Max current of a CM300, place your bets now! / Tesla Coils / Forums

Forums

4hv.org :: Forums :: Tesla Coils

« Previous topic | Next topic »

Max current of a CM300, place your bets now!

previous 1 2 3 next

Move Thread

LAN_403

Steve Conner

Mon Sept 22 2008, 09:04AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

As you increase the current, the MTBF will go down. It might last 100 years at its rated current, 10 years at twice its rated current, 1 year at 3 times Icm, 1 month at 4 times Icm, 1 minute at 10x Icm (with suitable gate overdrive) and so on.

So you can get any answer you want by doing the experiment for an appropriate length of time.

The only figure that the manufacturer actually warrants is 1x Icm, though, and we don't know the actual scaling law of MTBF vs. overdrive with any degree of accuracy. Steve's latest explosion is just one more data point, and we would need hundreds, and the data wouldn't necessarily be valid for other brands of IGBTs.

aonomus

Mon Sept 22 2008, 11:36AM

Registered Member #1497 Joined: Thu May 22 2008, 05:24AM
Location: Toronto, Ontario, Canada
Posts: 801

Killing that many IGBT bricks would be a horrible thing to do....

The biggest explosion I got was of unknown energy when a 555 timer relieved itself of its plastic DIP8 casing and fried the comparator next to it... no idea why either. I'd hate to be *near* a 1kJ 'event'.

HV Enthusiast

Mon Sept 22 2008, 12:53PM

Registered Member #15 Joined: Thu Feb 02 2006, 01:11PM
Location:
Posts: 3068

Steve Ward wrote ...

We have a result!

After increasing the gate drive to 30V, the IGBT exploded VIOLENTLY at 5230A, thats right FIVE THOUSAND amps. The last scope trace captured shows some signs of desaturation as the last 2 current peaks are equal in amplitude. Im still a bit shaken up after having a 1kJ explosion go off about 6 feet away.

Pictures coming shortly!

Steve,

It may be more beneficial to measure the capability in energy, rather than peak current.

I would make a guess that your peak current is strongly dependent on pulsewidth. You could probably put 100,000 amps peak current through that thing with a short enough pulse!

Steve Ward

Mon Sept 22 2008, 11:13PM

Registered Member #146 Joined: Sun Feb 12 2006, 04:21AM
Location: Austin Tx
Posts: 1055

As you increase the current, the MTBF will go down. It might last 100 years at its rated current, 10 years at twice its rated current, 1 year at 3 times Icm, 1 month at 4 times Icm, 1 minute at 10x Icm (with suitable gate overdrive) and so on.

So you can get any answer you want by doing the experiment for an appropriate length of time.

The only figure that the manufacturer actually warrants is 1x Icm, though, and we don't know the actual scaling law of MTBF vs. overdrive with any degree of accuracy. Steve's latest explosion is just one more data point, and we would need hundreds, and the data wouldn't necessarily be valid for other brands of IGBTs.

I understand this, there are mechanical failure mechanisms involved with power semi-conductors, modules in particular. The wire bonds can fail, the thermal stress on the dies can make them de-laminate from the ceramic insulator, etc... But, these failures seem to be more pronounced with low frequency thermal cycling from what ive heard. I havent yet found much information on what goes on with very short, but intense thermal cycling.

But instead of being overly pessimistic on this topic of overdriving semiconductors, im going to actually do real tests! My thermal calculations suggest that the die is not really peaking at unreasonably high temperatures, in fact its quite modest. So, how does the IGBT really know that its running higher than Icm provided its not getting hotter than what is specified as "safe" by the datasheet? Does a faster dT/dt cause more problems mechanically? Because really, thats the only difference here. The absolute die temp isnt breaching anything considered un-safe, even at 2000A peak.

I'll also reference the 10's of HOURS of run time on my CM300 bridges at 1500A peak. I know statistically they will fail at some time, but if that statistic says its 1000 hours from now, then i dont care. And as i said, instead of being so worried that i dont even try, im gonna see what happens when you run a CM300 at 2500A peak in tesla coil use.

Steve,

It may be more beneficial to measure the capability in energy, rather than peak current.

I would make a guess that your peak current is strongly dependent on pulsewidth. You could probably put 100,000 amps peak current through that thing with a short enough pulse!

Well, i was testing it with pulses that i expected to see with the application im designing for. Really, the current is limited by de-saturation of the die, and that began to show up at 5200A with 30Vge. So even if you pushed the gates to 35V or 40V, you'd still hit a brick wall at maybe 7000-8000A would be my guess, regardless of the time scale.

But, in any case, i have a clean captured current waveform, from which i can possibly use with pspice to model the energy consumed by the IGBT in that shot. Because indeed, the die melting is in fact the real failure mode here! Indeed, i want to know how much heat the die can soak up in a short time frame before it fails. I just made sure that the time frame was relevant to tesla coiling.

And of course the other "fault" of my test was that the low pulse rate allowed the dies to cool back down to room temperature, but that is being considered.

EDIT:

Ok Mr. Conner, id like to get your opinion on some things... First, you may want to look at this document that discusses how to calculate the junction temp of an IGBT based on peak power dissipation and time constants:

Now, assuming my Pspice model is pretty close to correct (and i think it is, the Vce sat looks right, and the switching losses match up OK with the switching loss table provided by pwrx), then with a linearly rising primary current peaking at 3500A at 400uS gives me about 1.743J of loss in the IGBT die (per shot). The peak power is roughly then 1.743/.0004 = 4357.5W. According to that PDF, the delta Tj = Zth*Rj-c*P. Zth at 400uS is about .04 and Rj-c is about .06C/W, this makes Tj spike up only 10.5 degrees Celsius. Pshhhh!

Now have a read here:

Page 24 mentions that if delta Tj is less than 30*C, then your IGBTs will basically never have a problem with mechanical things like de-lamination from the base plate, or bonding wires failing. So even if my power dissipation is off by a factor of 3, im still pretty much OK according to the guys who make these things.

So what am i missing? Why cant we run these IGBTs at 2000 or 3000A peak provided its relatively good soft switching? I mean, the switching current isnt really that much worse than it was designed for (few hundred amps). They dont latch up. They dont desaturate if you drive the gate right. The junction really aint that hot.

Thoughts?

I might just contact a powerex engineer and ask what they think, and settle this once and for all. Running these IGBTs within their peak ratings, at our duty cycles and soft switching seems to be an absurd waste of silicon.

One more EDIT:

I just ran a simulation of the IGBT falling out of saturation. In one quarter cycle, the energy consumed by the die jumps from about 1J to 12J. If that happens in say 10uS, the peak power is roughly 1,100,000W. The calculated delta Tj is 144*C, on top of 25C room temp plus 10.5C from the normal die heating. Well surely enough, that exceeds 150*C on the die, and it fails. So the numbers all make pretty good sense, even as rough calculations.

Steve Conner

Tue Sept 23 2008, 11:00AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Hi Steve,

Running within the peak rating might be an absurd waste of silicon, but it saves a lot of time and effort in trying to figure out what a new safe peak current might be. My time and effort is worth money, so maybe I just chose to use the money to buy extra silicon and get on with making sparks.

When I worked on DRSSTCs, my main interest was improving driver reliability and studying how driver problems affect reliability. I chose to run my experimental coil within its peak ratings to eliminate overdrive as a variable. In other words, if the IGBTs blew, there would be nothing to blame but the driver.

I assume that the manufacturer knows a lot more about IGBTs and their failure modes than I do or ever will (for instance, did you know about electromigration? there will be many other failure modes which they probably keep confidential) and he chose the Icm rating wisely based on that knowledge, and likely on lab tests that involved destroying hundreds of IGBT dice.

So, to make any claim for a peak current rating that I'd take seriously, you'd have to repeat those experiments yourself, and that would be an absurd waste of silicon.

So I say by all means overdrive your IGBTs as hard as you like, but don't claim that you've found "The real peak current rating that Powerex somehow missed" and don't be surprised if they mysteriously blow out now and again for "no apparent reason". Also, Powerex might not want to tell you anything, if they couldn't answer your question without revealing the kinds of trade secrets I hinted at earlier. Then again, they might, so go for it

FWIW, I ran the CM600HA-24H bricks at about 4000A peak each in my old OLTC2 and never blew any up. That's only about half of the current density you used there, though, and I started nearer 2500A and cranked it up over the life of the coil.

I was partly inspired to build the OLTC2 by Greg Leyh's papers on testing IGBT bricks for pulsed service at SLAC.

aonomus

Tue Sept 23 2008, 02:19PM

Registered Member #1497 Joined: Thu May 22 2008, 05:24AM
Location: Toronto, Ontario, Canada
Posts: 801

So.... pics?

Sulaiman

Tue Sept 23 2008, 07:09PM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

Could the failure be due to dI/dt x L(emitter) ?

This would cause an effective reduction in gate drive voltage,
maybe even to the point of linear operation (not saturated).

This may also be related to why 'excessive' gate drive voltages help drsstc stuff,
rather than the assumed benefit of lower Rds(on). ( Vce(sat) for igbt )

Steve Ward

Wed Sept 24 2008, 03:42AM

Registered Member #146 Joined: Sun Feb 12 2006, 04:21AM
Location: Austin Tx
Posts: 1055

Here's some pictures of the test setup including scope shots and blown IGBT pics:

Could the failure be due to dI/dt x L(emitter) ?

Not too likely, they provide a kelvin emitter connection.

This may also be related to why 'excessive' gate drive voltages help drsstc stuff,
rather than the assumed benefit of lower Rds(on). ( Vce(sat) for igbt )

I didnt think the higher gate V was to have lower Vce sat necessarily, but to keep the device from desaturating at really high currents. You really dont gain a whole lot of Vce at lower currents by having more gate drive. If you look at the traces labeled 5230Apk in that folder above, you will notice that the last 2 negative current peaks are just equal in amplitude. This is almost definitely because of desaturation of the IGBT. The losses go up enormously as Vce drop approaches 600V to keep the current from growing higher.

Steve Conner

Wed Sept 24 2008, 09:24AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

The L*di/dt thing is another failure mode that Greg Leyh explored in his papers but Steve likes to ignore. :P It causes uneven gate voltage distribution between the dice in a package, such that some of them desaturate and blow out prematurely, and a single Kelvin emitter terminal can't do anything about that.

The manufacturer tries to minimize it, but he can't eliminate it, short of arranging the dice in a circle, with both power and Kelvin emitters in the middle, to make the stray inductance and mutual coupling symmetrical. Back on the TCML, we debated ways of arranging TO-247 devices on PCBs to get better pulsed current sharing, and hence hopefully more RF amps per dollar, than a brick. It soon becomes a 3-D nightmare of multilayer boards and heatsinking.

Steve Ward

Wed Sept 24 2008, 11:10PM

Registered Member #146 Joined: Sun Feb 12 2006, 04:21AM
Location: Austin Tx
Posts: 1055

The L*di/dt thing is another failure mode that Greg Leyh explored in his papers but Steve likes to ignore. :P

Ok, the internal module inductance is 60nH max (this was told to me by a powerex engineer some time ago). At 5200A, the max di/dt is 5200A*28000hz = 145.6A/uS. The voltage dropped by 60nH is 8.74V or so. I guess that is enough to cause some issues with gate drive, but still i think the mutual emitter inductance is less than 60nH.

Greg's stuff was studying fault currents right? I guess really, anything over 10X the modules rating is in fault-current land, so i should probably read his paper again.

Damn physics getting in the way again.

previous 1 2 3 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob