sketched in Figure 3. The main difference between ideal and
real devices can be found during phase III, where the conduc-
tivity modulation takes place. This phase depends on the IGBT
technology, and its length is inversely proportional to the gate
current.
Depending on the device selected and its operating conditions,
the additional losses
can be significant. For the case
shown in Figure 2, the ideal losses
calculated according
to (1) amount to 83 mJ, but the total losses
equal to 142
mJ. Hence,
is equal to 59 mJ, 20% less than
. Note
that the example shown in Figure 2 represents an extreme
case, in which a 1200-V IGBT is operated at a voltage of 470
V.
SPICE models: Are they good enough?
After realizing that ideal waveforms are not sufficient for accu-
rate estimation of switching losses, it is important to evaluate
the accuracy of SPICE-based IGBT models, such as the one
provided by ROHM. However, when comparing experimental
waveforms with simulated ones, it is necessary to consider
that the IGBT in simulation will switch faster than in reality for
the same
value. In the case of the example shown in Fig-
ure 2, the simulation predicts a switching time of 38 µs, while
the measured value is 45 µs. This discrepancy can lead to un-
derestimation of losses if not compensated for. Specifically,
using the original
value in the simulation results in losses
of 82 mJ, while the measured losses are 142 mJ.
To match the experimental switching time, the value of
in
the simulation model was increased from 23 kΩ to 34 kΩ, as
shown in Figure 2. With this modification, the simulation repro-
duced a switching time of 45 µs, which matched the experi-
mental value. However, the SPICE model still did not accu-
rately reproduce the conductivity modulation phase, resulting
in 19% lower
losses compared to the experimental values.
While more accurate than ideal waveforms, standard SPICE-
based IGBT models remain limited in its ability to provide a
proper estimation of switching losses. The limitations of these
models have been well-documented in previous studies [1].
Proposed estimation method
Given that both ideal and simulated waveforms do not accu-
rately reflect the behavior of the IGBT for estimating switching
losses, a hybrid method is proposed here. This method in-
volves extracting device losses based on measurements and
reproducing the temperature swing based on worst-case op-
erating conditions using an equivalent RC thermal network.
Part 1: Experimental waveforms
To capture the
and
waveforms during turn-on and
turn-off, a single pulse test of the IGBT driving a resistive load
is carried out, using the circuit shown in Figure 4. To reflect the
temperature at which the device will be operating, the IGBT
should be heated up externally to a selected value between
100 and 150 °C. However, if the
value is high, setting a