I. INTRODUCTION

In the semiconductor industry, traditional miniaturization processes have been used to improve performance by shrinking each new generation of transistors in accordance with Moore's Law. However, due to the difficulties of small-scale patterning and the increasing complexity of integrated circuits, transistor scaling has reached its limits. As a solution, 3D packaging technology, which vertically stacks chips, is gaining attention. This technology reduces the distance between chips, thereby improving signal delay, power consumption, and transistor density within the same space. Notably, hybrid Cu bonding technology enables direct bonding between Cu-to-Cu and dielectric-to-dielectric without the use of micro-bumps, allowing for ultrafine-pitch implementation for higher I/O capacity. In hybrid Cu bonding, the dielectric material must act as a Cu diffusion barrier and achieve sufficient bonding strength to support Cu-to-Cu bonding during the annealing process.

Previous extensive research on Cu/SiO${}_{2}$ hybrid bonding has revealed reliability issues, caused by Cu pad misalignment and defects such as erosion during the CMP process ^[1]. Consequently, SiCN has emerged as a promising material for hybrid Cu bonding. SiCN offers superior Cu diffusion barrier properties compared to SiO${}_{2}$, exhibits low surface roughness after CMP (chemical mechanical polishing), and remains high bonding strength even under low-temperature annealing conditions ^[2,3,4,5,6].

Typically, SiCN is deposited using PECVD (plasma-enhanced chemical vapor deposition) at a temperature of approximately $350^\circ$C, and a high-temperature annealing process is required afterward. Also, there is a potential issue of residual precursors remaining in the film after deposition. If residual precursors are present, they may prevent sufficient bonding strength which is necessary for hybrid copper bonding ^[7]. By comparison, SiCN deposited by PVD (physical vapor deposition) does not require high-temperature annealing after deposition, which eliminates precursor residue issues, and also simplifies the process.

Based on these advantages, this study evaluated SiCN deposited by PVD for hybrid copper bonding applications. The dielectric constant and hydrophilicity of PVD SiCN were analyzed, and their effects on SiCN-to-SiCN bonding were investigated.

II. EXPERIMENTAL METHODS

In hybrid Cu bonding technology, dielectric-dielectric bonding typically involves plasma treatment followed by deionized (DI) water cleaning to increase the adhesion of the surfaces prior to the alignment and bonding process. In this study, we aim to develop a Cu bonding process without DI water rinsing to reduce Cu oxidation and contamination, resulting in a more CMOS-friendly bonding process flow. Fig. 1 is a schematic diagram of the PVD SiCN bonding process conducted in this study.

Fig. 1. Schematic diagram of PVD SiCN bonding flow.

SiCN was deposited on a 100 mm p-type Si (100) wafer using a reactive sputtering method. The deposition target was SiC, and Ar and N${}_{2}$ were used as reactive gases. The deposited SiCN wafer was diced into $5\times5$ mm$^2$ pieces using an automatic dicing saw (DAD3350, DHK). The refractive index of the SiCN thin film was measured using ellipsometry (UVISEL, HORIBA), and the dielectric constant of the SiCN thin film was calculated as the square of the refractive index. The chemical composition of the SiCN thin film was analyzed using auger electron spectroscopy (PHI-710, ULVAC-PHI), and the chemical state of the SiCN thin film was analyzed using an X-ray photoelectron spectrometer (Nexsa, Thermo Fisher Scientific Brno s.r.o).

Prior to bonding, the CMP (Chemical Mechanical Polishing) process was utilized for a surface planarization of the SiCN. For the CMP process, polishing was performed using a polisher (MULTIPREP SYSTEM-8, Allied high tech) under a pressure of 39.2 KPa, with both the pad and wafer rotating at 30 rpm for 90 seconds. The CMP pad used was the KONI pad, and the polishing utilized a colloidal silica-based slurry (CBS-7000, CNS) with a particle size of 35 nm. In general CMP is crucial for hybrid Cu bonding as it ensures a flat, smooth, and oxide-free surface, which is essential for strong adhesion and reliable bonding. In particular, the key to excellent electrical and mechanical reliability in hybrid Cu bonding is control of Cu dishing around 1 nm.

After the CMP process before bonding, the plasma pre-treatment was carried out, which has a crucial role in enhancing bonding strength by activating the surface, promoting chemical bonding, and removing oxides. In this study, a 2-step Ar/N${}_{2}$ plasma treatment was used as the plasma source. The 2-step Ar/N${}_{2}$ plasma treatment was performed using a sputtering system (SRN-110, Sorona). Detailed procedures for the 2-step Ar/N${}_{2}$ plasma treatment can be found in other literature ^[8,9]. After the 2-step Ar/N${}_{2}$ plasma treatment, the SiCN was bonded and annealed using a die bonder (Accura100, SET). The bonding was carried out at 260$^\circ$C for 1 hour, and the annealing was at $200^\circ$C for 1 hour. The bonding interface was examined using a transmission electron microscopy (JEM-2100F, JEOL).

III. RESULTS AND DISCUSSION

1. Thin Film Characteristics of PVD SiCN

AES (Auger Electron Spectroscopy) was employed to confirm the chemical composition of the PVD SiCN. Peaks of Si, C, N, and a small amount of O were identified, as shown in Figure 2. Table 1 shows the chemical composition of the PVD SiCN at a depth of approximately 40 nm from the surface. PVD SiCN has a carbon content of 24.1%, as shown in Table 1. In general, as the carbon content increases, which accompanied by an increase in the Si-C bonds, the refractive index increases ^[10,11]. It has been reported that when the carbon content of the SiCN film exceeds 35%, the density of SiCN decreases significantly, and porosity increases ^[12,13].

Fig. 2. AES measurement of PVD SiCN.

Table 1. Chemical composition of PDV SiCN.

Si %	C %	N %	O %	Empirical formula
48.6	24.1	22.0	5.3	Si 1:C 0.50: N 0.45:O 0.11

Table 2 shows the refractive index and dielectric constant at a wavelength of 633 nm as measured by ellipsometry. The dielectric constant of PVD SiCN is 3.93, while that of PECVD SiCN is 3.64. Although the dielectric constant of PVD SiCN is somewhat higher than that of PECVD SiCN, it shows that PVD SiCN films have full potential for use as dielectric layers in hybrid Cu bonds. The reference dielectric for hybrid Cu bonding applications is a SiO${}_{2}$ layer with a dielectric constant of 3.9 to 4.0. So generally, the target dielectric constant needs to be similar to or lower than this.

Table 2. Ellipsometry measurements of PVD SiCN and PECVD SiCN.

	Refractive index	Dielectric constant
PVD SiCN	1.983	3.932
PECVD SiCN	1.909	3.644

To further control the surface roughness of SiCN, the CMP process was utilized before the 2-step Ar/N${}_{2}$ plasma treatment. The amount of SiCN removed by the CMP process was approximately 15 nm. Fig. 3 shows the surface roughness of the PVD SiCN before and after the CMP process. Before the CMP process, the surface roughness of SiCN was 1.16 nm, but after the CMP process, the surface roughness of the PVD SiCN decreased to 0.34 nm. This low surface roughness of PVD SiCN can result in reduced voids at the bonding interface.

Fig. 3. Surface roughness of PVD SiCN before and after the CMP process.

2. Effect of 2-step Ar/N${}_{2}$ Plasma Treatment on PVD SiCN

XPS analysis was conducted on the SiCN surface immediately after deposition and following the 2-step Ar/N${}_{2}$ plasma treatment to evaluate the effects of the plasma treatment. All of the XPS peaks were calibrated based on the C-C bond at 284.5 eV before the deconvolution process.

Figs. 4(a) and (b) present the Si2p spectra results before and after the 2-step Ar/N${}_{2}$ plasma treatment. In the Si2p spectra, the peaks observed at 101.7 eV, 102.8 eV, and 103.7 eV correspond to Si-C, Si-N, and Si-O bonds, respectively. While the AES analysis shown in Fig. 2 detected only a small amount of oxygen in the bulk of the SiCN, the XPS analysis, which is more sensitive to the surface revealed the presence of silicon oxides (SiO${}_{2}$) on the surface of the SiCN film. This finding aligns with results from other studies ^[14,15].

After the 2-step Ar/N${}_{2}$ plasma treatment, the Si-C peak (shown in red) decreased, while the Si-O peak (in pink) and the Si-N peak (in green) increased (Figs. 4(a), 4(b)). This indicates that the 2-step Ar/N${}_{2}$ plasma treatment terminates the Si-C bonds, forming Si and C dangling bonds. Some of the terminated silicon bonds form Si-O and Si-N bonds due to exposure to air and N${}_{2}$ plasma. These Si and C dangling bonds are anticipated to play a crucial role in SiCN-SiCN bonding ^[16,17].

Fig. 4. XPS spectra of the top surface: (a) Si2p as deposited, (b) Si2p after 2-step Ar/N${}_{2}$ plasma treatment, (c) C1s as deposited, and (d) C1s after 2-step Ar/N${}_{2}$ plasma treatment (The black lines represent the XPS spectra (raw data), while the blue lines indicate the combined spectra of the deconvoluted multi-peaks).

(a)

(b)

(c)

(d)

For the C1s spectra shown in Figs. 4(c) and 4(d), the peaks at 283.9, 284.5, 285.5, and 287.1 eV correspond to C-Si, C-C, C-N, and C=N bonds, respectively. It can be seen that the overall C peaks decreased after the 2-step Ar/N${}_{2}$ plasma treatment. This indicates that the plasma treatment terminates bonds associated with C. As a result, some C dangling bonds react with oxygen in the air, leading to the appearance of the C-O bond in light blue at 288.6 eV, which was not seen in Fig. 4(c) but is present in Fig. 4(d). Additionally, the decrease in the C-Si bond in red after the 2-step Ar/N${}_{2}$ plasma treatment in the C 1s spectra supports the analysis of the Si2p spectra.

Fig. 5. Change in contact angle of PVD SiCN before and after the 2-step Ar/N${}_{2}$ plasma treatment.

The contact angles shown in Fig. 5 were measured in five times for each sample. Before the 2-step Ar/N${}_{2}$ plasma treatment, the contact angle of the PVD SiCN was approximately $41.3^\circ$. However, after the 2-step Ar/N${}_{2}$ plasma treatment, the contact angle of PVD SiCN decreased to less than $10^\circ$, consistent across all five measurements. This suggests that the PVD SiCN, after the 2-step Ar/N${}_{2}$ plasma treatment, has the appropriate surface energy for application in hybrid bonding. According to the study by F. Inoue et al. ^[18], PECVD SiCN exhibited a contact angle of $25^\circ$ before plasma treatment and less than $13^\circ$ after N${}_{2}$ plasma treatment. The reduction in the contact angle of PVD SiCN after the 2-step Ar/N${}_{2}$ plasma treatment is likely due to the low surface roughness and the formation of Si and C dangling bonds on the surface, as indicated by the XPS results.

3. Evaluation of PVD SiCN-SiCN Bonding

Fig. 6 shows the SAT image of PVD SiCN-SiCN bonded at $260^\circ$C and annealed at $250^\circ$C. The SAT image indicates a mostly well bonded interface shown in black color, but an unbonded area was existed at the center of the chip, which is due to air trapped at the interface during the bonding process. Air trapped in the center of the chip can generally be resolved during bonding by pressing from the center of the die outward.

Fig. 6. SAT image of PVD SiCN. (The red frame represents the edge of the $5\times5$ mm$^2$ chip).

Fig. 7. (a) TEM and (b) STEM images of the bonding interface of PVD SiCN.

The PVD SiCN-SiCN bonding interface was further analyzed by TEM and is shown in Figure 7. Through TEM images, it was observed that the PVD SiCN-SiCN bonding interface was very uniform and void-free. The EDS line scanning results shown in Fig. 8 indicate that there is almost no C and N at the SiCN-SiCN bonding interface, with O being predominant. These EDS line scanning results support the XPS results shown in Fig. 4, which describe the ultra-thin SiO${}_{2}$ layer present on the SiCN surface.

Fig. 8. EDS Line Scanning of the Bonding Interface of PVD SiCN.

IV. CONCLUSION

We investigated PVD SiCN for hybrid Cu bonding applications. PVD SiCN was uniformly deposited to a thickness of approximately 150 nm using reactive sputtering, and showed a low dielectric constant and low surface roughness of 0.34 nm after CMP. After the 2-step Ar/N${}_{2}$ plasma treatment, the PVD SiCN exhibited a contact angle of less than 10 degrees, proving its hydrophilicity prior to the bonding process. XPS and EDS results revealed the termination of C bonds, including Si-C, and the formation of Si and C dangling bonds and Si-O bonds, and confirmed that a decrease in C and N and an increase in O at the bonding interface, forming an ultra-thin SiO${}_{2}$ layer on the SiCN surface. Lastly, uniform and void-free low-temperature bonding of the PVD SiCN-SiCN was confirmed, suggesting that PVD SiCN can be applied as a dielectric material for hybrid Cu bonding.