S.W. Russell

Enhanced adhesion of copper to dielectrics via titanium and chromium additions and sacrificial reactions

Abstract Ti and Cr as both interposed layers and alloying components were found to enhance copper adhesion to dielectrics. Films deposited on SiO2, phosphosilicate glass (PSG) and boronphosphosilicate glass (BPSG) were annealed in 95%Ar-5%H2 over the temperature range 400–600 °C. The force required to separate films from substrates was measured by scratch testing. Optical and scanning electron microscopies provided detection of substrate exposure. In the Cu Ti and Cu Cr bilayer systems the force decreases with temperature on all substrates, generally exhibiting better adhesion on SiO2 than on PSG or BPSG. In the Cu(Ti) and Cu(Cr) alloy systems the force increases with temperature with less systematic difference among the three substrates. These results correlate well with tape testing. Ti and Cr segregate out of the Cu layer and react both with the dielectrics and with the ambient gases, as observed by Rutherford backscattering and secondary ion mass spectroscopy. These reactions appear to improve adhesion; however, only a small amount of this reaction is required for the enhancement to occur. We surmise that stress in the copper and/or voiding at the Cu-dielectric interface may play a role as well. We observe a correlation between adhesion and the degree of Cu texturing.

Passivation of Cu via refractory metal nitridation in an ammonia ambient

Cu-(27at.% Ti) and Cu-(26at.% Cr) alloys codeposited on silicon dioxide substrates were isochronally annealed for 30 min at 400–700 °C in a flowing NH3 ambient. In the Cu-Ti alloy, Ti segregates to the free surface to form a TiNx(O) layer and also to the alloy-SiO2 interface to form a Ti5Si3TiOw bilayer structure. Therefore the resulting structure is an almost completely dealloyed Cu layer located between a surface oxygen-rich Ti nitride and Ti-silicide/Ti-oxide bilayer interfacial structure. In the Cu-Cr alloy system, Cr seems to migrate only to the free surface to form a CrNx passivation layer. A 45 nm Al film was deposited after nitridation, whereupon a second anneal was performed to evaluate these nitride layers as diffusion barriers. The Cr-nitride diffusion barrier is stable up to 600 °C compared with Ti nitride that fails at 500 °C. The Cu-Cr nitrided samples also showed an overall lower sheet resistance.

Effectiveness of Ti, TiN, Ta, TaN, and W2N as barriers for the integration of low-k dielectric hydrogen silsesquioxane

The interactions between low-k (dielectric constant) material hydrogen silsesquioxane (HSQ) and barrier layers, Ti, Ta, physical-vapor deposited (PVD), and chemical-vapor deposited (CVD) TiN, PVD TaN, and CVD W2N, have been investigated by using sheet-resistance measurement, x-ray diffraction, transmission electron microscopy, Rutherford backscattering spectrometry, elastic resonance scattering, and forward recoil spectrometry. The conventionally used dielectric PETEOS [plasma-enhanced chemical vapor deposition tetraethylorthosilicate (TEOS)] was also studied as a control. The results show that none of these barriers except Ti can react with HSQ and PETEOS at elevated temperatures. However, significant outdiffusion of hydrogen due to the degradation of the HSQ films upon annealing exists for all barrier/HSQ structures, and exhibits a strong barrier dependence. Metal barriers Ti and Ta, and CVD TiN can retard the H outdiffusion, whereas PVD TaN and PVD TiN induce the H outdiffusion from the HSQ films. The ...

Formation of passivation and adhesion layers for Cu via nitridation of Cu-Ti in an ammonia ambient

Abstract Passivation and adhesion layers for Cu have been formed by nitridation of Cu(90 nm)/Ti(20 nm)/SiO 2 (90 nm) bilayers and 163 nm thick Cu(Ti 27 at.%) alloys at 400–700 °C for 30 min in an ammonia ambient. In both systems Ti segregated during annealing to the free surface to react with the NH 3 and formed an oxygen-rich Ti nitride passivation layer. The thickness of these layers was ~ 12 nm for the bilayers and ~ 20 nm for the alloys. Evaluation of the Ti nitride surface layers as diffusion barriers for Al and Cu showed stability up to 500 °C. The interfacial reaction between Ti and SiO 2 substrate resulted in the formation of a Ti 5 Si 3 /TiO w structure. Adhesion results obtained from a scratch test showed that the bilayers exhibited good adhesion in the as-deposited state and were stable up to 400 °C, but displayed only fair adhesion above that temperature. The alloys exhibited poor to fair adhesion in the as-deposited condition and at 400 °C, but adhesion improved considerably after a 500 °C anneal. The behavior of the adhesion properties of both the bilayer and alloy system were related to the interfacial reaction.

Influence of interfacial copper on the room temperature oxidation of silicon

Thick (∼1.3 μm) oxide films were grown by room‐temperature oxidation of silicon after low‐energy copper‐ion implantation. The structural properties of the silicon dioxide layer and the implanted silicon were characterized by Rutherford backscattering spectrometry and transmission‐electron microscopy. During the room temperature oxidation a portion of the implanted copper resided on the surface and a portion moved with the advancing Si/SiO2 interface. This study revealed that the oxide growth rate was dependent on the amount of Cu present at the moving interface. The surface copper is essential for the dissociation of oxygen at the surface, and it is this oxygen that participates in the oxidation process. The resulting oxide formed was approximately stoichiometric silicon dioxide.

Quantification of carbon in Si1-x-yGexCy with uniform profiles

Methods to quantify the carbon concentration of CVD grown Si1−x−yGexCy (0.25 < x < 0.37 and 0.01 < y < 0.12) layers on (100) Si with uniform composition profiles were investigated. Two analysis techniques were used: Rutherford backscattering spectrometry (RBS) using a 4.295 MeV He+2 incident ion and elastic recoil detection (ERD) using a 24 MeV Si+5 incident ion. For the RBS measurements the 12C(α,α)12C elastic resonance reaction near 4.265 MeV was used to enhance the scattering cross section of carbon. These carbon concentrations were calculated by either integrating the resonant scattering cross section across the energy width of the layer or by using a Lorentzian fit to estimate the area. The backscattering data were additionally analyzed with the program RUMP. These different analysis techniques resulted in a large scatter in the RBS predictions for the carbon concentrations depending on how the resonant cross sectional area was calculated. The appropriateness of each technique was judged by comparing the predicted concentrations to those obtained by ERD. The divergence between the carbon concentration predicted by using the Lorentzian approximation and the ERD values was great enough to deem this method as inappropriate. The values obtained by RUMP were systematically greater than the ERD concentrations, however the percent difference was never more than 20. The predicted carbon concentration that had the closest correlation to ERD was found by integrating an appropriate scattering cross section across the energy width of the layer.

Thin film interaction between low-k dielectric hydrogen silsesquioxane (HSQ) and Ti barrier layer

The interaction between low-k dielectric hydrogen silsesquioxane (HSQ) and Ti barrier layer has been studied using four-point-probe sheet resistance measurement, X-ray diffraction, conventional Rutherford backscattering spectrometry (RBS), nuclear resonance analysis (NRA), elastic recoil detection (ERD), secondary ion mass spectrometry (SIMS), Auger electron spectroscopy (AES) and thermal desorption spectroscopy (TDS). The conventional intermetal dielectrics SiO2 and plasma-enhanced tetraethylorthosilicate (PETEOS) have been studied also for the purpose of comparison with HSQ. In the low temperature regime (300–550°C), a considerable amount of oxygen atoms, from various sources, diffuses into Ti film to form a Ti(O) solid solution, raising the resistivity of Ti significantly and causing the expansion of the Ti lattice. A good correlation between the oxygen composition in the Ti film, the sheet resistance variation of Ti and the change of Ti lattice parameter C0 have been observed. At the same temperature, there are more oxygen atoms incorporated into the Ti film in Ti/HSQ than those for Ti/PETEOS, suggesting that additional HSQ-related oxygen sources, such as the moisture uptake and the conversion reaction of HSQ, may be attributed to this. In the high temperature regime (550–700°C), HSQ reacts with Ti to form a final TiO/Ti5Si3/HSQ stack structure. It is assumed that a few competing reactions occur in this regime. At 550–650°C, HSQ reacts directly with Ti; in the meantime, part of HSQ undergoes conversion reactions, with the reaction products SiO2 and SiH4 reacting with Ti to form Ti silicide. At 650–700°C, HSQ is almost completely converted into SiO2, so the dominant mechanism is Ti reaction with SiO2. Before HSQ is completely turned into SiO2, the Ti/HSQ system is more reactive than both Ti/PETEOS and Ti/SiO2. The initiating temperature for the Ti/HSQ reaction exhibits no obvious Ti thickness dependence.

Titanium Nitridation on Copper Surfaces

Cu{sub 74}Ti{sub 26} alloy films were annealed in flowing NH{sub 3} at temperatures of 450 to 700 C. Ti segregates to the free surface to form TiN, leaving nearly pure Cu in the remaining metal film. Elastic recoil detection and Auger electron spectroscopy were applied to investigate the chemistry of the as-formed nitride as a function of anneal temperature. The use of these two techniques in tandem offers a combination of standardless compositional determination, excellent mass and depth resolution, and chemical bonding information. The Ti:N ratio is nearly 1.0 for T {ge} 550 C, and the nitride contains {approximately}6 atom percent O. The amount of nitrogen incorporated increases with temperature with an activation energy of 0.6 eV. A native Ti oxide remaining at the TiN/Cu interface suggests Ti is the dominant diffusing species during nitridation.

Guidelines to the application of nuclear resonance to quantitative thin film analysis

Abstract The use of elastic nuclear resonances, including 12 C(α, α) 12 C, 14 N(α, α) 14 N and 16 O(α, α) 16 O, provide a useful means of enhancing the sensitivity toward light elements using the same experimental setup as for Rutherford backscattering. Quantitative information about light element concentrations is only obtainable under certain conditions, and the use of simulation programs in conjunction with resonance analysis may often lead to erroneous results. By using resonance near the peak value of the cross section one may enhance sensitivity; however, this may result in a loss of precision, defined as day-to-day repeatability of the measurement. Conversely, using resonance in an energy regime in which the cross section varies less rapidly may more accurately predict the actual composition than standard RBS but prove less useful for low concentrations. We explore the importance of film thickness and composition and the variation in the incident beam energy toward the selection of the appropriate resonance regime, as well as discuss common sources of error.

Comparison of elastic resonance and elastic recoil detection in the quantification of carbon in SiGeC

Abstract The carbon concentrations of chemical vapor deposition grown Si 1− x − y Ge x C y (0.25 x y 0.12) layers on (100) Si with uniform composition profiles were quantified by two ion analysis techniques. Measurements made with backscattering spectrometry using a 4.295 MeV He 2+ incident ion were compared to compositions predicted by elastic recoil detection (ERD) using a 24 MeV Si 5+ incident ion. To enhance the carbon scattering cross section for the backscattering measurements, the 4.265 MeV 12 C(α, α) 12 C elastic resonance reaction was used. The carbon concentrations of the films were calculated by integrating the resonant scattering cross section using the energy width of the layer as the limits of integration. The results of this backscattering analysis technique were compared to the predicted carbon concentrations obtained by ERD. It was found that the predictions of these techniques correlated within the uncertainty of each method.