Electroless Pd Nanolayers for Low-Temperature Hybrid Cu Bonding Application: Comparative Analysis with Electroplated Pd Nanolayers

Abstract

As 3D stacking technologies advance, low-temperature hybrid Cu bonding has become essential for fine-pitch integration. This study focuses on evaluating Pd nanolayers deposited by electroless plating (ELP) on Cu surfaces and compares them to electroplated (EP) Pd to assess their suitability for hybrid bonding. Pd nanolayers (5~7 nm) were deposited on Cu films, and their surface morphology, crystallinity, and chemical composition were characterized using AFM, TEM, GIXRD, and XPS. EP-Pd layers exhibited lower roughness and larger grain size, acting as effective Cu diffusion barriers. In contrast, ELP-Pd layers showed small grains, higher surface roughness, and partial Cu diffusion and oxidation. At 200 °C, both Pd layers enabled bonding, but ELP-Pd samples achieved more uniform and continuous interfaces with thinner copper oxide layers. Shear testing revealed that ELP-Pd samples exhibited higher average bonding strength (20.58 MPa) and lower variability compared to EP-Pd (16.47 MPa). The improved bonding performance of ELP-Pd is attributed to its grain-boundary-driven diffusion and uniform interface formation. These findings highlight the potential of electroless Pd as a passivation layer for low-temperature hybrid Cu bonding and underscore the importance of optimizing pre-bonding surface treatments for improved bonding quality.

1. Introduction

Three-dimensional stacking technology offers a transformative approach to overcoming the limitations of traditional 2D planar scaling by vertically stacking chips, which allows for increased functionality within a smaller physical footprint. This approach enables higher integration densities, leading to improved chip performance, better data transfer rates, and significant reductions in power consumption, all while maintaining or even improving the overall size constraints [1,2,3]. 3D stacked devices are particularly valuable in applications where space is limited, such as in mobile devices, high-performance computing (HPC), and internet of things (IoT) systems. However, to realize the benefits of 3D stacking, several complex processes must be carefully executed, including through-silicon vias (TSV), wafer grinding, and chip or wafer bonding. Among these, the bonding process is particularly crucial, as it directly influences the electrical, thermal, and mechanical properties of the final stacked device. With the increasing density of input/output (I/O) counts, bonding with the conventional micro-bumps becomes difficult to support continuous fine-pitch scaling. This difficulty can be overcome by the use of hybrid Cu bonding, which provides low resistivity, high density interconnects and improved power efficiency and better thermal management [4,5].

In recent years, low-temperature Cu bonding has emerged as a critical technique for enabling reliable and effective 3D stacking, particularly when coupled with noble metal passivation layers. The use of a metal passivation layer offers several advantages in advanced Cu-to-Cu bonding applications, particularly as bonding pitches continue to scale down. These nanolayers enable reliable low-temperature bonding below 200 °C without the need for vacuum conditions, which are often required in surface-activated or plasma-assisted bonding methods. In addition to passivating the Cu surface and preventing oxidation, the metal layer also promotes atomic interdiffusion at the interface, which is essential for forming strong bonds at low temperature. Moreover, the metal nanolayer can also help mitigate the challenges associated with achieving uniform and precise Cu dishing control during the CMP process. By acting as a compensating layer, it can reduce the sensitivity of the bonding interface to local height variations, thereby improving planarity and enhancing bonding yield, especially in fine-pitch applications. The passivation layers, such as palladium (Pd), Gold (Au), and ruthenium (Ru), have demonstrated some promising results in enhancing the bonding process, and offering suitable for strength in Cu-to-Cu bonding. Noble metal passivation layers are especially beneficial in fine-pitch bonding applications, where the need for high-precision and reliable connections is paramount. These layers serve to improve the overall reliability and performance of the bonding interface, reducing issues such as electrical resistance, thermal instability, and mechanical failure. Numerous studies [6,7,8,9,10,11,12,13,14,15,16,17] have already explored the use of noble metal nanolayers in bonding applications shown in

Table 1. Various nanometal layers used in Cu-to-Cu bonding.

Ref	Metal	THK (nm)	Bonding			Annealing	Shear Strength (MPa)	Contact Resistance (Ω·cm2)	Remarks
			Temp (°C)	Pressure (MPa)	Time (min)
This work	Pd (ASD)	7	200	15	60	200 °C, 1 h	20.59	-	D2D, Air
[6]	Pd	10	150	1.91	50	No annealing	125 kg (Pull test)	2 × 10−7	W2W, Vacuum
[7]	Ti	5–10	180	1.91	30–50	No annealing	-	7 × 10−4	W2W, Vacuum
[8]	Ru	10	200	15	60	200 °C, 2 h	17.16	1.78 × 10−7	D2D, Air
[9]	Au	3	140	0.3	30	No annealing	200		W2W, Vacuum
[10]	Au	5	180	15	30	200 °C, 1 h	21.85	-	D2D, Air
[11]	Ag	~20 (nanoparticle)	250	20	5	250 °C, 25 min	14.4	-	D2D, Air
[12]	Ag	15	180	0.8	30	200 °C, 1 h	6.5	-	W2W, Vacuum
[13]	Unknown (ASD)	5–7	-	-	-	>300 °C, 2 h	-	-	W2W
[14]	Pt, Ti, Ta, Cr	10~12	200	-	60	280 °C, 1	-	-	D2D, Air
[15]	Au/Ag	12/15	180	0.8	30	200 °C, 1 h	5.4/6.6	-	W2W, Vacuum
[16]	Ag (ASD)	18–40	-	-	-	180 °C, 200 °C. 3 min	-	2 × 10−9
[17]	Au (ASD)	10	150–200	200 (N)	3	-	26.122	2 × 10−7	D2D, Air

.

Among the various candidate materials, Pd is the primary focus of this study. Although Pd nanolayers present some disadvantages in hybrid bonding compared to other metals, they remain a promising option. While Pd offers better oxidation resistance than reactive metals such as Ti or Cr, it is not as chemically inert as Au or Pt, and its surface may still be susceptible to contamination. Additionally, Pd can diffuse into adjacent materials at elevated temperatures, potentially affecting long-term interfacial stability. Its relatively high cost is also a consideration. Despite these limitations, Pd exhibits several advantageous properties that make it suitable for hybrid bonding. It offers adequate electrical conductivity, good resistance to oxidation, and compatibility with low temperature bonding. Furthermore, in terms of bonding conditions, Pd performs similarly to other metals and is well-suited for low temperature bonding below 200 °C. The material properties of Pd are summarized in

Table 2. Materials properties of Copper and Palladium.

	Name	Copper (Cu)	Palladium (Pd)
Atomic Property	Atomic Number	29	46
	Electron configuration	[Ar] 3d10 4s1	[Kr] 4d10
	Atomic radius (nm)	0.128	0.137
Structural Property	Crustal Structure	FCC	FCC
	Lattice constant (a) nm	0.3615	0.38902
Thermal Property	Thermal expansion at 20 °C (10 × 10−6/K)	16.64	11.77
	Melting point (°C)	1084.6	1554.9
Electrical Property	Electronegativity	1.9	2.2
	Electrical resistivity (nOhm-m)	16.78	105.4
Mechanical Property	Young’s modulus (GPa)	110~128	121
	Vickers Hardness (MPa)	343~369	400~600
	Brinell Hardness (MPa)	235~875	320~610

. Both Pd and Cu possess a face-centered cubic (FCC) crystal structure; however, Pd has a slightly larger lattice constant. This lattice mismatch can enhance atomic interdiffusion at the bonding interface, which may contribute to stronger and more uniform bonding.

Pd nanolayers can be deposited using electroplating (EP) or electroless plating (ELP). EP involves the whole surface deposition driven by metal ion reduction process and typically results in dense, oriented crystal growth. Since EP deposits metal non-selectively across a wafer, it is not easy to apply in hybrid Cu bonding to passivate Cu pads because of post-plating processes such as CMP or lithography and etching to remove undesired Pd nanolayer on dielectric areas. ELP, on the other hand, is a chemically driven process that enables are-selective deposition only on catalytic metal surfaces, effectively preventing deposition on dielectric areas as shown in Figure 1. This selective deposition capability is particularly critical in hybrid Cu bonding application. ELP offers simple processing, lower equipment cost, and higher throughput, while maintaining compatibility with low-temperature bonding environments.

This study investigates the characteristics of ELP-Pd nanolayers for low-temperature hybrid Cu bonding applications, with a particular focus on comparison with EP-Pd nanolayers. The structural and chemical properties of the Pd nanolayers are examined, along with their interfacial interaction with Cu after bonding. Notably, this study demonstrates that ELP-Pd nanolayers yield stronger and more uniform Cu-to-Cu bonds than those formed with EP-Pd nanolayers.

2. Materials and Methods

A thermally grown SiO₂ layer with a thickness of approximately 0.7 µm was formed on an 8-inch Si wafer. Subsequently, 50 nm of Ti and 1 µm of Cu were deposited by RF magnetron sputtering (SRN-110, Sorona @ S-FAB, Seoul Tech, Seoul, Republic of Korea). Then a wafer was diced into 2 cm by 5 cm sample for plating processes (DAD3350, Disco, Tokyo, Japan). The schematics of process flow is shown in Figure 2. Pd nanolayers were deposited via both EP and ELP methods, with thicknesses of 5.90 ± 0.43 nm for EP-Pd and 6.78 ± 0.58 nm for ELP-Pd, respectively. The thickness was measured at 17 different sites using Hitachi FT160 XRF Analyzer (Hitachi High-Tech, Tokoy, Japan). The 5–7 nm thickness range was carefully selected to ensure a balance between two critical requirements in hybrid bonding applications: effective surface passivation and promotion of low-temperature bonding through interdiffusion. If the Pd layer is too thin, it may not provide complete and uniform coverage of the Cu surface, resulting in incomplete oxidation prevention during the bonding processes. On the other hand, an excessively thick Pd layer may impede solid-state interdiffusion between Cu and Pd atoms, which is essential for achieving strong bonding at low temperatures. Therefore, the thickness range of 5–7 nm was chosen as a compromise that ensures both adequate oxidation resistance and sufficient atomic interdiffusion. All plating processes were conducted at EEJA Ltd. All sample was pretreated by both O₂ plasma process (100 W for 2 min) and soft wet treatment (Microfab 74 at 25 °C) prior to the plating process. The PRECIOUSFAB Pd-ST3 plating solution (EEJA Inc., Kanagawa, Japan) was used for the electroplating of the Pd nanolayer, while a newly developed plating solution currently under development was used for electroless Pd deposition. The detailed plating recipe cannot be disclosed due to the company confidentiality policy. After plating, structural and chemical analyses of the Pd nanolayers were carried out using X-ray photoelectron spectroscopy (XPS), grazing incidence X-ray diffraction (GIXRD), and transmission electron microscopy (TEM). For surface roughness atomic force microscopy (AFM) was utilized.

For die bonding, the plated sample was diced into 5 × 5 mm² dies. Prior to bonding, Ar plasma treatment was performed on the diced dies at an RF power of 100 W, with an Ar flow rate of 150 sccm under a pressure of 7.5 mTorr for 10 s, in order to clean and activate the metal surface. Die-to-die bonding was subsequently performed at 180 °C and 200 °C, respectively, for 1 h under a bonding pressure of 15 MPa (SET, Accura100, Saint-Jeoire, France, shown in Figure 3). Post-bond annealing was conducted at 200 °C for 1 h. The bonding temperature range (180–200 °C) used in this study was selected based on conditions relevant to DRAM applications, where such temperatures are considered acceptable for memory packaging processes. The bonding interfaces were evaluated by scanning acoustic tomography (SAT), scanning electron microscopy (SEM), and TEM. For mechanical strength evaluation, the bonded dies were further diced into 2 × 2 mm² samples, and shear tests were conducted using a Nordson 400Plus system (Nordson Corporation, Westlake, Ohio, USA) at a test speed of 338 μm/s and a shear height of 50 μm. Five samples were tested for each condition. During the test, the system automatically determined the failure endpoint when the real-time applied force dropped below 70% of the peak force, as identified from the load–displacement curve, indicating a force decrease of more than 30% from the peak force.

3. Results and Discussion

3.1. Characterization of Pd Nanolayer on Cu Layer

To compare the morphology of electroplated (EP) and electroless plated (ELP) Pd nanolayers, surface roughness measurements were conducted over a 1 μm × 1 μm area, and the resulting Rq values are presented in Figure 4a. The surface roughness of the bare EP-Cu layer was measured to be 4.56 nm. It is noted that a Cu wafer has not been polished by CMP in this study. After Pd deposition on Cu layer, the EP-Pd nanolayer exhibited an increased roughness of 9.42 nm, while the ELP-Pd nanolayer showed an even higher roughness of 11.88 nm. As shown in Figure 4b,c, the surface roughness of the EP-Pd nanolayer was lower than that of the ELP-Pd nanolayer, and it was composed of relatively larger grains. This difference arises from the fundamental deposition mechanisms of each method. In electroplating, Pd ions are reduced on the cathode surface by an externally applied current, resulting in a high deposition rate and a growth-driven process. In contrast, electroless plating proceeds via chemical reduction without an external current, which promotes the formation of numerous nucleation sites, making it a nucleation-driven process. As conceptually illustrated in Figure 5, the EP-Pd nanolayer shows a continuous film formed by large grains, whereas the ELP-Pd nanolayer displays a slightly less-dense structure composed of aggregated small grains, attributed to the suppressed grain growth and Pd core formation inherent in the ELP process. This structural difference is further confirmed by the TEM images in Figure 6.

To investigate the structural characteristics of the Pd nanolayers, grazing incidence X-ray diffraction (GIXRD) and transmission electron microscopy combined with fast Fourier transform (TEM-FFT) analyses were performed. GIXRD measurements were conducted at an incidence angle (ω) of 0.4°, and the corresponding results are presented in Figure 7. For the EP-Pd nanolayer, distinct diffraction peaks corresponding to metallic Pd (111) and Pd (200) were observed, along with a minor peak attributed to PdO. In contrast, the ELP-Pd nanolayer exhibited negligible Pd-related peaks, while prominent peaks associated with metallic Cu and copper oxides were detected. This indicates significant diffusion of Cu toward the surface, likely facilitated by the fine-grained microstructure of the ELP-Pd layer. These findings are further corroborated by the TEM-FFT analysis shown in Figure 8. In the EP-Pd nanolayer, crystalline planes corresponding to both Pd and PdO were identified. Meanwhile, the ELP-Pd nanolayer displayed diffraction patterns indicative of Cu₂O mostly, suggesting that surface oxidation had already taken place prior to the bonding process. The interplanar spacings for Pd, Cu, Cu₂O, and PdO have been tabulated in

Table 3. Interplanar spacings for Pd, Cu, Cu₂O, and PdO.

	Plans	d-Spacing (Å)		Plans	d-Spacing (Å)
Pd	(111)	2.245	Cu	(111)	2.086
	(200)	1.745		(200)	1.805
	(220)	1.375		(220)	1.276
PdO	(111)	3.04	Cu2O	(111)	2.466
	(200)	5.33		(200)	2.135
	(220)	2.68		(220)	1.509

as references.

Figure 9 presents the X-ray photoelectron spectroscopy (XPS) analysis performed to investigate the surface chemical composition of the Pd nanolayers. The XPS measurements were conducted at an approximate analysis depth of 2 nm from the surface. As shown in Figure 9a, the EP-Pd nanolayer was primarily composed of metallic Pd, with no detectable Cu signal. According to other reports [18,19], the binding energies of metallic Pd (Pd3d) appear near 335.3 eV and 340.6 eV, while Pd oxide peaks are typically observed in the range of 335.4 eV to 336.8 eV. In the O1s spectra of Figure 9, the peak observed around 529.6–530.4 eV corresponds to Pd oxide, while additional peaks are attributed to other oxygen species or adsorbed oxygen. These results indicate that the EP-Pd nanolayer functioned effectively as a diffusion barrier, suppressing Cu diffusion and subsequent oxidation. However, partial surface oxidation of Pd was still evident. In contrast, the ELP-Pd nanolayer displayed distinct XPS peaks corresponding to Cu and Cu₂O, indicating that Cu had diffused to the surface and undergone oxidation. This suggests that the ELP-Pd layer provided limited protection against Cu oxidation, which can be attributed to its microstructural features, such as many grain boundaries due to smaller grain size.

3.2. Characterization of Bonding Interface

Figure 10 presents SAT images of the samples bonded at 180 °C and 200 °C. In the case of the EP-Pd samples, interfacial bonding is observed at both temperatures, although the bonding integrity appears suboptimal. For the ELP-Pd samples, negligible bonding is observed at 180 °C, possibly due to rough surface and excessively oxidized surface. At 200 °C, however, the ELP-Pd samples exhibit a notably improved interfacial bonding quality. Furthermore, the SAT images reveal the presence of surface contaminants across bonded interfaces, which likely contributed to the non-uniform bonding and reduced bonding integrity. These findings underscore the need for further optimization of pre-bonding plasma surface treatments and tighter contamination control measures during the die bonding process under ambient conditions.

Figure 11 shows the interfacial microstructure of the EP-Pd sample bonded at 200 °C. As seen in Figure 11a, the bonding interface appears highly non-uniform, with regions where Cu has significantly diffused toward the interface and other regions where little to no diffusion has occurred. Although negligible Cu was present on the surface prior to bonding, non-uniform Cu diffusion has occurred due to solid-state thermomechanical behavior at bonding temperatures. In Figure 11b,c, a non-uniform copper oxide layer with a thickness ranging from approximately 10 to 130 nm was observed at the bonding interface, accompanied by a substantial presence of oxygen. This is attributed to the incomplete removal of the oxide layer formed prior to bonding by the Ar plasma pre-treatment, suggesting that the surface preparation was inadequate. Beneath this oxide layer, a thin Pd-rich Pd–Cu alloy layer was detected.

In the case of the ELP-Pd sample shown in Figure 12, partial Cu diffusion to the surface was already observed prior to bonding. During bonding, Cu further diffused along numerous grain boundaries, resulting in the formation of a copper oxide layer at the interface. Compared to the EP-Pd sample, this copper oxide layer was more uniform and thinner, and exhibited a lower oxygen concentration. In contrast, the underlying Pd–Cu alloy layer appeared Cu-rich and was more widely distributed than in the EP-Pd sample. This behavior is attributed to the lower film density and smaller grain structure of the ELP-Pd layer, which facilitated more efficient interdiffusion between Cu and Pd. Although the ELP-Pd layer exhibited higher surface roughness (approximately 2.4 nm greater Rq than EP-Pd) and Cu₂O was detected on the surface by XPS analysis prior to bonding, a relatively uniform interfacial layer was still formed after bonding. This outcome can be attributed to the small grain size and slightly agglomerated morphology of the ELP-Pd layer, which likely enhanced surface activation by Ar ion bombardment and promoted subsequent atomic interdiffusion; however, the Ar plasma process itself still requires further optimization.

Furthermore, as shown in the EDS line scan profiles in Figure 11c and Figure 12c, Pd diffusion into the Cu side extends up to approximately 65 nm in the ELP-Pd sample, whereas the corresponding diffusion depth in the EP-Pd sample is about 40 nm. This suggests that Cu–Pd interdiffusion is more pronounced in the ELP-Pd sample, which can be attributed to the smaller grain size and higher density of grain boundaries characteristic of the ELP-Pd layer.

The shear test was conducted using samples diced into 2 × 2 mm² pieces, and the shear strength values are shown in Figure 13. The average shear strength of the EP-Pd samples was 16.47 MPa with a standard deviation of 7.54 MPa, while the ELP-Pd samples exhibited a higher average shear strength of 20.58 MPa and a lower standard deviation of 5.62 MPa. Based on our experimental observations, the ELP-Pd samples exhibited higher shear strength (20.59 MPa) compared to the EP-Pd samples (16.47 MPa). This difference can be partially attributed to the improved bonding interface morphology observed in ELP-Pd. As confirmed by TEM imaging, the ELP-Pd samples demonstrated a more uniform and continuous bonding interface, which likely contributed to better mechanical integrity. Additionally, the lower oxygen content at the surface of ELP-Pd may have further facilitated stronger interfacial bonding. Although the surface roughness of ELP-Pd is slightly higher than that of EP-Pd, the presence of smaller grains with a higher density of grain boundaries at the bonding interface in ELP-Pd appears to have outweighed this factor, contributing positively to the bonding performance. To improve the bonding quality, it is essential to optimize the pre-bonding plasma treatment process to effectively remove surface oxides and ensure the formation of an activated surface.

4. Conclusions

This study aims to investigate an electroless-plated Pd nanolayer process for application in low-temperature hybrid Cu bonding. The material properties of the ELP Pd nanolayer were systematically compared with those of an EP Pd nanolayer, and their performance in low-temperature bonding was evaluated. The EP-Pd nanolayer exhibited a relatively continuous and uniform structure, with excellent Cu passivation characteristics, as indicated by negligible Cu detection on the surface. In contrast, the ELP-Pd nanolayer exhibited a smaller grain structure and slightly higher surface roughness, with detectable amounts of Cu and Cu oxides present on the surface. Nevertheless, during bonding at 200 °C, the ELP-Pd nanolayer achieved a uniform bonding interface, likely due to its small grains, and demonstrated high shear strength.

To achieve hybrid Cu bonding via a metallic nanolayer at temperatures below 200 °C, solid-state diffusion between the constituent metals is essential. In this context, the selectively deposited ELP-Pd nanolayer emerges as a promising candidate. However, for the effective implementation of ELP-Pd nanolayers, it is necessary to enhance the nanolayer’s structure to passivate Cu diffusion toward surface and to develop appropriate plasma pre-treatment conditions specifically optimized for Pd surfaces.