Investigation of the Fundamental Reliability Unit for Cu Dual-Damascene Metallization

Abstract

An investigation has been carried out to determine the fundamental reliability unit of copper dual-damascene metallization. Electromigration experiments have been carried out on straight via-to-via interconnects in the lower metal (M1) and the upper metal (M2), and in a simple interconnect tree structure consisting of straight via-to-via line with an extra via in the middle of the line (a "dotted-I"). Multiple failure mechanisms have been observed during electromigration testing of via-to-via Cu interconnects. The failure times of the M2 test structures are significantly longer than that of identical M1 structures. It is proposed that this asymmetry is the result of a difference in the location of void formation and growth, which is believed to be related to the ease of electromigration-induced void nucleation and growth at the Cu/Si₃N₄ interface. However, voids were also detected in the vias instead of in the Cu lines for some cases of early failure of the test lines. These early failures are suspected to be related to the integrity and reliability of the Cu via. Different magnitudes and directions of electrical current were applied independently in two segments of the interconnect tree structure. As with Al-based interconnects, the reliability of a segment in this tree strongly depends on the stress conditions of the connected segment. Beyond this, there are important differences in the results obtained under similar test conditions for Al-based and Cu-based interconnect trees. These differences are thought to be associated with variations in the architectural schemes of the two metallizations. The absence of a conducting electromigration-resistant overlayer in Cu technology allows smaller voids to cause failure in Cu compared to Al. Moreover, the Si₃N₄ overlayer that serves as an interlevel diffusion barrier provides sites for easy nucleation of voids and also provides a high diffusivity path for electromigration. The results reported here suggest that while segments are not the fundamental reliability unit for circuit-level reliability assessments for Al or Cu, vias, rather than trees, might be the appropriate fundamental units for the assessment of Cu reliability.