Application Note Revision: 02

AN 20-001 Issue date:

Prepared by:
2020-01-29
P. Drexhage, A. Wintrich
Approved by: P. Beckedahl

Keyword: IGBT module, temperature sensor, thermal impedance

Calculating Junction Temperature using a

Module Temperature Sensor
1. Introduction .............................................................................................................................1
2. Temperature Prediction Goals .....................................................................................................2
2.1 Over temperature protection .................................................................................................2
2.2 Performance optimization ......................................................................................................2
2.3 Lifetime prediction ...............................................................................................................2
3. Integrated Temperature Sensors .................................................................................................3
3.1 Why are no Rth(j-r)/Zth(j-r) values specified in the module data sheet?............................................3
3.1.1 Influence of heatsink ......................................................................................................3
3.1.2 Influence of operating point.............................................................................................4
4. Determining Rth(j-r)/Zth(j-r) ............................................................................................................6
4.1 Measurement methods..........................................................................................................6
4.1.1 Thermocouple (Rth) ........................................................................................................7
4.1.2 Infrared camera (Rth) .....................................................................................................8
4.1.3 Vce method (Rth or Zth) ....................................................................................................9
4.1.4 Finite Element Analysis (Rth or Zth) ...................................................................................9
5. Simplified Method for Periodic Functions (Quasi-Steady State Conditions) .......................................10
5.1 Required circuit parameters (inverter example) ......................................................................10
5.2 Loss calculation..................................................................................................................10
5.3 Junction temperature calculation ..........................................................................................11
5.4 Correction factor for low output frequencies ...........................................................................12
5.5 Example (Three-phase PWM inverter) ...................................................................................13
6. Thermal Coupling ....................................................................................................................14
6.1.1 Determination of Rth/Zth matrix ......................................................................................16
7. Complex Method, Step-by-Step (Short, High Overload and Stall Torque Conditions) .........................17
7.1 Required circuit parameters (inverter example) ......................................................................17
7.2 Loss calculation..................................................................................................................17
7.3 Junction temperature calculation ..........................................................................................18
7.4 Example............................................................................................................................19
8. Summary ...............................................................................................................................20

1. Introduction

A legitimate but complex question is how does one use the integrated temperature sensor inside a power
semiconductor module to determine the virtual junction temperature? There are several answers depending
on the required accuracy and the goal of the temperature prediction. This application note will explain two
possible methods: one at the lower end and one at the higher end of the complexity scale. It is first necessary

© by SEMIKRON / 2020-01-29 / Application Note Page 1/22

PROMGT.1023/ Rev.7/ Template Application Note
to define the goals of the temperature prediction because it has an impact on the required device parameters,
qualification effort, and computing power for calculation. This document uses the example of a three-phase
2-level inverter circuit with IGBT and freewheeling diodes but these methods can be applied to other circuits
and semiconductors as well. Though the proposed procedures refer to module-integrated temperature
sensors, they can also be used for external sensors (e.g. on the heatsink). The intention is that the calculation
methods are implemented using a digital processor on a converter control board.

Table 1: Comparison of the two considered methods for junction temperature prediction

Simplified solution for quasi-steady state Complex solution for short high overload and
conditions stall torque conditions

Calculation of losses in one switch and assuming Calculation of the losses for any switch according to
that the other switches in a symmetric circuit have actual circuit conditions (VCC, Vout, Iout, cos(φ), fsw,
equal losses (e.g. IGBT1…6 in a three-phase T j)
inverter)

Only one thermal model (Tj to sensor Tr) per type of Calculation of the junction temperature using a Zth(j-
switch is required (e.g. one for diode, one for IGBT) r)-matrix including coupling of all relevant switches

Sampling rates on the order of 100ms…1s which Sampling rate on the order of 1/fsw (or multiples of
allows use of Rth instead of Zth that)

Result: average losses of periodic function and Result: instantaneous function of losses and device
average junction temperatures temperatures

Correction factor used to account for temperature Temperature swing with inverter output frequency
ripple due to fundamental output frequency is inherently calculated

“Envelope curve” for temperatures usable for device

protection

Advantage: Low computing power and data Advantages: Protection at low frequency or stall
volume torque possible. Information about temperature
cycle stress available.

Disadvantages: Limited protection in stall torque Disadvantages: High effort for model
or at short high overload. No usable information for implementation. High computing power required.
temperature cycle stress calculations. High data volume.

2. Temperature Prediction Goals

2.1 Over temperature protection

The most common use of temperature monitoring is simply to protect the semiconductors from operating
above their thermal limits. With gradually increasing junction temperature situations (e.g. increases in
ambient temperature or low magnitude/long duration overloads), this can be accomplished by comparing the
sensor temperature with a pre-determined (at the design stage) set point at which the system faults (or
issues a warning). For dynamic loads, power fold back curves can be developed which maximize the output
current at a given temperature (e.g. high current at low ambient temperature).

2.2 Performance optimization

The maximum output power of a system at a given operating point can depend on a variety of factors, among
them environmental (ambient air temperature, altitude) as well as electrical (fundamental output frequency).
The output current can be maximized for a given operating point while ensuring that the junction temperature
does not exceed its limits. However, this approach carries the risk of reducing power module lifetime if the
additional temperature stress is not considered (see explanation of power cycling in [2]).

2.3 Lifetime prediction

Most power cycling lifetime models are based primarily on the mean junction temperature and number of
junction temperature excursions (sec. 2.7.6 in [2]). In theory, this means that a system that is able to
calculate and store the value of actual junction temperature during operation could actively apply a lifetime

© by SEMIKRON / 2020-01-29 / Application Note Page 2/22

PROMGT.1023/ Rev.7/ Template Application Note
model to determine the “life remaining” in a system. In practice, this has proven difficult to achieve due to
the uncertainty of lifetime models coupled with the complexity (and cost) of storing, processing, and
evaluating such data in the field.

3. Integrated Temperature Sensors

Modern power semiconductor modules incorporate a temperature-sensitive resistive element (thermistor;

NTC or PTC) soldered on the DBC substrate. Due to layout restrictions (e.g. voltage isolation), the
temperature (Tr) of this sensor does not represent the actual IGBT or diode junction temperature.

Figure 1: Location of temperature sensors in baseplate-less MiniSKiiP (L) and baseplate-d

SEMiX3p (R)

For SEMIKRON product, the temperature of the sensor may be specified as approximating an existing
reference point (e.g. Tc or Ts). This is specified in the Technical Explanations for the product. For example:
 SEMiX3s: Sensor is on a separate DBC substrate  Tr is close to Ts
 SKiiP4: Sensor is on the same copper trace as IGBT and diode  Ts < Tr < Tj

However, this simplification should be used with caution as Tr can vary depending on a number of conditions.
In some cases, the sensor temperature may be lower than the heatsink temperature beneath the hottest
chip.

The temperature sensor can be treated as a node within a simplified Foster thermal network. If only
overtemperature protection for slowly changing loads is required and the sensor temperature is equal to
heatsink temperature, then the datasheet values for Rth(j-s) can be used with safety margin to estimate
junction temperature. For more accurate results under dynamic conditions, a dedicated Zth(j-r) must be
determined.

3.1 Why are no Rth(j-r)/Zth(j-r) values specified in the module data sheet?
The sensor temperature and resulting thermal impedance from “junction to sensor” depend on many
conditions that are outside the design of the module itself. These conditions influence both lateral
temperature spreading and vertical heat conductivity in the system and can be divided roughly into two parts.

3.1.1 Influence of heatsink

The mechanical design of the heatsink influences the distribution of heat beneath the module, due in part to:
 Cooling medium (air or liquid)
 Material and root thickness of the heatsink
 Conductivity and thickness of thermal paste layer
 Module and sensor position on heatsink (distance to edges and flow direction of cooling medium, see
Figure 2)
 Distance to other modules on the same heatsink

© by SEMIKRON / 2020-01-29 / Application Note Page 3/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 2: Effect of module mounting position on sensor temperature

Tj=142°C Tr=92°C

Tr=102°C Tj=142°C

Airflow Airflow

Therefore, power semiconductor modules are specified in their data sheet with Rth(j-s) or Rth(j-c) and not Rth(j-
r), except for modules that are qualified and delivered together with a heatsink (SKiiP).

3.1.2 Influence of operating point

The electrical operating point of the system determines the magnitude and distribution of losses between
switches within the module.

Figure 3 shows two simulations of a 50A/1200V Converter-Inverter-Brake (CIB) module on an air cooled
heatsink (Ta = 25°C, Rth(s-a) ≈ 0.13K/W). The left side shows the condition of “inverter” operation (22kW,
cos() = 0.85) with high load on the IGBT and the DC-link fed by the input rectifier. The right side shows the
same module during “brake” operation (22kW, cos() = -0.85) with the braking energy dissipated by the
brake chopper. The temperature sensor is at the lower left corner.

© by SEMIKRON / 2020-01-29 / Application Note Page 4/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 3: Top view of a MiniSKiiP3 CIB module on air-cooled heatsink under two operating
conditions: inverting with cos() = 0.85 (L) and braking with cos() = -0.85 (R)

Input Rectifier

Brake
3 Chopper
6
2
5 Inverter
1 4

T-Sensor FWD
IGBT

Airflow Airflow

Table 2 indicates that there is a noticeable variation in thermal resistance between two operating conditions.

Table 2: Variation in Rth(j-r) between two operating conditions

Inverting Operation Braking Operation

(cos() = 0.85) (cos() = -0.85)

Calculated Rth(j-r): hottest IGBT to sensor 0.81K/W (IGBT5) 0.73K/W (Brake IGBT)

Calculated Rth(j-r): hottest inverse diode to sensor 2.25K/W (diode5) 1.18K/W (diode2)

Calculated Rth(j-r): hottest rectifier diode to sensor 0.96K/W (diode9) N/A (No losses)

Furthermore, Rth(j-r) can vary considerably from Rth(j-s), indicating that the “Tr ≈ Ts” assumption mentioned
earlier has limitations in certain modules. This is shown in Table 3.

© by SEMIKRON / 2020-01-29 / Application Note Page 5/22

PROMGT.1023/ Rev.7/ Template Application Note
 Table 3: Effect of “Ts = Tr” assumption on calculated temperatures

Inverting Operation
(cos() = 0.85)
IGBT: 60W
Inverse diode: 12W
Rectifier diode: 20W

Rth(j-r): hottest IGBT to sensor 0.81K/W (IGBT5)

ΔTj-r_IGBT 60W * 0.81K/W = 48.6K

Datasheet Rth(j-s)_IGBT 0.71K/W

Calculated ΔTj-r (assumption: Tr ≈ Ts) 60W * 0.71K/W = 42.6K (6K too low)

Calculated Rth(j-r): hottest inverse diode to sensor 2.25K/W (diode5)

ΔTj-r_inverse_diode 12W * 2.25K/W = 27K

Datasheet Rth(j-s)_diode 0.95K/W

Calculated ΔTj-r (assumption: Tr ≈ Ts) 12W * 0.95K/W = 11.4K (15.6K too low)

Calculated Rth(j-r): hottest rectifier diode to sensor 0.96K/W (diode9)

ΔTj-r_rectifier 20W * 0.96K/W = 19.2K

Datasheet Rth(j-s)_rectifier 0.9K/W

Calculated ΔTj-r (assumption: Tr ≈ Ts) 20W * 0.9K/W = 18K (1.2K too low)

4. Determining Rth(j-r)/Zth(j-r)

4.1 Measurement methods

For the reasons listed above, the thermal resistance/impedance from junction to sink must be determined
for each application. Three experimental methods can be used to determine the thermal resistance between
each individual IGBT or diode switch and the thermal sensor. These results should then be compared with a
computer-based finite element model to verify the tests and allow for quicker derivation of thermal resistance
under other operating conditions.

In each test, a current is fed through a single switch and the temperatures of the sensor and a target switch
are measured (Figure 4).

For Rth(j-r), the current is a fixed direct current and the temperature of the sensor measured once stabilized.
For Zth(j-r), the current is a step function and the temperature is sensor measured continuously. The current
and the voltage across the switch are used to calculate the losses. In practice a step down function is used
(turn-off power dissipation) and the measured temperatures are inverted later as this is the only way to
reach steady state for temperature dependent losses.

© by SEMIKRON / 2020-01-29 / Application Note Page 6/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 4: Application of current and resulting temperatures

Transient Steady-state
(to calculate Zth) (to calculate Rth)

Phase U TOP IGBT

160

140
Iload
120

100

T [°C]
80

Switch Iload
Vdevice 60
(BOT IGBT) (Step)
40
+ Tj [°C]
15V -
20 Tr [°C]
dT(j-r) [K]
0
Pdevice = Iload · Vdevice
0,001 0,01 0,1 1 t [s] 10 100 1000

4.1.1 Thermocouple (Rth)

A specially prepared module is obtained from the manufacturer where thermocouples have been glued to the
surface of the chips with thermally conductive epoxy (Figure 5). A regulated direct current (at low voltage)
is put through the IGBT or diode and the temperature of the thermocouple and temperature sensor are
measured to calculate the temperature difference.

Tj_device_thermocouple ‒ Tr
Rth(j ‒ r) = Pdevice (1)

Due to the slow time response of thermocouples, only determination of static thermal resistance Rth(j-r) is
possible. Furthermore, the thermocouples themselves can introduce 5…15°C of error as the metal wire acts
as a heatsink on the top surface of the chip. Caution must be taken to isolate the thermocouples where they
connect with measurement equipment.

© by SEMIKRON / 2020-01-29 / Application Note Page 7/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 5: SEMiX3p module with thermocouples added to the IGBT chips

Tr = internal NTC sensor

Tj_IGBT4 Tj_IGBT3 Tj_IGBT2 Tj_IGBT1

4.1.2 Infrared camera (Rth)

A specially prepared module without silicone soft mould is used. The housing cover is removed and the
interior of the module coated with a matte paint with uniform emissivity to prevent reflections. A regulated
direct current (at low voltage) is put through the IGBT or diode and the temperature reported by the camera
and the temperature sensor are measured to calculate the temperature difference (Figure 6).

Tj_device_IR_camera ‒ Tr
Rth(j ‒ r) = Pdevice (2)

If properly calibrated a high resolution infrared camera gives accurate temperatures although at a slow
enough refresh rate that only determination of static thermal resistance Rth(j-r) is possible. Furthermore,
modules with internal busbars system cannot be easily measured as the view of the chips is blocked.

© by SEMIKRON / 2020-01-29 / Application Note Page 8/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 6: SEMiX3p module under infrared imaging

4.1.3 Vce method (Rth or Zth)

Semiconductors have a linear relationship between junction temperature, Tj, and forward voltage drop, Vce/Vf,
at low currents. Using uniform heating in a lab environment, a calibration curve for a particular semiconductor
type can be derived (Figure 7). The module is then placed in a fixture where a high current pulse is put
through the semiconductor to generate losses and is followed quickly by a low current to measure the forward
voltage drop (and hence the junction temperature.)

Figure 7: Test configuration and example calibration curve

500
450
400
350
V CE0 [mV]

300
250
200
150
100
50
0
20 40 T60 80 100 120
j [°C]

This method provides highly accurate results and can be used to determine the transient thermal impedance
(Zth) of the junction-to-sensor interface. However, it usually requires purpose-built test equipment.

4.1.4 Finite Element Analysis (Rth or Zth)

Finite Element Analysis (FEA) is achieved by modelling the module-heatsink system in software. In order to
construct this model, two items are required from the module manufacturer:
 The X-Y locations of the chips within the module (“die map,” “chip layout”).
 The thickness, density, thermal conductivity, and thermal capacity of the layers making up the
module in the Z-axis (“material stack-up”).

© by SEMIKRON / 2020-01-29 / Application Note Page 9/22

PROMGT.1023/ Rev.7/ Template Application Note
Once the model is constructed, losses can be applied to each switch and the junction, heatsink, and sensor
temperatures determined. This method should be done in conjunction with one of the other test methods to
verify the accuracy of the model (and vice versa).

5. Simplified Method for Periodic Functions (Quasi-Steady State Conditions)

An integral solution calculates average losses of the devices over a period (e.g. one fundamental cycle for a
PWM inverter). The device losses are the sum of conduction losses, Pcond, and switching losses, Psw. The
sampling rate is low, for example, in the range of 10/fout up to 1s. Therefore, a static thermal resistance,
Rth(j-r), is used. The losses are temperature-dependent which means an iterative calculation with Tj as an
additional input is required. If the losses do not change too much from time step to time step then Tj from
the previous time step can be used.

5.1 Required circuit parameters (inverter example)

Irms - Fundamental output current, RMS
M - Modulation depth
cos(φ) - Power factor
Vcc - DC link voltage
fsw - Switching (carrier) frequency
fout - Fundamental output frequency

5.2 Loss calculation

Cycle average losses for an IGBT in a three-phase PWM inverter:

Pcond_IGBT = ( 1
2π
+
M ∙ cos(φ)
8 ) ∙ (V CE0_25°C + TCVce ∙
(Tj - 25°C)) ∙ Ipk + (8 +
1 M ∙ cos(φ)
3π ) ∙ (r
CE_25°C + TCrce ∙
(Tj - 25°C)) ∙
2
Ipk (3)

Ipk Ki Vcc Kv
Psw_IGBT = fsw ∙ Eon + off ∙ 2π ∙
1
() ( )
Iref
∙ Vref (
∙ 1 + TCsw ∙ (Tj - Tj
ref
)) ∙ γ(Ki) (4)

Cycle average losses for a diode in a three-phase PWM inverter:

Pcond_D = ( 1
2π
‒
M ∙ cos(φ)
8 ) ∙ (V F0_25°C + TCVf ∙
(Tj - 25°C)) ∙ Ipk + (8 ‒
1 M ∙ cos(φ)
3π ) ∙ (r
f_25°C + TCrf ∙
(Tj - 25°C)) ∙ Ipk2
(5)

Ipk Ki Vcc Kv
Psw_D = fsw ∙ Err ∙ 2π ∙
1
() ( )
Iref
∙ Vref (
∙ 1 + TCsw(Tj - Tj
ref
)) ∙ γ(Ki) (6)

Ki - Exponent of current dependency (IGBT ≈ 1, FWD ≈ 0.6)

Kv - Exponent of voltage dependency (IGBT ≈ 1.35, FWD ≈ 0.6)
TCsw - Temperature coefficient (IGBT ≈ 0.003, FWD ≈ 0.006)
π+φ Ki
γ(Ki) - Integral ∫φ sin (α - φ)dα (IGBT: γ(1) = 2, FWD: γ(0.6) = 2.3)

Where the following are given in the module datasheet:

VCE0_25°C
rCE_25°C
Eon+off (measured at Iref, Vref, Tjref)
TCVce (calculated as a linear relationship between VCE0(low temp) and VCE0(high temp))
TCrce (calculated as a linear relationship between rCE(low temp) and rCE(high temp))

VF0_25°C
RF_25°C

© by SEMIKRON / 2020-01-29 / Application Note Page 10/22

PROMGT.1023/ Rev.7/ Template Application Note
Err (measured at Iref, Vref, Tjref)
TCVf (calculated as a linear relationship between VF0(low temp) and VF0(high temp))
TCrf (calculated as a linear relationship between rf(low temp) and rf(high temp))

See also [2].

5.3 Junction temperature calculation

This “simplified” method relies on a single static thermal resistance measurement for each type of switch and
does not consider the coupling effects between switches. Therefore:
1. The tested switch of each type (IGBT or diode) should be closest (best coupled) to the sensor
2. This approach is only valid for “balanced” losses (e.g. 3 phase inverter where all six IGBTs (or all six
diodes) are presumed to have similar losses)
a. This approach is likely not suitable for stall conditions (e.g. where the half-bridge is operating
as a buck converter and the lower IGBT is not close to the temperature sensor)

The standard Foster thermal network is used with thermal resistances referenced to the measured sensor
temperature, Tr, at a given time step (Figure 8).

Figure 8: Thermal network referenced to temperature sensor

PIGBT
Tj_IGBT Pdiode Tj_IGBT(avg) = Rth(j - r)_IGBT ∙ PIGBT + Tr (7)

Rth(j-r)_IGBT Tj_diode

Rth(j-r)_diode

Tj_diode(avg) = Rth(j - r)_diode ∙ Pdiode + Tr (8)

Tr
+
-

Junction temperature calculation during a single time step (sample) is an iterative process because many
chip parameters are temperature dependent.

© by SEMIKRON / 2020-01-29 / Application Note Page 11/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 9: Calculation process during a single time step

Sample circuit parameters and

sensor temperature

Set initial Tj(k) = Tr

Calculate device losses

Pdevice[Tj(k)] = Pcond[Tj(k)] + Psw[Tj(k)]

Calculate junction temperatures

Tj[Tj(k)] = Pdevice[Tj(k)] • Rth(j-r) + Tr

N
Tj[Tj(k)] ≈ Tj(k)? k=k+1

Adjust Tj for fundamental

frequency  Tj(max)

Output Tj
Next time step

5.4 Correction factor for low output frequencies

The method above yields an average junction temperature and does not indicate the peak temperatures that
occur as the junction temperature oscillates at the fundamental output frequency. This is mainly of concern
at low (<10Hz) frequency operation. A simple correction factor (as shown in Figure 10) is used to adjust the
calculated average temperature. The correction factor depends on the thermal impedance of the devices in
use.

© by SEMIKRON / 2020-01-29 / Application Note Page 12/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 10: Correction factor for Tj(max) = f(fout)
Frequency correction factor SKAI-HV
3
IGBT
2,5 Diode

2
Fcorr

1,5

0,5

0
0 20 40 60 80 100
fout [Hz]

Therefore, the maximum junction temperature during a fundamental cycle can be estimated as:

Tj_IGBT(max) = Fcorr_IGBT ∙ Rth(j - r)_IGBT ∙ PIGBT + Tr (9)

Tj_diode(max) = Fcorr_diode ∙ Rth(j - r)_diode ∙ Pdiode + Tr (10)

5.5 Example (Three-phase PWM inverter)

Device parameters from SKiiP39AC12T4V1 datasheet:
IGBT: VCE0_25°C=0.8V, rce_25°C=7mΩ, Esw=36.5mJ, TCVce = -0.0008V/K, TCrce = 2.67E-5Ω/K
Diode: VF0_25°C=1.3V, rf_25°C=5.6mΩ, Err=11.4mJ, TCVf = -0.0032V/K, TCrf = 1.76E-5Ω/K

Measured Rth(j-r) from testing:

Rth(j-r)I = 0.3 K/W
Rth(j-r)D =0.6 K/W

Initial time step: measured values during operation

Iout = 76Arms = 107.48Apk
M=1
cos(φ) = 0.85
VCC = 650V
fsw = 4kHz
fout = 20Hz
Tr = 100°C

Calculated losses
(First iteration shown)
Pcond_IGBT

= ( 1 1 ∙ 0.85
2π
+
8 ) 1 1 ∙ 0.85
∙ (0.8V ‒ 0.0008V/K ∙ (100°C ‒ 25°C)) ∙ 107.48A + +
8 3π ( )
2
∙ (0.007Ω + 0.0000267Ω/K ∙ (100°C ‒ 25°C)) ∙ 107.48A = 43.49W

1 107.48A 1 1.35
Psw_IGBT = 4000Hz ∙ 0.0365J ∙
2π
∙ (
150A ) ( )∙
650V
600V
∙ (1 + 0.003 ∙ (100°C ‒ 150°C)) ∙ 2 = 31.53W

© by SEMIKRON / 2020-01-29 / Application Note Page 13/22

PROMGT.1023/ Rev.7/ Template Application Note
Pcond_D

= ( 1
2π
‒
1 ∙ 0.85
8 ) ∙ (1.3V ‒ 0.0032V/K ∙ (100°C ‒ 25°C)) ∙ 107.48A + ( 1 1 ∙ 0.85
8
‒
3π )
2
∙ (0.0056Ω + 0.0000176Ω/K ∙ (100°C ‒ 25°C)) ∙ 107.48A = 8.81W

1 107.48A 0.6 0.6

Psw_D = 4000Hz ∙ 0.0114J ∙
2π
∙
150A ( ) ( ) ∙
650V
600V
∙ (1 + 0.006 ∙ (100°C ‒ 150°C)) ∙ 2.3 = 10.04W

Tj(avg)_IGBT = (43.49W + 31.53W) ∙ 0.3K/W + 100°C = 122.5°C

Tj(avg)_D = (8.81W + 10.04W) ∙ 0.6K/W + 100°C = 111.3°C

Table 4: Example showing stabilization of temperature after two iterations

K Pcond_IGBT Psw_IGBT Pcond_D Psw_D Tj(avg)_IGBT Tj(avg)_D

0 - - - - 100°C 100°C

1 43.49W 31.53W 8.81W 10.04W 123°C 111°C

2 44.47W 34.04W 8.68W 11.01W 124°C 112°C

3 44.51W 34.16W 8.68W 11.05W 124°C 112°C

4 44.52W 34.16W 8.68W 11.06W 124°C 112°C

Tj(max)_IGBT = 1.65 ∙ (44.52W + 34.16W) ∙ 0.3K/W + 100°C = 139°C

Tj(max)_D = 1.3 ∙ (8.68W + 11.06W) ∙ 0.6K/W + 100°C = 115°C

6. Thermal Coupling

In operation, the heat generated by adjacent switches affects the temperature of the examined switch, and
hence the effective thermal impedance between junction and temperature sensor (Figure 11). These thermal
impedances between chips must be quantified to use the “Complex Method” of junction temperature
calculation described below.

© by SEMIKRON / 2020-01-29 / Application Note Page 14/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 11: FEA model of half-bridge module illustrating thermal coupling between chips and
temperature sensor

TOP BOT

I I

D D

I I

I I
Temp.
sensor
D D

I I

The relationship between any one switch and the temperature sensor is thus defined by the effect the losses
in the other switches have on the switch in question (Figure 12). The switch for which you are trying to
determine the ultimate junction temperature in the application is referred to as the “Self” switch. Note that
in this document the definition of “switch” pertains to a single electrical element (e.g. IGBT or diode) as
opposed to other documents that refer to a single switch as being composed of an IGBT and diode.

© by SEMIKRON / 2020-01-29 / Application Note Page 15/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 12: Definition of static coupling for one switch in a hypothetical half-bridge

Tj_Switch1

PSwitch1
Switch1 Switch3 Rth(j-r)_Self
(TOP IGBT)
(TOP IGBT) (TOP diode)

PSwitch2
Rth(j-r)_Switch2
(BOT IGBT)
Switch2 Switch4
(BOT IGBT) (BOT diode)
ΔTj-r

PSwitch3
Rth(j-r)_Switch3
(TOP diode)

PSwitch4
Rth(j-r)_Switch4
(BOT diode)

Tr
+
-

Tj_Switch1 = Tr + PSwitch1 ∙ Rth(j ‒ r)_Self + PSwitch2 ∙ Rth(j ‒ r)_Switch2 + PSwitch3 ∙ Rth(j ‒ r)_Switch3 + PSwitch4 ∙
Rth(j ‒ r)_Switch4 (11)

6.1.1 Determination of Rth/Zth matrix

In a lab setting, losses must be applied to each switch individually and the junction temperature measured
using one of the methods in section 4.1. The following example uses the half-bridge circuit from Figure 12.

A. Apply target losses to Switch1 (Self) only. Measure Tj_Switch1 and Tr.
Calculate:

Tj_Switch1_A ‒ Tr_A
Rth(j ‒ r)_Self = PSwitch1 (12)

B. Apply target losses to Switch2 only. Measure Tj_Switch1 and Tr.

Calculate:

Tj_Switch1_B ‒ Tr_B
Rth(j ‒ r)_Switch2 = PSwitch2 (13)

C. Apply target losses to Switch3 only. Measure Tj_Switch1 and Tr.

Calculate:

© by SEMIKRON / 2020-01-29 / Application Note Page 16/22

PROMGT.1023/ Rev.7/ Template Application Note
 Tj_Switch1_C ‒ Tr_C
Rth(j ‒ r)_Switch3 = PSwitch3 (14)

D. Apply target losses to Switch4 only. Measure Tj_Switch1 and Tr.

Calculate:

Tj_Switch1_D ‒ Tr_D
Rth(j ‒ r)_Switch4 = PSwitch4 (15)

Repeat steps A through D for the remaining three switches. The resulting values can be placed in a matrix
as shown in Table 5.

Table 5: Rth matrix for hypothetical half-bridge

Apply losses at:

TOP IGBT BOT IGBT TOP Diode BOT Diode
(Switch1) (Switch2) (Switch3) (Switch4)
Measure Tj at:

TOP IGBT Rth(j-r)_Switch1,1 Rth(j-r)_Switch1,2 Rth(j-r)_Switch1,3 Rth(j-r)_Switch1,4

(Switch1) (Self)

BOT IGBT Rth(j-r)_Switch2,1 Rth(j-r)_Switch2,2 Rth(j-r)_Switch2,3 Rth(j-r)_Switch2,4

(Switch2) (Self)

TOP Diode Rth(j-r)_Switch3,1 Rth(j-r)_Switch3,2 Rth(j-r)_Switch3,3 Rth(j-r)_Switch3,4

(Switch3) (Self)

BOT Diode Rth(j-r)_Switch4,1 Rth(j-r)_Switch4,2 Rth(j-r)_Switch4,3 Rth(j-r)_Switch4,4

(Switch4) (Self)

In the case of transient thermal impedance, the term Rth(j-r)_Switch#,c is replaced by the Foster model elements,
Zth(j-r)_Switch#,c. It should be noted that it is often possible to simplify the matrix if, for example, no thermal
coupling between switches is present (zero entries) or if the step response can be modeled by one Rth/Tau
element only.

7. Complex Method, Step-by-Step (Short, High Overload and Stall Torque Conditions)

During system operation, the losses for any switch are calculated instantaneously using measured values.
The sampling rate is high: for example 1/fsw or some multiple thereof. If fsw is high compared to fout and the
current does not change much during several periods of fsw then it is possible to combine several periods into
one calculation step to lower the computing effort.

Prior to system implementation a Zth matrix must be created as described previously. During calculation, this
may be simplified to an Rth matrix if the sampling rate > 0.5s.

7.1 Required circuit parameters (inverter example)

i(t) - actual value output current
v(t) - actual value of output voltage (line-to-neutral)
M - modulation depth to calculate the actual duty cycle
Vcc - DC link voltage
fsw - switching (carrier) frequency

7.2 Loss calculation

Loss calculations are based on a step-down DC/DC (“buck”) converter with instantaneous values. For variable
definition, see 5.2.

v(t)
DCIGBT = 0.5 + Vcc (16)

DCdiode = 1 ‒ DCIGBT (17)

© by SEMIKRON / 2020-01-29 / Application Note Page 17/22

PROMGT.1023/ Rev.7/ Template Application Note
 [
Pcond_IGBT = DCIGBT ∙ i(t) ∙ VCE0 ( 25°C
) 2
+ TCVce ∙ (Tj ‒ 25°C) + i(t) ∙ (rCE_25°C + TCrce ∙ (Tj - 25°C)) ] (18)

Vcc Kv
Psw_IGBT = fsw ∙ Eon + off ∙ ( ) ( ) (1 + TC
|i(t)| Ki
Iref
∙ Vref
∙ sw ∙ (Tj - Tj )) ref
(19)

[
Pcond_diode = DCdiode ∙ i(t) ∙ (VF0_25°C + TCVf ∙ (Tj - 25°C)) + i(t) ∙ (rf_25°C + TCrf ∙ (Tj - 25°C))
2
] (20)

Vcc Kv
Psw_D = fsw ∙ Err ∙ ( ) ∙ ( ) ∙ (1 + TC
|i(t)| Ki
Iref Vref sw (Tj - Tj ))
ref
(21)

7.3 Junction temperature calculation

Temperature for any one of the N switches in a module can be calculated at a moment, tm+1, as follows:
‒ ∆t ‒ ∆t

[ )] (22)
m m

( ) ( ) N n
Tj_Switch# tm + 1 = Tr tm + ∑c = 1∑i = 1 ∆Tj_Switch#,c,i(tm) ∙ e
τ
Switch#,c,i
(
+ Rth_Switch#,c,i ∙ PSwitch#,c(tm + 1) ∙ 1 ‒ e
τ
Switch#,c,i

Where:
Switch#: switch under investigation (also row index)
c: column index of switch under investigation
N: total number of switches/rows/columns
i: index of Foster element
n: total number of Foster elements for switch under investigation

For a fixed tm, e-X and (1-e-X) become a set of constants that can be included in the Zth matrix.

© by SEMIKRON / 2020-01-29 / Application Note Page 18/22

PROMGT.1023/ Rev.7/ Template Application Note
 Figure 13: Calculation process per time step

Time step: tm

Sample circuit parameters and sensor temperature

Calculate losses for all switches

Set:
Tj(tm) = Tr(tm) for all switches

Time step: tm+1 = tm + 1/fsample

Sample circuit parameters and sensor temperature

Calculate losses for all switches

Set Switch# = c = i = 1

N
i = n? i i + 1

Y
Set i = 1

N
c = N? c c + 1

Set c = 1

N
Switch# = N? Switch# Switch# + 1

Y Next time step

7.4 Example
In the following example, the temperature in the TOP switch of a half-bridge module is calculated using the
losses (Table 7) for a theoretical system operating over 1s. Constant losses and a constant temperature

© by SEMIKRON / 2020-01-29 / Application Note Page 19/22

PROMGT.1023/ Rev.7/ Template Application Note
sensor are used for simplicity but the approach is valid for varying values as well. Temperatures at
subsequent time steps can be derived using the results from the preceding time step.

Table 6: Example Zth matrix for SEMiX603GB12E4p on watercooler

c IGBT IGBT Diode Diode
TOP BOT TOP BOT
Switch#
i Rth(j-r) Tau i Rth(j-r) Tau i Rth(j-r) Tau i Rth(j-r) Tau
1 0.0054 0.0028 1 0.0063 3.7000 1 0.0248 1.2 1 0.0087 4.7
IGBT
2 0.0086 0.025 2 0 1 2 0.0024 3 2 0 1
TOP 3 0.0190 0.1 3 0 1 3 0 1 3 0 1
4 0.0224 0.5 4 0 1 4 0 1 4 0 1
IGBT
Zth(j-r)_IGBT_BOT:IGBT_TOP Zth(j-r)_IGBT_BOT:Self Zth(j-r)_IGBT_BOT:Diode_TOP Zth(j-r)_IGBT_BOT:Diode_BOT
BOT
Diode
Zth(j-r)_Diode_TOP:IGBT_TOP Zth(j-r)_Diode_TOP:IGBT_BOT Zth(j-r)_Diode_TOP:Self Zth(j-r)_Diode_TOP:Diode_BOT
TOP
Diode
Zth(j-r)_Diode_BOT:IGBT_TOP Zth(j-r)_Diode_BOT:IGBT_BOT Zth(j-r)_Diode_BOT:Diode_TOP Zth(j-r)_Diode_BOT-Self
BOT

Table 7: Example run-time parameters used for calculating junction temperature

Time step 0.0s 1.0s
PIGBT_TOP 300W 300W
PIGBT_BOT 300W 300W
PDiode_TOP 100W 100W
PDiode_BOT 100W 100W
Tsensor 80°C 80°C
Tj_IGBT_TOP 80°C Tj_IGBT_TOP(1s)

Tj_IGBT_TOP(1s)
‒ 1s ‒ 1s

= 80°C + 0 ∙ e
[ 0.0028s
+ 0.0054K/W ∙ 300W ∙ 1 ‒ e ( 0.0028s
)]
‒ 1s ‒ 1s ‒ 1s ‒ 1s

[
+ 0∙e
0.025s
+ 0.0086K/W ∙ 300W ∙ 1 ‒ e ( 0.025s
)] + [ 0 ∙ e 0.1s
+ 0.019K/W ∙ 300W ∙ 1 ‒ e ( 0.1s
)]
‒ 1s ‒ 1s ‒ 1s ‒ 1s

[
+ 0∙e
0.5s
+ 0.0224K/W ∙ 300W ∙ 1 ‒ e ( 0.5s
)] + [ 0 ∙ e 3.7s
+ 0.0063K/W ∙ 300W ∙ 1 ‒ e ( 3.7s
)]
‒ 1s ‒ 1s ‒ 1s ‒ 1s

[
+ 0∙e
1.2s
+ 0.0248K/W ∙ 100W ∙ 1 ‒ e ( 1.2s
)] + [ 0 ∙ e 3s
+ 0.0024K/W ∙ 100W ∙ 1 ‒ e ( 3s
)]
‒ 1s ‒ 1s

[
+ 0∙e
4.7s
+ 0.0087K/W ∙ 100W ∙ 1 ‒ e ( 4.7s
)] = 97.8°C
In the example, the TOP IGBT has increased in temperature by 17.8°C after 1s of operation. Of this
temperature rise, 15.7°C was due to self-heating of the switch (red terms) and 2.08°C is contributed by the
remaining three switches (blue, green, violet terms). In this case, the terms are positive but it could be that
terms are negative if the losses in another switch reduce the temperature difference between the sensor and
the investigated switch.

8. Summary

Using the integrated temperature sensor to calculate Tj is possible but the accuracy depends on the level of
characterization that the designer is able to undertake during the design process. The most basic protection

© by SEMIKRON / 2020-01-29 / Application Note Page 20/22

PROMGT.1023/ Rev.7/ Template Application Note
can be achieved by taking a high safety margin and initiating an overtemperature fault when the sensor
reaches a fixed temperature.

A more advanced “simplified” approach involves measuring a thermal impedance Rth(j-r) and assuming even
loss distribution amongst the switches with average losses for periodic functions. This method requires little
computing power and can provide effective temperature protection for converters with well-characterized
operation and slow-moving transient overloads.

For protection against fast overloads and special conditions such a “0Hz” inverter operation, a detailed
thermal model defining transient thermal impedances between the chips and temperature sensor is required.
With careful measurements, individual models can be created for each switch that defines a transient thermal
impedance matrix for the entire module. Coupled with strong processing power, this matrix yields a large
amount of runtime temperature data that can be used for dynamic protection.

In all cases, it is important to understand that temperature measurement method is only valid for a particular
converter design.

Figure 1: Location of temperature sensors in baseplate-less MiniSKiiP (L) and baseplate-d SEMiX3p (R) .....3
Figure 2: Effect of module mounting position on sensor temperature .....................................................4
Figure 3: Top view of a MiniSKiiP3 CIB module on air-cooled heatsink under two operating conditions:
inverting with cos() = 0.85 (L) and braking with cos() = -0.85 (R) .....................................................5
Figure 4: Application of current and resulting temperatures..................................................................7
Figure 5: SEMiX3p module with thermocouples added to the IGBT chips ................................................8
Figure 6: SEMiX3p module under infrared imaging ..............................................................................9
Figure 7: Test configuration and example calibration curve ..................................................................9
Figure 8: Thermal network referenced to temperature sensor .............................................................11
Figure 9: Calculation process during a single time step .....................................................................12
Figure 10: Correction factor for Tj(max) = f(fout) ..................................................................................13
Figure 11: FEA model of half-bridge module illustrating thermal coupling between chips and temperature
sensor.........................................................................................................................................15
Figure 12: Definition of static coupling for one switch in a hypothetical half-bridge ................................16
Figure 13: Calculation process per time step ....................................................................................19

Table 1: Comparison of the two considered methods for junction temperature prediction .........................2
Table 2: Variation in Rth(j-r) between two operating conditions ...............................................................5
Table 3: Effect of “Ts = Tr” assumption on calculated temperatures .......................................................6
Table 4: Example showing stabilization of temperature after two iterations...........................................14
Table 5: Rth matrix for hypothetical half-bridge ................................................................................17
Table 6: Example Zth matrix for SEMiX603GB12E4p on watercooler ....................................................20
Table 7: Example run-time parameters used for calculating junction temperature..................................20

References
[1] www.SEMIKRON.com
[2] A. Wintrich, U. Nicolai, W. Tursky, T. Reimann, “Application Manual Power Semiconductors”, 2nd
edition, ISLE Verlag 2015, ISBN 978-3-938843-83-3

IMPORTANT INFORMATION AND WARNINGS

The information in this document may not be considered as guarantee or assurance of product characteristics
("Beschaffenheitsgarantie"). This document describes only the usual characteristics of products to be expected in typical
applications, which may still vary depending on the specific application. Therefore, products must be tested for the
respective application in advance. Application adjustments may be necessary. The user of SEMIKRON products is
responsible for the safety of their applications embedding SEMIKRON products and must take adequate safety measures
to prevent the applications from causing a physical injury, fire or other problem if any of SEMIKRON products become
faulty. The user is responsible to make sure that the application design is compliant with all applicable laws, regulations,
norms and standards. Except as otherwise explicitly approved by SEMIKRON in a written document signed by authorized
representatives of SEMIKRON, SEMIKRON products may not be used in any applications where a failure of the product or
any consequences of the use thereof can reasonably be expected to result in personal injury. No representation or warranty
is given and no liability is assumed with respect to the accuracy, completeness and/or use of any information herein,
including without limitation, warranties of non-infringement of intellectual property rights of any third party. SEMIKRON
does not assume any liability arising out of the applications or use of any product; neither does it convey any license under
its patent rights, copyrights, trade secrets or other intellectual property rights, nor the rights of others. SEMIKRON makes

© by SEMIKRON / 2020-01-29 / Application Note Page 21/22

PROMGT.1023/ Rev.7/ Template Application Note
no representation or warranty of non-infringement or alleged non-infringement of intellectual property rights of any third
party which may arise from applications. This document supersedes and replaces all information previously supplied and
may be superseded by updates. SEMIKRON reserves the right to make changes.

SEMIKRON INTERNATIONAL GmbH

Sigmundstrasse 200, 90431 Nuremberg, Germany
Tel: +49 911 6559 6663, Fax: +49 911 6559 262
[email protected], www.semikron.com

© by SEMIKRON / 2020-01-29 / Application Note Page 22/22

PROMGT.1023/ Rev.7/ Template Application Note