There has recently been an increasing interest in using electroless processes to produce the metal thin film stacks that make up the bond pads in solder ball grid array (BGA) packages. Electroless processes produce a more uniform and better controlled film thickness, but the resulting metal stacks have exhibited less mechanical reliability than that of traditional electrolytic stacks. This paper addresses a layered system consisting of a eutectic Sn-Pb solder sandwiched between stacks of Cu, Ni, and Au thin films. The samples were tested to determine cyclic fatigue behavior and fracture toughness values using traditional linear elastic fracture mechanics techniques. Surprisingly, fatigue crack propagation occurred in the middle of the solder layer even though the electroless interface was expected to be very weak. In contrast, fracture tests did appear to cause failure near an interface, but not necessarily at the electroless interface. Fracture toughness values measured for these samples were much lower than those reported for bulk metals. These results are discussed in terms of the microstructures present in order to determine possible relationships between the microstructure of the solder joint and its fracture behavior.

The mismatch between the coefficients of thermal expansion of silicon chips and their organic substrates has been mitigated through the practice of using underfill in flip-chip packages. Yet solder fatigue and package failures still occur. This is particularly true for flip-chips on organic substrates that are thermally cycled between low (−55 °C) and high temperatures (125 °C). In this study, I used electron-beam moir6 to measure displacements and calculate strains in a solderball contained in a flip-chip plastic-ball grid array package. A crossed-line grating with a pitch of 450 nm was used to allow detailed measurements of the local displacements from a cross section of a flip-chip package. Elastic displacements were observed and measured, and the v-field displacements, out of the plane of the chip, were more significant than the shear or u-field, in the plane of the chip, displacements. Larger v-field displacements were measured near the center of the silicon chip than at the edge of the chip.

The increasing demands on solder joints have made it imperative that they perform not only their traditional role of electrical connection but also possess good mechanical integrity. One such key mechanical property is the shear strength of the solder. A number of specimen geometries can be used to evaluate the shear strength of solders, each with its advantages and limitations. However, these geometries do not provide information about the notched fracture strength of the solder under shear. Studies on a number of ductile materials have indicated that in many of these materials shear is the preferential mode of fracture even when the imposed load is a combination of tensile and shear. This study, which is part of a larger project on the failure of materials under combined tensile-shear loading, uses a modified compact tension specimen to measure the appropriate failure criterion. It is shown that the fracture of solders under these conditions follows the general principles of a mixedmode fracture mechanism map and shear is the preferred mode of failure.

We have studied the effect of a Ti interlayer on the behavior of a TiN diffusion barrier for Al and Cu metallizations. Thermal stability of Al/Ti/TiN/<Si> and Al/TiN/<Si> samples annealed at 400–600°C for 30 min was investigated using Auger electron spectroscopy (AES), glancing angle X-ray diffraction and scanning electron microscopy (SEM). Sheet resistance was measured for electrical characterization.

After annealing at 400°C and 500°C, the AI/TiN/<Si> samples exhibited the same sheet resistance as the as-deposited one, while the sheet resistances of the Al/Ti/TiN/<Si> samples increased upon annealing. After annealing at 600°C, pyramidal pits developed on the surface of the Al/TiN/<Si> sample, but not on the Al/Ti/TiN/<Si> sample. Sheet resistance measurements for the 600°C-annealed Al/TiN/<Si> sample resulted in a more scattered distribution and a higher average value than for the Al/Ti/TiN/<Si> sample. The results clearly indicate that the performance of the TiN barrier layer is significantly improved by including a thin Ti film between the TiN and the Al. The Ti interlayer also improves the TiN barrier performance for the Cu metallization system.

The interfacial reactions that may occur from the ageing of integrated circuits during operation will change their properties and produce failures. It is crucial to understand the phenomena that controls these reactions and their kinetics.

The object of this work was to establish the role of impurities and grain size on the growth rate of TiAI3 from Al/Ti interfaces. The kinetic study consisted of different annealings followed by electrical measurements. These data were correlated with microstructural analysis and composition evolution from Transmission Electron Microscopy (TEM), Secondary Ions Mass Spectroscopy (SIMS).

The kinetic data were fitted using a combination of classical models (parabolic, Avrami, Aronson) and complemented by a simulation, which accounts for the nucleation frequency and the grain boundary diffusion. The growth simulations were compared to microstructures observations using TEM.

TiB2, films deposited by co-sputtering from a boron and a TiB, target are evaluated as the diffusion barrier for Cu metallization. Material characteristics of the TiB, films and metallurgical interactions of the Cu/TiB2/<Si> system annealed at 400−700°C for 30 min, in a 80%Ar+20%H2 flow, were investigated by glancing angle X-ray diffraction, Auger electron spectroscopy (AES), and scanning electron microscopy (SEM). Sheet resistance was measured for electrical characterization.

The composition and resistivity of the sputtered TiB1 films varied with the bias applied on the substrate. To obtain a low film resistivity, a negative bias of 200V was applied during sputtering. The resulting TiB2 film is nanocrystalline with a resistivity of 300 μΩcm. After copper deposition, the Cu/TiB2/<Si> samples have a constant sheet resistance after annealing up to 600°C for 30min. The overall sheet resistance of the sample increases by five orders of magnitude after annealing at 700°C, and scanning electron micrographs reveal that the sample surface is severely deteriorated after annealing at 700°C.

Low resistivity and fully conformal ZrB2 thin films are deposited by remote plasma chemical vapor deposition using zirconium tetrahydroborate, Zr(BH4)4. The problems with thermal CVD using this precursor – excess B incorporation, oxygen contamination, and high resistivity – are eliminated by injecting atomic hydrogen from a remote microwave plasma source onto the substrate. Using this technique, the films are stoichiometric, have ∼ 40 μΩ-cm resistivity, < 4 at.% oxygen contamination, and are fully conformal in deep trenches and vias. We show that a 50 nm thick ZrB2 film on c-Si (100) prevents Cu in-diffusion after 1 hour annealing at 650°C.

Molybdenum nitride thin films were prepared by N2 + implantation with acceleration energy of 20 keV and the ion dose of 3×1017 ions/cm2. The structural property and thermal stability of the films were investigated by XRD, AES, AFM and RBS. The crystal structure of N2 + implanted molybdenum thin films (Mo-N2 +) which had microcrystalline state was transformed to γ-Mo2N phase with a preferred (111) orientation after a post-annealing at 500 °C for 30min. However, a silicide reaction was not observed even after the annealing at 700 °C, which is due to the modification of the interface between Mo thin film and Si substrate by N2 + implantation. Also, Cu diffusion did not seem to be induced by the annealing at 700 °C for 30 min. The internal stress of the Mo-N2 + thin films during post-annealing at 600 °C for 30min was found to change from highly compressive stress to low tensile stress.

TiN and TaN thin films are proposed as barrier layers between copper interconnects and low dielectric constant (low-k) polymers. As the barrier layer thickness is scaled down, the uniformity and morphology of these films will be severely affected by the nitride-polymer interface and become an important issue for the reliability of the whole interconnect structure. In order to evaluate nitride formation and the interfacial chemistry we deposited TiN and TaN on fully cured low-k polymers by two different techniques: reactive evaporation of the metal in nitrogen ambient and by ion assisted reactive deposition using a low energy (100 eV) nitrogen ion beam during evaporation. Photoelectron spectra were recorded in situ for metal coverages from 0.1 nm until bulk like metal or metal nitride spectra were obtained. Nitride concentrations, extracted from the photoelectron spectra, show that even though very similar nitride films are produced by both techniques for thicker films (>5 nm) we only find significant amounts of nitride at the interface in the ion assisted case. Thinner films formed in nitrogen ambient were very similar to those where the pure metal was deposited and were dominated by the formation of compounds with carbon and oxygen from the polymer. This shows that the composition of barrier layers can be drastically altered near the polymer interface. Low energy ions in contrast allow the growth of more homogeneous films which can significantly improve the reliability of copper based high density interconnects.

A tool for simulation of electromigration and electromigration-induced damage, MIT/EmSim, has been used to investigate interconnect reliability, focusing on transitions in failure mechanisms associated with void nucleation, growth, and growth saturation. Conventional scaling of electromigration test results assume that the median time to electromigration-induced failure scales with the current density j to the power −n. The effects of transitions in failure mechanisms have been studied by characterizing the apparent current density exponent. When failure is limited by void nucleation, n=2 scaling is observed, and when failure requires substantial void growth, n=1 scaling is observed. When lines end at diffusion barriers such as W or liner-filled vias, void growth saturates in short lines at low current densities, and, depending on the failure criterion, lines under these conditions can be ‘immortal’. As growth saturates, apparent current density exponents increase above 2. Failure mechanism maps can be constructed to illustrate the failure mechanisms and scaling behavior as a function of line length and current density. Failure maps can be used in accurately scaling test results to service conditions, to suggest layout strategies for optimized circuit reliability, and to assess the reliability of new interconnect materials and structures.

To improve the fundamental understanding of alloying effects in electromigration, in particular of Cu addition to Al conductor lines, the electromigration process in the grain boundary of an Al-Cu alloy is simulated using a 2D kinetic Monte Carlo method. These simulations give the fluxes of Al and Cu from microscopic parameters, which determine the probability of each atomic jump. The parameters used in these simulations are the diffusion barriers, the attempt frequencies, the electromigration driving force, the temperature, the Cu concentration, and the binding energy of a Cu-vacancy pair. Values for the electromigration driving force on Al and Cu atoms are calculated ab initio. A very interesting result of the kinetic Monte Carlo studies is that the Al flux is reversed due to the addition of a small amount of Cu if the binding energy for the Cu-vacancy pair is larger than 0.12 eV. Such an Al back flux can explain the strongly inhibiting effect of Cu on electromigration damage initiation and transport in Al.

This paper considers an aluminum line in a multilevel interconnect structure. Upon cooling from the processing temperature, differential thermal contraction causes a triaxial tensile stress state in the aluminum line; voids may initiate and grow to relax the stress. When a direct voltage is applied, the electric current causes aluminum atoms to diffuse. The interconnect will evolve to a state with a high pressure at the anode, and a large void at the cathode. The pressure may crack the surrounding insulator or debond an interface, extruding aluminum. The void may uncover the via contact area, substantially increasing electrical resistance. Provided neither failure mode occurs, aluminum electromigration will stop and the interconnect will function forever. This paper examines the conditions under which the interconnect is immortal.

Numerical modeling of electromigration stress buildup and flux divergence was undertaken. The objective is to provide a mechanistic understanding of the question: Do preexisting stress voids grow during later electromigration, to become potentially fatal? Conflicting experimental results have been reported. In this work the stress field in aluminum interconnects containing voids was first quantified using the finite element analysis. The averaged stress field then served as the initial condition in the one-dimensional partial differential equation of electromigration stress buildup. The finite difference method was employed to solve the evolving stress profile and the associated atomic flux and flux divergence along the conducting line. It was found that a large preexisting void suffers greater flux divergence and is therefore more prone to growth during electromigration. A single large stress void is more detrimental than populous small voids. Published experimental observations seem to support these findings.

A microscopic nucleation model for electromigration is proposed. The solution to the master equations for site occupancy produces a variable current exponent with 1 < n < 2. These results disprove previous claims that n = 2 in nucleation models. We also address the long-standing speculation regarding an additional temperature dependence in Black's equation. Under the premise that the activation energy in Black's equation, E a , should be equal to the energy barrier for atomic hopping, we provide evidence for an additional T 2 dependence of the failure time. Microscopic modeling provides an explicit connection between the macroscopic model behavior and the microscopic parameters of the model. A continuum treatment of our model is derived and compared with standard continuum treatments.

In this work, the electromigration (EM)-behaviour of bamboo RIE and damascene Al(Cu) interconnects has been studied by drift experiments. For RIE lines, both lattice and interface EM can be operative, depending on the Cu-distribution. Cu-alloying only retains its retarding effect for Al-diffusion at metallic interfaces, while the accompanying presence of an incubation time was found to become rate-controlling for the EM-performance at operating conditions. Contrary to the case of polycrystalline lines, the higher EM-threshold still observed for bamboo damascene relative to RIE is now insufficient to compensate for the significantly increased Cudepletion rate in bamboo damascene due to both geometrical and metallurgical reasons.

Electromigration is a complex process consisting of the nucleation, growth, and movement of voids. While interconnect microstructure plays a major role in determining void nucleation time and location of void nucleation, recent studies have shown that this is not the only contributing factor. Thermodynamics and kinetics-based calculations by Flinn and Gleixner et al. have shown that electromigration void nucleation by vacancy condensation at both homogeneous and heterogeneous sites will not occur at a reasonable rate without a mechanism for reducing or altogether eliminating the energy barrier for nucleation. By implanting argon ions into the aluminum interconnect lines, we introduce an initial defect population (argon bubbles) of controlled size and location.

Tests were performed in a high-voltage SEM (120 keV), which enables in-situ observation of the voiding process through the passivation layer. Images taken throughout the in-situ tests were analyzed to determine void nucleation times and locations. In the argon-implanted interconnects, ten of the 15 voids that nucleated were within implanted regions. Voids nucleated in the interior of the line within the implanted regions, as well as at the passivation/sidewall interface where voids are typically seen in conventional electromigration tests. In addition, voids in implanted regions nucleated much more quickly than those in unimplanted regions. These observations support the idea of argon bubbles reducing the nucleation barrier. TEM was used to analyze the microstructure of both control and implanted interconnect lines.

Void motion can be an important part of electromigration failure in interconnects. Investigating this process is difficult due to the large variation in behavior in identical samples tested under identical conditions. To deal with the large variability, we have tested 50 samples to failure at four different accelerated test conditions. Because our automated, in-situ technique images an entire test structure, the complete history of a test line is recorded. By testing many samples under direct observation, the complex motion of electromigration voids was comprehensively studied.

In this work the authors want to report some experiments concerning unpassivated Al interconnect lines of 8 and 1.4 microns width which have been damaged by in-situ electromigration in the SEM (temperature 230°C, current density 2 and 4×106 A/cm2, respectively). The wider line represents a polygrained structure with few blocking grains spanning the whole width, whereas the narrow line shows bamboo structure. Before electromigration, the local orientation and thus the position of all grain boundaries was mapped by EBSD technique along the entire interconnect line. During and after in-situ current loading in the SEM, the damaged sites were correlated with the grain boundary map to locate where the diffusion paths are situated most likely. It was found that not the deviation from <111> fibre texture, but the misorientation class of the grain boundaries is essential for the localization of the fatal defects.

RF-magnetron sputter deposited copper/zirconium alloy films have been characterised in terms of resistivity, composition, microstructure, texture and stress. The properties and electromigration reliability of these alloys have been compared to those of pure sputtered copper films. The surface segregation of zirconium within the films after annealing has been studied. We have proposed a mechanism in which zirconium at the copper surface can influence the electromigration damage process, enhancing reliability.

To investigate the cause of the lower electromigration (EM) resistance in Al/Al3Ti lines compared to that in Al/TiN lines, we studied the Al microstructure and the EM characteristics. Even after fine-grained Al3Ti formation through a reaction of the Al and Ti, the Al microstructure of the unreacted Al on the AI3Ti remained a bamboo grain structure with a high 111 orientation. Thus, the Al grain structure and texture did not affect the EM resistance. The critical current density for voiding was 0.64 MA/cm2. The Al drift velocity in the Al/Al3Ti bamboo lines was two orders of magnitude higher than that for Al/TiN bamboo lines from 180°C to 280°C. This higher Al drift velocity decreased the EM resistance in the lines. The activation energy of Al drift in the AI/AI3Ti lines was 0.6 to 0.7 eV, which is lower than that for Al lattice diffusion. Thus, the Al lattice is not the dominant transport path in Al/Al3Ti lines. The dependence of the Al drift velocity on the line size showed that the Al/Al3Ti interface was the dominant path in the Al/Al3Ti bamboo lines. Thus, we concluded that the rapid Al transport through the Al/Al3Ti interface decreases EM resistance in Al/Al3Ti lines.