Increase of Output Power from IGBTs in High Power

High Frequency Resonant Load Inverters

Frode Kleveland, Tore M. Undeland

Norwegian University of Science and Technology (NTNU)
Faculty of Electrical Engineering and Telecommunications
0.S.Bragstads plass 2E, N-7491 Trondheim, NORWAY
fax: i-47 7359 4279
emails: <[email protected]>,<[email protected]>

John K. Langelid
Elva Induksjon A/S
Pb 760 Sentrum, N-3701 Skien, NORWAY
email:<[email protected]>

Abstract - The primary goal of this work is to replace The IGBT does not have an internal diode. This is because
MOSFETs with IGBTs in high frequency (-300kHz) high power the junction providing for conductivity modulation of the
(0.1-lMW) inverters with series resonant load, a topology collector drift region functions as a series diode. As a result a
commonly used for induction heating. good extemal free wheeling diode can be chosen without the
In this application, the IGBT losses are totally dominated by need of any series diode.
turn-off losses, and the turn-off losses decay less than linearly The IGBT is a widely used device, and is produced in large
with the current. This is an invitation to develop a new control numbers. It is a cheap component compared to the MOSFET
strategy which makes the necessary derating of the IGBTs less for this voltage and power level. This difference in cost
severe. For the conditions in this work, the new strategy can give becomes even larger if the series diode is needed for the
3 times more output power compared to a simple current MOSFET application. The IGBT-solution could be more
reduction destressing for the same number of paralleled IGBTs. attractive than a MOSFET-solution even if severe derating is
necessary for the IGBTs due to the high switching frequency.
I. INTRODUCTION Although an IGBT-solution seems far from ideal seen from a
technical point of view, it is the cost per kW that counts, and
The use of MOSFETs is not straightforward in these that dictates the technology in the end.
applications, due to the voltage level required (approximately
700 V at the DC link). Its internal diode can not be used as a 11. DESTRESSING
OF IGBTS: AN IMPROVED PRINCIPLE
free wheeling diode because a current through this diode sets
up the conditions for possible turn-on of the parasitic BJT. The In this paper, a more effective way of destressing IGBTs to
steep voltage rise across the reverse conducting MOSFET, make them capable of handling the losses at this high
caused by the complementary MOSFET turn-on, leads to frequency, is presented. In this application, the tum-off losses
destructive failure in this case (ref. [1,2]). Unwanted are totally dominant. Turn-off loss measurements show that
MOSFET tum-on can be avoided by carefully monitoring the the losses do not vary linearly with the current. At low current
converter, but the only completely reliable solution is a diode levels, the turn off loss to current ratio is larger compared to at
in series with the MOSFET drain and an external free higher current levels, as shown in fig. 6. This indicates that
wheeling diode. This latter approach is expensive. simply lowering the IGBT current level by increasing the

I 6-
A Load
current
,
\
Current
measurement
ooint
\ --U --U

Fig. I . Test circuit topology and current waveform

0-7803-6401-5/00/$10.00 0 2000 IEEE

2909

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.
 number of paralleled chips is a poor way of performing a 111. EQUIPMENT
AND MEASUREMENT
TECHNIQUE
destressing.
The sequence control method is described in fig. 2. By this A. Laboratory equipment
method, a destressing by reduced switching frequency rather
than by current reduction is achieved. Fig. 2a shows how the 1) Inverter. The test circuit is presented in figure 1. The
switching frequency of the IGBTs is lower than the resonant laboratory inverter is an H bridge with a damped resonant
frequency. The reduced switching stress allows the IGBTs to load. The resonant frequency is about 300 kHz, depending on
operate at much higher current levels than if the IGBTs were the damping which sets the current level. The DC voltage is
switched each period, like in a standard parallel operation 700V.
with equal current sharing in every cycle. The high current
operation of the IGBT secures the most effective tum off, or 2) Devices. The IGBT modules used in this work are Eupec
more precisely, the minimum turn off loss to current ratio. FF200 R 12 KS4 and Semikron 150 GB 125D, both equipped
This increases output power for a given IGBT loss level. with the fastest chips from Infinion Technologies. The FF200
Fig. 2b shows the coupling of the different sections. One is a 1200V 200A module equipped with the latest, thinner
section conducts the entire output current, but is relieved with transparent emitter NPT type IGBT. GB 150 is a 150.4 device,
currentless periods. The number of currentless periods equals and the chips fitted were the fastest 1200V NPT devices until
the number of sections minus 1. With 4 sections as in fig. 2a, the new technology with the thinner emitter was available. It
the sections have 3 currentless periods. has the standard transparent emitter thickness, but is equipped
A traditional inverter is an inverter with one section, and with lifetime killers, and is therefore temperature sensitive.
with all the IGBTs paralleled in this section. When using The thermal resistances from junction to case used in the
more than one section, parallel connection of several IGBTs calculations are 0.075 WW for Eupec FF200 and 0.095 WW
within each section will still be desirable to achieve the output for Semikron GB150.
power wanted. The IGBT current will then be less than the The external gate resistance is zero, but there is an internal
section current, which equals the output current. resistance of 2.5 ohm for the FF200 and 0.78 ohm for the
The aim of this work is to estimate, on basis of GB150. The chip temperature is 125OC for all measurements.
measurements, the output power of one H bridge containing
two IGBT modules when it is put into inverters with different 3) Control mode. The inverter runs semi continuously. This
section numbers. An inverter will consist of several H bridges. means that the inverter runs for some periods every second.
The IGBT stress, the power loss, will be equal for the This stabilizes all transient conditions, and generally emulate
different section numbers when doing the comparison. a real life situation, without consuming a lot of power causing
excessive problems on the supply side, in the inverter itself

I
I- I A I I . section
1___1

Output
n - - L - -
voltage
__
I
- -i -i - -

2910

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.
 and in the load. A real life situation was not fully achievable the device, multiply them to find the power loss, and integrate
since the measurement bridge leg consists of only one to find the energy loss. For each measurement, an average loss
module, and not as many modules as sections. The value based on 10-12 sweeps is used.
measurements are taken when the output voltage switches When turning off a transistor, the C,, of the
each period, like they also do in a real situation, as shown in complementary transistor will be discharged into the load
fig. 2. The difference is that in the test equipment the IGBTs during the blanking time, see fig. 4. The energy in ($,s
have no currentless periods. The losses are measured over one recovers every cycle and can be measured as a negative loss
period, which mirrors the current carrying period shown in during turn on. This recovered energy must be subtracted
fig. 2. The losses in the currentless periods are approximated from the measured tum off energy to find the switching losses
to zero. because recovered energy will be measured as a positive loss
during turn off. The amount of recovered energy is
B. Measurement technique independent of the current level as long as the switching is
performed in the minimum loss inductive mode case, which is
1) Current measurements. Making accurate current described in the next paragraph.
measurements in this case is not so simple. The traditional
Tektronix current probe with current transformer gives offset 4) Switching trajectory. It has been paid attention to
problems when measuring currents containing DC achieve a minimum loss case for all measurements presented
components. The test equipment runs semi-continuous, which here. References [4] and [5] give a good description of the
gives large DC components. In addition, the tail current lasts minimum loss tum off, including numerous measurements.
the entire off-period. These two facts gives both offset errors The turn off in fig. 3 and 5 is examples of such a turn off. For
and small possibilities in detecting the error value to make a comparison, fig. 4 presents an almost minimum loss turn off
correction. Because of this it was decided to develop a case. Minimum loss occurs when the IGBT voltage just
resistive measurement shunt. A resistive transducer will not reaches the DC rail voltage as the complementary IGBT tums
give offset errors, but quite a lot of other problems must be on, as seen in figure 3 . To make the turn off loss in fig. 4 the
overcome to make the shunt accurate and reliable. This work smallest possible, the turn off instant should be slightly
will be closer described in a later publication. earlier, and the blanking time slightly smaller. If the turn off
occurs at a too high current level, the voltage rise will be
2) Conduction voltage drop measurements. The conduction steeper and linear. As the current rises, a hard turn off will
losses is also very important to investigate. The forward occur, and the voltage rise will be restricted only by the Miller
recovery process taking place as the device turns on gives effect, like in a typical motor drive application. The losses
excessive losses compared to the datasheet due to the high will increase.
switching frequency. An active diode clipping network bolted No additional capacitors are used. The parabolic voltage
directly on the IGBT module was made to measure the rise seen in fig. 3 and 4 is only due to the nearly linearly
conduction voltage drop. decaying current flowing through the IGBT after turn off.
This current is used to remove the stored charge caused by
3) Loss calculations. Examples of tum-off and conduction conductivity modulation. This process is better known as tail
loss measurements are shown in fig. 3. The measurements are current. The tail current will continue to flow also after the
performed by measuring current through and voltage across voltage has reached the rail level. During the voltage rise, the

Fig. 3. An example of loss measurements for the Eupec FF200 device, Left figure shows tum-off loss, right shows conduction loss
Channel I is IGBT voltage (200V/div left fig., 2OVldiv right fig.)
Channel B is IGBT current (ZOA/div)
Channel D is dissipatedenergy for the current sweep
(2mUdiv left fig., 0.4mJ/div right fig.)
Channel A is average dissipatedenergy for IO sweeps
(2mUdiv left fig., 0.4mUdiv right fig.)

2911

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.
 7-Mar-00
9 :39:12

Complementary
IGBT turn-off IGBT tlllll-on

17-Mar-00

i< y
i Blanking time I

Fig. 4. An almost minimum loss turn off case for comparison with minimum
loss turn off in fig. 3. Fig. 5 . Switching trajectory for current values of 40 A and 200 A.

depletion zone must also be created, which is the same as current, compared to the peak value, is smaller for high peak
charging the intrinsic capacitance. These two processes current values then for smaller. All these observations point
determines the voltage rise. In addition, the channel will be towards relatively smaller turn off losses at higher peak
slightly in a conducting state, dependent on the gate resistor. current values.
The channel will act like a leakage current parallelling the
voltage rising process and hence slow down the voltage rise IV. MEASUREMENTS
rate. This influence will be weak at an ideal turn off, but
become more and more important as the turn off current A . Measurement results
becomes larger than optimal.
When inspecting fig. 5 , it is easier to justify the strong non Fig. 6 shows loss measurements for the Eupec FF200 and
linearity seen in the switching losses in fig. 7. At 40 A peak the Semikron GBl50 IGBT modules. The non linearity of the
current, the turn off current value is 25 A. At five times the turn off loss as a function of current is clearly seen, the losses
peak current, 200 A, the turn off current value is 80 A, a 3.2 are relatively larger at low currents.
times increase. It is also seen that the voltage rise rate is larger It could be worth to notice that in addition to the device
at the 200 A peak current case compared to the 40 A case. By itself, the resonant load and the use of minimum loss turn off
inspecting the tail current, it is also possible to see that the tail

Turn-off(o), turn-on(x) and conduction(") losses for FF200 Turn-off(o), turn-on(x) and conduction(") losses for GBI 50
6 6

5 5

4 4
=i
E E
z3 3
::
-1
U)

2 2

1 1

0
0 50 100 150 200 0 50 100 150
Current [A] Current [A]
Fig. 6. Turn-off, turn-on and conduction loss measurements. Left, Eupec FF200, right Semikron GB150.
Current is peak current during on state. Conduction loss is conduction energy loss pr cycle at 300 kHz.

2912

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.
 instant will influence the loss curves. The approach used in 2) Results. In fig. 7 and 8, Current is the IGBT peak current
this paper may not be valid for other applications. value for the current conducting period, Output power is the
The continuous lines are approximations achieved by power delivered per H bridge containing two IGBT modules,
fitting polynomial functions. and Power loss is the losses in one IGBT valve.
It is underlined that the data given is for the smallest units
B. Treatment of the measurement results of the inverter, namely one single H-bridge or one single
IGBT valve. The reader has to multiply the quantities given to
The continuous approximation in fig. 6 makes it possible to get the right values for an inverter with a certain section
do interesting calculations not restricted to only the current number, and for a certain number of paralleled modules in
values of the measurement points. To evaluate the derating each section. The principle of the section shared inverter was
principle in a section shared inverter, it is illustrating to explained in section 11, destressing of IGBTs: An improved
calculate the output power for each IGBT module as a principle. There is also an example in the following.
function of the number of sections. As fig. 7 and 8 show, a significant improvement is
1) Calculations. The output power is of course restricted by achieved by using a section shared inverter’s frequency
loss demands. In the calculation, the IGBT current for each derating rather than a traditional current derating. A
section number is chosen so the power loss equals the loss traditional inverter uses one section.
demand, which is calculated from the modules thermal Fig. 7 shows the results for the Eupec FF200 device. The
resistance and the desired temperature fall from junction to output power per H bridge is raised from 4.5 kW in a
case. The current fulfilling the loss demand is basis for the traditional inverter to 14 kW when using 5 sections. The
output power calculation, which is calculated as the DC IGBT current will then be around 170 A, compared to below
power minus the IGBT power loss. 20 A when using 1 section.
It is worth to notice that the loss measurements are only Fig. 8 shows the results for the Semikron GB150 device.
taken up to a certain current level, approximated data are only Here, the output power increases from 3.5 to 9 kW, and the
available for current levels slightly above this. In calculations IGBT current increases from 10 to 160 A.
for large section numbers, the current level becomes fixed, The measurements and calculations show that the FF200 is
and both output power and power losses per H bridge the best device for this application with 50% more output
decrease. power compared to GB 150. It has also the benefit of not being
The continuous loss approximations and the calculations temperature sensitive. However, the chip area and thereby
are performed in MATLAB. capacitances are greater for the FF200.

Device: FF200, Tj-Tc = 25 [K], load freq. = 300 [kHz1 Device: FF200, Tj-Tc = 25 [K]. load freq. = 300 [kHz] Device: FF200, Tj-Tc = 25 IKI, load freq. = 300 [kHzI
15 400

350
200
300

-i50 210 5250

s
I

53 200
E
5100 6 5150
a 5
100
50
50

0 0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Number of Sections Number of sections Number of sections

Fig. 7. D a t a for Eupec FF200: IGBT current level, output power per H bridge and power loss per IGBT for different number of current sections.

Device: GB150. Tj-Tc = 25 [KI, load freq. = 300 [kHrl Device: G8150. Tj-Tc = 25 [K], load freq. = 300 IkHz] Device: G8150, Tj-Tc = 25 [K], load freq. = 300 [kHzl
10
180
250
160
8
140 200
-120 5 L
-3 150
9
e 100
2E 80
B f
34 g 100
60 a
40 2 50
20
0 0 0
0 2 4 6 8 2 4 6 8 0 2 4 6 8
Number of sections Number of sections Number of sections

Fig. 8. Data for Semikron GB150: IGBT current level, output power per H bridge and power loss per IGBT for different number of current sections.

2913

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.
 One of the differences between these two devices is that IV. CONCLUSIONS
the GB I50 device benefits from a higher number of sections
than the FF200. The FF200 device seems to reach an optimum If IGBTs are used for high frequency applications where a
for 5-6 sections. This is maybe not so surprising taking into destressing of the devices has to be performed in order to
account that the FF200 with its thinner emitter is more handle the switching losses, a section shared inverter is
MOSFET like. desirable compared to a standard parallel connection where all
elements switches in time and conducts the same current.
C. Example By letting each section conduct more current, and instead
relief them with currentless periods, a far more effective turn
Let us take an example with an inverter with five FF200 off is achieved. This gives 3 times more output power
modules paralleled in each inverter leg, which gives five compared to a standard current reduction for the conditions
paralleled H-bridges. A temperature fall of 25 OC from investigated in this paper. The improvement is of course
junction to case is desired. dependent on switching frequency and also cooling
Using the traditional inverter control with one section, the engineering.
IGBT current becomes 13 A. The DC power from one H As for the devices used, the Eupec FF200 device seems to
bridge becomes then 13A.(2/.rr). 700V = 5 . 8 k W . The be favourable compared to the Semikron GB 150.
output power from one H bridge becomes Positive second order effects is severe reduced driving
5.8kW-(4. 330W) = 4.5kW when subtracting the power power and less problems caused by large amounts of chips in
losses. This gives a total inverter power output of parallel. A drawback is restricted numbers of paralleled
5 4.5kW 22.5kW. modules and hence output power for a designed inverter.
The inverter control is now rearranged, and just one
module takes the load current. However, the load current
circulates between the 5 modules. (The inverter is shared into
5 sections.) The IGBT current is now 170 A, but floats only
every fifth period. This gives a DC power per H bridge of ACKNOWLEDGMENT
170A. 2 / ( n . 5). 700V = 15.2kW .Theoutputpowerfrom
one H bridge becomes 15.2kW-(4’ 330W) = 13.9kW We want to thank Terje Rogne at Sintef Energy Research
when subtracting the power losses. The output power is for numerous discussions.
increased to 5 . 13.9kW = 69.5kW with the same power
loss as in the first case.
D. Second order effects
REFERENCES
1) Drawbacks. In practical use, a large number of sections
is not so desirable because then a determined number of [l]. Mohan N., Undeland T. and Robbins W. Tower
modules must be used. 6 sections, for instance, give 6 Electronics: Converters, Applications and Design’:
paralleled modules for each leg. The next allowable numbers Second Edition. John Wiley & Sons, 1995.
are 12 and 18, and so on. The freedom of choosing output [2]. Baliga B. J. “Power Semiconductor Devices”. PWS
power is restricted, and the principle is absolutely best fitted Publishing Company, 1995.
for high power applications, which also is our aim. 131. J. Yamashita, T. Yamada, S. Uchida, H. Yamaguchi, S.
Ishizawa. “A relation between dynamic saturation
2) Benefits. When the IGBTs switch more seldomly, the characteristics and tail current of non-punchthrough
driving power decreases. Driving power is surprisingly large IGBT”. IAS conference oct.1996.
with heavy parallel connection at these frequencies, and a [4]. Kleveland, F., Undeland T, Langelid, J. K., and Rogne T.
decrease to 1/4 using 4 sections is a convenient property. “IGBTs and Schottky diodes in high power high
Another positive property is that current sharing between frequency applications”. EPE 99.
paralleled modules and chips probably cause less problems. [5]. Johansen J. K., Jenset F. and Rogne T. ‘Characterization
The main reason for this is that the current level and thereby of High Power IGBTS with Sinewave Current’: IEEE
conduction voltage drop will increase, which will stabilize the Transactions on Industry Applications, Vol. 30, No. 5,
current sharing, In addition, fewer chips will share the current, 1994.
and they will be closer located

2914

Authorized licensed use limited to: UNIVERSITY OF LJUBLJANA. Downloaded on November 28,2021 at 17:17:47 UTC from IEEE Xplore. Restrictions apply.