A buyer sent me a photo last week: two reels of MLCCs, same case size, same 10 µF marking, same voltage rating. One came off the approved vendor list. The other came from a broker. “Same spec,” he wrote. “Can we just use it?”
You cannot tell from the label. Between those two reels sit nine checks — dielectric class, capacitance at operating voltage, ESR in the circuit it actually sits in, termination type, qualification level — and a part that fails any one of them will pass every incoming inspection you run, then fail in the field a year and a half later. In a year when high-capacitance and automotive-grade MLCCs are quoted at 20 to 40 weeks, that gap between “same spec” and “same part” is where the expensive mistakes are being made.
2026 is not one market. It is three.
Lead-time reporting through the third quarter puts high-capacitance MLCCs for AI server applications at 20 to 40 weeks, with some series closed to new orders. Automotive-grade parts qualified to AEC-Q200 sit at 16 to 28 weeks. Ordinary small-case, low-capacitance parts — the 100 nF decoupling capacitors that make up most of a BOM’s line count — are still quoted at 2 to 6 weeks, with stock on the shelf.
That split changes what a substitution decision is. If your part is in the first two groups, someone will offer you an alternative, and you will have to judge it. If it is in the third, an “alternative” offered at a premium is more likely a warning sign than an opportunity.
The second-order effect is the one engineers feel. Through 2026, buyers locked capacity three to six months ahead. The BOM approved at design freeze and the parts available at ramp are no longer the same list. Substitution has moved from a once-per-program exception to a recurring step in the release process, and most teams never built a process for it.
Two capacitors can share a case size, a nominal value and a voltage rating, and still behave as different components. The reasons are physical.
Capacitance is a curve, not a number. Class II ceramics — X5R, X7R, X6S and their relatives — lose capacitance as DC voltage rises across them. On a small case size at high rated voltage, the loss can exceed half the nominal value. It is not a defect; it is the material. Class I ceramics, C0G and NP0, do not behave this way, which is why they cost more and why they are used where the value has to survive the circuit — filters, timing, matching networks.
The dielectric code is a temperature promise, and a limited one. X5R is ±15% from -55 °C to +85 °C. X7R is ±15% to +125 °C. X6S is ±22% to +105 °C. Moving from X7R to X5R changes nothing on the label and everything about the temperature at which the part stops being the part you designed.
ESR is a design parameter. Lower is not automatically better. On a switching rail, low ESR is what you want. On the output of a regulator specified around a tantalum capacitor, it is not: many older LDOs need the ESR to sit inside a stability window, and a ceramic with an ESR a hundred times lower can turn a stable loop into an oscillator. This is the failure that shows up on the bench in five minutes, which makes it the cheap one.
The package is mechanical, not just dimensional. Going from 0402 to 0603 changes solder-joint geometry and the stress that board flex puts on the termination. On boards that bend — automotive, handheld, anything with a connector nearby — a soft-termination part and a standard one are not interchangeable even when the electricals match.
Crystals have their own version of the same problem. The frequency on the label is specified at a load capacitance. Change the load capacitance and the frequency moves; on a typical 12 pF design, an 8 pF part can pull it by tens of ppm. Nothing on the label changes. The UART works on the bench at room temperature and starts dropping characters in a cold truck.
1. Dielectric class. Compare the class, not the capacitance. A C0G and an X7R part can carry identical values and footprints. Red flag: any move down the ladder — C0G to X7R, X7R to X5R, anything to Y5V — in a circuit where the capacitance is doing work rather than just sitting there.
2. Capacitance at your operating voltage. Read the DC-bias curve at the voltage the part actually sees, not the rated voltage on the label. If the datasheet does not publish the curve, the part is not qualified for a substitution decision. Red flag: a smaller case size or a thinner dielectric offered as “the same 10 µF”.
3. Temperature range and drift. Confirm the upper limit against the hottest point on your board, not the ambient in the room. Red flag: a -55 to +85 °C part proposed for a design that already runs at 95 °C inside the enclosure.
4. Package, pad and termination. Check the land pattern, the thickness and the termination style, not only the length and width. Red flag: a thickness change that alters reflow stand-off on a board that flexes.
5. ESR and ESL in your circuit. Ask what the part is doing: bulk decoupling, high-frequency decoupling, or loop compensation. The requirement is different in each case. Red flag: a low-ESR ceramic dropped in as the output capacitor of a regulator that was characterised with a higher-ESR part.
6. Qualification level, and the paper behind it. AEC-Q200 is a test suite, not a sticker; the grade and the supporting PPAP or PSW documents are what you actually rely on. Red flag: “automotive grade” offered with no qualification report, or an industrial part described as “equivalent to automotive”.
7. For crystals: load capacitance, tolerance, drive level, negative resistance. Four numbers, not one. The tolerance at 25 °C, the stability across temperature and the aging rate are three separate datasheet lines, and the frequency you need is the sum. Then check that the oscillator circuit has enough negative-resistance margin — three to five times the crystal’s maximum ESR is the usual target — or the part may not start reliably at temperature extremes. Red flag: a substitution argued on frequency alone.
8. Batch consistency and change history. Ask whether the manufacturer has issued a product change notification for this part in the last two years, and what it changed. Red flag: no PCN tracking at all, from either side of the table.
9. Lifecycle, and what kind of second source it is. A second source can mean a second production line at the same manufacturer, a licensed alternate, or a different brand that happens to meet the same outline. Those are three different risk profiles. Red flag: replacing a part that is nearing end-of-life with another part that is nearing end-of-life.
Six cases that come up often. This table is meant to be argued with — that is the point of it.
| # | Change | Verdict | Why |
|---|---|---|---|
| 1 | C0G 100 pF 50 V → another brand’s C0G 100 pF 50 V | Green | Same class, no DC-bias loss; check ESR/ESL only if the part sits in an RF path |
| 2 | X5R 10 µF 6.3 V 0603 → X6S 10 µF 6.3 V 0603 | Amber | Higher temperature ceiling, but a different DC-bias curve; re-measure at the rail voltage |
| 3 | X7R 100 nF 50 V → X5R 100 nF 50 V | Red | Upper limit falls from 125 °C to 85 °C; the label does not change, the design margin does |
| 4 | 12 pF load crystal → 8 pF load crystal, same frequency | Red | Frequency pulls by tens of ppm; nothing else in the design compensates for it |
| 5 | Tantalum output cap on an older LDO → ceramic of the same value | Amber → Red | ESR falls outside the regulator’s stability window; check the datasheet’s ESR range first |
| 6 | Industrial-grade part → “automotive equivalent”, no PPAP | Red | Qualification is a document trail, not a claim; without it there is nothing to audit |
The verdicts are not the deliverable. The habit is: every substitution gets a row, a reason and a name next to it.
Approval is not the end of the process, it is the middle of it.
Take 30 to 50 pieces from the new batch and measure them at the conditions the circuit runs at. Compare the distribution, not the average — a batch with the same mean and twice the spread will pass a spot check and fail in production. Keep the measured set as a reference, and measure the next incoming batch against it; that is how a slow drift in dielectric or termination gets caught before it becomes a field return.
Then run the boundaries: minimum and maximum supply voltage, both ends of the temperature range, and the load step the circuit actually sees. If the part sits in a timing or RF path, check it at temperature, not at room temperature, because that is where the pull shows up. Confirm the moisture sensitivity level and the reflow profile match what your line runs, and write the decision down — part number, batch, measurements, who approved it. Six months later, that record is the only thing that will tell you why a board from a particular week behaves differently.
Some substitutions should not happen at any price.
If the part sits in a safety chain, in a certified module, or in a medical or automotive path that has already been through qualification, the cost of re-certifying the assembly is almost always larger than whatever the alternative saves. If the only available source is a broker lot with no traceable chain, the comparison is no longer between two components — it is between a component and a risk. And if the “second source” is a part that is itself approaching end-of-life, you have moved the problem, not solved it.
The honest answer to “can we just use it” is sometimes no, and saying so is cheaper than being wrong later.
Send us the manufacturer part number, the rail voltage, the ambient temperature and one sentence on what the part does in the circuit. We will put the two datasheets side by side and tell you where they diverge — including the cases where the answer is that they do not.
Lead-time ranges cited above reflect industry distribution reporting for the third quarter of 2026 and are indicative, not a quotation.
ODK (HK) Electronics Technology Co., Limited is a Hong Kong-based distributor of electronic components, supplying integrated circuits, passive components and electromechanical parts to OEMs, contract manufacturers and design houses. The company supports sourcing for allocated, long-lead-time and end-of-life parts, and works with engineering teams on part-number-level substitution review.
Cover photo: Filip Wahlberg / Wikimedia Commons, CC BY-SA 4.0 — cropped and colour-adjusted.