Oh yikes... 25nC vs. 8 max Qg, for the same Rds(on)... IRF7306 is ancient!
Hold on, what Vgs(on) was that? 5 and 10V? Ah, so it is worse, but not that much worse (see Fig. 6
https://www.infineon.com/dgdl/irf7306pbf.pdf?fileId=5546d462533600a4015355f1fc421b0a ).
Si4947:
https://www.vishay.com/docs/71101/71101.pdfOK, so I would be inclined to expect similar behavior here. It's not a perfect match, but it's fairly close.
As for capacitive divider: yes, this is indeed an effect, though usually small enough not to matter. The relevant parameters are Crss (Cdg) and Ciss (Cgs + Cdg). Which, if you look at the graphs, Ciss looks almost identical to Crss, shifted up a bunch, and this is pretty much the case for modern (vertical/trench) MOS. Basically there's a constant Cgs baseline value in there, and the nonlinear (non-constant / dependent) part is simply Crss. (Likewise, Coss is Cds + Cdg; you can do the same subtraction to find Cds alone, and as it happens, Cds itself depends on Vds, hence why Coss has a steeper slope.)
Evaluating the capacitive divider is nontrivial, because of the dependency. For example in the first volt, Ciss ~ 760pF and Crss ~ 380pF (IRF7306), the difference (Cdg) being ~380pF, a ~0.5 division. Whereas up at 30V, Crss is down to ~100pF and Ciss ~ 430pF, or a divider of 100 into 330pF, a ~0.25 division. The total figure for a 0-12V swing will be somewhere inbetween, though that already gives us enough information: Vgs(th) ~ 1V, but the best-case division is 0.25 or 3V, so the poor bastard's turning on, all right(!!).
The original Si4947 has Ciss ~ 800pF and Crss ~ 250pF at 0V, dropping to ~590 and ~60pF respectively at 30V. Again we see they have similar slopes, and mentally subtract, so Cgs ~ 530pF (constant) and Cdg 60 to 250pF, for a ratio of about 0.33 to 0.1, again still rather high but certainly will dissipate less power in the same conditions.
So that does seem consistent with the observed waveform. There's a small turn-on glitch (also visible in the unused channel; this will have frequency content in the 10MHz+ range, so is easily picked up by open probes, ground clips, etc.), then it goes from about 6V, rising gradually from there (presumably the resistor doing its work). This is consistent with the 0.25-0.5 capacitive divider (indeed closer to 0.5, even at the 12V level, which is even more concerning). When the NMOS switches off, the voltage rebounds above +12V, since the capacitors have been discharged a bit during the first part.
The downside to the shunt capacitor is, it slows switching in general, which may be a problem for other phases of operation not shown here. So, I don't know about that. But if nothing is worse than what's shown here, then it is indeed an effective solution.
Another caution: low impedances in the gate-source path, tend to cause parasitic oscillation; and this is further exacerbated by the slow turn-off, which may allow it to sit there grinding away, oscillating for quite some time -- greatly increasing EMI. This can be hard to see, even with a good oscilloscope (the frequency corresponds to the G-S loop dimensions -- it can be 100s of MHz!). The easiest way to avoid this, is by damping that loop with some series resistance, just a few ohms will be enough. So, you'd need to cut a trace to do that here (if you can fit a resistor at all!). And, it's not a guarantee: perhaps the environment just isn't reactive enough to cause oscillation, or the MOSFET isn't high enough performance to start it up (PMOS have ~2.5x poorer performance than NMOS; it might indeed simply not be capable of going that fast).
The preferred design solution of course is a proper driver, or at least a BJT boosted pull-up (like this:
https://www.seventransistorlabs.com/tmoranwms/Circuits_2010/12-24_Converter.pngthe 1N914 + 2N4403 part, note the 1k pull-downs in front of them; for driving PMOS, these are swapped, so, diode pointing the other way, and NPN e.g. MMBT3904). But you almost certainly won't be able to fit that on an existing PCB, so this isn't too helpful for present purposes.
So -- given the above -- the capacitor is indeed a surprisingly adequate engineered solution. The kind of thing that, it's mediocre at best, and it's stupid [it's just one component, that generally worsens performance otherwise], and it has side effects -- but when all that works out, yeh, there you have it.
Or to put it another way: if it works, it ain't stupid. Or, let's say the above analysis counts as smartening it up.
As for going the other direction, there are some better-performing types today, though they're all non-stock at DK, alas... Options include Diodes Inc. DMP3085LSD and Panjit PJL9807_R2_00001. And they're surprisingly comparable to the original in terms of Crss (overall, and steepness of slope), so it seems PMOS haven't improved much after all!
Tim