| Episodes | 113 |
| Mentions | 197 |
| Cited here | 70 |
| First — last | #4 — #723 |
| Top guests | Jeri Ellsworth, Ken Burns, Pete Staples |
| Related | pick and place machine · pcb assembly · pcb fabrication · component sourcing · kickstarter |
A contract manufacturer (CM) is a firm that fabricates or assembles products to another company’s design, ranging from bare printed circuit board assembly through box build, parts purchasing, inventory holding, and drop-shipping of finished units to the customer’s own customers.[315][255] The business is structurally low-margin: price pressure from the customer’s purchasing function on finished goods means the model depends on volume and machine utilisation rather than markup.[104] Its attraction as a business is the cash-flow shape — a contract shop builds nothing until an order exists and can ship before its inventory is paid for, which makes it startable without capital or a marketing operation.[109] Because a typical shop is built around two or three large accounts making up roughly eighty percent of its business, smaller customers’ work is fitted around the big accounts’ schedules.[411]
Business model and economics
Margin in contract manufacturing is taken at the prototype stage; once a purchasing function is involved on production goods, price is driven down and the economics rest on keeping placement machines running.[104][411] The governing question on the floor is what work can be moved off the placement machine, since anything done on the machine stops it doing its valuable work; the corollary practices are automated feeding and unloading, prepared kits, and staged changeovers so the line never stops.[411] A board with many unique part numbers can exceed the feeder capacity of one placement machine, which is why lines run several machines in series after a single stencil print.[411] The ideal plant is laid out as a long rectangular building with receiving at one end and shipping at the other, so that parts enter one side and finished product leaves the other.[411]
Manufacturers protect their own margins by relocating operations between countries, which makes a customer’s production location a moving target rather than a settled decision.[453] A manufacturer can also be forced into owning capacity it did not want: where partners cannot deliver required quality at modest and bursty volumes — a customer needing a hundred thousand units before a deadline and nothing for nine months afterwards is a lot of disruption for little payoff to a factory — the customer may end up in manufacturing itself, a position that is easy to enter and hard to exit.[544]
House parts carry a commercial dimension alongside the technical one: a board built entirely from the manufacturer’s own stocked components may earn the house nothing, so specifying part numbers where the choice matters is both a technical and a relationship decision.[243] Bare-board suppliers in some markets bundle free common components and assembly with a board order, provided the design uses generic parts.[1] Retailers of third-party designs operate a related arrangement, manufacturing a design in-house under royalty to the designer rather than having the designer build it.[157]
Scope of services
The outsourcing ladder runs from buying finished assemblies from a design shop that manufactures for others, through a contract manufacturer doing varying levels of assembly and integration, up to a large electronics manufacturing services firm that will also design the product.[255] An original design manufacturer (ODM) is a contract manufacturer that also holds design capability in-house — board layout, firmware development, and prototype builds — which distinguishes it from a pure assembly services company.[544] Full-service assembly can extend past the board to buying the parts, holding inventory, and drop-shipping finished units to the customer’s customers, with the whole flow driven from a web interface.[315] Handing the complete build, including part purchasing, to the manufacturer also functions as a continuity measure: production and customer supply continue if the founder is unavailable, and lot sizes as small as ten finished units can be run that way.[607]
Contract manufacturers now absorb a large part of design finishing: a half-built product can be handed over and the manufacturer will complete the design-for-manufacture work and the enclosure.[114] The manufacturers rarely do that work themselves — their speciality is the assembly line, so layout and enclosure design are farmed out to independent designers, which is where a substantial share of contract layout work originates, and consultants also pick up work directly from manufacturers whose clients hit obsolete parts and board problems.[114][409] Enclosure machining is best bought from the manufacturer that assembles the product, because routing cases through a third party adds receiving and shipping steps and its own minimum volumes, and a design change then strands punched stock.[102]
Volume thresholds
The boundary between in-house assembly and contract manufacture is drawn at characteristic quantities. Below about a hundred boards it is usually cheaper to assemble in-house, and above that the case for sending work out becomes clear; at a thousand units there is no reason to be building them oneself.[319] The arithmetic that settles the question is time: eight hours to hand-build five boards, valued at a consultant’s hourly rate, comes to roughly the price of having the same boards assembled commercially.[372] The narrow band where a personal placement machine pays for itself is around fifty boards done repeatedly; into the hundreds the work is farmed out.[364] Benchtop placement equipment serves that band between hand assembly and contract manufacture — internal prototype runs from around fifteen to a few hundred boards — with contract manufacturers existing for the point at which a product is stable and running.[686]
A very simple board, on the order of five parts and all through-hole, is one of the few cases where contract assembly is not worth arranging.[178] Arranging a contract build carries its own lead time of five to ten days before anything is made, so a schedule measured in days forces an in-house build regardless of cost.[16] Kits are hard to justify as a product where local contract assembly is cheap, since the assembled board carries a working guarantee a kit cannot; short-form kits with the surface-mount work already done are the surviving middle ground.[178]
A working definition of high volume in the industry is a million units shipped; below that a product is not a high-volume product regardless of how large it feels to the company making it.[328]
The low-volume gap
A recognised trap sits between garage assembly and contract manufacture, where the volume is too large to build by hand and too small to interest a factory, and the project has not raised enough to buy its way out.[175] The trap has a concrete shape at a few thousand units: a project at roughly twenty-five hundred pieces may find no manufacturer willing to take the whole job, having planned around hand assembly at five hundred to a thousand.[350] At low volume the setup charge dominates unit cost: twenty or thirty dollars to set a machine up, amortised across fifty boards, can exceed the selling price of a small board, which is why a company shipping many low-volume board types may assemble in-house.[458] Very small overseas runs can undercut local ones on the non-recurring engineering charge rather than unit price, with setup quoted at about a hundred dollars for runs of five or ten units.[239]
Keeping assembly in-house in this band buys just-in-time building: instead of ordering five hundred or a thousand boards that sit as inventory for months, production can run a couple of days ahead of orders with parts bought against demand.[237] The countervailing cost is that working through a subcontractor imposes a documentation burden that in-house assembly avoids, which is itself a reason short-run shops keep placement in-house.[224] Placement lines also carry ongoing costs beyond the machine — floor space, noise, and hours of programming — on the basis of which owners have sold entire lines and moved the work to contract manufacturers.[613] Choosing placement equipment badly is a due-diligence failure with a known remedy: open channels with several vendors at once, have the actual board built on the actual machine before buying, and prefer machines that manufacturers run in daily production.[237] A machine that places a small part but does so slowly still costs throughput, and cheap machines built on stepper motors rather than closed-loop encoders cannot hold placement accuracy; because boards are never perfectly flat or square, a vision system is needed, since a few thousandths of an inch of misalignment decides whether a passive lands on its pad or tombstones.[237][299][153]
Domestic versus offshore manufacture
Below roughly a thousand units the labour content matters far less than proximity: building domestically buys a faster design-build-test cycle, better protection of intellectual property, and the ability to sit down with the shop after a short drive or domestic flight.[113] Offshore manufacture starts to make sense at around five thousand units and above, because factories earn on volume through labour content and markup and are hard to interest below that; the offsetting advantage is a dense and diverse component supply chain concentrated in one region.[113] For a foreign customer approaching a Chinese factory, a hundred units is the point at which the factory will engage at all — it does not want the order, but takes it on the prospect of growth.[414] The practical choice is binary: either pay more and manufacture locally, or manufacture overseas and have somebody on the ground managing it, because an unattended offshore build is the arrangement that fails.[113]
Automation has narrowed the cost gap for mid-volume assembly, to the point where good local manufacturers can cover runs in the thousands without the offshore labour advantage applying.[232] Domestic assembly in some markets splits between quality-certified shops built around ISO and aerospace standards and small shops with a maker mindset; a shop in the second group must decide whether it wants to compete for pennies, knowing that volume work tends to move offshore regardless.[458] Miniaturisation has left domestic capability behind in some regions: houses that once built large consumer electronics can still handle small volumes but not extreme ones, which pushes work to a nearby lower-cost country or overseas.[363] Assembly capability also clusters geographically around large former employers, so a region that once hosted a major manufacturer may still support several independent board shops within a short drive.[509] Local capacity can be substantial without being enormous: two assemblers in one region sufficed to build up to two hundred thousand units a year, enough to carry three product lines before expansion was needed.[475]
For a first batch it is almost always better to build somewhere the team can walk down the street and look at it, because the first batch of a new product is where the problems are.[314] Physical presence is what closes the loop: customers who fly over achieve a great deal in a week and then stall the moment they leave, so momentum depends on continuous presence or a knowledgeable representative on the ground.[113] The designer’s presence carries authority a representative’s does not: with the person who can approve a change standing on the line, decisions that would take days or weeks happen in an afternoon.[279] Factory visits matter most when nothing is going wrong — a smoothly running programme invites complacency, and the things to verify in person are that the line has not been moved somewhere worse and that the account still has priority.[328]
Selecting a manufacturer
Referral is the practical way to find a manufacturer, because it borrows someone else’s filtering; a supplier that performs is then kept across projects rather than re-tendered.[161] Bidding a job to several shops and letting one win repeatedly produces a partner rather than a vendor, to the point where odd requests are simply absorbed.[305] Time should not be spent persuading a manufacturer to take work it does not want: a shop that says it will not talk until the customer is spending a million dollars a year is doing both parties a favour by saying so.[305] Stating the volume trajectory honestly at the outset — ten boards now with the hope of hundreds and then thousands — makes suppliers self-select quickly instead of discovering the mismatch later.[305]
Matching the manufacturer’s size to the job matters more than finding the best one: very small subcontractors disappear when they lose a single customer, large ones will not take small jobs, and a small customer cannot buy priority by offering to pay double because the interruption is not worth the shop’s while.[135] Sending a small company to a tier-one manufacturer because a large brand uses it is a bad decision, of the kind that circulates from people who have not run the process themselves.[402] Manufacturing houses differ in what they are actually good at, and that specialisation is not visible from outside, which is the main difficulty for a first-time customer.[363] Some smaller houses invest engineering time in promising startups without charge for a set number of iterations, treating it as a bet that the customer will become a volume account.[363] Design choices narrow the field further: choosing the smallest possible parts to make a product compact rules out many manufacturers, and fine-pitch BGA packages make reliable placement of that part on every board the qualifying question for a supplier.[237][517] One small company took the relationship further still, co-investing with its manufacturer to grow that manufacturer’s capability, which bought it a cost curve normally reserved for very high volumes and predictable build-to-build costs it could plan hiring and new designs around.[302]
A pilot order of a few dozen units before committing to volume is the standard way to test both the product and the supplier.[362] A very small paid run — even four assembled boards with the manufacturer sourcing all parts — works as a rehearsal that exercises the handoff and returns the same kind of feedback a higher-volume manufacturer would give.[445] Pilot quantity is usually set by cash rather than demand, with quotes gathered across volumes to model what gross margin becomes at scale; those models are fragile to supply-side events, such as a vendor acquisition that doubles a processor’s price and raises the bill of materials by ten percent.[330]
Design handoff and documentation
In the conventional flow the customer sends files and the manufacturer checks them, sets up the placement machine, and runs the job; web-based services that make the customer approve every part before the run invert that responsibility.[243] The handoff has to be complete and confident, because changes after it are where the manufacturer earns margin, and the informal loop of walking onto the floor to clarify a detail with a manufacturing engineer does not exist once production is external.[255] Revision identifiers become load-bearing once an outside party is building: a designation has to be unique and specific enough to serve as an address both sides can point at, and manufacturers act on revision changes through the part numbering, routing the change to the responsible team, updating the placement job, and adjusting test coverage.[445][577]
Where the manufacturer buys the parts, the bill of materials is best exported with generic specifications rather than one manufacturer part number, with approved alternates listed in an additional column or a separate substitutes report.[542] Assembly houses stock common passives across the usual case sizes and will supply them from their own reels, which avoids buying a full reel for a handful of boards; the caution is that a house part may differ in temperature coefficient or power rating from the part specified.[17] A product can be designed around the parts a reasonably sized assembly house already has loaded, and designing to a machine’s loaded part set turns placement into a fast iteration loop in the same way print orientation is designed for on a 3D printer — a tighter constraint than a manufacturer would impose, but one that removes changeover entirely.[216][686]
Asking a supplier in advance for the full set of process limits does not produce them; the only reliable way to learn what a process can actually do, as opposed to its stated safe limits, is to build the thing.[229] Design-for-manufacture review by the assembler catches problems before the first build — vias inside a fine-pitch package, or a footprint whose name says one case size while the parts ordered against it are another — and that feedback is a service customers routinely forget they are entitled to.[350][644] Designing for volume is largely a negotiation about what to give up: which compromises on shape, form, and function are acceptable in order to be manufacturable, settled before samples are requested from bidding factories.[592]
Supply chain and component management
Larger manufacturers employ supply-chain staff whose job is continuity: making sure commodity parts have several qualified alternates specified in the design and flagging single-sourced components early, because a manufacturer may need to build a line or open a factory to meet a demand ramp and that cannot be started late.[279] When a component becomes unobtainable during a build, the design house may buy the scarce parts itself and supply them free of charge to the manufacturer so the run can continue, and must explain to the client that a finished, paid-for design can still be unmanufacturable.[645] Parts going missing inside a factory is a tracking problem rather than a loss problem, solved by extending the customer’s own inventory system to follow components through the manufacturer’s process.[722]
The risks of delegation are concrete. An unsupervised factory may substitute components it has a supply relationship with, or buy through a broker carrying counterfeit or poor-quality parts; the resulting failures appear months later at volume, when tracing the cause backwards through a thousand fielded units is close to impossible.[60] A field failure traced through several layers of subcontracting showed how remote assembly compounds small differences: a cable assembly was subcontracted, the connector housings used were not the specified parts, their weaker keying let workers mate the connector reversed, and one reversed orientation killed the first driver in the chain — a failure a protection resistor would have prevented.[412]
Test and quality
Placing acceptance testing at the manufacturer keeps failing boards inside the factory, which gives the manufacturer its own reason to fix the process because it wants high yield; yield must be monitored, and failures escaping to the customer mean a test is missing.[522] A security step delegated to the manufacturer as a final production action can silently not happen: a shipped consumer device was found with its debug interface unlocked because locking it had been pushed to the contract manufacturer.[552] Large manufacturers employ engineers whose entire career is finding out why components do not solder correctly; a small company scaling up has none of that, and a defect appearing on a few percent of tens of thousands of boards consumes the company in rework rather than production.[11]
The worst event on a contract line is the line going down, because every idle day costs money and earns none for either party; a stoppage can justify flying an engineer in from wherever they are, and one such stoppage took four days on site to clear.[474] Sustaining engineering is the function that owns fielded product: tracking return trends, finding root cause, designing the failure mode out, and keeping the supply chain stable as volumes and vendor relationships change.[474] Production ramps in stages: the first builds are run by the design engineers alongside the manufacturer, more staff are trained as volume grows so the designers are not permanently on site, and once units are in the field the difficult failures route back to the design team.[363]
Failure modes
The recurring failure of first-time hardware companies is the assumption that a factory will fix an unfinished design: factories employ engineers, but there is no reason to expect them to be better than the customer’s own, and trusting an unknown party to redesign a board is a risk taken blind.[60] A manufacturer that quietly works around a design problem at prototype quantities — for instance by hand-soldering a part the machine cannot place — leaves the fault to surface at volume when the same shop declines the order; feedback about what will not scale is what separates a good manufacturer from a poor one.[229] A manufacturer that did all the design work may refuse, at the exit, to release the data needed to move production elsewhere, and ordering a custom part from a factory for decades without ever holding the design data leaves the customer unable to reproduce a part it has shipped in every product once the individual who arranged the original order leaves.[114][716]
Recovering money from a manufacturer for its own error is close to impossible; the realistic best outcome is an agreement to fix it in the manufacturer’s own time, costing the customer a week or two of schedule.[174] The relationship can also be ended by the supplier, whether because the account is too small to be worth keeping or because of friction with particular people at the customer.[139] Blaming a vanished contract manufacturer is a common account of a failed crowdfunded hardware project, and the underlying error is treating funding as the end of the work rather than the start of a process that has to be supervised; crowdfunded projects attract unsolicited offers from overseas firms proposing to handle everything for around a tenth of the money raised, with full-service arrangements of that kind quoted at ten to fifteen percent.[142][345]
A further cost is borne by the customer’s own engineering capability. Design work moving to external manufacturers costs younger engineers the experience of seeing production, which was previously how they learned why a board behaves the way it does.[180] Small process improvements come from proximity — an engineer next to the line notices that moving a resistor would ease assembly or stop a capacitor cracking, and an external manufacturer has no comparable incentive to feed that back — and engineers and product companies outsource earlier than they should, given the range of volume a team could build itself to improve both its ability to ship and its understanding of its own product.[305][405]
Web-based and automated services
Instant online quoting from one to ten thousand units, with lead time and price returned immediately and payment by card, replaced the request-and-wait cycle for small assembly orders, initially served through a network of other people’s factories.[699] Quoting speed is part of a manufacturer’s efficiency rather than separate from it: a traditional shop takes about two weeks to return a quote, while automated quoting requires a cost model built from thousands of past orders, and modelling assembly cost is much harder than modelling bare-board cost.[411] Web services that make the customer approve every part before a run invert the conventional division of responsibility, and bringing bare-board fabrication in-house is justified only to enable next-day turnkey service — going further and selling bare boards to others leads into a business with little money in it.[243] Full-service flows can extend through purchasing, inventory, and drop-shipping to the customer’s customers, all driven from a web interface.[315]