No abstract

The texture of electroplated Cu lines of 0.375, 0.5 and 1.5 μm widths with Ta and TiN barrier layers was analyzed using x-ray pole figure and electron backscatter diffraction (EBSD) techniques. Both techniques indicate a strong (111) fiber texture relative to the bottom surface of the trench for samples with a Ta barrier layer and a 400 °C, 30 min, postelectroplating anneal. Samples with a TiN barrier and no anneal exhibit a weak (111) texture. For both barrier layers the quality of the texture, as measured by (111) peak intensity, fraction of randomly oriented grains and (111) peak width, degrades with decreasing linewidth. EBSD data also indicate (111) texture relative to the sidewalls of the trench in samples with a Ta barrier and postelectroplating anneal. Electromigration tests at 300 °C of 0.36 μm damascene Cu lines with the same process conditions show that samples with very weak (111) texture have median time to failures that exceed those of the strongly textured Cu lines. These results indicate that diffusion at interfaces, such as the Cu/barrier and Cu/overlayer interfaces, along with diffusion along an electroplating seam play more dominant roles in electromigration failure in damascene-fabricated lines than diffusion along grain boundaries within the interconnect.

We describe a method for directly determining the strain state of passivated metal lines. Synchrotron radiation in the grazing incidence geometry is used to directly measure the in-plane interplanar spacing along the length and width of the lines, while the strain normal to the surface of the line is measured using conventional diffraction methods. The entire strain state is thereby defined. Previous work has measured out-of-plane reflections, fit them to a straight line as a trigonometric function of the angle of orientation, and extrapolated to determine the principal strains. The equivalence of the two x-ray methods on the same sample is demonstrated at room temperature before and after thermal cycling. For short time strain relaxation experiments during thermal cycling, measurement of the three principal strains leads to the direct calculation of the stress relaxation. We apply the strain determination technique to Al-0.5% Cu lines passivated with Si3N4 as the lines are thermally cycled from room temperature to 450 °C and back. The strain state, stress state, and strain relaxation of the lines are calculated at several temperatures during thermal cycling.

No abstract

No abstract

No abstract

The mechanical stress state of conventional Al and damascene Cu lines of a 0.18 pm logic technology flow have been determined using a novel X-Ray diffraction method that permits measurement of stress on an array of critical-dimension lines on the product die. The effect of high density plasma oxide deposition and the influence of low-K dielectrics on the stress state of the Al lines is described. The effect of materials properties and fabrication methodology on the stress state of damascene Cu lines is shown with measurement of mechanical stress and strain in passivated lines at room temperature and during annealing. The effect of underlayer on the damascene Cu stress state is also quantified.

The dynamic behavior of electromigration (EM) voids has been studied in situ using a field-emission scanning electron microscope fitted with a Robinson backscatter detector. A high-temperature stage has been used to minimize the temperature gradients associated with Joule heating and to allow independent control of temperature and current density. No evidence of pre-existing voids was found. The formation, growth, and motion of electromigration voids were observed and recorded photographically. The voids moved dynamically against the electron wind. No correlation between void size and void velocity was found. The static growth of EM voids was observed in some instances; however, this did not precede void motion nor did it lead to failure. Moving voids formed late in the test dominated final failure. Comparison of experimental results with void motion models reveals that the models for dynamic void motion are not consistent with experimental observations.