IGBT driver update: I called International Rectifier

[Jamie]

Ignoramus1797 wrote:




Be careful with them !

When the IGBTs fry - the driver normally goes too .

Graham

i know that one, had little hot balls falling out of 690+ series
today! bye bye IGBT's

 

[Ignoramus24489]

If I were to develop a unit with those chips, I would get an entire
*TUBE*. And I would probably use them, too. Until your gatedrive works
properly, it doesnt work properly. And gatedrives that dont work
properly cause explosions. Even when you are experienced and careful.

But there is a *LOT* to be learned from such an exercise.

Yes, worst case, I will learn something. I do not have the time to go
through an entire tube of stuff, blowing them up. If it does not work
on the first and second time, I will just give up.

i

 

[Ignoramus24489]

solid groundplane underneath the circuit. And keep IGBT gatedrive
connection inductance low - 50nH is really too much, 10nH is better.

if you use a DIP proto-board, it will go *BANG* the first time you try
to use it on the welder.

That's what I was, in fact, planning on doing. :-(

Is there some proto board type thing that is acceptable?

i

 

[Terry Given]

That's what I was, in fact, planning on doing. :-(

Is there some proto board type thing that is acceptable?

i

no, not really. the problem is the inevitable large physical loops that
exist. these pick up H fields, and convert them to troublesome voltages.
that in a gate drive always lead to a loud bang.

honestly, it would be far cheaper, at US$50 a pop, to buy a couple of
semikron drivers. the advantage here is that (provided you connect it
well to the IGBTs, minimising inductance) even if your logic goes
screwy, the gatedriver will prevent you from being able to blow the
IGBTs up.

or you could dead-bug your hardware, on top of a Cu-clad PCB as a ground
plane. I've built some pretty tricky stuff this way, and have seen
pictures on the web of some truly amazing gear - IIRC google "manhattan
style"

keep circuit loop inductances small. current flows in loops. even logic
signals.

Cheers
Terry

 

[Ignoramus25589]

no, not really. the problem is the inevitable large physical loops that
exist. these pick up H fields, and convert them to troublesome voltages.
that in a gate drive always lead to a loud bang.

honestly, it would be far cheaper, at US$50 a pop, to buy a couple of
semikron drivers. the advantage here is that (provided you connect it
well to the IGBTs, minimising inductance) even if your logic goes
screwy, the gatedriver will prevent you from being able to blow the
IGBTs up.

or you could dead-bug your hardware, on top of a Cu-clad PCB as a ground
plane. I've built some pretty tricky stuff this way, and have seen
pictures on the web of some truly amazing gear - IIRC google "manhattan
style"

keep circuit loop inductances small. current flows in loops. even logic
signals.

Thanks. I am reading about it now. If I have to do it, I have to do it.

i

 

[Tim Williams]

yep. Yuck!
:~p

always ensure base currents are limited. lest stuff fail during wierdo
start-up, shut-down or interference conditions.

Good point, so that'd be um, 0.15 ohm emitter resistors or something in the
gate drive circuit then? (If not for: )

in general, such linear
circuits have piss-poor performance. actual thresholds vary widely with
time, temperature and sock colour (or so it seems....). Use a real
reference voltage, and a proper comparator with suitable hysteresis.

Ok. But why? I get good results (into MOSFETs, granted they don't handle
as much current) using an emitter follower and transformer for high and low
side gate drive. (Three state voltage output: +12V for high side, -12V for
low side, 0V inbetween states for both off.)

I can't do that here since I want a square wave that goes from +10 or +15V
to -5 or -10V, and I need some sort of active circuitry because just passing
that waveform through a transformer is going to poop on the DC offset,
especially as duty cycle is varied (from perhaps 25 to 48%; ultimately it
will be set to the upper limit, but for testing I want control). A
requirement of 15V on-state gate drive doesn't work when duty cycle is less
than 50% and your negative gate voltage limit is -20V!

likewise the desat detector is equally yuck. use an LM339

and just
measure Vc. there are 2 methods: a resistive divider, or an HV diode and >

a pullup to (say) +15V.

Eh, that MIGHT work...

So what of the high side Vsat? Do I need to float a 393 by the +rail?

Either way, note that Vce can be very, very
negative - even though the IGBT has a reverse diode, there is an
inductor between it and where you measure Vc.

Ah, true. Although, I don't really see just how high it can be if I'm
switching it over say 0.5 to 1uS. That's 50-100A/uS, but still... hm, but
wait, diode you say? Ah yes, that will want to switch at around 1A/nS, good
point! So a small resistor and 1N4148 on the comp?

Tim

 

[Terry Given]

Good point, so that'd be um, 0.15 ohm emitter resistors or something in the
gate drive circuit then? (If not for: )

?!

if its a complementary emitter-follower output stage, driving Cg thru
Rg, then you can happily use a base resistor Rb = Beta*Rg without really
buggering up your output. pick a seriously pessimistic beta though,
about 25% rated....

I'm referring to the ratshit desat circuit here. that threshold will be
truly awful.

Ok. But why? I get good results (into MOSFETs, granted they don't handle
as much current) using an emitter follower and transformer for high and low
side gate drive. (Three state voltage output: +12V for high side, -12V for
low side, 0V inbetween states for both off.)

[snipped bit was here]

I can't do that here since I want a square wave that goes from +10 or +15V
to -5 or -10V, and I need some sort of active circuitry because just passing
that waveform through a transformer is going to poop on the DC offset,
especially as duty cycle is varied (from perhaps 25 to 48%; ultimately it
will be set to the upper limit, but for testing I want control). A
requirement of 15V on-state gate drive doesn't work when duty cycle is less
than 50% and your negative gate voltage limit is -20V!

then you have to do something trickier. the basic idea of the icky
gatedrive would work, but dont be afraid to hiff an LM339 or suchlike in
there. If you are careful, you can also suck your supply rails out of
the same xfmr - eg voltage doubler for +ve and -ve rails, and a couple
of regulators (the -ve regulator can be pretty crude).

Or run the gatedrive xfmr at 10MHz, MSK and toss in a demodulator. that
way you get nice supply rails for a diode and a cap....

a pullup to (say) +15V.

Eh, that MIGHT work...

I have seen and used both approaches.

So what of the high side Vsat? Do I need to float a 393 by the +rail?

maybe. depends on how good a gate drive you want. I'd not bother, and
just sense desats on the lower igbts, its not very high power.

its probably not that hard to have a ground-referred high-side desat
detector, either. you would be looking for Vc of the lower transistor
failing to get within about 10V of Vdc when the upper transistor is on.

Ah, true. Although, I don't really see just how high it can be if I'm
switching it over say 0.5 to 1uS. That's 50-100A/uS, but still... hm, but

100A/us * 1uF = 100V negative spike. watch something break.

100A/us * 100nH = 10V negative spike.

100A/us * 10nH = 1V negative spike. thats enough to make an LM339 output
go bonkers if its running from a unipolar supply. And I'd like to see
you keep the inductance between the diode cathode and your Vc takeoff
point below 10nH for *any* package.

 

[Tim Williams]

if its a complementary emitter-follower output stage, driving Cg thru
Rg, then you can happily use a base resistor Rb = Beta*Rg without really
buggering up your output. pick a seriously pessimistic beta though,
about 25% rated....

That's true. Not always thinking clearly, as you've noticed. <g>
Yeah I'd go for say beta = 5 or 10.. typical saturation figure for power
transistors. Obviously that depends on what type I pick, which I haven't
picked yet...

I'm referring to the ratshit desat circuit here. that threshold will be
truly awful.

Well okay, then swap the zener for something else...

I was told by a friend it would work, but he deals more with AF amps than HF
class D, too...

then you have to do something trickier. the basic idea of the icky
gatedrive would work, but dont be afraid to hiff an LM339 or suchlike
in there.

Well, what the heck do I need an extra fourteen pins for when I can use a
couple transistors to do the same thing? (Okay, so in the circuit I put
down four transistors which makes 12 pins which would probably cover a wider
footprint, but,,, that's beside the point!)

If you are careful, you can also suck your supply rails out of
the same xfmr - eg voltage doubler for +ve and -ve rails, and a couple
of regulators (the -ve regulator can be pretty crude).

<snip interesting modulator>

What, you mean use the gate drive for supply *and* switching? I don't think
the SG3524 would like that, although it does have 300mA output which would
go reasonably far at 12V (=4W). Haven't estimated gate power yet...

maybe. depends on how good a gate drive you want. I'd not bother, and
just sense desats on the lower igbts, its not very high power.

So what if there's a ground fault condition? I can hear a shotgun shell
going off yet the bottom transistor desat is aloof.

its probably not that hard to have a ground-referred high-side desat
detector, either. you would be looking for Vc of the lower transistor
failing to get within about 10V of Vdc when the upper transistor is on.

If I had some 400+ Vceo PNP's this could be a lot easier...
(Yeah, so buy some... I'm buying plenty of other things after all)

Hm.. I'm not seeing it work with a grounded op-amp that doesn't have 300V
common mode input range..

100A/us * 1uF = 100V negative spike. watch something break.

Um, abuse of units? Unless you typo'd uF instead of uH.

100A/us * 10nH = 1V negative spike. thats enough to make an LM339 output
go bonkers if its running from a unipolar supply.

Yeah... >>So a small resistor and 1N4148 on the comp?

If you mean not so much keeping things working as getting erroneous output
bunnies, I was going to integrate it a little since the gate voltage is at
least 200nS ahead of the collector.

And I'd like to see
you keep the inductance between the diode cathode and your Vc takeoff
point below 10nH for *any* package.

Izzat fine for IGBT + diode packages then? Well, not really... hmm in fact
it would be better to have an external diode since I could connect very near
the diode junction and keep it to a few nH.

Or I could put an integrator on to snub the 100nS-ian rectifier switching
hubub. Desat detector won't mind another half uS lag.........

Tim

 

[Terry Given]

That's true. Not always thinking clearly, as you've noticed. <g>
Yeah I'd go for say beta = 5 or 10.. typical saturation figure for power
transistors. Obviously that depends on what type I pick, which I haven't
picked yet...




Well okay, then swap the zener for something else...

I was told by a friend it would work, but he deals more with AF amps than HF
class D, too...

and what do you use as a detector? R + comparator = beta-dependent
threshold....

Well, what the heck do I need an extra fourteen pins for when I can use a
couple transistors to do the same thing? (Okay, so in the circuit I put
down four transistors which makes 12 pins which would probably cover a wider
footprint, but,,, that's beside the point!)

all depends on what you are trying to achieve. If you want a floating
desat detector, a dual or quad comparator is probably the way to go. a
quad allows simple UVLO too. I have designed gatedrives without desat
detectors, but *never* without UVLO. Low Vg is one of the best ways to
destroy FETs/IGBTs, and it usually happens at power-up (or down).

<snip interesting modulator>

What, you mean use the gate drive for supply *and* switching? I don't think
the SG3524 would like that, although it does have 300mA output which would
go reasonably far at 12V (=4W). Haven't estimated gate power yet...

Pg is tiny cf 4W.

So what if there's a ground fault condition? I can hear a shotgun shell
going off yet the bottom transistor desat is aloof.

very true. so add desat to the upper gatedrives. or feed both DC bus
wires (in opposite directions) thru a DCCT, and trip on overcurrent. or
think up some other scheme.

If I had some 400+ Vceo PNP's this could be a lot easier...
(Yeah, so buy some... I'm buying plenty of other things after all)

not hard to hang a low-power supply just below +Vdc - eg a zener in
series with your cap "balancing" (OK, discharge) resistors. Then a
simple opto (low dV/dt) can give you a logic sigbnal for Vceupper >
(say) 10V, which you can combine with gatedrive signals and filter to
develop a floating desat circuit. Hell, you probably dont even need the
supply....

Hm.. I'm not seeing it work with a grounded op-amp that doesn't have 300V
common mode input range..




Um, abuse of units? Unless you typo'd uF instead of uH.

oops. I suffer from dailysex

Yeah... >>So a small resistor and 1N4148 on the comp?

If you mean not so much keeping things working as getting erroneous output
bunnies, I was going to integrate it a little since the gate voltage is at
least 200nS ahead of the collector.

usually best to filter prior to the comparator, and use Vg_off to
discharge the filter cap. but either one would work. can do desat delay
digitally too.

Izzat fine for IGBT + diode packages then? Well, not really... hmm in fact
it would be better to have an external diode since I could connect very near
the diode junction and keep it to a few nH.

then the inductance between the diode and igbt will cause problems.
IGBTs have piss all reverse-blocking capacity. If you look inside a
half-bridge module, you will find that the top diode is closest to the
bottom IGBT and vice-versa, to minimise the current commutation loops....

Firstly, optimise the power path. Then deal with the effects that has on
the signal paths.

Or I could put an integrator on to snub the 100nS-ian rectifier switching
hubub. Desat detector won't mind another half uS lag.........

you have 5-10us to get around to desatting, depending on IGBT. for
little ones its less.

Tim

Cheers
Terry