Low Temperature Curable Polyimide Film Properties and WLP Reliability Performance with Various Curing Conditions

2017 IEEE 67th Electronic Components and Technology Conference Low Temperature Curable Polyimide Film Properties and WLP Reliability Performance with Various Curing Conditions Yu Chuan Chen (Steven), Katch Wan, Chen An Chang and Rick Lee Siliconware Precision Industries Co., Ltd No. 19, Keya Rd., Daya,Taichung, Taiwan 428, R. O. C. E-mail: stevenchen4@spil.com.tw, phone: +886-911-850847 Abstract Polyimide (PI) is a curable polymeric dielectric, which is a popular material for wafer level packages (WLP). Low-curable-temperature polyimide (LTP) attracts increasing interest in assembly applications and is one of the critical materials for 2.5D IC and fan-out (FO) applications because of its curing temperature constraint during the packaging process. As the possible curing temperature is approached, the curing level percentage is an important index of the polymer film. Herein, we aimed to study the thermal properties of PI under various curing conditions (curing level 30 100%) and choose the curing conditions (curing level 80 100%) to apply on WLP for component-level reliability (CLR), chemical resistance and board-level reliability (BLR) tests. According to the CLR results, the lower curing level (about 80%) passes scanning acoustic tomography (SAT) inspection for the 1000-h temperature humidity test (THT), the 96-h u-bias highly accelerated stress test (uhast), and the 1000-h high-temperature storage test (HTST) despite its lesser thermal stability and chemical resistance than the fully cured In the BLR test, ball drop test results show no gap between the 80% and 100% curing level conditions. In the thermal cycle test (TCT) the lower curing level gives a better TCT result. Finally, since reliability issues minimize chemical resistance capability and thermal process reduction, curing level criteria might be redefined in the polyimide curing process. Keywords Curing level, Polyimide, Chemical resistance, BLR Introduction In the semi-conductor industry, aromatic polymers are widely applied in high-temperature processes and those involving acids and bases because of their excellent chemical stability [1-3]. With the outstanding mechanical properties of resistance to cracking under high temperature, an aromatic organic polymer is often used as the dielectric layer in the bumping packing process. Especially in wafer level packing (WLP) structures, the organic dielectric layer has been a widely studied topic with its properties of low dielectric constant, high ductility, radiation resistance, high glass transition temperature, high chemical resistance, and alpha particle barrier [4]. Polymeric dielectrics have different molecular structures and functional groups, and thus different curing conditions and film properties. For better metal adhesion, polyimide (PI) is of interest in ewlp and 2.5DIC packaging. For thermal resistance and stress in fan-out (FO) and 2.5DIC products (prepared on epoxy and glass substrates), seeking a lower curing temperature with shorter operation time is essential for PI material development [5-8]. Nowadays, studies of commercial low-temperature PI material are based on fully cyclized PI (cyclization = 100%) to measure the curing condition and analyze the properties of the PI film; there is no discussion on the properties of thin films with lower cyclization and the impact on reliability in real wafer level chip scale package (WLCSP) products. Hence, these studies focus on low-temperature-curable polyimide (LTP) under different curing conditions, analyze film properties, and apply the films on WLP products for reliability studies. Experimental Method In this study, a 50-μm thin film was coated on a 12" wafer and cured under different conditions (i.e. table 1), and soaked in 20% hydrofluoric acid (HF) for about 60 min. The thin film sample was washed with deionized water and dried under nitrogen prior to analysis. During polyimide cyclization analysis, a 20 mm 20 mm thin film was used to measure the intensity of the absorbance spectrum of different functional groups using Fourier-transform infrared spectroscopy (FTIR); the polyimide cyclization rate was estimated from the signal strength ratio of the amide to benzene ring. The thermal stability was estimated by observing the loss of weight of sample (original weight 5 mg) during heating from room temperature to 700 C using a thermogravimetric analyzer (TGA). The glass transition temperature is determined from the amount of stretch in the heating process by thermomechanical analysis (TMA). For reliability testing, this study chose three different PI curing conditions and adopted a SAC405 commercial 250-mm solder ball to a WLP structure sawed into 8 mm 5 mm components and mounted to 132 mm 77 mm 1 mm printed circuit board (PCB) (certified JEDEC standard) for surface mount technology (SMT) process followed by board-level reliability (BLR) tests like the drop test and the thermal cycle test. The die component was also tested by component-level reliability (CLR) tests like u-hast (96 h), THT (1000 h), and HTST (1000h) after pre-conditioning; the reliability conditions are shown in table 2. Different samples were soaked in the common alkaline solution used in the semiconductor packaging process, 1-methyl-2-pyrrolidone (NMP), and adjusted to 50 C (temperature required by the packaging process) to observe (every 60 s) if there were any abnormalities on the PI surface like delamination or cracks. The study also adopted the pinhole test the toughest condition in chemical resistance testing known in semi-conductor processing to observe the surface change of samples soaked in 10% KOH solution at 100 C for different durations. All the test and judgment criteria in the analysis are shown in table 3. 2377-5726/17 $31.00 2017 IEEE DOI 10.1109/ECTC.2017.337 2040

Table 1 PI curing condition in this study. Curing Condition Time Temperature High Middle Low Long A-1 B-1 C-1 Middle A-2 B-2 C-2 Short A-3 B-3 C-3 Table 2 BLR and CLR reliability test Item Condition Method Result and Discussion PI film property analysis of different curing conditions Figure 1 shows the PI cyclization results under different curing conditions. Irrespective of curing time, PI film can achieve 100% cyclization under condition A, which is the highest curing temperature. In contrast, regardless of curing time, the cyclization of PI can reach only 30% (same as uncured PI) under curing condition C, which is the lowest curing temperature. Under curing condition B (curing temperature was between those under conditions A and C), the degree of cyclization and curing time are in direct proportion. BLR Drop 1000count TCT 1000cycle 1500g ; 0.5ms -40 ~ 125 JESD22-B104C Condition B JESD22-A104G, Condition G Pre- Condition CLR Bake Soak Reflow u-hast 1000hours THT 1000hours HTS 1000hours 125 85 85%RH 260 130, 85%RH J-STD-020 JESD22-A118 85 85%RH -- 150 JESD22-A103 Table 3 Analysis (test) item and index in this study. Film Property Test Item FTIR TGA TMA Index Cyclization 5 wt % lose temp. Tg Figure 1 PI cyclization rate of different PI curing Figure 2 shows that PI cyclization (polymerization) is a molecular dehydration reaction. According to collision theory [8], when the reaction temperature is too low and cannot reach the threshold energy, reaction particles cannot react even with prolonged operation time (i.e. curing conditions C1, C2, and C3). When the temperature is increased to reach the threshold energy, longer curing time increases the degree of completion. Further, the Arrhenius equation suggests that the reaction rate will increase significantly when the temperature is higher; this means that PI can complete cyclization in a very short time. This also describes that under the curing condition A, PI can complete cyclization in both the shorter duration A1 and the longer duration A3. Reliability Test Chemical Resist BLR- Drop BLR- TCT CLR- uhast CLR- THT CLR- HTST 50,NMP 100, 10% KOH Electrical test (Resistance 1000 ) SAT and SEM (No de-lamination) Optical Microscope (No surface abnormal) Figure 2 Reaction mechanism of PI cyclization. Figure 3 shows the glass transition under different curing conditions. The glass transition temperature was lowest at C3 which had the lowest curing temperature and shortest time and increased with increasing curing temperature or duration. The highest glass transition temperature was found at A1 (highest PI curing temperature and longest curing time). 2041

Figure 3 TMA result of different curing condition The TGA studies found that PI decomposition temperature is higher (more stable) when the curing temperature is higher or the curing time is longer, shown in figures 4 and 5. Taking the decomposition temperature of 5% weight loss as index, in the A series (higher temperature), there is no significant correlation between decomposition temperature and curing time (decomposition temperature increases 10 C when the PI curing time increases by 10 times). In the B series, the decomposition temperature is significantly related to the operation time (decomposition temperature increases by 70 C when the PI curing time increases by 10 ). In the C series (lowest temperature), the decomposition temperature and operation time had a positive relationship, but the relevance is significantly less than for the B series (decomposition temperature increases by 20 C when the PI curing time increases 10 times). As shown in figure 6, comparing the 5 wt% decomposition temperature, glass transition temperature, and cyclization, when the cyclization reaches 100%, even though the curing time of A1 is 10 times that of A3, the decomposition temperature and glass transition temperature are similar (the difference is less than 3%). When the B1 condition (lower curing temperature and cyclization rate is only 80%) was compared with the A condition (cyclization rate is 100%), the difference between 5 wt% decomposition temperature and glass transition is less than 5%. This implies that blind pursuit of a higher cyclization rate is not an appropriate index of thermal stability in a PI material. Figure 6 5 weight, Tg and cyclization result summary of different curing condition. Reliability test for different curing conditions in WLP products Different thicknesses of the dielectric layer have an influence on the reliability of WLP products. Before reliability testing, the PI and metal RDL thicknesses were measured after the bumping process in a WLP product (i.e. figure 7); it was found that there is no significant difference in PI thickness and profile due to the different curing conditions. Curing Conditiom x400 x2000 A2 Figure 4 TGA result of different curing A3 B1 Figure 5 Weight lose from 99 to 94 of TGA result PI1 10 um THK. PI2 8 um Figure 7 X-section SEM inspection before reliability test. 2042

Figure 8 is the Weibull distribution of the BLR drop test, the characteristic life of the A2 condition (2600 counts) is better than the A3 condition, which is at the same curing temperature but longer curing time than A2. B1, which has a lower curing temperature than A2 and A3, has a drop result similar to A3. Based on failure sample analysis after drop test, the result is shown as figure 9. The failure modes are caused by the cracking of the metal gasket at the PKG end; this is also the classical drop failure model in WLCSP products. 899 cycles, respectively. When analyzing the failure model using cross-section scanning electron microscopy (SEM) images (figure 11), the failure modes of all conditions are the same: cracking in the middle of the solder ball. This is a common failure model of thermal cycle in WLCSP products. Figure 10 TCT weibull distribution of different curing Figure 8 Drop test weibull distribution of different curing Figure 9 Drop test failure analysis of different for different PI curing Figure 10 shows the Weibull distribution of the board-level thermal cycling test. The characteristic life of B1 (lower curing temperature) is the best at 970 cycles, and those of A2 and A3 (higher curing temperature) are 839 and Figure 11 TCT test failure analysis of different curing 2043

As shown in table 4, the characteristic life of sample A2 is 3.5 times that of A3. Although all the A series samples reached 100% cyclization, the A2 result is much better than A3. The B1 sample reached only 80% cyclization but had the same characteristic life as A3. In the board-level thermal cycling test, B1 (with lowest cyclization) had the best characteristic life of ~970 cycles, greater than A2 (839 cycles) and A3 (900 cycles). The above results prove that curing conditions and cyclization do not always have a positive relation to reliability. Table 4 BLR and CLR reliability test result. Characteristic life PI curing condition Cyclization Drop TCT No. Temp. Time. Test A2 High Middle 100% 2600 893 A3 High Short 100% 745 900 B1 Middle Long 83% 739 970 In addition to the BLR test, we also applied CLR tests like u-hast, THT, and HTST tests, and inspected the interface between RDL and the upper and lower PI by SAT and cross-section SEM. Figure 12 to 14 shows SEM images obtained at different magnifications after the reliability test. According to the SEM images, only the IMC layer between tin-silver and copper plating increased slightly, the interface of the plating RDL and PI layers still adhere closely without any de-lamination failure. Figure 13: X-section after THT 1000 hours of different curing Figure 12: X-section after u-hast 96 hours of different curing Figure 14: X-section after HTST 1000 hours of different curing 2044

Chemical resistance under different curing condition in WLP products It has been known that PI after full cyclization has superior chemical resistance against acid. It is only susceptible to decomposition by strong base; therefore, strong base is used to etch PI in industry. Hence, this study adopted a rigorous pinhole test, which involves soaking different samples in 10 wt% KOH solution at 100 C to evaluate the chemical resistance capability. After the experiment, it found that B1 (cyclization rate 80%) shows significant delamination and discoloration after soaking for 180 s. In other samples where cyclization reached 100%, cracks emerged after soaking for 300 s, inspection result shown in figure 15. It also shows that the higher degree of cyclization and increased aromaticity of PI make it more stable, i.e. less susceptible to decomposition by base. Figure 15 Pin hole test of different PI curing Further, to certify different curing conditions that can be applied in the WLP process. We chose a base that is often used in the bumping process (NMP at 50 C) and observed the change in the PI surface after different soaking times. The result is shown in figure 16. There is no cracking and delamination for A2, A3, and B1 (cyclization 80% to 100%) when soaked for 360 s. Besides, in the actual bumping process(i.e. figure 17), the PI layer covered by a metal layer would not come into direct contact with the base. This means that the PI layer having cyclization rate 80% can survive the base etching in the real packaging process. Figure 16 Chemical resist result of 50 NMP Figure 17 Process comparison between chemical resist experiment and real bumping process Conclusions In WLP products, the choice of low-temperature polymer dielectric layer is often determined by the cyclization rate; thus, over-curing often occurs in the process. Although it has no impact on the qualities of products, it will be a disadvantage in terms of the cost and the pursuit of reducing the thermal process. This research focuses on low-temperature-curable PI for different curing conditions and analyzes film properties and test reliability for WLP products. The results are as follows: 1. In terms of film properties, low-temperature-curable PI has good thermal stability when cyclization reaches 80%. Irrespective of curing time, there is no obvious impact on thermal stability when PI cyclization reached 100%. 2. The characteristic lives of 100% and 80% cyclized samples exceed 700 counts in the board-level drop test, and cyclization does not have a positive correlation with the result of the board-level temperature cycling test. 3. PI has the same level of chemical resistance when cyclization reached 100%. When the cyclization rate is 80%, although the sample has chemical resistance of just 180 s in 10 wt% KOH solution at 100 C (shorter than 300 s when the cyclization rate is 100%), it can still resist decomposition in the base that is used in the bumping process. According to the above results, it is not necessary to choose 100% cyclization or a low-temperature-curable PI to lower the thermal stress in the total process. Cyclization at 80% can pass the bumping process when there is no particularly rigorous base material involved in the process. References [1] Dennis E. Curtin, Advanced materials for improved PEMFC performance and life, Journal of Power Sources. 131 (2004) 41 48. [2] M. K. Ghosh and K. L. Mittal, Ed., Polyimides: Fundamentals and Applications, Marcel Dekker, New York, N.Y., 1996.2. [3] M. I. Bessonov and V. A. Zubkov, Ed., Polyamic Acids and Polyimide: Synthesis, Transformation, and Structure, CRC Press, Boca Raton, Florida, 1993. [4] M. Toepper, T. Fischer, A comparison of thin film polymers for Wafer Level Packaging, in Electronic Components and Technology Conference, Proceedings 60th, Las Vegas, NV, USA, 2010. [5] R.L. Hubbard, Z. Fathi, J. Wander, T. Hattori, H. Matsutani, T. Ueno, C.E. Schuckert, Low Temperature Cure of Aromatic Polyimides, Symposium on Polymers for Microelectronics, Winterthur, DE, May 5-7 2003. 2045

[6] R.L. Hubbard, Z. Fathi, I. Ahmad, T. Hattori, Low Temperature Curing of Polyimide Wafer Coatings, International Electronics Manufacturing Technology Symposium, San Jose, CA, 2004. [7] Y.K. Lee and J.D. Craig, Polyimide Coatings for Microelectronic Applications, Polymer Materials for Electronic Applications, ACS Symposium Series, Washington, 1982. [8] Bui, Matthew. "The Arrhenius Law: Activation Energies". Chemistry LibreTexts. UC Davis, 2017 2046