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

Re: OLTC update



Original poster: "Terry Fritz" <twftesla-at-qwest-dot-net>

Hi Jan,

Since The gate capacitance and the reverse transfer capacitance form a
voltage divider (45pF / (45pF + 3300pF) = .0135 we need a vastly greater
dV/dT on the collector to drive the gate high too.  Since the output
capacitance is 200pF (note these capacitances change with voltage but...)
and we have ten IGBTs, the resonating capacitance is 2nF.  We now know the
frequency is 5.6MHz.  Well slow enough for bipolar transorbs to catch.

To extend such fun, we also know the theoretical peak voltage of the spike.
 Assuming no loss, we are transferring 650VDC at 47uF (9.9J) to 2nF...
Only 99.6kV ;-))

The dV/dT is 4 x Vp x F = 2.2MV/uS   That is a "REAL" dV/dT :o))  With the
capacitive voltage divider from above, the gates see about 1350V!!  So the
question is, why don't they blow right up?!!  First, the losses are
enormous.  Q = L /(C x R) so if the resistance is say 1000 ohms the Q is
0.2.  Then you have the secondary, transorbes, and strays eating at it too.
 But the real key are those transorbs.  They smack that pulse like a ton of
bricks at about 750 Volts so the gates may see a 10 Volt rise.  It is all
sort of messy, but really very beautiful how it all works :-))  Apparently,
that 10 joules gets eaten up in the CE resistance of the IGBT as it
switches since that is a far better looking load to the energy.  Only the
very first few 10s of nS of the pulse really gets to try to make high
voltage.  Of course, the coil is "eating" 1200 watts in such a condition
but that "should" go to warming the heatsink as opposed to blowing the
thing apart.  It has all worked so far ;-))  More below...

At 01:17 PM 10/1/2002 +0300, you wrote:
>Hi Terry,
>
>> The high gate voltage (27 volts = three 9V batteries) is needed since I am
>> running about 500 peak amps per IGBT.  2.5X their rating :-))
>
>Yeah works as long as max power dissipation at specific junction
>temperature is not exceeded. Something like 10 x Ic_peak_max possible,
>AFAIK, but only for a few microseconds.
>If your igbts survive longer than a few runs that'd be quite promising! ;)
>Those buggers really seem to get a hard life, 2.5x the rating :-)
>Well at least it seems they're latch-up free pretty high over ratings, so
>one less worry...
>

I have run the IGBT to 750 amps.  That is about the limit where they loose
all control of themselves.  The gate just can no longer control the
current.  So I guess I have 50% "headroom" condition as opposed to 250%
"over spec" condition :oD

Here are some fun scope pics:

http://hot-streamer-dot-com/temp/OLTC08-18-08.gif

That is turning the IGBTs off at current "peak".  Spark gaps don't do that
;-)  Note the "scream" it makes as the system wildly tries to dissipate the
stored energy (no secondary).

http://hot-streamer-dot-com/temp/OLTC08-21-02.gif

This is just a little low power spike.  Note how slow it really is:

http://hot-streamer-dot-com/temp/OLTC08-21-03.gif

Here it is being hit by a transorb:

http://hot-streamer-dot-com/temp/OLTC08-21-04.gif

http://hot-streamer-dot-com/temp/OLTC08-21-05.gif

The transorb is highly effective as long as it does not get blown away ;-)

Here is an Icollector graph for a single IGBT that the data sheet does not
have ;-)

http://hot-streamer-dot-com/temp/OLTC08-30-02.gif

At about 750 amps it just starts to flat top.  I could easily play with
that point if I did not let the IGBT get too hot.  Never blew the IGBT even
as the current tore it out of saturation (Vce about 500 volts too...).
They are REALLY tough ;-) 

>> So I need
>> "extra" gate voltage to insure the CE gate stays in saturation.  The Ic vs.
>> Vge graph shows to get "really high" current you have to have pretty high
>> gate voltage.  Of course, I really should have used 15 IGBTs in retrospect
>> to keep the gate voltage and current per device down.
>> I am not worried about gate failure.  I will test a spare IGBT sometime
>> but I bet they are solid to 50 volts.  The gate bonds are big and these
>> are copper metalized ICs.  They are built for big currents and fast gate
>> switching ;-))
>
>Peak gate current and maximum gate voltage depend on two different things.
>Using real metallization instead of poly on the gate "just" reduces gate
>series resistance. The maximum gate voltage is mainly restricted by the gate
>oxide thickness...
> http://www.irf-dot-com/technical-info/appnotes/an-936.pdf
> http://www.fairchildsemi-dot-com/an/AN/AN-9016.pdf
> (some nice I_c peak calculations there, too...)
>
>I've never heared anyone use more than 20V, really. While it seems that
>typical gate oxide punchthrough voltage _might_ be a stunning 70V..80V for
>_some_ devices, no datasheets specifically mention that. Probably
>an unreliably controllable process parameter.
>I.e. if you get all your current igbts to work at 50V for an extended
>time, that'd be a small miracle... ;o)
>
>Better stack up more IGBTs. That would solve a lot of problems.
>Maybe there's still room on the back side of the heatsink? :)

Next time ;-)

>
>Anyways, what I get from the is IRG4PC50FD datasheet, as an estimate, is
>300A at 12V Vge. Then around 400A at 14V. 500A at 16V? 600A at 18V?
>Before c-e saturates. Hard to tell as the curves end "too early" and some
>are missing. Voltages look lower for IRG4PC50WD (again, just
>an "extend the graph" estimate :-)

Have to remember that the batteries wear down too!  If they are at 7 volts,
then I have 21 volts still very safe.  With two batteries I have only 14
volts and hot IGBTs!

>
>Have you measured that they really do saturate?

750 amps.  I don't think more gate voltages matters after that...

>And it isn't due to unequal current sharing?

No, too much current, not enough IGBT active region.  The little IGBT just
can't control any more current with its gate structure doping regions and all.

BTW - Even though I am pushing enormous currents, the total power
dissipation is only 5 - 10 watts (low duty cycle).  Thus the IGBTs run very
cool.  That really helps :-))  But the pulse may superheat the wire bonds,
shatter the die (thermal shock), or generally tear the IGBT apart.
Hopefully, relatively long term failure mechanisms...  Even the magnetic
forces could do damage.

>
>> The IGBTs "can" turn on or off an "any" time.  The control circuit turns
>> them on and off nicely but the coil is designed to withstand errant
>> triggers too.  So I could turn the IGBTs on and turn them off in the middle
>> (peak voltage) of the next cycle or something like that.  So to keep the
>> thing reliable, even under fault conditions, it has to take any IGBT timing
>> signal and survive.  I clamp the gates with high voltage transorbs now but
>> they are on the "other side" of the gate resistors.  Thus, a high dV/dT on
>> the collector might be able to drive the gates high if the resistors are
>> too high of value.  Of course, lowering them increases dV/dT too....  The
>> TP250s are no light weights either.
>
>Negative gate bias may be helpful there. Negative bias is a good thing in
>any case. :o) Should be easy with the TP250s too, just one 9V battery
>more.
>
>> One "very bad" signal occurs if you turn the IGBTs off while the primary
>> current is maximum.  That give a BIG positive voltage kick right across the
>> IGBTS!!  ""7 Joules!!"" stored on a single turn 404nH coil!!  BIG dV/dT
>> there!!!  I have transorbes on that too but the voltage can jump from say
>> zero to the clamp voltage terribly fast and perhaps drive the gate voltage
>> up with it.
>
>Yeah even fast transorbs are like >5ns clamping response time so with
>really huge dV/dt it may not clamp fast enough to prevent Vc-e
>overvoltage. And then the capacitive coupling to the gate too (again,
>negative bias helps a bit)...
>
>OTOH, you could trade this in with slower igbt turn off (but still use
>fast turn on). If interrupted (over)current doesn't happen repetetively
>then just _maybe_ the igbts won't explode. More igbts, anyone?  :)

Next time, more IGBTs and more TP250's so I can drive a bit more gate
current.  But I am still fairly safe for the "first" OLTC try ;-))  But if
the IGBTs do fail, I should know right away :o)))  Probably at about
4,000,000VA!!!!


BTW - I just got my TEK P5205 High-voltage differential probe :-))  So gate
and CE voltages will be real easy to see now!!  Differntial two probe
measurements of high speed things just don't go well and digital scopes.
Really needs a real differential probe. 

Cheers,

	Terry



>
>cheers,
> - Jan
>
>--
>*************************************************
> high voltage at http://www.hut.fi/~jwagner/tesla
> Jan OH2GHR
>