Characterization and modeling of molding compound properties during cure

K.M.B. Jansen1, C. Qian1, L.J. Ernst1, C. Bohm2, A. Kessler2, H. Preu2, M. Stecher2 1 Delft University of Technology, Mekelweg 2, 2628 CD Delft, The Netherlands 2 Infineon Technologies AG (AIM AP), 81726 Munich, Germany [email protected] application conditions is only possible if one understands the relevant material properties. The present work is part of a larger project focussing on the characterization and modeling of molding compound material properties. In previous work we reported the cure kinetics of a series of molding compounds [1] as well as the bulk modulus and cure shrinkage [2,3]. In this study we characterize the viscoelastic mechanical properties and use the previously determined cure kinetics to develop a full, cure dependent material model. In this paper we will first discuss the establishing of the limiting glassy moduli and the fully cured viscoelastic behaviour. In the second part we focus on the cure dependent modulus behaviour. 2. Experimental The materials used are commercially available molding compounds used for chip encapsulation. The materials are reported to have filler percentages of 82 to 90 wt% and glass transitions ranging from 100 to 140 C (exact values depend on the definition used as well as on the experimental details). We will refer to the molding compounds as MC1, MC2 and MC3 throughout this paper. All mechanical tests were performed on a TA-Instruments Q800 Dynamic Mechanical Analyzer (DMA). The viscoelastic properties of the fully cured materials were performed in a tensile setup applying continuous frequency sweeps with frequencies from 0.32 to 60 Hz (log spacing) in combination with heating at 1 C/min. The sample was kept thin to minimize frame stiffness effects. The cure dependent mechanical properties were determined in a special sandwichbeam tool (see section 4 for details) mounted in the Q800 DMA. 3. Fully cured material behaviour The mechanical properties for the fully cured materials should serve as a limiting case of the cure dependent models to be derived in section 4. It is therefore useful to pay some attention in deriving a consistent set of fully cured mechanical material properties. This is not as simple as it seems since molding compounds are extremely stiff which complicates the exact determination of modulus values. A molding compound strip of 30x4x0.5 mm for example has a stiffness E*w*b/L = 2 106 N/m which is about as stiff as the mechanical frame of most tensile testers. The apparent measured modulus will therefore be half of the true value. If then these

Abstract During the encapsulation of electronic components stresses are generated due to curing effects and the difference in thermal shrinkage between molding compound and die. These residual stresses add up to the stresses generated during thermal cycling and mechanical loading and may eventually lead to product failure. In this paper we focus on three commercial molding compounds and and analyze in detail the increase in elastic modulus and the change in viscoelastic behaviour during cure. This was done with a special shear tool which allows to measure mechanical properties with sufficient accuracy in the liquid as well as in the solid state. The cure dependent viscoelastic material behaviour was modeled using a cure dependent shift factor and rubber modulus. The viscoelastic behaviour of the molding compounds is also shown not to be stable. During postcure the materials slowly continue to crosslink thereby systematically changing their viscoelastic behaviour. The material models presented here therefore only account for the initial curing stage and do not include postcure. 1. Introduction Integrated circuits are protected against shocks, vibrations and environmental conditions like moisture by encapsulating them with an epoxy resin, the so-called molding compound. During this encapsulation process the molding compound pellets are melted and injected into a preheated mold which contains the die(s) and leadframe. After the injection the epoxy starts to crosslink (cure) and eventually forms a rubbery substance which vitrifies upon cooling. The problem is that both during the curing and the cooling stage residual stresses are generated which may lead to failure during subsequent thermal cycling conditions or mechanical loading. Undesired package failure can in principle be avoided by using numerical simulations to predict these package stresses and then optimize the package design. The issue is, however, that for reliable predictions highly accurate material data is needed. For the inorganic materials like the die and leadframe this is not a real problem but for the molding compound it is. In the present paper we will discuss a new technique to measure the change in viscoelastic properties of three commercial molding compounds and propose a material model which can be implemented in simulation software. Material selection for future products and transfer of results from accelerated testing to product assembly and

978-1-4244-2128-2/08/$25.00 C 2008 IEEE

9th. Int. Conf. on Thermal, Mechanical and Multiphysics Simulation and Experiments in Micro-Electronics and Micro-Systems, EuroSimE 2008

values are used to calculate the bulk modulus or Poisson ratio, significant errors may occur (e.g. leading to negative Poisson ratio values). For isotropic elastic materials the following relations exist between bulk modulus (K), shear modulus (G), elongation modulus (E) and poisson ratio (): 1 E 3KE , . G= = Eq.1 9K E 2 6K For each of these materials we determined the glassy tensile modulus by performing a series of tensile tests on strips of different thickness. A plot of the measured compliance versus the sample geometry factor then yields both the frame stiffness and the true glassy tensile modulus. Table 1 lists the glassy modulus values for the three materials together with the glassy bulk modulus values [3] and the calculated shear modulus and Poisson ratio (Eq.1). The shear modulus is about 11 GPa for all materials whereas the Poisson ratio varies between 0.18 and 0.27, which can be considered as realistic values since they are in in between the values for pure silica (0.16) and pure epoxy (0.4). Compound Eg Kg g Gg [GPa] [GPa] [GPa] 282 192 0.25 11.1 MC1 252 232 0.18 10.6 MC2 272 192 0.27 10.7 MC3 Table 1: Glassy modulus values The next step was to determine the viscoelastic curves of the fully cured materials. This was first done after a standard postcure of 4 hours at 200 C. The corresponding tan curves (not included) showed a second peak above 200 C which indicates additional cure. We therefore also studied different postcure histories and found remarkable differences in material behaviour. The glass tranition shifts upwards by 20-30 C (Fig.1) and the rubber modulus increases with increasing postcuring. For postcuring at 300 C the glass transition temperature dropped again, indicating degradation (not shown in Fig.1). Similar effects were found for the other materials and the conclusion thus must be that molding compounds are not stable, not even after postcuring.
1E+05 5 min at 250C 4 hour at 200 C 4 hour at 250 C 1E+03

This observation can be an explanation for the apparent inconsistency of reported glass transition temperatures and viscoelastic data as measured in different laboratoria. It also means that our idea of using the fully cured viscoelastic curve as a limiting case and reference for the cure dependent material modeling must be abandoned. More details about the postcure effects and the actual viscoelastic curves will be presented in a separate study. 4. Mechanical properties during initial cure stage The change in mechanical properties during cure was studied in the shear tool depicted in Fig.2. It consists essentially of two thin steel leaf springs in between which the molding compound is injected. In that way the full range from fluid to high stiffness solid could be analyzed. Details can be found in [4]. The material injection process was done at 140 C in a specially designed brass mould, which could be cooled down rapidly directly after injection to prevent cure during the sample preparation step. The leaf springs (0.1 mm) were separated by two steel spacers at the end points and the inner surface of the leaf spring package was rinsed with diluted sulphuric acid to enhance adhesion. The ends of the package were then clamped in the single cantilever setup of the Q800 DMA (Fig.2) and an isothermal cure test was started.

Fig.2: Leaf spring package clamped in DMA An isothermal cure experiment consists of heating at 10 C/min to the desired cure temperature, curing for 400 minutes, cooling to 40 C, stabilizing for 60 minutes followed by a rescan to 280 C at 1 C/min. During the whole experiment frequency sweeps between 1 and 60 Hz were applied. A typical example is shown in Fig.3. During the first 5 minutes the material heats up resulting in a rapid stiffness drop. The minimum stiffness value is mainly determined by the leaf springs. After about 10-20 minutes the material starts to gel and the sample stiffness rapidly increases. The spreading of the different frequency curves in Fig.3 indicates that there are severe viscoelastic effects during the isothermal cure stage. The reheat scan from below to above Tg is used to determined the glassy and rubbery modulus values as well as the glass transition temperature. This Tg is a measure for the actual cure state after isocuring. The stiffness increase above 250 C is due to additional cure (see above). For each material 4 or 5 tests with curing temperatures between 80 and 140 C were performed to generate enough data.

E' [MPa]

1E+04

1E+02 50 100 150 200 250

T [C]

Fig.1: Effect of postcure on 1 Hz curve for MC3

2
9th. Int. Conf. on Thermal, Mechanical and Multiphysics Simulation and Experiments in Micro-Electronics and Micro-Systems, EuroSimE 2008

The data analysis consists of 3 steps. Firstly the measured stiffness curves were converted to shear modulus values (see [4] for details). In that way the stiffness of the leaf spring tool is corrected for. The other tool stiffness parameters were determined such that the glassy shear modulus was equal to the value in table 1. This was done to enforce consistency between the different tests. Note that these corrections only stretch the isocure curves in vertical direction and therefore have no effect on the desired cure shift factors (shifting on horizontal time scale). The second step of the analysis process consists of replacing the time scale during isothermal cure by the conversion level. This could be done quite easily since the cure kinetics were determined in separate experiments before [1]. Based on the conversion level the rubbery modulus was estimated using the Martin and Adolf scaling law:
2 2 gel , Eq.2 G r ( ) = G rfinal 2 1 where and gel denote the conversion level at time t and at the gel point, respectively. Notice that this previously determined kinetic model does note yet include the slow postcuring effects mentioned above. In the third step, for each material the data of all isothermal curing experiments and the calculated conversions were added in a single data array and fitted to the model listed below. The fitting was done in MatLab using a non-linear Levenberg-Marquardt least-squares procedure which resulted in a set of fit parameters and their 95% confidence interval values. 8/3

C (T Tr ) /(C 2 + T Tr ) T > Tc log aT = 1 Eq.4 A1 (T Tc ) + A2 T Tc Parameter A2 simply follows by demanding that the two parts coincide at T=Tc. The cure dependent part is a function which only depends on the conversion which should vanish at full cure (=1). It should decrease with increasing since conversion (crosslinking) slows down all viscoelastic effects. Ideas for the shape of this curve were obtained by performing a cure-frequency shift to the experimental data similar to the well-known temperaturefrequency shifts for fully cured materials. The chosen shift factor function is log a = d1 (1 exp[ d 2 (1 )]) Eq.5
This curve had the advantage above a second degree polynomial function that it ensures monotonic behaviour. The shape of the modulus mastercurve was approximated with a Havriliak-Negami (HN) type of equation log G g log G r log G ( , T , ) = log G r + Eq.6 n 1 + ( 0 / aT , ) m

where 0, m and n are fit parameters associated with the typical relaxation time at the reference temperature and the shape of the viscoelastic region, respectively. This type of function is known to be a good approximation for the shape of fully cured thermosets. Apart from the (known) glassy and rubbery plateau values it has only three additional fit parameters which simplifies the curve fitting procedure. C1 13.32 0 1.86E-5 C2 85.64 m 0.212 A1 0.105 n 1.245 Tc 101.8 d1 27.36

1 0

v is co e la s tic
1 0 s t if f n e s s K '
4

f u lly cu r ed

Table 2: Paramters of MC3 for Eqs.4 to 6. The model parameters for MC3 are summarized in Table 2. The other matrials yilded similar sets of parameters. For all materials a gel conversion of 0.40 was assumed. A comparison of the modulus rise of the model fits and the experimental data showed that this was a reasonable estimate. Examples of experimental data and model fits for material MC3 are shown in figure 4, 5 and 6. Note that only three of the 5 isocure experiments are shown while all five are included in the fit procedure. The upper and lower full green lines indicate the upper (glassy) and lower (rubbery modulus) limits. The other full lines are the experimental stiffness curves for different frequencies whereas the dashed lines are the model predictions. It can be seen that the models captures quite well the initial modulus (or and stiffness) decrease due to melting, the modulus increase after gelation as well as the final cooling and reheating steps (600-800 minutes in Fig.4). Postcuring effects like the modulus increase after 750-800 minutes (Fig.4, 5) and the slow increase after 400 minutes (Fig.6) are not captured by the model. Also note that the present model captures the modulus changes during

1 0

is o cu re

c oo l
400 t im e [ m in ] 500 600

re h ea t
130 7 00 800

1 0

100

200

3 00

Fig.3: Stiffness increase during isothermal cure We modeled the cure dependent shift factor for the present materials as the sum of a temperature dependent and a cure dependent part: log aT , = log aT + log a , Eq.3 where log aT is the temperature shift factor of the fully cured material. Fortunately when Tg is taken as the reference temperature the shape of this curve did not vary much with the postcure conditions. It was modeled as a WLF function for higher temperatures (T>Tc) and as a log-linear curve below Tc:

3
9th. Int. Conf. on Thermal, Mechanical and Multiphysics Simulation and Experiments in Micro-Electronics and Micro-Systems, EuroSimE 2008

curing process for a wide range of temperatures involving cure temperatures below (Fig.4) as well as above (Fig.6) the final glass transition. In the latter case the material is close to the rubbery state for a large part of the curing process which can be observed from the converging of the lines for different frequencies. For materials MC1 nad MC2 similar observations were done.

splitting of the viscoelastic shift factor in a temperature and a conversion part. Additional experiments with different postcure conditions showed that the standard postcure temperatures of 175-200 C are not sufficient to completely finish the curing reaction. This means that after each high temperature exposure (above 250 C) the glass transition, viscoelastic behaviour and the associated residual stress field will change. These postcure effects are not included in the present model an deserve further study.

stiffness K'

10

Rubber modulus
10
2

80 0 100 200 300 400 500 time [min] 600 700 800

Fig.4: Data and model fit for material MC3 at 80C

glassy modulus

10 stiffness K'

Rubber modulus
10
3

10

100 0 100 200 300 400 time [min] 500 600 700 800

Fig.5: Data and model fit for material MC3 at 100C

References 1. K.M.B. Jansen, C. Qian, L.J. Ernst, C. Bohm, Kinetic characterization of molding compounds, EurosimE 2007 conference, April 16-18, 2007, London, UK, pp. 574-578 2. K.M.B. Jansen, M.D. Patel, L.J Ernst, J. Zhou, C. Bohm, PVT Measurements and Bulk Modulus Study of Epoxy Molding Compound, McMat 2007 conference, Texas, 2-6 June 2007 3. M.K. Saraswat, K.M.B.Jansen, M.D. Patel, L.J Ernst, C. Bohm, A Characterization Method for Viscoelastic Bulk Modulus of Molding Compounds, EurosimE 2008 conference, April 21-23, 2008, Freiburg, Germany 4. C. van't Hof, K.M.B. Jansen, G. Wisse, et al., Novel shear tools for viscoelastic characterization of packaging polymer, Microelectronics Reliability 47, p.240-247, 2007 5. D.B. Adolf, J.E. Martin et al., Stresses during thermoset cure, J. Materials Res. 13, 1998, p.530

10 stiffness K' 10

130 0 100 200 300 400 time [min] 500 600 700

Fig.6: Data and model fit for material MC3 at 130C 5. Conclusions The change in viscoelastic properties during cure was monitored using a specially constructed tool which encloses the unreacted materials between two flexible leaf springs. A full cure dependent viscoelastic model for the shear modulus was then developed and the model parameters were detrmined for three commercially available molding compounds. The basic characteristics of the model are a cure dependent rubbery modulus and a

4
9th. Int. Conf. on Thermal, Mechanical and Multiphysics Simulation and Experiments in Micro-Electronics and Micro-Systems, EuroSimE 2008