Copper was deposited onto TiN by low pressure metal-organic chemical vapour deposition, using hfacCu(I)TMVS and argon carrier gas. The effects of the deposition temperature on the growth of copper films were investigated by observing the surface morphology and the cross sectional morphology of copper films. At the initial stage of growth, copper films tended to have the island-like growth mode, irrespective of the deposition temperatures. It was also observed that the aspect ratio(=height to width) of the islands gradually increased as the deposition temperature increased. The poorer movability of the copper atoms at the higher deposition temperature was evaluated on the basis of hindering effect by the following copper deposits.

Surface passivation of Cu by annealing Cu/Al multilayer films (Cu/Al/Cu/Al/Cu) on SiO2 has been demonstrated. Cu/Al multilayer films were annealed from 300°C to 500°C. The buried Al is transported to the Cu surface during annealing, forming A12O3 at the exposed surface. The Cu in annealed Cu/Al multilayer films shows no oxidation when heated in air at 300°C for 24 hrs. The electrical resistivity of these films is 2.3 μΩ cm.

The feasibility of Si-implanted TiN as a diffusion barrier between Cu and Si was investigated. Barrier effectiveness was evaluated via reverse leakage current of Cu/TixSiyNz/Si diodes as a function of post-deposition annealing temperature and time, and was found to depend heavily on the film composition and microstructure. TiN implanted with Si28, l0keV, 5xl016ions/cm2 formed an amorphous ternary TixSiyNz layer whose performance as a barrier to Cu diffusion exceeded that of unimplanted, polycrystalline TiN. Results from current-voltage, transmission electron microscopy (TEM), and Auger depth profiling measurements will be presented. The relationship between Si-implantation dose, TixSiyNz structure and reverse leakage current of Cu/TixSiyNz/Si diodes will be discussed, along with implications as to the suitability of these structures in Cu metallization.

In order to clarify the relationship between Al line reliability and film microstructure, especially grain boundary structure and crystal texture, we have tested three kinds of highly textured Al lines, namely, single-crystal Al line, quasi-single-crystal Al line and hypertextured Al line, and two kinds of conventional Al lines deposited on TiN/Ti and on SiO2. Consequently, the empirical relation between the electromigration (EM) lifetime of Al line † and the (111) full width at half maximum (FWHM) value ω is described by † ∝ ω-2 [1]. This improvement of Al line reliability results from as following reasons; firstly, homogeneous microstructure and high activation energy of 1.28eV for the single-crystal Al line (ω=0.18°); secondly, sub-grain boundaries which consisted of dislocation arrays found in the quasi-single-crystal Al line (ω=0.26°) has turned out to be no more effective mass transport paths because dislocation lines are perpendicular to the direction of electron wind. Although there exist plural grain boundary diffusion paths in the newly developed hypertextured Al line (ω=0.5°) formed by using an amorphous Ta-Al underlayer {1], the vacancy flux along the line has been suppressed to the same order of magnitude of single crystal line. It has been clarified that the decrease of FWHM value has promoted the formation of sub-grain boundaries and low-angle boundaries with detailed orientation analysis of individual grains in the hypertextured film. The longer EM lifetime for the hypertextured Al line is considered to be due to the small grain boundary diffusivities for these stable grain boundaries, and this diffusivity reduction resulted in the suppression of void/hillock pair in the Al lines. These results have confirmed that controlling texture and/or grain boundary itself is a promising approach to develop reliable Al lines which withstand higher current densities required in future ULSIs.

The microtexture of thin film aluminum has been studied using Electron Backscatter Diffraction (EBSD). The results indicate that, while the overwhelming majority of polycrystallite grains have a <111> fiber texture with the < 111 > axis within -5° of the surface normal, 1-2 % of the grains deposited on SiO2 show large deviations off of this axis. The off-texture grains are typically much smaller than those that are <111> textured. These observations have been made for Al(Cu) films with several different Cu concentrations and for different substrate materials. The statistical distributions of the orientations in the different cases are presented. Electrical stressing of 0.8 um Al(0.5%Cu) lines resulted in elimination of the small number of grains with off-<l 11> orientation.

Al-Cu lines of submicron width ordinarily have quasi-bamboo microstructures that contain both bamboo grains and polygranular segments. The usual mechanism of internal electro-migration failure in such lines involves the formation of a transgranular void across a bamboo grain at the upstream end of a long, polygranular segment. The voiding is preceded by the depletion of Cu from both the polygranular segment and the upstream bamboo grain. At least three metallurgical mechanisms are available to inhibit this failure mechanism and improve the useful lifetime of the line. First, the line can be annealed after patterning to break up long polygranular segments. Second, the line can be aged at relatively low temperature to stabilize the distribution of Cu and retard its depletion from polygranular segments. Third, the line can be subjected to reverse currents to restore the distribution of Cu and provide a Cu reservoir. Each of these methods has been demonstrated in the laboratory environment.

The reliability of Al-Cu interconnects is a strong function of both grain structure and the Cu content and distribution. Because pre-patteming and post-patterning grain growth are affected by the presence of Al2Cu precipitates, the effects of grain structure and Cu distribution are interrelated. We have studied the effect of Al2Cu precipitate spacing on interconnect reliability independent of grain size effects. This was accomplished by varying the thermal history of the continuous films to produce samples with identical grain structures, but different precipitate spacings. The precipitate spacing in continuous films was varied by an order of magnitude, from 2.6 μm to 25. pm. Lines were then patterned from each sample and tested. Lifetime testing results show that the reliability of lines with smaller precipitate spacings can be roughly an order of magnitude higher than that of lines with a larger precipitate spacing. We have also qualitatively studied the effects of Cu distribution on post-patterning grain growth. We found that precipitates retard post-patterning grain evolution, leading to reduced reliability. These results suggest strategies for developing thermal histories which will lead to optimal reliability.

Comparison of theory and experiment is critical in microelectronic reliability. In this work, a model which has been previously used for electromigration time to failure predictions (developed by Harrison) is extended to include predictions for numbers of voids and void sizes. Predictions from the model are compared to experimental measurements. The model makes reasonable predictions for mean lifetimes, numbers of voids, and void areas with very few free parameters. The model, however, does not adequately reproduce the range of lifetimes found in experiment.

The volumes of slit, edge, erosion and erosion/slit voids in stressed and electromigrated aluminum conductor lines were quantitatively determined with low resolution standard and high resolution enhanced tips by atomic force microscopy. These three-dimensional results were compared to semiquantitative determinations of void volumes extrapolated from two-dimensional backscattered scanning electron microscopy area determinations of the passivated aluminum conductor. After the passivation was removed by plasma etching, void volumes were also determined from two-dimensional scanning electron microscopy micrographs. The volumes of the nearest hillocks on the anodic side of the voids were quantitatively determined by atomic force microscopy, and these hillock volumes were determined to be independent of the respective void volumes.

In this paper, experimental results for electromigration void morphology and failure time distribution are presented as a function of test condition, grain size distribution and conductor linewidth. Unpassivated Al-Cu(1.5%)-Si(1.5%) conductors with linewidths in the range lum ≤W ≤l0μm were subjected to accelerated temperature stresses of T=150° C and 250°C and current stresses in the range j= 0.5-4X106 A-cm-2, for each temperature stress. Unlike previous reports, erosion-like voids dominated at all current densities and temperatures for both W/D50≤1 and W/D50 > 1. The void distribution within a line was strongly dependent on test conditions and linewidth. The critical current density was extracted from electromigration lifetimes measured as a function of current density. The results show that the linewidth dependence of electromigration lifetimes depend strongly on the Blech relation.

The work reported here concerns the effect of grain structure on electromigration failure in pure Al and Al-2wt.%Cu-lwt.%Si lines. The grain structures of fine lines were controlled by annealing after patterning to promote the formation of "bamboo" structures. Significant improvements in the median time to failure (MTF) and the deviation of the time to failure (DTF) were observed with the development of near-bamboo structures with polygranular-segment lengths shorter than ~ 5 μm. The most common failure sites are voids or slits across bamboo grains at the upstream ends of polygranular segments. The time-to-failure decreases with the polygranular segment length, and can be significantly enhanced by controlling the grain structure.

Unpassivated 2.1μm and 1.1μm wide lines of Al-4wt.%Cu have been electromigration-stressed to failure under accelerated testing conditions (up to 300 °C and 2 x 1010 A m-2). The line resistance and microstructure and the development of electromigration damage are discontinuously recorded. Distributions of damage sites (hillocks and voids) along the line lengths are measured, and the evolution of the distributions with stressing time is followed. The distances between damage sites show a clear correlation with the grain structure. Damage and failure sites are concentrated near the end of the lines at which electrons enter.

Microstructure is an important factor determining the lifetime of Al metallization lines. Deposition conditions, substrate material, alloying elements, and anneal treatments are the key parameters that influence microstructure. In this work we explore the use of graphoepitaxy as a tool for additional control over the grain structure of metallization lines. Onto a submicrometer topography in SiO2 (viz., a large number of parallel grooves), a pure Al film is grown by dc magnetron sputtering, followed by an in situ rapid thermal anneal. The topography of the annealed Al is investigated by cross section SEM. It is observed that if it is heated up to its melting point, Al fills the grooves and leaves the ridges between the grooves uncovered. X-Ray Diffraction, TEM, and Backscatter Kikuchi Diffraction are used to determine the global as well as the local crystallographic orientation of the grains in the quenched aluminum. The analyses are performed for various anneal and cool down treatments. Depending on the treatment, the Al lines in the grooves are either polycrystalline with an almost perfect (111) texture, or single crystalline but with a gradual change of 0.067°/μm in orientation. In the latter case, there is no preferred orientation. The single crystalline Al lines will be used as a starting point for the fabrication of model systems for fundamental electromigration studies.

Copper interconnect test stripes were fabricated by electron-beam lithography, evaporation and lift-off to linewidths as small as 140 nm. Grain structure in the as-deposited films exhibited smaller standard deviation than long-term annealed films. Grain growth during the anneal step increased maximum grain size by 2.5 times. Electromigration lifetime testing indicated these interconnects have 2-4 orders-of-magnitude greater normal-use lifetimes than large cross-sectional area, large-grain Al-alloy interconnects.

Stress-voiding and electromigration have become urgent reliability concerns at the decrease of interconnect dimensions to submicron size. Severe stress voiding may arise in multilevel metallizations, particularly at contact and via regimes where the thermal stresses are highest. Via and contact structures with the attached refractory layers usually block entirely the electromigration flux; hence, EM voids are likely to form here. In this paper we first model the migration of atoms due to gradients in composition, stress, and electric potential. We seek a formulation in the spirit of irreversible thermodynamics, i.e. without reference to any specific micromechanisms of diffusion. We then apply the general developments to predict the stress evolution and void growth in Al based interconnect lines confined between entirely blocking Wstuds. We also analyze the effects of Cu depletion at vias, either by interfacial or grain boundary diffusion, on the stress evolution and void growth. Finally we demonstrate that the change in the line resistance due to void growth will display apparent incubation times even if the void grows linearly with time.

Failure of the tungsten-filled via interconnect structure is modelled. Two mechanisms contributing to void growth are considered: relief of stress due to differential thermal expansion, and electromigration. A self-limiting void volume is predicted, resulting in a self-limiting resistance increase which is a function of structure geometry and the void morphology. The relative contributions of the two mechanisms change significantly as the temperature and current density are reduced from accelerated test conditions, affecting the extrapolation of test results to service conditions. Modified procedures for extrapolating lifetimes are discussed, as well as suggested process changes to improve reliability.

TiN/Ti films employed as glue layers for a blanket/etchback W plug technology show a significant surface contamination after the W-etchback process. An Al film deposited directly on a contaminated TiN/Ti underlayer exhibits a very weak texture, increasing the susceptibility to electromigration. Reliability tests confirm that the electromigration performance of a TiN/AlCu/TiN/Ti metallization system with a W plug technology, is significantly inferior to interconnecs that do not employ this plug technology. To improve the electromigration performance of this metallization system, several TiN surface cleaning treatments were evaluated. Also, the impact of the implementation of refractory metal buffer layers, before Al deposition was investigated. Analysis was performed by means of several surface analysis techniques, Al texture measurements and wafer-level reliability tests. It has been demonstrated that the contamination needs to be removed completely, or to be covered by a refractory buffer layer in order to restore the Al texture and to recover the proper electromigration resistance.

Continued miniaturization of conductor geometry below 0.5μm has by a concomitant decrease in Al deposition temperature, typically below 300°C. The degree of reliability exhibited by these films is strongly dependent on the grain size distribution and metallurgical configuration. This investigation focuses on the impact of post-deposition processing on changes in conductor microstructure and electromigration for films deposited at low temperature. AlCu(l%)Si(l%) was deposited at 300°C on PECVD phosphosilicate glass. The impact of post-deposition thermal budget on the as deposited grain size and distribution, preferred orientation and stress/strain states were analyzed using scanning electron microscopy, X-Ray diffraction and bending beam technique, respectively. These characteristics were also measured on unpassivated films subjected to the same thermal budget as SiO2 passivated films so that the geometric confinement and Al-SiO2 surface interaction could be quantified. Electromigration characteristics were measured for linewidths in the range W=1.0μm to W=10μm, for passivated and unpassivated films. A direct correlation between passivation and grain size was observed for both failure modes. This paper will also discuss the relationship between film stress and preferred orientation, the observed failure modes and their linewidth dependence.

Failure mechanisms of TiSi2/W-Ti(15%)/Al-Si(l%)-Cu(2%) multilayer thin film contact to silicon shallow junction (0,25 μm) have been investigated. Both a fast temperature cycling (with 300 ms thermal cycle period) and a high constant temperature were used as means for the degradation processes acceleration. We have shown that stress time could be reduced by the factor of 10 or more if the temperature of adjacent contact region was cycled by a pulsed electrical current passed through the semiconductor region lying under the investigated contacts. Our data have also indicated that W-Ti barrier layer serves to reduce drastically the electromigration degradation of thin film contacts for failures caused by an increase in junction leakage on the one hand and on the other hand it serves to intensify the electromigration degradation caused by an increase in contact resistance. These distinctions have been explained through structural-morphological investigations of the contact layers surface morphology. They showed a reduction in size of crystallines of the sputter deposited Al-Si-Cu top layer with W-Ti barrier layer present.

2 mm long, 0.8 μm wide and 0.8 μ thick Al-Si-Cu lines, passivated with a 1.14 μm thick PETEOS SiO2 were found to have a lower electromigration lifetime compared to identical unpassivated lines. The high tensile stress in the passivated lines is assumed to accelerate electromigration failure. This is confirmed by the very low activation energy for electromigration failure in the passivated lines and by the multiple events of recovery during and after electromigration testing, suggesting failure sites are slit voids.