When interconnect feature sizes (e.g., widths) and grain sizes are comparable, their average electromigration-lifetime increases. However, reliability increase occurs for a population of lines only if microstructural weak links, such as polygranular clusters, become unlikely throughout the population. Lifetime statistics and grain structure statistics are linked in a way which must be understood in order to take advantage of lifetime improvements through line-width-dependent and line-length-dependent current density limits. An analytic model has been developed which allows calculation of the polygranular cluster length distribution as a function of line width, line length and current density, given continuous-film median grain sizes and grain size deviations. A simple analytic model for the effect of post-patterning anneals on polygranular cluster length distributions is also presented. These models are shown to be consistent with detailed computer simulations of grain structures and grain structure evolution in interconnects. Improved grain structure models will enable less-conservative integrated circuit design practice, allowing design of higher performance circuits.

With a view to analyzing failure mechanisms in interconnects, we have developed a finite element method that may be used to compute the effects of diffusion and deformation in an electrically conducting, deformable solid. The analysis accounts for large changes in the shape of the solid due to surface diffusion, grain boundary diffusion, and elastic or inelastic deformation within the grains. To illustrate the capabilities of the method, we have analyzed void migration and collapse due to current induced surface diffusion; the drift of a strip of conducting solid on a substrate; and have calculated intergranular stresses and plastic deformation induced by electromigration in a passivated, polycrystalline interconnect line.

Electromigration-induced failure of near-bamboo interconnects depends strongly on the number and lengths of both polygranular clusters and bamboo segments in a line. We have used numerical techniques to calculate the stress profile as a function of position and time for near-bamboo lines with a variety of microstructures. Our results show that coupling of stress profiles between nearby clusters can affect the time to reach a pre-defined critical, or failure, stress. The degree of coupling depends both on the spacing between clusters and the ratio of the diffusivities between polygranular and single crystal regions. Electromigration failure is therefore sensitive to the distribution and lengths of bamboo segments, and the likelihood of long polygranular clusters coinciding with adjacent short bamboo segments.

In near-bamboo interconnect lines used in advanced integrated circuits, electromigration flux divergences occur at the intersection between polycrystalline cluster segments (where grain boundaries offer a fast diffusion path), and bamboo segments (where there are no grain boundaries along the line length). For confined, passivated metal interconnects, these flux divergences are linked to the evolution of significant mechanical stresses in the metal. A quasisteady state stress distribution builds up relatively quickly in the cluster segments and remains unchanged until there is significant diffusion into the bamboo segments. The stress profile of a given cluster then becomes dependent on neighboring clusters as well as the diffusivity and flux in the separating bamboo segments. Previous analyses of electromigration failure in interconnect lines have focused on the distribution of cluster lengths and the stress build up in isolated cluster segments. In this paper, we show that the bamboo length distribution can strongly affect the interaction between clusters and the evolution of stresses in a near-bamboo interconnect during electromigration. We present simulation results, using a ratio of cluster to bamboo diffusivity Dc/Db=100, which show greater interactions and larger maximum stresses in cluster segments as the average bamboo segment length decreases and as the bamboo segment length distribution widens.

The diffusion barrier performances of MOCVD TiN films with different thicknesses and various contact layers are compared by both ex situ and in situ TEM experiments. MOCVD TiN films grown at 425°C and 10 torr without a titanium underlayer show a semi-columnar structure while films grown with a titanium layer are columnar. It is found that the former has better diffusion barrier performance for Al metallization in Si contact holes. In addition, it is found the MOCVD TiN films combined with a TiSi2 contact layer are stable up to 600°C.

The diffusion barrier properties of 100-nm-thick TiN films, both as-deposited and "stuffed", were investigated in both Al/TiN/Si and Cu/TiN/Si metallization systems. In Al/TiN/Si systems, the TiN barrier fails by the formation of both Al spikes and Si pits in the Si substrate. However, in Cu/TiN/Si systems, the failure of TiN diffusion barriers occurs by the predominant diffusion of Cu into the Si substrate, which forms dislocations along the projection of Si {111} plane and precipitates (presumably Cu-silicides) around the dislocation. In Al/TiN/Si systems, it is shown that the diffusion barrier property of TiN is significantly enhanced by "stuffing" in N2 ambient prior to Al deposition. However, in Cu/TiN/Si systems, it is found that the "stuffing" of TiN does not improve the diffusion barrier property as it does in Al/TiN/Si systems.

In this paper, material properties of low pressure, metallic mode Coherent TiN and poison mode Coherent TiN have been examined. The material properties include resistivity, crystal orientation, stress, lattice spacing, density, texture, atomic composition, bonding, roughness, hydrogen content, and grain size. Various analytical tools have been used to analyze the TiN films.

The properties of platinum silicide have been evaluated in the form of blanket films and confined lines with linewidth down to 0.15 μm. Pt films, ranging in thickness from 150Å to 1000Å, were prepared by sputter-deposition onto Si (100) blanket substrates or substrates patterned with windows of various sizes in SiO2. The samples were then annealed in a rapid thermal annealing system up to 550°C to form PtSi. The sheet resistance of silicide lines did not change significantly with linewidth. The thermal stability of the sheet resistance of PtSi was also measured for different linewidths and film thicknesses. The sheet resistance remained stable on annealing up to 850°C for a silicide film made of 250Å Pt and did not appear to be sensitive to the linewidth. The thickness dependence of the thermal stability of resistance was also evaluated.

The stress of the silicide films was measured using a laser deflection system. The asdeposited metal films were under compressive stress, but the stress turned into tensile upon annealing when silicide started to form. After PtSi was formed, the stress remained stable with annealing temperature until approximately 900°C when the stress exhibited a sharp decrease. Unlike electrical conductance, however, the breakdown temperature for stress did not strongly depend on the film thickness.

The interdependence of mechanical stress and electromigration behavior has been recognized for quite some time, but has only begun to be fully appreciated and dealt with in a meaningful manner. There have been a number of recent advances in understanding that also may lead to a change in the way that electromigration failure is viewed in real applications.

We present a new approach to understand the mechanism of "homogeneous", or Blech electromigration (EM). This phenomenon describes macroscopically homogeneous displacement of the up-wind edge of thin film lines in microelectronic devices and is responsible for openings at contact windows, "vias" and other sites of perfect diffusion flux divergence.

Our SEM, EPMA and EM drift velocity experiments have revealed the gradual transition from the microscopically homogeneous EM displacement to the highly nonhomogeneous mode wherein copious islands of residual material remain behind the drifting cathode edge of aluminum stripes. The transition is shown to occur due to an increase in either the current density, j, or in the stripe length, 1. The latter case suggests, that the transition results from the growth of the net grain boundary (GB) diffusion flux, I=le-Ib ,where Ie∝j and 1b∝1/1 are the EM flux and stress-gradient-driven back flux, respectively.

Based upon recent progress in the theory of GB grooving under "external" GB fluxes, with surface diffusion acting as the healing mechanism, grooves' propagation along the line and their merging is considered to be the micromechanism of the "homogeneous" EM. In terms of the simple model described, the transition from the slow receding of the cathode butt edge slightly wrinkled by shallow grooves (A-regime of EM) to the fast extension and merging of slot -like grooves (B-regime) accounts for the transition observed in EM mode, while in both regimes the EM displacement velocity, V, is presumed to represent the groove propagation rate.

The theory developed reduces to Blech formulae for V for the truly homogeneous A-regime and predicts quite different EM kinetics for the B-regime of microscopically nonhomogeneous EM. The latter is expected to dominate for films loaded by high current density with large grains and low surface diffusion.

The dependence obtained for the residual mass left behind the drifted edge vs the displacement velocity, V, for unpassivated aluminum stripes of various lengths, loaded by j=2-106 A/cm2 at 548K provides a good evidence in support of a new approach.

Direct real-time observations of electromigration (EM) in submicron Al interconnects were made using a special sample-stage which allowed TEM observations to be recorded while simultaneously heating and passing current through the sample. The samples consisted of 4000 Å thick Al(0.5wt%Cu) patterned over a TEM-transparent window into five runners in parallel, with linewidths 0.2, 0.3, 0.5, 0.8, and 1.0 μm. Both passivated and unpassivated samples were examined. A current density of 2 x 106A/cm2 was used with temperatures ranging from 200 - 350°C. The experiments were done using constant voltage testing, and we used a special sample design which dramatically minimized Joule-heating. Our approach has allowed us to directly observe voids form, grow, migrate, pin, fail a runner, and heal, all with respect to the detailed local microstructure of the runners.

This paper presents new results concerning diffusion at the interface between Al and its oxide, due to electromigration in Al interconnects. The results show that the phenomenon is an important and general one. Significant large-scale modification of the surface of Alalloy conductors was observed after electromigration stress under a wide range of different conditions: 1) in a number of different alloys (Al-Si, Al-Cu, Al-Si-V & Al-Si-V-Pd), 2) with and without passivation, 3) over an extended temperature range, 4) over an extended range of current density, and 5) with a number of different underlayers (SiO2, W-Ti (no vacuum-break before Al deposition) and W-Ti (oxidized surface before Al deposition)). After electromigration stressing, the surface showed height variations: at certain locations the original thickness remained, while at adjacent locations the thickness was reduced; as much as 150nm in 2500 hrs, for Al-Si on oxidized W-Ti metallization. Damage by thinning has received little attention in the past. Based on the results presented, a model for the underlying mechanism is proposed. In this model Al atoms diffuse concurrently along the interface between the Al-metal and its oxide, and through the grain boundaries, even at low stress conditions and in non-bamboo lines.

As the interconnect line width reduces to the sub-micron range, the grain distribution of Al(Cu) alloy lines reaches a bamboo structure. Understanding of the electromigration resistance of near-bamboo and bamboo structures is essential to the sub-micron reliability issue. In this study, the mass transport and failure mechanism of 1 μm wide Al(1%Cu) two-level lines with near-bamboo and bamboo structures were investigated by a drift velocity technique [1]. Our results show that interfacial (A10X/A1 interface) diffusion coupled with grain boundary diffusion determines the failure mechanism of bamboo lines.

We report on the role of surface diffusion involved in relaxation of electromigration (EM) induced compressive stresses in relation to hillock growth and EM behavior of interconnects. Two competing mechanisms of EM stress relaxation by material transport onto the surface are considered. The first is hillocking by threshold diffusional creep (TCH), with rather large blocks of material (grains or group of grains) involved in plastic flow. The second mechanism, atomic diffusion hillocking (ADH), is presumed to be a nonthreshold one, and represents atomic grain boundary (GB) diffusion stimulated by the hydrostatic stress gradient in the direction normal to the film surface. The latter process involves surface diffusion because GB diffusional flux onto the surface must be coupled with the flux of redistribution of the atoms over the surface. If ADH acts rapidly, this should prevent the build-up of the matter at the down-wind (anode) end of the stripe, and thus, eliminate the Blech EM threshold resulting from the stress-gradient along the stripe. The question as to whether GB diffusion capable of transporting atoms pushed by electron wind along the stripe is also effective in relieving compressive stress by GB migration of the surplus atoms in the normal direction, has remained open up to now. The problem is especially acute for short or/and narrow lines separated into short polycrystalline segments, where the Blech threshold effects are critical to EM reliability.

We derived the main features of the EM behavior in drift velocity test geometry assuming that both TCH and ADH are operative. The result can be compared with available and future experimental observations in order to reveal if and when the ADH mechanism with surface diffusion involved works.

The dominant mode of electromigration in polycrystalline Al thin - film conductors is along grain boundaries when the conductor width is significantly larger than the grain size. Integrated circuit feature sizes, however, have now decreased to the point where microstructures are no longer polycrystalline, but are near - bamboo. Electromigration must operate along pathways other than grain boundaries in the bamboo segments. Here drift velocity data is presented for bamboo microstructures with widths down to 0.6 μm and compared with drift data available in the literature for thin films with a variety of microstructures and bulk Al. Bamboo films show lower drift velocities and higher activation energies for drift than polycrystalline films. The data for bamboo microstructures is consistent with drift measurements performed on bulk Al indicating that the transport mechanism in bamboo films is identical to that in bulk Al.

The changes of microstructure in Al and Cu thin film lines due to electromigration have been studied using transmission electron microscopy. Grain boundary migration, inclination and dislocation activity were found to be critically involved in the electromigration induced hillock formation that can be described as three dimensional grain growth.

In order to understand the role of Sc on electromigration (EM) failure, Al interconnects with 0.1 and 0.3 wt.% Sc were tested as a function of post-pattern annealing time. In response to the evolution of the line structure, the statistics of lifetime evolved. While the addition of Sc greatly reduces the rate of evolution of the failure statistics because the grain growth rate decreases, the MTF variation was found to be very similar to that of pure Al. These observations seem to show that Sc has little influence on the kinetics of Al EM; however, it has some influence on the EM resistance of the line since it is an efficient grain refiner. Unlike Cu in Al, Sc does not seem to migrate, which may explain its lack of influence on the kinetics of Al EM.

The effect of surface and grain-boundary diffusion on interconnect reliability is addressed by extending the theory of thermal grooving to arbitrary grain-boundary flux. For a periodic array of grain boundaries, three regimes are identified: (1) equilibrium, (2) global steady state, and (3) local steady state. These regimes govern the stability of polycrystalline materials subjected to large electric (electromigration) or mechanical (stress voiding) fields, especially in thin films where grain size approximates film thickness.

Preferred crystallographic orientation and grain size distribution were characterized as a function of processing for sputtered Cu films on Ta underlayers. The Ta barrier layer was deposited at two temperatures, 30 and 100 °C. Cu was deposited at 30, 150 and 250 °C on the 30 °C Ta, and at 100, 150, 200 and 250 °C on the 100 °C Ta. In the first set of samples, with increasing deposition temperature, the Cu (111) fiber texture grew weaker and the volume fraction of randomly oriented grains increased from 0.23 to 0.74. In contrast, for the films deposited on the 100 °C Ta, with increasing deposition temperature, Cu (111) fiber texture strengthened and the fractions of randomly oriented and twinned grains decreased. Grain size was lognormally distributed in all samples and varied approximately parabolically with deposition temperature. At a given deposition temperature, median grain size in the Cu was larger in the films deposited on the 100 °C Ta. These results will be related to the microstructure of the Ta underlayers. Cu microstructure on the 100 °C Ta is shown to be influenced by textural inheritance from the Ta underlayer. Microstructure of the Cu on 30 °C Ta is discussed in terms of trace contaminants.

Substrate curvature and X-ray technique were used to study the mechanical properties of Cu films. Stress-temperature curves were measured using both methods. An additional analysis of the X-ray peak width allows us to estimate grain size and dislocation density as a function of temperature. It can be shown that a capping layer changes the mechanical properties of a Cu film strongly and that in capped films dislocation processes seem to be more important than diffusion at high temperatures.