Modeling the Effect of Microstructure in BGA Joints

Transcription

1 Intl. Journal of Microcircuits and Electronic Packaging Modeling the Effect of Microstructure in BGA Joints S. J. Holden and E. R. Wallach Department of Materials Science and Metallurgy Cambridge University Pembroke Street, Cambridge, CB 3QZ United Kingdom Phone: Fax: s: Abstract When modeling the effects of thermal strains in soldered joints using Finite Element Analysis, the solder is generally treated as a single phase. In practice, solder joints comprise two phases, lead-rich and tin-rich, and there are often intermetallic layers at the copper pads. A preliminary assessment of the effects of these microstructural features on the predicted strain distributions within BGA joints is presented in this work, and the consequences on service lives are estimated using phenomenological equations. Comparisons with models based on treating the solder as homogeneous and with experimental data are included in this paper. The models were run with a number of different microstructures, showing the effects of different phase distributions, and the presence of defects such as voids in the joint microstructure. Key words: BGA, Finite Element Analysis, Microstructure, and Modeling.. Introduction Rapidly changing designs and the introduction of new materials mean that the Electronics Industry requires a fast, convenient method to predict component reliability. Finite Element (FE) Analysis is one such technique that can predict lifetimes of electronic components based on their geometry, thermal history, and materials properties. Most previous research using the FE method has modeled the solder as a homogeneous material, generally using data for the eutectic SnPb alloys. Work has also been carried out on lead-free alloys, although the data available for these alloys are less complete. While reasonable agreement with measured lifetimes has been achieved, the current research has been undertaken to examine the effects of modeling the actual microstructure of the solder, including details of the phases present in the solder, in order to observe the effects that these factors have on lifetime predictions. This approach has the potential advantage that it is possible to model effects that cannot be assessed by treating the balls in Ball Grid Array as homogeneous, such as the grain size, the phase distribution in the microstructure, and the presence of voids. This paper concentrates on the effects of changing the phase size, and including voids in the solder balls. The overall aim of the work is to understand how the solder microstructure affects the service lifetime that can be expected from a component in the field, based on test data and the predicted microstructural evolution.. Phase Size Variation The initial model used to consider the effects of microstructure on lifetime prediction, by treating lead and tin as separate phases, used a D slice of a 33 dummy BGA, specifically an 33 I/O Anam- Amkor BGA. The slice was taken along one of the diagonals of the package, as this sample models the solder balls which are most highly strained. Only half of the package was modeled due to the symmetry of the geometry. Inner balls were modeled as homogeneous eutectic SnPb solder. The outermost ball, which experienced the peak 80

2 Modeling the Effect of Microstructure in BGA Joints strain range in all previous work on dummy (die-less) packages, was modeled with 600 elements in a 40 by 40 grid. Approximate data were used for the individual phases (lead, tin and an Sn 6 Pb 5 intermetallic). Although this model was initially useful, a number of changes were judged necessary to make it suitable for further more accurate work. The resolution of the outer ball, at 40 by 40, was not sufficiently fine to accurately represent the microstructures observed. The subsequent models all have an 80 by 80 outer ball, giving much finer control over the microstructure represented. The phase distribution initially adopted is an extrapolation of a previous model, see Table for Creep properties of materials used, Table 3 for Strain range variation with phase size, Table 4 for Strain ranges for void models, Table 5 for Variations of mean peak creep strain range for different void locations and sizes, and Table 6 for Predicted and actual lifetimes. Figure illustrates the advantage that the previous model, which was an accurate representation of a real microstructure, incorporated the correct ratio of Sn to Pb, and the amount of intermetallic present. the trend observed was an artefact of the individual phase distributions; the phase size for this run was 60 µm. Figure. Phase distribution for run 3. Materials were modeled as purely elastic for the copper pad, interconnect, intermetallic, epoxy cap, and substrate board; properties are provided in Table. The eutectic solder, comprising the lead and tin rich regions, was modeled with elastic and creep properties. This simplifying assumption is valid as the solder materials will creep at stresses considerably below the yield stresses of all other materials present. The materials were modeled using the sinh creep expression, dγ c n Q = A[ sinh( ασ )] exp, () dt RT where c c = creep strain, t = time, r = stress, R = gas constant, T = temperature, Q = activation energy for creep, and A, a & n are material constants. The creep data used are shown in Table, Darveau & Banerji 5. The data are the same as those used in the earlier work,,3,4. Table. Solder joint phase properties. Material Temp Ref. E* ν α ( C). 63%Sn %Pb Lead Figure. Phase distribution for run. Tin Cu 6 Sn The thermal cycle modeled was an 80 minute cycle from 0 C to * Young s Modulus (GPa) Poisson s ratio 00 C, with 0 minute high and low temperature holds, and 0 minute Thermal expansion coefficient (x 0-6 / C) linear ramps, as was also used in the initial practical work carried out at Nortel. In all cases, 80 cycles were simulated in the FE modeling. Table. Creep properties of materials used. As a test of how a variation in phase distribution affects the strain range predicted, another script was used that increased the sizes of Parameter Tin-lead Lead Tin the tin and lead phases by factors of. and.6, respectively. It has eutectic the disadvantage that the phases so produced are increasingly more A (s - ) 9.6x0 4.x x0 5 blocky, but has the advantage that a realistic phase distribution is α (MPa - ) produced. Thus, initially, three models were run, at resolutions of n x80, with microstructures magnified by,. and.6, corresponding to phase (grain) sizes of 30, 60, and 0 µm, respectively, (the Q (J mol - ) difference in phase size can be seen in Figures and ). As a result of examining the results of these runs, a further run was made, with a slightly different phase distribution, in order to investigate whether 8

3 Intl. Journal of Microcircuits and Electronic Packaging 3. Strain Distributions and Lifetimes A typical example of the creep strain distributions observed is shown in Figure 5, top. The precise configuration of the two different phases (lead and tin phases) affects the local strains, as would be expected from their different properties. Nonetheless, the important point to note is that the highest strain ranges are in the ball corners. Although, these ranges give an indication of how the creep strains vary throughout the ball, it is also necessary to determine the location of the peak strained element, and thus where the solder joint is most likely to start to fail. Given the number of elements involved, a script was used to work out the strain range for all elements in the outer solder ball, and so find the peak strained element and the magnitude of the range, CE max - CE min. In all cases, the peak strained element was in the same location, being an element in the bottom right corner of the model adjacent to the intermetallic layer, suggesting a strong effect of the intermetallic or copper pad stiffness on the strain in the neighboring material. As an additional comparison, the mean value of the shear creep strain range throughout the ball was calculated to give an alternative method of comparison not so dependent on one possibly uncharacteristic element. As can be seen from Table 3, the magnitudes of the maximum creep strain range for the single element, and also the mean for the entire ball are lowest for the finest microstructure, with the coarser microstructures having considerably larger strain ranges, and thus by implication, shorter lifetimes. A possible reason for this effect, based on the phase distributions, is the proximity to this element of tin phases in the two coarser microstructures modeled, and the effect on the strain in the lead from this additional constraint. Hence, a further run (run 4) was carried out with the tin phase closest to the element that exhibited peak strain moved elsewhere in the model, in order to remove this influence. As can be seen, again in Table 3, this reduces the change in the peak strain range considerably, thus suggesting that there is a strong influence of the local environment on the peak strained element. Interestingly, a similar change is exhibited in the mean strain range over the whole of the outer ball. It appears that a small change in the modeled microstructure, in this case, by just moving one part of a phase of material within the model, can have a large effect on the predicted lifetime of the component. Work is continuing by the researchers on this aspect. The effect of observed imperfections on solder joint lifetime has not often been modeled. However, the presence of imperfections in joints are likely to have a strong effect on the lifetimes, both predicted and observed. A commonly observed imperfection in BGA packages is voiding at the package/joint and joint/substrate interfaces. The results of previous studies have shown that, in the case of dummy packages, the peak strain range, and thus most likely point of failure, is the outermost ball. Thus, the models in this work focus on the effects of voids in this most sensitive ball, although voids may occur in any of the balls in the package. The first model considered was as described for the microstructural models, but with the addition of a void to the outer ball as shown in Figure 3. The curvature of the void results in a sharp corner next to the interface which is difficult to mesh accurately with quadrilateral elements, so triangular elements were used. Apart from this behavior, efforts were made to keep the element distribution as close as possible to that of the other models. Figure 3. Location of void in outer ball. Further models were constructed similar to this first model. Clearly, there are a large number of different possible geometries that can be modeled with equal justification. Those geometries selected were initial choices. They differed in size and placement of the void in the ball, with three different void sizes and also the void located at either the top and bottom of the ball. Four different sites were used to show how the horizontal distance from the most highly strained element affects the lifetime predicted. The location of the voids at the interface with the pad material is based on micrographs of real joints, as well as their sizes. The models were subjected to the same artificial thermal cycles Table 3. Strain range variation with phase size. as described earlier for variation of microstructure. It was observed that, as before, the creep strain range reached a constant value after Run Scale Phase Mean strain Peak strain around 50 cycles. Representative results are presented in Table 4. size range range These results are for the same horizontal positioning of the void, so /µm that in this case, the variables are the vertical position, and the size G of the void. The data show a number of effects. A void near the G bottom surface of the ball has a detrimental effect on the package G lifetime. The most highly strained elements in the ball without a G void are towards the bottom of the ball, thus any change in this case would be expected to have a considerable effect. Figure 4 shows the inelastic strain distributions in the same structure with and without a void. As can be seen, the void has two effects: the local strains are 8

4 Modeling the Effect of Microstructure in BGA Joints higher around the surface of the void, and an influence on all elements in the ball, including those previously most highly strained. It can be seen that the areas of highly strained material at the top of the ball are in approximately the same position in both cases, whereas the distribution at the bottom is changed more substantially due the proximity of the void. The effect on the mean strain range appears to be influenced by the nature of the material removed by the presence of the void. If a high proportion of lead is removed, the mean strain range may be lowered since the greater part of the deformation occurs in the lead-rich phases. Accordingly, the interpretation of the mean strain range must be viewed with care. Nonetheless, it can be concluded that the closer the void to the already highly strained elements in the ball, the larger the effect. It is observed that the larger the void, the shorter the component lifetime, as would be expected simply from the total volume of material over which the strain is distributed. Further data, not shown in this publication, for a greater number of models (4) distributed over three void sizes and eight void sites evenly distributed between the top and bottom of the joint, give a mean variation of creep strain range between the top and bottom of the joint, and also between the three different void sizes modeled. Summaries of the results are shown in Table 5. Table 5a shows that the effect of a void on strain range and thus lifetime is far greater for a void at the bottom of the joint, closer to the location of the peak strained element, or region of highest initial failure probability. Table 5b confirms that, as has been stated above, a larger void leads to an increase in peak creep strain range, and therefore a decrease in joint lifetime. Table 4. Strain ranges for void models. Run Position Size/ µm Mean strain range Peak strain range G none V top V top V3 bottom V4 bottom Figure 4. Inelastic strain distribution. Bottom: void. Top: identical microstructure without void. Table 5. Variations of mean peak creep strain range for different void locations and sizes. a. Variation in position of void vertically b. Variation in size of void Void position Mean peak strain range Void size/ µm Mean peak strain range Top Bottom Life Predictions The methods used to predict the lifetimes of modeled components are all based on treating the number of cycles to failure as a function of the inelastic strain range. The most well known method of the type used is the Coffin-Manson equation although this is not included as, to date, predictions, using this equation, are not in good agreement with experimental results. An alternate approach is based on the use of calculated acceleration factors to extrapolate between temperatures. This does not suffer as heavily from the problem of requiring data for exactly the same conditions as does the Coffin-Manson method, but does require the acquisition of a standard set of measurements from which extrapolation is then made. A major assumption is that the same deformation and damage mechanisms operate in both the standard case and the situation of interest. Two methods of this type are applicable in this case. The first method was initially developed by IBM for predictions of fatigue life for controlled collapse chip connection (C4) devices 6, equation (), 83

5 Intl. Journal of Microcircuits and Electronic Packaging N = N f f z γ f γ f w H exp k T T, () where the expression considers the cycles per day (f), the peak cycle temperature (T), and the inelastic strain range Dc. The subscripts & denote the cycle considered (standard and cycle of interest). The constants used are the Boltzmann constant (k), the activation energy (DH), and z and w which are determined empirically by curve fitting. The method has been adapted for use in determining BGA lifetime by Katchmar 7, 8, who determined modified constants experimentally for plastic BGAs. The equation of this type often used is due to Norris & Landzberg 9, shown as, the following equation, be noted is that the FE method is a continuum method, and that boundary effects, in this case those at grain boundaries, are not considered. Investigations to include grain boundary effects must be undertaken in future research work. The effects of voids predict a shorter lifetimes for larger voids, as might be expected. The model is very sensitive to the positioning of the void in the solder ball, and clearly further work on this issue needs to be carried out to confirm the resulting lifetime predictions. Acknowledgments 84 N = N f f γ f γ f 3 T, (3) T which is similar to that of the modified IBM equation. The major difference from the modified IBM model is that this equation includes the temperature range of the test, DT, rather than peak temperature. These models can be used to predict lifetimes for the models described above, and the predictions are shown in Table 6. These data are compared to an experimentally measured value for an identical geometry, which is equivalent to the first phase size model. The two lifetimes correspond to two different failure criteria. The first is that 0 consecutive readings of the logging device record a fail (M), and the second is that half the readings in one particular cycle are a fail (M). Table 6. Predicted and actual lifetimes. Model Run Katchmar Norris- Landzberg Measured M 377 Measured M 733 Phase size G Phase size G Phase size G3 764 Phase size G Void V Void V Void V Void V Conclusions and Continuing Work The help of Andrew Liu, Martin Coleman, Eddie Knight and Brian Wright at Nortel Laboratories, Harlow is gratefully acknowledged. Special appreciation is dedicated to Paul Winter, who initiated much of the work discussed in this publication. The financial support of the EPSRC is also appreciated. References. P. Winter, and E. Wallach, Microstructural Modeling and Electronic Interconnect Reliability, Presented at The International Symposium on Hybrid Microelectronics, ISHM 96, Minneapolis, pp , T. Pan, Thermal Cycling Induced Plastic Deformation in Solder Joints, Transactions of ASME, Vol. 3, pp. 8-5, ITRI, Solder Alloy Data -Mechanical Properties of Solders and Soldered Joints, International Tin Research Agency Report 656, R. L. Fields, S. Low, and G. Lucey, Physical and Mechanical Properties of Intermetallic Compounds Commonly Found in Solder Joints, in The Metal Science of Joining, M. P. Ceslak, J; Kang,S; Glicksman,M, Ed. Warrendale, Pa.: TMS, pp , R. B. Darveaux, and K. Banerji, Reliability of PBGA Assembly, in Ball Grid Array Technology, J. Lau Editor, McGraw- Hill, pp , International Business Machines, C4 Product Design Manual - Volume : Chip and Wafer Design, Issue A, IBM Technology Products. 7. R. Katchmar, BGA Solder Joint Reliability, : BNR Kanata, R. Katchmar, BGA Fatigue-Life Update, : BNR Kanata, C. F. Ramirez; S. Fauser, Fatigue Life Comparison in the Perimeter and Full Array PBGA, Presented at Advances Packaging Technology, Motorola, USA, 995. A number of concluding remarks can be drawn from the work presented although, as has been mentioned before, the structures considered are quite specific. The trend evident in the modeling of phase size suggests that large phase size causes a shorter lifetime. This is counter to what is normally expected for creep in the range of homologous temperature 0.6 to 0.8, which is the case considered. What should

6 Modeling the Effect of Microstructure in BGA Joints About the authors Steve Holden completed a B.A. Degree at the Department of Materials Science & Metallurgy at the University of Cambridge in 995. He is currently working on a Ph.D. Degree in the same Department, modeling aspects of lifetime prediction in the field of solder joint reliability. Rob Wallach has been a lecturer in the Department of Materials Science, University of Cambridge for over twenty years following a short period in industrial research; he also is a Fellow of King s College. One of his main research interests is the joining of materials, and he has numerous publications in this area, including modeling of processes, joint formation and materials behavior. Recent projects have focused on diffusion bonding, brazing and high energy welding processes (such as electron beam and laser welding), as well as the reliability of soldered joints which is the subject of the current paper. Dr. Wallach is also interested in making Science and Engineering more accessible to schoolaged children. and has recently started a scheme to promote this task. 85