Electric Calculator Practical circuit and PCB design guides
English

Ceramic capacitors lose capacitance when you apply voltage

Put 3.3 V on a 0402, 10µF/6.3V part, about half its rating, and only about 3µF is left. The "10µF" on the datasheet is the value with no voltage applied.

Last updated: 2026-10-28 Ceramic capacitorsMLCCDC biasX7R

Why voltage reduces capacitance

X5R and X7R MLCCs use a barium-titanate-based ferroelectric. Under an electric field its polarization approaches saturation and the permittivity drops. The permittivity under a DC bias is lower than at zero volts, and that is the DC bias characteristic.

How much it drops depends on the field strength across each dielectric layer. A smaller case or lower rated voltage means a thinner dielectric, so the same voltage costs more capacitance (Figure 1).

C0G (NP0) uses a paraelectric material, so its shift is essentially zero. Its permittivity is low, though, so for the same case size it reaches only 1/10 to 1/100 of X7R's capacitance.

A ceramic capacitor losing effective capacitance under DC bias
A ceramic capacitor losing effective capacitance under DC bias. AI-generated conceptual illustration; see the article’s diagrams and text for circuit details and conditions.
Dielectric thickness and field strengthSchematic cross-sections of two capacitors at the same 3.3V. The thin-dielectric 1005/6.3V part sees a strong field and loses 60-75% at 3.3V. The thick-dielectric 3216/25V part sees a weak field and loses 5-15%. Same 3.3V, but a thinner dielectric means a stronger field 10µF X5R/X7R. One dielectric layer between two plates (loss values from the table) 1005 / 6.3V Thin dielectric 3.3V Thin Dielectric Strong field Capacitance loss at 3.3V -60 to -75% 3216 / 25V Thick dielectric 3.3V Thick Dielectric Weak field Capacitance loss at 3.3V -5 to -15%
Figure 1: a thinner dielectric means a stronger field.

How much it actually drops

This varies by part, so check the curve. Here is what a 10µF part retains at 3.3 V.

Approximate retained capacitance for 10µF X5R/X7R parts at 3.3 V (varies significantly by part)
Case sizeRated voltageRetained at 3.3VLoss
1005 (0402)6.3V2.5-4µF-60 to -75%
1608 (0603)6.3V4-5µF-50 to -60%
1608 (0603)10V5-6.5µF-35 to -50%
2012 (0805)10V6-7.5µF-25 to -40%
2012 (0805)25V7.5-9µF-10 to -25%
3216 (1206)25V8.5-9.5µF-5 to -15%

A 10µF part can differ threefold in effective capacitance at 3.3 V between a 1005/6.3V and a 3216/25V. If the circuit needs 10µF, pick the part that still measures 10µF at your operating voltage. That may be a 22µF or 47µF part number.

More margin in the rated voltage means a smaller loss. As a guide, use 10V or higher for 3.3V, 16V or higher for 5V, 35-50V for 12V. Even so, 10µF parts rated 10V lose 25-50% at 3.3V.

MLCC DC bias trendX-axis: DC voltage 0-10V. Y-axis: retained capacitance 0-100%. Three curves for 1005/6.3V, 1608/10V, 2012/25V, with points marking 32%, 55%, 85% retained at 3.3V. Same 10µF, but retained capacitance differs 3x by size and rated voltage Model curves showing typical trends. Check the part-specific curve for actual values 0V 2V 4V 6V 8V 10V 0% 25% 50% 75% 100% C remaining DC voltage 1005 / 6.3V 1608 / 10V 2012 / 25V 3.3V 32% 55% 85% Curves use C/C0 = 1 / (1 + (V/Vk)^2), fitted to typical retention at 3.3V.
Figure 2: retained capacitance of 10µF parts (model curve).

Temperature, aging and tolerance stack on top

Three more factors multiply in. Temperature coefficient. X7R holds ±15% over -55 to +125°C; X5R over -55 to +85°C. Y5V runs +22/-82% over -30 to +85°C, so it should not be used for power decoupling.

Aging. Capacitance falls logarithmically: for X7R, 1-2.5% per decade of time (-3 to -7% at 1000 hours). Soldering above the Curie temperature (about 125°C) resets it.

Tolerance. K = ±10%, M = ±20%.

Stacked worst-case for a 1608/6.3V/10µF X5R, grade M part at 3.3V, 85°C, one year in: 10µF × 0.8 × 0.45 × 0.85 × 0.95 ≈ 2.9µF, 30% of the catalog value (Figure 3).

Only tolerance is not on the curves.

Estimating effective capacitanceBar chart: multiplying the 10µF catalog value by tolerance 0.8, DC bias 0.45, temperature 0.85 and aging 0.95 in turn leaves about 2.9µF, 30% of the catalog value. Worst-case stacking turns 10µF into 2.9µF 1608/6.3V/10µF X5R, grade M, used at 3.3V, 85°C, after 1 year 10µF Catalog ×0.8 Tolerance Grade M ×0.45 DC bias 3.3V ×0.85 Temperature 85°C ×0.95 Aging After 1 year ≈ 2.9µF Effective 30% of catalog
Figure 3: effective capacitance with worst-case factors applied in turn.

Where the lost capacitance matters

LDO stability. The output capacitor is part of the phase compensation, and the datasheet minimum means effective capacitance. A 0402/6.3V/1µF part on 3.3V gives an effective 0.5µF and can oscillate.

Switching regulators. Halve the output capacitance and ripple voltage doubles, while crossover frequency rises and phase margin shrinks.

Time constants and filters. RC time constants and LC cutoffs shift. Use C0G for timing, analog filters, and oscillators.

Acoustic noise. Ferroelectrics are also piezoelectric, so an AC voltage makes them vibrate and ring the board ("singing"). It shows up with X7R on a supply with audible ripple or in audio coupling. The fix is C0G, a film capacitor, or an anti-singing part.

What lost capacitance affectsLeft: an LDO with a 1µF minimum gets only 0.5µF effective from a 1µF part at 3.3V output, which falls short. Right: a switching regulator whose output capacitance is halved sees ripple voltage double, crossover frequency rise and phase margin shrink. Two cases where the lost capacitance matters Judge by effective capacitance, not nominal LDO output capacitor Datasheet minimum 1µF Short Nominal 1µF Effective 0.5µF 0402 / 6.3V at 3.3V output Below spec, so it can oscillate Switching regulator output capacitor Output C is halved Ripple voltage doubles Crossover frequency rises Phase margin shrinks Ignoring effective C, the real board can oscillate
Figure 4: effects on an LDO and a switcher.

Selection procedure

When choosing a ceramic capacitor, check the capacitance remaining at the operating voltage, not just its nominal value.

  1. Determine the effective capacitance needed at the operating voltage and maximum temperature, read the retained fraction off the candidate's curve, then apply the temperature coefficient (±15%) and tolerance.
  2. If it falls short, raise the rated voltage, the case size, or the nominal capacitance.
  3. Use C0G for timing, filtering, oscillators, and audio. C0G above 1µF is large and expensive, so consider a film capacitor.
  4. Measure effective capacitance under DC bias with an LCR meter, or confirm it through LDO transient response and ripple.

Frequently asked questions

What fraction of the rated voltage keeps the DC bias effect small?
A third of the rating or less is a common guide, but small, high-capacitance parts can still lose 30-40% at a third, so check the curve.
Should I use X5R or X7R?
X7R guarantees ±15% up to 125°C, X5R up to 85°C. DC bias behavior is similar, and X5R more often offers higher capacitance in the same case.
Are codes like 104 or 106 the effective capacitance?
No, they are the nominal capacitance with no voltage applied (104 = 0.1µF, 106 = 10µF).
Would tantalum or aluminum electrolytic be better?
Tantalum and conductive-polymer aluminum capacitors don't lose capacitance under DC bias and age little, but their ESR is higher (10-50mΩ polymer, hundreds of mΩ wet electrolytic), so they don't suit high-frequency decoupling. A common split is polymer for bulk and MLCC for high frequency.

Standards and references

  • EIA RS-198 / IEC 60384-9 — Definitions of temperature characteristic codes (X7R, X5R, C0G, etc.)
  • MLCC manufacturers' datasheets and characteristic curves — Per-part DC bias curves, temperature characteristics, and aging rates
  • Murata, TDK, and Taiyo Yuden technical notes — The mechanism behind DC bias behavior and mitigating acoustic noise

Related calculators

← All design guides