TRANSIENT THERMAL ANALYSIS OF AN ANISOTROPIC CONDUCTIVE FILM PACKAGE ASSEMBLY PROCESS

Transcription

1 9-th International Flotherm User Conference October 16-19, Orlando, Florida TRANSIENT THERMAL ANALYSIS OF AN ANISOTROPIC CONDUCTIVE FILM PACKAGE ASSEMBLY PROCESS Victor Adrian Chiriac 1 and Tien-Yu Tom Lee 2 Interconnect Systems Laboratory Semiconductor Products Sector, Motorola, Inc E. Elliot Road, Mail Drop EL725 Tempe, Arizona (480) (480) (480) (Fax) victor.chiriac@motorola.com 1 tom.lee@motorola.com 2 ABSTRACT Transient thermal simulation was performed to analyze thermal response of the assembly process for a package using Anisotropic Conductive Film (ACF). Two assembly processes were modeled: a simplified process where the package was fixed at two different temperatures during assembly, and a detailed process where the package experienced a ramping heating process, followed by a constant temperature curing process. A 3D conjugate Computational Fluid Dynamics (CFD) study was first conducted, followed by a 3D conduction-only analysis due to the minimal effect of convection and radiation. Results from the detailed process modeling indicated that during the initial ramping, within 0.02 second, the die and nozzle head experienced a small temperature drop due to the cooling effect of the ACF material and substrate. The ACF material also displayed a steep increase in temperature after contacting the die, followed by a short decay, then ramped up again. At the end of the 10-second ramping process, the ACF reached a temperature of almost 203 C, while the die was at 206 C. During the 5 seconds of curing, all parts reached steady state in less than 2 seconds. NOMENCLATURE P Power dissipation (W) T Temperature ( C) t Time (s) k Thermal conductivity (W/m-K) c p Specific heat (J/kg-K) ρ Density (kg/m 3 ) INTRODUCTION Anisotropic Conductive Films (ACF) are strong candidates as die attach materials for flip chip package interconnects, offering simplified structures and environmentally compatible processing compared to solder joining and underfilling (Delaney and White, 2000). Among the main advantages of using ACF instead of the solder joints and underfill are the finer pitch (Watanabe et al., 1996), weight and size reduction, simplified structures and lower costs. ACF is a better candidate for low power applications compared to others, and a better understanding of its thermal behavior motivated recent work. This technology was recently applied to a wide range of mobile and portable electronic products (Watanabe et al., 1996) such as cellular phone and personal digital assistance, in addition to traditional applications, such as the wristwatch and electronic calculator. Furthermore, ACF based flip chip technology is lead free and environmentally friendly (Watanabe et al., 1996). During previous mechanical analyses (Yamaguchi et al., 1989; Zonghe and Liu, 1996; Lam et al., 1997; Kristiansen and Hiu, 1998; and Yim and Paik, 1999), it was learned that the advantage of using the ACF package was affected by the interconnect reliability under different real life testing scenarios. In this respect, there are several reliability issues to be considered, such as mechanical stresses, elevated humidity and detrimental temperature effects within the ACF package (Nagai et al., 1998). 1

2 The transient thermal response of a package is a critical issue, as it is not always useful to predict timeaveraged temperatures, especially when manufacturing, assembly or duty cycles lead to temperature variations that can have a negative effect/impact on the package structure and functionality. Large temperature gradients and peak temperature values pose serious limitations on manufacturing and assembly processes. The literature review revealed that the thermal behavior of the ACF package during its assembly process is not at all well known. Hence, the need of understanding the package thermal behavior is critical for accurate prediction of the package general mechanical performance under certain constraints and conditions. The present study investigates the package thermal behavior during the assembly process. The thermal resistance of the ACF material has to be small in order to prevent excessive die heating. The assembly and manufacturing conditions should be such that the temperature gradients inside the packages, and their evolution in time, are not detrimentally high. The continuous growth in component power dissipation combined with the system size reduction leads to the need for simple, yet accurate methods of estimating package assembly/operating temperatures. In order to address some of these issues, a transient thermal study of the ACF package was therefore performed. PACKAGE AND SYSTEM DESCRIPTION The ACF material normally consists of an epoxy resin filled with conductive particles approximately 3-10 µm in size. During the assembly, the substrate is first pre-laminated with the conductive film (ACF) at elevated temperature and pressure (Lam et al., 1997). Next, the die is flip chip bonded to the substrate with appropriate bonding temperature, pressure and cure time. The conductive particles between the die bump and the substrate pad make electrical contact while leaving the particles outside of the bonding area dispersed (Delaney and White, 2000). Because of film's different out- and in-plane electrical behavior, it was named anisotropic conductive film (ACF - terminology currently adopted by the industry). In order to investigate the package transient thermal response during the assembly process, a model was created. The simplified cross-sectional structure is shown in Fig. 1. The substrate is composed of two metal layers (top and bottom) and FR-5 core material. The properties of the metal layer were estimated by the volumetric ratio of copper and solder mask. The ACF film has nickel conductive particles dispersed in epoxy resin. The silicon die has 68 peripheral nickel bumps. The substrate trace pad is copper. The dimensions of package assembly and their thermal properties are listed in Table 1. Descr. Solution Domain ACF Material ACF Ni Cu Metal Layer Figure 1: Cross-sectional view of ACF package Table 1: ACF Package Geometries and Properties Dimensions (mm) Material Th. Cnd. (W/ mk) Dens. ρ (kg/ m 3 ) Spec. Heat c p (J/kg K) 6.8x6.2x0.73 Silicon Var w/ T ACF 82x82x0.052 Epoxy Material Ni Ni 0.15x0.15x0.02 Ni Bump 6 Cu Pad 0.15x0.15x0.02 Cu Substr. Cu Metal Layer thick Solder mask Substr. 100x100x0.4 FR Core core Top 30x1.02x34 Alumina Nozzle (96%) Bot Nozzle 10x1.02x6.2 Alumina (96%) PROCESS FLOWCHART This study consists of two parts: a simplified assembly process and a detailed assembly process (Fig 2). First, for a simplified assembly process, the backside of the die was fixed at a constant temperature of 220 C (representing the bonding head temperature) and bottom of the substrate was fixed at 70 C (0). The rest of the solution domain was assigned to an open boundary condition and initially was set to an ambient temperature of 25 C. The total assembly time is 10 seconds. The second part of the study uses the pure conduction analysis to model a detailed assembly process. It consists of an initial separation phase (I) when the die was picked up by the nozzle and maintained at 100 C, while ACF layer is maintained at 2

3 70 C. Then, a 10-second temperature ramping-up process (at nozzle top surface) from 100 C to 220 C during assembly (II), followed by a 5-second cure at a constant temperature of 220 C (III). The detailed process modeling examines the package thermal behavior before, during, and after the bonding and curing procedures. (0) T= 100 ºC Bottom Nozzle T= 70 ºC (I) ACF 70ºC 220ºC ACF T= 220 ºC ACF T= 100ºC - 220ºC T= 70 ºC Bottom Nozzle T= 70 ºC Bottom Nozzle Figure 2: Schematics of Assembly Flow (II) (III) Radiation heat loss was considered from all package surfaces. View factors for the identified surfaces were calculated in the software. All radiative surfaces were assumed to have an emissivity of 0.8. THERMAL AND HYDRODYNAMIC MODELING Computational Fluid Dynamics (CFD) software Flotherm V2.2 (Flotherm, 1999) was used for the simulation. It solves both momentum and energy equations for two/ three-dimensional steady or unsteady flow regimes. The discretized equations are derived from the differential equations using a finitevolume discretization procedure (Patankar, 1980). In the simulation, due to the small gap (filled with ACF particles, ~2 µm) between the nickel and copper pads (Fig. 1), a "plate" model was inserted in the gap, applying a "conduction resistance" in the y-direction while ignoring conduction in both x, z directions. A detailed grid resolution study was performed, with non-uniform grids used in x, y, respectively z directions. Three grid structures are considered: 99 x 20 x 99, 129 x 24 x 129, and 159 x 28 x 159, grid numbers in x-, y-, and z-directions, respectively. The grid nodes were appropriately refined near package and wall surfaces in order to capture the large temperature gradients. Results of the study indicate that the grid structure of 129 x 24 x 129 accurately captured the physics of the flow and was selected for further investigation. The 68 peripheral bumps were modeled as four rows of solid blocks (one along each die edge) with effective thermal conductivity. A complementary time-step study was performed and revealed an initial step t = 1E-03 s to be adequate. Initial investigation involved a full CFD study, with the package enclosed in a cuboid 0.1 x 0.1 x 0.1 m. The ambient air (cooling fluid) surrounding the package is initially at 25 C. At solution domain faces, first derivative of the velocity, respective temperature fields is enforced as the appropriate boundary condition (open faces). No-slip condition (zero-velocity) is considered at package faces. At time t = 0, overall temperature is at ambient value, while velocity field is 0 m/s. The full CFD study indicates that the maximum value for the velocity field surrounding the package was m/s (Fig. 3). This value is much smaller than the normal natural convection value of ~0.2 m/s, indicating weak buoyancy effects and negligible impact of related flow field on assembly process thermal characteristics. Based on this, as the total convection and radiation loss from the package surfaces accounted for less than 5% of the total heat flow, it was concluded that the conjugate heat transfer analysis is not required. This will lead as well to a reduction of the computational domain, which benefits the overall CPU time and related computational costs. A separate study was conducted by replacing all the open boundary conditions with the adiabatic boundary conditions (insulated surfaces) and a pure conduction analysis was performed. Table 2 shows the changes in temperature are less than 1% between two different boundary conditions. For the pure conduction modeling, the computational runs were about 30-40% faster than the conjugate cases. Based on these results, it was decided to model the manufacturing processes using pure conduction analysis. Figure 3: Velocity field - induced by natural convection Table 2: Open vs. Adiabatic Boundary Conditions Study Location Conjugate Case Temperature ( C) Adiabatic Case Temperature ( C) 3

4 Mid Mid ACF Top Metal Layer Mid & Bottom Mid Metal Layer RESULTS AND DISCUSSIONS Simplified Assembly Process An initial transient run of 10 seconds indicated that the simplified assembly process reached steady state in less than one-second timeframe. Hence, the total solution time was reduced to 1 second with 100 time steps. A power law was applied to discretize the time grid. A very fine time step was created at the initial start-up (~1 ms for the first time step), and gradually increased close to the end of the one second (Fig. 4). Fig. 5 illustrates the temperature field of the top metal layer (in the substrate) at the end of 1 second of bonding (ramp) time. The metal layer temperature was uniform at ~70 C except under the die/acf area. Detailed information is shown in Fig. 6. A typical radial heat spreading from the center of the package to the surroundings is observed. The temperature field in the ACF layer showed a ~30 C decrease from the center (218 C) to the edge of the die (188 C) and a further 80 C drop from the edge of the die to the edge of the ACF (Fig. 7). Figure 5: Temperature field - Top metal layer, simplified process (after one second) Figure 6: Temperature field - Top substrate metal layer under die/acf region, simplified process (end of bond time) Figure 4: Time-step distribution, simplified process Figure 7: Temperature field - ACF layer, simplified process (end of bond time) 4

5 z Middle of Top Metal Layer Middle of ACF Middle of Core Layer Between Top Metal Layer and Middle of In the detailed assembly process, a two-layer bond lead was placed on top of the die, as seen in Fig. 9. The bond head is fabricated from Alumina (Al 2 O 3, 96% pure), and its properties are listed in Table 1. The initial conditions are different from the simplified process. Prior to the assembly, the ACF is attached to the substrate, maintained at 70 C. The die is picked up by the bond head, and maintained at 100 C. When the die is assembled to the ACF/substrate, the bond head begins to ramp up to the curing temperature of 220 C. Once it reaches 220 C, the curing process begins. It was assumed that the ACF and substrate were initially at 70 C, while the die and nozzles were at 100 C. During the bonding, a linear ramping profile is assumed on the top nozzle being heated up from 100 C to 220 C in 10 seconds. After 10 seconds, the top nozzle temperature is fixed at 220 C for 5 seconds to represent the final stage of curing (Fig. 2). The overall time step distribution for the detailed assembly process is shown in Fig. 10. A power law was applied to discretize the time grid. A fine time step was created at the initial start-up (~1 ms for the first time step), and increased gradually close to the end. The actual bonding and curing temperature and time may be varied; however, the above assumptions will give a general trend of the thermal responses of the ACF assembly. Between Middle of and Bottom Metal Layer Bot Nozzle y x ACF Figure 9: ACF assembly-bond head structure, detailed process Figure 8: Transient temperature response, simplified process Fig. 8 illustrates the temperature response within the package as a function of assembly time. Several monitor points are assigned across the package. At time zero, die backside was set at 220 C of the bonding temperature, while substrate bottom was at 70 C. Results showed that the temperature field in the package became stable after only ~0.35 second, indicating fast heat diffusion inside the package. It also indicates that the die, ACF and top metal layer reach steady state in less than 0.06 second, much faster than the substrate, due to the combination of large thermal diffusivity and small material thickness. The largest temperature drop occurred across the substrate as compared to the die and ACF, mainly due to the poor thermal conductivity and relatively thicker substrate core material. Figure 10: Time-step distribution, detailed process Detailed Assembly Process 5

6 field Between Top Metal Layer and Middle of (Subst Top2) Between Middle of and Bottom Metal Layer (Subst Bot1) Top and Mid of Mid of Bot Nozzle Lateral of Bot Nozzle Top and Middle of Middle of ACF Top Metal Layer (Subst Top 1) Middle of Core Layer (Mid Substr) Bot of (Bot Face) Figure 11: Monitoring points for temperature The results of the full 3-D transient thermal response were documented by monitoring the temperature at several locations inside the packagebond head assembly. A total of 12 monitoring points are located across the package, as seen in Fig. 11. The temperature profiles of the monitoring points during the 15-second assembly process are shown in Fig. 12. Figure 12: Transient temperature response, detailed process During the first 10 seconds of the bonding process, all parts above the substrate (including the top metal layer in the substrate) followed the similar transient response of the ramping profile. In the substrate, the thermal responses were much slower than in the ACF and die due to the small diffusion rate across the substrate (with low thermal conductivity). At the end of 15 seconds, the maximum temperatures in the die and ACF were at ~209 C and ~206 C, respectively, approximately 11 C cooler than in the previous simplified study. During the 5 seconds of cure at 220 C, the temperature profiles for all monitoring locations leveled out after ~2 seconds, as compared to the simplified study when it took only ~0.35 second for the profiles to reach steady state. This indicates that for the temperature linear-ramping process, the gradual increase in top surface temperature induces a lower heat diffusion rate inside the package, as opposed to the case when there is a stepchange in temperature at the top of the package. A different trend was observed for the first 0.5 second (Fig. 13). Results indicated that during the initial ramping, the die first experienced a small temperature drop (less than 5 C) due to the relatively large heat capacity of the ACF and substrate. The die then began to heat up in less than 20 ms. The bottom surface of the bond head took longer time (~0.2 second) to recover from the initial temperature drop. The ACF material displayed a steep increase in temperature from its initial 70 C to 97.3 C after contacting the die, followed by a short decay (less than 20 ms) then ramped up again. The thin ACF has a very quick temperature response to the ramping condition. The substrate layers also experienced a quick increase in temperature during the first 0.2 second, then slowed down the heating process after ~0.2 second. Mid Mid Mid ACF Subst. Top1 Subst. Top2 Top Lateral Bot Nozzle Mid Bot Nozzle Mid Substr Mid Lateral Bot Nozzle Subst. Bot1 Mid Bot Nozzle Mid ACF Top Subst. Top1 Subst. Top2 Bot Face Mid Substr. Subst Bot1 Figure 13: Start-up for temperature profiles (first 0.5 sec) for detailed assembly process Bot Face 6

7 Bottom Nozzle Figure 14: Cross-sectional temperature field, at curing cycle end (after 15 seconds) assembly process, results indicated that during the initial ramping of 10 seconds from 100 C to 220 C, the die and bond head experienced a small temperature drop due to the cooling effect of the ACF and substrate. During the same initial ramping, the ACF material displayed a steep increase in temperature from its initial 70 C to 97.3 C, followed by a short decay (less than 2 C in 0.02 second), then ramping up again. The temperature drop across the nozzles was ~11 C, while across the die and ACF material it approached 3 C. The substrate experienced the largest drop in temperature, almost 136 C. As a result of the study, it was learned that the manufacturing process induces large temperature gradients inside the substrate, while the anisotropic film encounters only small changes in temperature value. The peak temperatures encountered by the film during the assembly process are within acceptable manufacturing limits, hence the procedures being used do not impact negatively the overall package structure and good functionality. ACKNOWLEDGMENTS The authors would like to express their gratitude to Lei Mercado and Jerry White, for providing package and material information and general guideline on ACF assembly. Thanks also go to Robert Hapke who maintained the computer network. REFERENCES Delaney, D and White, J., 2000, "Flip Chip Assembly Utilizing Anisotropic Conductive Films: A Feasibility Study", Proceedings of the 50 th ECTC Conference, May 2000, pp Figure 15: Top view of the ACF temperature field, at curing cycle end (after 15 seconds) The temperature contours of the mid-center crosssectional area of the ACF package at the end of 15 seconds are represented in Fig. 14. An almost 6 C temperature drop occurred across the top ceramic bond head layer, followed by another 5 C drop across the bottom bond head layer. The temperature drop (in the y-direction) across the die and ACF material was ~3 C, while ~136 C of temperature drop occurred across the substrate. In a similar fashion, the top view of the ACF (Fig. 15) showed that the center of the ACF was almost 206 C while the edge of the ACF layer was at ~116 C which was close to the value in the simplified process. CONCLUSIONS This work demonstrated the capability of using a CFD tool to model a transient thermal analysis for the ACF package assembly process. In the detailed z X Flotherm, 1999, Version 2.2, Flomerics Ltd., Surrey England. Kristiansen, H. and Hiu, T., 1998, Overview of Conductive Adhesive Interconnection Technologies for LCD s, IEEE Trans. Comp. Packaging, Manufacturing Tech. Part A, Vol. 21, pp Lam, D. C. C., Shen, C., Xie, J. F., Karim, Z., and Tong, P., 1997, Structural Reliability of Direct-Chip-Attaches Bonded with Anisotropic Conductive Film, Mat. Res. Soc. Symp. Proc., Vol. 445, pp Nagai, A., Takemura, K., Isaka, K., Watanabe, O., Kojima, K., Matsuda, K., and Watanabe, I., 1998, Anisotropic Conductive Adhesive Films for Flip-Chip Interconnection Onto Organic s, Proceedings of IEMT/IMC, pp Patankar, S.V., 1980, "Numerical Heat Transfer and Fluid Flow", Hemisphere Publishing Corp., Mc-Graw Hill Book Co., New York. Watanabe, I., Shiozuka, N., Takemura, K., and Ohta, T., 1996, "Flip Chip Interconnection Technology using Anisotropic Conductive Adhesive Films," Flip Chip Technology, Ed. J. Lau, McGraw-Hill, New York, pp Yamaguchi, Y., Tsukagoshi, I., and Nakajima, A., 1989, "Anisotropic Conductive Film," Circuit Technol., Vol. 4, No. 7, pp

8 Yim, M. and Paik, K., 1999, The Contact Resistance and Reliability of Anistropically Conductive Film (ACF), IEEE Trans. on Adv. Packaging, Vol. 22, No. 2, pp Zonghe, L. and Liu, J., 1996, "Anisotropically Conductive Adhesive Flip-chip Bonding on Rigid and Flexible Printed Circuits," IEEE Trans. CPMT B, Vol. 19, pp