A numerical study on mitigation of flying dies in compression molding of microelectronic packages

Fan-Out Wafer Level Packaging (FOWLP) is one of the latest packaging trends in microelectronics, enabling high degrees of integration and miniaturization. Known-good-dies are placed on a temporary carrier before compression molding, forming the reconfigured wafer (Fig 1). In this study, a 200 mm reconfigured wafer with 845 dies, with a form factor of 2×2×0.2 mm³, is considered (Fig. 2). Forces from pressure and shear acting on the dies' surfaces are determined for two liquid encapsulants under varying process conditions, such as temperature and compression profile.

If forces exerted onto a die by viscous epoxy flow exceed the adhesive strength of the adhesive thermal release tape (TRT) during the filling stage, so-called flying dies (Fig. 3) occur. Thereby, dies come loose from the adhesive tape and move into adjacent packages, decreasing yield. Two cases are discussed: First, the melt front of the EMC arriving parallel to the front surface and secondly arriving at an angle of 45°, approaching right towards the edge of the dies.

For embedding with liquid encapsulants, highly filled systems of liquid reactive epoxy resins with a filler content of spherical SiO₂ of around 90 wt.-% are commonly used. Their viscosity is a function of temperature T, shear-rate γ˙ and conversion α, which itself is a function of T and time t, as curing takes place faster at elevated temperatures. To capture the highly nonlinear interplay of these characteristics, rheo-kinetic behavior is measured.

In order to capture the curing kinetics, DSC is used at different heating rates (2, 5, 10, 20 and 40 K/min). The obtained data is fitted using Kamal's Model [1], where model parameters A₁, A₂ and respective activation energies E₁ and E₂ yield k₁ and k₂, which are reaction rate constants of Arrhenius type temperature dependence. n is the order parameter, which describes the reaction dependent only on concentration of epoxide. The autocatalytic part of the reaction is taken into account by introducing m.

The experimental data and the fitted curves are in very good agreement. Calculated isothermal conversion curves, calculated with the fitted model, can be seen in Fig. 4 for EMC A. They show the strong influence of T on the crosslinking reaction of the EMC.

For modelling the shear-thinning behavior in combination with, experimental data, obtained from a plate-plate rheometer, under varying temperatures (80, 90, 100 and 110 °C) and shear-rates (40 shear-rates between 0.1 and 100 rad/s), is fitted to the Cross Castro Macosko Model [2]. Thereby, α_G denotes the degree of cure at the gel point, η₀ is the zero-shear-rate viscosity, τ^*is the critical shear stress and n is the power law index, while E_A is the activation energy. c₁, c₂ and η* are fitted constants. Conversion α as a result form solving (1) forward in time is also used as an input for (3), which yields the viscosity, including both conversion and shear-thinning properties of the EMC.

Fig. 5 shows the shear-thinning behavior of both EMCs. In Fig. 6 the viscosity over time under a constant shear-rate is plotted.

Shear tests with dummy Si dies on a commonly used TRT are performed to determine the critical force F_crit, thus shear strength τ.

An analytical model to estimate the melt front velocity during the compression molding process is used. A principle sketch can be seen in Fig. 7.

With A_top = πr², A_meltfront = 2πrh and the conservation of mass, v_compression · A_meltfront, the following relationship is deduced:

For better prediction of v_meltfront with the analytical approach, the total volume of the 845 dies is added to the final wafer, resulting in 295 μm height instead of 300 μm. The initial epoxy charge height is 3.3 mm with a radius of 29.5 mm.

A constant value for v_compression leads to a strong increase of v_meltfront towards the peripherals. Higher velocities typically mean higher forces, therefore more stress and, ultimately, lower yield. To compensate for this acceleration, a compression profile with step-wise decrease in v_compression is realized, which is depicted in Fig. 8, while the melt-front velocities over time of the three different compression cases in this study can be seen in Fig. 9.

In the present study, the commercially available software Moldex3D R15.0 is used for the simulations. The filling analysis is performed using a Finite-Volume-Method (FVM) to solve the Navier-Stokes-equations for the EMC with the material models discussed above. Moldex3D allows to define virtual Sensor Nodes (SN), which log all variables at a specified point over time. In the present case the focus is on pressure, shear rate, viscosity and shear stress.

For saving computational effort and due to symmetry in the model, a 45 ° slice of the full wafer is simulated. It is presented in Fig. 10 with the respective Sensor Nodes on the dies along the 0 ° and 45 ° lines from the center of the wafer (red boxes). The dashed lines show the continuation of the dies, as only half of them are simulated due to the symmetry boundary conditions, which are imposed on the radial planes (compare Fig. 10 and 11). To compensate for the increase in velocity with a simultaneous decrease in height, resulting in an extensive increase of the shear-rate, the computational mesh is refined towards the peripheral.

Assuming perfect adhesion between the dies and underlying adhesive tape, as well as symmetrical flow around them, forces can be calculated as follows:

The front and back of the die are perpendicular to the melt front, therefore the resulting force only has a pressure contribution, while the sides and top only have shear contributions:

In this case, both sides have a pressure and shear contribution. Forces perpendicular to the surfaces are computed via F_i =p_i · A_i, for the parallel ones it reads F_i = τ_i · A_i.

The resulting force then becomes:

With the computation simplifies to:

Fig. 11 shows the computational mesh, with the initial EMC charge, the cavity with dies, as well as the compression zone.

For two EMCs temperature (115 °C, 125 °C and 135 °C) and compression speed (0.1 mm/s, 0.2 mm/s and the profile discussed above) are varied to study their influence on flow behavior, hence forces on dies.

From the shear strength, values at the respective temperatures (compare Table I) and the dies' base area, which is 4 mm², critical forces onto the dies read:

• 1.96 N at 115 °C

• 1.65 N at 125 °C

• 1.25 N at 135 °C

Tables II and III show a compilation of the overall simulation results, for EMC A and B, alongside with the different angles 0 ° and 45 °. For each combination, the table shows the number of dies subjected to drag forces larger than the critical value during the filling stage. Hence, 0 is desired, indicated in green.

For EMC A, the slowest v_compression of 0.1 mm/s leads to forces exceeding the critical values. This is due to the faster conversion at elevated temperatures, which increase the rate of crosslinking in the epoxy resin, combined with a longer compression time of 30 s.

The compression speed of 0.2 mm/s yielded uncritical forces, combined with a fast cycle time of 15 s.

However, further speeding up compression will lead to excessive forces on the dies. The compression profile, see Fig. 9, yields smaller forces at a only slightly increased compression time of around 23 s.

A representative comparison of the force profiles of EMC A at 135 °C is illustrated in Fig. 12. The red line shows the force limit of 1.25 N at 135 °C, which is exceeded for v_compression = 0.1 mm/s. If can be seen that, even though the other two graphs both are uncritical, the compression profile with step-wise decrease in v_compression yields lower maximum forces at an only slightly increased compression time. Therefore, regarding the issue of flying dies, the compression profile is well suited for meeting the challenges of achieving high yield at reasonable process times, as the maximum force is smaller and the process therefore is more tolerant to process and material variations.

The comparison of the 0 ° and 45 ° case (Fig. 13) shows that drag forces are slightly higher for 45 °, possibly due to the increased projected cross-sectional area. Since the increase is almost negligible, it seems as if the generic streamline form of the 45 ° almost completely compensates for the increased area.

The findings are consistent with the overall results and show similar behavior for both cases for both EMC A and B (compare tables II and III).

For EMC B, the significantly lower viscosity, compare Fig. 5 and 6, leads to acceptance of all tested sets of parameters. This indicates, that even faster cycle times could be used without lowering yield by flying dies.

A framework to determine forces on individual dies during encapsulation of micro-electronic packages is presented. It aims to mitigate the issue of flying dies during encapsulation, therefore contributing to maximizing yield and ultimately saving materials, cost and shorten development times. Within the presented methodology pressure and shear contributions can be evaluated separately.

In a compression molding setup for 200 mm wafers, flow simulations for two widely used EMCs are performed analyzing forces on 16 dies (0 °) along the radius (11 for the 45° case). Process temperature and compression speed are varied to gain insight into the interplay of shear-thinning behavior and curing kinetics, increasing viscosity. Forces are then contrasted to critical values resulting from shear strength determined by shear tests at process temperatures. It was found that EMC B had a sufficiently low viscosity for all the parameter combinations in the present study.

For EMC A, however, it was shown that for the slowest compression speed at 135 °C, forces exceed the limits. The proposed compression profile, aiming to find a sweet spot between melt front velocity and process time, showed significantly smaller forces onto the dies.

For future developments, a more generalized approach is desired. It would offer insight beyond the both extreme (geometrical) cases of 0 ° and 45 °. Additionally, for situations with asymmetrical flow around the dies, resulting torque can be quantified and analyzed.

For a deeper understanding of the interplay between chemorheology and geometrical conditions, different aspect ratios of the dies, as well as fine-pitch setups with small gaps, where flow around one die influences adjacent ones, should be studied in more detail.

As the framework is easily scalable to arbitrary geometries, application within current efforts of moving to larger substrates, such as from FOWLP to FOPLP [3], is desired.

Furthermore, the force profiles extracted from the simulations could be realized in a modified force-controlled-shear-test to determine flow impact on die-shift. Subsequent investigations on temperature-dependence of the shear strength of different thermal release tapes are necessary for a more reliable optimization of parameter windows.