...And sometimes, ceramic caps aren't so bad

[T3sl4co1l]

I mean, type II ceramic caps are still funny for all their quirks, but I was a little surprised this worked out:

dV/dt snubber application.



Driver: UC3843 with a little pull-down assistance (measured 40ns t_f at the gate pin).  Transistor STP70N10F4.

Pic:



It's kind of crowded for a picture, but suffice it to say, the schematic says what's going on in the power section here.  It's dead-bugged, more or less, as low inductance as it can get.

The power transformer is 2.3uH (primary, magnetizing), with 55nH leakage inductance.  So when the transistor turns off (at up to 15A peak), without a snubber, the 55nH would discharge into the transistor.  At 350kHz, that's 2.2W, which isn't terrible, but I'd rather burn that in a resistor than risk the excess peak voltage on the transistor (though with a 100V device pulled from the parts box, that's not really a concern here, is it?).

So, to limit risetime to no less than 20ns, we need C = Ipk * dt / dV or (15A)(20ns)/(30V) = 10nF.  That needs to be switched to ground with a diode during the rising edge, and discharged through a resistor during the falling edge.  The resistor can be whatever, but it has to be small enough that ~2*RC elapses during a typical on period, so that it's mostly discharged before the next rising edge.  It can also help dampen the primary during ringdown (in which case R ~= sqrt(Lp / Csnub), which is pretty much the case here).

But enough about design, on to the observations.

Stray inductance is critical in this path: even with a mere 5nH across the capacitor and diode to ground, the transistor (which has about 5nH lead inductance) will ring like a bell during that slope.  With peaks at 30 and 80MHz (corresponding to Cdss with Ls and LL), that's a mess for conducted and radiated emissions.  (Again, not a problem on this project, but: practice makes perfect, and I want to have clean waveforms!)  The 2.2 ohm + 10n damper (hanging off the end of the transistor tab, thus, at the end of that inductive 'springboard' as it were) keeps those squigglies down.

Now, component choice is obviously a concern.  So:
- I started with 10n (50V X7R 0805) for the snubber cap.  Burned a hole right through.  Well, yeah.
- Then I tried 2n (2 x 1000pF 630V C0G 1206), being the largest C0G I happen to have on hand.  Not enough, but it does the job, at least at lower currents.  Way too spiky/ringey at full load.
- So I put 10n film (radial, 7.5mm lead spacing) in parallel.  This has way too much inductance, and resonates at ~30MHz (~2nF + 10nH?).  The transistor is basically connected in series with a parallel-resonant tank, and it rings like a bell.  Well, yeah...
- On a larf, I put a green "103Z" ceramic disk in parallel (in place of the film cap), absolutely minimal lead lengths (maybe 3nH worth?).  It still rings, but not as much; the R+C damper covers that well enough.

Odd thing is, I was honestly expecting it to either: go really nonlinear (it's probably a 50V part, so, it should at least be visibly speeding up by 30V), or heat up so much that it "turns off" (just from tempco alone, or from reaching Curie temp).  Probably not enough to burn a hole in it though.

Survey says:



Here's the overall waveform; the rise looks pretty smooth!  A little twiddle when it gets to the top, but nothing horrible (of course, it looked worse when I started).

Zoom on the rising edge:



Now here's the weird part: it rises slightly faster, then flattens out, then rises again.  The effect is more exaggerated at high temperatures (i.e., holding the soldering iron on the capacitor to warm it up), where the flattish part remains about as flat, but the slanty bits get way slantier.

It would be tempting to attribute this to a low frequency resonance (it kind of looks like a hump-dip-hump), but I can't find anything in the circuit that should produce such a period on this waveform.

It would also be tempting to attribute perhaps some of the most initial rising edge to diode forward recovery (I did use a PN junction diodes, for whatever reason, instead of schottky).  But this isn't borne out by the diode waveform,



It's a 1 div spike (note the finer scale -- oh, and these are all 10x scale by the way, so 2 and 5V/div), and it's not nearly wide enough to account for the observed effect.

Now, the oddest thing about this observation, to me, is the normality of it: of ferromagnetic materials, low initial permeability is not a well-advertised feature, but it is totally a thing.  Steel is best known, but many materials exhibit this effect, including powdered iron cores (which sometimes have a permeability vs. AC flux density graph, or something like that, which shows the permeability peaking).  The B-H curve, in these cases, isn't so much the conventional sigmoidal hysteresis loop, but commonly called a "butterfly" curve, containing two loops that meet in the middle.

This is an even less well-advertised feature of ferroelectrics (i.e., type II dielectrics), only occasionally having any evidence in the C(V) curve (if provided), where there may be the most subtle of toe-in near 0V.  For sure, no manufacturer publishes the D-E curves of their materials (displacement and electric field are the complements of flux density and field intensity).  Neither material is even measured consistently, either: the capacitor's plot is a dynamic C versus static V plot (small signal + bias), while the magnetic core is measured as dynamic L versus dynamic dI/dt (large signal average, no bias).  Occassionally the former is given for magnetic materials as well, but never the other way around.

That said, this is all occurring above 0V bias, so with hysteresis besides, there should be no way it's getting very far into the "initial permittivity" region of this capacitor, if it be such a thing at all -- but I guess this example, at least, is just that crazy!

And besides everything else, this is a strongly dynamic thing (high dV/dt, high frequency / harmonics), so it need not bear any relation to a static plot at all -- operation up here could be governed by an entirely alien D-E curve, compared to the static one.  Who knows.

When all that is said and done, I am glad that the average capacitance comes out right: that's what matters the most to this energy transfer application; meanwhile, the losses are acceptable (the capacitor doesn't seem to be heating up at all!) and the inductance is minimal.  Works for me!

Tim

[paulie]

Yikes! What an excellent graphic demonstration of LC behaviors. More educational than a semester of university courses. With chip ceramics encroaching on electrolytic territory with 22uf in 0603 and costing a penny or two I though they were the cats meow. But interesting to see the old disc caps still have a place.

That circuit looks like a perfect candidate for planar inductor application with it's PCB track turns. Any hints on nominal efficiency?

[T3sl4co1l]

This exact circuit, mmm, with a 4 layer board it might be pretty lossy, but 8 layers would surely handle all the current flowing.  Without taking too large of a core set, that is.  Nice thing is the leakage inductance can be even smaller, which will actually begin to increase losses because capacitance will then rear its head!  Or to put it another way, the transformer's characteristic impedance will be lower than the switching impedance (approx. 12V / 15A, say).

The disc cap here is mainly because I don't have a 10n 250V chip handy; even a 1206 might not actually handle the peak current, though.  But C0G are readily available in those sizes, if you don't mind the cost -- that would be the ideal choice.

Tim

[TerraHertz]

That 2.2 ohm resistor is a metal film type, aka spiral inductor masquerading as a resistor.
You might want to try replacing it with a flat SMD resistor?

[T3sl4co1l]

Almost exclusively, remarks about spiral resistors are only relevant in radio, UHF and up -- wirewound resistors are the exception.

Carbon film actually (the 15 ohm may be metal film or metal oxide), but spiral cut all the same.

At all of a few nanohenries, the Q at any frequency in the circuit (up to 100MHz, say) is low enough for it to still be a resistor.

Or, to put it another way: I have the waveforms, you tell me.

Tim