Why is this MOSFET turning off 20x slower than specified?

Question

I use this MOSFET:

https://cdn-reichelt.de/documents/datenblatt/A200/ZETEX_DMG1012UW_ENG_TDS.pdf

in a basic boost converter topology in DCM mode (L=100µH). The gate is driven by an ATtiny402 through a 100 Ohm resistor. My oscilloscope shows some odd behaviour: the MOSFET turns off more than 500ns after the gate has reached 0V. Turn-on is practically instantaneous.

I also measured the current into the gate. It shows some constant-ish current coming out of the gate in the delay phase.

As a test I used a different MOSFET, where the waveforms look exactly as expected:

Unfortunately this well-behaved MOSFET is not suitable for the final application.

Now to my question(s):

What is happening here? At Vgs=0.7V Rds>30kOhm so it isn't just a low threshold voltage. I tried soldering in a new one of the same type, but the problem is the same. If possible, how can I improve turn-off performance? And no, shorting the gate directly to the pin of the ATtiny did not make a difference.

Update with some clarifications and some additional experiments:

I reduced the gate resistor to 10 Ohm, it did not change the turn-off delay. It did, however, slightly change the gate current waveform:

For those that requested a schematic, it really is just your standard textbook boost converter:

other tests I did:

1. added a 1.5nF capacitor across the 100 Ohm gate resistor, no effect on turn off delay.
2. reduced VCC of the circuit from 3.8V to 2.5V, the delay was shortened by 25%.
3. reduced Rg to 10 Ohm, no effect. (I/O pin resistance ~ 30 Ohm)

clarifications:

1. yes, I did measure directly at the gate, not at the MCU pin.
2. yes, the MOSFET is supposed to be able to be driven from a MCU pin (700pC)
3. where I bought the weird behaving MOSFET: reichelt.com
4. the second MOSFET I tested (that did not behave weird) was a BSS138 https://www.onsemi.com/pdf/datasheet/bss138-d.pdf

clarification edits in the original section above:

1. changed the Ig waveform picture to be more clear and show the remaining current better.

Answer 1

According to measurements, internal gate resistance is around 400 Ω. Datasheet doesn't specify R_G but from the switching times, it can't be much more than 100 Ω. You got dud parts, I'm afraid; buy new elsewhere.

This is a good demonstration by the way -- and kudos for taking complete measurements! -- of how to determine device parameters. Once you know input (gate) voltage and current, given a suitable waveform (like, a sharp step), you can solve for R_G and Q_g.

To clarify / for reference of those popping in, the GS and GD charges are evident here:

It's not obvious why Vds drops ~instantaneously (implying Crss has little effect at turn-on), but in any case it does, and this is accommodated by the positive pulse being a little taller/longer.

The negative pulse shows a mostly exponential RC decay as well, but with the Miller blip more evident. Therefore we can reasonably separate the D-G charge as the hatched area.

...Could Crss itself have considerable resistance, say? Perhaps that explains the asymmetry. I have no physical explanation for that, not based on any normal MOSFET fabrication process anyway, let alone failure mode.

There is quite a lot of ringing in the circuit (the layout is probably poor, long wires, breadboarded?), but that's evident as the squiggly lines, and all at a much shorter time constant (the ringing is over within ~100ns).

Also, the asymmetry is definitely not something like a rectification effect (say if the internal resistance had a parallel diode), because the exponential tails are fairly symmetrical (similar peak current and time constant), but for the Miller blip.