Вы находитесь на странице: 1из 8

Pd-coated Cu Wire Bonding Technology: Chip Design, Process Optimization, Production Qualification and Reliability Test for High

Reliability Semiconductor Devices


Inderjit Singh1,*, Ivy Qin2,**, Hui Xu3,***, Cuong Huynh2, Shin Low1, Horst Clauberg2, Bob Chylak2 and Viola L. Acoff3 1 Xilinx Inc., 2100 Logic Drive, San Jose, CA 95124, U.S.A. 2 Kulicke and Soffa Industries Inc., Fort Washington, PA 19034, U.S.A. 3 Department of Metallurgical and Materials Engineering, The University of Alabama, Tuscaloosa, AL 35404, U.S.A *inderjit.singh@xilinx.com; **iqin@kns.com; *** hxu14@bama.ua.edu Abstract Due to the increasing cost of Au, the semiconductor industry has gone through a dramatic shift away from Au wire bonding to Cu wire bonding. Pd-coated Cu (PdCu) wire has become the favored wire choice over bare Cu wire for fine pitch applications. The advantages of PdCu wire include better HAST reliability results and a more stable stitch bond process window. This paper will examine the critical aspect of bringing Cu and PdCu wire technology from chip design to wire bonding process development to qualification, reliability test and final production. Chip design needs to take into consideration the requirement of Cu wire bonding. Compared to Au wire bonding, Cu wire bonding has many more challenges including pad material push out (pad splash), Al layer peel off (pad peel) and crack in the barrier and dielectric layer (pad crack). This paper considers the impact of various pad structures & metallization to PdCu wire assembly quality and reliability. Because of the unique challenges facing Cu wire bonding processes, wire bonding process development becomes one of the bottlenecks in the assembly process. To shorten the development cycle and ease the requirement for extensive Design Of Experiments (DOEs), a new model based approach was developed and implemented on the latest Cu wire bonder. With this approach, the bonding parameters are calculated based on numerical models. It reduces the need for parameter optimization. This paper will also discuss how modifications and newly developed metrology tools are used in the optimization of PdCu wire. Because of the narrow process window in Cu and PdCu wire bonding, achieving reliable production and high yields requires the process to have a sustainable process window and be portable and transferable to multiple machines and be robust across material variations commonly encountered in a production environment. This paper examines the methods used in quantifying process windows and portability of the wire bonding equipment to ensure high reliability for the end product. Development and optimization of PdCu wire bonding requires a fundamental understanding of interfacial reaction in bonds during both bonding and reliability tests, as proper formation of intermetallic compounds is beneficial to bonding strength. While the knowledge of intermetallic growth in bare copper wire bonds on aluminum has become well established in the past five years, there is little literature in terms of the morphology and growth of intermetallic phase between PdCu wire and Al pad. This paper ascertains the intermetallic growth and phase transformation, retention of aluminum oxide, and behavior of palladium in PdCu wire bonds using high resolution transmission electron microscopy (HRTEM) of dual beam focused ion beam (FIB) thinned specimens. Introduction Wire bonding technology has been around for over 60 years. Knowledge about Au wire bonding process is extensive and mature. Design, equipment and process, from the front end of chip design and manufacturing to the back end wire bonding, molding and reliability tests, were developed and optimized for the Au wire bonding process. With the quick adaptation of bare Cu and PdCu wire, there are knowledge gaps in our understanding of Cu wire bonding that need to be closed. Cu wire bonding introduces new requirements and considerations including chip design and manufacturing, packaging material, equipment and process, and production control in packaging and assembly. There has been a lot of research and development work done in the Cu and PdCu wire bonding area. This paper will present and discuss our findings in chip design, process optimization, intermetallic formation and the critical parameters to control in the assembly process. The challenges in Cu and PdCu wire bonding are similar. To simplify the text in this paper, when Cu wire process is mentioned, it generally implies both bare Cu wire and PdCu wire bonding process. Some of the main challenges are oxidation free Cu ball formation, limiting pad damage, obtaining a robust process recipe, maintaining a robust process in production and passing reliability in a humid environment. Details on these challenges and potential solutions are discussed in [1]. Some of the recent advances include a CFD (Computational Fluid Dynamics) aided cover gas delivery system to ensure oxidation-free ball formation, understanding the root cause of the pad damage issues, developing the metrology and evaluation criteria for Cu wire bonding, choosing the right wire bonding process per application and setup, and calibration of the wire bonding equipment to ensure production stability. There are some differences between Cu and PdCu wire bonding. PdCu wire has shown better reliability under humid conditions such as the HAST test than bare Cu wire [2, 3, 4]. PdCu wire has been shown to have a significantly greater stitch bond window and a more robust 2nd bond [5, 6, 7]. PdCu wire also shows more stable ball formation than Cu wire. In fact, Cu wire tends to need forming gas for consistent, round free air ball formation. In contrast, round and repeatable balls can typically be formed in nitrogen for PdCu wire [2, 8]. However, the PdCu free-air ball is harder [6, 9] and as a result, may cause more pad damage including peeling and cratering [5, 7]. Due to the benefit of PdCu wire, Pd-coated Cu (PdCu) wire has become more commonly used than Cu

978-1-4673-1965-2/12/$31.00 2012 IEEE

1089

wire in fine pitch Cu wire bonding where the challenge of process stability and passing reliability test becomes greater. To our knowledge, PdCu is more popular than Cu wire for 0.9 mil and finer wire diameters. In this paper, the work is done with 0.8 mil wire and PdCu wire is chosen for its reliability performance, more stable free air ball formation and 2nd bond performance. In the session below, we will discuss several key aspects of developing a robust Cu wire bonded application including chip design, wire bonding process optimization using newly developed metrology, understanding key variables that can impact the process stability and methods to control these variables. We will also consider the fundamentals of IMC growth. Chip Design Consideration for Cu wire bonding Most chip design today includes bond pads on active circuit or CUP (Circuit Under Pad). These pads are sensitive to bonding parameters and can be easily damaged if the bonding parameters are not optimized, especially for Cu wire bonding where the Cu wire is harder than previously used Au wire. The pad damage can be in the form of pad cratering which is most severe, or minor cracks in the low K ILD layers which overtime can results in metal migration and cause short circuits. In both cases, it is a major reliability concern. So the pad structure design consideration is very important for the success of Cu wire qualification results. A robust pad design could consist of dummy metal structures connected by seas of vias which bridge the bonded Cu wire to inner circuits within the chip. Other factors that could impact the reliability of Cu wire bonds are related to Al pad thickness. The pad thickness is a critical factor related to long term reliability. Thinner Al pads are known to have very little Al remaining under the bonded ball and over time can separate and cause an open. Another reason to have a thicker Al structure is for its ability to absorb the bonding force and ultrasonic energy so that the underlying pad structure is not damaged. Al pad thicknesses of less than 8000 are known to have shown reliability margin issues. Al pad thicknesses of 12000 to 15000 tend to be optimum and highly recommended. Some caution that needs to be highlighted with thicker Al is that it can result in much higher splash occurrence. So bond parameter control and optimization is extremely critical. The newer generation Cu wire bonders are well equipped with necessary bond parameter algorithms and controls that have shown wider process windows and have achieved higher reliability margin capabilities. Wire Bonding Process Optimization Because of the many challenges facing Cu wire bonding processes, wire bonding process development requires experience, a systematic approach, as well as good quantification methods (metrology). A review of the wire bonding metrology is given Table I and a comparison is made between Au and Cu wire bonding processes.

Table I. Summary of wire bond metrology for Au and Cu.


Metrology Ball Diameter and Ball Height Aluminum Splash Ball Shear Wire Pull (at 1st bond) Intermetallic Coverage (IMC) Al Remain Dielectric Crack Au and Cu Difference Key dimensional measurement for Au. Secondary measurement for Cu Not typically done for Au. Key dimensional measurement for Cu Key bond strength measurement for Au. Secondary for Cu. Secondary bond strength response for Au. Key bond strength and pad damage measurement for Cu. Key bond strength measurement for both Au and Cu. Not typically done for Au. Commonly required for Cu through cross section measurement. Not typically done for Au. Commonly required for Cu through etching the top metal layer.

Reliable Cu wire bonding processes need to meet the requirements in three areas: dimension, bond strength and pad damage. In Au processes, dimensional requirements are determined by meet the ball size and height spec. In Cu processes, Al splash is usually the limiting factor because normally Al splash is bigger than the ball diameter. Bond strength is evaluated by ball shear, pull and IMC formation. In Au processes, shear is normally the key measure of bond strength; in Cu wire bonding, pull strength and IMC coverage are more important because shear strength is strongly influenced by the splash amount [1, 5]. Higher shear does not necessarily correspond to a stronger bond. Cu wire bonding also has the particular challenge of pad damage, which can be assessed by dielectric crack, pad peeling during pull test, and Al remains. Al remains are measured on the cross section of the Cu ball. The final determination of a good Cu process is the reliability testing. A few examples of the Cu metrology are given in Figure 1.

1090

further optimize the process. For example, the Bond Strength Adjust parameter will adjust the ultrasonic current level to achieve the best bonding strength for the given device, capillary and wire.

Figure 2. Graphic user input (partial) for the new model based ProCu processes. An alternative to this approach is to build up a process recipe library. The recipe may be recorded along with information such as bonded ball diameter, device type and capillary dimensions. Future application development can leverage the previously developed process recipe. The process library can be used in conjunction with the new ProCu models to further reduce the complexity of process optimization. The model based ProCu process is used on the two Xilinx devices (Type A and Type B). The Al metallization is the same except that Type A is thinner than Type B. The wire is Nippon Micro Metal EX1 PdCu wire of 0.8 mil (20 m). The capillary is Peco B0809.5-26-11.5-05. The target ball diameter in this case is set to 38m. After inputting the target ball diameter, the bond strength adjustment is set according to the capillary type. For ProCu processes, we developed a step by step approach to help process engineers quickly derive the optimal parameters for their application. The first step is to adjust the free air ball size to make sure the free air diameter is the correct setting to the application. This is done by the Height Adjust parameter. In second step we check for the deformation amount, and modify the Diameter Adjust parameter to achieve the desired deformation. The third step involves a fine tuning of the Bond Strength Adjust (BSA) to make sure that ultrasonic level is the optimal setting for this process. Table II shows the bonding results for both device types for different Bond Strength Adjust (BSA) offsets. This test was performed as the third step of the process optimization using ProCu process. The DOE was performed around the baseline BSA setting, which is 78% for this application. BSA at 78% means the ultrasonic current is at 78% of a calculated ultrasonic energy level. 80% is a typical setting for this capillary type. 0, 5% and 10% offset to the USG energy are tested in this case. In term of BSA values, the settings are 70%, 74%, 78%, 82% and 86%. The process responses starting from top of the table are ball diameter, ball height, Al splash, Shear, Pull average, Pull minimal, IMC percentage, Ball lift, Pad peel and Pad crack percentage. The superscripts denote the device type. For example, Ball DiaA is ball diameter measurement for device type A. In the last column, the difference between the 5 cell average of device A and device B is given as Device A average Device B average.

Figure 1. Examples of Cu metrology Optimization of a Cu wire bonding process is generally more complicated. Au wire bonding processes mainly need to meet the ball size and bond strength spec. Cu wire bonding needs to meet the additional requirement of Al splash and pad damage including Al remain and dielectric crack. Some of the metrology is time consuming. For example, the cross section of the bond is a time consuming process even with the latest cross section equipment such as the ion beam milling machine. In the case of IMC measurements, K&S vision and process engineers developed an in house measurement system to improve the accuracy and speed of the IMC measurement. The output is shown in Figure 1. The area inside the outer green line shows the ball to pad contact area. The areas inside the red lines shows the non IMC area, while the rest of the area is the IMC area. The IMC is commonly defined as the percentage of the IMC area over the contact area. Due to the increased complexity of Cu processes, the process recipe development can be very challenging especially for inexperienced process engineers. To develop the process recipe quickly and achieve more optimal outcome, a new model based approach is developed and implemented on the latest Cu wire bonders. The numerical models are derived from data obtained for different size bonded balls with different wire diameters and capillaries. Instead of inputting traditional bonding parameters such as bond force, impact speed and ultrasonic current levels, the main input is the target bonded ball diameter. All other bonding parameters are calculated based on the Target Ball Diameter. This approach reduces the need for parameter optimization and helps a user to converge to a well optimized process in less time. Figure 2 shows an example of a user interface. Because of the many variables in wire bonding processes (e.g., wire type, capillary dimensions and material, device metallization, etc.) the default bonding parameters may need adjustment. A set of fine tuning adjustments are provided to give user the control of the bonding parameters in case the user needs to

1091

Table II. Ultrasonic level (BSA) test results for two devices (Type A and Type B). Last column shows device A average device B average.
BSA Offset Ball Dia (um) Ball Dia (um) Ball Height (um) Ball Height (um) Al Splash (um) Al Splash (um) Shear (g) Shear (g) Pull Avg (g) Pull Avg (g) Pull Min (g) Pull Min (g) IMC (%) IMC (%) Ball Lift (%) Ball Lift (%) Pad Peel (%) Pad Peel (%) Pad Crack (%) Pad Crack (%)
B A B A B A B A B A B A B A B A B A B A

-10% 38.7
38.6

-5% 39.1
38.5

0 38.9
38.5

5% 38.8
38.5

10% 38.7
38.3

Device Avg Diff 0.3 0.3 -1.5 -1.8 0.4 0.8 0.7% 0.0 0.0 0.0

10.3
10.0

10.3
10.0

10.3
9.9

10.3
9.9

10.3
9.9

40.2 41.5 11.8


14.1

40.9 42.5 12.4


14.5

42.0 43.9 13.2


15.0

42.7 44.0 13.8


15.3

43.4 45.0 13.9


15.2

9.9
9.5

9.8
9.5

9.8
9.5

9.8
9.5

9.9
9.2

Figure 3. Al splash amount and shear value vs. Ultrasonic energy level (BSA offset) for Device A and Device B. There are other less significant differences in the response between the two devices. Figure 4 illustrates ball diameter and ball height differences for the two devices. The thinner pad consistently has a larger diameter and height than the thicker pad, although the difference is small (0.3m in each case). There is less compliance in the thinner pad, so the ball has to absorb more force, and is therefore slightly larger. For ball height, the measurement is a top view optical measurement from the top of the ball to the top of the bond pad. So a ball height measurement is equal to the real ball height minus the pad vertical deformation. There is less vertical deformation in the thinner pad, and therefore the measured height is slightly taller. In other words, we measured a taller ball because the ball sunk less in the pad. These differences tend to have a minor impact to the wire bonding outcome. So in this example case, the same process recipe can be shared with both device types.

9.6
9.0

9.5
8.8

9.6
8.9

9.6
9.2

9.8
8.3

82% 83% 0% 0% 0% 0% 0% 0%

88% 87% 0% 0% 0% 0% 0% 0%

92% 90% 0% 0% 0% 0% 0% 0%

95% 93% 0% 0% 0% 0% 0% 0%

97% 96% 0% 0% 0% 0% 0% 0%

The two device types have similar responses to ultrasonic energy (BSA). At the low ultrasonic level, Al splash, shear, and IMC are lower than at the high ultrasonic level. Ball diameter, height and pull strength do not vary significantly with changing ultrasonic levels. For an as bonded sample, the ultrasonic process window between ball lift and pad peeling is at least 20% for this application because there is no lift and peeling occurrence in the 20% tested range. Concerning the results, pull strength is the measurement of wire break load and does not vary according to USG settings. There are no cracks for both devices throughout the ultrasonic range. This is an indication that these two devices are well designed and suited for the example Cu wire bonding. There are differences in the response between the two devices. The major differences are in Al splash and shear value. Figure 3 illustrates Al splash and shear for the two devices. As expected, the thicker Al pad has more splash than the thinner pad. The average difference is 1.5m which can be significant for fine pitch applications. The shear value is 1.8 grams higher on the thicker pad. However, the IMC % is very similar on the two devices, which indicates the level of bonding is very much the same. In other words, there is no indication that the thinner pad is bonded less than the thicker pad even though the shear value is more than 10% lower. As we discussed earlier, in Cu wire bonding, a larger Al splash contributes to the higher shear value. Because in Cu wire bonding, shear strength is influenced by pad thickness and Al splash amount, a higher shear is not always an indication of stronger bond.

Figure 4. Ball diameter and ball height vs. Ultrasonic energy level (BSA offset) for Device A and Device B. Figure 5 shows the cross section of the devices. From the cross section, we can see that the interface is fairly flat for both devices. For all 5 settings, over 80% IMC are achieved. 80% is normally considered as acceptable. Higher IMC levels are good for bond strength, however, Al splash and pad damage tend to be worse. The optimal final process parameters should be determined in consideration of bond strength, pad damage, Al remains and Al splash. In this case, the baseline setting (BSA=78%) is about the optimal value.

1092

Table III. Free Air Ball (FAB) test results.


-1.50 36.9 Ball Hgt (um) 6.5 Al Splash (um) 43.6 Shear (g) 14.3
Ball Dia (um) Free Air Ball Offset (um)

-1.00 37.2 7.6 42.9 14.0

-0.50 37.6 8.1 43.0 14.5

0.00 38.3 8.8 43.3 14.0

0.50 38.1 9.4 42.8 14.0

1.00 38.7 10.3 42.5 14.5

1.50 39.1 10.9 42.5 14.0

5% USG offset Baseline USG +5% USG Figure 5. Cross section of device A and device B at the baseline settings, and 5%. Top row is device A, bottom row is device B. Robust Cu wire bonding production Developing an optimal process parameter recipe is only one step in achieving reliable production and high yields for Cu wire bonding. Reliable Cu wire bonding production requires the process to have a sustainable process window and be portable and transferable to multiple machines and be robust across material variations commonly encountered in a production environment. It also requires material and equipment to be well controlled and calibrated. Next we will look at a few of the process variables, their source of variation and its impact to the process. The factors we examined including free air ball size variation, ultrasonic energy level variation and force level variation. In each case, the parameter range we chose to test is two times or more of the normal machine to machine variation. For example, for free air ball size, we expect a machine to machine variation range of 1% of the free air ball diameter. In this application, it is about 0.3m. In our test, we looked at free air ball range of 1.5m to account for both equipment and material variation. Device type B was used for the testing in this session of the paper as well as the TEM analysis in next session. Bonding results for free air ball diameter tests are shown in Table III. Free air ball sizes of 1.5, 1.0, 0.5m around the baseline settings are tested. To save space, only a subset of data was shown including ball dimension, splash and shear. There is no ball lift, pad peeling and crack data shown. All pull failures are wire breaks, so pull test results are not shown. For a 3m free air ball diameter increase, there is a 2.2m increase in ball diameter, a 4.4m increase in ball height, a 1.1m decrease in splash and no significant change in shear value. The most sensitive response was ball height, with 1.4 times the rate of change of FAB size change. Although shear and splash are still considered acceptable across the FAB diameter range tested here, the ball height has exceeded target tolerance. In this example, the wire bonder has good free air ball control, 0.3m in this application, as mentioned before. The major source of free air ball variation is wire diameter change from spool to spool. We recommend checking FAB diameter when changing wire spools. This may be done by measuring the ball height since it is very sensitive to FAB size change. If the ball height is outside spec after a spool change, the FAB settings may be adjusted to bring ball size in spec.

Next we examine the process sensitivity to bond force change. Bonding results for bond force adjustments from -1.5 gram to +1.5 gram are shown in Table IV. For a 3 gram bond force increase, there are about a 1 um increase in Al splash, a 1 gram increase in shear and no significant change in ball diameter and height. In this case, the process is tolerant to a 3 gram bond force variation which is at least two times the bonder to bonder variation. As long as the wire bonder is setup and calibrated according to bonder setup and calibration procedures and the hardware is sound, force variation is typically not an issue for production and process stability. Table IV. Bond force offset test results.
-1.50 38.4 Ball Hgt (um) 9.0 Al Splash (um) 42.2 Shear (g) 13.6
Ball Dia (um) Bond Force Offset (gr)

-0.75 38.0 8.7 42.8 13.7

0.00 38.2 8.9 42.3 14.0

0.75 37.9 9.1 42.9 14.0

1.50 38.0 9.1 43.3 14.4

The ultrasonic level was studied in a previous section of the paper. For ball lift and pad peeling windows, the process can typically withstand a 10% variation, 20% range in total. At higher USG levels, splash tends to be excessive, and the ultrasonic level may be too low and will cause low IMC and ball lift. This makes the ultrasonic level one of the most sensitive parameters for this process. This highlights the need to ensure the ultrasonic portability from machine to machine. With this understanding, we performed a USG window verification test across multiple machines with the ProCu process developed in previous session. Table V illustrates the results across 3 machines. The superscript denotes the machine number. The 1st machine was the engineering lab machine used to develop the process recipe. The next two machines are newly manufactured ProCu machines that have gone through the proper calibration and setup. The process responses are measured and compared. There are no ball lifts, pad peeling and pad crack after etching for all 5 ultrasonic levels and for all 3 machines. The 2nd to last column in Table V shows the average response of the 5 cells BSA settings for each machine. The last column shows the range of the 3 machines. There are some variations across the machines as expected. The largest variations are in ball diameter which is 0.6m and shear which is 0.6 gram. They are inside the tolerance of the process portability, which is 1m in ball diameter and 1 gram in shear in this example case. The two new production machines (machine 2 and machine 3) are very similar to each other. The lab machine (machine 1) can be adjusted to more closely match the production machines. This can be done by reducing the ultrasonic current factor of machine 1 by 5 to 10% in the calibration menu.

1093

Table V. Machine to machine portability test results.


BSA Offset Ball Dia (um) Ball Dia (um) Ball Dia (um) Ball Hgt (um) Ball Hgt (um) Ball Hgt (um) Shear (g) Shear (g) Shear3 (g)
2 1 3 2 1 3 2 1

-10%
38.6 39.2 38.6 10.0 10.1 9.8 14.1 13.0 13.5

-5%
38.5 39.0 38.5 10.0 10.2 9.8 14.5 13.8 13.8

0
38.5 39.0 38.3 9.9 10.2 9.8 15.0 14.2 14.6

5%
38.5 38.6 38.3 9.9 10.2 9.8 15.3 14.8 15.0

10%
38.3 38.9 38.1 9.9 10.2 9.8 15.2 15.3 15.1

Machine Machine Average Range


38.5 38.9 38.4 10.0 10.2 9.8 14.8 14.2 14.4

0.6

0.4

0.6

TEM analysis results Our previous TEM study [10, 11] shows that existence of a native oxide layer on original Al pads has a significant effect on bonding, and the initial formation of IMCs during the bonding process must overcome this relatively inert thin film. For these Al surface areas where the aluminum oxide is not broken, the IMC will not form. The degree of breakdown of the aluminum oxide layer is influenced by bonding parameters, wire type and Al pad itself. In this study, we also find two distinct regions at contact area: (1) areas containing IMC, and (2) non-IMC areas where an oxide layer in direct contact with Cu (Fig. 6). Fig. 6b gives a bright field (BF) TEM image taken in region A of Fig. 6a, which presents a ~5 nm thick uniform aluminum oxide layer between the copper ball and the aluminum pad. A scanning (S)TEM-EDX line scan crossing such layer (Fig. 7a) shows the oxygen peak at bond interface, consistent with existence of an aluminum oxide layer, which remains intact after bonding. The STEMEDX line scan crossing IMC (Fig. 7b) shows distinct element distribution, without oxygen peak at bond interface. Fast Fourier Transformation (FFT) analysis of the interference lattices of the IMC was consistent with CuAl2 (I4/mcm, a = 6.067 and c = 4.877 ) aligned along [210] orientation (Fig. 6d).

Fig. 6 BF TEM images of Cu-Al interfaces in the as-bonded state with baseline bonding parameters: (a) bond interface consisting of IMC and non-IMC areas; (b) details of region A in (a) showing aluminum oxide layer between copper and aluminum; (c) details of region B in (a) showing an IMC particle; (d) lattice image and Fourier reconstructed pattern of region C in (c) with CuAl2 [210].

Fig. 7 STEM-EDX line scanning crossing bonder interface in the as-bonded state: (a, a) Cu/aluminum oxide/Al; (b, b) Cu/IMC/Al. When isothermally baked for 24 hours at 175C, a continuous and non-uniform IMC layer forms (Fig. 8a). Highermagnification BF TEM (Figs. 8b) confirms that the IMCs consist of two layers. Those two layers of IMC are even distinguishable under STEM mode (Fig. 9a). STEM-EDX line

1094

scan (Fig. 9, line scan #1) crossing such IMC layers identified Cu9Al4 close to Cu and CuAl2 abutting the Al pad. The CuAl2 is the first IMC which is initially formed prior to baking. Cu9Al4 is the second IMC nucleated in the early stage of baking. Both IMCs grow during baking, but CuAl2 is still the dominant intermetallic phase in terms of thickness. Identification of Cu9Al4 was performed by FFT analysis of the lattices of such IMC. As given in Fig. 8c, a lattice image and Fourier reconstructed pattern of region D in (b) is with Cu9Al4. A few void-like bright areas in BF TEM image (such as region C in Fig. 8b) are present after a 24-hour baking period. Lattice image of Fig. 8d shows the details of region E in Fig. 8b, suggesting that those bright areas are thinner Cu9Al4. This is consistent with EDX line scan result crossing thinner Cu9Al4 (Fig. 8, line scan #2). Under STEM mode, thinner Cu9Al4 areas appear dark in color. The distribution of Pd, especially at the interface, is of great interest in the electronics industry. The existing reports show inconsistent results. Some researchers pointed out that Pd is absent at bond interface. However, Qin et al. [5] detected Pdrich IMCs. In this study, STEM-EDX line scans crossing the interface of a bond show the absence of Pd at the interface (Figs. 7 and 9). Different results in terms of Pd existence are probably due to variation in EFO setting and wire type.

Fig. 9 (a) STEM images of Cu-Al interface after baking for 24 hrs at 175C; (b) details of region F in (a); (c) EDX line scan #1; (d) EDX line scan #2. As baking progressed further, both CuAl2 and Cu9Al4 layers continued to grow simultaneously until the Al pad was depleted, as seen in both BF TEM and STEM images (Fig. 10) after baking at 175C for 192 hours. The aluminum oxide debris is mostly located between CuAl2 and Cu9Al4 layers (Fig. 10d)

Fig. 8 BF TEM images of Cu-Al interfaces after baking for 24 hrs at 175C: (a) continuous IMCs; (b) two layers of IMCs; (c) lattice image and Fourier reconstructed pattern of region D in (b) with Cu9Al4 [111]; (d) details of region E in (b) showing thinner Cu9Al4 area. Fig. 10 TEM images of Cu-Al interface after baking for 192 hrs at 175C: (a) BF TEM; (b) higher-magnification BT TEM showing details of region G in (a); (c) STEM; (d) details of region H in (c) confirming two layer of IMCs Cu9Al4 and CuAl2.

1095

The thickness of layers of CuAl2, Cu9Al4 and overall IMCs (CuAl2 + Cu9Al4), as a function of baking time at 175C are given in Fig. 11. CuAl2 is the only IMC in the as-bonded state, with a thickness of ~20 nm. Cu9Al4 nucleates at the early stage of baking. CuAl2 and Cu9Al4 grow simultaneously, with CuAl2 being the dominant IMC until the Al pad is depleted. It is expected that CuAl2 will be transferred into Cu9Al4 when the Al is completely consumed and baking continues.

magazines, heat block and clamps, assembling the parts, and conducting the wire bonding and reliability tests. The authors also want to thank Jim Suttie for the cross section, Virginia Mota and Jeremy Toner for metrology support, Dr. Gordon Tang for developing the IMC and splash measurement equipment and software. References 1. Chylak, R. et al. Copper High Volume Manufacturing R&D to Production, 13th Electronic Packaging Technology Conference, (2011). 2. T. Uno, S. Terashima, T. Yamada, SurfaceEnhanced Copper Bonding Wire for LSI, Elect. Comp and Tech, Conf. (2009) pp. 1486-95. 3. H. Abe, Reliability Elvaluation of Copper Wire Package Semicon Japan (Dec 2010, Tokyo). 4. Carson, F. et al. Die to Die Copper Wire Bonding Enabling Low Cost 3D Packaging, 61st Electronic Components and Technology Conference, (2011) pp. 1502-1507. 5. Qin, I. et al. Wire Bonding of Cu and Pd Coated Cu Wire: Bondability, Reliability, and IMC Formation, 61st Electronic Components and Technology Conference, (2011). pp. 1489-1495. 6. J. Yeung, D. Stefan and F. Wulff, Comparison of Bare Copper and Palladium Coated Copper Wires in Wire Bonding, IMAPS Workshop at Semicon West (San Francisco, July 2010). 7. H. Clauberg, B. Chylak, N. Wong, J. Yeung and E. Milke, Wire Bonding with Pd-Coated Copper Wire, IEEE CPMT Symposium, Aug. 2010. 8. Rezvani, A. et al. Free-air Ball Formation and Deformability with Pd Coated Cu Wire, 61st Electronic Components and Technology Conference, (2011). pp. 1516-1522 9. D. Stefan, F. Wulff, E. Milke, Reliability of Palladium Coated Copper Wire, EPTC (Singapore, 2010). pp. 343-348. 10. H. Xu, C. Liu, V. V. Silberschmidt, S. S. Pramana, T. J. White, Z. Chen A re-examination of the mechanism of thermosonic copper ball bonding on aluminium metallization pads, Scripta Materialia, 2009, Vol. 61 (2), pp. 165-168. 11. H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, M. Sivakumar , V. L. Acoff. A micromechanism study of thermosonic gold wire bonding on aluminium pad. Journal of Applied Physics. 2010, Vol 108 (11), Art. No. 113517. 12. H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, V. L. Acoff. Behavior of intermetallics, aluminum oxide and voids in Cu-Al wire bonds. Acta Materialia. 2011, Vol. 59(14), pp. 5661-5673.

Fig. 11 CuAl2, Cu9Al4 and overall IMC thinkness at different baking times at 175C. Conclusions There are a few critical factors in Cu wire bonding including chip design, process optimization and equipment and material control to ensure production stability and device reliability. This paper shows the example optimization steps to derive good bonding parameters using a new model based approach. The results illustrate a good process response for a wide ultrasonic current window for two devices with different Al pad thickness. The thicker Al pad has more Al splash and a higher shear value; however, the IMC is similar to the thinner pad. The paper shows how the process varies as a few key input variables change. As the free air ball diameter changes, the ball height correspondingly changes. A potential source of free air ball variation is wire diameter variation due to spool to spool variation. We recommend monitoring free air ball and ball height at wire spool change and to adjust free air ball formation parameters accordingly. Study of the force variation shows the example process we developed is insensitive to a 3 gram force change. Study of ultrasonic level shows Al splash, shear and IMC all increase with ultrasonic levels and wire bonders should be calibrated to ensure process portability. TEM study shows IMC CuA2 (~20 nm think) forms at the areas where aluminum oxide are broken during bonding. Cu9Al4 appears as a second IMC between CuAl2 and copper ball at early stage of baking. CuAl2 and Cu9Al4 grow simultaneously, and the former is the dominant phase until the Al pad was depleted. Pd is absent at bond interface. Acknowledgments The authors are grateful for support from SPIL Taiwan for providing the material including substrate, capillary,

1096

Вам также может понравиться