You are on page 1of 4

Proceedings of the 2020 32nd International Symposium on Power Semiconductor

Devices and ICs (ISPSD) September 13 – 18, 2020, Vienna, Austria

Tools for Broadband Electromagnetic Modeling of


Power Semiconductor Packages and External
Circuit Layouts
Ivana Kovacevic-Badstuebner and Ulrike Grossner Dan Popescu
Advanced Power Semiconductor Laboratory, ETH Zurich Infineon Technologies AG
Zurich, Switzerland Neubiberg, Germany
Email: kovacevic@aps.ee.ethz.ch Email: danhoria.popescu@infineon.com

Abstract—Advanced power electronics applications using fast and exporting of the broadband models to circuit simulators
switching power semiconductor devices imply the needs for affect the model of parasitics.
electromagnetic (EM) tools for extracting the parasitics of two-
and three-dimensional geometries (i.e. PCB layouts and packages)
in an accurate and computationally efficient way. This paper II. E LECTROMAGNETIC M ODELING T OOLS
presents a comprehensive analysis of commercially available
The modeling domains in power electronics (PE) applica-
EM tools (ANSYS Q3D and HFSS) for broadband modeling of
circuit and package layout parasitics. The coupling of layout EM tions have to be typically treated as a mixture of two-(2D)
models and device models are performed in a circuit simulation and three-dimensional (3D) geometries. In the 3D domain,
environment. It is demonstrated that an accurate modeling of the EM modeling tools have to be able to accurately predict
layout parasitics in the range above 100 MHz is required in order the HF effects, i.e. skin and proximity effects, and as well
to predict the fast switching performance of power semiconductor
as the conduction losses in a wide frequency range. With
devices more precisely and this cannot be achieved by the low-
frequency EM tools. emerging fast switching power devices, the simulation ac-
curacy from dc to hundreds MHz and above becomes more
important. Commercially available EM modeling tools offer
I. I NTRODUCTION the numerical solvers specialized either for low frequency
(LF) or HF domain. For modeling from dc to HF frequencies,
Energy losses, electromagnetic (EM) interference and de-
two numerical solvers should be used, which requires the
vice stress are the key design aspects of power converters im-
results to be properly merged [7]. LF simulation tools such
plementing fast switching power semiconductor devices. Un-
as Q3D Extractor have been frequently used for extracting
derstanding dynamic effects strongly influenced by frequency-
HF effects contained in fast switching transients. However,
dependent parasitics is required for system-level optimization
the needs to resort to radio frequency modeling techniques for
and as well as highly beneficial for system designers in terms
predicting rather fast switching transients of power devices
of improving system performance and efficiently utilizing the
have been recognized [7], [8] and HF EM tools such as e.g.
next generations of power devices. Broadband EM modeling
ANSYS HFSS attract more and more attention. In this paper,
tools such as ANSYS HFSS and Q3D are frequently used
ANSYS HFSS is used as a representative of Finite Element
for the prediction of dynamic system performance. The ver-
Method (FEM)-based 3D EM simulators. It was shown that
ification of modeling can be performed indirectly by cur-
a broadband macromodel of layout/package parasitics can
rent/voltage measurements; however, the accuracy of dynamic
be built up by merging the S-parameter results gained from
simulations depends not just on the parasitics model but also
Q3D and HFSS [7], mainly based on adopting the Q3D LF
on the device models. Direct experimental verification of the
modeling and HFSS modeling in the HF range. However, the
parasitics modeling in a wide frequency range is practically not
inaccuracies of Q3D at HF and HFSS at LF were not shown.
feasible for more complex PCB layouts and power modules.
This paper analyzes the modeling accuracy of Q3D and HFSS
The parasitics of power modules at high frequency (HF) were
in a wide frequency range for fast switching PE applications
extracted neglecting the mutual couplings [1]–[3], while for
including packages and PCB layouts.
discrete components, the HF self- and mutual-parasitics were
extracted based on impedance measurements [4] and time-
A. Broadband Model Generation
domain reflectometry [5]. The mismatch between measure-
ments and modeling results is not well understood in practice Q3D Extractor, a specialized EM tool for the extraction of
[5] and can be ascribed to unknown internal package structure parasitics, generates an EM model of a 3D geometry structure
[6]. The purpose of this work is to show how the simulation based on separate calculation of capacitive and inductive
parameters as e.g. excitation ports, mesh, boundary conditions, effects in a form of an equivalent electrical circuit (RLGC

388
978-1-7281-4836-6/20/$31.00 ©2020 IEEE
Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 08:48:42 UTC from IEEE Xplore. Restrictions apply.
circuit). Using the defined ports (source and sink), the equiv- respectively, were used. The E4990A measurements are shown
alent circuit can be represented as a linear electrical network up to 10 MHz, since the E4490A features very low accuracy
described by S-parameters with a user defined characteristic for HF measurements of low resistive/high inductance com-
impedance, Z0 . An equivalent RLGC circuit can be used ponents; the measurements below 100 kHz are omitted from
only in ANSYS, while S-parameters can be exported to any the plot due to high noise levels. The HF effects seen in the
circuit simulator that allows a flexible multiphysics modeling. measured resistive behavior using the E4991B are not captured
A challenge for PE applications with exporting S-parameters in the HFSS and Q3D simulations. Fig. 4 implies that a better
from Q3D is a proper selection of Z0 , which should be matching between the impedance measurements and the Q3D
typically less than 50 Ω in order to keep a high accuracy in a simulations are achieved by setting up the ports to resemble
wide frequency range up to dc. the measurement conditions defined by the test fixtures.
The HFSS 3D-FEM solver takes into account the full
physics of electromagnetic fields and directly calculates S- ire late
dw ckp
parameters of the ports with the user-defined Z0 . As a FEM- n ba
bo
te
based simulator, HFSS requires a boundary air-region to be Dext ga
Din source
defined to limit the modeling domain. HFSS 3D-FEM simula-

die
Gext d bondwires
Gin lea
tions (solving EM fields inside of metal objects with arbitrary ga
te d
Sin lea
thickness) are typically computationally expensive, especially in
Sext dra e
if S-parameters have to be calculated in a wide-frequency urc
Port 1: Dext-Sext d2 so ad
l e
range without compromising accuracy. With this respect, in
order to speed up the calculations of broadband S-parameters
in HFSS, the interpolating sweep option is recommended. This Fig. 1. Modeling of a TO247-3 package in HFSS for the extraction of the
method fits S-parameter data to a rational polynomial transfer drain-source loop parasitics.
function using a minimum number of discrete FEM solutions.
Accordingly, a trade-off between simulation time and accuracy
of HFSS 3D-FEM simulations has to be made, which mainly
depends on the discrete FEM solutions calculated in LF range,
the specified error tolerances and the boundary region size.

B. HFSS vs. Q3D Ports


The ports model the connection of circuit components to the the le
ads
e
p art of test fixtur
layout of interconnections. The definition of Q3D equipotential a)
d1 th e
e of the
in s id
surface ports and their impact on the extraction of parasitics
was described in [6]. HFSS 3D FEM supports single-ended
and differential lumped ports [9]. A lumped port in HFSS is
defined between two conductive objects, where one object has d1
to be defined as a reference node. When the reference node d1
belongs to the global ground current path then it refers to a D lead Source Drain/Source lead
b) excitation ports c) excitation ports d) excitation ports
single-ended port. Another option is a differential port, which
assumes that a series component is connected at such a port Fig. 2. Q3D 3D model of a TO247-3 package mounted on the 16047E test
and any possible current path towards the ground plane is not fixture of Keysight E4990A impedance analyzer, b) a photo of the 16047E test
considered by the extracted S-network. Modeling a connection fixture of E4990A impedance analyzer, c) Q3D excitation ports on the drain
(D) and source (S) leads resembling the E4990A measurement conditions, d)
to a real ground path in a given geometry structure is however Q3D excitation ports at the cross section of the D/S leads.
not always feasible. Additionally, the length of lumped ports
should be kept as short as possible and/or de-embedded from The verification of the modeling results by measurements
the simulation in a post-processing step in order to compensate can be performed only for current loops such as e.g. the
for the port intrinsic inductance, which is not fully physically commutation and gate driver current loops. As Q3D calculates
accurate [9]. separately the capacitive and inductive behavior of 3D model,
Several ways of defining the Q3D and HFSS ports are a circuit simulator has to be used to verify the Q3D impedance
illustrated in Figs. 1-3 for a TO-247-3 package housing a matrix in a wide frequency range. The Q3D impedance results
1.2 kV SiC MOSFET. The impedance measurements of the plotted in Fig. 4 are extracted in the ANSYS circuit simulator,
drain-source current path inside of the TO247-3 package with Simplorer, using S-parameters for Z0 = 1 Ω, which allows
the device in on-state are compared with the Q3D and HFSS to match the L and R calculated directly in Q3D in a
simulations to show how the definition of ports affects the wide frequency range. In the LF range, the Q3D and HFSS
extraction of power loop impedance (Fig. 4). Two Keysight simulations show a very good matching in this experiment, as
impedance analyzers, E4990A and E4991B, operating in the the lowest frequency calculated by HFSS is 1 Hz, which was
frequency range of (20 Hz-120 MHz) and (1 MHz-500 MHz), acceptable with respect to the simulation time for this smaller

389
Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 08:48:42 UTC from IEEE Xplore. Restrictions apply.
TO247-3 package in Fig. 6. The commutation loop impedance represented by the
Signal equivalent series inductance L and resistance R is simulated
Ground Pin
Reference in the ANSYS Nexxim Circuit simulator importing the S-
Ground Fixture parameters extracted from the HFSS and Q3D solvers. Fig. 6
shows that the HFSS ports modify the actual Lloop simulated
without removing the 3D (conductive) device models, i.e.
a) b) Leq,loop �= Lloop , to a smaller extent (4.6 %) than the equi-
potential Q3D ports (21 %). Both the HFSS and Q3D ports

excitation port
have a more significant influence on the modeling of LF Rloop .

Source
In the LF range, i.e. below ≈ 1 MHz, HFSS simulations show
excitation port

a significant deviation from the Q3D simulations. On the other


Drain

hand, the Q3D modeling of the commutation loop in the


c) HF range, i.e. above ≈ 100 MHz, is not highly accurate. It
was observed that the size of the boundary region in HFSS
Fig. 3. Q3D 3D model of a TO247-3 package mounted on the spring clip mainly affects the modeling in the LF range. For the frequency
fixture of Keysight E4991B impedance analyzer (b) a photo of the spring clip
fixture, c) Q3D excitation ports on the drain and source leads resembling the range between ≈ 1 MHz and ≈ 100 MHz, the difference
E4991B with the spring clip fixture measurement conditions. between the HFSS and Q3D modeling for this geometry is
not significant. Accordingly, HFSS is a recommended tool for
E4990A Meas Q3D E4990A + Radd modeling the HF effects (> 100 MHz) occurring in the fast
E4991B Meas Q3D E4991B+ Radd switching transients as it will be further demonstrated in the
Q3D Cross Sec. + Radd HFSS d2 = 2mm + Radd
next section.
11 350
10 300
PCB board with Cdec
250 DC+ DHS decoupling
port1
9
R [mŸ]
L [nH]

200 capacitor (Cdec)


8 150 SHS
port3

7 100 DLS
DCB
port2

50
102 104 106 108 102 104 106 108 SLS substrate low-side (LS)
DC-
a) f [Hz] b) f [Hz] high-side (HS) switch
a) b) backplate
Fig. 4. Inductance (L) and resistance (R) of the drain-source path inside
of a TO247-3 package: impedance measurements using Keysight E4990A
and E4991B impedance analyzers (MOSFET turned-on) vs. HFSS/Q3D
simulations (10 Hz-1 GHz) using: the equipotential ports at Fig. 2c) (Q3D
E4990A), at Fig. 2d) (Q3D Cross Sec.), at Fig. 3c) (Q3D E4991B) and the
Port 1 at Fig. 1 (HFSS). c)
HFSS lumped Port Q3D Eq. Potential Q3D Eq. Potential
between DC+ and DC- DC- Port DC+ Port

modeling task. At the frequencies above ≈ 100 MHz when lumped


the parasitic capacitive behavior of the drain-source current port 1 lumped port 3
path is increasing, the differences between the Q3D and HFSS
simulations are more significant.
source-plate
HS switch
III. HFSS VS . Q3D F REQUENCY D OMAIN d)

The advantage of EM modeling is a possibility to evaluate Fig. 5. Simplified model of the half-bridge power module [10] in HFSS and
partial current paths and their mutual couplings, which cannot Q3D: a) the equivalent circuit, b) 3D model, c) the comparison between the
be measured typically by available measurement equipment. excitation ports in Q3D and HFSS, and d) HFSS excitation ports.
When extracting the layout parasitics, 3D models of power
semiconductors are removed and the ports are set to extract
the equivalent electrical network. The ports as boundary IV. HFSS VS . Q3D T IME D OMAIN
conditions of EM simulations can affect the calculation of For modeling the switching transients, device models are
the current distribution inside of the package layout. The simulated in time-domain circuit simulators together with
significance of their impact on the modeling results depends layout and package EM models. Accordingly, first the time
mainly on the modeling geometry, as it was demonstrated domain representation of frequency-dependent layout and
for Q3D ports in [6]. The definition of HFSS and Q3D package models has to be generated. This can be achieved
ports of a wirebondless half-bridge power module from [10] either by extracting a circuit netlist from S-parameters (e.g. by
are described on Fig. 5. The differences in extracting the using IdEM [11]) or directly using S-parameters in a circuit
commutation loop impedance using HFSS and Q3D are shown simulator, which implements advanced algorithms for the

390
Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 08:48:42 UTC from IEEE Xplore. Restrictions apply.
Q3D Loop (Lloop, Rloop) Q3D Partial (Leq,loop, Req,loop) between the cells due to the internal gate layout at HF.
HFSS Loop (Lloop, Rloop) HFSS Partial (Leq,loop, Req,loop)
8 105
7
Q3D HFSS
104 30 30
6 103 20 Id [A] 20 Id [A]
10

R [mŸ]
L [nH]

10 0
5 102 0 -10
145.0 145.1 145.2 145.0 145.1 145.2
4 101
20
t [ȝs] t [ȝs]
3 100 40
10 Vgs [V] 20 Vgs [V]
2 10-1 0 0
-20
103 104 105 106 107 108 109 103 104 105 106 107 108 109 -10 -40
145.0 145.1 145.2 145.0 145.1 145.2
a) f [Hz] b) f [Hz]
t [ȝs] t [ȝs]
0.4 0.4 time [s] 10-4
0.3 0.3
Fig. 6. HFSS and Q3D simulation of the commutation loop path between DC+ 0.2 0.2
Vds [kV]
0.1 Vds [kV] 0.1
and DC- inside of the the half-bridge power module [10]: a) the equivalent 0 0
145.0 145.1 145.2 145.0 145.1 145.2
series inductance (L) and b) the equivalent series resistance (R).
a) P7 t [ȝs] b) C7 t [ȝs]

Fig. 7. Simulated switching transients for Vgs = 0-15 V, Vds = 400 V,


S-parameter-to-impulse response conversion and convolution Id = 24.5 A, Rg,ext = 10 Ω comparing (a) the slower switching device
IPW60R180P7 and (b) the fast switching device IPP60R180C7.
[12], such as e.g. ADS. A special care has to be taken to ensure
that the starting S-network extracted from the HFSS and Q3D
V. C ONCLUSION
solvers do not violate passivity and causality [13] or to use
the enforce passivity option when generating the time-domain The presented analysis provides a deep understanding of
models. 3D FEM HFSS solver has an option to minimize the the EM modeling using the available commercial software
passivity and causality violation using the interpolating sweep, tools, which is highly important for more accurate virtual
while ADS and IdEM can enforce passivity on the given S- prototyping.
parameters. However, enforcing passivity can have a negative R EFERENCES
impact on the accuracy of the generated time-domain models. [1] T. Liu et al., “A new characterization technique for extracting parasitic
The multiphysics coupling between the device models and the inductances of SiC power MOSFETs in discrete and module packages
layout/package EM model in a circuit simulation environment based on two-port S-parameters measurement,” IEEE Tran. on Power
Electronics, vol. 33, no. 11, pp. 9819–9833, 2018.
is described using an example of a test board from [7] with [2] L. Pace et al., “Extraction of packaged GaN power transistors parasitics
ADS and IdEM for generating the time-domain models. The using S-Parameters,” IEEE Tran. on Electron Devices, vol. 66, no. 6,
current commutation between a CoolMOS superjunction (SJ) pp. 2583–2588, 2019.
[3] A. Lemmon and R. Graves, “Parasitic extraction procedure for silicon
MOSFET (IPP60R180C7, IPW60R180P7) and a SiC power carbide power modules,” in Proc. of IEEE Int. Workshop on Integrated
diode (IDH16G65C5) is simulated in ADS together with the Power Packaging (IWIPP), 2015, pp. 91–94.
HFSS/Q3D EM 3D model including a PCB board and two [4] Y. Mukunoki et al., “Modeling of a SiC MOSFET with focus on internal
stray capacitances and inductances, and its verification,” IEEE Tran. on
TO-packages. The simulated switching transients of the fast Industry Applications, vol. 54, no. 3, pp. 2588–2597, 2018.
switching high efficiency device IPP60R180C7 and the slower [5] H. Iida et al., “Mutual inductance influence to switching speed and
switching device IPW60R180P7 are presented in Fig. 7. Fig. TDR measurements for separating self-and mutual inductances in the
package,” in Proc of 31st Int. Sym. on Power Semiconductor Devices
7 shows that the fast switching device causes significantly and ICs (ISPSD), 2019, pp. 503–506.
higher oscillations, and the differences between the circuit [6] I. Kovacevic-Badstuebner et al., “Parasitic extraction procedures for
simulations implementing the Q3D and HFSS EM models are SiC power modules,” in Proc. of 10th Int. Conf. on Integrated Power
Electronics Systems (CIPS). VDE, 2018, pp. 1–6.
then more pronounced. It can be also concluded that the EM [7] D. Popescu and M. Treiber, “Broadband TCAD mixed-mode simulation
modeling accuracy above ≈ 1 MHz is the most relevant for es- framework for predictive modeling of fast dynamic switching events,”
timating fast switching transients, however, the inaccuracies of in Proc. of 31st Int. Sym. on Power Semiconductor Devices and ICs
(ISPSD), 2019, pp. 327–330.
HFSS S-parameters in the LF range can impact the estimation [8] M. Mazzola et al., “Behavioral modeling for stability in multi-chip
of conduction losses and introduce voltage/current offsets. An power modules,” in Proc. of IEEE Int. Workshop on Integrated Power
ADS simulation directly using S-parameters was faster than Packaging (IWIPP). IEEE, 2015, pp. 87–90.
[9] W. Sun, “Accurate EM simulation of SMT components in RF designs,”
an ADS simulation using a netlist generated in IdEM, and a in Proc. of IEEE Radio Frequency Integrated Circuits Symposium
smaller mismatch between the simulations was observed. (RFIC), 2017, pp. 140–143.
[10] M. Guacci et al., “Analysis and design of a 1200 V All-SiC planar in-
Very high oscillations in the switching transients are com- terconnection power module for next generation more electrical aircraft
mon for fast switching devices featuring low integrated gate power electronic building blocks,” CPSS Tran. on Power Electronics and
resistance such as high efficiency C7 CoolMOS devices. These Applications, vol. 2, no. 4, pp. 320–330, 2017.
[11] 3DS, “IdEM,” 2020. [Online]. Available: https://www.3ds.com/
oscillations are determined by both parasitics and the device products-services/simulia/products/idem/
physics. A highly accurate prediction is typically not feasible [12] F. Rao et al., “The need for impulse response models and an accurate
by using common compact device models. A 2D TCAD device method for impulse generation from band-limited S-parameters,” in
DesignCon 2008, 2008.
model of a single device cell limits the HF modeling as [13] P. Triverio et al., “Stability, causality, and passivity in electrical inter-
well, and therefore, TCAD modeling of several device cells in connect models,” IEEE Tran. on Advanced Packaging, vol. 30, no. 4,
parallel would be required to consider the signal propagation pp. 795–808, 2007.

391
Authorized licensed use limited to: Cornell University Library. Downloaded on September 01,2020 at 08:48:42 UTC from IEEE Xplore. Restrictions apply.

You might also like