even though they use the same terminal names and
behave similarly (though not exactly the same, as we’ll
soon see). In fact, the first GaN switches (formally re-
ferred to as HEMTs, for High Electron Mobility Transis-
tors) had more in common with JFETs as they operated
in depletion mode - that is, they were normally on with
0 V applied to the gate - and the gate would conduct
significant current if forward-biased too much, lead-
ing to device destruction. Depletion mode operation is
intrinsically unsafe in most power converter circuits for
the obvious reason that one generally doesn’t want their
switches to be turned on by default, but to enable en-
hancement mode operation requires building the GaN
switch on top of a silicon substrate (to incorporate a
control MOSFET). This complicates the manufacturing
process, but it results in a switch that power electronics
engineers will at least consider using.
Although GaN switches don’t have a body diode, they
will still conduct current in the reverse direction when
off, and with a diode-like characteristic, except that the
voltage drop is the sum of the gate threshold voltage
and the voltage biasing the gate. Note, then, that biasing
the gate with a negative voltage will increase the volt-
age drop during reverse conduction! Since negative gate
bias is practically a requirement for reliable operation of
IGBTs and MOSFETs at high power, in order to counter-
act spurious turn-on via the “Miller” capacitance (that
is, the parasitic capacitance between the gate and drain
terminals), this can be quite a gotcha, indeed. GaN
switches do exhibit much lower parasitic capacitances
overall - a major reason they are so fast - but they also
MAY/JUN 2019 31
have a relatively low gate threshold voltage compared to
Si or Si-C MOSFETs (and IGBTs have higher thresholds
still). Whether the reduction in Miller capacitance is
sufficient to counteract the lower threshold (and faster
switching speed) will likely have to be evaluated on a
device-by-device basis, and, while slowing the switch-
ing speed down will also help, that sort of eliminates
the primary reason for choosing GaN over any other
technology.
Finally, GaN switches are very particular about how
their gates are driven, though the current generation
is much improved in this respect compared to the first
devices that were made commercially available in 2010.
The biggest issue is that there isn’t much margin be-
tween the threshold and maximum allowed voltages
for the gate, the latter being 6 to 10 V, depending on
the manufacturer. Also, the current through the drain-
source channel is strongly dependent on gate voltage- much more so even than in IGBTs, and also way more
so than in MOSFETs. This could be a plus or a minus in
that deliberately under-driving the gate will limit fault
current - much as the desaturation mechanism in IGBTs
does - but this results in higher conduction losses over-
all, which, again, defeats one of the reasons for select-
ing GaN in the first place. In the end, SiC is plenty fast
enough for any power conversion application in an EV,
and its overall higher ruggedness is a real plus, too. GaN
is really too fast for its own good here, and seems better
positioned to applications in which space is at a pre-
mium, and where MHz switching speeds can actually be
used, for example in low-voltage DC-DC converters.
GaN Switch Structure