Some time ago I built a H-bridge with 5 paralleled IRFP250N and after a couple of blown-up IR2104s I came up with this driver design that was more about protecting the IC than anything else:


enter image description here Where:

  • HO is upper MOSFET gate
  • LO is lower MOSFET gate
  • Vs is half bridge output

(R2 is not connected. It was designed so I can put another chip in there. R1, R4, R5 are shorted if I recall. R17, R18, R19, R21, R22, R23 are there so I could experiment with different values. Gate resistors are directly soldered to the transistors).

It worked well, and I never had a problem with it. Switching frequency was 400 Hz with about 700 nS turn on and off times. Then I bought two 2MBI200U2A-060 modules for cheap and hooked up these drivers. When no power is applied to the half-bridge this is the waveform on the gate of the lower IGBT:

No power waveform

It looks good to me. Then after applying 30 V to the bridge this is what happened:

enter image description here

The yellow waveform is Vge and blue Vce of lower IGBT. This is without the load. There are snubbers added but there is still some ringing on the output, but this is not what I am concerned about.

The voltage on the gate suddenly rises well above the turn-on threshold and the IGBT actually starts conducting because I can see increased power consumption when this spike goes above 7 V. I tried different gate resistors and diodes as well as adding a PNP gate discharging circuit (using I think it was TIP42C paired with the second transistor connected in Darlington) and the max spike voltage was around 5 V.

When I connect the load it gets a bit better but still not pretty. With a 100 Ω resistor and diode on the gate the turn on time was about 7 μs and turn off about 1 μs but at this point having these times so large is just a loss of power.

I tried different combinations of gate resistors, diodes, and even a second totem pole driver but I can't get rid of this spike. I know I can use negative voltage at the gate but this makes the circuit so much more complicated. How can I prevent this?

I think that this has something to do with Miller capacitance but I'm not sure. The datasheet says that this module has 14 nF input capacitance, so quite high. In the end I plan on using this bridge with 150 V, 180 A running at 400 Hz and I fear that with 150 V both transistors will turn on and blow up.

Any help is appreciated.

  • \$\begingroup\$ Welcome! Please show layout. Main culprit door destroyed IR2104 is bad layout causing peak negative voltage to go out of spec. Ask me how I know :-| \$\endgroup\$
    – winny
    Aug 27, 2022 at 12:20
  • \$\begingroup\$ I've edited the post so the layout is shown. You are right about peak negative voltage and in my current design are protective diodes and the chip no longer blows. Now I'm trying to get rid of this spike at turn off but I don't know how \$\endgroup\$ Aug 27, 2022 at 13:10
  • 2
    \$\begingroup\$ No, show the inverter itself, the module. The driver layout isn't very important here. \$\endgroup\$ Aug 27, 2022 at 13:19
  • 1
    \$\begingroup\$ Is there any reason -I'm currently missing - for the flipped transistors in the lower halves of gate drivers discrete push-pull? This way they will be giving very weak pull down and struggle to keep IGBTs off. BTW for safe and reliable operation you definitely need isolated gate drivers (both upper and lower) \$\endgroup\$
    – carloc
    Aug 27, 2022 at 17:41
  • 1
    \$\begingroup\$ @carloc Yeah, those are erroneous. Seems to be laid out that way, too (whoops!). Those transistors aren't doing anything much anyway, with 500mA ratings; PBSS303NX/PX or relatives are more likely candidates here. \$\endgroup\$ Aug 27, 2022 at 18:44

1 Answer 1


A bootstrap driver is wholly unsuitable for this type of IGBT. Notice Cres vs. Coes curve in the datasheet (also see edit below): this means for a voltage step on the collector, half that voltage appears on the gate. They're pretty awful as transistors go these days; but what do you expect from a 20-ish year old design, I guess. (Plenty fine for say motor drives, which was probably the main application they were used in.)

Devices like this, must operate with low enough gate drive impedance, and slow enough dV/dt on the switching node, that the differentiator thus formed (between the C-G-E capacitor divider, and total equivalent gate resistance), does not drop enough voltage to turn on the transistor, thus causing destructive switching (shoot-through). Vge(off) could be as high as a couple volts below Vge(th), but it is for this reason that it must be considerably less: the datasheet recommends -15 V.

Turn-on dV/dt is likewise controlled by RG and the same equivalent capacitance, so you simply wire up whatever gate drive you have, and adjust RG until switching is just safe enough under all operating conditions.

Your circuit may still work, but RG must be much higher to get dV/dt low enough to support it. This is a direct consequence of VGE(off) being only a few volts below VGE(th), rather than more than 15 below.

Typically an isolated gate driver is used, with 30 V total supply (or 15 V supply and an H-bridge output, but this makes desat protection difficult to add -- another highly recommended feature!). This solves both the issue of isolation between control and, usually mains voltage (600 V devices are typically used on 240 V industrial equipment), and the difficulty (or impossibility!) of using bootstrap type drivers.

Do you really need IGBTs for such low voltages? -- is this a convenience thing, this module is cheaper than alternatives? Is it really cheaper when the lost efficiency (VCE(sat) ~ 2 V, max. ~97% efficiency) is accounted for? Switching loss at 400 Hz at least isn't a problem, either way. (Indeed with MOSFETs, your application might benefit from the reduction of harmonics, or component size, by using PWM at higher frequency; this will require more care in the inverter layout, and more design of the controls, of course, so I understand if this isn't an option at this time.)

I guess you were originally working with MOSFETs, but blew the driver, which is most likely be due to poor layout. This will not be affected by going to IGBTs. The switching frequency doesn't matter: it's speed that kills, how fast it goes between on/off. Layout is very much a part of the circuit and cannot be ignored!

Edit: Manufacturer explains the parameters here:
Technical Terms and Characteristics
Particularly Fig. 2-13: the capacitances are separated, using a different definition from most MOSFETs. This explains why Coes can be lower than Cres. This also means you must use the sums of these values, as appropriate, in calculations.

Also note the typo on page 3, they forgot to edit the "Output capacitance" description to say C-E capacitance. Presumably, they are using the latter definition.

  • \$\begingroup\$ How is it even possible for Cres to be higher than Coes as in that datasheet? My understanding is that Coes = Cres + Ccs, and Ccs is obviously positive... \$\endgroup\$
    – Hearth
    Aug 27, 2022 at 15:26
  • \$\begingroup\$ Cce, not Ccs. Mixing up MOSFET and IGBT terminology. \$\endgroup\$
    – Hearth
    Aug 27, 2022 at 15:50
  • \$\begingroup\$ So that explains why they were so cheap. I abandoned MOS because I needed higher current capability and I didn't want to add more transistor in parallel because my heatsink was too small and dealing with parasitic inductances was a nightmare at this point. IRFP250N are rated at 200V and my supply voltage is 150V with no load (it goes down with the load - CC power supply aka. welder) so there is not much headroom. Also this driver design worked flawlessly for me for a long time. The earlier version didn't had diodes and negative voltage spikes killed it as someone mentioned in the comment above \$\endgroup\$ Aug 27, 2022 at 17:43
  • \$\begingroup\$ @Hearth Yeah, not sure how that happens....... Aha, found a reference, see edit. \$\endgroup\$ Aug 27, 2022 at 18:52
  • 1
    \$\begingroup\$ @TimWilliams I'd say less "more believable" and more "actually possible", but yes. \$\endgroup\$
    – Hearth
    Aug 27, 2022 at 19:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.