[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Snubbing an IGBT in a sstc

Original poster: "Steve Ward" <steve.ward@xxxxxxxxx>

Hi Ken,

I will comment where i can.

On 1/23/07, Tesla list <tesla@xxxxxxxxxx> wrote:
Original poster: "K. C. Herrick" <kchdlh@xxxxxxx>

Steve (& all)-

As always, your comments are very cogent!  The repetitive avalanche
energy maximum for the IRFBE30 MOSFET is 13 mJ.  My understanding is
that 1 mJ  is 1 mW-second, which would be 1 W-millisecond, which
would be 1000 W-microsecond...is that right?

Yeah, conservation of energy (or work if you want).  But, it might not
be that simple.  The avalanche rating might assume some sort of
time-frame, so that localized heating of the die is not too critical
(ive heard this is why older IGBTs dont have avalanche ratings, since
their dies dont spread out the heat properly, and small sections fail
from the transient thermal rise).  You might want to look at what
parameters they gave that 13mJ rating at.

 So over 1 us, the
IRFBE30 should accomodate an energy of 13 KW-microseconds.  (Seem
like a lot, for a teeny little TO-220 transistor!)  And as you point
out, 2 A x 800 V yields 1.6 KW which is a lot less, over that 1 us.

Right.  You have to be careful though, since the 1.6kW might not be
constant over the whole 1uS.

Have you looked at TVSs at all? (im not sure if they would work for
your design or not).

Am I missing something?...  Does one need snubbers, as such, in an
H-bridge configuration?  Is not the best you can do to utilize
lowest-inductance bus structuring plus capacitive bypassing of V+ and
V- right at the transistors?

Well, there is some uncertainty here for myself as well.  For very
large IGBT circuits, which are inherently large, you can only minimize
the bus inductance so far.  Even with the low bus L, manufacturers
still suggest snubber circuits.  For low current applications you can
get by with simple decoupling capacitors on the rails.  But, here is
the biggest problem i have found: the decoupling caps can resonate
with the bus inductance!  Under this condition, my rail voltage was
building up to maybe 160% of the "nominal" input voltage!  In the end
i eliminated most of the small poly propylene decoupling capacitors,
as they only seem to cause ringing on the supply (they no doubt
increase the Q of this circuit). I now only have the large inverter
grade electrolytics present, along with 2uF of decoupling per
half-bridge.  The 2uF improved the switching noise, but does still
cause the rail voltage to oscillate to some degree (not as bad as the
24uF i used to have).

Powerex recommends for applications using larger than 300A devices,
RCD snubbers are a must.  RCD (resistor, capacitor, diode) snubbers
simply charge up a capacitor through a diode (this blocks out any sort
of resonance).  The capacitor is discharged to the *supply voltage*
through a resistor (this way, the RCD only takes action when Vce
exceeds the supply).

For push pull / flyback converters, you might want to look at a
different snubbing solution that i came up with for my boost converter
(which has to hard switch 100A regularly):


More specifically:


I explain the "Active snubber" in the webpage (i use the term in a
different manner than your idea).  This is a proven design!  Even
after a few failures of the tesla coil, the boost power supply
continues to live.  Ive only pushed it to about 8500W though as thats
about all my coil can cope with ;-).

I must warn that the snubber design required extensive pspice
simulation to optimize, and its current values are probably all wrong
for your 150khz operating frequency!  But, it is a very efficient
design compared to normal RCD snubbers where the extra energy is just
burned off.

 In a single-ended or push-pull
configuration such as I'm contemplating there's no reverse-diode(s)
to "catch" the overshoot so snubbing is essential.  But in a 1/2- or
full-H, overshoot--absent stray inductance--is absorbed by the
reverse diode of the "off" transistor(s).  Enlighten me on this.

Its the non-ideal details that are to blame.  When the diodes shut off
(because the opposing IGBT has turned ON) they have a "snappy" turn
off and this causes the "switching noise" ive seen (and at the time,
thought was a major problem, but now im not as convinced).

I've bought some used Powerex CM300DY-24H half-bridges (and also some
CM300DU-24Fs) which I'm hoping to use.  Nothing in the data sheets on
lead-inductance, of course.  But I do notice the boast, "High
Frequency Operation (20-25 KHz)", so...we'll see what happens (if
anything at all) at 150 KHz.

In the simulation, I got h-f oscillation also--until I added the R-C
elements at the base of Q3.

There is, of course, IGBT current during the snubbing event.  But, in
simulation at any rate, that current diminishes to zero in 600 ns at
which time the IGBT's gate voltage returns to its negative-off
level.  So no current-overshoot appears at that time.

In a push-pull arrangement, I think there might be harm in allowing
both transistors to be on at the same time.  Mutual inductance
between the primary-halves would allow excess current to be drawn,
limited, I think, only by the coupling coefficient and the circuit
resistances and stray inductances.  Is that right?

You might have a good point there.  In this case, you are energizing
both inductances at the same time, but their magnetic fields would
cancel mostly (As you said, due to mutual L), so the current would
probably rise very fast indeed!  Ok, better avoid that situation.

As to the active snubber acting fast enough...  The rate of rise of
the IGBT collector voltage at turn-off simulates at 1800 V/us with
the 100 nF value of C3, the 2 uH primary-half and the 20 m-ohm
circuit resistance.  So I would have, in a real implementation,
1200/1800 = 2/3 us absolute max. in which to effect the
snubbing.  Not a whole lot of time, I agree.

I guess its not quite as bad as i was imagining.  Probably do-able.

...And what is an RCD snubber?  Would that be what I show as the
D6/C3/D1/R9 configuration?  I know that's called a snubber
circuit...but to me that's a misnomer: it doesn't snub, i.e. clamp,
the voltage but merely slows down its rise.

Yes, what you show there is in fact an RCD snubber of some sort.
Somehow i overlooked that initially.  And yes, that is the point, to
slow down the rise time, but also, the capacitor works to remove
energy that was stored in the inductance which causes the transient,
thus "snubbing" the voltage spike.

I may well need to do something more with this design other than turn
it into hardware & hope for the best:  I simulated 2 uH of inductance
in the drain lead of the avalanche-MOSFET Q1 and...sure
enough--oscillation during snubbing, shooting the IGBT's collector
voltage to 1.2 KV or so during some of the events.  However...I
question whether a real "brick" IGBT will respond that fast, i.e.
support the feedback loop; the oscillation is at about 16 MHz.

Well, the gate certainly wouldnt respond much to 16mhz noise (you
would need a heck of a lot of gate drive current!).

you have a thought on that?  Also, in regard to reducing
bus-inductances as you suggest...I wonder if that's very important in
a single-ended or push-pull configuration?  Any bus inductance is
merely going to add to the inductance of the primary per se; the main
concern is to keep the loop inductance in the snubber circuit itself
to a minimum--not so?

I see it exactly the same way.

So, best to go on to spark gaps and pole pigs instead?  What's to
fail there??  I even have a pole-pig--which I'd love to get rid
of.  But not for me, I'm afraid...I'm too old.

Never owned a pole pig, and i dont plan on it ;-).