The primary driver for performance improvements comes from increasing the number of transistors, while improvements in the performance of individual transistors only play a supporting role in maintaining Moore's Law. As the size of the SoC gradually approaches the limit of the photomask aperture size (858 mm²), and process miniaturization becomes extremely difficult and cost-effectiveness faces challenges, multi-chip packaging technology has taken center stage as a key to further improving chip performance. Flip-chip bonding technology has become one of the most important technologies in advanced multi-chip packaging.
While some wafer-level or panel-level fan-out packaging processes can avoid flip-chip bonding, this technology has its own limitations. For example, the number of redistribution layers in mass production is mostly within five layers, making it less suitable for server chip packaging. Currently, the most common flip-chip bonding technologies based on solder balls are three: mass reflow, thermocompression bonding (TCB), and laser-assisted bonding (LAB). Copper-copper direct bonding flip-chip packaging is not the focus of this article; we will discuss the most promising methods for further reducing bump pitch (bump pitch scaling down) in subsequent articles, such as hybrid bonding and copper-copper direct thermocompression bonding.
The basic principle of hot-press bonding is the same as that of traditional diffusion bonding, where the Cu bumps of the upper and lower chips are aligned and then directly contact each other. The main parameters affecting atomic diffusion bonding are temperature, pressure, and time. Because the surfaces of the electroplated Cu bumps are rough and have a certain height difference, they need to be planarized before bonding, such as through chemical mechanical polishing (CMP), to ensure sufficient contact between the Cu surfaces during bonding. Based on current research, methods for achieving low-temperature Cu-Cu bonding via hot-press bonding can be mechanistically divided into two categories: increasing the Cu atomic diffusion rate and preventing/reducing oxidation of the Cu surfaces to be bonded.
Table 1 summarizes the advantages and disadvantages of these three solder ball-based flip-chip bonding packaging technologies at a high level. It can be seen that no bonding method is perfect. For any given product, the task is to find the most suitable bonding method. As the focus of this article, the biggest advantage of thermocompression bonding is its more precise control over the die and substrate (which could be the substrate, another die, wafer, or panel).
Table 1: Flip-chip bonding technology based on solder balls
Next, we will focus on discussing why the bonding method for high-end logic chips has gradually transitioned from reflow bonding to thermocompression bonding. Figure 1(a) shows a common reflow chip bonding process. First, flux is applied to the bumps on the chip, or a measured amount of flux is sprayed onto the C4 (controlled-collapse chip connection) area on the substrate. Then, a chip placement device is used to place the chip relatively precisely onto the substrate. Finally, the chip (die) and substrate are placed together in a reflow oven. Common reflow temperature control is shown in Figure 1(b). The entire reflow process typically takes 5 to 10 minutes. Although the time is long, because this is batch processing, a single reflow oven can accommodate a very large number of products being processed simultaneously. Therefore, the overall throughput is still very high, typically reaching several thousand chips per hour, or even higher.
Depending on the type of reflow solder paste used, the peak reflow temperature is generally controlled between 240ºC and 260ºC. After reflow bonding, it is usually necessary to remove the flux and add CUF (Capillary Underfill) to fill the gaps between the bumps to improve product reliability. In common reflow soldering processes, we don't impose many restrictions on the chip and substrate. This leads to ineffective control of chip and substrate warpage, resulting in significant variations in the chip gap height below the chip surface. Excessive warpage leads to the two most common defects: NCO (noncontacted opening) and SBB (solder ball bridging). The yield of complex multi-chip bonding based on reflow soldering can be very low, making reflow soldering no longer the most suitable bonding method.
The yield rate of reflow soldering chip bonding is related to many factors, commonly including chip size and thickness, bump pitch, substrate thickness, and mismatch in coefficient of thermal expansion (CTE), as shown in Figure 2. Further reduction in bump pitch also increases the probability of surface-mount block (SBB). It is also worth noting that even without NCO and SBB defects, excessive chip gap height variation and excessive warpage pose significant challenges to downstream packaging steps.
Figure 1: Common reflow soldering chip bonding process
Figure 2: Relationship between yield rate and chip size and thickness in common reflow soldering chip bonding
Figure 3 illustrates the structure of a common thermoforming bond. Both the substrate and the chip have their own heating devices. Depending on the type of solder paste, the substrate is typically heated between 150ºC and 200ºC. The substrate is vacuum-bound on a very flat base, thus effectively controlling substrate warpage. The chip is similarly vacuum-bound on a very flat bond head, resulting in well-controlled chip warpage as well.
The alignment between the chip and the substrate requires extreme precision, including alignment in the XY plane, distance between the chip and the substrate (control in the Z direction), and relative tip tilt. Typically, alignment accuracy needs to be ±3 µm 3 sigma. Excessive oxygen concentration in the bonding region can negatively impact bonding, such as causing voids and affecting bond strength. Unlike reflow soldering, which takes 5 to 10 minutes, thermocompression bonding takes only about 1 to 5 seconds. However, because thermocompression bonding is performed chip-by-chip rather than in batches like reflow soldering, its throughput is only about 1/5 that of reflow soldering. Furthermore, thermocompression bonding equipment is generally significantly more expensive than reflow soldering equipment, making thermocompression bonding more costly than reflow soldering.
Furthermore, because the heat required for bonding is primarily provided by the heater in the bonder head, the heat from C4 diffuses towards the substrate edges. This results in a significantly lower temperature at the C4 edges compared to the center. Consequently, the peak temperature of hot-press bonding is much higher than the melting temperature of solder paste, typically exceeding 300°C. Compared to the peak temperature of reflow soldering, this excessively high hot-press bonding temperature poses considerable challenges to the selection of bonding materials, the stability of the bonding process, and the reliability of the product.
As mentioned earlier, in reflow soldering, the molten solder paste has a self-correcting ability to reduce surface energy, which helps the chip bumps align more accurately with the substrate bumps after reflow soldering. However, in thermocompression bonding, the chip and substrate are bound together while the solder paste is molten, thus losing this advantage. Fortunately, most thermocompression bonding equipment currently available can achieve very high placement accuracy between the chip and substrate, with some reaching ±2 μm accuracy with 3σ. ASM Pacific, Kulicke & Soffa, Besi, and Toray are among the most common thermocompression bonding equipment suppliers.
Currently, domestic equipment manufacturers are also actively developing in this field, such as Huafeng and Tangren Manufacturing. While each manufacturer's thermosetting bonding equipment may have its own unique features, they all optimize the following aspects: position control precision, tilt/parallelism control precision, fast and accurate temperature control, precise measurement and control of bonding force, vacuum adsorption control of the chip and substrate, equipment stability, reduction of differences within the same model of equipment, increased throughput, reduced equipment costs, and reduced equipment footprint, among others.
Figure 3: Structure of a common hot-pressed bond
The chip gap height (CGH) differences between reflow soldering and thermocompression bonding are shown in Figure 4. For this specific product, the CGH ranges from 70 µm to 100 µm for reflow soldering-based bonding. Such a large CGH variation results in a very low bonding process margin. Even slight variations in incoming materials can lead to a decrease in production yield. In contrast, the CGH variation for thermocompression bonding is only about 5 µm. This smaller CGH variation not only helps improve the bonding process margin itself but also helps reduce incoming material variability in downstream packaging and testing processes, thus making downstream packaging and testing steps more stable.
Figure 4: Chip gap height based on reflow soldering and thermocompression bonding
Next, we will briefly describe the most common TCCUF (Thermo Compression Bonding with Capillar UnderFill) hot compression bonding process, which typically takes 1-5 seconds.
The substrate is vacuum-adsorbed onto a very flat pedestal and typically heated to 150ºC to 200ºC. The substrate temperature is set as high as possible to reduce bonding time.
Apply sufficient flux to the C4 region of the substrate.
Heat the bond head to between 150ºC and 200ºC, and use the bond head to pick up the chip.
Up-looking and down-looking cameras are used to determine the relative positions of the chip and the substrate. The required spatial position adjustment of the chip is calculated using a calibrated algorithm to perfectly match the bumps of the substrate. This step is accomplished through precise mechanical control on the equipment.
Then, the bond head, along with the adsorbed chip, is brought close to the substrate with a sub-µm precision. At this point, both the chip and the substrate are below the solder ball's melting temperature, so the solder balls are solid. The solder balls can be on the substrate, on the chip, or both.
During the descent, the bond head remains under pressure-sensitive control, performing highly sensitive and real-time force measurements.
The moment the chip and the substrate come into contact, the system detects a change in pressure, thereby determining that contact has occurred and quickly switching the bond head from pressure-sensitive control to joint pressure and position control.
At this point, the chip is rapidly heated to over 300ºC using a heating element on the bond head. It's worth noting that the temperature change rate for thermoforming bonding is typically around 100ºC/s. In contrast, the temperature change rate for reflow bonding is much lower, usually around 2ºC/s.
When the solder balls are molten, precise positioning control of the chip via the bond head ensures that each pair of bumps bonds together and keeps the chip gap height within a reasonable range. It's worth noting that the entire system undergoes thermal expansion during heating, and this expansion needs to be offset by precise control of the bond head position.
The bond head is rapidly cooled to below the melting point of the solder ball, causing the solder ball to solidify. The rate of temperature change during cooling is typically lower than that during heating, usually around −50 °C/s.
The vacuum adsorption of the chip by the bond head is turned off, and the chip separates from the bond head. The chip is then bonded to the substrate and removed from the thermosetting bonding equipment, completing the bonding process.
To better illustrate the key steps of thermocompression bonding, we use the bonding profile shown in the figure below as an example. In fact, bonding profiles can vary significantly depending on the product. The implementation of all these bonding profiles relies on the precise control of temperature, pressure, and position by the thermocompression bonding equipment. As shown in Figure 5(a), the red, blue, and black lines represent the curves of temperature, pressure, and displacement of the bond head over time, respectively. When the bond head detects a pressure change, it indicates that the chip and substrate have made contact, and it rapidly heats both the chip and substrate above the melting point of the solder balls. For typical SAC305 solder balls (96.5% Sn, 3% Ag, and 0.5% Cu), this temperature is approximately above 300 °C. Due to the large temperature gradient, even though the bumps at the center of C4 may already be above 300 °C, the solder balls at the edges of C4 may barely be above their melting point.
For this reason, the peak temperature of the bond head is usually much higher than the melting point of the solder ball. The blue line in the diagram shows that during the heating process, the bond head is in a constant pressure control mode until the solder ball melts. Because the pressure drops instantly upon the solder ball melting, depending on the selected chip gap height, the bond head may change from pressure to tensile force. At this point, we adjust the height of the bond head to maintain constant pressure control; this adjustment also compensates for the effects of thermal expansion of the entire device. Typically, after detecting solder ball melting, we continue to press the chip down by 5 to 10 μm. The main reason for this is that the height of solder bumps is not uniform; coplanarity is usually within the range of 5 to 10 μm. The bond head may continue to press down further to ensure there is no NCO (on-contact open). Subsequently, the bond head may rise to control the chip gap height within a reasonable range. Then, the bond head is rapidly cooled below the melting point of the solder ball to complete the bonding.
Figure 5: Bonding profile of hot-press bonding, showing the bonding process.
Depending on the filler material, thermocompression bonding can be further divided into TCNCF (Thermo Compression Bonding with Non-Conductive Film), TCNCP (Thermo Compression Bonding with Non-Conductive Paste), TCCUF (Thermo Compression Bonding with Capillar UnderFill), TCMUF (Thermo Compression Bonding with Molded UnderFill), and so on. Depending on the substrate material, thermocompression bonding can also be divided into Chip-to-Substrate (C2S) and Chip-to-Wafer (C2W), Chip-to-Chip (C2C), and Chip-to-Panel. We will discuss these further in future articles.
