MOCVD BaSrTiO3 for >1-Gbit DRAMs
07/01/1997
MOCVD BaSrTiO3 for =1-Gbit DRAMs
Steven M. Bilodeau, Ralph Carl, Peter Van Buskirk, Jack Ward, Advanced Technology Materials Inc., Danbury, Connecticut
Dynamic random access memory (DRAM) bit density has increased nearly fourfold every three years, from 4 kbit in 1972 to 64 Mbit now. Each successive generation has maintained the same stored charge within ever shrinking cell areas. Prior to the 4-Mbit generation, this was achieved by reducing the thickness of traditional SiO2 and Si3N4 oxide-nitride-oxide (ONO) capacitor dielectrics.
Subsequent DRAM generations have employed increasingly complex three-dimensional cells to maintain adequate capacitance in ever smaller projected areas. This complexity requires additional mask levels that ultimately impact yield and manufacturing cost. If this trend is followed to the 1-Gbit generation, ONO stacks will require a film area to wafer-surface area ratio of approximately 40:1 to maintain the required 35 fF storage capacity/cell.
An alternative approach is to use new materials with higher permittivities than conventional ONO dielectrics. Barium strontium titanate (BST) is of high interest for this application because of its unique combination of high dielectric constant, low dc leakage, low dispersion up to high frequencies, and stable operation at high temperatures.
Although the high dielectric constant properties of BST allow simpler storage node geometries, the capacitance/unit area of this material is not adequate for planar storage capacitors at 1-Gbit DRAM densities. Since three-dimensional geometries are still required, a conformal deposition process is essential. The drive to qualify high-permittivity BST materials for advanced DRAMs has intensified over the past four years, and most major DRAM manufacturers have established R&D programs to develop processes and integration methodologies for CVD BST thin films.
CVD process requirements
Several materials properties and process requirements determine the usefulness of BST capacitors for ULSI DRAMs. These include dielectric constant, capacitance, leakage current density, film uniformity, variations in electrical properties with respect to time, and wafer throughput (Table 1).
 |
First, the storage capacity must be high enough to significantly simplify the storage node geometry, since decreased process steps and increased yield are the main economic advantages of this approach. A value of 80 fF/?m2 allows 1-Gbit density [1] with a realistic storage node and an aspect ratio near 1. In practice, manufacturing issues such as the slope of the capacitor sidewall will reduce capacitor area, and BST nonconformality will reduce the storage density. Even higher storage densities are therefore desirable.
Next, the maximum tolerable leakage current is determined by the minimum charge needed at the end of a refresh cycle. The design therefore requires a trade-off between higher storage capacities at lower film thicknesses and lower leakage current at higher thicknesses.
Finally, films must be thin enough to avoid cross coupling. For 1-Gbit DRAMs, the spacing between adjacent storage nodes will be small enough for bridging between nodes to occur if the dielectric film is thicker than ~600 ?. If bridging occurs, the top electrode is no longer able to shield adjacent nodes and they become coupled [2, 3]. Thinner films also increase process throughput.
MOCVD of BST
One of the more challenging aspects of BST CVD processing is delivery of the precursors in a way that enables reproducible film stoichiometry. CVD processes require source reagents with sufficient volatility and stability to permit gas-phase transport to the substrate surface. For multicomponent materials, transport must be accomplished without gas-phase association, ligand exchange, premature reaction or premature decomposition, all of which can cause uncontrolled fluctuations in element delivery rates.
The best source reagents available for BST deposition are low vapor pressure solids that are temperature, air, and moisture-sensitive. These precursors can undergo undesirable reactions when mixed or premature decomposition when heated to the temperatures required to sustain usable vapor pressures.
The ATMI LDS-300B liquid delivery system was developed to provide point-of-use flash vaporization of solid precursors dissolved in solvents [4]. Highly accurate precursor delivery is achieved by the precise source mixing and metering provided by the liquid delivery technique.
Vapor transport of Group IIA b-diketonate precursors is confounded by the tendency of these sources to form multinuclear aggregates that have poor volatility and thermal stability. The multinuclear properties of [Ba(thd)2]4 and [Sr(thd)2]3 mean that source volatility can be enhanced by reducing molecularity. For this reason, barium and strontium bis(2,2,6,6-tetramethyl-3,5-heptanedionate) tetraglyme "adducts" (combinations, by means of van der Waals` forces, coordinate bonds, or covalent bonds, of two or more independently stable compounds) were developed for the CVD of BST films.
 |
Figure 1. Composition precision for a set of 600 runs employing a standard process deposited over a three-month period. The precision in Ti concentration is 0.36 At%.
Many titanium sources available for the CVD of multicomponent ferroelectric oxides react with Ba and Sr b-diketonate sources to yield nonvolatile alkoxides. To eliminate this problem, bis(isopropoxy)bis(2,2,6,6-tetramethyl-3,5-heptanedionate)titanium [Ti(i-OPr)2(thd)2] was developed since it does not undergo detrimental ligand exchange with the preferred Group IIA sources.
Nitrogen-donor ligands form stronger bonds with the Group IIA b-diketonates than with analogous oxygen-donor ligands (e.g. tetraglyme) due to increased basicity. Consequently, barium and strontium bis(2,2,6,6-tetramethyl-3,5-heptanedionate) pentamethyldiethylenetriamine (pmdeta) adducts demonstrate enhanced vaporization robustness as compared with the tetraglyme adducts.
The thermal transport of a solution of Ba(thd)2 (pmdeta) (0.235 M), Sr(thd)2(pmdeta) (0.107 M) and Ti(OiPr) 2(thd) 2 (0.258 M) was studied. All the runs used an ATMI liquid delivery system with a solution flow rate of 0.08 mliter/min and an Ar flow rate of 200 sccm. The pressure above and below the vaporizing element and the liquid pressure were monitored every minute for 10 hours. Reproducibility at a vaporizer temperature of 200?C was found to be 0.79 torr/100 mliter.
At 245?C, essentially no increase in the upstream pressure is observed for the pmdeta solution - indicating little or no decomposition of the precursors. For the solution containing identical concentrations of M(thd)2 tetraglyme adducts, the pressure rise corresponded to 2.65 torr/100 mliter of solution at 245?C. Clearly, the molecular structure and thermochemical properties of the precursor mixtures are responsible for the large performance variation observed in these vaporization experiments.
The chosen solid source reagents, when they are dissolved in solvent media, form liquid solutions. The chemical suite design minimizes intermolecular association between chemicals, while the liquid delivery technique significantly reduces the residence time of the thermally unstable source reagents under heated conditions. The combination of these effects enhances source volatility and reduces premature thermal decomposition.
This technique introduces a solvent into the MOCVD process, making it more chemically complicated. However, with careful choice of the precursors, solvent system, and CVD process conditions, a wide variety of precursors can be used and state-of-the-art BST film properties can be obtained.
Beginning in 1992, a Watkins-Johnson Select 7000 CVD reactor and liquid delivery system produced BST films for evaluation. A temperature controlled gas mixing manifold and showerhead type injector modified the reactor for compatibility with low vapor pressure precursors.
Varian Associates and ATMI subsequently designed an improved reactor for multicomponent ferroelectric oxide deposition. Delivered in 1994, the Varian reactor incorporates temperature controlled walls and showerhead injector, load lock wafer loading, and has provisions for an in-situ plasma clean. Both systems produced films for evaluation, with only minimal adjustments needed to transfer process from one system to the other.
The LDS-300B liquid delivery system mixes, meters, and transports precursor(s) at room temperature and high pressure to a heated zone, where they are flash vaporized and mixed with a carrier gas to produce a controlled temperature, low pressure vapor stream. The gas stream flows into the reactor`s mixing manifold, where it is mixed with oxidizer gases and passes through a showerhead injector into the deposition chamber.
Both the ratio of the concentrations of the metalorganic compounds in the vaporized liquid and the deposition conditions determine the final film composition. Chemical solution concentrations for this system are varied by mixing from three reservoirs. For mature applications, one or two solutions with pre-mixed metalorganic compounds in a predetermined ratio can be used.
 |
Wavelength dispersive XRF [5] data show that the film composition precision for this process is reasonable (Fig. 1). More than 900 mliter of solution were passed through the reactor over a three-month period to generate data. During this period, both a standard and nonstandard process produced films (Table 2), and some solution passed through the reactor without a substrate present to mimic additional deposition. Deposition rates of up to 150?/min were achieved.
The total simulated and actual deposition was equivalent to 1300 wafers of 300-? thick BST films, with no observed drift. The precision for Ba, Sr and Ti composition was 0.44, 0.39, and 0.36 at% respectively (1 s), close to the composition measurement precision of 0.31 at% Ba, 0.10 at% Sr, and 0.22 at% Ti. The measured thickness precision was 2.4% (1 s).
Film electrical properties
Capacitance/unit area and leakage current density determine the suitability of a BST film for DRAM storage capacitor applications. As processing techniques have improved, the minimum thickness consistent with acceptable leakage currents has decreased and the dielectric constant, at relevant film thicknesses, has increased. The result is a substantial increase in storage densities.
 |
Figure 2. Dielectric constant as a function of titanium content for 600-? thickness films with Ba/Sr = 70/30 deposited by MOCVD.
Dielectric properties of BST films [6, 7] are summarized in Figs. 2 to 5. The dielectric constant as a function of film composition for 600-? thick films (Fig. 2), reaches a maximum of 430 near the stoichiometric composition and drops off to 230 with a 3 at% increase in Ti content. The trend is qualitatively similar for films of all thicknesses. However, the dependence of macroscopic permittivity on composition is weaker for thinner films. This composition dependence is believed to be due to low dielectric constant second phases located at grain boundaries in BST with excess titanium [8].
The dielectric constant of thin BST doesn`t change within the range of process control. Also, since the fundamental dielectric constant is relatively invariant, small changes in composition can be used to optimize the leakage current.
Capacitance as a function of film thickness (Fig. 3) for the MOCVD process previously described and for sputtered films [9], shows that storage densities of >100fF/?m2 are achieved with MOCVD. Both processes produce acceptable leakage currents for DRAM applications (Fig. 4). Film properties as a function of thickness are quite similar for these very different processes (Figs. 4 and 5).
 |
Figure 3. C/A vs. thickness for MOCVD and sputtered planar films with Pt electrodes.
The storage density rises slower than expected with reducing film thickness (Fig. 3). This behavior can be explained by a thickness-dependent dielectric constant [10] or by a low dielectric constant interfacial layer at the electrode [11]. It is still unclear how much of the drop in permittivity with thickness is intrinsic, but the relatively rapid pace of properties improvements suggests that further increases in storage density are possible.
 |
Figure 4. Current density vs. film thickness for MOCVD and sputtered BST films with Pt electrodes. The current density for the sputtered films (solid green line) was measured at 1.65 V and was independent of Ba/Sr ratio (from[11]). The current density for MOCVD films was measured 10 sec after application of a 1.6-V step. For MOCVD grown films the leakage current of films with thickness >200 ? are suitable for DRAM applications.
For a 600-? thick BST film, higher temperature results in slightly higher current at all voltages, with the Shottky emission dominated region shifting to lower voltages (Fig. 5). 80?C is roughly the upper operating temperature for DRAM operation. The upper operating voltage for 600-? thick films is around 2.2 V at room temperature, but decreases to 2 V at 80?C. The expected operating voltages for 1-Gbit DRAMs are 1-1.5 V.
Film conformality
Since more than 80% of the area on a structure like this is on the sidewall, good sidewall thickness uniformity is more important than conformality (top/sidewall thickness). For isolated 2 ? 0.4 ?m lines, the conformality is approximately 75% with highly uniform sidewall thickness. For closely spaced 0.29 ? 0.38 ?m lines with a gap aspect ratio of 1.25, the conformality is 60% (Fig. 6) with a 10-15% decrease in sidewall thickness at the bottom compared to the top of gaps. This conformality is typical at temperatures significantly above the kinetically limited regime (Table 2). Difficulties in producing conformal top electrodes precluded sidewall electrical tests.
 |
Figure 5. Current density vs. voltage curves measured at room temperature and 80?C for a 600-? thick film with 1.3 at% excess Ti and Ba/Sr = 70/30 deposited by MOCVD. There is a 0.2-V shift in the high field behavior at 80?C relative to room temperature.
 |
Figure 6. SEM micrograph showing BST conformality on closely spaced 0.29 ? 0.38 ?m SiO2 lines with a gap aspect ratio of 1.25.
The highest conformality is achieved in a CVD process when precursor incorporation is surface-reaction rate limited. The incorporation efficiencies for the precursors are strongly dependent on the deposition temperature (Fig. 7).
 |
Figure 7. Incorporation efficiency vs. substrate temperature. A transition from mass transfer limited to surface reaction rate limited is indicated near 520?C for Ti incorporation and near 475?C for Ba and Sr incorporation.
Film thickness and composition data, measured by XRF, allows for the calculation of the number of incorporated atoms. Incorporation efficiency sharply drops between 450?C and 490?C for the Ba(thd)2 ? tetraglyme and Sr(thd)2? tetraglyme adducts; it drops between 490?C and 540?C for Ti(i-OPr)2(thd)2. This suggests a transition from mass transfer limited to surface reaction limited below 475?C. These results are consistent with reports of higher conformality below this temperature [12].
Conclusion
MOCVD BST films significantly simplify DRAM storage capacitor architectures for 1-Gbit and higher densities. Planar storage densities of over 100 fF/?m2 (oxide thickness equivalent <0.35 nm) have been achieved with leakage current of <10-7A/cm2 at ?1.6 V. These properties allow for processing variations and optimization of other parameters.
A CVD BST pilot-scale manufacturing reactor is currently available. The next step is to demonstrate that the geometrical simplicity enabled by this material can be translated into lower manufacturing costs.
Acknowledgments
This work was performed within the ARPA Ultra-Dense Capacitor Materials Processing Partnership that includes ATMI, IBM, Micron Semiconductor, NCSU, Texas Instruments, and Varian Associates as members. We would also like to acknowledge contributions from Angus Kingon, Stephen Strieffer, and Cem Basceri at NCSU who performed most of the electrical characterization and analysis; Dave Kotecki and the staff at IBM who contributed the cross-sectional SEMs; Jeff Roeder at ATMI who contributed XRF method development used to characterize film thickness and composition; and Tom Baum and Gautam Bhandari at ATMI who contributed vaporization transport studies performed with the M(thd)2.(pmdeta) precursors.
References
1. D. Kotecki, " High-K Dielectric Materials for DRAM Capacitors," Semiconductor International,, p. 109, Nov. 1996.
2. P. Lesaicherre et al., "SrTiO3 Thin Films by MOCVD for 1-Gbit DRAM Applications," Integrated Ferroelectrics, 8, p.
201, 1995.
3. P. Fazan, "Trends in the Development of ULSI DRAM Capacitors," Integrated Ferroelectrics, 4, p 247, 1994.
4. P. VanBuskirk, S. Bilodeau, J. Roeder, P. Kirlin, `Metalorganic Chemical Vapor Deposition of Complex Metal Oxide Thin Films by Liquid Source Chemical Vapor Deposition," Jpn. J. Appl. Phys., 35, p. 2520, 1996.
5. The methodology we have used for composition measurement and feedback to aid in determining the proper solution concentrations for a desired film concentration is summarized in: P. Vanbuskirk, J. Roeder and S. Bilodeau, "Manufacturing of Perovskite Thin Films Using Liquid Delivery MOCVD," Integrated Ferroelectrics 10, p. 9, 1995.
6. S. Streiffer, et al., "Dielectric Behavior of CVD (Ba,Sr)TiO3 Thin films on Pt/Si," Mat. Soc. Proc., Fall 1996.
7. C. Basceri, et al., "Resistance Degradation of CVD (Ba,Sr)TiO3 Thin Films for DRAMs and Integrated Decoupling Capacitors," Proceedings of the 8th International Symposium on Integrated Ferroelectric, April 1996 (to be published in Integrated Ferroelectrics).
8. R.F. Pinizzotto et al., `The Effect of Stoichiometry on the Microstructural and Electrical Properties of BST Thin Films,"presented at MRS Symposia Dec. 1994 (unpublished).
9. T. Kuroiwa, et al., "Dielectric Properties of Ba(x)Sr(1-x)TiO3 Thin Films Prepared by RF-Sputtering for DRAM Application," Jpn. J. Appl. Phys., 33, p 5187, 1994.
10. M. Yamamuka, T. Kawahara, A. Yuuki, K. Ono, "Reaction Mechanism and Electrical Properties of (Ba,Sr) TiO3 Films Prepared by Liquid Source Chemical Vapor Deposition", Jpn. J. Appl. Phys., V 35, p. 2531, 1996.
11. S. Yamamichi, J. Yabuta, T. Sakuma, Y. Miyasaka, "(Ba+Sr)/Ti Ratio Dependence of the Dielectric Properties for (Ba0.5Sr0.5)TiO3 Thin Films Prepared by Ion Beam Sputtering," Appl. Phys. Lett., 64, 13, p.1644, 1994.
12. T. Kawahara, et al., "Step Coverage and Electrical Properties of (Ba, Sr)TiO3 Films Prepared by Liquid Source Chemical Vapor Deposition," Jpn. J. Appl. Phys., 33, p. 5129, 1994.
STEVE BILODEAU received his PhD degree in materials engineering from Rensselaer Polytechnic Institute in 1986. He is a research engineer at Advanced Technology Materials Inc. Prior to ATMI, Bilodeau engaged in thin film research and development at Perkin Elmer and Hughes Danbury Optical Systems. At ATMI, he has developed deposition processes for a wide range of ferroelectric multicomponent oxides for electronic and optoelectronic applications. Bilodeau is the author of 25 technical papers and is a member of SPIE and MRS.
PETER VAN BUSKIRK recieved his BS degree in physics and his BA degree in history from the University of Bridgeport and the University of Redlands, respectively. He is director of technology for oxide thin films at ATMI. Since 1990, Van Buskirk has been responsible for developing MOCVD hardware and processes for depositing high permittivity and ferroelectric thin films. He is a co-inventor of the liquid delivery technology. Prior to ATMI, Van Buskirk was employed at Perkin Elmer in the the Optical Operations Directorate, where he was an engineering manager. He holds two patents.
RALPH CARL received his masters of engineering degree in mechanical engineering from Rennsselaer Polytechnic Institute in 1995 and is pursuing a master`s degree in electrical engineering from Polytechnic University. His work involves developing processes and hardware that will enable the large-scale manufacture of BST and other ferroelectric thin films in next-generation memory devices.
JACK WARD received his BA degree in economics from Indiana University. He is director of sales and marketing for the NovaMOS Division of ATMI. He has 20 years of experience in the semiconductor thin film materials industry. Advanced Technology Materials Inc., 7 Commerce Drive, Danbury, CT 06810-4131; ph 203/794-1100.