SuperJunction Pulse Generator (of sorts)

[T3sl4co1l]

Observed a curious effect just playing with a transistor here.

Setup:



Suffice it to say, it's absolutely babied, purely just charge controlled switching, inspecting the edges with resistive load at low rates (low current).

Gate rising impedance is 100 ohms total in this configuration.  Nothing impressive, and the generator isn't particularly fast either (but more than fast enough).  The diode turn-off network doesn't seem very useful here, but I was also testing at higher R_G(ON).  And there's a 100k load resistor and 0.1uF bypass cap for the high voltage supply, just to keep it in regulation (it doesn't perform well at low voltages).  I put them on the schematic for clarity, but they obviously don't do anything for the effect below.

Photo: https://www.seventransistorlabs.com/Images/SJ_Test.jpg
It's a bit of a mess so I've outlined the input in blue and output in orange.  It's a clippy lead hooked around the tab -- NO CONNECTION, that's an insulated ("full pack") device.  This just gives a little capacitance, a few pF if that, to the drain.  Lead goes back to a BNC cable, terminated at the scope.

Waveform:



Ch3: gate. Three stages left to right: initial rise, Miller plateau, final rise (RC slope to 8V).
Ch2: drain. Most of the falling edge (>90%), then a very slow settling to 0V.  Leading corner coincides with Miller plateau leading corner; heel corresponds with spikiness on Ch4; Miller plateau trailing corner coincides with 0V settling point.
Ch4: capacitive probe.  Clearly the ~square pulse corresponds to the negative dV/dt, but there's unexpected ringing, or oscillation, or something, at low voltages.




Zoomed closer.  The plateau leading corner is shown, and it's also evident that, around this critical point, Vgs actually rises a bit.  This rise cannot be attributed to any inductance in the circuit, it's not due to external currents -- at most there's 10pF of capacitance in the solderless breadboard and from device body to surroundings.  The bump is consistently present even at low Ig(on) (R2 = 1k, 10k, 100k).  I suspect it has to do with the ratio of Cds to Cdg (do they track at slightly different rates?), and how that affects gain while in Miller effect.

It could also be that this is actually an oscillation phenomenon, and it's at frequencies beyond what I can resolve (>600MHz?), and what manifests is effectively a rectified envelope of that oscillation.  Rectification could be by device nonlinearity (more or less as usual, when RF gets detected by a circuit), or perhaps even parametric (since the capacitances are changing so aggressively in this region).




Zoomed on the sharpest, tallest edge.  There's definitely some lumpiness going on here, which may be characteristic of the connecting wires and solderless breadboard (couple ns period), but there's also much more of an edge here, that I doubt can be explained by a coincidence of resonances and random excitation.

dV/dt during the bulk of the drain voltage slope is about 4.3 V/ns, and 120mV into 50 ohms is 2.4mA; evidently, C2 = 0.56pF.  Which also gives a cutoff frequency of 5.7GHz (at least nominally; there may be peaks in the low 100s MHz from the shitty wiring), so this should be a reasonable passive differentiator at signal frequencies, at least those within the scope's (350MHz) bandwidth.

Which further implies the step may be taller than shown here.  Instead of fractional volt, probably several volts are present at the junction.

What's weird is, if this is a sort of step-recovery phenomenon, it's not a reverse-recovery effect; it's turning on (depletion region shrinking), not off (depletion region forming and growing)!  More like the junction is vacuuming up free charge as fast as it can, as the voltage falls, and the depletion region shrinks and suddenly snaps almost utterly out of existence; and this snappiness it seems is fast enough to be visible from the outside.

Which should have EMI implications as well, at frequencies that aren't necessarily characteristic of circuit elements necessarily (10s to 100s MHz?), but as a very broad excitation (because it's steppy) and from very high frequencies (i.e. extending out to 100s or even 1000s of MHz?, if probed very close by) which in turn excite component body/lead resonances as well as circuit elements.  It's all synchronized to the switching edge of course, so you'd almost never know from the outside that it's this effect.  Perhaps with a very carefully time-synchronized method, you could resolve the edge / tone bursts (wavelets) from this effect, mingled within the overall switching edge.

Also, notice this is technically hard switching.  I need to do another test to see if this occurs in soft switching too.  And also see what the rising edge is like (evidently they're not symmetrical, there's a hysteresis loop here).

Tim

[jonpaul]

At these fatst risetimes and widebands, you need a 500 MHz ..1 GHz BW probe and scope to see the true signals.

With the  breadboard, in the photo , and long probe ground leads , that poor technique  explain some of the ringing.

I Suggest to repeat with a minimium lead legnth, low capacitance construction, eg PCB or dead bug and , no breadboard.

Avoid the Chinese 10X probes !

Suggest to use a Zo probe into a 50 Ohm scope or make a simple Zo probe, a 450 or 4950 Ohm 1/4W axial R in series at tip of cutr BNC><50 Ohm cable (eg RG174/U)

Jon

[T3sl4co1l]

And here's the falling edge (interesting part) when driven by a resistor.

So, uh...

That's some capacitance?





Ch2 is the DUT probe (Q3 drain), Ch3 is the driver (Q2 drain).

It's being pulled down by 10k, and just freezes solid at ~17.8V for a good 17us.  Varying R9, I measure:
2.2k: 4.42us
10k: 17us
56k: 79us
100k: 127.6us
Seems like it leaks away a bit on its own (which makes sense..?).

Ed: here's a comparison of five devices.  I'm guessing the lumpiness is characteristic of the doping profile in the SJ structure, or something like that.  Averaged over the whole die of course.




Tim