How manufacturing processes can affect warpage, and how compression molding can help.
By Eric Ouyang, Yonghyuk Jeong, JaeMyong Kim, JaePil Kim, OhYoung Kwon, and Michael Liu of JCET; and Susan Lin, Jenn An Wang, Anthony Yang, and Eric Yang of CoreTech System (Moldex3D).
Abstract
System-in-Package (SiP) technology has been used for a wide range of electronic devices, but the warpage behavior of the package can be difficult to control and predict due to complex manufacturing parameters and processes [1,2]. Previous research on the warpage primarily focused only on the SiP module unit, while the consideration of strip warpage as a function of manufacturing processes has not typically been studied theoretically and experimentally. In this paper, the impact of manufacturing processes, mainly the compression molding process, on the warpage is investigated experimentally and numerically. To better understand the advantages of compression molding, we will also compare compression molding with transfer molding using a computer simulation. The paper will point out the pros and cons of these two different manufacturing processes.
I. Introduction
In this paper, a SiP strip was manufactured with a compression molding process. An advanced material characterization method is conducted to study the curing reaction and the Pressure-Volume-Temperature-Cure (PVTC) kinetics of the mold compound. The curing reaction of epoxy resins, as a function of temperature and activation energies, is experimentally determined. During the curing process, the viscosity of epoxy resins change with temperature and conversion rate. The Castro-Macosko model is adopted to describe the rheological properties of epoxy resins.
Experimentally, we have prepared many substrate strip samples. Each strip comprises nine hundred and sixty eight SiP units. The manufacturing parameters of compression molding process, such as temperature, pressure, and process times, were carefully controlled and monitored in order to study the impact of warpage and voids. The warpage and voids were carefully measured using Shadow Moire and Scanning Acoustic Tomography (SAT).
In additional to the experiment, two different Finite Element Method (FEM) approaches were used to simulate the warpage of the strips. The strip warpage along various directions were measured and compared with the simulation. The first simulation approach uses traditional mechanical simulation to consider only the thermal stress, and the second simulation approach uses the PVTC simulation to consider the transient kinetic reaction and the viscoelastic behaviors of epoxy compound.
With the work described in this paper, we will provide a unique characterization and numerical scheme to better understand the warpage behaviors of strips using the compression molding process. The comparison of compression molding and transfer molding processes provides crucial insights for the development of better and more reliable next generation IC packages.
II. Preparation of SiP substrate
The SiP substrate strips, which have six metal stack-up layers, are prepared to accommodate nine hundred and sixty eight SiP units. Table 1 shows the thicknesses and materials of the strip layers, where SR stands for solder resist, M1 to M6 are metal layers, and PP stands for prepreg.
The percentage of copper material on each layer is calculated in order to study the warpage of the strip. The effective material properties of each strip layer, based on percentage of copper distribution, are considered to simplify the simulation procedures. The size of strip is about 240mm by 95mm with SiP units evenly populated on top. Each SiP has many passive and active components, such as capacitors, resistors, inductors, and flip chip dies. The effective material properties of passive components are again considered based on the volume percentages of components inside.
III. Process flow
The compressing molding of the SiP strip involves complex procedures, and Figure 1 illustrates the basic process flow. For this paper, our main reliability concerns are the warpage and voids of SiP strips, and we focused on “Molding”, “C-SAM”, and “Post Mold Cure”.
Figure 1. Process flow of compression molding
Figure 2 is a graphical representation of the implementation of compression molding, with one mold holding the substrate strip and other IC components, and the other mold holding the melting resin. With the two molds clamped and heated together, the mold compound is cured and bonded with the IC components and substrate strip to form a plurality of SiP modules. The temperatures of the mold compound and the steel mold were carefully controlled according to the recommended specifications for high volume production of SiP modules. Besides the temperature control, the compression speeds, or the moving distances of steel molds as a function of time, were designed to ensure the maximum uniformity and lowest shear stress on the mold compound. A proprietary multi-step compression process is implemented and Figure 3 is a schematic demonstration of the compression molding steps, where t is the time in seconds and d is the distance in mm. After compression molding, the final thickness of the mold compound samples is about 500um.
Figure 2. Schematic of compression molding
Figure 3. Compression molding processes
After the molding, our next step is to inspect the voids of strips. We have observed that the voids are either too small or they had not occurred at critical locations. Figure 4 is a SAT image of the SiP package [3]. The insignificant voids are obviously not a reliability concern, thus we shifted our focus to the warpage of the compression molded strip.
Fig. 4: Inspection of void with SAT
IV. Traditional mechanical simulation
To analyze the warpage behavior, our first approach is to implement a traditional FEM mechanical simulation. The traditional mechanical simulation here refers to the applying of a simple thermal loading of the entire strip components from a reference temperature to the final state of the device. With the thermal loading, the differences of material properties, such as CTE and Young’s modulus, will cause the deformation of the strip. The mold compound, which is the main component of the device, is assumed to be what dominates the warpage of strip. Therefore, the curing temperature of mold compound is selected as the reference temperature, or the stress-free temperature, and the final step of thermal loading is usually the room temperature. For the traditional mechanical simulation, kinetic behaviors of mold compound cannot be considered, and the warpage of the strip is affected only by the thermal stress.
Table II shows the material properties used for traditional mechanical simulation and Figure 5 is an example of the simulated warpage of a strip in the out-of-plane direction. In order to compare the simulated warpage data with experimental data, warpage magnitude as a function of distances along three different paths, namely width, length, and diagonal directions, as shown in the Figure 6, are defined.
Figure 5. Warpage from traditional mechanical simulation
Figure 6. Warpage of the strip along different directions
V. PVTC simulation
In additional to traditional mechanical simulation, we also implement a mold compound process flow simulation.
A. Governing Equations and Chemorheology
The 3D mold flow modeling tool, Moldex3D R17, was used to study the molding process. Theoretically, the microchip encapsulation process is a three-dimensional, transient, reactive problem with moving resin front. The non-isothermal resin flow in the mold cavity can be mathematically described by the following equations:
where μ is the velocity vector, T is the temperature, t is the time, p is the pressure, σ is the total stress tensor, ρ is the fluid density, k is the thermal conductivity, Cp is the specific heat, and Φ is the energy source. In this work, the energy source contains two contributions:
Where η is the viscosity, ϒ is the magnitude of the rate of deformation tensor, α is the conversion rate and ΔΗ is the exothermic heat of polymerization.
The epoxy specific volume is a function of pressure, temperature, and curing kinetics (PVTC). The molding simulation will provide the pressure and temperature, and the curing degree should be calculated by curing kinetics model. The advantage of the PVTC is that it can describe the overall specific volume change. The two domain modified Tait model is used here to describe PVTC:
where b1~b5 are model parameters.
The curing reaction of epoxy resins has received much attention using different analyses. In this work, a combined model was applied to investigate the curing kinetics of the given EMC because of its ability to accurately predict the experimental data. The combined model can be expressed as follows:
Where α is the conversion rate of the reaction, A1, A2, E1, E2, m, n are model parameters. During the curing process, the viscosity of epoxy resins changes with the temperature and conversion rate. The Castro-Macosko model was adopted to describe the rheological properties of epoxy resins:
where Α,Εa,C1,C2 are model parameters, αg denotes gelation conversion at which the viscosity curve grows upward because of the formation of three-dimensional network structure of the epoxy resins.
After the product was ejected from the mold, a free thermal and cure shrinkage happened due to the temperature and conversion rate difference. The mechanical properties are described in Table II. The equilibrium equation with representing the stress is expressed as follows:
And the relation between stress and strain,
where σ is the stress, C is a 4th tensor and function of relaxation modulus E(t,T,α), ε is the strain tensor and υ is the displacement vector, representatively. The stiffness tensor could be express as:
Thermal and cure induced strains can be expressed in the following:
αCLTE is CTE tensor, VS(P,V,T,C) can be calculated from the chemical volume shrinkage by measurement.
B. Visco-elastic constitutive behavior [4]
Assuming the constitutive behavior of the epoxy molding compound is linear visco-elastic, the visco-elastic properties of epoxy molding compound are defined in the form of Prony series with time-temperature shift factor. A Prony series (also known as Generalized Maxwell model) was used to fit the master relaxation curve of the shear and bulk modulus of the mold as follows:
Where G∞ and K∞ represents the equilibrium shear and bulk modulus after the molding compound has fully relaxed; Gi, Ki, and λi are the Prony coefficient and relaxation time for each element respectively. The shift factor is determined based on time-temperature superposition principle (TTS) which assumed that visco-elastic behavior of polymer materials are a reference temperature (T0) can be related to the temperature T by multiplying the time or frequency scale aT. Generally, there are two models to describe the aT. They are Arrhenius type equation and WLF equation. The Arrhenius equation is indicated in
Where ΔΗT is chemistry activation energy and R is the gas constant. TheWLF equation is derived from free volume theory as described in Eq. 21 where C1 = 17.44 and C2 = 51.6 for many polymers:
The Arrhenius equation is used for temperature above the reference temperature, while the WLF equation is used for temperature below reference temperature.
For viscoelastic property measurement, a DMA (Dynamic Mechanical Analyzer) is used to measure the material modulus as a function of relaxation time and temperature.
C. Characterization of material properties
In this section, the details of material characterization are explained. Figure 7 shows the viscosity as a function of shear rate, temperature, and heating rate (Q). The Cross Castro Macosko Model is implemented to describe the relationship. For thermoset material, the viscosity increases as temperature increases. At high temperatures, the cross-linking of material causes the rapid increment of viscosity.
Figure 7. Viscosity as a function heating rate and temperature
Figure 8 and 9 show the specific volume as a function of pressure and temperature of cured and uncured mold compounds. The PVTC simulation considers the changes of specific volumes from the two curves.
Figure 8. Specific volume of cured mold compound as a function of pressure and temperature
Figure 9. Specific volume of uncured mold compound as a function of pressure and temperature
Figure 10 is the curing kinetic of mold compound. The curing kinetic was described by a combined model, and the figure shows the conversion rates increase at higher temperatures.
Figure 10. Conversion rate as function of temperature and time
Figure 11 is the shear modulus of a fully cured mold compound as a function of relaxation time and temperature. The relationship is described by Generalized Maxwell model. The temperature shift factor (TTS) is used to describe the material softening as a function of temperature and time.
Figure 11. The modulus of fully cured mold compound as a function of relaxation time and temperature
VI. Comparison of warpage
The warpage data from both experiment and simulation were compared, and Figure 12 is one of the experimental warpage measurements with Shadow Moire. The deformed strip is concave in the middle along the length direction. Figure 12 and Figure 5 look similar but it is difficult to compare the differences without showing warpage magnitudes together. In this regard, we placed all warpage data in the same graph, and Figure 13, 14, and 15 illustrate the warpage data along the width, length, and diagonal directions. To have a clear comparison, the lowest warpage value of each curve is offset to be zero in Y axis, thus all the curves have the same reference point. In the figures, only two sets of experimental data, which are named exp1 and exp2, were placed to compare with traditional mechanical and PVTC simulation data.
In general, the warpage directions of simulation data match with that of experimental data. However, the data of traditional mechanical simulation, those along the length and diagonal directions, overshoot the experimental data significantly compared to that of PVTC simulation data. One of the possible reasons is that traditional mechanical simulation is not able to consider the kinetic behaviors of mold compound during the molding process and the relaxation of the stresses of mold compound during the annealing process. Another observation is that traditional mechanical simulation matched better with experimental data in the width direction compared to that of PVTC simulation.
Figure 12. Experimental warpage data
Figure 13. Warpage of strip along width direction.
Figure 14. Warpage of strip along length direction.
Figure 15. Warpage of strip along diagonal direction.
VII. Compression vs. transfer molding
The manufacturing mechanisms of compression and transfer molding are very different, for example, for transfer molding, the mold compound is injected into the mold cavity from the gates along the long edge of strip, as shown in Figure 16, and then the mold compound gradually flows and fills the entire mold cavity. In this section, we will compare the pros and cons of the compression molding and transfer molding processes.
Figure 16. Mold compound flow of transfer molding
A. The needed pressure of compression molding shall be much more uniform and lower than that of transfer molding, and Figure 17 is an example. The placing of molding gates of transfer molding at one edge of strip leads to a longer flow path. Therefore, a higher injection pressure and pressure gradient across the filling direction occurred. On the other hand, for compression molding, mold compound is evenly placed inside the mold cavity before the molding process, such that the required molding pressure requirement is much lower, and the pressure will be very uniform.
B. The compression molding process requires lower clamping force for the devices which have larger dimensions and many thinner parts, and Figure 19 shows an example.
C. To enforce the mold compound flow of the transfer molding process, higher pressure and higher flow speeds may induce higher shear rates of mold compound. The higher shear rates of compound introduce great risks of not only cracks of bumps or dies but also weaker encapsulation.
D. If the IC packages are very thin, transfer molding will require very high pressure to fill the mold cavity, and it may not be achievable or practical.
E. Compression molding may be a better option if the device’s dimension is larger because the molding process is more energy efficient and has less material waste.
Figure 17. Comparison of pressure distribution.
Figure 18. Comparison of shear rate.
Figure 19. Clamping forces for transfer molding and compression molding.
VIII. Conclusion
The warpage of compression molded SiP substrate strips was studied numerically and experimentally. The warpage of SiP modules was measured with Shadow Moire, and two different modeling approaches, traditional mechanical and PVTC simulations, were used to calculate the warpage and to compare with experimental data. The PVTC material properties of mold compound are carefully characterized for simulation. The conclusions or the suggestions of continued works are listed below:
A. In general, both traditional and PVTC simulation can predict the warpage direction of a strip. However, traditional mechanical simulation will overshoot the warpage magnitude significantly along the strip length direction. In this regard, we believe the traditional mechanical simulation is not appropriate to predict the magnitude of warpage of strip, and a process flow simulation, such as PVTC simulation, is more appropriate to capture the warpage behavior of compression molded strip.
B. It is not practical to manufacture a SiP strip with two different molding processes, the transfer molding and compression molding, as discussed in this paper. For transfer molding, we have to use numerical simulation to obtain the data of warpage, pressure, and shear stresses. To better understand the pros and cons of two molding processes, we recommend selecting an appropriate device to be manufactured by both molding processes so that we can also compare them experimentally.
C. The behavior of warpage along the width direction of strip is not clearly understood at the time of finishing the paper, and the PVTC simulation data does not match nicely with the experimental data. The possible reasons are (a) the anisotropy material properties of strip components, including substrate layer and mold compound, are not considered. (b) the intrinsic stresses of substrate strips are not included in the simulation. (c) the used visco-elastic model is not able to capture the real warpage behavior and a better visco-elastic model is needed.
D. Better material characterization methodology and the development of advanced simulation schemes are still needed, and these two things are critical to solve the issues stated above. The deeper and more thorough fundamental studies of materials and processes are the keys to having reliable IC packages.
References
 |
 |
 |
 |
 |
 |
 |  |
Leave a Reply