SiC Inverter Losses: Conduction, Switching & Reverse Recovery

Why inverter loss is worth getting right

The losses in a traction inverter's power devices set two things that dominate an EV powertrain's design: the efficiency (range) and the cooling (how big and heavy the heatsink and coolant loop have to be). Getting the loss estimate right early tells you whether a given SiC device, paralleling, and switching frequency will stay inside its junction-temperature limit — or cook. The losses split into a few distinct mechanisms, each with its own scaling.

Conduction loss

When a SiC MOSFET is on, it behaves like a resistor: it drops I·R_ds(on) and dissipates I²·R_ds(on). Two things to respect. First, R_ds(on) rises significantly with junction temperature, so conduction loss and temperature chase each other and you have to iterate. Second, running synchronous rectification (the normal way to run SiC — channel conducts in both directions instead of the body diode) makes the total conduction loss symmetric between motoring and generating; the motor/generator distinction only shows up explicitly when sync rect is off and the body diode carries part of the current.

Switching loss

Every turn-on and turn-off dissipates a packet of energy while voltage and current overlap during the transition. Datasheets give these as Eon/Eoff at a test point, and they scale two ways: linearly with the switched current, and with voltage by (V_dc / V_test)^kv. Total switching loss is that energy times the switching frequency — which is why raising f_sw for smaller passives directly buys you more switching loss. A useful SiC fact: Eon/Eoff barely change with junction temperature (unlike silicon IGBTs), so they're treated as temperature-independent.

Reverse-recovery loss

When the opposing device turns on, it has to sweep out the charge stored in the freewheeling device's diode — reverse recovery. SiC's big advantage over silicon is how small this is. It's taken from the datasheet Err where published, or approximated as Q_rr·V_dc / 4 for a soft-recovery device.

What doesn't heat the die

Gate-drive loss is real power, but it's dissipated in the driver and gate resistors, not the silicon — so it's excluded from the junction-temperature solve (though it still matters for the gate-drive supply budget).