Sei sulla pagina 1di 10

Multi-scale analysis of polysilicon MEMS sensors subject to accidental

drops: Effect of packaging


Aldo Ghisi
a
, Fabio Fachin
b
, Stefano Mariani
a,
*
, Sarah Zerbini
c
a
Dipartimento di Ingegneria Strutturale, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milano, Italy
b
Technology Laboratory for Advanced Materials and Structures, Dept. of Aeronautics and Astronautics, Massachusetts Institute of Technology,
77 Massachusetts Avenue, Cambridge, MA 02139-4307, USA
c
MEMS Product Division, STMicroelectronics Via Tolomeo 1, 20010 Cornaredo, Italy
a r t i c l e i n f o
Article history:
Received 4 December 2007
Received in revised form 5 November 2008
Available online 7 February 2009
Keywords:
Polysilicon MEMS
Accidental drops
Multi-scale approach
Finite element analysis
Packaging
a b s t r a c t
The effect of packaging on the impact-carrying capacity of micro electro-mechanical systems (MEMS) is
investigated, with specic reference to a translational accelerometer. By exploiting the small ratio
between the masses of MEMS and package/die (typically 10
3
or less) a decoupled two-scale, nite ele-
ment approach is adopted: at the package/die length-scale the dynamics of whole device after the impact
against a at target surface is studied; at the sensor length-scale the response of the MEMS to the drop-
induced loading is investigated, and MEMS details where the stress state can exceed the tensile strength
of polysilicon are identied. Two drop orientations are considered, here termed bottom and top; in the
rst case, package and die strike the target with their bottom surfaces; in the second case, they fall
upside-down, and strike the target with their top surfaces.
By comparing the simulation outcomes in terms of maximum attained tensile stress, it results that
package does not always lead to benets in term of capability of the studied sensor to sustain drops.
In the bottom drop conguration, e.g. MEMS failure may be triggered by the package.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
The effect of impacts on microelectronics have been recently
investigated focusing on instrumented boards or solder joints,
see e.g. [13]. As for the mechanics of MEMS subject to accidental
drops, a smaller amount of experimental and numerical researches
has been published, see e.g. [47]. In [8] we showed that a numer-
ical estimation of the impact-carrying capacity of these devices is
in need of tools able to accurately describe how loading at the
package length-scale (here termed macro-scale) affects the failure
mechanisms at the sensor length-scale (here termed meso-scale).
Because of the difference between the characteristic sizes of a
MEMS and of the whole package, a multi-scale approach is re-
quired to assure the required accuracy of the results.
A further length-scale (here termed micro-scale) should be con-
sidered if a detailed micro-scale description of the damaging/
cracking phenomena at polysilicon level has to be furnished.
According to the methodology proposed in [8], we avoid running
micro-scale analyses; local failure is assumed to take place every-
where the stress eld, as obtained at the meso-scale, exceeds the
polysilicon strength. Additional details and insights into links of
forthcoming outcomes with the stress eld at the micro-scale
can be found in [9].
In an industrial environment the MEMS response to shocks is
customarily deduced from vibration tests featuring high accelera-
tion peaks, or by dropping the device from a reference height
[10,11]. After the test, eventual variations of the working frequency
or malfunctioning of the device are adopted as evidence of a
mechanical failure of the sensor, judged on the basis of a pass/
not pass classication. Quality of information collected with this
test protocol is too poor from our perspective, since the failure
mechanisms cannot be argued at all. A predictive model, either
analytical or numerical, seems necessary to get insights into the ac-
tual sensor impact-carrying capacity.
To model impacts against a massive target body, the accelera-
tion experienced by the device while bouncing off can be adopted
of a half-sine waveform (see e.g. [4,12,13]). Amplitude of the accel-
eration pulse as well as pulse duration are usually assumed known.
This approach can be adopted if failure occurs in device compo-
nents at the same length-scale of the whole device, like for
instance in the case of solder joints. When failure occurs at
length-scales smaller than the device one by orders of magnitude,
this approach leads to an over-simplication of the loading condi-
tions: the complex interactions among stress waves propagating
inside the device, sensor dynamics and micro-mechanics driven
failure modes are neglected [9]. In Section 4 we will show that
0026-2714/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.microrel.2008.12.010
* Corresponding author. Tel.: +39 02 23994279; fax: +39 02 23994220.
E-mail addresses: aldo.ghisi@polimi.it (A. Ghisi), ffachin@mit.edu (F. Fachin),
stefano.mariani@polimi.it (S. Mariani), sarah.zerbini@st.com (S. Zerbini).
Microelectronics Reliability 49 (2009) 340349
Contents lists available at ScienceDirect
Microelectronics Reliability
j our nal homepage: www. el sevi er . com/ l ocat e/ mi cr or el
the features (like e.g. the amplitude) of the aforementioned accel-
eration pulse might not be an objective indicator of MEMS failure,
since failure is actually linked to the local stress eld while accel-
eration is dened at the device scale (see also [14]).
The effect of packaging on the impact-carrying capacity of a
translational polysilicon MEMS accelerometer is here assessed by
investigating two device congurations: in the rst one the MEMS
is assumed to be already packaged when it experiences the acci-
dental drop, whereas in the second one the naked die is assumed
to fall. Furthermore, two possible falling orientations are envisaged
(see Fig. 1). The rst one (termed bottom drop conguration) is
characterized by the package falling with the layered support plate
downwards; before impinging upon the sensor anchor, the result-
ing stress wave gets dispersed by inner surfaces. The second one
(termed top drop conguration) is characterized by the package
falling upside-down, with the mold compound downwards; in this
case, before arriving to the sensor anchor, the stress wave has to
propagate inside the compliant mold phase. Since the paths fol-
lowed inside the package and the materials traveled across by
the stress waves are very different in the two mentioned cases, it
is expected that also the way the sensor is loaded will differ in
the two explored congurations.
To assure accuracy of the numerical results at low computa-
tional costs, the multi-scale framework offered in [8,9] has been
adopted. At the macro-scale the dynamics of the whole package
or of the naked die after the impact is modeled. At the meso-scale
the response of the sensor to the shock loading is simulated, and
the MEMS details where the stress state tends to get increased
are identied. The link between the analyses at the two length-
scales is represented by the displacement histories at sensor an-
chors, which are continuously stored at the macro-scale and later
used as loading/boundary conditions at the meso-scale. Account
taking of the brittleness of polysilicon at room temperature, MEMS
failure can then be judged by comparing the obtained maximum
principal stress eld and the polysilicon tensile strength [8].
Capabilities of this multi-scale framework to describe counter-
intuitive experimental evidences have been already proved in [8],
and further extended here. Specically, it will be shown that in
case of a bottom drop from an height h
drop
150 cm the amplitude
of the stress eld in the suspension springs gets increased by the
package. This means that, according to a local failure criterion for
polysilicon, failure of the whole MEMS more likely occurs in the
packaged case than in the bare die one. It will be also shown that
this results matches available experimental data obtained through
laboratory drop tests.
It is worth emphasizing that in this work we focus on the
assessment of the effect of drops on a MEMS already available on
the market, without addressing the interesting topic of shape opti-
mization of the failing regions so as to enhance the shock-carrying
capacity of the whole device.
2. Multi-scale modeling of drop-induced MEMS failure
An accurate assessment of the stress eld in MEMS subject to
extreme loading conditions like impacts is not trivial, because of
the several length-scales involved. It has been already pointed
out in the Introduction that, in case of polysilicon sensors, at least
three length-scales can be identied and should be properly inves-
tigated: a macro-scale, at the package/die level (size on the order of
mm); a meso-scale, at the sensor level (size on the order of hun-
dred of lm at most, but details with characteristic size on the order
of lm); and a micro-scale, at the polycrystal level (size on the order
of hundred of nm at most).
Fig. 1. Lateral views of the studied drop orientations: (a) top; (b) bottom.
support plate
I.C.
MEMS die
cap
MEMS die
sensors
Fig. 2. Sketch (a) of the package geometry (mold compound is not displayed to
show where MEMS die and I.C. are located) and (b) of the MEMS die geometry (cap
is not displayed to show where the accelerometers are placed).
Table 1
Elastic properties of the materials constituting the package (at the macro-scale).
Die and I.C. Mold Support plate
Youngs modulus E (GPa) 130 28 117
Poissons ratio m 0.22 0.3 0.3
Table 2
Homogenized elastic properties of polysilicon (at the meso-scale). Subscripts refer to
the local reference frame shown in Fig. 3.
E
11
E
22
(GPa) E
33
(GPa) G
12
(GPa) G
13
G
23
(GPa) m
12
m
13
m
23
152.9 130.1 63.7 79.6 0.2 0.28
A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349 341
With specic reference to the analysis of drop impacts, at the
macro-scale the impact-induced propagation of stress waves in-
side the device must be tracked. Because of the geometry of the de-
vice (see Fig. 2), wave reections at any interface between
dissimilar materials occur; these reections will be shown in Sec-
tion 4 to affect the MEMS failure.
At the meso-scale the inuence of the impact-induced stress
waves propagating inside the device on sensor dynamics and on
possible failure has to be captured. At this scale two kinds of failure
mode can be envisaged: a rst one is linked to the vibrations of the
sensor as a whole, and is caused by an excessive deformation of the
movable parts of the MEMS; a second one is instead linked to the
propagation of shock waves inside the sensor, and shows up as
very localized cracking close to MEMS anchor(s). Former analyses
[8] have shown that the actual failure mechanism of the MEMS
accelerometer here investigated falls within the rst category
mentioned above. While at the macro-scale materials can be as-
sumed to be isotropic (see Table 1), at the meso-scale polysilicon
behaves as a transversely isotropic solid, with overall elastic mod-
uli as gathered in Table 2 (the relevant reference frame in depicted
in Fig. 3).
These overall elastic properties of the polysilicon lm constitut-
ing the aforementioned movable parts have been determined on
the basis of an ad hoc homogenization procedure; even though this
procedure has been already explained in [15,8], its fundamentals
are collected in Appendix A. Likewise, the overall strength and
toughness of polysilicon can be estimated through homogenization
techniques for nonlinear solids.
At the micro-scale, details of the failure mechanism, involving
intergranular and/or transgranular cracking, can be studied
[16,9]. Analyses at this length-scale can be avoided if an accurate
and calibrated failure criterion for polysilicon is available at the
meso-scale. According to experimental evidences [17], polysilicon
is extremely brittle and very sensitive to imperfections at room
temperature; hence, its tensile strength may depend signicantly
on the loading condition and, therefore, on the MEMS layout. In
our analyses we dene at the meso-scale, once for all, a reference
480
7
5
0
6
5
0
A
B
C
D
suspension springs
seismic plate
MEMS anchor
x
1
x
2
x
3
Fig. 3. Sketch of the geometry of the two uni-axial accelerometers (measures in lm; thickness of the seismic plates th = 15 lm).
Fig. 4. Package drop, bottom orientation. Time evolution of the acceleration a3 felt
by the center of gravity of the seismic plate in the rst 100 ls after the impact (here
g = 9.81 m/s
2
is the gravity acceleration).
C
A
B
D
Fig. 5. Package drop, bottom orientation. Time evolution of the relative displace-
ments between corners of the seismic plate and die/cap surfaces: (a) damped case
and (b) undamped case.
342 A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349
polysilicon tensile strength r
t
[18]: anytime and anywhere the
maximum principal stress exceeds this reference tensile strength,
local failure is assumed to occur according to a Rankine-type fail-
ure criterion. The aforementioned characteristic tensile strength
r
t
has to account for all the possible micro-mechanical failure
mechanisms, involving transgranular as well as intergranular
cracking caused by either tensile or shear stress states. r
t
can be
determined with an experimental campaign: because of the brit-
tleness of silicon, it must be ensured that during the tests the same
loading condition (i.e. tensile, torsional or exural) of the actual
case holds in the polysilicon, to avoid size effects (see e.g. [18]).
Since analyses at the micro-scale are not run, this two-scale
procedure can detect the most dangerous drop congurations
and the MEMS details which are expected to break; on the con-
trary, it does not allow to track all the micro-mechanical features
of the failure mode (i.e. crack path at percolation). First results col-
lected in [9], where a characteristic crystal morphology in terms of
grain size and shape has been allowed for, showed that in view of
the layout of the sensor here studied, the two-scale approach actu-
ally furnishes robust and accurate estimates of the failure mode.
As for coupling between macro- and meso-scales, in the case
here investigated it results negligible. In fact, after the impact the
sensor is loaded by forces due to the contact with the die surface
(surface loading), and due to the acceleration/deceleration history
(bulk loading). The interaction between length-scales, i.e. the inter-
action between the dynamics of package/die and sensor is repre-
sented by tractions transmitted at the anchor point(s), which are
primarily affected at the MEMS level by the inertia of the sensor it-
self. Because of small ratio between the mass of the sensor and the
mass of package/die (in the studied case on the order of 10
4
for
Fig. 6. Comparison among the maximum acceleration peaks felt by the center of gravity of the seismic plate in the studied congurations.
Fig. 7. Effect of package on the stress envelopes at spring end cross-sections. Top row: damped case; bottom row: undamped case. (a and c) top drops and (b and d) bottom
drops.
A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349 343
package and 10
3
for die), length-scale interaction can be ignored.
This means that macro- and meso-scale analyses do not need to
run concurrently; instead, macro-scale analyses can be run rst,
and their outcomes exploited in order to dene the boundary
(loading) conditions for meso-scale ones. At the macro-scale the
presence of the sensor does not need be accounted for, whereas
at the meso-scale the loading conditions are represented by the
stress waves impinging upon the anchor point(s).
In what follows we assume package/die to fall from an assigned
height h
drop
150 cm . This reference drop height has been chosen
in order to understand what happens in case of an accidental drop
during handling of a portable device. If drag resistance during the
free ight is disregarded, i.e. if the drop is assumed to occur in vac-
uum, than impact velocity would be v
drop

2gh
drop
_
. If instead
drag resistance is accounted for, impact velocity gets slightly smal-
ler: details are furnished in Appendix B.
The struck target body is assumed to have a at surface and to
be so massive as compared to the device to be considered a rigid
body. The amplitude of sensor vibrations thus turns out to be
slightly overestimated, and drop features (like, e.g. drop height)
leading to failure accordingly under-estimated. Regardless of this
approximation, results to follow provide an objective classication
of drop severity for the MEMS.
It is nally worth emphasizing that, while the package/die
bounces off the target, the sensor is shaken and the magnitude of
the displacement eld results to be as great as the polysilicon lm
thickness. Therefore, in the simulations nite strain effects have
been fully taken into account.
3. Analytical estimation of the maximum acceleration felt by
the sensor
Drop severity can be estimated through the maximum acceler-
ation experienced by the sensor after the impact. A reference value
of this acceleration is (see [19]):

v
6
drop
R
m
1m
2
t
Et

1m
2
d
E
d
_ _ _ _
2
5

_
1
where v
drop
is the velocity of the falling device when it strikes the
target; m is the mass of the whole device; E
t
and E
d
are the target
and device Youngs moduli; m
t
and m
d
the target and device Poissons
Fig. 8. Effect of damping on the stress envelopes at spring end cross-sections. Top row: naked die drop; bottom row: package drop. (a and c) top drops and (b and d) bottom
drops.
Fig. 9. Experimentally determined effect of package on failure occurrence.
344 A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349
ratios; R is a characteristic size of the falling device. In Eq. (1) the
target surface has been assumed at, whereas the device is viewed
as an homogeneous, spherical-like body.
This approach has been already shown to be ineffective for reli-
ability assessment of MEMS (see e.g. [14]), since both the package
and the die are not spherical in shape. Furthermore, it does not al-
low to distinguish among the falling orientations of the device.
According to [14,8,9] and to the forthcoming results, it turns out
that the only effective way to establish an objective criterion for
drop severity is to look at the impact-induced stress eld inside
the sensor, since this stress eld is the real cause of the MEMS
failure.
4. Results
The geometry of the studied device is shown in Figs. 2 and 3. As
far as package is concerned, a layered plate supports die and I.C.,
which are both wrapped around by the mold compound. Twin uni-
axial accelerometers are supported by the die and protected from
the external environment by the cap; each sensor is constituted
by a movable seismic plate, connected to the die via two suspen-
sion springs. MEMS failure, which is obviously linked to its layout,
is expected to be affected by the slenderness of the suspension
springs. While the shape of these springs is designed to enhance
in the working regime the sensitivity of the accelerometer to the
out-of-plane (i.e. along axis x
3
, see Fig. 3) acceleration, excessive
deformation and therefore possible localized failures can be caused
by impact-induced severe shaking of the seismic plate.
To fully assess the effect of packaging on failure of the transla-
tional accelerometer, both a package and a naked die are assumed
to strike the target, falling from a drop height h
drop
150 cm.
In case of packaged sensor subject to a bottom drop, Fig. 4
shows the history of acceleration a
3
in the sensing direction x
3
,
as experienced by the center of gravity of the seismic plate. Fig. 5
instead shows the evolution of the relative displacements (again
in the sensing directions) between the four plate corners and the
die/cap surfaces; results are here reported for both the damped
and the undamped case. Damping has been accounted for in the
simulations through a viscous term in the equations of motion of
the sensor; according to the product datasheet, a rate-independent
quality factor Q = 1.5 was adopted. In Fig. 5 the two horizontal
(dashed) lines represent the maximum relative displacement each
point of the plate can experience, corresponding to the situations
in which the plate corners get in contact with die Du
3

1:8 lm or cap Du
3
15:9 lm surfaces.
If we assume that, after the impact against target, plate vibra-
tions are linked to exural oscillation of the two suspension
springs, it results that the fundamental sensor deformation mode
has period T
u
7:7 ls. This value is very similar to the period asso-
ciated with the acceleration peaks showed in Fig. 5b; almost the
same period of oscillations can be recognized in Fig. 5a, even
though damping here partially shadows uctuations beyond
t 20 ls. The difference between T
u
and the period of oscillations
in Fig. 5 is related to the eccentricity of the center of mass of the
seismic plate with respect to the spring longitudinal axes. The rst
deformation mode of the sensor thus displays a coupling between
exural and torsional spring vibrations, leading to a slightly differ-
ent vibration period. In fact, tilting of the seismic plate, which
causes the coupled exural/torsional deformation mode of the
springs, can be recognized in Fig. 5 by looking at the different evo-
lution of the relative displacements at corners A and D, and at cor-
ners B and C. Higher-frequency vibrations can be seen in the
history plots: these vibrations are linked to higher order deforma-
tion modes of the springs, since the seismic plate is massive en-
ough to behave almost rigidly under the impact loadings (even if
it is modeled as a deformable body, whose elastic properties are
collected in Table 2).
Histories of acceleration a
3
and of relative displacements Du
3
at
plate corners have been obtained also in all the other cases inves-
tigated, but they are not reported here for brevity. The major out-
come, in terms of post-impact maximum acceleration peaks are
gathered in Fig. 6, where a comparison among the drop congura-
tions is reported. These results offer a very low content of informa-
tion as for the denition of a critical acceleration peak threshold,
beyond which MEMS failure should be expected.
If the maximum acceleration felt by the sensor is estimated
according to Eq. (1), results on the order of a 10
5
g are obtained:
they turn out to be two orders of magnitude smaller than those
collected in Fig. 6. As already emphasized above, this formula ap-
pears not able to account for wave propagation phenomena inside
the package/die and therefore it is not able to assess how danger-
ous an impact could be for MEMS.
By looking, inside the suspension springs, at the so-called stress
envelopes, which represent the time evolution of the maximum
and minimum principal stresses, an unambiguous way to dene
MEMS failure can be established. Because of the sensor layout,
in the studied drop congurations the stress eld turns out to be
highly localized around the end cross-sections of the suspension
springs; for this reason, Figs. 7 and 8 show the stress envelopes
at the end cross-sections of both springs. Specically, in Fig. 7
the effect of package on the stress envelopes is shown for each
Fig. 10. Comparison between post-impact vertical displacements of the striking surfaces of package and naked die: (a) top drops and (b) bottom drops.
A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349 345
drop conguration, both when damping is allowed for and not. In
Fig. 8 results are organized so as to understand the effect of damp-
ing. In these plots the horizontal dashed lines represent the exper-
imentally determined strength of polysilicon, which amounts for
this device to r
t
4 GPa . From these plots it can be assessed that
damping slightly affects the results, since the maximum tensile
stress peaks are attained a few ls after the impact, while damping
effectively reduces the envelope size at later instants. In these
graphs the fundamental period of vibration is T
r
15 ls, which
is different from the previously mentioned T
u
. The difference be-
tween the two periods is due to the fact that stress envelopes
are relevant to a whole spring cross-section: therefore, local uc-
Fig. 11. Top drop. Snapshots of the impact-induced deformed sensor conguration up to 25 ls: (left) packaged MEMS; (right) naked die supported MEMS. Die and cap are
shown in light blue, the seismic plate is shown in green, and suspension springs are shown in violet. (For interpretation of the references in colour in this gure legend, the
reader is referred to the web version of this article.)
346 A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349
tuations at each material point are shadowed by the sectional
effects.
From these plots it also emerges that in the studied cases pack-
aging increases the tensile stress state in the suspension springs;
this increment is marginal in the top drop conguration, whereas
it looks detrimental of the sensor impact-carrying capacity in the
bottom drop conguration. In fact, if the naked die falls the maxi-
mum tensile stress never exceeds the characteristic tensile
strength of polysilicon; on the contrary, if the whole package falls,
soon after impact against the target, the polysilicon tensile
strength is exceeded and failure, according to our criterion, occurs.
This outcome agrees well with the experimental data reported in
Fig. 9 in terms of percentage of devices broken by laboratory drop
tests: these results show that packaging actually increases failure
Fig. 12. Bottom drop. Snapshots of the impact-induced deformed sensor conguration up to 25 ls: (left) packaged MEMS; (right) naked die supported MEMS. Die and cap are
shown in light blue, the seismic plate is shown in green, and suspension springs are shown in violet.
A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349 347
occurrence. In the experimental campaign devices either packaged
or supported by a bare die were subjected to drops from the same
h
drop
150 cm considered in the numerical simulations. Post-mor-
tem inspection of the broken devices revealed that the typical fail-
ure mode actually consists of cracking at the end cross-sections of
the suspension springs, see also [9]. To explain this outcome, unex-
pected on the basis of the maximum acceleration peaks shown in
Fig. 6, Fig. 10 depicts macro-scale results in terms of post-impact
vertical displacement of the package/die surface striking the target.
These plots allow to determine t
impact
, namely the time interval
during which the striking surface remains in contact with the tar-
get while compressive stress waves start traveling inside the pack-
age/die. In case of top drops (Fig. 10a), t
impact
is not substantially
affected by the presence of the package. On the contrary, for bot-
tom drops (Fig. 10b), package is associated with an increase of
t
impact
, which means that the elastic energy stored in the device be-
cause of the travelling compressive stress waves, increases too;
consequently, the bounce off the target, and the shaking of the sen-
sor turns out to be more severe.
This rationale is conrmed by Figs. 11 and 12, where lateral
views of the time-evolving congurations of the sensor, as ob-
tained from meso-scale analyses, are represented. In case of top
drops (Fig. 11), the sensor dynamics and, specically, spring deec-
tion/torsion do not signicantly differ in the die and package cases:
therefore, failure is only marginally inuenced by packaging. In
case of bottom drop of the whole package, the seismic plate is soon
pushed towards the cap and, after striking the cap surface, it
bounces off and continuously vibrates in the cavity between die
and cap. On the contrary, in case of bottom drop of the naked die
the elastic energy stored in the die is not sufcient to cause impact
of the seismic plate against the cap surface. Since we have assumed
an elastic-perfectly brittle behavior of the sensor, the larger the
deformation of the suspension springs the higher the stress peaks
at their end cross-sections. This further justies the detrimental ef-
fect of packaging on the impact-carrying capacity of the MEMS in
case of bottom drops.
A comparison between results in Fig. 6 and graphs in Fig. 7 fur-
ther shows that the maximum acceleration does not constitute an
objective indicator of drop severity. In fact, in case of die drops the
value of a
max
3
associated with a bottom drop is about twice that
associated with a top drop; on the contrary, Fig. 7 shows that the
tensile stress envelope slightly exceeds the adopted reference ten-
sile strength r
t
of polysilicon in case of a top drop (Fig. 7a),
whereas it does not approach r
t
in case of a bottom drop
(Fig. 7b). Fig. 10 again allows to understand this unexpected result:
t
impact
is by far longer in case of a top drop of the bare die, and the
elastic energy stored within this interval inside the die causes a
more severe bounce off the target. Furthermore, in the packaged
case a
max
3
results to be grater in case of a top drop (see Fig. 6),
whereas the maximumpeak of the tensile stress envelope is higher
in the bottom drop case (see Fig. 7a and b). According to our crite-
rion, the maximum acceleration thus furnishes a wrong classica-
tion of drop severity.
5. Concluding remarks
In this work the effect of packaging on the failure of a uni-axial
polysilicon MEMS accelerometer available on the market, subject
to accidental drops have been explored. By considering the multi-
ple length-scales involved in MEMS failure, a multi-scale nite ele-
ment procedure has been proposed: at the macro-scale, drop
features like drop height and falling orientation are fully accounted
for to dene how sensor is loaded; at the meso-scale the links be-
tween the drop-induced loading and the failure occurrence are
elucidated.
Sensor failure has been assumed to occur as soon as a character-
istic tensile strength of the polysilicon is attained anywhere in the
MEMS bulk. Because of sensor layout, the stress eld gets always
amplied at the end cross-sections of the suspension springs, inde-
pendently of the falling orientation and of the presence of the
package. Therefore, the drop-induced failure mode is very localized
in these regions.
It has been shown that packaging is not always benecial in
enhancing the sensor impact-carrying capacity. In case of top
drops, when sensor is falling upside-down (i.e. when the top sur-
face of the package or of the naked die strikes the target surface),
packaging does not affect much MEMS failure. On the contrary,
in case of bottom drops (i.e. when the bottom surface of the pack-
age or of the naked die strikes the target surface) packaging
strongly increases the occurrence of failure. This detrimental effect
has been shown to be linked to the post-impact compressive defor-
mation of the whole package, which stores enough elastic energy
to render the bounce off the target much more severe for the
sensor.
In future works we will further investigate the effect of packag-
ing on MEMS failure by coupling the proposed numerical scheme
with stochastic micro-scale simulations. Moreover, the effect of
tilting, i.e. the effect of a skewed impact of the device against the
at target, will be also allowed for.
Acknowledgements
The support of EU NoE Design for Micro & Nano Manufacture
(Patent DfMM), contract No. 509255 is gratefully acknowledged.
A.G. wishes to thank Cariplo Foundation for the nancial support
of the project Innovative models for the study of the behaviour of sol-
ids and uids in micro/nano electro-mechanical systems. F.F. is grate-
ful to STMicroelectronics, as part of this work has been developed
while the author was a visiting student working at STMicrolectron-
ics BU in Singapore. S.M. gratefully acknowledges the nancial sup-
port of Italian MIUR through PRIN-Con2005 programme Interfaces
in innovative micro and nano structured materials and devices.
Appendix A. Homogenized elastic properties of polysilicon
Fundamentals of the procedure to obtain the overall elastic
properties of polysilicon at the meso-scale are here discussed.
According to homogenization theory, the size of the material
representative volume (RV) rst needs to be dened, ensuring
that the orientation of the axes of elastic symmetry of each sili-
con grain can be treated as a continuous-like eld. Such details
of homogenization theory will not be discussed here; readers
are referred to, e.g. [20,15]. As a rule of thumb, Cho & Chasiotis
[16] suggested that the RV should contain at least a few hun-
dreds of grains to be representative; this obviously constitutes
a sufcient condition. Since the suspension springs of the MEMS
under study contain only a few grains along their in-plane thick-
ness, the aforementioned requirement is not satised. Hence, the
homogenized elastic properties of polysilicon can be adopted
only to model the structural behavior of the whole MEMS at
the meso-scale [8,9].
By assuming a random orientation in the x
1
x
2
plane of the
axes of elastic symmetry of silicon grains, the polysilicon lm con-
stituting the movable parts of the MEMS turns out to behave as a
transversely isotropic material, the axis of transverse isotropy
being the direction x
3
of epitaxial growth (see Fig. 3). As standard
practice in homogenization procedures, Voigt- and Reuss-like
bounds on the effective elastic properties are respectively obtained
via [15,8,9]:
348 A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349
S
e

1
2p
_
2p
0
T
T
e
S
m
T
e
d#
S
1
r

1
2p
_
2p
0
T
T
r
S
1
m
T
r
d#
A:1
where
T
stands for transpose; S is the matrix of homogenized elastic
moduli, which links at the meso-scale the stress and strain vectors
(respectively r and e) through r Se; S
m
is the matrix of elastic
moduli of the single FCC grain in its privileged reference frame; T
e
T
r
is the orthogonal mapping that denes the variation of strain
(stress) components due to rotation fromthe local privileged crystal
reference frame to the overall x
1
x
2
one.
As already pointed out in [15,8], because of the small deviation
from elastic isotropy of FCC crystals, bounds (A.1) turn out to be
tight for polysilicon. In-plane Youngs modulus E
11
E
22
and Pois-
sons ratio m
12
actually result to be bounded by, respectively:
E
e
158:7 GPa , E
r
147:1 GPa ; m
e
0:18, m
r
0:22. Results of
the simulations here presented have been obtained by adopting
the relevant average values gathered in Table 2.
Appendix B. Dynamics of a falling object: drag effects
The dynamics of a falling object, whose size is far smaller than
the displacement its center of mass is experiencing, is governed by
the differential equation:
dv
dt
g 1
v
2
V
2
_ _
B:1
Here: t is time; v is the velocity of the falling object, assumed to be a
rigid body; g = 9.81 m/s
2
is the gravity acceleration. The last term at
the right-hand side of Eq. (B.1) accounts for the interaction between
the moving object and the surrounding air (treated as a non-dense
uid); accordingly, V is [21]:
V

2mg
.
air
AC
d

B:2
where m is the mass of the falling object; .
air
% 1:2 kg=m
3
is the air
density at room temperature; A is the projected frontal area of the
falling body; C
d
is the drag coefcient (C
d
1:28 for plates).
Keeping as reference a drop height h
drop
150 cm, in case of
drops in vacuum (drag resistance null) any body strikes the target
surface with velocity v
drop
5:425 m=s. In case of drops in air,
assuming the body is not rotating during ight, results in terms
of velocity v
drop
at impact for MEMS either supported by a bare
die or packaged are reported in Table 3. It results that drag forces
reduce v
drop
by about 20% and 8% with respect to v
drop
, when the
naked die or the whole package is respectively falling.
To effectively compare failure indicators for MEMS and for ease
of presentation, we have always assumed that device strikes the
target with velocity v
drop
. According to the results of Table 3, the
drop height h
drop
should be appropriately increased to achieve
the stress levels in the suspension springs reported in this paper.
At any rate, it clearly emerges that in case of drops in air from an
assigned height, the presence of package leads to worse loading
conditions for the MEMS; in fact, drops of a bare die, which is light-
er in weight, are characterized by a smaller impact velocity, and
therefore by tensile stress peaks of a smaller amplitude if com-
pared to those relevant to package drops.
References
[1] Varghese J, Dasgupta A. Test methodology for durability estimation of surface
mount interconnects under drop testing conditions. Microelectron Reliab
2007;47:93103.
[2] Luan J, Tee TY, Peck E, Lim CT, Zhong Z. Dynamic responses and solder joint
reliability under board level drop test. Microelectron Reliab 2007;47:45060.
[3] Ghaffarian R. CCGA packages for space applications. Microelectron Reliab
2006;46:200624.
[4] Srikar VT, Senturia SD. The reliability of microelectromechanical systems
(MEMS) in shock environments. J Microelectromech Sys 2002;11:20614.
[5] Cheng Z, Huang W, Cai X, Xu B, Luo L, Li X. Packaging effects on the
performances of MEMS for high-g accelerometer: frequency domain and time-
domain analyses. In: Proceedings of HDP04, 2004. p. 282289.
[6] Wagner U, Franz J, Schweiker M, Bernhard W, Mller-Friedler R, Michel B, et al.
Mechanical reliability of MEMS-structures under shock load. Microelectron
Reliab 2001;41:165762.
[7] Li GX, Shemansky FA. Drop test and analysis on micro machined structures.
Sensors Actuat A 2000;85:2806.
[8] Mariani S, Ghisi A, Corigliano A, Zerbini S. Multi-scale analysis of MEMS
sensors subject to drop impacts. Sensors 2007;7:181733.
[9] Mariani S, Ghisi A, Fachin F, Cacchione F, Corigliano A, Zerbini S. A three-scale
FE approach to reliability analysis of MEMS sensors subject to drop impacts.
Meccanica 2008;43:46983.
[10] Choa SH. Reliability of vacuum packaged MEMS gyroscopes. Microelectron
Reliab 2005;45:3619.
[11] Tanner DM, Walraven A, Hegelsen K, Irwin LW, Brown F, Smith NF, et al. MEMS
reliability in shock environments. In: 38th Annual international reliability
physics symposium, San Jose, California, 2000. p. 129138.
[12] Shi BJ, Shu DW, Wang S, Luo J, Ng Q, Lau JHT, et al. Shock response and power
spectrum analysis of a head actuator assembly. In: Impact loading of
lightweight structures. WIT Transactions on Engineering Sciences, vol.
49. WIT Press; 2005.
[13] Jiang D, Shu D. Predication of peak acceleration of one degree of freedom
structures by scaling law. ASCE J Struct Eng 2005;131:5828.
[14] Suhir E. Is the maximum acceleration an adequate criterion of the dynamic
strength of a structural element in an electronic product? IEEE Trans Compon,
Packag Manifact Technol 1997;20:5137.
[15] Mullen RL, Ballarini R, Yin Y, Heuer H. Monte Carlo simulation of effective
elastic constants of polycristalline thin lms. Acta Mater 1997;45:224755.
[16] Cho SW, Chasiotis I. Elastic properties and representative volume element of
polycrystalline silicon for MEMS. Exp Mech 2007;47:3749.
[17] Chasiotis I, Knauss WG. The mechanical strength of polysilicon lms. Part 2:
size effect associated with elliptical and circular perforations. J Mech Phys
Solids 2003;51:155172.
[18] Boroch R, Wiaranowski J, Meller-Fiedler R, Ebert M, Bagdahn J.
Characterization of strength properties of thin polycrystalline silicon lms
for MEMS applications. Fatigue Fract Eng Mater Struct 2007;30:212.
[19] Falcon E, Laroche C, Fauve S, Coste C. Collision of a 1-D column of beads with a
wall. Euro Phys J B 1998;5:11131.
[20] Ponte Castaneda P, Suquet P. Nonlinear composites. Adv Appl Mech
1998;34:171302.
[21] Clancy LJ. Aerodynamics. Pearson Higher Eduction; 1998.
Table 3
Effect of drag resistance on the impact velocity v
drop
5:425 m=s.
v
drop
vdropvdrop
vdrop
%
Bare die 4.331 20.17
Package 4.988 8.06
A. Ghisi et al. / Microelectronics Reliability 49 (2009) 340349 349

Potrebbero piacerti anche