Công nghệ cắt bằng tia laser tạo nên cuộc cách mạng cho máy cưa wafer siêu mỏng vượt xa khả năng cắt bằng lưỡi dao

The demand for consistently high electrical performance in the power discrete semiconductor market has driven component developers to continuously enhance semiconductor assembly packaging technology through advanced package design and wafer fabrication methods. Among the cost-effective approaches are increasing the die area size and decreasing the die thickness, which minimize electrical resistivity within the die, while improving heat dissipation away from the product. This leads to enhanced temperature and power cycling reliability performance [1].

However, the die area size remains restricted by the pad size on the base of the lead frame. Although die thickness is currently the primary challenge, semiconductor companies are exploring silicon wafer thicknesses below 50 µm, potentially reaching the capability of blade dicing. Issues such as chipping at the side walls of dies (as shown in figure 1) and large bottom-side die chipping sizes (larger than half of the die thickness) persist when using conventional blade dicing technology. This challenge has been previously reported [2-3]. The mechanical wafer dicing process, which employs diamond grit, induces stress and mechanical damage during dicing on ultra-thinned dies, resulting in larger bottom-side and sidewall chipping and significantly reduced process yield [4-8]. As die thickness decreases to less than 50 µm, die strength becomes more critical. A study suggests that die strength should be consistent across all thicknesses, but different loads are required to cause fracture in dies of varying thickness [9]. In other words, the pressure load should be lower when the die is thinner; otherwise, die fracture or cracks may occur due to serious chipping along the die’s side wall. This stress or pressure loading may not only occur during the dicing process but also in downstream processes such as molding, wire bonding, and die pick-up and placement.

Fig. 1: Die side wall chipping.

To address these issues, the industry is actively seeking effective replacement technologies and laser ablation has emerged as a focal point for ultra-thinned silicon wafer dicing [10-13]. Unlike conventional blade dicing, laser ablation offers advantages such as reduced chipping, less cracking, and higher throughput. In laser ablation, material is removed from the top side of the wafer to the bottom side using a laser beam. This full-cut method focuses laser energy onto a minimal area, overlapping the oval beam shape in short pulses (approximately 10,000 per second). The process sublimes and evaporates the solid material. Laser ablation employs a level scrubbing method, starting from the top surface, and progressing downward. Figure 2 illustrates wafer scrubbing via laser ablation from the top surface.

Fig. 2: Photo of laser ablation scrubbed start from top surface.

The scrubbing process occurs in levels, as depicted in figure 3. Initially, the first laser pulse scrubs the top of the wafer while the wafer moves toward the laser as figure 3(a) illustrates. Subsequent pulses cause deeper scrubbing as shown in figure 3(b) and 3(c). Multiple laser irradiations continue until the wafer is fully cut through, as figure 3(d) illustrates, resulting in a sawing line (figure 3(e)). However, excessive feed speed or insufficient laser power may cause incomplete wafer cutting.

Fig. 3: The wafer sawed mechanism by laser ablation.

During the laser ablation sawing process, resolidified polycrystalline and amorphous silicon can form at the side of the die [13]. Additionally, debris may accumulate when recasting material is ejected and lands on the wafer’s top surface, as depicted in figure 4. This debris poses a potential risk of damaging the circuit on the die’s surface.

Fig. 4: Schematic of debris formed and the recast area on the surface of a partially sawn silicon wafer.

To safeguard the wafer surface during laser ablation wafer sawing, upfront preparation is essential. Figure 5 provides an overview of the laser ablation process on the wafer. In figure 5(a), a polyvinyl alcohol (PVA) layer is coated on the wafer’s surface for protection after loading [13]. Subsequently, the laser ablation process (similar to figure 5(b) or the explanation in figure 3) is performed. The PVA coating shields the wafer surface from loose particle residues or debris during laser ablation. Debris forms on the PVA coating layer without damaging the die surface. After deionized (DI) water rinsing, both the debris and the PVA coating are removed together, as shown in figure 5(c). However, even with the availability of the PVA layer coating, the potential for debris formation still exists if incorrect coating process parameters are used.

Fig. 5: Overall process of laser ablation on a wafer.

With this level of understanding the fundamentals and challenges of laser ablation, five remedy approaches are covered in this study. Initially, the laser ablation process was characterized and optimized. A structural design of experiments (DOE) was used to identify critical process parameters for optimization, focusing on 50 µm thick wafers. This was followed by the feasibility of 30 µm thick wafer sawing and quick check studies by evaluating the feasibility of laser ablation for dicing 30 µm thick wafers with quick check studies conducted on the saw streets. Next is a threshold limit study to determine the threshold limit of laser ablation capability by evaluating its performance on 30 µm and 75 µm thick wafers. Fourth is a die strength study and improvement proposals to address concerns related to die strength when using laser ablation for dicing. Improvement proposals were also explored. Finally, a comparison of blade versus laser ablation dicing in productivity and operation cost was performed.

Process characterization

A study on characterization followed by optimization of the laser ablation process aimed to identify critical-to-function (CTF) parameters for achieving full-cut wafers. The study focused on three parameters: laser defocus level, laser ablation power, and wafer feed speed. For the laser defocus height study, six levels of laser focus were considered: top surface level (0%), followed by -20%, -40%, -60%, and -80% levels down from the top surface. Additionally, the bottom level of wafers, corresponding to -100% from the top surface, was studied for both 50 µm and 75 µm thick wafers. The study used constant laser power and measured the fastest wafer feed speed required to complete the full cut. Results in figure 6 indicated that reducing the laser defocus level led to faster wafer feed speed to complete the full cut wafer process. Specifically, for both 50 µm and 75 µm thick wafers, there was an approximate increment of 2.09 mm/s and 1.37 mm/s in the wafer’s feed speed for every µm reduction in defocus level.

Fig. 6: Graph of fastest wafer’s feed speed in different laser defocus levels.

In addition to measuring the wafer feed speed as an output, the kerf width size is also evaluated. The results are shown in figure 7. The study utilized six levels of laser defocus, considering three different conditions: wafer type, laser power, and feed-speed differences. Across all conditions, it was observed that the kerf width size decreased when the laser defocus was set at a lower level. Specifically, in this experiment, the maximum and minimum kerf width sizes were 11.2 µm and 4.9 µm, respectively, when the defocus was set at the bottom of the wafers.

Fig. 7: Graph of kerf width versus laser defocus level in three factors: wafer type, laser power, and wafer’s feed speed.

The next study focused on evaluating massive recast defects or burrs, specifically comparing laser defocus to wafer feed speed. In this experiment, a maximum laser power is applied to simulate the worst-case scenario for massive recast defects. Figure 8 illustrates that slow and ultra-slow wafer feed speeds, regardless of the laser defocus level, result in massive recast defects. Only average feed speeds show no massive recast defects. Furthermore, slower wafer feed speeds and lower laser defocus levels exacerbate the massive recast condition.

Fig. 8: Graph of wafer’s feed speed versus laser defocus level.

As a rule of thumb, the laser defocus level should always be set at the bottom of the wafer during the laser ablation process. When the laser defocus level is positioned at the wafer’s bottom, the wafer can be cut deeper, allowing for increased feed speed and improved productivity. Additionally, this deeper cut results in a narrower kerf width. However, it is important to avoid excessively slow wafer feed speeds combined with a low laser defocus level, as this could potentially lead to massive recast issues on the side of the die. The combination of laser power and wafer feed speed parameters is discussed further in the threshold limit study session.

Feasibility study of sawing ultra-thinned wafers

A feasibility study was conducted on 30 µm thick wafers. A wide range of laser power levels from the lowest to the highest were used to dice the 30 µm thick wafers based on insights from previous sessions. Despite the absence of massive recast defects and successful cutting of all experimental wafers, burn mark defects were observed, as shown in figure 9. These defects persisted across the entire range of laser power settings. However, reducing the laser power may mitigate the severity of the burn marks.

Fig. 9: Burn mark photos of the impact of different laser power levels from lowest to highest on a 30 µm thick wafer.

However, burn mark defects can be resolved by optimizing the PVA coating process as shown in figure 10. By fine-tuning the coating duration and rotation per minute (rpm), the burn mark issue can be effectively addressed.

Fig. 10: Burn mark photos in before and after PVA coating process optimization.

In addition to the 30 µm thick wafer feasibility study, a quick check of laser ablation on kerf width was performed. An actual 50 µm thick power discrete wafer with a saw street width of 68 µm was used. An offset laser ablation full-cut sawing process was used to simulate a 30 µm saw street. Results indicate that no chipping occurs during offset laser ablation sawing, as shown in figure 11.

Fig. 11: Photo of quick check study on 30 µm saw street. Simulated by performing offset laser ablation sawing.

Threshold limit study

This study aims to establish a process window for laser ablation by examining the interaction of two critical process parameters: laser power and wafer feed speed. The goal of this study is to define their threshold limits using 30 µm, 50 µm, and 75 µm thick wafers as samples. Figure 12 displays the relationship between wafer feed speed and laser power. It highlights the threshold limit for full cut sawing of wafers with thicknesses of 30 µm, 50 µm, and 75 µm using laser ablation technology. Considerations include machine limitations, such as maximum wafer feed speed and laser power stability, as well as addressing the massive recast issue. Notably, this graph not only defines the potential process window for these three wafer thicknesses but also demonstrates that laser ablation is capable of sawing wafers ranging from 30 µm to 75 µm thick using the full-cut method.

Fig. 12: Threshold limit of laser ablation technology on 3 different wafer thicknesss: 30, 50, and 75 µm.

Die strength and improvement study

The next study focuses on die strength, which is a critical consideration in dicing technology. For this experiment, actual 50 µm thick power discrete die with back metal deposited were used. All die strength results mentioned in this study were measured using a three-point die strength tool, both on the top-side surface (active structure up) and the bottom-side back metal (structure up). An example photo of the three-point die strength measurement tool is shown in figure 13.

Fig. 13: Three-point die strength measurement tool.

In this study, a total of three sets of actual die were examined. These dies were processed using laser ablation, both with and without process optimization, as well as blade dicing as control. Each set consisted of 20 pieces. Figure 14 shows the experimental results.

From the topside die strength test in figure 14A, blade diced samples had an average die strength of 829 MPa and laser diced samples before optimization exhibited lower die strength, averaging only 159 MPa. However, after optimization, the laser-diced die strength increased to 290 MPa. This represents an 82% improvement compared to the non-optimized laser-diced samples, but it is still only 32.6% of the blade-diced sample strength.

In the bottom side die strength test shown in figure 14B, blade-diced samples showed an average die strength of 824 MPa. Without laser process optimization, the diced samples had an average die strength of 429 MPa, representing only 52% of the blade-diced sample strength. Remarkably, the laser-optimized samples significantly improved die strength, reaching an average of 877 MPa. This represents a 104% improvement compared to the non-optimized samples. Interestingly, the bottom side die strength of the laser-optimized diced sample is comparable to that of the blade-diced samples, with the laser-optimized sample having a slightly higher average die strength of 5.8%.

A) Top side die strength test in 3 different die diced conditions.

B) Bottom side die strength test in 3 different die diced conditions.

Fig. 14: Box plot graphs on the distribution of die strength under different dicing method for samples of a) top side and b) bottom side structure up.

Laser ablation versus blade dicing

The last study compares the laser ablation and blade dicing manufacturing capabilities and estimated operation cost differences by using actual 50 µm power discrete wafers with die size of 3.2 x 1.75 mm².

Table 1 presents the manufacturing capability comparison between blade dicing and laser ablation. Specifically, blade diced die yielded a kerf width of 27 µm, while laser ablation resulted in a narrower 15.4 µm kerf width. This represents a 57% difference favoring laser ablation. Additionally, laser ablation exhibited smaller top-side chipping which is average 2.3 µm compared to blade dicing 6.2 µm in average. The bottom-side chipping for laser ablation was 13.8 µm, smaller than the blade diced counterpart which was 24.3 µm. Notably, laser ablation samples showed no side wall chipping, whereas blade-diced samples did exhibit such chipping. Overall, laser ablation outperformed blade dicing in terms of chipping criteria.

From a productivity perspective, laser ablation could complete wafer sawing in 3 to 5 minutes. In contrast, blade dicing would take approximately 25 minutes, making laser ablation roughly 6.3 times faster in terms of productivity. Regarding die strength, laser ablation diced samples had only 32.6% of the die strength of blade-diced samples in the top side surface (active structure up) measurement. However, when measuring the bottom side back metal structure up, laser ablation becomes comparable to blade dicing if the laser ablation process is optimized.

Items	Blade	Laser Ablation
Manufacturing Quality
-Kerf width (µm)	27	15.4
-Top side chipping (µm)	6.2	2.3
-Bottom side chipping (µm)	24.3	13.8
-Side wall chipping/peeling	Fail	Pass
Productivity (per wafer)	25 min	3-5 min
Die Strength
-Top surface active	829 MPa	290 MPa
-Bottom back metal	824 MPa	877 MPa

Table 1: Blade versus laser ablation in dicing capability.

Although laser ablation offers superior capabilities in terms of chipping quality and productivity, it also comes with higher operational costs. Laser ablation dicing incurs approximately 2 to 3 times the operation cost due to the need for laser ablation dicing tape, PVA coating material, and considerations of laser head deterioration over time in contrast, blade dicing has lower operational costs, even when using ultraviolet (UV) dicing tape and dicing blades for comparison.

Laser ablation dicing technology is specifically designed for ultra-thin wafer sawing, addressing limitations associated with blade dicing, such as side wall chipping. To achieve optimal laser ablation dicing performance, process characterization and optimization are essential.

Process characterization involves determining the ideal laser defocus level, which is typically at the bottom of the wafer. If the wafer’s feed speed is too slow, it can lead to significant recast or burr issues. Laser ablation has been demonstrated to successfully dice wafers with thicknesses ranging from 30 µm to 75 µm in the full cut method, as confirmed by threshold limit studies. Additionally, a quick assessment indicates that laser ablation should not encounter issues when dicing 30 µm saw street wafers.

Comparing actual dice produced by blade dicing and laser ablation, laser ablation results in smaller chipping on both the top and bottom sides of the die, as well as a narrower kerf width. Laser ablation also offers higher productivity, but it also comes with increased operational costs compared to blade dicing. While the die strength of laser-diced samples is lower than that of blade-diced samples, optimizing the laser ablation process can improve the bottom-side die strength providing comparable performance as blade-diced units.

References

1. H. Fan,T. Boettcher, C. Li, E. Cheam, J. Yang and H. Chen, “Solder Crack Improvement For A Power Package During TCT And PTC By Simulation,” 21st International Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems (EuroSimE) , 2020.
2. J. Li, H. Hwang, E. C. Ahn, Q. Chen, P. Kim, T. Lee, M. Chung and T. Chung, “Laser Dicing and Subsequent Die Strength Enhancement Technologies for Ultra-thin Wafer,” 57th Electronic Components and Technology Conference, 2007, doi: 10.1109/ECTC.2007.373883
3. F.A.T. Schraub, “Automated Handling of Ultra-Thin Silicon Wafers,” Solid State Technology, vol.45, no.9, pp. 59–64, 2002.
4. W. S. Lei, A. Kumar and R. Yalamanchili, “Die singulation technologies for advanced packaging: a critical review,” Journal of Vacuum Science & Technology B, vol.30, no.4, pp. 40801(1)–40801(27), 2012.
5. M. R. Marks, Z, Hassan and K. Y. Cheong, “Ultrathin wafer pre-assembly and assembly proces,s technologies: a review,” Crit Rev Solid State Mater Sci, vol. 40, no. 5, pp.251–90, 2015.
6. W. Kroninger, “Thin die production,” in: Lu D, Wong CP, editors. Materials for advanced packaging. New York: Springer; ch. 6, pp. 219–235. 2009.
7. W. Kroninger and F. Mariani, “Thinning and singulation of silicon: root causes of the damage in thin chips,” 56th Electronic Components and Technology Conference (ECTC). pp. 1317–1322, May/June 2006.
8. M. R. Marks, K. Y. Cheong and Z. Hassan, “A review of laser ablation and dicing of Si wafers,” Science Direct, Precision Engineering, vol. 73, pp. 377-408, January 2022, doi: 10.1016/j.precisioneng.2021.10.001.
9. D.Y.R. Chong, W.E. Lee, B.K. Lim, J.H.L. Pang and T.H. Low, “Mechanical Characterization in Failure Strength of Silicon Dice,” Inter Society Conference on Thermal Phenomena, pp.203-210, 2004, doi: 10.1109/ITHERM.2004.1318284.
10. C. M. Dunsky, “Laser material processing in microelectronics manufacturing: Status and near-term opportunities” Proceedings of SPIE, doi:10.1117/12.600913.
11. V. R. Marinov, O. Swenson, Y. Atanasov and N.Schneck N, “Laser-assisted ultrathin die packaging: Insights from a process study,”  Microelectronic Engineering, vol. 101, pp. 23-30, January 2013.
12. M. H. Hong, Q. Xie, K. S. Tiaw and T. C. Chong, “Laser singulation of thin wafers and difficult processed substrates: a niche area over wafer sawing,” Journal Laser Micro, vol. 1, no.1, pp. 84–8. 2006.
13. H. Theuss, A. Koller, W. Kröninger, S. Schoenfelder and M. Petzold, “Assessment of a Lasersingulation Process for Si-Wafers with Metallized Back Side and Small Die Size,” 58th Electronic Components and Technology Conference, May 2008, doi: 10.1109/ECTC.2008.4550178.

Facebook Comments Box