Filling high-AR structures using pulsed nucleation layer deposition
09/01/2002
The requirements for tungsten deposition at sub-100nm technology nodes place stringent demands on gap fill. At 90nm, the ITRS roadmap shows that a 12:1 aspect ratio (AR) contact will need to be filled for a DRAM contact, with 14:1 needed at 45nm [1]. For logic applications, the aspect ratios are smaller, but the requirements for seamless fill and CMP coring resistance to prevent copper contamination of the device are also demanding. Barrier layers proposed for these technologies include tetrakis-dimethylamido titanium (TDMAT)-based CVD TiN with plasma treatment to densify the films and reduce film impurities; titanium tetrachloride (TiCl4)-based CVD TiN; and various types of PVD TiN.
Each barrier film has unique challenges. The plasma-densified TDMAT films have a different composition on the field, sidewalls, and via bottom, creating a difficult surface for W growth initiation [2]. TiCl4-based TiN has more uniform properties, but has microcracking due to stress, and higher chlorine content in the barrier film [3]. PVD barrier films, while having uniform film properties in the field and via, have low step coverage along the feature sidewalls due to the directional nature of the PVD deposition [4].
Integrating different barriers on aggressive AR structures requires innovative solutions to W technology. A technique developed for W nucleation, pulsed nucleation layer (PNL), uses sequential pulses of reactants with purge cycles between the reactant cycles [5]. It provides conformal layers with minimal surface sensitivity to allow successful integration with various types of TiN, including topological variances in TDMAT CVD TiN composition.
 Figure 1. 45Å/ pulsed nucleation W, on plasma-densified TDMAT TiN, deposited using a C3 ALTUS system on a 16:1 via. |
null
CVD W: Initiation, nucleation, and fill
The CVD W deposition process consists of three steps: initiation, nucleation, and fill. The initiation process consists of wafer exposure to silane. It is theorized that SiH4 exposure enhances the barrier performance of the TiN by stuffing the grain boundaries with SiH4 molecules, or reacting with unreacted species in the barrier film [6]. The initiation step is followed by the nucleation process, which traditionally involves the simultaneous flow of silane (SiH4) and tungsten hexafluoride (WF6). The use of silane in nucleation prevents the reaction of WF6 with silicon or with the titanium contact layer and also promotes nucleation. The third step, bulk fill, is the hydrogen reduction of WF6. H2 is used for fill instead of SiH4 in order to give a conformal film with the lowest possible resistivity and particle levels. The reaction for the nucleation layer can be represented as follows [7]:
 |
null
This reaction occurs both in the gas phase (homogeneous), as well as on the surface. The nucleation process, as described above, shows adequate step coverage for larger features (>80% with opening diameters >0.13µm, 5:1 AR). The step coverage, however, starts to degrade for narrow and higher-aspect-ratio features. This is due to a combination of gas phase and depletion effects in the feature. The PNL process in which the same WF6 and SiH4 reactive gases are used for the nucleation process addresses these limitations.
The gases are introduced and purged sequentially (silane exposure, argon purge, WF6 exposure, Ar purge), such that the reaction occurs only on the surface similar to an atomic layer deposition (ALD) technique. While ALD processes have low deposition rates (<2.5Å/cycle) and are also generally slow, the PNL process is optimized for a higher deposition rate per cycle, making it a viable manufacturable process for future device generations [8]. PNL tends to have deposition rates on the order of 8-10Å/cycle.
Experimental
This work was performed in a 300mm ALTUS tungsten CVD reactor; the multistation sequential deposition process module consists of four active deposition stations where a portion of the W film gets deposited. Each deposition station consists of an independent showerhead, heater, and individual pumping and station isolation using an argon gas curtain. The W deposition is cleaned from the system using an NF3/Ar gas combination and an inductively coupled, remote plasma source. W is excluded from the wafer bevel using a patented minimum overlap exclusion ring (MOER) [9].
Tungsten nucleation layers with thicknesses between 45 and 100Å/were deposited using the pulsed nucleation technique with SiH4 and WF6 to serve as a nucleation layer for subsequent in situ fill deposition of CVD tungsten from WF6 and H2. In some cases, hydrogen/argon was used as a carrier gas. The effects of deposition temperature, gas flows and doses, and deposition timing sequences were systematically studied to characterize tungsten growth rates, surface roughness, impurities, resistivity, and step coverage.
Although other pressures were investigated, the work reported here was performed at 40torr, which is also the optimum pressure for CVD W bulk fill. This enabled the PNL process to be performed simultaneously on the first two stations of the reactor while the CVD fill was performed on the last two. The processes took place in the same chamber without the need for pressure cycling. CVD-TiN films deposited from both tetrakis-diethylamido titanium (TDEAT) and TDMAT were used as underlayers for the PNL deposition.
null
TEM was used to measure tungsten thickness and step coverage. Focused ion beam secondary electron microscopy was used to characterize the step coverage of plug fill after full deposition. Atomic force microscopy, x-ray diffraction (XRD), and secondary ion mass spectroscopy were used to measure roughness, microstructure, and impurity concentrations, respectively.
Results
The tungsten films formed using PNL exhibited high growth rates and step coverage of >95% in a 17:1 aspect ratio, 70nm. This deposition was done using the first two stations of the reactor for PNL, with stations 3 and 4 on the tool used for fill. The thickness of the nucleation layer is 45Å/for the 75nm CD, AR 17 structure, and the plug-fill thickness is 2000Å/. A close-up of nucleation on a 16:1 via shows nearly perfect step coverage and smooth, crack-free coverage on the bottom corners (Fig. 1). These results were obtained on plasma-densified TDMAT CVD TiN.
A comparison of CVD and PNL-based nucleation was performed in order to demonstrate the importance of nucleation layer quality and step coverage on overall fill (Fig. 2). The pulsed nucleation method gives very low surface sensitivity to the differences between the treated field region and via bottom, which are nearly stoichiometric TiN, and the sidewalls, which contain relatively high levels of carbon and oxygen and are prone to outgassing.
null
In contrast, conventional CVD nucleation is difficult on high-aspect-ratio structures with this type of barrier because the deposition rate on the sidewalls is lower, which leads to voiding. The voiding that can occur with conventional nucleation manifests itself as "coring" after CMP planarization. Coring is particularly problematic in copper applications as second-level copper can deposit in the cores and cause stress-related failure.
Conformal deposition and fill have also been demonstrated on other barriers, including TDEAT and TiCl4-based CVD TiN, as well as PVD TiN. Figure 3 shows a comparison of film roughness on treated and untreated TDMAT TiN. The untreated film is similar in composition to the untreated sidewalls of a via or trench. CVD nucleation, besides giving a rougher film overall, produces a very rough film on untreated CVD TDMAT TiN, which can help explain the larger amount of seaming and voiding observed with CVD nucleation. Bulk fill CVD tungsten exhibited ≥50% reduction in surface roughness when deposited on PNL tungsten compared to CVD nucleation.
PNL, being less surface sensitive, gives lower roughness on all types of TiN and is therefore not expected to create roughness inside the via and trench. Smoother growth results in a very smooth minimal seam, even in re-entrant structures as shown in Fig. 2b. Figures 3a and 3b (right-hand side of both) show a TEM image with the morphology of CVD and PNL films of similar thickness being compared. The TEM image shows much smoother films for PNL with smaller grains. This is confirmed by the XRD analysis of PNL films, which shows that these layers are microcrystalline with a grain size of 130Å/. The structure of the film is primarily alpha (a) W.
A CVD and PNL layer composition comparison is shown in Fig. 4. This Auger profile shows that, while PNL films have a higher level (1%) of silicon incorporation, the level of fluorine, in particular at the TiN interface, is lower. WF6 and SiH4 flow rates and dose times have been optimized so the deposition rate is saturated (Fig. 5). This is shown by the slope change of the dosing curves in Figs. 5b, 5d.
For a high-step coverage, dense film to be deposited with PNL, the film deposition should saturate, or stop, after a sufficient exposure of a given reactant. This ensures that a complete reaction occurs with a totally covered surface. WF6 flow and SiH4 and WF6 dose time all show clear saturation points. Saturated growth conditions result in uniform reactant concentrations in both deep contact features and in the field, which is expected to be the mechanism behind the improvement in step coverage observed with PNL nucleation.
PNL deposits several monolayers or up to 10Å//5-sec cycle and can deposit a 40Å/layer using 2 deposition stations in ‡20 sec. Pulsed nucleation combined with a multistation sequential deposition architecture, in which several stations can be used to perform deposition simultaneously, can give single module throughputs of >85wph steady state for a 2000Å/W film with a nucleation layer ≥75Å/thick.
As well as depositing conformal nucleation layers, the technology enhances film uniformity, reduces roughness, and increases reflectivity. This means more repeatable fill over the entire wafer (Fig. 6), even when etch variations give different via profiles. The averaging effect of performing some of the deposition at each station gives uniformity and repeatability of <2% on a 300mm wafer (Fig. 6).
null
Discussion
In the ALD process, silane alone is introduced into the reactor, during which time a single monolayer is adsorbed onto the surface. It is then purged or replaced by an inert gas. WF6 is introduced and reacts with the silane only at the wafer surface. The reaction described above is carried out, leaving a tungsten monolayer on the surface. In this way, gas phase reactions are minimized and films with near 100% step coverage may be deposited.
Typically, ALD is a low-pressure (∼1torr) process that gives low deposition rates of one monolayer or less/cycle. Using the PNL process, higher pressures (∼40torr) and temperatures are used to enhance the decomposition of silane to a limited number of silicon atoms and silane sub-hydrides (SiHx) on the reaction surface [10]. These species, with absorbed silane, provide additional sources of silicon for WF6 reduction, while maintaining self-limited growth. It is thus possible to deposit several layers of tungsten/cycle while still achieving conformal step coverage. Pulsed nucleation is high-throughput ALD, accomplished by appropriate process design.
null
null
Conclusion
The improvements to the basic tungsten nucleation process described in this article extend tungsten fill capability into the foreseeable future. The fill for extremely aggressive structures is enabled by pulsed nucleation technology. PNL technology compensates for limitations with the barrier film, and has film properties superior to CVD W. Systems in the field can be upgraded with this technology, which is standard on new 300mm systems.
 Figure 6. Repeatability of a)PNL and b)PNL with H2-based fill. |
null
References
- International Technology Roadmap for Semiconductors (ITRS), SIA, http://public.itrs.net/Files/ 2001ITRS/Home.htm.
- C-K. Wang et al., Proceedings, MRS '96 Spring meeting.
- S.R. Kurtz, R.G. Gordon, 140, 277-290, 1986.
- S.M. Rossnagel, Thin Solid Films, 263, 1-12, 1995.
- S-H. Lee et al., Proceedings of the Annual Meeting of the AMC 2001.
- S.L. Lantz, A.E. Bell, JVSTA, 12(4), 1032-1038, 1994.
- A. Hasper et al., Mat. Research Soc. Proc. VLSI V, 127-134, 1990.
- J.W. Elam et al., Thin Solid Films, pre-print, 2001.
- U.S. Patent # 5,238,499.
- N. Kobayashi et al., J. Appl. Phys., 73(9), 4637-4643.
Acknowledgments
Additional authors of this article are Kaihan Ashtiani and Luis Gonzalez. ALTUS is a registered trademark, and pulsed nucleation, PNL, and MOER are trademarks of Novellus Systems Inc.
For more information, contact: Gerard Chris D'Couto, product marketing, W Business Unit, Novellus Systems Inc., 3011 North First Street, San Jose, CA 95134; ph 408/570-2674, fax 408/545-3080, e-mail [email protected].
Thomas Omstead, Gerard Chris D'Couto, Sang-Hyeob Lee, Panya Wongsenakaum, Josh Collins, Karl Levy, Novellus Systems Inc., San Jose, California