In this study, a simplified fluid visualization experiment was designed to simulate the actual transfer molding process. A visualized mold was developed to represent the flow-front merging phenomenon beneath the chip gap. Silicone oil was used as the working fluid instead of an EMC. Abbott and Higgins [53] reported that the surface tension of curing epoxy decreases linearly with temperature, with a coefficient of −0.074 dyne · cm⁻¹ · °C⁻¹. At a typical molding temperature of 175°C, the estimated surface tension is approximately 29.75 dyne/cm. In comparison, the surface tension of the selected silicone oil is about 21.3 dyne/cm, which is reasonably close to that of EMCs. Although silicone oil does not exhibit the same rheological behavior as EMCs—such as shear-thinning—its viscosity can be precisely selected from a wide range to match that of EMCs under molding conditions. This makes it a practical substitute for flow visualization purposes. Similar substitution has also been adopted in prior studies [51], supporting its validity for qualitative analysis of flow behavior and void formation. This substitution allowed the observation of the flow behavior in the MUF process without considering the rheology of the EMC. Therefore, experiments were conducted at room temperature. In this fluid visualization experiment, the flow front and air pockets were observed. In addition, a 3D mold flow simulation was performed for comparison. This study aims to provide suggestions for accurately determining air pocket sizes using mold flow simulation. Building upon the validated modeling solution, sensitivity analyses were conducted to examine the influence of variations in several critical design factors; these analyses provide valuable insights that are particularly useful for qualitatively evaluating the relative impact of each key factor under investigation, especially when considered within the broader context of design optimization and performance prediction. Beyond the analytical results presented in this study, the simulation framework developed herein offers practitioners a flexible and robust tool; it enables them to conduct case-specific analyses tailored to their actual product geometries and material characteristics, thereby enhancing the applicability of the methodology across a wide range of engineering scenarios and facilitating more informed decision-making in real-world design processes.

The fluid visualization experimental setup is shown in Fig. 1, and the dimensions of the visualization mold are shown in Fig. 2. The mold was designed with nine chip-like protruding objects, hereafter referred to as “chips”. Each chip measured 10 mm × 10 mm with a thickness of 1 mm. The chips are spaced 4 mm apart. A glass plate was placed over the mold to create a 0.1 mm gap between the chips and glass, replicating the flow space of the MUF process, with thinner gaps beneath the chips and thicker flow channels around them.

A stepping motor was used to transfer the working fluid from the syringe through the gate to the mold cavity. The presence of air creates flow resistance, leading to larger air pockets in the chip gap. To mitigate this, a high vacuum is preferred during molding. In this experiment, a vacuum pump was connected to the end opposite the gate to extract air and achieve a vacuum level of 19 kPa.

The viscosity of the working fluid significantly influences the flow behavior during the filling phase of the transfer molding process. For this reason, three silicone oils with different viscosities—1000, 3000, and 10,000 mPa · s—were selected for the experiments. The viscosities of silicone oil were selected based on the typical viscosity range of EMC during the encapsulation filling process. The filling process was captured using a high-speed camera to document the changes in the flow front. The captured images are shown in Fig. 3a. Edge detection was then applied as the image processing method to obtain a positional image of the flow front, as shown in Fig. 3b, with a particular focus on the air pockets in the chip gap. The experiment was conducted three times to ensure repeatability. The central air pocket was selected as the reference void for evaluation, and the results showed that the standard deviation of the void size was 2.43%, which is within an acceptable range for experimental repeatability.

In this study, the flow behavior of the working fluid was evaluated. Three conservation laws of fluid dynamics—conservation of mass, conservation of momentum, and conservation of energy—were considered in the modeling. The mass-conservation equation is expressed as follows: (1)∂∂tρ+∇⋅ρu=0,where u is the velocity vector of the fluid, ρ is the density of the fluid, and t is time. The momentum conservation equation is expressed as (2)∂∂t(ρu)+∇⋅ρuu=−∇p+∇⋅μγ˙+fσ,where p is the fluid pressure, μ is the fluid viscosity, γ˙ is the shear stress tensor of the fluid, and fσ is the capillary force. The capillary force is expressed as follows: (3)fσ=−σκ∇φ,where σ is the surface tension of the fluid, κ is the curvature at the flow front, and φ is the volume fraction of the working fluid at the flow front. The curvature and volume fraction of the working fluid at the flow front were calculated iteratively using 3D mold flow simulations. The fluid volume conservation equation was used to determine the position of the flow front as follows: (4)∂∂tφ+∇⋅φu=0.

In addition, the energy conservation equation must be satisfied, which can be expressed as (5)∂∂t(ρh)+∇⋅ρuh=∇⋅k∇T+μγ˙2,where h denotes the entropy of the fluid, k denotes the thermal conductivity of the fluid, and T denotes the fluid temperature. The two-phase flow behavior can be described based on these equations. The mathematical model also accounts for the effects of capillary forces.

In this study, Moldex3D software was used for mold flow simulation. The geometric model for the simulation was constructed based on the mold design used in the fluid visualization experiment. The mesh model is illustrated in Fig. 4. Fig. 4a shows the top view of the mesh model, where a triangular mesh with a size of 0.1 mm was applied. Fig. 4b presents a side view of the mesh model. For the chip region, the mesh size was set to 0.1 mm, whereas for the chip gap area, a finer mesh size (0.01 mm) was applied to ensure accuracy. A mesh independence study was performed, as shown in Fig. 5, where the void size beneath the center chip was used as the evaluation index. The results show that the void size becomes convergent when the mesh size is smaller than 0.4 mm, indicating sufficient mesh resolution for numerical accuracy.

The process conditions used in this study were as follows: The transfer time was set at 10 s at a constant transfer speed. The transfer pressure was maintained after the mold cavity was completely filled. The material properties of the silicone oil, which were assumed to be constant, are summarized in Table 1. In this study, the material parameters were considered constant.

After validating the simulation model, an experimental design was developed to evaluate the influence of the design factors. The air pocket size at the end of the filling phase of the molding process eventually became a void in the molded product; therefore, this was selected as the response. However, the air pocket sizes of the nine chips varied. The air pocket under the central chip was selected as the response index. Table 2 summarizes the evaluated factors, including viscosity, contact angle, transfer time, transfer pressure, and chip gap height.

Viscosity represents material selection. The contact angle reflects process control. In general, plasma activation is performed prior to molding. This process results in a lower contact angle. However, the contact angle increases when the queue time of the semi-manufactured product between the plasma activation and molding processes is prolonged. The transfer time and pressure are process parameters controlled by the process engineer, whereas the chip gap height is a product design factor that is beyond the control of the process engineer.

To further assess the impact of key factors on void formation, sensitivity analyses were conducted as part of the design of experiments (DOE). The variations in viscosity, contact angle, transfer time, transfer pressure, and chip gap height were examined at different levels, and their effects on void size were systematically quantified. This approach enabled us to evaluate the relative importance of each factor and gain a deeper understanding of how these parameters influence the encapsulation process. The results of these sensitivity analyses provide valuable insights for optimizing the molding process and minimizing void formation in practical applications. Two or three levels were selected for each factor, as listed in Table 2. The one-factor-at-a-time method was selected for the experimental design. The reference parameter combination consists of 3000 mPa · s for the viscosity, 19.32 degrees for the contact angle, 10 s for the transfer time, 3 MPa for the transfer pressure, and 0.1 mm for the chip gap height, which are highlighted by underlining in Table 2. The range of each factor was determined based on practical experience and process knowledge. Specifically, level 1 of the contact angle (19.32°) corresponds to the condition used in our validation experiment, which simulates the surface of a flip-chip substrate immediately after plasma activation. Levels 2 (50°) and 3 (70°) represent typical values observed when the substrate has been left in queue for varying durations prior to molding. These values are consistent with real-world process variations and thus are representative of EMC-substrate interactions in practical flip-chip packaging scenarios.

The capillary force is influenced by the surface tension of the fluid and the contact angle between the fluid and the contact surfaces. Capillary effects are typically observed in small-scale systems. The comparison results regarding the effect of the capillary force are illustrated in Fig. 6. Fig. 6a shows the simulation results without considering the capillary force; Fig. 6b shows the simulation results including the capillary force, and Fig. 6c displays the experimental results.

When the capillary force was not considered, the fluid was driven solely by the transfer pressure. In the thin gap (under the chip), the flow resistance owing to viscosity becomes dominant, making it difficult for the transfer pressure to drive the fluid forward. This explains why the simulation results without the capillary force exhibited larger air pockets than the experimental results.

Conversely, the simulation results those accounts for the capillary force shows smaller air pockets compared to the result without the capillary force. This is due to the additional driving force provided by capillary action. These results highlight the necessity of considering the capillary force in the mold flow simulation of the MUF process. In addition to the qualitative agreement between simulation results (Fig. 6b) and experimental observations (Fig. 6c), root mean square error (RMSE) and error percentage were employed to objectively evaluate the level of agreement. The RMSE was calculated to be 0.224, and the error percentage was 12.8%, using the experimental void volume as the reference.

Generally, the capillary number (Ca) is used to assess the relative importance of the viscous and capillary forces and is defined as (6)Ca=μuσ,where the viscosity (μ) and surface tension (σ) are material properties of the fluid, and the fluid velocity (u) is obtained from numerical simulation. A cross-section of the fluid velocity distribution during the MUF filling process is shown in Fig. 7. Fig. 7a shows a top-view figure with the A-A broken-section line indicating the location of the cross-section, as depicted in Fig. 7b.

Fig. 7b shows that the velocity in the chip gap is 0.481 mm/s, which is significantly smaller than the velocity outside the chip (14.091 mm/s). Therefore, the capillary number was calculated as follows:

At the chip gap: Ca=0.068=O(10−2),

Wolf et al. [54] indicated that the capillary number for a capillary-dominated flow between parallel plates is typically in the range of O(10−4) to O(10−2). The results of the present study explain why the capillary force should be considered in the MUF simulation, especially for the flow in the chip gap.

Based on the validated simulation model, an experimental design was developed to evaluate the influence of the design factors. Fig. 8 shows the flow-front positions for fluids with different viscosities, with all other material parameters kept constant, as listed in Table 1. The blue (smaller) circles indicate the flow fronts around the chips for fluids with varying viscosities. The positions of the flow fronts remain the same and are unaffected by the fluid viscosity. This is because the flow is governed primarily by the transfer speed rather than the transfer pressure. The same fluid volume must fill the mold cavity regardless of the viscosity; thus, the flow-front positions are identical at the same time.

However, when focusing on the red (larger) circles in Fig. 8, it is evident that the positions of the flow fronts differ. The lower the viscosity, the greater the extension of the flow-front. This is because the capillary force is independent of the viscosity, and a fluid with lower viscosity resistance is more easily drawn by the same capillary force. These results demonstrate that the capillary force is significant in the gap under the chip and insignificant on the side of the chip. Therefore, it can be concluded that mold flow simulations for general transfer molding do not need to account for capillary force, whereas mold flow simulations for MUF must.

The overall numerical modeling results for the experimental design are summarized in Fig. 9, which was obtained using the one-factor-at-a-time method. A black horizontal line in Fig. 9 indicates the void size corresponding to the reference parameter combination, which is 2.05 mm³. A table was added to Fig. 9 to present the range and ranking of all studied factors. The range for each factor was calculated as the difference between the maximum and minimum void sizes resulting from its variation. Based on this data, the optimization potential is comparable to the reference void size, highlighting the significance of parameter tuning. The most influential factors are viscosity, chip gap height, transfer time, contact angle, and transfer pressure, in that order. The optimal combination includes lower viscosity, a larger chip gap height, longer transfer time, lower contact angle, and higher transfer pressure. This result underscores the critical role of product design. Although miniaturization trends may exacerbate the void issue, improvements can still be achieved through careful EMC material selection and process control. Further discussion on the influence of each factor and future prospects is provided in the following section.

Furthermore, the viscosity of the working fluid was crucial. Selecting the EMC with the lowest viscosity during the filling phase is vital. However, the viscosity of the EMC changes owing to rising temperatures and the reactivity of the material. Thus, viscosity is influenced not only by the EMC’s intrinsic properties but also by the process temperature and time, which are factors within a process engineer’s control.

Regarding the contact angle, the results indicated that smaller angles were better. However, the difference in performance between 19.32° and 50° was much smaller than that between 50° and 70°. This suggests that there is no need for excessively stringent standards requiring contact angles to be significantly less than 50°.

The results also showed that a longer transfer time was beneficial but only if the viscosity remained constant. For EMC, the viscosity increases if the transfer time is too long. This is an important consideration for process designers. Therefore, there is an optimal transfer time for materials with reactive viscosities, such as EMC. This represents the maximum allowable time before the EMC viscosity begins to increase dramatically owing to the chemical reactions.

In summary, with the trend of miniaturization, several approaches can be considered to address the void issue. These include selecting an EMC with a lower initial viscosity and longer low-viscosity duration, extending the transfer time, and managing the queue time between plasma activation and molding to maintain the contact angle at or below 50°. Low viscosity during the flow process, or an extended low-viscosity window, can be achieved through both material selection and process condition optimization. From the perspective of material selection, a lower initial viscosity may be obtained by using low molecular weight resins or adjusting the filler content. A longer low-viscosity duration can be achieved by choosing lower-reactivity curing agents, reducing catalyst concentration, or incorporating rheology modifiers. From the perspective of process adjustments, reducing the mold temperature (e.g., from the typical 175°C to 165°C or 160°C) can decelerate the curing reaction and prolong the low-viscosity phase.

Real EMC materials exhibit temperature-dependent viscosity variations during the molding process. In this study, a constant-viscosity silicone oil was deliberately selected as a simplified substitute for EMC in the validation experiment. This choice enabled controlled and repeatable testing conditions, allowing us to focus on the quantitative investigation of flow behavior and void formation mechanisms without the added complexity of thermal effects. By introducing this simplification, the study was able to effectively capture the key flow mechanisms of the MUF process under controlled conditions. We recognize that assuming constant viscosity may limit the quantitative accuracy of void predictions for real EMCs. Nonetheless, the insights gained from this simplified model offer a valuable foundation for understanding flow behavior, which can inform more detailed, material-specific simulations in future work.