500W, CLASS E 27.12 MHz AMPLIFIERUSING A SINGLE PLASTIC MOSFET.pdf

(63 KB) Pobierz
By:
APT9903
Richard Frey, P.E.
APPLICATION NOTE
500W, CLASS E 27.12 MHz AMPLIFIER
USING A SINGLE PLASTIC MOSFET
1
APT9903
500W, Class E 27.12 MHz Amplifier
Using A Single Plastic MOSFET
Richard Frey, P.E.
Advanced Power Technology, Inc.
Bend, Oregon 97702 USA
ABSTRACT
In this paper, we report on the design and evolution
of a 500W, 27 MHz Class E amplifier. It doubles
the operating frequency of previous high efficiency
amplifiers using MOSFET transistors in the TO-247
package. Device criteria, circuit design, and
amplifier performance characteristics are presented
and compared to a HEPA computer model.
INTRODUCTION
As the semiconductor industry moves to larger size
wafers to raise its productivity, they need more
control of the processing. This translates into the
need to produce a “harder” and more uniform plasma
in vapor deposition and etching operations that rely
on RF plasma. Typical operation is at 13.56 MHz
but 27.12 MHz produces better results and is gaining
increased popularity.
One of the challenges facing designers is finding
devices suitable for service at the higher frequency.
Around 1990, plasma equipment designers
discovered that some of the inexpensive plastic high
voltage MOSFETs they used in their switchmode
power supplies were capable of operating in high
efficiency Class E service at 13.56 MHz. Since the
switchmode devices operate at high voltage and cost
much less than the alternative purpose-built RF
parts, acceptance was almost immediate.
Purpose-built RF devices operate on supply voltages
below 50 Vdc. Suitable switchmode devices with
drain breakdown voltage of 1 kV can run on supply
voltages up to 300 Vdc. They also demonstrated
APT9903
system level benefits operating at higher voltage.
Power Combining and matching are easier since the
drain impedance is higher. Lower RF current is less
stress on the series elements and enables the power
supply design to be more efficient, smaller and
lighter.
This RF application challenges most standard
switchmode devices due to their packaging and die
layout. The drain of the MOSFET is connected to
metal back heat spreader, used for heat sinking the
device. This requires either an insulator between
the package heat spreader and the heat sink or
grounding the drain and changing the circuit
topology to accommodate it. Either method adds
assembly cost of the PA and increases the potential
for failures due to incorrect installation not to
mention the poor thermal transfer characteristics of
the insulator.
The package construction results in high source
inductance, limiting the frequency response of these
devices. In some cases, the circuitous gate
metallization layout caused gate signal propagation
delay, limiting the frequency response and reducing
the effective power dissipation of the device.
Raising the bar to 27 MHz reduces the number of
suitable switchmode parts. This is primarily due to
resistive losses associated with the polysilicon gate
structure in the closed cell geometry used in most
switchmode devices. Input capacitance is essentially
fixed by the device's power rating. Doubling the
operating frequency quadruples the gate dissipation.
If the ESR of the gate capacitor is large, it cannot
carry the RF current required to drive the device.
2
The use of metal gate conductors dramatically
reduces gate ESR. It is not unusual to reduce the
gate resistance by a factor of 100 -- from 4 to 0.04
ohms -- on devices of equal power rating. This
complicates RF matching to the gate, but it is now
rugged enough to operate reliably.
At 13 MHz, the device C
OSS
is combined with an
external capacitor to obtain the required shunt output
capacitance needed for proper class E operation [1].
At 27 MHz, the C
OSS
is typically larger than that
required for optimum Class E service. This
compromises optimum Class E efficiency and output
power capability. However, this would be a design
consideration for any RF device, not just
switchmode MOSFETs.
Vdd
C4
C9
L4
L1
Q1
J1
C1
T1
C6-C8
R1
L2
C5
C10
J2
PSPICE MODEL RF N-CHANNEL POWER
MOSFET
* ARF448A/B
27 July 1998
**
GDS
* .SUBCKT ARF448 6 4 1
CISS 3 5 1450P
CRSS 5 2 65P
LG 7 6 4 6N
M 8 5 3 3 125-050M L=2U W=1.4; DGSB LEVEL 1
J1 8 3 2 125-050J
D 3 2 125-050D
LS 1 3 2 3N
REGATE 7 5 .29
LD 4 2 4 5N
.MODEL 125-050M NMOS (VTO=3.4 KP=14u
Lambda=1m Gamma=.2 RD=130m RS=13m)
.MODEL 125-050J NJF (VTO=-25.5 BETA=.01
Lambda=.5)
.MODEL 125-050D D (BV=550 RS=230M CJO=422P
VJ=670M M=330M)
.ENDS
*
C3
Figure 2.
SPICE model for ARF448 MOSFET
Figure 1:
27 MHz Class E Amplifier Schematic
High voltage MOSFETs are now available that
combine the best practices from the RF world with
the economy of the switchmode devices and
packaging. They are available in mirror image pairs
and the heat spreader of the plastic TO-247 package
is connected to the source. They operate up to 300V
V
dd
, at frequencies up to 100 MHz.
AMPLIFIER DESIGN
The device used for this amplifier is an APT
ARF448A. It has a BV of 450V and R
θJC
of 0.55°C/
W. The class C gain is more than 25dB at 27 MHz.
C
OSS
is 125 pF at V
dd
=125V. R
D(ON)
is 0.4Ω. These
data sheet parameters were entered in the HEPA [1]
3
C1,C3
C4-C8
C9,C10
L1
L2
L4
Q1
R1
T1
75-380 pF mica trimmer, ARCO 465
.01 uF 1 kV disc ceramic
.1 uF 500V disc ceramic
6 uH. 25t #24 ga.enam. 0.5" dia.
210 nH. 4t #8 ga. .75" id, 1" long
2t #20 PTFE on .5" ferrite bead
µ=850
APT ARF448A
25Ω 5W non-inductive
Pri: 4t #20 PTFE, Sec: 1t brass tube
on 2 hole balun bead
Fair-Rite #2843010302
µ=850
Table 1:
27 MHz Class E Amplifier Part Values
APT9903
Class E design program and it generated starting
design values for the circuit. The circuit is common
to most Class E amplifiers. [2] See Figure 1. In
Class E amplifiers there is a capacitor shunted across
the drain to source. There is none used here because
the output capacitance of the device is slightly larger
than the optimum value for that component. This
defines the upper frequency limit for efficient Class
E operation of a particular device. The 3Ω ESR of
C
OSS
is one of the primary loss mechanisms in the
circuit. However, C
OSS
exhibits little of the parasitic
inductance that caused VHF ringing in the circuit
described by Davis. The output circuit values were
adjusted slightly for maximum efficiency at an
output power of 490 W but basically, it worked "as
advertised" right off the bat.
While this description of the output circuit sounds
straightforward, it was not reduced to practice very
easily. The problem was providing enough rf voltage
to the gate to drive the drain into saturation. The
input impedance of the gate at 27 MHz is 0.1 -j2.7.
C
iss
is 1400 pF. If 10V of peak gate drive is needed,
a reasonable match between the drive source and
gate is required. There is approximately 9 nH of
parasitic gate inductance. This is enough inductance
to make it impossible to observe the actual voltage
applied to "the gate" with an oscilloscope.
SPICE was used to model the gate drive circuit. A
SPICE macro model for the device is shown in
Figure 2. The goal was to design a network to match
the gate impedance sufficiently to permit sine wave
drive as in reference [2]. A circuit using a 4:1
transformer and an L-network was designed using
winSmith. [3] This worked, but the circuit was not
stable. The parallel equivalent of the gate is 2200
pF in parallel with 210 ohms. A 25 ohm 5W padding
resistance was placed across the gate. This raises
the effective input impedance to 0.38 -j2.6, lowers
the network Q, and makes it much easier to match
and drive properly.
The input transformer has been used before [4]. It
is made from a two hole ferrite "binocular" bead
balun. The secondary winding consists of two 7/8"
pieces of 3/16" diameter brass tubing connected with
APT9903
copper shim stock.
Figure 4:
Drain voltage waveform of 500W
Class E amplifier
The four turn primary is wound inside the tubes for
maximum coupling and minimum leakage which
measured 19 nH referenced to the secondary side.
High quality passive components are required in the
output network. Most important of these is L2. It
was wound from #8 ga. bare copper wire, its Q
measured 375, and the calculated dissipation is 4.2
W. This coil is not capable of continuous duty
operation unless it is attached with high temperature
solder and/or separate mechanical termination
support is used. It was necessary to parallel three
10 nF Z5U ceramic coupling capacitors to carry the
rf current.
COMPUTER SIMULATION
A simple SPICE model of the amplifier, similar to
that used by Davis in [2], was compared with HEPA
results and the measured results of the operating
amplifier. Overall, the agreement was good,
especially between HEPA and the ideal circuit
SPICE model. Attempts to insert the SPICE macro
model of the transistor into the amplifier was not
successful. A much more sophisticated model is
4
needed to adequately simulate the effects of the
nonlinear capacitances of the MOSFET. However,
the SPICE model was very useful for understanding
the gate drive problem mentioned earlier.
current foldback would be needed in the power
supply of a practicable amplifier.
90
1000
135
800
600
400
200
180
0
rel ef f
P out
% Id
0
45
Figure 3.
Drain voltage waveform from HEPA
HEPA assumes that the device is being driven into
saturation, the only input parameter it considers is
input drive power -- used for the overall efficiency
calculation. The gate drive was the biggest problem
in the design because a large RF-capable power
MOSFET has a very small input impedance.
One of the best tests for reliability of an amplifier is
mismatch load testing. Called "load pull" in some
circles, it is a test which describes what the amplifier
does when operating into a load other than 50 ohms.
The ARF448 has about 175 watts of available
dissipation in the test amplifier with its air cooled
heat sink. The performance at eight points around a
2:1 VSWR mismatch circle was calculated using
HEPA. Then the same test was run on the amplifier
itself with the drive duty cycle reduced to 50%. The
various loads were obtained using a variable L-
network and adjusting each load impedance with a
vector impedance meter. The differences between
the calculated and measured results were very small.
The measured load pull results are shown in Figure
5. The most obvious conclusion is that there is a
region in the high impedance inductive quadrant that
should be avoided. Protection would be required to
keep the transistor within its safe operating area if
operated in this region.
Without protection, the output power soared to 750
W and the drain current was almost twice normal.
The efficiency stayed quite constant, never losing
more than 11% at any load angle. Some form of
5
-135
-45
- 90
Figure 5:
Load pull result for 500 W nominal output
at 2:1 VSWR. Efficiency is displayed as % x 10.
Class E amplifiers are used commercially in RF
sources of plasma generators and as exciters for CO
2
lasers. In plasma generator applications it is known
that the load varies over a particular range. The
feed line length is chosen so the load presented to
the amplifier does not enter the +45° quadrant. Other
plasma generators have some form of active
matching network between the amplifier and the
load to handle the load pull issue. The drive or the
drain voltage is reduced until a reasonable match is
achieved. In CO
2
lasers, the drive is pulse width
modulated and under high VSWR conditions, the
pulse width is reduced until the load stabilizes.
EXPERIMENTAL RESULTS
The efficiency of the Class E amplifier is 13% better
with 30% more output power than was obtained
from an initial Class C amplifier designed by
classical methods using the same components. [5]
The schematic did not change, only the values of
L2 and C3 are different.
The performance of the amplifier is summarized in
the table below along with the calculated values from
HEPA. The results are fairly close.
APT9903

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika:

Zgłoś jeśli naruszono regulamin