| Episodes | 122 |
| Mentions | 160 |
| Cited here | 81 |
| First — last | #3 — #699 |
| Top guests | Elecia White, Chrissy Meyer, Eric Klein |
| Related | kickstarter · pcb fabrication · 3d printing · component sourcing · design for manufacturing |
A prototype is a preliminary physical realisation of a design, built to be evaluated, tested or demonstrated before a product is committed to manufacture. Its usual output is a decision to abandon the idea: most ideas are unworkable, and building a prototype rather than entering product development directly is what makes that discovery cheap.[550] Prototypes are conventionally described by stage — proof of concept, looks-like and works-like — and communicating which stage a given unit represents is a substantive part of presenting it.[350] The distance between a working prototype and a manufacturable product is where hardware programmes most often fail: by the time a design reaches manufacturing, what is built typically differs from the original prototype by nearly every measure.[628]
Function and stages
The role of most prototypes is to demonstrate that an idea is unworkable, because most ideas are; the prototype exists so that this conclusion is reached before product development money is spent.[550] A demonstration prototype does not need working features, but everything shown must be internally consistent and plausible, because a viewer who stops to ask why an element is present has stopped evaluating the idea.[550]
Prototypes progress through distinguishable stages — proof of concept, looks-like and works-like — and stating which stage a unit represents is part of presenting it, since each stage supports different conclusions.[350] A looks-like, works-like unit assembled from development modules is a legitimate deliverable in its own right when the quantity needed is small and size is unconstrained: on one programme, Elecia White’s team delivered a prototype built from development modules fitted together for a run of about twenty units.[422] The term “flight hardware” denotes the final prototype, the version considered ready to go.[119]
A demonstration unit that both functions and looks finished produces a step change in what a company can do with it, because outside parties — customers, investors, buyers — can interact with the product rather than with a description of it.[402] What counts as that unit extends beyond the physical part: the website, data path or consumable through which a user interacts with it is part of what must exist.[402]
The physical build
Holding the physical board exposes errors that extended review of the design does not: a first assembled prototype of a design that had been reviewed on and off for months revealed roughly ten errors immediately on handling.[134] Modelling is not a substitute; a 3D model and a design review are not equivalent to having a board in hand and working with it.[79] The same effect appears at the layout stage, where decisions that looked acceptable on screen read as obviously wrong once the physical object exists.[127] Automated design-rule and electrical-rule checks can catch much of what prototypes catch, but hardware still requires the physical build before production; there is no equivalent of pushing a validated change straight through.[546]
Cycle time is the governing variable: where a part cannot be exercised at all until it is on a board, the sooner it is on a board the sooner it is known whether the design is broken, and the sooner that knowledge feeds back into a fix.[166] Some designs cannot be prototyped on a breadboard at all, which sets a floor on the time and money any attempt requires.[209] Thorough verification accordingly means building every circuit rather than trusting the design on paper: Ron Quan validated the projects in his book by physically constructing roughly twenty-six breadboards, close to one per chapter.[133] Comparative evaluation is likewise done by building rather than reading; one assessment of ten embedded Linux platforms was carried out by physically constructing about thirty-five prototypes.[514]
Iteration count is itself a defensible measure of engineering progress. One university course graded student projects directly on prototype count — one prototype for a D, two for a C, three for a B — provided each revision represented real change rather than a component substitution.[302]
Iteration strategy
Where board respins are cheap, spinning a board a couple of times deliberately beats insisting on first-time correctness, because the delay of getting the design perfect is typically followed by a respin anyway.[79] On a cheap prototype the correct discipline is to release the design before it is finished, accept that some of it will be broken, and repair it with bodge wires; where only one or two expensive fabrication turns are available, the calculation changes and more verification time is warranted.[287] A partially finished circuit is still worth fabricating, because a board that can be populated in part, fitted into a case and handled answers questions the unfinished design cannot.[287]
Simulation earns its cost where placing the circuit on a board is expensive or a respin is unavailable, so that a day of simulation buys a higher probability that the first prototype works.[17] Going from simulation straight to production without a prototype does occur — a power converter simulated in LTspice has been put directly into production — but it is a step taken under schedule pressure rather than a recommended path.[310] Conversely, the proof-of-concept stage can be skipped entirely in some workflows: once a circuit exists on breadboard or veroboard, going straight to the production board avoids an intermediate one-off build that teaches little.[360]
Rigour, simplification and part selection
Practice on prototype rigour varies widely between engineers, from deliberately rough first articles to a standard that accepts nothing less than a final-quality board and bill of materials even for a one-off.[291] Hand-built first prototypes are expected to be crude enough to be embarrassing, and that is not evidence of poor work.[461] The relative crudeness of early prototypes is characteristic of genuinely new work rather than a mark against it.[328]
Deliberate simplification is a valid technique: brute-forcing every ground together produces a functional board with a higher noise floor, which is preferable to blocking progress on a grounding scheme that would take a full review to get right.[68] At the prototype stage a reflexively chosen part carries little consequence and can be revisited later; the exception is the processor, where memory ceilings and switching costs make the choice expensive to reverse.[349] The danger is not the reflexive choice itself but carrying it into production without ever re-examining it.[349] Adding a function through a separate peripheral module rather than an integrated part may not be the optimal design, but the speed it buys is usually worth it at the prototype stage.[629]
Sourcing, fabrication and cost
Component distribution splits by role: catalogue distributors, which hold single-piece stock, are where prototype and small-run quantities of ten to a hundred units are obtained, while volume distributors are approached for reels at production quantities.[116] Prototypes built from the research-and-development parts drawer or from vendor sample bins produce a bill of materials that does not correspond to parts anyone can actually buy, which becomes a problem the moment the design moves forward.[273] An individual without a registered company can find that suppliers will not sell them parts at all, a practical barrier to building a first prototype independently; Christina Cyr encountered this when sourcing parts for her own device.[475] Sourcing — finding parts, buying them, opening packaging and organising them — is a substantial and usually unaccounted share of the effort in prototype work.[219]
Scarcity changes purchasing behaviour. Buying a stock of a scarce connector ahead of need is calculated in prototypes rather than products: two hundred parts covers roughly five to ten prototypes per project across ten or twenty projects, and rework with hot air consumes them faster than that.[645] Twenty-week component lead times can dominate a prototype programme to the point where waiting is the correct decision rather than redesigning around the shortage.[601]
Prototype debugging can uncover errors in the component vendor’s own documentation: repeated destruction of a linear regulator during one project was traced to undocumented back-to-back diodes between its set and output pins, with the vendor’s own application circuit exceeding the device’s limits, and later variants of the same architecture carried a series resistor the original part lacked, implying an undisclosed fix never applied to the original device.[140] A returned prototype must be verified against the design before it is trusted, because the designer is the one who confirms that the parts fitted are the parts specified.[140]
Fabrication methods and cost benchmarks
In-house board milling suits quick, rough prototypes only, since it produces neither solder mask nor plated-through holes; for simple analogue work a board can be cut in about two hours.[176] A solder-paste dispensing machine at around ten thousand dollars is flexible enough to paste a board directly from its Gerber data without a stencil, but is slow enough to suit one-off prototypes and rework rather than production.[506] In-house reflow equipment is justified against the prototypes it saves rather than its price alone, and increasingly complex boards, including ball-grid-array assemblies, are built locally that would previously have been sent to an assembly house.[608] Flexible circuits earn their place in prototypes that must fit a small, awkward volume, and are worth fabricating in house only if the turnaround is on the order of an hour; otherwise ordering a flex board is the better route.[415] Custom enclosures are the expensive element of a prototype: where one is genuinely required, it costs both money and time to produce.[50]
A complete prototype — a one-off fabricated board, a laser-cut case, printed mechanical parts and the components, all delivered — can be produced for on the order of a hundred dollars, a benchmark stable enough to be tracked over time like a basket of goods.[619] The same hundred-dollar figure was estimated for a cheaply fabricated board with inexpensive parts and a printed mechanical section once free schematic and modelling tools became available, at the cost of considerable effort and an unpolished result.[49] A convincingly fit-and-finished prototype adequate for a promotional video can be produced for well under ten thousand dollars using laser cutting and printing, unless the object is large or made from formed metal.[576] First prototypes do not require large capital: a handful of first articles can be built for around a thousand dollars, and runs of a dozen units have been built entirely by hand.[87]
Prototyping as professional practice
High prototype throughput is itself a learning mechanism: the value comes from touching many kinds of problem and being able to discard a failed circuit and move on with the lesson rather than defending it.[194] At the toy invention house where Todd Bailey worked, the shop built roughly two hundred prototypes a year, of which about ten sold and two to four made money, and treated the ability to build almost anything in house — machinists, model makers and sculptors alongside the electronics staff — as the core of the business.[194] The standard held there was that a prototype had to be better than the production article, and a prototype that could not be built as a product was generally not worth making at all.[194] Showing the physical object is also what closes a sale in a pitch-driven business, because a buyer responds to the thing rather than to a drawing.[194]
Large companies build prototypes at a steady institutional rate — twenty or thirty a year within a single design group — which is what makes a prototyping service a repeatable business rather than a one-off one.[699] Some consultancies keep prototype construction in house deliberately: Bob Davidson’s practice builds every prototype it needs on the premises, on the view that the hands-on connection measurably improves design work.[232] A consultant unsure whether a job is achievable prototypes the doubtful part before accepting it, under the rule that work should never be accepted without knowing it can be delivered; this makes a small speculative prototype an instrument of risk management rather than of design.[135] Building a prototype oneself before engaging a specialist firm is likewise a way of formulating and communicating the requirement, with the professional engaged afterwards to make the result robust.[615]
An engineer whose electronics knowledge stops at wiring modules together can legitimately deliver a prototype, provided the boundary is stated and a suitably qualified engineer is engaged before production.[281] For a build of about three units, assembling the product from off-the-shelf components is cheaper and faster than a custom design, and the custom product is only worth discussing once volume can amortise the engineering.[635] A personal project built to prototype standard functions as evidence of capability when approaching an employer or client, even without commercial work behind it.[143] It is feasible for a single person to take a design as far as a working prototype or proof of concept without a multi-person company behind them.[606] Photographing every prototype builds a dated record that makes a year’s engineering progress legible after the fact.[613]
From prototype to production
The gap between a working prototype and a manufacturable product is where hardware programmes fail: the recurring situation at consultancies is a client arriving with a development-board prototype that must then be turned into a real thing, and consulting value at that stage lies in reaching a manufacturable product quickly, which is only possible once it is settled what that manufacturable product is.[628] A prototype optimised for hand assembly, printing and locally available parts is effectively a different design from the version a factory will build, so the transition has to be planned as a deliberate design freeze followed by a manufacturable second version.[437] Where a design should sit on the spectrum between quick-and-dirty and fully protected against future manufacturing problems has no general answer, and is among the most contested judgements on a startup engineering team.[437]
The transition is not a single step. There is a real step change between a first prototype and a run of even a hundred units, distinct again from the step to volume manufacturing.[291] One consultancy characterises client readiness as the eighty percent point: a prototype exists and demonstrates the concept, though it might run only two or three minutes before breaking, and the remaining work is taking that into tens or hundreds of thousands of units lasting one or two thousand hours.[113] Making prototypes and mass manufacturing are different problems of different magnitudes; a vehicle programme that could build prototypes without difficulty was nearly destroyed by manufacturing them.[104] Development boards and finished products are different engineering problems, the former being less complex and more narrowly specified than a general product.[219]
Mechanical work follows the same pattern. A printed mechanical prototype achievable in a month is not evidence that the mechanical design is close to done; the manufacturable design plus its several spins is a different order of work.[173] An enclosure acceptable as a prototype fails once tens of thousands of injection-moulded parts are contemplated, because the mould wears, the plastic sags and the appearance shifts, and the finished product then has to be certified.[287] Regulatory certification is a stage most first-time hardware developers underestimate, and going through it once tends to discourage choosing a custom part over a pre-certified finished module the next time.[287] Modular sub-assemblies with castellated edges work well in a prototyping context because functions can be swapped in and out, but the space, cost and complexity they add are usually not justified in the final product.[541]
A well-finished prototype creates a false sense of completion in people who do not build hardware: a solid block of machined ABS matching the industrial design drawings prompted an executive to order twenty-five thousand units of what was still six months from being manufacturable.[422] The same false sense of completion afflicts the designer, since a printed prototype that works can mask defects such as delaminated layers, which get deferred as things to be tweaked later.[337] A part that performed correctly in the prototype can still be the wrong part if the qualification work — thermal testing and the like — was not done, and at volume the error is discovered only after hundreds of thousands of the part have been committed.[502]
A consultant handling a prototype programme has to state the full sequence to the client in advance — design, prototype build, qualification, and the real possibility that something will need adjustment — rather than quoting the design alone.[601] The prototype-to-production handoff is what makes production work unattractive to some consultants: the customer returns months later, by which time everything has to be recalled and the documentation produced.[294] When a prototype is handed to an engineer for productionisation, the engineer produces the engineering drawings and the design-for-manufacture model from a collaborative CAD model rather than from the prototype itself.[592] White-label development inverts the usual order: the starting point is an existing manufacturer’s prototype seen in a showroom, which is then modified against the manufacturer’s stated capabilities.[364]
Manufacturing readiness
Prototypes coming off the contract manufacturer’s actual pick-and-place machine, rather than from hand assembly, are the signal that a design has reached manufacturing readiness; a pilot run of tens of units follows.[362] Iterating through a series of prototype units with a supplier is also how a buyer verifies what hardware that supplier is actually capable of producing before committing to a purchase.[362] The rapid-iteration ethos of prototyping — discarding last week’s state and moving forward — stops working at the point a contract manufacturer is engaged, where identifiers and records become load-bearing: prototype revisions need unique identifiers meaningful enough to name when talking to a manufacturer, because a description in terms of a revision plus whatever firmware happened to be loaded conveys nothing.[445]
A prototype should never be a single unit: an engineering prototype is built in tens so that the assumptions made during development can be tested against feedback from real users.[402] Selling a unit at engineering-prototype maturity by the hundreds or thousands to paying customers, rather than to ten or twenty people who know they are testing a beta, is a recognised failure pattern.[402] Having produced a prototype is not evidence of the ability to manufacture: an applicant may arrive with a working unit and still lack the experience, or carry the feature-creep tendency that sinks funded projects once money arrives.[207] A stronger requirement than a prototype is a completed pilot run and a bill of materials, since only a build produces realistic costs.[289] At one product company, hiring a second engineer was what allowed the effort on design detail needed to have a product rather than a prototype; the programme, led by Jason Cerundolo, built on the order of hundreds of prototypes and concentrated its limited engineering resource on the hardest subsystem while pushing the simpler assemblies onto a manufacturing partner.[340]
Thinking about prototype spend in unit prices conceals the cash requirement of the transition: a hundred-dollar unit cost becomes twenty thousand dollars at a two-hundred-unit pilot, and running out of cash at that point is a common cause of failure.[417]
Funding, sales and demonstration
A single prototype that both functions and looks good produces a step change in a company’s valuation, because an investor can interact with the product; reaching that point before tooling, hiring a large factory and spending millions is the objective of early funding.[402] Early-stage investment is made against a proof of concept that barely works — described as bubble gum, rubber bands and a development board — with the ability to communicate the vision wrapped around it carrying as much weight as the hardware.[495] Having working prototypes to show distinguishes a hardware startup from the substantial proportion that are presenting an idea alone, and bootstrapping the early units is treated as evidence of seriousness.[147] A prototype in hand is what secures the meeting: having something to show is the difference between competing on a proposal and competing on a demonstration.[334]
The counter to scepticism about a hardware claim is physical inspection rather than more video material: Jeri Ellsworth’s successful approach was to invite doubtful parties to handle the prototype in person, where supplying videos and materials had not worked.[147] A recognised startup pattern is to build a rough prototype, secure press coverage of it, and use that coverage as a springboard to further meetings.[394] Investor meetings are a non-renewable resource: during the CastAR programme, Ellsworth’s company went out to raise without working prototypes, showed hardware the same investors had already seen, and consumed the whole pool of available investors.[394] Time spent on activities such as repeated rebranding is time not spent on prototypes, and that displacement is a direct engineering cost in a startup.[394] An applicant to an accelerator without a working prototype is advised to assemble a half-working one and show it to people directly before interviews, or failing that to bring other hardware as evidence of capability.[268]
Customer engagement follows similar rules. Christina Cyr’s customers repeatedly failed to engage with paper, cardboard and metal models of her device, and only began treating the project seriously once a prototype existed that could make a call and receive a text — even though the internals were not the intended design; the outcome runs against the taught sequence of validating with a drawing, then a cardboard prototype, then progressively more resilient models before building anything complete.[475] Smaller and less established customers are the ones willing to trial an unproven prototype in their operations, while larger and better-resourced ones require substantially more assurance before allowing it near their equipment.[266]
Crowdfunding platforms came to require that campaigns show real hardware rather than renderings or simulations, forcing applicants to do the up-front work and spend their own money on a prototype before raising; the rationale is that an appealing product concept costs nothing to render, so requiring photographs of real hardware is what separates projects with engineering behind them from those without.[114] The recognisable failure signature before that rule was a campaign with an aggressive delivery date, a rendered video and no hardware, with requests to see any earlier prototype going unanswered; campaigns that repeatedly describe spinning more boards while never showing an earlier one are describing work that does not exist.[142] Choice of platform can itself signal prototype maturity, since a campaign lacking hardware may run on a platform that does not require it.[156] The obligation defended is that the developer funds the prototype themselves rather than raising money to build the first one, however crude the result.[244] A prototype requirement is only as strong as the standard applied to it: a demonstration whose performance falls far short of the campaign’s claims, on the argument that funding will close the gap, satisfies the letter of the rule and not its purpose.[271]
Historically, some advertised products existed only as prototypes or proof-of-concept models, placed in the market to test for interest on the assumption that a buyer would justify making them work.[540] In custom silicon, treating an early spin as an intentional throwaway is a recognised strategy: Andreas Olofsson built Adapteva’s first chip deliberately as a prototype rather than a product — a “version zero” and an intentional punt — with the explicit purpose of raising enough money to assemble a team, and a later respin was caused by a vendor’s error rather than the designer’s, with the vendor bearing the cost.[254] Cheap access to fabrication moves the money spent from an idea to a prototype, so a customer or investor can evaluate a real design rather than a theoretical one without funding a toolchain first.[503]