Académique Documents
Professionnel Documents
Culture Documents
-
Same Die Size - Stacked Chip Scale - Ball Grid Array Package
Bret A. Zahn
Vice President, Worldwide Design & Characterization - ChipPAC Inc.
2775 North Arizona Avenue, Chandler, Arizona 85225
Voice: 480.632.3721 Fax: 480.632.0650 Email: bret.zabn@chippac.com
initiate a crack, and the number of cycles for the crack to The diagonal slice model passes through the thichess
propagate across a solderjoints diameter. of the package assembly, capturing all major components
Darveaux's methodology has been previously and a full set of solder joints. The use of a slice model
presented in the successful analyses of various electronic involves a choice on the part of the analyst on the boundary
assemblies from multiple indusay sources. Amagai [I31 constraints to be applied at the slice plane. The plane is
generated lifetime predictions for a chip scale package on neither a free surface nor a true symmetry plane. The
an organic printed circuit board. Fusaro and Darveaux [14] reasonable compromise of coupling the y-displacements of
used the viscoplastic properties of eutectic solder to the nodes on the slice plane was chosen. This has the effect
analyze the reliability of a copper baseplate attachment for that the slice plane is free to move in the y-direction, but
a power module. Dougherty et al. [15] analyzed a micro- that the surface is required to remain planar. Boundary
miniature electronic package. Johnson [I61 and Pitaressi et constraints applied to a typical slice model are shown in
al. [I71 utilized the methodology to predict board-level Fig. IO. Note that for all analyses presented in this paper,
solder joint reliability of multiple ball grid array packages. the printed circuit board x-dimension (or length) was set at
Recently, Zahn [I81 and Goetz and Zahn [19] extended the 1.25X that of the modeled package slice x-dimension. UX
methodology to predict both solder ball and solder bump was not coupled at the printed circuit board +X vertical
reliability of a multi-chip silicon based system-in-package. surface as shown in Fig. 10. Coupling this surface is a
In many of these publications, the authors have presented good choice if another package is assumed to exist adjacent
reliability test data that validates the accuracy of to the modeled package. When assuming an adjacent
Darveaux's methodology within +/-2X, which is package, the printed circuit board length should be set to
considered state of the art for this type of complex physical extend one-half the distance between the modeled package
analysis. and the assumed adjacent package. The y-dimension (or
width) of the slice model is one-half the solder ball pitch.
Same Die Size - Stacked - Chip Scale - Ball Grid Array Note that for a diagonal slice model, the ball pitch is the
Package hypotenuse (1.1314mm) of the true ball pitch (0.80") as
A 1Ox8mn1, 72-ball (9x8 Full Matrix), 0 . 8 0 " pitch is evident by the ball separation along the bold print dashed
package was analyzed using a same size stacked die line in Fig. 1.
configuration. The die size was measured at 7.9x6.4".
The package outline drawing is displayed in Fig. 1. The
basic structure of the stacked die package is shown in Fig. Material Properties & Modified Anand Constants
2. A "spacer" die is placed between the two active die to Linear and non-linear, elastic and plastic, time and
facilitate wire bonding as displayed in Fig's 3 and 4. The temperature independent and dependent material properties
layer dimensions of the printed circuit board and package were incorporated in the f ~ t element
e models as displayed
substrate are given in Fig's 5 and 6. Dimensional details of in Tables 1 and 2. As an alternative to a rate-independent
the re-flowed solder ball along with the package substrate plasticity approach, Darveaux [l 1,201 has presented solder
and printed circuit board solder re-flow pads are shown in constitutive relations based on Anand's [I21 model for rate-
Fig. 7. dependent plasticity. Anand's constitutive model
incorporates viscoplasticity, a time-dependent plasticity
Solder Ball Fatigue Models phenomenon, where the development of plastic strains is
Viscoplastic finite-element simulation methodologies dependent on the rate of loading. Viscoplasticity is defined
were utilized to predict solder ball joint reliability of the by unifying plasticity and creep. Anand's model does not
same size stacked die chip scale package under accelerated consider rate-independent plasticity. Therefore, Darveaux
temperature cycling conditions (-40C to +125C, 15min modified the constants in Anand's constituitive relation to
rampdl5min dwells). Due to the complex physics that account for both time-dependent and time-independent
encompass this type of non-linear tmsient finite element phenomenon. These modified Anand constants are given
analysis, only a diagonal slice of the package was modeled in Table 3 [20]. Using ANSYS as the finite element
in order to facilitate reasonable model run times. The analysis tool, the Anand plasticity data table was activated
utilization of a diagonal slice assures that a worst-case for the solder ball material and incorporated the constants
situation is simulated where the perimeter solder ball is the as given in Table 3. Solder ball materials were meshed in
fwthest distance from the package center neutral point. ANSYS using the VISC0107 elements, whereas all other
The diagonal slice is shown by the bold print dashed line in package materials were meshed using SOLID45 elements.
Fig. I . Although this is not considered a perfect diagonal
slice, it will provide results that are within the +/-2X Solder Joint Fatime Life Prediction Methodology
published accuracy of Darveaux's methodology. The By measuring the crack growth rate of actual solder
resulting diagonal slice model is shown in Fig's 8 and 9. joints, Darveaux [20] was able to establish four crack
The entire model utilized a mapped (or structured) f ~ t e growth correlation constants (KI through K4) along with
element mesh that consisted of 6504 nodes and 4764 two equations by which finite element simulation results
elements. Typical solution run times were 2 hours (Dual could be used to calculate thermal cycles to crack initiation
Processor, 800 MHz Pentium In, lGByte RAM, NT along with crack propagation rate per thermal cycle.
Operating System). However, the methodology is sensitive to the finite element
modeling procedure. First, care must be taken in
controlling the element thickness at the interface between ANSYS Solution Methodology
the eutectic solder and copper pad. Second, element
volumetric averaging of the stabilized change in plastic Once the slice model has been completed as displayed
work witbin this controlled eutectic solder element in Figs 8 and 9, and the boundary constraints have been
thickness must he used. This procedure reduces singularity applied as indicated in Fig. IO, the ANSYS solution setup
issues whereby the size of the fnite element mesh affects commands are as follows:
plastic work simulation results.
Although equation constants for varying interface ! SET SOLUTION OPTIONS
element thicknesses are provided by Darveaux [20], the Isolu ! enter soh processor
element interface thickness utilized by all models discussed eqslv,pcg,l .Oe-08 ! set solver and tollerance
herein was 0.0254mm (Imil). This thickness equates to the antype,static,new ! set analysis type
first two layers of solder ball material elements at the nlgeomon ! set large def and strain
package substrate and printed circuit board interface joints nropt,auto,,off ! set newton-raphson soh
(see Fig. 9). It should also he noted that Darveauxs outres,all,last ! write data to .rst file
methodology requires that the solder ball and solder mask
material elements not be joined in the finite element model. The thermal cycle temperature and time variables can be
This is due to the non-adhesion between solder and mask set in ANSYS using variable names and equations as
materials. Given a solder mask defined solder joint at the follows:
package substrate, Darveaux recommends a 0.0127mm
(0.5mil) gap between the solder hall and solder mask ! SET THERMAL CYCLE VARIABLES
material in the finite element model. This gap is visible in hitmp= 125+273 ! set hi cycle temp (K)
Fig. 9. Corresponding KI through K4 crack growth hirmp=l5*60 ! set lo-hi ramp time (sec)
correlation constants for a 0.0254 (Imil) solder joint hidwl=15*60 ! set hi dwell time (sec)
element thickness are given in Table 4. The equations for
the calculation of thermal cycles to crack initiation WN, lotmp=-40+273 ! set lo cycle temp (K)
and crack propagation rate per thermal cycle da/dN are lormp=l5 *60 ! set hi-lo ramp time (sec)
shown below as (1) and (2) respectively. lodwl=15*60 ! set lo dwell time (sec)
Note that the analysis will use one substep for every IO
degrees K of temperature change in a thermal ramp load
step as suggested by Darveaux [ l l ] and as calculated by
Where AW,. is the element volumetric average of the the variable mpstp above.
stabilized change in plastic work within the controlled Once the solution setup is complete two thermal cycles
eutectic solder element thickness. The characteristic solder are simulated The ANSYS zero strain reference
joint fatigue life a (number of cycles to 63.2% temperature is set to the high temperature hitmp of the
population failure) can then be calculated by summing the thermal cycle sequence. Each thermal cycle consists of
cycles to crack initiation with the number of cycles it takes four load steps (ramp low, dwell low, ramp high, and dwell
for the crack to propagate across the entire solder joint high). Thus a complete simulation of two thermal cycles
diameter a as shown in equation (3). consists of eight load steps. Other publications that
incorporate Darveauxs methodology have indicated the
a simulation of three thermal cycles (twelve load steps).
a=No+- (3) However, it has been this authors experience that the
daiw difference in predicted fatigue life when simulating two
thermal cycles, as opposed to three thermal cycles, has
It should be noted that material intermetallic layers, been less than 5% and in the conservative direction (i.e. a
along with intermetallic spikes directed perpendicular to reduced number of predicted thermal cycles to failure). By
the intermetallic layers, typically form at the solder pad I only simulating two thermal cycles, simulation m times
solder ball interfaces. The mechanical effects of these can be reduced by 30-35%.
intermetallics on the fatigue life of the solder joints are not The below sequence of ANSYS commands indicates
directly included in the fnite element models. However, the setting of the zero strain reference temperature along
since the fatigue life prediction methodology developed by with those required for the f m t thermal cycle (i.e. first four
Darveaux was derived using measurement data taken from load steps).
actual solder joints, which presumably contained similar
intermetallic structures, their influence is believed to be tref,bitmp ! set zero strain temp
indirectly incorporatedinto the predicted results.
! RAMP LOW (LOAD STEP I)
autots,off ! tum off auto time step
Microcircuits and Electronic Packaaging, Vol. 20, [I81 Zabn, B.A., Comprehensive Solder Fatigue and
No. 2, 1997, pp. 81-88. Thermal Characterization of a Silicon Based Multi-
Chip Module Package Utilizing Finite Element
Dougherty, D., Fusaro, J., and Culbertson, D., Analysis Methodologies, Proceedings of the gh
Reliability Model for Micro-Miniature Electronic Intemational ANSYS Conference and Exhibition,
Packages, Proceedings of International Symposium August 2000.
on Microelectronics, 1997, pp. 604-61 1.
[19] Goetz, M. and Zabn B.A., Solder Joint Failure
Johnson, Z., Implementation of and Extensions to Analysis Using FEM Techniques of a Silicon Based
Daweauxs Approach to Finite-Element Simulation System-In-Package, Proceedings of the 2 f h
for BGA Solder Joint Reliability, Proceedings of IEEECPMT Intemational Electronics
49 Electronic Components & Technology Manufacturing Technology Symposium, October
Conference, June 1999, pp. 1190-1 195. 2000.
Pitarresi, J.M., Setburaman, S., and Nandagopal, B. [20] Dmeaux, R, Effect of Simulation Methodology
Reliability Modeling of Chip Scale Packages, on Solder Joint Crack Growth Correlations,
Proceedings of the 25 IEEE/CPMTlnternational Proceedings of 5 t hElectronic Components &
Electronics Manufacturing Technology Symposium, Technology Conference, May 2000, pp. 1048-1058.
2000, pp. 60-69.
~~~~~ ~
c1
~ ~~~~
I S. ( m a ) I 12.41 I Initial Value of Deformation Resistance
c2 QiR (1iKelvin) 9400 . Activation EnergyBoltzmanns Constant
c3 A (]/sec) 4.OE-06 Pre-ExponentialFactor
c4 6 (dimensionless) 1.5 Multiplier of Stress
Constant Value
K1 22400 cycles/psi
K2 -1.52
K3 5 . 8 610.
~ in/cycle/psi
west 2002
SEMICOP
SEMP Technology Symposium: International Eleetmntcs Manufarmring Technology (IEMV Symposium
ozoozIEEE
o.iam-no~m21f1~.00
Data Description Option A Option B Option C Option D
Table 7 Derailed Simulurion Resulrs - 0 1/43" (4 5mdJ Spacer and 0 70" Mold Cap
I
Data Descnptlon I Option A I Option B I Option C I Option D
I
q
, ,----+it
A9 !
c 0 0 0 0 0 0 0 0 ~
0oopooooo
F 0 op'0 Q 0 0 0
,,,' 1;
0,do 0 0 0 0 0 0
c'0 0 0 0 0 c-+ I
I
1
LL
w
0
N
Die 'A"(0.1397mm)
Mold Cap 3" Die Attach (0.0381mm)
(0.8 or 0.7") D i e " B / SpacerDie(0.1397 ar0.1143mm)
.2d Die Attach (0.0381")
Die"C'(O.1397mm)
I'DieAttach(0.0318mm)
S E M I C O P West 2002
SEMP Technology Symposium: lntemational Elechonics Manufacturing Technology (IEMT) Symposium
02002 IEEE
C-7~0575014M17.00 ~ 2002 SEMlllEEE IEMT
283
Fig. 5 & 6. Layer dimensions ofprinted circuit board and package substrate respectively.
Fig, 7. Dimensional details of rejlowed solder ball along withpackage andprinted circuit boardpads.
Fig. 8 & 9. Diagonal slice finite element model and resulting mesh.
SEMICOP
west 2002
SEMT Technology Symposium: htemati-1 Electmnics Manufacturing Technology (IEMT) Symposium
c ~ ~ ~ w I - u ~ ~ I IEEE
~ . o o ~ ~ ~ L v 1 M 2 SEYl"PFF m u l
284
Fig. I1 & 12. Package deformation trend at the maximum temperahwefrom the zero strain reference temperature.
wet 2002
SEMICOP
SEMImTechnology Symposium:lntemational Elecmnics Manufachuing Technology (IEMT) Symposium