Effects of microstructure on the formation, shape, and motion of voids during electromigration in passivated copper inte
- PDF / 829,043 Bytes
- 9 Pages / 612 x 792 pts (letter) Page_size
- 22 Downloads / 157 Views
In situ scanning electron microscope observations have been performed on passivated damascene Cu interconnect segments of different widths during accelerated electromigration tests. In some cases, voids form and grow at the cathode. However, an alternative failure mode is also observed, during which voids form distant from the cathode end of the interconnect segment and drift toward the cathode, where they eventually lead to failure. The number of observations of this failure mode increased with increasing linewidth. During void motion, the shape and the velocity of the drifting voids varied significantly. Postmortem electron backscattered diffraction (EBSD) analysis was performed after in situ testing, and a correlation of EBSD data with the in situ observations reveals that locations at which voids form, their shape evolution, and their motion all strongly depend on the locations of grain boundaries and the crystallographic orientations of neighboring grains.
I. INTRODUCTION
It is well established that electromigration (EM) in confined metal interconnects can lead to tensile stresses that can cause void nucleation and growth.1 In situ observations of voiding in aluminum-based interconnects revealed that once voids nucleated, they sometimes became mobile and drifted toward the cathode end of interconnect segments, often changing shape as they migrated.2–5 In Al-based interconnects, electromigration along grain boundaries is faster than electromigration at interfaces, and studies of passivated interconnects were carried out in lines with multiple grain boundaries across their width so that voids drifted along grain boundaries.2–4 In unpassivated Al-based interconnects with bamboo structures, voids were also observed to drift, and shape changes were thought to be associated with the motion of a void from one grain into another.5 EM in copper-based interconnects differs from EM in Al-based interconnects in a number of important ways. In Cu-based interconnects, EM is faster at the interface between the Cu and the overlying dielectric diffusion barrier than at grain boundaries.6 EM at the top surface
a)
Present address: Institut für Materialforschung II, Forschungszentrum Karlsruhe, Postfach 3640, 76021 Karlsruhe, Germany. b) Address all correspondence to this author. e-mail: [email protected] DOI: 10.1557/JMR.2008.0054 J. Mater. Res., Vol. 23, No. 2, Feb 2008
http://journals.cambridge.org
Downloaded: 02 Apr 2015
dominates because the Cu is in contact with refractory metal diffusion layers on the sides and at the bottom, and EM at these interfaces is slower than at the Cu– passivation layer interface. In Cu-based via-terminated interconnect segments, EM is usually blocked by refractory metal liners found at the base of the vias. Therefore, when an electrical current is applied, EM leads to stress evolution such that a maximum tensile stress occurs at the cathode-end via. Voids are typically located at the interface between Cu and the top passivation layer, very close to the via at the cathode. Nucleation occurs at t
Data Loading...
