Model-Based Power Converter Design Using High-Confidence
Model-Based Power Converter Design Using High-Confidence
lation time for one controller PWM cycle is 2.3 seconds on an Intel®
Core™ i7-9700 CPU @ 3.00GHz running MATLAB version R2021b.
This is fast enough to analyze circuit performance at current op-
erating state, but not to assess circuit sensitivity to design param-
eter sweeps or to directly optimize the circuit parameters. Further-
more, it is not fast enough to simulate to a periodic steady state
which, given a thermal time constant of approximately 10 seconds,
equates to 200,000 20kHz PWM cycles.
The model reflects that current flowing through the MOSFET causes
changes in semiconductor temperature which, in-turn, influences
the MOSFET electrical parameters such as charge carrier mobil-
ity, voltage threshold, drain resistance, gate-drain capacitance and
gate-source capacitance. Referring to Figure 2, thermal behavior is
modeled in the following way: a current source (Pv) representing
MOSFET dissipated power injects the heat into the PN-junction (Tj),
and that heat is then propagated all the way through MOSFET pack-
age to the case (Tc). The thermal dynamics is modeled as a Cauer
network made up of lumped thermal resistances (Rthi) and thermal
capacitances (Cthi). By means of analog simulation of the thermal
Figure 4: Detailed model of the buck converter
model, the optimum cooling/heat sink (Rth HS and Cth HS) can be de-
termined for given design parameters like load current, maximum
To address the need to explore the design space efficiently, a sys-
allowed junction temperature (Tj), ambient temperature (Tamb) and
tem-level version of the buck converter model created. For this, the
thickness/number of PCB layers (Rth PCB and Cth PCB).
imported MOSFET device models are replaced with ideal switches
Importing a subcircuit into Simscape with fixed on-resistance set to the datasheet Rds(on) value. This is
Simscape [5] from MathWorks provides a block diagram environ- shown in Figure 5. Some of the faster parasitics are also omitted,
ment to model multi-domain systems, including electrical, mechan- such as the MOSFET lead inductances. This system-level model is
ical, magnetic and thermal aspects. The accompanying Simscape fixed temperature, the user setting an appropriate Rds(on) value for
language expresses the underlying physics using differential equa- the assumed junction temperature. The model takes around 0.05
tions, associated algebraic constraints, events and mode charts. seconds to simulate one PWM cycle, 46 times faster than the de-
tailed model. As there are no thermal time constants, the slowest
dynamic is now associated with the voltage regulation and is of the
order of 5ms or 100 PWM cycles. Hence simulation to steady state
takes approximately 5 seconds.
With this simulation performance, the system-level model can be
used to thoroughly explore the design space and optimize the con-
troller. With the main design decisions made, the final step is to
validate the design using the detailed simulation model that makes
use of the Infineon MOSFET models. This validation is typically
Figure 3: Infineon’s Automotive MOSFET IAUT300N08S5N012 in TOLL
reported at a set of operating points defined by load power and
(PG-HSOF-8)
ambient temperature. However, we have seen that to simulate the
detailed model to steady state requires 200,000 PWM cycles which
Simscape Electrical™ [6] is able to import a targeted set of SPICE is not practical if each cycle takes 2.3 seconds to simulate.
device models, such as MOSFETs, into an equivalent Simscape lan-
guage implementation [7]. Simscape’s tight integration with Simu-
link then enables simulating both the digital controller and the ana-
log electronics with a single solver, this resulting in a more efficient
bodospower.com
Simulation workflow
Having imported the Infineon device into Simscape, the next step is Figure 5: System-level Simulink model of the buck power converter
to create a Simulink model of the complete converter including the
To initialize the detailed model at a specified operating point, an
imported Infineon devices, remaining analog components and the
iterative approach involving multiple models is proposed. Overall,
controller. This is shown in Figure 4.
the idea is to separate out slower time constants into separate
The controller is implemented using Simulink discrete-time library models that run faster. Before explaining in more detail, one more
blocks, and the complete model is simulated using a variable-step model is required which is one that models the MOSFET and envi-
solver so that the faster time constants associated with parasitics ronment thermal states only. This is shown in Figure 6.
and the MOSFET charge model are accurately captured. The simu-
28 Power Management July 2022
To construct the thermal-only model, the imported Infineon SPICE implement two or three MOSFETs in parallel for both the high-side
subcircuits are edited to leave just the Cauer network. The input and low-side switches. The important thing to note is that there can
to the two Cauer networks are two constant heat flow sources Q1 be a high-level of confidence in this result given that the validated
and Q2 which represent the average junction heat flow per PWM foundry SPICE MOSFET models were used to generate them and
cycle. This thermal-only model can either be run to steady state, or that the results are for the actual circuit. This gives a higher-level
the Simscape start from steady state option used. Either way, the of confidence than the sometimes-used alternative based on data-
time to solve the Cauer network node temperatures is negligible sheet plots of on-state and switching losses for a representative
compared to everything else. test circuit.
www.infineon.com
Figure 7: Losses in power switches and efficiency of overall system www.mathworks.com