September 2016
29
P
ower
E
lectronics
nonlinear. This not only makes it difficult to compare devices based on
datasheet capacitance values (as they can change 3 orders of magnitude
depending on voltage), but also makes a big difference between devices
that are optimal for hard switching (low Eoss), versus ZVS soft switch-
ing (low Qoss). The graph in figure 3 clearly illustrates this difference. A
650 V, 70 mΩ rated high performance superjunction FET is shown in
blue, compared to a GaN HEMT with the same rated on resistance in
red. The dotted lines indicate charge (left axis), and the solid lines indi-
cate energy, both versus Vds voltage on the horizontal axis.
Note that the superjunction charge (dotted blue line) rises steeply up
to 90% of its final value within the first 20 volts applied. The slope then
abruptly changes so the final 10% charge takes the remaining 380 volts.
This is a result of how the charge distributes in the columnar structures
of a superjunction FET. This behavior has an interesting effect: since the
energy required to pump charge into the Coss at low voltage is smaller
due to the V2 relationship (E = ½ CV2), even though 90% of the charge
occurs in the first 20 volts, a far smaller portion of the total energy is
stored by this point. This characteristic nonlinearity is why superjunc-
tion FETs can have a relatively low Eoss for a given Qoss, and therefore
is what makes them excellent (low Eoss) for hard-switching applica-
tions compared to the other silicon alternatives.
In stark contrast, the GaN HEMT is a lateral device and has a low,
nearly linear capacitance versus voltage. In figure 3, the red dotted line
shows this as a shallow slope rising to a value an order of magnitude
smaller than the superjunction Qoss. But since this charge is evenly
distributed along the voltage axis, the integration of charge times volt-
age brings the final value of Eoss almost the same as the superjunction.
This relationship is summarized in table 1, comparing Si superjunction
FET to a GaN HEMT. The HEMT is still better in Eoss, but by a much
smaller margin than Qoss where it is 10X improved. To further illus-
trate the effect of Qg and Qoss, the well-known LLC circuit of figure 4 is
used to evaluate the ZVS performance of both devices. The same circuit
is used for comparison of both Si and GaN on the primary side running
at approximately 350 kHz, delivering 750 W from a 385 V bus.
Figure 5a shows the waveforms for the superjunction FETs on the pri-
mary of the LLC, and figure 5b shows the GaN HEMT in the same
circuit, same conditions. In figure 5a, the upper gate turns-off and the
drain voltage takes more than 350 ns to slew from bus to zero volts.
The superjunction nonlinear charge creates long, shallow tails on the
voltage that mandate the long deadtime. As shown, the deadtime in
Figure 5a is 350 ns, and even then, when the lower gate turns-on, the
voltage has not yet reached zero (so it is near ZVS). This may seem to be
a small compromise, turning on slightly early before the voltage across
the switch is really zero, but it is not: don’t forget that nearly half the
Eoss still remains at only 20 V on the drain because of the nonlinearity
(refer to figure 3 blue Eoss curve). In other words, the deadtime shown
is as short as it can be without significant compromises in power loss
(and efficiency).
Figure 5b shows the same waveforms for the GaN HEMT. Note that the
gate voltage has a much faster rise and fall time than the superjunction
device. This is a result of the gate driver having a much easier time driv-
ing the low charge gate of the GaN device. Moreover, due to the low
Qoss of the HEMT, the drain voltage is linear and much faster as well.
Because of this, the deadtime can be 3X shorter, and have no additional
loss from non-ZVS.
A real-world LLC example: figure 6 shows two LLC power supplies.
They are both 3 kW telecom supplies operating from a 385 V DC bus.
Both were designed for high efficiency, and their efficiency curves are
nearly identical, hitting peak efficiency of 98.3%. The difference is that
the power supply on the left is operating at nominally 130 kHz using
superjunction silicon MOSFETs on the primary side, and silicon low
voltage MOSFET synchronous rectifiers on the secondary side. The
power supply on the right is operating at 350 kHz using 600 V GaN
HEMTs on the primary side, and silicon synchronous rectifier FETs.
Because of the ~3X higher operating frequency, the GaN power supply
on the right achieves 140 W/in3, nearly 3X higher density than the 50
W/in3standard Silicon-based power supply on the left.
The key to increasing power density is to operate at higher frequency,
MHz instead of 100 kHz. This reduces the size of magnetic components
and puts them in a range where circuit-board integration is feasible – in
other words integrating the windings into the PCB and using planar
core structures that can be machine-inserted. Including the windings
on the PCB also helps minimize the termination effects that cause hot-
spots on conventional magnetic structures. Termination effects occur
when all of the AC current sums into a connection point between the
magnetic and the PCB. Even if the transformer is manufactured with
Litz wire, it has to be soldered at some point, and that is where the
skin-effect makes the losses really high. With PCB winding on the
other hand, the synchronous rectifier and capacitor components can
be literally mounted right on the winding, completely eliminating the
termination losses, and additional losses due to parasitic impedance.
This technique can work at power levels from a few watts up to several
kW. At higher power levels, the transformer is generally divided into
several segments like a matrix transformer. Each transformer element
can operate at power levels up to 500 watts or more. The primaries are
placed in series (to ensure current sharing), and the rectified second-
aries are placed in parallel to sum the current just as they are on the
toroidal transformer pair in figure 6 right.
n
