4.9 Advanced Concepts .1 Fine Pitch
4.9.3 Higher Frequency Bonding
Stacked die (See Fig. 4.26) present their own set of issues, but in general, the problems involve multiple geometries in a given component package with closely spaced wirebonds that can overlap. In addition, sometimes the bonding must be done to chips that are cantilevered over another chip without a means of mechan- ical support under the bonding pad areas. Fixturing and very careful control of bonding parameters (reduced force and power, higher frequency, and tempera- ture) has allowed successful wirebonding to stacked geometries with as many as six chips. A full discussion of the details of wirebonding to stacked chips is not possible in this work, but some insight can be gained by reading Yao, et al., [88].
made several decades ago based on transducer (bonding head) dimensions for microelectronic assemblies and stability during the bonding (transducer load- ing) operation [38]. Other frequencies from 25 to 300 kHz have been used to attach wires. Ultrasonic welding and material softening have been reported in the range between 0.1 Hz [87] and 1 MHz [53]. Today’s interest in higher frequency bonding stems from reports by various authors [69, 75, 82, 40, 44, 34] that using higher ultrasonic frequencies produces better welding at lower temperatures in shorter bonding times (dwell times). It has also been indicated that higher frequency wirebonding improves bonding to pads on soft polymer layers such as Teflon1, unreinforced polymide, and flexcircuits. While all these improvements were real for the particular situations in hand, few if any con- trolled studies (systematic, side-by-side experiments on the same substrates with an attempt to control all variables except frequency) have been performed. The following material presents excerpts from the one such study [20, 21].
Three metallization schemes were used in this study: (1) aluminum (99.99%
pure) with a titanium/titanium nitride (Ti/TiN) adhesion layer; (2) aluminum plus one percent silicon alloy (Al + 1% Si) again with a Ti/TiN adhesion layer; and (3) gold metallization with a titanium-tungsten adhesion layer (TiW). The metal bonding pad formation layers were sputter deposited to thicknesses between 1 and 2mm on silicon base layers. The silicon wafers were p-type with a nominal resistivity of 30–50.cm. The wafers were thermally oxidized to achieve a SiO2 thickness of 1mm prior to metal deposition or spin coating with polymide. The polymide layers were between 5 and 20mm in thickness. The gold metallization was also deposited on highly polished ceramic (99.6% pure alumina) substrates.
Various test structures were photolitographically patterned on each of the metal layers [16, 18]. The patterns included: arrays of bonding parts of varying sizes (150–25 mm square), a daisy chain pattern consisting of almost 650 wirebonds with the resistance of the wirebonds accounting for over 60% of the total resistance of the circuit, and a radially distributed wirebond pattern for shock and vibration testing.
All wirebonding for the study was performed with two semiautomatic thermo- sonic ball bonders (Marpet Enterprises, Inc., Model 827) equipped with negative electronic flame off (Uthe Technology, Inc., Model 228-1) for uniform control of free air ball size. The flame offs were adjusted to produce 602mm diameter free air balls as shown in Fig. 4.10. Free air balls as small as 22mm in diameter have been formed with 15mm diameter wire. One of the MEI Model 827 wirebonding machines was equipped with a UTI Model 25ST (64.1 kHz) transducer driven by a standard UTI Model 10G ultrasonic generator. The other Model 827 wire- bonding machine was equipped with a UTI Model 4ST (99.5 kHz) transducer which was driven by a UTI 10G generator tuned for 100 kHz. In order to make both transducer waveforms similar, since the Model 25ST transducer is much larger than the Model 4ST, a short 60 kHz transducer (Model 17STL (63.1 kHz) was also used. A comparison of the transducer dimensions has been given previously [18]. Squashed ball sizes were also quite uniform with diameter ran- ging between 76 and 82mm depending upon substrate.
This study has yielded a large amount of data. Key observations and findings include the following. It is clear that significant differences exist between bonding at nominally 60 kHz and bonding at 100 kHz. In addition to differ- ences in transducer electronic waveforms between the standard 60 kHz (long) and the 100 kHz transducer, there exist differences in bonding machine opti- mization behavior. The 60 kHz system appeared to have a larger bonding window (i.e., for a given force and substrate temperature, a wider range of ultrasonic power and dwell times produced acceptable bonds (strong, yet not over bonded or with wire damage)) when compared to the bonds produced by the 100 kHz system. The 100 kHz bonding window, in addition to being smaller than the 60 kHz window, was also sharper (i.e., a smaller change in ultrasonic power and/or dwell in relationship to the window edge was required to go from acceptable bonding to either a no-bond condition or to an over-bonded condi- tion when compared to the 60 kHz system). Despite the smaller, sharper fall-off of the bonding window, the 100 kHz system has one obvious advantage. It formed strong bonds in times that are 30 to 60% shorter than comparable dwells for the 60 kHz system. Comparison of both bonding systems and their transducer waveforms indicate that the 100 kHz system has much faster bond- ing pulse rise and fall times, along with a more stable voltage (or current) amplitude envelope than that of the 60 kHz system. Switching to a short 60 kHz transducer with dimensions comparable to those of the 100 kHz trans- ducer produced ultrasonic drive parameters (voltage and current) similar to those of the 100 kHz transducer.
Shear test data on gold substrate metallizations on high polished ceramic showed that an optimized 100 kHz system produced much stronger bonds than the 60 kHz system (See Table 4.10). As can be seen from Fig. 4.10 and Table 4.11, this difference cannot be accounted for by average ball diameters (either pre- or post-bonding), which were essentially the same for both the 60 and 100 kHz systems. When the data was analyzed for the Al + 1% Si metallization (on oxidized silicon), the 60 kHz bonds appeared stronger. Although the difference between the 60 and 100 kHz test results was relatively small (less than 7%).
However, when analysis of variance techniques were applied, the difference was significant at the 99% confidence level. Similar results were observed on thermo- sonic ball bonds attached to an integrated circuit chip (Al + 1% Si Table 4.10 Gold thermosonic ball bond shear strength (grams (force)) on gold and aluminum (1% S1) metallizations at both 60 and 100 kHza
Metal 60-kHz 100-kHz means Significantb
Au (on ceramic) 68.43.7 84.86.5 16.4 Yes (highly)
A1 + 1% Si (on silicon) 54.03.2 50.62.9 3.4 Yes
means 14 34.2
Significantb Yes (highly) Yes (highly)
aNominal sample size at each frequency was 100
b99% confidence that the difference in the means is significant using analysis of variance with the F-test
metallization), on which both the ball shear test and the wirebond pull test gave a small edge to the 60 kHz system. Although this data set was relatively small, the student’s t-test indicated that the results were significant at the 99% confidence level. Independent of frequency, the difference in ball bond shear strengths between metallization types, were relatively large and highly significant. Bonds on gold were always stronger than bonds on Al + 1% Si metallization consistent with the results shown in many previous studies [12] (Charles, 1986 and Charles, et al., 1999).
Other differences were observed such as asymmetry of ball shape with metallization type. While no differences in average ball diameters [(X-diameter + Y-diameter)/2] were observed with frequency (Table 4.11). Any variations in average ball diameters, even those between metallizations (Table 4.11) could be accounted for by variations in the free air ball size between experimental series.
On the other hand the differences in the X and Y diameter measurements are highly significant and appear to depend on metallization type (Table 4.12). On gold metallization, the as bonded ball diameter in the Y-direction or the direc- tion of the ultrasonic scrub is larger than the orthogonal nonscrub diameter (X- direction) with consistent measurements for both 60 and 100 kHz. On Al + 1%
Si, the non-scrub direction (X-direction) is larger than the Y-direction by a Table 4.11 Gold thermosonic ball bond average diametersa(mm) on gold and aluminum (1%
Si) metallizations at both 60 and 100 kHzb
Metal 60-kHz 100-kHz means Significantc
Gold 89.14.0 88.32.9 0.8 No
A1 + 1% Si (on silicon) 91.32.3 92.02.0 0.7 No
means 2.2 3.7
Significantc Yes Yes
aAverage diameter ¼ 1nPn
i
Xi þ Yi
ð Þ=2
½ :
bNominal sample size at each frequency was 100
c99% confidence that the difference in the means are significant using analysis of variance with the F-test
Table 4.12 Gold thermosonic ball bond diameters (in mm) in directions perpendicular (X-direction) and parallel (Y-direction) to the direction of the ultrasonic scrub on gold and aluminum 91% Si) metallizations at both 60 and 100 kHza
Metal Frequency
X- direction
Y- direction
means Significantb Gold (on ceramic) 60 kHz 84.94.8 93.25.2 8.3 Yes (highly) 100 kHz 82.83.4 93.84.0 11.0 Yes (highly) A1 + 1% Si (on
Silicon)
60 kHz 93.92.9 88.73.4 5.2 Yes (highly)
100 kHz 98.72.8 85.42.6 13.3 Yes (highly)
aNominal sample size at each frequency was 100
b99% confidence that the difference in the means are significant using analysis of variance with the F-test
significant amount for both the 60 kHz and 100 kHz bonding systems. Similar behavior was also observed for pure aluminum metallization. The cause of this phenomena is not well understood, but is believed to be associated with the dynamics of the weld formation process. On gold there is the single interdiffu- sion of the gold wire and gold pad materials. On aluminum and aluminum alloys the formation of gold-aluminum intermetallics is key to the bonding process. The formation of the relatively hard intermetallics may tend to lock the developing bond in the direction of the scrub on the aluminum and aluminum alloy metallizations while on the gold (being relatively ductile) the bond may be able to fully expand in the scrub direction.
Table 4.13 shows results for both 60 and 100 kHz bonded samples under conditions of thermal aging (120 h at 1508C). Aging at 1508C has been shown to be very effective [16] for assessing wirebond (ball bond) quality and relia- bility without introducing unwanted effects caused by substrate interactions and other heat-related phenomena. Table 4.13 again illustrates the significant improvement in shear strength using 100 kHz bonding on gold metallization, this time for gold on a silicon substrate as compared to the gold on ceramic data given in Table 4.10. The small observed differences on the Al + 1% Si metallization for 60 kHz versus 100 kHz is also consistent with the results in Table 4.10, although in this case the difference is statistically insignificant at the 99% confidence level. Again, large and significant differences were observed in the shear strengths between the two metallizations with bonds to gold being much stronger than bonds on Al + 1% Si metallization. These results are consistent regardless of the bonding frequency. After aging, the shear strength of the bonds on gold, at both frequencies, remained essentially unchanged. On the Al +1% Si metallization the strength of the bonds increased significantly for both frequencies. Again, 100 kHz bonding pro- duced stronger bonds on gold metallization, while 60 kHz bonding appeared to have a slight edge on Al + 1% Si. The increase strength for the aged bonds Table 4.13 Gold thermosonic ball bond shear strength (grams (force)) on gold and aluminum (1% Si) metallizations at both 60 and 100 kHz under conditions of thermal aginga
Metal Agedb 60 kHz 100 kHz means Significantc
Gold (on silicon) No 81.44.6 97.43.7 16.0 Yes (highly)
Yes 82.13.3 96.44.6 14.3 Yes (highly)
means 0.7 1.0
Significantc No No
A1 + 1% Si No 47.03.7 46.54.3 0.5 No
(on silicon) Yes 57.83.3 56.14.1 1.7 Yes (slightly)
means 10.8 9.6
Significantc Yes (highly) Yes (highly)
aNominal sample size at each frequency was 100
b120 hours at 1508C
c99% confidence that the difference in the means are significant using analysis of variance with the F-test
on the Al + 1% Si metallization is consistent with similar increases reported previously under aging [14], but the timeframe for the existence of the increased strength above the as-bonded condition appears to be longer in this particular experimental series.