| Episodes | 80 |
| Mentions | 108 |
| Cited here | 40 |
| First — last | #17 — #718 |
| Top guests | Chrissy Meyer, Simone Giertz, Jeff Keyzer |
| Related | contract manufacturer · injection molding · digi key · kickstarter · prototype |
Design for manufacturing (DFM) is the practice of shaping a product’s design around the capabilities, constraints, and economics of the processes that will build it, so that the design can be produced at the intended volume and cost.[33][437] Effective DFM depends on the designer having visibility into the specific equipment and supply ecosystem of the chosen manufacturer, because layout, component selection, and mechanical geometry are optimised around what a particular assembler or fabricator can actually do rather than around abstract best practice.[33][450] The discipline spans printed circuit board fabrication and assembly, mechanical processes such as machining, casting, and injection moulding, test fixturing, and supply chain decisions, and it determines whether a working prototype can become a product at a viable price point.[268][644]
Process visibility as a design input
The central premise of DFM is that the designer must know what is on the machines that will build the product, because the layout and bill of materials are optimised around the assembler’s actual configuration.[33] Designing next to where a product is built exposes problems invisible on paper, such as a component that consistently falls off the board or an enclosure that takes three and a half hours to assemble where a single change would reduce that to half an hour.[124] Visiting several comparable factories, rather than one, establishes a sense of what is normal and what is unusual for a process, and watching parts being made leads directly to better designs.[450]
Firsthand assembly experience plays a similar role: hand placing the first boards of a design generates most of its manufacturability insights, because the designer directly experiences the consequences of choices such as an unnecessarily small passive size.[612] Physical mechanisms also make design rules self-evident; the minimum via size a fabricator accepts follows directly from the fact that the hole is mechanically drilled, so the smallest available drill bit sets the limit.[414]
Cost structure and volume
Non-recurring engineering (NRE) cost is roughly flat between a thousand and a hundred thousand units, so the volume a product is designed for changes the design itself far more than it changes the up-front engineering charge, and a large swing in expected orders can force substantial redesign.[147] The correct trade between design time and unit cost follows volume: at five million units, several days of engineering to remove ten cents of cost pays for itself many times over.[699] Most electronics designs, however, are industrial or medical products built in the low thousands, where optimising for manufacturing cost is the wrong objective and accessibility, reworkability, and robustness matter more.[699] An electronics manufacturing services provider handed a design will often re-engineer circuit topologies for cost at consumer volumes, for example replacing a switching regulator with a linear one where the additional power loss is tolerable, because saving twenty cents on the bill of materials is significant at those quantities.[291]
Component pricing does not halve with each order-of-magnitude increase in quantity: a jump from single units to a hundred can roughly halve the price, but a further jump to a thousand generally does not repeat that saving except on generic jellybean parts.[81] Part selection during design should therefore be made against volume price breaks rather than single-unit prices, even when only prototypes are being built, because the chosen parts are difficult to change once the design is frozen.[81] A component repeated hundreds of times dominates a bill of materials, so its unit cost effectively sets the product price; a calendar product with one mechanical switch per day of the year would have reached roughly fifteen hundred dollars retail.[592]
The threshold for moving production overseas is better judged by annual revenue than by unit count, since an expensive product justifies the supply chain at low volume; one contract manufacturer put its own threshold at a million dollars a year, low relative to the tier-one houses.[544] Hardware has no equivalent of recompiling, so the design entering production must be correct; from completed DFM work and a hundred verification units, reaching very large quantity fulfilment typically takes up to eighteen months rather than the one to three months backers are often promised.[268]
Component selection and the bill of materials
Specifying parts an assembly house already keeps loaded on its feeders or holds in stock can be cheaper overall than specifying a nominally cheaper part, because it avoids feeder changes and procurement effort; the saving justifies a slightly higher unit price on the component itself.[33] Designing from a standard in-house part library, and quoting against that library rather than against specific manufacturer part numbers, lets an assembler stock a single set of parts that covers every board in the product line and shortens procurement.[125] Constraining a design to a fixed personal part library, in the way a printed part is designed around print orientation, allows boards to be iterated on in-house without feeder changes; the constraint is tighter than a contract manufacturer would impose but is chosen deliberately for speed.[686]
Catalogue distributors list every part but give no indication of which parts are commonly used or available at manufacturing scale, so a bill of materials assembled purely from distributor part numbers can be unbuildable or expensive in a factory’s local supply ecosystem.[121] Substituting a locally standard component for a rarely used equivalent can cut that line item to roughly a tenth of its cost, which is why factories routinely return a submitted parts list with alternates drawn from what is common in their own market.[121] An unexplained jump in an assembly quote often traces to sourcing: rather than substitute, the factory imported the exact specified part and paid duties on it; specifying generic parts, or explicitly authorising local equivalents, prevents this.[125]
A cost-down pass during DFM typically has the contract manufacturer propose substitutes from its local supply ecosystem, which the designer then evaluates and approves; this is how designs prototyped with locally available parts end up populated with regional brands.[580] A bill-of-materials export checked before release against the assembler’s own catalogue and stock levels acts as a lightweight manufacturability review, flagging parts the assembler does not stock and which will therefore carry a sourcing surcharge.[700]
Multiplying per-board component counts by planned annual volume exposes availability limits early: fifty LEDs on each of fifty thousand boards is two and a half million parts, a quantity few sources hold, with lead times of at least twenty weeks that must be built into the schedule.[502] Committing to a long-lead part at volume locks the design to it, so an engineering shortfall found later, such as inadequate thermal testing, cannot be fixed by substitution without writing off the parts already ordered.[502]
Printed circuit board fabrication
Fabricators price by process class, so relaxing a design from three-thousandths track and space to four moves it into a cheaper class and widens the set of fabs that can build it; designing to the finest available geometry as a point of pride removes that option.[502] Where a fabricator’s cheapest process class is known, designing to it rather than to the finest capability available keeps a board in the low-cost tier; a designer whose regular fab quoted six-thousandths-of-an-inch minimum track and gap standardised all small boards on 6/6 for that reason.[33] Bare-board cost falls with layer count, so escaping a ball grid array’s pins on four layers rather than six is worth pursuing where the routing allows it.[502] Selecting a layer stackup from a fabricator’s published standard set, rather than specifying an arbitrary one, avoids a custom quote and its cost and lead-time penalty; comparing available stackups and prices across several fabs to find where the cost curve bends is a routine part of the work.[718] Buying a fine-pitch subsystem as a module confines the tight tolerances to a small board built by someone else, and the module often arrives pre-certified for radio compliance, at the price of a higher unit cost.[502]
A fabricator should not silently alter a customer’s copper, for instance by widening traces, because the design of record then diverges from what was built; corrections belong in the designer’s own files.[299]
Printed circuit board assembly
Placement machine setup is quantised by feeder count: a board with one more unique part than the machine has feeder slots forces a second setup pass, which is why unique part count, not total placement count, drives assembly setup cost.[153] Minimising the number of unique parts on a board is a core DFM lever, and a typical machine setup charge for a small board runs around 250 dollars.[153] A design constrained from the outset to hand-solderable geometry, with 0603 passives and no package finer than SSOP, needs little preparation before it can be machine assembled, because every such limit is looser than what a pick-and-place line requires.[125]
Solder paste deposition sets a practical floor on component size: below a certain size jet printing is no longer usable and a stencil is required, and very small apertures demand an expensive stencil, so shrinking passives raises assembly cost as well as risk.[411] Assemblers routinely flag vias placed in a QFN’s thermal pad area before build, because the via draws solder away during reflow; changes of this kind are identified in the contract manufacturer’s design review rather than by the designer.[350] Through-hole pins soldered by a selective soldering machine need a clear area around them on the underside of the board, on the order of ten millimetres, because the selective solder head cannot reach a pin surrounded by surface-mount parts; violating the keep-out forces the joints to be hand soldered at much higher cost.[447] Components falling off the first side of a double-sided board during the second reflow pass is a predictable failure, prevented at the design review stage by checking part mass and placement against the second-pass profile rather than by adjusting the process.[716]
A DFM review performed before a board is released checks clearance between parts for the pick-and-place head to reach, spacing between adjacent components, and which side of the board each part sits on, and may require the board to be laid out again.[716] At quantities of ten or twenty an assembler can absorb a poor layout with custom jigging and manual intervention, but by a few hundred units the process must run unattended, so design faults that were merely inconvenient become the limiting cost.[716] An assembler prices a difficult board with a risk premium rather than a cost increase: the run may go through cleanly and simply return higher margin, but the probability of rework and scrap is what the surcharge covers.[411] A factory reviewing a finished design frequently identifies trivial changes, such as reorienting or shifting a component, that would have removed several dollars of unit cost had they been made during layout.[451]
Mismatches between the ordered component package and the footprint placed on the board, such as 0805 parts against 0402 land patterns, are a routine finding in a first DFM report and are caught before build rather than at assembly.[644] Errors cluster in the final stage of a design, when late manufacturability-driven part substitutions are made under schedule pressure and without review; on one board the last five percent of the work produced the large majority of the mistakes.[652]
Test and fixturing
Testability work can be moved earlier in the design cycle by routing test access through a JTAG chain or onto silicon, which reduces the number of physical test points a board needs; the physical handling, fixturing, and probing of boards through the production line remains regardless.[219] Automated test and handling equipment is built around a fixed board geometry: custom trays, pneumatic ejectors, and bed-of-nails fixtures all assume the board’s size and test-point positions, so moving a test point or changing board dimensions invalidates the fixturing investment.[321] A company masters production by being present on the line, designing its own assembly and test fixtures, handing them to the contract manufacturer, and iterating on them, rather than delegating the whole process to a large supplier.[402]
Mechanical design
A machined feature that cannot be reached on a two-axis machine forces the job onto a five-axis machine at substantially higher cost, so the axis count of the intended machine is a design constraint rather than a shop-floor detail.[153] General-purpose mechanical CAD packages do not check whether a modelled part can actually be machined, so an inexperienced designer discovers unmachinable features only through repeated iterations with the shop.[379] A quick manual test for machinability is to imagine the cutting tool as a pencil held perpendicular to the part and to trace how each feature would be reached; anything that cannot be reached that way needs a tilted head or a machine with more axes.[379] A manufacturability check written by the shop that will cut the part can flag geometry no general CAD package will, such as an internal corner radius smaller than the smallest available router bit, and highlight the offending feature in the model.[685]
A part must be designed for the process that will make it, since the geometry suited to additive printing differs from that suited to sheet-metal bending or turning; a designer unaware of the available processes produces generic curved shapes that suit none of them.[414] Choosing a fabrication process to match the geometry matters more than defaulting to the most flexible one: a folded sheet-plastic part made over a heated nichrome wire is straightforward, while printing the same shape struggles with undercuts and thin walls.[437]
Injection-moulding design guidelines such as permissible draft angle are tiered by mould budget: cheap tooling imposes the most conservative rules, expensive tooling relaxes them, and only operations running very high volume tooling can attempt geometries that are otherwise impossible.[218] Increasing a moulded part’s draft angle from two to three degrees lets the part release from the tool faster and raises yield, with no functional or aesthetic consequence, which is the kind of change a moulding or machining shop will identify in minutes.[577] For moulded plastic parts the decisive threshold is the cutting of tooling steel: design changes must be made before the mould is cut, because afterwards they mean modifying or replacing the tool.[437]
Volume products reserve internal clearance for the battery to swell as it ages, because a predictable fraction of the fleet will show expansion; consuming that allowance to fit extra hardware is acceptable in a one-off but would cause field failures across a production run.[414] Connector pin assignment should leave spare pins as margin between high-voltage supply pins and low-voltage data and ground pins, so that moisture ingress bridging adjacent contacts does not put the supply directly onto signal lines.[447] Teardown comparisons of a Tesla against established volume car makers found substantially more fasteners for the same function, for example five bolts where Toyota used two, attributed to designers without experience of designing for low-cost volume production; the consequence appears as production cost and throughput problems.[447]
From prototype to production
Custom mechanical parts should be designed so they can be produced by a prototype-compatible process as well as the production one: a heatsink destined for die casting should still be machinable, and an enclosure destined for injection moulding should still be printable, so prototypes can be built before tooling exists.[436] A prototype optimised for hand assembly, local sourcing, and printing is effectively a different design from the production version, so schedule and iteration must be budgeted for the transition, including several runs with the real suppliers to find and fix issues.[437] An enclosure developed as a printed part frequently cannot be produced by the intended volume process, because printing imposes almost none of the constraints that moulding or casting does, and the mismatch surfaces only when the design reaches the production house.[682] Starting a product from an off-the-shelf enclosure and designing the board to fit it removes tooling cost and mechanical risk entirely, and is a common origin for successful small-volume products.[682]
A design can be complete, working, and indistinguishable from a finished product while still failing to reach production, because it is manufacturable only at a cost above any viable price point.[644] The turnkey contract manufacturer’s role extends beyond assembly into completing the design, including redesigning portions of it for manufacturability and producing the enclosure, so that a customer can hand over a hand-built prototype and receive finished units.[114]
The DFM review and its automation
DFM cannot be performed alone from published best practice; on a moulded part it consists of an extended exchange with the supplier about what they can and cannot do and what their mould flow analysis shows, which on a complex product can run for months before tooling is ordered.[437] A DFM review on a mechanical part at consumer scale produces a report of roughly twenty pages, worked through line by line with the supplier over hours of calls; this is engineering work, not a project management task.[437] Withholding CAD files from the manufacturing partner until after internal design review delays their feedback until the design is frozen; a five-minute change identified at that point cost two weeks of schedule that could not be recovered.[577]
Board houses have exposed their internal incoming-inspection checks as self-service web tools that accept uploaded Gerbers and a bill of materials and return a manufacturability assessment with a visualiser, removing the need to route a simple quote through a sales engineer.[504] An automated board service can interpose a DFM layer that inspects an uploaded design and rejects the order outright when the design would fail, rather than accepting it and building a board that cannot work.[299] Automated manufacturability checking covers simple two- and four-layer boards without human involvement, but the number of possible combinations of via structures, track and gap, and stackups grows so large on complex boards that human review is still required.[699] Manufacturing constraints ideally belong inside the design tool rather than in the fabricator’s checking stage, following semiconductor practice where the fab publishes a process design kit expressing its rules as executable code rather than as a specification sheet for the designer to transcribe.[699]