Flex Cracking of Multilayer Ceramic Capacitors Assembled With Pb-Free and Tin\u0026#x2013;Lead Solders

182 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

Flex Cracking of Multilayer Ceramic CapacitorsAssembled With Pb-Free and Tin–Lead Solders

Mohammadreza Keimasi, Michael H. Azarian, Member, IEEE, and Michael G. Pecht, Fellow, IEEE

Abstract—In this paper, an experimental study was conductedto study susceptibility to flex cracking of multilayer ceramic ca-pacitors (MLCCs), in which a comparison was made betweenidentical samples which were assembled using either Pb-free(Sn3.0Ag0.5Cu) or eutectic tin–lead (Sn37Pb) solder. Flex testingwas performed on MLCCs with different sizes (1812 and 0805)and on different dielectric materials (C0G and X7R) obtainedfrom three different manufacturers. Experimental results showedthat MLCCs mounted on printed circuit boards (PCBs) withPb-free solder crack less with board flexing than those mountedon boards with eutectic tin–lead solder. For 1812-size MLCCsassembled with tin–lead solder, the PCB strain at 10% failureranged between 1700 and 2000 microstrains. The PCB strain at10% failure for 1812-size MLCCs assembled with Pb-free soldervaried between 2300 and 9600 microstrains, depending on theMLCC manufacturer. C0G MLCCs are more resistant to flexcracking than X7R MLCCs. Out of 96 samples tested, none of theC0G MLCCs showed evidence of flex cracking up to the maximumstrain level on board of about 137 000 microstrains. In contrast, inthe case of X7R MLCCs, from the same manufacturer, assembledwith tin–lead solder, 94 out of 96 capacitors failed.

Index Terms—Capacitors, ceramic capacitors, crack detection,Pb-free solder, reliability testing.

I. INTRODUCTION

CAPACITORS are an indispensable component in elec-tronic circuits. The worldwide market in 2004 for tanta-

lum, aluminum, and ceramic capacitors was almost 9 billionU.S. dollars per year [1]. Revenues of ceramic-capacitor man-ufacturers reached 5.5 billion dollars in 2004 [2], with a salesgrowth of 25% per year [3]. However, based on the analysisby the Center for Advanced Life Cycle Engineering (CALCE)of over 150 electronic-product failures over a four-year period,capacitors are responsible for a larger proportion of failuresthan any other component or failure site.

Multilayer ceramic capacitors (MLCCs) consist of formu-lated ceramic dielectric materials, which have been fabricatedinto thin layers, interspersed with metal electrodes which are

Manuscript received March 28, 2006; revised April 14, 2007. This work wassupported by more than 50 members of the CALCE Electronic Products andSystems Consortium at the University of Maryland.

M. Keimasi was with the Center for Advanced Life Cycle Engineering,Department of Mechanical Engineering, University of Maryland, College Park,MD 20742 USA. He is now with the Dell, Inc., Round Rock, TX 78682-2222USA (e-mail: [email protected]).

M. H. Azarian and M. G. Pecht are with the Center for Advanced LifeCycle Engineering, Department of Mechanical Engineering, University ofMaryland, College Park, MD 20742 USA (e-mail: [email protected];[email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TDMR.2007.912256

Fig. 1. Flex cracking of MLCCs due to PCB bending.

alternately exposed on opposite edges of the laminated structure(Fig. 1). The entire structure is fired at high temperatures,typically above 850 ◦C, to produce a block which provides thedesired capacitance values in a small physical volume. Afterfiring, conductive terminations, typically made of copper orsilver, are applied to opposite ends of the chip to make contactwith the exposed electrodes. The end terminations are usuallyoverplated with a nickel-barrier layer and tin to provide goodsolderability.

Bending of a printed circuit board (PCB), as shown inFig. 1, will cause stresses to be transmitted through the solderfillets to the surface-mounted capacitor. When the capacitoris on the convex side of the PCB, tensile stresses exhibit amaximum at the bottom of the capacitor near the edge of thecapacitor end-termination. The imposed tensile stress insidethe capacitor is reduced due to plastic deformation in thesolder. If the tensile stress applied to the capacitor body exceedsits fracture strength, the capacitor will crack. The stress atwhich the capacitor cracks is dependent upon the ceramicmaterial in the capacitor body, the solder material and filletsize, the termination materials, and defects within the ceramicstructure [4].

Stable temperature-compensating dielectrics or C0G (EIAclass I [5]) are made with less barium titanate (BaTiO3) contentin comparison to EIA class-II dielectrics, but are basically othertitanates, particularly neodymium titanium oxide (Nd2Ti2O7),TiO2, or CaTiO3 with additive materials. EIA class-II (e.g.,X7R) general-purpose capacitors use BaTiO3 as a base, towhich stabilizing materials are added, because of its highdielectric constant [6], [7]. For X7R dielectrics, the bariumtitanate content is about 90%–98%, while for C0G dielectric,the barium titanate content varies from 10% to 50% [7]. Capac-itors with X7R dielectric are known to be more susceptible toflex cracking than those with C0G dielectric, which has greaterfracture toughness [8], [9].

Fig. 2 shows a flex crack at one of the MLCC end-terminations. The crack typically starts near the edge of thetermination margin and, then, extends toward the termination

1530-4388/$25.00 © 2008 IEEE

KEIMASI et al.: FLEX CRACKING OF MLCCS ASSEMBLED WITH Pb-FREE AND TIN–LEAD SOLDERS 183

Fig. 2. Flex-cracking failure of a multilayer ceramic chip capacitor.

Fig. 3. Mechanical failures in MLCCs.

face. The angle of this crack is often around 45 ◦. The crack mayextend into the termination face, thereby separating a cornersection. It may also turn toward the top of the capacitor, where itwill usually turn out toward the top termination margin’s edge.In some cases, this crack can cause the entire end of the capac-itor to be separated from the main body of the capacitor [4].In most cases, cracks are visually undetectable at the exteriorof the cracked capacitor, and cross sectioning is necessary tovisualize a flex crack.

A breakdown of about 40 different MLCC mechanical fail-ures that were analyzed over several years by the failure-analysis laboratory at the CALCE, University of Maryland,is shown in Fig. 3. The largest root causes of failures inMLCCs are defects introduced during capacitor manufacturingprocesses, including voids and cracks. The second most com-mon cause of failure of MLCCs is cracking due to excessiveboard flexure, or “flex cracking,” and includes about one quarterof failures. Cracking of MLCCs due to thermal shock in the

soldering process or field is the third most common cause offailure.

Flex cracking can occur during assembly processes andhandling, as a result of excessive flexure of PCBs. Examples ofprocesses causing board flexure include PCB warping duringsolder reflow, depaneling, handling during processing, testing,assembly, insertion or removal of a board from edge-mountconnectors, attachment of a board to other structures (e.g.,support plates, heat sinks, and chassis), connection of cables,thermal expansion of a board with respect to a cabinet orchassis, and vibration or shock in the end-use environment.

II. BACKGROUND

Prymak and Berganthal [4] conducted flex tests on different-size MLCCs. They used a three-point bend configuration withone capacitor mounted in the middle of each board. In situmeasurement of capacitance was used as means of detectingthe occurrence of flex cracking. They showed that, for some ca-pacitors, a capacitance drop observed during testing recoveredafter removal of flexure. This suggested that in situ monitoringof capacitance during bend testing was necessary. Their resultsalso showed that the larger the capacitor, the more susceptibleit was to flex cracking. Their experiments were repeated forcapacitors from production batches spanning a 25-week period.Their results remained consistent, which led them to believethat the materials and manufacturing processes were consistent.

Berganthal [10] assessed the dependence of flex crackingof ceramic capacitors on several different parameters. Solderfillet size was found to impact bending strength, with largerfillets causing a reduction in strength. Pad size had a similareffect, with narrower pads increasing the bending strength.This is in agreement with Kemet, which recommended thatwave solder pads should be narrower than reflow solder pads.Chip length was also found to affect bending strength, withlarger chip capacitors being more susceptible to flex cracking.Dielectric materials were found to impact bending strengthbased on their fracture toughness; the C0G dielectric is superiorto the X7R dielectric, and X7R is better than Z5U and Y5Vdielectrics. The PCB thickness also impacts bending strength.Thicker PCBs are more rigid and transmit more stress to acapacitor at a given deflection. Therefore, thicker PCBs resultin flex cracking at lower deflections than thinner boards of thesame materials. The PCB material affects the bending strengthprimarily through the Young’s modulus. Some data indicatethat larger capacitor thicknesses increase bending strength, butthere appears to be some confusion about this parameter. Somecapacitors have thicker cover layers around the active layers.These parts may have greater bending strength than those withthinner cover layers. These results may also arise if a capacitorhas cracked prior to an observable drop in its capacitance.Finally, capacitors with wider termination bands typically havegreater bending strength. The wider band has the same effect asreducing the capacitor length.

Prymak [11] investigated reverse flex testing of MLCCs. Ina typical flex test, the capacitor is on the convex side of thePCB, but in a reverse flex test, the capacitor is on the concaveside of the board. In each instance, the capacitor failed during

184 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

the withdrawal of the ram and relaxation of flexure. During theapplication of reverse flex, the capacitor is predominantly undercompression. The compressive strength of ceramics is largeenough that these stresses will not usually damage the capacitorunless it contains a defect, such as an internal delamination. Inthis configuration, the solder fillets deform outwardly to relievethe stress applied to them. When the board flexure is rapidlyreduced, the deformed fillets cause tensile stresses to be appliedto the capacitor, and failure occurs with a similar characteristiccrack to that observed with normal flex testing. Prymak [11]mentioned that it is unlikely that reverse flex testing providesadditional chip-related information beyond that obtained withnormal flex testing.

Prymak [12] exposed capacitors to a humid environmentafter flex testing to detect “hidden” cracks as low insulationresistance (IR). He showed that, of the capacitors specified as“cracked” by a sudden capacitance change during flex testing,80% were found to be IR rejects when exposed to humidity.The reduction in IR is due to penetration of moisture into theexisting cracks, possibly accompanied by capillary condensa-tion. Destructive physical analysis revealed internal cracking inthe remaining 20%.

Al-Saffar et al. [13] investigated the flexure strength, ormodulus of rupture (MOR), of MLCCs using three-point bendtests. Their samples included blanks (dielectric material withoutelectrodes) and MLCCs with X7R and Z5U (an EIA class-IIIdielectric [5]) dielectrics containing different numbers of elec-trodes and ink-laydown concentrations (electrode material,which is screened onto the dried ceramic dielectric duringmanufacturing of MLCCs). It was found that, for blanks, MORdecreased as the specimen thickness increased, and the X7Rdielectric had greater flexure strength than the Z5U dielectric.For X7R capacitors, the MOR increased with the number ofelectrodes, while for Z5U capacitors, the MOR was indepen-dent of the number of electrodes, and all values were less thanthe values obtained for X7R capacitors. Electron microscopyshowed major differences in the mode of fracture of Z5Uand X7R MLCCs. In their study, for as-fired X7R capacitorswithout terminations, the MOR increased with the ink-laydownconcentration to a plateau level (∼140 MPa), but in capacitorswhich had been terminated and annealed, the flexure strengthwas uniformly greater than capacitors without terminations andindependent of ink-laydown concentration.

Maxwell [14] compared flexure strength of similarly sizedmultilayer ceramic and film capacitors. Ceramic capacitorsare brittle and crack in response to excessive PCB bending.Multilayer film capacitors are made with polymer films, arenot brittle under normal conditions, and are more flexible whenstressed on a bending PCB. He concluded that surface-mountfilm capacitors did not exhibit failure or degradation whentested at or beyond deflection values that cracked ceramiccapacitors of similar size and capacitance value.

Syfer Technology Limited [15], a manufacturer of ceramiccapacitors, reported that, when capacitors are broken, the prob-lem usually manifests itself at a very late stage of board assem-bly. Syfer’s observation is likely due to the fact that capacitancedrops resulting from flexure are frequently not detectable afterremoval of the bending stress. Another factor that can lead to

this phenomenon is the growth of the small cracks introducedby flexure as the board is exposed to additional thermal andmechanical stresses during assembly. An immediate and lastingchange in any key capacitor parameter is rare. A decline in theIR of a cracked capacitor requires penetration of a conductivemedium, which is often atmospheric moisture, into the crackstructure, and this takes time. Hence, detection usually occurslate, such as in the field. Typically, 60% of parts damaged dueto flex cracking do exhibit a detectable change in IR but only asmall minority of these parts is preidentified by a user as havingpotentially failed. Cracks are visible at the exterior of less than2% of affected parts, and immediate detection of change of thecapacitance is a feature of about 10% of cracked capacitors.Syfer conducted experiments to study the effects of differentparameters on flex cracking of ceramic capacitors. Their resultsshowed that the only significant difference in strength, acrossa broad matrix of capacitor design and build parameters, liesbetween barium-titanate-based capacitors (the material usedin X7R and Y5V dielectrics) and neodymium-titanium-oxide-based capacitors (the basis for their C0G dielectric). The C0Gdielectrics failed at deflections, which were twice those of theX7R and Y5V dielectrics. Most ceramic formulations withina given dielectric category were found to have similar bendstrengths. Their results indicated that small capacitors are notstronger than large capacitors, and thin capacitors are notweaker than thick capacitors. It was shown that large solderjoints do have a negative effect on bend strength. A minor dif-ference was observed for flexure strength of parts soldered withSn36Pb2Ag versus Sn40Pb alloys; however, the performanceof a “soft” solder, In50Pb, was much better. Average deflectionat failure was more than double for this soft solder. They alsostudied the effects of solder-pad geometry on flex performance.Pad widths narrower than the chip width were found to increasebend strength.

Nies and Maxwell [16] studied factors in board flexuretesting of surface-mount ceramic capacitors. They studied theeffects of PCB material, board thickness, and solder materialon flexure strength of ceramic capacitors. They used boardswith different constructions (six, seven, and eight ply) anddifferent elastic moduli. They concluded that MLCCs withstandlarger deflections when the board is thin and compliant, sincelower strains are generated and, thus, less stress is transferredto the capacitor. On the other hand, only a relatively smallload is needed to deflect thin boards, so flexible boards arenot appropriate for boards experiencing high loads. In contrast,thicker less flexible boards may cause fewer failures for agiven load because their deflection is less but would causemore failures at a given deflection because this translates to alarger strain on a thick board than on a thin one. Therefore,information on either load or deflection alone is not sufficientto generalize results obtained using a specific board in orderto establish the allowable range of conditions to minimize therisk of cracking. The authors also studied the effect of soldermaterial on flex strength of ceramic capacitors. Four soldercompositions as follows: Sn37Pb, Sn40Pb, Sn36Pb2Ag, andSn4Ag were used in their experiments. They used both high-fire (> 1300 ◦C) capacitors with palladium electrodes and low-fire (< 1200 ◦C) capacitors with silver–palladium electrodes.

KEIMASI et al.: FLEX CRACKING OF MLCCS ASSEMBLED WITH Pb-FREE AND TIN–LEAD SOLDERS 185

Even though silver-doped solders showed somewhat betterperformance, their standard deviations were large enough tomake the statistical significance of this difference questionable.

Long et al. [17] conducted experiments with both eutectictin–lead (Sn37Pb) and Pb-free solders (Sn3.0Ag0.5Cu) tocompare effects of solder material on flex cracking ofMLCCs with X7R dielectric manufactured by Kemet. Theirresults were mixed, because for some case sizes, Pb-freesolder performed better, and for others, the eutectic tin–leadsolder performed better. Their results for 1812-size capacitorswith X7R dielectric showed that Pb-free solder gave betterflex performance. For 0603 capacitors with X7R dielectric,Pb-free and tin–lead solders produced very similar results. For0805-size capacitors, although the capacitors assembled withPb-free solder performed better, the difference between thetwo solders at 100-ppm failure rate was less than 0.3 mmof deflection. For 1206-size capacitors, the tin–lead solderedparts performed better at low deflections, whereas above2.6 mm of deflection, the Pb-free soldered parts performedbetter. Based on their investigation of the solder joints, theyfound that tin–lead solder had better wetting over the wholepad area, while Pb-free solder wetting was limited to part ofthe solder pad. The reduced wetting with the Pb-free soldersreduced the effective pad width. The authors hypothesized thatthis affects the flex performance of ceramic capacitors.

In this paper, the main focus was to investigate differencesin flex cracking of MLCCs assembled with Pb-free solder ascompared with eutectic tin–lead solder through a systematicexamination of factors which had not been studied previouslyin this context. Capacitors from several manufacturers wereincluded in order to incorporate variations in materials andmanufacturing processes, which provides insight into the rangeof behavior possible from commercially available components.MLCCs with C0G dielectric were tested in addition to thosewith X7R dielectric, which is assembled with both types ofsolder. Finally, experimental results from two different capaci-tor sizes were compared.

III. EXPERIMENTAL PROCEDURE

In order to study flex cracking of ceramic capacitors, PCBswere designed and built using high glass transition tempera-ture (Tg = 170 ◦C) FR4 material (glass-fiber-reinforced epoxy-resin matrix). On each board, 24 capacitors were assembled.There were two board designs (similar to each other but withdifferent pad sizes) for 0805 and 1812 capacitors. The layoutof the test board for 1812-size MLCCs is shown in Fig. 4.A Pb-free solder (Sn3.0Ag0.5Cu) and eutectic tin–lead solder(Sn37Pb) were used to assemble capacitors on the boards usingconvective solder reflow. The temperature profiles of the reflowprocess for these two different solder materials were different.The peak reflow temperature for eutectic tin–lead solder wasbetween 210 ◦C and 230 ◦C, while for Sn3.0Ag0.5Cu Pb-freesolder, it was between 235 ◦C and 255 ◦C.

After assembling capacitors, boards were subjected to deflec-tion based on IPC/JEDEC standard 9702 [18] using the four-point bend test shown in Fig. 5. Vertical displacement of loadrams versus time on an MTS (r) hydraulic tester during a flex

Fig. 4. Test board for flex-cracking test of 1812-size MLCCs. The thicknessof the boards is 1.57 mm.

Fig. 5. Schematic of the experimental setup for flex testing of MLCCs.

Fig. 6. Vertical-load ram displacement versus time during flex testing ofMLCCs.

test is shown in Fig. 6. The ram displacement was increased in0.2-mm steps at a rate of 4 mm/s and then held at that positionfor 10 s to allow capacitance measurement of all 24 capacitorsassembled on the board at each step. Fig. 5 shows a photographthat was taken during actual bend testing at the maximum ramdisplacement of 10 mm. After the ram reached its maximumdisplacement for 10 s, it was brought back to zero displacementin one step at a rate of 4 mm/s.

186 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

TABLE IEXPERIMENTAL DESIGN FOR FLEX TESTING OF MLCCS

In order to measure PCB strain during flex testing, straingauges were mounted on each board. Since the net verticalforce between the two load rams of the four-point bend test iszero, the radius of curvature between the load rams, where thecapacitors were mounted, is theoretically constant.

The capacitance of each capacitor was monitored in situthroughout the flex test in order to detect the occurrence offlex cracking. A multiplexing switch was used to scan all thecapacitors mounted on the board and measure each capacitorusing an LCR meter. Capacitance is expected to drop uponcracking of a capacitor if a crack passes through the electrodesof the capacitor. Recovery after the end of the flex test forsome capacitors necessitates in situ monitoring of capacitance.In addition, after each capacitance measurement, the strain onboard was measured to determine the strain level that causeda capacitor to crack. A software program was developed tocontrol the LCR meter, switch, and digital multimeter.

Strain was measured for each column individually on oneboard, with strain gauges placed in the locations shown in thecenterline of the board in Fig. 4. From these measurements, itwas found that strain varied by a maximum of about 12% fromthe average, although this difference remained below 10% fornearly all deflection levels. Analysis of variance for strain-to-failure data from each of the three columns of capacitors on aPCB revealed that these three distributions belong to the samepopulation.

A. Experimental Design for Flex Test of MLCCs

The experiment addressed the following three parametersknown to affect flex cracking of MLCCs: solder material, di-electric material, and capacitor size. Two solder materials wereused to assemble the capacitors on PCBs, which are eutectictin–lead solder (Sn37Pb) and a Pb-free solder (Sn3.0Ag0.5Cu),to compare the effect of assembly with Pb-free solder toeutectic tin–lead solder. Two capacitor sizes were selected,

0805 and 1812, to allow comparison of susceptibility to flexcracking of a larger size capacitor (1812) with a smaller sizecapacitor (0805). Finally, two dielectrics were considered, C0Gand X7R, in order to compare flex cracking of two commonlyused dielectric materials with different mechanical properties.

The sample matrix for the flex tests is shown in Table I.Capacitors from three manufacturers, AVX, Kemet, and Vishay,were studied. These manufacturers have been referred to in thispaper as “A,” “B,” and “C” in arbitrary order. All capacitorsused in this paper were obtained from authorized distributors,and they are standard production parts. Initially, 1812 capaci-tors from manufacturer “A,” with two different dielectrics (X7Rand C0G), were tested using both solder materials (Pb-free andtin–lead). As shown in Table I, four boards with 24 capacitorson each board were tested for each case. In addition, capacitorsfrom manufacturer “A” with size 0805 and dielectric X7R weretested using both solder materials.

In order to compare flex cracking of ceramic capacitors fromdifferent manufacturers, capacitors with size 1812 and dielec-tric X7R from the two other manufacturers (“B” and “C”) weretested using both solder materials. The larger capacitor size wasselected, because it is more susceptible to flex cracking.

This experimental design is equivalent to a combinationof three full factorials, including a 2 × 3 design on soldercomposition (Sn37Pb versus Sn3.0Ag0.5Cu) and manufacturer(“A,” “B,” and “C”) for X7R 1812 capacitors, a 2 × 2 designon solder composition and size (1812 versus 0805) using man-ufacturer “A,” and a 2 × 2 design on solder composition anddielectric (X7R and C0G) also with manufacturer “A.”

Some parameters affecting flex cracking of MLCCs were ex-cluded from the experimental design. These parameters includethe PWB thickness, the PWB material, the ratio of pad width tocapacitor width, and the ratio of pad-to-pad separation distanceto capacitor length. The capacitors which have been selected fortesting have a capacitance of 0.1 µF except capacitors with C0Gdielectric, for which a value of capacitance has been chosen to

KEIMASI et al.: FLEX CRACKING OF MLCCS ASSEMBLED WITH Pb-FREE AND TIN–LEAD SOLDERS 187

Fig. 7. Capacitance change and strain on the board as a function of timeduring flex testing of MLCCs. The capacitance drops due to flex crackingand recovers after removal of the strain. C0 is the nominal capacitance of thecapacitor, which is equal to 0.1 µF for the capacitor shown here.

Fig. 8. Capacitance change and strain on board as a function of time duringflex testing of MLCCs. The capacitance drops due to flex cracking and does notrecover after removal of the strain.

keep the thickness of the capacitors the same as that for 1812-size capacitors with X7R dielectric selected for testing (seeTable I).

IV. EXPERIMENTAL RESULTS

This section presents plots of cumulative percent failure asa function of PCB strain for capacitors assembled using twodifferent solder materials. The results are compared to examinethe effects of solder material, capacitor size, dielectric material,and manufacturer.

The capacitance of each capacitor was measured duringboard bending to detect crack formation. Figs. 7 and 8 showcapacitance change and strain on the board as a function oftime during flex testing. In Fig. 7, capacitance drops uponthe occurrence of the flex crack and recovers after removal ofthe strain. This behavior would be indicative of a “walking-wounded” in field applications, because a crack exists. If acracked capacitor is used in a field application, a conductivemedium, which is often atmospheric moisture, can penetratethrough the crack into the capacitor and cause leakage currentof the capacitor to increase. This may also lead to the shortingof the opposing electrodes of the capacitor, ultimately causingcatastrophic failure in applications such as those involving high

Fig. 9. Comparison of cumulative percent failure as a function of strain onthe board for 1812-size X7R capacitors from manufacturer “A” mounted on theboards with Sn37Pb and Sn3.0Ag0.5Cu solders.

power, in which the short circuit may initiate a fire. In Fig. 8,capacitance drops upon the occurrence of the flex crack anddoes not recover. Cross sections and optical microscopy oftested capacitors verified conclusions of in situ measurementsof capacitance. The failure criterion for flex cracking waschosen to be a 10% drop in capacitance relative to the nominalvalue of capacitance according to the manufacturer’s datasheetfor the part, indicating the point at which the capacitance fellout of specification. In practice, the high rate of crack growthfor these brittle components meant that the final results werealmost insensitive to whether capacitance drop was calculatedrelative to the nominal value or to the initial measured value.

A. Comparison of the Effects of the Solder Material on FlexCracking of MLCCs

Flex cracking of MLCCs mounted on boards with eutectictin–lead solder and Sn3.0Ag0.5Cu Pb-free solder was studied.Fig. 9 shows the effects of Sn37Pb and Sn3.0Ag0.5Cu solderson the flex cracking of capacitors with 1812 size and X7Rdielectric from manufacturer “A.” MLCCs mounted on boardswith Pb-free solder showed less susceptibility to flex crackingthan capacitors mounted with eutectic tin–lead solder, and thiswas true for capacitors from the other two manufacturers. Asimilar trend to that observed for 1812-size capacitors was alsoobserved for 0805-size capacitors.

Figs. 10 and 11 show flex-cracking susceptibility of capaci-tors with 1812 size and X7R dielectric from the three differentmanufacturers mounted on boards with tin–lead and Pb-freesolders, respectively. The results from three manufacturers arecomparable with each other in the case of tin–lead solder,while among capacitors mounted on boards with Pb-free solder,capacitors from manufacturer “A” showed greater susceptibilityto flex cracking as compared to capacitors from manufacturers

188 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

Fig. 10. Comparison of cumulative percent failure as a function of strain onthe board for 1812-size X7R capacitors from manufacturers “A,” “B,” and “C”mounted on the boards with Sn37Pb solder.

Fig. 11. Comparison of cumulative percent failure as a function of strain onthe board for 1812-size X7R capacitors from manufacturers “A,” “B,” and “C”mounted on the boards with Sn3.0Ag0.5Cu solder.

“B” and “C.” The differences observed with Pb-free solderamong capacitors from different manufacturers may be relatedto solder-fillet variations or to their response to the greatersolidus temperature for Pb-free solder.

The solder-fillet geometries for MLCCs from different man-ufacturers were investigated to explore possible correlationswith flex-cracking susceptibility. Three parameters, which arethe maximum height of the solder fillet, solder wetting angleat the top of the fillet, and stand-off height of the capac-itor body above the solder pad surface, were measured forcross-sectioned MLCCs. Upon comparison of these solder-fillet parameters for MLCCs assembled with eutectic tin–lead

solder and Pb-free solder, none of these parameters exhibited astatistical difference for these two solders. Furthermore, solder-fillet geometrical variations among capacitors from differentmanufacturers were not statistically significant.

During the solder reflow process, the PCB and capacitorsexpand and contract as a result of temperature changes. Thecoefficient of thermal expansion (CTE) for an X7R MLCC wasmeasured to be between 6.5 and 9.6 ppm/◦C for the temper-ature range of 25 ◦C − 200 ◦C. For an FR4 PCB in the planeof the board, the CTE below the glass transition temperature isaround 15 ppm/◦C, and the corresponding value above the glasstransition temperature is around 20 ppm/◦C [20]. Therefore,the FR4 board expands more than the capacitors during theheating phase of the reflow process, particularly when thetemperature is greater than the glass transition temperature.During the cool-down phase of the solder reflow process, theboard and capacitors contract. Before the solder completelysolidifies, the amount of stress that is transmitted to the capac-itor body is negligible. Below the solidus temperature of thesolder when the last remaining liquid has solidified, stresses aretransmitted through the solder material to the capacitor body.Franken et al. [19] used finite-element analysis to show that,after solder reflow, the bottom-cover layer (PCB side) of anMLCC is under compression, and the top cover layer is intension.

The eutectic temperature for tin–lead solder is 183 ◦C, whilethe corresponding value for tin–silver–copper Pb-free solderis 217 ◦C [20]. For the near-eutectic Sn-3.0Ag-0.5Cu alloyused in this paper, the solidus temperature is approximatelythe same as the eutectic temperature. Therefore, during thecool-down process, the Pb-free solder completes solidificationat about 50 ◦C above the glass transition temperature of theFR4 board (Tg = 170 ◦C), while eutectic tin–lead solder iscompletely solidified at only 13 ◦C above the glass transitiontemperature. As a result of this difference, boards assembledwith Pb-free solder impose more residual compressive stresson the capacitors than boards with tin–lead solder.

Bending of the PCB generates tensile stresses in the bottom-cover layer and compressive stresses in the top-cover layer [19].The stresses due to PCB bending are offset by stresses due tothe soldering process. If this results in a net tensile stress greaterthan the fracture strength of the dielectric, a crack will initiate atthe bottom of the capacitor at the edge of the end-termination.Since the capacitors assembled with Pb-free solder have greaterresidual compressive stresses, they require more applied bend-ing stress to crack than those assembled with tin–lead solder. Inaddition, capacitors from different manufacturers are composedof different dielectric, electrode, and termination materials, andthey have different constructions (dielectric thickness, num-ber of electrodes, margin layers, porosity in end-terminationand dielectric layers, etc.). This will result in differing levelsof compressive stress upon cooling from the greater meltingtemperature associated with Pb-free solder assembly. This cancause capacitors from different manufacturers to behave dif-ferently in flex testing when they are assembled with Pb-freesolder.

Another solder property which affects flex cracking of capac-itors during bending is solder elastic modulus. Tin–silver-based

KEIMASI et al.: FLEX CRACKING OF MLCCS ASSEMBLED WITH Pb-FREE AND TIN–LEAD SOLDERS 189

TABLE IISTRAIN ON BOARD AT 1% AND 10% FAILURE DUE TO FLEX CRACKING OF MLCCS. THE SHADED REGION SHOWS MLCCS WITH X7R DIELECTRIC AND

1812 SIZE FROM THREE MANUFACTURERS. WITHIN THE LAST TWO COLUMNS, THE BOLD AND UNDERLINED NUMBERS INDICATE THE LOWEST AND

THE HIGHEST VALUES, RESPECTIVELY, OF STRAIN ON THE BOARD AT THAT PERCENT OF FAILURE AMONG MLCCS IN THE SHADED REGION

Pb-free solders are stiffer than eutectic tin–lead solder. Theelastic modulus of Pb-free (Sn-3.0Ag-0.5Cu) solder is about41 GPa and that of eutectic tin–lead solder is about 36 GPa [20],so the elastic modulus of Pb-free solder is about 14% greaterthan eutectic tin–lead solder. During bending, the greater modu-lus for Pb-free solder allows more stress to be transmitted to thecapacitor body. If one considers only the effect of elastic modu-lus, then the Pb-free solder would be expected to produce lowerstrain-to-failure. However, the Pb-free solder was found, in thispaper and a previous study [17], to produce greater strain-to-failure. Thus, it is concluded that elastic modulus of the solderis not, by itself, sufficient for prediction of the tendency toundergo flex cracking. In addition to elastic modulus, otherproperties of the solder material, such as solidus temperature,also affect flex cracking of capacitors.

In the study by Syfer [15], they used Sn36Pb2Ag, Sn40Pb,and In50Pb solder alloys for assembling capacitors on PCBs.The solidus temperature is 184 ◦C for In50Pb solder and thecorresponding values for Sn36Pb2Ag and Sn40Pb solders are179 ◦C and 183 ◦C, respectively. Since the solidus temperatureof these solder alloys are very close to each other and to that ofeutectic tin–lead, the residual compressive stress should havesimilar values. However, in this paper, the elastic modulus ofthe solder alloy plays the dominant role. The elastic modulusof In50Pb is significantly lower than that of the other twosolder alloys (based on our literature investigation, we expectthe modulus to be less than half of that for the Sn–Pb alloys withwhich its performance is contrasted), and capacitors assembledwith this soft solder failed at greater PCB deflections.

Nies and Maxwell [16] investigated four solder alloys in theirexperiments, which are Sn37Pb, Sn40Pb, Sn36Pb2Ag, andSn4Ag. The solidus temperatures for the three tin–lead-basedsolders (Sn37Pb, Sn40Pb, and Sn36Pb2Ag) are very closeto each other. The solidus temperature for eutectic tin–lead

solder (Sn37Pb) is 183 ◦C and the corresponding values forSn36Pb2Ag and Sn40Pb solders are 179 ◦C and 183 ◦C, re-spectively. However, the solidus temperature for Sn4Ag solderis greater than that of the other three solders and is around220 ◦C. In their results, MLCCs assembled with Sn4Ag showedsomewhat greater deflection to failure than the other solderalloys. Based on the discussion earlier regarding residual stressfollowing reflow, the greater solidus temperature for Sn4Agsolder implies that more deflection is required to overcomethe residual compressive stress on the capacitor and reach thefracture stress in order to crack a capacitor. However, the greaterelastic modulus of Sn4Ag causes more stresses to be transmit-ted to the capacitor body during PCB bending and increasessusceptibility to flex cracking. Thus, the greater elastic modulusof Sn4Ag partly mitigates the effect of the solidus temperature.

Table II shows the strain on board which causes 1% and10% failure for all capacitors used in this paper. Althoughthe strain on board for 1812-size X7R capacitors showed abig difference between tin–lead and Pb-free solder for 10%failure, the differences are less at 1% failure (see Table II).The Weibull three-parameter distribution was used to generatethe curves that are fit to the flex-test data. Table III shows theWeibull parameters for these distributions. It is clear that scaleparameters for MLCCs assembled with Pb-free solder are muchlarger than those for MLCCs assembled with tin–lead solder,which confirms that MLCCs assembled with tin–lead solder aremore susceptible to flex cracking.

Table IV shows the strain on board at 10% failure due toflex cracking of MLCCs and the 95% confidence bounds ofstrain at 10% failure. This parameter was selected as a metricfor comparing the effects of solder alloy, capacitor size, andmanufacturer, although the conclusions are valid over a widerange of cumulative percent failure. The X7R MLCCs withsize 1812 from different manufacturers assembled with tin–lead

190 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

TABLE IIIWEIBULL DISTRIBUTION PARAMETERS FOR FLEX-TEST RESULTS ON MLCCS

TABLE IVSTRAIN ON BOARD AT 10% FAILURE DUE TO FLEX CRACKING OF MLCCS AND 95% CONFIDENCE BOUNDS OF STRAIN AT 10% FAILURE

solder showed very similar results in flex testing, and ouranalysis shows that the 95% confidence intervals for differentmanufacturers overlap. In the case of assembly with Pb-freesolder, Table IV shows that 95% confidence intervals of strainat 10% failure for MLCCs from manufacturers “B” and “C”overlap, whereas the 95% confidence interval for MLCCs frommanufacturer “A” does not overlap with those of the other twomanufacturers. For all cases, the 95% confidence interval ofMLCCs assembled with tin–lead solder does not overlap withthat of MLCCs assembled with lead-free solder.

B. Comparison of the Effects of the Dielectric Material onFlex Cracking of MLCCs

Flex testing of 1812 capacitors with C0G dielectric mountedon boards with eutectic tin–lead and Sn3.0Ag0.5Cu Pb-freesolder was conducted to compare with X7R dielectric. For eachsolder material, 96 capacitors with C0G dielectric were tested.None of the capacitors with C0G dielectric showed evidence offlex cracking up to the maximum strain level on board of about13 000 microstrains (the value of the strain given in microstrainsshould be multiplied by 10−6 to get strain). In contrast, in the

case of X7R capacitors from the same manufacturer assembledwith tin–lead solder, 94 out of 96 capacitors failed. Our resultsare in agreement with previous results [10] that capacitorswith X7R dielectric are known to be more susceptible to flexcracking than those with C0G dielectric, which has greaterfracture toughness [8], [9].

C. Comparison of the Effects of the Capacitor Size on FlexCracking of MLCCs

In order to study the effect of capacitor size on flex crackingof MLCCs, flex tests of two different capacitor sizes (0805 and1812) from manufacturer “A” were conducted. Fig. 12 showsthe cumulative percent failure of 1812 and 0805 capacitorswith X7R dielectric assembled on boards with eutectic tin–leadsolder. Smaller capacitors (0805) showed better resistance tocracking as compared to larger capacitors (1812) for tin–leadsolder. A similar trend was observed for MLCCs assembledwith Pb-free solder. Our results are in agreement with pre-vious literature [4] and [10] that smaller capacitors are lesssusceptible to flex cracking. Strain on board at 10% failurewas 5100 microstrains for 0805-size capacitors assembled with

KEIMASI et al.: FLEX CRACKING OF MLCCS ASSEMBLED WITH Pb-FREE AND TIN–LEAD SOLDERS 191

Fig. 12. Comparison of cumulative percent failure as a function of strainon the board for capacitors of size 1812 and 0805 with X7R dielectric frommanufacturer “A” mounted on the boards with Sn37Pb solder.

tin–lead solder. In comparison, for 1812-size capacitors fromthe same manufacturer, the strain on board at 10% failure was1700 microstrains (see Table II). As Table IV shows, the 95%confidence interval of 0805-size MLCCs does not overlap withthat of 1812-size MLCCs.

V. CONCLUSION

An investigation was performed to compare the flex-cracking susceptibility of MLCCs assembled with Pb-free sol-der (Sn3.0Ag0.5Cu) and eutectic tin–lead solder (Sn37Pb). Thedesign of experiments included MLCCs with two differentsizes (1812 and 0805), two different dielectric materials (C0Gand X7R), and three manufacturers. PCBs were flexed usinga four-point bend test, during which the strain on the boardand the capacitance of each MLCC were monitored in orderto determine the strain-to-failure. In situ capacitance measure-ment provided data on the strain-to-failure. For some failedcapacitors, the capacitance recovered after removal of flexure.This behavior could allow an electronic assembly containinga cracked capacitor to pass electrical functional tests and en-ter the application environment as a “walking-wounded” part.Eventually, this could lead to a field failure and possibly a fire.

Experimental results showed that MLCCs assembled withPb-free solder are less susceptible to flex cracking as comparedto MLCCs assembled with eutectic tin–lead solder. This wasthe trend for all three manufacturers included in this paper. Theflex test results for MLCCs with 1812 size and X7R dielectricfrom three manufacturers are comparable with each other in thecase of tin–lead solder. In contrast, among MLCCs assembledwith Pb-free solder, MLCCs from one manufacturer showedgreater susceptibility to flex cracking as compared to those fromthe other two manufacturers. For 1812-size MLCCs assembledwith tin–lead solder, the PCB strain at 10% failure ranged

between 1700 and 2000 microstrains. The PCB strain at 10%failure for 1812-size MLCCs assembled with Pb-free soldervaried between 2300 and 9600 microstrains, depending on theMLCC manufacturer.

Cooling of a capacitor after solder reflow assembly places thecapacitor under compressive stress. For a higher melting solderalloy, cooling of assembled capacitors places capacitors undergreater compressive stress. Therefore, a capacitor assembledwith a higher melting solder should require more bendingstress to crack. Since the solidus temperature for Sn3.0Ag0.5CuPb-free solder (about 217 ◦C) is higher than that for eutectictin–lead solder (183 ◦C), Pb-free solder places more residualcompressive stresses on a capacitor after assembly than tin–leadsolder. Therefore, capacitors assembled with Pb-free soldershould require more applied bending stress to crack than thoseassembled with tin–lead solder.

Among all the MLCCs tested in this paper, MLCCs withC0G dielectric had the lowest susceptibility to flex cracking. Nofailures of C0G MLCCs out of 96 samples tested for each soldermaterial were detected within the range of deflections investi-gated. This result is attributed to the greater fracture toughnessof C0G as compared to X7R. Unfortunately, MLCCs with C0Gdielectric (EIA class I) have several drawbacks including higherprice and narrow capacitance range. For example, 1812-sizecapacitors with C0G dielectric are almost twice as expensiveas those with X7R dielectric. In addition, the maximum capac-itance values for which C0G MLCCs are available are muchlower than those for classes II and III dielectrics, since thedielectric constant of C0G dielectric is significantly lower.

Comparison of flex-test results for 1812- and 0805-sizeMLCCs showed that larger MLCCs are more susceptible toflex cracking. The strain on board at 10% failure for 0805-sizecapacitors assembled with tin–lead solder was three times thecorresponding value for 1812-size capacitors from the samemanufacturer. Among all the MLCCs with X7R dielectric,0805-size MLCCs assembled on boards with Pb-free soldershowed the least susceptibility to flex cracking. It is, therefore,highly recommended to use smaller size MLCCs in situationswhere similar capacitance and rated voltage MLCCs are avail-able to decrease flex-cracking failures, cost, and PCB realestate.

REFERENCES

[1] D. M. Zogbi, Shortages Reveal Sweet Spot in Global CapacitorMarkets, Mar. 2004, TTI, Inc. (Last visited on 12 July, 2006). [Online].Available: http://www.ttiinc.com

[2] D. Zogbi, Fastest Growing Ceramic Capacitor Manufacturers: 2003–2004, Dec. 6, 2004, TTI, Inc. (Last visited on 19 October, 2005). [Online].Available: http://www.ttiinc.com

[3] T. Zennicek, S. Zednicek, Z. Sita, C. McCracken, W. Millman, and J. Gill,“Technology of niobium oxide capacitor,” Adv. Microelectron., vol. 31,no. 3, pp. 6–10, May/Jun. 2004.

[4] J. D. Prymak and J. Berganthal, “Capacitance monitoring while flex test-ing,” IEEE Trans. Compon., Packag., Manuf. Technol. A, vol. 18, no. 1,pp. 180–186, Mar. 1995.

[5] EIA Standard, Ceramic dielectric capacitors classes I, II, III and IV—Part I: Characteristics and requirements, Nov. 2002. EIA-198-1-F.

[6] C. A. Harper, Passive Electronic Component Handbook, 2nd ed.New York: McGraw-Hill, 1997.

[7] H. Kishi, Y. Mizuno, and H. Chazono, “Base-metal electrode-multilayerceramic capacitors: Past, present and future perspectives,” Jpn. J. Appl.Phys., vol. 42, no. 1, pp. 1–15, Jan. 2003.

192 IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY, VOL. 8, NO. 1, MARCH 2008

[8] C. R. Koripella, “Mechanical behavior of ceramic capacitors,” IEEETrans. Compon., Hybrids, Manuf. Technol., vol. 14, no. 4, pp. 718–724,Dec. 1991.

[9] C. R. Koripella, “Mechanical properties of ceramic capacitors,”Kemet Tech Topics, Sep. 1991. vol. 1, No. 5, Kemet ElectronicsCorporation, Document No.: 9/91. [Online]. Available: http://www.kemet.com/feedback&RequestType=Literature

[10] J. Berganthal, Ceramic Chip Capacitors “Flex Cracks”: Under-standing & Solutions, Jan. 1998. Kemet Electronics Corporation,Document No.: F-2111. [Online]. Available: http://www.kemet.com/feedback&RequestType=Literature

[11] J. D. Prymak, “Flex testing II,” Kemet Tech Topics, Oct. 1994. vol. 4,No. 6, Kemet Electronics Corporation, Document No.: 10/94. [Online].Available: http://www.kemet.com/feedback&RequestType=Literature

[12] J. D. Prymak, “Flex or bend testing,” Kemet Tech Topics, Sep. 1993. vol. 3,no. 7, Kemet Electronics Corporation, Document No.: 9/93. [Online].Available: http://www.kemet.com/feedback&RequestType=Literature

[13] R. Al-Saffar, R. Freer, I. Tribick, and P. Ward, “Flexure strength ofmultilayer ceramic capacitors,” Br. Ceram. Trans., vol. 98, no. 5, pp. 241–245, 1999.

[14] J. Maxwell, “Board flexure comparison between surface mount multi-layer ceramic and film capacitors,” in Proc. CARTS, IL, Mar. 1999.

[15] Syfer Technology Limited, Mechanical Cracking of Chip Capacitors.[Online]. Available: www.Syfer.com

[16] C. Nies and J. Maxwell, “Important factors in board flexure testingof surface mount capacitors,” in Proc. CARTS Asia, Singapore, 1991,pp. 75–82.

[17] B. Long, M. Prevallet, and J. Prymak, “Effects of lead-free solderson flex performance,” in Proc. CARTS, Palm Springs, CA, Mar. 2005,pp. 97–101.

[18] IPC/JEDEC-9702, Monotonic Bend Characterization of Board LevelInterconnect, Jun. 2004.

[19] K. Franken, H. R. Maier, K. Prume, and R. Waser, “Finite-elementanalysis of ceramic multilayer capacitors: Failure probability caused bywave soldering and bending loads,” J. Amer. Ceram. Soc., vol. 83, no. 6,pp. 1433–1440, Jun. 2000.

[20] T. Siewert, S. Liu, D. R. Smith, and J. C. Madeni, Database for solderproperties with emphasis on new lead-free solders. Golden, CO: Nat.Inst. Standards Technol. Colorado School Mines, Feb. 2002. Release 4.0.

Mohammadreza Keimasi received the B.S. andM.S. degrees in mechanical engineering from SharifUniversity of Technology, Tehran, Iran, and thePh.D. degree in mechanical engineering from theCenter for Advanced Life Cycle Engineering, Me-chanical Engineering Department, University ofMaryland, College Park.

He is currently with Dell, Inc., Round Rock, TX,where he is a Reliability Engineer. He has publishedin accredited journals and conference proceedings inthe fields of reliability of capacitors, effects of lead-

free solder on reliability of electronic products, low-temperature effects onelectronic components, and computational fluid dynamics.

Dr. Keimasi is a member of the academic honor society Phi Kappa Phi.

Michael H. Azarian (M’96) received the B.S.degree in chemical engineering from PrincetonUniversity, Princeton, NJ, and the M.S. degree inmetallurgical engineering and materials science andthe Ph.D. degree in materials science and engineer-ing from Carnegie Mellon University, Pittsburgh, PA.

He has spent 13 years in data-storage, advanced-materials, and fiber-optics industries. He is currentlya Research Scientist with the Center for AdvancedLife Cycle Engineering, University of Maryland,College Park. His research on reliability of electronic

products has led to publications on electrochemical migration, capacitor relia-bility, electronic packaging, and tribology. He is the holder of five U.S. patentsfor inventions in data storage and contamination control.

Dr. Azarian is a Technical Editor of the draft IEEE standard 1624 on orga-nizational reliability capability, which is for assessing suppliers of electronicproducts.

Michael G. Pecht (S’78–M’83–SM’90–F’92) re-ceived the B.S. degree in acoustics, the M.S. degreein electrical engineering, and the M.S. and Ph.D. de-grees in engineering mechanics from the Universityof Wisconsin-Madison, Madison.

He is currently a Chair Professor and Founderand Director with the Center for Advanced Life Cy-cle Engineering (CALCE), University of Maryland,College Park, where he is also the Founder andthe Director of the CALCE Electronic Products andSystems Center. He has been leading a research team

in prognostics for the past ten years and formed a Prognostics and HealthManagement Consortium within CALCE. He is the author of 18 books onelectronic-product development and use and on supply-chain management.

Dr. Pecht is a Professional Engineer and an American Society of MechanicalEngineers (ASME) Fellow. He is the Chief Editor for Microelectronics Relia-bility, Associate Editor for the IEEE TRANSACTIONS ON COMPONENTS AND

PACKAGING TECHNOLOGY, and has served as the Chief Editor for the IEEETRANSACTIONS ON RELIABILITY for eight years. He has also served on theadvisory board of the IEEE Spectrum. He has also edited a series of books onthe Asian electronics industry. He was the recipient of the 3M Research Award,the IEEE Undergraduate Teaching Award, and the IMAPS William D. AshmanMemorial Achievement Award for his contributions.