1
A
PP
LI
C
AT
IO
N
N
O
TE
APT9703
By: Ken Dierberger
A New Generation of Power MOSFET Offers
Inproved Performance at Reduced Cost
2
A New Generation Of Power MOSFETs Offers Improved
Performance At Reduced Cost
Ken Dierberger
Application Engineering Manager
Advanced Power Technology Inc.
Bend, Oregon 97702
ABSTRACT
In today’s power electronic market place, as in other areas of electronics, reducing cost is
necessary to stay competitive. A new generation of high voltage Power MOSFETs offers
lower on-resistance (RDS(ON)) using the same chip area as a previous generation of devices.
Smaller die MOSFETs are also being produced with this technology that have the same RDS(ON)
as a previous generation but now at a lower price.
This paper covers the use of these devices in a phase shifted control mode full bridge DC/DC
converter featuring zero voltage switching. A comparison was made between the performance
of an older generation device to the new generation device. The comparison shows that the
new smaller die device, while providing reduced cost, produced equal performance to the
older generation.
Introduction
Advances in resonant and quasi-resonant
power conversion technology offer alternative
solutions to a conflicting set of square wave
conversion design goals; obtaining high
efficiency operation at a high switching
frequency from a high voltage source.
Presently, the conventional approaches are
by far, still in the production mainstream.
However, an increasing challenge can be
witnessed by the emerging resonant and
quasi-resonant technologies, primarily due to
their lossless switching merits.
The concept of quasi-resonant, “lossless”
switching is not new, most noticeably patented
by P. Vinciarelli [1]. Numerous efforts focusing
on zero current switching ensued, first
perceived as the likely candidate for
tomorrow’s generation of high frequency
power converters. In theory, the ON-OFF
transitions occur at a time in the resonant
cycle where the switch current is zero,
facilitating zero current switching, hence zero
power switching. While true, two obvious
concerns can impede the quest for high
efficiency operation with high voltage inputs.
By nature of the resonant tank and zero
current switching limitation, the peak switch
current is significantly higher than its square
wave counterpart. In fact, the peak of the full
load switch current is a minimum of twice that
of its square wave kin, significantly increasing
conduction losses. In its OFF state, the switch
returns to blocking a high voltage every cycle.
When activated by the next drive pulse, the
MOSFET output capacitance (Coss) is
discharged by the FET, contributing a
significant power loss at high frequencies and
high voltages. Instead, both of these losses
are avoided by implementing a zero voltage
switching technique.
APT9703
3
Zero Voltage Switching
Conventional zero voltage switching can
best be defined as square wave power
conversion during the switch’s ON-time with
“resonant” zero volt turn-ON switching
transitions. For the most part, it can be
considered as square wave power using
either a constant OFF or ON time while
varying the conversion frequency to maintain
regulation of the output voltage Fixed
frequency conversion using variable OFF and
ON times can also be used to maintain
regulation of the output voltage.
The benefits of lossless Zero Voltage
Transition (ZVT) switching techniques are well
known through the power supply industry.[2]
The parasitic circuit elements can be used
advantageously to facilitate resonant
transitions rather than snubbing dissipatively.
The resonant tank functions to put zero
voltage across the switching devices prior to
turn-ON, eliminating power loss due to the
simultaneous overlap of switch current and
voltage at each transition. High frequency
converters operating from high voltage input
sources gain significant improvements in
efficiency with this technique. The full bridge
topology as shown in Figure 1 will be the
specific focus of this paper, with emphasis
placed on the fixed frequency, phase shifted
mode of operation.
Phase Shifted Control
The phase shifted mode H-bridge
functions by applying two square waves to
the primary of a transformer. For an output
duty cycle of D=1 the two square waves would
be 180 degrees out of phase as shown in
Figure 2. For an output duty cycle of D<1
one of the square waves would be phase
shifted resulting in the primary of the
transformer being short circuited during a
portion of the period, Figure 3.
+VIN
-VIN
QA QC
QDQB
T1
LO
CO
D1
D2 VO
A phase shifted mode power supply can
be implemented using a conventional H-
Bridge circuit, Figure 1. The switch drive
signals are fixed frequency square waves with
QA and QB switched 180 degrees out of phase
and QC and QD switched 180 degrees out of
phase. Using the switch drive signal of QA
and QB as a reference, the switch drive signal
of QC and QD will be phase shifted to create
the duty cycle needed to produce the correct
DC output.
Figure 2. Phase Shifted Mode for D=1.
Figure 3. Phase Shifted Mode for D<1.
Figure 1. Full Bridge Topology.
APT9703
4
The gate drive signals for switch operation
is shown in Figure 4. Interval 1, QA and QD
are ON and QB and QC are OFF resulting in
a positive voltage output from the transformer
and transferring power to the inductor and the
load. During interval 2, QC is switched ON
and QD is switched OFF resulting in QA and
QC shorting the primary of the transformer
producing no output voltage and no power is
delivered to the inductor from the power
source. The inductor will free wheel during
interval 2 and along with CO will continue to
deliver stored energy to the load. During
interval 3, QA and QB are switched OFF and
ON respectively resulting in a negative voltage
being applied to the transformer. Again the
power source will deliver power to the inductor
and the load. The final interval 4, QC and QD
are switched OFF and ON respectively
resulting in a short across the transformer and
no output voltage is produced. Again no
power is delivered to the output and the
inductor, along with CO will continue to deliver
stored energy to the load. The next interval
QA
QB
QC
QD
1 2 3 4 5INTERVAL
GATE
DRIVE
OUTPUT
T1
Figure 4. Gate Drive and T1 Output Signals.
5, QA returns to ON and QB is returned to OFF.
With QB and QC OFF and QA and QD ON this
is the same conditions that existed in interval
1 and one cycle has been completed.
It is noteworthy to point out that this circuit
is hard switched and considerable power will
be lost as a result of switching losses. The
circuit, as described, will suffer from shoot-
through current and resulting losses as QA-
QB or QC-QD will most likely conduct current
simultaneously during the transition between
intervals.
One method commonly used to prevent
shoot-through current power losses is to delay
the turn-ON of one switch until the other
switch in the leg has completely turned OFF.
If we examine the use of this technique during
the transition from interval 1 to 2 we note
something interesting.
Zero Voltage Switched Phase Shifted
Control
Figure 5 shows a MOSFET H-Bridge with
output capacitances and intrinsic diodes
included. Also shown as a separate
component is the transformer leakage
inductance LR.
Examining the switching transition
between intervals 1 and 2 in more detail, with
a short delay added between the turn-OFF of
QD and the turn-ON of QC, reveals it more
than just limits the shoot-through current.
When QD is turning OFF the voltage at the
drain of QD will only rise as fast as the primary
current of T1 can charge the capacitance CD,
and discharge the capacitance CC. If the turn-
OFF speed of QD is much faster than the
charge/discharge time of CD/CC the drain
voltage of QD will remain low during turn-OFF
and this will minimize the turn-OFF losses of
QD. The drain voltage will continue to rise,
but will not dissipate any power as energy is
being stored in CD and the energy stored in
APT9703
5
CC is returned to the input supply +VIN, until it
reaches a voltage slightly above the positive
supply voltage (+VIN). At this point the intrinsic
diode DC will begin to conduct clamping the
drain voltage to VIN. After DC is conducting
the voltage across the drain-source of QC is
essentially zero and QC can be turned ON
with a Zero Voltage Transition (ZVT) resulting
in lossless switching. A MOSFET channel can
conduct current in both forward and reverse
directions and turning the MOSFET ON while
the body diode is conducting will result in
reducing the conduction losses of the diode.
The current driving the transition of the
right leg can be approximated as the reflected
load current in the primary of transformer T1
(IP). During the right leg transition the current
IP will remain relatively constant as current
will continue to flow through the output diode
(D1) into the output inductor (LO). The
resulting right leg transition time is a linear
ramp calculated from equation (1) :
PI = RC
dv
dt
or dt = RC
dv
PI
(1)
where:
IP is the primary current of T1 at the end of
interval 1
dv=VIN
dt is the transition time
CR is the effective value of CC+CD which is:
RC =
4
3 CC
+ DC( ) (2)
The 4/3 factor is used to estimate the average
output capacitance over the drain voltage
range.
At the end of the right leg transition the
primary voltage of T1 will be zero as both QA
and QC are conducting, resulting in a short
circuit of the primary. With the primary of T1
at zero volts the secondary will also be at zero
volts, stopping transfer of power to the output.
The presence of LR in the primary circuit
of T1 will steer the output current to continue
to flow in D1 during the clamped freewheeling
interval. Note that the current in the
secondary windings never splits in half during
the OFF period of power transfer as it does in
a conventional, non-phase shifted H-bridge.
During interval 2 the current in LR will remain
nearly constant as the current in LO is still
being reflected to the primary.
At the end of interval 2, QA will turn-OFF
and QB will turn-ON, with a short delay added
between the events. When QA is turning OFF
the voltage at the source of QA will only drop
as fast as the current supplied by LR can
charge CA and discharge CB. If the turn-OFF
+VIN
-VIN
QA QC
QDQB
T1
LR
LO
CO
D1
D2 VO
DA DC
DDDB
CA
CB CD
CC
Figure 5. H-Bridge With Parasitic Diodes
and Capacitances Shown.
APT9703
6
speed of QA is much faster than the charge/
discharge time of CA/CB the drain-source
voltage of QA will remain low during turn-OFF
and this will minimize the turn-OFF losses of
QA. The source voltage will continue to drop
until it reaches a voltage slightly below the
supply voltage return (-VIN). This transition
will not result in any power dissipation as
energy is being stored in CA and the energy
stored in CB is returned to the input supply
+VIN. At this point the intrinsic diode DB will
begin to conduct clamping the drain voltage
to -VIN. After DB is conducting the voltage
across the drain-source of QB is essentially
zero and QB can be turned ON with a Zero
Voltage Transition (ZVT) resulting in lossless
switching.
Note that the left leg transition is resonant
and not linear like the right leg transition. As
soon as the left leg transition begins the
reflected output inductor disappears from the
primary circuit. Reverse voltage being
impressed on the primary of T1 will begin the
output diode commutation from D1 to D2
resulting in a short circuit of the secondary
winding during the left leg transition thus
removing the reflection of LO from the primary
circuit.
The left leg resonant transition must be
fueled by energy stored in LR. The exact
circuit to describe this transition is a series L/
C circuit with an initial current flowing equal
to IP at the start of the transition. LR is the
total series primary inductance and CR is the
effective circuit capacitance, 4/3(CA+CB). If
the leakage inductance, LR, of T1 is too low
and has insufficient energy to complete the
resonant left leg transition a small inductor
will need to be added in series with the primary
of T1. The primary current during this
transition has a sinusoidal shape with the
peak amplitude occurring at the beginning of
the transition. Because of this, solving for the
exact transition time will require taking the
arcsin of the function describing the transition
parameters at the beginning of the transition.
TRAN(LEFT)t =
1
Rω
arcsin INV RZ
PI
(3)
Where ZR is the resonant tank circuit
impedance and ωR is the resonant tank self
oscillating frequency in radians:
RZ = RL
RC
(4)
Rω =
1
RL RC
(5)
At the end of the left leg transition T1 will
have VIN impressed across the primary as
both QB and QC are conducting. With the
primary of T1 at VIN the secondary will be at
VIN multiplied by the turns ratio of T1,
resuming transfer of power to the output.
At the end of interval 3 an identical
analysis can be done to facilitate the transition
from the conduction of QB and QC back to
the conduction of QA and QD. The only
difference is QC will be exchanged with QD
and QB will be exchanged with QA in the
discussion. The right leg will transition first in
a linear mode as before and the left leg will
transition second in a resonant mode as
before. The shorted primary function will be
done by the lower switches QB and QD.
A more detailed discussion of the
preceding can be found in reference [2].
Evaluation Power Supply
To demonstrate the Phase Shifted ZVT
Controlled Power Conversion technique we
obtained an evaluation board from Unitrode
Corp. which uses their UC3875 PWM
controller IC. The evaluation board was
APT9703
7
designed for 250 watts +12 to +32 volts
output. Advanced Power Technology (APT)
made some modifications to the evaluation
power supply to convert it to 1000 watts at 48
volts output. Specifications for the power
supply are:
VIN 400VDC
VOUT 48VDC
IOUT 20.5ADC
POUT 1000W
Efficiency >93%
400VDC input voltage was chosen
because this allows the power supply to be
operated from a Power Factor Corrected
(PFC) universal input (85 to 265VAC)
preregulator. The 48VDC output voltage was
chosen to match the requirements of the
telecom industry.
Design Overview
The circuit used for the evaluation power
supply is shown in Figure 6 with the parts list
shown in Appendix 1.
This design conforms in principle with the
Unitrode application notes References [2, 3,
4]. Details of the differences between this
design and the aforementioned application
notes can be found in reference [5]. In
summary the main differences in the
evaluation power supply are: 1. The control
circuitry is ground referenced to the output
DC ground not the input DC ground. This
improves the regulation of the output voltage
and allows easier implementation of over
current protection. 2. Coupling capacitors C1
and C2 are inserted in series with the driver
transformers primary to improve the symmetry
of the v-t product for the transformer cores
and protects against saturation. A coupling
capacitor C3 has been added in series with
the power transformer T4 and has the same
function as C1 and C2. 3. The v-t product
can be asymmetrical if there are differences
in coupling of both secondary halves to the
primary, differences in the forward voltage
between the rectifier diodes or by modulation
of the error amplifiers output by clock-
synchronous distortion. The disadvantages
of this method are: The gate drive
transformers may require additional insulation
to meet safety standards. An isolated auxiliary
power supply must be provided to power the
UC3875. This is easily implemented if a PFC
Preregulator is used to drive the converter.
Three major modifications made by APT
to increase the evaluation power supply
design to 1000W were: The main power
transformer was changed back to a planar
transformer similar to the design in reference
[3]. APT30D20K diodes were used for the
output rectifiers to accommodate the
increased output current requirements.
APT5020BNFR, APT5020BVFR and
APT5017BVFR FREDFETs were all
evaluated in the circuit. The APT5020BNFR,
APT5020BVFR were found to provide equal
performance. The APT5017BVFR showed a
few tenths of a percent better efficiency. All
devices offer low RDS(ON) to minimize
conduction losses which are the dominant
losses of the switching transistors. The
APT5020BVFR is the better choice due to its
lower price. The FREDFET was selected to
provide better safety margin for commutative
dv/dt of the body diode during no-load and
short circuit operation of the power supply.[6]
Several minor modifications were needed
to adjust the operation of the controller to
accommodate the higher power and output
voltage. The clock frequency was increased
to 300KHz to provide a conversion frequency
of 150KHz. The ramp and slope were
adjusted for the new frequency. The current
sense resistor was changed to 5 milliohms.
The voltage divider for the error amplifier was
adjusted for the 48V output voltage.
APT9703
8
The power supply board layout is shown
actual size in Figure 7. Note the small heat
sinks required to cool the power MOSFETs.
In fact the power dissipation of each FET is
estimated to be under 2W at full load. A small
fan was required to cool the power supply as
the resonant inductor L1 was operating hot.
Additional refinement of this component
appears to be required. All other components
were cool to the touch after several minutes
of run time. The overheating of L1 prevented
long run time testing. Once L1 is optimized it
is believed that the better performance of the
APT5017BVFR will be more significant and
will make this device the better choice for the
power supply. The parts list is given Appendix
1.
+VIN
-VIN
-VAUX
+VAUX
R1
10
R4
10
R9
30
R16
15.8K R18
220K
R21
4.7K
T1
R2
2.7K
R5
2.7K
R10
270
R11
70.5K
R12
4.7K
R24
1KR25
1K
R23
33K
R28
33K
R22
470K
R13
0.005
R3
2.7K
R6
2.7K
L2 L3
L1
R
14
1
5.
8K
R
15
4
.7
K
R
17
4
70
R
19
1
1K
R
20
7
.5
K
R
26
1
K
R
27
1
K
L4
39µH
C28
2.2µF
C1
.22µF C2
.22µF
C3
.68µF C7
470pF
C25
47nF
C27
33nF
C9
1000µF
C10
1000µF
C11
1.0µF
C5
1500pF
C6
1500pF
C13
1000µF
C12
1µF
C14
0.1µF
C21
47pF
C20
47nF
C15
1.0nF C
16
0
.1
µF
C1
7
0.
1µ
F
C1
8
22
0p
F
C1
9
27
0p
F
C2
2
4
7n
F
C2
3
47
nF
C2
4
47
nF
11
16 17 1 6 5 2 3 15 7 418 19
20
1289131410
FR
EQ
. S
ET
SY
NC
VR
EF
SL
O
PE
SS RA
M
P
CS
+
E/
A
O
UT
E/
A-
D
EL
A/
B
D
EL
C/
D
E/
A+
S
G
ND
PW
R
G
ND
O
UT
D
O
UT
C
O
UT
B
O
UT
AVC
VI
N
IC1 UC3875
IC2 UC3610
D3,4,5,6
1N4148
D7
APT30D20B
D8
APT30D20B
D9
1N4148
T3
T4
T2Q1
APT5020BVFR
Q3
APT5020BVFR
Q2
APT5020BVFR
Q4
APT5020BVFR
5
4
6
8
7
IC3
1/2 LM258
+ -
6
2
1
8
C4
0.1µF
C8
4.7µ
Figure 6. Schematic of the Evaluation Power Supply.
Performance
The evaluation power supply was tested
over a load range of 100W to 1000W. The
unit maintained ZVT to a load of under 50%
and achieved an efficiency of over 93% from
a load range of 750W to 1000W. Load
regulation was ±0.18% from 100W to over
900W and was ±0.56% from 100W to 1000W.
The loss of regulation above 900W was due
to the required duty cycle reaching very near
100%. This could be corrected by an
adjustment in the transformer turns ratio.
A graph of the measured converter
efficiency is shown in Figure 8 for loads above
100W. Efficiency remains above 91% above
450W. The efficiency drops under 91% with
APT9703
9
Figure 7. Power Supply Board Layout.
C13
L4 C12 IC1
C1
4 R2
0
C2
3
C1
7
C1
8
C2
4
C2
0
C2
1 R1
8
R
27
R21 R25
R11
C11
R13
R26
R17
R9
C25
R23
C10
C9
IC2
C1 R4 C2
D6
D4
D5
D3
T1 T2
T3
T4
C6 C5
L1
C8
D8
D7
Q2 Q1 Q3 Q4
-VAUX
+VAUX
-VIN
+VIN
C28
C3
C7
R
19 C2
2
C1
5
R
14 R
15 C1
8