| Episodes | 56 |
| Mentions | 63 |
| Cited here | 58 |
| First — last | #16 — #727 |
| Top guests | Jeff Keyzer, Chrissy Meyer, Greg Charvat |
| Related | digi key · bill of materials · component sourcing · fpga · analog to digital converter |
Component obsolescence is the withdrawal from production of an electronic part that an existing design depends on, and the discipline of keeping a product buildable as its components are withdrawn is known as obsolescence engineering.[716] A vendor end-of-life notice is one of the strongest triggers for reopening a finished design, because unlike a cost or performance improvement it removes the option of leaving the design alone.[64] Engineering work therefore does not stop when a product reaches volume production: sustaining effort continues for as long as the factory is building, because components go obsolete and materials become unavailable across the product’s life.[279] Availability is bound up with the working definition of a commodity part, since a jellybean component that can no longer be obtained ceases to function as one regardless of how standard its specification is.[567]
Causes
Semiconductor vendors that change ownership are a longevity risk to designs, because a company under financial pressure typically trims its product portfolio and cancels individual chips that customers have already designed in.[7] Mergers are consequently treated as a leading indicator of obsolescence notices on the combined portfolio, prompting closer watch on the affected parts.[382] Consolidation of legacy fabs drives obsolescence in batches rather than one part at a time: a single last-time-buy announcement covered roughly a hundred devices built in one old National Semiconductor plant, including long-lived audio parts such as the LM386.[271]
Passive manufacturers make the same decisions by category economics rather than per part, assessing projected volume across an entire package and value category and withdrawing the whole category together.[391] More generally, manufacturers decide whether to keep a legacy line running from the shape of its demand distribution: if the largest customer is not a government or an industrial buyer taking on the order of a hundred thousand units a year, the marginal cost of keeping the line usually loses.[613] Technical merit does not protect a device with a small installed base, as with the Intel 196 microcontroller, a capable part that almost nobody used and that became completely obsolete, leaving surviving designs without a supply route.[489] Specifying an obscure variant carries a double penalty on the same logic: an unusual package or the fastest speed grade with only a handful of customers is the first candidate for withdrawal, and during allocation its buyers sit at the back of the queue with no distributor stock to buy through.[384]
A conservative policy at the other extreme is possible. At one large analogue vendor, parts are withdrawn only when production becomes physically impossible, such as a fab shutting down or the test equipment for the part no longer being usable, at which point a last-time buy is offered.[270]
Loss of supply without obsolescence
A part can become unobtainable without ever being declared obsolete, when another buyer absorbs the available distributor stock.[88] Supply can also be lost without any commercial decision at all, through the destruction or shutdown of the sole fab producing a part, as happened to hard-drive production affected by flooding and earthquakes in Asia.[393] Lead time alone can strand a design: multi-layer ceramic capacitors reached roughly 72-week lead times, long enough to make a milestone unachievable if the exposure is discovered only after design-in.[451] A production line going down pushed the lead time on one component from two weeks to thirty, and the documented response was for a customer to buy the entire world supply of it.[425] Free and unrestricted sampling is not evidence that a part can be bought in production quantities; Maxim sampled generously while its chips ran to roughly 40-week lead times.[502] Long-term purchase contracts offer no better guarantee: a memory device specified in a mid-2000s medical design became unavailable in the marketplace despite contracts being in place, leaving replacement as the only option.[488]
A part can also change substantively without changing part number. The LMH6518 variable-gain amplifier used in the front end of many modern oscilloscopes was redesigned so that it no longer reliably supports the 400 mV input offset the original handled, with the change noted only deep in the datasheet, and instruments relying on that capability have been rendered unusable.[727] Silent specification changes between datasheet revisions are a recognised hazard, with vendors treating the errata and the revision letter as sufficient notice to the customer.[322]
Warning signals
A “not recommended for new designs” status is effectively an early obsolescence warning, typically leaving two to five years of remaining availability before the part is withdrawn.[126] A product change notification, or PCN, is the formal vendor mechanism announcing a change or discontinuation of a part, and major distributors forward PCNs and last-buy notices to customers who previously purchased the part, effectively acting as an obsolescence monitoring service.[554] Those notification services cover obsolescence and discontinuation but not errata or specification changes, so datasheet revisions must be tracked by some other means.[322]
Signals can also point the wrong way. A last-time-buy announcement is not always final: a sufficiently large order can cause a manufacturer to reverse the decision and keep the line stocked, which makes end-of-life dates an unreliable planning input in both directions.[351] Manufacturers do not always offer a last-time buy either; a part can simply disappear, leaving the buyer to purchase remaining world stock as the only mitigation.[178] Price signals are similarly ambiguous, since sorting candidate parts by distributor price can select a device whose low price reflects a distributor clearing stock after a line-card change, leaving the design with long lead times or a much higher price from other sources even though the manufacturer has not obsoleted the part.[211]
Design-in practices
Supply confidence is itself a design-in criterion: engineers select vendors partly on the expectation that the device will still exist long enough to avoid a forced redesign within two years of launch.[102] A common rule is to avoid chips from startup vendors unless there is a compelling technical reason, because the probability of the company failing and the part disappearing within five years is high.[106] Because defence programmes require twenty to thirty year supply commitments, identifying which devices are designed into those programmes serves as a longevity heuristic for commercial designers choosing parts expected to stay in production.[588] Because any production line can be shut down, the standard mitigations taught to designers are to prefer high-demand parts, keep firmware portable so a processor change is survivable, and document the design thoroughly.[351]
Sole sourcing concentrates the exposure. A sole-source part leaves a design at the mercy of the vendor’s internal restructuring, since if the manufacturer stops producing that device there is no drop-in alternative and the product cannot be built.[49] A small semiconductor vendor can only justify a sole-source design-in by offering a part so functionally unique that the customer’s specification cannot be met any other way, and without that differentiation the supply risk is not worth taking.[191] The same differentiation becomes a liability when lead time stretches, since no compatible alternative exists and replicating the function may take a dozen discrete devices.[317] Where only one off-the-shelf power supply meets a product’s requirements the design is locked to a single vendor, whereas specifying a custom magnetic component restores portability because the design files can be taken to any transformer manufacturer.[406] Building a product around a commercial single-board computer transfers the whole business’s continuity risk to another organisation’s decision to keep manufacturing that board.[351]
Line-item count matters independently of component count. Reimplementing a function from discrete parts when an integrated single-chip solution exists increases board area, cost and engineering time, and raises obsolescence exposure because every additional component is another part that can be withdrawn.[230] Offloading a function to an external dedicated driver chip can be cheaper than moving up to a larger microcontroller in the same family, but it adds a bill-of-materials line item and therefore one more independent obsolescence and supply exposure.[393] Minimising the number of distinct line items, rather than the total component count, is also a design goal for hand-assembled products because each line item adds kitting and stocking labour.[229]
Selection driven by convenience rather than supply recurs as a failure. Choosing a device because a development board was supplied by a vendor repeatedly surfaces at production time as a twelve-week lead time or an obsolete part.[135] Part selection can also be overtaken by the component lifecycle itself, with a month spent researching parts ending with some of the shortlisted devices already unavailable.[16] Sourcing questions of this kind, such as whether a part can be obtained and whether it is heading for obsolescence, trip up more projects on the electrical side of design for manufacture than layout rule violations or pick-and-place capability limits.[437] A formal bill-of-materials health risk assessment addresses this by screening every line item against component databases and flagging parts that are not recommended for new designs, scheduled for end of life, single-sourced, or carrying long lead times, so those decisions can be corrected before layout.[451]
Verbal assurances are not a substitute. Longevity promises given by a manufacturer at design-in time carry no weight years later, so a designer cannot defend a part choice on the basis that the vendor once said supply was secure.[96] Cheap application-specific chips sold through the Shenzhen market carry no availability guarantee at all, making them unsuitable for a commercial product with a multi-year service life unless supply is secured in advance, and there is no consolidated catalogue or datasheet repository for those custom ASICs, so parts that would be useful in a design are effectively invisible to engineers working from Western distributor catalogues.[317]
Buying ahead
One hedge for a cheap but uncertain part is a lifetime buy sized far above forecast demand, on the order of fifty times the expected build quantity, with the inventory cost absorbed into the product’s cost structure.[317] Under a fixed-quantity supply contract the equivalent move removes the risk entirely: buying the full chip quantity up front at contract signature means the parts needed for every unit are already in inventory.[496] For a small-volume product with a stable design, lifetime buys of discontinued line items and physical storage of that inventory are cheaper than redesigning each time a component goes obsolete.[613] Small-volume and one-off designers can go further and sidestep the risk by selecting only parts currently in distributor stock and ordering components at the same time as the bare boards, an option unavailable to corporate programmes with long production lives.[135]
Monitoring and organisational handling
At companies whose boards carry hundreds to over a thousand components, monitoring component availability is a standing full-time role, because a manufacturing business earns nothing during any period it cannot build product.[290] In large firms the workflow is that a dedicated component obsolescence engineer identifies an unbuyable part and hands it to a design engineer, who then reworks the affected product.[72] Entry-level engineering roles at instrument manufacturers often consist largely of sustaining or maintaining engineering, dealing with obsolete parts in shipping products rather than new design work.[727] Obsolescence work is also a recurring source of engagements for electronics consultants, arriving through contract manufacturers and distributor field application engineers whose clients need an unbuyable part designed out.[409]
Smaller operations rely on periodic checks. Re-quoting the whole bill of materials, on the order of every couple of months, is a practical way to detect parts that can no longer be bought before a build depends on them.[382] Holding the bill of materials inside a distributor’s own system lets a designer reorder in one step and see immediately which line items have gone obsolete or moved to a new version.[106] Guaranteeing a multi-year supply commitment requires an owner inside the company, ideally multiple qualified vendors for every part on the board, and an automated tie-in to vendor end-of-life notification systems so that a last-time buy can be placed in time.[514]
Supply chain information is structurally opaque to design engineers because it sits with a separate team and surfaces weeks or months after the part selection decisions have already been made.[577] When a selected part fails an availability, price or approved-vendor requirement, procurement staff generally cannot choose the alternative themselves and must return to the engineer, forcing a context switch back into a design decision made months earlier.[577] Product lifecycle management tooling closes that loop by notifying on obsolescence or unacceptable lead time, letting the engineer select an alternate, and auto-populating a change order from the CAD integration with approvers and an automatic visual diff of the revisions.[577] Contract manufacturers work against an approved vendor list attached to the bill of materials, which authorises the factory to use named alternative manufacturers and component substitutions, and that list is normally incomplete at production start and filled in over the product’s life.[279] Some turnkey assemblers additionally maintain a house-parts library already loaded on their machines, offered below typical small-volume pricing with placement labour waived, with a guarantee of availability and a direct six-month notice before any house part is discontinued.[243] Crowdfunding and product-launch platforms that manage supply chain for hardware creators, by contrast, typically handle only the cash-flow side, placing and paying purchase orders rather than performing part substitutions as components go obsolete.[314]
Responding to an end-of-life notice
The escalation ladder for handling an obsolete component runs from dropping in a pin-compatible part and recompiling, through relaying out the board, to fitting a mod board such as a small flex PCB when existing built units must be modified without touching the original layout.[496] A functional equivalent frequently comes in a different package, so a substitution that looks like a paper change on the schematic can force a full PCB respin with its associated cost and schedule impact.[577] Within a microcontroller family a single character of the part number can halve memory and peripheral count and change package, so the nearest available family member may demand a layout change, a firmware rewrite and a revised system architecture rather than a simple substitution.[587] Even staying within one family carries cost penalties, since higher pin counts are only available in larger packages that also carry more memory and a larger die, so a design needing pins alone can end up paying roughly three times the cost for memory it does not use.[393] Footprint naming is no substitute for checking dimensions during a substitution: a component whose part name contains 0805 does not fit an 0402 land pattern, an error a contract manufacturer’s design-for-manufacture report will catch.[644]
Qualification effort scales with the function of the part. A passive can be cleared by a paper qualification from the datasheet, whereas swapping something like a DC-DC converter requires a sample build and testing before approval.[279] Recording why each component value was chosen pays off during this work, because the substitution decision is usually made months or years later when the original constraints behind a value have been forgotten.[279] Commodity shortages leave more room to manoeuvre than unique parts do, since values can be substituted or in some cases components depopulated, whereas the loss of a unique controller chip has no equivalent workaround.[393] Shifting an entire product to smaller passive case sizes purely to hedge against shortages is nonetheless a poor trade, because it changes the assembly process technology, lowers yield and may force a change of assembly supplier; package choice should follow the product’s own process requirements.[391] Large passive packages persist despite mobile-driven miniaturisation in any case, because voltage and creepage requirements cannot be met in the smallest case sizes and a mains-rated capacitor cannot be had in an 0201 package.[391]
Once a shortage has forced a redesign, the rational move is to select the best part available at that moment rather than the closest match to the old one, and to secure physical stock of the new part before committing to the change.[587] During a shortage the calculus shifts in favour of redesign generally, because the cost of waiting for an unavailable part exceeds the cost of revising the design.[541] Where the obsolete device defines a plug-in platform that installed modules depend on, substitution would break backwards compatibility, and the remaining route is to commission a manufacturer to copy the existing design and supply a compatible part.[178]
In product lifecycle management systems, obsoleting a component means creating a replacement item, revising the drawing that calls it out, and rolling that revision up through every assembly and higher-level drawing that contains it.[502] Management approval for a disruptive engineering change is far easier to obtain when framed as a supply deadline, with a named obsolete part and a stated number of months of remaining inventory, than as an argument from technical merit.[413]
Economics
Redesign economics scale with annual volume. On a product selling 100 units a year carrying a 2,500 a year, so an engineering change taking more than roughly 25 hours destroys the saving.[64] For a low-volume product that is selling steadily, the practical trigger for reopening the design is therefore a component becoming unavailable rather than any accumulated wish list of improvements.[229] The same logic drives second generations of open hardware projects, where inability to source the original parts is one of the routine forcing functions independent of any desire for new features.[98]
Obsolescence is costlier in modern systems than in discrete-logic eras: replacing a 74-series logic chip meant at worst a logic change, whereas a modern device carries firmware and application software whose changes propagate through the whole system.[126] Obsolescence discovered after certification is more expensive still, because a substitution can invalidate existing regulatory compliance such as FCC approval in addition to stranding fabricated boards and partly populated assemblies.[393] Long-lived products often defer any refresh until the processor becomes unbuyable, at which point the end-of-life notice turns a deferred plan into an emergency.[363] A design left unmaintained for a decade accumulates a large backlog of obsolescence notices, so the absence of sustaining engineering is itself a source of technical debt on long-lived products.[488] A rising count of obsolete components, including basic items such as through-hole resistors, is in turn a practical signal that a long-running low-volume product is approaching the end of its economic life after more than a decade in production.[655]
Obsolescence is not the only sustaining problem that high-volume manufacturing exposes: incoming parts arrive bad, devices drift across their specified ranges, and multiple devices in one package that usually match are not guaranteed to.[328]
Long-life and regulated programmes
Programmes with a contractually mandated ten to twenty year service life bar designers from specifying parts whose vendors may discontinue them or go out of business, regardless of technical merit.[322] Defence system development runs on timescales of thirty to forty years, and much of the work on fielded systems consists of upgrades forced by degradation and by parts that can no longer be sourced.[588] A published multi-year availability promise, such as an eight-year lifetime on a compute module, converts every end-of-life notice into a financial obligation, since if no equivalent part can be engineered in the vendor must fund an eight-year buy of the discontinued device.[514]
Under stringent regulatory regimes such as FDA oversight, medical device designs are deliberately frozen and changed only when a change is forced, which turns most ongoing engineering into obsolescence maintenance.[486] Qualifying a replacement for a single obsolete component in a regulated product can absorb around six months of engineering effort, even for a part as simple as a transistor.[486] Components aimed at certified high-volume systems face a two to three year lag between launch and integration, because an integrator will not requalify a working certified system to save a few hundred dollars on a component, so obsolescence cycles overlap the design-in cycle.[425]
Legacy hardware and the aftermarket
A secondary market exists in which specialist manufacturers keep producing long-obsolete processors for military and aerospace programmes, at order quantities of a handful of units and prices on the order of $10,000 per chip.[20] Where an electronic module for a long-lived product is no longer manufactured, repair depends instead on a stockpiling aftermarket in which a large fraction of hoarded units have already failed from age and surviving units command high prices.[93] One established service model for legacy hardware is to reverse engineer an older board whose transistors and footprints can no longer be purchased and reimplement it with modern components that are currently available.[716]
Documentation is often the binding constraint on such work. For very old parts the surviving material may exist only as scanned PDFs with no CAD source, so reproducing an evaluation board can require toner-transferring the layout image printed in the original application note.[270] Analogue building-block parts used in instrumentation designs have also gone out of production, including the XR2206 function-generator IC used in hobby and educational waveform generators.[179] Large through-hole radial capacitors are among the passive families disappearing from distribution, which raises the maintenance burden on older equipment designed around them.[613] Age rather than use is frequently the failure mechanism in surviving equipment: mechanical step attenuators in vintage RF test gear stop switching reliably because the tiny internal O-rings degrade as all rubber parts do.[613]