Cu interconnects with Ru diffusion barriers
10/01/2005
Extensive characterization of physical vapor-deposited ruthenium and RuxNy films was carried out to evaluate their performance as a copper diffusion barrier. Barrier strength, crystal structure, film stress, adhesion, and electrical resistivity were studied using Rutherford backscattering spectroscopy, scanning electron microscopy, x-ray diffraction, stress gauge, tape-pull test, and a four-point electroprobe.
Traditionally, TaN/Ta or TiN/Ti bilayer barrier films have been used for copper diffusion barriers - the main reason is that metallic Ta does not bond well to the nonmetallic Si/SiO2 substrate and Cu does not bond well to nonmetallic TaN [1]. However, such usage requires a physical-vapor deposition (PVD) or atomic-layer deposition (ALD) Cu seed layer to facilitate Cu electrochemical plating (ECP). As the feature size of interconnects becomes smaller, the composite thickness of the barrier/Cu seed layer is becoming too thick relative to via/trench size.
Recently, ruthenium has emerged as an alternate barrier material because Cu can be plated directly onto Ru without a Cu seed layer [2]. In this article, both Ru and RuxNy films, which were prepared by direct and reactive PVD, were evaluated to determine their applicability. Although Ru has shown excellent barrier strength up to 700°C, its adhesion to the substrate (Si and SiO2) is found to be unacceptably poor. Adhesion is known to be one of the most important factors in microelectronic interconnects because device failures are often associated with stress and electromigration that are enhanced at poorly bonded interfaces [3, 4]. This work reports on the strength and weakness of Ru base barriers and suggests potential ways to use Ru for seedless Cu ECP.
Experimental method
Ru and RuxNy films were prepared in a physical-vapor deposition (PVD) cluster tool that allowed direct deposition of ruthenium, reactive deposition of ruthenium nitride, and copper separately or in tandem, without breaking the vacuum. All films were prepared on 200mm wafers. Specific deposition conditions are addressed with the data. Thin Ru films were electrochemically plated with Cu to verify direct plating capability. Rutherford backscattering spectroscopy (RBS) and scanning electron microscopy (SEM) were employed to determine the ranges of Cu diffusion. X-ray diffraction (XRD) was used to ascertain the crystal structure of the films. A stress measurement system was used to establish the stress level of the deposited films. Adhesion strength was evaluated using the ASTM Standard Tape Test Method [5]. Film sheet resistance was measured by a commercial four-point electroprobe. Film thickness was derived from the weight of the film and specific gravity and from a well-calibrated deposition rate.
 Figure 1. XRD peaks for PVD a) Ru and b) RuxNy. |
null
Results
Ru films were deposited at 100°C with power at 1kW and with Ar pressure at 2.5mtorr; RuxNy films were deposited with power at 3kW and an Ar+N2 pressure of 4.8mtorr. Both Ru and RuxNy films revealed typical columnar crystalline structures. The XRD peaks in Fig. 1 show well-defined HCP peaks for Ru with d-spacing of 2.34Å for crystal planes with (100) Miller indices, 2.14 (002), 2.06 (101), 1.22 (103), 1.13 (201), and 1.07 (004), but poorly defined broad peaks for RuxNy with d-spacing near 2.65, 2.31, 2.21, 2.07, and 0.99. (All d-spacings are in angstroms, and the numbers in parentheses are Miller indices.) The crystal structure of RuxNy could not be determined. Ruthenium nitride is known to be very unstable, so thermodynamic data is not available. In our separate quantum mechanical calculation, the internal energy of the Ru + 1/2 N2 → RuN reaction is found to be slightly exothermic with ΔEf = -1.68kcal/mole, Ru3N2 endothermic at +1.02 kcal/mole, and Ru4N2 highly endothermic at +19.7kcal/mole, confirming the unstable nature of the nitride phase. This instability is further confirmed later in this study.
 Figure 2. a) SEM cross-section and b) RBS profile of a Ru film. |
Figures 2 and 3 compare SEM cross-sections and RBS profiles for Ru and RuxNy, respectively. There was no surprise for the Ru film. However, the SEM cross-section of RuxNy showed an unusual diffusion layer that extended sporadically beyond a depth of 200nm, although diffusion was unlikely at such a low deposition temperature (100°C). RBS could not detect this diffusion layer because of the thick nitride film on the surface, so analysis was carried out after stripping off the nitride film. The stripped surface was optically clean under 200× magnification. A ~5nm-thick layer of Ru was identifiable in a magnified RBS scale (Fig. 3). It is believed that unstable RuxNy molecules dissociate into atomic Ru at the wafer surface and the atomic Ru diffuses rapidly into the substrate initially. Further diffusion is prevented, however, as Ru atoms cluster themselves as dimers and trimers, and eventually form a barrier layer at the interface. It appears that RBS detects only the clustered thin barrier layer at the surface; therefore, the Ru concentration beyond this thin layer is considered to be <0.1 atomic % below the RBS resolution limit. RBS revealed that nitrogen incorporation in the RuxNy phase was <18 atomic %, confirming the instability of the nitride phase and difficulty in its formation. An important finding is that RuxNy cannot be used as a barrier film.
 Figure 3. a) SEM cross-section of a RuxNy film, and b) an RBS profile of Ru in the substrate after stripping off the surface nitride film. |
For the evaluation of barrier strength of Ru against Cu diffusion, film stacks were prepared as Si/SiO2/5nm Ru/Cu/TaN (Fig. 4). Ru was deposited at 100°C, while Cu was deposited at ambient temperature. Finally, TaN film was deposited on Cu to protect the Cu from oxidation during the thermal treatment. The result showed no indication of Cu diffusion across the 5nm thin Ru barrier after annealing at 700°C for 1 hr. After exposing the specimen to 750°C for 1 hr, sporadic patches of diffused area were observed as shown in the SEM cross-section. This is an impressive barrier strength, superior to TaN and TiN, according to the database.
 Figure 4. SEM cross-section of Si/SiO2/Ru/Cu/TaN stacks exposed to a) 700°C and b) 750°C, for one hour. |
Despite the excellent barrier strength, experimental results suggest that poor adhesion between the layers would be a potential problem for the application of Ru as a barrier material. From Fig. 4, it is apparent that debonding had occurred at the weakly bonded Ru-Cu and Cu-TaN interfaces during the cleavage of the wafer for SEM examination.
A tape-pull adhesion test also revealed film stress played an important role in film integrity. Ru deposits initially as a compressive film, but the stress trend changes from compressive to tensile with increasing film thickness. This reversal in stress trend, or buckling, appears to cause adhesion failure as marked by the square dots in Fig. 5a.
 Figure 5. Film stress as a function of a) Ru thickness and b) Cu thickness on 20nm Ru. The squares show the points that failed the tape-pull test. |
Next, Cu was deposited in various thicknesses on 20nm-Ru films at ambient temperature. Although the 20nm-Ru film stress was compressive, the Ru/Cu composite film stress became increasingly tensile with increasing Cu thickness. This “buckling effect” is found to be the most detrimental factor for adhesion failure for 20nm-Ru/Cu composite films. All Ru/Cu composite films failed tape test (Fig. 5b). The association between a buckling and adhesion failure has been documented consistently in many other experiments conducted by this research team.
In general, the PVD film nucleates as fine clusters at the substrate surface, and thereafter, crystals begin to grow in columnar structure with increasing sizes widening upward (Figs. 2 and 3). The initial fine-scale crystallites render generally high compressive stress, and the stress is reduced as grains begin to grow. Exceptions are metals with a low melting point, such as Al and Cu. Although not shown here, careful examination of stress data indicates that Al and Cu do deposit as compressive films but become tensile due to dynamic recovery (i.e., recrystallization and grain growth) during deposition, which can be confirmed by depositing the film at a very low temperature.
The origin of compressive stress can be attributed to the shot-peening effect that stretches the film by bombardment with sputtered atoms. The speed of sputtered atoms can be >10km/sec if 400eV Ar+ transfers a kinetic energy to a sputtered atom. The substrate temperature is substantially higher than ambient temperature due to plasma heating during deposition. Cu is known to anneal even at room temperature [6]; thus, it is likely that varying degrees of thermally driven recovery occur in Cu depending upon deposition condition, so it is desirable to use materials that render very low compressive, or better, low tensile stress, for barrier application.
Figure 6 illustrates Ru film stress as a function of deposition temperature for 100nm-thick films deposited at 1kW power. Film stress changes from compressive to tensile with increasing deposition temperature, showing a “sweet spot” near 300°C.
 Figure 6. Temperature dependence of Ru film stress. |
For ECP, there was no plating problem. Cu could be electroplated directly even on 10nm thin Ru films without any difficulty. However, adhesion failure occurred when ~1μm-thick Cu was plated on Ru. Delamination occurred at the Ru-SiO2 interface when a tape-pull test was applied, indicating that the bond strength of Ru-SiO2 is weaker than that of Ru-Cu. This failure is also attributable to ECP Cu recrystallization [6]. In general, metal-to-nonmetal bonding is weaker than metal-to-metal bonding. When the stress exceeds a certain critical value, deformation begins to occur by triggering dislocation motion. As dislocations move, successive debonding and rebonding occur along the slip plane. Complete rebonding occurs in a material with the same crystal structure and no lattice mismatch. At the interface with two dissimilar crystal structures, however, complete rebonding is not warranted, resulting in a partially bonded state and degradation of adhesion.
Electrical resistivity of a 10nm thin Ru film was ~25μΩ-cm, substantially higher than the bulk value of 7.13μΩ-cm. The inability to plate Cu on Ta or Ti results from persistent oxide that prevents adhesion, not high electrical resistivity. Cu can be plated on Ta but it does not stick well. Both Cu and Ru form oxide but in a less stable state because of relatively low oxygen affinity or large electronegativity that makes binding to oxygen less favorable compared with Ta and Ti. The electronegativities of O, Ru, Cu, and Ta are 3.5, 2.2, 1.9, and 1.5, respectively. Ru also has weak bond strength with oxygen: 43 (Ru-O), 96 (Cu-O), and 198 (Ta-O) kcal/mole. Thin copper oxide is known to dissolve easily with sulfuric acid. Ruthenium oxide appears to dissolve readily as well, since Ru is a more noble metal than Cu. The low electrical resistivity and low oxide stability are therefore in favor of direct Cu plating on Ru.
 |
The table lists the typical mean free-path length and resistivity data for interconnect metals. Cu has no advantage over Al as the feature size approaches <5nm. As films become thinner, electrons scatter more often at the surface and interface - an effect more pronounced for metals with a long mean free path that, by nature, is also responsible for low electrical resistivity in the bulk material. As film thickness decreases, surface/interface scattering becomes important, and effective resistivity therefore increases (Fig. 7). The resistivity of thin film varies with thickness and the mean free path length of the material as ρfilm ∝ ρbulk (1 + λ/t), where λ is the mean free-path length, and t is the film or wire thickness [7]. (For calculation methods, see [7-9]).
Figure 7 shows the effect of film thickness on electron drift range and the resistivity variation with film thickness for Ru deposited at 100°C and 400°C, and for Cu deposited at ambient temperature. Experiments show that the scattering effect begins at much thicker films than predicted, suggesting that electron scattering would be a primary factor in determining the line resistance as feature size diminishes. Overall, the resistivity of Ru is higher than the value found in literature: 7.13μΩ-cm. The excess resistivity is attributable to the enhanced electron scattering at finer columnar grain boundaries and high-density dislocations in the grains. This effect is more pronounced for films deposited at 100°C, as expected.
Conclusion
Ruthenium has shown excellent barrier strength against Cu diffusion up to 700°C and can be plated directly with Cu without a Cu seed layer, making Ru a possible diffusion barrier material for future Cu dual-damascene applications. However, Ru’s poor substrate adhesion demands that new application schemes be developed. Incompatible stress between Ru and Cu is identified as the main factor that jeopardizes film integrity. This disparity in stress can be minimized by the choice of deposition temperature. The Ru-SiO2 bond also is rather weak. Perhaps a glue layer should be placed between the substrate and Ru to improve adhesion. Such efforts are already underway in various sectors.
Acknowledgments
The authors wish to thank Werner Hort for x-ray data and Mike Pinter for providing the Cu electroplating demonstration.
References
- N. Iwamoto, N. Truong, E. Lee, “New Metal Layers for Integrated Circuit Manufacture: Experimental and Modeling Studies,” Thin Solid Films, 469-470, p. 431, 2004.
- I. Goswami, R. Laxman, “Transition Metals Show Promise as Copper Barriers,” Semiconductor International, 27(5), p. 49, 2004.
- P. Besser, A. Marathe, L. Zhao, M. Herrick, C. Capasso, et al., “Optimizing the Electromigration Performance of Copper Interconnects,” IEDM Tech. Dig., p. 119, 2000.
- K.Y. Lim, Y.S. Lee, Y.D. Chung, I.W. Lyo, C.N. Whang, et al., “Grain Boundary Diffusion of Cu in TiN Film by X-ray Photoelectron Spectroscopy,” Appl. Phys. A 70, p. 431, 2000.
- “Cross-cut tape test by ASTM D3359-95,” Annual Book of ASTM Standards, Vol. 06.01.
- Q.-T. Jiang, M.E. Thomas, J. Vac. Sci. Technol. B 19(3), p. 762, 2001.
- F.J. Blatt, Physics of Electronic Conduction in Solids, McGraw-Hill, New York, p. 236, 1968.
- S.O. Kasap, Principles of Electronic Materials and Devices, McGraw Hill, New York, Chapter 2, 2002.
- C. Kittel, Introduction to Solid State Physics, 6th ed., John Wiley & Sons Inc., New York, Chapter 6, 1986.
For more information, contact Eal Lee at Honeywell Electronic Materials, 1349 Moffett Park Dr., Sunnyvale, CA 94089; ph 408/962-2130, e-mail [email protected].
Nicole Truong, Nancy Iwamoto, Bob Prater, Janine Kardokus, Honeywell Electronic Materials, Sunnyvale, California