Voiding Phenomena in Copper-Copper Bonded Structures: Role of Creep

Wednesday, 8 October 2014: 15:40
Expo Center, 1st Floor, Universal 9 (Moon Palace Resort)
P. Gondcharton, B. Imbert (CEA, LETI, MINATEC Campus), L. Benaissa (CEA-LETI MINATEC Campus), and M. Verdier (CNRS, SIMaP)
For its good resistivity and reliability, copper remains the most widely used material for metal interconnections since the early 1990s [1]. Concerning the 3D-integration technological transition for next generation integrated circuits, copper still remains interesting as a bonding material. In this specific configuration, this metal allows wafer or die level assembly of active component strata with high bonding strength and it also ensures interlayer electrical conduction, reducing interconnection delays and increasing overall performances [2-3]. In the process temperature range from 300°C to 400°C, the voiding mechanism that drives the structure failure [4-5] shows similar characteristics to stress-induced voiding previously reported in basic encapsulated copper interconnections [6]. However, regardless of the bonding technique, void density, shape and location exhibit variations which cannot be explained by the mechanisms previously established. In this paper, we investigate several factors, including copper surfaces roughness, copper layer microstructure and bonding processing parameters to weight their impact on the voiding phenomenon. Better understanding of void formation has enabled the achievement of wafer assemblies exhibiting much higher reliability and bonding quality.

200mm thermally oxidized silicon substrates are used in this study. 500nm-thick copper layer are then deposited on a 20nm-thick titanium nitride layer which is used to prevent copper diffusion into silicon substrate. Copper-copper bonding is studied through two different bonding techniques: direct bonding at room temperature under atmospheric pressure after surface activation by chemical and mechanical polishing (CMP activated bonding) and thermocompression bonding (TC bonding) using a combination of a thermal budget and compressive stress. In both cases, additional post-bonding annealing is accomplished under nitrogen atmosphere.

In our bonded copper layers configuration, voiding phenomenon is supposed to have different origins:

- Residues of the copper surface roughness which is brought into contact during bonding process and trap some volume

- Presence of metal oxide on sample surfaces and its subsequent degradation by annealing

- Nucleation and growth of cavities due to standard metallurgical creep mechanisms of copper layers during its thermomechanical loading

In order to evaluate roughness impact on the voiding phenomenon, bonding was performed thanks the two different techniques on copper surfaces exhibiting identical surface topology properties (post CMP treatments) (Fig. 1a). In order to aggravate void creation, TC bonding was also performed with extreme parameters in terms of pressure values and copper surface roughness (Fig. 1b). Void densities and volumes were estimated by high resolution Secondary Electron Microscopy (SEM) cross-sections observations (Fig. 2). A special attention was paid to imaging conditions and analysis. Indeed we operated data collection over large sections of about 40µm. In the context of this study we were able to appreciate significant differences between bonded samples. More specifically we focused on volume variations and size distributions which enabled us for instance to highlight the role of roughness in voiding phenomenon.

Other possible causes for void generation were also investigated using dedicated experiments:

- Gold-gold TC bonding was performed as a bonding reference system involving oxide free metallic layers

- Bonding of copper layers carefully deposited on copper bulk substrates to highlight the influence of thermomechanical stress (Fig. 3)

In the light of these results, the impact of the different factors will be discussed and the main source of voiding due to creep deformation will be demonstrated.


Fig 1: 20 x 20 µm² AFM scans of copper surface. a- In CMP bonding configuration: RMS roughness is 0.2 nm. b- In TC bonding configuration: RMS roughness is 22 nm.

Fig 2: SEM cross section. a – On CMP activated bonding after a post bonding annealing at 400°C. b – On TC bonding performed at 400°C under a pressure of 2.9 MPa on high-roughness surfaces. Voids are arrowed.

Fig 3: SEM cross section of TC bonding involving copper layers deposited directly on copper bulk substrate. Voids are arrowed.


The authors would like to thanks the French Direction Générale de l’Armement (DGA), the French Direction Générale de la Compétitivité de l’Industrie et des Services (DGCIS) and SOITECSA for its financial support.


[1] P. C. Andricacos, et al., IBM J. Res. Dev., vol. 42, no. 5, pp. 567–574, Sep. 1998.

[2] N. Sillon, et al., pp. 1 –4., IEDM 2008.

[3] R. Taibi, et al., pp. 219–225, ECTC 2010.

[4] C. S. Tan, et al., Appl. Phys. Lett., vol. 87, no. 20, pp. 201909, 2005.

[5] P. Gondcharton, et al., MRS Online Proc. Libr., vol. 1559, 2013, doi : 10.1557/opl.2013.718.

[6] H. Okabayashi, Mater. Sci. Eng. R Rep., vol. 11, no. 5, pp. 191–241, 1993.