.

Effects of Board Surface Finish

on specially designed for specific missions. Space applications
may rangefromshuttlemissionslastinga few days to
Failure Mechanisms missions outer solar systems requiring multi-decade service
and Reliability of BGAs life.

The difference in reliability requirements plays a key role

Dr. Reza Ghaffarian, Ph.D.
in theselection/designandinspection/testingprocesses.
Jet Propulsion Laboratory
Eventhoughthehighestreliabilityisdemanded,itis
California Instituteof Technology
extremely
difficult
quantify
to thereliability
level
Tel: (818) 354-2059
requirement and demonstrate its indices.One reason is due
email: Reza.Ghaffarian@ JPL.NASA.Gov
to the unavailability of solder joint field data and failure
statistics. This limits the exactdefinition of reliability
Abstract
assurance, Assurance must depend on qualification testing
BallGridArray(BGA)isanimportanttechnologyfor methodologiesunique to acceleratedenvironmentsalong
utilizinghigherpincounts,withoutthehandling and with credible analytical prediction.
processing problems of the peripheral array packages. The
assembly reliability of BGAs was investigated under a JPL- The current constraint on budgets and changes in military
led consortium. A large number of test vehicles with two specificationsalongwiththeapplications of boldnew
ceramic (361 and 625VOs) and five plastic (256,313,352- technologiesforspacemissionsnecessitatestheuse of
twotypes,560 VOs) packagesandthreesurfacefinishes advanced
commercial
electronic
packages.
Use of
wereassembledandsubjectedto two thermalcycling advanced packages including BGAs and CSPs, specifically
environments. Thenumber of
cycles
to failure was theirplasticversions, are contingent uponresolutionof
determined. many issues of space applications such as radiation damage
and considerationfor thermal/vacuum environment.
This paper will present cycles to failure in Weibull plots as
well as their failure characteristics up to 7,000 cycles. The Tounderstandandaddressmany commonqualityand
effect of surfacefinishes(HASL,OSP,andNi/Au)on reliability issuesof BGAs, JPL organized a consortium with
cyclestofailuresandtheirfailuremechanismswillalso sixteen
members in
early
1995.
Diverse
membership
discussedindetails. The SEMmicrophotographs and including
military,
commercial,
academia,
and
elementalsurfacechemistryforrandombrittlefailure of infrastructure
sectors
which
permitted
concurrent
a
NUAu will be presented. Possible reasons for ductile and engineeringapproach to
resolving
many
challenging
brittle failure behavior will also be discussed. technical issues.

Interimtestresultspresentedpreviously [l]. 'This paper

Introduction will discussthe following:
BallGridArray(BGA)electronic
packages are now The update cycles to failure test results in Weibull plots
become in common use for commercial applications and is and failure mechanismsto about 7,000 cycles
being persuade actively by aerospace and military industry.
Thesepackages are replacingQuad Hat Pack(QFP) Differences
in
ceramicand
plastic
package failure
packages at higherpincounts. Inadditiontoimproved mechanisms
thermalandelectricalperformance, BGAs higherpitch The effect of three types of surface finishes on failures
(0.050 inch typical), better lead rigidity, and self-alignment and failure mechanisms
characteristicsduringreflowprocessingmake them also
very attractive from the manufacturing aspects.
The finish effect will be emphasized. The surface finishes
BGAs' solder joints cannot be inspected and many aspects included: (1) Hot Air
Solder
Leveling
(HASL),the
of manufacturingincludingrework,anddoublesided predominant surface finish in industry; (2) Organic Solder
assembly processing and their effects on reliability are not Preservative (OSP), the new and cost-effective coating; and
well characterized. In high reliability SMT assembly (3) Ni/Au, an application specific finish.
applications, e.g. space and defense, the ability to inspect
the solder joints visually has been standard and has been a Generally,ceramic
assemblieswithdifferent
surface
key factor in providing confidencein solder joint reliability. finishes showed similar ductile (creep) failure mechanisms
whensubjectedto -30°C to100°Ccycles.Unexpectedly
Unlike commercial applications which measures reliability
brittlefailures wereobserved in twoinstances.Onewas
failures in partspermillion,theacceptableriskforthe
noticedduringaroutinevisualinspectionandtheother
success of a space mission is associated with the first failure
whenthepackagesubjectedtopulltests after cycling
of electronicpartsorinterconnection. In addition, every
exposure.Visual
inspections
wereverified by cross
spacemissionisuniqueandtheelectronicboards are
.'
sectioning and observation at higher magnifications using THERMAL CYCLING RESULTS
Scanning
Electron
Microscopy
(SEM).
The
SEM Figure 1 shows cycles
to
first hilure forall plastic
microphotographsandelementsurfacechemistry.will be packages: PBGA 313and256,OMPAC352,SBGA352
presented.Possiblereasons for ductileandbrittlefailure 'and SBGA 560. The results are for assemblies on FR-4 and
behaviorwillalso be discussed. , PolyimidePWBs.Preliminaryresultsfortwoassemblies
showedgenerallyhighercyclesto failure forpolyimide
underthe B conditions [I]. Themostrecent PBGA313
BGA Test Vehicle Configuration assemblies that failed under the cycle A conditions are also
included in the plots for comparison.
The two test vehicle assembly types were plastic (PBGAs)
and ceramic (CBGAs) packages. Both FR-4 and polyimide
PWBs with six layers, 0.062 inch thick, were used. Theseplotsweregenerated by rankingcycles tofailures
fromlowtohighandthenapproximatingthefailure
Plastic
packages
covered
the
range
from
OMPAC distribution percentiles using a medi'an plotting position, Fi
(Overmolded Pad Array Carrier) to SuperBGAs (SBGAs). = (i-0.3)/(n+0.4). No attemptsweremade to generate
These were: Weibull parameters since the sample sizes for B conditions
Two peripheral SBGAs, 352 and 560 VO was small and all assemblies under A conditions are being
Peripheral OMPAC 352 YO, PBGA 352 and 256 VO analyzed.
DepopulatedfullarrayPBGA313 I/Os
256 QFP (Quad Flat Pack), 0.4 mm Pitch Severaltrends were identified.The PBGA313,with a
depopulated full array with solder balls under the die, were
InSBGA,theIC die isdirectlyattached to anoversize the first plastic packages to show signs of failure at both A
copperplateprovidingbetterheatdissipationefficiency and B cycling conditions. The scatter for both was narrow
than
standard PBGAs. The solder
balls
for
plastic
packages were eutectic (63Sd37Pb). indicating large m Weibull parameter.

Ceramicpackageswith625110sand361 UOs were also The next package assembly failure was SBGA 560. This
included our
evaluation.
in Ceramic
solder
balls package and thePBGA 256 with failure data at the extreme
(90Pb/lOSn) with 0.035 inch diameters had a high melting cycles showed very large scatter, indicating a low m value.
temperature.Theseballswereattached to theceramic Scatter inPBGAs,in general, could be attributed toslow
substratewitheutecticsolder(63Sn/37Pb).Atreflow, failureprogress.Scatterinfailure data forSBGA560
packagesideeutecticsolderandthe PWB sideeutectic could be dueto the immaturity of the package at the time of
pastewill bereflowed to providetheelectro-mechanical use. These packages were taken from those assembled in
interconnects. theearlystage of production.Thereasonsforalarge
scatterfor thePBGA256cyclesto failure data arenot
Plasticpackageshaddummyanddaisychains with the known at this time.
daisy chains onthePWBdesignedtobe able to monitor
critical solder joint regions. Most packages had four daisy BothOMPACandSBGA352 failed at l a m cyclesthan
chain patterns, 560 I/O hadfive, and the QFP had one. SBGA 560 and they were between a low tail for PBGA 313
and a hightailforPBGA256. The 352 VO package
showed low scatter. Further test results being gathered and
analyzed at JPL could shed
somelight
on
failure
Thermal Cycling distribution and behavior on various plastic packages.
Two significantly different thermal cycle profiles were used
at two facilities, conditions A and B.
The
cycle A
conditions
ranged
from -30 to 100°C and had Effect of Ball Planarity on Failure
increase/decrease heating rates of 2°C and dwells of about
20 minutes at the high temperature to assure near complete Forassembliesundercycle A condition,a SBGA560
creeping. The duration of each cycle was 82 minutes. The packageshowednumerouselectricalinterruptionswith
cycleBconditionsrangedfrom -55 to 125°C.Cycle B signs of electrical opens at an unexpectedly low number of
could be also considered a thermal shock since it used a cycles(1,810cycles). To assurefailure, joints weredie
threeregions ch'amber:hot, ambient,andcold.Heating penetratedand testedtofailurein pull testing.Stain on
and cooling rates were nonlinear and varied between 10 to surfacerepresentsexistence of crackpriortopulltest.
15 "C/min. with dwells at extreme temperaturesof about 20 Fullycovered mined areameans joint failure. Photos of
minutes. The total cycle lasted approximately 68 minutes. surface failures for 560 SBGA with stains are compared to
BGAtestvehiclesweremonitoredcontinuouslytodetect a PBGA 313 in Figure 2. There much less stain on SBGA
electrical
failure
and
their
failure
mechanismwere 560 that PBGA 313. Nonof detached surfaces ofSBGA
characterized. BGA test vehicles
were
continuously
560 showedcomplete shins an indication of no joint
monitored througha LabView system at both facilities.
failure. DeLxhed surfaces also included a fewpad pulling
sites. Therefore, it waspostulated thatfailure could have
caused by packageinternaldaisychainpossibly by dierandomlydistributed. Ilridomncss thought to be due non-
debonding or otherpackage intend failuremechanisms.optimumsolderjointsformedfromthedistributiontail of
ball heightvariation. If this postulate is m e , then the
Solder joints were detached from package, board, and hoard solder height distribution for the balls that separated at the
traces. 'me majority of detachmerqswere from thepackageboardinterfaceshould be different fromthosethat
side- 468 out of 560. This means only 92 (560-468=92)separatedatthepackageinterface.
detachedfromthe board side in which22werefrom the
board traces. The 70 (92-22) solder detached surfaces were

CYCLES TO FAILURE DATA FOR 1 +PBGA 313- JPL Cycle

"c.OMPAC 352- Boeing Cycle
+SBCA 560- Boeina Cvcle
+PBGA 252- Boeing Cicle
PLASTIC PACKAGES .)I SBGA 352- Boeing Cycle
"4%- PBGA 313- Boeing Cycle
100

..........................................................................
90
./ SBGA 560 P
-Q
5
00

70
t..

E
t
s 60 ........c...........

50

-i
P
w 40

30

20

10

0
1500 2000 2500 3000 4500 3500 4000
JPUNASA Number of Thermal Cycles Reza Ghaffarian

PBGA 3 13 SBGA 560

Figure 2 Critck Propagation (Stained) of PBGA 313 and SBGA 560 after 1,810 A Cycles.
 Wewere able
to
verify
this
postulate
since we
had Planarity distributions forboth all balls and those failed from
. . documented package
ball
height
and
location
for
each the board under pull test are similar. Planarities were in the
packageprior to assembly. Solder
ball
heights
wen: range of 0.5 to 4 mils,averaging at 2.5 mils. For this
measured using a 3-D laser system. Then, planarity of each specific package, it wa5 concluded that there was no strong
ball calculated using the height of the ball minus the height correlation between site scparation and solder ball heights.
ofthe virtualseatingplanegenerated by thethreetallest 'Illis exercise did not consider the effects of package warpage
balls.
Theball
planarity
distribution
for
this
specific anditsdistribution as well as localizedboardwarpage
SBGA560packageisshowninFigure 3. Thisfigurealso distribution. Mostprobably either of thesevariableswould
include planarity distribution forjoints separated at the board have caused a more localized rather than random separation
interface. distribution.

50
45
40
35
30
25
Percentage
20
15
10
5
0
0 0.0005
0.001
0.0015 0.002
0.0025 0.003 0.0035 0.004
SEGA 560 Solder Ball Planarity (inch)

Figure 3 Solder Height Variation of SolderBalls for Board and PackageInterface Failure for a SBGA

at loOx by SEM,nogrossdamagewasobserved, as
evidenced from the top left SEM photo
of Figure 4.
Failure Mechanisms for Plastic Assemblies
Amongthe plastic
packages,the
PBGA 313 with Figure 5 includes a collection ofSBGAandPBGAsolder
depopulated full array balls was first to fail under both B and joints withvoids at different locations in the solder joints.
A thermal cycling conditions. Figure 6 shows various cross- Similarly toPBGA 313, most of voidsareclosetothe
sections of this package's balls from comer to the center at packageside.Thereisalsoanindicator ofhow cracks
4,682 A cycles. These photos are for the balls under the die propagate in and out of the void. For the upper right hand
wheremostdamagesoccurduetolocalCTEmismatch. photo of this figure, a crack initiated at the package interface
Photoswithandwithoutvoidswerealsoincludedfor and propagated into a void at a 45 O angle. There is another
comp'xison.Voidsappear to
haveconcentrated at
the voidwhichisparallel to thepackageinterface.The two
packageinterfaceunderthe die. Cracking
occurred at types of cracks representa mix mode stress condition.
package or board interfaces for solder balls with or without
voids. Note also that the large void in the lower left photo shows
evidence of mounting materialsfrom
metallugraphic
preparation. This is puzzling since there is no evidence of a
Thevoidswereconnectedtointerfacenotapparentfrom
through-crack. Perhaps a crack, normal to the photograph's
photos of cross-section, but it is evidenced from the seepage
cross section, has interconnected the void to the outer balls'
of mountingmaterialsintovoids.Exceptfor theinterface
surface. Again. this reveals the complexity of failure modes
connecting cracks, there appeared to be no crack propagation
'amongthevoids. Thesolder joints forthispackagewere
andremindusofcautionthatmust be exercisedwhen
interpretingfailures by two dimensionalcross-sectioning.
alsovisuallyinspectedtoidentifyindicators of defects.
Ideally, a three dimensional characterization is need to better
There were none apparent. Even when these were inspected
interpret the results.
 1 I
Figure 4 SEM and Cross-sections of PBGA 313 after
4,682 cycles under A (-30°C to 100°C) Conditions

Figure 5 Voids in PBGA. Cross-section photos for PBGA 352 and SBGA 352 of Solder Bal; with Void and Cracking
after 4,682 cycles (-3OOC to 100°C)

Package attachment were very weak nCter 4,500 B cycles :utd board, might bc due to higher thermal stability of polyimide
were easily detached from the board. Cycled palck;l,w s were compare to IX-4 i n the n thermal cycle.
detached and balls were inked and their
images were
mapped.Theballmapdistributions for several dctatchctl Note also. separation o f onewholeedgerowfromthe
PBGA 313 I/O packages shown elsewhere [21 aurti Cor package side. 'l'his mightrepresentstheexistence of high
SBGA560 are shown in Figure 6 . AI1 joint failures were not warpage Cor outer p;tckage edge generally observed for the
from the board interface :uld thcrc appeared to some trcntl i u plastic p;~ck:tgcs. Similarly, mappings for otherplastic
their distribution. p;tck;lgcs \vcrc carried out and trends were studied.

Mostballswereseparated f r o m the boartl i1rtcrl';tcc li)r la'or ccrmric p:tck;~gcs, matpping was not possible since most
SBGA560on FK-4 boards ;IS cvidcwcti I'rorn tlrc large of tlrc p;tck;t~cswere Arcadyseparatedfrom theboard at
number of darkspots which rcprcscnts tlrc hAls 0 1 1 tl~c 4,500 cycles. Scpanttions were mostly at thepackage
package. 'This isnottrue tor 11rosc on polyirrritlc ho;lrtl. inlcrCtccs since dctaclrctl packages generally had no balls on
Morejointseparations :I( the p;tck;~~c side li)r polyitrridc tlrcm. 1 1 1 a d d i t i o n , scvcr;d testvehicleboardshad a large
 . number of balls
on
the
location
ceramic
of package attachments.
* -

. .. . . ... .......
. ......... -._.
:..a
.
. .. ..... .. . .. .. .. ......
.... .,.. ..
'I. .?!.. '...:::!:""I: ... :
.
. .. . . . . :...;:,.:::..
:*,.

......
. ...,
,I

.- *.
' ' _

.......
- 11 . . ..... .
....
...
.... .. 1 .
_ .
;I..
.. ....:!
...
.. .. .....,
..... , ,
....
. . . . ... .. .. .........
. . . .:.!:. .
. . ... . ... . . ...... .. ......
P1002-134 P978-178 P167-170
Polyimide Polyimide Polyimide
......... i Solder Balls
........ ..........
.............:+:..:. ................
.... ..: : . ,. .... . .. i
.............
....
...... .......... 'i: I ....................
..:.., On Package
....
.....
...
.....
..
, .. . ._ ._
.....
......
-,,.
-.
.,_... I
....,, ...
.....
a .

-. ...
,..
. .
:
....
(-55"C<>125"C)
.. ., ......
.... ... .. .., . ( 4500 Cycles)
,:.:: .
*.
*..
1"

...... .-
....
.. .. ...... .. ... .. .. .. .. .. .. .. .....
. . . . ........
.....
.......
,.:.. .:.. .. . :..............
....... ........,... .............. . .. .. ..
I. 7.

:a:
...
:>, . ::;.:,,.: ..........
I . :

..rtt:.:;.

P1008-97 P987-182
P101.5-96
FR-4 FR-4
FR-4

Surface Finishes OSP

The surface finishes considered for evaluationwere 180 TVs OSPs, also knownas anti-tarnish, are gaining popularity as a
with Organic Solderability Preservative (OSP), 14 electroless low-cost,highvolumealternativetoHASLandprovide
nickellimmersion gold, and 5 HASLs.Thecriticalprocess excellentsurfacecoplanarityrequiredforfinepitch SMT
factors for BGA assembly as well as reliability issues were applications. They are. waterbasedorganiccompounds
consideredduringtheselection of surfacefinish.Factors (typically
azole-
benzotriazole or benzimidazole)that
such as surfacecoplanarity,soldervolume,solderability, selectivelybondwithcoppertoprovide an organometallic
moisture sensitivity, thermal aging behavior (shelf life) and copperlayer.OSPprocesstemperaturesusuallydonot
second reflows were also considered when the PWB surface exceed 150 "C, markedly lowerthan HASLprocess
finishes were selected. temperatures. Key drawbacks of OSPs include susceptibility
to stain (water,flux, fingerprint) anddeterioration as a
result of temperatureheatcyclestypicallyfound inmixed
HASL technology.
The predominantsurfacefinish in the PWB industryhas
been HASL. In HASL, the tinhead plating is accomplished For example, compatibility of OSP finish with aramid PWB
by dipping the fluxed PWB into or by passing it over molten might bean issue. This is primarily due to the hydroscopic
solder. The excess solder is removed by high velocity hot air nature of aramid whichrequires
prebaking
high
at
which levels the remaining solder on exposed surfaces. The temperaturebeforereflow.ThisprocessdegradestheOSP.
solder coatings are dense and have good adhesion, but are Anotherissuethatmust be considerediscompatibility of
non-uniform and
are
not
suitable
for
someemerging OSPwithflux.Watersolublefinishesaremorelikely to
technologies. form solder balls if the temperature profile and solder mask
design
are
not
optimized. No clean flux with use of
This process is being replaced for several reasons including nitrogenatmosphereflowmight bethechoiceforsome
environmental and
safetfissues
(hazardous wastehead applications.
exposure),
technological limitations(finepitch
device
assembly), and equipment maintenance expenses. Alternate The OSP materials recommended by the PWB fabrictor. was
processeshavebeendevelopedthatareenvironmentally chosen for the test vehicle surface finish. The fabricator had
friendly,andprovideasurfaceplanarityequivalent tothe usedthecoatingforvariousBGAapplications with no
platedcopperfinish,andrequireverylittleequipment reportedproblems.Oneadvantage ofthiscoilting was its
maintenance. more tolerant to higher temperatures commonly experienced
during the drying stage.
 Vendor data sheets indicate that the coating is a water-bnsed A cross-section photo with elements in each phase shown in
~ OSP used to protectandmaintainthesolderability of printed Figure 8. Qualitative elements
detected by probing at
circuitboardcopper surfaces. Theactiveingredientis various sections ofthe joint are also determined. No traces
designed to react with copper and copper alloys forming a of goldwasdetected,possiblybecausethe low amount of
thin uniform coating on circuitry. It is also report that the goldcouldhavebeendissolvedduringthereflowprocess.
coating was compatible with R , RMA, OA, No-Residue, and Other elements were as expected and included Cu, Ni, P, Pb,
No-Cleansolderpastesandfluxes. It has excellentsolder and Sn. Existence of phospl~orus traces indicates
the
wetting even after multiple reflow operations. Thickness is chemistry ofthebathusedforNi plating. This element is
typically 6-8 microinch (0.15 to 0.18 micron). found in a common chemistry used for the Ni surface finish.

NilAu Evaluation
ofFurtherCause of Brittle Failure
Although electroless
nickelhmmersion goldare
more Elemental characterization of brittle surfaces was obscured
complex thanOSP,theysatisfy
the
application-specific by the metallugraphic mounting materials. Results were not
requirements for fine pitchandchip-on-board (COB) accurate. A virginsurfacewiththebrittlebehavior was
assemblies. This process is atwo-layer,gold-over-nickel needed for a more accurate surface analysis.
metallicsurfacefinishplated onto thecopperbase by
chemical vapor deposition. Nickel is deposited on copper by Remain of this package was pull tested under sustaining load
an electroless process resulting in a deposit thickness of 100 of 7 Ib. The load was small and no separation was observed
to 200 microinches. after two monts at roomtemperature. No separationalso
after 5 daysexposure at 100°Cunderthesameload. The
The gold is then deposited in an exchange reaction replacing package was finallypriedoff.Thistechniques was not
nickel on the entire surface with a thin layer of pure gold. attempted at thebeginninginorder toavoid potential
Thegoldthickness is typicallyintherangeof 3 to 5 damage to solder joint.
microinches. Eventhoughthe
gold
thickness
not
is
significant compared to that typically used for solder paste Out of 525 separated balls, only 13 balls remained on board
volume,thepotential of solder joint embrittlementdue to andthemajority(525-13= 512) were stillattachedto
formation of intermetalliccompounds withtinmay be of package.Out of 512padseparation onboard,470were
concern. Ni/Au coatings are not reworkable, and are pulled from traces and 42 wereby solder joint separation.
relativelyexpensivein
terms of waste treatmentand Out of these 42, only 3 pad showed partial signs of brittle
processing. The thickness of the gold was 3-5 microinches. failure. The crescent surfaces were from 1/3 to about 1/4 of
Ni/Au surface finish was carried out by the same company the pad area.
that did theOSP coating.
Figure 9 showstheSEMphotographsforbrittlefailure
feature on ball and package pad. Figure 10 shows elemental
Failure Mechanisms ForDifferent SurfaceFinishes analysis ofthe ball and pad surfaces by ESCA. For ductile
Ceramicpackagefailureunder A cyclingconditionfor failureregions,theelements werePbandSnofeutectic
PWBs with different surface finishes are shown in Figure 7. soldercomposition as expected. This was notthecasefor
BothHASL andOSPfailures,initiated at the PWB side, brittle
surfaces. At
the
smooth
ball
surface, dements
showed ductile
featuresevidenced from
the
zig zag included Ni, P, and Sn whereas at the pad surface, they were
propagation through the eutectic solderjoint. Generally, this mostly Ni. Pimples on the ball surface were Sn.
was not the case for Ni/Au. For Ni/Au, several joints of a
packageshowedbrittlefailure,whichwerefirstnoticed
during a routine visual inspection of assemblies. Failure of PlasticPackage with NVAu, OSP, HAPL
Figure 1 1 , shows cross-sectional micrographs ofPBGA 256
Brittlefailure by thegoldembrittlementcould be one after 6,800 A cycles.Photos are forthreetypesofsurface
possible explanation, ifthe amount ofgoldexceededthree finishes,OSP,Ni/Au,and HASL.Thefailurearevery
volume percentages of the solder joint. This is not the case, similar to those show previously for assemblies after 4,800
since thethicknessofgoldwascontrolled by immersion cycles with the OSP finish. There is no apparent differences
techniqueto3-5 pinch. Toassurethis was thecase,a on failuremechanismsamongthethreesurfacefinishes.
sample joint of Ni/Au surface finish was analyzed for various Ductilefailure is tile principalfailuremechanismforthe
trace
elements at PWB interfaces using
the
Energy threesurfacefinishes.Similiuly to otherPlasticpackages,
Spectroscopy for Chemical Analysis(ESCA) feature of SEM. failures were either from package or board side. Generally,
voids were accumulated at the package interface.
 HASL NVAu
I
Figure 7 Failure Mechanisms for Joints on OSP, HASL, and Ni/Au Surface Finishes for CBGA 625 under A Conditions
(-30°C to100°C)

Figure 8 Magnified Photos of Ni/AuInterface Surface Finish

HASL NVAu
 Ni/Au, Ball Failure Feature
1 I
Figure 9 SEM photomicrograph of detached ofceramic package on Ni/Au surfacefinish with brittle areas

Ni/Au, Board Surface Chemistry Ni/Au, Ball Surface"Chemistry

Figure 10 Elemental chemistry of brittle surface shown in Figure 9

cycling temperature ranges of -55 to 125 "C and

-30 to 100lx:.

' l l ~ c 1'1X;As with 313 I / O depopulated full

arrays, were first anlong the PBGAs to fail
(SIKiA 560, SMiA 352, OMPAC 352, PBGA
3 13, and I W i A 256) with both cyclingranges
( A and 11 contlitions). I t has been
well
esl:lblishcd h l t Ihisconfiguration,withsolder
b;1lls untlcr thc die, is notoptimumfroma
rcli;thility point ol'vicw.
 Forperipheralplasticpackages,thesizeand Pulltest of assembly with Ni/Aubrittlefailure
number of I/Os appeared to be the key factors in revealed only three localize brittle failure out 525
cyclestofailurefor B conditions. As thesize separated sites. This means potential of tinding
increased,cyclestofailuredecreased,Le.,the such
randomfailure by a few cross-section
SBGA 560 was first to fail in this category and examination is very low. Brittle
failure
has
PBGA256wasthe last.Thesetwoassemblies shown by bend test by others.
alsoshowedlargescatters incyclesto failures.
Elementalanalysis of onebrittle Ni/Ausurface
The
reasons
for
the
large
failure
scatters
finish revealed no detectable gold, but reasonable
especiallyfor
the
PBGA 256,
are
not
well
amount of phosphorand Nionthedetached
understood.
solderballsurface.Phosphoroussegregation at
A SBGA 560 packageassembly
showed interface,surfacecontaminationduringNi/Au
numerouselectricalinterruptionswithsigns of plating or after plating by diffusion, and brittle
electrical opens at an unexpectedly low number fracture of Ni-PandNi3S&wereshownthe
of A cycles(1,810).Destructivedyepenetrant threepotential of failure mechanisms [3]. This
testingrevealednosolder joint failure. It was is forthecasewhentheNi/Aucontentisless
postulated that failure could have caused by die than 1% andthereisnopotentialofgold
debonding or other package internal failure. embrittlement.
PulltestingperformedonaSBGA560after
1,810 A cycles.Failuresitesonboardwere
identifiedandthedistribution of solderball REFERENCES
height attached to these sites were compared to 1. Ghaffarian,R.“Reliability of
BGA Packages
the distribution of all other balls. No correlation forHighlyReliableApplicationsandChipScale
between height and failure sites was found. Package Board Level Reliability,” Proceedingsof the
ChipScalePackagingSymposium,SurfaceMount
Failure Mechanisms International (SMI),Sept. 7-11, 1997
For plastic assembly, there were no indicators for 2. Ghaffarian,
R.,
Kim,
N.P.
“Reliabilityand
early failure warning. This was not the case for FailureAnalyses of ThermallyCycled Ball Grid
ceramicpackages.IndicatorsforCBGAswere ArrayAssemblies,”48thElectronicComponents &
due to differences in the metallurgy of ball and Technology Conference (ECTC), May25-28,1998
solder
and
also
because
the
first
signs of 3. Mei, Z., etal“BrittleInterfacialFracture of
damagesoccurinthecomer joints withthe PBGA
Packages
Electroless
Soldered
on
highest DNP. The ceramic joints showed signs Nickel/Immersion
Gold,” 48th Electronic
of graingrowth,pinholeformation,creeping, Components & Technology Conference (ECTC),
microcracking and cracking prior to failure. May 25-28,1998
plastic
For
packages,
crack
initiation,
propagation,andfailureoccurred at either the
package or board interfaces for sections with or ACKNOWLEDGMENTS
without voids. These were true for A or B cycle Theresearchdescribedinthispublicationisbeing
conditions.Generally, voidswereconcentrated carried
out by the JetPropulsion
Laboratory,
near thepackageinterfaces.Thereappearedto California Institute of Technology, under a contract
be no crackpropagationsamongthevoids, National
with
theAeronautics
Space
and
except
for
the
voids
interconnected
the
at Administration.
interface..
I would like to acknowledge the in-kind contributions
For a ceramicassemblyfailure,brittlefailure andcooperativeefforts of
BGA consortium te‘am
wasobservedforNi/Ausurface finishes. OSP membersincludingandwhohadbeenorarebeing
and HASL showed ductile
failure
through contributing to the progress of the progrm.
eutectic solder joint. For plastic packages, there
were no distinction for the three surfaces.