In-situ electromigration experiments were carried out recently to investigate the electromigration failure mechanisms in dual-damascene Cu test structures. It was found that electromigration-induced void first nucleates at locations far from the cathode, then moves along the Cu/dielectric cap interface in opposite direction of electron flow, and eventually causes void agglomeration at the via in the cathode end to open the interconnect. In the present study, two methods have been explored in order to improve the electromigration reliability in real Cu damascene interconnects. (1) Immersion Sn surface treatment was employed after CMP and before SiN deposition. A 20nm thick Cu3Sn intermetallic compound overlayer on Cu interconnect surfaces was found to be effective in blocking the dominant surface diffusion path, thus resulting in close to one order of magnitude improvement in electromigration lifetime. This improvement may be explained on the basis of the terrace-ledge-kink (TLK) model in which the supply of Cu adatoms by the dissociation of atoms from the kinks on the Cu surface steps is hindered by a stronger chemical binding of Sn atoms to the kink sites. The mode of electromigration failures seems to have changed from surface diffusion induced void formation at the cathode via corner to interfacial and grain boundary diffusion induced void formation in the interconnect line. (2) A dual damascene structure with an additional 25 nm Ta diffusion barrier embedded into the upper Cu layer was fabricated. This thin layer of diffusion barrier blocked voids from propagating into the via, thus eliminating the previously reported failure mechanism. With this structure, a lifetime improvement of at least 40 times was achieved. Analysis on failed samples suggested that failures in samples with the embedded Ta barrier layer occurred at the bottom of the via, which were caused by void migration along the bottom of the Cu lines. Furthermore, with this test structure, electromigration induced drift displacement can be accurately measured with a linear dependence on time. Measurement was conducted at a series of temperatures to obtain the Cu/capping interface diffusion controlled activation energy.
展开▼