LGAs VS. BGAs – LOWER PROFILE AND BETTER RELIABILITY S. Joshi1, B. Arfaei1, A. Singh2, M. Gharaibeh2, M. Obaidat1, A. Alazzam1, M. Meilunas3, L. Yin3, M. Anselm3, and P. Borgesen1 1 Department of Systems Science & Industrial Engineering 2 Department of Mechanical Engineering Binghamton University Binghamton, New York, USA 3 Universal Instruments Corporation Conklin, NY, USA
[email protected]
ABSTRACT Land Grid Array packages are attractive for a number of reasons, including the small overall outline of assemblies and the ability of the user to choose the solder alloy. In general, life in thermal cycling may also not be reduced by as much as might be expected based on the standoff alone. In fact, we have shown packages soldered with SAC305 to survive as long as, or sometimes longer than, corresponding Ball Grid Array components in accelerated thermal cycling. This was found to be associated with a superior microstructure formed in the smaller solder volumes. This is, however, sensitive to materials and design parameters. So far we have been able to control it in our work. A major separate effort is focused on how to do so under general manufacturing relevant conditions. Even assuming the solder microstructure can be controlled questions remain as to whether it is also likely to be superior under realistic service conditions, whether it remains stable over time, and whether it has disadvantages for alternative damage and failure mechanisms such as solder pad cratering. This is the focus of the present report. Key words: Pb-Free Solders, Interlaced Twinning, Land Grid Array (LGA), Ball Grid Array (BGA), Pad Cratering INTRODUCTION Due to the ever increasing demand for very small outline assemblies Land Grid Array (LGA) packages are gaining popularity, particularly in portable electronics. The LGA style of packaging is similar to that of Ball Grid Array (BGA) except that the ball-attach process is left out. Relying only on solder paste for connection to the mother board the solder joint height is noticeably smaller than for BGAs [1]. This would normally be expected to be accompanied by a considerable reduction in reliability, at least when it comes to thermal cycling. However, we have reported on the superior thermal cycling performance of SAC305 solder joints in LGA assemblies compared to the corresponding BGA versions [2-10]. The observation of a comparable or longer life appears to be unique to lead free solders. Very short SnPb solder joints
are often found to last longer in thermal cycling than predicted by extrapolation of results for taller joints, but in service, they still fail considerably faster than the latter. We have shown the longer life of shorter SAC305 solder joints to be associated with the formation of a different solder microstructure, so-called interlaced twinning, in those. [2] According to unpublished industry reports SnAgCu based LGA assemblies far from always perform that well, presumably because interlaced twinning does not always occur [8, 10]. Indeed, formation of this structure is a result, among other, of the relatively strong undercooling of small solder volumes during cool-down from reflow [2, 11]. This is sensitive to a number of factors and a systematic ongoing investigation addresses the question of whether formation of this microstructure can be controlled under practical circumstances [2, 5]. Emphasis of the present effort is on a preliminary assessment of whether our accelerated test results reflect performance under realistic service conditions as well, i.e. on whether acceleration factors and effects of aging are likely to be similar, and whether other damage and failure mechanisms may be enhanced instead. As far as the latter is concerned, the interlaced twinning structure appears to be harder than the common single Sn grain or ‘beach ball’ structures [2]. A concern is that this, together with the less compliant joint shape, may significantly enhance the propensity for solder pad cratering in assembly, test, handling or service. LGA and BGA assemblies were therefore also compared in cyclic bending experiments. Comparisons of SnPb based assemblies reflected the mechanical effect of solder joint shape, while comparisons of SAC305 based assemblies revealed effects of interlaced twinning as well. BACKGROUND The much lower standoff between component and board in LGA than in BGA assemblies would, as said, normally be expected to be accompanied by a considerable reduction in reliability, at least when it comes to thermal cycling. Indeed,
that is what is commonly observed for SnPb solder joints (Figure 1) .
Figure 2. Weibull plot showing the cycles to failure of SAC305 CABGAs and corresponding LGAs in -40/125°C thermal cycling test with 60 min dwell time [10] Figure 1. Weibull plot showing the cycles to failure of SnPb CABGAs and corresponding LGAs in -40/125°C thermal cycling test with 60 min dwell time. Note the better performance by the BGAs [10].
Comparisons of three different Amkor PBGA components with SAC305 solder balls to their corresponding LGA versions for the same thermal cycling conditions showed the life of the LGA assemblies to be as good, or up to 50% longer than, the life of the BGA ones (Figures 3-5).
Actually, the 3x difference in Figure 1 is considerably lower than predicted based on the nominal cyclic strain range
where Δ is the difference between the coefficients of thermal expansion of component and printed circuit board (PCB), ΔT is the temperature range, DNP is the distance to neutral point of the joint subject to the highest strain (at the corner of the array or under the die edge, depending on the component construction). In cases where these three parameters remain constant, like for the components compared in Figure 1, we would expect the life to vary slight faster than the inverse square of the standoff, h. The ‘better than predicted’ performance of the LGAs in Figure 1 may be understood based on the competition between the effective rigidities of the component, the PCB and the solder joints. This leads the actual strain on the joints to be lower than the nominal one predicted by the equation above. Since the LGA joints are less compliant the reduction in strain, relative to the prediction, will be greater than for the corresponding BGA joints. It is, however, obvious that the actual strain on the LGA joints will still be greater than for the BGA joints, as will the associated stress. This effect alone would therefore not be enough to explain a longer fatigue life of the LGA joints.
Figure 3. Weibull plot showing the cycles to failure of SAC305 PBGA 324 assemblies and corresponding LGA ones in -40/125°C thermal cycling test with 60 min dwell time [10]
That is important because we have found the thermal cycling performance of SAC305 based LGA assemblies to be comparable to, and sometimes considerably better than, that of the corresponding BGA assemblies [2-10]. The same comparison as in Figure 1, but with SAC305 solder joints and on a different PCB, showed [10] the life of the LGA assemblies to be almost as long as for the BGAs (Figure 2).
Figure 4. Weibull plot showing the cycles to failure of SAC305 PBGA 676 assemblies and corresponding LGA ones in -40/125°C thermal cycling test with 60 min dwell time [10]
An impression creep test on selected samples also showed a lower compliance of the interlaced structure (Figure 7). Still, a strong dependence of the properties of beach ball or single grain joints on Sn grain orientation provides for a variability that is considerably greater than these differences. The reduced compliance alone is therefore still not enough to explain the difference in Figure 5.
Figure 5. Weibull plot showing the cycles to failure of SAC305 PBGA 1156 assemblies and corresponding LGA ones in -40/125°C thermal cycling test with 60 min dwell time [10] The above differences cannot simply be ascribed to the less compliant nature of shorter solder joints. That alone would, as said, not be enough to provide for a longer life for LGA assemblies. Indeed, we have shown that effect to be associated with the Sn grain morphology in the SAC305 joints [2-10]. Individual area array scale solder joints invariably exhibit a microstructure formed in a single solidification event during cool-down from reflow. Figure 6a shows a cross section of a typical BGA solder joint. Cross polarized optical microscopy reveals the typical “Kara’s Beachball” configuration characteristic of cyclic twinning [12]. LGA scale joints, on the other hand, often exhibit the so-called interlaced twinning structure shown in Figure 6b. Like the beach-ball structure this only includes three distinct Sn orientations, all formed by cyclic twinning from the same nucleus in solidification, and the extremely low energy twin boundaries between them do not facilitate crack propagation.
Figure 6. (a) Cross polarized image of a 30 mil SAC305 solder joint exhibiting so called “Kara’s Beachball” with three large grains, (b) cross polarized image of a SAC305 LGA joint exhibiting the interlaced twinning grain structure [2]. Micro-hardness testing showed the Knoop hardness of interlaced twinning structures to typically exceed that of beach ball or single Sn grain ones by an average of 25% [2], a difference that may be due to a higher density of Ag3Sn precipitates in the interlaced region rather than the interlaced Sn grain structure itself.
Figure 7. Impression creep graph for interlaced and beach ball structures of 12 mil SAC305 having Cu/ELNG pad finish [2]. Indeed, it appears that the main reason for the better performance of the interlaced structure in thermal cycling is that it delays recrystallization [13]. A major ongoing research effort has led to the establishment of a mechanistic model for the evolution of damage and failure in area array SnAgCu solder joints in thermal cycling [14]. According to this, alternating between the build-up of dislocation cell structures at low temperature and the coalescence and rotation of these at a higher temperature eventually leads to the formation of a continuous network of high angle grain boundaries across the high strain region of the joint. This is followed by the relatively rapid growth of cracks along the boundaries. Although the crack growth stage takes longer than the completion of recrystallization, the latter was found to take up a fixed fraction of life, independently of cyclic strain range, temperatures or dwell times [14]. LGA joints with an initial interlaced twinning structure are found to fail by the same mechanism. However, indications are that this is delayed by the expenditure of a considerable amount of cycling energy in strain enhanced migration of the twin boundaries [13]. EXPERIMENTAL PROCEDURE Experiments addressed solder fatigue in thermal cycling and solder pad cratering in cyclic bending. Thermal Cycling Model packages were designed and constructed which differed only in terms of solder volumes and solder joint pitch. 64 solder mask defined pads, each with a 0.381mm diameter and a pad finish of electrolytic Ni/Au (ELNG), were arranged in an 8 x 8 full array and populated with 0.254mm (10 mil), 0.30mm (12 mil) or 0.40mm (16 mil) diameter SAC305 solder spheres. Combining these with five different pitches (1.0 mm, 1.2 mm, 1.4 mm, 1.8 mm and 2.2 mm) and corresponding total package sizes provided for a total of 15 variations. The packages had 0.5mm thick silicon
die sandwiched between two 0.4mm thick FR4 substrates with thin layers of a commercial flip chip underfill. The balanced construction served to minimize warpage. The motherboard used in the experiment was designed to support five samples of each package pitch (Figure 8a). The board was a glass fiber reinforced FR4 substrate with four signal layers and a total thickness of 1.575mm. The nonsolder mask defined 0.30mm diameter pads had a copper with Organic Solderability Preservative (OSP) surface finish. Top and bottom side surface routing was completed using 1/2oz copper which was plated up to a total thickness of approximately 0.072 mm. Trace widths were approximately 0.10mm. The daisy chain pattern used for each assembly is identical except the pad pitch is varied. Nine probe points are provided to help locate eventual failures (Figure 8b). A flux printing process was used in both ball attach and assembly in order to minimize solder joint voiding and maintain a tightly controlled solder joint volume. Using a no clean tacky flux, test vehicles were reflow soldered in a forced convection oven containing a nitrogen atmosphere. The peak temperature recorded was 244°C.
found to be 5.5 ppm/°C whereas the motherboard CTE was 14.1 ppm/°C. Table 1: (a) Pitch and distance to the neutral point (DNP) (b) Solder ball size and respective standoff distance for two different body sizes.
(a) (b) Isothermal Cycling Separate cyclic bend test experiments employed a test board based on the JEDEC drop test board [15], but modified to a 2 layer design. Four-point bend testing involved support and load spans of 110mm and 75mm, respectively, as shown in Figure 9.
Figure 9. 4-point bending test setup as per JEDEC JESD22B113 [15] The high Tg filled-phenolic resin FR4 board measured 132 x 77 mm, was 1mm thick and had non-solder mask defined pads with a Cu OSP surface finish and a nominal diameter of 14.5 mil. Three locations on the center of the board were populated as shown in Figure 10. Figure 8. (a) Assembled board containing 5 different pitch sized packages; (b) daisy chain pattern Dark lines: component connections. Faint lines: board connections. Three accelerated thermal cycle conditions were utilized for this experiment. Each cycle had 10 minute dwell times at the temperature extremes and 9°C per minute transition rates between the extremes. The three cycles were: (a) 0/100°C (b) -20/100° (c) -40/125°C. The test specimens were monitored using event detection with AnaTech® testers set to record resistance exceeding 500 ohms and lasting 200 nanoseconds or longer. Failures were based upon the criteria defined in IPC-9701. The distance to neutral point (DNP) was measured from the package center to the center of a corner solder bump. To determine the CTE, PEMI Moiré interferometry was used between room temperature and 90°C. The package CTE was
The test components were commercially available Amkor CABGA208 with Electroless Nickel Immersion Gold (ENIG) surface finish. Some of the components were supplied with 0.45 mm solder balls while some were not. The solder balls were arranged in a 4-row perimeter array. The components with solder balls attached were used to build Ball Grid Array (BGA) samples while the others were used for assembling Land Grid Array (LGA) samples. The stencil aperture size used for the LGAs was 0.362 mm in diameter. The resulting solder joint standoff height was recorded after cross section analysis as 0.077mm. Two different kinds of solder alloys, SAC305 and SnPb were employed, so we had four variations (two alloys and two perimeter arrays). Before placing the components on the motherboard, the strain values during bending of the virgin board were measured on the three locations. A two sample t-test with
95% confidence interval was conducted using Minitab®. The results of the test indicated that there is no statistical difference among the three locations.
Figure 10. Assembled bend test board. Two boards per variation were assembled. The daisy chain on the board is depicted in Figure 11. SAC305 components were reflow soldered onto the PCBs in nitrogen atmosphere with a peak temperature between 242°C and 245°C while tin lead components were reflowed with a peak temperature of 216-220°C.
Figure 11. Daisy chain on board side. Note seventeen probe points for the resistance detection. For bend testing the event detectors were manually set to record resistance exceeding 300 ohms and lasting 200 nanoseconds or longer. Failures were based upon the criteria defined in JEDEC JESD22-B113 [15]. Most of these assemblies were cycled well past the detection of electrical failure before they were subjected to failure analysis. Some components were dyed using Dykem Steel Red Layout Fluid, dried, and then pried off to identify the failure mode. Relatively low magnification (5X) micrographs were taken with the use of oblique lighting. Both component and board side surfaces were analyzed for the failure. RESULTS AND DISCUSSION As discussed above, we have found the thermal cycling performance of SAC305 based LGA assemblies to be comparable to, and sometimes considerably better than, that of the corresponding BGA assemblies [2-10]. This was shown to be a result of the interlaced twinning structure (Figure 6b) often formed in LGA joints. The present effort addresses the questions of whether the superior solder
properties are stable over time, whether acceleration factors are the same as for the beach-ball and single Sn grain structures found in BGA assemblies, and whether the less compliant nature of the LGA joints leads to an enhanced risk of intermetallic bond failure or solder pad cratering under conditions (shock, isothermal cycling) favoring these mechanisms. Although too often ignored it is clear that accelerated tests may be seriously misleading if their results do not somehow correlate with life under realistic service conditions. For our present purposes this poses the question as to whether acceleration factors are the same for LGA as for BGA assemblies. This requires, at a minimum, that the damage and failure mechanisms are the same. As outlined above we have shown the thermal mismatch induced fatigue and failure of SnAgCu BGA joints to be controlled by the ongoing recrystallization of the large Sn grains [14]. The same is found to be the case for interlaced LGA joints [2, 4]. However, indications are that the superior fatigue resistance is due to the fact that some of the deposited energy is consumed by cycling enhanced migration of the twin boundaries. This process may of course affect the effective acceleration factors. In a recent paper [4] we showed the ratio of life in 0/100°C cycling with a 10 minute dwell to that in -40/125°C cycling with a 60 minute dwell to be very similar, but as we shall see below acceleration factors are in fact slightly different if we vary only the temperatures, keeping dwell times constant. This is desirable as far as the relative performance of LGA assemblies under realistic service conditions is concerned. Separate experiments will be required to assess the sensitivity to aging. Thermal Cycling The present experiments employed specially built model BGAs with different size SAC305 solder spheres on pad sizes originally intended for the largest (16 mil diameter) ones. As a result, only the 16 mil spheres led to solder joint dimensions typical of a commercial BGA (Figure 12). The 10 mil spheres led to a configuration similar to that found in an LGA assembly. The relatively large standoff may be the reason why the distance between boundaries is greater than in Figure 6b. The intermediate size (12 mil) spheres led to a combination of interlacing, presumably near the point of nucleation in cool-down, and beach ball structure. To the extent that interlacing does in fact delay recrystallization, expending some of the thermal cycling induced work on grain growth, we might expect the interlaced joints achieved with 10 mil spheres to compete less effectively with beach ball structures than for the cases mentioned above because of the greater initial distance between boundaries. On the other hand, the difference in standoff is less in the present case (Figure 12).
Figure 12. Cross polarized images of solder joints formed by 10, 12 and 16 mil SAC305 solder spheres on Cu/ELNG substrates just after reflow [2]. In fact, the 10 mil sphere joints did quite well. Figure 13 shows the results of 0/100°C cycling with 10 minute dwells. The smallest; 10 mil, solder spheres gave joints that lasted slightly longer than the ones achieved with 16 mil spheres and considerably longer than the ones achieved with 12 mil spheres. As we shall see, this is partially a result of the milder cycling conditions.
Figure 13. Weibull plot showing the cycles to failure of 1.4mm pitch SAC305 BGAs in 0/100°C thermal cycling test with 10 min dwell time. Note the better performance by 10 mil solder spheres. In general failure occurred by solder fatigue in one of the corner joints. Figure 14 shows the characteristic life, N63.2, of our assemblies versus the distance to neutral point (DNP) of the corner joints in 0/100°C cycling. Overall, the largest and smallest solder joints give very similar results while the 12 mil spheres give a longer life for the smallest DNPs and a shorter one for the largest.
Figure 14. Characteristic Life (N63.2) versus DNP of the corner joints for 10, 12 and 16 mil SAC305 solder joints in 0/100°C thermal cycling test with 10 min dwells.
Figure 15. Characteristic Life (N63.2) versus DNP of the corner joints for 10, 12 and 16 mil SAC305 solder joints in 20/100°C thermal cycling test with 10 min dwells. Switching to a harsher thermal cycle, -40/125°C, the interlaced structure doesn’t compete quite as well (Figure 16)
A very similar trend was observed when the minimum temperature was reduced to -20°C (Figure 15).
Figure 16. Characteristic Life (N63.2) versus DNP of the corner joints for 10, 12 and 16 mil SAC305 solder joints in 40/125°C thermal cycling test with 10 min dwells.
Nevertheless, the smallest solder joints still do much better than we would have predicted for beach ball or single Sn grain structures, especially for the larger pitch sizes. (Figure 16) Figure 17 shows a log-log plot of N63.2 vs. DNP for the 10 mil sphere assemblies in the different thermal cycles. The data agree very well (R2 values of 0.97-0.99) with power dependencies, N63.2 ~ DNP-n, with n ranging from 2.8 to 3.2. In fact, all three curves are seen to agree quite well with N63.2 ~ DNP-2.8 (total R2 = 0.9857).
this effect, but it is repeated in each of the three thermal cycles. Results for the three largest DNPs by themselves would be in good agreement with a N63.2 ~ DNP-1.9 dependence, but definitely not with N63.2 ~ DNP-2.8 like for the other solder sphere sizes. The meaning of this is not clear at present.
Figure 19. Log(N63.2) versus Log(DNP) for 12 mil SAC305 in 0/100°C, -20/100°C and -40/125°C thermal cycling test with 10 min dwell time. Figure 17. Log(N63.2) versus Log(DNP) for 10 mil SAC305 in 0/100°C, -20/100°C and -40/125°C thermal cycling test with 10 min dwell time. Figure 18 shows a log-log plot of N63.2 vs. DNP for the 16 mil sphere assemblies in the different thermal cycles. The data agree reasonably well (R2 values of 0.89-0.97) with power dependencies, N63.2 ~ DNP-n, with n ranging from 2.5 to 2.8. Like for 10 mil spheres all these three curves also seem to agree with N63.2 ~ DNP-2.8.
Figure 18. Log(N63.2) versus Log(DNP) for 16 mil SAC305 in 0/100°C, -20/100°C and -40/125°C thermal cycling test with 10 min dwell time. The 12 mil SAC305 data, however, are not nearly as well represented by a power dependence (Figure 19). Rather, there is no significant difference between N63.2 for the two lowest DNP values. We have no immediate explanation for
Overall, it thus appears that completely interlaced LGA joints will exhibit the same dependence on cyclic strain range as the conventional BGA ones. That would mean that the comparison between LGA assemblies and their BGA counterparts does not depend on the nominal strain range. There must therefore be another reason why the comparison depends on the component (Figures 2-5). One possibility might be the abovementioned positive effect of the lower compliance of the shorter LGA joints. The competition between the rigidities of the solder joints, board and component favors the joints increasingly (strain on the corner joint is reduced) as the number of joints sharing the overall load on the array goes up. It is well established that the presence of joints further in, may reduce the strains on the corner joints appreciably [16, 17]. This might explain why the LGAs do comparatively better as the total number of I/Os increases (Figures 2-5). While the nominal cyclic strain range does not matter for the comparison, however, the thermal cycling parameters still seem to do to a minor extent for some other reason. Previous experiments showed the ratio of life in 0/100°C cycling with a 10 minute dwell to that in -40/125°C cycling with a 60 minute dwell to be very similar for LGAs and BGAs [4]. However, a comparison between Figures 17 and 18 seems to suggest a stronger overall effect of cycling temperatures for the LGA joints. To the extent that this is true, LGAs may thus compete better under realistic service conditions than indicated in accelerated thermal cycling. Isothermal Cycling Shock and isothermal cycling tends to favor solder joint failure through the intermetallic bond to one of the pads, or
by solder pad cratering. In this case an enhanced fatigue resistance of the solder is not particularly useful, while greater compliance of the solder would be. The greater rigidity of the shorter LGA joints together with the hardness of the interlaced structure may therefore be a concern. BGA and LGA assemblies soldered with both SnPb and SAC305 were compared in cyclic bending with amplitudes chosen to ensure solder pad cratering.
show significant change in the microstructure as the standoff height was reduced. Figure 22 shows backscattered SEM images of regular cross sections, clearly revealing the cracks in the traces. Continued cycling eventually led to cratering of a number of pads.
Solder pad cratering is not always detectable by nondestructive means, depending on the location and direction of the associated traces. However, in our case electrical failures were detected in-situ and all tracked to the failure of a trace at one corner of the assemblies. A few assemblies were flat sectioned, ground and polished using colloidal silica and alumina suspensions. Figure 20 shows the location of the cracked trace in the flat section. After removing the printed circuit board completely, a cross polarized image revealed de-lamination. In each of the samples tested, the same trace was found to fail first. Figure 22. Backscattered SEM images of (a) SnPb BGA (b) SnPb LGA (c) SAC305 BGA (d) SAC305 LGA. The failure mode was trace fracture in all the samples.
Figure 20. Flat section of assembly showing trace failure. Note that this is the location of first electrical failure in each of the samples tested.
Figure 21. Cross polarized microscope images of (a) SnPb BGA (b) SnPb LGA (c) SAC305 BGA (d) SAC305 LGA. Note the microstructural change as a function of standoff height in the SAC305 alloy. Figure 21 shows cross polarized optical micrographs of the failed samples clearly reflecting the difference in the microstructure of the SAC305 BGA (single grain) and LGA (interlaced twinned). As expected the SnPb alloy did not
Figure 23 shows typical dye and pry results for BGA and LGA joints near the corner after the same number of cycles. Cracks have evolved in the laminate under each pad, but the cracks are clearly smaller for SnPb than for SAC305, and for LGA compared to BGA joints.
Figure 23. Cracks in (a) SnPb BGA; (b) SnPb LGA; (c) SAC305 BGA; (d) SAC305 LGA pads after dye and pry. Note much smaller cracks in LGAs as compared to BGAs. The current sample sizes were very limited, and the rate of cratering varied quite strongly with the component location in spite of the similarity of strains noted above. Rather than interpreting the overall failure distributions we therefore
relied on paired comparisons of times to electrical failure for similar components in the same locations. Figure 24 shows first a paired comparison of the times to failure for BGA assemblies soldered with SAC305 and SnPb. The number of cycles to failure of individual components soldered with SnPb are plotted against values for SAC305 soldered components in the exact same location on a different board. All but one of the points lie above straight line expected if there was no difference, i.e. in spite of the small size we can say, with 95% confidence, that the greater ductility of the SnPb reduces the rate of cratering.
Figure 25. Paired comparison for LGA components involving SnPb and SAC305 solder alloys. Note better performance of SAC305 LGAs.
Figure 24. Paired comparison for BGA components involving SnPb and SAC305 solder alloys. Note better performance of SnPb BGAs. Now, Figure 25 shows a similar paired comparison for the LGA assemblies. In this case we seem to discern a tendency for cratering to be delayed with SAC305 solder compared to SnPb. However, our t-test shows that we can only have 80% confidence in this, i.e. we do not consider the difference significant. Still, this does indicate that the greater ductility of the SnPb offers no significant advantage for the LGAs. It would seem reasonable to assume that a beach ball or single Sn grain structure of the SAC305, if that could be achieved for the LGA joints, would ‘fall in between’ the two, i.e. it would make no difference for these either. Finally, Figure 26 offers a paired comparison between BGAs and the corresponding LGAs. Overall the LGAs seem to fail later than the BGAs, in good agreement with the apparent trend in cratering represented in Figure 23. In the case of SnPb a t-test shows only 85% confidence in this, but at least the LGAs do not give significantly faster cratering. Focusing on the SAC305 we can say, with 95% confidence that the LGAs lead to slower cratering.
Figure 26. Paired comparison of BGAs and LGAs. Circles represent SnPb components and triangles represent SAC305 components. Note better performance of LGAs in general. Overall it thus appears that the reduced compliance of the LGA joints will not be a problem for cratering (and thus presumably not for the intermetallic bonds either). This seems surprising. A combination of finite element modeling and testing of individual pads is ongoing to assess effects of the different stress concentrations and stress distributions (tension vs. torque) at the pad surfaces. Until these issues are resolved, we caution against generalizing the present results. SUMMARY AND CONCLUSION Given a choice between an LGA and a BGA version of the same component the much smaller standoffs of the former offer a lower assembly profile, but this is usually expected to come at the cost of a much reduced life in thermal cycling. Indications are, however, that the robust nature of the shorter joints tends to counteract this reduction somewhat, an effect that is likely to depend on the layout and rigidities of component and printed circuit board. In the case of SAC305 solder the smaller volume LGA joints also tend to undercool more than corresponding BGA
joints after reflow, offering a chance of interlaced twinning. This microstructure is more resistant to thermal cycling induced fatigue, presumably because energy is absorbed into cycling enhanced twin boundary migration. As a result LGA assemblies may perform similar to, or sometimes even better than, corresponding BGA assemblies in accelerated thermal cycling tests. The sensitivity to cycling temperatures may be slightly greater (larger acceleration factors), suggesting that comparisons under realistic service conditions may favor the LGAs even more. Interlaced twinning is far from always ensured for LGA joints, and a major ongoing research effort is focused on the assessment of whether it can be controlled under manufacturing relevant conditions. For cases where interlaced twinning is achieved for LGAs current expectations are that the performance compared to the corresponding BGA version will depend on pad sizes and solder volumes, as well as on the rigidities of component and PCB. It does, however, not seem to depend on the component layout (nominal cyclic strain range). The interlaced microstructure is not stable in aging, but the consequences of this and precipitate coarsening for comparisons remain to be assessed. The interlaced twinning structure also appears to be harder and more creep resistant than the beach ball and single Sn grain structures found in BGA joints, although it remains to be ascertained whether this is really a result of differences in precipitate distributions. So far the less compliant LGA joints do not seem to enhance solder pad cratering in isothermal cycling. In fact, in the case of SAC305 they seem to delay cratering compared to BGA assemblies. This may be associated with differences in the stress distributions at the pad surfaces, but that is something we would need to document and explain convincingly before generalizing the present results. Work is ongoing to address that. ACKNOWLEDGEMENT This work was funded by the AREA (Advanced Research in Electronics Assembly) Consortium and the U.S. Department of Defense, through the Strategic Environmental Research and Development Program (SERDP). The authors would like to thank Antoinette Chvatal, Universal Instruments and Luke Wentlent, Binghamton University for preparation of selected samples. REFERENCES [1] A. Kujala, A. Peltola, S. Akkanen, V. Halkola, “Land Grid Array Packaging Technology in Portable Electronics”, Journal of SMT, Vol. 15, No. 2, (2002), pp. 11-17 [2] Arfaei B., Wentlent L., Joshi S., Alazzam A., Tashtoush, T., Halaweh M., Chivukula S., Yin, L., Meilunas M., Cotts E., and Borgesen P., “Improving the Thermomechanical Behavior of Lead Free Solder Joints by Controlling the Microstructure”, Proceedings of ITHERM (2012) pp. 392-8
[3] L. Wentlent, B. Arfaei, and P. Borgesen, “Damage Mechanisms and Acceleration Factors for No-Pb LGA, TSOP and QFN Type Assemblies in Thermal Cycling”, Proc. IMAPS Mid-Atlantic Microelectronics Conf., Atlantic City, NJ, June 23-24, 2011 [4] L. Wentlent, L. Yin, M. Meilunas, B. Arfaei, and P. Borgesen, “Damage Mechanisms and Acceleration Factors for No-Pb LGA, TSOP, and QFN Type Assemblies in Thermal Cycling”, Proc. SMTA Int. (2011), pp. 101-110 [5] B. Arfaei, L. Wentlent, S. Joshi, M. Anselm, and P. Borgesen, “Controlling the Superior Reliability of Lead Free Assemblies with Short Standoff Height Through Design and Materials Selection”, accepted for publication in the Proc. IMECE 2012 [6] B. Arfaei, L., E. Cotts , and P. Borgesen, “Effect of Sn Grain Morphology on the Reliability of Lead Free Solder Joints”, presented at TMS 2012 [7] B. Arfaei, S. Joshi, P. Borgesen, “Thermal Cycling Induced Failure Mechanisms for Different Lead Free Solder Microstructures”, presented at MS&T 2012 [8] M. Meilunas and B. Berger, “Effects of SAC305 Solder Joint Dimensions on Accelerated Thermal Cycle Reliability”, AREA Consortium Report, 2010 [9] P. Borgesen, “Lead Free Solder Microstructure and Reliability”, presented at AREA Consortium meeting, June 2010 [10] M. Meilunas, “Lead Free Land Grid Array Review”, AREA Consortium Report, 2009 [11] B. Arfaei, N. Kim, E. J. Cotts, “Dependence of Sn Grain morphology of Sn-Ag-Cu Solder on Solidification Temperature”, Journal of Electronic Materials, Vol. 41, No. 2, pp. 362-374, Feb. 2012 [12] L.P. Lehman, S.N. Athavale, T.Z. Fullem, A.C. Giamis, R.K. Kinyanjui, M. Lowenstein, K. Mather, R. Patel, D. Rae, J. Wang, Y. Xing, L. Zavalij, P. Borgesen and E.J. Cotts, “Growth of Sn and Intermetallic Compounds in Sn-Ag-Cu Solder.” Journal of Electronic Materials, vol. 33, no. 12, pp. 1429-1439, Dec. 2004 [13] Yin L., Meilunas M., Arfaei B., Wentlent L. and Borgesen P., “Effect of Microstructure Evolution on Pb-free Solder Joint Reliability in Thermomechanical Fatigue”, Proceedings of the 62nd Electronic Components & Technology Conference (2012) pp. 493-9 [14] L. Yin, L. Wentlent, L. Yang, B. Arfaei, A. Qasaimeh, and P. Borgesen, “Recrystallization and Precipitate Coarsening in Pb-free Solder Joints during Thermo-
mechanical Fatigue”, J. Electronic Materials 41, Issue 2 (2012) pp. 241-252 [15] JEDEC Standard JESD22-B111, “Board Level Drop Test Method of Components for Handheld Electronic Products”, 2003. [16] P. Borgesen, C.Y. Li, and H. D. Conway: “Analytical estimates of thermally induced stresses and strains in flipchip solder joints”, Proc. Joint ASME/JSME Conf. on Advances in Electronic Packaging (W. T. Chen and H. Abe, eds., 1992) 845-854 [17] P. Borgesen, C.Y. Li, and H. D. Conway: “Mechanical Design Considerations for Area Array Solder Joints”, IEEECHMT 16 (1993) pp. 272-283 [18] V. Venkatadri, L. Yin, Y. Xing, E. Cotts, K. Srihari, and P. Borgesen, “Accelerating the Effects of Aging on the Reliability of Lead Free Solder Joints in a Quantitative Fashion”, Proc. ECTC 2009 [19] Z. Yifei, C. Zijie, J. C. Suhling, P. Lall and M. J. Bozack, "The effects of aging temperature on SAC solder joint material behavior and reliability," in 2008 58th Electronic Components and Technology Conference, 27-30 May 2008, Piscataway, NJ, USA, 2008, pp. 99-112. [20] B. Arfaei, T. Tashtoush, N. Kim, L. Wentlent, E. Cotts, and P. Borgesen, “Dependence of SnAgCu Solder Joint Properties on Solder Microstructure”, Proc. 61st ECTC, 2011, pp. 125-132
