Thomas Schreier-Alt , Frank Rehme , Frank Ansorge , Herbert Reichl
Fraunhofer IZM, Micromechatronic Systems, Argelsrieder Feld 6, 82234 Oberpfaffenhofen, Germany
EPCOS AG, Anzingerstr. 11, 81671 Munich, Germany
Technische Universität Berlin, Fakultät IV – Elektrotechnik und Informatik, Gustav–Meyer–Allee 25, 13355 Berlin, Germany
Abstract
This paper presents new experimental and numerical methods to characterize the transfer overmolding
of substrates with epoxy polymer. We investigated Multi Chip Modules on ceramic panels as well as on
printed circuit boards encapsulated as a Mold Array Package (MAP). Experiments show that the polymer
flow during the overmolding process depends significantly on the mold height: While standard MAP-type
mold cavities are filled homogeneously and symmetrically in most cases, low cavity heights (<500 lm)
can cause the flow front to concentrate on a few flow paths (flow front fingering). We developed a numerical
method to describe this inhomogeneous polymer flow. The reason for flow front fingering seems to
be local variations of polymer viscosity which enforces a necking on distinct flow paths. Fingering can
cause the formation of air traps and excessive wire sweep. We also developed new experimental methods
to measure the pressure distribution within cavities: our sensor is based on commercially available, passive
pressure sensitive films from FUJUFILM and is operational at temperatures up to 180°.
1. Introduction: MAP-type overmolding
Transfer molding with epoxy molding compound (EMC) is a
standard process for reliable chip encapsulation since decades.
Currently the market trends clearly drive towards System in Package
(SIP) [1]. They allow much higher component integration rates
compared to classical single chip technologies. Additionally they
have significant advantages for the manufacturers as they can offer
complete functional modules to the OEM instead of single components,
increasing the added value and uniqueness of their product.
Its main advantage over pure front-end technologies is its flexibility
of design by integrating discrete passive and active components
within one package. Modules for mobile phones (Fig. 1) can integrate
more than 50 components in a single package. In comparison
with conventional solutions based on discrete components, the SIP
solution reduces board space requirements by 95%.
Fig. 1. Front-end module for mobile phones before encapsulation to a system in package.
One upcoming encapsulation standard for SIP can be identified
to be the substrate overmolding process. The substrate to be overmolded
typically is a Printed Circuit Board (PCB), a ceramic panel, a
leadframe or a wafer [2,3]. The substrate consists of several electronic
modules arranged in a MAP-type configuration. Each of
them is composed of numerous passive or active components.
The assembled surface of the substrate is encapsulated by EMC
and singulated by dicing. A simple tool for substrate overmolding
and an overmolded PCB are pictured in Fig. 2. Details of molding
tool and process are given at the end of this section.
Fig. 2. Left: Molding steel tool. Label 2: Inlet for EMC pellet. Label 3: Film gate with
shaping structures. Label 4: PZT cavity pressure sensor. Right: Overmolded FR-4
substrate with film gate. Label 1: Overmolded FR-4 substrate. Label 5: Mark of
pressure sensor surface within EMC.
Main advantages of these Mold Array Package (MAP-type)
geometries for production are the following:
- One single mold tool can be used for multiple module dimensions
as long as their height remains.
- Length of polymer flow can be minimized because runners for
material supply can be omitted.
- Increased integration density of components within the mold
tool leading to reduced packaging costs.
Typically the singulated modules are Quad Flat No Leads (QFN)
or ball less package types. A very promising technology derived
from ball less MAP-type molding is the embedded Wafer Level
BGA (eWLB) approach [4,5]. Both encapsulation technologies,
MAP-type molding as well as eWLB, combine the overmolding of
large areas (Ø= 25–100 mm) with a relatively thin mold cover
(height < 1 mm).
The MAP-type overmolding process of PCB substrates consists
of several production steps: A substrate (labeled "1" in Fig. 2)
and an epoxy pellet (inlet for pellet labeled "2") are inserted into
the mold tool (step A in Fig. 3). Both must be pre-heated, to
guarantee initial plasticity and flowability of the pellet. Pre-heating
of the substrate increases thermal uniformity and adhesion between
substrate and polymer. The pellet is crushed by a plunger,
increasing thermal contact to the hot tool and melting (step B).
After this preconditioning the plunger presses the polymer through
the film gate and into the cavity on top of the PCB (step C). Specially
designed shaping structures (labeled "3") can be used to
achieve sufficient shear heating within the EMC and a homogeneous
flow front at the gate. After filling of the cavity the EMC is
pre-cured for about a minute to guarantee sufficient gelation and
therefore adequate contour accuracy during removal of the substrate
(step D). In most applications the parts are post-cured afterwards
at the mold temperature for several hours. Our mold tool
was equipped with a PZT pressure sensor from Kistler (labeled
"4"). Small marks of the sensor can be seen on the overmolded substrate
(labeled "5").
Fig. 3. Transfer molding process with central injection of the molding compound. Step A: Pellet and substrate are inserted into the mold tool and preheated. The substrate can be equipped with electronic components or a pressure sensitive film. Step B: The plunger presses on the pellet, the pellet melts. Step C: The melted polymer is pressed into the cavity. After filling the cavity the material is pre-cured for 2 min. Step D: The plunger is removed together with the cull. The substrate is removed.
The integration of sensors and actuators within MAP-type modules
narrows the molding process window significantly, because
their pressure sensitivity is much higher compared with monolithic
silicon chips or passive components. Often the stress during
the mold filling process can be neglected, considering the electronic
part to be stress free at mold temperature. Consequently
the simulation of thermo-mechanical stresses is restricted to temperature
changes and subsequent failures like delaminations and
popcorning [6,7]. Some research groups started to implement polymerization
stresses into their models in order to investigate substrate
warpage after molding [8–10]. Nevertheless there is the
need to monitor also the pressure acting on the electronic parts
during the polymer encapsulation process considering e.g. the
spreading usage of MEMS sensors within automotive industry
and their increased demand for signal stability, production tolerances
and cost [11–13].
2. Simulation: Polymer flow during MAP-type overmolding
Numerical simulation of the polymer flow within MAP-type
cavities is until now a challenging task for a process engineer. First
the large amount of small gaps between electronic parts, mold wall
and PCB has to be meshed precisely, heavily increasing the FEM
computation time. Secondly, the filler particles within the polymer
(up to 90 wt.%) can increase the flow resistance of the epoxy within
small gaps and cause separation of filler and polymer matrix. Non
uniform filler distribution can be a cause of failures such as cracks
[14].
Commercially available tools for simulation of transfer molding
are – as far as we know – not able to predict the filling behavior of
polymers within thin (<500 μm), but wide-stretching gaps
(>40 mm) precisely. Chen et al. [15] have observed viscosities of
unfilled polymers within 200 μm wide gaps differing up to 75%
from standard FEM predictions. Simulations mostly show a more
or less symmetrical flow pattern around the gate which does not
change significantly if one of the following parameters changes:
filler degree, viscosity η, injection pressure p, temperature of the
melted epoxy T or cavity height h. This fact can be seen in Fig. 4
where velocity distribution derived by standard FEM tools within
gaps of 250 μm and 1000 μm thickness do not differ significantly
from the Newtonian flow derived by Poiseuille’s theory. Fig. 5
shows that standard FEM simulations also show no difference in
flow patterns when the mold height is reduced from 750 μm to
450 μm.
Fig. 4. Numerical (Moldflow MPI 3D)and analytical (Poiseuille) calculation of
polymer velocity within gaps of 250 μm and 1000 μm.
Fig. 5. Standard simulation of polymer velocities on substrates with a mold cavity height of 750 μm (left) and 450 μm (middle). No significant difference between the flow fronts can be observed. The picture on the right shows the polymer velocity after modification of the FEA tool by using a corrected viscosity η* according to formula (2).
The governing laws of polymer flow used by FEM models (Reactive
Cross-Arrhenius-Macosco, Herschel-Bulkley-WLF) show no
correction factors when narrow gaps are analyzed. Standard models
without correction factors get definitely inaccurate if gap sizes
are of the same dimension as the filler particles. The effect of temperature
and filler degree on viscosity is theoretically described
since the works of Einstein [16] as well as Ball and Richmond
[17]. Our experiments indicate that the relation k between maximum
filler particle size d and gap size h is of significant
influence on flow resistance. λ will be defined as:
To take into account the maximum filler size of a typical EMC
(with 75–90 wt.% filler particles) we defined a corrected viscosity
η*. It depends on the geometrical parameter λ (described by a correction
factor C) and material parameters typically used within
Herschel-Bulkley-WLF or reactive viscosity models (temperature
T, shear rate ý, pressure p and fitting constants).
The correction factor C was derived by fitting numerical simulations
to our experimental results that will be described within Section
5. Until λ = 0.4 there is no significant difference between
theories only considering interactions between particles and walls,
e.g. the Bungay and Brenner model [18], and our fit. However for
λ > 0.5 we get a much faster increase of viscosity with increasing
filler to gap ratio (Fig. 6). The reasons are interaction forces between
filler particles that have to be considered within highly filled
epoxy molding compounds. These interactions increase flow resistance
in narrow gaps much more than in wide cavities, especially
for λ > 0.5 when only one maximum sized filler particle after the
other can enter the gap and the whole filler distribution within
the polymer matrix has to be rearranged.
3. Measurement of pressure distribution with sensitive films
Fig. 6. Increase of C with increasing relation of maximum filler size against gap size.
In order to examine polymer flow behavior we measured the
pressure within the cavity online by PZT pressure transducers
mounted flushing with the mold cavity wall, chronologically
recording pressure values at the measurement spot (labeled "4"
in Fig. 2). Disadvantages of this sensor are the missing ability to record
an areal pressure distribution and the occurrence of unattractive
marks on the part’s surface (labeled "5" in Fig. 2). As the
sensor’s mark can be even larger than one single singulated module,
their usage within MAP-type transfer molding is rarely accepted.
Another disadvantage of active sensors – electrical as
well as optical – is that the signal transmission requires holes or
grooves within the metal tool. This is undesirable, especially if
the measurement should take place during the developmental
stage only.
To overcome these disadvantages of conventional pressure sensors,
a PrescaleMWpressure sensitive film (sensitivity 10–50 MPa)
from FUJIFILM [19] has been modified to enable its usage within
transfer molding. Pressure sensitive films are based on a PET matrix
that contains small microcapsules. Pressure upon the film
causes the microcapsules to crack, starting a chemical reaction that
is producing an instantaneous and permanent color change across
the contact area. Pressure sensitive films record only the maximum
pressure value applied on the film and are pressure sensitive up to
80 °C. To overcome the temperature restriction, we appended an
isolating polyimide coating and an adhesive layer onto the film. We
mounted this multilayer on top of the PCB in such a way that the
liquid molding compound is flowing over the multilayer (Fig. 3).
As the softening temperature of the PET resin within the Prescale
film is 80 °C, the additional layers have to guarantee that the cavity
pressure is almost entirely transferred to the pressure sensitive
particles within the Prescale film. For examination of the color
change, the substrate can simply be separated from the adhesive
film of the multilayer (Fig. 7A). Also the adhesion between EMC
and PI film is weak enabling their separation. The color intensities
of the separated pressure sensitive film is analyzed by a commercial
scanner and transferred into a color spectrum by software.
With our multilayer film we can enable test durations at 180 °C
of several minutes without detraction of the sensor function.
The first process under investigation is an FR-4 board overmolded
with EMC through a film gate as shown in Fig. 2. The polymer
melt has been injected directly onto a Prescale-IZM-multilayer
mounted onto the PCB within the cavity. The color distribution
on the FR-4 board (Fig. 7B) is very homogeneous with a slight pressure
drop with increasing distance from the gate. The pressure distribution
is in accordance with numerical simulations, which
expect slightly lower values than measured (~10% deviation).
The pressure values captured by the multifilm measurement
technique have been compared with values from piezoelectric
pressure sensors mounted within the same mold tool. Apart from
FR-4 substrates we tested the technology with ceramic substrates
(Fig. 7C) and used different transfer molding machines with different
positions of the PZT sensor. We derived a semi-empirical formula
that describes the sensitivity S of the pressure sensitive
film at temperature T and pressure p. For transfer molding with
T ≈ 175 °C and p ≈ 5.7–12.5 MPa the following linearization was
used for pressure derivation.
Fig. 7. A: PCB with imprints of the mold tool and removed EMC with pressure sensitive film. B: Pressure distribution of a film gate by simulation (left) and sensitive film. C:
Pressure distribution of a central pin gate by pressure film and short shot experiment (Fujifilm Prescale data analysis by "Fujifilm Pressure Distribution Mapping System FPD-8010E").
where p is the pressure measured by the Prescale film and p is the
real pressure, e.g. recorded by the PZT sensor. T(RH), ΔT(T) and Δp(p, T) are material specific values derived from experiments. The measured values of p do not necessarily represent the maximum pressure distribution at the beginning of the post pressure as the
sensitivity S depends on temperature which can change during
the transfer molding process. Therefore the time dependent thermal
behavior on the substrate’s surface should be regarded. Numerical
simulations are advantageous because they also give a rough value
for the expected pressures and simplify the correct choice of Prescale
film. In most cases, with a proper pre-heating of the substrate
and the polymer pellet, the temperature remains constant and the
Prescale color pattern will represent the maximum pressure
distribution.
The maximum pressure values have been varied between
5.7 MPa and 12.5 MPa. The measurement values of the PZT pressure
sensors are displayed in Table 1 and compared with the multilayer
film values. The percentage values refer to the multilayer
values in relation to the PZT sensor. We found both values to agree
within ±10%, which is also in accordance with the finite element
analyses performed by Autodesk Moldflow 3D.
Table 1
Relation of cavity pressure values recorded during transfer molding by multilayer film "p (film)" in relation to values of standard PZT pressure sensors from Kistler "p (PZT)".
In order to control the pressure distribution during transfer
molding also within mass production, the authors invented a
new in-mold measurement method. Basic principle of the invention
is to monitor the polymer pressure on the substrate by subjacent
fiber optic Bragg gratings (patent pending). The wavelength of
the light reflected at the grating depends linearly on the molding
pressure (Fig. 8). As it also depends on the temperature, methods
for temperature compensations have to be taken [11]. Main advantage
of the new sensor is its ease of integration and the time
dependent measurement of the areal distribution by the fiber optic
sensor network (multiplexing of dozens of Bragg sensors within
one fiber).
Fig. 8. Wavelength shift of a fiber optic pressure sensor with increasing molding pressure.
4. Experiments: MAP-type overmolding of FR-4 substrates
The flow behavior of polymers on Multi Chip Module (MCM)
substrates is far more complicated than the classical single chip
encapsulation. Even on small FR-4 substrates (edge length:
25 mm) without any assembled components surprising results
can be observed. Fig. 9 shows the progress of EMC during substrate
overmolding by subsequent short shots. They have been produced
by increasing the amount of EMC material within the plunger step
by step. The flow front splits into two separate flow paths shortly
after injection into the cavity (state A). First, the material flows uniquely
along the edges of the cavity. Then an additional flow path
starts in the centre of the substrate (B and C), but keeps sharply
separated from the boundary flow paths. They reach the PCB corner
opposite of the gate slightly before the central flow path and
continue their way along the cavity border (D). If the cavity is
not evacuated the enclosure of air traps is unavoidable (E).
Fig. 9. The filling behavior of EMC on an FR-4 substrate without components can be surprisingly complicated which can be illustrated by short shots.
The reason for this behavior is rooted in the reduced stiffness of
the substrate above its glass transition temperature. As the coefficient
of thermal expansion (CTE) is rising sharply above Tg even
small temperature differences between the mold tool (~175 °C)
and the PCB cause a significant elongation of the substrate after
closing of the mold tool. In this case the substrate is mechanically
fixed between the halves of the mold tool and the elongation of the
substrate leads to a convex, ∩-shaped bending and forces the polymer
to flow around this central elevation. With increasing pressure
at the gate and further increasing temperature of the substrate,
buckling of the PCB starts in the middle leading to a ∩∩-shaped
substrate with three distinct flow paths. This curvature of the substrate
can still be measured after ejection, because its shape is frozen
by the cured epoxy polymer. If the experiment is conducted
with a substrate assembled with electronic components, the warpage
is suppressed by the components (Fig. 10). They work as spacers
between PCB and mold tool and are preventing the soft
substrate to buckle by mechanically touching the upper mold tool.
This can be demonstrated by short shots: the highest components
on the board are not covered by EMC but exposed.
Fig. 10. Left: Short shot of PCB overmolding with assembled components. Right: PCB with reduced substrate thickness shows homogeneous, straight flow front.
A reliable way to prevent the buckling of the PCB is to use substrates
with reduced thickness. Without being rigidly clamped between
the two mold tools the substrate can slightly slip through
them during elongation. The disadvantage of this method is the
possibility of polymer flash (Fig. 10).
5. Discussion: Comparison between experiments and simulation of MAP-type overmolding
Our experiments show another important effect during MAPtype
overmolding that is not predicted correctly by standard
numerical simulation software, because an appropriate flow model
is not implemented up to now. Overmolding with a film gate produces
a straight flow front that is moving over the panel. We have
analyzed this for flow path lengths up to 40 mm. Experiments with
evacuated molds with pin gates show that the straight flow front
starts to ripple after ~20 mm (Fig. 11). These flow front instabilities
can also be observed on plane substrates without any assembled
components, which is especially important for overmolding
of large wafers (Fig. 12). Table 2 gives dimensions of the cavities
used within the experiments.
Fig. 11. Flow of highly filled EMC over an assembled ceramic panel, injected with a pin gate.
Fig. 12. Short shots with an EMC with filler cut 53 μm (left) and with filler cut 45 μm (right). Different intensities of flow front fingering can be observed. Both polymers had a filler content of 90 wt.%.
Table 2
Geometry of the cavities.
This behavior is called ‘‘flow front fingering instability” within
the literature. Within filled polymers, this effect can be caused
by local variations in filler concentration (e.g. agglomerations) that
locally change the viscosity [20]. The effect has also been reported
with unfilled shear thinning viscoelastic fluids [21]. Here, the flow
front should get smoother if the filling velocity is lowered in such a
way that the polymer is always in the Newtonian range. Without
the shear thinning effect no inhomogeneities should appear. Because
of the fast polymerization of our material, we were not able
to lower injection speed sufficiently to observe this effect. Flow
front fingering can also be observed when the advancing flow front
is increasing in length, e.g. when spreading symmetrically from a
pin gate. The occurrence of this effect seems to be linked to shear
stresses perpendicular to the flow direction [21]. If a pin gate is
used, the velocity of the melt decreases monotonously with
increasing distance from the gate. Consequently the shear rate at
the spreading flow front decreases steadily until the Newtonian regime
is reached (if applicable), where shear rate will be independent
from the polymer’s viscosity (Fig. 13). Until this point small
variations in shear rate ý can change the viscosity of the material.
This shear thinning effect is typically used when the polymer is
forced through the narrow mold gate. A similar effect can happen
when the material is forced to flow through narrow gaps between
components. This locally increased shear rate increases flowability
within these regions and leads to a necking of the flow to few distinct
paths with reduced viscosity. Consequently the circular flow
front splits up into several flow paths. Little random variations of
polymer velocity seem to be enough to break the symmetry of
the filling behavior. At the contact area of the separate flow paths a
weld line is formed where unfilled voids remain.
Fig. 13. Viscosity of ShinEtsu KMC-184 with Newtonian regime y < 5 s and non-Newtonian regime y > 5 s.
The inhomogeneous filling behavior has been observed in short
shots, where we stopped the overmolding process during the flow
to inspect the flow front. It has to be mentioned that the filling pattern
of a short shot might change after the injection pressure is removed.
However, we were able to correlate an inhomogeneous
polymer flow with the occurrence of small voids (diameter
>200 μm on the polymer surface, examined by visual inspection)
of completely overmolded substrates.
It could be demonstrated, that the following parameters
contribute to the occurrence of flow front fingering instabilities:
- Decreasing thickness of overmolding.
- Decreasing distance between electronic parts and between parts and mold wall.
- Increasing viscosity of the polymer.
- Increasing size of the filler particles.
This fingering of the flow front during transfer molding can not
easily be implemented into standard transfer molding simulation
tools. Based on the thixotropic index [22], we numerically adjusted
the material behavior by a thixotropic shear thinning parameter
that is highest at the gate location and monotonically drops with
proceeding time. Not only could the detailed form of the flow front
be predicted, but also the locally reduced viscosity within a few
preferred flow paths between the components (Fig. 14). At the
beginning of overmolding the regions with slow material movements
are additionally provided with molten resin by the neighboring
preferred flow paths, but the occurrence of these shear
thinned regions seems to be already the forerunner of an asymmetrical
filling behavior where regions far away form the flow
paths remain unfilled.
Fig. 14. Simulation of filling behavior on a substrate with successively increasing filler/gap relation (left: filler cut 45 μm with silicone additives, middle: filler cut 53 μm without additions, right: substrate with components and locally reduced gap sizes).
6. Summary
Homogeneous material flow and pressure distribution within
the cavity is of particular importance for encapsulation of mechanically
sensitive parts, e.g. MEMS. With the use of pressure sensitive
films a fast and inexpensive estimation of the pressure distribution
can be achieved. Our multilayer film design is based on commercially
available color changing papers (e.g. Prescale), an adhesive
layer and an isolating film. This assembly enables tests for several
minutes at 180 °C without detraction of the sensor function. The
accuracy of this pressure measurement method is ~10%, checked
by conventional pressure sensors located at different positions
within the mold walls. The pressure results are also fitting well
with numerical simulations.
By using a standard molding process, the polymer melt has
been injected directly on this multilayer, generating red color
changes on the Prescale film depending on the local maximum
pressure value. Also the contact pressure between solder pads
and bottom mold wall could be measured with the pads emerging
clearly from the flat PCB surface.
We could clearly distinguish two different kinds of flow behavior
during MAP-type overmolding: 1. Homogeneous filling of the
cavity with a closed flow front and a continuous pressure drop
along the flow path. 2. Contraction of the polymer flow within
few, narrow paths with increased shear rate, reduced viscosity
and excessive velocity.
Additionally we could show that modifications of the computational
fluid simulation tool increase the accuracy of the flow
behavior prediction, especially of the filling patterns with concentrated
flow paths. The models have been validated with short shots
within the transfer molding machine.
Acknowledgement
We would like to thank Ralf Andussies from FUJIFILM Recording
Media GmbH for valuable support concerning measurement strategies
at elevated temperatures. The authors wish to thank VDI/
VDE-IT, especially R. Schliesser for their financial support.
References
- Ardebili H, Pecht MG. Encapsulation technologies for electronic
applications. Elsevier; 2009.
- Marenco N et al. Investigation of key technologies for system-in-package
integration of inertial mems. In: Symposium on design, test, integration and
packaging of MEMS/MOEMS; Rome, Italy: 1–3 April, 2009.
- Falat T et al. The influence of process parameters and materials properties on
stress distribution in MEMS – ASIC integrated systems after molding –
numerical and experimental approach. In: International conference on
thermal, mechanical and multi-physics simulation and experiments in
micro-electronics and micro-systems, EuroSimE 2009, Delft, The
Netherlands; 2009.
- Plieninger R, Dittes M, Pressel K. Modern IC packaging trends and their
reliability implications. Microelectron Reliab 2006;46(9–11):1868–73.
- Meyer T, Ofner G, Bradl S, Brunnbauer M, Hagen R. Embedded wafer level ball
grid array (eWLB). In: Proceedings of the 2008 electronic packaging technology
conference (EPTC), Singapore, December 9–12; 2008. p. 994–8, .
- Lall P, Pecht M, Hakim EB. Characterization of functional relationship between
temperature and microelectronic reliability. Microelectron Reliab
1995;35(3):377–402.
- Gannamani R, Pecht M. An experimental study of popcorning in plastic
encapsulated microcircuits. IEEE Trans Compon, Packag, Manuf Technol – A
1996;19(2):194–201.
- Jansen KMB et al. Modeling and characterization of molding compound
properties during cure. Microelectron Reliab 2009;49(8):872–6.
- de Vreugd J, Jansen KMB, Ernst LJ, Bohm C. Prediction of cure induced warpage
of micro-electronic products. Microelectron Reliab 2010;50(9):910–6.
- Kim YK, Park IS, Choi J. Warpage mechanism analyses of strip panel type PBGA
chip packaging. Microelectron Reliab 2010;50(9):398–406.
- Schreier-Alt T, Ansorge F. Process optimization: throwing light on electronics
encapsulation. Kunstst Int 2009;99(6).
- Müller-Fiedler R, Knobloch V. Reliability aspects of microsensors and micromechatronic
actuators for automotive applications. Microelectron Reliab
2003;43:1085–91.
- Feiertag G, Winter M, Leidl A. Packaging of MEMS microphones. In:
Proceedings of the 2009 SPIE Europe international symposium on
microtechnologies for the new millennium, Dresden; May 4–6, 2009.
- Huang Y, Bigio D, Pecht MG. Fill pattern and particle distribution of underfill
material. IEEE Trans Compon Packag Technol 2004;27(3):493–8.
- Chen SC et al. Preliminary study of polymer melt rheological behavior flowing
through micro-channels. Int Commun Heat Mass Transfer 2005;32:501–10.
- Einstein A. Eine neue Bestimmung der Moleküldimensionen. Annalen der
Physik 1906;324(2):289–306.
- Ball RC, Richmond P. Dynamics of colloidal dispersions. Phys Chem Liq
1980;9(2):99–116.
- Bungay PM, Brenner H. The motion of a closely fitting sphere in a fluid-filled
tube. Int J Multiphase Flow 1973;1(1):25–56.
- http://www.fujifilm-prescale.eu/.
- Becker KF, Vollbrecht H, Braun T, et al. Basic design guidelines for a robust
underfilling process. In: Proceedings of the international acousto microscopic
imaging symposium – IAMIS 2000, 19–20.10.00, San Jose, CA, US.
- Kabanemi K, Hétu JF. Elastic flow-front fingering instability in flowing polymer
solutions. Rheol Acta 2006;45(5):693–704.
- Okuno A et al. Properties and reliability of liquid epoxy resin for LSI fabrication
with the printing encapsulation system (PES). Electron Commun Jpn, 2 1997;80(2):431–9.
|