The microstructure of conductor lines depends on many parameters, such as the core material deposition technique, the barrier layer material and its deposition technique, the copper seed layer deposition technique and thickness, and the line width. As an example, for a given line width, the grain size may differ significantly, when the copper deposition technique changes from chemical vapor deposition (CVD) to electroplating [36].
In polycrystalline lines the network of grains provides grain boundaries running parallel to the electric current, so that grain boundary diffusion may significantly contribute to material diffusion. Moreover, the microstructure provides grain boundary triple points, which are sites of flux divergence, where voids are seen to nucleate. Grain boundary diffusion was certainly the dominant transport mechanism in Al interconnects. For Cu lines, however, this is not so evident, because diffusion along interfaces appears as the dominant mechanism. Nevertheless, it has been shown that larger Cu grains lead to longer EM lifetimes compared to lines with smaller grains [36].
The microstructure influence on EM failure tends to be reduced as the copper grain sizes are comparable to the line width. In this case, the interconnect presents a bamboo-like structure and, therefore, grain boundaries cannot provide a continuous path for fast diffusion. However, local variations of the microstructure affect Cu diffusivity along the copper/capping layer interface, which dependens on the orientation of individual grains. In this way, the difference in interfacial diffusion of neighboring grains leads to flux divergences located at the intersection of the copper/capping layer interface with the grain boundary formed by these grains.
From the above observation, it is argued that the texture of the line can have a significant impact on the electromigration behavior [37]. It was observed that lines with a stronger (111) texture have longer EM lifetimes [36,38]. This is attributed to the lower Cu diffusivity at (111) oriented grains in comparison to (100) and (110) surfaces [39]. Moreover, the texture distribution is seen to have an important effect for both void nucleation [40,41] and evolution [42].
Fayad et al.[43] showed that the lognormal standard deviation of EM lifetimes can be explained by the dependence of the diffusion along the copper/capping layer interface on the orientation of the grains. Moreover, copper grain sizes seem to follow lognormal distributions in typical dual-damascene process technology [20]. Therefore, it has been argued that the lognormal distribution of EM lifetimes is related to microstructure features [44].