Development of 3D Wafer Level Hermetic Packaging with Through Glass Vias (TGVs) and Transient Liquid Phase Bonding Technology for RF Filter

The development of 5G mobile communication created the need for high-frequency communication systems, which require vast quantities of radio frequency (RF) filters with a high-quality factor (Q) and low inband losses. In this study, the packaging of an RF filter with a through-glass via (TGV) interposer was designed and fabricated using a three-dimensional wafer-level package (3D WLP). TGV fabrication is a high-yielding process, which can produce high precision vias without masking and lithography and reduce the manufacturing cost compared with the through silicon via (TSV) solution. The glass interposer capping wafer contains Cu-filled TGV, a metal redistribution layer (RDL), and the bonding layer. The RF filter substrate with Au bump is bonded to the capping wafer based on Au-Sn transient liquid phase (TLP) bonding at 280 °C with a 40 kN (approximately 6.5 MPa) bonding force. Experimental results show that shear strengths of approx. 54.5 MPa can be obtained, higher than the standard requirement (~6 MPa). In addition, a comparison of the electrical performance of the RF filter package after the pre-conditional level three (Pre-Con L3) and unbiased highly accelerated stress (uHAST) tests showed no difference in insertion attenuation across the passband (<0.2 dB, standard value: <1 dB). The final packages passed the reliability tests in the field of consumer electronics. The proposed RF filter WLP achieves high performance, low cost, and superior reliability.


Introduction
As more components are incorporated into radio frequency (RF) front-end modules, devices are more densely packed, and more heat is produced in the modules. The integration of RF filters in the front-end module demands further miniaturization and higher temperature stability [1].
RF filter demands excellent frequency band selectivity, high quality, and low insertion loss. Surface acoustic wave (SAW) and bulk acoustic wave (BAW) filters become the essential technique route. SAW filter combines low insertion loss with good suppression performance. However, it typically works below 1.5 GHz-band and is easily affected by temperature [2]. BAW filters are favored in many high-frequency applications due to their excellent frequency band selectivity, especially in 4G and 5G communication terminals. A BAW filter is usually packaged using thin-film cavity acoustic resonator (FBAR) technology. Its typical structure is two metal electrodes clamped to the piezoelectric film [3]. The piezoelectric film is sensitive to any additional mass loading on the surface, such as humidity or any corrosion. Thus, efforts in packaging are increasingly important. For propagation of the bulk wave, the package requires a cavity structure above the chip surface to prevent moisture corrosion and provide a stable/safe environment [4]. In addition, the parasitic effects of packaging for RF or microelectromechanical system (MEMS) devices should be minimal. based on laser-induced deep etching (LIDE) technology. It can produce high-precision vias without masking and lithography and reduce the manufacturing cost compared with the TSV fabrication. And glass material is very attractive because it exhibits excellent hermetic performance and doesn't cause outgassing compared to film lamination packaging. The glass is transparent, and we can track the packaging yield in real-time during processing. It is not advantageous for tracking the packaging yield for wafer-level packaging of RF MEMS devices by transfer bonding of silicon caps, especially after the bonding process. Therefore, it is commonly used in MEMS or RF field WLP. Yang et al. [27] presented a MEMS fabrication process with TGVs by laser drilling technology, and reliability concerns were overcome during the whole packaging process. Lee et al. [28] presented a waferlevel RF MEMS packaging structure with Cu-filled TGVs, and mechanical reliability was confirmed through a thermal shock test. However, studies on the 3D WLP with TGVs vertical feedback and cavity hermetic for RF filter were rarely reported in the available literature. Therefore, we are working on developing a 3D WLP for RF filter with a low cost, high performance, and superior reliability.
In this paper, we present a novel process development of wafer-level hermetic packaging with TGVs structure for RF filters, especially FBAR devices, which allows the cointegration of on-chip passive devices. To seal the internal structure, the packaging uses a TLP bonding with Au-Sn solder. A closed square loop of the Au-Sn bumping ring is fabricated at the chip's edge area to make a cavity. The device's pads inside the ring are used to support the weight of the back end of the line (BEOL) and the I/O interconnection. In Section 2, we propose the RF filter WLP structure and optimize the bonding structure. Section 3 introduces the TGV fabrication and bonding process for RF filter 3D integration. Section 4 discusses reliability evaluation strategies and the final packaging of RF filters using 3D/TGV technologies. Lastly, we conclude with some important findings and remarks in Section 5. Figure 1a shows the schematic of the proposed AlN RF filter packaging structure. The device component, which has a size of 720 µm × 545 µm with a thickness of 300 µm (with ball), is composed of a two-part glass cap and device wafer. They are bonded through a closed square loop of Au-Sn soldering ring at the chip's edge area using a wafer bonder (EVG 520). Full filling of the TGVs and metal traces were formed on both sides of the glass. Meanwhile, in Figure 1b, four metal pads inside the seal ring are used for I/O interconnection between the device and external signal source via the TGVs and RDL. The width of the Au and Sn layer is set to 43/27 µm, respectively. The cap wafer is an 8-inch glass wafer with a thickness of 100 µm. It has an internal cavity array with a depth of 20 µm to provide clearance for the device to work properly. TGV is fabricated by LIDE and filled by copper electroplating for electrical through. We retained the Au-Sn wafer bond proven to provide reduced size and a robust hermetic seal.

Optimization of the Thickness of the Sn and Ni Layer
As the bonding process is fluxless, systematic experimental work has been carried out to study the effect of thickness of the Sn layer in the seal ring and bump metallization on the reliability of the RF filter package. Firstly, too thick seal ring closed square will result in molten Sn overflow from the edge into the die area, which will result in a short circuit. Conversely, a too-thin Sn layer will cause poor wetting, affecting the bonding strength. Figure 2 shows the cuboid micro-joint, the effective Au concentration, set at 1.5 µm, and all Au atoms are dissolved in the solder, which can be calculated by Equation (1) [29].

Optimization of the Thickness of the Sn and Ni Layer
As the bonding process is fluxless, systematic experimental work has been carried out to study the effect of thickness of the Sn layer in the seal ring and bump metallization on the reliability of the RF filter package. Firstly, too thick seal ring closed square will result in molten Sn overflow from the edge into the die area, which will result in a short circuit Conversely, a too-thin Sn layer will cause poor wetting, affecting the bonding strength. Figure 2 shows the cuboid micro-joint, the effective Au concentration, set at 1.5 μm, and all Au atoms are dissolved in the solder, which can be calculated by Equation (1) Figure 2. The simplified Au/Sn/Ni/Cu micro-joint. Here, C Au is the effective Au concentration. d Sn and d Au are the thickness of the Sn and Au, respectively. ρ Au and ρ Sn are the densities of the Au and Sn, respectively. The critical thickness of the Sn layer is set to 3 µm, 5 µm, 6 µm. Applying Equation (1), the effective Au concentration C Au : To control the diffusion process, a thin Ni-buffer layer was introduced into the bonding structure. The Ni layer has two main effects. First, it can prevent fast diffusion between the low-temperature Sn and Cu components during storage and the step of heating up during the bonding process. Second, the thin buffer layer dissolves into the Sn at the beginning of the soldering reaction. Then the diffusion between the solder materials and the Cu started, and finally, all solder was converted into intermetallic compounds (IMCs). As for improving reliability, control of Ni layer thickness becomes very important. The thickness of the Ni layer varied from 0.35~2.8 µm, while the Au layer was set at 1.5 µm to simulate the interfacial reactions during solid-state aging. We found that all the layers showed a decreasing trend in the shear strength during the early stage. Meanwhile, for the thinner Ni layer, the shear strength decreased to a lesser extent [30]. Based on the actual packaging process, the actual Sn layer thickness was set to 5 µm. Figure 3 shows the cross-sectional image of Au/Sn/Ni/Cu micro joint with 5 µm Sn layer after the bonding process. It can be seen that Au is consumed and formed IMCs. The Au effective concentration is 44.23 wt.% in the seal ring, and bumps if complete mixing is assumed. To control the diffusion process, a thin Ni-buffer layer was introduced into the bonding structure. The Ni layer has two main effects. First, it can prevent fast diffusion between the low-temperature Sn and Cu components during storage and the step of heating up during the bonding process. Second, the thin buffer layer dissolves into the Sn at the beginning of the soldering reaction. Then the diffusion between the solder materials and the Cu started, and finally, all solder was converted into intermetallic compounds (IMCs). As for improving reliability, control of Ni layer thickness becomes very important. The thickness of the Ni layer varied from 0.35~2.8 μm, while the Au layer was set at 1.5 μm to simulate the interfacial reactions during solid-state aging. We found that all the layers showed a decreasing trend in the shear strength during the early stage. Meanwhile, for the thinner Ni layer, the shear strength decreased to a lesser extent [30]. Based on the actual packaging process, the actual Sn layer thickness was set to 5 μm. Figure 3 shows the cross-sectional image of Au/Sn/Ni/Cu micro joint with 5 μm Sn layer after the bonding process. It can be seen that Au is consumed and formed IMCs. The Au effective concentration is 44.23 wt.% in the seal ring, and bumps if complete mixing is assumed. In Table 1, energy-dispersive X-ray spectroscopy (EDX) analysis shows that the effective Au concentration is about 42 wt.% to 62.3 wt.%. According to the AuSn binary phase diagram [30], the Sn layer on the capping wafer is almost consumed, and the final microstructure should consist of Au5Sn ( ς ′ phase, as shown in spectrum 4) + AuSn ( δ phase, as shown in spectrum 1) + AuSn2 (ε phase, as shown in spectrum 2, 3). There is no void at the interface between the Ni layer and the IMC layer. Meanwhile, Ag is detected in EDX shown in Table 1 (see spectrum 1 and spectrum 2), the main reason being that the electroplating liquid contains "Ag" impurity.  In Table 1, energy-dispersive X-ray spectroscopy (EDX) analysis shows that the effective Au concentration is about 42 wt.% to 62.3 wt.%. According to the AuSn binary phase diagram [30], the Sn layer on the capping wafer is almost consumed, and the final microstructure should consist of Au5Sn (ς phase, as shown in spectrum 4) + AuSn (δ phase, as shown in spectrum 1) + AuSn2 (ε phase, as shown in spectrum 2, 3). There is no void at the interface between the Ni layer and the IMC layer. Meanwhile, Ag is detected in EDX shown in Table 1 (see spectrum 1 and spectrum 2), the main reason being that the electroplating liquid contains "Ag" impurity.

Packaging Process
The fabrication processes of the test vehicle for RF filter packaging involve a TGV cap wafer which includes transmission lines and RF filter fabrication process on the silicon substrate. The characteristic of RF filter WLP is determined by an accurate delta value that should not be greater than 1 dB in peak insertion loss (IL) of the passband. Figure 4 illustrates the process flow of the present 3D WLP for the RF filter. In this process flow, key processes include TGV cap wafer formation with transmission lines and through vias interconnection, Au-Sn TLP bonding, under bump metallization (UBM) formation, and solder ball.

Packaging Process
The fabrication processes of the test vehicle for RF filter packaging involve a TGV cap wafer which includes transmission lines and RF filter fabrication process on the silicon substrate. The characteristic of RF filter WLP is determined by an accurate delta value that should not be greater than 1 dB in peak insertion loss (IL) of the passband. Figure 4 illustrates the process flow of the present 3D WLP for the RF filter. In this process flow, key processes include TGV cap wafer formation with transmission lines and through vias interconnection, Au-Sn TLP bonding, under bump metallization (UBM) formation, and solder ball.  Figure 5 shows the fabricating process flow of the dummy wafer. Ti/W is sputtered on the silicon substrate as adhesive & seed layer. An organic photoresist 6μm thick is spincoated and patterned. Figure 6a,b show that the pattern of the dummy device wafer and the opening of the seal ring and bump are 43.5 μm and 85 μm, respectively. In Figure 5 (4,5), 1.5 μm Au is electroplated as the bonding layer. Finally, the photoresist is removed, and the seed layer can be etched by the wet process.   Figure 5 shows the fabricating process flow of the dummy wafer. Ti/W is sputtered on the silicon substrate as adhesive & seed layer. An organic photoresist 6µm thick is spin-coated and patterned. Figure 6a,b show that the pattern of the dummy device wafer and the opening of the seal ring and bump are 43.5 µm and 85 µm, respectively. In Figure 5 (4,5), 1.5 µm Au is electroplated as the bonding layer. Finally, the photoresist is removed, and the seed layer can be etched by the wet process.

Packaging Process
The fabrication processes of the test vehicle for RF filter packaging involve a TGV cap wafer which includes transmission lines and RF filter fabrication process on the silicon substrate. The characteristic of RF filter WLP is determined by an accurate delta value that should not be greater than 1 dB in peak insertion loss (IL) of the passband. Figure 4 illustrates the process flow of the present 3D WLP for the RF filter. In this process flow, key processes include TGV cap wafer formation with transmission lines and through vias interconnection, Au-Sn TLP bonding, under bump metallization (UBM) formation, and solder ball.  Figure 5 shows the fabricating process flow of the dummy wafer. Ti/W is sputtered on the silicon substrate as adhesive & seed layer. An organic photoresist 6μm thick is spincoated and patterned. Figure 6a,b show that the pattern of the dummy device wafer and the opening of the seal ring and bump are 43.5 μm and 85 μm, respectively. In Figure 5 (4,5), 1.5 μm Au is electroplated as the bonding layer. Finally, the photoresist is removed, and the seed layer can be etched by the wet process.

Glass Cap Wafer Fabrication Process
The fabrication process of TSVs is comparatively complex. Its main fabrication processes are listed below.

•
Step 2. Apply photoresist and mask, then use photolithography techniques to open vias on the SiO2; • Step 3. RIE of SiO2;
To lower costs, LIDE is a high-yield process. It can produce high-precision vias without using masking and lithography and reduce the manufacturing cost compared with blind TSV fabrication. Meanwhile, LIDE technology can be used for different kinds of glass, such as quartz and borofloat. By selecting laser pulses, the concentration of HF solution and the chemical composition, fabricating TGV with the different profile can be achieved. In this paper, the key technology for fabricating a glass capping wafer includes blind TGVs formation, Cu electroplating to fill vias without voids, and metallization on the glass wafer. In Figure 7a,b, according to the designed packaging structure, blind TGVs with a certain depth and diameter are formed economically while retaining all excellent properties. Here, we show that the picosecond laser affected zone (LAZ) has a higher etching rate while reacting with the hydrofluoric acid (HF) solution compared with the laser unaffected zone. At the LAZ, a series of nanovoids along the path of a laser beam propagating are observed, which contribute to the enhanced etchability of LAZ when the glass sample is immersed in HF solution [31]. Besides, it is easy to prepare the one-side blind TGVs by controlling the depth of focus and HF solution concentration. Here, the process parameters are set at laser beam: single-shot laser: E = 55 J μ /pulse, p τ =16 ps, concentration of HF solution: 10%.

Glass Cap Wafer Fabrication Process
The fabrication process of TSVs is comparatively complex. Its main fabrication processes are listed below.

•
Step 2. Apply photoresist and mask, then use photolithography techniques to open vias on the SiO 2 ; • Step 3. RIE of SiO 2 ; • Step 4. Strip off the photoresist.
To lower costs, LIDE is a high-yield process. It can produce high-precision vias without using masking and lithography and reduce the manufacturing cost compared with blind TSV fabrication. Meanwhile, LIDE technology can be used for different kinds of glass, such as quartz and borofloat. By selecting laser pulses, the concentration of HF solution and the chemical composition, fabricating TGV with the different profile can be achieved. In this paper, the key technology for fabricating a glass capping wafer includes blind TGVs formation, Cu electroplating to fill vias without voids, and metallization on the glass wafer. In Figure 7a,b, according to the designed packaging structure, blind TGVs with a certain depth and diameter are formed economically while retaining all excellent properties. Here, we show that the picosecond laser affected zone (LAZ) has a higher etching rate while reacting with the hydrofluoric acid (HF) solution compared with the laser unaffected zone. At the LAZ, a series of nanovoids along the path of a laser beam propagating are observed, which contribute to the enhanced etchability of LAZ when the glass sample is immersed in HF solution [31]. Besides, it is easy to prepare the one-side blind TGVs by controlling the depth of focus and HF solution concentration. Here, the process parameters are set at laser beam: single-shot laser: E = µJ55 /pulse, τ p = 16 ps, concentration of HF solution: 10%. Figure 7c,d shows the cross-section view of the glass capping wafer (SCHOTT BOROFLOAT ® 33) after the LIDE. In this stage, the glass wafer with a thickness of 600 µm is thinned down to 550 µm. The feasibility of via array with 45 µm in diameter and 145 µm in depth is formed.  Glass is an insulating material and does not require a barrier layer before plating. After the TGVs are formed, the titanium (Ti) and Cu seed layer are sputtered on the wafer surface, including via sidewall. A negative organic film is affixed to the glass wafer and patterned. In Figure 8, the patterned photographic film was used to define the seal ring opening with a width of 45.6 μm. Meanwhile, it has high alignment accuracy (shift < 7 μm). Then the vias are deposited with Cu by full filling plating. After last, the seed layer at the non-exposed area is etched away. In Figure 9, the dimensions of TGV are 144.1 μm depth (spec value: 150 ± 10 μm), 45 μm in diameter at the via opening (spec value: 45 ± 5 μm). The aspect ratio of the via (depth-to-width of via) is ∼3:1. We have tried several combinations of plating parameters, including plating current, time, additives consisting of accelerator (A), suppressor (S), and leveler (L), to develop the void-free full-via plating. In this paper, Figure 9b shows the scanning electron microscopic (SEM) image of voidfree Cu TGV plating. Large thermomechanical stresses develop at the copper-glass interface because of the significant coefficient of thermal expansion (CTE) mismatch between the copper and glass, which can lead to various reliability issues at elevated temperatures. We control the thermal mismatch between glass and copper interaction by setting proper annealing temperature with a reasonable duration. The copper via has more significant creep at higher temperatures. Therefore, a relatively higher temperature (ramp rate: 4 °C/min, peak temperature: 320 °C) is preferred in the TGV annealing process to improve the material properties of the coppe. Glass is an insulating material and does not require a barrier layer before plating. After the TGVs are formed, the titanium (Ti) and Cu seed layer are sputtered on the wafer surface, including via sidewall. A negative organic film is affixed to the glass wafer and patterned. In Figure 8, the patterned photographic film was used to define the seal ring opening with a width of 45.6 µm. Meanwhile, it has high alignment accuracy (shift < 7 µm). Then the vias are deposited with Cu by full filling plating. After last, the seed layer at the non-exposed area is etched away. In Figure 9, the dimensions of TGV are 144.1 µm depth (spec value: 150 ± 10 µm), 45 µm in diameter at the via opening (spec value: 45 ± 5 µm). The aspect ratio of the via (depth-to-width of via) is ∼3:1. We have tried several combinations of plating parameters, including plating current, time, additives consisting of accelerator (A), suppressor (S), and leveler (L), to develop the void-free full-via plating. In this paper, Figure 9b shows the scanning electron microscopic (SEM) image of void-free Cu TGV plating. Large thermomechanical stresses develop at the copper-glass interface because of the significant coefficient of thermal expansion (CTE) mismatch between the copper and glass, which can lead to various reliability issues at elevated temperatures. We control the thermal mismatch between glass and copper interaction by setting proper annealing temperature with a reasonable duration. The copper via has more significant creep at higher temperatures. Therefore, a relatively higher temperature (ramp rate: 4 • C/min, peak temperature: 320 • C) is preferred in the TGV annealing process to improve the material properties of the coppe.
After the first RDL layer formation, the bonding layer was fabricated by photolithography and electroplating. This not only served as a sealing layer for preventing humidity or any corrosion but also acted as an electrical interconnection layer between chip pad and backside solder ball. Figure 10 presents the outlook of the chip after the bonding layer formation. The opening and thickness of the sealing layer are 27 µm and 9 µm (Cu/Ni/Sn (2/2/5 µm)), respectively. After the first RDL layer formation, the bonding layer was fabricated by photolithography and electroplating. This not only served as a sealing layer for preventing humidity or any corrosion but also acted as an electrical interconnection layer between chip pad and backside solder ball. Figure 10 presents the outlook of the chip after the bonding layer formation. The opening and thickness of the sealing layer are 27 μm and 9 μm (Cu/Ni/Sn (2/2/5 μm)), respectively.  After the first RDL layer formation, the bonding layer was fabricated by photolithography and electroplating. This not only served as a sealing layer for preventing humidity or any corrosion but also acted as an electrical interconnection layer between chip pad and backside solder ball. Figure 10 presents the outlook of the chip after the bonding layer formation. The opening and thickness of the sealing layer are 27 μm and 9 μm (Cu/Ni/Sn (2/2/5 μm)), respectively. After the first RDL layer formation, the bonding layer was fabricated by photolithography and electroplating. This not only served as a sealing layer for preventing humidity or any corrosion but also acted as an electrical interconnection layer between chip pad and backside solder ball. Figure 10 presents the outlook of the chip after the bonding layer formation. The opening and thickness of the sealing layer are 27 μm and 9 μm (Cu/Ni/Sn (2/2/5 μm)), respectively.

Bonding Procedure
Wafer-level bonding is achieved with optimized Au-Sn TLP bonding technology with a void-free seal ring. The functional area of the RF filter, the AlN electrode, is protected in a hermetic cavity from humidity or any kind of corrosion. In the glass cap wafer, RDL and TGV vertical interconnects are processed to realize the electrical interconnection from the chip's pad to the solder.
Prior to bonding, the surface of Sn is easily oxidized in the open air at room temperature, an important step for a good bonding procedure to remove the oxide layer on the Sn surface. Surface treatments by soaking the wafer in the dilute sulfuric acid followed by a forming gas purge (typically 5%H 2 : 95%N 2 , by volume) in the bonding chamber at 100~200 • C are typically applied. There are many advantages, such as low or no outgassing from the solder layer, and it can withstand high temperature after bonding using Au-Sn bonding technology. That is, generally, the reaction between low melting temperature 232 • C Sn solder and Au will completely convert to IMC, including AuSn (δ phase) and Au5Sn (ς phase). In the package structure of this paper, the bonding structure is a 27 µm wide, 704 µm × 529 µm closed square ring with four round bumps. The design bonding interface metallization and bonding parameters are shown in Figure 11. The bonding is performed in EVG 520 vacuum bonder under a vacuum pressure around 3 × 10 −5 mbar at 280 • C with 40 kN bonding force (approximately 6.5 MPa) for 5 min, making it advantageous for temperature-sensitive devices. Consequently, the pressure inside the sealed cavity is estimated at 3 × 10 −5 mbar (far less than atmospheric pressure). Figure 12 shows the OM and cross-section SEM image of the final RF filter WLP. The solder ball is dropped on the opening of the UBM layer by stencil printing and reflow process. The stencil printed solder bump dimension is 100 µm, and the material is SAC320. Finally, the diameter and height of the solder ball are 100 µm and 70 µm, respectively. The wafer is diced into individual packages after the final solder balling procedure, as shown in Figure 12a. Figure 12b displays the bonding structure. No void is found in the closed square ring.
with a void-free seal ring. The functional area of the RF filter, the AlN tected in a hermetic cavity from humidity or any kind of corrosion. In the RDL and TGV vertical interconnects are processed to realize the electrica from the chip's pad to the solder.
Prior to bonding, the surface of Sn is easily oxidized in the open air ature, an important step for a good bonding procedure to remove the o Sn surface. Surface treatments by soaking the wafer in the dilute sulfuric a forming gas purge (typically 5%H2: 95%N2, by volume) in the bond 100~200 °C are typically applied. There are many advantages, such as lo sing from the solder layer, and it can withstand high temperature after bo Sn bonding technology. That is, generally, the reaction between low mel 232 °C Sn solder and Au will completely convert to IMC, including AuS Au5Sn ( ς ′ phase). In the package structure of this paper, the bonding μm wide, 704 μm × 529 μm closed square ring with four round bumps. T ing interface metallization and bonding parameters are shown in Figure  is performed in EVG 520 vacuum bonder under a vacuum pressure arou at 280 °C with 40 kN bonding force (approximately 6.5 MPa) for 5 min, m tageous for temperature-sensitive devices. Consequently, the pressure cavity is estimated at 3 × 10 −5 mbar (far less than atmospheric pressure). the OM and cross-section SEM image of the final RF filter WLP. The solde on the opening of the UBM layer by stencil printing and reflow process. Th solder bump dimension is 100 μm, and the material is SAC320. Finally, t height of the solder ball are 100 μm and 70 μm, respectively. The wafer i vidual packages after the final solder balling procedure, as shown in Fi 12b displays the bonding structure. No void is found in the closed squar

Shear Strength Test
The bonding strength of the Au-Sn joint is measured using die shear testing equipment. The test scheme is made and illustrated in Figure 13a. The bottom silicon substrate is fixed to the holder, and the shearing tool is applied to the glass cap for testing shear strength. The shear direction of the applied force is perpendicular to the glass cap die. The fracture picture after the die shear test is shown in Figure 13b. In Figure 14, the shear strength of Au-Sn bonding is measured with an average of 54.5 MPa. According to the test method standard: MIL-STD-883 (~6 MPa, method No. 2019.5Die shear strength) [29][30][31], this shear force is higher than the standard requirement. Meanwhile, J. Peng et al. [32] suggests that the mechanical reliability of Au-Sn joints during TLP bonding remains stable at 50 MPa at room temperature. In this paper, the shear strength of the TLP-bonded joint is stable, with a value of 54.5 MPa.

Shear Strength Test
The bonding strength of the Au-Sn joint is measured using die shear testing equipment. The test scheme is made and illustrated in Figure 13a. The bottom silicon substrate is fixed to the holder, and the shearing tool is applied to the glass cap for testing shear strength. The shear direction of the applied force is perpendicular to the glass cap die. The fracture picture after the die shear test is shown in Figure 13b. In Figure 14, the shear strength of Au-Sn bonding is measured with an average of 54.5 MPa. According to the test method standard: MIL-STD-883 (~6 MPa, method No. 2019.5Die shear strength) [29][30][31], this shear force is higher than the standard requirement. Meanwhile, J. Peng et al. [32] suggests that the mechanical reliability of Au-Sn joints during TLP bonding remains stable at 50 MPa at room temperature. In this paper, the shear strength of the TLP-bonded joint is stable, with a value of 54.5 MPa.

Shear Strength Test
The bonding strength of the Au-Sn joint is measured using die shear testing equipment. The test scheme is made and illustrated in Figure 13a. The bottom silicon substrate is fixed to the holder, and the shearing tool is applied to the glass cap for testing shear strength. The shear direction of the applied force is perpendicular to the glass cap die. The fracture picture after the die shear test is shown in Figure 13b. In Figure 14, the shear strength of Au-Sn bonding is measured with an average of 54.5 MPa. According to the test method standard: MIL-STD-883 (~6 MPa, method No. 2019.5Die shear strength) [29][30][31], this shear force is higher than the standard requirement. Meanwhile, J. Peng et al. [32] suggests that the mechanical reliability of Au-Sn joints during TLP bonding remains stable at 50 MPa at room temperature. In this paper, the shear strength of the TLP-bonded joint is stable, with a value of 54.5 MPa.

Shear Strength Test
The bonding strength of the Au-Sn joint is measured using die shear testing equipment. The test scheme is made and illustrated in Figure 13a. The bottom silicon substrate is fixed to the holder, and the shearing tool is applied to the glass cap for testing shear strength. The shear direction of the applied force is perpendicular to the glass cap die. The fracture picture after the die shear test is shown in Figure 13b. In Figure 14, the shear strength of Au-Sn bonding is measured with an average of 54.5 MPa. According to the test method standard: MIL-STD-883 (~6 MPa, method No. 2019.5Die shear strength) [29][30][31], this shear force is higher than the standard requirement. Meanwhile, J. Peng et al. [32] suggests that the mechanical reliability of Au-Sn joints during TLP bonding remains stable at 50 MPa at room temperature. In this paper, the shear strength of the TLP-bonded joint is stable, with a value of 54.5 MPa.

Deflection Assessment
The sealed cavity collapse of a micro-package often becomes the determining factor of the performance of encapsulated RF filters. Sealed cavity pressure comes from the molding process (appearing in the RF front-end module package), which means the maximum pressure is 3~5 MPa [33]. To predict the cavity deflection under pressure, we introduce a deflection reference model to simulate the collapse of the package capping layer [22,34].
where ω max is the maximum deflection of the glass capping layer; ∆y and ∆x are the length and width of the cavity (475 µm × 650 µm), respectively; P total is the total pressure on the glass capping layer; D is the flexural rigidity of the glass capping layer where t is the thickness of the thin glass capping layer (t is set as 100 µm); E is the young's modulus; v is the Poisson's ratio. Applying Equation (3) to our package (with parameters in Table 2), when its pressure increases from 0 to 3 or 5 MPa. The change of the deflection of the capping layer can be derived: 0.264 µm or 1.399 µm, respectively. The highest permitted deflection change of our package is 19 µm. Our theoretical results show that the designed cavity size based on the glass capping bonding package can meet the requirements.

Reliability Assessment Results
According to the joint electron device engineering council (JEDEC) reliability test standards (specific details see ref. JESD22-A110E.01 [35]), in Table 3, we have also performed reliability tests on the RF filter packages to evaluate the integrity of the parts. Twenty chips are soldered onto a printed circuit board (PCB) and molding. The samples are first baked at 125 • C/24 H to remove the moisture and then soaked at 30 • C/60%/192 H. At last, reflow and uHAST (130 • C under 85% RH for 96 H) are conducted for reliability evaluation. The results show that no samples are failed. The hermeticity of seal rings for patterned dies was evaluated by exposing them to the wafer and humidity stress test and exposure time was 96 H. The compared results before and after packaging are shown in Figure 15. We found our bonding approach for the electrical performance of the RF filter package provides the structure of the hermetic package. There is no difference in insertion attenuation across the passband (<0.2 dB, standard value: <1 dB) after the WLP process.

Conclusions
In summary, a novel RF filter wafer-level packaging solution based on Au-Sn TLP bonding technology has been proposed and realized. The proposed technology has significant advantages compared to other established processes in terms of integration of a microstructured wafer, which can be achieved in an ultra-thin package size with a thickness of 300 μm, 1.5 μm Au bonding layer, bonding frame width as narrow as 27 μm, bonding temperatures as low as 280 °C, and a bonding force of approximately 6.5 Mpa at the same time. Due to its effective packaging process, it can provide solid reliability and reduce the manufacturing cost alternative to conventional packaging for RF filter applications.
The main conclusions are described as follows: (1) The glass interposer capping wafer is fabricated by LIDE, Cu plated blind TGVs, and RDL process to provide protection and electrical access to the RF filter. Following, the prepared glass capping wafer is bonded to the RF device wafer by Au-Sn TLP bonding. Shear strength's of approx. 54.5 MPa can be determined, which is higher than the standard requirement (MIL-STD-883: method No. 2019.5 Die shear strength). (2) The glass interposer capping wafer is fabricated by LIDE, making improving process efficiency compared to the FBAR device package using silicon lid bonding technology. In addition, the glass is transparent, we can track the packaging yield in realtime during processing. (3) Through comparing the electrical performance after standard reliability tests, there is no difference in insertion attenuation across the passband (<0.2 dB). It is proved that the RF filter WLP with TGVs connection and cavity hermetic is a very promising solution due to its high robustness.

Conclusions
In summary, a novel RF filter wafer-level packaging solution based on Au-Sn TLP bonding technology has been proposed and realized. The proposed technology has significant advantages compared to other established processes in terms of integration of a micro-structured wafer, which can be achieved in an ultra-thin package size with a thickness of 300 µm, 1.5 µm Au bonding layer, bonding frame width as narrow as 27 µm, bonding temperatures as low as 280 • C, and a bonding force of approximately 6.5 Mpa at the same time. Due to its effective packaging process, it can provide solid reliability and reduce the manufacturing cost alternative to conventional packaging for RF filter applications.
The main conclusions are described as follows: (1) The glass interposer capping wafer is fabricated by LIDE, Cu plated blind TGVs, and RDL process to provide protection and electrical access to the RF filter. Following, the prepared glass capping wafer is bonded to the RF device wafer by Au-Sn TLP bonding. Shear strength's of approx. 54.5 MPa can be determined, which is higher than the standard requirement (MIL-STD-883: method No. 2019.5 Die shear strength). (2) The glass interposer capping wafer is fabricated by LIDE, making improving process efficiency compared to the FBAR device package using silicon lid bonding technology. In addition, the glass is transparent, we can track the packaging yield in real-time during processing. (3) Through comparing the electrical performance after standard reliability tests, there is no difference in insertion attenuation across the passband (<0.2 dB). It is proved that the RF filter WLP with TGVs connection and cavity hermetic is a very promising solution due to its high robustness.