Scanning electron microscopy (SEM) is a frequently used tool for establishing reliability where potential causes of failure are related to structural aspects that show up on a submicron scale. Conventional SEMs, however, even those equipped with field emission sources, can provide high-quality photomicrographs only up to a magnification of about 100,000×. For many purposes greater resolution (several nanometers or below) is required, in which case the usual alternative is to turn to transmission electron microscopy (TEM), in spite of the laborious sample preparation required and limited plan-view images obtained. We describe here an underutilized alternative to TEM for many applications. The in-lens field emission SEM (ILFESEM) can provide pictures above 500,000× magnification with sub-nanometer resolution, with the simple sample preparation and surface viewing advantages of the SEM. Magnification in this range is necessary to verify modern day tolerances on surface morphology and roughness, oxide and thin film structure, thickness, and step coverage, and pore sizes in adhesion layers or diffusion membranes.

The use of Tungsten plug technology to obtain good step coverage in vias and contacts is a well know fact. The effect of TiW as an adhesion layer between the plug and the overlying Aluminum is shown in this paper to have a significant impact on the electromigration characteristics of this scheme. A comparison between plug and non-plug technology is also made keeping the aspect ratios the same. Also discussed in this paper will be the effect of the first metal (Tungsten vs Aluminum) on electromigration characteristics. Constant current DC electromigration stressing was performed on via chains and the results are discussed. Discussed, also, are SEM cross sections (XSEM) of the stressed vias indicating the failure modes. An idealized metal scheme will be presented and discussed.

Stress-induced voiding in microelectronics chips has been reported to exhibit a variety of dependencies on temperature and on passivation stress. The dependence of line failure is reported here at four different temperatures (150, 225, 285, and 315 °C) for AI-0.5%Cu-2%Si, AI-0.5%Cu and AI-1%Si lines with passivation thicknesses (silicon oxide or silicon nitride) ranging from 0.1 to 2.5 times the metal thickness. Failure distributions change in a complex manner with changing passivation thickness, and with temperature for a specific passivation thickness, raising questions on the validity of using the conventional median time to failure (t50) and lognormal slope (σ) to project field failure rates from data generated by accelerated life testing.

Strain relaxation in passivated Al-0.5% Cu lines was measured using X-ray diffraction coupled with in-situ observation of the formation and growth of stress induced voids. Samples of 1 μm thick Al-0.5% Cu lines passivated with Si3N4 were heated to 380°C, then cooled and held at 150°C. During the test, principal strains along the length, width, and height of the line were determined using a grazing incidence x-ray geometry. From these measurements the hydrostatic strain in the metal was calculated and strain relaxation was observed. The thermal cycle was duplicated in a high voltage scanning transmission electron microscope equipped with a backscattered electron detector. The 1.25 μm wide lines were seen to have initial stress voids. Upon heating these voids reduced in size until no longer observable. Once the samples were cooled to 150°C, voids reappeared and grew. The measured strain relaxation is discussed in terms of void and θ-phase (Al2Cu) formation.

Stress and current induced degradation of the interconnects may well define the ultimate limits on device density and total on-chip power in microelectronic circuits. The two damage mechanisms are found to be mutually interdependent in a fashion determined by the line microstructure, as well as by design features such as W-studs and refractory metal back up layers. We have developed a comprehensive model for the synergistic effects of stress migration and electromigration, taking such factors into account. The present work reviews the modelling of the statistical failure distributions, with emphasis on the temperature and current density dependencies, for the case of ‘near-bamboo’ submicron lines.

Accelerated electromigation tests on unpassivated, pure aluminum interconnects were performed. The failure mechanisms were observed by interrupting the tests and exanming the conductor lines using an SEM. Because the metal thin film was subjected to a so-called laser reflow process before patterning, grain boundaries were visible in the SEM as thermal grooves. Voids were observed to move along the line and to grow in a transgranular manner, and a characteristic asymmetric void shape was identified which seems to be related to the failure mechanism. It is argued that substantial progress in modelling and understanding of electromigration failure can be made by consideration of such void shape effects.

Earlier experimental findings concluded that electromigration voids in these meandering stripe test structures were not randomly distributed and that void nucleation frequently occurred sub-surface at the metal/thermal oxide interface. The data showed a strong correlation between void area, void growth rate and stripe segment length [1]. The influence of mechanical stress on electromigration damage in these test structures has been examined by applying tensile stresses to both passivated and unpassivated samples. The stress distributions are calculated using finite element analysis for each of the test conditions. The resulting impact on electrornigration voiding, as well as mechanical stress voiding, and lateral hillock formation is discussed.

An iterative method describing the stress state at the vicinity of singularity point is presented here. This approach is based on concepts of rupture mechanics, uses classical solutions from finite element calculations and concerns plane problems. Through this approach, the stress state at the vicinity of several types of singular points in a patterned structure were analysed. The obtained results can be discussed according to failure mecanisms.

The determination by X-ray diffraction of the elastic strain tensors and the corresponding stress tensors in uniform films and patterned lines of tungsten was used to investigate the effect of line width. The stresses were found to increase with increasing line width. These experimental results are discussed with respect to the values obtained from a model using a distributed force in the line. The results of the calculations are in agreement with the X-ray measurements. The edge effects appear to be significant for tungsten lines.

Recent work in thermal stress induced voiding in passivated lines has focussed on the isothermal, quasi-steady state growth of existing voids. In this work, the transient growth and shrinkage of voids during the thermal cycling of narrow, passivated lines was studied. Analysis of experimental data for passivated lines demonstrates the quick buildup and evolution of a backstress at the grain boundary during heating and cooling. This backstress greatly reduces the driving force for void growth at higher temperatures and must be accounted for in stress induced void growth models for the case of appreciably varying temperatures in short time scales.

Stress induced voiding in passivated Cu lines was investigated by x-ray strain analysis and in-situ high voltage scanning electron microscope (HVSEM) techniques. Cu lines on a Ta underlayer and Cu lines on an Al underlayer were patterned by a trilayer liftoff technique and passivated with Si3N4. For direct observation of stress voiding, specimens were heated to 350°C in the HVSEM and then cooled and held at 150°C. Identical samples were subjected to the same thermal cycle for strain state determination using x-ray techniques. The hydrostatic stress state at each temperature was calculated from the measured strains. Few initial voids were observed after passivation in either sample. After heating to 350°C and cooling to the dwell temperature, no new voiding was seen in the Ta/Cu lines. Measured hydrostatic strains were half those measured in the Al/Cu lines. Heavy voiding was observed in the Al/Cu lines after cooling to the dwell temperature.

A systematic investigation of the anelastic relaxation of thin Al films on Si substrates has been carried out. It was found that both the relaxation in bulk and thin film material can be explained by a model involving glide of grain boundaries (GBs). The mass transport necessary for the glide occurs via GB diffusion in the thin films and via lattice diffusion in the bulk material the different behavior being due to the more of two orders of magnitude smaller grains in the films. Internal friction thus provides a technique to measure diffusional parameters of GB diffusion in thin films.

Finely dispersed, stable Al-oxide particles were produced in Al films on Si substrates by oxygen ion implantation. A laser reflow technique was employed to vary the grain structure of some of the films. Transmission electron microscopy (TEM) was used to characterize the oxide particles and the grain size in the films, and a wafer curvature technique was employed to study the influence of microstructure on the deformation properties as a function of temperature.

For coarse grained laser reflowed films, ion implantation increased the strength considerably, both in compression and in tension. Weak beam TEM techniques showed that the strengthening is most likely caused by attractive interactions between dislocations and particles. As-deposited and ion implanted films showed a stable grain size of only 0.35 μm after annealing, which caused significant softening to occur in compression, especially at high temperature. However these films showed very high stresses in tension at temperatures below 130°C. In these films the presence of the oxide particles stabilizes the small grain size and this causes a weakening effect which can be attributed to diffusion controlled grain boundary relaxation mechanisms. The high tensile stresses at temperatures below 130°C can be explained by direct and indirect particle strengthening.

Synchrotron radiation has been used in the grazing incidence geometry to determine the stress gradients and stress relaxation in 0.6 μm thick Al uniform films. We have examined both pure Al and Al implanted with 3 at% oxygen. The films were cycled between 23°C and 400°C. The surface of the pure Al films was more highly stressed, on average, than the film as a whole, both in tension and in compression. In contrast, the surface of the oxygen-implanted film was relaxed relative to the average film stress in compression but was more stressed in tension. The larger gradient in stress in the implanted films reduces during subsequent thermal cycling. These differences are attributed to the microstructures of the films. The 0-implanted film develops a small, non-columnar grain structure and forms small AI2O3 particles during annealing. The small grain size allows diffusional relaxation to occur in compression, relieving the stress in areas close to the film surface.

Wafer curvature measurements of a trilayer (SiO2 / AlSiCu / Si) structure are compared to that predicted by a weighted sum of individual measurements of SiO2 and AISiCu films on Si, and significant differences are found to exist for temperatures above 200°C. A straightforward analysis of the stresses in each layer has been modeled using an extension of a model by Feng et al. which assumes uniform plastic deformation throughout the Al. The modeling results suggest a straightforeward method for determining stresses in deformable thin films that are confined by elastic overlayers. A comparison of the stress-temperature behavior for unpassivated and passivated AISiCu films reveals that the confined films exhibit less plastic deformation and both higher tension and compression during thermal cycling.

Wafer curvature and grazing incidence x-ray scattering (GIXS) techniques were used to investigate the biaxial stresses induced in blanket Cu films during a thermal cycle to 460°C and back to room temperature. Cu was deposited by DC sputtering at ambient temperature. Several different barrier layer materials — SiO2, W, Ta, TiN, and Si3N4 — were used to compare any effect barrier choice might have on Cu microstructure evolution and mechanical behavior. Ta and Si3N4 encouraged a strong (111) Cu texture. A W barrier led to an untextured microstructure which underwent large, uneven grain growth during thermal cycling. Several samples were capped with a Ta layer which affected the stress behavior during cooling by inhibiting dislocation motion. An inverse relationship between strength and thickness was also documented.

The stress distribution about a series of deliberately fabricated defects in a Cu/Ni/Au/Cr redundant metal stack deposited on a Cr/Cu/Cr serpentine has been studied via x-ray microbeam diffraction using the sin2 ψ, technique. Strong directional anisotropies in the stress were found for both the second level nickel and copper. The extent of nickel stress relaxation from the edge of a defect far exceeded that predicted by a simple metal on substrate model. The size of the defect did not appear to significantly influence the stress distribution in the second level metal; however, the data suggest that the linewidth may influence the magnitude of the stress along the length of a line when near a defect or other sharp discontinuity.

Non-uniform hot-carrier degradation in n-channel polycide-gate MOSFET's with different thicknesses of the poly-Si film, and in p-channel polycide-gate MOSFET's with TiSi2- or CoSi2-gate-silicide, is studied. The n-MOSFET's with the thinnest poly-Si film, show an increased interface trap generation, while the influence of the gate-silicide material on the degradation behaviour of the p-MOSFET's is found to be very small. The results are evaluated in terms of the effect of mechanical stress on the degradation characteristics: favourable for compressive mechanical stress and unfavourable for tensile stress. A correlation with stress measurements by micro-Raman spectroscopy is made.

We have measured the principal strain state of AI-0.5%Cu lines passivated with silicon nitride directly and used it to calculate the stress state. The stress was determined as the lines were thermally cycled from room temperature to 450°C. The general stress-temperature behavior shows good fundamental agreement with that calculated using finite-element methods, although the magnitude of the stresses measured with x-rays is less than that predicted by modeling due to stressinduced voiding in the lines. This is shown with a high voltage scanning transmission electron microscope (STEM) operated in the backscattering mode.

An empirical model for electromigration-induced early resistance changes is given. The basis olthe model are two coupled partial differential equations, one for vacancies, and one for mobile atoms. These equations are then solved numerically. For model verification purposes model calculations are compared with experimentally obtained resistance change curves.