In this study, we have evaluated electromigration performance of W-plug vias, fabricated with process variations in via hole etching as well as in via barrier deposition, on a TiN(barrier)/AlCu/TiN metallization system. We found that via etch profile scheme as well as deposition conformity for via barrier can significantly affect W-plug via lifetimes. The results indicate that the quality of contact with AlCu layer of bottom level metal is a critical factor in determining via lifetime. Precise control of via etching or by use of conformal via barrier to ensure a good barrier between W-plug and the AlCu of bottom level metal is essential for achieving reliable via structures.

This work investigated the electromigration and failure behaviour of layered tungsten vias. It has long been recognized that the flux divergence caused by the tungsten-aluminum interface is an area of high reliability risk. In an attempt to lower this risk, several process experiments were initiated. The thickness (500Å, 800Å, and 1200Å), composition (TiN and Ti), and etch and rinse rates of the underlaying layers of the tungsten plugs were varied. Test structures consisting of a chain of eight M1 to M2 vias were placed on electromigration testing. Their variability in lifetimes, resistance vs. time characteristics and failure behaviour is reported. The fail sites were characterized using TEM analysis and FIB cross-sections. Based on these results process improvements for improved reliability were developed.

Recently, device reliability has been an important issue since the design rule is scaled down and complicated processing is involved. To improve the reliability, many problems should be solved. One of the problems in interconnection failure is short or open circuits at interconnection lines. Contact or via hole failure has been ignored while the metal line failure has been strongly studied by many researchers.

In this study, electromigration (EM) phenomena at the submicron size of contact hole are presented with respect to the type of dopants, boron and arsenic, implanted by high dose. In both cases, the failure mode at the contact hole, the values of the activation energy and the mean-time-to-failure (MTTF) are much different. For example the value of MTTF in boron is lower than that in As. The contact failure by EM is analyzed by SEM micrographs, FIB micrographs, EM measurement. One of the interesting results is void formation at the interface between implanted surface and silicon dioxide film, which is located between contact holes. This fact indicates that the interface migration of Si is much faster than the self-diffusion of Si because the current path is mainly formed along the interface.

The influence of a TiW barrier layer on the stress-voiding behavior of AlCu and AlSiCu interconnects is investigated. The results are compared to the same alloys deposited on SiO2- In both cases, AlCu exhibits a notably better voiding behavior compared to AlSiCu. In the case in which the alloys are directly deposited on the TiW barrier without breaking vacuum between TiW and Al(Si)Cu depositions, a significant improvement of the voiding behavior of both alloys is observed. Compared to AlCu, AlSiCu shows worse voiding behavior due to the presence of Si precipitates, which introduce significant extra dislocations and defects in the Al grains. These dislocations and defects are diffusion paths which assist stress relaxation and void formation. In the presence of a TiW barrier part of the Si content of the AlSiCu is consumed during the anneal by the Al-TiW interface, which results in a decreased number of dislocations and defects within the Al grains, and an improved voiding behavior compared to AlSiCu on SiO2-Furthermore, W and Ti diffuse into the grain boundaries of both alloys. The presence of W and Ti in the grain boundaries reduces the amount of Cu depletion from within the grains, which makes both alloys more resistant to stress voiding. The above is supported by Auger and TEM results. The electromigration results of the alloys in question are presented. These show that also with respect to electromigration AlCu is the preferred alloy both on TiW and SiO2.

A selective range of hydrated silicon oxynitride thin films (SixOyNz:H) have been characterized in terms of their stress, hardness, and modulus in order to mechanically qualify them for use as an encapsulation layer for memory devices (e.g., Flash and EPROM memories). These films are analyzed by RBS and HFS for stoichiometry. The films exhibited stress values between 1.86 x 109 to -3.54 x 109 dyne/cm2 and showed a linear correlation with the hydride ratio (N-H/Si-H). An Ultra Micro-Indentation System (UMIS) measured hardness values between 10.5 GPa to 16.2 GPa while the elastic modulus varied between 119.1 to 141.2 GPa. The monatomic increase of modulus with hardness is attributed to increased amounts of nitrogen and nitrogen hydride bonding in the silicon oxynitride samples.

Continuous measurements have been made of the changes in several electrical quantities of Al 4%Cu tracks during the lifetime under electromigration stress. The changes in resistance, second harmonic and the second harmonic delay have been used to determine the changes in the track cross-section, resistivity and temperature. During most of the lifetime the results are consistent with a linear increase in the microscopic damage occurring uniformly throughout the track. During the final phase of the life there is evidence of large void formation with subsequent healing. Considerable information can be obtained about these processes and the extent to which they are reversible. The observed events are compared with the changes which have been predicted by specific mechanisms.

We have examined small changes in the resistance of a single aluminum via test structure with a resolution of 10-8 ohms. The via is roughly one cubic micron and contains a well defined TiN diffusion barrier. The high resolution of these resistance measurements allows the observation of atomic drift rates corresponding to roughly 100 atoms/second into and out of the well defined via structure. Resistance changes are observed with moderate current densities in the temperature range of 100°C to 200°C. We observe reversible increases and decreases in the resistance caused by the accumulation of atoms at the diffusion barrier within the via. Comparisons can be made to recent models for the transient electromigration induced vacancy drift. We are able to extract microscopic electromigration parameters as well as differentiate between copper and aluminum as the initial diffusing element.

Two series of experiments were carried out to investigate the line length dependence of changes in the electrical resistance caused by passing a DC current through thin film Al conductors. First, the effect of adding side branches with a spacing of 5 μ m to 150 μ m long lines was studied. Experiments performed at T = 148 °C and at a DC current density of 0.5 MA/cm2 showed three effects due to adding the side branches: i) the magnitude of the resistance changes is reduced, ii) the resistance changes saturate on a time scale of one hour, and iii) the resistance changes fully recover after switching off the current. In the second series of experiments the current, temperature and line length dependence of resistance changes induced in short (3-100 μ m) lines between bonding pads were studied. For current densities smaller than a certain length dependent critical value the resistance saturates after applying the current and relaxes back to the original value after removing the current. The time to relax back, which is comparable to the time to reach saturation, is proportional to the sample length squared. This and the activation energy derived from the temperature dependence of the relaxation process suggest strongly that the observed time dependence of the resistance changes is determined by grain boundary diffusion (electromigration) along the whole length of the line.

This paper is focused on the non-linear resistance behaviours often detected by means of high resolution resistometric methods both at the very beginning of an electromigration (EM) test, and after the high stressing current inducing EM is switched off. It is shown that temperature steps, always present at the beginning or after EM, are the triggering events for different, often reversible, physical phenomena contributing to resistance changes. Precipitation-dissolution of alloyed elements is perhaps the most important one, while other mechanisms, like the combined effect of hydrostatic stress relaxation and void volume change, should cancel out. These effects, however, don't exclude a possible simultaneous effect on the resistance of accumulation/relaxation of EM damage. Experimental results are collected by means of different, complementary techniques.