growing Improved Silicon Crystals for VLSI/ULSI
11/01/1996
Cover Article
Growing improved silicon crystals for VLSI/ULSI applications
Kyong-Min Kim, MEMC Electronic Materials Inc., St. Peters, Missouri
Producers of blank silicon wafers routinely meet existing specifications - in terms of resistivity, oxygen level and uniformity, dislocations, and stacking faults. But achieving the higher quality levels and larger wafer diameters required for emerging VLSI and ULSI device applications will be much more difficult. To grow silicon crystals of the required quality, we must first solve increasingly complex problems involving point defects, oxygen content, secondary crystalline defects, melt hydrodynamics, and heat flow in the crystal. Each of those phenomena needs better understanding and control. It is essential to understand the cause-and-effect relationships between crystal growth conditions and grown-in defects - along with their subsequent impact on device performance.
Czochralski (CZ) silicon-crystal technology has shown continuous progress over the years in terms of crystal size, metallic-impurity levels, uniformity of dopant and oxygen distribution, and the levels of crystalline defects. Silicon wafers with diameters up to 200 mm are now routinely used, while production of 300-mm wafers is just beginning. The quality of silicon wafers produced today is impressive, but it will improve further.
It is essential to understand the interrelationships of crystal growth conditions, grown-in defects, and their impact on device performance in order to develop the high-quality crystals required for the ULSI era. In this article, we shall examine CZ silicon-crystal growth technology and materials, and also look at the role of computer simulation in process design.
 |
Figure 1. Schematic of a Czochralski silicon crystal puller. The size of the crucible must increase substantially for systems to grow 300-mm crystals. Typical one-meter-long crystals currently require about 100 kg of silicon melt, but that will increase to 150-300 kg for the larger-diameter crystals.
Today`s 200-mm silicon wafers
Silicon wafer requirements for VLSI/ULSI applications from 1995 to 2010 have been assessed by the Semiconductor Industry Association (SIA)[1]. Most of the requirements for 200-mm wafers are already being met (see silicon "Starting Materials Requirements" assessed by the SIA on the opposite page). With regard to mechanical properties, the site flatness for 200-mm shaped silicon wafers is 0.23 ?m for a site area of 22 ? 22 mm2. To enable high-resolution lithography, particles are controlled at <0.2 cm-2, with a size of =0.12 ?m. The front surface micro-roughness is 0.2 nm.
Heavy metallic impurities in the bulk of the crystal are far below the one-ppba level - for example, with total Fe of 5 ? 109 at/cm3. Individual surface levels of key metallic impurities fall in a range from 5 ? 109 to 1 ? 1010 at/cm3.
Recombination and generation lifetime of the minority carriers in wafers are typically 500 and 2000 ?s, respectively. The oxygen concentrations, at levels from 24 to 33 ppma (old ASTM, 1979), are controlled to within ?2 ppma, with a radial uniformity better than 5%. Dislocations are completely eliminated in large-diameter crystals. The density of oxidation-induced stacking faults (OISF), which occur when the wafers are oxidized at about 1000?C during device processing, is typically <2/cm2.
 |
Crystal growth and oxygen content
A schematic diagram for a CZ silicon-crystal puller is shown in Fig. 1. A quartz or fused-silica (SiO2 ) crucible contains molten silicon. Today`s 200-mm diameter crystals (typically one meter in length) are grown from a 100-kg silicon melt contained in a 22-in. crucible. Within a few years, 300-mm crystals are expected to be produced from 150-300 kg silicon melts.
Oxygen enters the silicon melt via dissolution of the crucible. Most dissolved oxygen (over 95%) escapes from the free melt surface as silicon monoxide, and only a small amount (< 5%) is incorporated into the growing crystal [2].
In IC processing, oxygen has been employed beneficially for internal gettering (IG) of metallic impurities [3], thus improving the minority-carrier lifetime and device yield. Control of the oxygen concentration (as well as its axial and radial uniformity) is of major importance in silicon-crystal growth and device processing.Though device processing has become less contaminating, IG will continue to be important along with external-gettering (EG) techniques. Oxygen precipitation is determined by both the oxygen concentration and the pre-existing crystal defects and precursors prior to the IC processing. Defects which cause oxygen precipitation include point defects and their secondary defects (with or without oxygen) formed during CZ crystal growth and post-growth processing.
The oxygen level of silicon wafers currently being processed in the semiconductor industry is in a range from 24-33 ppma. Oxygen concentration in the silicon crystal is determined by the oxygen distribution in the melt, at the crystal-melt interface, and in the bulk melt. Those sources are affected, in turn, by the dissolution rate, transfer in the melt (via convection and diffusion), and the ratio (m) of the free melt surface to the melt in contact with the crucible wall. Dissolution of the SiO2 crucible into the silicon melt is influenced by crucible wall temperature, crucible rotation rate, and by fluid-flow instabilities in the melt.
Fluid flow is turbulent in large CZ growth systems, such as 100 kg with a 22-in. (or larger) crucible. As crystal growth proceeds, the ratio m increases, and the axial oxygen level tends to decrease unless corrective steps are taken. The radial oxygen gradient, or uniformity, depends on the oxygen distribution at the crystal-to-melt interface boundary layer. That distribution, in turn, is influenced by the fluid flow in the melt near the interface. Crystal rotation, which drives a centripetal flow under the crystal-to-melt interface, helps to achieve radial uniformity of the oxygen and the dopant.
Many techniques have been developed over the years to control the axial and radial gradients of oxygen in CZ silicon. Several approaches apply a magnetic field during the basic CZ process [4-6]. Various magnetic CZ (MCZ) techniques have been developed and are being used effectively to control the oxygen in 200-mm silicon.
Efforts to control oxygen segregation have led to remarkable production capabilities in the silicon-wafer industry. With today`s 200-mm silicon, typical specifications of ?2 ppma can be routinely met, with a radial gradient <5%. Lower ranges (?1.5 and even ?1.0 ppma) are becoming possible, though usually at great difficulty and/or cost. With 300-400 mm crystals, it is expected that similar specifications will eventually be met.
As device processing continues to introduce less contamination, the need for internal gettering will diminish, so there is a trend toward lower oxygen specifications. Another incentive for this trend is to prevent any presence of oxygen precipitation in the active device region - a few microns from the surface.
Available thermal processes can establish a precipitate-free surface region for active devices. Also, denuded zone (DZ) separation from bulk silicon that contains precipitates or bulk micro-defects (BMDs) is effective in gettering metallic impurities from the active devices. The higher the oxygen level, however, the greater is the possibility of detrimental oxygen precipitates occurring in the active device region.
Control of the oxygen level, uniformity, and precipitation behavior in CZ silicon has been and will continue to be a challenge in the face of ever increasing crystal diameters - from 200 mm today to 300-400 mm in the near future. Optimization of crystal-growth parameters, using such approaches as the MCZ technique [4-6], has been largely effective in meeting the challenge so far; but, with the larger crystal-growth systems (with 150-600 kg melts in 24-36 in. crucibles) required for larger wafers, the problems will persist.
Recently, there has been significant progress in computer simulation of industry-scale CZ crystal-growth systems. Simulation will assist us by computing thermal conditions, fluid flow, and oxygen transfer/segregation. Computer simulation may provide further insight and guidance for oxygen control, and it may provide some insights into point-defect dynamics and the formation of secondary defects.
Point defects in silicon wafers
Point defects in silicon consist of vacancies and interstitials. Both point defects and secondary defects become increasingly important as we march toward the era of Gbit devices.
Concentrations and distributions of point defects are affected by the crystal growth conditions, especially the ratio, v/G, of the growth rate (v) and the thermal gradient (G) at the crystal-melt growth interface. Depending on the v/G ratio, either vacancies or interstitials will predominate. Recently, a critical value, Ccrit, of 1.3 ? 10-3 cm2 min-1 K-1 was derived empirically [7, 8]. Depending on the growth conditions of the crystal segment, i.e., on whether v/G is greater or less than Ccrit, the crystal segment is vacancy or interstitial dominant. CZ crystal-growth processes that provide vacancy dominance in the crystal are usually preferred and pursued because oxygen precipitation and extended defect formation are more easily controlled during subsequent wafer processing.
Point defects in float-zone (FZ) silicon (which has oxygen levels about 2-3 orders of magnitude lower than CZ silicon) form some distinct secondary defects. Interstitials form A-defects [9, 10], and vacancies form D-defects [10, 11], as a result of agglomeration due to super-saturation in situ during the crystal growth. Point defects (and especially their secondary defects) in CZ silicon are somewhat different from those in FZ silicon because of the higher oxygen concentration in the CZ silicon.
In vacancy-rich CZ-silicon sections, flow-pattern defects (FPDs) often occur [12]. Those defects are similar to the D-type defects in FZ silicon caused by vacancy clustering. On the other hand, A-type defects in FZ silicon are not observed in CZ silicon. However, at the boundary between the vacancy and interstitial-rich region in a CZ-silicon wafer, an annular segment of high OISF density (commonly called an OISF ring or edge swirl) can occur. The section inside the OISF ring is vacancy rich, so one finds a relatively high density of FPDs, as well as crystal-originated points (COPs) or light point defects (LPDs) [13]. COPs are believed to have similar physical origins to FPDs. Enhanced precipitation of oxygen provides a further indication of excess vacancies inside the OISF ring. The outside area of the wafer, a region of interstitial dominance, has relatively few defects. There is a low density of dislocation loops - in an area where it is somewhat more difficult to nucleate oxygen precipitation [14].
Performance of MOSFET devices, especially the gate oxide integrity (GOI) and the refresh times of high-density DRAMs, is degraded by grown-in secondary defects or micro-defects. As discussed previously, these secondary defects evolve from agglomeration of the vacancies, interstitials, and/or oxygen, which become super-saturated in situ during the crystal-growth batch process. Secondary defects or micro-defects that degrade the GOI include FPDs, COPs, and other grown-in defects. In general, the grown-in defects are called light-scattering tomography defects (LSTDs), after the detection technique [14-16]. LSTDs are affected by the in-situ dwell time during crystal growth at about 1050?C. Formation of LSTDs is believed to be facilitated by the vacancies at that high temperature. The oxygen super-saturation alone cannot provide a sufficient driving force for homogeneous nucleation at high temperatures. Figures 2 and 3 show the dependence of the GOI yield as a function of the FPD density [12] and the LSTD distribution [16], respectively.
 |
Figure 2. Dependence of gate-oxide integrity (GOI) yield on flow-pattern defect (FPD) density. FPDs occur in vacancy-rich CZ silicon and can severely impact device yield at high defect densities.
 |
Figure 3. Dependence of yield on light-scattering tomography defects (LSTDs). Charts show the axial distributions of a) GOI yield, b) LSTD density, and c) averaged square root of the scattering intensity. In a), solid blocks indicate total GOI yield, while the half-shaded blocks indicate GOI yield inside the oxygen-induced stacking fault ring, where the silicon is vacancy rich.
Refresh times for high-density DRAMs are, however, degraded by oxide (SiO2) polyhedral precipitates [17, 18]. The problem is especially severe for wafers cut from the tail-end section of the crystal, due to anomalous oxygen precipitation (AOP) [16, 19]. The dwell time of the tail-end section at high temperatures below the freezing point of silicon (1412?C) is much shorter than for the bulk of the crystal, due to fast cooling after completion of the CZ-crystal growth process.
Crystal-growth conditions thus play a substantial role in the generation of crystalline defects. Those defects degrade device performance - in terms of the GOI and the refresh time in MOSFET devices, for example. Efforts are being made to minimize or eliminate defects associated with the crystal-growth process. A better understanding of the relationships between the crystal-growth conditions and defects in the material, and their impact on the device performance, will help development and optimization of the crystal growth processes and equipment and so provide high-quality silicon for VLSI/ULSI device applications.
Today, we understand the causes of point defects and the secondary defects sufficiently well to allow optimal control of heat transfer during CZ pulling, and thus grow high-quality silicon crystals.
Computer simulation of crystal growth
Computer simulation of the CZ silicon-crystal growth system guides optimization of the growth parameters, thus allowing oxygen and defect control in large-diameter silicon crystals. In recent years, comprehensive models have been developed for heat transfer, melt convection, and oxygen/dopant transfer in large CZ silicon-growth systems [20]. Software simulates fluid flow in the CZ silicon melt, temperature distribution (both in the melt and crystal), and the shape of the crystal-to-melt interface.
Fluid-flow instability and oxygen transfer/segregation depend substantially on the size of the CZ crystal-growth system. The Grashof number (Gr), which indicates fluid-flow instability, is proportional to the temperature difference on the melt surface times R3, where R is the radius of the crucible. For a relatively large CZ system (150 kg of silicon in a 24-in. crucible), the estimated value of Gr is about 3 ? 1010. Note that the melt flow is turbulent with a Gr of 3 ? 107. Thus, a turbulent flow should prevail in the large CZ melts required for growing 300-mm crystals.
An existing simulation of melt hydrodynamics, based on the k-e turbulence model, may provide guidance for achieving axial and radial uniformity of oxygen and dopant in large CZ systems [20]. Fluid flow in the melt has several physical origins, including buoyancy surface-tension gradient on the free melt surface, and forced motion from rotation of the crucible and the crystal.
Based on a laminar-flow model [2, 6, 21], fluid flow in the CZ melt is governed by the Navier-Stokes equation with Boussinesq approximation and a magneto-motive body force when a magnetic field (B0) is applied:
- v = 0
Dv/Dt = -(1
) -(p + rgz) + ag(T- Tz)iz + j ? B0
+ u -2u
where
v = the fluid velocity
p = the pressure
g = the acceleration due to gravity
j = the induced current in the melt due to flow
iz = the unit vector upward
r = the density
a = the volumetric expansion coefficient
u = the kinematic viscosity of the melt
From the conservation of energy and mass of the oxygen/dopant, the melt temperature T and the concentration of a dilute solute c are governed by:
DT/Dt= k-2T
Dc/DT= D-2c
where
k = the thermal diffusivity of the melt
D = diffusion coefficient of the solute
Turbulence is the most complicated of fluid motions. It differs from laminar flow in that streamlines fluctuate randomly over small distances with high temporal frequencies. It is thus inherently three-dimensional, and is characterized by a self-reinforcing cascade of energy from larger to smaller flow structures through a continuous spectrum. The k-e turbulence model [22, 23] has been applied to the fluid flow in CZ silicon crystal growth, and it has proved very useful.
In the turbulence model, two new parameters are introduced to the Navier-Stokes equations: i.e., the kinetic energy k in the fluctuating velocity of the turbulence, and the dissipation rate of the kinetic energy e associated with turbulent viscosity ?T. The partial differential equations, along with the boundary conditions and the physical properties of the CZ silicon crystal growth, are solved numerically with the finite-element or finite-difference method. The finite-element method (FEM) is the more general and versatile. Among other advantages, the method allows complicated realistic geometries (such as the crucible shape and the crystal-to-melt interface) to be entered and solved.
A two-dimensional, steady-state turbulence simulation was solved with FEM - using an efficient Newton method with quadratic convergence [20]. Close agreement was found between the simulated and experimental measurements in the thermal fields and for the crystal-melt interface shape. Also, there was a qualitative agreement for the radial distribution of oxygen in the crystal.
Though significant progress has been made in the computer simulation of CZ silicon crystal growth, further developments are needed before simulation can be effective in developing crystal-growth processes for 300-400 mm silicon wafers.
Point-defect dynamics
Unfortunately, computer simulation of point-defect dynamics is still in an early stage of development [24, 25], partly because defect simulation requires that other simulations be done first. Results of the computer simulations of thermal and fluid flow are applied as inputs for the defect simulation. The thermal field in the crystal (especially at the crystal-to-melt interface) and the interface shape are especially important.
When simulating point-defect dynamics during the crystal-growth process, based on a continuum-balance model, each species may be considered to diffuse, convect, and react within the evolving silicon lattice. Within this framework, the balance equation for a species, such as an interstitial point defect, can be written as:
?c1/?t = {Fickian diffusion} + {thermo-migration} + {bulk convection} - {loss by bulk recombination} - {loss by precipitation}
The bulk-convection term accounts for the motion of the crystal through the imposed thermal and thermoelastic stress field. The expressions for the Fickian diffusion and the thermo-migration are known [24]. Recombination of vacancies and interstitials is usually considered in terms of the departure from the equilibrium concentrations [10]. The dynamics of the formation and growth of precipitates can be divided into two steps: the nucleation of a precipitate, and its growth by diffusion of point defects and oxygen and their incorporation into the cluster or agglomerate.
Simulation of defect dynamics [24, 25] may provide a quantitative understanding of the spatial distribution of point defects in situ during CZ crystal growth, as well as their recombination and formation of secondary defects (with or without participation of the oxygen).
Crystal growth for 300-mm wafers
For cost-effectiveness, large (300-400 mm) wafers will soon be required for ULSI devices.The CZ pullers for emerging 300-mm crystals will be scaled-up versions of the present 200-mm pullers, but with some modifications and improvements. The 300-mm puller is expected to accommodate much more silicon (over 300 kg) in a 32-in.-dia. (or larger) crucible. Control of point and secondary defects in 300-mm crystal growth will pose an important challenge.
In addition to the crystal-growing issues examined in this article, various steps must be taken to achieve and maintain the required purity and quality during the final stages in preparing and handling 300-mm silicon wafers for semiconductor fabrication. After the crystal has been grown, wafer shaping [26], polishing [27], and cleaning [28] are other crucial steps that must be controlled and optimized to provide high-quality silicon.
Remarkable progress in CZ silicon-crystal size, oxygen uniformity, and purity has led to high minority-carrier lifetimes and low OISF densities. In addition, we now have an improved understanding and better control of point defects and secondary defects, and their impacts on device performance. Today`s 200-mm wafers meet all the starting-material requirements for subsequent device applications. As we approach the ULSI era, and as wafer sizes increase to 300 mm or larger, the linkage between the crystal-growth conditions, defects in silicon material, and their impact on the device performance needs to be continually identified. Then we shall be able to develop and optimize crystal-growth processes, wafer preparation, and device-fabrication processes to achieve superior performance, reliability, and yields for semiconductor chips.
Acknowledgment
The author would like to thank Harold Korb for his careful reading, suggestions, and contributions. This article is based on an invited talk at the "4th International Conference on VLSI and CAD (ICVC `95)," October 16-18, 1995, in Seoul, Korea.
References
1. The National Technology Roadmap for Semiconductors, edited by the Semiconductor Industry Association, 1994.
2. K. M. Kim and W. E. Langlois, J. Electrochem. Soc., Vol. 138, p. 1851, 1991.
3. T. Y. Tan, E. E. Gradner and W. K. Tice, Appl. Phys. Lett. 30, p. 561, 1977.
4. S. Takasu, S. Takahasi, M. Ohwa, O. Suzuki and T. Higuchi, Semiconductor Silicon 1990, p. 45, eds., H. R. Huff, K. Barraclough and J. Chikawa, Electrochem. Soc., Pennington, NJ, 1990.
5. H. Hirata, K. Hoshikawa, Jpn. J. Appl. Phys., 21L, p. 545, 1987.
6. W. E. Langlois, K. M. Kim, J. S. Walker, J. Crystal Growth 126, p. 352, 1993.
7. W. Von Ammon, E. Dornberger, H. Oelkrug, H. Weidner, J. Crystal Growth 151, p. 273, 1995.
8. J. Holzer et al., private communication.
9. P. M. Petroff, A. J. R. de Kock, J. Crystal Growth, Vol. 30, p. 117, 1975.
10. V. V. Voronkov, J. Crystal Growth, Vol. 59, p. 625, 1982.
11. P. J. Roksnoer, M. M. B. Van den Boom, J. Crystal Growth, Vol. 53, p. 563, 1981.
12. H. Yamagishi, I. Fusegawa, N. Fujimaki, M. Katayama, Semicond. Sci. Technol., Vol. 7, p. 135, 1992.
13. J. Ryuta, E. Morita, T. Tanaka, Y. Shimanuki, Jpn. J. Appl. Phys., Vol. 29, p. L1947, 1990.
14. M. Sano, M. Hourai, S. Sumita, T. Shigematsu, Crystalline Defects and Contamination: Their Impact and Control in Device Manufacturing, p. 3, eds., B. Kolbesen, P. Stahlhoffer, C. Claeys, F. Tardif, ECS, Pennington, NJ, 1993.
15. K. Nakajima, J. Furukawa, H. Furuya, T. Shingyouji, Semiconductor Silicon 1994, p. 156, eds., H.R. Huff, W. Bergholz, K. Sumino, ECS, Pennington, NJ, 1994.
16. M. Hourai, T. Nagashima, E. Kajita, S. Miki, S. Sumita, M. Sano, T. Shigematsu, ibid, p. 156.
17. K. Marsden, S. Sadamitsu, M. Hourai, S. Sumita, T. Shigematsu, Semiconductor Silicon, ibid, p. 684, 1994.
18. S. S. Kim, W. Wijaranakula, J. Electrochem. Soc., Vol. 142, p. 553, 1995.
19. H. Harada, T. Abe, J. Chikawa, Semiconductor Silicon 1986, p. 76, eds., H. R. Huff, T. Abe, B. O. Kolbesen, The Electrochem. Soc., Pennington, NJ, 1986.
20. T. A. Kinney, D. E. Bornside, R. A. Brown, K. M. Kim, J. Crystal Growth, Vol. 126, p. 413, 1993.
21. H. Hirata, K. Hoshikawa, J. Crystal Growth, Vol. 125, p. 181, 1992.
22. B. E. Launder, Turbulence Topics in Applied Physics, Vol. 12, ed., P. Bradshaw, p. 231, Springer, Berlin, 1978.
23. C. G. Speciale, Ann. Rev. Fluid Mech., Vol. 23, p. 107, 1991.
24. T. Sinno, R. A. Brown, Semiconductor Silicon,, p. 625, 1994.
25. Z. K. Jiang, R. A. Brown, Chem. Eng. Sci., pp. 3001-3014, 1994.
26. E. Steffen, J. Schandl, J. Jung, H. Lundt, J. Crystal Growth, Vol. 126, p. 207, 1993.
27. I. J. Malik, S. Pirooz, L. W. Shive, A. J. Davenport, C. M. Vitus, J. Electrochem. Soc., Vol. 140, p. L75, 1993.
28. W. Kern, J. Electrochem. Soc., Vol. 137, p. 1887, 1990.
KYONG-MIN (GEORGE) KIM received his BS degree in chemical engineering from Seoul National University in Korea, and Diplom-Chemiker and PhD degrees in physical chemistry from Braunschweig Technical University in Germany. He is a senior member of the technical staff at MEMC, where he develops advanced CZ/MCZ crystal growth techniques for high quality silicon. Prior to joining MEMC, Dr.
Kim was with IBM, where he was involved in the R&D of CZ/MCZ silicon crystal growth and materials for VLSI applications and computer simulation and the process modeling for advanced bipolar and BiCMOS devices. He was awarded 15 US patents (five pending), and received four Invention Achievement awards from IBM and a Skylab Achievement award from NASA. Dr. Kim is a member of the Electrochemical Society, the American Physical Society, and the American Association for Crystal Growth. MEMC Electronic Materials, Inc., 501 Pearl Drive, P.O. Box 8, St. Peters, MO 63376; ph 314/279-5943, fax 314/279-5157, e-mail [email protected].