Thin Solid Films 262 (lYY5) 46-51
Chemical additives for improved copper chemical vapour deposition processing John A.T.
Norman”, David A. Roberts”, Arthur K. Hochberg”, Paul Smithb, Gary John E. Parmeterb, Chris A. Apblettb, Thomas R. Omsteadc “Schumacher, 1969 Palomar Oaks Way, Carlsbad, CA 92009, USA bSandia National Laboratories, Albuquerque, NM 87185-5800, USA ‘CVC Products, Inc., 5.25 Lee Road, PO Box 1886, Rochester, NY 14603-1866,
Abstract Techniques for improved copper chemical vapour deposition (CVD) processing by the addition of trimethylvinylsilane (tmvs) and hexafluoroacetylacetone (Hhfac) during copper deposition from the volatile liquid precursor Cu(hfac)(tmvs) are described. The tmvs enables stable high vaporization rates of precursor by direct liquid injection and the Hhfac permits higher deposition rates of smoother copper films. The resistivity of the copper films averages approximately 1.8 p.0 cm as deposited. Combined together, these results mark an important advance toward a manufacturable copper CVD process. Keywords:
The development of a robust chemical vapour deposition (CVD) copper technology for thin metal film applications is now a high research priority for the semiconductor industry. Much of the recent copper CVD development work aimed at this goal has focused upon the use of the volatile liquid organometallit copper precursor Cu( hfac)( tmvs) (hfac = hexafluoroacetylacetonate; tmvs = trimethylvinylsilane) [ 11. This precursor yields pure copper films by the overall thermally driven disproportionation reaction 2 Cu( hfac)( tmvs) 4 Cue + Cu( hfac), + 2(tmvs)
Copper metallization utilizing this compound has already been successfully demonstrated using commercially available CVD reactors . However, to fully transition this CVD technique into a productionworthy technology will require the implementation of key process refinements. Amongst these are the development of a reliable precursor vapor delivery technique and suitable procedures for optimizing copper film morphology and growth rate. Here we report our progress toward reaching these goals by using the individual tmvs and Hhfac ligands during the CVD processing of Cu( hfac)( tmvs). Essentially the tmvs ligand is used to stabilize the precursor during its vaporization stage whereas the Hhfac ligand is used to Elsevier Science S.A. SSDI 0040-6090(94)05808-3
activate the precursor toward a higher metallization rate at the wafer surface. Neither of these techniques, as described below, impairs the quality of the resulting copper films. 1.1. Precursor stabilization and delivery by direct liquid injection (DLI)
Although the precursor Cu(hfac)(tmvs) is completely stable under ambient conditions of temperature and pressure, it reacts according to Eq. (1) at elevated temperatures and reduced pressure. Therefore, if it is used in the traditional “bubbler” mode for vapor delivery, some degree of disproportionation is likely to occur. This can lead to difficulties in controlling the mass of precursor vapor flowing from the bubbler. We have now shown that Cu(hfac)(tmvs) can be stabilized toward disproportionation by first dissolving it in a measured amount of tmvs and evaporating this solution by DLI. In this technique the bulk of the precursor is maintained under ambient conditions of pressure and temperature from which it is flowed into an evaporator zone maintained at a higher temperature and reduced pressure to induce vaporization. This approach for precursor delivery is especially appropriate for thermally sensitive compounds such as Cu(hfac)(tmvs). The addition of tmvs tends to suppress the metallization reaction (l), which thus helps
to provide a stable vapor delivery rate. A Cu(hfac)(tmvs)/ tmvs solution can also be safely vaporized at higher temperatures to deliver more vapor to the reactor and hence achieve a higher copper deposition rate. The use of the DLI technique to continuously the Cu(hfac) (tmvs)/tmvs solution is vaporize essential. Traditional bubbling is unsuitable, since there is a tendency to preferentially sweep tmvs from the Cu( hfac)( tmvs) / tmvs solution before evaporation of Cu(hfac)(tmvs) can occur. We have shown that this addition of tmvs as a DLI stabilizing agent for Cu(hfac)(tmvs) does not compromise the purity or resistivity of the resulting CVD copper films. Asdeposited films are shown to be 99.99% pure by secondary ion mass spectrometry (SIMS), with resistivities of approximately 1.85 p.fi cm.
1.2. Optimization of film growth To provide an acceptable wafer throughput, a commercially viable CVD copper process must have a proven capability of rapidly filling high aspect ratio submicron interconnect structures. It is critical that this can be achieved without adversely affecting the film properties of resistivity, purity, uniformity and conformality. In a technique that is complimentary to the tmvs addition, we found that the copper deposition rate from Cu(hfac)(tmvs) can be increased and the film uniformity and smoothness improved by the addition of Hhfac ligand vapor to the CVD reactor chamber during copper film growth. It is proposed that this phenomenon is due to the surface of the growing copper film being supplied with (hfac) species which facilitate the formation of Cu*+(hfac),. This drives Eq. (1) to the right-hand side and thus accelerates metallization. The observed growth rate enhancement is very sensitive to the type of metallic substrate used but does not appear to be a function of a simple metallic substrate cleaning of the surfaces by Hhfac etching of metal surface contaminants [3-51. No measurable increase in resistivity was discerned for the resulting copper films, which on average, after correcting for surface scattering, measured 1.85 @cm at 20°C as deposited.
2. Experimental 2.1.
Norman et al. I Thin Solid Films 262 (1995) 46-51
The precursor Cu( hfac)( tmvs) was synthesized and packaged in stainless steel vessels at Schumacher. The Hhfac and tmvs were dried by distilling from phosphorus pentoxide. Cu( hfac)( tmvs) / tmvs solutions
were prepared by mixing under anaerobic at room temperature. 2.2.
A Vactronics PECVD 2000 custom-built single-wafer reactor fitted with a loadlock as described previously  was used for the Hhfac addition experiments. A Watkins Johnson SELECT 7000P fitted with an ATM 300 DLI system and a prototype CVC single-wafer reactor configured with an MKS DLI system were used for the tmvs addition experiments. 2.3. Deposition conditions for Hhfac addition Reaction conditons were: wafer temperature 160170 “C, chamber pressure 500mTorr, He carrier gas 40 seem, bubbler source temperature 36°C. A separate ampoule of degassed Hhfac was used to introduce Hhfac vapor directly into the CVD chamber via a mass flow controller (MFC) while simultaneously delivering the copper precursor vapor through its normal injector ring. The quantity of Hhfac added was 15 vol.% of the copper precursor used in a run except where otherwise noted. Typically a section of onequarter wafer portions of different substrates would be metallized simultaneously to enable an accurate comparison of copper film morphologies and deposition rates for a given set of CVD reaction conditions. 2.4. Deposition conditions for tmvs addition Film purity and resistivity vs. film thickness studies utilized the SELECT 7000P system: wafer temperature 220 “C, chamber pressure 500 mTorr , nitrogen Cu(hfac)(tmvs)/tmvs carrier 50 seem , gas (50:50 vol.%) 0.5 ml min-‘, DLI vaporizer temperature 55 “C. Substrates used were silicon wafers with an overlayer of approximately 1200 8, of thermal Si3N,. Deposition rate and resistivity vs. temperature studies used the CVC system: wafer temperature 175215 “C, chamber pressure 500 mTorr, hydrogen carrier gas 200 seem for 0.3 ml min-’ DLI flow of Cu(hfac)(tmvs)/tmvs (80:20 vol. %) (DLI vaporizer temperature 65 “C) or hydrogen carrier gas 140 seem for 0.2 ml min-’ of pure Cu(hfac)( tmvs) (DLI vaporizer temperature 60 “C). Substrates used were titanium nitride. 2.5. Film morphology
Reflectivity measurements were accomplished using a Process Control Corp. Quik Test 1170 A apparatus in reflectance mode. SIMS measurements were performed by Charles Evans & Associates, Redwood
J.A. T. Norman et al. I Thin Solid Films 262 (1995) 46-51
City, CA using a CAMECA mass spectrometer.
2.6. Chemical mechanical polishing (CMP) Samples were polished on a Strasbaugh tool using an experimental slurry (XJFW7355) and an IClOOO/ Suba 4 pad.
3. Results and discussion 3.1. Addition of tmvs
Table 1 SIMS analysis (hfac)(tmvs)/tmvs
CVD copper (50:5Ovol.%)
Oxygen Nitrogen Carbon Hydrogen Chlorine Fluorine Iron Silicon Chromium Nickel
35 24 <6 <2 <2
In our early experiments using direct liquid injection for pure Cu(hfac)(tmvs), we found that if the vaporizer temperature was set higher than 55-60 “C, the flow of precursor would eventually become impeded owing to the formation of copper from premature thermal decomposition of the precursor. Since the metallization reaction (1) is in part driven by the release of tmvs, we reasoned that adding tmvs to the precursor prior to evaporation would suppress this undesired reaction from occurring in the evaporator. We found that this approach was very effective in promoting the free flow of precursor through the evaporator and provides a practical and controllable vaporization technique. Although in this study the added tmvs levels ranged from 50:50 to 80:20 vol.% Cu(hfac)(tmvs)/tmvs, we have recently found that a ratio as low as 9O:lO vol. % is still effective in maintaining a DLI delivery at a 65 “C evaporator temperature. Using the lowest level of added tmvs is important, since for a given total flow rate of pure precursor the added tmvs suppresses the metallization rate in the reaction chamber vs. that achievable for pure Cu( hfac)( tmvs)  under identical conditions. However, this deposition rate depression is readily overcome by the higher evaporation temperatures and precursor throughput achievable by the tmvs addition. 3.1.2. Effect upon film purity To determine any effect that the added tmvs might have upon the quality of the resulting copper films, we carefully measured their atomic composition at various thicknesses after CMP. A representative SIMS analysis is listed in parts per million (ppm) in Table 1. Oxygen and nitrogen represent the major impurities at approximately 35 and 24 ppm respectively. Iron, nickel and chromium were sought after, since these elements are present at high levels in the stainless steel precursor vessel, DLI system, reactor walls and in the vapor delivery lines leading to the reactor. However, the measured concentrations of these three metals in the copper films is negligible. This therefore indicates that
the Cu(hfac)( tmvs) precursor does not react with stainless steel in a manner that leads to an appreciable accumulation of these elements in the CVD copper films. Thus the films are 99.99% copper, with total estimated impurity concentrations being less than 80 ppm. Notably the values for carbon and silicon contamination are low, indicating that the added tmvs does not significantly degrade these CVD conditions. 3.1.3. Effect on film resistivity To be production worthy, a copper CVD process must deliver metal films that are as close as possible to bulk resistivity (1.67 l~.Klcm). To determine whether added tmvs adversely affected this key film property, we measured the resistivity of copper films grown by the DLI of 50:50 vol. % Cu( hfac)( tmvs) / tmvs solutions in the SELECT 7000 system after CMP, as shown in Fig. 1. For resistivity to be accurately established, it is essential that the film thickness d is accurately determined. Measuring film thicknesses by cross-sectional scanning electron microscopy (SEM) of CMP films
1.5 I o.2
Fig. 1. Resistivity Cu(hfac)(tmvs)/tmvs
vs. film thickness (50:50 vol.%).
Norman et al. I Thin Solid Films 262 (1995) 46-51
was found to be much more accurate than using stylus profilometry on unpolished films, especially at higher temperatures where the films tend to be rougher. The latter technique tends to overestimate the film thickness (and hence the resistivity) owing to surface roughness. In Fig. 1 the filled and open circles respectively represent the measured resistivities (p) and the corrected values for surface scattering according to
where p is the measured resistivity, A is the electron mean free path, d is the film thickness and pb is the resistivity of an equivalent bulk copper sample . A value of 400 A was assumed for A . Within the range shown, the resistivity of the films is approximately independent of film thickness, with an average value of 1.86 ~0 cm which reduces to 1.82 PSLcm by Eq. (2). 3.1.4. Effect on copper deposition rate With the 80:20 vol.% Cu(hfac)(tmvs)/tmvs blend we found that an evaporation temperature of 65 “C could be used with no blockage occurring. This permitted a 0.3 ml min- ’ DLI flow with complete evaporation, corresponding to a total flow of 0.24 ml min-’ of pure precursor. This improved the overall deposition rate from that achievable from pure Cu(hfac)(tmvs) evaporated at 60 “C, as can be seen from a comparison of Figs. 2 and 3. 3.2. Addition of Hhfac 3.2.1. Effect upon deposition rate Our initial impetus for this work stemmed from an empirical observation that batches of Cu( hfac)( tmvs) which had become contaminated with water gave
Fig. 3. Dependence Cu( hfac)( tmvs).
of copper film growth (80:2Ovol.%).
(“C) rate vs. temperature
unusually high copper deposition rates. By analysis we discovered that this contamination had led to the formation of Hhfac ligand by a hydrolysis reaction. (Similar hydrolysis of Cu’i(hfac)2 has been reported by Chiang et al. [lo].) Therefore we proceeded to deliberately add measured quantities of Hhfac vapor during metallizations using Cu( hfac)( tmvs) and measure its effect on the resulting copper films. Fig. 4 shows the effect of Hhfac addition on the rate of deposition of copper on Pt, TIN and Au-Cr alloy. Clearly the degree of rate enhancement depends strongly upon the substrate utilized. Only a small enhancement is seen for Pt, an approximate doubling for TiN and an increase from zero to 200 i$ min - ’ for Au-Cr (approximately 5% Cr). In general we found that the degree of accelerated deposition varied great-
Wafer Temperature Fig. 2. Dependence Cu(hfac)(tmvs)itmvs
(“C) rate vs. temperature
15 % Hhfac
Fig. 4. Effect of hexafluoroacetylacetone lization from Cu(hfac)(tmvs).
J.A. T. Norman et al. I Thin Solid Films 262 (199.5) 46-51
ly with the type of substrate used. One trend that emerged was that surfaces known to give the highest deposition rates (pure Pt and Au) [ll] tended to give the lowest relative rate increase. This substrate dependence would suggest that the Hhfac affects the rate of nucleation on these surfaces.
possibly linked together in the form of an intermediary activated complex. However, for simplicity we have separated the two.
3.2.2. Effect upon reflectivity The enhancements in reflectivity for films grown with Hhfac added are listed in Table 2. We believe that this increase in smoothness is due to an increase in the nucleation density of the growing copper film. In support of this we have shown that a greater density of copper nuclei is grown on a TiN substrate after a brief exposure to the Hhfac-accelerated copper CVD conditions described herein than under the same conditions without added Hhfac.
3.2.3. Effect upon resistivity All the copper films grown with added Hhfac were measured to have resistivities less than 2.0 $2 cm, measuring thicknesses by stylus profilometry. Measuring very smooth films and correcting for thin film effects gave resistivities of 3.8 @ cm. However, care must be taken to deliver the Hhfac in an uncontaminated condition to achieve near-bulk resistivities. In an initial set of experiments we discovered that if the MFC through which the Hhfac vapor is delivered was not thoroughly evacuated between runs, the result was copper films of resistivity greater than 2.0 @ cm. Our speculation is that residual Hhfac caused internal corrosion of the MFC to generate volatile contaminating species which were inadvertently added to the Hhfac vapor stream during subsequent runs. 3.2.4. Mechanism of the effect Although the precise mechanism by which the Hhfac-induced accelerated deposition occurs has yet to be determined, the following thesis is presented as a possible explanation of the effect. Simplistically, the mechanism for metallization by thermal disproportionation using Cu(hfac)(tmvs) can be thought of as occurring in the sequence of events shown in Eqs. (3)-(7). The electron exchange, Eq. (5), and the movement of the (hfac) ligands, Eq. (6), are quite Table 2 Enhanced Wavelength
of CVD copper Specular
films by Hhfac
2 Cu( hfac)( tmvs)( g) -+ 2 Cu( hfac)( tmvs)(s) 2 Cu(hfac)((tmvs)(s)+=
2 Cu(hfac)(s) + 2(tmvs)(g)
Cu”(hfac)(s) + Cu2’(hfac)(s)
Cu’(hfac)(s) + Cu*‘(hfac)(s)+Cu~~,
+ Cu2’(hfac),(s) (6)
where (s) denotes a metallic substrate and (g) denotes the gas phase. In Eq. (3) the precursor absorbs on to the metallic substrate. It has been shown that when molecules of Cu( hfac)( tmvs) are absorbed on to a metallic substrate under CVD conditions, the dissociation and subsequent desorption of (tmvs), Eq. (4), are extremely facile . Electron exchange between surface Cu( hfac) species, Eq. (5), should also be rapid. However, the formation of Cu2’(hfac),, Eq. (6), requires the migration of an (hfac) group to ultimately form Cu*+(hfac),. This is thought to be the ratedetermining step . Finally, the Cu2’(hfac)2 desorbs from the surface in Eq. (7), a process that should also be rapid. If these assumptions are correct, the addition of free Hhfac aids in the more facile formation of Cu2+(hfac), and should thus accelerate the growth rate of the copper film. A simplistic view is that by adding Hhfac, it acts as an acid to provide both protons and (hfac) anions to the surface of the growing copper film. Protons could facilitate the dissociation of (hfac) groups from Cu’(hfac) species to produce Hhfac, while the (hfac) anions could migrate to surface Cu2+ centers and drive the formation Cu2’(hfac),. Our data show that the introduction of Hhfac leads to the formation of smaller grained copper films and higher deposition rates compared with identical runs at the same temperature with no Hhfac added. Therefore the higher concentration of Hhfac in the reactor leads to increased opportunities
440 5.50 630
Cu on Au
Cu on Al
0.429 0.366 0.348
0.447 0.577 0.889
0.366 0.381 0.694
0.344 0.468 0.811
0.286 0.374 0.697
0.366 0.502 0.841
0.366 0.487 0.811
Cu on Pt
J.A. T. Norman et al.
t Thin Solid Films 26-3 (1YY.y) 46-51
for surface nucleation. This is also consistent with a lowered sensitivity of deposition rate to the type of substrate used and an enhancement in film uniformity, both of which are observed. Since p-diketone species are known to be effective vapor phase etchants for the removal of oxides and halides from metallic surfaces , we also investigated whether the enhanced deposition characteristics observed could be related to a “cleaning” of the substrate surfaces. In these experiments, substrates were first exposed to pure Hhfac vapor under the same CVD conditions and for the same duration employed for a copper metallization run. A copper CVD run was then made under identical conditions without added Hhfac. The latter runs showed no increase in copper deposition rate, film smoothness or uniformity, indicating that the cleaning effect was not contributive. We also evaluated the effect of adding the lesser fluorinated homologues of Hhfac, i.e. 2,4-pentanedione and l,l,l-trifluoro2,4-pentanedione, but found these to be much less effective than Hhfac. Similarly, adding either trifluoroacetone or trifluoroacetic acid (potentially, hydrolysis products of Hhfac) had little or no effect. Since trifluoroacetic acid is a strong acid (pK, = 0.23), the latter result indicates that the deposition acceleration observed for Hhfac is not solely due to its acidity. It has previously been reported that the addition of low levels of water vapor to Cu(hfac)(tmvs) vapor in CVD processing results in an increased deposition rate of copper [ 141, although the chemistry responsible for this acceleration has not yet been determined. Since the hydrolysis of Cu(hfac)(tmvs) in solution is known to yield Hhfac ligand [lo] and we have demonstrated here its effectiveness as a deposition rate enhancer, it is possible that Hhfac is released during the water addition process and plays an important role in the reported rate enhancement when using water vapor. The water addition process is also reported to reduce the delay time for the onset of nucleation on TiN, which is also consistent with in situ Hhfac generation.
metallization at vaporization temperatures so that the resulting precursor solution can be delivered and vaporized under controlled conditions by direct liquid injection. The thermal stability that the added tmvs imparts to the precursor enables it to be vaporized at higher temperatures. This permits a higher precursor input to the CVD chamber. with subsequently higher copper deposition rates. Both the tmvs and Hhfac addition techniques deliver copper films of approximately 1.8 ufl cm, as deposited, and together represent a significant advance toward a manufacturable CVD copper process.
4. Conclusions [lOI
We have shown that adding the ligands tmvs or Hhfac to the copper precursor Cu(hfac)(tmvs) during CVD respectively provides excellent control of precursor delivery and improved copper deposition characteristics. The Hhfac is added into the CVD chamber to accelerate copper film growth by a mechanism that we believe facilitates the formation of Cu”‘(hfac),, a byproduct of metallization. Film uniformity and smoothness are also enhanced. By contrast, tmvs is added to the precursor to suppress
J.A.T. Norman and B.A. Muratore, Volatile liquid precursors for the chemical vapor deposition of copper. US Pufent 5085 731. lYY2. A.V. Gelatos. S. Poon. R. Marsh. C.J. Mogab and M. Thompson. CVD of copper from a Cu ” precursor and water vapor and formation of TiN-encapsulated submicron copper interconnects by chemical-mechanical polishing, Proc. VLSl Symp.. Kyo~o. May lY93, pp. 123-124. H.-K. Kang, I. Asano. C. Ryu, S.S. Wong and J.A.T. Norman. Grain structure and electromigration properties of CVD Cu metallization, Proc. /EEE VMIC. Santa Clara. CA, 1993. IEEE, New York. J.C. Lvankovits. D.A. Bohling. J.A.T. Norman and D.A. Roberts, US Patent 5028 724, 1991. J.C. Ivankovits. D.A. Bohling. A. Lane and D.A. Roberts, Chemical vapor cleaning for the removal of metallic contamination from wafer surfaces using, 1 .l .I .5,5,5-hexafluoro-2,4pentanedione. Proc im. Symp. on Clerrnrng Technology in Semiconductor Device Manufacture, Vol. 9212. Electrochemical Society. Phoenix, AZ, 1992. pp. IOS-Ill. J.A.T. Norman, J.C. Ivankovits, D.A. Roberts and D.A. Bohling. US Parent 5OY4 701, 1092. J.A.T. Norman, D.A. Roberts and A.K. Hochberg, Surface and reactor effects on selective copper deposition from Cu(hfac)tmvs. MRS Symp. Proc.. 282, (lYY3). 347-352. G.A. Petersen. T.R. Omstead. P.M. Smith and M.F. Gonzales, ECS Proc.. 93-25 (1993) 225. S. Middleman and A.K. Hochberg, Process Engineering Analvsis in Semiconducror Device Fabrication. McGraw-Hill. New York. 1993. p. 675. C. Kittle. Introduction to Solid Stare Ph.ysws. Wiley. New York. 5th edn., 1976, p. 169. C.M. Chiang, T.M. Miller and L.H. Dubois, J. Phys. (‘hem.. Y7 (1993) 117x1. J.A.T. Norman, D.A. Roberts and A.K. Hochberg. New precursors for blanket and selective CVD copper. Proc. Electrochem. Sot., Y3-Y2 (1993) 221-230. L.H. Dubois, P.M. Jeffries and G.S. Girolami, Proc. Conf. on Advanced Metallization for ULSI Appliwtions, lYY2. MRS. Murray Hi/l, NJ, 1992, pp. 375-381. G.S. Girolami, P.M. Jeffries and L.H. Dubois. J. Am. Chem. Sot., 115 (1993) 1015-1024. A.V. Gelatos. R. Marsh. M. Kottke and C.J. Mogab, Appl. Phys. Len. 63 (20) (1993) 2842-2844.