Induction heater questions

Check my current progress here:

formatting link

Looking for suggestions on fast IGBTs suitable for offline operation (my goal is a 240V operated unit capable of perhaps 10kW), so 400Vce or so, and frequency up to 20kHz, possibly more, so they'll need to be fast.

I heard IGBTs have a bad habit of latching, is this true? What other hazards should I expect, and how to handle them?

My driver circuit probably won't like the extra Cg so I'll accept any suggestions for half or full bridge driver circuits or ICs too.

Speaking of bridge, what is the advantage to half vs. full bridge in the final version?

I'd also like something more comfortable than a scope and crude frequency control - anyone have suggestions for a good PLL (to control phase some amount above resonance, to control power output) and/or voltage regulator (to keep tank voltage constant - without dropping below resonance!)?

Oh, and what inductance should I expect to use for Lmatch, and since I won't like the size of an air core unit, what kind of core and how many turns would you suggest?


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams
Loading thread data ...

While we are on the subject, can some kind person tell me the equation for the theoretical mutual inductance (or coupling coefficient) between a vertical circular loop mounted above a horizontal ground plane of virtually infinite extent and thickness and with known conductivity.

From the geometry, I feel it in my bones, it will be simple in form, like -

M = Function( D, H, d, C ) micro-henrys

where -

D = loop diameter, H = height of centre of loop above ground surface, d = diameter of loop conductor, C = conductivity, or resistivity of loop material.

It may be the conductor diameter is not required, the conductor being merely a very thin filament.

Eventually, I wish to calculate the mutual inductance between the loop and the ground and the loss resistance induced in the loop.

And I shall be eternally grateful.

Reply to
Reg Edwards

Nothing eh? Too hard? Too easy? Too boring? Not boring enough?

*Sniff*... and I even took pictures of *glowing metal*!...


-- Deep Fryer: a very philosophical monk. Website:

formatting link



Reply to
Tim Williams

Thanks for the reply :)

Yea... too bad the camera sucks in low light. Most do..

Ok. I've been looking at turn-off times and there's always a 60-200+us delay in there, does that exist or do I care about it?

Ok. I'll have to use two transformers then (if I use transformers). Hmm, duty cycle will do weird things to the voltage levels if I try that...

Hmm, okay, good thing I've been eyeing 600V devices then. Those should be better than 400V devices which "should" be fine for 240V mains.

About like I've been doing, but at the transistor? But humm, negative gate volts...floating power supply, charge pump???

For sure. BTW how much value is there in snubbing? At the moment I've just got junction capacitance of the 6 FETs and that works fine for the speed they switch at, apparently. Snubbing dV/dt would certainly reduce Imiller.

*Gong sound* (parasitic ringing, that is.)

I've got a whole impedance matching system going here, so that doesn't really matter. If I took the other half bridge and paralelled the transistors to one side, I could double the signal current with half the Vds loss, at the expense of rewinding Lmatch with heavier wire and fewer turns. Or is the better idea to go with higher signal voltage?

For sure it'll be rather uncomfortable to have the coil bouncing between goalposts...

Um... okay. Anything *I* can do?

'Course the problem is I'm sticking things in the coil and f changes ;-)

1uH, humm I'm looking at maybe 100uH.

I seem to remember Watt Sun hacksawed(!) either ferrite or powdered iron cores at one time. Had to be P.I., ferrite is ceramic...

My thoughts exactly. :o(

Howzabout this, assume Q = 5 to 10 (with unloaded tank Q circa 30-50). For

20uF x 4uH at 18kHz, that's around 2-5 ohms effective impedance, no?

Of course Ctank is subtracted by Lmatch, since it's series resonant feeding parallel resonant, sharing the same cap...I smell a simultaneous equation...


Ah interesting. I did something like that with a MOSFET and two power supply rails (think flyback converter pumping current from -V to +V), but nothing so dangerous to a cap as that. ;-)

Too bad I have a Tek 475. Guess I'll be getting out the FETs and function gen.

Yup, ultimately ending in the truth that a core can only handle so much energy until you have to buy a larger one. ;)


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

very good. turn the lights out, disable flash and take some more, that would look really neat.

look at the switching energy vs switching frequency curves to get an idea of how fast the IGBT is.

?! dunno about latching. But you must ensure you win the pissing contest against Cmiller. -ve gate bias is invariably used for all but the smallest IGBTs, for exactly that reason - who cares if Cmiller gives a

10V jump in Vg if it starts out at -15V.

IGBTs really dont like being over-voltaged - kaboom. ultra-low inductance from die to bus cap is the key here, along with generous safety margins. google groups will show you a discussion I had with Win about a 6kV IGBT used with rectified 3.3kV mains....

formatting link

I've successfully used some IR half-bridge driver chips with little IGBTs. make sure you have plenty of capacitance on the high-side power supply, and a complementary emitter-follower is a good idea too.

depending on the IGBT, different Rgon and Rgoff can be useful. switching energy curves provide the clues here.

gate drive inductance exacerbates miller effect, by increasing the effective impedance seen by the IGBT for fast transients coupled thru Cmiller.

clamping max Vg is a good plan too. I worked on one design where 6 300A IGBTs were direct paralleled, at the end of 1m of ribbon cable. A sizeable active clamp was required to prevent IGBT destruction. A later version used an emitter follower at the IGBT (1mm from terminals) with bucketloads of supply rail capacitance, clamp diodes etc, with a source-terminated ribbon cable feed. And a humungous TVS whose job was to blow the gate bond wires in the event of a C-G short, thereby preventing destruction of the upstream circuitry.

2x power from full bridge (2x voltage). Design-wise there is little or no difference, the full bridge is just two half-bridge circuits. With a half-bridge you may need to balance the cap voltage (one winding connects to half-bridge, the other to DC bus cap mid-point).

thats a bit trickier. a pulse-by-pulse current limit circuit is pretty much mandatory if you want a robust design, but even that can snot IGBTs. micro + V,I,T sensors + detailed thermal model = best solution

as for the actual controller, cant really say without a detailed analysis. the problem is the symmetric nature of the resonance curve. at a guess, design the VCO so it cant ever produce a frequency below resonance, then convert AC Vtank to DC and control with error amp.

?! at those frequencies you'll probably be stuck with gapped ferrite, kool-mu or MPP. Vishay have a range of smt high current chokes designed for VRMs (cpu buck converters), you can get a 1uH part that saturates at

100A or so. ETD cores are easy to wind and gap.

Toroids are a pain to wind, and cant be gapped without special cutting equipment. I once tried a wide variety of diamond and tungsten-carbide saws to do this. All I did was rip the teeth off, barely scratching the ferrite. I gave up, and found a company that specialises in that.

as for what Lmatch you need, I'd need a bit more information before guesstimating that.

its easy to cobble together a choke without any calculations. Just do a "splat test" - google groups search me as author, s.e.d. and "splat".

That will give you both the inductance and saturation current in one easy measurement. a few tests with different N shows NI = constant. As N increases, L increases but Isat decreases. Then bung in a gap (0.1mm -

1mm) and watch L plummet as Isat skyrockets. Finally, take the cores away, to see Lair_cored which wont saturate at all.

you'll find there is a law of diminishing returns - more turns means higher L but lower Isat requiring a bigger gap, thus lowering L. IIRC there will be a maxima for any given core. Plus of course the R eventually gets in the way, as does the finite winding area.

Cheers Terry

Reply to
Terry Given

That probably explains why I've seen designs for heaters running full coil current through the transistor (or SCR). Current is low because voltage (and turns) are high. I'm doing ZCS at turn-on, but ZVS in general (assuming it turns off faster than swing increases Vce, which can be tweaked with snubbers until slower than the transistors turn off).

Ah, true. I'll have to get some bigger zeners. (It has to handle turn-off and miller current, right? Or is that forward-biased?)

for some reason the IGBT desats, Idesat will increase about ten-fold, so >

Lboom drops to 200nH. Thats not so hard to achieve.

Ok... so IGBTs don't have the happy avalanche characteristic that I'm so used to in BJTs and FETs... Suppose that's what the TVS's are for eh?

Ok, so if I slowed the collector swing massively and made a relatively stiff drive transformer, I could keep the zero-volts-is-off deadtime circuit I use now?

Wow, doesn't that result in some hefty dissipation?

I don't mean zobels, I mean raw capacitance, which I can do since I'm not taking a big bite out of +340V. ;)

- Yeah, that's what I mean. Not a problem as I don't intend on running anything but an induction coil with the induction heater!

You sure? The inverter itself has a peak V, I output (namely 340V 30A real power, who knows how many switched VAr from the matching inductor) and the load is a resistance.

Meh, winding is just fine. I'll probably find myself with 3/8" tubing in the 10kW version anyway. If you mean by the inverter, it's not going to be more than 30A or less than 340V since that's all it's getting!

Yeah, but my three-pronged o-scope would have a heart attack, to put it mildly!

...Anything *else* I can do?

Heh heh! Ever try some sharpened ferrite for a tool bit then? Sounds like it could make quick work of alloy steel. ;-)

Well, there ought to be a pic of the coil on my site. Calculated Q of the coil is about 60, I don't know what the caps are like (hmm wonder if the datasheet says, Digi-Key PF2224-ND) but I'd assume as much. That would bring parallel Q to 30, no?

Gapping does make it gradually less of a core after all. The point of no returns is only when a gapped core suddenly turns to no core at all, something which doesn't happen in good, continuous reality!


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

Needs to be a big cap I guess?

Oh, I mean a TVS mounted at the device terminals, C-E. To protect it from those evil voltage spikes...

...Eww... PC boards...

I prefer to run wires through air, and have no problem with soldering things as close to the die as possible... didja see the hookups on my heatsink? I put the gate drive twisted pair right on the transistors.

The source/drain hookups leave a bit to be desired, okay a lot to be desired, but it's out of the control loop, just acting in series with Lmatch.

I will be using much copper tubing and/or sheet later. :)

Ok, so let's see, if I setup the 3524/494/7500 (and whatever other equivalents) SMPS chip to do 25 to 75% duty cycle, with positive peak voltage always above 15V, then it'll be +15/-45V at 75% duty cycle, +/-30V (60Vp-p) at 50% and +45/-15V at 25%. With DC restorer that could come down to 0V negative peak, but positive is still loopy without something to cap it.

I could always rig a small DC-DC converter to supply +/-15VDC for each half of the totem pole and use my existing circuit to run a driver, but geez...

I'd think that would be even worse?

Thanks :)


Grinning, ducking and running...

Ok, that's one... but I need one for each transistor, unless LM339 can handle +325V on one section's inputs. ;)

At that rate I could make a HV diff pair out of suitable TO-92's or TO-5's (if they're fast enough... low current so probably). Let's see, 2N3439T x 4 sits in my junk tray, 450V 20 or 100mA as I recall.

I heard boiling water can crack SMPS transformer varnish, but I've never tried it myself.

- No CT'd supply, so I return to ground instead. Hence the coupling capacitor between bridge and Lmatch.


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

no worries.

without double-checking (IOW I might be wrong), its because IGBTs turn on like a FET but turn off like a BJT, so its due to having to sweep out the minority carriers.

Again it depends on the IGBT, but ZCS and/or ZVS alter things. In general ZCS is best for IGBTs. VPEC have a number of good papers on doing this.

not necessarily. crank up the voltage a bit, use a series zener with a cap across it, and voila - DIY -ve bias sans extra supplies.

the calcs in that link are the real clue. A 600V device will happily shit its pants at 400V, if you give it a 200V spike. At 10A switching in

100ns that'll require 2uH of inductance, which you'd have to work at. If for some reason the IGBT desats, Idesat will increase about ten-fold, so Lboom drops to 200nH. Thats not so hard to achieve.

it all boils down to how stiff the gate drive is. If it aint stiff enough, -ve bias will help. If it is stiff enough, -ve bias is unnecessary. IR wrote a paper on this a few years back.

As an aside, I once worked on a FET UPS with Rg = 47k. the idea was to slow dV/dt this way, and it did work. But one day ST changed the FET (new process, "improved" parameters) and the circuit just latched at Vt. A far better approach would have been an emitter-follower with a decent RC delay on its input, and a nice low Zout. In the end we bought a whole bunch of the old FETs.

in theory snubbers aint required at all - square RBSOA. In practice small RC snubbers can be convenient to stop any parasitic ringing. If you try and control dV/dt that way, watch them catch fire - draw the original V,I waveforms (I is roughly constant at switching edge), and the desired V waveform. the difference between the two V waveforms, multiplied by I, is how much power you'll dump into the snubber on each transition.

Ideally you can just add C across the IGBT and adjust the deadtime, as you are driving an inductive load which will charge/discharge the Cds for you during dead-time. It'll go horribly wrong if you ever run a non-inductive load.


OTTOMH I'm not sure thats right re. impedance matching.

beware how Vce changes with Ic. IGBTs optimised for switching tend to have ratshit Vce-vs-Ic curves, and vice-versa - IOW fast switching or low conduction losses, pick one. And at these low powers, just use a bigger IGBT.

IMO until its too high, higher voltage will cause you less problems. As current increases, skin/prox effect becomes a real pain, as does winding the suckers.

Einstein says its OK, and the coil dont care.

have LM339, will travel.

indeed. I suspect its non-trivial.

try it yourself, and watch all the teeth fall off the hacksaw blade. We absolutely amazed the guy at Trade Tools by showing him his diamond-toothed hacksaw blade sans teeth after a couple of swipes.

I dont know what your LC circuit looks like, but the impedance of each of those is about 0.45 ohms at 18kHz.

the repetitive splat tester is really just a no-power smps. I now have a much nicer one, made out of a little induction heater PCB :)

I built a couple dozen of these for glueing planar cores - its a diagonal half-bridge (TL & BR FETs, BL & TR diodes) driven by a UCC3801 & HIP2100. The D-HB means all inductive energy is returned to the +13.8V supply sans losses (except 2*schottky drops worth). The coil is a few dozen turns wound around a machined steel heater, that is spring-loaded and presses up against the lower core to be glued (thermosetting epoxy). A bit of machined formica, and voila.

IME the energy storage curve looks like an RC rise, so the point at which it becomes useless to gap further is somewhat arbitrary.

Cheers Terry

Reply to
Terry Given

So yeah, a few uF or so, which is huge at this frequency.

Er, oh yeah, the diodes clamp it within buss limits... Just have to make a stiff buss. Film caps any use here (maybe a filmlytic?) or are standard comp-grade caps fine? (I was thinking of 4 or so Nippon D77421B 780uF 450V caps, if that means anything.)


Yeah, but not much in comparison to the intended Lmatch that's there anyway.

Yeah, that's why I wonder if any film caps would be useful? They'd have to be pretty large to do much good though.

Ok, will check them out...

Naturally...sigh! ;)

Think I could get away with 25VCT AC transformer with the secondary bouncing along at 18kHz 240VAC? That'd save the trouble of making yet another switching supply inside this thing...

So just detect a smaller margin (say 1V) off a 24Vp-p signal? Yeah, I suppose that *COULD* work...


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

AIUI ZVS doesnt solve the minority carrier problem (aka tail current). so the improvement is nowhere near as great as with FETs.

Likewise ZCS turn ON. This eats the turn-on switching loss Eon, but does nothing for Eoff. The ratios of Eon to Eoff vary dramatically with Rg, with Eoff dominating for small Rg - which is what you would expect, as lowering Rg wont reduce tail current

Done this way the zener carries the turn-on gate current, and the cap carries the turn-off gate current.

charge pumps can be used to good effect also, but UVLO is pretty much mandatory for a gate drive with such a wobbly power supply (unless of course you want random failures)

you shouldnt need a TVS across your DC bus, and even if you do have one, the inductance will again have to be low. In general the DC bus capacitance is huge compared to the energy sloshed around per switching cycle (by about the ratio of Fsw/Facline), so it "looks like" a TVS anyway. its the parasitic L that will get you. I would use a 2-layer PCB, pref. 0.8mm, for the DC Bus - a plane on either side, with suitable creepage/clearances. Look at Semikrons mini-SKiiP for a neat low-L package. TO-xxx devices can have very high lead inductance if mounted poorly (IOW without consideration to minimising L), and its not hard to get 1uH with a birds-nest prototype using hookup wire.

If Zg is low enough, you wouldnt need to slow the collector swing at all. If you use a transformer, -ve bias should be a doddle. Unlike the

+ve supply (which controls Vcesat and Idesat) you dont really care about the value, as long as its not enough to snot the gate (+/-Vgmax).

not at 60Hz. they wanted slow edges to drive a big X-cap.

OTTOMH I was wrong :)

nice work coil BTW.

Spiral-wound Cu strip would be nice. Then slip ETD cores on and voila.

good point.


desat sensors. If Vce > 10-15V while gate on, trip.

no, havent tried that. But its incredibly brittle, so I doubt it would be much use. I've cut myself with smashed 3F3 cores though - the disadvantage to using a good glue is that when someone glues the wrong cores together, its hard to fix. Luckily I have a hammer :)

so you have a C in parallel with the work coil, one end to DC bus center-point, the other fed from Lmatch?

yep. continuous but nonlinear due to fringing, which can be neglected for small gaps.

Cheers Terry

Reply to
Terry Given

just plenty bigger than Qgate/Vgate = Cgate

stray L issues again, but if you must....beware the monstrous capacitance of some of these though. Its cheaper to use your bus caps.

get some double-sided copper-clad pcb, and DIY-it. all ya gotta do is drill holes for the caps, then remove (sharp knife + soldering iron) a big piece around the "other" terminal.

thats good.

and acting to give LdI/dt voltage drop every time the IGBTs switch.

I'd do 2-bits of PCB. one as above, and a single-sided bit for the cap center-point. It'll also help the current sharingh, at the moment the closest caps do a lot of the HF work (for edges, L dictates current sharing)

ST have some nice app notes showing a few tricks, as does Siliconix.

thats how real men do it ;). Unnecessary at low power, but above a few kW its mandatory. Though shalt not f*ck with gate supplies....

lower IGBTs only is a good start, but think about it, it can be done quite easily. A (compensated) voltage divider for Vout, as well as Vdc. A bit of logic....

we changed glue, and now I use my hot-air rework station. The cores often crack though.

Ah, Lseries + Cseries + Lwork//Cshunt

Cheers Terry

Reply to
Terry Given

Gack, SMT! Even worse!

But how much good does the "really small" do when you've got edges only a half uS or so wide? (I'm thinking up to 1uS is acceptable so I should be able to use a variety of IGBTs as far as speed...)

That's what line filters are for. Nah, I wouldn't use a RadioSnack transformer...I'd at least make sure it has an electrostatic shield and hi-pot test.

Ah, even better, if I'm going to be putting in gate supplies.

How about current/pulse transformers?

No, you don't understand... it's a miracle I'm even *considering* transistors in the first place. If it were up to me, and the having of $1000, I'd have bought a 3CX3000A, set it up and be done two years ago! But nooo, to save the $900 I have to do all this crap instead. An LM339 is just another shovel of dirt dug out of my grave... ;^)


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

Hmm, ain't got an FT convienient... or simulator, or digital scope...or DMM (now that last one needs to be replaced, yes)...

I'm sure they do. Hum, I meant that in terms of detecting Ic/Ie, sorry for being ambiguous :)

Yeah, the whole thing has been so far!

Hmm.. I'm suprised no one else has replied to this thread.


-- Deep Fryer: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

a typical big electrolytic has > 30nH or so inductance, which is huge compared to the inductance of your DC bus if using 2 large planes with small separation (eg PCB or 1.6mm Cu sheets with 0.5mm nomex etc), so paralleling N of them reduces L by N (cheesy math craps out as L' tends to Lbus) which is a good start. SMT caps as close as possible to the IGBT is a common trick, N in parallel is a good plan too. Beware HV smt ceramic caps, if you thermally stress them (say use a soldering iron) they develop micro-cracks, and eventually explode :)

for real big IGBTs, there are film caps you can buy that bolt directly onto IGBT terminals, with L around the 2-3nH mark. A sneaky bugger I worked with in the US developed an astonishingly low inductance (3nH or so) 150uF film cap, we used about 12 of them. Another approach (you probably wont be able to get it) is super-high-permittivity PCB material, can get as high as Er = 300 (cf 5 or so for FR4). wasnt that a horrible tease :)

indeed, so you can ignore it there. just dont ignore it wrt Vce spikes.

big//small//really small

in theory yes. In practice, no - a standard 50Hz xfmr probably wont like the wicked high dV/dt it'll see - it'll cer5tainly couple metric shitloads of noise straight into the national grid, there may be insulation breakdown issues (analogous to first-turn failure in motors) and the insulation might not be up to scratch. But a carefully (re-)wound toroid ought to be fine. Before my time, but thats how we did all our tralington bipolar gate drive supplies (EI actually)

in practice when an IGBT desats, Vce swings rapidly to full bus volts, and I sits at about 10*Irated (depends on IGBT type, Vg) so P is wicked high - thermal inertia of die (at this speed its adiabatic) gives you about 10us. But the huge voltage swing makes your job a lot easier.

When you build bigger ones, the desat circuitry sits with the gatedrive cct at the emitter potential of the IGBT, so life is a lot easier (but you need a fault feedback opto). And of course when the IGBT costs $400 who gives a shit about a few LM339s etc.

Cheers Terry

Reply to
Terry Given

only until you get used to it, then you'll never want to design with leaded parts again. An order-of-magnitude reduction in package L.

fourier transform + reactance paper = oh, thats why.


little gate drive pulse xfmrs make nice HF smps xfmrs for gatedrives.

teaspoon, more like. you obviously dont value your time....but its one hell of a learning experience.

Cheers Terry

Reply to
Terry Given

I would never build a circuit that wasnt current-controlled. With a decent sensor & limit/trip circuit, its easy to make the damn thing short-circuit proof.

When doing desat tests on a 100kW drive (short an IGBT with a bit of wire and press start), we had to use a wide, flat strap across the IGBT to actually make the other one desat, lifting it 100mm above the IGBT gave enough inductance for the current trip to catch it before a desat.

for years Danfoss used to short the drive outputs while running, at tradeshows. Very impressive (until you notice they have large output inductors)

I was very lucky, my first job was for a crowd that built 250W-1MW drives, and had done so for years. Wicked fast learning curve, without having to make most of the common mistakes myself (so I found uncommon mistakes)


Cheers Terry

Reply to
Terry Given

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.