Obsolescence is a schedule problem wearing a purchasing costume
When a part goes end-of-life, the purchasing team is handed a problem that was created three to five years earlier, in a schematic review that nobody remembers. The question that matters is not “can we still buy it” — it is “how many months of production does our remaining stock cover, and what happens in month thirteen”. By the time a distributor sends an end-of-life notice, the design has no second source and no drop-in alternate, so the schedule is exposed to a single reel.
That is why we treat lifecycle status as a design input, alongside pitch, voltage and temperature. A design review that signs off a bill of materials without recording the lifecycle status of every active device has deferred a schedule risk to the production phase, where it costs the most to remove.
The second thing to accept is that most of the risk is not in the exotic parts. It is in the cheap, boring ones — a specific 0.1 µF 0402 in a particular dielectric, a 16 V regulator in a package the board was routed around, a connector whose mating half was discontinued quietly. Every one of those is replaceable in principle and expensive in practice, because replacing it means re-laying a board that is already in production.
How to read lifecycle status and change notices
Lifecycle codes are not standardised across manufacturers, which is precisely why they get misread. The four you must distinguish:
- Active. In production, no notice issued. Still check the second-source column, because active today is not active in five years.
- Not recommended for new designs. The part is still made and still sold, but the manufacturer is signalling a future end-of-life. Designing it in today means scheduling its replacement in two to four years.
- Last-time buy. A final order date or a final ship date has been published. This is the decision point, and it is a hard date, not a suggestion.
- Obsolete. No longer produced. Everything from here is brokered stock, and brokered stock is where counterfeits enter a supply chain.
Change notices are the early warning system. JESD46 is the JEDEC framework for product change notification, and it defines the categories: a process change, a fabrication site change, a material change, a package change, a datasheet change. Not every PCN is an obsolescence signal, but three of them are effectively a design change on your board: a change of die fab or assembly site, a change of termination finish, and a change of dielectric or construction inside a capacitor. Those can alter solderability, moisture sensitivity and derating behaviour without any change to the part number.
The notice window is usually shorter than the design cycle. A 90-day notice is common; a 30-day notice happens. Anything requiring a board respin, a new stencil and a first-article report cannot be absorbed inside 30 days, which is why the second-source footprint decision — covered below — matters more than reacting fast.
Any PCN that names a different assembly site, a different termination finish, or a different dielectric material gets reviewed as a design change, not as a purchasing substitution — even when the manufacturer says the form, fit and function are unchanged. Form and fit are visible; the function that shifts is the one that fails at 1,000 temperature cycles.
Last-time-buy arithmetic: quantity, storage and the cost of carrying it
A last-time buy is a small logistics problem with a large capital consequence. The quantity calculation itself is straightforward; what goes wrong is that teams buy to the annual usage and forget everything else. Work through a real example.
Take a mixed-signal device at 12,000 pieces a year on a programme with six years of remaining life. Gross remaining demand is 72,000 pieces. Attrition — assembly fallout, rework, field spares and the damaged reel nobody planned for — runs at 3 %, which brings the requirement to 74,160 pieces. Reels come in 2,000-piece multiples, so the order is 76,000 pieces. If the part is moisture-sensitive, multiply the storage problem by its floor life, not by its price.
| Step | Basis | Quantity |
|---|---|---|
| Gross remaining demand | 12,000 pcs/year × 6 years | 72,000 pcs |
| Attrition allowance | +3 % for fallout, rework and spares | 74,160 pcs |
| Reel multiple round-up | Nearest 2,000-piece reel above | 76,000 pcs |
| Carrying cost on a $1.00 part | 76,000 × $1.00 × 15 % per year × 3 years average holding | $34,200 |
Two things fall out of that table. First, the carrying cost is a real budget line: on a nominal $1.00 part, holding the buy for an average of three years costs about a third of the material value again, before storage, insurance and the labour to re-bake stock. Second, the quantity is a programme-level decision, not a component-level one. If the programme is likely to end two years early, buying for six years converts a part shortage into an inventory write-off. If the programme extends, the last-time buy is short and there is no second chance.
Storage life is the constraint that kills naive buys. Moisture sensitivity is governed by J-STD-033: an MSL 3 part has a floor life of 168 hours out of the bag, MSL 4 is 72 hours, MSL 5 is 48 hours and MSL 5a is 24 hours. Sealed in a moisture-barrier bag with a humidity indicator card, the shelf life is 12 months from the seal date; every time the bag is opened the clock restarts only after a documented bake. A 125 °C bake of 24 to 48 hours, depending on body thickness and MSL, restores the parts to a dry state — and it also consumes handling cycles, consumes floor space and consumes the solderability margin of the terminations.
Termination finish sets the outer limit. Most distributors guarantee solderability for 12 months from the date code, and some manufacturers will not warrant it beyond two years. A pure tin finish carries a whisker risk that storage does not reduce; an anneal of 150 °C for one hour mitigates it, and a nickel-palladium-gold finish avoids it at a price premium. Where a programme genuinely needs a decade of stock, the right answer is usually to redesign around a part with a NiPdAu finish rather than to buy ten years of a matte-tin part and hope.
Alternates and cross-references: what must be re-qualified
Cross-reference tables are useful for finding candidates and useless for qualifying them. A pin-compatible alternate with a similar datasheet is a starting point for a paper review, and nothing more. What has to be re-qualified depends on what changed, and the changes sort into tiers.
- Q1Same die, different assembly siteSolderability, moisture sensitivity and thermal resistance may all move. Re-qualify with a sample build, X-ray on the critical joints and a first-article report.Sample build
- Q2Different manufacturer, same functionFull electrical characterisation at the temperature extremes the product sees, plus EMC and timing checks. Pin compatibility does not imply tolerance compatibility.Full char.
- Q3Passive with a changed dielectricAn X5R to X7R swap moves the capacitance over temperature and the DC bias derating curve. Re-check the value at the operating voltage, not at 1 kHz and 0 V.Derating
- Q4Package or footprint changeNothing about this is a substitution. It is a layout change, a new stencil, a new panel, a new first article and — in a regulated programme — a change control submission.Respin
In regulated work the paperwork is the long pole. A medical programme changes under ISO 13485:2016 change control and may need a fresh verification record; an automotive programme runs the change through IATF 16949:2016 with a PPAP submission; an aerospace programme adds AS9100D traceability and, where stock comes from outside the authorised channel, an AS6081-style counterfeit inspection. Budget the qualification at 8 to 12 working days for a sample build with first-article data, and considerably longer if the customer's own change board has to convene.
The sourcing desk can shorten this. We hold 4,000 unique part numbers on the shelf in 12,000 sq ft of ESD-controlled store across two plants, so a candidate alternate is often already in stock and can be built into a sample board without a distributor lead time. What we will not do is silently substitute: every alternate is proposed with its change tier, and the customer decides.
Design changes that buy years
Obsolescence is cheapest to design out at layout, and four decisions do most of the work.
- Decision one
Put two footprints where one would fit
A dual footprint costs a few square millimetres and buys an escape route. Lay out the regulator for both the 3 × 3 mm DFN and the SOT-23-5 alternative; lay the crystal for two package sizes; put the connector on 2.54 mm pitch even if the current part is 2.00 mm. This is the single highest-value obsolescence measure available at layout.
- Decision two
Standardise on package families, not part numbers
Choose 0402 or 0603 rather than mixing three chip sizes, SOIC-8 and TSSOP-16 rather than a bespoke package, and one logic family across the board. Generic package families have many manufacturers and a long tail of life; a bespoke package has one.
- Decision three
Genericise the passives on the bill of materials
Specify 0.1 µF, 0402, X7R, 16 V, ±10 % — and then approve three manufacturers against it rather than releasing one manufacturer's part number. The same applies to resistors and to the common inductor values. This removes the majority of line items that would otherwise need a cross-reference.
- Decision four
Record a lifecycle status and a second source per active device
A bill of materials with a lifecycle column and a named alternate for every active device is a design deliverable, not a purchasing spreadsheet. If the column is empty at design review, the review is not finished. Automotive parts additionally need AEC-Q200 or AEC-Q100 qualification to be a real alternate.
None of this is free. A dual footprint consumes board area, and on a 20 × 20 mm minimum board area is the scarcest resource there is. The trade is deliberate: spend 5 mm² at layout, or spend a respin later.
What we do when a part disappears mid-programme
It happens on running programmes, not only at design review. The response is a sequence, and the first step is always the same: quantify the runway before touching the design.
- Freeze and count. How many pieces are on the shelf, on order, and in transit? Divided by the monthly build rate, that is the runway in months. A four-month runway and a six-week cross-reference cycle is survivable; a six-week runway is not.
- Search the authorised channel first. Manufacturer stock, franchised distributors, and our own 4,000 part numbers. Anything outside the authorised channel is brokered stock and gets AS6081-style inspection — visual, X-ray, X-ray fluorescence on the terminations, and electrical test on a sample — before it is allowed into a build.
- Propose a tiered alternate. Come back with the candidate, its change tier from the table above, and the sample-build cost. Do not come back with a substitution already made.
- Buy time with a partial last-time buy. A buy covering 12 to 18 months is often enough to qualify an alternate properly, and it caps the capital at risk if the programme changes direction.
- Respin if the tier says respin. Package and footprint changes go through layout, a new stencil in four hours, a new panel, a first article with X-ray and cross-section, and a pilot run 15 working days after the DFM sign-off.
A checklist for the next design review
Put these eight questions to the bill of materials before the layout is released. Each one has cost a programme somewhere.
- Does every active device have a recorded lifecycle status, and a date it was checked?
- Does every active device have a named second source, even if the footprint is single?
- Are the passives specified by value and family rather than by one manufacturer's part number?
- Is there a dual footprint anywhere it is cheap — regulator, crystal, connector?
- What is the moisture sensitivity level of each moisture-sensitive part, and is the floor life longer than one shift?
- Is the termination finish pure tin, NiPdAu or something else, and is whisker risk managed?
- What is the planned production life of the product, in years, and does any single part have a shorter life than that?
- If the worst part disappeared tomorrow, what is the runway in months and what would the respin cost?
Takeaways
- Obsolescence is a schedule risk created at design review and paid for in production.
- A last-time buy is arithmetic: demand × remaining years, plus 3 % attrition, rounded to the reel multiple.
- For 12,000 pcs/year over six years, that is 76,000 pcs after attrition and rounding.
- Carrying a buy for an average of three years costs about 15 % of material value per year — roughly a third again on a $1.00 part.
- J-STD-033 floor life is the real storage limit: 168 hours at MSL 3, 72 hours at MSL 4, 24 hours at MSL 5a.
- Solderability is typically guaranteed 12 months from the date code; buy for a decade only with a NiPdAu finish.
- Re-qualification tier follows what changed, not how similar the datasheets look.
- Dual footprints and generic passives are the two cheapest design measures available.
- Brokered stock is never a shortcut: it enters the build only after AS6081-style inspection.