RC snubber calculator

Two oscilloscope readings give the loop inductance and the capacitance it is ringing against. From those, TI SLYT465's snubber values — and the overshoot each resistor actually leaves, integrated rather than assumed.

Fig 1 — Switch node after turn-on. Bare it peaks at 22.7 V; with 1.00 nF and 4.77 Ω it peaks at 16.0 V.

Parasitic C from

Ringing frequency (MHz)

Capacitance that halved it (nF)

Input voltage (V)

Switching frequency (kHz)

Snubber C multiple (×)

Resistor to try (Ω)

Parasitic capacitance: 333 pF
Power-loop inductance: 7.60 nH
R = √(L/C): 4.77 Ω → E24 4.70 Ω
Snubber capacitor: 1.00 nF (3× the parasitic)
Dissipation: 86.4 mW — C·V²·f, at any load
Ringing frequency: 50.0 MHz once the capacitor is fitted
Peak node voltage: 16.0 V, against 22.7 V bare

How this is calculated

Standard: TI SLYT465 — Controlling switch-node ringing in synchronous buck converters

$C_{par} = \frac{C_{added}}{3}$: SLYT465: "When the frequency is half the original value, the parallel capacitor is equal to three times the parasitic capacitance." Frequency goes as 1/√C, so halving it needs four times the total — and the part added is three times the part already present.
$L = \frac{1}{(2\pi f_0)^2\,C_{par}}$: The note gives f = 1/(2π√(LC)); with the frequency measured and the capacitance now known, the power-loop inductance is the only unknown left. It is also the number a better layout reduces.
$R = \sqrt{\frac{L}{C_{par}}}$: The ringing circuit’s characteristic impedance, and the value SLYT465 publishes. A resistor matched to it develops both voltage and current, which is what lets it absorb; far above or below, one of the two collapses and the ringing returns.
$P = C_{snub}\,V_{in}^2\,f_{sw}$: The capacitor is charged to the input voltage and discharged every switching cycle. The resistor does not appear: it decides where the heat goes, not how much there is.
$V_{over} \propto I\sqrt{\frac{L}{C}}$: Why layout comes first. Overshoot follows the characteristic impedance, so halving the loop inductance removes 29 % of it — free, and with nothing dissipated.

Assumptions

The ringing frequency is measured, not computed. A MOSFET’s output capacitance is strongly voltage-dependent, so the only trustworthy way in is the scope — and it should be read at the highest input voltage, where the overshoot is worst.
“Critically damped” is SLYT465’s description of R = √(L/C). A snubber is an R and C across the node, which makes the network third order rather than a series RLC, so that phrase is the note’s and not a claim about a damping ratio of one. The peak voltage shown is integrated from the real topology so the resistor can be judged by result.
The simulation carries three elements: loop inductance, parasitic capacitance and the snubber. No reverse recovery, no gate-drive detail, no non-linear Coss. Use it to compare resistor values, not to predict a waveform.
Dissipation assumes the capacitor charges and discharges fully each cycle, which holds when the snubber time constant is short against the switching period — true for every sensible value here.
A snubber is the third remedy, not the first. Shrink the power loop, then consider a boot resistor, then this.

What sets an RC snubber's values

The two parasitics that make a switch node ring belong to your layout, not to a datasheet, so they cannot be looked up. TI'sSLYT465 measures them instead, with an oscilloscope and one capacitor, and this tool is that procedure: note the ringing frequency, add capacitance until it halves, and both the parasitic capacitance and the power-loop inductance fall out.

From those two the snubber follows — the resistor from the ringing circuit's characteristic impedance, the capacitor as a small multiple of the parasitic one, and the dissipation from the energy the capacitor moves each cycle. The peak node voltage is not a formula: the tool integrates the real network so the resistor can be judged by the overshoot it leaves rather than by the label on the equation.

Worked example: halving a 100 MHz switch-node ring

A synchronous buck from 12 V at 600 kHz. The switch node rings at 100 MHz; fitting 1 nF across the low-side FET drops that to 50 MHz.

C_par  = 1 nF / 3                       = 333 pF
L_loop = 1 / ((2 pi x 100 MHz)^2 x 333 pF) = 7.6 nH
R      = sqrt(7.6 nH / 333 pF)          = 4.8 ohm   -> E24 4.7 ohm
C_snub = 3 x 333 pF                     = 1.0 nF
P      = 1.0 nF x 12^2 x 600 kHz        = 86 mW
peak   = 22.7 V bare, 16.0 V snubbed

The halving step is arithmetic rather than a rule of thumb. Frequency goes as one over the square root of capacitance, so halving it needs four times the total; the part added is therefore three times the part already there. The unit tests re-derive that by bisection instead of trusting the three.

Where the snubber model stops being valid

"Critically damped" is SLYT465's word, not a damping ratio of one.A snubber is a resistor in series with a capacitor placed across the node, which makes the circuit third order — not a resistor in series with the loop. R = √(L/C) is the ringing circuit's characteristic impedance and is the value the note publishes; sweeping the resistor here shows it lands close to the empirical minimum, which is the honest justification for it.

The simulation is a model of three elements. Loop inductance, the FET's output capacitance and the snubber. It has no reverse-recovery charge, no gate-drive detail, no non-linear C_oss— and C_oss is strongly voltage-dependent in a real MOSFET, which is why the measured ringing frequency is the input rather than a computed one. Use it to compare resistor values, not to predict a waveform.

Measure at the highest input voltage. The overshoot grows with the step that excites the loop, so a snubber tuned at nominal input is tuned at the wrong point. SLYT465's own table shows it: three remedies all optimised at 12 V, and at 16 V the snubber is no better than the bare circuit.

The loop comes first. Overshoot scales with √(L/C), so halving the power-loop inductance takes 29 % off it, costs nothing and dissipates nothing. A snubber that achieved the same is burning C·V²·f for the life of the product. SLYT465 is explicit that these techniques "can be nullified by poor power-supply layout".

Common RC snubber mistakes

Measuring the ringing frequency with a probe ground lead. A few inches of wire is an antenna at 100 MHz and will show a ring that is partly its own. Tip-and-barrel, or a coax pigtail.
Reading the dissipation as depending on the resistor. It does not: the capacitor is charged to the input voltage and discharged every cycle whatever the resistor is. The resistor decides where the heat goes and how the ring is damped, not how much energy is lost.
Forgetting that the loss is the same at no load. A snubber is proportionally worst at light load, which is where a battery-powered converter lives.
Fitting a snubber when the converter's current limit sits above the inductor's saturation current, or when the input capacitor is a centimetre from the FETs. Both are larger problems than the ringing, and one of them caused it.