Switching Frequency: What Actually Changes

Switching frequency is the one parameter that touches everything else. Change it and the inductor changes, the capacitors change, the losses change, the thermal design changes and the EMI profile moves.

It is also the parameter most often chosen by accident, because the device came with a default and nobody asked.

Quick answer. Raising the switching frequency shrinks the passive components, roughly in proportion, because each cycle moves less energy. It also raises switching losses, which scale with frequency, and moves the noise spectrum upward. Lower frequency means larger magnetics and better switching efficiency. There is no universally correct answer, because the right choice depends on which constraint is binding on your board.

What gets smaller

The inductance needed for a given ripple current is inversely proportional to frequency. From the buck relationship:

L = VOUT · (1 − D) / (ΔIL · fsw)

Take the running example of this course: 12 V in, 3.3 V out, 2 A load, targeting 30 % ripple, so ΔIL = 0.6 A and D = 0.275.

At 500 kHz:

L = 3.3 × 0.725 / (0.6 × 500 000) = 7.98 µH

At 2 MHz:

L = 3.3 × 0.725 / (0.6 × 2 000 000) = 1.99 µH

Four times the frequency, one quarter the inductance. In practice that is a visibly smaller part, usually a lower-profile one, and often cheaper.

Output capacitance behaves similarly. The ripple contribution from capacitance is ΔIL / (8 · fsw · C), so for the same ripple voltage you need proportionally less capacitance as frequency rises, with the caveat that the ESR contribution does not shrink the same way. How a capacitor's own impedance behaves across frequency is the subject of the bulk and ceramic lesson in the power integrity course.

This is the argument that sells high-frequency converters, and it is real.

What gets worse

Switching losses scale with frequency. Every transition costs energy: the overlap of voltage and current while the switch changes state, the energy in the gate charge, and the energy stored in the switch node's own capacitance. Those are losses per transition, so doubling the frequency roughly doubles them.

Conduction losses do not change with frequency. So as you climb, the loss budget tips from conduction-dominated to switching-dominated, and at some point efficiency falls despite everything else being better.

The thermal problem concentrates. Switching loss happens inside the converter IC or the transistors, not spread across the board. A design that is 2 % less efficient overall may put most of that extra heat in one small package.

Dead time and minimum on-time start to bite. At 500 kHz with D = 0.275, the on-time is around 550 ns. At 2 MHz the period is 500 ns and the same duty cycle gives roughly 137 ns. Controllers have a minimum controllable on-time, and if your required on-time approaches it, the converter starts skipping pulses or behaving irregularly. High input voltage with a low output is exactly the case that gets there first.

The spectrum moves. Higher fundamental means the harmonics that matter land in different parts of the EMC spectrum. This is not automatically worse, and sometimes it is better, but it is never neutral. If you have a radio on the board, the relationship between its band and the converter's harmonics is something to check rather than discover.

The trade-off, laid out

Nothing in that table is a recommendation. It is a list of what moves, and in which direction, so that you can see which row is the one that constrains your board.

How to actually choose

Most of the time the honest answer is that the device chose for you. Many integrated regulators have a fixed internal frequency, and your choice was made when you picked the part. That is fine, provided you know what you accepted.

When it is adjustable, work from the binding constraint:

• Height limited? Go higher and get a thinner inductor.
• Efficiency critical at high load? Go lower, where switching losses

matter less.

• A sensitive radio or analog band on the board? Choose a frequency whose

harmonics avoid it, and check several harmonics rather than the fundamental alone.

• Large step-down ratio? Check the minimum on-time before anything else.

This constraint will decide for you regardless of your preferences.

• Multiple converters on one board? Running them at the same frequency,

or at deliberately spread frequencies, changes whether their noise adds coherently. Some devices support synchronisation for exactly this.

The claim worth refusing

"Higher switching frequency is better" is a sales argument, not an engineering one. It is better for component size and worse for switching loss, and which of those matters is a property of your product rather than of converters in general.

The same goes for the reverse. A 100 kHz converter is not more robust by virtue of being slow. It is bigger, and it has lower switching losses, and whether that is a good trade depends on what you are building.

What comes next

Lesson 5 takes the inductance number produced here and turns it into an actual part, with the four current ratings you have to check and the arithmetic worked in full.

Sources and further reading

• Texas Instruments: Basic Calculation of a Buck Converter's Power Stage (SLVA477) — application report, the inductance and ripple relationships used here.
• Analog Devices: Selecting the Right Inductor Current Ripple — technical article, the effect of ripple choice on size and loss.
• Texas Instruments: LM60430 / LM60440 datasheet — datasheet, minimum on-time and its consequences for high step-down ratios.
• Texas Instruments: Power Supply Design Seminar, loss mechanisms in switching converters (SLUP340) — seminar paper, how switching and conduction losses divide.