Buck Input Capacitor: RMS Current, Ripple and Derating

Quick answer

The input capacitor of a buck converter carries a pulsed current, not the DC output current. Its RMS current is roughly I_OUT × √(D(1−D)), where D is the duty cycle, and its ripple voltage is roughly I_OUT × D / (f_SW × C) plus the ESR term. Size the capacitance with the effective value after DC-bias derating, check the ripple-current rating, and place the capacitor so the input loop is as small as possible.

Executive takeaways

At 12 V to 5 V and 3 A, the input capacitor sees about 1.48 A RMS — half the output current, but in a much harsher waveform.
A 22 µF ceramic at 12 V may retain only 55–65% of its nominal capacitance; the ripple you calculate with 22 µF can be nearly double in the real board.
Electrolytic ESR dominates the ripple before capacitance does: 50 mΩ at 3 A is 150 mV before the capacitor even charges.
The capacitor only works if the loop is short: the input loop area sets the high-frequency impedance, not the capacitance value.

1. Why the input capacitor is not a “DC” capacitor

In a buck converter the input current is a series of pulses: during the on-time the switch draws current from the input, and during the off-time it draws almost nothing. The input capacitor supplies the difference, so its current waveform is a square wave with a DC average equal to the output current.

In a 12 V to 5 V converter at 3 A, the input current alternates between roughly 3 A and 0 A at the switching frequency, and the capacitor supplies the difference every cycle.

That has two consequences. First, the capacitor's RMS current can be a large fraction of the output current even though the average is small. Second, the high di/dt makes loop inductance — not just capacitance — part of the problem, which is why placement matters as much as the value.

2. The RMS current formula

For a buck converter in continuous conduction mode, the input capacitor RMS current is well approximated by:

I_RMS ≈ I_OUT × √(D × (1 − D)), with D = V_OUT / V_IN

The factor peaks at D = 0.5, where it equals 0.5. That is the worst case for input ripple current, and it is the number to use when you are not sure of the operating point.

V_IN → V_OUT	Duty D	√(D(1−D))	I_RMS at 3 A
12 V → 5 V	0.417	0.493	1.48 A
12 V → 3.3 V	0.275	0.447	1.34 A
24 V → 5 V	0.208	0.406	1.22 A
5 V → 3.3 V	0.660	0.474	1.42 A

Input capacitor RMS current divided by output current versus duty cycle, peaking at 0.5 — The RMS factor peaks at 0.5 when the duty cycle is 50%; 12 V to 5 V sits at 0.493.

Two practical notes. A capacitor rated for 1 A RMS at 100 kHz may be rated lower at 500 kHz; check the frequency correction in the datasheet before comparing parts. First, this is the capacitor current, not the inductor current; the inductor ripple adds a smaller term. Second, at light load in discontinuous mode the waveform changes, but the worst case for capacitor heating is still near full load and D ≈ 0.5.

3. Ripple voltage: capacitance, then ESR

The ripple across the input capacitor has two parts: the charge/discharge component and the ESR component.

ΔV ≈ I_OUT × D / (f_SW × C) + I_OUT × ESR

Take 12 V to 5 V at 3 A with a 500 kHz switcher and 22 µF of ceramic capacitance. The capacitive term is 3 × 0.417 / (500 k × 22 µ) = 114 mV. With 2 mΩ of ESR the second term is only 6 mV, so the capacitor dominates.

Now replace the ceramic with an aluminum electrolytic of the same 22 µF and 50 mΩ ESR: the ESR term alone is 150 mV, larger than the capacitive term. This is why “same capacitance” is not the same capacitor.

22 µF, 12 V input, 3 A, 500 kHz	ESR	ESR ripple	Capacitive ripple	Total
Ceramic (1210 X7R)	2 mΩ	6 mV	114 mV	120 mV
Polymer	10 mΩ	30 mV	114 mV	144 mV
Aluminum electrolytic	50 mΩ	150 mV	114 mV	264 mV

A second operating point shows how the numbers move. For 24 V to 12 V at 2 A, the duty cycle is 0.5 and the RMS factor is 0.5, so the capacitor sees 1.0 A RMS. With 10 µF effective and 500 kHz, the capacitive ripple is 2 × 0.5 / (500 k × 10 µ) = 200 mV — twice the ripple of the 12 V example with the same capacitance, because the duty cycle is at the worst point and the capacitance is smaller.

4. The value on the label is not the value on the board

Class II ceramics (X5R, X7R) lose capacitance under DC bias. A 22 µF, 25 V, 1210 X5R part may retain only 55–65% of its nominal value at 12 V, and a smaller case loses more. If you calculate with 22 µF and the board delivers 13 µF, the ripple is 114 mV × 22/13 ≈ 193 mV.

Three rules follow:

Use the effective capacitance at your DC bias and temperature, not the catalog value.
Derate for temperature as well: X5R is specified to 85 °C, X7R to 125 °C, and capacitance falls toward the top of the range.
Check the ripple-current rating separately. A small case can have the right capacitance and still overheat at 1.5 A RMS.

5. Ceramic, polymer or electrolytic?

Comparison of ceramic, polymer and aluminum electrolytic capacitors for buck input filtering — Same capacitance, different physics: ESR, DC-bias loss and lifetime decide which one belongs at the input.

Ceramic gives the lowest ESR and the smallest size, but loses capacitance with DC bias and can be piezoelectric — it may sing at the switching frequency and it can couple board vibration into the output. Polymer has stable capacitance and low ESR with a higher cost and a softer failure mode than ceramic. Aluminum electrolytic offers large capacitance in one package, but ESR and lifetime dominate the design; it is often used in parallel with ceramic rather than alone.

A common and defensible pattern is a 10–22 µF ceramic close to the switch plus a larger bulk capacitor (electrolytic or polymer) a few millimetres away. The ceramic handles the high-frequency pulse; the bulk handles the lower-frequency component and the cable inductance.

6. The bulk capacitor is a different job

If the ceramic handles the switching pulse, the bulk capacitor handles the slower component: the input cable inductance, the source impedance and the load steps that the upstream supply cannot follow. The two are not interchangeable, and removing the bulk because “the ceramic is enough” is a common way to create a hot-plug ring or a supply sag.

A practical starting point is a bulk capacitor of at least ten times the ceramic value near the input connector, with the ceramic close to the switch. For a 22 µF ceramic, that means roughly 220 µF of bulk, often an electrolytic or polymer. The bulk does not need low ESR at 500 kHz; it needs enough capacitance and a ripple rating for the lower-frequency current it actually carries.

Check the two together, not separately. A 220 µF electrolytic and a 22 µF ceramic in parallel form a filter with the cable inductance; the damping depends on the ESR of the bulk. Too little ESR and the filter can ring at hot-plug; too much and the ripple rises. The datasheet ESR range is part of the design, not a detail.

7. Source impedance and hot-plug

A bench supply with short leads behaves nothing like a real input: a metre of cable adds roughly 1 µH, and the source may be several metres away. At hot-plug, the input capacitor charges through that inductance, and the current can ring well above the steady-state value.

Two numbers make the point. Charging 22 µF to 12 V stores ½ × 22 µF × 12² = 1.6 mJ; if that energy transfers in 10 µs, the average current is about 13 A. With 1 µH of cable inductance, the resonant frequency is about 34 kHz, and without damping the input can ring to nearly twice the supply voltage.

That is why the input often needs a bulk capacitor plus some damping, and why hot-plug connectors on 12 V rails are specified with inrush limits. The ceramic alone makes the ring faster and sharper, not smaller.

8. When ceramic is the wrong choice

Audio or precision analog nearby. Piezoelectric ringing can show up as noise; a polymer or a different dielectric may be better.
Very high capacitance in a small space. Ceramic loses too much under bias; a polymer or electrolytic is more predictable.
High-voltage inputs. Above roughly 50 V the available ceramic capacitance per case drops sharply and the bias derating becomes severe.
Wide-temperature industrial products. X5R is only rated to 85 °C; check X7R or polymer above that.

9. Layout: the loop is the capacitor

The input capacitor only clamps the input if the loop through it is short. The high-frequency current path is: input capacitor → high-side switch → ground → back to the capacitor. Every millimetre of that loop adds inductance, which shows up as a voltage spike at the switch node and as conducted EMI.

Place the input capacitor on the same side as the switch, with the shortest possible connection to the switch source and the ground plane.
Use multiple vias per pad; a single 0.3 mm via adds roughly 1 nH, and 1 nH at 1 A/ns is 1 V.
Do not route the input current through a narrow trace shared with the output or the feedback network.
Keep the feedback tap away from the input loop; the switching node is the noisiest copper on the board.

10. Measuring input ripple

Bench measurements of input ripple are easy to get wrong, and a wrong measurement leads to a part change that does not fix anything. Use a short ground spring at the capacitor pins, set the scope to 20 MHz bandwidth, and measure differentially if the probe allows it.

Three checks separate a real problem from a measurement artefact. First, the ripple should be measured across the capacitor, not from a distant ground point. Second, the switching edge should be visible as a fast spike on top of the slower ripple; if the “ripple” is all spike, the loop inductance is the problem, not the capacitance. Third, compare the measured value with the calculation: if the measurement is more than 50% higher, suspect the effective capacitance or the probe.

A good practice is to measure at minimum and maximum input voltage, and at light and full load. The worst case in the formula is at D = 0.5, but the worst measured spike often appears at the highest input voltage, where the edge is fastest and the loop is least damped.

11. Five mistakes

1. Using the nominal capacitance. DC bias can remove 30–45% of a ceramic's value before temperature does anything.

2. Ignoring the RMS current rating. A part with the right capacitance can still exceed its ripple-current rating and run hot.

3. Sizing for the wrong operating point. The worst case is near D = 0.5, not at the nominal input voltage.

4. Using one capacitor for everything. A single bulk electrolytic cannot supply the high-frequency pulse; a single small ceramic cannot hold the low-frequency ripple.

5. Placing the capacitor far from the switch. The loop inductance, not the capacitance, then sets the input spike.

12. Selection checklist

Input voltage range and the worst-case duty cycle.
Output current and the resulting input RMS current.
Target input ripple and the resulting capacitance at the effective value.
ESR contribution, especially for electrolytic parts.
DC-bias and temperature derating curves.
Ripple-current rating at the switching frequency and at temperature.
Dielectric class and temperature rating.
Piezoelectric or audio sensitivity of the surrounding circuit.
Loop area, via count and ground return.
Verification: measure input ripple at the pins with a short ground spring, not a long probe.

FAQ

How do I calculate the input capacitor RMS current?

Use I_RMS ≈ I_OUT × √(D(1−D)) with D = V_OUT/V_IN. The factor peaks at 0.5 when D = 0.5, which is the worst case.

Is more input capacitance always better?

It reduces ripple, but it also increases inrush current at hot-plug and can interact with the input filter. Beyond a point, the loop inductance and the source impedance dominate.

Why does my 22 µF capacitor behave like 13 µF?

DC-bias derating. Class II ceramics lose capacitance as the applied voltage rises; the datasheet curve gives the effective value at your operating point.

Can I use an electrolytic instead of ceramic?

Yes, but check ESR first. At 3 A, 50 mΩ of ESR produces 150 mV of ripple before the capacitance contributes anything.

Does the input capacitor need a voltage rating above the input?

Yes, with margin for transients. A 12 V rail usually uses 25 V parts; the derating at 12 V is then part of the effective-capacitance calculation.

Is a 25 V capacitor enough for a 12 V input?

Usually yes, with transients in mind. A 25 V part gives roughly 2× headroom on a 12 V rail, and the higher voltage rating also reduces DC-bias loss compared with a 16 V part of the same case size. Check the actual transient and the derating curve before deciding.

Can I use one large ceramic instead of a ceramic plus bulk?

Sometimes, but check DC bias and hot-plug. A single large ceramic loses more capacitance under bias and makes the hot-plug ring sharper; the bulk capacitor is there for the cable and source, not for the switching frequency.

Why does the input voltage ring when I plug in the board?

The input capacitor charges through the cable inductance. Without damping, that LC forms a resonant circuit and the voltage can overshoot to nearly twice the supply; the fix is bulk capacitance plus damping, not more ceramic.

How do I measure input ripple correctly?

Use a short ground spring at the capacitor pins, not the long ground lead of a probe. A long lead picks up the switching node and shows ringing that is not in the capacitor.

Glossary

Duty cycle (D): The fraction of the switching period the high-side switch is on; approximately V_OUT/V_IN for a buck.
RMS current: The heating-equivalent current of a waveform; what the capacitor's ripple rating refers to.
DC-bias derating: Loss of ceramic capacitance as the applied DC voltage increases.
ESR: Equivalent series resistance; produces ripple proportional to current.
Hot loop: The high-di/dt path through the input capacitor and the switch; its area sets the high-frequency behaviour.