Choosing the Output Capacitor

Output capacitor selection gets reduced to a number more often than any other part of a switching design. The datasheet suggests 22 µF, somebody fits 22 µF, and the question is considered closed.

Three things make that unsafe. The capacitance you fitted is not the capacitance you have. Ripple is not the only requirement. And on many regulators the output capacitor is part of the control loop, so changing it changes stability.

Quick answer. The output capacitor has three jobs: keep ripple within budget, supply the load during a transient before the loop responds, and satisfy whatever the controller requires for stability. Ripple comes from both capacitance and ESR. Ceramics lose much of their value under DC bias, so the effective capacitance is what counts. Some controllers need a minimum ESR, and most specify a range rather than a single value.

Ripple has two sources

The inductor current ripple has to go somewhere. The load takes the average, and the capacitor absorbs the alternating part. That produces ripple voltage by two mechanisms that add.

From the capacitance. Charge flows in and out each cycle, and a finite capacitance means a finite voltage change:

ΔV_C = ΔIL / (8 · fsw · C)

From the ESR. The same ripple current flowing through the capacitor's equivalent series resistance produces a voltage directly:

ΔV_ESR = ΔIL × ESR

Using the running example, where lesson 5 produced ΔIL = 0.5835 A at 500 kHz, and taking a 22 µF ceramic whose effective value at the operating voltage is 12 µF:

ΔV_C = 0.5835 / (8 × 500 000 × 12 × 10⁻⁶) ΔV_C = 0.5835 / 48 ΔV_C = 12.2 mV

With a low-ESR ceramic at 3 mΩ:

ΔV_ESR = 0.5835 × 0.003 = 1.75 mV

So the capacitive term dominates and total ripple is around 14 mV. Now repeat with an electrolytic at 30 mΩ:

ΔV_ESR = 0.5835 × 0.030 = 17.5 mV

The ESR term now exceeds the capacitive one. Same capacitance, same ripple current, an order of magnitude more ripple voltage from one parameter.

The practical rule that falls out: with ceramics you are usually capacitance-limited, and with electrolytics you are usually ESR-limited. Which term dominates tells you which parameter to spend money on.

The capacitance you did not get

A 22 µF X5R ceramic in a small package may measure close to 22 µF with no voltage on it, and well under half that at its rated voltage. The dielectric loses permittivity as the DC field across it rises.

This is why the arithmetic above used 12 µF for a 22 µF part. The exact derating depends on dielectric, package size and applied voltage, and the only reliable source is the vendor's curve or tool for that specific part.

Three habits that follow:

• Design with the effective capacitance at your operating voltage, not

the printed value.

• Prefer a larger package or a higher voltage rating when you need the

capacitance to survive the bias. A 25 V part at 3.3 V keeps much more of its value than a 6.3 V part at the same voltage.

• Remember that X5R and X7R also drift with temperature and age. If a margin

matters, it needs to survive all three effects.

The power integrity course works through the same derating problem when choosing decoupling values.

Transient response is the harder requirement

Ripple is steady-state and usually easy to satisfy. The demanding case is a load step.

When the load suddenly demands more current, the inductor cannot change its current instantly. For the time it takes the control loop to respond and the inductor current to ramp, the extra charge comes from the output capacitor alone. The output voltage dips by roughly the charge removed divided by the capacitance, plus an immediate step equal to the current change times the ESR.

That ESR step is instantaneous and no amount of capacitance removes it. For a 1 A load step into a 30 mΩ capacitor it is 30 mV, appearing as fast as the load changes.

So a transient specification usually drives capacitor choice harder than a ripple specification does, and it drives it towards low ESR and adequate capacitance together, which is why several smaller ceramics in parallel often beat one larger part.

Where the controller gets a vote

On many regulators the output capacitor is inside the feedback loop, and its value and ESR help set the loop's response. This has two consequences that are easy to miss.

More capacitance is not automatically more stable. Adding output capacitance moves the output filter pole down in frequency, which changes phase at crossover. On some designs that improves things; on others it reduces phase margin and makes the response ring. Lesson 10 explains the mechanism.

Some controllers require a minimum ESR. Older voltage-mode and some current-mode designs rely on the ESR zero for stability, and replacing an electrolytic with an all-ceramic bank removes the zero and destabilises the loop. The failure appears as oscillation that was not there with the original part.

Both are reasons to read the controller's own requirements before choosing, and to treat "the datasheet says 22 µF to 47 µF" as the real constraint rather than a suggestion. Where a range is given, it is a range because outside it the device's compensation no longer matches.

A selection order that works

1. Read the controller's stated requirement: type, range, and any ESR

constraint.

1. Compute the ripple from both mechanisms at your ripple current, using

effective capacitance.

1. Check the transient requirement, including the instantaneous ESR step.
2. Derate for DC bias, temperature and ageing.
3. Choose the arrangement, usually several parts in parallel, that meets all

of the above with the lowest total inductance to the load.

Step 1 first, because it can invalidate everything downstream.

What comes next

Lesson 8 turns to the node that decides whether all of this regulation actually works: the feedback divider, which is quiet, high impedance, and sitting next to the noisiest node on the board.

Sources and further reading

• Texas Instruments: Basic Calculation of a Buck Converter's Power Stage (SLVA477) — application report, output ripple from capacitance and ESR.
• Texas Instruments: LM60430 / LM60440 datasheet — datasheet, output capacitor range and its relationship to loop compensation.
• Analog Devices: Buck Power Stage Design Equations — application note, output capacitor behaviour in the power stage.
• Texas Instruments: Control Loop Cookbook (SLUP113) — seminar paper, how the output filter interacts with the control loop.
• Murata: DC bias characteristics of ceramic capacitors — manufacturer FAQ, why effective capacitance falls with applied DC voltage.
• onsemi: LC Selection Guide for the DC-DC Synchronous Buck Converter (AND9135) — application note, output capacitor selection against ripple and transient.