Transfer Molding Encapsulation of Flip Chip Array Packages

Transcription

1 Intl. Journal of Microcircuits and Electronic Packaging Transfer Molding Encapsulation of Flip Chip Array Packages Louis P. Rector*, Shaoqin Gong, and Tara R. Miles Dexter Corporation 11 Franklin Street Olean, New York Phone: Fax: s: Kevin Gaffney Amkor Technology 1900 South Price Road Chandler, Arizona 8548 Phone: Fax: *Author to whom correspondence should be addressed Abstract Epoxy molding compounds have been developed which can simultaneously underfill and overmold the Flip Chip die in a single transfer molding process. Transfer molding is a well-defined industry process; with suitable mold design, these materials can utilize the currently installed capital base. The ability to apply pressure during the molding process can reduce the void rate under the die as well as achieving production efficiencies which are typical of transfer molding processes. Transfer molding compounds offer enhancements in thermal expansion coefficients and moisture absorption levels relative to traditional liquid underfills as well as low levels of package deformation due to cure shrinkage and thermal mismatch effects. These improvements are achieved through the use of unique resin chemistries and filler package compositions. This paper will provide an overview of molding compounds which are useful as transfer molding underfill/ encapsulant materials. Key words: 1. Introduction and Background Flip Chip, Transfer Molding, Encapsulation, Underfill, and Epoxy Molding Compound. 400 Increases in I/O densities and processing speeds continue to drive the evolution and applicability of Flip Chip packaging for small, lightweight components. 1 To improve solder joint and device reliability, Flip Chip packages are conventionally underfilled with a liquid material capable of filling the small (5-75micron) gap between the chip and the substrate. In the majority of cases, the chip is subsequently protected with either a liquid encapsulant or a transfer molding compound. The liquid underfill process (dispense, flow, and cure) is typically slow and prone to defect formation, such as voids and filler streaking or settling, if the process is not

2 Transfer Molding Encapsulation of Flip Chip Array Packages carefully designed and controlled. Advanced liquid underfill compositions have been developed which mitigate some of these concerns. Alternative underfill technologies continue to be explored and have utility depending on the application. For example, no-flow underfill materials have been investigated extensively. In this case, the undefill layer is deposited before the solder interconnect step. However, reliable interconnect formation during solder reflow and the development of materials with suitably high filler levels may be a technical challenge with the no-flow approach. 3 Employing a single-step transfer molding process to underfill and overmold a Flip Chip device offers a number of technical and process advantages. The initial development of this process has been previously reported, as well as the potential for four fold improvement in rates of production versus a conventional underfill process. 4 Other workers have also reported initial investigations of the molded underfill process 5-6. The application of external pressure to drive the underfill process allows the use of higher filler content materials compared to conventional liquid materials. It is well known that increases in filler content lead to improvements in moisture absorption and thermal expansion coefficients of epoxy molding compounds. Increases in filler content also commonly translate to improved device performance in JEDEC and thermal cycling evaluations. In addition, high performance liquid underfill compounds are also typically expensive in terms of both material cost and floor life. The introduction of an underfilling transfer mold compound can substantially improve the economy of the encapsulation process in high volume applications. The development of a successful transfer molded underfill/ overmold encapsulation material and molding process present a number of technical challenges. The judicious design of the mold and choice of molding parameters are equally critical to the implementation of this technology. These materials have been designed to have the ability to flow significant distances in small channels (5-50µm). The long flash character of these materials and the potential need for vacuum assist present challenges for mold chase and vent design, which must be addressed by development with the mold manufacturer. The performance of the molding compound is also crucial. The filler particle shape, size, and size distribution must be optimized to provide good gap filling capabilities. The choice of the epoxy resin and phenolic hardener is partially governed by the need for low viscosity, which allows maximization of the filler content, while maintaining other properties, such as low warpage. As expected, there are interaction effects present between the various components in a molding compound which influence device reliability. It is the objective of this paper to present material developments of molded Flip Chip (MFC) compounds as well as device reliability results.. Experimental Work Several experimental designs were conducted as part of this development effort. The first such study was a simple lattice mixture design focusing on the performance of three epoxy resin types. In this study, the extremes in composition were prepared (100% of each epoxy), along with two axial blends of each pair (1/3:/3 and /3:1/3 ratios) and the overall midpoint composition (1/3 of each component). Spherical silica (100% sized below 15 µm) was used as the filler. A second study investigated the effect on epoxy molding compound (EMC) properties of the variations in filler composition. Other formulation ingredients (phenolic hardener, flame retardants, stress modifiers, coloring agent, release agents, coupling agents, and catalyst) were also present, the levels of which were held constant. The molding compounds were prepared by initially dry-blending the ingredients followed by melt-mixing. The material was extruded in sheet form, allowed to cool, and ground to a fine powder. Standard transfer molding techniques were used to fabricate various test specimens. In-mold cure times were s at 165 C followed by a four hour post-cure, also, at 165 C. Flexural properties were evaluated in a three-point bending mode on an Instron 406 unit. The flexural testing conformed to ASTM D790-96a. Flexural bar dimensions were 5.0 x 0.5 x 0.5; a cross-head speed of 0.10 in/min was used. To assess moisture uptake, molded disks.0 inches diameter by 0.15 inches in thickness were conditioned in an 85% RH/85ºC environment. The disks were periodically removed, weighed, and returned to the test environment. Shrinkage before and after post-cure was measured on 5.0 x 0.5 x 0.5 bars. The glass transition temperature and thermal expansion coefficients were measured on a Thermal Analysis TA 100/940 unit with a heating rate of 10C /min. Flash and channel flow were measured in industry standard molds. Data analysis was conducted using Design-Expert, a statistical analysis software package produced by Stat-Ease, Inc. 3. Results and Discussion The flow of the molding compound in narrow channels, which has been found to correlate with the ability of the material to underfill a die of corresponding offset from the board, was measured for channel heights of 5 µm, 50 µm, and 75µm. The dependence of 50µm channel flow length upon epoxy composition is shown in Figure 1. In this case, an empirical linear model fits the data quite well (R = ), 50µm channel flow (mm) = 7.65*A *B *C (1) where A, B, and C represent fractions of the three epoxies. The magnitudes of the coefficients trend with the values reported by the material suppliers for the epoxy melt viscosities. For the 5µm 401

3 Intl. Journal of Microcircuits and Electronic Packaging channel flow, interaction effects become significant (Figure ). As might be expected from such a multi-parameter model, the fit is quite good (R = 0.990), 5µm channel flow (mm) = 17.19*A *B * C *A*B *A*C *B*C () A Shrinkage of the molding compound upon post-cure was a second parameter to be optimized, since it can be related to the warpage of the encapsulated package (see, for example References 7,8.) As trends towards packages with larger, thinner die, and thinner, more flexible substrates become more prominent, the minimization of package warpage becomes increasingly important. This is especially critical in array format packaging, where in addition to the above trends, the overmold is becoming larger and thinner as well. The measured warpage is strongly package and cure schedule dependent, as well as being a function of the viscoelastic relaxation character of the encapsulant. As an example of the dependence of warpage on cure schedule, the deformation of a 35mm BGA package was measured as a function of in-mold cure time. The results are summarized in Table 1, where the warpage measured via Shadow Moire is strongly dependent upon in-mold cure time. This is related to the degree of cross-link density achieved prior to the removal of pressure on the system and to relaxation effects which occur at the molding temperature Table 1. Dependence of room temperature warpage (µm) upon in-mold cure time. The two EMCs were used to encapsulate a 35mm BGA Package. B C In-Mold Cure (s) (sec) EMC A EMC B Figure 1. Dependence of 50µm channel flow (in mm) at 165 C on epoxy composition. A The shrinkage measurement was chosen as an initial screening tool to separate the effects of molding compound performance (cure shrinkage and thermal contraction effects) from the effects of package structure. The best fit empirical relation between epoxy composition and shrinkage upon post-cure is given in equation (3), however, the goodness of fit is low (R = ), Shrinkage (%) = *A *B *C (3) 40 B Figure. Dependence of 5µm channel flow (in mm) at 165 C on epoxy composition. C Comparison of the dependencies of shrinkage (to be minimized) and channel flow lengths (to be maximized), a binary blend of resins A and C is preferred. The measurements of other molding compound physical properties not reported in this publication (such as, moisture absorption under conditions of 85%RH/ 85 C, high temperature modulus and strength), support this conclusion.

4 Transfer Molding Encapsulation of Flip Chip Array Packages The physical properties of one of the optimized compositions are summarized in Table. This material has a relatively high Tg and a low modulus. Compared with conventional liquid underfill materials, it also has decreased moisture absorption, due to its higher filler content. As discussed above, the material was formulated with a relatively long flash and good flow characteristics in order to promote uniform package encapsulation. Table. Properties of MFG material (-19A). It is well known that the filler loading level, particle size and distribution, and particle morphology have profound effects on EMC material properties such as viscosity, spiral flow, channel flash, coefficient of thermal expansion, mechanical properties, and moisture absorption. The effects of filler loading on the material properties of a MFC material were reported previously 4. To study the effects of particle size and distribution on MFC performance, three 1 spherical silica fillers A, B, and C with median particle sizes of µm, 0.9 µm, and 0.4 µm, respectively, were selected for the study. 76 Filler System 1 consists of mixtures of fillers A and B; Filler System consists of mixtures of fillers A and C. Both filler systems 7 74 examined ratios of the large filler (A) to the small filler (B or C) 70 ranging from 90:10 to 75:5. The base formulation used for this 68 study had a 61% by volume filler loading. Additional components present in the formulation are identical to the epoxy mixture study % Filler A discussed above. Figure 3 shows the Shimadzu viscosity versus the percentage of Filler A for both filler systems. The viscosity decreases with the Figure 4. Dependence of the maximum packing density for increase of Filler A for both filler systems. The viscosity of Filler the indicated filler systems on filler A level. System is observed to be higher than that of Filler System 1 below the level of 85% of Filler A. To describe the viscosity of a It has been demonstrated that Mooney equation holds for the highly filled dispersion, numerous approaches have been proposed suspension viscosity for a range of volume fractions near the close in the literature. These treatments are efforts to extend Einstein s packing density 1. Mooney s theory predicts that the viscosity of classical analysis of the viscosity of a dilute suspension of rigid the dispersion will decrease with increasing maximum packing spheres in a viscous liquid to suspensions of higher filler concen- density at a given filler loading. For both systems, the viscosity is tration 9. Among these approaches, the semi-empirical equation developed by Mooney has been widely used 10,! =! o exp[k e " f /(1-[" f /" max ]) (4) where! is viscosity of the suspension,! o is the viscosity of the binder, K e is the Einstein s coefficient, " f is the volume fraction of filler, and " max is the theoretical maximum packing density. Viscosity (poise) % Filler A 1 Figure 3. Minimum Shimadzu viscosity for filler systems 1 and at 165 C as a function of filler A (expressed as a percentage of total filler) level. The maximum packing density of the filler system is calculated via a computer algorithm developed by Lee based on the size distribution of the filler. 11 Figure 4 shows the calculated maximum packing density versus the percentage of filler A for both filler systems. In this case, it is seen that the maximum packing density decreases with the amount of filler A in both filler systems. The maximum packing density of Filler System 1 is consistently lower than that of Filler System. These results agree with previous observations of the effects of the volume and diameter ratios of the large to small particles on the maximum packing density. 11,1 Maximum Packing Percentage 403

5 observed to increase with increasing maximum packing density. This is due to the fact that the semi-empirical Mooney equation neglects the effects of particle-medium and particle-particle interactions on the suspension viscosity. For a system with strong interactions, viscosity will increase with increasing surface area or decreasing particle size.10,1 The silica fillers used in the molding compound were treated with epoxy-functional silane which is reactive with the binder system, increasing the particle-medium interactions. Therefore, viscosity is determined by both the maximum packing density factor and the particle surface area/size factor. At a constant filler loading, the larger the maximum packing density or the larger the particle surface area, the lower the viscosity. The BET surface areas of fillers B, and C are 6, and 13 m/g, respectively. The differences in BET surface areas may account for the variation of the viscosity with the amount of the large filler for both filler packages and the difference (below 85% of Filler A) between the two filler systems. These two factors (maximum packing density and BET surface area) compete with each other within the range of this study; the surface area factor appears to be dominant. Figure 5 shows the relationship between spiral flow and the percentage of filler A for both filler systems. For both filler systems, the spiral flow roughly increases with the percentage of filler A, and, therefore, decreases with viscosity. However, spiral flow is a complex rheo-kinetic event, as has been previously reported13, and filler loading effects can be masked by curing effects. Figure 6 shows the dependence of channel flow on the percentage of Filler A for both filler systems. The channel flows of Filler System 1 are consistently higher than that of Filler System, except at 90:10 filler ratio. Other material properties such as the glass transition temperature, thermal expansion coefficient, and flex strength, and modulus (both room temperature and 15 C) have also been measured. No substantial difference has been observed among these samples. The moisture absorption of Filler System 1 after 1 week 85 C/85%RH conditioning is slightly lower than that of Filler System. This effect may be related to the fact that the saturated level of moisture absorption increases with filler specific surface area. Channel Flow (mm) Intl. Journal of Microcircuits and Electronic Packaging µm (1 ) 7 5 µm ( ) 5 0 µm (1 ) 5 0 µm ( ) 5 µm (1 ) µm ( ) 90 % of Filler A Figure 6. Dependence of channel flow length at 165 C filler A level. Filler system is indicated in parentheses. In the epoxy mixture study outlined above, two promising candidates were identified which provided the best balance of properties (such as shrinkage, moisture absorption, channel flow). These candidates (-13A and 19A) were evaluated in two Amkor Technology prototype array format packages. The underfill characteristics of a material similar in composition to that of -13A was evaluated in a previous study using a test die and molding facilities provided by Fico.4 In that study, the capability of this class of compounds to effectively transfer underfill/overmold devices with small gap thicknesses (5-50µm) was demonstrated. The Amkor package characteristics are given in Table 3. In both packages, the die thickness is 0.30mm, the gap height is 40µm, the bump diameter is µm, the soldermask thickness is 0µm, the substrate thickness is 0.mm, and the total mold cap height is 0.6mm. A typical device cross-section is shown in Figure 7 for material 19A. Table 3. Amkor test device characteristics. Package Package Die Dimensions* Dimensions (mm) (mm) I 8 x 8 x x.6 II 17 x 17 x x 10.5 # of Solder Peripheral Bumps 64 (single row) 56 (three rows) # of Die per strip Spiral Flow (cm) *Package dimensions are given without BGA balls attached. 100 silicon µm % Filler A 90 Figure 5. Dependence of spiral flow length at 165 C for the indicated filler systems on filler A level. solder joint soldermask Figure 7. Cross-section of Amkor Package I using material -19A showing complete filling of the Flip Chip gap. 404

6 Transfer Molding Encapsulation of Flip Chip Array Packages There were some voids noted in the package in the moldcap area; the elimination of these voids is a subject of future mold process and materials development. However, both materials are noted to have uniformly underfilled the area beneath the die, demonstrating the feasibility of this concept, although the substrates were mechanically vented. Collaboration with mold chase manufacturers is currently in progress to develop a molding process that would not require a mechanical vent in the substrate. The molded parts were post-cured four hours at 175 C in a stacked configuration with load of approximately Kg on the stack. After substrate singulation, the parts were evaluated by CSAM, then subjected to JEDEC Level 3 preconditioning (30 C, 60%RH, 168 hrs) followed by either a 0 C or 40 C reflow. After preconditioning, the parts were inspected by CSAM. Popcorn failures, which indicate macroscopic delamination, can be identified by black circles on the CSAM images, as in Figure 8. Failure rates for all four package populations are summarized in Table 4. The performance of the -13A material in Package I at both temperatures and in Package II at the 0 C reflow are statistically equivalent. In Package II at 40 C reflow, the failure rate of 13A becomes substantial. The 19A material exhibits no failures in either package type at either reflow temperature. The flatness (warpage) of each population was sampled after preconditioning. The deformation was measured optically by focusing on BGA pads at the corners and one in the center of the BGA matrix. A regression plane was fit to the five data points and the maximum deviation above and below the best fit plane was summed to get the flatness value. The statistical variation among the populations is shown graphically in Figure 9. This warpage data show no difference between the materials for the type I pack- age. However, for the larger package, the coplanarity decreases with respect to the smaller package. The large package with material -13A shows the maximum warpage. Figure 9. Comparison of the flatness measurements for the four material/package combinations. The Tukey-Kramer circles indicate that populations II/13A and II/19A are significantly different from every other population, including each other. However, the Package I populations have the same flatness. 4. Concluding Remarks Figure 8. CSAM images of Package II after JEDEC Level 3 exposure plus 40 C reflow for the A(-13A and B) -19A materials. Note the popcorn failure (large black circle) in image (A). Table 4. Failure levels of Amkor test devices after MRTL3. The technical feasibility of the use of a solid epoxy molding compound to fully encapsulate a Flip Chip device has been demonstrated. Uniform underfill was achieved with accompanying good JEDEC performance. The optimum resin and filler particle compositions have been identified which provide the excellent performance of these molding compounds. Technical challenges remain in the area of suitable mold design which will require the close interaction between mold compound supplier, mold designer, and the component manufacturer. Dexter Package Reflow Popcorn Material Temperature Failures ( C) -13A I 0 0/6-13A I 40 1/8-13A II 0 1/4-13A II 40 5/ -19A I 0 0/8-19A I 40 0/8-19A II 0 0/19-19A II 40 0/18 Acknowledgments The authors would like to acknowledge the assistance of Henk Peters, Walter de Munnik, and Wilfred Gal at FICO in molding and mold design. The authors would also like to acknowledge the efforts of technicians at Dexter Electronic Materials who made this work possible; John Barrett, Pete Parker, and Bill Cochran. Useful discussions with Dr. Do Ik Lee of Dow Chemical are gratefully acknowledged. Finally, the authors appreciate the support of the management teams at Dexter Electronic Materials and Amkor Technology in these studies. 405

7 Intl. Journal of Microcircuits and Electronic Packaging References About the authors 1. R. R. Tummala, E. J. Rymaszewski, and A. Klofenstein, Microelectronics Packaging Handbook, Chapman & Hall, M. Todd, K. Desai, and L. Hoang, Evaluation of Key Underfill Formulation Parameters on the Performance of Flip-Chip Devices, Proceedings of the Pan Pacific Microelectronic Conference, Kaanapali, Hawaii, pp. -6, S. H. Shi and C. P Wong, Recent Advances in the Devlopment of No-Flow Underfill Encapsulants A Practical Approach towards the Actual Manufacturing Application, Proceedings of the 49th IEEE Electronic Components and Technology Conference, ECTC 99, San Diego, California, pp , T. R. Miles, L. P. Rector, S. Gong, and T. LoBianco, Transfer Molding Encapsulation of Flip Chip Array Packages : Technical Developments in Material Design, Semicon West 000, San Jose, California, pp. E1-E6, K. Gilleo, B. Cotterman, and I. A. Chen, Molded Underfill for Flip Chip in Package, HDI, pp. 8-31, June Y. Lin, K. Chai, T. D. Her, and R. Lo, Transfer Molded Underfill for FC-BGA, Proceedings of Semicon Taiwan, pp , September, K. Kuwata, K. Iko, and H. Tabata, Low-Stress Resin Encapsulants for Semiconductor Devices, IEEE Transactions on Components, Hybrids, and Manufacturing Technology, Vol. 8, No. 4, pp , L. Rector, S. Gong, T. Miles, and J. Zhang, Themoset Encapsulant Performance Requirements for BGA 1C Applications, Proceedings of the 1999 Workshop on Polymeric Materials for Microelectronics and Photonics Applications, Paris, France, EEP, Vol. 7, pp , American Society of Mechanical Engineers Press, D.H.Everett, Basic Principles of Colloid Science, pg. 115, Royal Society of Chemistry, M. Mooney, The Viscosity of a Concentrated Suspension of Spherical Particles, J. Colloid Science, Vol. 6, pp , D. I. Lee, Packing of Spheres and its Effect on the Viscosity of Suspensions, Journal of Paint Technology, Vol. 4, pp , R. K. McGeary, Mechanical Packing of Spherical Particles, Journal of American Ceramic Society, Vol. 44, No. 10, pp , A. Hale, M. Garcia, and C. W. Macosko, Computer Simulation of the Spiral Flow of a Commercial Epoxy Molding Compound, SPE RETEC, pp , Louis Rector is a Research Associate in the Microelectronic Molding Powders Department at Dexter Electronic Materials. He is team leader for molded Flip Chip program and also assesses new materials technologies for Dexter s advanced packaging efforts. He received B.S. Degrees in Chemistry and Chemical Engineering from the University of Illinois at Champaign-Urbana, and he has a M.S. Degree in Manufacturing Engineering, and a Ph.D. Degree in Materials Science, both from Northwestern University. Shaoqin Gong is a Senior Materials Technologist in the Microelectronic Molding Powders Department at Dexter Electronic Materials Division. She received a B. S. Degree in Materials Science and Engineering, a B.S. Degree in Economics and Management, and a Master s Degree in Materials Science and Engineering all from Tsinghua University in China. She also received a Ph.D. Degree in Materials Science and Engineering from The University of Michigan at Ann Arbor. Since joining Dexter, she has been developing molding compounds for advanced BGA and molded Flip Chip applications. Tara Miles is the Product Manager for the electronic formulated liquid and optoelectronic product lines for Dexter Electronic Materials. Prior to assuming this role, she has worked for Dexter as a formulating chemist in the Research and Developement group for eleven years. She has published numerous papers in the electronics industry, focusing on adhesion science and the effects of delamination on device reliability. She received her Bachelor s Degree in Chemistry from Alfred University and a Masters in Business Administration from St. Bonaventure University. Kevin Gaffney is the Sr. Process Development Engineer for Flip Chip CSP products at Amkor Technology s Advanced Product Development Center in Chandler, Arizona. His focus is prototyping Flip Chip assembly processes for a variety of new products including stacked die and System in Package applications. He is also responsible for the introduction of new assembly processes, equipment, and materials for CSP assembly. He received his B.S. Degree in Ceramic Engineering from the University of Illinois at Urbana-Champaign, and his M.S. Degree in Material Science from the University of Illinois at Chicago. 406