See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286258679

Extreme Thermal Transient Stress Analysis with Pre-Stress in a Metal Matrix

Composite Power Package

Article  in  Additional Conferences (Device Packaging HiTEC HiTEN & CICMT) · January 2012
DOI: 10.4071/HITEC-2012-THA25

CITATIONS READS
11 9,293

4 authors, including:

D.C. Hopkins James Pitarresi

North Carolina State University Binghamton University
128 PUBLICATIONS   1,804 CITATIONS    71 PUBLICATIONS   1,054 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Electrohydrodynamics of Dielectric Fluids for Insolation in HV Power Modules View project

Predicting Solder Joint Reliability View project

All content following this page was uploaded by D.C. Hopkins on 08 February 2016.

The user has requested enhancement of the downloaded file.

Extreme Thermal Transient Stress Analysis with Pre-Stress in a Metal Matrix
Composite Power Package

Douglas C. Hopkins Theodore Baltis, James M Pitaress Donald R. Hazelmyer

North Carolina State University Binghamton University DensePower, LLC
1791 Varsity Dr., Suite 100 Room: ES- 1312 1201 E. Fayette St.
Raleigh, NC 17606 Bingahmton, NY 13902 Syracuse, NY 13210
d.hopkins@ieee.org jmp@binghamton.edu don.hazelmyer@densepower.com

grid technologies, SiC- and GaN-SSCBs can provide

Abstract
the fastest protection with better controllability in
This paper culminates several years of power distribution.
development of a SiC MOSFET-based solid-state
The overall SSCB module design
circuit breaker power module designed and fabricated
incorporating all four dies in parallel, would provide an
for aluminum-based composite metal-ceramic
operating capability of 120 Adc nominal with 1200 Adc
packaging. The aluminum composite structure was used
fault current. The application was primarily focused on
for high temperature thermal management >350˚C and
270 Vdc aircraft electrical systems, though all devices
high reliability. Testing of the final module surpassed
and packaging also applies to consumer electrical
750 total cycles.
vehicle and DC Smart Grid building applications at 380
The module was optimized for four dies, Vdc. Two power module designs are reported. One,
4.1mm X 4.1 mm each for SiC or GaN. The electrically which is physically realized, uses four 4.1 mm X 4.1
loading per die was 48 A for 5 ms with a di/dt of 2.1 mm dies in an approximate 6cm X 6cm module (with
kA/us (23 ns opening time). An internal snubber 2.5 cm X 2.5 cm active area). The second design,
increased the response to 390 ns. The die absorbed ~4.6 developed through simulation, uses four 7 mm x 8 mm
J causing a transient junction temperature increase of dies for 1200 Adc.
~245 °C. Two ambient temperatures were used in
Both designs use a newly developed cast-
testing and set at 25 °C and 105 °C. The maximum
Aluminum composite module capable of managing
junction temperature was conservatively projected to
transient temperatures to >350°C. The electrical
reach 350 °C during the 5 ms pulse. Electrical, thermal
transient capability is directly correlated to the thermal
and mechanical design and testing results are presented.
and mechanical transient capability of the packaging.
For high reliability, a cast Aluminum(Al) interconnect
on Aluminum nitride ceramic captured in a cast
Introduction of the Application
Aluminum-composite baseplate was developed,
Solid-State circuit breakers (SSCBs) (and simulated and tested. The results of the module
power controllers, SSPCs) are replacing many development from package development to electrical
conventional mechanical breakers, and provide smaller, performance are reported here.
lighter, faster and more reliable characteristics [1-5].
SSCBs, using silicon devices, are commercially
available, e.g. for aircraft applications, but limited to A. The Development of SSPCs
secondary distribution systems due to relatively low
The concept of solid-state power control in
power ratings and high power loss. Compared to
power systems was originally introduced in military
silicon, silicon carbide (SiC) and gallium nitride (GaN)
aircraft and aerospace applications [6-8]. The first
devices provide lower on-state resistance, higher
SSPC demonstration was in an aerospace demonstrator
breakdown voltage, and superior transient capabilities
– X-33, and firstly applied on military aircraft,
due to higher thermal die conductivity, higher
Lockheed C-130J, in 1995-96 [6]. It was then applied to
crystalline strength and lower expansion coefficient.
the space station and several commercial aircraft, such
The first two advantages significantly increase the SiC-
as Airbus A-380, Bombardier Global Express light
SSPC power rating. The latter advantages provide
passenger jet, etc.
greater reliability, particularly at higher junction
temperatures. This research reports designs for Tj > 350 For 270 Vdc onboard electrical systems, the
˚C, that provide greater than 35% current clearingT present commercial SSPC products are available for
capability or density. With the development of smarte 30A and below due to size, weight, power loss, and
xt
High Temperature Electronics Conference (HiTEC), Albuquerque, NM, May 8-10, 2012
T
e
xt
economic considerations. A 50 A Si IGBT-based SSPC layers to keep the die under compression. Stress
is reported in the literature as an experimental prototype translation in a function of a materials CTE and Young’s
[9]. A 100 A Cool MOS-based SSPC was proposed modulus.
through simulation [10], to determine the required Table 1 Structure dimensions
number of silicon die for transient operation, but Layer Thickness(um) Half-Width (um)
packaging issues are not addressed. Wide-bandgap
semiconductor devices are firstly proposed to SSPCs in Silicon 350 5100
[11] using SiC JFETs and capable of carrying 30A Solder (63Sn/37Pb) 100 5600
nominal in 270Vdc systems with a maximum 200°C Copper, Aluminum 356 75, 200, 380, 510 6600
junction temperature, but the advantages of wide-
Al2O3, AlN, BeO 75, 255, 635 7100
bandgap semiconductor were not extensively explored.
Al 356 50 7100
635, 1270, 1905,
AlSiC 8400
B. Advantages of SSCBs 2540
Thermal Paste 50 8400
The SSCB is capable of interrupting fault
currents in sub-microseconds, far below the maximum Copper Heat Sink 3200 10900
short-circuit current that effects wire gauge or cable Traditionally, Copper (Cu) is used in directly
sizeing. Hence, a significant system-wide weight bonded metal to ceramic substrates, i.e. direct bonded
reduction in wires and cables can be achieved. Also, a copper (DBC) or direct copper bonded (DCB)
fast fault-current response protects generators, substrates. Direct Bonded Aluminum (DBA) or Direct
Aluminum Bonded (DAB) ceramic substrates were
developed recently to replace DBC as an industry
standard.
Table 2. Material Properties
Conductivity Young’s
TCE
Layer -6 Conductivity Modulus
(10 /˚C)
(W/m˚C) (GPa)
Silicon 3.1 105 110
Solder
24.7 50.6 14.9
(63Sn/37Pb)
Cu, Al-356 17.3, 6.61 383, 159 125.7, 69.0
Fig. 1 The structure represents the proposed cast metal-
Al2O3, AlN, 7.36, 3.8, 372.7, 345.0,
ceramic structure for power module applications. The 25, 170, 280
BeO 5.8 344.7
module was analyzed with half symmetry.
Al 356 6.61 159 69.0
transformers, and terminal equipment. Solid-state AlSiC 7.2 165 226
devices, as the name implies, do not have an arc Thermal Paste - 2.5 -
problem. Reduction of the fault-current magnitude and
Copper Sink 24.7 383 125.7
clearing time greatly reduces the possibility of electrical
system fires. Studies show that saving 1 lb (0.454 kg) The Cu has a high CTE to ceramic, as does Aluminum
on each aircraft in a fleet of 800 widebody aircraft in (Al). However, the Young’s Modulus for Al is lower
service over a 15-year period is worth $830,700 [12]. that Cu and can help better manage stress in a metal-
For military aircraft whose agility is critical, weight ceramic layered structure. The DBA substrates have the
saving brings improvements on acceleration and potential to provide the needed reliability and have a
maneuverability. tolerance to thermal cycling that is substantially better
than DBC substrates.

Metal Material: Al Vs Cu Thermal cycling causes stress in the bond

During fault clearing, the GaN or SiC device
must absorb transient thermal energy, and as such
experiences rapid heating. This directly affects its
mechanical dimension due to CTE. This causes severe
stress on the die, die-attach material and interconnects.
Each structural layer, with its own CTE, must be
harmonized to translate stress, yet constrain adjacent
between layers in a substrate. The stress is a result of the
expansion and contraction of adjacent layers that have
differing thermal expansion coefficients. Aluminum,
being a more malleable metal than Cu, better withstands
thermo-mechanical stresses.
 Copper-Silicon Aluminum-Silicon

180 180
160 160
Silicon Max Stress (MPa)

Silicon Max Stress (MPa)

140 140
120 120
100 AlN 100 AlN
Al2O3 Al2O3
80 80
BeO BeO
60 60
40 40
20 20
0 0
0 50 100 150 200 0 50 100 150 200
Silicon Max Temp (C) Silicon Max Temp (C)
Figs 2a and 2b. Shown are maximum values of stress versus temperature in the silicon layer for every simulation. Copper
interconnects on the left and aluminum interconnects on the right. The colors reflect the type of ceramic used: Blue for AlN, pink
for Al2O3 and yellow for BeO.
Copper-Copper Aluminum-Aluminum

300 300

Aluminum Max Stress (MPa)

Copper Max Stress (MPa)

250 250

200 200
AlN AlN
150 Al2O3 150 Al2O3
BeO BeO
100 100

50 50

0 0
0 50 100 150 200 0 50 100 150 200
Copper Max Temp (C) Aluminum Max Temp (C)
Figs 3a and 3b. Shown are maximum values of stress versus temperature in the metal interconnect layer for every simulation.
Copper interconnects on the left and aluminum interconnects on the right.

Copper-Ceramic Aluminum-Ceramic

500 500
Ceramic Max Stress (MPa)

Ceramic Max Stress (MPa)

450 450
400 400
350 350
300 AlN 300 AlN
250 Al2O3 250 Al2O3
200 BeO 200 BeO
150 150
100 100
50 50
0 0
70 90 110 130 150 70 90 110 130 150
Ceramic Max Temp (C) Ceramic Max Temperature (C)

Figs 4a and 4b. Shown are maximum values of stress versus temperature in the ceramic layer for every simulation. Copper
interconnects on the left and aluminum interconnects on the right.
 The trends seen in this report may provide
To determine a DBA or DBC preference, the
insight into substrate optimization for power modules.
structure shown in Fig. 1 was analyzed with ABAQUS
An optimum combination of interconnect types,
for 228 combinations of varying thickness of Al and Cu
ceramic types, and layer thicknesses could be chosen to
interconnect; Al2O3, AlN and BeO ceramic; and
create a substrate that is customized for the
thickness of metal composite baseplate as noted in Table
environment in which it will be operating. This will
1. Material parameters are given in Table 2.
enable the production of more cost effective and
The structure was modeled after a 1.04 cm2 reliable power modules.
IGBT. A half-symmetry 3D model was created using
the ABAQUS finite element analysis software. This
model was compared to a half symmetry 2D model that Power Substrate Development
was extruded 1 µm in the z-axis (into the page in Fig. 1)
The SSCB package approach is document in
creating a very thin 3D cross section. The results
previous publications. In summary, an Al composite is
showed that for equivalent meshing, there was less than
shaped as a base carrier. An AlN ceramic is placed on
3% difference between the two models both thermally
the carrier, and the entire structure is cast such that the
and mechanically. The extruded 2D model was chosen
AlN is captured onto the composite while an
to reduce computing time.
interconnect pattern, which is machined into the casting
The model used 8-node thermo-mechanical die, is cast onto the AlN ceramic. The resulting module
solid elements. Adiabatic boundary conditions were baseplate with Integrated Ceramics (MOSAIC [14]) is
applied to all surfaces except the bottom of the copper shown in Fig. 5 after casting. The center starts as a
heat sink layer, which was held at a constant sink ~350um thick aluminum pad that can be etched or pre-
temperature of 25°C. Half-symmetry boundary cast to form interconnects.. A post process was
conditions were applied to the left-hand surfaces of the developed to provide an electrically clean ceramic
model and the bottom corner of the AlSiC layer was surface for high voltage withstand.
pinned at the symmetry line. The structure was meshed
to a resolution that would give minimal thermal stress
error and a reasonable computation time. The
simulation was run for every combination of the chosen
variables given in Table 1, using the material properties
given in Table 2. This resulted in 288 simulations.
The maximum stresses in each layer for each
simulation case are plotted in Figures 2, 3 and 4, for Cu
and Al respectively. The colors reflect the type of
ceramic. Blue data are for AlN, pink data for Al2O3,
and yellow for BeO.
The minimum metal interconnect thickness is
determined by the electrical designer based upon the
minimum electrical resistance needed [13]. The
thickness of the interconnect layer directly affects the Fig. 5 Cast patterns with 1 mm lines and spacing. The cast-
magnitude of the thermo-mechanical stresses that occur process allows direct definition of interconnect patterns and
in the other layers during operation. The data present in components. Also shown is a 1 mm thru-ceramic filled via.
this report show the different options that power
packaging engineers have to control the thermo- Outer to the AlN, is a full Al-Metal Matrix
mechanical stresses. composite baseplate for thermal and mechanical
The results of this report show persuasive management. Traces that can be “cast-in” to form
evidence that aluminum 356 interconnects are a viable interconnects and electrical components. e.g embedded
replacement for copper in power modules. Aluminum current shunts. The casting resolution was demonstrated
356 interconnects greatly reduced stresses in every to be 1mm line widths and 1mm line spacing. Also
layer studied. Although further research is needed to shown is the capability to insert 1mm diameter through-
form a definitive conclusion, aluminum 356 ceramic vias. The vias are filled during the casting
interconnects could lead to a more robust power module process.
that is better suited for today’s severe power demands.
 Thermal Simulation and Characterization of Table 3. Thermal Properties and Thicknesses of Module
Module Layers
Silver Al
A. Setting up the Thermal Model 4H SiC AlN Baseplate
Glass trace
Density
The overall geometry of the module is not 3.21 7.52 2.7 3.3 2.4
(g/cm3)
complicated and was easily built by adding and Thermal X - 230
overlapping volumes. With the model built, the Cond. 490-160 200 250 180 Y - 230
material properties and loads/conditions could be (W/m*K) Z - 120
applied to properly set up the simulations. In contrast Specific
to the heat generation load previously mentioned, the Heat 690 310 230 740 852
initial conditions and constraints on the model are very (J/kg*K)
simple. An initial condition of 105°C was applied to all Thickness
0.4 0.08 0.381 0.635 6.35
domains of the model, and to allow for thermal (mm)
equilibrium a thermal ground of 105°C was applied to
are as pronounced as that of the SiC thermal
the entire bottom surface of the baseplate, which is
conductivity. Also, because this is the layer where all
where the module would be in contact with outside
of the heat is generated it is also the most important
components when bolted into place. All of the
layer to accurately represent in the model. In addition,
simulations performed as part of this research assume
this relation is best represented by a second-order
adiabatic conditions where heat is only transferred by
polynomial which places a considerable burden upon
conduction through the module layers. The simulations
the computational effort needed, and limiting the
also assume that all module layers have even surfaces
and the effects of imperfect bonds are not considered.
Also note that all of the simulations assume equal
power density amongst the dies operating in parallel.
As previously mentioned, the loading in the
simulations is a temperature-dependent heat generation
following the form of i2R, where i is the input current
and R is the temperature-dependent resistance of the
semiconductor device. Using the resistance versus
temperature function
Rds−on (T ) = ( 7E − 07 × T 2 ) − (0.0004 × T ) + 0.0821 Eq 1,

which was previously determined for the SiC Fig. 6 Thermal Conductivity of MOSFETs vs Temperature
MOSFETs in earlier research [14], the entire heat
€ generation function (for the maximum fault current load number of variable-dependent material properties helps
of 240ADC per die) is given below in Eq 2. In the keep this computational burden to a minimum.
semiconductor layer of the model, a separate volume
was created inside of each device so that this load could Eq 3
be applied to only the active volume of each device.
Again, please note that the unit of temperature used in This equation describing the non-linear
Eq. 2 is Kelvin. behavior of the SiC thermal conductivity is given below
in Eq. 3, along with Fig. 6, which visually demonstrates
(240) 2 ((6E − 07 × T 2 ) − (0.0003 × T ) + 0.0732) the relationship between thermal conductivity and
Heatgen (T ) =
3.68 temperature. Note that the units of temperature in the
equation are again given in Kelvin.
[W/m3] Eq 2
€ With all of the loads and constraints set up in
the model, the material properties could be applied to B. Steady-State Thermal Response
each respective layer. The material properties and
With the non-linear heating effect properly
thickness of each layer can be found below in Table 3.
represented in the model, simulations were performed
As seen in the table, the thermal conductivity of the SiC
to investigate the thermal and mechanical behavior of
devices is given as a range because this property
the module during operation. Although the focus is the
changes considerably over the operating temperature
transient behavior of the module in response to the
range of the module. While other properties may also
maximum fault current, it was important to first verify
be dependent upon temperature, none of these relations
that the module could operate indefinitely at the
nominal current, i.e. that the module could reach DensePower module consisting of smaller (4.1mm x
thermal equilibrium with a maximum temperature 4.1mm) dies [14]. The extreme improvement in this
under that of the junction temperature so that there parameter helps demonstrates the impressive power and
would be no chance of tripping under these conditions. heat management capabilities of the module.
A steady-state simulation was run where the i2R relation
In the time it takes the MOSFET to reach the
was applied to each of the MOSFETs in the model with
junction temperature the produced heat is only able to
a nominal current value of 24A (96A distributed evenly
diffuse down through the module just slightly into the
across all 4 dies). The results of this simulation are
baseplate, but again the temperature gradient produced
documented in reference [14,15]
is very uniform as seen in Fig. 4.7. This even
The simulations show a steady state distribution of temperature demonstrates that the
temperature of approximately 111°C, a 6°C temperature generated heat is evenly spreading through the layers,
rise. Naturally, the temperature is greatest in the again due to the near uniformity of thermal conductivity
semiconductor where the heat is generated and of the layers.
gradually varies down to the ambient temperature at the
bottom of the baseplate, which acts as a thermal ground
in the module. Due to the near-uniformity of the
thermal conductivity of the layers the temperature
gradient is very even throughout the entire thickness of
the module. A quarter-symmetry surface temperature
plot is also provided below in Fig. 7 to demonstrate
how the temperature distribution is uniform in both the
x and y-directions.

Fig. 8 Transient Cross-Sectional Temperature Distribution

Mechanical Simulation and Characterization of

Module
Setting up the Thermal Stress Model
For the thermal stress simulations, all of the
thermal loads and conditions were kept the same as
before and the only new condition was that of a zero-
displacement condition that was applied to the bottom
surface of the baseplate. Note that quarter-symmetry
Fig. 7 Steady State Quarter-Symmetry Temp. Distribution
was implemented in these simulations to reduce the
computational burden and accelerate the simulations.
The mechanical properties of each of the module layers
C. Transient Thermal Response
are provided below in Table 4
To study the transient behavior of the module
Table 4. Mechanical Properties of Module Layers
in response to the maximum fault current of 240ADC
per die, the simulation was manually iterated due to the Silver Al
lack of a temperature-based cutoff option in the 4H SiC AlN Baseplate
Glass trace
software. After several runs it was found that the Young's
module is capable of sustaining a maximum fault Modulus 410 11.1-3.0 70-50 330 98.6
current rating of 240ADC for approximately 3ms (GPa)
before reaching the maximum junction temperature of Poisson's
350°C. With the sustaining time determined, it is 0.14 0.33 0.35 0.24 0.30
Ratio
possible to calculate the i2t constant for the trip profile X-4
of this specific solid-state power controller. CTE
4 16 23 4.5 Y-4
Multiplying the sustaining time by the square of the (ppm/°C)
Z - 24
fault current per die, it is found that the semiconductors
have a trip constant of 172.8, which is over 4 times Note that Young’s Modulus of two of the
greater than the constant determined for the layers, the silver glass adhesive and the aluminum trace,
are given as a range. This is because it was found that higher-order polynomial may still not sufficiently
these properties change considerably over the operating describe the behavior of the layer, and applying such an
temperature range. equation in the program would definitely add a
significant computational burden to the simulations. In
While the relevant material property data for
order to circumvent this problem an alternative
aluminum could only be found for temperatures up to
technique was used in order to capture the dynamics of
200°C, extrapolating this data out to 350°C indicates
the silver glass behavior. Using data from the
that the Young’s modulus of this layer will lower by
manufacturer, an extrapolation function was created
approximately 20% over the operating temperature
where lower-order equations could be used to find the
range of the module. This demonstrates that the
value of Young’s modulus at any temperature. This
aluminum becomes significantly softer as temperature
allows us to incorporate the overall non-linear behavior
increases in response to a fault current. This is
in a way that will not overburden the available
beneficial to the module for it will mitigate the
computing resources.
accumulation of stresses in this layer and therefore it is
important to include this material behavior in the
simulations. A plot demonstrating the relation between
temperature and Young’s modulus of aluminum is
provided in Fig. 9. This behavior closely approximates
by a linear relation as presented in Eq. 4. Note that this
equation uses temperature units in Kelvin to produce a
Young’s modulus in Pascal so that it may be used in the
simulation software.

Fig. 10. Young's Modulus of Silver Glass vs Temp.

Transient Response
Figure 11 shows the stress tensor in the x-
direction at 3ms. As indicated by the legend, most of
the module is experiencing compressive stresses. This
may be alarming at first due to the general
Fig. 9. Young's Modulus of Aluminum vs Temperature understanding that materials tend to expand as they heat
up, but this is actually why the module is experiencing
compression. The most important parameter in the
Eq. 4 module is the coefficient of thermal expansion as it is
directly responsible for the behavior of the module
The same could be said for including the layers.
temperature dependent Young’s modulus of the silver
glass layer, but this material’s behavior is a bit more
complicated. This is because the silver glass adhesive
becomes much softer at higher temperatures and
actually undergoes a glass transition at approximately
170°C. At this temperature, the material enters a more
rubber-like state and its Young’s modulus significantly
and rapidly decreases with further heating. The
relationship between the Young’s modulus of the silver
glass and temperature was provided directly by the
manufacturer and is presented in Fig. 10.
As seen from the figure, the behavior of this Figure 11. X-Tensor Stress Distribution at 3ms
property can be described as a fairly linear section To understand the behavior of the module
followed by a non-linear section, therefore making it layers one could look, for example, at the AlN layer and
difficult to define the entire relation with a single those adjacent to it. From looking at Table 4 above,
equation unless it is of very high-order. The use of a one can see the relation of the CTEs of the Al trace,
ceramic AlN, and composite baseplate layers.
Disregarding the z-direction CTE of the baseplate
(these results are more focused on the interaction of the
layers in the xy-plane), there is a reverse-hierarchy in
this parameter from the baseplate up to the Al trace
layer that causes the compression observed through the
simulation results.
Figure 12. Cross-Sectional X-Tensor Stress Distribution
Through AlN at 3ms
Due to the temperature rise in the module,
each of the layers wants to expand, and the amount of
this expansion is determined by the thermal expansion
coefficient of each layer. Because the baseplate has the
lowest CTE, it naturally wants to expand less than the
layers on top of it, so when the AlN layer tries to
expand further (due to its higher CTE) its movement is
limited by the baseplate with its lower CTE. The
ceramic wants to expand, but because it is bonded
directly to the baseplate its movement is restricted and
therefore the layer experiences a compressive stress.
The AlN also has the Al trace bonded directly to it, and
the much higher CTE of the Al trace makes it want to
expand much more than any of the other layers in the
module. This great urge to expand pulls outward in all
directions on the top surface of the ceramic, resulting in
a slight alleviation of stresses in the area where the two
layers are connected, but also resulting in a greater
compressive stress seen in the surrounding areas of the
ceramic.
Figure 5.5 also demonstrates this behavior by
providing a xy-plane cross-section through the top of
the ceramic layer. Looking at the figure, once can
easily discern the area where the Al trace layer is
bonded to the ceramic by observing the sudden increase
of compressive stresses. One can even make out the
area where the die would be located in the stack-up.
This is due to the interplay of the CTEs between the Al
trace, die attach adhesive, and the aluminum layers.
The generated stresses between these layers are
transferred down into the AlN ceramic layer.
Although most of the module is in
compression, some parts of the module are able to
expand more and experience tensile stresses. The
greatest tensile stress observed in the module is
approximately 62 MPa and is observed in the Al trace
layer and in the SiC MOSFETS. The Al exhibits this
stress in its free edges where it’s allowed to expand the
most. In slightly different behavior, the MOSFETs
experience this stress at the corners of the active region
where the material wants to expand the most and is
causing the dies to want to bulge out, but by an amount
that can only be observed through great magnification.
These two materials have approximate tensile strengths
of 248 MPa and 200 MPa, respectively, so each
material has a factor of safety of at least three during
this mode of operation. The baseplate is the only other
layer to experience any tension, and experiences a
maximum stress of about 10 MPa in the top portion of
the protruding baseplate. The manufacturer of the
baseplate states that it has a tensile strength of
approximately 100 MPa, so it is also well under its limit
due to the fault current conditions.
Although the tensile stresses experienced in
the module are easily manageable, the majority of the
module experiences compression and the magnitude of
these stresses are considerably greater than their tensile
counterparts. The greatest compression in the module
is observed in the AlN ceramic at the boundary where it
is bonded to the aluminum and has a magnitude of
about 382 MPa, which is well under its compressive
strength. The AlN can easily withstand these stresses,
and this behavior was expected as ceramics always
perform well under compression. Also, note that this
stress is greater than the maximum compression seen in
the cross-section provided above. This helps
demonstrate the general trend that the greatest stresses –
compressive or tensile – are observed near or at the
surface where they are bonded to other layers. This
further validates the role of the CTE mismatch between
layers and how important it is to minimize this
mismatch with proper material selection.
The next greatest compression observed in the
module is approximately 360 MPa and is seen in the
aluminum trace layer and the MOSFETs. The
aluminum experiences this compression at the boundary
where the MOSFETs are bonded to it, and even though
the simulations take into account the softening of the
aluminum, the stresses in this layer are still relatively
close to its compressive strength of 530 MPa, resulting
in a factor of safety of about only 1.5. The SiC dies
experience their greatest compression in the center, but
have a compressive strength that is even higher than the
ceramic layer.
Inclusion of Pre-Stress
Although the initial simulations showed that
there is not failure in any of the layers during operation,
these results did not include any previous stress state
that may exist in the module. Stresses may be present
in the module before it is ever put into operation due to
the manufacturing process; these would be known as
the pre-stresses. The first three layers of module – the
baseplate, AlN ceramic, and Al trace – are first
fabricated together, and then the SiC semiconductor is
attached using the silver glass epoxy. The first three
layers are fired together at 700°C and then allowed to
cool down to room temperature in approximately two to
four hours.
To determine the stresses in the module due
this fabrication process, a model was first built that
included only the bottom three layers. Although these
layers are fired at 700°C, they start to set during the
cool down at approximately 660°C, so this was used as
the starting temperature in the simulation. Convection
was applied to all free surfaces of the module, and
because the cool down time is given as a range of two
to four hours, various convection coefficients were used
in order to let the module layers cool down to room
temperature in different lengths of time.
From post-processing these simulations it was
seen that the plots of specific stress tensors
corresponded identically to those from the previous
simulations, but only differed in the magnitudes of the
stresses. Tension was still observed in the free edges of
the module while the greatest compression was seen in
the boundary between the AlN ceramic and Al trace
layers, but the values of these stresses are minimal
compared to those that exist in the module during
operation. The stresses observed in these simulations
are so small in comparison to those seen during
operation that the previous results can be considered
valid and accurate. For example, the compressive
stresses – which are the ones we are most concerned
about during operation – resulting from the fabrication
process are only a fraction of one percent of the stresses
observed in the module during operation.
Through the use of different convection
parameters, it was found that the stresses in these layers
change only minimally according to the cooling rate.
Therefore it can be said that the pre-stresses in the
module do not have a strong second-order dependence
upon the temperature change, but are more dependent
upon the temperature change itself. Table 5.4 below
summarizes these results by comparing the maximum
and minimum stresses observed in the module due to
different cooling rates.
Table 5. Cool Down Times and Stresses
Cooling Convection Maximum Minimum Final
Time Coefficient Stress Stress Temperature
(Hours) (W/m2K) (MPa) (MPa) (°C)
2 5 1.000 -3.703 20.479
2.5 4 0.740 -3.655 20.479
4 2.5 0.713 -3.662 20.479
Harsh Environment
Although the simulations performed up until
this point all considered that the module starts from an
ambient 105°C, it is of great importance to study the
behavior of the module when subjected to very low
temperatures. The lower spectrum of general military
aerospace applications reaches as low as -65°C, so
simulations were performed to study the behavior of the
module when subjected to very low ambient
temperatures. Setting the entire module to room
temperature and applying convective boundary
conditions on all free surfaces in order to allow the
module to cool to the desired temperature did this.
Thermal stress simulations were performed where the
module was subjected to ambient as low as -65°C.
Fig. 13. Cross-Section X-Tensor Stress Distribution Through
AlN at -65°C
From observing the results of these
simulations, it was found that the module has an
inflection point at approximately room temperature. As
temperatures drop below this point, the behavior of the
module ‘flips’ from what it was due to heating. This is
understandable, as generally materials tend to shrink or
contract as they drop in temperature. Also in contrast
to the behavior due to previous conditions, most of the
module is now experiencing tension, which is a result
of the mismatch of thermal expansion coefficients of
the layers. To help demonstrate the behavior of the
module, a cross-section of the x-tensor stress
distribution at -65°C through the AlN layer is shown in
Fig. 13. Using the ceramic layer as an example once
again, it can be understood by explaining that the
contraction of the ceramic layer is limited by the
baseplate because it has a lower CTE than the ceramic.
Because it cannot contract as it would like to it
experiences an initial tensile stress, and then the Al
layer on top pulls inward with even greater force on the
ceramic because it is attempting to contract much more
due to its much higher CTE. This interplay of CTEs
creates the tensile stress observed in the AlN.
Compression is observed only in the free
edges of the semiconductors and the aluminum layer, as
well as in the top of the protruding wall of the
baseplate. The magnitudes of these compressive
stresses increase with lower temperatures with the
greatest compression of about 70 MPa seen in the
aluminum at -65°C. At this temperature, both the
MOSFETs and the baseplate experience compressive
stresses of approximately 10 MPa. The compressive
behavior of these layers at -65°C is not of concern for
the observed stresses are well below their respective
material limits, but unfortunately compression is not the
dominant behavior in the module.
As stated before, most of the module
experiences tension under these harsh conditions, with
the greatest tensile stresses located in the AlN ceramic
where it is bonded to the aluminum trace layer. The
stresses observed in this layer during high temperature
operation were tolerable because they were
compressive and ceramics perform exceptionally well
under this type of loading, but ceramics do not perform
nearly as well under tensile conditions. The maximum
tension observed in the ceramic is 404 MPa, which is
roughly 50% times greater than its tensile strength of
270 MPa. Note that again the maximum stress seen in
the cross-section of this layer provided above is less
than near the surface of the material where it is bonded
to the Al, which again emphasizes the importance of
minimizing CTE mismatch between layers.
The AlN is not the only layer that exceeds its
material’s strength, however, as it is observed that the
MMC baseplate experiences a tensile stress of about
130 MPa. This is approximately 30% greater than the
tensile strength of 103 MPa provided by the
manufacturer. The greatest tension in the aluminum
layer is observed at the boundary where it is bonded to
the ceramic and has a magnitude of about 344 MPa.
The aluminum experiences a tensile stress of about 270
MPa at the boundary where the MOSFETs are bonded,
which also exceeds its tensile strength.
Although most of the layers in the module
experience stresses at -65°C that will cause failure, the
SiC MOSFETs and the silver glass are the two layers
that do not bring up concern. The tension in the middle
of the dies is about 70 MPa, which is about three times
lower than the tensile strength of SiC, and although
material stress limits for the silver glass are unknown
this layer experiences a very uniform tensile stress of
about only 20 MPa.
Although the module cannot survive at the
lowest target temperature, it was still important to
identify the lowest temperature at which the module can
safely exist. Simulations were performed where the
entire module was subjected to ambient temperatures of
0°C, -20°C, and -40°C. From the results of these
simulations it was observed that the module behavior is
always identical to that at -65°C, but the stresses – both
compressive and tensile – increase as the ambient
temperature decreases. Figure 14 below demonstrates
the trend of stresses in each layer according to the
temperature of the module. Note that only the tensile
stresses are included in the figure because even the
greatest compressive stresses observed at any
temperature were well under their material strength and
are not a cause of concern. According to the figure, and
comparing these values to the strength of each material,
it is seen that the module is limited to a temperature of
about -38°C, where at this temperature the AlN layer
reaches its tensile limit and will experience some form
of failure.
Fig. 14. Max Tensile Stresses in Each Layer vs Temp.
Note that the goal of this work is not to
provide usage guidelines for this module, and no
recommendations for what can be considered a safe
ambient temperature will be provided. With no
previous studies performed on the mechanical response
of the module, the goal of this research is to instead
provide a first account of the expected module behavior
under these very low temperatures. Also note that these
simulations represent a non-operational state. When
operating even at the nominal current the
semiconductors will produce heat that will spread
throughout the module and – based on previous
observations – will most likely alleviate the tensile
stresses that are the cause of failure in the module under
these low-temperature conditions. This work does not
include these considerations and only presents the stress
state of the module when in storage or in a non-
operational state.
Conclusion
As the aviation and aerospace industries move
towards more electric aircraft, the demands of the on-
board power distribution systems are consistently
growing. In order to meet these demands, the
development of reliable, fast-interrupting circuit
breakers is essential to continue protecting the wires
and components of these electrical systems. Because
conventional electro-mechanical circuit breakers cannot
meet these requirements, the solid-state power
controller has been developed for fault current
protection of these systems. These devices are able to
detect and interrupt a fault current in nanoseconds,
preventing this current from surging through the rest of
the PDS and damaging the components of the system.
This research presented a SSPC of the highest
power density currently in development, designed to
operate at a nominal current rating of 96ADC and able
to reliably protect against a maximum fault current
rating of 960ADC. This module has been developed in
conjunction with DensePower, LLC, and is able to
operate at such high power densities by being able to
operate up to a junction temperature of 350°C from
105°C. Operation up until this temperature is achieved
through the use of silicon carbide-based (SiC)
MOSFET and a heavily aluminum-based packaging
that promotes steady and efficient heat flow away from
the switching device where it is generated. Similar
work can also focus on GaN devices.
Through simulation work using COMSOL
Multiphysics, the transient thermal and mechanical
behavior of the module was studied, with emphasis on
determining the sustaining time of the MOSFETs
during a fault current and the stresses in the module due
to the expansion of the layers as a result of heating.
From the results of the transient thermal
simulations, it was determined that the module (more
specifically the MOSFETs) can operate for a sustaining
time of 3 milliseconds when subjected to the maximum
rated fault current of 960ADC. Cross-sectional
diagrams of the temperature distribution at 3ms were
used to demonstrate the efficient spread of heat through
the module layers and calculations were performed to
quantify the ability of the module to provide efficient
thermal management. Through these calculations it
was found that approximately 80% of the energy
produced within the active volume of each MOSFET is
dispersed through the module, and this efficient thermal
management is directly responsible for the sustaining
time of the power switch devices. These results
demonstrate the effectiveness of the Al-based
packaging and justify the decision to go with a mono-
material approach to the packaging.
Simulations were also performed to study the
thermo-mechanical behavior of the module under the
same conditions. As the module heats up due to
resistive losses, stresses are produced due to the
mismatch of thermal expansion coefficients between
the layers. All of the stresses observed in the module
layers were under the compressive or tensile limits of
their respective material, with the aluminum layer
causing the greatest concern as it exhibited a factor of
safety of roughly 1.5.
In order to fully understand the behavior of the
module, simulations were performed to study the
mechanical response of the module due to the
fabrication process, where the first three layers are fired
together at approximately 700°C and allowed to cool
down to room temperature. It was determined that the
pre-stresses in the module layers are not very sensitive
to the cooling rate, and were found to be well under the
limits of their respective materials. These stresses are
low enough to be admissible and the results from the
transient simulations can be considered accurate and
credible.
Lastly, the entire module was cooled down to
temperatures as low as -65°C so that the behavior of the
layers could be studied at this military application
range. Under these conditions of very low ambient
temperatures tension dominated the module and limited
its usage to a temperature of about -38°C. Tensile
stress versus temperature data was provided for each
layer in a graph so that the behavior of the module in
response to low temperatures could be easily
understood. It was observed that the AlN ceramic
experienced the greatest stresses and would be the first
to failure as the ambient temperature decreased.
This research was a necessary step in the
development of a solid-state power controller. No
previous work has investigated the behavior of a
complete SSPC module, especially in response to its
true operating conditions, namely the resistive losses
resulting from a very high fault current. While much
more work can be performed on this particular module,
this research was important in that it established a
precedent for a solid-state power controller that can
operate to a very high junction temperature at an
extremely high power density. This work provided
some of the very first characterization of a full SSPC
module by documenting its thermal management
capabilities and the resultant sustaining time of its
power switching semiconductors. This work also
 characterized the mechanical behavior of the module by
documenting the stresses generated in the module under
various conditions. This allowed for identification of
areas in the module that are most susceptible to failure
and for determination of the lowest temperatures that
this module may be subjected.
Acknowledgment
The authors acknowledge the work done by
Troy McKay providing extensive simulation of Cu and
Al systems.
References
[1] D. Izquierdo, A. Barrado, C. Raga, M. Sanz, P.
Zumel, A. Lazaro, “Protection devices for aircraft
electrical power distribution systems: a survey”, in
Proc. IECON, pp. 903 – 908, Nov. 2008
[2] S. N. Friedman, “Solid-state Power Controller for
the Next Generation”, in Proc. NAECON, vol. 1, pp.
155 – 162, 1992
[3] S. Friedman, “Robust/low cost SSPCs are finally
available”, in Proc. NAECON, vol. 1,pp.145–150, 1997
[4] J. J. Simon, “State of the art in solid state power
controller for aeronautical application”, Eur. Conf.
Power Electronics and Applications, vol. 8, pp. 121 –
126, 1993
[5] R. Y. Kinoshita, “Solid State DC power switch”,
U.S. Patent 4709160, Nov. 24, 1987
[6] J. R. Wilson, (2002, July). End of the Circuit
Breaker: Smart Power Makes its Mark, Military
Aerospace Electronics, vol. 13, issue 7
View publication stats
[7] M. Thek, (2002, Sept.). Leach International
Pioneerd SSPC Development, Military Aerospace
Electronics, vol. 13, issue 9
[8] D. S. Hart, G. B. Prickett, C. R. Schwarz, M.
Mohadjer, “A Solid State Power Controller Module for
the International Space Station Express Rack”, in Proc.
IECEC, 1997, vol. 1, pp. 280 – 285
[9] Wei Liu, Alex Q. Huang, “A Novel High Current
Solid State Power Controller”, in IECON’2005. 31st
Annual Conference of IEEE, Nov. 2005,pp.1306–1310
[10] M.M.R. Ahmed, P.A. Mawby, “Design
specification of a 270V 100A solid-state power
controller suitable for aerospace applications”, In 13th
European Conference on Power Electronics and App.
2009, Sept. 2009, pp.1 – 8
[11] X. Feng, A. V. Radun, “SiC based solid state
power controller”, in Proc. APEC, pp.1855–1860, 2008
[12] L. J. Feiner, “Power Electronics for transport
aircraft applications”, in Proc. IECON, vol. 2, pp. 719 –
724, 1993
[13] D. C. Hopkins, J. M. Pitarresi, and J. A. Karker,
"Thermal Impedance and Stress in a Power Package
Due to Variations in Layer Thickness," Spie Int’l Soc
for Optical, 1997, pp. 72-77
[14] D C Hopkins, Yuan-Bo Guo, H E Dwyer J D.
Scofield, “Development of a SiC SSPC Module with
Advanced High Temperature Packaging,” HiTEC’10,
Albuquerque, NM 10-13, May 2010
[15] T. B. Baltis, Transient Thermo-Mechanical
Investigation of a SiC-Based Solid-State Power
Controller, Masters Thesis, Binghamton University,
May 2011.