SEMIKRON Apon With Central and Individual Driver Board EN 2017-01-27 Rev-00
SEMIKRON Apon With Central and Individual Driver Board EN 2017-01-27 Rev-00
SEMIKRON Apon With Central and Individual Driver Board EN 2017-01-27 Rev-00
1. General
Parallel circuits are always necessary when the performance criteria of a single component are insufficient.
This starts at the microscopic chip level with several 100,000 individual IGBT cells, then further in the
module by the parallel connection of chips and continues at the circuit level by parallel connection of
modules and entire inverter units. [2]
This application note gives an overview of the causes that can be attributed to an asymmetrical current
distribution. It also serves as an aid to the effective parallel connection of IGBT power modules. The focus
is on the influence of the driver concept used (individual or central driver) as well as the impedances
contained in a system.
For further information, please refer to the SEMIKRON® "Application Manual Power Semiconductors" [2].
2. Definition of Terms
The investigations regarding the current distribution were carried out on a phase module, consisting of four
SEMIX®603GB12E4p modules connected in parallel. Figure 1 shows the individual driver concept on the
left and the concept with a central driver on the right-hand side.
Figure 1: Phase module with individual drivers (left) and central driver (right)
The individual driver concept is based on the SKYPER12 press-fit driver especially developed for the
SEMIX® press-fit module. Each of the four drivers has its own primary and secondary side with its
associated output stage. Every one of the output stages controls one of the four parallel
SEMIX®603GB12E4p modules. The four individual drivers are connected on the primary side via an
adapter board, which interfaces the signals from the higher-level control unit to the individual drivers.
The core of the concept with the central driver is a SKYPER® 42 LJ R, which is directly connected to the
higher-level control unit. An adapter board serves as an interface to the four SEMIX® modules. On this
board are located the gate and emitter resistors as well as the gate protection circuits. Each module switch
has its own gate circuit, which is controlled by the central secondary stage of the SKYPER® 42 LJ R.
Figure 2: Modules with individual drivers (left) and central driver (right)
The coloured markings of the module positions in Figure 2 serve as an orientation aid. They correspond to
the colours of the traces in the diagrams of this application note.
The influence of the various factors on the current distribution was determined using two different
measuring methods. On the one hand, the double pulse method, which is well suited for characterising the
switching behaviour of semiconductors. On the other hand, the inverter operation, which maps the load of
the semiconductors or modules in close reference to the application.
𝐼𝐶 𝑉𝐺𝐸
𝑉𝐶𝐸
𝐿𝑙𝑜𝑎𝑑
𝑉𝐶𝐸
𝑉𝐺𝐸
𝑉𝐷𝐶 𝐼𝐶 𝑉𝐷𝐶
𝐿𝑙𝑜𝑎𝑑
𝑉𝐶𝐸
𝑉𝐺𝐸
𝐼𝐶
𝐼𝐴𝐶
𝐼𝐴𝐶 𝐿𝑙𝑜𝑎𝑑
𝑉𝐷𝐶
𝑡
Table 2: Test conditions for double pulse test and inverter operation
𝑉𝐺(𝑜𝑛) 15 15 [V]
𝑉𝐺(𝑜𝑓𝑓) -8 -8 [V]
The impedances contained in a system and thus the mechanical design of the system bears considerable
influence on the current distribution between parallel connected modules. The reasons for this are, on the
one hand, the differences in the impedance values of the individual current paths, which result from the
asymmetry of the design. On the other hand, the spatial position of the individual components relative to
each other and the resulting inductive couplings influence the current distribution.
By choosing a smart design, the effects of the influencing factors can be significantly reduced. The more
symmetrical the structure of the system, the more symmetrical is the current distribution.
In this chapter the influence of the system design is considered independently of the driver concepts. The
influences that the mechanical design exerts only in connection with one of the two driver concepts are
explained in chapter 7. The measuring results shown below are determined using the individual driver
setup, but are valid for both driver concepts. For the theoretical considerations, it is always assumed that
only semiconductors or modules with identical characteristics are connected in parallel.
The static and also dynamic differences are minimised for chips within a given production lot, as they have
been manufactured from similar silicon under the same manufacturing conditions. Therefore, it is wise to
select modules with similar date codes for paralleling.
900 900
VGE = 15V Tj = 25°C
[𝐴] [𝐴]
Tj = 25°C Tj = 150°C
750 Tj = 150°C 750
600 600
PTC
450 450
150 150
𝐼𝐶 𝐼𝐹
0 0
𝑉𝐶𝐸 1 2 3 4 [𝑉] 5 𝑉𝐹 1 2 3 [𝑉] 4
The influence of the parasitic inductances is described mathematically by means of the simplified equivalent
circuit diagram shown in Figure 6. It consists of two parallel current paths and a current source from which
the current 𝐼𝑙𝑜𝑎𝑑(𝑡) flows. 𝐿𝜎1 and 𝐿𝜎2 represent the sum of all parasitic inductances of a current path, 𝑅1 and
𝑅2 represent the sum of all ohmic resistances.
I ∆𝐼
𝐼𝑙𝑜𝑎𝑑(𝑡) 𝐼𝑙𝑜𝑎𝑑(𝑡) 2
∆𝐼 𝐼1(𝑡)
𝐼1(𝑡) (𝑡 ≫ 𝜏)
𝐿𝜎1 𝐿𝜎2
𝐼1(𝑡) (𝑡 ≪ 𝜏)
𝐼1(𝑡)
𝑅1 𝑅2 𝐼1(𝑡) (𝑡 ≫ 𝜏)
𝐼1(𝑡) 𝐼2(𝑡) 𝐼2(𝑡) (𝑡 ≪ 𝜏)
𝑡=𝜏
𝐼𝑙𝑜𝑎𝑑(t) = 0 𝑓ü𝑟 𝑡 ≤ 0
In order to describe the current distribution with the terms listed below, three assumptions must be made.
(1) The rising current 𝑑𝐼𝑙𝑜𝑎𝑑/𝑑𝑡 is constant during the common conductance phase of the IGBTs and is
determined by the behaviour of the current source 𝑉1.
(2) The influence of the ohmic resistances on the current distribution is negligible 𝑅1 = 𝑅2 = 𝑅.
(3) The load current 𝐼𝑙𝑜𝑎𝑑(t) is zero for the period 𝑡 ≤ 0 [3].
𝐿𝜎1 + 𝐿𝜎2
𝜏 =
2𝑅
𝐿𝜎2 𝐿𝜎1
𝐼1(𝑡) ≈ ∙𝐼 (𝑡) 𝐼2(𝑡) ≈ ∙𝐼 (𝑡) 𝑓𝑜𝑟: t≪τ
𝐿𝜎1 + 𝐿𝜎2 𝑙𝑜𝑎𝑑
𝐿𝜎1 + 𝐿𝜎2 𝑙𝑜𝑎𝑑
𝐼𝐿𝑜𝑎𝑑(𝑡) 𝐼𝐿𝑜𝑎𝑑(𝑡)
𝐼1(𝑡) ≈ ‒ ∆𝐼 𝐼2(𝑡) ≈ + ∆𝐼 𝑓𝑜𝑟: t≫τ
2 2
The formulas show that the current divider of the branch inductances determines the current asymmetry
for times t ≪ τ. As the time increased t ≫ τ the two branch currents, at a distance of ∆𝐼, extend parallel to
half the load current 𝐼𝑙𝑜𝑎𝑑(𝑡) 2. The magnitude of ∆𝐼 is dependent on the difference of the branch
inductances 𝐿𝜎1 ‒ 𝐿𝜎2, the sum of the ohmic branch resistances and the current rise time 𝑑𝐼𝑙𝑜𝑎𝑑/𝑑𝑡. The
current rise time, in turn, in the real application is decisively determined by the level of the DC-bus voltage
and the magnitude of the load inductance.
IGBT turn on; AC current; 100A/div; 20µs/div IGBT turn on; AC current; 100A/div; 20µs/div
The left oscillogram in Figure 8 shows the current distribution with the double pulse test with a load
inductance reduced to half, compared to the measurement from chapter 6.1.1. (right diagram). As in the
case of the measurement from chapter 6.1.1 the load is also connected off-centre here and results in an
asymmetrical current distribution. The raise in the current through the right module increases to about
30% from approx. 20% (measurement from chapter 6.1.1) due to the lower load inductance.
Considering the results of the converter operation, however, which was carried out with the same off-
centre position of the load connection, an asymmetrical current distribution of only approximately 2% rms
relative to the nominal current 𝐼𝐴𝐶 4, occurs.
Due to the positive temperature coefficient of the IGBT collector-emitter voltage (VCE) and the higher load
inductance, the current distribution in the inverter mode is much better than in the double pulse test with a
low load inductance.
IGBT turn on; AC current; 150A/div; 20µs/div inverter mode; AC current; 150A/div; 2ms/div
𝐼𝑙𝑜𝑎𝑑(𝑡)
𝑅1 𝑅2 𝑅3 𝑅4
𝐼1 𝐿 𝐼2 𝐿 𝐼3 𝐿 𝐼4
𝜎12 𝜎23 𝜎34
Assuming that all inductances depicted in the equivalent circuit have the same value and there is no
inductive coupling between the load cable and the AC busbar, the current path with the smallest impedance
is located on the right and the current path with the largest impedance on the left side. Accordingly, a
current distribution should result corresponding to the right hand trace of Figure 7. The actual current
distribution, which results from the inductive couplings 𝑀1, 𝑀2 and 𝑀3 between the load cable and the AC
busbar, is depicted in Figure 10.
𝐼𝑙𝑜𝑎𝑑(𝑡)
𝐿𝜎1 𝐿𝜎2
𝐼12
𝑅1 𝑅2
𝑑𝑖𝐿𝑜𝑎𝑑
𝑀1 ∙ 𝐿𝜎12
𝑑𝑡
𝑑𝑖𝑙𝑜𝑎𝑑 𝑑𝑡 𝐿'𝜎12
The equivalent circuit diagram on the left-hand side in Figure 10 is used to illustrate this effect. It shows
two parallel current paths with a coupling between the parasitic inductance of the AC busbar 𝐿𝜎12 and the
parasitic inductance of the load cable 𝐿'𝜎12. The coupling between the two inductances is symbolised by a
voltage source with a terminal voltage 𝑀1 ∙ 𝑑𝑖𝐿𝑜𝑎𝑑 𝑑𝑡.
If a current changing over time flows through 𝐿'𝜎12 this induces a voltage along the AC busbar due to the
inductive coupling. Caused by this voltage, a circular current 𝐼12 flows counter clockwise through the
network consisting of 𝐿𝜎1, 𝑅1, 𝐿𝜎12, 𝑅2 and 𝐿𝜎2. The superposition of the circular current with the load
currents leads to the current distribution shown in Figure 10.
The outcome of this effect is proportional to the factor of the inductive coupling. This, in turn, is dependent
on the distance between the current-carrying conductors and their position relative to one another. If the
distance between the AC busbar and the load cable increases, the coupling factor decreases due to the
magnetic field lines becoming weaker with the distance. The effect can be completely eliminated by moving
the load cable perpendicularly away from the AC busbar, since the magnetic field lines extend parallel to
the AC busbar.
This application note is based on the case of module-internal commutation, in which the mechanical design
of the system is not the cause for the asymmetrical current distribution but affects the degree of
asymmetry.
Figure 11 on the left side shows the typical current profile for modules which are operated in parallel but
not exactly simultaneous. To highlight this effect, the BOT switch of a module (red curve) was switched on
with a delay of 100ns. The time-delayed switching leads to an asymmetrical dynamic current distribution
between the modules in the negative part of the cycle, whereas the current distribution is symmetrical in
the positive half-wave. The right side of the figure shows a turn-on sequence of the parallel BOT IGBTs
during the negative half cycle of the output current. The IGBT, which last changes from the non-conducting
to the conducting state, takes up significantly less current at the beginning of the common conduction
phase, since the IGBTs, which turn on first, take over a part of the total current.
Figure 11: Measurement of the influence of different switching time on the dynamic current
distribution
inverter mode; AC current; 200A/div; 2ms/div inverter mode; AC current; 200A/div; 200ns/div
The influence of the mechanical design on the dynamic current asymmetry is described by the equivalent
circuit diagram shown in Figure 12. This consists of two parallel connected modules with identical
characteristics. 𝐿1 and 𝐿2 symbolise the sum of the inductances, 𝑅1 and 𝑅2 the sum of the resistances which
are located in the AC branches. For the sake of simplicity, both 𝐿1 and 𝐿2, as well as 𝑅1 and 𝑅2 have
identical values, where: 𝐿1 = 𝐿2 and 𝑅1 = 𝑅2. 𝐿𝑙𝑜𝑎𝑑 represent the inductive load of the common AC output
through which the total current 𝐼𝐴𝐶(𝑡) flows.
𝑇1 𝐷 1 𝑉𝐿12(𝑡)
𝑉1(𝑡) 𝐷2 𝑇2 𝑉 (𝑡)
2
𝑉𝐿1(𝑡) 𝑉𝐿2(𝑡)
𝑅1 𝑅2
𝐿1 𝐿2
𝐾1 𝑉𝐷𝐶
𝐾2
𝐼1(𝑡 ≥ 𝑡0) 𝐼2(𝑡 ≥ 𝑡1)
𝑉3(𝑡) 𝑇3 𝐷 3 𝐿𝑙𝑜𝑎𝑑 𝐷4 𝑇4 𝑉4(𝑡)
𝐼𝐴𝐶(𝑡)
𝐼1(𝑡 < 𝑡0) 𝐼2(𝑡 < 𝑡1)
𝐷𝐶 ‒
The diagram in Figure 13 shows the calculated current profiles of 𝐼1(𝑡) or 𝐼2(𝑡) for three different values of
the inductances 𝐿1 or 𝐿2 respectively. The inductance for the calculation of 𝐼1(𝑡) or 𝐼2(𝑡) corresponds to the
' ' ~ ~
value 𝐿, for the calculation of 𝐼1(𝑡) or 𝐼2(𝑡) to the value 3.3 ∙ 𝐿 and for the calculation of 𝐼 1 (𝑡) and 𝐼 2 (𝑡) to the
value 67 ∙ 𝐿. The current profiles show a switch on process of IGBT 𝑇1 or 𝑇2 in which IGBT 𝑇1 switches on
fist. The calculation is also valid for the switch off process of IGBT 𝑇1 or 𝑇2, but with reverse signs for ∆𝐼.
𝐼1(𝑡)
'
𝐼1(𝑡)
~
𝐼 1 (𝑡)
𝐼2(𝑡)
'
𝐼2(𝑡)
~
𝐼 2 (𝑡)
𝑡0 𝑡1
𝑡
The basis for the consideration is a current 𝐼𝐴𝐶(𝑡) distributing uniformly to the diodes 𝐷3 and 𝐷4 up to the
time 𝑡0.
𝐼𝐴𝐶(𝑡)
𝐼1(𝑡) = 𝐼2(𝑡) = 𝑓𝑜𝑟 𝑡 < 𝑡0
2
At the time 𝑡0 IGBT 𝑇1 turns on, diode 𝐷3 takes up blocking voltage and raises the voltage at the node 𝐾1 to
the value 𝑉𝐷𝐶 ‒ 𝑉1(𝑡). Until turn-on of IGBT 𝑇2 at the time 𝑡1 diode 𝐷4 keeps the voltage at node 𝐾2 at 𝑉𝐷𝐶 ‒ 𝑉4
(𝑡). The voltage 𝑉𝐿12(𝑡) thus occurs across the inductances 𝐿1 and 𝐿2.
𝐼𝐴𝐶(𝑡0) 𝑈𝐿12(𝑡)
𝐼1,2(𝑡1) = ± ∙ (𝑡1 ‒ 𝑡0) 𝑓𝑜𝑟: 𝑡0 ≤ 𝑡 < 𝑡1
2 𝐿1 + 𝐿2
The term shows that by introducing additional inductance, for example by means of longer load cables at
the AC terminals of the modules, the rise time 𝑑𝑖1 𝑑𝑡 bzw. 𝑑𝑖2 𝑑𝑡 can be reduced. Thus, with an identical
delay time and an identical voltage drop across the inductances 𝐿1 and 𝐿2, a smaller difference between 𝐼1
(𝑡1) and 𝐼2(𝑡1) results.
6.3 Current sharing symmetry effect during the common conduction phase of the
semiconductors
If the current has been unevenly distributed between the parallel current paths during the commutation
phase, the currents will completely or partially recombine during the common conducting phase of the
IGBTs.
The effect of current sharing is explained below using the example of two identical modules which are
operated in parallel. Figure 14 shows the corresponding simplified equivalent circuit diagram, analogue to
the one in Figure 12.
𝐷𝐶 +
𝑉𝐿12(𝑡)
𝑇1 𝐷 1 𝐷2 𝑇2
𝑉1(𝑡) 𝑉2(𝑡)
𝑉𝐿1(𝑡) 𝑉𝐿2(𝑡)
𝑅1 𝑅2
𝐿1 𝐿2
𝐾1 𝐾2 𝑉𝐷𝐶
𝐷𝐶 ‒
The diagram in Figure 15 shows the calculated current profiles of 𝐼1(𝑡) or 𝐼2(𝑡) for three different values of
the inductances 𝐿1 or 𝐿2 respectively. The inductance for the calculation of 𝐼1(𝑡) or 𝐼2(𝑡) corresponds to the
' ' ~ ~
value 𝐿, for the calculation of 𝐼1(𝑡) or 𝐼2(𝑡) to the value 3.3 ∙ 𝐿 and for the calculation of 𝐼 1 (𝑡) and 𝐼 2 (𝑡) to the
value 67 ∙ 𝐿. The resistance values 𝑅1 bzw. 𝑅2 are identical for all three calculations.
𝐼1(𝑡)
'
𝐼1(𝑡)
~
𝐼 1 (𝑡)
𝐼2(𝑡)
'
𝐼2(𝑡)
~
𝐼 2 (𝑡)
𝑡1
𝑡
The basis for the consideration is that IGBT 𝑇1 turns on first, both IGBTs 𝑇1 and 𝑇2 are turned on at the
time 𝑡1 and the currents are divided according to the situation described in section 6.2.
𝐼𝐴𝐶(𝑡1)
𝐼1,2(𝑡1) = ± ∆𝐼(𝑡1)
2
The inductances 𝐿1 and 𝐿2 can be neglected at time 𝑡1, the point of inflection of the currents 𝐼1(𝑡) and 𝐼2(𝑡).
At the node 𝐾1 a voltage of 𝑉𝐷𝐶 ‒ 𝐼1(𝑡1) ∙ 𝑅1 will result and at the node 𝐾2 a voltage of 𝑉𝐷𝐶 ‒ 𝐼2(𝑡1) ∙ 𝑅2, relative
to the DC- potential. The resulting voltage 𝑉𝐿12(𝑡) between the nodes can be calculated as follows.
The voltage 𝑉𝐿12(𝑡) increases the driving voltage across the inductor 𝐿2 and simultaneously reduces the
driving voltage across the inductor 𝐿1. This leads to both currents converging to the value 𝐼𝐴𝐶(𝑡) 2 . The
course of the currents can be expressed by the following terms.
𝑅1 + 𝑅2
𝐼𝐴𝐶(𝑡1) 𝐼𝐴𝐶(𝑡) ‒ ∙ (𝑡 ‒ 𝑡1)
𝐼1(𝑡) =
2 [
+ 𝐼1(𝑡1) ‒
2 ] ∙𝑒
𝐿1 + 𝐿2
𝑓𝑜𝑟: 𝑡 > 𝑡1
𝑅1 + 𝑅2
𝐼𝐴𝐶(𝑡1) 𝐼𝐴𝐶(𝑡) ‒ ∙ (𝑡 ‒ 𝑡1)
𝐼2(𝑡) =
2 [
+ 𝐼2(𝑡1) ‒
2 ] ∙𝑒
𝐿1 + 𝐿2
𝑓𝑜𝑟: 𝑡 > 𝑡1
The equations show that the convergence speed of the currents 𝐼1(𝑡) and 𝐼2(𝑡) is determined by the ratio of
the sum of the resistors 𝑅1 and 𝑅2 to the sum of the inductances 𝐿1 and 𝐿2.
𝐿1 + 𝐿2
𝜏=
𝑅1 + 𝑅2
If the sum of the inductances 𝐿1 and 𝐿2 increases, then the time constant 𝜏 increases and the convergence
speed of the currents 𝐼1(𝑡) and 𝐼2(𝑡) decreases. The time available for the currents 𝐼1(𝑡) and 𝐼2(𝑡) to approach
the value 𝐼𝐴𝐶(𝑡) 2 is limited by the next switching operation of the semiconductors and thus by the clock
Figure 16: Current offset at the end of the IGBT conduction period
𝐼1(𝑡)
|∆𝐼1| + |∆𝐼2|
𝐼2(𝑡)
The effect of the current sharing is superimposed by the effects of the asymmetrical static current
distribution, described in chapter 6.1, which works against the convergence of the currents.
𝐼 𝐼 𝐼
① ② ③
𝐿1 𝐿2 𝑡 15 ∙ 𝐿1 15 ∙ 𝐿2 𝑡 7000 ∙ 𝐿1 7000 ∙ 𝐿2 𝑡
The inductances 𝐿1 and 𝐿2 limit the slew rate of the currents 𝐼𝐴𝐶1(𝑡) und 𝐼𝐴𝐶2(𝑡) during the time-shifted
switching of the modules. The smaller the sum of the inductance values is, the greater is the resulting
Figure 18: Simplified equivalent circuit diagram for two parallel modules for large AC
inductors
𝐼1(𝑡) 𝐿1 𝐿2 𝐼2(𝑡)
𝐼𝐴𝐶
The difference between the output voltages 𝑉𝑉1 and 𝑉𝑉2 is dependent on the switching time differences ∆𝑡,
the switching frequency 𝑓𝑠𝑤 and the DC-bus voltage 𝑈𝐷𝐶. The difference can be determined by the following
formula.
The occurring voltage difference 𝑉𝑉1 ‒ 𝑉𝑉2 drives a compensation current ∆𝐼 through the inductances 𝐿1 and
𝐿2. The compensating current ∆𝐼 is superimposed with output currents 𝐼1(𝑡) and 𝐼2(𝑡). The output current is
limited by the inductances 𝐿1 and 𝐿2.
𝑉𝑉1 ‒ 𝑉𝑉2
∆𝐼 =
2 ∙ 𝜋 ∙ 𝑓𝑜𝑢𝑡 ∙ (𝐿1 + 𝐿2)
The ohmic resistances of the current branches were not taken into account for these considerations.
Just like the impedances of a system, the characteristics of the drivers also influence the current
distribution. In contrast to the mechanical design, which mainly affects the static current distribution, the
driver characteristics predominantly affect the dynamic current sharing. The reasons for this are, on the
one hand, the differences in the signal propagation times and the gate-emitter voltages, as well as the
influence of the jitter. On the other hand, the common emitter path, of the central driver unit, affects the
current sharing.
With regard to the driver concepts, this means that only a small difference between the signal propagation
times is much more probable when a central driver is used than in the case of the parallel connection of
individual drivers.
The left diagram in Figure 19 shows the measurement results of IGBT current of the pulse test for the
parallel SEMIX603GB12Ep modules. Each of the modules was controlled by a dedicated SKYPER12 press-fit
driver. One of the drivers was prepared in such a way that the associated IGBT (red trace) switches off
with a delay time of approx. 100ns. Due to the longer conduction phase of the IGBT, the currents of the
parallel IGBTs do not immediately commutate to the corresponding freewheel diodes, but partly flow
through the still conducting IGBT. In this test arrangement, a current increase of approx. 48% occurs,
based on the nominal value 𝐼𝐶,𝑠𝑢𝑚 4. The distribution of the currents, during the commutation sequences of
the IGBTs, takes place according to the concept described in chapter 6.2.
The results for the inverter operation can be found in the right diagram in Figure 19, in which one of the
BOT IGBTs (red trace) is switched on approximately 100ns earlier.
The real time offset between parallel SKYPER12 press-fit drivers, which occurs due to the signal
propagation time tolerances, is significantly less than 100ns. The influence of the switching under realistic
delay times can be seen on the positive half-wave of the right diagram.
IGBT turn off; collector current; 150A/div; 200ns/div inverter mode; AC current; 200A/div; 2ms/div
7.1.2 Jitter
The influence of the jitter occurs with digital drivers connected in parallel, when they have their own
system clock. The maximum time offset due to jitter is determined by the frequency of the system clock. If
the input signal changes simultaneously at the parallel driver stages, the change in the corresponding
output signal may vary in time by the length of one system clock period.
The time offset variance due to jitter approximately shows a Gaussian normal distribution over the values
0 ≤ 𝑡𝑗𝑖𝑡𝑡𝑒𝑟 ≤ 𝑇𝑐𝑙𝑜𝑐𝑘 . It is therefore not foreseeable which driver provides a valid output signal at which time.
Looking at the effect of the jitter in isolation, it can be assumed that with an increasing number of signal
changes at the input of the parallel drivers, the summed time offset variance between the output signals
approaches zero.
𝑛 𝑛
In the real application, the time offset of the jitter is added to the time offset of the different signal
propagation times. The effect of the jitter can, however, be neglected for the central driver concept.
𝑡𝑑 = ∆𝑡𝑝 + 𝑡𝑗𝑖𝑡𝑡𝑒𝑟
Figure 20 shows the measured results of the pulse test with four parallel individual drivers. The delay time
due to the jitter is approximately 25ns, which corresponds to the reciprocal of the system frequency of the
SKYPER12 press-fit driver of 40Mhz. The measured values of both diagrams were recorded under identical
conditions and in immediate succession. The left diagram shows a nearly symmetrical current distribution.
The delay between the driver outputs is almost zero at this time. The right diagram shows the effect on the
dynamic current distribution with a time delay of approx. 25ns between the outputs of the parallel drivers.
In this measurement, a time delay of approx. 25ns results in a current increase of approx. 12% (yellow
trace), based on the nominal value 𝐼𝐶,𝑠𝑢𝑚 4. The currents distribute according to the concept described in
chapter 6.2.
IGBT turn off; collector current; 100A/div; 200ns/div IGBT turn off; collector current; 100A/div; 200ns/div
IGBT turn off; collector current; 150A/div; 200ns/div inverter mode; AC current; 200A/div; 2ms/div
In order to avoid this, a separate gate resistor is provided for each IGBT. The gate voltage on the IGBTs
can then rise independently of one another, which results in only slight switching differences.
Figure 22: Equivalent circuit diagram for two parallel IGBTs without individual gate resistors
𝐷𝐶 +
𝑉𝐺𝐸
𝑇1 𝐷1 𝑇2 𝐷2
𝑅𝐺 𝑉𝐺𝐸1 𝑉𝐺𝐸2
𝑉𝐺𝐸
𝑉𝐺𝐸(𝑝𝑙)1 𝑉𝐺𝐸(𝑝𝑙)2
𝑡
∆𝑡
𝐷𝐶 ‒
𝐷𝐶 + 𝐷𝐶 +
𝑇1 𝑇2 𝑇1 𝑇2
𝑅𝐺1 𝐷1 𝑅𝐺2 𝐷2 𝑅𝐺1 𝐷1 𝑅𝐺2 𝐷2
𝑅𝐺 𝑅𝐺
𝑅𝐸1
𝑅𝐸2
𝐷𝐶 ‒ 𝐷𝐶 ‒
If IGBT 𝑇1 switches earlier than 𝑇2, a voltage drop occurs at the emitter inductance 𝐿1. This voltage drop
leads to a current (𝐼12, blue line) through the emitter resistors 𝑅𝐸1 and 𝑅𝐸2. The current causes a voltage
drop at the emitter resistors which counteracts the gate voltage at 𝑇1 and adds to the gate voltage at 𝑇2.
This means that the gate voltage is reduced on the IGBT which switches first (𝑇1), causing this to be
delayed. This effect is referred to as negative feedback. On the IGBT which switches later (𝑇2), the gate
voltage increases, whereby this switches faster (positive feedback). This means that the currents are
balanced during switching.
𝐷𝐶 +
𝐼1
𝑉𝐺𝐸1 𝑉𝐺𝐸2
𝑉𝑅𝐸1 𝑉𝑅𝐸2
𝐼12
𝑉𝐿1 𝐿1 𝑉𝐿2 𝐿2
𝐷𝐶 ‒
For this reason, in the design low and symmetrical emitter inductances are a must. This is more critical
with the TOP switch than with the BOT switch because the AC connection usually has a higher inductance
than the minus of the DC-bus.
This also shows the limits for a central driver. For systems with higher power, many modules must be
connected in parallel so that the emitter connections become longer and thus the emitter inductance is
large and unbalanced. As described, this leads to uneven current distribution and oscillations. This is why
individual drivers are used for large systems.
𝑫𝒓𝒊𝒗𝒆𝒓 𝑫𝑪 +
𝑻𝑶𝑷
𝑅𝐶𝑥
𝑅𝐺𝑜𝑛 𝑅𝐺𝑜𝑛𝑥 𝑅𝐺𝑜𝑛𝑥
𝑅𝐶𝑥
𝑅𝐺𝑜𝑓𝑓𝑥 𝑅𝐺𝑜𝑓𝑓𝑥
𝑅𝐺𝑜𝑓𝑓
𝑅𝐺𝐸𝑥 𝐶𝑥 𝑉𝑥 𝑅𝐺𝐸𝑥 𝐶𝑥 𝑉𝑥
𝑅𝐸𝑥 𝑅𝐸𝑥
𝑩𝑶𝑻 𝑨𝑪
𝑅𝐶𝑥
𝑅𝐺𝑜𝑛 𝑅𝐺𝑜𝑛𝑥 𝑅𝐺𝑜𝑛𝑥
𝑅𝐶𝑥
𝑅𝐺𝑜𝑓𝑓𝑥 𝑅𝐺𝑜𝑓𝑓𝑥
𝑅𝐺𝑜𝑓𝑓
𝑅𝐺𝐸𝑥 𝐶𝑥 𝑉𝑥 𝑅𝐺𝐸𝑥 𝐶𝑥 𝑉𝑥
𝑅𝐸𝑥 𝑅𝐸𝑥
𝑫𝑪 ‒
In order to minimise the influence of the semiconductor properties on the current distribution, modules
from one batch should be used. In general, a current unbalance of 5% should be expected. This current
asymmetry must be taken into account when calculating the conduction and the switching losses.
In order to achieve a symmetrical current distribution it is important to ensure that the impedances of the
parallel current paths are as equal as possible. To ensure this, the AC connection should be routed centrally
to the parallel-connected modules with the same length to the individual modules. The load cable should be
routed at a distance from the modules so that the inductance of the modules to the load stays as balanced
as possible.
In the inverter test, the current distribution is significantly better than in the double pulse test due to the
positive IGBT temperature coefficient and the high load inductance. The double pulse test is well suited for
investigating the switching behaviour and the symmetry but not the current distribution of parallel
connected modules.
When selecting the driver concept, the size of the system is crucial. For compact systems, a solution with a
central driver is preferable. The negative influences on the current distribution caused by the jitter, the
differences in the signal propagation times, as well as the differences in the gate-emitter voltages are
largely done away with in this concept. In addition, there is the positive effect of emitter feedback.
In the case of larger systems, the shared emitter path increases by design, and oscillations between the
parallel semiconductors or modules can occur due to the emitter feedback. In this case, the concept with
individual drivers must be used.
If individual drivers are used, care must be taken that the differences between the signal propagation times
of the drivers and the time differences due to the jitter are as low as possible. The time lag between the
drive signals of the IGBTs and the differences in the switching speeds of the IGBTs themselves are critical
when dimensioning the AC inductances. With low jitter and low differences in the signal propagation times,
an additional AC inductor is generally not required and the modules can be connected hard parallel.
If the current asymmetry of a hard parallel circuit is higher than the desired value, the parallel current
branches must be decoupled via as large inductances as possible. Particularly separate motor or line filter
windings are suitable for this because, in the case of high power, they are frequently already composed of
parallel-connected winding systems.
∆I Delta of current
∆t Delta of time
C Capacitor
D Diode
I Current
iC Collector current
K Branch point
L Inductance
Lσ Stray inductance
M Magnetic coupling
R Resistor
τ Tau
t Time
T IGBT / temperature
td Delay time
Tj Junction temperature
tjitter Jitter
V Voltage
VDC DC-voltage
VF Forward voltage
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
[3] M. Spang, “Current sharing between parallel IGBTs in power modules during short circuit with
unsymmetrically connected load”, 2016