Selecting a MOSFET for a switching power supply is usually presented as a parts hunt — filter the catalog by voltage, pick a low RDS(on), done. The parts hunt fails in practice because two different loops close at the same time. One loop is thermal: the design’s loss budget closes on RDS(on) and the switching losses, and a part with a beautiful on-resistance can still fail the loop when its switching loss is counted. The other loop is frequency: the design’s switching frequency closes on the gate charge Qg and the layout’s ability to drive it, and a part that works at 100 kHz can be un-drivable at 500 kHz. This article builds the specification from the load up, runs both loops, and ends with a four-row comparison table you can fill from any catalog.
Writing the Requirement Sheet: Voltage, Current, Frequency, and Budget
Before any catalog opens, write the requirement sheet — four numbers that every MOSFET selection depends on. First, the operating voltage and its extremes: the rail voltage, the maximum reverse and transient voltage the part must hold, and the margin you will accept. Second, the current: average, peak, and the pulse shapes the application imposes. Third, the switching frequency, including the range if the design modulates. Fourth, the loss budget: the watts the power stage may dissipate in the MOSFET before the thermal design breaks.
The requirement sheet is what makes the hunt honest. A MOSFET selected without the loss budget will win on headline numbers and lose on the thermal loop; one selected without the frequency range will win at the datasheet’s test frequency and fail at the application’s real one. Writing the four numbers down also gives procurement a spec sheet instead of a part number, which is the difference between a sourcing conversation and a guess.
The voltage number carries the most subtle requirement. On a switching supply the MOSFET sees not just the steady rail but the transients of switching — the overshoot of the stage, the reverse-recovery behavior of the output, and the margins the insulation test demands. Select the part for the maximum the stage can produce, not the nominal rail, and the voltage ratings method applies the same margin logic that rectifier designers apply to diodes.
Conduction vs Switching Loss: The Split That Guides the Part
The loss budget splits into two families with opposite dependencies, and the split is the core of the selection. Conduction loss is RDS(on) × I² — it grows with the square of current and does not care about frequency. Switching loss grows with frequency and with the energy each switching event dissipates — roughly 0.5 × Coss × V² per event, or the switching-loss term from the manufacturer’s curves. A part selected for low RDS(on) alone often has a larger Coss, and at high frequency the switching term dominates and defeats the low-resistance advantage.
The practical tool is to compute both losses at the design’s operating point and compare the split. At low switching frequency and high current, conduction dominates and the part with the lowest RDS(on) wins; at high frequency, the switching term overtakes, and the part with the lower Coss and gate charge wins even at higher RDS(on). The crossover is where “low RDS(on)” stops being a virtue and starts being a red herring, which is precisely the D-series trade the title names — the family trade between conduction and switching performance.
The loss split also decides the test you need. A conduction-dominated design can be verified on the bench with a DC current and a temperature measurement; a switching-dominated design needs the scope and the frequency sweep to see the real loss. Building the split into the requirement sheet before the catalog hunt prevents the two being conflated at selection time.
A numbers example makes the split concrete. A 10 A, 12 V buck stage switching at 80 kHz with a candidate MOSFET at RDS(on) = 8 mΩ (hot) and Coss = 500 pF has a conduction loss near 0.8 W and a switching loss near 24 W at 100 kHz if estimated by the simple term — clearly frequency-bound, and the catalog’s low-resistance headline is irrelevant next to that switching term. The same stage at 20 kHz cuts the switching loss to about 6 W, where the RDS(on) choice begins to matter again. The split is not decoration: it flips the winning part between two reasonable operating points of the same design.
Gate Charge Qg and the Frequency Ceiling
Every switching event charges and discharges the MOSFET’s gate, and the charge moved per event — Qg — is the number that sets the frequency ceiling. The gate driver must push Qg through the loop resistance within the switching period, so the maximum frequency is bounded by Qg, the driver current available, and the layout’s ability to move charge quickly. A large device with a quoted low RDS(on) carries a large Qg, and at high frequency its gate demand can exceed what the driver and layout can supply.
The frequency ceiling shows up as two failure modes. The first is a visibly degraded waveform: at high frequency the gate edges soften, the switching loss climbs, and the part runs hotter than the loss split predicted. The second is an un-drivable part: the driver current or the loop inductance cannot move Qg fast enough, and the design must either lower the frequency, accept a smaller part, or invest in a faster driver with a dedicated gate-drive layout.
The design habit that manages Qg is to size the driver and the gate loop from the target frequency before the part is locked. A part chosen for the perfect RDS(on) that demands a driver beyond the board’s space is a failed selection; one chosen in the frequency loop — Qg within the driver’s reach, loop inductance budgeted — survives the first prototype. The gate-drive design guide covers the timing and layout side of this loop in detail.
The crossover logic answers the “D-series” question directly: D-series families process the conduction-versus-switching trade inside the part design, and the buyer’s job is to buy the family whose crossover sits at the design’s operating point, then verify it with the thermal closure. No family is universally best at both ends of the loss split, which is why the four-row table — not the brand — decides the selection.
Package and Thermal Checks Before the Datasheet
The package is part of the spec, not decoration, and its thermal capability is the closing term of the first loop. A MOSFET’s RDS(on) is quoted at 25 °C, and its real on-resistance rises with temperature — typically 0.4–0.7%/°C for the material — so a part that dissipates its budget at 25 °C can run 20 °C hotter on-resistance in a warm enclosure, compounding the loss. The package determines how much heat can leave: a TO-220 with a heatsink tab, an SMD with a board-facing pad, and a leadless package each have different thermal paths and different practical dissipation.
The check before the datasheet is the thermal closure, using the same equation the rectifier guides apply: junction temperature = ambient + dissipation × thermal resistance. Run the loss split’s worst number through the package’s Rth, verify the junction stays under the datasheet limit at the worst ambient, and the package decision is made before any electrical refinement. A part that passes the electrical tests but fails the thermal closure is out, and the envelope forces the selection toward a larger package or a lower-loss part.
The package check also answers the mounting question, because the available heatsink space and orientation decide how much of the dissipation the design can actually shed. The thermal design guide works the full Rth math, and the MOSFET’s datasheet comparison should include the thermal resistance numbers the same way any rectifier comparison would — as part of the specification rather than a footnote.
Shortlisting Parts: A Four-Row Comparison Table
With the requirement sheet, the loss split, and the package check done, the shortlist writes itself as a four-row table. Each candidate row records the operating voltage margin, the RDS(on) at the real junction temperature, the switching-loss term at the design frequency, and the thermal closure result — the four numbers that the earlier steps produced.
| Candidate | Voltage margin | RDS(on) at temp | Switching loss @ f | Thermal closure |
|---|---|---|---|---|
| A (low RDS(on)) | Meets | Low | High at frequency | Marginal |
| B (balanced) | Meets | Moderate | Moderate | Passes |
| C (low Qg) | Meets | Higher | Low | Passes |
| D (large package) | Meets | Low | Moderate | Passes, larger |
The table makes the trade visible: candidate A wins the first column and loses the thermal closure at the design frequency; candidate B is the balanced pick; candidate C wins where frequency dominates; candidate D passes everything by brute size. The selection reads the table against the design’s dominant loop — conduction, switching, or thermal — and the choice falls out without a parts-hunt argument.
The SMPS components explainer places the MOSFET in the full supply architecture, and the gate drive design guide owns the timing side of the Qg loop this article closes. The MOSFET categories on the site provide the datasheet fields for filling the table. Shortlisting is now a bookkeeping step rather than a gamble, which is the point of building the spec before opening the catalog.