Available online at www.sciencedirect.com ScienceDirect Procedia Engineering 7 (14 ) 134 139 MRS Singapore - ICMAT Symposia Proceedings Synthesis, Processing and Characterization III Hardness Measurement of Copper Bonding Wire Johnny Yeung a, Loke Chee Keong a a Heraeus Materials Singapore Pte Ltd, Block 002, Ang Mo Kio Ave, #04-07, Singapore 69871 Abstract In semiconductor packaging industry, thin metal bonding wires, such as gold and copper, in the diameter range of 18 to 2 microns, are commonly used to connect between IC chip and connector pins through thermosonic and ultrasonic welding (bonding) in various types of packages. Copper bonding wire posts great challenges to industry users when they need to bond it onto IC bond pad of sensitive construction underneath. Any hard impact due to the material property, i.e. hardness, or excess ultrasonic parameter setting in the bonding process can easily cause the pad to crack. It is therefore of interest to the users on how hard the copper wire is and most importantly, the subsequent solidified molten ball as a result of melting the tip of the wire in the process. A Vickers micro-indentation hardness tester with minimum load of 0. gmf () capability is used to measure the hardness of copper wire along its length and in the free air ball (FAB). To determine the suitable load to be used to measure hardness of such fine diameter and minimize variation in the measured results due to different sample preparation effects, a range of load from 1 gmf to less than 0 gmf was studied. 13 The Authors. Published Published by by Elsevier Elsevier Ltd. Ltd. Selection and/or and/or peer-review under under responsibility responsibility of the of the scientific scientific committee committee of Symposium of Symposium [Synthesis, [Advanced Processing Structural and and Functional Characterization Materials III] for ICMAT. Protection] ICMAT.. Keywords: Vickers; micro indentaion hardness; copper bonding wire; free air ball Nomenclature 4N 4 Nine (99.99%) FAB Free Air Ball HV Vickers Hardness HAD Hard As Drawn 1. Introduction In semiconductor packaging industry, the use of thin metal wires such as gold and copper to connect between IC chip and connector pins through thermo- and ultrasonic welding (bonding) are common for various types of packages [1]. These so called bonding wires are mainly gold, copper and aluminium with silver beginning to be used for low end LED devices. In late 08 to early 09 when gold price was soaring after the economic downturn, the demand to replace gold bonding wire with much lower cost material was unavoidable and copper being the natural choice due to its proven application in IC packages. Not only has the demand of copper increases over the years to replace gold, the size of the wire also sees a decrease from over 2 microns in diameter to less microns as package design requires tens and hundreds of wires to be bonded in a small landscape layout. However, copper being not malleable material compared with gold and work hardens when force is applied to it, it post a challenge to user of the wire to avoid damage to the sensitive structure of the IC bond pad when switched from gold to copper. Many questions arise on how hard the copper wire and the subsequent solidified * Corresponding author. Tel.: +6 671 78; fax: +6 671 7793 E-mail address: johnny.yeung@heraeus.com 1877-8 13 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the scientific committee of Symposium [Advanced Structural and Functional Materials for Protection] ICMAT. doi:.16/j.proeng.13.11.029
Johnny Yeung and Loke Chee Keong / Procedia Engineering 7 ( 14 ) 134 139 13 molten ball when it was required to thermosonically bonded onto the IC pad. As the dimension of the wire and the ball are in the range of less than 0 microns, conventional hardness testers like Rockwell, Brinell and even normal Vickers cannot be used due to the respective size of indenter or the load used. With the size of a bonding wire in micron range, the load required to measure the hardness has to reduce significantly in the range of less than 0 gram force and a hardness tester of specially capability to apply load in such a range is necessary. Bonding wires are made from casting into millimeter diameter sized wire followed by progressive multi-die drawing and annealing to achieve final diameter of 0 micron or below. As copper will work harden, annealing at high temperature might be needed during the course of drawing, depends on the material characteristics. However, an annealing process at the final diameter must be carried out to induce recrystallization and softening of the wire before it is being put to use. Industry users are most interested in knowing the hardness of the copper wire for ultrasonic wedge bonding and also of remelted wire that forms a ball for thermosonic ball bonding that will allow best bondability and least or no damage to the device. Hence, to determine the hardness of the wire, they have to be measured at the exact size and configuration in order to reflect the correct property. 2. Sample Preparation and Measurement Bonding wires of 99.99% (4N) purity are wound onto U-shape frame made with leadframe material and cold mounted with epoxy resin by arranging the wires facing the bottom of the mount (Fig. 1), before subjecting to fine grinding and polishing to achieve scratch-free finish. The free air ball is obtained using a special program in the wire bonder to melt the tip of the wire (Fig. 2) and then bond the other end of the wire to a leadframe die paddle to form an array of FABs (Fig. 1). The FABs are pressed down before cold mounting, again with the wire faced down for subsequent grinding and polishing. Wire cross-section FAB cross-section Figure 1. Wire and FAB sample preparation prior to grinding and polishing. Figure 2. Free air ball (FAB) formation in a bonding wire. Indenter x > x Indenter Wire Cold mount resin (relatively soft) FAB Wire Possible air gap due to resin shrinkage upon curing Figure 3. Constrain rule for hardness measurement with bonding wire which is not of constant thickness both in the wire and in FAB. Indentation must be done at the centre and in the FAB, although thicker cross-section is enabled due to the ball size of usually 2X wire diameter, care must be taken not to put the indentation too close to the edge. Hardness test of the wire and FAB was done with a Fischerscope HC Vickers microhardness tester using a minimum test load of mn (1 gram) to 0 mn for μm and 0 mn for 80 μm wire. Hardness value was determined by measuring the indentation s diagonal lengths in the SEM and uses the average length of the two diagonals to calculate the Vickers hardness value according to the following equation:- HV = F/A (0.1891F)/d 2 (1)
136 Johnny Yeung and Loke Chee Keong / Procedia Engineering 7 ( 14 ) 134 139 where F is the test load in Newton and A is contact area of indentation, d is the average diagonal length of the indentation in mm [2]. The contact area can be calculated from the geometry of the indenter and the depth of penetration to the material [3]. With the known tip angle of 136 for a Vickers indenter and the measurable diagonal length, the contact area can be calculated and the HV value can be simplified as shown in equation (1). During hardness test, there are criteria on how close adjacent indentations can be applied [4-]. Referring to Figure 3 on the proximity of adjacent indentation and to the edge of sample, the indentation depth should not be more than 1/ of the thickness of the sample and adjacent indentation must be greater than or equal to 3 time the length of the diagonals. It must also not be too close to the sample edge and must be at least 2. times the length of the diagonal [4]. In actual testing of the bonding wire, the criteria will post further constrain due to the cylindrical shape of the wire and the FAB. The thickness of the sample changes from edge to the centre due to the curvature of the wire shape. There is also complication of possible air gap between the wire and the cold mount resin upon curing and when a higher load is applied, the sample may not be stable enough for proper test. 3. Results and Discussion 3.1. Hard As Drawn 80 μm Copper Wire Hardness The hardness of a metal can be measured with different test load. To understand the effect of test load on hardness, a range of load was used to see the effect and trend. Figure 4 shows the load displacement curve and respective hardness value of an 80 μm diameter hard as drawn (HAD) 4N purity copper wire tested with a load from mn to 0 mn at the centre of a longitudinally cross-section wire. The measured Vickers hardness (HV) value decreases as the load increases. There is more variation of HV value at very low load possibly due to measurement accuracy variation of the small indentation as well as stability of the test load. The HV value decreases in a somewhat hyperbolic manner and tends to stabilize at higher loading of 10 mn and beyond. However, at this test load, the depth of indentation is more than 2 μm. With a perfectly cross-sectioned 80 μm wire at its diameter, the maximum thickness is μm. Running a normality test of the data (probability plot) for each test load, all showed P value higher than 0.0 (Table 1) indicating that the data falls into normal distribution. To select the suitable test load for this wire diameter, and consider the criteria and dimension constrains in hardness test, a test load of up to 0 mn is possible. The indentation depth at this load is about 2.6 μm. If the indentation landed at 3/4 of a way across the μm diameter, the thickness of the wire is calculated to be just over 26 μm. This is just sufficient to meet the criteria. However, in order to minimize fluctuations and possible air gap between the wire and epoxy resin in the cold mount, a lower load is preferred. Load (mn) 0 20 0 10 0 0 Load vs. Displacement - 3.1mil Hard As Drawn (HAD) Wire Variable mn mn mn mn mn mn 10mN 0mN 0mN Hardness, HV ( Sec Creep Time) 1 180 1 160 10 1 1 1 3.1mil Hard As Drawn (HAD) Wire Hardness 0.0 0. 1.0 1. 2.0 2. Displacement (μm) 3.0 3. mn mn mn mn mn mn 10mN 0mN 0mN (a) (b)
Johnny Yeung and Loke Chee Keong / Procedia Engineering 7 ( 14 ) 134 139 137 Percent Probability Plot of,,,,,,, 10, 0, 0 Normal - 9% CI Variable 99 9 80 10 0 60 0 0 Mean StDev N AD P 171.7 9.719 19 0.38 0.38 166. 6.791 18 0.283 0.93 18.6 4.160 0.3 0. 13.8 3.072 0.49 0.23 11.1 2.394 0.16 0.944 143.9 3.669 0.3 0.3 139.6 3.244 0.0 0.179 1 1.4 3.716 0.618 0.093 1 1 1 10 160 1 180 1 0 2 129.1 4.164 0.268 0.646 127.0 2.684 0.497 0.187 Data (c) (d) Figure 4. (a) Load displacement curve of different test loads used. (b) Minitab box plot of test load on Vickers hardness of a 3.18 mil (80 μm) HAD 4N Cu showing decreasing trend with increasing test load. (c) Plotting with regular scale, the hardness appears to stabilize at and beyond 10 mn. (d) Probability plot of all loads showing P values higher than 0.0 indicating data fall into normal distribution 3.2. Hard As Drawn μm Copper Wire Hardness The 80 μm HAD drawn wire was drawn further down to μm and the hardness of the wire was measured at different test load to see if similar effect is observed. Figure shows the load-displacement curve and box plot of the hardness results from mn to mn. Again the effect of lower hardness registered with increasing test load has been observed. There is a sudden drop in the hardness beyond 1 mn of which the root cause was not clear. As the dimension of the wire is small with a hemispherical cross-section along the wire, the stability of hardness registered depends very much on the surface stress as a result of grinding and polishing as well as the rigidity of the wire sitting in the cold mount epoxy where no air gap should be present. Should there be variation in the condition of the sample, the measurement accuracy might be affected. Nevertheless, the trend of hardness with test load is obvious. The indentation depth ranges from 0.3 μm to about 1.4 μm at this load range. The material thickness at 3/4 way across the diameter of the wire is calculated to be just over 13 μm and so test load of mn with the indentation depth just over 1 μm can be used. Normality check of the entire test load data showed at 1 mn, the P value (Table 1) is less than 0.0 indicated that at the data of this test load do not fall into normal distribution and hence the data cannot be trusted. Still, too high a test load would violate test rule while too low a load could lead to instability. Therefore, a test load of to 1 mn should be used for this wire diameter. HAD 0.8mil 4N Cu Wire Hardness HAD 0.8mil 4N Cu Wire Hardness 1 1 Hardness, HV ( Sec Dwell Time) 1 1 1 Average 1 1 1 1 0 (a) (b) (c) Figure. (a) Load displacement curve of HAD 0.8 mil ( μm) 4N Cu wire. (b) & (c) Minitab box plot and line plot of test load on Vickers hardness showing decreasing trend with increasing test load. The hardness appears to drop suddenly beyond 1 mn load and continue to decrease with load.
138 Johnny Yeung and Loke Chee Keong / Procedia Engineering 7 ( 14 ) 134 139 3.3. Annealed μm Copper Wire and FAB Hardness The hardness of the μm annealed copper wire also showed decreasing trend as HAD one and the hardness tend to show stabilization at 1 mn load and beyond (Fig. 6). At 1 mn, the indentation depth reached about 0.9 μm, close to the limit of test rule. However, if the indentation landed again 3/4 way to the edge of the wire, there is only about 6.6 μm thick material, and therefore a lower load of mn must be used. At mn load, the indentation depth is about 0.6 μm and this allows more flexibility in location of indentation in the wire. However, normality check showed the P-value of this test load is less than 0.0 (Table 1), again, indicating the data does not fall into normal distribution. Great care must be employed in conducting the test at this load to ensure consistency. Load vs. Displacement - 0.8mil Annealed 4N Cu Wire 0.8mil Annealed 4N Cu Wire Hardness Load (mn) 2 1 0 Variable mn 1 mn 2 mn Hardness HV ( sec Creep Time) 1 1 80 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 mn 1 mn 2 mn Displacement (μm) (a) (b) Figure 6. (a) Load displacement curve of annealed 0.8 mil ( μm) 4N Cu wire. (b) Minitab box plot of test load on Vickers hardness showing similar decreasing trend with increasing test load as HAD wire. The hardness appears to stabilize at and beyond 1 mn load. The FAB hardness can be measure by cross-sectioning the ball parallel to the wire where it was formed or perpendicular to the wire direction. Figure 7 shows images of cross-sectioned FABs, longitudinally and transversely, with indentation mark from various test loads and their respective FAB hardness. Both cases show decreasing hardness with test load as seen in wire hardness. For longitudinal cross-sectioned FAB, hardness decreases from an average of about 9HV at mn test load to just above 60HV at mn. The hardness appeared to start stabilizing from mn onwards. For transversly crosssectioned FAB, the hardness tested under the same loads differs significantly and dropped from 80HV to 0HV across the same range. There is 16.6% difference in hardness obtained at mn test load between logitudinal and transverse crosssection The lowest difference is about 8% at mn test load. Viewing from the cross-sectioned image, the grain morphology between the two cross-sections are different with large, columnar grains appeared at longitudinal direction. The grains seen in transvers direction is just a slice of the large columnar grain seen in perpendicualr direction. The reason why a lower hardness was registered with transverse cross-sectioend FAB is the fact that the indentation will mostly landed onto the bulk of a large single grain and reflects the characteristic of a single crystal s hardness. In the longitudinal crosssection, the indentation could land onto a grain boundary not far below the surface. The presence of grain boundary, which has high dislocation density will respond with higher hardness. P values for both test directions are shown in Table 1 and indicate all data are in normal distribution. mn 1 mn mn mn FAB Hardness Comparison between Longitudinal and Transverse 1 FAB Hardness HV 80 60 0 -L _T 1_L 1_T _L _T _L _T _L _T Figure 7. (a) (b) (a) Image of hardness indentation from different test load on longitudinal and transversely cross-section FAB. (b) Minitab box plot of test load on Vickers hardness of the FAB showing decreasing trend with load and higher HV for longitudinally cross-sectioned balls.
Johnny Yeung and Loke Chee Keong / Procedia Engineering 7 ( 14 ) 134 139 139 Table 1. Normal probability plot s P-value of various test loads used in copper wire and FAB hardness test Test Load (mn) 1 2 10 0 0 80µm HAD 0.38 0.93 N/A 0. N/A 0.23 0.944 0.3 0.179 0.093 0.646 0.187 µm HAD 0.29 0.806 0.044 0.361 N/A 0.323 0.128 N/A N/A N/A N/A N/A µm Annealed 0.4 0.017 0.86 0.4 0.88 N/A N/A N/A N/A N/A N/A N/A FAB (longitudinal) N/A 0.1 0.6 0.839 N/A 0.31 0.296 N/A N/A N/A N/A N/A FAB (transverse) N/A 0.491 0.827 0.798 N/A 0.81 0.662 N/A N/A N/A N/A N/A 4. Conclusions Measuring Vickers hardness of copper bonding wire is a challenge due to the physical size of the sample and the preparation method applied which would lead to variations. Only very low test load of 1 mn or less can be applied to the μm diameter sample. Even the larger diameter of HAD wire at 80 μm, the load used should be less than 0 mn, consider the test criteria and constrains according to standard. It was found that in all tests conducted, the hardness values decrease with increasing test load. The rate of decrease in hardness will slow down significantly when higher loads are applied. For the μm wire, much lower test load of 1 mn or mn must be used to achieve good statistical results. The hardness of FAB depends on direction of measurement. Hardness at transverse cross-section has hardness value of 8% to 16.6% lower than in longitudinal direction. This is likely due to indentation on large grain than on shallower grain with adjacent grain s boundary beneath it. Therefore, for μm wire, test load still needs to be kept at 1 mn or mn. Since the hardness of wire and FAB is dependent on test load, the HV value convention of stating test load and dwell time must be use when quoting hardness value. Acknowledgements The authors would like to thank our former industrial attachment student from School of Materials Science and Engineering, NTU, Jessica Oh Si Jia, for the support of sample preparation and conduct testing and measurement of the wires and FABs. References [1] G.G. Harman, Wire Bonding in Microelectronics: Materials, Processes, Reliability, and Yield, McGraw Hill, NY, 1997 [2] Instrumented Indentation, N. P. Laboratory May, 07 [3] D.I.h. G. Michalzik, "Determination of the Hardness and Other Characteristic Materials Parameters Using the Instrumented Indentation Test Using Hardness Testing Equipment from Fischer Instrumentation," Helmut-Fischer instrumented indentation test, pp. 1-1, 04. [4] ISO 607 1: 0 Metallic Materials Vickers hardness Test Part 1: Test Method [] ASTM E384: Standard Test Method for Microindentation hardness of Materials