Fine Pitch Probing and Wirebonding and Reliability of Aluminum Capped Copper Bond Pads

Transcription

1 Intl. Journal of Microcircuits and Electronic Packaging Fine Pitch Probing and Wirebonding and Reliability of Aluminum Capped Copper Bond Pads Tu Anh Tran*, Lois Yong*, Bill Williams**, Scott Chen***, and Audi Chen*** * Motorola Semiconductor Products Sector 31 Ed Bluestein Boulevard Austin, Texas Tu.Anh.Tran@motorola.com, Lois.Yong@motorola.com ** Motorola Semiconductor Products Sector 13 North Alma School Road Chandler, Arkansas 82 RXFP6@ .sps.mot.com ** *Advanced Semiconductor Engineering (ASE), Chung-Hwa Road Sec. 1, Chung-Li Taiwan, R.O.C. s: Scott.Chen@asecl.com.tw, Audi.Chen@asecl.com.tw Abstract The requirement for improved electrical performance and reduced silicon area has driven Copper to replace Aluminum interconnection as silicon technology is scaled beyond.2µm. The front-end change, in turn, pushes wirebond pad pitch from above 1µm to 8µm-66µm range. This creates challenges for back-end to probe and wirebond at fine pitch geometry onto a readily oxidized Copper surface. After several re-metallization structures and types of metallurgy were evaluated, capping Copper bond pads with Aluminum was selected as the primary approach for probing and wirebonding Copper devices. Aluminum re-metallization structure offers many advantages that help leverage existing tooling and knowledge in fab, probing, and wirebonding processes. This paper will describe probe and wirebond experiments used to select the proper adhesion and diffusion barrier between Copper and Aluminum, and Aluminum thickness that can withstand the mechanical stress during probing and wirebonding. Probe mark depth and the impact of probe marks to the underlying barrier and Copper pad were examined. Ball shear, wire rip, and corresponding failure modes, intermetallic coverage and cratering analysis were evaluated at various readpoints of thermal aging study to evaluate the integrity of the re-metallization structure as well as the quality of ball bonds onto the new structure. Contact resistance measurement and reliability assessment were also performed. One re-metallization structure was recommended for Copper High Performance wirebonded devices. Key words: 1. Introduction Copper Interconnection, Aluminum Capped Copper bond Pads, Adhesion/Diffusion, Probe Process, Wirebond Process, and Package-level Reliability. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 332 The requirements for improved electrical performance, higher speed, and increased packing density have driven Copper to replace Aluminum interconnection. The challenges for back-end probe and assembly- in processing Copper (Cu) wafers are as follows: (1) the back-end has to accommodate much finer pitch geometry in the range from 8µm to 66µm pad pitches; (2) Cu is readily oxidized in air at low temperature (<2 o C) and forms no

2 Fine Pitch Probing and Wirebonding and Reliability of Aluminum Capped Copper Bond Pads self-passivating layer to stop further oxidation and erosion; and (3) in some wireless applications, the last Cu layer can be as thin as.42µm. There are three main approaches to treating Cu bond pad surface to provide a reliable, probe-able and wirebondable surface: (1) replacing the last Cu layer with an Aluminum (Al) layer; (2) re-metallizing Copper pads with noble metal (for example Gold and Palladium) or self-passivated metal (such as Aluminum); and (3) no additional treatment so that probing and wirebonding directly onto Copper pads must be developed. Capping Cu bond pads with Aluminum was selected as the primary approach for Cu wirebonded devices 1. This approach helps leverage well- established Al probe and assembly flows, and wirebonders using Gold wire. Since Cu diffuses quickly into Al forming very brittle intermetallics that would degrade the device reliability, Cu-Al structure requires a complete isolation from each other by a thin film layer that functions as an adhesion promoter and a diffusion barrier for both Cu and Al. Ta, TaN, Ti/TiN and Tungsten are various barrier materials available for Cu interconnect processing due to their excellent barrier and adhesion properties 2. The process for capping Cu pads with Al begins with a Cu oxide removal process. An adhesion/diffusion barrier is deposited across the wafer, then Al is subsequently deposited. A photoresist layer is applied, exposed, and developed. The barrier and Al layers are etched out, leaving an Al cap slightly larger than the bond pad. The construction of Al capped Cu pads is illustrated in Figure 1. Passivation Dielectric Al Cu Pad 2. Probe Experiments to Evaluate Aluminum Cap Thickness The purpose of the probe experiment was two-fold; (1) to establish the probe process window for Al capped Cu bond pads, and (2) to recommend an Al cap thickness. The Al cap thickness must be sufficient such as to withstand worst case probe conditions without disturbing the underlying adhesion/diffusion barrier. Due to brittle Cu-Al intermetallics and related long-term degradation of the metal stack, the barrier film must be kept intact to ensure separation between Cu and Al layers. Three probe variables were considered, probe tip diameter, probe overdrive, and number of probe touchdowns. Two tip diameters were evaluated, 1.2 mil as the standard production size and 1. mil in anticipation of finer pitch geometry. Three settings of needle overdrive were selected, 4µm, 6µm, and 8µm. The center value, 6µm, is the nominal overdrive used in most production probe. Four settings of number of touchdown were 2 touches (low), 4 touches (normal), 8 touches (high), and 16 touches (expected to force fails). Probe marks were visually inspected and profiled using a profilometer. The effect of probe mark on the underlying barrier was examined using Scanning Electron Microscopy (SEM) images after stripping off the Al cap using the well-established NaOH etching technique. Two methods were employed to detect Cu exposure: EDX to analyze the deepest area of the probe mark for Cu pad exposure, and 2:1 H 2 O 2 :HNO 3 etchant for minutes to detect potential corrosion/migration path between the barrier and passivation. One Al cap thickness would be recommended. Wafers with the selected Al cap thickness would be probed, and sent for assembly evaluation. Figure 1. Structure of Aluminum capped Copper bond pad. 3. Assembly Experiments to Evaluate The test vehicle was a 6.7mm x 6.7mm daisy chain device with one.42µm thick Cu metal layer. Bond pads were designed at 76µm wire pitch with 72µm x 96µm passivation opening. Wafer splits were processed with two adhesion/diffusion barrier metals (barriers A and B) and three Al cap thickness (thin, thick, and The scope of the assembly experiment was five-fold: (1) to thickest). Thin Al thickness had revealed stained or corroded establish a wirebond process window for Al capped Cu pads, (2) Cu lines after probing and acid dipping test. The non-hermetic to assess the quality of ball bonds on the Al capped Cu structure, cap caused a concern in latent reliability degradation. The thickest (3) to evaluate the integrity of the adhesion/diffusion barrier, (4) Al cap was preferred for back-end processing. However, this to assess the reliability of packaged parts, and () to ascertain required the longest metal etching time. whether one barrier is superior to another. This paper will describe probe and wirebond experiments designed The wafers were processed in a standard production flow, to evaluate the integrity of the adhesion/diffusion barri- which includes a plasma clean process prior to wirebonding. Dice ers, and the Al cap thickness that can withstand the mechanical were wirebonded with 1 mil wire (Au > 99.99%) on a machine stress during probing and wirebonding, while minimizing Al with a 6kHz transducer. Power and force settings from the standard etching duration and maintaining the lowest system cost solution. Al interconnect process were used as a baseline to establish One Cu re-metallization structure, one adhesion/diffusion the wire bond process window for Al capped Cu pads. The low, barrier and one Al cap thickness, will be recommended for Cu normal, and high levels of power were-, 7, and 9mW, and interconnect wirebonded devices. three levels of force were- 2, 4, and g. The time parameter was held constant at 1ms. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 333

3 Intl. Journal of Microcircuits and Electronic Packaging Wirebond responses were non-stick at first bond, percentage of pad lift/ball lift during wire rip test, Au-Al intermetallic coverage, and pad cratering analysis. Plastic Ball Grid Array (PBGA) substrate strips wirebonded at normal and high wirebond settings were then submitted for barrier evaluations. In order to determine the quality of Au bonds and the integrity of the barrier as an adhesion/diffusion layer between Cu and Al layers, thermal aging test was chosen. Heat treatment for an extended period of time promotes inter-diffusion between Cu and Al layers, thus making this a suitable test to expose weaknesses in the barrier layer or in the deposition method. Substrate strips of wirebonded but not molded devices were aged in a N 2 chamber at 17 o C and removed at,,, 1, 2,, and 1 hours for evaluation. Au-Al ball shear strength and failure mode were obtained. The height of the shear tool was programmed at 3µm above the bond pad. Since fine pitch wirebonding results in low squash height, 3-µm tool height is a widely accepted setting to ensure that the ball will be sheared through the range of fully grown intermetallics 3. Au-Al intermetallics coverage and growth during thermal aging was determined using SEM imaging of the underside of Au ball bonds after they were removed from the Al cap by the NaOH etching solution. Au-Al intermetallics coverage and thickness was also examined in cross-sectional SEM images. Grain boundary staining was achieved using an aqua-regia based etchant (ratio of : by volume of HNO 3 :HCl). The effect of thermal aging to the integrity of the barrier and of the Al cap structure was evaluated by four methods: NaOH etching, Au wire rip test, four-point circuit resistance measurement, and cross-sectional Transmission Electron Microscopy (TEM). Since the NaOH etching solution used to etch off Au bonds from the Al cap dissolves the Al cap and exposes the underlying adhesion/diffusion barrier, inspection of the barrier could be conducted. The extent of barrier disturbance was recorded. The wire rip test was used in place of the standard wire pull test. Although the absolute pull strength was not recorded, visual inspection of failure modes could be used to determine the strength of the pad structure. A strong pad structure is indicated by a break at the annealed area above the ball (typically termed as break at neck ). Two less desirable failures modes are ball lift, where the Au ball bond is detached from the bond pad leaving the Al pad intact; and pad lift, where the Au ball bond is detached from the bond pad, pulling Al and the underlying pad structures with it. Devices submitted to package-related reliability testing were assembled in mm x 17 mm x 1.3 mm 1 mm pitch PBGA. The parts underwent Moisture Sensitivity Level 3 (MSL3), preconditioning (1 cycles of temperature cycling -6 o C/1 o C, bake 12 o C/ hrs, moisture soak 3 o C/6%RH/192 hrs, and 3 cycles of Vapor Phase Reflow). After preconditioning, some parts were subjected to Temperature Cycling (-6 o C/1 o C for 2 cycles), and Autoclave (121 o C, 1%RH, 1PSIG for 144 hours). 4. Impact of Probe Mark to Underlying and Cu Pads Figure 2 shows an example of the profilometry output measuring probe marks in the x-direction. Table 1 summarizes the depths of probe scrubs measured across multiple wafers. The depth of probe mark was heavily dependent on probe conditions. In effect, the heavier overdrive, higher number of probe passes, and smaller probe tip diameter resulted in a deeper probe needle penetration into the Al cap. At the small 1-mil tip diameter, the probe scrub was measured about 8-9kÅ deep with 8X probe touches. The depth of probe mark was independent of the underlying barrier. Probe Scrub 9kÅ Al Cap above Passivation Opening Bond Pad Level Passivation Level Figure 2. Profilometry graph of probe mark. Table 1. Depth of probe marks as a function of tip diameter, overdrive, and probe passes. (depth in kå) Probe Passes Overdrive (um) 1X 2X 4X 8X 16X Black: 1.2 m il tip G re y: 1. m il tip The effect of probe marks on barrier types was evaluated after the Al cap was stripped and revealed the barrier. Although the probe scrub did not penetrate through the entire Al cap thickness, barrier disturbance was found. The deeper the probe marks were, due to either a thin Al cap or heavy probe conditions, the harder the force propagated through the Al cap to disturb the barrier. Figure 3 depicts the progression of barrier disturbance as a response of probe conditions. At normal probe condition, no barrier disturbance was detected. At high probe condition, barrier cracking or trenching was found under thin Al cap whereas only barrier trenching was found under thick Al cap. Furthermore, less disturbance was found on barrier A than on barrier B given the same probe conditions. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 334

4 Fine Pitch Probing and Wirebonding and Reliability of Aluminum Capped Copper Bond Pads A Al Cap Thickness = Thin Probe Tip = 1 mil B Al Cap Thickness = Thick Probe Tip = 1 mil The low power/low force combination used in the standard Al process did not produce acceptable bonds on Al capped Cu pads. This indicated that the wirebond process has shifted to the right. Additional process characterization and optimization have been planned. (a) Overdrive = 4 um, # Touches = 4x 6. Aging Behaviors of Au-Al Bonds on Al Cap Structures (b) Overdrive = 6 um, # of Touches = 8x Figure 3. disturbance observed on probed wafers after NaOH etch. No Cu staining or corrosion was found around the barrier and passivation on the thin Al cap, indicating that the Al cap structure was hermetic. At the deepest area of the probe marks, EDX did not detect any Cu exposed from the underlying bond pad. Although no Cu metal was exposed, the barrier disturbance during probing was a concern since this could potentially lead to Cu-Al inter-diffusion path. The thickest Al cap could resolve this concern; however, this would entail longer metal etching time. Therefore, after considering front-end and back-end implications, the mid-range Al cap thickness was recommended as the minimum thickness to ensure pad integrity across the range of probe conditions and as a compromise with fab s metal etch. Wafers with the selected Al cap thickness were probed with a 1 mil probe tip, at 6µm overdrive, six touchdowns, and sent to assembly for evaluation. Through the initial probe experiments, probe process window for Al capped Cu pads was similar to that for standard Al pads. However, further probe optimization was planned to characterize probe parameters to ensure no barrier disturbance at finer pitch geometry. Figures 4 and show ball shear strength as a function of thermal aging time for normal and high wirebond settings for barriers A and B, respectively. Both barriers met the minimum shear strength of 1gf. As the intermetallic thickness increased due to greater inter-diffusion with extended aging, the ball shear strengths at either wirebond setting increased with thermal aging up to hours. Beyond hours, the ball shear strengths decreased. The shape of the curve depicting the shear strength versus time confirms what is found in the literature, in which the ball shear strength increases dramatically from hour to hours, and then increases only slightly beyond hours. Ball Shear Tim e (h rs ) Figure 4. Ball shear strength versus thermal aging at 17 C for barrier A. Ball Shear ( Time (h rs). Wirebond Process for Al Capped Cu Pads An initial five-cell wirebond process evaluation was conducted. At low force setting (2g), non-stick at first bond and ball lift during the wire rip test were found. This was caused by a low Au-Al intermetallic coverage of less than 2%. At a combination of low force and high power settings, pad cratering was seen. Cells with normal and high settings resulted in no failures. These two wirebond conditions were used to process strips for subsequent thermal aging study. From this initial study, it was concluded that the wirebond process for Al capped Cu was different from that for standard Al. Legend for Figures 4 and : Normal Wire Bond Settings High Wire Bond Settings - 1 Std Dev (Normal Wire Bond Settings) x 1 Std Dev (High Wire Bond Settings) Figure. Ball shear strength versus thermal aging at 17 C for barrier B. (sample size/sell: 16 balls/unit x units). At normal wirebond settings, shear strength was similar for both barriers and reached a peak at 2- hours. At high wirebond settings, shear strength for barrier A was higher than that of barrier B. It was verified that there was a statistically significant difference between ball shear readings for the two barriers at high wirebond settings for all temperature readpoints except for the readpoint at hours. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 33

5 Intl. Journal of Microcircuits and Electronic Packaging The superior performance of barrier A over B is further illustrated in Figures 6 and 7 where the shear strength per unit area (gf per mil 2 ) is plotted. The ball shear per mil 2 data give an indication of the stability and robustness of the Al Cap Cu process across different wirebond parameter settings. The minimum shear strength per mil 2 of.gf/mil 2 was met by both structures. Ball shear strength per unit area of barrier A was high and stable across the wirebond process window whereas ball shear per unit area of barrier B decreased at the high wirebond setting, approaching the minimum strength limit. Lower values for barrier B could be attributed to lower shear readings and larger ball diameters. Ball Shear per mi (gf/mil2) um ball size 72um ball size The distribution of ball shear failure modes of the two barriers at normal and high wirebond settings is shown in Figures 9 and 1. At normal wirebond conditions, mode 1 (ball lift) was prevalent at time zero for both barriers due to thin Au-Al intermetallic phases (IPs). With increased thermal aging from hours to hours, the failure mode switched to mode 2 (ball shear) as the IP zone grew thicker. A small percentage of mode 3A (pad lift exposing SiO 2 ) was also observed. This failure mode occurred when there was a mixed failure of a fracture through the IPs and an IP-SiO 2 interfacial fracture. It was speculated that mode 3A was device related. At 1 hours, mode 3B with pad lift exposing barrier was more prevalent. The Al cap was completely consumed into the Au-Al IPs. Shearing through the Au- Al IPs would peel off the depleted Al cap, resulting in weaker ball shear strength. For both barriers at high wirebond conditions, mode 2 (ball shear) dominated at all thermal aging intervals until 1 hours where more mode 3B was observed. Norm alw ire B o n d High W ire B o n d Figure 6. Ball shear strength per mil 2 versus thermal aging at 17 C for barrier A. Ball Shear per mil2 (gf/ um ball size 77um ball size Norm alw ire B o n d High W ire B ond % Failure Mode Normal WB Time (hrs) Figure 9. Distribution of ball shear failure modes for barrier A. Figure 7. Ball shear strength per mil 2 versus thermal aging at 17 C for barrier B. 1% Normal WB Figure 8 illustrates four failure modes encountered during ball shear testing: ball lift (mode 1), ball shear (mode 2), pad lift exposing SiO 2 (mode 3A), and pad lift exposing barrier (mode 3B). % Failure Modes 8% 6% 4% 2% % Time (hrs) Figure 1. Distribution of ball shear failure modes for barrier B. Mode 1 Mode 2 Mode 3A Mode 3B The progression of IP growth was also monitored through SEM images of the underside of the bonded balls after etching off the Al metallization and through cross-sectioning of the bonded balls (Figure 11). Balls bonded onto Al cap Cu pads with different barriers exhibited similar behaviors. SEM analysis of the underside of the Au ball at readpoint of 2 hrs showed the progressively increasing intermetallic thickness. Since NaOH etches Al faster than Au, 2 phases of the IP are clearly seen in Figure 11 at 2 hour thermal aging. Figure 8. Failure modes after ball shear and rip tests. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 336

6 Fine Pitch Probing and Wirebonding and Reliability of Aluminum Capped Copper Bond Pads After Hour Bake (Higher Magnification) After 2 Hour Bake (Higher Magnification) Al rich region Au rich region After the Al Cap was etched off, the barrier metal left on the bond pad was inspected. No disturbance was found on both barrier metals. This indicates that a ball bond could be created between the Au wire and the Al Cap, without disturbing the under- The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) lying barrier material. The four-point circuit resistance measurements were similar for both barriers. The resistance measurement was taken as a response to forced current through the pad structure. The resistance was measured at ohms consistently across the thermal aging duration. The stability of the circuit resistance and the integrity of the barrier after thermal aging were also confirmed with TEM images of the adhesion/diffusion barriers across thermal aging (Figure 14). Both barriers appeared to be intact after 2 hours. Figure 11. SEM images showing underside of Au ball and ball cross-section through thermal aging. B Al Cu B Cu 7. Aging Behaviors of Cu/ A/Al and Cu/ B/Al Pad Structures Figure 14. TEM images of barrier B structure (barrier A structure looked similar and is not shown). Figures 12 and 13 present results of the wire rip test for barriers A and B at normal and high wirebond settings. Minimal lifted balls or pad lifts were recorded for both barriers at thermal intervals less than hours, indicating a strong pad structure. At hours and longer, the progressive transformation and depletion of Al metallization into the Au-Al IPs was recorded as ball lift exposing barrier (failure mode 3B). Pads with barrier A were more robust than those with barrier B, with less pad lifts and ball lifts across thermal aging. % of Pad Lifts 12% 1% 8% 6% 4% 2% % Normal WB 2 1 Figure 12. Wire rip test results for barrier A. % of Pad Lifts 12% 1% 8% 6% 4% 2% % Normal W B A 1 46% 3B Figure 13. Wire rip test results for barrier B. (sample size/ cell: 222wires/unit x 2 units). 3A 3B 8. Reliability Results Table 2 shows the results of reliability testing for packaged units with the two Al cap structures. Both structures passes 2 cycles of temperature cycling. Some packages failed open or short testing during 96 hours and 144 hours of autoclave. Failure analysis indicated that these failures were not associated with the Al cap structure or the barrier. Bond pads were corroded as a result of fab processing. With subsequent change in fab processing, packages passed autoclave testing without failures. Table 2. Reliability results. A MSL3 Pre-conditioning /8 (a) Air-to-Air Temperature Cycling -6C Autoclave to 1C hours Cycles Cycles Cycles Cycles hours /4 (a) /38 (a) /3 /3 2/38 (c) 3/36 (d) B MSL3 Pre-conditioning 1/8 (b) Air-to-Air Temperature Cycling -6C Autoclave to 1C hours Cycles Cycles Cycles Cycles hours /4 (a) /38 (a) /3 /3 2/39 (b) /37 (a) Parts Removed for Physical Analysis (b) Open - Missing Solder Ball (c) 1 Open - Recovered After Clean, 1 Short - Corroded Bond Pads (d) 1 Short, 2 Open - Corroded Bond Pads 337

7 Intl. Journal of Microcircuits and Electronic Packaging 9. Selecting One adhesion/diffusion The Au-Al intermetallic growth behavior on Al capped Cu pads was in very good agreement with that on standard Al pads. Packaged parts have passed production-level reliability evaluations. Through thermal aging test, Cu pads re-metallized with Al film separated by either barrier performed similarly in many aspects: stable circuit resistance, good adhesion and diffusion property, and strong pad structure over time. There exist, however, some statistically significant differences between the two barriers that indicate the more robust barrier A/ Al structure for back-end processing. Less disturbance was found on barrier A than on barrier B given the same Al cap thickness and probe conditions. A/Al structure provided statistically improved performance over B/Al structure in ball shear strength across the evaluated wirebond process window, and more stable Au-Al intermetallic strength over thermal age duration. Less pad lift was observed on barrier A than on barrier B. About the authors Tu Anh Tran and Lois Yong are presently working at Motorola Final Manufacturing Technology Center focusing on wirebond package development. Bill Williams is in Motorola Final Manufacturing Technology Center focusing on fine pitch probe development for wirebonded packages. Scott Chen and Audi Chen are currently working in the process engineering group at Advanced Semiconductor Engineering (ASE) in Chung-Li, Taiwan. Acknowledgments The authors would like to thank the following people for their support in their respective expert areas: Scott Pozder and Thom Kobayashi in leading the Al Cap Processing team; Gloria Estrada, Dennis Kiffe, Gordon Fowkes and John Milton in conducting thermal age testing; Van Ho and Jeff Bosworth for electrical testing and reliability testing; Chuck Miller, Kevin Hussey, Gary Clark, Ava Morrision, Steve Heineke and Chongnan Kim for surface and physical analyses ; Joan Sibbitt, Tony Angelo, Xuefeng Liu and Mary Di for probe testing and Failure Analysis work; and Andrew Mawer and Thomas Koschmieder for providing N 2 oven support. References 1. T. Tran, Approaches to Probing and Wire bonding onto Copper Bond Pads, Motorola 1999 Hermes Symposium, Austin, Texas, September 2-, C. Ryu, H. Lee, K.W. Kwon, A. Loke, and S. Wong, s for Copper Interconnections, Solid State Technology, pp. 3-6, April K. Dittmer, S. Kumar, and F. Wulff, Influence of Bonding Conditions on Degradation of Small Ball Bonds due to Intermetallic Phase (IP) Growth, Proceedings of SEMICON Singapore, Test, Assembly & Packaging, May, The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number3, Third Quarter 2 (ISSN ) 338