Burning FET's and driver troubles

Things at the limits.

Re: Burning FET's and driver troubles

Postby JonathanH13 » Fri Dec 10, 2010 7:16 pm

@Doug,

>the bottom one

Yes, that is precisely the shot that is bothering me (pulling the fets out of the on state). I am making all of these measurements with the ground clip of the scope lead connected to the negative line of the power supply
(marked '- Smoothed DC line' in the schematic), because this is a floating supply. Sorry about the focus - I have ordered an optical RS232 cable for that scope, I should have it next week. The blue trace in that bottom scope image is 10V/div.
Thanks for your perseverance with this.

Here is the propaganda from the SPW47N60C3 datasheet:

VDS @ Tjmax 650 V
RDS(on) 0.07 Ω
ID 47 A

* Worldwide best RDS(on) in TO247
* Ultra low gate charge
* Periodic avalanche rated
* Extreme dv/dt rated
* Ultra low effective capacitances

@Chris

The supply rail is fed by 230VAC -> 10Amp Variac -> 10Amp Rectifier -> 2X 4700uF 400VDC electrolytic capacitors in parallel. So that is a fairly direct and hefty supply, I would have thought? I can increase the cap bank to 10X 10000uF 450VDC with some nice 700amp bus bars if you think it would help? What is happening here really cuts to the heart of the problem that I have been experiencing with this project from day one. That is, the plasma shorts out the power supply; the high voltage drops from 40kV down to around 1kV. It's a dead short, effectively. That is why I chose a fairly high frequency (40KHz) when I was doing the voltage multiplier ladder calculations - to prevent the voltage from sagging (before I built the stack). But it seems to make no difference. I guess I need larger caps in the stack, to move more current through at this pressure. I have no idea what current the load is pulling, all attempts to measure it have been meaningless, and I am not keen on trying to measure it at the stack output like Farnsworth did. I measure the current draw on the variac, and have seen this go quite happily up to 10amps, which is scary. I know that this whole issue goes away at lower pressure, but it would be great to have a supply that could deliver, regardless of the pressure.
I have two 3-phase 500VDC 50amp power supplies standing by, with a 7kW 3-phase generator and 12 amplifier modules which weigh 50kg's each. With the CT transformer that is on route, that is over 800kg's, which I think is a nice way to measure power 8-) This arrangement will surely vaporize the grid and probably a fair portion of the tank as well. Doug is trying to assure me that finesse is the better option, but after 3 years my patience is wearing thin.

>a low-ish ESR cap of a uF or two up close to the FET

This is interesting - I have two .47uF low ESR bootstrap caps on the driver circuit, but they are nowhere near the fets. The driver is shielded and is around 6inches away from the H-bridge circuit. If I needed more capacitance near the fets wouldn't we see the gate waveform being pulled down? I have added some details to the schematic. I will post some pics of the entire setup tomorrow...
User avatar
JonathanH13
 
Posts: 53
Joined: Tue Aug 10, 2010 3:20 pm

Re: Burning FET's and driver troubles

Postby chrismb » Sat Dec 11, 2010 3:05 am

Yes, I think you have plenty of capacitance/low inductance there. The proximity of the Low ESR cap does seem to make a difference in my MHz setup, but I can't say if it is critical in yours, maybe not. Are you running a ballast resistor to your plasma load (sorry if I have missed that, if you've already said)?
chrismb
 
Posts: 620
Joined: Thu Aug 05, 2010 6:32 pm

Re: Burning FET's and driver troubles

Postby Doug Coulter » Sat Dec 11, 2010 12:01 pm

If you're measuring that bridge output against ground, the "pull out of on" might not be that, but simply ripple on your supply caps. I keep having to remember than in all my setups, the negative rail IS ground, but in yours, it's not. It would be typical in that case -- no caps are perfect -- and even if they are, when you use energy out of them, they droop.

It should be obvious -- if you're pulling a fet 5v or so out of "off" with 3 amps -- it's going to get warm fairly fast (15w). You could also just scope the power rails and see what it looks like there.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.
User avatar
Doug Coulter
 
Posts: 3515
Joined: Wed Jul 14, 2010 7:05 pm
Location: Floyd county, VA, USA

Re: Burning FET's and driver troubles

Postby JonathanH13 » Sat Dec 11, 2010 2:14 pm

Doug,

Yes, of course, that makes sense. I would like to tie that negative line to ground, but I need to figure out what's happening with the residual current trip circuits in my apartment - they don't like it. I will measure again - I need to go pick up the scope. Myself and a friend are renovating an old pinball machine. It has some solid 80's electronics and some serious electromechanics...

The low down on the transformer:

This transformer has two primaries. They are made from 4mm diameter litz wire, what looks like about 10 or 12 windings each. I thought they were interlaced, but now that I look carefully, I see that primary one is actually wound around the ferrite core first, then a layer of insulation, and then primary two is wound over primary one. I never noticed this before and I guess that it means the primaries do not contribute equally. The dc resistance of the primaries is 0.1ohm each.

The secondary was designed to carry about 0.5 amps - so I presume the wire is around 0.2mm diameter, although I cannot see it as it is potted. The secondary is centre tapped to earth, and this earth wire is connected to the ferrite core by copper plates in contact all the way around the core. The dc resistance of the secondary is 11.5ohms each to the centre-tap, 23ohms all together. The secondary is wound above the primary, on the same side of core.

The core itself is made up of 2 square '[' shaped halves, forming one large rectangle '[]'. There is a 3mm air gap top and bottom of this rectangle - so the '[' shapes do not touch - they have a some sort of 3mm hard plastic shim keeping them apart.

I have some other figures here but I do not know if they are accurate: The input impedance of the primary is low, so that would require the driver to have a low input impedance to match. I have a figure of 19uH for the input, with a self capacity of around 115pF. The self capacity of the secondary winding is more like 35pF. I'm not sure if that makes sense or if these inductance and capacitance values are manufacturer measurements or just estimates. It was designed to resonate at 40KHz, but it looks more like 63KHz based on those measurements in the previous post. The people who built it never had any means of testing it.

The next obvious step here is to run it again with just one primary.
Attachments
xfmer.jpg
High voltage transformer
litz.jpg
Primaries
gap.jpg
Air gap
driver and bridge.jpg
Shielded driver and H-bridge with smoothing caps
User avatar
JonathanH13
 
Posts: 53
Joined: Tue Aug 10, 2010 3:20 pm

Re: Burning FET's and driver troubles

Postby Doug Coulter » Sat Dec 11, 2010 3:30 pm

You won't be able to ground the negative rail without having an isolation transformer as the negative rail is driven below earth ground when either AC input is negative of the earth ground, which I assume in you country is simply a centertapped 240v with the centertap grounded in your electrical service box (that's how our 240v is here BTW, but most of our house wiring goes between one of the mains wires and ground/neutral, we only wire for 240 for real big appliances like stoves). So trying to "earth" that point will result in shorting the line through your diodes, not good for either one. But you said you could float the scope, and hook its "ground" to the negative rail, so as to get a signal measured source to drain on the fet anyway.

(this is why I ragged on you to get an iso transformer -- every shop should have one, it makes things safer to work on anyway)

In other words, your two mains wires are the outside of a 240v center-tapped transformer (out on the street most likely) with its center tap wired to earth ground either outdoors or in your service box (In USA, both places). Either wire then goes both positive and negative of earth by that much -- roughly 168v peak either way for 120v RMS AC on either half.

If you didn't connect your scope ground to a rail, then the bad waveform above is explained by capacitor ripple (which you can measure on either rail vs earth gnd, and you should see some when the bridge is drawing current -- that's normal and unless it's too big, no problem). In normal operation with your bulk DC circuit, you should see roughly equal and opposite voltage rails with "real ground" as a center-tap -- about the aforementioned 168v no load each at full input voltage. So shorting either rail to ground in your setup won't fly -- you're just shorting out the power company in a potentially expensive way.

In an iron core circuit, the fact that one primary is outside the other isn't a big deal at all -- the "iron" in this case being ferrite, but still hundreds of times more magnetically conductive than air, so a little air here and there just doesn't make a noticeable difference -- it's really how many times a wire goes through the hole that matters most. The gap spacing does -- those shims you mentioned. They are there to control how much net "mu" the core has, and to prevent it from being "too good' and saturating so easily. So in fact by changing this shim thickness you can tune the thing all over the place -- a tiny air gap has more "reluctance" (which in magnet-speak is roughly analogous to resistance) than the length of core material does, because the core material is hundreds of times more magnetically conductive than air is, or thousands, depending on the material. The gap is used to tune that parameter overall as far as what the windings see. It's not normally used in AC circuits, but is often used when there is supposed to be a DC component in say, a switching supply inductor, so the core doesn't saturate. In this case it's cheaper to use more wire to get the inductance back up than tons more core material.

I couldn't easily tell what the square area is on that core, but it looks like about 1 sq inch (sadly, I think distances in "English" while the English have gone metric, but I can understand either).

I'm able to get about 8v/turn without saturation on cores here about that size with no gap, for what it's worth -- if yours is bigger than that in square area, you should be able to get about that much more volts/turn without saturation. In your case, reducing the gap makes it easier to saturate re current (ampere-turns), but willing to take more volts/turn before the same current is drawn (because of the higher inductance you have with more mu). If it turns out that you can't get high enough in drive frequency to make this guy truly happy, reducing the gap would bring the resonance down to where the driver could be happy. In realty, you can't really get "zero" gap as nothing ever fits perfectly, and even .001" is significant with high mu ferrites.
For fanatics about that one, they do make a ferrite loaded epoxy but that's over the top for this -- a small gap is probably what you want anyway to keep the resonance F high.

I have a similar transformer from a Spellman, and while it has more than one resonance, the main one for a lower voltage secondary than I think you have, is much lower than you are reporting (more like 20khz). Something isn't quite right here, yet. You might be wise to extend your measurement with an oscillator down to 10khz or less, and up to a couple hundred. That sloppy made Wallis I have shows about 3 significant resonances, as in that case there enough weird airgaps and two huge (8" diameter) secondary pie windings that all interact with one another in "interesting" ways -- everything has various stray C's to ground and to each other and a lot of leakage inductance and even coupling between the two secondaries. As luck would have it in that case, the main resonance is well inside where the driver is happy, and there's enough leakage inductance for the other resonances to still occur -- right at 3 and 5x the lower one, so it's perfect for a square wave drive(!) -- my Spellman that looks more like yours doesn't have that "feature" and is much tougher a load on the drivers without that series resonant LC in series with it (which I'll have to add externally). It seems to be more or less "built in" to the Wallis transformer, which isn't potted and is very sloppy, made out of a bunch of short square sticks of ferrite, loosely held together with gaps all over the place.

I note you have the other leg of the magnetic circuit exposed and available. I've gotten fantastic results by putting a few of the primary turns on that other leg so as to have some leakage inductance between primary and secondary, and Glassman does that on purpose in their designs. Makes it a lot easier to drive that fast edge into what is otherwise a fairly high Q tuned circuit -- the leakage L there means it doesn't present such a low impedance load to the driver for that part of the spectrum. I've had it cut quiescent current from hundreds of ma to a few, and everything gets a lot happier at that point.

Until you get into heating issues, you can just use one primary for testing and it may work better that way anyway. No need to parallel them till you're getting to high enough input currents to heat the wire ohmically. At any rate, you shouldn't notice much difference in anything with just one compared to both in parallel unless something else is wrong.

(I'm going to find a nice looking hysteresis curve for ferrites and put it up here to illustrate something -- you really shouldn't be seeing that rise in resonance with voltage with that big core and big gap unless you're well outside the ratings of the transformer, or pushing them real hard). Your rise in resonance indicates (if too large) that you're getting to where the BH curve bends over and so the material looks like lower mu (a straight line from the origin has less slope plotted as B vs H).


Yes, plasma is a hard load to drive indeed, and the dependency of current draw on gas pressure is intense to say the very least. This is why we use a series ballast resistor in the 50-100k range and spend so much time with gas control. The way it works out, we're working on the very steep part of the left-hand side of the Paschen law here.

751px-Paschen_Curves.PNG
Paschen's law from wikipedia


If there's too much gas, current will rise more or less without limit until evaporated metal atoms get involved, at which point it gets to a pretty good short circuit (arc welder -- 20v, 100's of amps). As you go down in gas pressure, there's not as many ions and electrons to carry current, and less of the gas is ionized as a proportion of the total gas present too, so in a narrow range, you can drive it with reasonable kit.

A plasma has negative resistance -- the more current you try to put across it, the lower the voltage goes, due to more current carriers coming into play -- this is on top of Paschen's law, so the plot above isn't the whole story at all. We are trying for the opposite extreme from an arc welder here, and even a moment of high peak current (from the stack capacitors) can take you to "arc welder" conditions. You need an absolute current limit (series resistor) of under an amp or so at these sizes to prevent that happening every single time it lights off. Which is why you can make a neon lamp blink with a series R and a parallel C. The arc mode from peak capacitor current drives the voltage on the cap well below the original "strike" voltage of the lamp, which goes out as soon as the cap can't bring all the big currents into it.
The lamp then goes out, and the cap recharges to the strike voltage again, and the cycle repeats. Same thing can happen in a fusor, and often does. In the case of the lowly Ne-2 bulb, the strike volts are about 110, and the stable sustain volts at very low current is about 70, but during a capacitor discharge can go much lower while the current is high.
Here's a link to two books about just this phenomenon.

In my fusor, without an ion source to help, the entire range of gas pressure I can run does not change the least significant digit in a two digit (plus exponent) gas gage display. It's that critical. If you can't get to "can't light it off" vacuum, then that's your next issue -- you really want to get at least order of magnitude better vacuum than that when you're not putting in gas, to be running in reasonably pure conditions when you are. Two orders is better yet.

With an ion source, I can hugely increase the working range towards lower pressure, and therefore higher voltage at reasonable currents. In that case I can go down to 1.6e-2 mbar up to 2.2 e-2 millibar. Without the ion source, it just goes out below 2.2 e-2 mbar. This is so repeatable and so stable, you can use it to calibrate a gas gage against (in your case, directly as we have the same basic fusor dimensions). But that's another thread.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.
User avatar
Doug Coulter
 
Posts: 3515
Joined: Wed Jul 14, 2010 7:05 pm
Location: Floyd county, VA, USA

Re: Burning FET's and driver troubles

Postby Doug Coulter » Sun Dec 12, 2010 4:49 pm

Here's what you need to see on your scope to make me (and more importantly your semiconductors) happy.
DS0008.gif
Good operation traces
DS0008.gif (5.66 KiB) Viewed 5761 times


Upper trace is scope directly between source and drain of one of the lower fets. Lower trace is a probe held near one of the secondary output wires.
In phase, no nasty glitching, etc. Note while the secondary isn't a square wave or sine wave, it's closer to the latter.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.
User avatar
Doug Coulter
 
Posts: 3515
Joined: Wed Jul 14, 2010 7:05 pm
Location: Floyd county, VA, USA

Re: Burning FET's and driver troubles

Postby JonathanH13 » Sun Dec 12, 2010 5:02 pm

@Chris: Yes, I have a ballast resistor - it is 47K at 1000Watts

@Doug:

The ferrite core has a cross sectional area of 1 inch square, and there are actually two of these cores side by side, to
make a 1" x 2" dimension - which I think is a pretty respectable core. So you are saying then that I should be able to get about 16v/turn without saturation?

>You won't be able to ground the negative rail without having an isolation transformer

Right, that is what I suspected. I have 3 isolation transformers, but they are all 240:120 step-down. I suppose I could put two in parallel?

>In other words, your two mains wires are the outside of a 240v center-tapped transformer

I thought it was a standard (non-centre tapped) transformer with one of the wires (Neutral) tied to ground at the sub station? To be honest, English wiring is a mystery to me - it's almost as bad a English plumbing. :| If the mains wires were the outside of a 240v center-tapped transformer, then you would get shocked if you touched either wire, but you only get shocked when you touch the Live (and I seem to touch it more often than I would like!)

>the bad waveform above is explained by capacitor ripple (which you can measure on either rail vs earth gnd, and you should see some when the bridge is drawing current

I have checked the rails under load (30KV at 'high pressure' - around 200mbar) - see pic 1 & 2
I also remeasured those gate/output waveforms - looks like you are right, I just need to measure with respect to ground. (pic 3 - using only one primary)

>by changing this shim thickness you can tune the thing

Yes, that is exactly what the manufacturer said. Given the frequency limits of this chip, I'm inclined to leave it as is. I have extended the measurement with an oscillator at lower and higher frequencies as you suggested, in orer to see if there are any harmonics. Result - no, nothing jumps out at me - at 70KHz upwards (to 2MHz) there is no significant change.

>I've gotten fantastic results by putting a few of the primary turns on that other leg.

I guess I could find some litz wire and put in a primary extension without disturbing the existing windings. Worth a try?

>you really want to get at least order of magnitude better vacuum than that

Ah, yes, the vacuum is about the only thing that is OK:

I have an e2m12 edwards foreline pump that is rated to 2.5e-4 mbar. The entire system gets down to 8e-3 mbar with this pump alone (measured with a very respectable and calibrated vacuum gauge that was on loan).

The diff pump is a VHS-4 Varian with a flow rate of around 900 litres per second when the cryotrap baffle is fitted.
With this pump running, the entire systems drops to 1.2e-5 mbar. The plasma quenches long before this point, even above 40KV.

I reconditioned both pumps and primed them with appropriate high quality pump oils. The diff pump is connected directly to the chamber with only the -110 degree centigrade cryotrap in between (the butterfly valve has a 120mm diameter gate, so there is huge conductance. The pump is massively overated for the 14 liter chamber, it is difficult to throttle and pulls out any deuterium in short order - so I'm look forward to switching to turbos...

All these power tests are done at residual vacuum, without running the pumps. Is it correct to assume that if the supply can run at pressure, it is only going to perform better lower down?

I understand your point about an additional ion source - I think that will help a lot.

Oh, I fixed the focus on this camera.
Attachments
power rail 2.jpg
Capacitor rail under load, measured @ positive rail and scope gnd to negative rail
power rail 1.jpg
Capacitor rail under load, measured @ positive rail and scope gnd to earth
Red = positive rail, Blue = negative rail
output.jpg
Red = gate, Blue = output (drain) @66.5KHz, scope gnds to earth, with plasma load
User avatar
JonathanH13
 
Posts: 53
Joined: Tue Aug 10, 2010 3:20 pm

Re: Burning FET's and driver troubles

Postby Doug Coulter » Sun Dec 12, 2010 5:36 pm

I'll defer to a Brit on how your wiring is done there. So you're saying one side is effectively ground/neutral? News to me, but...easily verified with a voltmeter referenced to real earth. Here our neutral will typically read a couple volts off ground due to currents and R drops when the circuit is powering something else. But only a couple volts.

Yes, if you've got the more or less "standard" ferrite they use for this, you should be able to get on the order of 16v/turn. Depends on some other things like total length and gaps, but that should be close.

Until you get to full drive voltage, you could use an iso that had the stepdown, or for full output you could use two in stepdown with their primaries in parallel and the outputs in series.

I'd suggest also measureing down from where you are. I think I've never seen a HV secondary resonate as high as 60khz, but I've seen some secondary parasitic peaks up there. There should be some point where the drop across the series R in your test setup goes really close to "nothing" -- with nothing connected to the transformer outputs.

Using similar cores, and hand winding, the highest I've ever gotten is 70khz -- with "count them" 100 secondary turns. Not thousands. Glassman's little transformers that look like TV flybacks are 12.5 khz, spellman's big 10kv CT output -- 18 khz. Wallis -- 42 kv (cubic foot dual flyback, litz, not potted which means less C).

Right -- see how sharp that curve is rising? It's so steep as a make a good calibration point! Now, we are finding that in fusors this size, we do better nearer the lower pressures, all the way to "ion source only " currents are being drawn, in both Q and total output. Counter-intuitive, but I think Chris might chime in and say "because you are no longer working with collisions in a sea of stopped neutrals" and I'd agree. But we should talk fusors under "fusors".

Up to a point the supply will work well lower down, it's a different problem space there. Up at high pressure
it's nearly a short, as you noticed, which tests that aspect of things. But doesn't test if you're going to have arcing etc, or get it stable at light loads and higher volts. Not much free in this game, you gotta test it all nearly always.

Yes, scope traces far better, now I can read the legends too.

40v ripple is a bit excessive on the cap rails. Shoot for under 10v pk pk and things will like that better.

That bottom trace -- hopefully we're seeing mainly drop on the power supply if you're using 'earth' for the scope ground rather than the fet source terminal. If that's the case, you're fine now I'd guess. Otherwise looks "in good operating range" for these things.

I have a new ion source design to build the instant I finish getting the pump driver board running....
See recent post in RF.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.
User avatar
Doug Coulter
 
Posts: 3515
Joined: Wed Jul 14, 2010 7:05 pm
Location: Floyd county, VA, USA

Re: Burning FET's and driver troubles

Postby JonathanH13 » Sun Dec 12, 2010 5:53 pm

Ah, I just saw your post and ran that same measurement. It looks pretty close.

The first pic is at quiescent current, and is about as near as I can get to resonance (just below by a few hundred Hz without everything going haywire (no load). Upper trace is scope directly between source and drain of one of the lower fets. Lower trace is a probe held near one of the secondary output wires.


The second pic is at resonance - the tank actually 'whines' at a rather unsettling 'imminent death' kinda pitch, and the waveforms go crazy. (no load).

The last pic is under load at a pressure that my thermocouples read as 5 microns - looks like a phase shift...
Attachments
1.jpg
260mA on variac, -25KV
2.jpg
270mA on variac, -25.5KV
3.jpg
3amps on variac, -1KV
User avatar
JonathanH13
 
Posts: 53
Joined: Tue Aug 10, 2010 3:20 pm

Re: Burning FET's and driver troubles

Postby Doug Coulter » Sun Dec 12, 2010 6:25 pm

Excellent pix (and operation).

Hard to explain that whine being anything in the audible region at all, even with 100hz ripple on the supply. But yes, if it's making noise, that's not a happy condition for some reason. Could be a kind of loose primary turn plus some acoustic resonance in the construct that's getting hit at 100hz and ringing at some more audible frequency?
At any rate, that implies motion someplace that will eventually make trouble. "If your arm hurts when you do that, don't do that" ;)

The phase shift under heavy load is in the direction I'd expect -- this is actually operating according to "theory" now, great! When you draw secondary current, it cancels some of the primary volt/amps, which lets the ferrite run at lower magnetization, where the BH curve is steeper and the inductance is higher, so the resonance drops. Hence the lag under load. This is a good thing as now "theory" becomes useful as a tool again should anything else need work.

This means you're getting near magnetic limits at no load, which is worse than full load for that -- and eventually you might want to tune this for "typical load" but I'd make no load happy by default myself unless that caused other problems.

We learned this one the hard way, buying some X ray supply transformers that only wouldn't burn up if fully loaded, and even then only run for a second before smoking out of oil Perfectly fine in their intended use -- in oil, with fractional-second power-on times, always full load (preheated tube filaments) and plenty of time to cool between so some guy could come back in the room and change film -- not so good for this work.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.
User avatar
Doug Coulter
 
Posts: 3515
Joined: Wed Jul 14, 2010 7:05 pm
Location: Floyd county, VA, USA

Previous

Return to High Voltage, High Power

Who is online

Users browsing this forum: No registered users and 4 guests

cron