Reliability of copper interconnects in integrated circuits

Abstract

As dimensions shrink and current densities increase, the reliability of metal interconnects becomes a serious concern. In copper interconnects, the dominant diffusion path is along the interface between the copper and the top passivation layer (usually Si3N4). One of the predominant failure mechanisms in Cu has been open-circuit failure due to electromigration-induced void nucleation and growth near the cathode ends of interconnect segments. However, results from accelerated electromigration tests show that the simple failure analyses based on simple void nucleation and growth can not explain the wide range of times-to-failure that are observed, suggesting that other types of failure mechanisms are present. In this thesis, by devising and performing unique experiments through the development of an electromigration simulation tool, unexpected complex failure mechanisms have been identified that have significant effects on the reliability of copper interconnects. A simulation tool was developed by implementing the one-dimensional non-linear differential equation model first described by Korhonen et al. By applying an implicit method (Backward Euler method), the calculation time was significantly reduced, and stability increased, compared to previous tools based on explicit methods (Forward Euler method).

(cont.) The tool was crosschecked with experimental results by comparing void growth rates in simulations and experiments. Using this tool, one can simulate stress and atomic concentration states over the entire length of an interconnect segment or throughout a multi-segment interconnect tree, to identify analyze possible failure locations and mechanisms. Experiments were carried out on dotted-i structures, where two 25jim-lomg segments were connected by a via in the middle. Electrical currents were applied to the two segments independently, and lifetime effects of adjacent segments were determined. Using the simulation tool and calculations, it was shown that adjacent segments have a significant effect on a segment's stress state, even if the adjacent segment has no electrical current. This explains experimental observations. This also suggests that for reliability analyses to be accurate, the states of all adjacent segments must be considered, including the ones without electrical current. In a second set of experiments, the importance of pre-existing voids was investigated. Using in-situ scanning electron microscopy, voids away from the cathode were observed. These voids grew and drifted toward the cathode and the shape of the voids were found to be closely related to the texture and stress state of individual grains in the interconnect.

(cont.) The drift velocity of voids was shown to be directly proportional to surface diffusivity. Electromigration tests on unpassivated samples were performed under vacuum to obtain the surface diffusivity of copper and its dependence on texture orientations. Simulation results show that pre-existing voids cause void growth away from the cathode. Subsequent failure mechanisms differ depending on the location of the pre-existing void and the critical void volume for de-pinning from grain boundaries. If pre-existing voids are present, void-growth-limited failure is expected in interconnects at low current densities, due to growth of pre-existing void, and the lifetimes are expected to scale inversely with j. However, at higher current densities (typical for accelerated testing), failure can occur through nucleation of new voids at the cathode (so that lifetimes scale inversely with j2), or through a mixture of nucleation of new voids and growth of pre-existing voids. These effects must be taken into account to accurately project the reliability of interconnects under service conditions, based on experiments carried out under accelerated conditions.