2017 IEEE 67th Electronic Components and Technology Conference Dynamic Strain of Ultrasonic Cu and Au Ball Bonding Measured In-Situ by Using Silicon Piezoresistive Sensor Keiichiro Iwanabe, Kenichi Nakadozono, Mamoru Sakamoto, and Tanemasa Asano Graduate School of Information Science and Electrical Engineering Kyushu University 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan asano@ed.kyushu-u.ac.jp Abstract Dynamic changes in distribution of mechanical strain generated during wire bonding in Si under and near the bonding pad were measured by using a piezoresistive linear array sensor. The sensor was designed to be able to determine strains in the directions normal and parallel to the surface. Bonding dynamics of Cu and Au balls were investigated. We can clearly observe the oscillating strain according to the application of 150 khz ultrasonic vibration. It was also clearly observed that the position of the largest compressive strain moved from the center of the ball to the periphery according to the progress of bonding under the application of the ultrasonic vibration. Bonding of Cu was found to generate larger strain than bonding of Au. A large oscillating tensile strain generated at the periphery of Cu ball when ultrasonic amplitude is increased is found to cause fracture of Si. The largest residual strain is observed for Cu bonding at the location where the end of capillary tool was present during bonding. Keywords-wire bonding; ball bonding; ultrasonic bonding; Cu; strain; under pad damage; piezoresistance gauge I. INTRODUCTION Interest in advancement of the ultrasonic wire-bonding technology increases more and more in microelectronics packaging because it offers significant benefits to many kinds production in small quantities. Understanding of mechanical strain generated during bonding and remained after the bonding in Si under and near the bonding pad during wire bonding and remained after the bonding is a significant issue not only to produce reliable interconnection but also to develop new technology such as circuit under pad (CUP). Cu wire is being increasingly seen as a candidate to replace Au wire for fine-pitch interconnections of microelectronics packaging [1]. However, since Cu is harder than Au, high compression force and large ultrasonic energy are usually required to complete bonding. The high compression force induces under-pad damage such as pad peeling and bulk silicon cratering [2]. The large ultrasonic vibration increases under-pad splash or squeezing from the peripheries of the bonded ball [3, 4]. In-situ measurement of strain generated in the Si circuit layer during bonding will provide useful information to understand the bonding dynamics and to find a way to avoid under-pad damage. We have developed a piezoresistive liner array sensor made of Si to measure dynamic strain generated under the bonding pad during ultrasonic application. The sensor has been designed and fabricated in-house. It is able to determine the strains along the vertical and horizontal axes with respect to the Si surface. It is also able to measure insitu temperature rise during bonding. We have applied the sensor to investigation of bonding dynamics of flip-chip microbumps [5, 6] and wire bonding [7]. In this work, we investigate time evolution of dimension and distribution of strain generated in the Si layer during ultrasonic bonding of Cu and Au wires. Bonding was performed using the program composed of three steps; compression only, compression + ultrasonic at moderate amplitude, and compression + ultrasonic with an increased amplitude. We investigated how dimension and distribution of strain changes with the progress of bonding and what the critical bonding parameters are in generation of damage in the Si circuit layer. II. STRAIN SENSOR AND EXPERIMENTAL Fig. 1 shows schematic illustration of the in-situ measurement of strain and its distribution. The silicon sensor composed of an array of piezoresistance gauges was designed and fabricated in-house. The gauge array was covered with an interlayer dielectrics and a bonding pad Figure 1. Schematic illustration of measurement method and relative position between the wire ball and Si thin-film strain gauges. 2377-5726/17 $31.00 2017 IEEE DOI 10.1109/ECTC.2017.316 1786
made of Al. The interlayer dielectric was SiO 2 deposited by plasma enhanced chemical vapor deposition. A wire ball was whose diameter was 40 mm was bonded to the Al pad and change in resistance of the gauge array was measured during bonding. A single gauge was composed of a pair of n- and p- type Si thin films resisters to simultaneously measure the strain parallel to the surface along the ultrasonic vibration (horizontal direction, hereafter) and the strain perpendicular to the surface (vertical direction, hereafter). Principle and details of the strain measurement is described elsewhere [5, 8]. Fig. 2 shows a die photo (Fig. 2(a)), an enlarged view of the strain gauge array (Fig. 2(b)), and a schematic crosssection of a gauge (Fig. 2(c)). The sensor was fabricated on a SOI wafer to minimize isolation area. The size of each gauge is 10 m in pitch and 10 m in width. The layout was designed to maximize spatial resolution under the constraints of the resolution of in-house photolithography. These sensors were fabricated on silicon-on-insulator (SOI) wafer. The use of SOI provides perfect electrical isolation between the gauges and the substrate. The thickness of the Si layer of SOI was 200 nm. The sensor is covered with a 400 nm-thick SiO 2 layer and a 900 nm-thick Al pad. The gauge factors were determined from measurements of change in resistance with uniaxial strain applied by a cantilever bending tool [5, 8]. Fig. 3 shows electrical connection to measure change in resistance during wire bonding. A constant current was supplied to the gauges connected in series while voltage drop across each gauge was measured during bonding with using voltage meters with analog-to-digital converter having 12 bit resolution at 20 MHz sampling rate. Additional voltage source was applied between the low voltage end of the n- type gauge array and the ground to reversely bias the pnjunction of the gauge pairs. The in-situ measurements were carried out for the bonding programs shown in Fig. 4. Bonding tests being Figure 3. Electrical connection of the gauge array to measure the dynamic change in strain distribution. A dc voltage source was used to reverse bias the pn junctiion composed of the the n-and p-type gauges. Figure 2. (a) Die photo of the sensor fabricated. (b) Photo showing the array of piezoresitance gauges. (c) Schematic cross-section of the sensor composed of p- and n-type Si made of silicon-on-insulator. Figure 4. Load force and ultrasonic programs used in this study. (a) Loar force used for all bonding tests. (b) (d) Three ultrasonic programs. The ultrasonic vibration was applied while the load force was kept constant. 1787
described in this article were all carried out at room temperature. Fig. 4(a) shows compression force applied for bonding. The compression force was kept constant during the bonding. Figures from Fig. 4(b) to Fig. 4(d) shows the current applied to drive the piezoelectric element to produce ultrasonic vibration. The ultrasonic frequency was 150 khz. Application of ultrasonic vibration was started while the compression force is kept constant and the amplitude of the ultrasonic vibration was increased during the bonding action. Thus, the bonding program is composed of three steps; the first step where only the compression force is applied (step 1), the second step where the compression force and ultrasonic with a small amplitude is applied (step 2), and the third step where the compression force and ultrasonic with a large amplitude is applied (step 3). The duration each step was set at 10 ms. In Figs. 4(b) 4(c), the transducer current of 0.1 A corresponds to vibration amplitude of 1 m. In the ultrasonic program shown in Fig. 4(b), the ultrasonic with the amplitude of 1 m was firstly applied at the step 2 and it was increased to 2 m at the step 3. In the ultrasonic program shown in Fig. 4(c), the amplitude was increased from 1 m to 2 m but a large amplitude was applied only at the initial stage of step 3. In the bonding program shown in Fig, 4(d), the first and second amplitude was set at 1.6 m and 3.2 m, respectively. The ultrasonic programs shown in Figs. 4(b), 4(c) and 4(d) are designated hereafter as US condition 1, 2 and 3, respectively. The sensitivity of the Si strain gauges designed and fabricated is demonstrated by the test example shown in Fig. 5. The results were obtained for bonding of the Cu wire ball during the application of the compression force of 50 gf and ultrasonic with the amplitude of 2 m. Fig. 5(a) shows strains in the horizontal and vertical directions. The negative sign indicates compressive strain. Fig. 5(b) shows change in current to drive ultrasonic vibration. We find that the sensor is well detect the strains oscillating according to the ultrasonic drive at 150 khz. It is noteworthy that the strain signals contains noise. The signals shown in Fig. 5 was measured using the gauge at positon III which is connected to the ground. However, a largest noise was observed at the gauge connected to the ground. The electrical ground is supposed to be the noise route in the bonding machine used in this study. III. RESULTS A. Bonding Results Fig. 6 shows bonding results of Au observed with an optical microscope. Figs. 6(a) 6(c) show the results obtained from bonding under the US conditions 1-3, respectively. Fig. 6(a) indicates that bonding under the US condition 1 results in bonding failure as we find from the absence of the ball and/or wire on the bonding pad. Note that the bonding test was carried out without elevating the substrate temperature. Bonding under the US condition 2 results in bonding of the ball. Bonding under the US condition 3, on the other hand, the ball disappeared and a trace of the capillary end is clearly observed. This indicate that the US strength of the condition 3 is so strong for bonding of Au ball that the ball was punched with the capillary. Fig. 7 shows results observed for bonding of Cu wire ball. Figure 6. Oprical micrograph showing bonding results of Au wire ball. (a), (b), and (c) are the results obtained by applying the US conditions 1, 2, 3, respectively. Figure 5. (a) Example of horizontal (red line) and vertical (black line) measured using the strain sensor during bondign under application of 150 khz ultrasonic. (b) Current to drive the ultrasonic transducer. Note that signals in (a) were measured at 20 MHz sampling, while the signal in (b) was measured at 1MHz sampling due to constraint of the equipment. Figure 7. Oprical micrograph showing bonding results of Au wire ball. (a), (b), and (c) are indicates the same as above Fig. 6. 1788
Figure 8. Scanning electron micrograph showing cross section of the bonded Au wire ball (a) and Cu wire ball (b) under the US condition 2 and 3, respectively. Figs. 7(a)-7(c) show the results obtained from bonding under the US conditions 1-3, respectively. Similarly to the above mentioned results, bonding under the US condition 1 resulted in failure. Bonding under the US condition 2 completed the bonding. Bonding of Cu under the condition 3 also resulted in bonding, although the Al splash took place as is observed in scanning electron micrograph shown in Fig. 8(b). B. Strain Dynamics Figs. 9 and 10 show time evolution of strain measured during bonding under the US condition 1 3 at the positions I III for Au wire ball and Cu wire ball, respectively. The positive and negative sings in strain indicate tensile and compressive strains, respectively. As was mentioned above, noise appears to be larger in the signals obtained at the position III (i.e., periphery) than in the signals obtained at the position I (i.e., center). However, the signals give well Figure 9. Time evolution of horizontal (red) and vertical (black) strain generated duding bonding of Au wire ball. (a) (c) are the results obtained by gauges at the positons I III from bondings under US conditions 1 3, respectively, as is illustrated in the inssets. 1789
change in strain from the landing of the ball to the release of the load and the cut of the wire. Results of bonding of Au shown in Fig. 9 reveal the followings: 1. Strains generated during bonding are mostly compressive. Tensile strain occasionally appears only in the horizontal strain in a time period during application of ultrasonic. 2. The compressive strain is much higher in the vertical direction than in the horizontal direction. 3. At the position I (center of the ball), the average strain in the vertical direction decreases with the progress of the bonding step from step 1 to step 3. On the contrary, at the position III (periphery of the ball), the average strain increases as bonding step proceeds. In other words, as shown in Fig. 11 where evolution of the average strain with the process step is plotted, the position of the maximum strain in the vertical direction moves from the middle to the periphery of the ball. This phenomenon can be explained by considering spreading deformation of the ball in accordance with the progress of bonding. 4. The strain signal indicates that the deformation in each ultrasonic steps (step 2 and step 3) is almost completed within approximately 5 ms at the initial stage of each step. A similar deformation rate was observed with a high speed camera for Au stud bump of flip-chip bonding. [9] However, the deformation rate observed in this study (approximately 5 ms) is higher than that observed in the previous study (approximately 20 ms). This difference can be accounted for by taking the difference in ultrasonic frequency into account, that is, 150 khz in this Figure 10. Time evolution of horizontal (red) and vertical (black) strain generated duding bonding of Cu wire ball. (a) (c) are the results obtained by gauges at the positons I III from bondings under US conditions 1 3, respectively. 1790
Figure 11. Change with process step of the average value of vertical strain during bonding of Cu wire ball. study and 48.5 khz in the previous study. [9] 5. The amplitude of the oscillating strain under the application of ultrasonic is larger at the periphery than at the center of the ball. Si in the center area is clamped by the ball compressed by the ball and, therefore, strain generated by the oscillating movement of the capillary is relatively small. Si in the peripheral area, on the other hand, becomes sensitive to the movement of capillary because the capillary end is placed near the periphery of the ball. Most of the above mentioned characteristics are common for both Au bonding and Cu bonding. Comparing the dynamic strain of Au bonding (Fig. 9) with that of Cu bonding (Fig. 10), the difference between these two wire materials is seen in the dimension of strain at each step. This is true for not only in the average strain but also for the amplitude of oscillating strain. For example, the compressive strain as large as 0.4 0.5% is generated at landing of Cu ball at the position II, above which the inner wall of the capillary end is present, while it is approximately 0.3 0.4% at landing of Au ball. The difference can be simply explained by the differences in stiffness and yield strength of Au and Cu. C. Residual Strain A significant difference between Au bonding and Cu bonding appears in residual strain. The residual strain can straightforwardly be measured by the signals after the bonding event, for example, strain values at 50 ms in Figs. 9 and 10. The fact that the time evolution data of strain for the bonding under the US condition 1 shown in Figs. 9(a) and 10(a) indicate zero strain after the bonding event at all positions is reasonable because bonding was failed under the condition and ball is absent from the surface as shown in Figs. 6(a) and 7(a). The difference between Au and Cu can be found from Fig. 9(b) and Fig. 10(b) where time evolution data for the US condition 2 are plotted. The following characteristics in residual strain distribution are commonly seen for both Au and Cu: At the position I (center), horizontal and vertical strains are tensile and compressive, respectively. At the position III (periphery), both horizontal strain and vertical strain are compressive. The maximum residual strain appears at the position II, above which the inner wall of the capillary was present, and is compressive 0.2%. This residual strain may significantly change the current drive of a MOSFET. [10] D. Time and Place of Fracture When the thickness of Al pad was reduced from 900 nm to 200 nm and Cu wire ball was bonded to the thinned Al pad, breakdown of the gauge was often observed. In order to find the location of the breakdown, output of each gauge of the gauge array was individually monitored. Bonding was carried out using the US condition 2. It has been found that breakdown takes place more likely at the position II, i.e., near the periphery of the ball. [7] The time at which breakdown takes place was always the beginning of the step 3, i.e., the transition to the largest ultrasonic amplitude. Note that, in Fig. 9(c) for Au bonding, the breakdown of the sensor is found to take place in the middle of step 3. However, this breakdown was induce by the punch through of the capillary end to the surface of the Al pad. Thus the mechanism of the breakdown is completely different from the strain induced fracture. When we look at the time evolution in strain shown in Fig. 10(b), Position III, a large oscillating tensile strain along the horizontal direction is generated at the beginning of step 3. A similar peak in tensile strain is also found in Fig. 10(c), Position III, showing the result obtained from the bonding carried out using the US condition 3 where no overshoot was applied at the beginning of the application or the amplitude change. The large tensile strain may induce fracture of Si and generate under pad damage such as crack. IV. CONCLUSION Strain generated in Si under bonding pad during wire ball bonding was in-situ measured using our originally designed and fabricated Si sensor. The Si sensor is able to determine the strain in the horizontal direction and the vertical direction. It is also able to sense the variation of strain with time according to the application of 150 khz ultrasonic. Changes in dynamic strain with position, ultrasonic condition, and 1791
wire ball material (Au and Cu) were extensively studied. The followings summarize and conclude this work: The position of the maximum compressive strain moves from the center of the ball to the periphery of the ball with spread out of the ball. The above phenomenon is common for both Cu and Au. However, Cu gives larger strain than Au due to the difference in stiffness and yield strength. A large tensile strain along the horizontal direction is generated at the beginning of application of ultrasonic with a large amplitude. The location where the large tensile strain is generated is periphery of the ball. The large tensile strain mentioned above may induce fracture of Si during bonding of Cu wire ball, particularly, in case that the bonding pad is made thin. Residual strain under the bonding pad is larger for Cu wire ball than Au wire ball. The largest strain is compressive and appears at the location above which the end of capillary was present. To more accurately say it is the location above which the inner wall of the capillary was present. ACKNOWLEDGMENT The authors are grateful to Shinkawa Ltd., Tokyo, Japan for the use of the ultrasonic bonding machine and useful discussion. This work is supported in part by the Matching Planner Program (No. MP28116808463) of JST. REFERENCES [1] A. Shah, A. Rezvani, M. Mayer, Y. Zhou, J. Persic and J.T. Moon Reduction of ultrasonic pad stress and aluminum splash in copper ball bonding, Microelectronics Reliability, Vol. 51 (2011), pp. 67-74. [2] B. K. Appelt, A. Tseng, C. Chen and Y. Lai Fine pitch copper wire bonding in high volume production, Microelectronics Reliability, Vol. 51 (2011), pp 13-20. [3] K. Toyozawa, K. Fujita, S. Minamide and T. Maeda Development of copper wire bonding application technology, IEEE Transactions on Components, Hybrids, and Manufacturing Technology, Vol. 13, (1990), pp. 667-672. [4] N. Srikanth, S. Murali, Y. M. Womg and Charles J. Vath III Critical study of thermosonic copper ball bonding, Thin Solid Films, Vol. 462-463, (2004), pp. 399-345. [5] K. Iwanabe, K. Nakadozono, Y. Senda, and T. Asano, "Bonding dynamics of compliant microbump during ultrasonic bonding investigated by using Sistrain gauge", Jpn. J. Appl. Phys. 55, 06GP22 (2016). [6] K. Nakadozono, K. Iwanabe, Y. Senda, and T. Asano, "Sensing Local Dynamic Strain and Temperature Evolution during Ultrasonic Bonding of Microbumps", Proc. IEEE Electron. System-Integration Tech. Conf., 2016, p. 137. [7] K. Iwanabe, K. Nakadozono, Y. Senda, and T. Asano, "In-situ Strain Measurement of Ultrasonic Ball Bonding", Proc. IEEE Electron. System-Integration Tech. Conf., 2016, p. 110. [8] N. Watanabe and T. Asano, "Behavior of Plated Microbumps during Ultrasonic Flip-Chip Bonding Determined from Dynamic Strain Measurement", Jpn. J. Appl. Phys.42, 2193 (2003). [9] T. Shuto and T. Asano, In-situ observation of ultrasonic flip-chip bonding using high-speed camera, Jpn. J. Appl. Phys. 54, 030204 (2015). [10] N. Watanabe, T. Kojima, Y. Maeda, M. Nishisaka, and T. Asano, "Breakdown Voltage in Uniaxially Strained n-channel SOI MOSFET", Jpn. J. Appl. Phys.43, 2134 (2004). 1792