| Episodes | 34 |
| Mentions | 40 |
| Cited here | 32 |
| First — last | #22 — #715 |
| Top guests | Jeri Ellsworth, Chrissy Meyer, Travis Goodspeed |
| Related | kickstarter · arduino · pcb · contract manufacturer |
Mass production is the manufacture of standardized products in large quantities, organized around interchangeable parts built to tolerance rather than components hand-fitted in one-off batches.[670] It is a problem distinct from prototyping: an organization can build working prototypes without difficulty while nearly failing at the task of building the same product at volume.[104] The difficulty scales with quantity, because the number of things that can go wrong grows continuously as build counts rise, and small prototype runs mask component and process variation that only appears at scale.[279]
Historical development
The early automotive assembly line marked the transition from hand-fitting every part and building in one-off batches to assembling interchangeable parts manufactured to tolerance, an assumption on which modern electronics manufacturing rests.[670] By the Second World War, United States aircraft production had reached roughly one B-17 bomber every 45 minutes, with the entire facility layout and supply chain planned without computing assistance.[670]
In semiconductors, mature processes such as 180 nm allow performance to be obtained by paralleling many identical structures across a very high unit count, reducing the significance of die area on such nodes.[111]
Economics
The economics of mass production constrain design decisions from the outset, and part selection is made against large-volume pricing even when only a handful of units are planned near term.[81] A retail price serves as a rough upper bound on volume build cost: a consumer touchscreen Wi-Fi appliance retailing at 10 to $15 of volume cost to embed the same interface into another product.[23]
Volume price breaks are not linear; the reduction seen when moving from one unit to a hundred does not repeat at a thousand units unless the design uses generic commodity parts, and distributor pricing contains anomalies in which a quantity break is quoted above the single-unit price, so quoted breaks must be checked rather than assumed to fall monotonically.[81] At sufficiently high volume, small savings become large sums: at a million units, shaving ten cents off the bill of materials returns 100,000, which reframes tedious part substitutions as funding additional engineering headcount.<sup><a href="#ref-577" title="Ep 577: Product Lifecycle Management with Michael Corr (February 13, 2022)">[577]</a></sup> Cost reduction extends beyond part price into assembly economics, including qualifying unknown suppliers and reworking a circuit—sometimes placing two resistors where one would do—to cut the number of placements a pick-and-place machine must make.<sup><a href="#ref-577" title="Ep 577: Product Lifecycle Management with Michael Corr (February 13, 2022)">[577]</a></sup> Businesses selling modules under 10 accordingly survive on unit volume rather than per-unit margin.[226]
Non-recurring engineering charges at contract factories are typically kept low and recovered through margin on sustained volume; factories bid engineering effort up front and select programs expected to run for many years, because their return comes from improved margin in the later production years.[113] Factories advertising no minimum order quantity are rare under this model.[113] In offshore knitting, for example, factories quote roughly one dollar per unit for 5,000 identical acrylic scarves, with the minimum order and the single-material constraint constituting the real barrier rather than the unit price.[257] Similarly, injection mold tooling at around $50,000 per mold is a dominant line item that can consume most of a small hardware company’s seed funding when moving beyond 3D-printed parts.[104] Product generations can also be kept in full production maintenance mode alongside their successors when the older part remains slightly cheaper, which matters for extremely price-sensitive applications.[687]
The cost structure of the pre-manufacturing phase has shifted over time: off-the-shelf modules costing around twenty dollars now cover early validation work that once required custom hardware, shortening the phase before a product commits to manufacturing.[715] Hardware accelerators have also emerged that fund small teams roughly 20,000 to relocate to China and drive an idea to a mass production run within a couple of months.[87]
Design for mass production
For weight- and size-constrained products headed to volume, a custom board carrying only the required circuitry displaces any general-purpose development platform, which is not carried into the finished product.[22] General-purpose single-board computers similarly suit one-offs or runs near a hundred units but not finished products built in the thousands or tens of thousands.[372]
Cost, not robustness, sets the design budget for consumer products, and engineers moving from industrial or automotive work over-specify protection and margin until this is internalized.[458] Training in research laboratories teaches design against functional requirements with cost as a secondary concern, a mindset that does not transfer to volume products where cost is a primary driving function.[577] Hardware designed for personal or experimental use commonly minimizes component count while ignoring part price where free samples are available; making such a design mass producible requires a ground-up redesign around different priorities.[442] Converting a solder-it-yourself kit into a volume product likewise means redesigning it as a fully assembled, manufacturable unit rather than ordering more of the same parts.[121]
Design for manufacturability on programs running hundreds of thousands of units begins during early research and development and consists largely of unglamorous compromises, such as enlarging features so they can be probed and tested.[614] Reaching mass production on the first PCB spin is achievable, and has been reported for customers building wireless designs directly from vendor reference designs.[202] Deliberate violations of design rules are survivable only when the violation is engineered to hold across the part variation and process spread of mass production.[222]
Component variation and design margin
A build of a hundred units typically draws all of a given part from a single reel, so the components behave nearly identically and mask any sensitivity to part spread.[279] At high volume a design consumes many reels of each component, and parameters such as transistor threshold voltage shift from build to build in ways never seen in the laboratory.[279] A design overly sensitive to component spread can stop working entirely when a reel arrives at the high end of a parameter the designer implicitly assumed was low, and second-sourcing a component widens the spread the circuit must tolerate because two suppliers’ parts differ from each other as well as within their own distributions.[279] Monte Carlo and simulation coverage of component variation is routine at the chip level but rarely available for every board-level part, so board designs must instead carry deliberate margin.[279] Consumer products must additionally hold their operating margin across the environments customers actually use them in, with temperature the leading variation parameter down to well below freezing.[279] Design trade-offs chosen during prototyping remain unvalidated until a large build exercises them, and some volume behaviors cannot be predicted from smaller builds at all—the only real verification that a design works at a hundred thousand units is building a hundred thousand units.[279][229]
Enclosures and industrial design
Choosing an off-the-shelf enclosure shifts engineering cost elsewhere, typically into internal wiring plus dedicated front and back panel boards to reach the connectors.[124] 3D-printed enclosures scale to roughly one to ten units and then stop; they are not a path into mass production.[124]
On the CastAR augmented-reality programme, Jeri Ellsworth—whose background was in toy design and mass manufacturing—treated the manufacturability of the novel consumer device as the settled part of the risk, leaving the software experience as the open question, and identified the plastics as by far the largest manufacturing risk in a device combining optics, an ASIC and plastics, expecting schedule to be consumed reconciling industrial design against manufacturability.[147][173] Ellsworth had previously observed industrial designers ruling a design and producing shapes with compound curves that could not be mass produced, a failure pattern seen at more than one company.[173]
The production ramp
Validation builds
The standard build ladder before volume is an engineering validation build (EVT), then one or more pre-production builds, then mass production.[394] An EVT build uses production-intent hardware throughout—real plastic tooling, real circuit boards and real optical assemblies—so that it exercises the actual manufacturing processes rather than prototype substitutes.[394] Line bring-up begins with engineering trial builds well before volume, with the express purpose of getting the assembly line itself ready rather than producing sellable units.[279] Standing up a PCB assembly line for one consumer product took six separate trips to a factory near Shanghai across a single year, beginning with such early trial builds.[279]
Ramp versus steady state
The painful phase of a program is the ramp rather than steady state: climbing from one unit to ten, to a hundred, to a thousand per day is where the process problems surface.[437] Once a line is running at steady state, output flows smoothly because sophisticated factories carry heavy support infrastructure specifically to minimize line-down events and recover from them quickly.[437] Volume thresholds force discrete changes of method: a thousand units of a wearable can be built essentially by hand, but sixty thousand units requires a completely different manufacturing plan.[175] Some industries commit to mass producing a fundamentally new technology before the end applications are fully defined, as in communications silicon programs, which drives much of the engineering difficulty.[430]
Quality, testing and failure modes
In semiconductor production, every packaged die is individually tested after wafer sawing and packaging, and the resulting test data stream is monitored continuously as the primary production quality signal; mature chip production additionally runs periodic package qualification to confirm that packaging materials have not begun to introduce reliability problems.[687]
At board level, a marginal timing error does not stay theoretical in production; it materializes as a pile of thousands of dead boards traceable to the responsible engineer.[222] A defect discovered after thousands of boards are built converts into mass rework, including hand re-soldering of 0201-size passives.[279] Because the line never pauses in volume manufacturing, a discovered bug accrues defective units by the hour and forces same-day containment decisions.[279]
A contract manufacturer may quietly hand-solder a part or work around a bad footprint to deliver prototype boards, hiding a defect that only blocks the order once volume is requested.[229] The distinguishing feature of a good contract manufacturer is volunteering manufacturability feedback rather than silently building what was sent, and such feedback is difficult to obtain.[229] Manufacturing-experienced reviewers can also spot impossible production claims in funding campaigns, such as promises to machine 400,000 pieces in two or three days at a cent apiece.[176] Working inside a manufacturer that ships thousands of units exposes the gap between building one of something and building a hundred thousand, a lesson difficult to acquire from design work alone.[663]
Schedule and financing constraints
Crowdfunded hardware forces a promised delivery date of roughly three months, because a realistic one-to-two-year manufacturing schedule will not attract backers.[113] Handing an impossible schedule and a pre-sales budget to a team that has never manufactured anything at volume is structurally set up to fail regardless of the team’s ability.[113] A six-to-twelve-month build cycle before the first units reach customers means all program costs are carried up front with no revenue in return.[715] Tesla’s early history illustrates the separation of the prototyping problem from the volume problem: the company built working prototypes without difficulty, while mass manufacturing nearly ended it.[104]
The Pebble smartwatch programme faced an abrupt version of this transition when a hundred thousand pre-orders forced an immediate jump from garage-built hardware to mass production with a team that had never worked with a factory or visited China; bringing that first volume build up meant relocating a team of seven or eight people, one per discipline, to live at the factory in China for six months.[715] Earlier, after venture funding for the watch fell through, Andrew Witte’s team abandoned the plan to go overseas for full mass production and instead scaled up a local California prototype shop.[175]
United States government small-business research programs run in phases, with awards from tens of thousands up to over $200,000: a short proof of concept, a roughly two-year development phase, and a final phase aimed at mass production.[506] Mass-produced products also carry a certification burden that personal or experimental builds avoid, including FCC ID and RoHS compliance.[442]