correlation between mechanical properties and microstructure of...

Microelectronics Reliability xxx (2015) xxx–xxx

MR-11612; No of Pages 6

Contents lists available at ScienceDirect

Microelectronics Reliability

j ourna l homepage: www.e lsev ie r .com/ locate /mr

Correlation between mechanical properties and microstructure of different aluminumwire qualities after ultrasonic bonding

M.S. Broll a,⁎, U. Geissler a, J. Höfer b, S. Schmitz b, O. Wittler b, M. Schneider-Ramelow b, K.D. Lang a

a Research Center of Micropheripheric Technologies, Technical University of Berlin, Berlin 13355, Germanyb Fraunhofer Institute for Reliability and Microintegration (IZM), Berlin 13355, Germany

⁎ Corresponding author.E-mail address: [email protected] (M.S. Broll)

http://dx.doi.org/10.1016/j.microrel.2015.06.0610026-2714/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: M.S. Broll, et al., Coafter ultrasonic bonding, Microelectronics Re

a b s t r a c t
a r t i c l e i n f o
Article history:Received 18 May 2015Received in revised form 16 June 2015Accepted 16 June 2015Available online xxxx

Keywords:Ultrasonic wire bondingElectron backscatter diffraction (EBSD)HardnessAluminum

Three different heavy aluminumwire qualities were investigated regarding their microstructural evolution afterultrasonic bonding by electron backscatter diffraction and nanoindentation. The results complete the findings ofour recent research regarding the effect of bonding mechanisms on the wire bond microstructure and its localmechanical properties. Local elastic-plastic material parameters of the bonded wires were approximated onthe basis of the elastic anisotropy of crystals and a correlation between hardness and stress.

© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Ultrasonic bonding with heavy aluminium wires is well establishedin power electronic modules. To ensure a reliable functionality inapplication a variety of reliability studies have been realized [1–4]. Asa result, thewire lift-off caused by amismatch in the coefficients of ther-mal expansion (CTE) between thewire and substratematerial appearedas one of the main failure mechanisms. In this case, the cyclic occurringthermo-mechanical loads lead to crack propagation in thewirematerialslightly above the metallization layer. As another failure mechanism,the wire heel-crack caused by mechanical deformation of the wire hasbeen exposed [5,6]. These types of failure are influenced by the localmechanical properties of the bonded wire. During ultrasonic bonding,the wire material gets significantly changed by the deformation andultrasonic induced hardening as well as softening mechanisms [7,8].Due to the microstructural changes, the mechanical properties of thewire are modified.

To investigate the main failure mechanisms and to make an exactprediction of wire bond lifetime, it is crucial to analyse the microstruc-tural evolution during ultrasonic wire bonding and to determine theresulting mechanical properties. The mechanical properties are ofgreat interest since they can easily be used as input parameters for finiteelement simulation tools. Commonly, themechanical properties of wirebonds are specified by the Young's modulus and stress–strain datadetermined by tensile tests of the wire material [9–11]. Obviously, thisapproach does not consider the influence of the bonding process. First

.

rrelation between mechanicaliability (2015), http://dx.doi

efforts to determine the mechanical properties of the bonded state byusing nanoindentation are for instance presented in [12]. However,there is still a considerable ambiguity regarding a reliable determinationof the mechanical properties and their dependency from the ultrasonicbonding process. In this respect, this article deals with the microstruc-tural evolution during ultrasonic bonding and the resulting localmechanical properties of the wire.

2. Experimental procedure

2.1. Wire bonding

In this study, MOSFET chips wired with three different Al wirequalities are investigated. Fig. 1 shows the used MOSFET chip and ametallographically prepared longitudinal-section of a bonded wire.The nanoindentation test areas are schematically illustrated as blueframes. All wires were bonded with an Orthodyne wedge bonder on a5 μm AlSi1Cu0.5 metallization layer deposited on a silicon chip whichwas attached using silver sintering. The investigated commerciallyavailable wires and their breaking loads are listed in Table 1.

The bonding parameters were adjustedwith the aid of shear tests bygenerating shear forces and residues as high as possiblewithout fractur-ing the chip.

2.2. Microstructural and mechanical tests

2.2.1. Electron backscatter diffraction (EBSD)Microstructural investigations were carried out at metallographical-

ly prepared samples by EBSD. In a first step, the wire bonds were

l properties and microstructure of different aluminum wire qualities.org/10.1016/j.microrel.2015.06.061

http://dx.doi.org/10.1016/j.microrel.2015.06.061

mailto:[email protected]

Journal logo


http://www.sciencedirect.com/science/journal/

www.elsevier.com/locate/mr


Fig. 1.MOSFET chip and metallographically prepared longitudinal-section of the wire bond.

2 M.S. Broll et al. / Microelectronics Reliability xxx (2015) xxx–xxx

embedded near room temperature. After hardening at room tempera-ture, the sampleswere carefully cut,mechanically grinded andpolished.As a final preparation step, all samples were polished and etched by ionbeam. The EBSD measurements were conducted at an accelerationvoltage of 20 kV and step sizes between 0.03 and 1 μmusing a Digiview4 CCD camera (EDAX) and SEM–Quanta FEG 400 (FEI).

2.2.2. Tensile tests and nanoindentationTo correlate the hardness with stress, tensile tests were performed

up to several defined strains. The as-received and pre-strained wireswere prepared as described in Section 2.2.1.

To analyse the local mechanical properties, load controlled nanoin-dentation tests were performed at the prepared longitudinal sectionsof the bonded wires. Furthermore, the as-received and pre-strainedwires were examined. The used test system is a Triboindenter(Hysitron) with a Berkovich-tip. Each test was performed under thesame test conditions. The maximum force was set to 9 mN and the du-ration of loading and unloading to 9 s, respectively. Before unloading,the maximum force was kept constant for 2 s.

3. Results and discussion

3.1. Microstructure of the as-received wires

The as-received wires were analysed regarding their crystallographictexture. As can be seen in Fig. 2, all investigated wires exhibit distinct butdifferent textures.

These textures are the result of the specific wire drawing processes.In the case of Q1, the majority of grains exhibit a (001)-A1 orientation.The Q2 wire exhibits a (001)-A1 and additionally weak (111)-A1texture component. In contrast to Q1 and Q2, the pole figures of theQ3 wire show a clear (111)-A1 fibre texture. Furthermore, the Q3 wireshows aweak gradient in grain orientation regarding thewire diameter.Slightly below thewire surface, the (111)-A1 texture intensity seems todecrease.

Table 1Mechanical properties of the investigated heavy aluminum wires.

Wire Diameter [μm] Breaking load [cN]a

Quality 1 (Q1) 300 320–340Quality 2 (Q2) 300 560–570Quality 3 (Q3) 300 630–650

a From tensile tests (see Section 2.2.2).

Please cite this article as: M.S. Broll, et al., Correlation between mechanicafter ultrasonic bonding, Microelectronics Reliability (2015), http://dx.doi

3.2. Microstructure after wire bonding

Themicrostructural change of the differentwire qualities after ultra-sonic bonding is shown in Fig. 3.

After wire bonding a distinct gradient in grain orientation and sizehas evolved within the wedge area. Remarkably, each wire bond ex-hibits a distinct (101)-A1 textured area slightly above themetallization.This texture is considered a rotated cube (RC) texture. The black framewithin the inverse pole figure maps shows the areas used for the mea-surements of the RC texture. Within the investigated longitudinal sec-tions, Q1 seems to exhibit the highest magnitude of the RC texturedarea. Between Q2 and Q3 no obvious difference regarding the magni-tude of RC textured area can be observed. Above the RC textured areaand especially slightly below the wire upside the as-received wire tex-ture is still extant. Considering the heel area, no remarkable change intexture can be observed. These characteristic microstructuralgradients occur nearly in the same manner for Q1, Q2 and Q3.

The evolution of RC textures is already discussed in [13] as a result ofmicrostructural processes such as dynamic recrystallization, recoveryand plastic deformation during ultrasonic bonding. Since RC texturescan be observed for different Al wire qualities, it can be assumed thattheir evolution is generally possible for face centered cubic (fcc) metals.Beside the assumed transferability to other fcc metals, the occurring ofRC texture might have considerable influence on the results of sheartests which were commonly used for the quality control. In conclusion,in addition to geometrical aspects, the local material properties,especially the RC texture, should be considered in the interpretation ofshear results.

3.3. Mechanical characterization

3.3.1. Determination of elastic propertiesEach as-received wire exhibits a fibre texture. However, wire bond-

ing leads to a texture gradient in the wedge area. This texture gradientconsists of RC texture slightly above the metallization and fibre textureat the upside of the wedge. The intensity of both depends on the wirebonding process. In this case, all wire bonds exhibit a distinct RC textureand minor changed fibre texture. Also the texture of the heel areaappears to be equal to the as-received wires. Consequently, the textureof the heel area and area at the upside of the wedge can be assumed asequal to the texture of the as-received wires. In the following part, wecorrelate the observed fibre and RC textures with the respective elasticproperties.

Fibre textures lead to transversely isotropic behaviour. Using thesingle crystal compliance tensor Sijkl of Al and the texture informationfrom Fig. 2, the elastic properties can be approximated [14]. In thecase of transverse isotropy, the elastic properties are represented bythe Young's moduli parallel and perpendicular to the A1 axis (EA1, EA2,

al properties and microstructure of different aluminum wire qualities.org/10.1016/j.microrel.2015.06.061


Fig. 2.Microstructure and pole figures of the as-received wires a) Q1, b) Q2 and c) Q3.

3M.S. Broll et al. / Microelectronics Reliability xxx (2015) xxx–xxx

EA3), shear moduli (GA1A2, GA1A3, GA2A3) and Poisson's ratios (νA1A2,νA1A3, νA2A3). The Young's moduli, shear moduli and Poisson ratios fora specific crystallographic plane/direction (hkl)/[hkl] can be calculatedwith Eqs. (1), (2), and (3) and the single crystal constants S11 =0.0157; S12 = −0.0057; S44 = 0.0351 of Al [15–17].

1E hkl½ � ¼ S11−2 S11−S12−0:5S44ð Þn ð1Þ

1G hklð Þ ¼ S44 þ 2 S11−2S12−S44ð Þn ð2Þ

ν h�k�l�½ � hkl½ � ¼ S12 þ S11−S12−0:5S44ð ÞmS11− 2S11−2S12−S44ð Þn

n ¼ n21n

22 þ n2

1n23 þ n2

2n23

m ¼ n21m

21 þ n2

2m22 þ n2

3m23

ð3Þ

The characters n and m are the direction cosines for the respectivetension and contraction direction [hkl] [h*k*l*]. Considering the planeof symmetry, it can be simplified that EA2 = EA3, GA1A2 = GA1A3,νA1A2 = νA1A3. To approximate a representative Young's modulus forthe respective sample direction, we determined the area fraction ofeach crystal direction. The tolerance value for the misorientation anglewhich separates two different crystal directionswas set to 20°. It shouldbe noted that the used tolerance angle of 20° leads to a simplification ofthe plane and direction dependency of the elastic properties. After mul-tiplication of the area fractionwith the respective E[hkl], the representa-tive Young's modulus was approximated for each sample direction byaveraging overall crystal directions. In the case of Q2, the nearly samearea fraction of (001)-A1 and (111)-A1 orientated grains leads to a com-pensation of the minimum and maximum Young's modulus.

Please cite this article as: M.S. Broll, et al., Correlation between mechanicaafter ultrasonic bonding, Microelectronics Reliability (2015), http://dx.doi

Furthermore, no clear texture perpendicular to A1 can be observed.Therefore, Q2 is assumed as isotropic on a macroscopic level.

Complex mathematical expressions for the calculation of the shearplane and direction dependent shear modulus for cubic crystals aregiven in [17,18]. However, for simplification the general expression ofHearmon [16] seems to be more convenient. In this expression theshear modulus of a specific plane is related to torsion around all direc-tionswithin this plane. Using this expression (see Eq. (2)), the same ap-proach as used for the approximation of the representative Young'smoduli was applied to approximate the representative shear moduli.

The tension plane and direction dependent Poisson ratio was calcu-latedwith Eq. (3) [17]. Applying this equation,we receive the respectivePoisson ratios ν[100][010] = ν[001][010] = 0.36, ν½101�½111� ¼ ν½110�½111� ¼ 0:34, ν[001][110] = 0.41 and ν½110�½110� ¼ 0:27. As a conse-quence, the Poisson ratios referring to the in-plane contraction for theb010N and b111N crystal directions which were aligned parallel to thetension direction are set constant to 0.36 and 0.34. Although,ν[h*k*l*][110] varies corresponding to the respective in-plane directionwithin the crystal system, we assume that ν[h*k*l*][110] is nearly con-stantwithin the sample system. This assumption ismade by consideringthe random orientation of the in-plane directions within the samplesystem (see Fig. 2). Here, we note that this assumption should also bevalid for ν[h*k*l*][010] and ν[h*k*l*][111]. Averaging the minimal andmaximal Poisson ratios for the in-plane directions of the (110) plane re-sults in 0.34. Using this approach, it can be seen that the calculatedPoisson ratios referring to the (010), (111) and (110) planes varybetween 0.34 and 0.36. Under consideration of the corresponding areafractions the difference between the wire qualities is minor. The resultsfor the approximation of the elastic properties are summarized inTable 2.

In the case of Q1, the maximum difference between the sampledirection dependent Young's modulus is ~4 GPa which is ~6% of the



Fig. 3.Microstructure and rotated cube textured area of the wire bonds a) Q1, b) Q2 and c) Q3.


maximum value. Comparing Q1 and Q3, the maximum differenceoccurs in the A1 direction with ~8 GPa which is ~11%. Despite of therelatively low anisotropy of the Al single crystal, the calculated percent-age differences appear quite high. Therefore, considering the elastic an-isotropy and the difference between wire qualities should improve thesignificance of finite element simulations for wire bonds under variousloading conditions. Especially in the case of loading conditions with amajor elastic part, the consideration of the current results might haveconsiderable impact. Also, it should be mentioned that in the case ofmore anisotropic materials such as copper, the direction dependencywould be more significant by far.

Table 2Approximated elastic material parameters for the fibre textures of Q1, Q2 and Q3.

Elastic properties Q1 Q2a Q3 Unit

EA1 EA2 EA3 65 69 69 70 73 71 71 GPaGA2A3 GA1A3 = GA1A2 27 28 26 26 GPaνA2A3 νA1A2 = νA1A3 0.35 0.34 0.34 0.34 0.34 /

a Isotropic properties.


Due to the high intensity of RC texture, the area slightly above themetallization layer probably exhibits anisotropic behaviour. Simplify-ing, the elastic properties of this area can be described by a tilted singlecrystal. Under this assumption, the elastic properties can be given as a6 × 6 stiffness tensor which can be directly implemented into finiteelement simulation tools. However, it is necessary to tilt the stiffnesstensor according to the respective texture. The RC texture correspondsto a 45° tilt around the A3 axis. Applying the rules for transformationof the stiffness tensor, the 45°/A3 tilted stiffness tensor can be deter-mined. This tensor transformation results in S11 = S22 = 113.3, S12 =S21 = 56.7, S13 = S31 = S32 = S23 = 62, S33 = 108, S44 =S55 = 28.3and S66 = 23.

3.3.2. Correlation between hardness and microstructureFig. 4 summarizes the results for the nanoindentation tests of the as-

received and bonded wires. The diagrams show the wire hardnessmeasured in the wedge area in regard to the distance from the Si inter-face. Furthermore, the hardness of the as-receivedwires and loop area isindicated as a constant value (horizontal lines). For allwire qualities, thelowest hardness within the wedge area is located slightly above the Siinterface in the area of RC texture. This indicates the occurring of



Fig. 4. Hardness of the as-received wires and after bonding in the wedge and loop area for a) Q1, b) Q2 and c) Q3.

5M.S. Broll et al. / Microelectronics Reliability xxx (2015) xxx–xxx

dynamic recrystallization and recovery during the ultrasonic deforma-tion. Except for Q3, the wedge clearly exhibits higher hardness thanthe loop area. In all cases, the loop area exhibits slightly higher hardnessthan the as-received state. Surprisingly, in the case of Q3 the hardnessof RC textured area is nearly the same as of the loop area and theas-received state.

3.3.3. Correlation between hardness and stressThe hardness values measured at the as-received wires and defined

interrupted tensile test samples were correlated to tensile stresses onthe basis of the Tabor correlation [19]. In this correlation, each hardnessvalue corresponds to a defined strain level εwhich is the sum of the ini-tial plastic strain εpl adjusted by the tensile test and the representativestrain from indentation εr. The representative strain depends amongothers on the indenter tip geometry and the mechanical properties ofthe sample material. For simplification we set the representative strainto constant 0.02 which is within the range of the calculated values by[20] and [21] for a sharp indenter and a metallic sample material. Itshould be noted that this approximation leads to an uncertainty regard-ing the absolute stress value.

Due to the indentation size effect (ISE), the hardness and therewiththe constraint factor depend on the indentation depth [22]. Conse-quently, the calculated constraint factor is only valid for hardness valuesdetermined under the same test conditions as described in Section 2.2.2.The correlation between the experimental stress–strain curves and thehardness values results in a constraint factor of around c = 6.7–8.1 forQ1, c = 4.2–5.4 for Q2 and c = 4.2–5 for Q3 (see Fig. 5). Applying therespective constraint factor, it is possible to calculate the local represen-tative stresses for each measured location within the wedge and looparea. Due to the relatively small representative strain of 2%, the calcula-ble representative stresses can be approximated as flow stresses for theinitiation of plastic deformation. As an example, for the RC textured areaof Q1 we calculate flow stresses between 51–62 MPa which is twice as

Fig. 5. Correlation between hardnes

Please cite this article as: M.S. Broll, et al., Correlation between mechanicaafter ultrasonic bonding, Microelectronics Reliability (2015), http://dx.doi

much as the flow stress of the as-received wire. These results showthe necessity of considering the change of material properties duringthe ultrasonic bonding process.

4. Conclusions

Despite of the different initial texture, grain size and hardness, all in-vestigatedwires exhibit a rotated cube textured area afterwire bonding.Hardness measurements show that the bonding process increases thehardness of the wedge and loop area. Due to dynamic recrystallizationand recovery, the RC textured area exhibits the lowest hardness withinwedge for all wire qualities. The calculated local elastic and plasticmaterial properties indicate the importance of considering the bondingprocess and crystallographic wire texture for accurate lifetime predic-tions of wire bonds in power modules. Furthermore, the results form abasis to improve existing simulation models.

Acknowledgements

This workwas financially supported by the German FederalMinistryof Education in the research activity “RoBE” (FKZ 16N11464).

References

[1] J. Goehre, U. Geissler, M. Schneider-Ramelow, K. Lang, Influence of bonding param-eters on the reliability of heavy wire bonds on power semiconductors, InternationalConference on Integrated Power Electronics Systems (CIPS) 2012, pp. 1–6.

[2] A. Hamidi, N. Beck, K. Thomas, E. Herr, Reliability and lifetime evaluation of differentwire bonding technologies for high power IGBT modules, Microelectron. Reliab. 39(6-7) (1999) 1153–1158.

[3] H. Lambilly, H.O. Keser, Failure analysis of power modules: a look at the packagingand reliability of large IGBTs, IEEE Trans. Components Hybrids Manuf. Technol. 16(4) (1993) 412–417.

[4] S. Ramminger, P. Türkes, G. Wachutka, Crack mechanism in wire bonding joints,Microelectron. Reliab. 38 (6-8) (1998) 1301–1305.

s and stress for Q1, Q2 and Q3.


http://refhub.elsevier.com/S0026-2714(15)00153-5/rf0105













[5] Y. Celnikier, L. Benabou, L. Dupont, G. Coquery, Investigation of the heel crack mech-anism in Al connections for power electronics modules, Microelectron. Reliab. 51(5) (2011) 965–974.

[6] S. Ramminger, N. Seliger, G.Wachutka, Reliability model for AI wire bonds subjectedto heel crack failures, Microetectron. Reliab. 40 (8-10) (2000) 1521–1525.

[7] U. Geißler, M. Schneider-Ramelow, H. Reichl, Hardening and softening in AlSi1 bondcontacts during ultrasonic wire bonding, IEEE Trans. Compon. Packag. Technol. 32(4) (2009) 794–799.

[8] U. Geißler, M. Schneider-Ramelow, K.-D. Lang, H. Reichl, Investigation ofmicrostruc-tural processes during ultrasonic wedge/wedge bonding of AlSi1 wires, J. Electron.Mater. 35 (1) (2006) 173–180.

[9] L. Merkle, M. Sonner, M. Petzold, Lifetime prediction of thick aluminium wire bondsfor mechanical cyclic loads, Microelectron. Reliab. 54 (2) (2014) 417–424.

[10] A. Grams, T. Prewitz, O. Wittler, J. Kripfgans, S. Schmitz, A. Middendorf, W.H. Müller,K.D. Lang, Simulation of an aluminum thick wire bond fatigue crack by means of thecohesive zone method, EuroSimE (2013) 1–8.

[11] O. Schilling, M. Schäfer, K. Mainka, M. Thoben, F. Sauerland, Power cycling testingand FE modelling focussed on Al wire bond fatigue in high power IGBT modules,Microelectron. Reliab. 52 (9-10) (2012) 2347–2352.

[12] P. Agyakwa, V. Marques, M. Corfield, J. Li, L. Yang, C. Johnson, Room-temperaturenanoindentation creep of thermally cycled ultrasonically bonded heavy aluminumwires, J. Electron. Mater. 42 (3) (2013) 537–544.


[13] M. Broll, U. Geissler, J. Höfer, S. Schmitz, K.L.O. Wittler, Microstructural evolution ofultrasonic-bonded aluminum wires, Microelectron. Reliab. 55 (6) (2015) 961–968.

[14] T.C.T. Ting, Anisotropic Elasticity: Theory and Applications, Oxford University Press,Oxford, 1996.

[15] G. Grimvall, Thermophysical Properties of Materials, Elsevier, Netherlands, 1999.[16] R.F.S. Hearmon, The elastic constants of anisotropic materials, Rev. Mod. Phys. 18

(1946) 409–440.[17] E. Schreiber, O.L. Anderson, N. Soga, Elastic Constants and Their Measurement,

McGraw-Hill, New York, 1973.[18] J. Turley, G. Sines, The anisotropy of Young's modulus, shear modulus and Poisson's

ratio in cubic materials, J. Phys. D. Appl. Phys. 4 (1970) 264–271.[19] D. Tabor, The Hardness of Metals, Oxford University Press, Oxford, 2001.[20] N. Ogasawara, N. Chiba, X. Chen, Measuring the plastic properties of bulk materials

by single indentation test, Scr. Mater. 54 (2006) 65–70.[21] K.E. Prasad, N. Chollacoop, U. Ramamurty, Role of indenter angle on the plastic

deformation underneath a sharp indenter and on representative strains: anexperimental and numerical study, Acta Mater. 59 (11) (2011) 4343–4355.

[22] W. Nix, H. Gao, Indentation size effects in crystalline materials: a law for straingradient plasticity, J. Mech. Phys. Solids 46 (3) (1998) 411–425.












































correlation between mechanical properties and microstructure of...

Documents