Kind of odd P-MOSFET

[T3sl4co1l]

https://assets.nexperia.com/documents/data-sheet/PMV50UPE.pdf

Looks pretty ordinary, right?  But some things don't quite add up.  Maybe they're just unfamiliar?  Are they inconsistent?  Let's take a closer look:
- Digi-Key lists it as 24pF C_iss, despite the tiny Rds(on).  This caught my attention.
- Vgs(th) is quite low, and tightly controlled; this would be anomalous for a regular transistor, but see the Vgs(on) listed for Rds(on) -- it's rated down to 1.8V, nice!  They probably selected these very carefully, for just this reason.  (So this is okay.)
- This should give a slow turn-off time, since it's discharging through R_G at a plateau voltage around Vgs(th).  So there's SFA current through the gate resistance during that phase.  Fair enough.  (Turn-on should still be okay.)
- So it's somehow 24pF, but also 10.5nC Qg(tot) typ.  It's definitely a big junction: low Rds(on), high Id, high Qg.  What the heck is up with the capacitances?
- Look down at Fig.13.  C_iss is varying with V_ds, and the others are constant?!  Wait, is this actually a lateral transistor?  Would it vary with Vgs at all instead I wonder?  (No, it says "trench" up top.)
- Look at the switching times.  Wait, holy hell, whole microseconds turn off, and turning on in the better part of one microsecond?  But wait, RG = 6Ω, so they're not being a slouch about it (which has been seen before -- some Fairchild datasheets used RG = 25Ω for some reason).  So, what then: if the gate time constant is about twice the total switching time (this is roughly typical for MOSFETs), and the resistance is 6Ω, the equivalent gate capacitance is... 2 * (1.1us) / (6Ω) = 0.37uF?!  No, that'd be stupid.
- If Qg(tot) is right, then 10.5uC at 4.5V is 2.3nF equivalent (C = Q/V).  Is it actually that there's considerable internal gate resistance?  If C = 2.3nF and t = 2.2us, that's 0.95kΩ!
- And that time constant means RG is substantial at 1MHz -- the frequency where all the capacitances are measured.  Hmm.
- If RG is actually being measured rather than C_iss (or more specifically, the series combination), what would we actually measure?  If we measure |Zg| and assume it's a capacitor, we expect on the order of 1nF + 1kΩ equivalent circuit, which at 1MHz is -159j and 1000 ohms, so |Zg| = 1012Ω which would be a 157pF capacitor.
- Or if we use a vector impedance measurement, we get exactly that series equivalent, 1000 - 159j ohms; but if we took the parallel equivalent instead, we would get... 1025 ohms and 24.7pF?  HMMMM
(Play with it yourself: https://daycounter.com/Calculators/Parallel-Series-Imedance-Conversion-Calculator.phtml )
- Also, looking back at the switching times, yes, the turn-off values are almost ten times the turn-on values (implying the plateau falls around 0.5V, 1/10th the Vgs(on) used for this test).
- Beyond the datasheet, they do in fact provide a SPICE model for this part; it only shows 40 ohms R_G though.  It should be a lot faster, if these capacitances were to believed, or if the model is truthful; but it cannot be both!

So that's my story of critically reading a suspicious datasheet; I hope you were adequately bored by it!

Tim

[coppercone2]

what application is that good for if its so weird, the data sheet does not advertise anything beyond totally normal jellybean mosfet applications

Relay driver•High-side loadswitch•Switching circuits

the first two many people would not even look at specifications at all when choosing a part and go by looks of the package alone, and the third one I am sure you can list in a relay applications datasheet (just shove anything in there you can)

did this thing just come together as a business decision based on machinery and materials that were available to make something sell able?> a chip fab Saturday night special made with stuff no one wanted to throw out?

[rfeecs]

Typo?
Comparing values in Table. 7 with plot in Fig. 13, it looks like they swapped C_iss and C_oss in the table?

Comparing to a probably comparable part:
https://assets.nexperia.com/documents/data-sheet/PMV33UPE.pdf

Capacitances look 10 times too low,
Turn on, turn off, rise fall times look 100 times too high.

So yes, looks funny.

[T3sl4co1l]

what application is that good for if its so weird, the data sheet does not advertise anything beyond totally normal jellybean mosfet applications

Relay driver•High-side loadswitch•Switching circuits

the first two many people would not even look at specifications at all when choosing a part and go by looks of the package alone, and the third one I am sure you can list in a relay applications datasheet (just shove anything in there you can)

did this thing just come together as a business decision based on machinery and materials that were available to make something sell able?> a chip fab Saturday night special made with stuff no one wanted to throw out?

It can be intentional for reduced switching noise -- a similar application would be IGBTs for automotive ignition.  But they should specify why it is the way that it is.  Those IGBTs, they specify the internal gate resistors, and protection diode.  The slow switching time is no mystery, and quite intentional.

If they had specified R_G (internal gate spreading resistance), and measured capacitances at a low frequency (say 10, maybe 100kHz), that would explain it nicely.

Tim

[T3sl4co1l]

Typo?
Comparing values in Table. 7 with plot in Fig. 13, it looks like they swapped C_iss and C_oss in the table?

Comparing to a probably comparable part:
https://assets.nexperia.com/documents/data-sheet/PMV33UPE.pdf

Capacitances look 10 times too low,
Turn on, turn off, rise fall times look 100 times too high.

So yes, looks funny.

Haha, yeah, that looks like the same device with a somewhat bigger die: roughly proportional increase in Qg and decrease in Rds(on).  But the capacitances are right.

Swapped curves wouldn't even make sense; Ciss is the flat one, and the other two are highly curved.  Two flats isn't sensible.

What if it's, like, they characterized exactly one part, and on that particular day, someone stuffed a 1kohm gate resistor into the tester?  Or it was programmed for R_G other than what they thought it was, or there was a fleck of dirt on the socket that just happened to offer this resistance?  These are only typical values, after all, it's not like they need to test more than one.  (They might not need to measure them at all, using design calculations instead; but I would suspect that to give a much more reasonable result in both cases...)

Tim