unit.[320] Production stencils are mounted in frames that hold the foil under high tension in a small number of standard sizes, typically 23 by 23 or 29 by 29 inches with tubular or slim frame profiles, so that the stencil lands in the same position on every board and the deposition is consistent.[320] Stencils ordered alongside a board are made from whichever revision the fabricator processed first and are not revised when the design is, so a stencil can arrive matching an earlier layout; the mismatch is workable when only small parts have moved but has to be anticipated.[692]
Manual steps
Reel loading onto the feeders is the step that has stayed manual even in well-automated shops, so the visible difference between a good factory and a poor one is that in the good one people watch machines rather than handle boards.[121] Mechanised through-hole soldering, by contrast, repays its cost even where labour is cheap, on reliability grounds rather than labour cost: a line built in a low-wage location started with people doing through-hole work and moved to machines because the mechanised process was more consistent.[105]
Design for assembly
The assembler, not the standard, decides the footprint. IPC defines least, nominal and most land patterns for each package according to board density, and an assembly house will nominate the variant its equipment is qualified for on the grounds that it produces the lowest defect rate.[70] Pin-one markers placed directly over the pin-one pad, which many CAD libraries supply as the default, conflict with assembly, because the machine needs that pad; the purpose of the whole layout exercise is to get the board built, so assembly constraints govern the footprint.[408]
Panelisation must be agreed with the assembler before the panel is drawn, because a panel that a fabricator will make is not necessarily one the placement machine can handle. Inconsistent copper coverage across a panel also causes plating problems that surface as breaks and shorts, to the point where a fabricator may decline the job rather than hand-correct hundreds of defects.[149] Process differences between suppliers are real and undocumented: what works at one fabricator or assembler may not work at another, and environmental regulation drives some of it, so that in one region a via is plated only if it has copper on both sides, while elsewhere a different resist process plates in a single pass.[149]
Feeder capability constrains what can be sent for assembly at all: an unusually wide part such as a long connector may require a feeder the assembler does not own, so a design cannot simply be handed over on the assumption that any shop can build it.[96] The silkscreen has to carry the information the assembler cannot infer: a part whose polarity is not marked stops the build while the assembler tries to reach the customer, and a late substitution such as a tantalum capacitor for a ceramic one turns an unmarked orientation into a blocked job on hundreds of boards.[411]
Consolidating a product onto one board rather than several is a manufacturing decision as much as an electrical one, because a single board can be tested and calibrated in one fixture, which matters most at batch sizes below tens of thousands.[640] Industrial design can force the opposite decomposition: a consumer robot with sensors and indicators positioned on different planes inside its enclosure was built as twelve separate board assemblies joined by wiring harnesses, because that was the practical way to place each element where the geometry required.[461] Building the complex, shared part of a design once as a module and carrying variants on simple two-layer host boards moves the assembly difficulty into a single reusable item, so that changing a sensor becomes a trivial board change and the module itself can be developed by a separate team with the relevant expertise.[676]
Identification and traceability
A bare board and an assembled board are distinct items with distinct part numbers, generated at design time from a product lifecycle management system and printed on the silkscreen so both appear on the board itself.[445] Ordering from an assembly house needs a three-part identifier because the layers revise independently: the revision of the bare board, the revision of the assembly built on it, and the firmware version programmed at build time.[445] Assemblers commonly label boards at the end of the run rather than as they are built, so tracing which units received an alternate part requires an explicit instruction to label at that point; otherwise the boards are batched and the distinction is lost.[178]
Traceability is built by scanning at each transition: a code on the bare board identifies it, the populated board is tested in an in-circuit fixture and relabelled as an assembly, and the assembly serial numbers are scanned into the finished-unit test report. The record then carries which test fixture and which pogo-pin cassette were used, the timestamps, and whether the board passed first time, so a worn fixture showing rising failures can be identified as the cause.[544] Test rather than assembly dominates the schedule on high-reliability hardware: a satellite programme runs over four hundred unique tests spanning individual printed circuit board assemblies up through vibration and thermal-vacuum on the integrated system, and the test phase is the bulk of a twelve to eighteen month integration period.[679]
Economics and sourcing
Placement is charged per component, at around five to six cents per placement for medium to small volumes, which makes part count rather than board area the cost driver; a design with five hundred parts, as multi-channel analog products tend to have, turns placement into the dominant line item and can justify bringing assembly in-house.[237] At the upper end of the range, a board of roughly a thousand parts across a hundred and thirty distinct line items was assembled offshore for about a hundred dollars a board, at quantities of a few hundred where no meaningful component discounts apply.[325] Pricing assembly by board area rather than by part count, on the model already used for bare boards, is an alternative structure that shifts the cost driver away from component quantity.[285]
Setup charges dominate first-article costs, so an order of five boards can cost several hundred dollars; the comparison that matters is not the unit price but the cost and delay of doing it by hand, against which paying the setup is usually worth it.[434] Setup charges make low-quantity work disproportionately expensive, which is the structural reason a designer who mostly builds small numbers of things hand-assembles even while using assembly houses for the jobs that justify them.[194] An assembly run cannot start until every part is present, so the constraint that governs scheduling is completeness of the kit rather than the value or difficulty of any individual line.[666]
Assembly capacity is unevenly distributed to the point where whole countries have effectively none: one has a single remaining house specialising in high-end military and radio-frequency work, which will build a small two or four layer prototype but at around a thousand dollars.[494] For a large build destined for a distant market, assembly cost need not be the deciding factor: in one case the difference between local and offshore assembly was small, while shipping, logistics and part sourcing dominated, so the work was routed through a distributor with an office in the destination region.[294] Some board fabricators bundle components and assembly with a bare board order at no additional charge where the design uses generic parts they already hold, which changes the economics of a design constrained to commodity components.[1]
A working relationship with an assembler is worth more than the contract terms suggest: asked to deliver early against a fixed external deadline, one shop accepted a partially used reel the customer brought in, re-taped it, ran a hundred and twenty boards ahead of schedule and completed the balance on normal lead time without charging for splitting the run.[330]
In-house versus outsourced assembly
The economics of owning a placement machine rarely work for a designer: against roughly five dollars a board to have it assembled, a ten thousand dollar machine must also absorb the time to learn, run, tune and maintain it. The case it genuinely solves is being able to build at any hour on demand, which few people need, and where they do, overnight service from an assembly house is usually faster and cheaper.[612] The threshold at which in-house placement starts to make sense is not five boards but a few tens, and buying a machine for one’s own needs creates an obligation to run other people’s boards to justify the purchase.[372] Production placement machines require three-phase power and compressed air and occupy floor space rather than bench space, which is the practical obstacle for anyone contemplating a bench-scale line of stencil station, placement machine and reflow oven.[610]
Occasional in-house assembly never converges on a repeatable process, because each job starts from scratch: the thermal mass of a board changes with layer count, copper weight and the thermal relief on its pads, so the reflow profile has to be re-established every time.[610] The skills decay between builds, paste application most obviously, and a machine left unused for six months has to be relearned, which is why four days of setup to build fifty boards compares poorly with four days of hand assembly.[643] Where a placement service exists that will paste, place and reflow a handful of boards for a day’s wages, that is an order of magnitude cheaper than setting up automation for the same three to five boards.[341] Shared-facility placement machines need a dedicated operator who does nothing else, and the same holds for board milling machines; equipment that anyone may walk up and use does not produce reliable results.[341]
Small manufacturers taking assembly in-house inherit a problem that large manufacturers employ specialists to solve: working out why components do not solder correctly is an entire career in a large organisation, and while defects can be inspected and reworked at small scale, a defect affecting a few percent of a run of thousands cannot be recovered from without consuming the time that was supposed to produce the margin.[11]
Hand assembly
Designing a board and assembling it are separate skills, and competent designers frequently never assemble their own work, either from lack of interest or because the time is better spent elsewhere.[473] Hand assembly is nonetheless useful as a bring-up technique on an untested design, because the board can be populated in sections and each section powered and checked before the rest is fitted, rather than committing the whole board at once.[657] A prototype build that has to be done in days bypasses contract assembly entirely, since arranging an outside build takes five to ten days before anything is made; in one case twenty boards were laid out, fabricated overseas overnight and hand-built in-house to meet the date.[16]
Very small passive packages change what hand assembly means: 0201 parts make 0603 work look comfortable, and modules hand-soldered onto boards during early revisions were moved to professional assembly as soon as the design stabilised.[226] Fine-pitch wafer-level packages are within reach of low-cost fabrication and hand assembly only at the exact process limits: a 0.35 millimetre pitch part required via-in-pad at the minimum permitted 0.25 millimetre spacing, which works only if the layout grid is aligned to it, and not all rows of the package could be escaped; one of three hand-assembled boards came up.[692]
A hand assembly jig cut to the board outline and 3D printed holds the board flat and supported during paste application, which is the step where a small or irregularly shaped board is otherwise hardest to work with.[724] A one-person production process is built out of dedicated fixtures rather than general-purpose equipment: a lead-forming machine that cuts and bends component leads to the board, manual insertion, and a solder bath the populated boards are laid onto, with the whole flow specialised to one product at a time.[513]
Turnkey and online services
Turnkey services collapse the fabrication and assembly steps into one transaction: design files are synchronised from a folder, Gerbers are generated, parts are matched against a catalogue, the user selects which positions are populated and how many boards are wanted, and the service orders the bare boards, buys the parts and ships the assembled result.[182] Behind such a service, a single customer order becomes one to three orders placed with fabricators, assemblers and component suppliers with everything drop-shipped to the right place, so that an unassembled order simply arrives as two boxes, one of boards and one of parts.[163] Some services push placement verification back onto the customer, who checks component outlines against the Gerbers before the job runs, and bundle several customers’ designs onto a shared panel so that the setup cost is shared.[236]
Automated assembly quoting was the hard part of building an online manufacturing service and required a working partnership with an actual assembly house to calibrate the model; board cost was approached empirically, starting from a guess and refining it against thousands of real orders until the model predicted the invoice. The resulting ability to quote in minutes is what enables a three-day turnaround from finished design to board in hand, which a conventional contract manufacturer cannot match because it cannot quote in an hour.[411] Fast-turn assembly services impose a house parts library as the price of the schedule: three-day turnaround is achievable only because the stock is already held and the feeders already programmed for those parts, which removes both external lead time and setup error from the path.[255] A hybrid model lets a shared facility machine-place the standard parts from its library and then hand the pasted board back to the customer to place unusual components by hand before it goes through the oven, which extends automated assembly to designs the feeders cannot cover.[337] An assembly service can hold the component inventory itself and sell only the finished product: it does not sell bare boards and does not sell parts, on the reasoning that stocking the components in-house is what removes external lead time from a prototype order.[699]
Board assembly resists the automation levels reached in semiconductor manufacturing because the workpiece is not uniform: circuit boards vary in size and shape, carry large connectors and heavy components, and warp in ways that a wafer does not, so the handling mechanics differ for every job.[699]
Defects and failure modes
Failures concentrate in assembly rather than in silicon: solder that does not flow, a joint that does not make contact, a part that is wrong. Integrating a function onto the die therefore maximises the chance of it working at all, at the cost of removing every remedy, since a board allows a trace to be cut or a wire moved and a die does not.[483] Defects grow more expensive as boards move up the value chain: a fault discovered after four or five thousand boards have been populated cannot be answered by scrapping them at the cost of a bare board, because the parts and the labour are already committed, leaving only rework or scrap at the assembled value.[532] A bare-board parameter out of tolerance can present as a mechanical field failure rather than an electrical one: board thickness slightly outside specification produced an intermittent rotary range switch on about two percent of units, discovered only after the product had begun shipping.[532] Board runs of a few hundred sit at an awkward scale where hand assembly consumes whole nights but automated test jigs are only just worth developing, and every unit still has to be verified because downstream users depend on them working.[500]
Assembled boards as products
Selling an assembled board has become cheaper and easier than selling the equivalent kit, because the labour the buyer would have supplied now costs less than the labour of kitting; that reversal is why kits have declined as a product form.[217] Supplying parts kits for others to assemble fails on quality: recipients fitted chips and capacitors backwards, and troubleshooting their assembly errors consumed more time than the exercise was worth, which is what drove the move to selling assembled boards instead.[302] The collapse in bare board and assembly cost changed prototyping method rather than just its price: where point-to-point wiring was once the fast route, laying out a two or four layer board that carries off-the-shelf modules on headers is now both quicker and more repeatable.[422]