Diffusional mechanisms of electromigration and stress relaxation involve the flow of atoms in response to a gradient in chemical potential along an interface. This gradient in chemical potential may be provided by the component of an electric field parallel to the interface, or it may be established by the normal component of stresses along it. In either case, considerations of continuity of the potential dictate that diffusive flow must also be induced along any other boundary that intersects the interface. As an example, in this paper, a model system that contains grain boundaries normal to an applied electric field is analyzed. While the electric field does not directly induce diffusion along these grain boundaries, it is shown that a complimentary flux must be induced along them. The effect of this flux on electromigration is discussed in this paper. Furthermore, it is well-known that non-homogeneous diffusion of matter along boundaries induces elastic distortions and stress gradients. These in turn, influence the diffusion process. The effect of these elastic distortions on the atomic flux has been examined by considering diffusion along a single interface in an elastic medium. Prior studies of diffusional cavity growth have established the magnitudes of non-dimensional time-scales over which the deposition of atoms along the grain boundaries can be assumed to be essentially uniform. Such an assumption considerably simplifies analyses for stress evolution in these problems. The appropriate time-scales over which such a simplification can be made for electromigration are discussed in this paper, and illustrated by some model calculations.

A finite-element model is used to simulate electromigration voiding and failure in Al-Cu interconnects. The modeling includes the fluxes of Al and Cu atoms in the grain-boundary network, and is used to study the mechanisms by which Cu as a solute improves the reliability of Al metallization. A possible mechanism for the Cu to reduce the grain-boundary diffusiv-ity of Al is explored, and it is shown that with appropriate parameters this can reproduce the observed increases in lifetime. Also investigated are the effects of electromigration-induced non-uniformity in the Cu distribution and the role of Al2Cu precipitates as reservoirs for Cu.

We present a modeling strategy for assessing the reliability cost for improved performance from modified interconnect structures. We have studied air gaps which have been deliberately introduced in the passivation between aluminum interconnect lines as a means for increasing transmission speed by decreasing dielectric capacitance. The models allow examination of tradeoffs between improved circuit performance and decreased reliability due to dielectric cracking. Stresses in the dielectric due to electromigration in the metal were modeled using finite element analysis. These stresses were used to compute an estimate of the mean time to failure relative to the case with no air gap using an electromigration failure model from MIT.

A tool for 3-D modeling of EM and SM in interconnect lines has been developed based on a commercial finite element code. After detailing the approach, we focus on the verification of the simulator by comparing the results of 1-D analytic and FEM simulations, and then we apply the simulator to interconnect line segments with a specified grain structure.

Finite element solutions to problems of electromigration and stress-driven diffusion in polycrystalline films are presented. The numerical formulation accounts for diffusion along grain boundaries and through the bulk. The procedures are validated by comparing the finite element solutions to an existing analytic solution of an idealized, isolated grain boundary lying parallel to current flow. A simple polycrystalline interconnect is then considered and the effects of diffusion along each grain boundary and through the bulk are each added separately to the analysis so that the importance of each mechanism can be ascertained. In some cases, the inclusion of additional diffusion paths result in significant reductions in predicted stress levels.

Stress voiding is caused by the difference in the thermal expansion coefficients of the metallization lines and the surrounding passivation. Volumes of individual stress voids are measured as a function of stress time at 200°C, 180°C, and 150°C in 1-μm-wide AlCu(1%wt) integrated circuit metallization lines. The time needed for void growth to saturate, and the total void volume depend on the stress temperature. If the void growth is regarded as an isothermal phase transformation in which voids are formed as precipitates at the void nucleation sites, the void volume growth is accurately described by the Avrami equation. Depending on the temperature, the time power n ranges between 0.5 and 1.5. First principle calculations are in excellent agreement with the measurements. The impact of the above results on electromigration testing is discussed.

We present in this work a series of detailed failure analyses of via-line structures to correlate void geometry and resistance behavior during accelerated electromigration testing. We find that while the void volume is dependent only on the time under stress, the resistance change is affected very strongly by void geometry and location. The primary effect of void location and geometry is to determine the availability of alternate current paths and how soon during the test these come into effect. The microstructure and crystallography of the connecting Al lines appears to play a secondary role in determining which samples fail early. The processing of small geometry vias can lead to high resistance interfaces between the via and underlying metal which can lead to large variations in electromigration lifetime depending on void geometry and location.

We observed the dynamics of Al2Cu precipitates, voids, and hillocks under electromigration (EM) stress (200–250° C, 1–2 MA/cm2) in Al-2wt%Cu lines using in-situ and ex-situ side-view TEM. The sub-micron wide (0.5 to 0.8 μm) bamboo Al-Cu lines were on a TiN underlayer in a drift velocity measurement structure. Large precipitates near the cathode ends of the lines shrank under EM, leaving voids before all the Cu was depleted from the cathode ends. We attribute the void formation at the cathode end to the stram induced by the shrinking precipitates in neighboring Al grams. Precipitates downstream of a sufficient Al source (eroding precipitate or Al grain) eroded without adjacent void growth. Size and location of precipitates were thus important in determining whether voids formed or not. Size and location of precipitates at the anode end, however, did not necessarily affect hillock growth. Hillocks grew near the anode end of the lines even with so-called blocking grain precipitates upstream in some cases.

This paper will show that scanning force microscopy (also known as atomic force microscopy (AFM)) is a good tool to observe the microscopic changes in the morphology of metallic conductor lines during in situ current stressing with a very high spatial resolution. AFM was used to image AlSi1%Cu0.5% meander structures (1 μm thick, 4 μm wide, 44 mm long) at room temperature during current stressing. The microstructure of the material showed significant changes already at a current density of 1.5 MA/cm2, still before voiding and hillock growth occurred. Growth and reduction of the grain sizes and also grooving and widening of the grain boundaries were observed.

Aluminum films were deposited from a high purity aluminum source by the self-ion assisted technique onto oxidized silicon wafers with TiN sub-layers. The ions were accelerated toward the substrate by potentials of 0, 3 and 6 kV. The films were patterned into strips 670 μm long and 8 μm wide using photo-lithographic procedures and wet etching. Average drift velocities were measured in the films tested under electromigration conditions. Electromigration activation energy was obtained for the films. It was found that electromigration activation energy increased with the acceleration potential. The strength of the (111) fiber texture, however, decreased with the acceleration potential. Therefore, the weaker textures resulted in higher electromigration activation energies. These results can be explained in terms of grain boundary structure, which controls electromigration behavior. By using orientation imaging microscopy to characterize the structures, it was shown that the weaker textured specimens contained a high fraction of low angle and low diffusivity grain boundaries.

Electromigration(EM)-driven mass transport in “near-bamboo” Al-lines, which consist mostly of “blocking grains” is an important topic of research on ULSI-metallizations. Because the most easy diffusion path, i.e. grain boundaries parallel to the line, is suppressed in bamboo-like Al-lines other paths have to be considered. In this work two other possible paths of diffusion were examined by in-situ observations in a transmission electron microscope (TEM). For these experiments a special sample holder had to be constructed.

One path is EM-driven intragranular diffusion in Al-lines. In this experiment, inert gas-filled voids with a mean diameter of about lOnm, so-called bubbles, which were created after gas implantation and annealing of the Al-lines, serve as indicators of mass (or vacancy) transport. The in-situ EM-tests reveal no intragranular void motion over a period of more than 100h at current densities of l-1.75MA/cm2 and temperatures of 150–225°C. This leads to an estimation of the maximum void diffusion velocity which was compared with calculated values of surface and volume diffusion controlled void motion, respectively. The second point of interest was the behavior of dislocations in Al-lines under an applied EM-force. The importance of their observed motion for intragranular mass transport will be discussed.

In-situ TEM observations on void formation and growth have been performed in passivated Al-lines in order to collect detailed microstructural data such as void number density, void size distribution, and growth rates as a basis for modelling stress relaxation and electromigration (EM) effects.

The samples were 0.5mm long lines of 200nm thick Al with different widths (0.8, 1.6 and 2.4μm) passivated with 200μm thick SiNx. Additional experimental parameters were average grain size (0.5μm to more than 3μm, i.e. nearly perfect bamboo structure in the lines), relaxation time (0.1 to 20h) and temperature (20 – 300°C) and electromigration current density (without current up to ∼2MA/cm2).

Measurements without current reveal that stress induced voids

– develop after a temperature step of 50°C during cooling from 450°C and above. Upon heating from RT they start shrinking at 150°C and disappear completely at 350°C.

– decrease in number and total relative volume but have constant mean void size with increasing number n of temperature cycles (n < 4)

– show a strong dependence of their total relative volume on the microstructure (grain size, bamboo structure) which is strongly dependent on the anealing temperature for T ≥450° C.

– show growth rates during isothermal annealing which are very dependent on relaxation time and temperature.

Additional measurement with an applied electrical current, i.e. during an EM-test, show that nearly all the preexisting stress-induced voids

– change their shapes and positions. This is dependent on electrical current density and time. Their frequently observed motion along the line (against the electron wind) was not neccessarily stopped by so-called blocking grains, i.e. grains without boundaries parallel to the line.

– slightly change their total number but nearly maintain their total relative volume which is independent of current density or time.

The collected statistical microstructural data are in good agreement with macroscopic stress relaxation results by wafer curvature and X-ray diffraction and recent in-situ stress measurement during an EM-test.

We have studied the effect of Cu concentration and initial distribution on the electromigration lifetimes of 0.3 – 0.8 μm wide, ≥ 99% bamboo Al interconnects. The mechanism by which Cu reduces electromigration in polygranular interconnects may not be important in bamboo interconnects, so it is not possible to predict whether Cu will have a positive or negative effect in this case. Samples containing 0.0, 0.5, and 0.9 wt. % Cu were tested at 225° and 2×106 A/cm2. For the Cu-containing samples, two different initial Cu distributions were tested: one in which the Cu was quenched into supersaturated solution within the Al grains (no precipitates) and another with many small, closely-spaced Al2Cu precipitates. We find that Cu still improves the mean time to fail of bamboo lines, although not as dramatically as in polygranular lines. More importantly, it significantly decreases the standard deviation in failure time, thus increasing the time to the first failure. Unlike in polygranular lines, increasing the Cu concentration from 0.5 to 0.9 wt. % has little effect on lifetime, and deliberately forming Cu precipitates prior to electromigration stressing reduces lifetimes. The best performance for submicron, highly bamboo runners was obtained from Al(0.5 wt. % Cu) with no precipitates prior to electromigration stressing.

The effect of grain boundary on the corrosion of Al(Cu 1%) etched using SiCl4/Cl2/He/CHF3 gas plasma has been evaluated with XPS (X-ray photoelectron spectroscopy), SEM (scanning electron microscopy) and AES (Auger electron spectroscopy). It was found with SEM that the surface of Al(Cu 1 %) mainly corroded at the grain boundary. Using AES point analysis, the cause of selective corrosion at the grain boundary of Al(Cu 1 %) has been investigated. The results of AES indicated that the contents of F and Cl have made a difference at the analyzed positions. This seems to result from the imperfect crystalline structure of Al(Cu 1%) grain boundary. It was also confirmed that F has passivated the Cl at the grain boundary. The SEM and XPS results implied that Cl incorporated in the grain boundary of polycrystalline Al(Cu 1%) film accelerated the corrosion and could not be easily removed by the subsequent SF6 plasma treatment.

Failure of test structures composed of pairs of W vias connected by lengths of Al/0.5%Cu conductor were compared as a function of the position of the via with respect to the conductor link. In one configuration, the current was passed from first level to a conductor link on second level and in the other the current was passed from second level to one on the first. In each case the links were identical in cross section, length and composition. There was a significant difference in the performance with respect to electromigration lifetime between these configurations as well as a difference in the failure mode. In the first level links, where the current passed first through a a 350A thick TiN ARC (Anti-Reflective Coating) layer and then a 150A Ti layer, the lifetime was significantly shorter than that of the second level samples where the current was passed across an interface of Al3Ti intermetallic compound. In addition, the failure mode was different, the first metal links failing directly under the via whereas in second metal link, the failures were away from the via in the metal link itself.

Precipitation and solid solution hardening have been studied in Al-Cu thin films with Cu content ranging from 0 to 2 wt%. We have changed the precipitate and dislocation structures of films having different Cu concentrations by varying the heat treatments prior to mechanical testing. Pure Al films showed the same values of tensile and compressive yield stresses at a given temperature during stress-temperature cycling. Al-Cu alloy films, however, showed larger stresses in tension than in compression. The compressive flow stress during heating could be changed by a factor of five by the initial heat treatment, but the differences disappeared above 250°C. Upon cooling from 480°C, solution hardening as well as precipitation hardening was observed in the Al-Cu films. Solution hardening is independent of Cu concentration, but for precipitation hardening, both the magnitude and the temperature range in which the mechanism is effective are sensitive to the Cu concentration. The microstructure was observed using transmission electron microscopy. The mechanical behavior is consistent with interactions between dislocations and precipitates which arise due to constraints on the film by the substrate.

Thermal stresses in pure Al lines passivated with a baseline 1000Å oxide and additional passivations of 0.5μm oxide, 1μm polymer, or 0.5μm, 1μm, or 2μ nitride were analyzed. Results from finite element analysis and X-ray measurements were compared, and samples were examined in a high voltage SEM for stress voids. For unvoided samples, calculated and measured results showed good correlation, while results for the voided samples showed little correlation due to stress relaxation through voiding. Initial in-situ electromigration test results showed that electromigration voids can occur at stress void sites.

The hillock formation of Al-0.5wt%Cu films on TiN/Ti/SiO2/Si and T1N/SiO2/Si multilayer structures was investigated by the observation of FESEM. Two types of hillocks, large and small, were observed in both multilayer structures with or without Ti after annealing at 400°C for 30 minutes. A wedge-like plateau appeared next to the large hillock in the multilayer structure without Ti interlayer. The small hillock also exhibited wedge-like morphology in the same multilayer structure. Both wedge-like plateau and small wedge-like hillocks oriented toward the same direction, which disappeared in the multilayer structure with Ti interlayer. The difference in hillock morphology on these two types of multilayer has been correlated to the difference in the extent of Al(111) texture, which was characterized by the off-axis angle and orientation factors in X-ray ψ scan analysis. The wedge-like hillocks in multilayer without Ti were aligned along the tilted direction with larger off-axis angle in ψ scan. The Al(111) texture was largely improved due to the insertion of Ti into multilayer structure.

In the present study, several different types of amorphous passivation layers such as PECVD-SiN and PECVD-TEOS were tested to learn how effectively they protect underlying Al interconnection lines. According to the experimental results, a thick monolithic passivation layer composing of PECVD-SiN was found to be highly susceptible to stress-related migration because it did not have sufficient elasticity. Moreover, since silicon nitride also has a high dielectric-breakdown strength, it exhibits an increased impedance to electric current due to parasitic resistance that exists in the path between the two passivated metal lines. On the other hand, passivation thickening through the use of PECVD-TEOS as an initial layer was estimated to be a more effective way to improve device reliability because of its better step coverage and smaller dielectric constant. The FEM simulation explains why the thick multilayer compromising an alternating sequence of mechanically dissimilar layers is an effective way to suppress stress-induced passivation damage during thermal cycling without having a significant effect on the IC pattern.

Stress-induced voiding and shear plasticity of passivated metal lines during thermal cycling should be dependent on various properties of both, metallization and passivation, such as geometry and elastic constants and on the flow stress of the metal lines. Therefore we performed strain measurements on passivated lines of various aluminum-alloys with different passivation materials by x-ray diffraction during thermal cycling between RT and 420°C.

As samples we used an array of 1μm wide and 0.8μm thick parallel lines, 1μm apart. As metallization sputter deposited Al, AlCu(0.5wt%), AlCu(1.0wt%), and AlSi(1.0wt%)Cu(0.5wt%) on a Ti/TiN diffusion barrier were used and passivated with different combinations of silicon-oxide/silicon-nitride and different geometries. To account for the interaction metal-passivation FEM calculations have been performed and the formalism of “eigenstrains” has been used to evaluate the voiding and shear plasticity separately.

First results show a significant dependence of the strain on the passivating temperature (higher passivating temperature induces higher strain at RT) as well as on the passivation material and geometry. Plastic shear deformation is mostly influenced by the passivation geometry. Passivation composition shows smaller but also visible effects for plasticity whereas alloy composition seems to have no influence on any measured elastic and calculated plastic behavior of the metal lines within the errors.