The electromigration damage in polycrystalline AlSiCu lines caused by dc current densities of the order of magnitude 1010 Am −2 at 500 K has been investigated by long-time high-resolution ac noise measurements over a wide range of temperatures. A multiple-probe setup allowed us to monitor the electrical resistance increase along the line and to show that sudden enhancements of the noise power and the resistance occurred in the same line segment. The peak in the distribution function N(E) of the activation energies E for the defect motion responsible for 1/f noise was found to shift from E0 = 0.7 eV with Gaussian width ΔE = 0.1 eV in the undamaged samples to E0 = 1.0 eV and ΔE = 0.2 eV in the damaged samples. The latter values are compatible with the Cu diffusivity along intermediate-misfit grain boundaries in Al, indicating that the damage process depends not only on the composition but also on the microstructure of the lines. It is proposed that electromigration causes Cu atoms to drift towards and get trapped in grain boundaries, that their thermally activated motion in the boundaries contributes substantially to the 1/f noise, and that eventually the Cu atoms assemble on the anode side of the sample. The dependence of the diffusivity on the grain-boundary mismatch may cause flux divergencies at boundary junctions and thus lead to void formation.

We report experimental results of measurements of the changes in the resistance and thermal non-linearity of a variety of samples during accelerated electromigration stress to failure. The value of the extra measurement of the second harmonic of the AC bridge signal is illustrated by observations of changes in the thermal conductance to the substrate and the temperature coefficient of resistivity which are not detected by the resistance measurement. These quantities are also larger than the resistance measurement.

In this work we use a wafer-level, High Resolution Resistance Measuring Technique (HRRMT) to detect fabrication faults of Al-Cu interconnections. Experiments have been performed on two distinct sets of metal lines. The first set includes two lots of 4 μm wide lines which, once tested at moderately accelerated stress conditions, gave largely different life times. A microstructural analysis confirmed a major defectivity of the lot with shorter life time. An accurate examination of the early resistance variations revealed the presence of two distinct and subsequent phases, namely an initial pseudo-parabolic resistance increase followed by a linear resistance drop. Significant differences between the resistance behaviour of the two lots were detected during the first stage, lasting a few hours. Measurable differences could even be detected in the first few minutes. A second group of experiments was launched in order to assess the capability of HRRMT as in-line monitors. Samples from four wafers, one reference wafer and three wafers with intentional process variations, have been tested using our HRRMT at constant temperature and current, simulating an in-line production test. The standard life time of the four wafers have also been collected. Preliminary measurements highlight that a change of life time due to process variation corresponds to changes of the resistance behaviour in the first hours of test. These results pave the way for a new application of high resolution methods to assess the quality of a metallization system in a reasonable amount of time.

A set of 10, 5 and 2 µm wide, 500 µm long and 0.7 µm thick Al/Si/Cu-lines (1% Si, 0.5% Cu) was investigated under different stress conditions. Typical stress conditions were current densities of 2 to 6 MA/cm2 and ambient temperatures of 125 to 225°C. For additional mathematical simulation it was important to observe the complete history of the resistance development from the beginning until the break down of the samples. Depending on the line width a number of different resistance developments and times to failure occured. Especially at smaller line widths, i. e. near to bamboo structures, the differences of the resistance development and times to failure are wide-ranging. This fact must be taken into acount for the simulation. Furthermore unpassivated samples were used to enable investigations of the structures by SEM and SPM. Measurements of the size of hillocks and voids has been carried out by AFM. Thereby twin-crystals were observed. This fact indicates that there are special strains, which could be estimated.

Another aspect of investigations was the exact measurement of the temperature in the break down area. For that purpose a special test structure with diodes underneath the stressed line was developed. These diodes are used to deduce local temperatures.

Electromigration-induced drift in sub-micron wide Al segments has been measured using energy dispersive x-ray analysis in a scanning electron microscope. This technique, which works on both passivated and unpassivated samples, is sensitive to uniform Al displacements as small as 100 nm. The samples were 0.3 and 0.5 µm wide runners consisting of a continuous Ti/TiN line on top of which were patterned 0.4µm thick Al (0.5wt% Cu) segments of lengths varying between 5 and 100 µm. The microstructure, as determined by focused ion beam imaging, was predominantly bamboo. We have measured the Al depletion at the cathode ends of these segments after stressing for various times at 2 MA/cm2 at 200'C. For segments shorter than roughly 15 µm, no depletion occurred during the entire 100 hours of the experiment, yielding a value of the current-length threshold product of roughly 3000 A/cm, in agreement with previously published data. Longer segments (20-25 µm) depleted at a roughly constant rate for the duration of the experiment, yielding a drift velocity of around 3 nm/hr in the 0.5 µm wide lines, also in good agreement with published data for near-bamboo lines. However, the 0.3 µm wide lines drifted faster at roughly 6.5 nm/hr. The longest segments (50-100 µm), which started drifting at the same rate as the shorter segments, accelerated after 50 hours of testing to drift velocities in excess of 40 nm/hr. This may be associated with the depletion of Cu from the drifting end.

We report the application of a unique photoemission spectromicroscope (MAXIMUM) to the study of electromigration phenomena in Al-Cu interconnects. MAXIMUM is a scanning type photoemission microscope that uses multilayer-coated optics to focus 130 eV x-rays to a sub-0.1 µm spot. An electron energy analyzer collects photoelectrons in a chosen spectral region of interest to form an image of the sample that is sensitive to chemical states of elements on the sample surface. Al-Cu lines were characterized by spectromicroscopy techniques before and after electromigration stressing in the UHV environment of the microscope chamber. We present spectro-micrographs showing the chemical and structural changes on Al-Cu line surfaces as a result of the electromigration process.

The technique of spatially resolved photoluminescence (PL) spectroscopy was used to determine the local channel temperatures on GaAs/GaInAs/GaAlAs pseudomorphic high electron mobility transistors. By focusing a laser beam onto the different regions of the DC-biased transistor, it is shown that the channel temperature can be determined from the energy shift of one of the peaks in the PL spectra, with a spatial resolution of about 1 µm and a temperature resolution in the order of 1 °C. In particular, an asymmetry in the temperature distribution between the drain and source sides is observed. Using this approach, detailed temperature maps of the devices were obtained, as a function of the gate-source voltage VGS. These experimental temperature values are also compared with predictions derived from an analytical model of the thermal resistance in these devices.

Earlier work revealed that the transverse thermal conductivity of thin films of amorphous SiO2 deposited on monocrystalline silicon decreased substantially when the film thickness was less than about 1 µm. When multiple interfaces were created by the intercalation of thin intermediate layers of polycrystalline silicon into the SiO2 film, the thickness effect was enhanced. This observation pointed to an interfacial effect. A model accounting for the interface effect is discussed.

In this paper a phenomenological characterisation of an anomalous grain growth, observed after classical electromigration lifetests, on Al-l%Si-0.5%Cu multigrain stripes, deposited at high temperature (460°C) on a Ti substrate, is reported. Failure analysis, carried out by Scanning Electron Microscopy (SEM) and Focus Ion Beam (FIB), has detected an abnormal single-grain growth in the vertical direction, starting from the Ti/Al interface. Medium Time to Failure (MTF) data are compared with those of stripes on Ti/TiN substrate, on which no grain growth was observed. The growth of the anomalous grains has been related to the Ti/Al interface properties. Transmission Electron Microscopy (TEM), X-Ray Diffraction (XRD) and Atomic Force Microscopy (AFM) analyses concurrently show a higher texture of the film on the Ti, compared with the film on the Ti/TiN substrate. The tight columnar orientation of grain boundaries strongly limits the mechanism of grain boundary diffusion during electromigration stress. The atomic flux is then forced to take place at the Ti/Al interface, where the epitaxial growth of the Al single grains is favoured.

High-resolution resistance measurements of metal stripes have been performed to study void growth during in situ annealing at 180°C. Void growth has been characterized by resistance monitoring over 14,000 hours. During the annealing of Al-lwt%Cu stripes, Cu atoms from solution migrate to grain boundaries to form Al2Cu precipitates leading to a drop in resistance. At the same time, relaxation of tensile stresses in metal stripes takes place in the form of void nucleation and growth, leading to an increase in resistance. The resistance drop due to precipitation was shown to obey Avrami's precipitation kinetics while void growth over this time period obeyed a diffusion-type equation. The resulting equation to describe this physical model was fitted to the measured resistance data. Extrapolation to failure condition (ΔR/R=10%) could therefore be obtained for all the devices under test and plotted on a cumulative probability plot. Using the measured value of activation energy for void growth, stressmigration reliability was then assessed by extrapolation to an operating temperature. Post-mortem microscopy was performed to correlate resistance increase with void density and size in these interconnects. Finite element simulations were performed to calculate resistance increases due to voiding in metal stripes and correlate these with the experimentally obtained data.

The electromigration-induced transport properties of Cu in Al-Cu alloys, and their effect on electromigration lifetimes in interconnects with bamboo grain structures are not well understood. To isolate and study the mechanisms and kinetics of Cu diffusion and electromigration in interconnects for which grain boundary transport is not dominant, we have developed a test structure consisting of parallel Al single crystal lines, with every alternate line terminating in contact pads. Cu is locally added to the same regions in all the lines, and the effect of temperature and electric field can be simultaneously characterized by analyzing the Cu concentration profile measured using electron-probe microanalysis. Comparison of the calculated values of diffusivities with the diffusivity of Cu through the Al lattice, and through dislocation cores in Al, suggests that the path of diffusion of Cu in Al single crystals is along the Al/AlOx interface.

In this paper, we have studied the segregation phenomenon of Cu on the surfaces of patterned lines, dry-etched films and non-etched films, by using X-ray photoelectron spectroscopy and lower energy Rutherford Backscattering Spectrometry. Significant enrichment of Cu is found on the sidewall of the lines. Annealing at 350°C and above cause the disappearance of this enrichment. Origin and evolution of this Cu enrichment have been investigated on films taken out from different steps of the etching process. It has been found that most of the Cu products induced by the plasma etching are CuCl and CuCl2 and they are removed mostly from the top Al oxide layer by the strip process. On the interface area between Al and the native oxide, considerable quantities of etched induced Cu are retained. This Cu is identified to be mainly metallic Cu. Different from the mechanism explained above, thermal annealing can also cause Cu segregation. We have found that Cu atoms diffuse into the native Al oxide where they form Cu2O.

Accelerated electromigration tests were performed in a High Voltage SEM (HVSEM). These experiments were conducted on 10 identical, passivated Al interconnect test structures at a temperature of 237°C and a current density of 30mA/µm2. Simultaneous testing and observation of the entire structure allowed the void nucleation times to be measured. Two normal distributions fit the nucleation times. The second of these distributions coincides roughly with the distribution of failure times of the test structures. Fracture of the passivation is hypothesized as the mechanism that causes the concurrent late nucleation and failure processes.

The electromigration (EM) behaviour of damascene Al(Cu)-interconnects has been studied for the first time by drift experiments and results directly compared to plasma-etched lines. For the latter EM is known to occur through sequential stages : an incubation period, needed for Cu-depletion, and an Al-drift stage. It is shown that using a damascene implementation, the incubation time significantly increases but the Al-drift velocity does as well. This is demonstrated to be an intrinsic effect, directly related to the technological nature of both metallization schemes. Implications on lifetime prediction at operating conditions are discussed.

Electron beam evaporated Ni thin films are found to have tensile stress of 1.2 Gpa. The stress is relieved upon heating, accompanied by grain growth. The calculated stress matches the experimental result and a mechanism of grain growth is proposed.

The 1 µm-wide Al-0.5wt%Cu/TiN/Ti interconnect, on the oxidized Si(100) wafer without Al sputtering pretreatment on the SiO2 surface prior to Ti deposition, failed after stressing with a current density of 1 × 106 A/cm2 for 190 hr at 175°C. In contrast, the interconnect with Ar sputtering pretreatment did not fail after stressing under the same condition for 550 hr. X-ray diffraction spectra indicated that the Ar sputtering pretreatment lowered Al(111) peak intensity down to 1/46 of its magnitude. This result suggests that the improvement of electromigration resistance by Ar sputtering pretreatment is not due to Al(111) texture, which is opposite to the result of Shibata et al. (Jpn. J. Appl. 32, 4479 (1993)). The elemental concentration distribution in depth was characterized on Al-Cu/TiN/Ti multilayers with and without Ar sputtering pretreatment by using secondary ion mass spectroscopy and Auger electron microscopy. A fairly large amount of oxygen and nitrogen was found to segregate near the Al-Cu/TiN interface. The phenomenon associated with the large amount of segregated oxygen and nitrogen is attributed to the enhancement of electromigration resistance.

This paper investigates Cu segregation and void morphology along AlCu alloy metal lines as a function of resistance change resulting from isothermal DC stressing at 225°C and a current density of J=2×106 A/cm2. The Al-1.5wt.%Cu alloy was deposited via DC magnetron sputtering onto a Si substrate at 525°C with a 1500Å TiW barrier layer. NIST test structures (length = 800µm, thickness = 1.2µm, width = 5 and 10 µm) were utilized in this study. BPSG was used as the insulation layer between Si substrate and conductor. The surface passivation layer was composed of Si3N4/PSG. Various failure criteria were selected to explore the correlation between Cu segregation and void morphology along the metal line and the relative percent resistance change (ΔR/R). The log-normal plots, mean-times-to-failure, and sigmas at each ΔR/R ( -−%, 2%, 5%, 10%, 20%, 100%, 250%) were plotted and listed. The microstructural evolution in terms of void morphology was monitored using SEM. SEM-EDS was used to analyze the Cu concentrations along metal lines tested at various %ΔR criteria.

The development of hillocks on metal films during annealing is detrimental to downstream processing of integrated circuit structures. This work focuses upon the local character of texture and grain boundary structure near hillocks in metal films. It is apparent from the results that local grain boundary structure and texture strength are important parameters in identifying locations in the films that are preferentially susceptible to failure under given conditions. Results in aluminum and platinum films indicate that non-(111) oriented grains preferentially contain hillocks. In addition, (111) oriented grains with boundaries characterized by high angle rotations about random axes are prone to hillock formation.

This paper reports the observations of the grain-refining mechanism found in Al-0.5Cu thin films that are subjected to hydrostatic compressive pressure during the annealing process. The films are deposited on Si substrates and subsequently placed under 60 MPa Ar gas pressure at 400°C. Transmission electron microscopy on these films reveals that plastic deformation occurs by dislocation slip and induces a refined grain structure. Polygonization is the primary mechanism for grain refining, resulting in the formation of sub-grain boundaries.

In previous work we proposed a simple constitutive equation that describes the electromigration failure kinetics in naturally passivated Al(Cu) quasi-bamboo lines. The time to failure, tf obeys the relation, tf = t0 exp (−l/t0), where t0 and l0 are characteristic constants and l is the length of the polygranular segment that causes failure (which is ordinarily the longest polygranular segment in the line). The present paper gathers the data supporting this relation, which includes tests on lines as-patterned and annealed at low and high temperature, and examines the dependence of the factors t0 and l0 on current density, test temperature, line geometry, and Al2Cu precipitate distribution. The experimental data suggest that to varies geometrically with j, with exponent n ≈ 2.8, and exponentially with T−1, with activation energy of ∼0.69 eV. It also varies with the line width and the distribution of Al2Cu precipitates. On the other hand, l0 is only weakly dependent on current density and temperature, but may vary with the ratio of line width to grain size (w/G).