At the MRS Spring Meeting in San Francisco this year, a special historical session was held in the symposium on Materials Reliability in Microelectronics to reflect upon thirty years of research into electromigration as a failure mode in integrated circuits. Six of the original “discoverers” of electromigration as a reliability issue were invited to speak of their work and to relate stories of the early days of these historic efforts.

To enhance the corrosion resistance and reliability of the proposed copper interconnections in silicon integrated circuits, alloying with small amounts thermodynamically favorable elements has been pursued. In the present investigation dilute copper (boron) alloy thin films (in boron concentration range of 0–4 at % in copper) were deposited by DC magnetron co-sputtering using a high purity copper and Cu-4 at. % B targets. Films were then annealed in Ar-3% H2, pure Ar, vacuum, and air ambients in the temperature range of 200–500°C. Sheet resistance, Rutherford backscattering, x-ray diffraction measurements were made to characterize the films. The residual resistivity of the as-deposited alloy films was found to be 5.3 μΩ-cm/at %. To obtain sufficiently low working resistivity, an alloy content below 0.5 at % is suggested for application as a potential metallization material. The addition of boron, which is the common dopant in Si. to the copper films offers considerable oxidation protection. The resulting oxidation rates are considerably lower than that for pure copper films. All this will be presented and discussed.

Co-deposition of copper-palladium alloy films was demonstrated using low-pressure CVD from individual precursors - (hexafluoroacetylacetonato)copper(I)(vinyltrimethylsilane) [(hfac)Cu(I)vtms] and palladium(II)bis(hexafluoroacetylacetonate) [Pd(hfac)2], for Cu and Pd respectively. High-purity alloy films with controlled composition, microstructure and morphology were prepared and their oxidation behavior was examined at different temperatures and compared with that of pure copper CVD films with similar morphology. As-deposited copper-palladium alloy films showed improved resistance to oxidation up to 300°C in air. Enhanced oxidation resistance compared to Cu CVD films was observed at concentrations of palladium as low as 0.5 at. %.

Copper stripes, current stressed under UHV, clean surface conditions, have an activation energy Q for electromigration (EM) of 0.5 eV, significantly lower than the 0.8 eV reported when the surface has been exposed to air. Application of positive or negative potentials to the stripes during EM testing raises or lowers the clean surface Q in a linear relationship in the range of 0.2 to 1. 0 eV. These results were obtained from in-situ electrical resistance measurements during which potentials up to +/− 800 volts were applied to the stripe. Thus the rate at which surface EM takes place is influenced by the sign and magnitude of the applied potential. The low Q for the in-situ clean surface studies also implies that surface diffusion is the dominant physical mechanism whereby EM damage occurs. Ex-situ SEM studies support this conclusion. This work has important implications for ULSI since numerous recent studies have indicated that surface and/or interface diffusion are present during current stressing of passivated and unpassivated fine line stripes.

Atom movements of Cu in pure Cu and Cu(0.5 to 2 wt.% Sn) alloys have been investigated using drift velocity and radioactive tracer techniques. The void growth rate in pure Cu at the cathode end, as a result of electromigration driving force, linearly increases with the applied current density. A marked decrease in the Cu grain boundary diffusivity and electromigration drift velocity is attributed to Sn trapping of Cu atoms and/or binding defects at grain boundaries. The effect is more pronounced at lower temperatures. The activation energies for electromigration and diffusion of Cu in pure Cu grain boundaries were found to be in the range of 0.7 – 0.9 eV. Addition of about 0.5 to 2 wt.% Sn increased these energies to 1.1 – 1.3 eV, respectively which resulted in enhancement of electromigration resistance by several orders of magnitude over that in pure Cu at the field operation temperature.

Activation energy of electromigration damage was determined as 1.2 eV for Cu lines, indicating grain boundary paths for electromigrationo The surface diffusion was found to play a role during electromigration in Cu and impeded the failure process.

Copper films were deposited onto oxidized silicon wafers by the self-ion assisted technique. A 0 and 6 kV bias was applied to the substrate during the deposition. The films were patterned into parallel line arrays of 20 lines 0.5 mm long, using electron lithography and dry etching. After patterning, the lines were covered by silicon oxide and annealed in vacuum for 1 hour at the temperature 450° C. Electromigration testing was performed in air in the temperature range from 280° to 350° C and at a current density 3.106 A/cm2.

It was found that the resistivities of the films deposited at 6 kV and without bias were 1.7 and 2.0 μΩ-cm, respectively. The median times to failure are 398.6 and 240 h and the deviations in the time to failure are 0.8 and 0.54 for 6 kV lines and 0 kV lines, respectively. An electromigration activation energy of 0.89 eV was found for 0 kV films.

CuCr alloys with varying Cr content were sputter deposited on polyimide films, and the metal/polyimide films were maintained under 350°C/N2 environment up to 10 hours for reliability measurements. Before exposure to 350°C, the peel strength increased proportionally with the Cr content in the alloy layer up to 17% and saturated. After exposure to 350°C, the peel strength dropped for all the specimens due to the formation of Cr2O3 at the metal/polyimide interface, and the effect was most drastic for the specimen with the lowest Cr content, which failed along the Cr-oxide/polyimide interface. Depletion of carbidic bonds at the metal/polyimide interface accompanied the oxide formation, and the result was quite devastating for the specimens with low Cr content in the alloy layer.

Electromigration lifetime dependence on crystallographic texture for AlCu interconnects is determined. It is found that enhancement of AlCu texture at <111> orientation improves EM endurance. But this beneficial effect is limited after a certain level of texture enhancement is reached. The effect of lifetime improvement is proved to result from a decrease in atomic diffusivity. Saturation of the lifetime improvement effect for highly textured AlCu indicates a change in the main diffusion mechanism for electromigration, possibly from the regular grain boundary diffusion to diffusion through edge dislocations.

Electromigration failures occur at locations where Al mass transport flux divergences exhibit a maximum. In VLSI circuits, these locations correspond to tungsten plug contact / via areas. Characterizing and understanding electromigration failures at tungsten plug contact / via areas are essential in assuring the reliability of 0.5 μm and smaller devices. In this study, a novel test structure with four vias connecting metal 1 and metal 2 levels was used to separate two stages in the Al-Cu interconnect electromigration process. These stages are the incubation time during which Cu in the Al-Cu alloy is swept away from the via area and an Al drift time which leads to interconnect failure. By the use of this structure, a non-destructive method has been established to measure incubation times and Al drift times separately, monitoring discrete increases in line resistance corresponding to a sequential depletion of all four vias. Activation energies for both processes have been determined. The temperature dependence of the incubation time is characterized by a thermal activation energy of Q=1.08 ± 0.15 eV, indicative of Cu grain boundary diffusion through Al grains. Al depletion was found to occur with an activation energy of Q=0.72 ± 0.12 eV. We therefore conclude that Al self-grain boundary diffusion constitutes the Al depletion mechanism. The application of this novel test structure enables distinct determination of dominating failure mechanisms under operating conditions and therefore provides a realistic reliability assessment of actual on-chip interconnect structures.

Effects of finite source material, in multilevel interconnect systems, on reliability has been investigated. Two types of novel test structures were used in this study: 1] asymmetrical design structure in width and thickness for the top and bottom conductor layers, 2] symmetrical design structure in width and thickness for the top and bottom conductor layers. The results of the asymmetric contact and via chains studies showed that the electromigration lifetimes were not effected by the difference in the magnitude of the finite source material. The electromigration lifetimes of the symmetrical via structure, in width and thickness for the top and bottom conductor layers, exhibited a strong dependence on the finite source material. A failure model is proposed to explain the effect of finite source material on electromigration performance in such interconnect systems.

Accelerated life testing has long been used to measure the electromigration reliability of integrated circuit (IC) metallization systems. In order to establish or verify electromigration design rules for IC products the measured data is extrapolated over a wide range of operating conditions using phenomenological models. These models assume that the components of the diffusional flux, thermal stress and resulting void morphology are independent of test/operating condition. In this paper, electromigration void morphology and failure criteria are studied over a range of stress conditions and microstructures for non-layered AlCu and AlCuSi metallurgies. The failure criteria, defined as the average change in conductor resistance prior to an open circuit condition, was strongly dependent on test current density. Moreover, the nature of the relationship between failure criteria and current density was governed by the linewidth (W) relative to the median grain size (D50). This dependence can be explained qualitatively in terms of the Blech Effect.

Electromigration (EM) void nucleation and growth is a failure mechanism of integrated circuit (IC) metallization. The time-to-failure of interconnect lines depends on the void nucleation time and the void growth time. The current understanding of the void growth stage is minimal, and characterization of the void growth stage is essential to further explain EM performance of IC metal interconnections. This work used high-resolution electron-beam lithography to define small dimension edge-voids into gold lines at various separation distances from each other, on the same side or opposite sides of the lines. The EM-induced interaction behavior of pre-defined voids was measured in a FESEM in-situ. Results showed that for small separation distances, void-void interaction enabled shape changes in the pre-patterned voids. For larger separation distance, void-void interactions could be characterized by secondary, induced void and hillock area measurements. As the separation distance increased, the void-void interaction diminished, and the voids acted independently of each other.

The effects of the duty cycle and frequency of a pulsed-current stress on the critical length-current density product, (jlc) were measured using edge-displacement techniques. In our study, Al edge-displacement segments were stressed at a temperature of 200 °C and at peak pulsed-dc current densities below 6×105 A/cm2 in a nitrogen ambient. jlc was measured for duty cycles of 25%, 50%, 75% and 100% at 100 kHz, and for 50% duty cycle at lower frequencies. jlc was observed to increase with decreasing duty cycle and increasing frequency. We measured a 2.6-fold increase in jlc for the 25% duty cycle at 100 kHz as compared to dc. We have used the vacancy supersaturation model to predict the jlc enhancement for lower duty cycle operation.

Electrical resistance data were gathered from AI(Cu) VLSI metallization test structures during pulsed DC current stressing (15 mA peak) at elevated temperature (200 °C). Duty cycles of 50%, 67%, and 80% were applied at a frequency (repetition rate) of 133 MHz. Also, a duty cycle of 50% was applied at frequencies of 100 KHz, I MHz, and 133 MHz. The resistance-to-starting resistance ratio, R/Ro, was logged over time, and the data were used to extract the median time dependence of the resistance change. This, in effect, reveals the dependence of the median “failure” time on the R/Ro (or ΔR/Ro) criterion. An “enhancement factor” was defined as the ratio of the median “failure” time from a given pulsed test to that from a DC test with the same peak current. A weak dependence of this enhancement factor on R/Ro was found in the range examined. The degree of enhancement corresponds more closely to the average current density prediction than to the on-time prediction, although there is some small variation according to duty cycle and frequency.

Changes in the electrical resistance induced by electromigration in short (< 20 μm) Al lines show a rather well-defined behavior. For current densities j below a critical value jc the resistance change saturates with time and the resistance fully recovers when the current is switched off. Above the critical current density the induced resistance changes do not saturate and vary approximately linearly with time. In this case the resistance changes recover only partially after removal of the current. We report (i) measurements of the current dependence of the magnitude of the reversible resistance changes and (ii) the results of atomic force microscopy (AFM) inspection of the lines after stressing with current densities above jc. The resistance measurements were made with a high-resolution AC bridge technique. The samples were pure, unpassivated Al lines with a film thickness of 100 nm and a line width of 2 gm. The results show a linear dependence between the magnitude of the reversible changes and the current density. The linear dependence is predicted by two models. The first is based on a description of the vacancy flux and the second on a description of the build-up of mechanical stress during an electromigration experiment. To study the origin of the irreversible effects, samples were stressed at current densities above jc, and the induced irreversible changes in the resistance were recorded. Both negative and positive changes of the resistance were observed. After six hours the experiment was stopped and the lines were inspected by atomic force microscopy. It was always possible to observe a void, a hillock or a hillock/void pair that was created during the passage of the DC current. Moreover, lines with decreasing resistance during stress always showed a hillock and lines with an increased resistance always showed a void.

The continuing scaling-down of integrated circuits leads to increased metallization reliability problems, especially electromigration. We used 1/f noise measurements to study the relation between electromigration and microstructure. These measurements are very sensitive to the microstructural attributes, such as grain boundaries and dislocations. Al lines were grown by graphoepitaxy: First, a pure Al film was grown by dc magnetron sputtering on a groove pattern etched into a SiO2 substrate. The growth was then followed by an in situ rapid thermal anneal that resulted in a complete filling of the grooves with Al. These Al lines were carefully characterized with SEM and Backscatter Kikuchi Diffraction. Depending on the presence of a temperature gradient during the anneal, the lines were either nearly single-crystalline or bamboo with one grain per ∼ 3 μm. The resistivity was ∼ 2.8 μΩcm, only slightly higher than for bulk Al. We measured the 1/f noise with the two-channel ac technique at RT. We found in both bamboo as well as the single-crystalline lines a very low noise intensity; a factor two lower than in conventionally sputter deposited and annealed Al lines. No clear difference between the noise spectra of the bamboo and the single-crystalline lines was observed. We concluded that grain boundaries are not the only contributor to 1/f noise; other types of defects must play a role as well.

A high resolution resistometric measurement technique (HRRMT) is used to study electromigration (EM) in different Al-based metal lines. Instead of using a bridge-method, the high resolution is now obtained by performing experiments in a very stable temperature environment. The main advantage of this technique is that resistance variations are measured, due to processes that occur in the tested metal line only. During EM, both resistivity and geometrical variations can occur, both resulting in a resistance change. A new method is proposed in order to separate these two contributions on-line during the whole EM-experiment. It is shown that even at low current stress, geometrical changes are the main cause for the EM-induced resistance increase.

When a metal conductor is driven by an AC current of frequency ω, its resistance will vary at a frequency 2ω, and a third harmonic voltage will be generated. If the conductor is also under a DC bias, a second harmonic voltage will arise as well. The magnitude of these harmonic voltages increases with damage because of nonlinear increases in the resistance caused by localized Joule heating. We use the harmonic technique to evaluate the damage in metal lines. The correlation between deliberately introduced defects of different sizes and the amplitude of the harmonic voltages has been studied experimentally. The harmonic technique shows higher damage sensitivity than the commonly used resistance method. Additionally, it has a better rejection to noise and resistance drifting.

Thin polycrystalline aluminum films were investigated by high-resolution ac noisemeasurements before and after damaging by high direct current. Immediately after the interruption of the dc-stress a transient noise component was observed that was inversely proportional to the square of the frequency f (1/f2 -noise). It was caused by discrete jumps in the resistance presumably due to the relaxation of mechanical stress. The second component of noise was stable and in all cases approximately proportional to 1/f. The spectral density of 1/f-noise showed characteristic discrete steps as a function of damaging time, in contrast to the resistance which increased almost continuously up to the failure of the film. This indicates that nucleation-and-growth processes of mobile defects were observed in the noise measurements. Thus noise measurements might help to understand the microscopic process of electromigration.