Every power design starts with a loss number, and most loss numbers are wrong in the same way: they count the obvious term and miss the one that grows with frequency. A diode in a power circuit loses power in three distinct ways — conduction, switching, and recovery — and each term maps to a datasheet number you already have. The one-page equation sheet that follows is the tool selection and thermal articles across this site expect you to hand them, and two worked examples show it closing on real circuits: an SMPS output rectifier and a PFC input stage.
The Loss Sheet: Three Terms, Three Datasheet Numbers
The total loss of a diode in a switching power circuit is the sum of three terms, each tied to a datasheet figure:
P_total = VF × I_avg + f × E_switch + f × E_recovery
- Conduction loss is VF × I_avg — the forward drop times the average current. It is the term the datasheet’s VF curve feeds, and it does not depend on frequency.
- Switching loss is the energy per switching event times the frequency f — the turn-on and turn-off transitions that dissipate energy each cycle. The datasheet’s switching or recovery-related curves supply the per-event energy.
- Recovery loss is the reverse-recovery energy per event times f — the stored charge the junction must sweep out at each turn-off.
The sheet’s structure is the reason to memorize it: each term has a different dependency, so each dominates a different operating regime. Conduction dominates at low frequency and high current; switching and recovery dominate at high frequency, even when they look small per event. The “right” diode for a design changes with the operating point, which is why the sheet — not the datasheet headline — is the selection tool.
The error-bound discussion belongs next to the sheet because the three terms do not deserve equal trust. The conduction term is the most certain: a VF curve measured at a defined current and temperature, folded through a multiplication, lands within a few percent if the operating point is known. The switching term is less certain, because the per-event energy depends on the di/dt of the actual circuit, the loop inductance, and the driver behavior — all places the datasheet’s test conditions and the real design part ways. The recovery term sits between: the datasheet’s recovery charge is well defined, but the circuit’s reverse di/dt modifies it. The fair use of the sheet is therefore to compute a central estimate and a high-side bracket, then hand the bracket to the thermal design rather than the point estimate.
The terms also have different error bars. Conduction loss is computable from the VF curve with small error; switching and recovery energies come from curves that are test-condition-dependent and carry wider uncertainty. The honest use of the sheet brackets the high-frequency terms and treats them as estimates, which is exactly the error-bound discipline the forward-voltage hub applies to VF.
Conduction Loss: VF × I_avg with the Right VF
The conduction term is deceptively simple and frequently misapplied. P_conduction = VF × I_avg, but the VF must be read at the operating current and junction temperature, not at the datasheet’s 25 °C test point. Using the headline VF understates the loss in a hot, high-current design by exactly the amount the temperature and current corrections add.
Three corrections complete the term. First, read VF at the average operating current on the curve, not at the nameplate current. Second, correct VF for the junction temperature, because a 100 °C junction drops its VF by roughly 0.15 V from the 25 °C figure. Third, confirm which current the design actually averages — a duty-cycled load uses the RMS or average current over the cycle, and mixing peak and average inflates or deflates the loss number.
A hand example shows the scale. A rectifier running 10 A at VF 0.8 V dissipates 8 W of conduction loss; at VF 0.5 V the same current dissipates 5 W. The 3 W difference is the entire budget that low-VF parts are bought to save, and it is invisible until the term is computed with the right VF. The forward-voltage physics hub owns the curve detail; the sheet’s job is to make the multiplication explicit.
Switching and Recovery Loss at Frequency
The frequency-dependent terms are where most loss estimates go silent. Each switching event stores and releases charge in the junction, and each turn-off sweeps the stored charges out through reverse recovery — both dissipate energy proportional to the event count. The sheet writes them as f × E_switch and f × E_recovery, and the per-event energies come from the datasheet’s switching and recovery data.
The frequency dependence is the design consequence. At 100 kHz, an event energy of 2 µJ multiplied by 100,000 events per second is 0.2 W; at 1 MHz the same event energy becomes 2 W — an order of magnitude shift that can flip a design from conduction-dominated to switching-dominated without any component change. The crossover between terms is where “low VF” stops mattering and “fast recovery” begins to.
The reverse-recovery term deserves its own discipline because it is the least intuitive. Reverse recovery is the time and energy the junction needs to clear stored charge when the current direction reverses, and its energy grows with the switched current and the recovery charge in the datasheet. The reverse-recovery hub covers the waveform and the mechanism; the sheet’s contribution is to fold the per-event energy into the total at frequency rather than leaving it as a datasheet curiosity.
The same double-entry habit applies to the frequency itself. Most power stages do not switch at a single frequency; an SMPS with a modulating control loop, a PFC stage sweeping near line harmonics, or a drive with variable duty all spread the switching rate across a range. Folding the loss sheet at the worst-case frequency overstates the loss, folding it at the nominal understates it. The practical compromise is to run the sheet at two or three representative frequencies — idle, nominal, and peak — and keep all three numbers in the design file, so the thermal design can size for the worst case while the efficiency claim uses the nominal. That is the difference between a number and a model.
Two Worked Examples: SMPS Output and PFC Input
Two examples prove the sheet closes on real circuits.
First, an SMPS output rectifier: 5 A average, VF 0.4 V (hot), switching at 200 kHz with a per-event switching-plus-recovery energy of 1 µJ. Conduction is 5 × 0.4 = 2 W. Frequency losses are 200,000 × 1 µJ = 0.2 W. Total ≈ 2.2 W, and the design is conduction-dominated — the low-VF choice matters, and the fast-recovery value is modest. The example mirrors the SMPS output-rectifier selection question, where a Schottky’s low VF and its recovery behavior are weighed together.
Second, a PFC input stage: 10 A peak, averaging 7 A, VF 0.5 V (hot), switching at 400 kHz with a per-event energy of 4 µJ. Conduction is 7 × 0.5 = 3.5 W. Frequency losses are 400,000 × 4 µJ = 1.6 W. Total ≈ 5.1 W, and the frequency terms now carry nearly a third of the loss — doubling the frequency would add another 1.6 W and flip the design’s sensitivity. The two examples show the same sheet landing on different regimes: conduction-bound in one, frequency-sensitive in the other, and the selection differs accordingly.
Each example documents its assumptions — the VF at temperature, the per-event energy, the frequency — so the number can be recomputed when the operating point changes. That is the spreadsheet discipline the sheet is meant to enable, and it is the same discipline the loss worksheet applies across candidates.
Handing the Loss Number to Thermal Design
The loss total is not an end; it is the input to the thermal loop. The total power dissipated feeds the junction-temperature equation — Tjunction = Tambient + P_total × Rth(j-a) — and the thermal design guide takes the number from here. The handoff is explicit: compute the loss with the sheet, find the thermal resistance from the package, and check the junction against the datasheet limit at the worst ambient.
The handoff also defines the data that must travel with the loss number: the operating frequency, the current, the VF at temperature, and the per-event energy assumptions. Without that context, a “5 W loss” number is meaningless to the thermal designer, because the same diode can dissipate 5 W in a way that is all conduction (easy to shed) or partly transient (harder to shed). Packaging the loss number with its terms is what makes the handoff a collaboration instead of a hand-wave.
Two published resources extend the sheet: the 250 V secondary loss worksheet runs the same three-term method on a concrete output stage, and the fast recovery diode guide develops the reverse-recovery term this sheet counts per event. The sheet, the two examples, and the thermal handoff together are the one-page method this site’s selection and reliability articles expect. When a design runs through it, the parts list stops being a catalog pick: the diode is chosen because its datasheet numbers, folded through the three terms at the operating point, close the loss budget and the thermal loop. The general rectifier category and the fast recovery category supply the datasheet fields for running the sheet on real candidates, and the loss worksheet reference ties the method to the published tools.