Failure Mechanisms of SAC/Fe-Ni Solder Joints …...tance24,25 and the degree of cracking2,26 are typi-cally used to measure failure. In our study, owing to the absence of a PCB in

Failure Mechanisms of SAC/Fe-Ni Solder Joints During ThermalCycling

LI-YIN GAO,1,2 ZHI-QUAN LIU ,1,2,3 and CAI-FU LI1

1.—Shenyang National Laboratory for Materials Science, Institute of Metal Research, ChineseAcademy of Sciences, Shenyang 110016, China. 2.—University of Chinese Academy of Sciences,Beijing 100049, China. 3.—e-mail: [email protected]

Thermal cycling tests have been conducted on Sn-Ag-Cu/Fe-xNi (x = 73 wt.%or 45 wt.%) and Sn-Ag-Cu/Cu solder joints according to the Joint ElectronDevice Engineering Council industrial standard to study their interfacialreliability under thermal stress. The interfacial intermetallic compoundsformed for solder joints on Cu, Fe-73Ni, and Fe-45Ni were 4.5 lm, 1.7 lm, and1.4 lm thick, respectively, after 3000 cycles, demonstrating excellent diffusionbarrier effect of Fe-Ni under bump metallization (UBM). Also, two deforma-tion modes, viz. solder extrusion and fatigue crack formation, were observedby scanning electron microscopy and three-dimensional x-ray microscopy.Solder extrusion dominated for solder joints on Cu, while fatigue cracksdominated for solder joints on Fe-45Ni and both modes were detected for thoseon Fe-73Ni. Solder joints on Fe-Ni presented inferior reliability during ther-mal cycling compared with those on Cu, with characteristic lifetime of 3441 h,3190 h, and 1247 h for Cu, Fe-73Ni, and Fe-45Ni UBM, respectively. Thisdegradation of the interfacial reliability for solder joints on Fe-Ni is attributedto the mismatch in coefficient of thermal expansion (CTE) at interconnectionlevel. The CTE mismatch at microstructure level was also analyzed by elec-tron backscatter diffraction for clearer identification of recrystallization-re-lated deformation mechanisms.

Key words: Fe-Ni under bump metallization (UBM), thermal cycling,microstructural evolution, lifetime, recrystallization, electronbackscatter diffraction (EBSD)

INTRODUCTION

Under bump metallization (UBM) is an essentialaspect of solder technology, especially for decreasedbump size. The main functions of UBM are toprovide reliable adhesion to the substrate and anexcellent wettability and good diffusion barrier forsolder material. In particular, the performanceunder thermal cycling, a critical aspect of solderjoint reliability, has been widely investigated forsolder joints on traditional Cu UBM. Thermal stressinduced by coefficient of thermal expansion (CTE)mismatch between adjoining materials fundamen-tally determines the reliability of solder joints under

thermal cycling.1–3 Moreover, the total amount ofCTE mismatch can be defined at different levels,e.g., interconnection and microstructural levels.4

The interconnection level is mainly investigatedbased on the CTE mismatch between the differentinterconnecting materials,5 while at the microstruc-tural level, the role of the recrystallization phe-nomenon in crack initiation and propagation isdiscussed.6,7

Cu is a widely used UBM material, but isconsumed relatively quickly during aging and gen-erally forms Kirkendall voids at the interface,degrading package reliability.8–10 To address thisissue, novel UBMs are always being researched,including Cu-Zn11 and Co-P12,13 films and compositeNi-P film with ZrO2 nanoparticles doped.14 In recent(Received December 28, 2016; accepted April 24, 2017;

published online May 9, 2017)

Journal of ELECTRONIC MATERIALS, Vol. 46, No. 8, 2017

DOI: 10.1007/s11664-017-5554-1� 2017 The Minerals, Metals & Materials Society

5338

years, Fe-Ni alloy films have been identified as anovel UBM for use with various lead-free sol-ders,15,16 and related research has reported thatboth electroplated and electroless-plated Fe-Ni filmsexhibit better wettability than pure Ni owing to theaddition of iron.17,18 Moreover, Zhu et al.19 showedthat a 6-lm-thick scallop-type Cu6Sn5 layer formedin conventional solder joints on Cu, while only 200-nm-thick intermetallic compound (IMC) formed onFe-Ni UBM after reflow, indicating an extremelyslow consumption rate of Fe-Ni UBM comparedwith Cu UBM. However, the performance underthermal cycling of solder joints on Fe-Ni UBM hasseldom been reported.

In this study, two types of Fe-Ni UBM wereelectroplated onto a wafer-level chip-scale package(WLCSP) to evaluate their reliability under thermalcycle testing. The interfacial reaction, microstruc-tural evolution, and characteristic lifetime of the Fe-Ni UBMs were investigated in detail in comparisonwith commercial Cu UBM. Moreover, the failuremodes of the different solder joints were character-ized, and the corresponding failure mechanism forboth SAC/Fe-Ni and SAC/Cu solder joints is dis-cussed based on the experimental observations.

EXPERIMENTAL PROCEDURES

Preparation of Solder Joints

For the thermal cycling tests, test vehicles wereprepared using three different UBMs. The structureof the test vehicles used was a bumped WLCSP witha 12 9 12 ball grid array, as shown in Fig. 1a, withbump diameter of 300 lm and bump pitch of500 lm. First, two Fe-Ni UBMs were electroplatedonto a patterned 8-inch-diameter wafer using acustomized wafer plating system, with Fe-Ni UBMcomposition of 73 wt.% and 45 wt.% Ni, denoted asFe-73Ni and Fe-45Ni, respectively. The electroplat-ing details for the Fe-Ni thin films were reportedpreviously.20,21 For comparison, solder joints were

also fabricated with commercial Cu UBM using thesame WLCSP structure.

Wafer-level reflow processing was conducted fol-lowing IPC/Joint Electron Device EngineeringCouncil (JEDEC) standard J-STD-020D. The soldermaterial was Sn-3.8Ag-0.7Cu (wt.%), referred to asSAC herein for simplicity. The temperature profileof the reflow process had peak temperature of 245�Cwith holding time above eutectic temperature(217�C) of 67 s to ensure reliable soldering. Finally,the wafer was cut into 6 mm 9 6 mm componentscontaining 144 bumps in a 12 9 12 ball grid array(Fig. 1a). Notably, the test vehicle used in this studywas a bump structure without a printed circuitboard (PCB), intentionally designed to highlight theeffect of the UBM. First, the CTE mismatchbetween a PCB and chip,22 which plays a crucialrole in crack propagation, is large. Second, becausethe thermal stress on the PCB side is generallytwice that on the chip side,23 the PCB in a packageis more vulnerable to damage and even opencircuits. As a result, both the deformation modeand lifetime are related more to the CTE mismatchbetween the silicon and PCB rather than the choiceof UBM. Therefore, in this study, the thermalcycling test was conducted on components withouta PCB, to focus on the effect of the UBM on the chipside. A cross-section of a single solder joint in thetest vehicle is shown in Fig. 1b.

Thermal Cycling Test and MicrostructuralCharacterization

The thermal cycling test was designed followingJoint Electron Device Engineering Council indus-trial standard JEDEC 22-A104D (condition G).Three different solder joints were thermally cycledat �40�C/125�C. Specifically, the ramping/coolingrate was set as 11�C/min, with 10 min dwell and15 min for soaking, corresponding to total durationof 50 min per cycle.

Fig. 1. Test vehicle: (a) 12 9 12 bumped WLCSP and (b) single solder joint.

Failure Mechanisms of SAC/Fe-Ni Solder Joints During Thermal Cycling 5339

Microstructural observations were conducted onsolder joints after 250, 1500, and 3000 thermal cyclesby scanning electron microscopy (SEM; Supra 55,Carl Zeiss AG). Cross-sections of solder joints wereprepared by standard metallographic method, apply-ing deep etching with combined solution of 5 g FeCl3,5 mL HCl, and 90 mL C2H5OH on cross-sections ofsolder joints after reflow and 3000 cycles to reveal thedistinct IMCs. Furthermore, at least four SEMimages obtained under identical conditions wereused to measure IMC thickness data. In each SEMimage, the thickness of the IMC layer was measureddigitally as follows: first, commercially availablePhotoshop software was used to distinguish the areain pixels of the individual IMCs. This total pixel areawas then divided by the length in pixels to obtain themean thickness in pixels of each IMC layer. Finally,although the thickness of the IMCs was not uniform,the average thickness of the different IMCs wasobtained by converting the thickness in pixels tophysical units using the image scale bar.

For lifetime calculations, 58 solder joints for eachtest condition were investigated to obtain failurestatistics. Seven test conditions were used (0, 250,500, 1000, 1500, 2000, and 3000 cycles), making atotal of 406 solder joints for each type of UBM. Insuch studies, both the increase of the initial resis-tance24,25 and the degree of cracking2,26 are typi-cally used to measure failure. In our study, owing tothe absence of a PCB in our test vehicle, the latterindication was adopted during microstructuralobservations. Moreover, both extrusion and crackswere counted as failures induced by thermal stress.As a result, deformation over 5 lm in length wasdefined as the failure criterion in our study, similarto reported in literature.26

Furthermore, three-dimensional x-ray microscopy(3D-XRM; Xradia Versa 500, Carl Zeiss AG) wasalso applied to characterize the failure mode of thedifferent solder joints. This technique provides aclear and relatively large-field view with resolutionof 0.6 lm, much better than typical x-ray detectorsused in industrial applications. Using 3D-XRM, 144solder bumps with each type of UBM were observed.Finally, the recrystallization phenomenon occurringwithin the solder joints was analyzed by electronbackscatter diffraction (EBSD) to reveal the failuremechanism of the solder joints on Fe-Ni. Prior toEBSD observation, cross-sectional samples werefirst ground and polished mechanically then cleanedusing an ion milling system (Leica EM RES101,Leica Microsystems) to enable high-resolutionobservation. The voltage used for ion milling was6 kV, with milling angle of 3�.

RESULTS

Formation of Interfacial IMCs in DifferentSolder Joints

The microstructural evolution of the interfacialIMCs in different solder joints during the thermal

cycling test is shown in Fig. 2. Figure 2a, b, and cshow the reflow state of solder joints on Cu, Fe-73Ni, and Fe-45Ni, respectively. Scalloped Cu6Sn5

formed at the SAC/Cu interface (Fig. 2a), and layersof rod-like (Cu,Ni)6Sn5 and thin FeSn2 both formedat the SAC/Fe-73Ni interface (Fig. 2b), whereasonly a thin FeSn2 layer formed at the SAC/Fe-45Niinterface (Fig. 2c). As reported previously, theFeSn2 layer was 0.25 lm thick after reflow, asobserved by transmission electron microscopy(TEM; JEM 2100, JEOL).20 After 3000 cycles, a1.3-lm-thick Cu3Sn layer and scalloped 3.2-lm-thick Cu6Sn5 formed at the SAC/Cu interface(Fig. 2d). Notably, a large number of Kirkendallvoids formed within the Cu3Sn interface, which isdetrimental to solder joint reliability.27,28 For theFe-73Ni UBM after 3000 cycles (Fig. 2e), however,a thin FeSn2 layer and (Cu,Ni)6Sn5 with varyingmorphology formed at the interface; the interfacebetween the FeSn2 and (Cu,Ni)6Sn5 layers isindicated in Fig. 2e (black arrowheads). Further-more, the growth of the FeSn2 layer was extremelyslow, reaching thickness of about 0.3 lm after 3000cycles. Some (Cu,Ni)6Sn5 grains integrated into acontinuous 1.4-lm-thick IMC layer at the externalside of the FeSn2 layer, while other, mostly rod-likedispersed (Cu,Ni)6Sn5 grains floated randomlynear the interface. Similarly, both FeSn2 and(Cu,Ni)6Sn5 formed in the SAC/Fe-45Ni solderjoints after 3000 cycles (Fig. 2f), with FeSn2 layerthickness of only about 0.3 lm. In contrast, the(Cu,Ni)6Sn5 (Fig. 2f, white arrowhead) exhibitedcharacteristics different from the rod-like shapeobserved in the solder joint on Fe-73Ni, typicallyshowing island shape at the external side of theFeSn2 layer. As also revealed in Fig. 2f, the(Cu,Ni)6Sn5 islands had varying dimensions withdiameter ranging from 0.6 lm to 2.5 lm, showing atendency to integrate into a continuous layer as aresult of grain coarsening. Notably, EDS pointanalysis (using tens of different points) indicatedthat (Cu,Ni)6Sn5 contained about 8 at.% to 16 at.%Ni in the solder joints on Fe-73Ni but only 2 at.%to 6 at.% Ni in the solder joints on Fe-45Ni. LessIMC was also detected for the solder joints on Fe-45Ni compared with Fe-73Ni, which can beattributed to the diffusion barrier effect of theFeSn2 layer. In our previous work,29 the grain sizeof FeSn2 within solder joints on Fe-45Ni wasdemonstrated to be larger than for those on Fe-73Ni. Meanwhile, small grain size is reported to bebeneficial for rapid grain-boundary diffusion.8,30 Asa result, solder joints on Fe-45Ni possess a lowerNi concentration gradient and larger FeSn2 grains,suppressing growth of (Cu,Ni)6Sn5.

After 3000 thermal cycles, the total IMC thick-ness was 4.5 lm for the solder joints on Cu,compared with 1.7 lm and 1.4 lm for the solderjoints on Fe-73Ni and Fe-45Ni, respectively. There-fore, the IMC growth rate was effectively sup-pressed by use of the Fe-Ni UBM materials.

Gao, Liu, and Li5340

Microstructural Evolution During ThermalCycling

The microstructural evolution of the differentsolder joints after 250, 1500, and 3000 thermalcycles is shown in Fig. 3. The SAC/Cu solder jointexhibited no failures after 250 cycles (Fig. 3a), whileextrusion of solder material was observed near theinterfacial region after 1500 cycles (Fig. 3b). After3000 thermal cycles (Fig. 3c), this extrusion wasmore extreme, and a small crack had initiated onthe extrusion. The solder extrusion area in Fig. 3c isshown at higher resolution in Fig. 3c¢, where it ismeasured as 11 lm in thickness and 24 lm inlength, and both Ag3Sn and Cu6Sn5 granules areseen to be distributed uniformly within the soldermatrix of the extrusion. Notably, no voids or crackswere observed by SEM within the extrusion area.This was further confirmed by 3D-XRM observation,as discussed in ‘‘Failure Modes for Different SolderJoints’’ section, indicating that the solder extrusionwas a kind of plastic deformation induced bythermal stress. More importantly, note that thesolder extrusion always initiated at the belt-shapedarea near the solder/UBM interface. For simplicity,we call this region the ‘‘belt area’’ in the discussionbelow. As reported in literature, there exist amaximum stress and strain, as revealed in the beltarea and within the entire solder joint throughfinite element analysis, where deformations usuallyoccur.5,31,32 It is experimentally confirmed, there-fore, that solder extrusion occurs in the belt area asa result of thermal stress.

Examining the Fe-73Ni solder joint, no failureswere detected after 250 cycles (Fig. 3d). After 1500cycles (Fig. 3e), crack initiation was detected within

the solder in the belt area, i.e., the same locationwhere solder extrusion occurred for the solder jointson Cu discussed above. The crack propagatedinward into the solder and reached nearly 50 lmin length as the number of thermal cycles wasincreased to 3000 (Fig. 3f). As shown in the higher-resolution image of the crack area in Fig. 3f ¢, thecrack had zigzag morphology and is obviously afatigue crack. Nevertheless, it is also supposed thatthe noncoherent IMC particles (edges marked withblack/white dots in Fig. 3f ¢) within the solderprovide favorable sites for crack propagation.

The solder joints on Fe-45Ni exhibited the mostserious deformation during the thermal cycling test.After 250 cycles, an indentation was already notice-able at the right corner of the belt area (Fig. 3g),and after 1500 cycles (Fig. 3h) cracks had initiatedon both sides of the belt area. The left crack wasmeasured to be as long as 50 lm, while the crack onthe right side was about 20 lm in length. Crackpropagation was seen to vary even within the samesolder joint. The orientation of the Sn grains, thenoncoherent IMC particles, and the grain bound-aries will affect the parameters of the thermalstress. After 3000 cycles (Fig. 3i), the cracks almostcrossed the entire solder joint. In the enlargedimage in Fig. 3i¢, IMC particles can be seen alongthe crack, as highlighted by white/black dots tracingthe particle edges. Notably, IMC granules not onlyinitiated the cracks (as illustrated by the middle andright grains in Fig. 3i¢) but also blocked the cracks(as illustrated by the left grain in Fig. 3i¢), accordingto our SEM observations.

It is obvious that crack initiation was easier forthe solder joints on Fe-Ni, especially for Fe-45Ni.Although use of the Fe-Ni UBMs slowed down the

Fig. 2. Interfacial microstructure of solder joints on (a) Cu, (b) Fe-73Ni, and (c) Fe-45Ni as reflowed, and (d) Cu, (e) Fe-73Ni, and (f) Fe-45Niafter 3000 thermal cycles.

Failure Mechanisms of SAC/Fe-Ni Solder Joints During Thermal Cycling 5341

interfacial reaction compared with the Cu UBM, asdemonstrated in ‘‘Formation of Interfacial IMCs forDifferent Solder Joints’’ section, it also inducedgreater thermal stress within the solder joint.Therefore, reliability under thermal stress shouldbe carefully considered before adoption of Fe-NiUBM in applications, which is also a common butcritical issue for introduction of other novel barrierlayer materials.

Failure Modes for Different Solder Joints

The new technology 3D-XRM has recently beenemployed to observe failures and defects inadvanced packages.33,34 Using 3D-XRM, compo-nents can be reconstructed in three dimensionsfrom thousands of cross-sectional x-ray imagesusing the Xradia reconstruction software. In thiswork, the failure modes of the different solder jointswere further studied using 3D-XRM. Because thedeformations in this work mainly occurred withinthe solder, only the solder material is displayed inthe 3D-XRM results in Fig. 4, showing two projec-tions for each solder ball.

As demonstrated in Fig. 4a, before thermalcycling the solder ball was well fabricated, beingused as a reference for the typical failures of thesolder joints on the different UBM materials after3000 thermal cycles shown in Fig. 4b, c, and d. Forthe SAC/Cu solder joints (Fig. 4b), the originalconfiguration changed due to solder extrusion. Theextruded solder lay around the belt area, resem-bling the brim of a cap with rough and irregularedge. In addition, the thickness of this brim rangedfrom tens up to hundreds of micrometers for thedifferent solder balls, and the extruded solderextended to as much as 50 lm in length, accordingto our 3D-XRM observations. According to analysisof the cross-sectional x-ray images, no voids orcracks were found within the SAC/Cu solder joints,confirming that the extrusion occurred via plasticdeformation of the solder joint.

In the solder joints on Fe-73Ni (Fig. 4c), solderextrusion occurred on the left side of the belt areaand a circular crack was detected on the right.Therefore, thermal stress was relieved via bothcrack initiation and solder extrusion in the solderjoints on Fe-73Ni. In the solder joints on Fe-45Ni

Fig. 3. Microstructures of: solder joints on Cu after (a) 250, (b) 1500, and (c) 3000 cycles and (c¢) enlargement of (c); solder joints on Fe-73Niafter (d) 250, (e) 1500, and (f) 3000 cycles and (f¢) enlargement of (f); solder joints on Fe-45Ni after (g) 250, (h) 1500, and (i) 3000 cycles and (i¢)enlargement of (i).

Gao, Liu, and Li5342

(Fig. 4d), more severe deformation was observed,wherein the circular crack propagated inwards andalmost reached the center of the solder. Further-more, the solder above the crack inclined towardsthe left side, accompanied by severe crack propaga-tion, suggesting that a slipping process of the Sngrains occurred under thermal stress.

According to 3D-XRM analysis, two deformationmodes were observed, viz. solder extrusion andfatigue crack formation. Specifically, solder extru-sion dominated for the solder joints on Cu, whilefatigue cracks dominated for the solder joints on Fe-45Ni and both modes were detected for the solderjoints on Fe-73Ni.

Characteristic Lifetime by Weibull Analysis

The Weibull distribution is one of the mostcommonly used methods to calculate the lifetimeof microelectronics from reliability tests.25 Thecumulative failure percentage of the Weibull func-tion is given by the formula31

F tð Þ ¼ 1 � exp � t

g

� �b" #

; ð1Þ

where F(t) is the cumulative failure percentage, g isthe scale parameter, b is the shape parameter, and tis the time to failure for each vehicle. The scaleparameter g is also defined as the characteristiclifetime at which 63.2% of the population has failed.

Figure 5 presents Weibull distribution plots forthe solder joints on the different UBMs underthermal cycling. The correlation coefficient q, scaleparameter g, and shape parameter b for the differ-ent UBMs are summarized in Table I. The q valuesfor the three UBMs tested are above 0.99, demon-strating good fitting of the failure data. Further-more, the value of b has a distinct effect on thefailure rate: when b> 1, failures become morefrequent as time elapses, and wear-out-type failuresare expected. The Weibull distribution in this studyis based on testing of thousands of solder joints (i.e.,

Fig. 4. XRM observations of (a) solder joint as reflowed, and typical failures for solder joints on (b) Cu, (c) Fe-73Ni, and (d) Fe-45Ni after 3000thermal cycles.

Fig. 5. Weibull distribution for different solder joints.

Failure Mechanisms of SAC/Fe-Ni Solder Joints During Thermal Cycling 5343

seven conditions with 58 solder joints for each type ofUBM, making a total of 1218 solder joints), provid-ing the fundamental evidence necessary to evaluatethe reliability of the different solder joints. ThroughWeibull analysis, the characteristic lifetimes for theCu, Fe-73Ni, and Fe-45Ni UBMs were calculated as3441 h, 3190 h, and 1247 h, respectively. Therefore,the degree of deformation under thermal stress forthe three solder joints is clearly determined to be inthe order Fe-45Ni � Fe-73Ni> Cu.

DISCUSSION

As stated above, the failure mechanism of a solderjoint during the thermal cycling test can beexplained based on the CTE mismatch betweenadjoining materials, which can be categorized intointerconnection and microstructure levels.1–7 In ourstudy, the interconnection level focuses on theinterconnections between different materials,including the solder, IMC layers, UBM, and Cupad. The different UBM materials used for thesolder joints result in significantly different CTEmismatches, further influencing the degree of fail-ure as well as the characteristic lifetime. Also, atmicrostructure level, the recrystallization phe-nomenon is fully discussed and reported to beclosely related to the initiation and propagation ofcracks. Therefore, these two levels are discussedseparately in the following sections.

CTE Mismatch at Interconnection Level

At interconnection level, the solder joint consistssuccessively of the solder, interfacial IMCs, UBM,and Cu pad. The CTEs for these materials within theinterconnections are summarized in Table II.35–40

Considering the strong CTE anisotropy of Sn andSn-based SAC solder, for simplicity we use the data

for pure Sn to represent the SAC data. Furthermore,the differences in the CTE mismatch among thethree solder joints originate from the different UBMmaterials and interfacial IMCs.

For the solder joints on Cu, interfacial IMCs ofCu3Sn and Cu6Sn5 are considered. The CTE ofCu6Sn5 and Cu3Sn is measured to be 18.3–19.4 ppm/�C and 18.2–19 ppm/�C, respectively, at60�C to 140�C by thermomechanical analysis(TMA),36 with similar values being obtained inother studies. For solder joints on Fe-Ni, mean-while, only the (Cu,Ni)6Sn5 interfacial IMC isdiscussed because the FeSn2 layer is so thin(300 nm after 3000 cycles) that its effect on theCTE mismatch can be temporarily ignored. Fur-thermore, physical properties of FeSn2 phase haveseldom been reported. In our previous work,20 theCu6Sn5 formed in solder joints on Cu was inmonoclinic phase (below 189�C in the phase dia-gram), while (Cu,Ni)6Sn5 formed in solder joints onFe-Ni had hexagonal structure (above 189�C in thephase diagram) owing to the stabilizing effect of Ni.Furthermore, Mu et al.38 indicated that the CTEvalues for hexagonal Cu6Sn5 and (Cu,Ni)6Sn5

(21.3 ppm/�C and 23.5 ppm/�C, respectively) areslightly higher than that of monoclinic Cu6Sn5

(18.7 ppm/�C) under the same conditions. Finally,the CTE of the interfacial IMC in the solder joint onFe-Ni is higher [23.5 ppm/�C for (Cu,Ni)6Sn5] thanthose in the solder joint on Cu (18.3–19.4 ppm/�C forCu6Sn5 and 18.2–19 ppm/�C for Cu3Sn).

Meanwhile, the CTE values for the Fe-Ni UBMmaterials (12.5 ppm/�C for Fe-73Ni and 9.5 ppm/�Cfor Fe-45Ni) are lower than that of Cu (13.5–14.3 ppm/�C). Based on the analysis above, theCTE mismatch between the Fe-45Ni UBM(9.5 ppm/�C) and (Cu,Ni)6Sn5 (23.5 ppm/�C) in thesolder joint on Fe-45Ni is the greatest, resulting inthe largest thermal stress. Hence, the most signif-icant degree of deformation and the shortest life-time (1247 h) were found for the solder joints on Fe-45Ni. In contrast, the solder joints on Cu showedsuperior interfacial reliability and the longest char-acteristic lifetime of 3441 h owing to their smallestCTE mismatch. The solder joints on Fe-73Ni exhib-ited reliability (3190 h) comparable to that of solderjoints on Cu owing to the small discrepancy betweentheir CTE mismatch.

Table I. Parameters for Weibull distributionanalysis

UBM q b g (h)

Cu 0.992 1.5 3441Fe-73Ni 0.997 1.8 3190Fe-45Ni 0.992 1.8 1247

Table II. CTE for materials in the interconnections

Material CTE (ppm/�C)

Sn Average: 21.2–27.8 (30–130�C); a axis: 16.4–21.5 (30–130�C); c axis: 30.9–40.4 (30–130�C)35

Cu6Sn5 18.3–19.4 (60–140�C)36; 20 (25–100�C)37; 18.7 (170�C)38

Cu3Sn 18.2–19 (60–140�C)36; 18.4 (25–100�C)37

(Cu,Ni)6Sn5 23.5 (170�C)38

Fe-Ni 12.5 (Fe-73Ni) and 9.5 (Fe-45Ni)39

Cu 13.5–14.3 (�40–125�C)40

Gao, Liu, and Li5344

CTE Mismatch at Microstructure Level

As stated above in ‘‘Microstructural EvolutionDuring Thermal Cycling’’ and ‘‘Failure Modes forDifferent Solder Joints’’ sections, the microstruc-tural evolution and failure modes of the three typesof solder joint revealed different characteristics.Because the recrystallization phenomenon is sup-posed to be the dominant effect for initiation andpropagation of cracks, as widely reported,4,6,25,41 the

EBSD technique was applied in this work to studythe deformation modes.

SEM images (a–h) and corresponding Euler ori-entation images (a¢–h¢) of typical solder joints arepresented in Fig. 6, where Fig. 6a and a¢ present thetypical morphology of a solder joint before thermalcycling. In the EBSD technique, the three Eulerangles (u1, U, u2) describe a minimum set ofrotations that can be used to orient the sample tocoincide with the crystal orientation. Crystal

Fig. 6. EBSD analysis of microstructural evolution: (a) as reflowed, (b) global recrystallization (solder joint on Fe-45Ni after 1500 cycles), (c)solder extrusion (solder joint on Cu after 3000 cycles) and (d) its enlargement, (e) crack initiation (solder joint on Fe-45Ni after 1000 cycles) and(f) its enlargement, (g) crack propagation (solder joint on Fe-45Ni after 3000 cycles) and (h) its enlargement; (a¢–h¢) Euler orientation imagescorresponding to (a–h).

Failure Mechanisms of SAC/Fe-Ni Solder Joints During Thermal Cycling 5345

orientations can then be represented by these Eulerangles using different colors in a Euler map.Accordingly, the colors for the three Euler anglesshown in the upper-right insets of Fig. 6a¢ indicatethe corresponding crystal orientation. For the solderjoint before thermal cycling (Fig. 6a¢), the entiresolder ball is almost entirely one color (green),indicating that only one grain was observed in thecross-section after reflow. In general, the as-re-flowed microstructure of a solder ball (300 lm indiameter) is composed of only one or several largetin grains.

The recrystallization phenomenon was observedwithin all three types of solder joint, and usuallybefore 1500 cycles. Figure 6b and b¢ show the typicalmorphology of the recrystallization phenomenonwhere, as shown in Fig. 6b¢, new grains nucleatein the belt area close to the interfacial region, wherethe solder material is severely stressed, as men-tioned above. Grain boundaries where the orienta-tion of adjacent grains exceeds 10� are indicated byblack lines in the Euler orientation images in Fig. 6.The recrystallization in Fig. 6b and b¢ is seen tooccur without solder extrusion or fatigue cracks,being termed global recrystallization to distinguishit from the local recrystallization that occurs accom-panied by cracks. With global recrystallization,stresses within the solder material are partiallyreleased and thus no extrusion or fatigue cracks aredetected. As the thermal stress continues toincrease during thermal cycling, however, a thresh-old is reached beyond which global recrystallizationcannot fully release all of the stresses and deforma-tions then occur.

Within most solder joints on Cu and some of thoseon Fe-73Ni, solder extrusion was observed as atypical failure mode. EBSD analysis of the relation-ship between recrystallization and solder extrusionis shown in Fig. 6c and d. As demonstrated in thecross-sectional images (Fig. 6c), solder extrusionwas detected on both sides of the belt area, wherespecifically the extruded solder on the lower left wasabout 30 lm in length while that on the lower-rightside was only 10 lm in length. Accordingly, severalfine grains were detected within the left extrudedsolder, while the other areas including the extrudedsolder on the right remained as a large tin grain(purple area in Fig. 6c¢). This therefore indicatesthat recrystallization occurred after serious solderextrusion. For further analysis, the belt area, ashighlighted by the black arrowhead in Fig. 6c, isshown enlarged in Fig. 6d. As demonstrated inFig. 6d¢, small contrasts are detected within a largegrain (purple) in the belt area, indicating slightdiscrepancy from the original orientation. Thissuggests formation of subgrains under thermalstress. It is possible that, as the thermal stresskeeps increasing, a slipping process of Sn grainsoccurs that further induces the solder extrusion.Consequently, the thermal stress is partiallyreleased during the slipping process and

recrystallization is thereby suppressed. As a result,recrystallization is only detected within the seri-ously extruded solder.

The failure mode of fatigue cracking is typicallyseen within solder joints on both types of Fe-NiUBM, for which the CTE mismatch is larger.Therefore, both the initiation and propagation ofcracks were analyzed carefully. The typical mor-phology of solder joints with an early initiatedfatigue crack (about 20 lm in length) is shown inFig. 6e, as indicated by black arrowheads. Asdemonstrated in Fig. 6e¢, the recrystallization phe-nomenon was not obvious (almost cyan color withinthe entire solder ball) in spite of the presence of thefatigue crack in the right corner. The recrystallizedgrain is shown in green color and indicated by ablack arrowhead in Fig. 6e¢, and enlarged in Fig. 6fand f¢. Within the large recrystallized grain (green-colored area in Fig. 6f¢), both the crack and severalnewly recrystallized fine grains along the crack areobserved. Thus, the initiation of the cracks can beunderstood based on the following three processes:First, recrystallization occurs preferentially in themost stressed area (i.e., belt area) to release aportion of the thermal stress during cycling. Second,as the thermal stress further increases to a level atwhich recrystallization and plastic deformationcannot release it completely, cracking will initiatein the most stressed area. Third, once a crack isinitiated, the distribution of stress and strainaround the crack is changed, and in particular,local recrystallization occurs easily at the tip of thecrack because the local stress is larger there.

In Fig. 6g, cracks in both sides of the belt areahave propagated to become nearly 50 lm longwithin the solder joint, and local recrystallizationis obvious along the propagation path. As can beseen in the enlarged images in Fig. 6h and h¢, theorientations of the recrystallized fine grains on bothsides of the crack are typically different, suggestingthat these grain boundaries provide favorable sitesfor the crack to propagate in the solder. Further-more, two slight contrasts (seen as light-purple anddark-purple areas in Fig. 6h¢) are detected withinthe grain at the tip of the crack, suggesting thatminor disorientation had already occurred prior tocracking within this grain. This implies that thelocal crystallization phenomenon is also promotedby the crack propagation. Nevertheless, the solderball was always inclined to one side with theappearance of serious cracking, as described abovein ‘‘Failure Modes for Different Solder Joints’’section, according to our observations. Once recrys-tallization occurs, more grain boundaries are intro-duced. In addition, the slipping process of newlyrecrystallized grains also occurs simultaneouslywith the crack, leading to the inclination of thesolder ball.

Based on the above experimental observations, itis found that the failure mode depends directly onthe level of CTE mismatch. When the CTE

Gao, Liu, and Li5346

mismatch is relatively small, thermal stress can bereleased by some form of plastic relaxation such asglobal recrystallization and solder extrusion. Whenthe CTE mismatch increases to a certain degree,cracks initiate in the most seriously stressed areaand give rise to local recrystallization along thecrack, thereby promoting further propagation of thecracks.

Although Fe-Ni UBM exhibited inferior thermalcycling reliability compared with commercial CuUBM in this study, its reliability for high-tempera-ture storage and electromigration are better than theCu UBM according to our systematic investiga-tions.29,42 To overcome the deficiency in thermalcycling reliability for use in applications, properdesign of the soldering pad is needed to reducethermal stress at the neck belt area. Also, thermalstress can be further reduced by choosing a propercomposition for the Fe-Ni UBM, such as the Fe-73Nicomposition in this study. Moreover, because the CTEvalue of Fe-Ni is much closer to that of silicon(1.5 ppm/�C at �40�C to 3.1 ppm/�C at 125�C22)compared with Cu, as shown in Table II, Fe-Ni candirectly replace Cu as pad or die-attachment material.

CONCLUSIONS

We investigated the performance of solder jointson Fe-Ni UBMs under thermal stress, a common butcritical issue for application of novel UBM materi-als. The interfacial microstructure of solder jointson Fe-Ni provided a better diffusion barrier com-pared with solder joints on conventional Cu. TheIMCs in the solder joints on both Fe-73Ni (1.7 lmthick) and Fe-45Ni (1.4 lm thick) were thinner thanon Cu (4.5 lm thick) after 3000 h of cycling. How-ever, the solder joints on Fe-45Ni exhibited apropensity to deform under thermal stress. Thecharacteristic lifetime as calculated from the Wei-bull distribution was 3441 h, 3190 h, and 1247 h forthe solder joints on Cu, Fe-73Ni, and Fe-45Ni UBM,respectively. Therefore, use of Fe-Ni UBM enhancedthe diffusion barrier effect at the SAC/Cu interface,but also induced greater thermal stresses within thesolder joints, degrading their reliability under ther-mal cycling.

The degree of deformation of the solder joints wasfound to depend directly on the degree of CTEmismatch of the interconnected materials. Thesolder joints on Fe-45Ni exhibited the greatestdeformation owing to their largest CTE mismatch,while the solder joints on Fe-73Ni exhibited relia-bility comparable to that of solder joints on Cuowing to the small discrepancy in their CTE mis-match. When the amount of thermal stress wasmodest, it could be released by some form of plasticrelaxation such as global recrystallization and sol-der extrusion. When the thermal stress increased toa magnitude such that plastic relaxation could notrelease it, cracks initiated in the area of higheststress in the belt area within the solder joints. Crack

initiation then gave rise to local recrystallization,which contributed greatly to crack propagation. Theinferior thermal cycling reliability of Fe-Ni UBMcan be improved by proper design of the solder jointand by choosing a composition that reduces thethermal stress in the neck belt area.

ACKNOWLEDGEMENTS

We gratefully acknowledge technical supportfrom Jiangyin Changdian Advanced Packaging Co.Ltd. (JCAP) in providing the WLCSP technology, aswell as financial support from the Major NationalScience and Technology Program of China (GrantNo. 2011ZX02602).

REFERENCES

1. J. Hokka, T.T. Mattila, H.B. Xu, and M. Paulasto-Krockel,J. Electron. Mater. 42, 963 (2013).

2. Y. Zhong, W. Liu, C. Wang, X. Zhao, and J.F.J.M. Caers, J.Mater. Sci. Eng. A 652, 264 (2016).

3. Q. Zhou, B. Zhou, T.-K. Lee, and T. Bieler, J. Electron.Mater. 45, 3013 (2016).

4. T.T. Mattila and J.K. Kivilahti, IEEE Trans. Compon.Packag. Technol. 33, 629 (2010).

5. H. Chen, M. Mueller, T.T. Mattila, J. Li, X. Liu, K.J.Wolter, and P.K. Mervi, J. Mater. Res. 26, 2103 (2011).

6. T.R. Bieler, B.T. Zhou, L. Blair, A. Zamiri, P. Darbandi, F.Pourboghrat, T.K. Lee, and K.C. Liu, J. Electron. Mater.41, 283 (2012).

7. B.T. Zhou, Q. Zhou, T.R. Bieler, and T.K. Lee, J. Electron.Mater. 44, 895 (2015).

8. S. Bader, W. Gust, and H. Hieber, Acta Metall. Mater. 43,329 (1995).

9. J.H. Ke, H.Y. Chuang, W.L. Shih, and C.R. Kao, ActaMater. 60, 2082 (2012).

10. K.N. Tu, A.M. Gusak, and M. Li, J. Appl. Phys. 93, 1335(2003).

11. C.Y. Yu, K.J. Wang, and J.G. Duh, J. Electron. Mater. 39,230 (2010).

12. N.D. Lu, D.H. Yang, and L.L. Li, Acta Mater. 61, 4581(2013).

13. C.H. Wang, C.C. Wen, and C.Y. Lin, J. Alloys Compd. 662,475 (2016).

14. X. Hu and Y.C. Chan, J. Mater. Res. 29, 2657 (2014).15. X.F. Zhang, J.D. Guo, and J.K. Shang, J. Alloys Compd.

487, 776 (2009).16. S. Narayan and K.N. Prabhu, Mater. Sci. Technol. 29, 464

(2013).17. H.F. Zhou, J.D. Guo, and J.K. Shang, J. Electrochem. Soc.

160, D233 (2013).18. J.J. Guo, L. Zhang, A.P. Xian, and J.K. Shang, J. Mater.

Sci. Technol. 23, 811 (2007).19. Q.S. Zhu, J.J. Guo, P.J. Shang, Z.G. Wang, and J.K. Shang,

Adv. Eng. Mater. 12, 103 (2010).20. L.Y. Gao and Z.Q. Liu, in Int. Conf. Electron. Packag.

Technol. High Density Packag., 16th (2015), pp. 715–718.21. H. Zhang, D. Wu, L. Zhang, Z.Z. Duan, C.M. Lai, and Z.Q.

Liu, Acta Metall. Sin. 48, 1273 (2012).22. M. Amagai, Microelectron. Reliab. 39, 1365 (1999).23. L. Yin, L. Wentlent, L. Yang, B. Arfaei, A. Oasaimeh, and

P. Borgesen, J. Electron. Mater. 41, 241 (2012).24. J. Hokka, T.T. Mattila, H.B. Xu, and M. Paulasto-Krockel,

J. Electron. Mater. 42, 1171 (2013).25. T.K. Lee, C.U. Kim, and T.R. Bieler, J. Electron. Mater. 43,

69 (2014).26. A. Persson, Scand. J. Metall. 33, 53 (2004).27. X. Li, F. Li, F. Guo, and Y. Shi, J. Electron. Mater. 40, 51

(2011).28. H.L.J. Pang, K.H. Tan, X.Q. Shi, and Z.P. Wang, Mater.

Sci. Eng. A 307, 42 (2001).

Failure Mechanisms of SAC/Fe-Ni Solder Joints During Thermal Cycling 5347

29. L.Y. Gao, C.F. Li, P. Wan, and Z.Q. Liu, J. Mater. Sci.:Mater. Electron. (2017). doi:10.1007/s10854-017-6576-4.

30. A.M. Gusak and K.N. Tu, Phys. Rev. B 66, 115403 (2002).31. T.T. Mattila, P. Marjamaki, and J.K. Kivilahti, IEEE

Trans. Compon. Packag. Technol. 29, 787 (2006).32. H. Chen, J. Li, and M. Li, J. Alloys Compd. 540, 32 (2012).33. L. Kong, A.C. Rudack, P. Krueger, E. Zschech, S. Arkalgud,

and A.C. Diebold, Microelectron. Eng. 92, 24 (2012).34. Y. Sylvester, B. Johnson, R. Estrada, L. Hunter, K. Fahey,

T. Chou, Y.L. Kuo, and IEEE, in IEEE/SEMI Adv. Semi-cond. Manuf. Conf. Workshop, 24th (2013), pp. 249–255.

35. V. Deshpande and D.B. Sirdeshmukh, Acta Crystallogr. 14,355 (1961).

36. N. Jiang, J.A. Clum, R.R. Chromik, and E.J. Cotts, Scr.Mater. 37, 1851 (1997).

37. W.J. Reichenecker, Insul./Circuits 27, 109 (1981).38. D. Mu, J. Read, Y. Yang, and K. Nogita, J. Mater. Res. 26,

2660 (2011).39. R.M. Bozorth, Ferromagnetism, 1st ed. (New York: Van

Nostrand, 1951), p. 102.40. F.C. Nix and D. MacNair, Phys. Rev. 60, 597 (1941).41. B. Zhou, T.R. Bieler, T.-K. Lee, and W. Liu, J. Electron.

Mater. 42, 319 (2013).42. L.Y. Gao, Z.Q. Liu, L. Zhang, H.Y. Guo, and C.M. Lai, in

6th Electronic System-Integration Technology Conference(ESTC)(2016), pp. 1–5.

Gao, Liu, and Li5348