Low Cycle Thermal Fatigue of Ball Grid Array Components

Transcription

1 Abstract Debbie Detch William Villers tentechllc.com Corporate Pointe, Suite 75 Culver City, CA (44) Low Cycle Thermal Fatigue of Ball Grid Array Components Non- linear finite element analysis employing I- deas model solution and ABAQUS is used to evaluate low cycle thermal fatigue in ball grid array (BGA) electronics packages. Prior analyses have investigated manufacturing residual warping and stress, as failure may occur during cooling from the first solder flow or later reduced temperature flows. This analysis incorporates creep and manufacturing residual stress with low cycle fatigue in order to estimate mean cycles to failure of the solder BGA. Key words: BGA, low cycle, fatigue, thermal Introduction BGA Description Ball grid array (BGA) or flip chip technology has become more common for interconnection of IC package as routing density requirements increase. A BGA multi- layer substrate is typically a ceramic with drastically lower CT than the PWB to which it is attached. This CT difference can lead to manufacturing issues and failures including warpage, substrate cracking and interconnect solder or pad failure. Predictions Lifetime predictions are necessary for many BGA applications; however, experimental and theoretical predictions for low cycle thermal fatigue in BGA devices are not strongly developed. ven when theoretical hand calculations may be performed, this does not take into account creep, relaxation and hysteresis that may occur in both lead and non- lead solders. This Analysis In this analysis, a method of predicting cycles to failure for low- cycle thermal fatigue is introduced. Creep, hardening, temperature- dependence and geometric factors are included, although inter- metallic layers, reflow coarsening and micro- cracking are not taken into account. This paper investigates a consumer BGA attached to a GR4/epoxy PWB with 6Sn- 7Pb eutectic solder. The BGA is BT (Barium Titanate) and package data has been utilized from a manufacturer s website. The objective of this analysis is to develop a simulation method that incorporates creep hysteresis applicable to a BGA and which may be extrapolated to other components. This study focuses on eutectic SnPb solder BGA components, although this method may be applied to any solder for which creep and temperature dependent material properties are available. This analysis is focuses on three aspects of the analysis: finite element model development, comparison with other studies with any available data and solder joint reliability prediction. BGA Selected for Analysis Development of a representative ball shape for the finite element model was done through review of research literature including SM images of BGA solder ball cross sections. For the most part, the profile is spherical as shown in Figure 1. () (1) Figure 1 Representative solder ball shape for SnPb and non- lead BGAs (,1) The general shape of the solder ball may then be defined by the height and width specified by the BGA manufacturer. Some previous studies have utilized cylinders with slight center bulges; however, this is not fully representative. Provided with the ball width and height from the BGA manufacturer, a formulation for the radius as a function of height from the neutral mid- plane can be derived with basic trigonometry and is provided in quation 1. The shape produced using the Altera

2 dimensions provided in Table 1 is shown in Figure with parameters indicated. H r( h) = ( D H) + h Where: D = maximum diameter (mm) H = solder ball height (mm) h = height used in equation from H/ to H/ (mm) r(h) = solder ball radius as a function of height (mm) (eqn 1) H Figure Generalized solder ball dimensions accompanying quation 1. The package is a 4- Pin FineLine Ball- Grid Array from Altera Corporation. (Data sheet D r(h) h Figure 5 Altera BGA package used for model dimensions, top view (5) Table 1 Dimensions corresponding to Figure, Figure 4 and Figure 5 (5). Dimension Millimeters Minimum Nominal Maximum A A A A 0.70 RF D BSC BSC b e 1.00 BSC Table Dimensions corresponding to Figure, Figure 4 and Figure 5 (5). Description Substrate Material Solder Ball Composition Maximum Lead Coplanarity Weight Specification BT 6Sn:7Pb in (0.0 mm).0 g The Altera BGA substrate is a BaTiO (BT) ceramic with a CT of 17 x Figure Altera fineline BGA package used for model dimensions, bottom view (5) Figure 4 Altera fineline BGA package used for model dimensions, side view (5) Table BT (BaTiO ) Properties Property Value Units Density (6) 6.0 g/cm lastic Modulus (6) 67 GPa Hardness (6) 5 Mohs CT (5) 17 ppm Table 4 FR4 PWB Properties Property Value Units Density (6) 6.0 g/cm lastic Modulus (6) 67 GPa TN TCH LLC ngineering Services

3 Hardness (6) 5 Mohs CT (5) 17 ppm Determination of an appropriate profile for the finite element solder interconnect was done using The finite element model is composed primarily of linear solid elements (quadx, quadx, etc.) and beam elements. The solid elements are utilized for the PWB, BGA housing and BGA finite element model contains a combination of solid meshed balls at the perimeter, where failure has occurred first in numerous papers, and density and thus the outermost solder balls will have detailed solid mesh. I- deas Model Solution will run non- linear solutions with beam elements, although the inner balls will be meshed with a beam configuration that is representative of the solder balls. The beam mesh may be seen in Figure 8 with solid round beams. Compared to the actual shape seen in Figure, the 10 element model is a reasonable representation of the actual geometry. The detailed mesh may also be found in Figure 8. Finite lement Model Development The - D finite element model should be detailed enough in critical regions to capture the behavior of the real geometry, without so many elements and nodes that run times and result files become prohibitive. Determining which methods for model size reduction will be used prior to beginning meshing will reduce meshing time. For this model, a combination of model size reduction techniques was employed. The BGA selected for this analysis is a 16x16 array, with all solder balls positions filled. The use of symmetric mirrored boundary conditions in the - D finite element model will quarter the total number of elements. This scheme is indicated in Figure 6 with the location of the vertical cutting planes indicated by arrows. Figure 8 Beam mesh of solder ball composed of 10 beams of varying cross section (L) and solid mesh composed of 960 solid elements (R). The outer 15 solder balls have detailed solid mesh composed of 960 solid elements and 1019 nodes each, while the inner 49 balls will be represented by the beam models. That is a total 490 elements for the interior solder balls. PWB BGA Arrows indicate planes for mirrored symmetric boundary conditions Figure 9 Solid mesh of the detailed and beam model solder balls shown with BGA component mesh The complete model is composed of Figure 6 Model with planes for symmetric mirrored boundary conditions indicated Material Properties for Simulation The eutectic 6Sn- 7Pb properties provided below were obtained through the National Institute of Standards & Technology website (NIST) (11), although attempts are made to site the NIST references directly as websites move and change. The original reference papers were not obtained in most cases. Figure 7 Quarter section of PWB (green) and BGA(gray) used for analysis Additionally, it is recognized that the exterior solder balls will experience the greatest stress and strain energy Representative temperature- dependent properties with some conservatism built in are desired for this analysis. The density, ρ, of the solder is shown in quation 1 below. This value is used for all temperatures (16). TN TCH LLC ngineering Services

4 g ρ = 8.4 cm (eqn 1) Figure 10 Normalized creep rate plot for 6Sn - 7Pb eutectic solder (11) The CT, α, has been indicated from 1 to 5 ppm/k (15,16,0) however the selection of 4 ppm/k has been made as the data was provided over a range of temperatures (14). α 6 6Sn 7Pb 4 10 = K 1 The Young s Modulus,, is temperature dependent with the relation from Knecht et al. shown in quation a below. This equation has been modified from its original form in order to represent temperature in K and Young s Modulus in Pascal as shown in quation b (18). ( T) = 0. 88T (Temp C, GPa) ( T) = ( 0.088( T 7.15)) 10 9 Pa The Poisson s Ratio, ν, has been indicated anywhere from 0.5 to 0.4 (0,14,15). The following value in quation 4 is used for all temperatures and was the most commonly referenced in reviewed literature. υ = 0.4 (unitless) (eqn ) (eqn a) (eqn b) (eqn 4) There is a recognized difference between creep in shear and tension loading (9,10), however that differentiation is not being made in this study. The I- deas Model Solution uses the Von Mises stress for creep in a state of uniaxial stress, with dependence on the effective stress (Von Mises), temperature, and time. A relationship for tensile creep is provided in Figure 10. The strain rate is normalized using the temperature- dependent diffusion parameter and the stress is normalized for the temperature- dependent Young s Modulus as supplied in quation above. From the study by Wong et al. (1), the average creep rates may be given by quation 4 below, utilizing the temperature- dependent diffusion parameter in quation 5. quation 5 was derived from data from several independent sources over a range of temperatures (- 60 C to 150 C), stresses and creep rates (11). ε SS D = 7 1 σ 4 σ Where the relation for D =D(T) is provided in quation 6 with an activation energy of 45kJ/mol/K or 0.47eV. D( T) Q = C 547 exp = exp RT T There are two distinct slopes present in Figure 10, reflecting different creep mechanisms. The dislocation glide and climb regimes belong to the low and high stress regions respectively. The lower and upper bounds for the correlation provided in Figure 10 are shown in quations 7 and 8 below from Clech et al. (19). ε SS D ε SS D = = 7 11 σ σ σ 5 σ This analysis is utilizing the assumption that the temperature ranges are extreme and that temperature transitions are relatively slow, such as in satellite electronics, causing lower temperature cycles to dominate, with lower strain rates. Thus making this even more severe, the minimum strain rate will be used. Currently shows Wong middle correlation in eqn 9. (eqn 5) (eqn 6) (eqn 7) (eqn 8) TN TCH LLC ngineering Services 4

5 Figure 11 Plot of quations 5, 7 & 8 Model solution Creep quation (17) was utilized in the form shown in quation 9. ε C + C C = C C C ( C σ ) ( C t) C 1 C exp T C T C7 C ( C σ ) ( C t) exp The constants have the following dimensions: C,C 4,C 7,C 9 dimensionless C 1,C 6 length / force C,C 8 1/time C 5,C 10 - temperature The following values have been assigned to the quation 8 constants in order to match quation 4. C, C 1 6 C = C = C 1 m = ( T) N , C9 = 5, C10 = 1 1 C 547 C = 7 7 C = (sec ( K) 4 (sec 1 1 ) ) Wong et al. (1) With (T) defined as in quation, the resulting expression is provided in quation 10, where the unknown time dependence is removed through use of a zero exponent. 4 Upper and lower curve from XXX et al (eqn 9) ε C = σ ( T ) 1 1 σ ( T ) 7 (*) 0 exp 547 T 547 T 0 (*) exp * The term in parentheses is unimportant as the 0 exponent renders the value of 1 for the term A number of successful non- linear electronic analyses have been performed utilizing the Garofalo implicit creep equation as provided in quation 11 (0). This equation has not been utilized in these analyses, however, because it applies only to secondary creep and the constants obtained and presented in Table 5 are of unknown accuracy. The quation 10 format utilized in the current analysis is accurate for primary and secondary creep. If one wishes to use the following data with I- deas Model Solution, it may be possible to use the quation 6 format. Table 5 Garofalo constants from Qi et al. for eutectic Sn- Pb solder for use in quation 11 (0). 18 (K) 98 (K) C C C.. C The constants provided in Table 5 are for use in quation 11, an ANSYS specific input, which was obtained from the ANSYS documentation website. ε cr C = C1 exp T C [ ( C )] 4 sinh σ The I- deas Model solution form of the equation is provided in quation 1, and it is assumed some algebraic kung fu would result in usable constants from the above information. ε C C = C 1 exp σ exp C T (eqn 10) (eqn 11) (eqn 1) TN TCH LLC ngineering Services 5

6 Fatigue Failure Prediction Methods Many fatigue failure indicators have been utilized in predicting fatigue damage. A list is provided in Table 6 More common fatigue failure prediction indicators Total Strain Plastic Shear Strain Indicator Creep Strain Range per Cycle Strain nergy Density Dissipated Strain nergy Density per Cycle Reference A second method uses more recent fatigue data () as shown in Figure 1, with a plot indicating the cycles to failure divided by the crack area (cycles/in ) vs. cyclic shear strain range. The incorporation of the solder joint cross sectional shear area in the equation is meant to provide some scaling for size and the y- axis term may be interpreted as the inverse of an area crack propagation rate (). For this instance where the minimum ball diameter is 0.0mm: A shear = π r = π 8 ( m) = m Fatigue Failure Prediction The fatigue failure prediction in this analysis is performed utilizing several methods with different failure data and these methods are them compared. The first method uses the FM Von Mises stress with the Steinberg S- N log- log curve (8) which is presented in Figure in SI units. The cycles to failure may be read directly off the plot or can be input in quation 1 to estimate the number of cycles to failure with relative ease Vibration b = 4.0 Figure 1 Cycles /crack area vs. cyclic shear strain for SnPb eutectic solder for thermal cycling converted to SI units () Thermal b = The last data included in Figure 14 and Figure 15 are based on isothermal mechanical fatigue and will be included for comparison to the thermal fatigue data. This is being considered in the event that thermal fatigue data is not available for the chosen solder. Figure 14 indicates the relation of inelastic strain energy and cycles to failure/crack area. Figure 15 indicates the relation for strain energy density to cycles to failure. Figure 1 S- N log- log vibration and thermal cycling fatigue curve for 6Sn- 7Pb solder, Figure. from Steinberg converted to SI (8) N 1 = N σ σ 1 b (eqn 1) TN TCH LLC ngineering Services 6

7 The total number of cycles starting at 0 C was XX, with a cycle defined as indicated in Figure 17. This accounts for a full stress reversal. Figure 14 SRS correlation and updated correlation median line (Clech, 1996 & 000) also corresponding to data by Darveaux (NIST). Figure 16 Thermal cycle profile for simulation Figure 15 Strain energy density vs. cycles to failure for 6Sn- 7Pb solder under structural loading (X) NIST Thermal Cycle Profile The thermal cycle profile for the simulation is provided in Figure 16. The assumption that the solder solidifies at the liquidus temperature XX C and is then at a state of zero stress/strain and is then cooled down to room temperature is made Figure 17 Definition of thermal cycle used in simulation TN TCH LLC ngineering Services 7

8 Results The finite element model results including maximum stress, strain, strain energy and XX after X cycles is provided in Table 7 Table 7 Values used for fatigue life prediction Parameter Value Units Stress Pascal Strain Unitless Strain nergy J Strain nergy Density J/m Figure 18 Solder ball Von Mises stress Figure 19 Solder ball strain Figure 0 Solder ball strain energy Figure 1 Solder ball strain energy density Table 8 Values used for fatigue life prediction Parameter Cycles % Difference Stress Reference Strain Strain nergy Strain nergy Density Conclusion The low cycle thermal fatigue for BGA components may be modeled with relative accuracy. TN TCH LLC ngineering Services 8

9 References Flexural strength of BGA solder Joints with NIG substrate finish using 4- point Bend Test. Bansal, Yoon, and Mahadev, SMTA Pan Pacific Microelectronics Symposium, Kauai, Hi, Jan. 5-7, 005. (altera- panpac- 005.pdf) ecifications/pkg- pin/spe- index.html Ag- Cu_Fit.htm 8. Dave S. Steinberg. Vibration Analysis for lectronic quipment, rd d. John Wiley and Sons, 000. Canada. 9. Darveaux, R., Banerji, K., Mawer, A. and Dody, G., Reliability of plastic ball grid array assemblies, Chap. 1, Ball Grid Array Technology, ed. J. H. Lau, McGraw- Hill, 1995, pp Wiese, S., Schubert, A., Walter, H., Dudek, R., Feustel, F., Meusel,. and Michel, B., Constitutive behavior of lead- free solders vs. lead- containing solders xperiments on bulk specimens and flip- chip joints, Proceedings, 51 st lectronic Components and Technology Conference, Orlando, Fl, 001, pp Pb_Creep.htm#Creep 1. Wong, B., Helling, D.. and Clark, R.W., A creep rupture model for two- phase eutectic solders, I Transactions on Components, Hybrids and Manufacturing Technology, Vol. II, No., Sept. 1988, pp Knecht et al Tim Wong and A.H. Matsunaga, Ceramic Ball Grid Array Solder Joint Thermal Fatigue Life nhancement, Proceedings: NPCON West Conference (February 8 March, 1995), Anaheim, CA. 15. J. Lau, C. Chang, R. Lee, T.- Y. Chen, D. Cheng, T.J. Tseng, D. Lin, Thermal- Fatigue Life of Solder Bumped Flip Chip on Micro Via- In- Pad (VIP) Low Cost Substrates D= I- deas Creep quation Documentation 18. Knecht et al Clech et al Qi et al., Simulation Model Development for Solder Joint Reliability for High Performance FBGA Assemblies, 0 th I SMI- THRM Symposium 1. Henshall et al., Manufacturability and Reliability Impacts of Alternate Pb- Free BGA Ball Alloys, Hewlett Packard Development June Clech, Jean- Paul, Lead- Free and Mixed Solder Joint Reliability Trends, Proceedings of the IPC Printed Circuits xpo, APX and Designers Summit 004, February 4-6, Anaheim, CA.. Dongkai Shangguan. Lead- Free Solder Interconnect Reliability. lectronic Device Failure Analysis Society. pp TN TCH LLC ngineering Services 9

10 900 Vibration b = 4.0 Thermal b = Figure BGA for XXX (9) really (1) 9 Figure 5 S- N log- log vibration and thermal cycling fatigue curve for 6Sn- 7Pb solder, Figure. from Steinberg (8) fab.com/documents.asp?grid=17&d_id=1176 Figure BGA for XXX () Package specifications 780- FBGAoption ecifications/pkg- pin/spe- index.html Initially, ceramic BGA packages are attached to PWBs at temperatures slightly higher than solder flow temperature. As the assembly returns to room temperature, the PWB reduces in size much at a much faster rate than the substrate ceramic and there is a residual shear in the solder. Many techniques are utilized to reduce stress in the ball grid arrays including underfill with epoxy materials. Figure 4 XXXX (4) TN TCH LLC ngineering Services 10