Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability

Transcription

1 Journal of ELECTRONIC MATERIALS, Vol. 33, No. 10, 2004 Special Issue Paper Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability HYOUNG-JOON KIM, 1,3 JONG-SOO CHO, 2 YONG-JIN PARK, 2 JIN LEE, 2 and KYUNG-WOOK PAIK 1 1. Nano Packaging and Interconnects Lab. (NPIL), Department of Materials Science and Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejon, , Republic of Korea. 2. R&D Center, MK Electron Co. Ltd., Geuneo-ri, Pogok-myeon, Yongin-Si, Gyeonggi-do hjkim76@kaist.ac.kr The main purposes for developing low-alloyed Au bonding wires were to increase wire stiffness and to control the wire loop profile and heat-affected zone length. For these reasons, many alloying elements have been used for the various Au bonding wires. Although there have been many studies reported on wire strengthening mechanisms by adding alloying elements, few studies were performed on their effects on Au bonding wires and Al pad interfacial reactions. Palladium has been used as one of the important alloying elements of Au bonding wires. In this study, Au-1wt.%Pd wire was used to make Au stud bumps on Al pads, and effects of Pd on Au/Al interfacial reactions, at 150 C, 175 C, and 200 C for 0 to 1200 h thermal aging, were investigated. Crosssectional scanning electron microscopy (SEM), energy-dispersive spectroscopy (EDS), and electron probe microanalysis (EPMA) were performed to identify intermetallic compound (IMC) phases and Pd behavior at the Au/Al bonding interface. According to experimental results, the dominant IMC was Au 5 Al 2, and a Pd-rich layer was at the Au wire and Au-Al IMC. Moreover, Au-Al interfacial reactions were significantly affected by the Pd-rich layer. Finally, bump shear tests were performed to investigate the effects of Pd-rich layers on Au wire bond reliability, and there were three different failure modes. Cracks, accompanied with IMC growth, formed above a Pd-rich layer. Furthermore, in longer aging times, fracture occurred along the crack, which propagated from the edges of a bonding interface to the center along a Pd-rich layer. Key words: Au bonding wire, Pd, intermetallic compound, Au stud bump, Pd-rich layer (Received March 09, 2004; accepted April 22, 2004) INTRODUCTION Due to the downscaling of electronic package dimensions, the size of chip pads has been reduced. To satisfy the requirements of the first level of packaging, technology has been changing; new gold bonding wires, such as fine diameter gold wires for fine pitch bonding and gold wires for gold stud bumping, have been introduced recently. Especially, gold stud bumping is a useful flip chip bumping method, easily obtained by modifying a conventional wire bonding machine, and gold stud bumps on aluminum bond pads can interconnect a die to a substrate using conductive adhesives. Therefore, stud bumping does not require under-bump-metallization (UBM) formation, reflow, and flux cleaning processes. Gold stud bumping technology has several advantages: (1) it requires simple manufacturing processes, such as no UBM, no solder reflow, and no flux; (2) it is cost-effective; (3) it is a well-established technology; and (4) it possesses good electrical and high frequency performances. An important issue of gold stud bumping is to control uniform tail heights, as shown in Fig. 1. To satisfy this requirement, gold wires for stud bumping have been developed by adding several doping or alloying elements, which can be classified as either solid solution dopants, such as Pd, Ag, Pt, and Cu, or interstitial dopants, such as Be, Ca, and rare earth 1210

2 Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability 1211 Fig. 1. SEM image of gold stud bumps. elements. 1 Actually, the purposes of doping were to increase wire stiffness, tensile strength for high wire sweep resistance during epoxy molding compound molding process, 1 5,7 and recrystallization temperature for shorter heat-affected zone formation during the electro-flame off process. 6 However, since the electrical resistivity of doped- or alloyed-gold wires was higher than that of pure gold wire (4 N gold wire), many previous studies were performed to investigate the effects of doping elements on mechanical and electrical properties of gold wires. However, during chip operation, a critical reliability issue of gold bonding wires on aluminum pads is the Au-Al IMC formation and crack formation at gold/ aluminum interfaces. Although the Au-Al IMC formations between 4 N gold wires and Au-1wt.%Pd wires were compared, the role of Pd on Au-Al IMC reaction was not explained. 8 Moreover, the thermal aging time was very short, only 1 h at 300 C. Au-Al IMC formation using Au-1wt.%Pd wire was investigated, and the formation of a Pd-rich layer on Au/Al IMC layers was reported. 9 To identify the phase changes, wavelengthdispersive spectroscopy was used and final IMC phases were Au 5 Al 2 and Au 4 Al. 9 However, the Pd effect on Au-Al IMC formation was not explained. During the Au/Al interfacial reaction, it was reported that Pd was accumulated, and formed a Pd-rich layer on an IMC layer. Consequently, if Pd affects Au/Al interfacial reaction, then the reaction rate of Au/Al IMC growth will be changed before and after the formation of a Pd-rich layer. Moreover, the bonding reliability will also be affected by the presence of a Pd-rich layer. Therefore, the purpose of this study is to investigate the effect of Pd addition on Au/Al interfacial reactions during a thermal aging treatment, and also on bonding reliability using Au-1wt.%Pd wire, generally used for gold stud bumping applications. EXPERIMENTAL PROCEDURE Materials A 1 mil. (25.4 µm) diameter 2 N gold wire (Au-1wt.%Pd type wire), manufactured by the MK Fig. 2. Optical microscopic view of a test chip after gold stud bumping. Electron Co. Ltd., was used to form gold stud bumps. The size of test chips was 1 cm 1 cm, and 124 Al bond pads were peripherally arrayed. The metallization of bond pads was 1 cm thickness of an aluminum layer, and 5 µm thick benzo-cyclo-butene was used for the chip passivation material. Gold stud bumps on aluminum bond pads were formed using a K&S 4522, Telephus Co. Ltd., manual wire bonder under established bumping conditions. An optical image of a test chip after gold stud bumping is shown in Fig. 2. Thermal aging was performed on the prepared test chips at 150 C, 175 C, and 200 C for 0 to 1200 h. IMC Growth Evaluation and Analysis Cross-sectional images of aged samples were investigated using an optical microscope and scanning electron microscopy (SEM), and energy-dispersive spectroscopy (EDS) was used to identify IMC phases. Due to the difficulty of clarifying the shape of a Pdrich layer, the focused ion beam (FIB) technique was performed on a cross-sectional specimen. Furthermore, distribution changes of Au, Al, and Pd elements, with various aging times, were examined using electron probe microanalysis (EPMA). Finally, to reveal the effect of Pd on the Au/Al interfacial reaction, IMC thickness data of each aging temperatures were plotted as a function of annealing times, and then diffusion coefficients of Au atoms were calculated before and after the formation of a Pd-rich layer. Bond Reliability Test Bump shear tests of the aged gold stud bumps were performed using a DAGE 4000 series, KAIST, under the following conditions: bump shear height: 10 µm stylus speed: 100 µm/s number of sheared bumps: more than 30 bumps at each aging condition Finally, fractured pads and bumps were analyzed using an optical microscope and SEM to investigate the failure modes.

3 1212 Kim, Cho, Park, Lee, and Paik Fig. 3. Cross-sectional SEM images of thermal-aged samples. RESULTS AND DISCUSSION Au-Al IMC and Pd-Rich Layer Figure 3 shows the formation of Au-Al IMCs and a Pd-rich layer. At 150 C, aluminum pad metal was not fully consumed by Au-Al interfacial reactions until 10 h thermal aging, and the thickness of IMC was about 1 µm. The IMC phases, identified using EDS, were Au 5 Al 2 and Au 2 Al. However, at 175 C and 200 C, the measured thicknesses of IMC were about 3 µm, and only Au 5 Al 2 IMC phase was identified. After 50 h aging at 150 C, aluminum pads were completely consumed by Au-Al interfacial reactions, and Au 2 Al phase was transformed to Au 5 Al 2 phase. Similar interfacial reactions were observed at crosssectional specimens of 175 C and 200 C. The major IMC phase at a center of Au/Al bonding interface was Au 5 Al 2. Actually, among the five Au-Al IMC phases (Au 4 Al, Au 5 Al 2, Au 2 Al, AuAl, and AuAl 2 ), Au 5 Al 2 showed the fastest reaction rate. 10 Therefore, this phase first formed at an early stage of thermal aging. Because of the limited supply of 1 µm thick aluminum pad metals compared with bulk gold stud bumps, overall equilibrium might not be reached. 9,10 Consequently, the expected final IMC phases, formed at the bonding interface, will be gold-rich IMC phases, and actually, Au 5 Al 2 was confirmed in these thermal-aging conditions. However, the aluminum-rich phase, AuAl 2, was formed at the edge of the bonding interfaces (Fig. 4.), where aluminum was continuously supplied from the unbonded remaining Al pad. A Pd-rich layer was formed at a gold stud bump and the Au-Al IMC (Au 5 Al 2 ) layer. Figure 5a showed the cross-sectional SEM of a thermal-aged specimen at 175 C for 700 h. In Fig. 5a, the Pd-rich layer looked like dots or cluster shapes. However, an FIB image, in Fig. 5b, clearly showed that Pd was accumulated between a gold stud bump and the IMC layer, and formed a continuous layer. According to an EDS quantitative analysis, a maximum of 7wt.%Pd was detected at a Pd-rich layer, even though the original Pd content in Au wire was just 1 wt.%. However, the exact phase of a Pd-rich layer could not be identified by EDS because of its submicron thickness. Diffraction pattern analysis, accompanied by TEM work, should be performed to investigate the exact phase of Pd-rich layer, whether it is IMC between Pd and Al or just accumulation of Pd atoms. Chang reported that after

4 Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability 1213 Fig. 4. Cross-sectional SEM of a thermal-aged sample at 175 C for 950 h. Au 5 Al 2 was the major IMC phase at the center of the bonding interface, but the aluminum-rich phase, AuAl 2, was formed at edges due to the continuous aluminum supply from the remaining Al pads. a Fig. 5. (a) Cross-sectional SEM image of thermal-aged specimen at 175 C for 700 h. (b) FIB image of the same specimen. The area of square ABCD was sputtered by Ga ions, and the section under line AB was observed. b a certain incubation time, Au atoms penetrated through a Pd-rich layer and formed a block-shaped Au 4 Al. 9 However, in this study, the Au 4 Al phase was not detected at IMC layers and EPMA mapping results showed that a Pd-rich layer acts as a good diffusion barrier to interrupt IMC growth (Figs. 6 and 7). At 150 C, a Pd-rich layer clearly appeared after 100 h of thermal aging. The Au-Al IMC grew rapidly up to 100 h, but the IMC thickness did not significantly increase after a Pd-rich layer formed (Fig. 6). In the case of 200 C, a Pd-rich layer already appeared even at 10 h, and the thickness of the IMC did not change even after 700 h, compared with that of 10 h (Fig. 7). This is presumably because Pd atoms, which did not join the interfacial reactions, were accumulated at the gold stud bump and the Au-Al IMC layer. Furthermore, it is clear that a Pdrich layer plays a diffusion barrier, and interrupts continuous Au-Al interfacial reactions for further IMC growth. Effect of a Pd-Rich Layer on IMC Growth Kinetics To investigate the kinetics of Au-Al IMC growth, thicknesses of IMC layers were measured at every thermal aging condition. The measurement of IMC

5 1214 Kim, Cho, Park, Lee, and Paik Fig. 6. EPMA mapping of thermal-aged specimens at 150 C. Fig. 7. EPMA mapping of thermal-aged specimens at 200 C.

6 Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability 1215 interdiffusion of metals will follow a parabolic law as follows: x 2 Kt Fig. 8. Measured thickness of IMC layers versus the square root of aging times. thickness was performed at the center of the bonding interface, and more than 20 gold stud bumps were used for measuring each cross-sectional specimen. According to Philofsky 10 and Kidson, 11 the growth rate of an intermediate phase during an where x is the intermediate phase layer thickness, t is aging time, and K is the rate constant dependent on the interdiffusion coefficients. Therefore, the growth of IMC thickness should be proportional to the square root of aging time. As shown in Fig. 8, graphs of IMC thickness versus t 1/2 were not perfectly straight, and deflection points were observed. At early aging times, Au-Al IMC grew rapidly, and then the rates of IMC growth slowed. This indicates that interfacial phenomena, which interrupted Au-Al interfacial reactions, occur at bonding interfaces, resulting in deflection points. According to the previous EPMA mapping results, times for a continuous Pd-rich layer formation were about 100 h and 10 h at 150 C and 200 C, respectively. Moreover, deflection points in Fig. 8 were located between 50 and 100 h at 150 C, between 25 and 50 h at 175 C, and between 10 and 25 h at 200 C. Figure 9 shows that the results of two-step linear fitting at each aging temperature and aging a b c Fig. 9. Two-step linear fitting of IMC thickness versus thermal aging time 1/2 at (a) 150 C, (b) 175 C, and (c) 200 C aging.

7 1216 Kim, Cho, Park, Lee, and Paik time at intersection points of fitting lines were about 83 h, 37 h, and 15 h at 150 C, 175 C, and 200 C, respectively. Consequently, the aging times at deflection points of two-step linear fitting corresponded well with the results of EPMA mapping, when a continuous Pd-rich layer formed. Therefore, IMC growth kinetics of 1wt.%Pd alloyed gold wire should be divided into two steps. (1) Initial Au-Al interfacial reactions are not affected by a Pd-rich layer. The obtained rate constant, K 1, and activation energy of Au-Al interfacial reaction are similar to those of previous Au-Al interfacial studies. (2) A continuous Pd-rich layer formed at gold stud bumps and Au-Al IMC layers; a Pd-rich layer slows the Au-Al interfacial reaction rate significantly. The obtained rate constant, K 2, depends on the diffusion coefficient of gold into palladium, because gold is the dominant diffusing element for the growth of Au 5 Al 2 phase. 10 Consequently, the K 2 values should be similar to the diffusion coefficient of gold into palladium. Measured rate constants, K 1, are compared with previously reported values 12 (Table I). Although there are about an order of magnitude difference, these are roughly fitted together. The deviations are presumably due to the compositional difference between 2 N ( 99%) and 4 N ( 99.99%) gold wire samples and the experimental errors for measurement of IMC thickness. The obtained activation energy in this study was 15.6 Kcal/mol, which is almost the same as that of Philofsky, Kcal/mol. In conclusion, the effect of Pd in an early aging step 1 can be negligible, and the Au-Al IMC growth kinetics corresponds well with the previous diffusion controlled results. In contrast, calculated K 2 values are shown in Table II. The K 2 values are almost the same at different aging temperatures. In other words, the kinetics of Au-Al IMC growth at step 2 is independent of temperature. It was reported that in the case of Au-Pd interdiffusion at C, defectenhanced diffusion and grain boundary diffusion were dominant processes, and the required diffusivity for satisfying the conditions was about cm 2 /sec. 13 Measured K 2 values agree well with the reported gold diffusivity in palladium, which is much lower than the self-diffusivity of gold, cm 2 /sec. 14 These results fit well with the previous statement. For the growth of Au 5 Al 2 IMC, the major diffusing element is gold, but after formation of a Pd-rich layer, gold diffusion is strongly inhibited Temperature Table II. Measured Rate Constants, K 2 Rate Constant, K 2 (cm 2 /sec) 150 C C C by the presence of a continuous Pd-rich layer. Gold atoms must diffuse through a Pd-rich layer and react with aluminum to grow Au 5 Al 2 IMC phase. Consequently, measured K 2 values represent the diffusion coefficient of Au into a Pd-rich layer. Bond Reliability Test and Failure Mode Analysis To evaluate bond reliability after thermal aging, bump shear tests were performed and the results are plotted in Fig. 10. The bump shear strength decreases as a function of aging time. Three types of failure modes were observed (Fig. 11), and failure mode changes were closely related with IMC growth and the presence of Pd-rich layer. (1) Failure mode A: the fractured interface was between the remaining aluminum pad metal, which did not join the interfacial reaction, and the IMC layer. Therefore, scratched aluminum pad metals were observed at fractured pad sites. (2) Failure mode B: There were two fractured interfaces. First, at bump edge, fracture occurred between the IMC layers and the gold stud bumps; and second, fractured interface was between the IMC layers and SiO 2 at bump centers due to complete consumption of aluminum pad metal. (3) Failure mode C: fracture occurred between gold stud bumps and a Pd-rich layer. Actually, cracks propagated just above a Pd-rich layer, as shown in Fig. 12. For the growth of Au 5 Al 2 phase, gold was the dominant diffusing element resulting in voids formed at gold stud bump sites. The EDS analysis, shown in Fig. 13, indicates that the fractured interface was between a Pd-rich layer and a gold stud bump, because Pd was detected only at the fractured pad sites. Line 1 is the boundary of failure modes A and B. Failure mode B appeared after 10 h, 50 h, and 100 h Table I. Comparisons of Measured Values of the Rate Constant, K 1, and the Reported Rate Constants for Au-Al IMC Formation 12 Measured Rate Reported Rate Rate Constant Constants Constants Temperature (cm 2 /sec) (cm 2 /sec) 150 C C C Fig. 10. Bump shear strength versus thermal aging times.

8 Effects of Pd Addition on Au Stud Bumps/Al Pads Interfacial Reactions and Bond Reliability 1217 a b c Fig. 11. Top views of observed failure modes: (a) failure mode A, (b) failure mode B, and (c) failure mode C. (d) Cross-sectional view. d Fig. 12. Cross-sectional view of crack propagation just above a Pd-rich layer. of aging at 200 C, 175 C, and 150 C, respectively. These times nearly correspond to the times of deflection points in Fig. 9. Failure modes B and C were divided by line 2. The failure modes change after 300 h, 700 h, and 950 h of aging at 200 C, 175 C, and 150 C, respectively. Between lines 1 and 2, IMCs grew slowly due to a Pd-layer formation, and cracks formed at the edges of the bonding interface and propagated to the center area of bonding. Therefore, fracture occurred along the crack during bump shear tests. CONCLUSIONS According to the results of cross-sectional SEM and EDS, the major IMC phase resulting in the Au-Al interfacial reaction was Au 5 Al 2. However, Pd, which did not join the reaction, was accumulated at the gold stud bumps and the IMC layer and formed a continuous layer. The incubation times for the formation of a Pdrich layer corresponded well to the measured times resulting in intersection points of two-step linear fitting. Early rapid growth of Au-Al IMCs during thermal aging treatment was slowed by the presence of a Pd-rich layer interrupting Au-Al interfacial reactions. The obtained activation energy in step 1, 15.6 Kcal/mol, agreed with the previously reported values, and calculated K 2 values were also very similar to the reported diffusion coefficient of gold atoms into the palladium matrix. Actually, these K 2 values were smaller than the selfdiffusivity of gold. Therefore, the formation of a Pd-rich layer slowed the Au-Al IMC growth.

9 1218 Kim, Cho, Park, Lee, and Paik Fig. 13. EDS analysis was performed at (a) fractured Al pads and (b) fractured Au bumps site. Palladium was only detected at fractured Al pad sites. Cracks formed at the edges of a bonding interface, and propagated to the centers along a Pd-rich layer. According to the SEM image, crack formed above a Pd-rich layer at longer aging times. Therefore, Pd was detected only at the fractured pad sites by EDS analysis. ACKNOWLEDGEMENT This work was supported by the Center for Electronic Packaging Materials (CEPM), Korea Science and Engineering Foundation. REFERENCES 1. J. Seuntjens, Z.P. Lu, R. Emily, C.W. Tok, F. Wulff, S.S. Aung, and S. Kumar (Paper presented at SEMICON Singapore Advanced Packaging Technologies Seminar I, May 2001). 2. Y. Ohno and Y. Ohzeki, Nippon Steel Techn. Rep. 59, 5 (1993). 3. C. Simons, L. Schrapler, and G. Herklotz, Gold Bull. 33, 89 (2000). 4. G. Rui, J. Xuan, L. Baoguo, and B. Lizhi, Rare Met. 14, 203 (1995). 5. G. Humpston and D.M. Jacobson, Mater. Technol. 6 7, p. 25 (1993). 6. G. Qi and S. Zhang, J. Mater. Processing Technol. 68, 288 (1997). 7. S. Liu and Y.C. Chao, J. Electron. Mater. 32, 159 (2003). 8. T.D. Hund and P.V. Plunkett, IEEE Trans. CHMT 8, 446 (1985). 9. H.S. Chang, K.-C. Hsieh, T. Martens, and A. Yang, J. Electron. Mater. 32, 1182 (2003). 10. E. Philofsky, Solid-State Electron. 13, 1391 (1970). 11. G.V. Kidson, J. Nucl. Mater. 3, 21 (1961). 12. T. Uno and K. Tatsumi, J. Jpn. Inst. Met. 63, 828 (1999). 13. W.J. Debonte and J.M. Poate, Thin Solid Films 25, 441 (1975). 14. A. Gangulee and F.M. d Heurle, Scripta Metall. 7, 1027 (1973).