Substituting a ceramic cap on long lead time

The board is designed, the BOM is frozen, and one line item, a 10 µF ceramic in a small case, shows no stock and a delivery date that has nothing to do with your schedule. The obvious move is to find another 10 µF part in the same size and order it. That move is how a working board turns into one that resets under load, and nobody notices until the substitute is on hundreds of units.

Multilayer ceramic capacitors (MLCCs) are the part where “same value, same size” hides the most difference. This note covers why, and the checks we run before a swap goes on a build.

Why this shortage is different from a normal one

Trade reports this year describe MLCC supply tightening under demand from AI server power delivery, automotive and energy infrastructure, and they describe it as uneven. The pressure is not on every capacitor. It is heaviest on high-capacitance values in small case sizes, on automotive-qualified series, and on the parts every other buyer wants at the same moment.

That unevenness is useful. A 100 nF decoupling capacitor in 0402 or 0603 is a commodity with many sources. A 10 µF part in 0603 is contested, because it is the densest thing a factory makes in that size. Two boards that look alike on a BOM can sit at opposite ends of the market.

Memory and some microcontroller families are also reported as tight this year. The reasoning below applies to any second-source decision, but a microcontroller swap is a firmware project, not a purchasing one. We cover that separately in our note on microcontroller end-of-life.

What a “10 µF” ceramic actually is

X5R and X7R ceramics are Class II dielectrics. Their capacitance is not fixed. It falls as DC voltage is applied across them, it moves with temperature, and it drifts with time. The number on the reel is measured with little or no bias. Your circuit does not run at zero bias.

Three things set how much capacitance you lose:

• Case size. For the same value and voltage rating, a smaller case has thinner dielectric layers under a stronger field, so it loses more. Moving from 0603 to 0805 at the same value can help a lot.
• Voltage headroom. A part run near its rated voltage loses more than the same part at a fraction of it. A higher-rated part in the same case generally holds its capacitance better, because its dielectric layers are thicker. It also usually comes in lower values.
• Manufacturer and series. The same nominal part from different vendors can show very different curves at the same bias. The label does not tell you which one you bought.

A 10 µF, 6.3 V part on a 5 V rail can deliver a fraction of 10 µF. Vendors publish the curves. Murata’s SimSurfing tool, for one, plots capacitance against DC bias for a specific part number, and other makers publish comparable data. Read the curve at your rail voltage. That number, not the reel label, is the capacitance your design has.

This is the whole risk of a swap. A substitute matches the label and misses the curve.

The checks before a swap

Run these in order. Stop at the first one that fails.

1. Find the requirement. What effective capacitance does the circuit need at its operating voltage and temperature? For a regulator or converter, the datasheet says. If the schematic only says “10 µF”, work it back from the original part’s curve and treat that as the minimum.
2. Read the curve for the candidate. Take the candidate’s effective capacitance at the real bias. If it is below the requirement, it fails, whatever the label says.
3. Match the dielectric where it matters. C0G (NP0) capacitors are stable with voltage and temperature. X7R and X5R are not. Swapping a C0G part for an X7R part in a filter corner, an oscillator load or a timing network moves the circuit. Swapping X5R for X7R, or a smaller-case part for a larger one, is usually fine.
4. Check the voltage rating. A higher rating is normally safe. A lower one is not, and a rating with no margin over the rail costs capacitance at the operating point.
5. Check the physical fit. Height, pad geometry and neighbors. If you go up a case size to buy DC bias margin, confirm the footprint and keep-outs allow it.
6. Check board flex. Larger ceramic bodies crack more readily when the board bends. Near connectors, mounting holes, screw bosses and depanelization lines, prefer a soft-termination (flexible-termination) part, which adds a compliant layer to the end termination. A cracked MLCC can fail as a short, which is a worse failure than a dead board.
7. Check qualification. Automotive qualification belongs to a specific series from a specific maker. It does not carry over to a substitute from another vendor.

Where the capacitor sits decides how much risk you take

Position	How forgiving a swap is
Bulk decoupling, away from flex points	Most forgiving. Confirm effective capacitance, then swap.
High-frequency bypass at an IC pin	Forgiving. Keep the same small case and dielectric.
Regulator or converter output	Not forgiving. The datasheet sets minimum capacitance, and some older LDOs also set an ESR window.
Timing, filters, crystal load	Not forgiving. Needs C0G or the same class, then a measurement.
Safety isolation, RF matching	Do not swap on parameters alone. Requalify.

The regulator output is where we see the most trouble. A control loop is compensated for a range of output capacitance. Take a tantalum or a large ceramic out and put in a lower-ESR ceramic that loses half its value at bias, and the loop can ring or oscillate. It may pass a bench check at room temperature and fail cold. Too much capacitance also causes trouble, through startup inrush and overshoot. Stay inside the window the datasheet gives.

Paralleling two smaller ceramics to make up the value is legitimate. It lowers ESR and adds to effective capacitance, but it takes board area, adds placements, and different values in parallel can produce an anti-resonance peak in the impedance. Check it in a simulation or on the bench, not by eye.

Do it in the design, not on the assembly line

The cheapest substitution is one decided before the board is laid out.

• Specify by parameter. In the schematic and BOM, write the capacitance, effective capacitance at bias, voltage, dielectric and minimum rating for termination, not only a single manufacturer part number. Whoever sources the part then has something to check against.
• List approved alternates by part number. Each one has had its curve read and its position checked. “Equivalent” written on a BOM is a decision handed to someone who cannot see the circuit — the same discipline we use when qualifying a second source for a single-sourced flash part.
• Design footprints that accept two sizes where the board has room, so a case-size change is a BOM edit and not a respin.
• Prefer the plentiful part. Where a 4.7 µF at a higher voltage rating or a larger case meets the requirement, it is the better design as well as the more available one.
• Send the BOM to your assembler early, and require that any substitution is approved by an engineer who can see the circuit before it is placed. A silent substitute on a first article is the hardest kind of fault to trace.

After any swap, test the rails that changed: ripple, load step, startup and at least a cold and a warm run. A substituted regulator capacitor is a changed design, and it needs the same bring-up as a new one.

Sometimes nothing fits. The part is short at every source and no alternative meets the curve. That is a redesign question, and it is the same one an end-of-life notice raises: change the circuit so it needs less capacitance, changes value or moves to a different regulator, and then verify it. Better to find that out in the design review than after the boards are built.

How we work on this

We treat a capacitor substitution as a small design change: read the curves, check the position, note the alternates, and test the rails afterward. If you have a stalled build, send the BOM, the schematic and the reference designators that cannot be sourced, and we will tell you which are safe to swap and which need a change to the circuit.