| The introduction of Cu and low-k dielectric materials and continuing scaling of on-chip interconnects raise reliability concerns on electromigration. Failures occur mainly by void growth at the line end. The corresponding failure times usually follow a lognormal distribution with the median lifetime depending on the quality of the interface, which controls the mass transport. Since electromigration experiments are conducted at higher current densities and temperatures compared to operating conditions, extrapolations are needed to assess reliability at operating conditions. The extrapolated lifetimes depend exponentially on the lognormal standard deviation sigma. The objective of this study is to identify intrinsic parameters that control the electromigration statistics and failure mechanism in Cu interconnects. To accomplish this task, lifetimes, void evolution and void size distributions were analyzed in detail as a function of failure criterion. Experiments were performed on 0.18mum wide Cu interconnects with tests terminated after certain amounts of resistance increases, or after a specified test time. Void size distributions of resistance-based and time-based electromigration tests were obtained using microscopy. The lifetime and void size distributions were found to follow lognormal distribution functions. The sigma values of these distributions decrease with increasing test time. The statistics of resistance-based void size distributions can be simulated by considering geometrical variations of the void shape. In contrast, the characteristics of time-based void size distributions require consideration of kinetic aspects of the electromigration process. For this purpose, a model was developed including geometrical and experimental factors of the electromigration test, as well as kinetic aspects of the mass transport process, such as differences in interface diffusivity between the lines. The variation in diffusivities at the line ends arises from differences in the interface structure as a result of varying Cu grain orientations. It will be shown that the statistics of electromigration lifetime distributions can be adequately modeled by combining the time-based and resistance-based void area measurements. To examine the validity of the model describing electromigration lifetime statistics, the effect of various process changes was analyzed. These include a variation in line height, a change in capping layer, and a comparison between single and dual damascene structures. |
