An evaluation of semiconductor-grade polysilicon furnace fixtures
08/01/2001
James Boyle, Raanan Zehavi, Integrated Materials Inc., San Jose, California
overview
Data show that polysilicon furnace fixtures can significantly reduce post-LPCVD process particle counts and the need for periodic furnace fixture cleaning, and increase wafer deposition rate. Thermal and purity characteristics of this material also help to reduce defect densities and metallic contamination on silicon wafers. These data have been developed from long-term evaluations at many major semiconductor manufacturers.
Use of polysilicon for diffusion fixtures, such as boats, towers, and paddles, is neither new nor novel. In the early 1970s, polysilicon boats were produced by a CVD process that deposited polysilicon from silane, trichlorosilane, or silicon tetrachloride onto quartz or graphite tubing. After deposition, the quartz or graphite substrate was etched or burned away, leaving a silicon tube. The tube was then machined into flat-bottomed, 120° clamshells with wafer slots at the desired pitch.
 |
Though rudimentary by today's standards of purity, complexity, and precision, many early polysilicon fixtures are still in use in discrete diode and transistor production after 20+ years of service. Neither quartz nor silicon carbide fixtures have achieved even a fraction of this long-term reliability and durability.
Given longevity and performance factors, it would have seemed a natural evolution for diffusion equipment suppliers to offer polysilicon furnace fixtures as alternatives to quartz and silicon carbide. Two decades ago, however, polysilicon was costly, with uncertain supply and no reliable method for joining or fusing parts.
Furthermore, specifications for purity, particles, and uniformity associated with past processing of small-diameter wafers were minimal compared to today. These factors relegated production and development of polysilicon fixtures to a long-term hiatus.
 |
More recently, an increasing number of IC manufacturers are beginning to evaluate furnace components manufactured entirely from semiconductor-grade silicon that has purity levels not achievable with quartz or silicon carbide (Table 1). In fact, these silicon fixtures are as pure as, if not purer than, the wafers being processed (Table 2). Our data show that fabricated silicon towers, boats, injectors, pedestals, and other furnace components can be zero-particle generating, seldom need cleaning, and provide long-term dimensional integrity in high-temperature processes.
Process furnaces and application
To collect the needed data, at a cooperating IC manufacturer we fitted an SVG AVP-1 vertical furnace, being used for an LPCVD nitride process, with a 200mm four-rail, 160-slot silicon tower fabricated from virgin semiconductor-grade polysilicon. This furnace's alternating two-boat system provided an ideal platform for a direct one-to-one evaluation of silicon vs. quartz components.
In parallel, we also installed silicon towers in Kokusai furnaces used for LPCVD polysilicon, SVG and Tempress furnaces for extreme high-temperature anneal and drive-in (1365°C for 12 hr), ASM and TEL furnaces for oxide growth, and Steag RTP equipment.
A prerequisite of this evaluation was to assure that our silicon furnace fixtures met or exceeded original furnace OEM tolerances compared to equivalent quartz silicon carbide components. With these parameters in place, we set out to look at particles, cleaning, and contamination control (i.e., the relative purities of competing furnace fixture materials). For example, we knew that as the deposition film thickness on a furnace tower increases, thermal stress between films and furnace fixtures results in film fracturing and subsequent flaking. We felt that it would be straightforward to determine if our silicon tower was resistant to this effect.
 Figure 1. Comparison of thermal expansion for silicon, silicon carbide, and quartz. |
In addition, cleaning cycles for quartz and silicon carbide fixtures used in LPCVD nitride processing are well documented and determined by an individual IC manufacturer's allowable particle specification. When particles exceed specifications, fixtures require cleaning or outright replacement. As a rule, newly installed quartz towers have a projected use period of ~17-34 cycles before particles exceed a pre-set limit requiring a fixture's removal for cleaning. The subsequent cleaning is time consuming and results in diminished tool availability. If particle data proves that silicon towers do not require scheduled cleaning, this would markedly improve process results and productivity. The International Technology Roadmap for Semiconductors provides many details about future process contamination control requirements. For example, surface metal purities <7.5 x 109 atoms/cm2 are needed for 130nm linewidths.
Thus, fab operations requiring temperatures >900°C may be subject to diffusion of any impurities in a furnace into wafers.
Silicon carbide vs. silicon
For many processes today, fabs have gone to silicon carbide fixtures as an alternative to quartz because differences in the coefficients of thermal expansion (CTE) between quartz furnace fixtures and silicon can introduce slip in wafers [1]. The CTE of silicon carbide and silicon are close (Fig. 1). In addition, neither silicon nor silicon carbide will sag at temperatures >1100°C.
This same line of reasoning leads to the conclusion that silicon is even more advantageous than silicon carbide as a replacement for quartz in furnace fixtures. Obviously, the CTE of a silicon fixture is identical to that of a silicon wafer, greatly reducing the possibility of induced slip. In addition, silicon boats can be used up to 1375°C without sagging. With advanced fabrication techniques, silicon fixtures can be specifically designed to use thermal and other unique physical characteristics of silicon.
Polysilicon characteristics
Prior to the use of polysilicon as a source material for diffusion fixtures, the adhesion strengths of LPCVD films were not only process-dependent, but dependent on a substrate's relative CTE and surface roughness. This resulted in our development of a patent-pending surface-finish treatment that has proven a key factor in the virtual elimination of delamination, cracking, and resultant particle generation. The special surface finish has proven to provide permanent mechanical adherence between the thin-film and the silicon fixture substrate.
 Figure 2. Comparison of 0.26µm particle data after a 2500. |
A nominal 10µm-thick film of CVD silicon carbide deposited over a heavily contaminated underlying silicon carbide fixture does help solve metal contamination issues as long as the fixture does not have or develop pin holes, cracks, or chips. It is also possible to use 100% CVD silicon carbide fixtures, but these are very expensive and difficult to manufacture. By contrast, fixtures fabricated from semiconductor-grade silicon can endure operational-induced wear and tear, such as scratches, chips, cracks, etc., and still contribute no impurities to the process. In addition to its high purity advantage, silicon's other natural physical and mechanical characteristics, such as lack of porosity, contribute to this result. Conveniently, silicon parts are also repairable.
Another interesting silicon vs. silicon carbide furnace fixture comparison involves annealing wafers in hydrogen: To achieve a "perfect" wafer surface with very low numbers of crystal originated particles (COPs or pits), one process requires a wafer anneal in hydrogen at >1200°C [2]. Clearly, quartz cannot be used at these temperatures because it will sag. Silicon carbide is also challenged in this process because the hydrogen environment causes the SiC metallic impurities to become quite mobile and diffuse into the wafers. The alternative is to use virgin polysilicon furnace fixtures fabricated from material that is free of alkali and heavy metal contaminants, thereby eliminating the potential for their mobility.
Initial runs LPCVD nitride and poly
After some initial runs to test our polysilicon material and fixture design in the LPCVD nitride furnace, which was used to deposit a 2500
Somewhat serendipitously, an unfortunate incident with the furnace provided additional comparative data. During our evaluation period, the furnace suffered a pump shut down (at run 38) with the resultant back-streaming contaminating the process chamber while the silicon tower was in use. We simply removed and rinsed the tower with hot DI water (no etch), reinstalled it, and did a standard nitride pre-coat before reusing it. Data captured by process engineering confirmed that the tower continued performing at the reduced particle levels of previous runs (see Fig. 2). The evaluation furnace subsequently suffered three additional pump failures over the course of the year; after each, our experience was the same.
Overall, our data showed that the polysilicon fixtures did not contribute any additional particles during the yearlong, back-to-back cycle evaluations. Significantly, this meant that the silicon tower did not need any cleaning during the year, neither as a result of increased particle counts attributable to the fixtures nor for scheduled maintenance in anticipation of particle count increases.
Fixture longevity
We supposed that we could calculate the polysilicon fixture's functional life span by forecasting when the accumulation of film deposition would begin to clog wafer slots and challenge robotic loading [3]. We assumed that a 25% reduction in slot dimension would provide an acceptable safety margin. Thus, we projected an operational life span of 8-12 months. Initially, this seemed accurate, but then four months into our evaluations we observed two interesting developments: Wafer deposition rate began to improve and continued epitaxial nitride film growth on the silicon fixture stopped. We devoted particular attention to confirm that measurable nitride depositions were no longer accumulating on the tower.
Based on the stoppage of epitaxial nitride film growth, we are still investigating the silicon tower's ultimate life potential. Because the original evaluation towers are in their 18th month of continuous use, we can only forecast their potential as indefinite. Based on the previously stated comprehensive data, a majority of evaluation participants, which includes the previously mentioned furnace OEMs and ten major IC manufacturers, have approved polysilicon fixtures for production. Those that haven't are still in the midst of their evaluations.
Conclusion
Extensive evaluations of silicon towers for both vertical and horizontal furnaces have shown that these fixtures do not add particles over an extended period of use. This leads to an extended time without cleaning. Further, we found that when using a polysilicon furnace fixture, film deposition rates to wafer eventually increased while deposition to the fixture itself decreased. A polysilicon fixture's functional life span can be many times longer than either quartz or silicon carbide. Overall, we found that polysilicon fixtures improved tool availability, and reduced defect density, cleaning time, and expense, which, in turn, reduced environmental issues.
Annual COO is also less than for quartz and silicon carbide. An even better COO can be expected when significant process improvements, elimination of cleaning, life span, etc., are considered. Finally, polysilicon is four orders of magnitude or more purer than quartz and silicon carbide.
References
- J. A. Tomanovich, "LPCVD Components Trend Toward SiC," Solid State Technology, pp. 135-143, June 1997.
- I. Matsushita et al., "300mm Diameter Hydrogen Annealed Silicon Wafers," JECS, Vol. 144, No. 10, Oct. 1997.
- H. R. Huff, R.K. Goodall, "Silicon Wafer Thermal Processing: 300mm Issues," Future Fab, Issue #3, 1997.
James Boyle received his PhD in chemistry from the University of Buffalo. He handles silicon mechanical and electrical applications at Integrated Materials Inc., 737 E. Brokaw Rd., San Jose, CA 95112; ph 408/437-7591, fax 408/437-7549, e-mail [email protected].
Raanan Zehavi has degrees in electrical and computer engineering from the ORT Institute Jerusalem, Tel Aviv University, and Ben Guryon University. He is cofounder and CTO of Integrated Materials Inc.