CMP of flourinated silicon dioxide: Is it necessary and feasible?
02/01/1997
CMP of fluorinated silicon dioxide: Is it necessary and feasible?
Wei-Tsu Tseng, National Nano Device Laboratories, Hsinchu, Taiwan Yuan-Tsu Hsieh, National Chiao-Tung University, Hsinchu, Taiwan Chi-Fa Lin, Winbond Electronics Corp., Hsinchu, Taiwan
The advent of ultra-large scale integration (ULSI) has prompted an intensified search for low-permittivity (low-k) dielectric materials to reduce inter-metal capacitance and thus enhance device speeds for multilevel-interconnection systems. Chemical vapor deposited (CVD) fluorinated silicon oxides (F-SiO2) have attracted strong interest, because dielectric constants as low as 3.0 can be achieved in these oxides if the fluorine doping concentration is carefully adjusted. Furthermore, the precursors for fluorine doping and the process tools for the oxides are readily available, paving the way for smooth and economical process integration [1].
Multilevel interconnections in ICs have created gigantic stacked-material structures consisting of alternating adhesion layers, chemical-diffusion-barrier layers, conducting metals, and isolation dielectrics. Accurate lithographic pattern transfer for very fine features requires that the surface of each layer be planar to achieve successful etching and deposition.
Chemical mechanical polishing (CMP) has been recognized as the most promising planarization method for IC fabrication, and the use of CMP in ULSI processing is showing rapid growth [2]. Therefore, CMP compatibility must be evaluated before a new material (dielectric or metal) can be used in the stacked structures.
Effects of fluorine ions
The addition of extremely electronegative fluorine ions to the SiO2 strongly affects the properties of the oxide. Electrons from the O-Si bonds in the O-Si-F structures move more toward the fluorine side under the influence of the strong electro-negativity. That electron shift leads to a decrease in bond length and a weakening of the neighboring Si-O bonds [3]. As a consequence, the entire oxide structure becomes less polarizable, which lowers the dielectric constant. However, weakening the Si-O bonds also reduces the hardness and mechanical strength of the fluorinated oxides. Furthermore, the hydrophilic fluorine increases the moisture absorption of the oxides [4]. Subsequent formation of hydrofluoric acid (HF) after moisture absorption raises concerns about possible corrosion of adjacent metals and degradation of the oxide itself.
Table 1 summarizes the material properties of CVD-grown F-SiO2 with tetraethylorthosilicate (TEOS) + O2 + C2F6 deposited in a dual-frequency plasma process [5]. The dielectric constant and refractive index both decrease as more fluorine is added to the oxide. Fluorine also switches the intrinsic stress from medium-compressive (176 MPa) in undoped TEOS oxides, to low-tensile (35 MPa) in heavily doped (~8.9 at. %) F-SiO2. That shift in stress may have technical merit, as it provides (with adequate F-doping) a simple way to fine tune (or alleviate) the stress level in the IMDs with a concurrent decrease in the dielectric constant.
Hardness of the oxides (determined from nano-indentation techniques) declines with increasing fluorine content - presumably due to the Si-O bond weakening postulated previously. The reduction in hardness is also coupled with a decline in elastic modulus calculated from the elastic recovery parameter R [6]:
where
E = the elastic modulus
H = the nano-hardness
n = Poisson`s ratio
K = the indentation shape factor (= 23.897 for Berkovich indenters)
In addition, wet-etch rates of the oxides (tested with a 10:1 BOE solution) increase with added fluorine content. This behavior can be explained by a self-assisted etching mechanism caused by the formation of HF in the fluorinated oxides when they are immersed in the BOE solution.
CMP oxide-removal rates
The alterations in material characteristics caused by the addition of fluorine into the PECVD oxides would certainly affect CMP performance. Since the F-SiO2 films become mechanically softer and more flexible as more Si-F bonds are incorporated into the O-Si-O structure, the rigid particles in the CMP slurry (usually silica for oxide CMP) would abrade the oxide surface more efficiently - leading to a higher mechanical abrasion rate. Similarly, the high chemical reactivity of fluorine with moisture (as manifested by the wet etching results) and alkaline-based solutions (common slurry chemistry) implies that the chemical erosion rate would increase as well. The combination of those two factors points to an increased CMP removal rate for F-SiO2 (compared with that for undoped oxides).
CMP removal rates at a fixed pressure of 7 psi (48,260 Pa) for F-SiO2 of different fluorine concentration are shown in Fig. 1. For this test, the oxide samples were deposited on 150-mm unpatterned silicon wafers to a thickness of 1 mm. Removal rates using Cabot`s SC-1 slurry, with pH = 9 and pH = 10, were then recorded. The presence of fluorine in the oxides not only significantly enhances the removal rate, but also sensitizes the oxides to the slurry`s pH value. Higher removal rates with increasing fluorine content in the oxides are observed with the slurry of pH = 10. This effect may result from accelerated chemical reactions between the higher-pH (more alkaline) slurry and the F-SiO2 with higher fluorine content (increased acidity).
Figure 1. CMP removal rates for fluorinated silicon dioxide versus the Si-F concentration in the oxide. Slurries with pH = 9 and pH = 10 are both used for polishing, and the removal rate increases more rapidly with the higher pH slurry.
Figure 2. Variations of CMP removal rate with applied normal pressure for both undoped and fluorine-doped oxides; a) using slurry of pH = 10; and b) with slurry of pH = 9. Removal rates are more sensitive to pressure with the fluorinated oxide.
The variations in removal rate versus applied downward pressure for undoped and F-doped TEOS oxides are depicted in Figs. 2a and 2b for slurries with pH = 10 and pH = 9, respectively. In both cases, removal rates are found to be more sensitive to the variations in pressure for F-SiO2 than for undoped oxides, further supporting the hypothesis that the increased removal rate for fluorinated oxides results from their enhanced chemical reactivity along with their reduced hardness and elasticity.
Post-CMP effects on the oxides
As mentioned previously, moisture absorption could degrade the stability of F-SiO2 films. Since, during CMP operations, the wafers are in intimate contact with the abrasive aqueous slurry, post-CMP property changes resulting from moisture absorption may be a critical issue in determining the feasibility of CMP for F-SiO2.
Figure 3a shows post-CMP changes in refractive index (RI) as a function of the fluorine concentration in the oxides. A remarkable increase in RI can be seen after CMP for F-SiO2 with Si-F concentration greater than ~5.8 at. %. Clearly, this change in RI could be caused by accelerated chemical reactions and exacerbated moisture attack during CMP, but it is apparently not caused by mechanical effects (such as changes in surface roughness/scratching due to increasing pressure).
Figure 3. The post-CMP refractive index (RI) changes of F-SiO2: a) RI change vs. Si-F concentration in F- SiO2; b) RI change versus applied normal pressure for F-SiO2 with a fixed 8.9% concentration of Si-F.
The plot of changes in RI vs. pressure in Fig. 3b provides indirect evidence supporting our hypothesis. As can be seen, the RI of the F-SiO2, with an 8.9% concentration of Si-F after CMP, is virtually unaffected by variations in downward pressure. Mechanical influence on the RI after CMP is thus minimal or nonexistent. Note that the post-CMP increase in RI due to moisture permeation may also correspond to an increase in dielectric constant - which would introduce unwanted stray capacitance for circuit interconnections.
Figure 4. Linear relationships exist between the CMP removal rate and the hardness and elastic modulus of F- SiO2. Removal rate, hardness, and modulus are all normalized with respect to those parameters for undoped TEOS oxides.
When we are investigating the mechanical part of CMP, measurements of hardness and elastic modulus for the films to be polished provide a quick and easy way to predict the removal rate (Fig. 4). All quantities are normalized with respect to those of undoped TEOS (i.e. their ratios are used to index the axes). As the figure shows, a linear relationship exists between oxide hardness and CMP removal rate. In addition, the CMP removal rate is inversely proportional to the elastic modulus. The latter result is consistent with the prediction using a model [7], which can be simplified as:
where:
Ea = the elastic moduli for the abrasive
Ef = the elastic moduli for the film
P = the downward pressure
V = the relative rotation speed
C = a constant related to slurry chemistry and other material characteristics
Figure 4 clearly indicates that, with fixed chemical and mechanical parameters, the CMP removal rate increases significantly with increasing fluorine content in the TEOS oxides, due to the lowered hardness and elasticity in the F-SiO2 films.
From a mechanical perspective, the CMP removal rates of the fluorinated oxides increase as the stress in the deposited film is increased from medium compressive to low tensile. A recent study identified a similar trend [8]. For most oxide films, the magnitude of intrinsic stress lies in a range from 100 to about 300 MPa, which is far below their tensile strength (from 5000 to about 7000 MPa). Therefore, assuming a minimum of cracks or other stress-increasing anomalies, those stresses would contribute little, if any, to the breaking of the bonds. Instead, the formation of softer and more malleable surface layers by chemical erosion may be the key to the removal process. As the fluorine content increases in the oxides, the intensified water permeation and chemical activity directly enhance the CMP removal rate - which coincides with the changes in intrinsic stress.
To summarize, both mechanical abrasion and chemical erosion are intensified by fluorine doping (relative to undoped TEOS oxides) because of the reduced hardness and enhanced chemical reactivity, respectively.
Capping the dielectri
To resolve the dilemma, one has to engineer an integrated process that can achieve global planarization, while maintaining the reliability and low-k characteristics of the F-SiO2. One approach is to avoid CMP of the dielectric, while depositing a thin cap-layer to protect the F-SiO2 from moisture permeation. PECVD silicon-rich oxides (SRO) are often chosen for the added layer; they contain excess silicon atoms, which capture and stabilize the fluorine by forming Si-F bonds along the SRO/F-SiO2 interface. On top of the cap-layer, another thicker IMD layer is deposited. Then a typical CMP process can be applied to planarize the overall topography (Fig. 5). That approach results in partial loss of the low-k dielectric characteristic, however, due to the presence of the cap-layer. The approach also leads to relatively complex process steps and higher operating costs.
Figure 5. Schematic showing a simplified intermetal dielectric planarization scheme using a low-permittivity dielectric of F-SiO2 with a thin cap-layer of silicon-rich oxide. The cap-layer captures and stabilizes the fluorine, but it also tends to increase the dielectric constant.
A sequential deposition-etch-deposition process offers another alternative to direct CMP of F-SiO2, allowing better gap filling and local planarization. Nevertheless, this approach suffers from the same moisture-absorption problem as direct CMP (Table 2).
What is the answer?
There is no definitive answer to the question of whether CMP of F-SiO2 is necessary and feasible. While CMP has been recognized as the most promising planarization technique for IMDs, direct CMP of F-SiO2 may not be feasible for the reasons described above. While modern process tools, such as ECR-CVD, have been demonstrated to produce denser and more thermally stable F-SiO2, the extent to which fluorinated oxides can withstand slurry attack during CMP is still under investigation. Despite the high CMP removal rate (i.e. high throughput) of F-SiO2, one has to be cautious when designing a process based on this low-k dielectric. In general, tradeoffs must be made between the required dielectric characteristics and the reliability specifications for the materials.
Acknowledgment
The authors appreciate contributions to this article by Dr. Ming-Shih Tsai and Dr. Bau-Tong Dai of NDL for their valuable comments and suggestions throughout the course of the work. This article is based on a presentation at the MRS Spring meeting (San Francisco, CA, April, 1996) on "Advanced Metallization for Future ULSI."
References
1. R.K. Laxman, Semiconductor International, p. 71 (May 1995).
2. M.A. Fury, Solid State Technology, p. 47 (April 1995).
3. T. Homma, R. Yamaguchi, Y. Murao, Journal of The Electrochemical Society, 140, 687 (1993).
4. Y.-T. Hsieh, et al., Proceedings, Second International Dielectrics for VLSI/ULSI Multilevel Interconnection Conference (DUMIC), Santa Clara, CA, p. 287 (February 20-21, 1996).
5. W.-T. Tseng, C.-F. Lin, Y.-T. Hsieh, M.-S. Feng, Materials Research Society Spring Meeting, San Francisco, CA (April 8-12, 1996).
6. W.C. Oliver, G. M. Pharr, Journal of Materials Research, 7, 1564 (1992).
7. C.-W. Liu, et al., Journal of The Electrochemical Society, 143, 716 (1996).
8. C.-P. Chen, et al., Proceedings, First International Chemical-Mechanical Polish for VLSI/ULSI Multilevel Interconnection Conference (CMP-MIC), Santa Clara, CA, p. 82 (February 22-23, 1996).
WEI-TSU TSENG received his BS degree in metallurgy and materials science from Cheng-Kung University in Taiwan and his MS and PhD degrees in materials science and engineering from the University of Texas at Austin. He joined NDL in 1994 and is currently an associate research scientist. He has published more than 25 technical papers. His main research interests include phase stability, materials reliability, metallization, CMP, and photolithography. National Nano Device Laboratory, 1001-1 Ta Hsueh Rd., Hsinchu 30050, Taiwan, Republic of China; fax 886/3-5713403, e-mail [email protected].
YUAN-TSU HSIEH received his BS degree in nuclear engineering from Tsing-Hua University and his MS degree in materials science and engineering from Chiao-Tung University. His research interests include dielectric materials and CMP. He is now a thin-film process engineer at Winbond Electronics in Taiwan.
CHI-FA LIN received his BS and MS degrees in materials science and engineering from Feng-Chia University and Chiao-Tung University, respectively. He is currently a senior engineer in Advanced Process Technology Development division at Winbond Electronics in Taiwan. His work is mostly concerned with CVD and CMP processes.