As this form, the wires for either three pins will be much longer (higher parasitic inductance) than the other mosfet types that is soldered on board, as we need to use quite relative long wires to be secured at the terminals. Does this mean for this type of power mosfet, relatively high switching speed is quite challenging ?

IIRC, the typical lead inductance is around 8nH, and the mutual is much less.

You bolt the PCB to the device(s), after bolting them to the heatsink. With several transistors, the PCB can be self supported, or you can use standoffs or brackets to hold things together as usual.

Made a 5kW inverter module, some 5 years ago, using a quad of these. Up to 400kHz. Think the expected limit was around 600kHz at modest derating, but we never pushed it.

Suppose you have a transistor with an equivalent gate of 2 ohms + 20nF. This must be driven at +15/0V (or some negative 'off' bias), as fast as needed (on the order of 2 ohm * 20nF = 40ns).

A typical driver would be IXDD614CI, which has under 1 ohm output ESR, and can switch in about 20ns, fast enough.

The gate driver loop includes these elements:

- Gate drive supply bypass cap ("big enough not to matter" C, some ESL and ESR)

- Driver IC (~10nH ESL due to pin inductance, ~1 ohm ESR due to Rds(on))

- Gate resistor (if applicable; no more than a couple ohms, unless you need it slower*)

- Transistor lead inductance (~8nH)

- GND return path

- And of course any stray inductance in all these paths along the PCB.

*Of course if you need it slower (say for EMI reasons), you can cheapen the driver a good bit.

Likewise the drain circuit has a loop including:

- Transistor D-S (~8nH, and ESR of Rds(on) when on, or C of ~nF when off)

- Opposite side transistor D-S (if applicable; take half bridge for instance)

- Supply bypass C (and ESR, ESL)

- And PCB strays.

For the gate circuit, you have a total of maybe 20nF + 3 ohms + 30nH. This RLC circuit has a resonant impedance of Zo = sqrt(30nH/20nF) = 1.5 ohms. ESR > Zo, so it is overdamped. There is no need to further minimize inductance, nor to increase RG.

For the drain circuit, you have a total of maybe 24nH + 4nF, for an impedance of 2.4 ohms. But ESR is maybe 0.3 ohm! So, if this resonance is excited, it will exhibit considerable overshoot, leading to failure!

The resonance has a quarter period of pi*sqrt(24nH*4nF) = 15ns, which is on the order of times we expect to see here (gate switching will occur in < 80ns, so the Miller plateau will easily be started, crossed or finished in this time period). So expect large overshoot (~60%?).

What to do about it?

Add inductance.

If the load is 300V and 50A, then the load impedance is around (300/50) = 6 ohms. Adding ESL to bring the total to ~100nH will raise the switching impedance Zo to the same level. This only makes things worse at first (more reactance means more parasitic energy stored during a switching cycle), but the first step is to make the inductance manageable, and balanced. Now, when the inverter is lightly loaded (in hard switching), it dissipates maximum power in junction capacitance; when loaded with rated inductive load, it dissipates maximum power in snubber inductance.

Now we add a clamp snubber. Since the supply inductance is modest (~80nH added), we have some opportunity to add an RCD clamp snubber circuit across it. The capacitor and diode may have ~20nH (a large part of which is equivalent from the diode's turn-on speed (forward recovery) -- you'd use a junction diode here for the lower cost and capacitance), which is 1/4th the supply loop inductance so we won't be doing too badly here (we can expect a similar reduction in overshoot).

The C value needs to be some times the loop capacitance (>3x, otherwise it charges too much during the spike), and R needs to be small enough so the capacitor is discharged (> 2 RC time constants) before the next pulse. The diode handles very little average current, but cannot be too small (I think the failure mechanism is electromigration; follow the repetitive pulse rating if present).

Now the resistor dissipates the majority of commutation losses. It doesn't need to be fancy; a tubular ceramic resistor is fine, or you can build a converter to recycle the clamp energy. (A quasi-resonant "lossless" snubber can be used to recycle its own energy as well, but this increases the turn-on peak current, and usually needs more voltage overshoot. In other words, solves the problem by making things worse again. They can be appropriate for some converters though.)

Tim