CCGA design variables and their effects on reliability: using full array, thin ceramic substrates and lidless packages can improve the thermal cycle fatigue life of CCGAs.
Because of the variety of functions and applications, CCGAs span a range of design variables. These variables include ceramic substrate thickness, column array format (fully populated versus depopulated), lidded or lidless and card thickness. In addition, process variables include the column attach process and the column supplier. Currently, the column last attach solder process (CLASP) is the most used production column attach process. Previously, the cast column attach process had been qualified. With an increase in production volumes, two column suppliers have been qualified. These design and process factors can all affect interconnect reliability. The card pad design and card assembly process also influence interconnect reliability.
The CCGA has been qualified and in production since 1995. (1-3) The early qualification activities assessed the reliability of the package interconnection as representing the worst case form factor. The worst case form factor for a CCGA is the largest body size, the thickest ceramic substrate and the presence of an aluminum lid contacting the ceramic surface.
In most product designs, the column array on a CCGA is fully populated to take advantage of wiring and electrical performance benefits. As the number of CCGA applications has increased, the form factors have become more varied. The functional requirements of the ceramic package tend to determine its ceramic thickness because of its multi-layer structure of wiring and voltage plane layers.
Because the variety of CCGA form factors has increased, design and process effects on the reliability have been investigated. An earlier study optimized the card pad design and card assembly process and also evaluated some package variables such as ceramic thickness and column attach process. (4) For this current study, the following design variables were evaluated across both the cast and CLASP column technologies using both qualified column suppliers:
* ceramic substrate thickness, 1.4 mm versus 3.0 mm
* column array format, fully populated versus depopulated
* lid versus lidless
* card thickness, 1.57 mm versus 2.79 mm.
Ceramic substrate test vehicles
Two 42.5 mm white alumina ceramic test vehicles, TV911 and TV951, were used in this study. Both had an array of 33 x 33 columns on 1.27 mm pitch with the standard 0.86 mm substrate pad diameter. The TV911 was 3.0 mm thick, simulating 20 layers for a typical application-specific integrated circuit (ASIC) using an area array flip-chip footprint. The TV951 was 1.4 mm thick, simulating nine layers for a typical ASIC using a peripheral flip-chip footprint. The daisy chain wiring between the package and the card included nine concentric rings of similar distance from neutral point (DNP) columns, plus one ring that included all of the columns not in the DNP rings.
Bond and assembly process
The packages were tested in both a lidless and a lidded configuration (Figure 1). The lid used was a standard aluminum lid joined to the ceramic surface with a silicone-based adhesive. A thermal compound filled the gap between the ceramic and the lid, as it would if a flip chip die were present. No die was included in the package because the daisy chain wiring is contained within the ceramic substrate. In previous studies on ceramic ball grid arrays (CBGAs), the presence of a chip on a ceramic substrate in this thickness range was shown not to influence the card interconnection reliability. (5)
[FIGURE 1 OMITTED]
Two versions of the CCGA solder joint structure of a Pb90-Sn10 solder column were evaluated. In both cases, the high lead solder column (solidus of 275[degrees]C, liquidus of 302[degrees]C) that does not melt during the standard eutectic Sn63-Pb37 surface-mount card assembly reflow provides a 2.2 mm tall solder joint. This high standoff helps overcome the mismatch in coefficient of thermal expansion (CTE) between the approximately 6.5 ppm/[degrees]C alumina ceramic chip carrier and the 18 to 21 ppm/[degrees]C epoxy glass printed circuit board (PCB).
The cast column process uses the Pb90-Sn10, 2.2 mm (87 mil) solder culunm, but it is totally melted during the column attach process to form a fillet in the shape of the casting mold. The CLASP column attach process joins the Pb90-Sn10 solder column to the ceramic substrate with a eutectic tin-lead solder that has been doped with a small percentage of palladium. (6) This new solder paste composition reflows during the column attach process using the same reflow profile as standard eutectic solder paste. During this process, the palladium forms palladium-tin intermetallics within the eutectic tin-lead matrix. The low temperature joining process of the CLASP column lends itself to a highly automated assembly process. (7) CCGAs of both column types were built using wire from both column suppliers.
Column array configuration
Most CCGAs use a full array of solder columns. The multi-layer structure of the ceramic chip carrier allows the package designer to choose signal, power and ground locations within the array. The design can then be optimized for electrical performance, wireability or a compromise of the two.
Most plastic ball grid array (PBGA) packages depopulate the center area of the array or use the center for only power and ground connections. When a CCGA is designed as a pin compatible alternative to this type of PBGA, the CCGA column array is also depopulated.
A previous unpublished evaluation of a CBGA with a depopulated center array found that depopulating the center degraded the thermal fatigue life. This study compared the reliability of a CBGA with a depopulated array to that of the typical fully populated array. The test vehicles with the depopulated array had 699 columns in the 33 x 33 1089 column array that corresponds to 36 percent depopulation (Figure 2). This pattern was selected to closely represent a proposed product.
[FIGURE 2 OMITTED]
Printed circuit board
The PCB ground rules and dogbone pad dimensions for this 1.27 mm pitch CCGA test card were the same as those developed and optimized during the previous 1.27 mm pitch CCGA study. (4) A design point of 0.75 mm was used for the card pad diameter in a non-solder mask defined configuration. An organic solderability preservative (OSP) surface finish was used. A wiring trace connected the dogbone joining pad to a 0.30 mm diameter plated through hole.
The 4.3 mm X 5.5 mm test card had six CCGAs wired to readout connectors. Two card thicknesses were evaluated: 1.57 mm and 2.79 mm, both with a cross section of four signal and eight power planes (4S8P). Both versions of the test card had the same layout of six CCGA sites wired as described above.
CTE measurements to compare the effect of the card thickness on the average expansion from 25[degrees]C to 100[degrees]C were taken with a dilatometer that was calibrated per fused silica and sapphire NIST standards. The x-values ranged from 16.5 to 17.6 ppm/[degrees]C with no significant difference between the two card types. The y-values ranged from 19.9 to 21.9 ppm/[degrees]C, again with no significant difference between the two card types. Site-to-site and card-to-card differences were essentially within sample and instrument variability and rather close to previous data.
Having equivalent CTEs for two cards of such different thickness may be unusual, but that occurred because the cross sections tend to change with thickness as well. In this experiment, the difference in card thickness was achieved by maintaining the same cross section and increasing the layer thicknesses. In another test vehicle experiment, the combination of an increased number of layers within the card and an increased card thickness combined to increase the CTE. (8)
Card Assembly Processes
The card assembly process for 1.27 mm CCGAs uses standard surface-mount tools and processes. The assembly for this study followed the standard CCGA process guidelines.
A water soluble solder paste was used with stainless steel solder stencils that had chemically etched circular tapered apertures on an automated paste screener typical of assembly manufacturing. The 0.20 mm thick stencil had 0.81 mm stencil apertures to meet the specified solder paste volume range of 3,000 to 7,600 cubic mils, with a 5,000 cubic mil nominal. An automated measurement tool was used to monitor the screened paste volume and record the data.
This solder paste volume range produces small fillets at the base of the column that do not increase the column's rigidity and give the best reliability performance in thermal cycling. CCGA assembly yield is high within this paste volume range, producing no opens or shorts due to inadequate or excessive paste at the joints.
An automated placement tool was used to place the CCGAs. Package pickup was performed from the IEDEC outline shipping trays. For quick placement, body outline alignment was used rather than column vision. The previous study had verified that a placement requirement of 50 percent of the solder column on the card pad resulted in good self-alignment and acceptable solder joints. (4)
A standard surface-mount reflow profile was used with a preheat, ramp and dwell that were recommended by the solder paste manufacturer. The reflow oven was a 12-zone convection furnace with nitrogen atmosphere. The two test cards were run with the same zone temperatures, but at a higher belt speed for the thin card to maintain the same reflow profile.
The trailing CCGA had a peak temperature of 222[degrees]C at the corner and 208[degrees]C at the center. The leading CCGA also had a peak temperature of 208[degrees]C at the center. The bottom and top surfaces of the card had peak temperatures of approximately 235[degrees]C.
The inspection of all test cards showed good solder joint wetting on the exterior CCGA rows. The reflowed solder was shiny, with a smooth transition to the solder column (Figures 3 and 4). No solder opens or bridges were observed.
[FIGURES 3-4 OMITTED]
A matrix of 22 different cells of test vehicle assemblies was thermal cycled at -55[degrees] to +110[degrees]C and 2 cph to evaluate four design variables. The test vehicles included both the cast and CLASP column technologies using both qualified column suppliers across the design variables evaluated. Typically, six samples were in each test cell that combined these variables (Table 1). The design variables were as follows:
* ceramic substrate thickness, 1.4 mm versus 3.0 mm
* column array, fully populated versus depopulated
* lid versus lidless
* card thickness, 1.57 mm versus 2.79 mm.
The effect on CCGA fatigue life for each of the variables was analyzed using the robust design of experiments; all other variables, except for the two being compared, were equally represented by both sets of data (Table 2). A minimum of 23 package test vehicles was included for each variable comparison. A statistical test was applied to determine if the two lognormal distributions being compared (N50 and sigma) were different.
The column supplier and column type were compared first. As expected, no statistical difference existed between the two suppliers or the cast and CLASP columns. Accordingly, the results from both suppliers and column types were combined in analyzing the effects of package rigidity, card rigidity and column array configuration.
The first design variable analyzed, package rigidity, did show a statistical difference. Modeling and experimental CBGA data had both indicated that a more rigid package would have a lower CCGA fatigue life. Two contributors to package rigidity were evaluated in this study. In both cases, the more rigid package had the lower fatigue life. First, increased package thickness increases the rigidity. This effect is more noticeable when no lid is used.
Second, the presence of a lid attached to the ceramic makes the package more rigid. The lid has less of an impact on the thicker package than the thinner package. The impact of a lid that does not attach to the ceramic surface, such as the direct lid attach (DLA) configuration, is expected to be much less.
The lognormal distributions of the four data sets described were plotted to show the relative ranking of the fatigue life (Figure 5). For this comparison, the data from the card thickness cells, in addition to the process variables, were combined. The packages were thermal cycled at -55[degrees]/110[degrees]C and 2 cph.
[FIGURE 5 OMITTED]
The card thickness was also expected to affect the CCGA fatigue life. The thick card was expected to degrade the fatigue life by a few percent due to its greater rigidity when compared to the thin card. For this study, the CTEs of the two test vehicle cards were equivalent, allowing a straightforward comparison of fatigue life effect due to thickness alone.
Two comparisons to evaluate the effect of card thickness were completed. The first used the thin, lidless packages that would be expected to show the most difference. The second used the thick, lidded packages that would be expected to show the least difference.
No statistical difference in fatigue life was observed for the thin package, but a small difference was observed when using the thick, lidded package. For the thick, lidded package, a large difference existed in the distribution of the data of the two types of cards. The sigmas of the lognormal distributions were set equal, and the N50s recalculated lot easier comparison.
Finally, the effect of column depopulation on column fatigue life was evaluated. From prior studies of depopulated flip chip and CBGA arrays, the depopulated array was expected to have a lower fatigue life. The results of this comparison of a depopulated and a full CCGA array also exhibited this same effect. Again, a large difference existed in the sigmas for the two sets of data, so they were set equal and the N50s were recalculated for an easier comparison (Figure 6). The packages were thermal cycled at -55[degrees]/110[degrees]C and 2 cph.
[FIGURE 6 OMITTED]
Experiments comparing several design features across two column types have furthered the understanding of the thermal fatigue life of 1.27 mm pitch CCGA card assemblies. Based on these studies, design recommendations such as using a full CCGA array, thin ceramic substrates and lidless packages can improve the thermal cycle fatigue life.
The trend for improved fatigue life with a less rigid CCGA by either using a thinner ceramic substrate or a lidless configuration was expected based on earlier CBGA studies. Likewise, the degradation in fatigue life when depopulating the column array followed the trend observed in testing both C4 flip-chip devices and CBGAs.
This study again demonstrated the equivalent reliability, independent of column supplier, of the cast and CLASP processes in the 1.27 mm pitch format. Finally, the two cards of different thicknesses, but with equal CTE, gave similar fatigue life, indicating that the card CTE may be the overriding influence.
The data generated here can be applied to many CCGAs used to package numerous chip functions in a wide range of system applications. This package will continue to grow in popularity as its benefits of high interconnection density, high I/O capability and compatibility with standard surface-mount processes are more fully understood.
TABLE 1: Test matrix, package and card design and process variables. Cell Card Lid? Supplier Column Ceramic Array Qty 1 Thick Yes 1 CLASP Thick Full 5 2 Thin Yes 1 CLASP Thick Full 6 3 Thick Yes 2 Cast Thick Full 6 4 Thin Yes 2 Cast Thick Full 6 5 Thick Yes 2 CLASP Thick Full 6 6 Thin Yes 2 CLASP Thick Full 6 7 Thick No 1 CLASP Thin Depopulated 6 8 Thin No 1 CLASP Thin Depopulated 6 9 Thick No 2 Cast Thin Full 5 10 Thin No 2 Cast Thin Full 6 11 Thick Yes 2 Cast Thin Full 5 12 Thin Yes 2 Cast Thin Full 6 13 Thick No 1 CLASP Thin Full 6 14 Thin No 1 CLASP Thin Full 6 15 Thick No 2 Cast Thick Full 6 16 Thin No 2 Cast Thick Full 6 17 Thick No 1 CLASP Thick Full 6 18 Thin No 1 CLASP Thick Full 6 19 Thick No 2 Cast Thin Depopulated 5 20 Thin No 2 Cast Thin Depopulated 6 21 Thick Yes 1 CLASP Thin Full 5 22 Thin Yes 1 CLASP Thin Full 6 TABLE 2: Fatigue life results for design of experiments. N50 Experiment Variables Cell No. (cycles) Sigma Column Supplier A 1,2 630 0.14 B 5,6 680 0.20 Column Attachment Cast 3,4 670 0.10 Process CLASP 5,6 680 0.20 Ceramic Thickness with Thick 15,16,17,18 760 0.19 No Lid Thin 9,10,13,14 980 0.16 Lid on Thick Ceramic Yes 1,2,3,4 650 0.12 No 15,16,17,18 760 0.19 Lid on Thin Ceramic Yes 11,12,21,22 610 0.14 No 9,10,13,14 980 0.16 Card Thickness on Thin, Thick 9,13 950 0.20 Lidless Package Thin 10,14 1,020 0.11 Card Thickness on Thick 1,3,5 690 0.20 Thick, Lidded Package Thin 2,4,6 600 0.04 Card Thickness on Thick 1,3,5 690 0.15 Thick, Lidded Package Thin 2,4,6 640 0.15 with Equal Sigmas Array Format Depop 7,8,19,20 800 0.04 Full 9,10,13,14 980 0.16 Array Format with Equal Depop 7,8,19,20 830 0.13 Sigmas Full 9,10,13,14 990 0.13
The authors gratefully acknowledge the contributions of Kim Berger, Armando Cammarano and Bob Rita to this project and article.
(1.) Caulfield, T., et al. 1993. Surface mount array interconnections for high I/O MCM-C to card assembly. Proceedings of the 1993 International Conference and Exhibition on Multichip Modules, pp. 320-325, April.
(2.) Banks, D., et al. 1993. Second-level assembly of column grid array packages. Proceedings of SMI, pp. 92-98, August.
(3.) Master, R., et al. 1995. Ceramic column grid array for flip chip applications. Proceedings of the 45th Electronic Components and Technology Conference, pp. 925-929, May.
(4.) Phelan, G., et al. 1995. Card assembly and reliability of 44 mm ceramic solder column array modules. Proceedings of Nepcon West, pp. 1048-1058, February.
(5.) Martin, G., et al. 1997. The effect of substrate thickness on CBGA fatigue life. Proceedings of Surface Mount International, pp. 172-177, September.
(6.) Ray, S., et al. 1999. CLASP ceramic column grid array technology for flip chip carriers. Proceedings of the 2nd Annual Packaging Symposium at Semicon West, pp. F1-F7, July.
(7.) Achard, L. and I. DeSousa. 2000. A new manufacturing process for high volume production of ceramic column grid array modules. International Journal of Microcircuits and Electronic Packaging, Vol. 23, No. 4, pp. 451-455.
(8.) Ingalls, E., et al. 1998. Improvement in reliability with CCGA column density increase to 1 mm pitch. Proceedings of the 48th Electronic Components and Technology Conference, pp. 1298-1304, May.
Marie S. Cole is a senior technical staff member, Ellyn Ingalls is an advisory engineer, Gregory B. Martin is a senior engineer, and Cynthia Milkovich is a senior engineer, all with IBM Microelectronics, Hopewell Junction, NY; e-mail: email@example.com.
|Printer friendly Cite/link Email Feedback|
|Date:||Jan 1, 2002|
|Previous Article:||0201s: an inspection nightmare? Both optical and x-ray inspection should be used to reduce scrap and rework in assembling 0201s.|
|Next Article:||2002 salary survey questionnaire.|