Académique Documents
Professionnel Documents
Culture Documents
power IGBTs
R. Azar, F. Udrea, W.T. Ng, F. Dawson, W. Findlay, P. Waind and G. Amaratunga
Abstract: A novel IGBT electrothermal model is implemented for the first time in PSpice for the
simulation of steady state and transient temperature dependent IGBT operation including self-
heating and latchup. A thermal circuit representing the characteristics of the IGBT package is
developed and validated against a finite element model and experimental results. A novel electrical
IGBT model based on the Kraus model is developed to account for the electrical impact of
instantaneous junction temperature variations owing to self-heating. The resulting electrothermal
model is validated against experimental DC and transient FBSOA measurements.
1 Introduction
22 0.0
on the Kraus model [2] has been developed for the study of 2
2 .0
0
.0
18 .0
dynamic performance variations due to self-heating effects
16 .0
10
14 2.0
and prediction of the temperature dependent forward bias 0
1 .0
safe operating area (FBSOA) and short circuit safe
10
y,
µm
cathode µm 200
8. 0
operating area (SCSOA). This complete model has been
0 anode
x,
6. 0
4. 0
0
0
IEE Proc.-Circuits Devices Syst., Vol. 151, No. 3, June 2004 249
The total equivalent junction to sink thermal resistance is 180
then found by considering all package layers in series and
160 Temperature reached
summing their respective thermal resistances. The total P/50 (W) after 50 µs pulse according
junction to case equivalent thermal resistance is found to be 140 to Zth measurements
20.0 K/kW.
A two-dimensional thermal finite element analysis in 120
temeperature, C
Pspice FEMLAB
FEMLAB is also performed. It is determined that lateral
100
thermal interference between adjacent IGBTs only has a
4% impact on the junction temperature. The junction-to- 80
case equivalent thermal resistance is found to be 19.4 K/kW.
Both the circuit and FEMLAB approaches do not consider 60
the three-dimensional nature of the heat flow profile in the FEMLAB-PSPICE
40 Average Error = 4%
package and therefore yield conservative thermal resistance
results when compared to the worse case experimental 20
measurement of 18 K/kW.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
2.3 Dynamic thermal analysis of the time, s × 10−4
package
A dynamic analysis of the package thermal response to an Fig. 2 Comparison of FEMLAB results with transient FBSOA
applied power pulse requires the consideration of each experimental calculations and with PSpice thermal circuit model
layer’s thermal capacitance. In the case of transient power
transfer to the package, say during turn-off or turn-on of one-dimensional PSpice model. The final RC equivalent
the IGBT, the energy transferred through the surface of the thermal circuit is displayed in Fig. 4.
package in time dt can be expressed as Pdt. This energy will
cause a temperature rise dT according to each material’s 3 Electrical modelling
specific heat c and mass M [7].
Currently, the Kraus IGBT model [1] remains the model of
Pdt ¼ cMdT choice for use with the PSpice simulator because it offers an
dT dT ð2Þ excellent trade-off between complexity and accuracy in DC
P ¼ rmd cV ¼ Cth and transient simulations. However as a purely electrical
dt dt
model, the Kraus model does not allow for the simulation
rmd is the material density in kg/cm3, c the specific heat in of any thermal effects.
J/kgK and V the volume in cm3. Equation (2) defines the
thermal capacitance Cth of each material. It is apparent 3.1 A novel self-heating Spice IGBT model
from (1) and (2) that the transient thermal impedance of A new temperature pin is added to the Kraus IGBT model.
the IGBT package can be represented by a multiple time Since PSpice does not allow dynamic variations of
constant network using current as the analogue of thermal parameters between bias point calculations, each semicon-
power and voltage as the analogue of temperature [9]. Every ductor parameter is translated to a voltage at a node inside
package layer is then represented by an RC circuit where R separate subcircuits that depend on the voltage at the
and C are extracted from material properties and represent temperature pin (Fig. 3). The voltage at a given node, say
the thermal resistance and capacitance of the material. The ni(T), will then represent the semiconductor parameter’s
power generated by the IGBT is inserted in the thermal value at the given instantaneous junction temperature, in
circuit as a current source, and the temperature of the this case the intrinsic carrier concentration. The equations
source junction is then given by the potential at a given
node in the circuit. tau 0
Ise 0
un 0
up 0
+
ni 0
temperature
power pulse is analysed in FEMLAB. According to the
experimental thermal transient impedance curve and the on- − VT
anode
state current measurements, it is calculated that the 0
ni (T)
un (T)
Ise (T)
up (T)
tau (T)
250 IEE Proc.-Circuits Devices Syst., Vol. 151, No. 3, June 2004
Ls Va
Rg Lg 1
temp
ano
2 0
gate cat
+
−
+ Vg Tj
S + IN1+
−
3
−
IN1-OUT+
+
−
Tsink HNOM
IN2+OUT-
IN2-
multiplier
Deltat Si 0
IGBT
power 3.39e-5 1.0e-3 6.7e-5 6.7e-5 2.7e-4 2.7e-4 2.7e-4 1.1e-3
2.1e-2
1.1e-3 1.1e-3 4.3e-3 4.3e-3 4.3e-3 27.6e-3 31e-3 1.9e-2 1.9e-2 2.4e-2
3.51e-3 3.51e-3 1.405e-2 1.405e-2 1.405e-2 3.66e-2 11.42e-2 16.46e-2 11.42e-2 2.74e-2
solder Cu tile Cu solder
Vg = 15 V
The novel IGBT model is coupled to the package thermal 400
circuit developed above through dynamic temperature
anode current, A
IEE Proc.-Circuits Devices Syst., Vol. 151, No. 3, June 2004 251
self-heating due to the package. Comparing the DC current accumulation subcircuit (Fig. 3) so that the n-p-n latchup
simulation results obtained with the original Kraus model current contributes to an increased base charge leading to
and this fully coupled electrothermal model (Fig. 5a and 5b) a reduced base resistance, thereby increasing Ipc. This
shows that the increased temperature due to self–heating feature allows one to model the thyristor action leading to
leads to a reduction in the injection efficiency and in the temperature dependent latchup.
channel mobility, thereby leading to earlier saturation of the The electrothermal on-state simulation is repeated this
I–V curves and subsequent reduction in current levels time using the additional latchup circuitry described above.
deeper in the saturation region. Note that the DC current is, The FBSOA is obtained based on both the junction
however, not significantly affected below Va ¼ 5 V. This overheating limit of 1251C and on temperature dependent
increases model accuracy deep in the saturation region latchup (Fig. 6a and b). It is found that in steady-state
allowing one to model the steady state and transient operation the IGBT module will always reach overheating
forward biased SOA and compare results with experimental well before reaching latchup. Thus the FBSOA is limited by
measurements. the thermal resistance of the package rather than by the
p-well resistance.
4.2 Steady-state forward bias safe As can be seen from Fig. 6a, very good agreement is
operating area modelling reached between the thermally limited FBSOA simulation
A controlled current source is added between the cathode and the experimental measurements. The temperature
and the base in Fig. 3 to model the parasitic n-p-n BJT dependent latchup simulation results in Fig. 6b are
effect. This current source models a temperature dependent consistent with the SCSOA measurements showing typical
BJT current where the temperature is set to correspond to latchup temperatures in the order of 4201C.
the instantaneous voltage at the temperature pin. Therefore,
the temperature dependent BJT threshold voltage is taken 4.3 Transient forward bias safe operating
into account as well as the increasing p-well resistance due area modelling
to lower mobilities at higher temperatures. Using the electrothermal IGBT model, the transient
The p-well resistance responsible for latchup is calculated temperature response of the IGBT under short circuit
from MEDICI simulations by plotting the hole quasiFermi conditions for different pulse times of 50 ms, 100 ms and 1 ms
level and the hole current density throughout the p-well. At is simulated using the circuit in Fig. 4. Having determined
room temperature, a value of Rs ¼ 75 mO is found. The that the module reaches overheating before latchup,
feedback effect leading to latchup is achieved by measuring dynamic thermal feedback is implemented to sweep the
the parasitic n-p-n BJT current and feeding it to the charge anode and gate voltages deep within the saturation region
to determine the maximum allowable value of Va for a
150A given Vg for which the overheating limit is reached. The
simulated FBSOA
results are in good agreement with the transient FBSOA
experimemtal results
100A experimental measurements as shown in Fig. 7.
current
50A
104
SEL>>
0A 0 −I (VA)
125v
Pspice model
Vg = 20
predictions
temperature
Vg = 15 103
Vg = 10
100v
Vg = 11
Vg = 12
Ic, A
75v 102
0v 1.0v 2.0v 3.0v 4.0v 5.0v tp = 50 µs
α v(Tj) −273 anode voltage tp = 100 µs
a
Ic(max) DC tp = 1 ms
101
500A
Vg = 20 V
anode current
375A
Vg = 15 V Vg = 12 V
100
250A
100 101 102 103 104
125A Vg = 10 V Vce, V
T = 407 C
375 T = 401 C T = 398 C T = 399 C 5 Conclusions
IEE Proc.-Circuits Devices Syst., Vol. 151, No. 3, June 2004 253