The email purchasing dreads is not the price increase; it is the PDN. A part used for three years is discontinued by a single supplier notice, and two thousand units sit on the line waiting for it. Worse, the notice usually names a "recommended replacement", as though the problem were solved — in reality it merely converts "cannot buy it" into "can we use it".
This article covers two things: what to do in the week the notice arrives, and what a substitute has to demonstrate before it counts as approved.
"Discontinued" is a loose word. The four supplier states have windows and costs that differ by an order of magnitude.
| Status | What the supplier says | Your window | What to do |
|---|---|---|---|
| Active | Normal supply | — | Buffer as usual, watch lead-time swings |
| NRND (not recommended for new designs) | "Still available, but new designs should use XX" | Typically 2–5 years, shrinking annually | Switch new designs immediately, start substitution work on existing ones |
| EOL / PDN (end-of-life notice) | "Last buy date YYYY-MM-DD" | Typically 6–12 months | Calculate the LTN quantity and start substitution in parallel |
| Allocation | "Supply allocated by historical usage" | Uncertain — may recover or turn into EOL | Push for authorisation, chase open-market stock, assess substitutes |
The most dangerous of the four is NRND, precisely because it does not hurt: stock is there, the price has not moved, and nobody convenes a meeting about it. By the time EOL arrives, the validation window has been squeezed into a few months and you are chasing samples while gambling.
Treat NRND as EOL with a longer clock. These are the parts that quietly turn into incidents.
A rough formula for the LTN quantity:
LTN quantity = forecast usage over the next 24 months × 1.15 + service spares (2–3% of shipped volume)
Adjust the multiplier and spares ratio to product life. Industrial and medical products have long lives, so 24 months is clearly too short; consumer products can be cut to 12. Over-buying ends in excess and obsolescence, under-buying ends in a stopped line. Both are expensive, but a stopped line usually costs more.
"Same specification" is a sales phrase, not an engineering conclusion. Fill in at least the following columns item by item, mark the differences, and then judge whether they are acceptable.
| Category | Items that must be checked |
|---|---|
| Electrical | Ratings, tolerance, temperature coefficient, leakage, ESR/ESL, switching time, gate charge |
| Thermal | Maximum junction temperature, Rθja / Rθjc, derating curve |
| Mechanical | Body dimensions, lead pitch and length, coplanarity, land pattern |
| Process | MSL level, peak reflow temperature, solderable finish (Sn / NiAu / OSP) |
| Environmental | Operating temperature range, RoHS / halogen-free status, SVHC content |
Two of these are skipped most often: MSL level and solderable finish. Neither affects electrical performance, so neither design nor purchasing looks at them — yet they directly determine whether you get opens and tombstoning after reflow.
Two SOT-23 parts from different makers can differ by 0.1 mm in lead width, coplanarity and thermal pad size. On a small board that difference becomes bridging or an open joint.
After a material change the following must be redone or re-confirmed — "run it the way we did before" is not an option:
"We tried five and they work" is a power-on test, not a validation.
| Validation | Reference sample size | Criterion |
|---|---|---|
| Solderability | ≥ 10 pcs | Good wetting, no dewetting, no solder shrinkage |
| Post-reflow visual + X-Ray | ≥ 30 pcs | No solder balls, no excess voiding, no shift |
| Function and parameters | ≥ 30 pcs across boundary conditions | All parameters in spec, including voltage and temperature extremes |
| Temperature cycling | ≥ 22 pcs (zero-failure criterion) | −40 to 125 °C or per product spec, 100–500 cycles |
| Damp heat (if applicable) | ≥ 22 pcs | 85 °C / 85% RH, 500–1,000 h |
| System-level validation | Per customer requirement | EMC, safety and environmental tests repeated to the original standard |
These sample sizes are the statistical lower bound for demonstrating zero failures, not arbitrary numbers. Where safety, medical or automotive standards apply, their criteria take precedence.
This is the most counter-intuitive class of problem in validation — every parameter of the new part is better, and the assembly fails.
The conclusion is plain: substitute approval is not a comparison table, it is putting the new part back into the original circuit and recalculating. Anywhere loop stability, timing or impedance matching is involved, treat it as a new design.
Substituting one part is, in essence, a small design change. Handle it as "purchasing switching brands" and the problem surfaces six months into volume; handle it as a change and the pain is concentrated in two weeks, after which it is clean.
