Ammonia Plasma Surface Treatment for Enhanced Cu–Cu Bonding Reliability for Advanced Packaging Interconnection

Abstract

With the emergence of 3D stacked semiconductor products, such as high-bandwidth memory, bonding-interface reliability cannot be overemphasized. The condition of the surface interface before bonding is important and can substantially affect product reliability. Plasma technology can be used to control the state of a bonding interface, but various factors of interest, such as surface roughness, chemical bonding state, and surface cleanliness, may depend on the type of gaseous plasma. These factors may increase voids at the interface, which can jeopardize the product reliability. In this study, NH₃ plasma surface treatment is investigated and compared with the conventionally preferred surface treatment under Ar plasma. Under the latter method, specific anomalies occurred and led to void formation at the interface during bonding. By contrast, NH₃ plasma treatment maintained higher uniformity, higher overall surface conditions, and a smooth reduction process. Furthermore, the formation of a nitride passivation layer effectively inhibited the oxidation of the metal surface, and the flat surface resulted in the decrease in voids compared with the Ar plasma treatment after the copper–copper bonding. From the experimental analysis, we achieved a 12% reduction in resistance in the samples treated with NH₃ plasma treatment due to the suppression of surface oxidation. However, it is unfortunate that the shear strength in the experimental samples treated with NH₃ plasma treatment needs to be further improved.

1. Introduction

The needs for excellent device density, high signal communication bandwidth, superior performance, and low manufacturing costs persist with the expansion of the role of 3D packaging in the field of semiconductors [1,2]. An example of early 3D packaging technology is stacked chip-scale packaging with copper wires, but an increase in the number of copper wires in a package leads to a power consumption problem, signal loss, and the increased package footprint [3]. Through-silicon vias (TSVs) have been devised to alleviate concerns of multiple wire bonding, but the high cost of TSV fabrication on wafers has hindered the emergence of the application in commercial products [4]. Likewise, direct Cu–Cu bonding with TSV interconnection is a promising interconnection method in 3D packaging [5]. Cu–Cu bonding involves connecting two copper bumps or pads back-to-back in a TSV to interconnect another semiconductor chip, and it can also be used for hybrid bonding along with SiO₂–SiO₂ bonding [6]. It plays a crucial role in advanced packaging, a stacked high-bandwidth memory (a type of dynamic random-access memory), and backside power delivery networks [7,8].

Cu–Cu bonding technology is currently being investigated in various research directions. Cu–Cu bonding is normally conducted at temperatures exceeding 400 °C. Its thermal impact on nearby components can be reduced by decreasing the bonding temperature. This can be achieved via surface-activated bonding using plasma surface treatment [9]. Additionally, various issues related to the bonding interface are being studied, including oxidation and diffusion problems [10,11]. Papers on oxidation problems focus on inhibiting oxidation itself and removing oxide layers through plasma and wet chemical surface treatments. As for diffusion problems, researchers are investigating the application of coatings to the copper surface to promote diffusion during bonding [12,13,14,15,16]. Plasma technology should be used effectively to address these challenges. Ar plasma–based surface treatment has been adopted in many studies; this process is driven by Ar ion sputtering, which physically removes the bond between copper and oxygen. Although this process performs well in surface activation and oxide film removal, it requires appropriate control because it may increase the surface roughness [17].

For successful Cu–Cu bonding, surface conditions should be controlled strictly. The key factors affecting surface conditions include surface roughness, surface chemical state, and surface cleanliness [18]. The surface roughness should be minimized, and any surface oxide layer must be removed. The presence of particles on the surface can disrupt proper bonding and reduce bonding reliability [19]. Plasma processes enable the control of these parameters. Surface roughness reduction decreases the probability of void formation during bonding and increases the contact area between different Cu surfaces, thereby improving the shear strength and electrical properties of devices [20]. The effective control of oxide layers can enhance electrical characteristics and copper atom diffusion. The roughness of a copper surface is improved by adding H₂ to Ar plasma during plasma surface treatment [17]. Furthermore, the use of N₂ plasma for copper surface treatment effectively suppresses oxidation by forming a passivation layer on the surface [21]. In this study, the copper surface treatment was performed using NH₃ plasma to obtain the passivation layer expected from N₂ plasma and the improved surface roughness expected from H₂ plasma. In the following section, we explain how the samples were prepared, bonded, and tested, including the plasma surface treatment procedure. In Section 3, we present our experimental results and discuss our observations regarding the bonding strength. Finally, the conclusion is presented in Section 4.

2. Experiment

2.1. Surface Activation

A total of 50 nm of Ti barrier layer and 1 μm of Cu film were deposited on a 4-inch Si wafer via physical vapor deposition. Plasma surface treatment was performed using the plasma-enhanced chemical vapor deposition (PECVD) equipment, as shown in Figure 1. The process recipes for the sample fabrication are presented in

Table 1. Plasma surface treatment feasibility test.

Sample Number	Gas	Power (W)	Flow Rate (sccm)	Pressure (Torr)	Time (s)
1	Ar	300	50	1	30
2	NH3 (50%)/Ar	300	NH3 25/Ar 25	1	30
3	NH3 (90%)/Ar	300	NH3 45/Ar 5	1	30

. Ar plasma primarily facilitated a physical process. Ar gas is considered an unreactive gas, as it does not readily react with other elements. This characteristic makes Ar suitable for processes involving sputtering, where physical effects dominate. In this physical bombardment process, we achieved a reduced oxidized surface, shown in Figure 2a. Unlike Ar plasma, NH₃ plasma was selected not for a physical effect but for its chemical effect, which minimized physical damage to the copper surface presented in Figure 2b [22,23]. Hydrogen and nitrogen radicals formed within the NH₃ plasma. When the hydrogen radicals reacted with the surface oxygen the copper oxide layer was removed from the surface. Then the removal proceeds to the interface of Cu/Cu₂O, then slowly moves from the surface region and, at last, the entire whole metallic layer is reduced [24]. After the reduction process, nitrogen-free radicals generated by N₂ plasma reacted chemically with pure copper atoms, devoid of oxygen, resulting in the formation of copper nitride. This copper nitride acted as a passivation layer, preventing copper oxidation before Cu–Cu bonding, thus protecting the copper surface [25].

2.2. Bonding Process

Cu–Cu bonding was conducted via thermal compression bonding (TCB), which involves the application of heat and pressure. The bonding process was carried out as shown in Figure 3.This bonding process was conducted at 420 °C for 1 h, followed by an additional annealing process at 400 °C for 30 min. Cu–Cu bonding using heat and pressure typically comprises three stages. The initial stage (plastic deformation) occurs at the peaks of a wavy surface, followed by diffusion within the voids between the bonded surfaces. This diffusion leads to the formation of grain boundaries. Finally, the voids at the bonding interface ripen, thus completing the bonding process [26]. The additional annealing process was conducted for stronger bonding [27].

3. Results and Discussion

3.1. Atomic Force Microscopy (AFM) Analysis of Cu Surface

After surface treatment using plasma, atomic force microscopy (AFM) was used to evaluate the state of the Cu surface. The AFM results were analyzed in terms of different characteristics: line roughness, surface roughness, and AFM peaks. Line roughness and surface roughness are expressed by the parameters R_a and R_q. R_a represents the arithmetic mean of the deviation in height from the measured data’s average height, while Rq is the root mean square of the variation in height across the surface profile. Both values are used to characterize surface roughness. For line roughness analysis, measurements were obtained at various positions. However, the measured results do not consistently exhibit a specific trend. Some measured line roughness values are high at certain positions, whereas others are low. This inconsistency shows that line roughness analysis is unsuitable for this study. The next roughness parameter considered was surface roughness, which represented the roughness value throughout the measurement area. Roughness is expected to increase after Ar plasma treatment; by contrast, in this study, surface roughness does not significantly differ between the samples treated using Ar plasma and NH₃ plasma, as observed in Figure 4a,b. However, other parameters exhibit significant differences. Figure 4c,d depict the data for the height of the highest peak from the reference plane throughout the measured total area ($R_p$), and the sum of the highest peak and the deepest valley across the measured total area ($R_y$), respectively. As shown in the graphs in Figure 4c,d, as the NH₃ content increases, $R_p$ and $R_y$ decrease. This phenomenon is also evident in the 3D AFM image in Figure 5. This finding is attributed to a partial, irregular phenomenon indicating the high peaks forming on the reference plane resulting from the argon sputtering effect, and implies that the likelihood of this phenomenon decreases with the influence of Ar during plasma treatment. Furthermore, this partial, irregular phenomenon may increase the probability of void formation between the bonded surfaces during bonding.

3.2. X-Ray Photoelectron Spectroscopy (XPS) Analysis of Cu Surface

We investigate the chemical bonding state of the surface based on the type of plasma used through XPS analysis. In Figure 6, the Cu 2p_3/2 peaks for the processes involving different types of plasma are evident. The peak at 932.2 eV corresponds to pure Cu, indicating the absence of oxidation or nitridation. The peak at 933.8 eV is associated with CuO [28,29]. As for the process using NH₃ plasma, the surface consists of nitride caused by the chemical reactions of the nitride radicals generated in the NH₃ plasma, resulting in the formation of Cu₃N [30]. The Cu₃N peak is observed at 933.2 eV in Figure 6b,c [31]. For the O1s peak, as shown in Figure 7, dominant wavelengths are observed between 529 eV and 531 eV. At 529.9 eV and 530.8 eV, Cu₂O and CuO are identified, respectively [32,33]. The significant reduction observed with NH₃ plasma treatment confirms visible improvement in overall oxide levels. Hence, recipes 1 and 3 from Table 1 were selected for the process, considering the substantial improvement in oxide levels achieved through NH₃ plasma treatment, as confirmed visually. In Figure 8, the N1s spectra, not observed in the Ar plasma process, are seen, and they increase with the NH₃ ratio. This confirms the presence of nitride components on the surface and can be inferred to be a passivation layer that prevents oxidation.

3.3. Scanning Acoustic Microscopy (SAM) Analysis

Following the TCB process, void inspection was conducted using scanning acoustic microscopy (SAM). In the SAM images in Figure 9, the dark regions represent well-bonded areas, whereas the bright regions indicate poor bonding. Differences in the bright areas, which represent voids, are observed between Figure 9a,b. As for the shades of the dark regions, the image in Figure 9b, which involves the formation of the nitride passivation layer, is brighter than that in Figure 9a. This difference is likely due to the possibility of the improper decomposition of copper nitrides during bonding. A single SAM image is insufficient for comparison, so additional electrical and physical experiments, such as current–voltage (I–V) and shear tests, were conducted to assess the reliability of the bonded copper effectively.

3.4. I–V Test

I–V tests were conducted to investigate the electrical characteristics of the bonded copper. I–V test samples were prepared as shown in Figure 10. To conduct the I–V test, after dicing the bonded copper samples, we applied silver paste to each end of the samples to establish an electrical connection between the copper and the measuring equipment. These tests were performed to utilize the deposited Cu layer as a metal layer and examine the electrical properties of the wiring. The I–V test results are in Figure 11. Resistance was calculated by dividing the applied voltage by the measured current. A comparison of the resistances obtained using this method shows that the samples treated using NH₃ plasma have lower resistance compared with those treated using Ar plasma. However, this result cannot be solely attributed to the difference between the pure Cu and the Cu surface with a passivation layer, as oxidation occurred during sample preparation. Relative to the Ar plasma process, the NH₃ plasma process is more effective in inhibiting oxidation, resulting in lower resistance in the oxidized Cu.

Simulations were conducted to validate the I–V test method. The sample shown in Figure 10 is represented as an equivalent circuit in Figure 12. We represent the resistances of the materials. For example, the resistance of wafer is represented as $R_{wafer}$ and Ti is $R_{Ti}$. The resistance of copper ($R_{Cu}$) was calculated and used in the simulation by regarding the previously obtained resistance as $R_{total}$ and applying it to the parallel resistor formula. The simulation results are shown in Figure 13.

3.5. Shear Test

For the bonding strength measurement, experiments where force is applied horizontally, such as shear tests, are more appropriate than experiments where force is applied vertically to the bonding interface, such as four-point bending tests. Samples sized 2 × 2 mm² were prepared for our shear tests. Equation (1) was used for the precise measurement of bonding strength [34].

$$\tau ={\displaystyle \frac{F_{\mathrm{f}\mathrm{a}\mathrm{i}\mathrm{l}\mathrm{r}\mathrm{u}\mathrm{e}}}{A}}$$

(1)

In the above equation, $\tau$ represents the shear strength, $F_{\mathrm{f}\mathrm{a}\mathrm{i}\mathrm{l}\mathrm{u}\mathrm{r}\mathrm{e}}$ corresponds to the force at which the bond is broken, and $A$ denotes the area of the shear test specimen. Shear tests were conducted following the method shown in Figure 14 and Figure 15. The results in Figure 16 show that Cu–Cu bonding using Ar plasma exhibits a higher shear strength compared with Cu bonding using NH₃ plasma. The passivation layer formed when using NH₃ plasma should decompose at temperatures exceeding 350 °C [35]. However, during the bonding process in this study, the copper nitride passivation layer was insufficiently decomposed, leading to bonding in the form of copper nitride. Consequently, the bonding strength under the use of NH₃ plasma is lower than under the use of Ar plasma. This issue has also been observed in other experiments. Therefore, additional research may be necessary for the effective removal or decomposition of the passivation layer when conducting Cu–Cu bonding with such layers.

3.6. Transmission Electron Microscopy (TEM) Analysis

A TEM analysis of Cu–Cu bonding was conducted for surface treatment using Ar plasma and surface treatment using NH₃ plasma, which is presented in Figure 17. For treatment using Ar plasma, voids are discernible at the bonding interface. In contrast, for treatment using NH₃ plasma, the bonding interface is hardly distinguishable, and the defects readily identified as voids in Figure 17a are not easily visible in Figure 17b. The image in Figure 18 schematically represents the shape of Figure 17a, clearly showing the presence of voids. This observation can be related to the AFM findings. According to the roughness findings in Figure 4 and Figure 5 and the 3D AFM image, during surface treatment using Ar plasma, nonideal peaks and valleys formed, unlike during treatment using NH₃ plasma. The initiation of bonding in this state may lead to the formation of significant gaps at the bonding interface, thus increasing the probability of void formation. This explains why more voids emerged during treatment using Ar plasma.

4. Conclusions

In this study, we investigated the effects of using Ar plasma and NH₃ plasma for Cu surface treatment during Cu–Cu bonding via TCB. We found that these treatments influenced the inhibition of oxidation and the suppression of void formation. The passivation layer formed during NH₃ plasma surface treatment played a significant role in inhibiting oxidation. Moreover, the plasma surface treatment mechanism had a substantial impact on the suppression of void formation. We compared the physical effects dominant in Ar plasma surface treatment, where physical processes prevailed, with the chemical reactions predominant in NH₃ plasma surface treatment. Notably, we observed significant differences in the surface roughness parameters $R_p$ and $R_y$, with NH₃ plasma surface treatment, which exhibited smaller values of $R_p$ and $R_y$, resulting in fewer voids. However, Ar plasma surface treatment led to higher bonding strength. This was attributed to the insufficient decomposition of the passivation layer formed during NH₃ plasma surface treatment. This incomplete decomposition compromised the bonding strength, resulting in Cu nitride–Cu nitride bonding rather than Cu–Cu bonding. In semiconductor manufacturing, where the production of a single semiconductor product consumes considerable time, copper oxidation should be inhibited. However, the passivation layer formed during NH₃ plasma surface treatment must eventually be removed to improve the electrical and physical properties of the product. In this study, the insufficiently decomposed passivation layer compromised the physical properties of the product. Therefore, in future research, the impact of this passivation layer should be minimized for improved performance.