Severe nonlinearity of Cds in enhancement HEMT FET?

Recently I've been using GaN FETs in amplifiers, and have found that some of them have extremely nonlinear drain-source capacitance, while others do not. For example, here are two plots from two similar devices (roughly same size, Rdson, Qg, etc).

First, the EPC1013:

The curve of Coss (the red curve) for this one is severe, but not terribly surprising.

And second, the EPC2012:

The nonlinearity here is much more extreme, and I can't think of any physical reasons to explain it. I'm wondering if it's due to different gate shielding, or to overlapping of the gate and drain/source terminals, but it's been a while since I've studied HEMT physics, and even then we never discussed nonlinearity.

I'm not a semiconductor designer, but I'm still very curious what the explanation for this behavior is. Any thoughts?

The Cdg ought to drop for a while as the drain junction
pushes back from the gate, but will eventually flatline
when the metallization fringing capacitance becomes all
that's left.

With these FETs perhaps representng different generations,
perhaps that corner is just off the chart for the EPC1013,
or the device's useful Vds range ends before that corner.

That, or the modeling dude did a sloppy fit on one or the
other.

dick_freebird said:

The Cdg ought to drop for a while as the drain junction
pushes back from the gate, but will eventually flatline
when the metallization fringing capacitance becomes all
that's left.

Click to expand...

Right, I expect Cds to drop off in an inverse square root manner to some flat minimum value as the depletion region grows, but that's not what it looks like.

With these FETs perhaps representng different generations,

Click to expand...

They are, but as far as I know the main difference is that the EPC2012 is RoHS compliant.

perhaps that corner is just off the chart for the EPC1013,
or the device's useful Vds range ends before that corner.

Click to expand...

Each plots shows the full range for the device, up to Vdsmax.

mtwieg said:

Each plots shows the full range for the device, up to Vdsmax.

Click to expand...

Right, my point being that maybe in this generation breakdown
(or degraded reliability) comes before the capacitance has
reached its fringing asymptote.

dick_freebird said:

Right, my point being that maybe in this generation breakdown
(or degraded reliability) comes before the capacitance has
reached its fringing asymptote.

Click to expand...

Yes, that could be the case. But what I'm more interested in is the huge drop in the Cds of the EPC2012 at around 15V. I don't think that can be explained merely by lengthening of depletion regions.

Looking through the literature, the manufacturer seems to address this briefly in one of their white papers (page 4):

These capacitances are a function of the voltage applied to various terminals. Figure
8 shows how the values change as the voltage from drain to source increases. The
reason for the drop in capacitance as VDS goes up is that the free electrons in the GaN
are depleted away. For example, the initial step down in COSS (Output capacitance)
is caused by the depletion of the 2DEG near the surface. Further increase in VDS extends
the depletion region deeper into the body of the device thus increasing the
distance between the plates of the capacitor.

Click to expand...

So I guess I can see how the 2DEG itself being depleted would cause a large change, but I don't see why some devices have a much more abrupt drop in Cds than in others.