Monolithic Nb5Si3 films and microlaminates consisting of alternating, equally thick layers of Nb and Nb5Si3 were synthesized by magnetron sputtering. Thick monolithic Nb5Si3 films (25,000 nm) were deposited on a sapphire substrate to set process parameters and evaluate the microstructure and mechanical properties of as-deposited crystalline films. Nb5Si3/Nb micro-laminates with modulation wavelengths (i.e., bilayer thickness) of 40 and 200 nm were deposited on Nb substrates. Mechanical properties (elastic modulus, microhardness, compressive yield strength) of the films and microlaminates were studied using the nanoindentation method and Vickers microhardness. Mechanical property test results are presented.

Thin film stress has been measured in (Ni66Fe16Co18 25 Å/Ag 50 Å) multilayer thin films that exhibit annealing-induced giant magnetoresistance (GMR) known as ‘discontinuous’ GMR1. This GMR results when NiFeCo layers are broken up into discontinuous ferromagnetic islands or platelets by immiscible Ag atom grain boundary diffusion normal to the film plane. As-deposited films display no GMR and are in compression. An MR value of 5 % with a relatively sharp profile can be induced in a ten bilayer sample annealed at 400°C for 15 min. Annealing generally results in a decrease in the as-deposited compressive stress and can result in tensile stress at high enough temperatures. Interestingly, at the annealing temperatures corresponding to the maximum induced GMR, there is a plateau in the otherwise monotonic decrease in compressive stress with increasing temperature.

Stress-temperature plots reveal a plateau in the stress value during the heating of a ten bilayer sample due to plastic deformation in the film via densification. Isothermal stress-time plots done on a single sample show increasing compressive stress relief after each successively higher temperature anneal. The stress was also measured in a single bilayer film with the same volume of film as in a ten bilayer sample but far fewer interfaces. High and low angle x-ray diffraction (HXRD, LXRD) was used to reveal the structural evolution with annealing. Magnetic (M-H) hysteresis measurements revealed increasing coercivity and decreasing induced as-deposited in-plane magnetic uniaxial anisotropy with annealing temperature.

A new method for calculation of superlattice (SL) piezoelectric properties is proposed. It is used for derivation of the piezoelectric constants of ultrathin [001] oriented (GaAs)m(AlAs)n SL. These SL are formally considered to result from complete ordering Ga (Al) atoms over a-positions of a hypothetical initial structure of a bulk mixed crystal of (GaxAl1-x)As with x=m/(m+n). This viewpoint allows us to develop a method which is close to the Landau phenomenological theory of second- order phase transitions and which is intended for calculation of the SL crystal-lophysical constants. It is shown that a nontrivial term has much significance in changing one of the piezoelectric constants, γz,xy, when the initial structure is transformed into a SL. This term is composed of coefficients at three translational invariants (Dzyaloshinskii invariants) in the expansion of the thermodynamic potential of the “ordering” system in a power series of the order parameter components. The values of these coefficients must be obtained from microscopic calculations which take account of spatial correlations in Ga(Al) atomic distribution. Specific calculations are carried out for a SL with m+n=3.

Unpassivated 2.1 μm wide Al-4wt%Cu interconnects with near-bamboo grain structure, are electromigration-tested to failure in-situ in transmission electron microscopy. Early stress-induced voids stop growing and are not fatal. Hillocking is associated with precipitates, fatal voiding with copper depletion. Electromigration-induced voids form at the upstream end of inclined grain boundaries. Healing events are analysed and it is shown that open-circuit failure can occur when the proximity of grain boundaries impairs the stress-driven healing.

A wafer curvature technique was used to measure the mechanical stresses during thermal cycling and the isothermal stress relaxation in passivated Ti/TiN/AlSi(l%)Cu(0.5%)-lines (aspect ratio: 0.92) in the in-plane directions, parallel and perpendicular to the lines. The evaluation of the measured curvature data is explained in detail. The evaluation procedure was tested with the help of passivated SiO2-lines in which the stresses can be calculated because both, the SiO2-lines and the passivation, behave elastically. Comparision to elastic Finite-Element-Method calculations show excellent agreement. The main results are:

1. a) During thermal cycling all stresses in the AlSiCu-lines vary linearly with temperature without significant hysteresis. The stress parallel to the lines is higher than perpendicular to the lines. X-ray stress data from the same sample confirm the wafer curvature data and show, additionally, that the stresses in the two directions perpendicular to the lines are equal.

2. b) The isothermal stress relaxation depends strongly on the temperature with a maximum at 250°C. Void formation and growth probably control the observed relaxation.

The orientational dependence of stress and strain was measured for (111) and (110) oriented grains in an Al thin film which hillocked upon heating, and (111) and (100) oriented grains in Cu thin films, one of which exhibits abnormal (100) grain growth. Results from the Al thin film were inconclusive in the evaluation of the Sanchez model for hillocking, however they indicate that the Nix model for the orientational dependence of yield stress is not applicable to individual grains at room temperature. Results from the Cu thin films suggest that suppression of (100) abnormal grain growth is due to isostress averaging among the grains of varying orientation.

The microstructure and mechanical properties of hot (h) and cold (c) sputtered Al-lwt%Si and Al-lwt%Si-0.5wt%Cu films were studied using transmission electron microscopy and wafer curvature stress measurements.

Stress/temperature curves of all films showed only slight differences in compression on healing once a stable grain size was established. However, on cooling several remarkable differences were observed. These observations cannot be explained by assuming dislocation glide/climb as the dominant relaxation mechanism. The results will be discussed in terms of grain boundary diffusional relaxation (Coble creep), which occurs in addition to dislocation glide.

Substrate curvature measurements were used to study stress relaxation in Al-Si-Cu films at temperatures between 45 and 165 °C. Dislocation glide with an average activation energy, resp. athermal flow stress of 1.7 ± 0.2 eV, resp. 600 ± 200 MPa could describe the relaxation data for temperatures up to 120 °C well. Stress relaxation at 92 °C was found to progress much slower in 1 μm wide nitride passivated lines than in thin films or unpassivated lines.

Blanket films (1 μm thick) of both A1-0.5Cu and A1-0.15Pd were deposited at room temperature, 150°C, and 300°C. Stress in the as-deposited wafers increased with substrate temperature, as expected from the thermal expansion mismatch on cooling. All conditions were tiicrmally cycled to 450°C three times while continuously monitoring stress. The shapes of the curves were different for the two alloys because precipitates dissolve and reprecipitate in AlCu, but are present over the entire temperature range in AlPd. Lesser differences were evident comparing the stress-temperature behavior for the various substrate temperatures within a single alloy. The precipitate structure also influences the grain growth during thermal cycling, where substantially larger median grain sizes are found in AlCu compared to AlPd.

The substrate curvature technique was employed to study the mechanical properties of 0.6 μm and 1.0 μm Cu films capped with a 50 nm thick Si3N4 layer and to compare them with the mechanical properties of uncapped Cu films. The microstructures of these films were also investigated. Grain growth, diffusional creep and dislocation processes are impeded by the cap layer. This is evident in the form of high stresses at high temperatures on heating and at low temperatures on cooling. At intermediate temperatures on heating and cooling, stress plateaus a relatively low stresses exist. This can be explained by the so-called Bauschinger effect. A film thickness dependence of the stresses in the film could not be observed for capped Cu films.

Models describing the effect of grain orientation on dislocation glide in a thin film are reviewed; it is predicted that differences in stress-temperature behavior should exist between grains of different orientation in a polycrystalline copper film. Direct x-ray evaluation of strains in {111} and {100| grains within primarily {111} textured copper films shows that these stresses are film thickness or grain size dependent. The {111} oriented grains behave as expected, with room temperature flow stress increasing linearly with 1/film thickness. The flow stress of grains of {100} orientation, however, is approximately constant with film thickness and is not well represented by the models. It is proposed that simplifying assumptions about orientation dependent yield may only be appropriate for the majority grain orientation population in a textured film.

The stress in an encapsulated metal line has been calculated using the finite element technique for two passivation stress conditions. It is found that the stress-state in the passivation has a negligible effect on the stress in the metal line. The dominant factors affecting metal line stress are the passivation temperature, the thermal expansion coefficient difference between the metal line and the substrate, and the passivation modulus.

Interconnect metallizations used in advanced integrated circuits typically use an Al-alloy sputterdeposited onto a Ti barrier layer. The Ti and Al react above ∼ 400°C to form TiAl3, which affects the stress evolution of the metal stack during thermal cycling. This paper describes results of thin film experiments performed on Ti/Al-alloy bilayer films. Two Al alloys were studied: Al-I%Cu and Al-0.5%Cu-1%Si. The rate of TiAl3 formation at 430°C was determined for both alloys and used to relate TiAl3 formation to the stress evolution of the film stacks during thermal cycling. The dominant effect of the TiAl3 intermetallic formation on stress arises from a change in the stress-temperature behavior of the film stack, due to a change in the yield behavior, effective modulus, and thermal expansion coefficient of the stack. The presence of Si in the Al-alloy markedly reduces both the rate of TiAl3 formation and the resulting change in composite stress.

The paper analyzes electromigration (EM) conditions and material properties that determine the maximum EM induced stress, σa, and stress gradient, ∇σ, which counteract EM flow in interconnects.

The first systematic data on the drift velocity vs. stripe length, L, current density, j, and temperature are presented for Al lines. In contrast to the conventional approach to the Blech problem with σa taken to be a material constant (“yield strength”), the observations suggest that σa increases with j. The stress adjustment is shown to result from the imperative coupling of the net flux of material directed to the downwind end of the stripe with the flux of plastic flow (creep) responsible for stress relaxation. The effect of parameters of the constitutive equation assumed to describe the plastic flow kinetics, namely that of strain rate exponent, threshold stress, and creep, effective viscosity, on the stress cya is considered. To account for the creep viscosity, η, obtained unpassivated aluminum stripes from EM experiments, a model for the attachment-controlled Coble creep is suggested.

It is known that an electric current can interact with dislocations in metals. Recent studies suggest that these interactions include current-induced dislocation motion and multiplication (electroplasticity) through both glide and climb mechanisms. Such interactions might be expected to effect both the mechanical properties and the electromigration behavior of the thin metal conductor lines used in integrated circuits. The mechanical properties would be directly modified through changes in dislocation density and mobility. Changes in microstructure and strength could affect electromigration lifetime and failure morphology. In order to investigate the effects of current-dislocation interactions, we conducted three sets of experiments in unpassivated aluminum conductor lines: First, a nanoindentation device was used to measure the mechanical properties of the lines during application of current densities up to 6 MA/cm2. Second, the mechanical properties of the lines were measured ex situ as a function of time in accelerated electromigration testing conditions. Third, the nanoindentation device was used to damage the lines at specific sites and damage morphology was examined during subsequent electromigration experiments by means of in situ scanning electron microscopy. To date we observe no direct effect of electric current on the mechanical properties of conductor lines and no clear correlation between mechanical damage and electromigration failure characteristics. Based on theoretical calculations and previous experiments on bulk samples, these results were not expected.

The growth of hillocks and voids in metal films was studied. The applicability of a model involving fractals and kinetic equations was examined on the basis of whether there is independent justification for using scaling arguments in the model and whether there is reason to connect the evolution of hillocks with that of voids. Hillocks and voids were found to be self-similar across about three orders of magnitude of variation in spatial scale with the same fractal dimension. Voids and hillocks were found to have the same fractal dimension whether studied using atomic force microscopy (AFM) or scanning electron microscopy (SEM). The parameters obtained from these fractal analyses demonstrate quantitative internal consistency with an earlier time dependent study of thermal annealing effects on hillock distributions. Remarkably, area-perimeter data obtained from either a long-time study of a single void or a spatial average of a large number of different voids both yield quantitatively identical results.

In multilevel metallization processes, stress induced void and hillock formations cause serious problems with respect to the reliability of metal line, such as pitting corrosion of metal during the pad etching step, mask misalignment due to surface roughness, and open or short failure of metal line. It is generally known that these types of metal-film related problems are due to stress relaxation process in the metal line stress-induced by thermal expansion mismatch between metal and over/underlayer oxide film or compressive stress of overlayered passivation dielectric film.

In this study, the stress behavior of phosphosilicate glass (PSG) film on the Al-l%Si film, deposited by different CVD methods, is investigated. Here we found that the degree of PSG film stress is related with the suppression of hillock growth, void formation, and cracking of passivation layer, depending on the heat treatment and heat treatment. Using stress measurement, surface profilometer, and SEM, the relation between passivated PSG film stress on metal line and the metal line failures is analyzed. In addition, an effective method of hillock suppression is suggested, which is the use of PECVD-PSG film on the metal line before heat treatment of 400°C or that of any type of PSG film before 350°C.

Pure Al films deposited by the partially ionized beam technique at bias Iavels of 0, 3 and 6 kV onto TiN sublayers were patterned for measurement for direct drift velocities. Then films were annealed in a vacuum 10−4 Pa for 1 hour. Electromigration tests were performed in an air atmosphere at a temperature of 280°C. The current density during tests was 6·105 A/cm2. Examinations in the SEM showed that the morphology of the damage on cathode and anode ends of the Al stripes depends on the bias during deposition. In the case of 0 kV bias, voids nucleate on the back and side edges of the cathode end and propagate toward the anode end. Hillock and whisker formation was observed on the anode end. The height and diameter of the whiskers were about 50 and 1–2 μm, respectively. In the case of 3 and 6 kV bias the drifting edge was smooth. Only hillocks (no whiskers) grew on the anode edge. The morphological differences of these films are discussed on the basis of self-ion bombardment effects.

The electromigration failure mode and failure rate during accelerated electromigration testing is expected to be strongly affected by the mechanical stress state of Al lines, since tensile stress and compressive stress states favor void growth and hillock formations (extrusions), respectively. During electromigration testing, the mechanical stress state or evolution of mechanical stress of an interconnect is a function of current density and temperature, the two principal variables in electromigration testing. In our experiments, we have observed two different electromigration failure modes by varying the current density and temperatures where (i) the passivated Al lines tested at high current density and high temperatures failed by extrusion or hillock type failure and (ii) the interconnect lines tested at low current density and moderate temperature failed by voiding. A mechanical stress model which incorporates both the thermally generated stress and electromigration induced mechanical stress is invoked to explain the electromigration failure mode selection observed in our experiments.

The edge delamination test (EDT) measures the energy required to cause a thin film, under biaxial tensile stress, to debond from a rigid substrate. Circular holes are etched through the films and if the stress is large enough, stable debond rings grow radially around the holes. A finite element analysis is used to find the applied strain-energy release rate as a function of debond crack length, hole radius and side-wall angle. However, the analysis may not be widely available, so, we present tables of reduced debond energy val ues, Gr, to facilitate the use of the EDT. With observed EDT results, Gr is extrapolated from the tables and then the critical debond energy is calculated by multiplying Gr by the maximum strain energy. Also, we have fit Gr with a semi-empirical equation, which is in good agreement over a broad range of geometries.