Diffusion barrier material for Cu metallization using ALD-WN_xC_y

07/01/2003

Overview

Currently, PVD-Ta-based thin films are the most commonly used materials as a diffusion barrier in the semiconductor industry. The maturity of the process makes it hard to replace or even change materials. Future requirements for ultra-thin diffusion barrier films, however, have called into question the universality of PVD technology [1]. Atomic layer deposition (ALD) is becoming attractive for interconnect applications because of its unmatched film conformality in deep submicron, high aspect ratio vias and trenches [2]. Ultra-thin ALD-WN_xC_y is capable of replacing current PVD-Ta-based materials and is a promising diffusion barrier solution for future technology nodes beyond 65nm.

At the 2002 IITC Conference, the authors reported a novel ALD-WN_xC_y process developed at ASM Microchemistry Ltd. for a diffusion barrier application using the Pulsar 2000 and 3000 ALCVD reactors [3]. The process uses WF₆, NH₃, and triethylboron (TEB) as precursors with deposition temperatures of 275–325°C. It was shown that the diffusion barrier properties of ultra-thin (<30Å) ALD-WN_xC_y films have surpassed their PVD-Ta based counterparts for dual damascene structures. The electromigration (EM) test results were particularly significant [4]. While the reaction mechanism is yet to be understood, it is believed that a unique surface reaction with highly reductive TEB in an ALD mode is the most crucial step:

Click here to enlarge image

null

WF₆ (g) + NH₃ (g) + B(C₂H₅)₃ (g) Æ WN_xC_y (s).

The resultant low resistivity ternary material contains ultra-low concentration impurities such as O (<0.5 at.%), F (<0.5 at.%) and B (<0.5 at.%); the H content (<4 at.%) is also low. Table 1 lists some of the material properties of these thin films. One of the important consequences of ALD growth is that a conformal WN_xC_y film with step coverage >90% can be achieved in deep trenches with an aspect ratio of 40:1 (Fig. 1). This performance is unmatched by current barrier technology.

Figure 1. SEM micrographs of ALD-WNxCy grown in deep trenches with an aspect ratio of 40:1 (130nm trench opening).Click here to enlarge image

null

Extensive thin film analysis with XRD, RBS, AFM, and HRTEM, etc., revealed that ALD-WN_xC_y is a dense material, close to an amorphous structure, with face-centered cubic WC_1-x or W₂N nanocrystallites. It is estimated that the density of ~10nm thin films is about 14g/cm³, close to that of bulk WC_x and WN_x at approximately 16g/cm³. A cross-section HRTEM shows the absence of a long range order of atomic arrangement in the film (Fig. 2). A path for Cu diffusion — even in the case of an ultra-thin film — is prevented because of the high density, amorphous structure; this is a highly desirable property for diffusion barrier applications. Furthermore, LEIS studies suggested that the full coverage of substrate with WN_xC_y thin film took place between 10–20 ALD cycles corresponding to a nominal thickness of 9–18Å.

The presence of tungsten carbide makes for the low resistivity of the ALD-WN_xC_y film. The material is highly stable at least up to 600°C, which is well within the 400°C thermal budget for Cu metallization. Unlike some of the Ta-based materials that are easily oxidized after deposition, no significant change in resistivity value has been observed with ALD-WN_xC_y samples stored at ambient conditions up to a year, indicating the stability against oxidation.

Figure 2. A high-resolution cross section TEM micrograph of ALD-WNxCy.Click here to enlarge image

null

Compatibility with dual damascene materials

ALD-WN_xC_y is a chemical process that relies on surface reactions, so issues regarding the interaction between process and under-layer materials must be addressed, especially in the case of barrier application in dual damascene structures that have great differences in their material properties [5]. These issues include: a) the barrier process should not alter or damage the underlying materials' properties, and vice versa; b) a conformal barrier film with similar growth rate on both sidewall and bottom of vias and trenches must be maintained despite the multiple surfaces in dual damascene structure; c) chemical penetration into pores is undesirable and a viable pre-treatment method is needed for porous low-k materials; and d) other properties such as adhesion and stress must fulfill the requirement for damascene processing.

ALD-WN_xC_y thin film growth on 200mm blank and patterned wafers coated with different dual damascene materials has been studied; Table 2 summarizes the compatibility of this film with these different materials. Surface reconditioning prior to barrier deposition ensures compatibility with all dual damascene materials tested, including porous low-k materials. The positive influence of the surface pre-treatment clears the way for integrating ALD-WN_xC_y into future damascene processes.

A large number of tests have shown that integration of current PVD-Cu seed, electrochemical deposition (ECD), and CMP steps with ALD-WN_xC_y in dual damascene processing is straightforward. No single failure has been observed due to the delamination of PVD-.Cu from ALD-WN_xC_y after CMP. The SEM studies showed that PVD-Cu has good wetting on ALD-WN_xC_y.

Click here to enlarge image

null

Figure 3. 1080k via chain test results of a set of samples with ~2.7nm ALD-WNxCy diffusion barrier deposited with different process parameters. A control wafer of PVD Ta with a nominal thickness of 25nm is shown for comparison.Click here to enlarge image

null

Figure 4. EM test results of 16 packaged samples with a 2.7nm ALD-WNxCy diffusion barrier. Half of them have not failed up to 1700 hrs. The reference PVD Ta samples all failed at ~100hrs.Click here to enlarge image

null

Electrical performance

The significance of the ALD-WN_xC_y as a diffusion barrier material is its superior electrical performances. Tests have been carried out with two-metal level dual damascene structures in SiO₂ having minimum feature sizes of 0.25ηm. These wafers were received from International Sematech (ISMT) without further treatment. Ultra-thin (<30Å) ALD-WN_xC_y barrier films were deposited in a stand-alone ALCVD reactor used for R&D. The wafers then were shipped back to complete the dual damascene process steps and electrical testing.

Figure 3 compares the 1080k via chain test results between the ALD-WN_xC_y (30 cycle, ~24–27Å) and PVD-Ta (nominal thickness 250Å) processes. Several different WN_xC_y process parameters were used but all resulted in high yields with similar via resistances — approximately half that of PVD-Ta — because of the ultra-thin films that were used.

Further EM tests have shown that samples with ultra-thin (<30Å) ALD-WN_xC_y diffusion barriers had lasted at least 10x longer when compared with the PVD-Ta samples with the same test conditions (Fig. 4). Demonstrating extraordinary EM properties, a large number of ALD-WN_xC_y samples did not fail even up to 1700 hrs of testing (tests were stopped). The ALD-WN_xC_y barrier films were deposited in the R&D tool and all experienced air break and shipment before the sample packaging. It is interesting to observe that the few samples that failed early were comparable to their PVD-Ta counterparts [4].

Conclusion

ALD tungsten nitride carbide (WN_xC_y) has shown excellent properties as a Cu diffusion barrier material. A high-density thin film with nanocrystalline WC_1-x and/or W₂N in an amorphous matrix is highly desirable for preventing Cu diffusion. Straightforward integration in a dual damascene structure is assured by low resistivity, a viable pre-treatment method, and a robust process on different substrate materials. Electrical characterization of ultra-thin (~27Å) ALD-WN_xC_y with a two metal level dual damascene structure in SiO₂ out-performed that of its standard PVD-Ta counterpart. EM tests have shown that a large number of samples having an ~27Å ALD-WN_xC_y diffusion barrier have not failed after 1700 hrs of testing compared to the standard PVD-Ta for which all samples failed at ~100 hrs.

Acknowledgments

The authors would like to thank colleagues at ASM for the development work. Cooperation partners at Top Analytica, International Sematech, Philips, LETI, IMEC and Seoul National University are acknowledged for sharing their measurement data. This work was partly performed in the ALAD1N+ project, a program in the MEDEA+ context. ALCVD, Pulsar and Aurora are registered or pending trademarks of ASM International. SiLK is a trademark of Dow Chemical Company. Black Diamond is a trademark of Applied Materials Inc.

Wei-Min Li, Marko Tuominen, Suvi Haukka, ASM Microchemistry Ltd., Espoo, Finland

Hessel Sprey, ASM Process Application Development, Leuven, Belgium,

Ivo J. Raaijmakers, ASM International, Bilthoven, The Netherlands

References

1. International Technology Roadmap for Semiconductors, 2001.

2. A.E. Braun, Semiconductor International, p. 52, October 2001.

3. W.-M. Li, et al., Proceedings of the IEEE 2002 International Interconnect Technology Conference, Burlingame CA, p. 191, 2002.

4. S. Smith, et al., IEEE 2003 International Interconnect Technology Conference, Burlingame CA, 2003.

5. S. Haukka, et al., Proceedings of the IEEE 2002 International Interconnect Technology Conference, Burlingame CA, p. 279, 2002

6. W. Besling, et al., Proceedings of the IEEE 2002 International Interconnect Technology Conference, p. 288, 2002.