The case is made for delivering liquid precursors from a central delivery system to the epi/dep tool as a vapor of precisely-controlled composition. 

By EGBERT WOELK, Ph.D., Dow Electronic Materials, North Andover, MA, USA and ROGER LOO, Ph.D., imec, Leuven, Belgium 

The epi and deposition processes for silicon-based semiconductor devices have used gaseous and liquid precursors. Gaseous precursors are compounds whose vapor pressure at room temperature is higher than 1500 torr (2000 mbar), which is sufficient to drive a mass flow controller (MFC). Using only one MFC, gaseous precursors can conveniently be metered to the process. Silane and dichlorosilane (DCS) have been used with that method. The industry has also used Trichloro silane (TCS) that boils at around 33°C and can be directly metered to a low pressure epi process using an appropriate MFC. For the epi of SiGe, germane, which is a gas, has been used.

Tetraethylorthosilicate (TEOS) has long been used for the deposition of SiO2 and has mostly been delivered using direct liquid injection (DLI). DLI meters the flow of the liquid precursor to a flash evaporator and provides good control, but flash evaporation requires high temperatures and care must be taken that the precursor compound does not break up prematurely. This can be a challenge for precursors that work at lower deposition temperatures.

More recently, trisilane (Si3H8) has been used for low temperature Si epi and deposition. The delivery of trisilane to the process uses the carrier-gas-assisted delivery method. In the most common implementation, it employs an on-board evaporation ampoule dedicated to one reactor. The same setup has been used for III-V compound semiconductor and LED epi with good success. Driven by cost pressure, however, the LED epi industry is moving from dedicated onboard ampoules to a central delivery system for high-volume precursors like trimethylgallium (TMGa). One part of the cost reduction simply comes from the economies of scale. Another aspect comes from the elimination of excessive hardware, such as thermal baths and pressure controllers, and their maintenance. Most importantly, a substantial part of the cost reduction comes from yield increases due to improved process control. The same central delivery system can be used for trisilane and other liquid CVD precursors for silicon-based CVD for similar cost reduction.

Carrier-gas-assisted precursor delivery

Liquid compounds with an RT vapor pressure between 1 and 400 mbar require carrier-gas-assisted delivery. Many liquid compounds within that vapor pressure range are excellent precursors for CVD and epi processes. For such compounds, the difference between the vapor pressure and the process pressure is too small to drive an MFC for straight metering. Adding a carrier gas increases the pressure to between approximately 760 and 1500 torr (1000 and 2000 mbar). The selection of a good delivery pressure depends primarily on the desired concentration.

The carrier-gas-assisted delivery method has long been used for trimethylgallium (TMGa) and trimethylaluminium (TMAl) for the growth of GaAs and GaN. For the growth of GaAlN and GaInN for LEDs, the composition ratio of the two group III precursors is extremely critical for the performance of the final product. Therefore, the precision of the evaporation and the metering has always been a concern.

FIGURE 1a shows the setup for a straight gas delivery and FIGURE 1b shows the setup for a carrier-gas-assisted delivery. The design shown in Figure 1b requires no modification of the epi/dep tool in order to accept a normally liquid precursor. From an epi/ dep tool perspective, the design shown in Figure 1b behaves just like the straight gas delivery of Figure 1a. As such, it allows the use of the gas mixture from one delivery system at several points of use, i.e. the output of the delivery system can be subdivided. In Figure 1b the precursor vapor is made on demand. While the output (mol flux of precursor per time) is theoretically unlimited, there are practical limits that restrict the output to approximately 20 standard liters per minute (slm). The main limitation is the dynamic range of the metering valve: the best units have a dynamic range of 1 in 104, which means that they can reliably control a flow between 0.002 and 20 slm. This is important for the mol flux precision at smaller flows, i.e. when only one or two tools draw precursor.

On-board ampoules and central delivery system

There are several designs of carrier-gas-assisted delivery sources. The traditional design meters carrier gas into the ampoule rather than the mixture into the process chamber. Such a delivery system is dedicated to one reactor because the mass flow is metered upstream of the evaporation vessel and the associated MFC is controlled by the epi/dep tool. The ampoule serves two functions: (1) as the transport vessel and (2) as an evaporation device. For cost reasons, the ampoule should be of simple design. This means that trade-offs for the evaporation performance have to be made. The trade-offs result in line-to-line delivery rate variations and a noticeable change of delivery rate over the life of the ampoule. For some products, such changes require run-to-run recipe adjustments. In some cases the on-board ampoule is connected to a central dispense unit that transfers liquid precursor into the on-board ampoule. The result is a complex system that is still subject to delivery rate shifts requiring recipe adjustments.

A new central delivery system design is shown in Figure 1b. The task-optimized evaporator is fitted with temperature, pressure and level sensors that hold the precursor output variation at less than +/-0.4% by use of special stability algorithms. The evaporator is a permanently-installed part of the central delivery system. It is fed from a supply canister and features two precision thermometers inside the precursor liquid and gas distribution baffles and strainers for entrained droplets. Once calibrated, the system delivers a precisely known rate to a number of epi/dep reactors in the fab.

FIGURE 2 shows the output concentration of two calibrated central delivery units under various loads [1]. The curve that is alternately dotted and solid represents the signal of the binary gas sensor, which was alternately connected to one or the other unit. The other curves represent the output of the two units in standard liters per minute. The results show that proper calibration of the temperature and pressure sensors results in error of the delivery of less than +/- 0.4%. This precision cannot be achieved with ordinary on-board ampoules.

Recently, the application of the VAPORSTATION Central Delivery system has been expanded to deliver SnCl4 to a new process for the deposition of GeSn. It was fitted to a gas delivery line that was available on a mainstream silicon epi tool.

GeSn epi using a SnCl4 as new precursor

There has been increasing interest in GeSn and SiGeSn as alternative Group IV semiconductor material for electrical and optical device applications. The continuing expansion of traditional silicon with Sn and Ge offers additional design options for band gap and stress engineering. Over the past years, stress engineering using Ge made a major contribution to the improvement in Si-CMOS device performance. More recently the use of GeSn as a stressor for Ge-CMOS and relaxed GeSn as a virtual substrate, which is used to create tensile strain in a Ge epitaxial film, have been considered. The creation of tensile strain in an epitaxial Ge film is expected to result in germanium with a direct band gap [5] for photonic devices. Epitaxial Ge1-xSnx itself has also been considered as a promising candidate material for lasers and photodetectors. It has been predicted that, for sufficiently high Sn content, relaxed Ge1-xSnx turns into a direct band gap semiconductor [6,7]. Recent work of imec and its partners describe the active functionality based on the heterogeneous integration of strained GeSn/Ge on a Si platform providing photo-detection in the mid-infrared [8].

Due to the poor solubility of Sn in the Ge matrix of less than 1%, the epitaxial growth of (Si)GeSn is very challenging. Low solubility demands out-of- equilibrium growth conditions and, from epitaxial growth point of view, extremely low growth temperatures. Until recently, GeSn was grown by molecular beam epitaxy — a technique that is not suited for mass production. More recently, deuterated stannane, SnD4 has been used as Sn precursor for a CVD process, but the practical application is questionable due to the instability of SnD4.

To eliminate the problems posed by SnD4, imec chose to investigate stannic chloride SnCl4 , a stable, benign, abundant and commercially-available liquid Sn compound. Currently though, most of the CVD reactors for SiGeSn epi are not designed to use liquid precursor sources. In order to facilitate the use of liquid CVD precursors at imec, Dow Electronic Materials provided an R&D version of the central delivery system. It features the output stability and other benefits described above. The use of one of these units enabled imec to use SnCl4 and develop a groundbreaking new CVD process using digermane (Ge2H6) and SnCl4 to grow GeSn epitaxial films in a production-compatible CVD reactor. The films are metastable GeSn alloys with up to 13% substitutional Sn [10,11].

FIGURE 3 shows a typical cross section transmission electron microscope (TEM) picture with associated (224) x-ray diffraction reciprocal space mapping (XRD RSM) of a fully strained GeSn layer, grown on top of a relaxed Ge virtual substrate. The deposition temperature for the GeSn growth was kept low (320°C) in order to allow Sn incorporation in Ge lattice without Sn precipitation or agglomeration.

The TEM picture in Fig. 3(a) exhibits a defect-free and high crystalline quality for the 40-nm-thick GeSn layer. Furthermore, the surface quality of the as-grown Ge0.92Sn0.08/Ge/Si heterostructure was investigated by reflection high-energy electron diffraction (RHEED) analysis after ex-situ transfer to a MBE system. An annealing in ultra-high vacuum up to 420°C resulted in an oxide-free GeSn surface showing a strong (2×1) surface reconstruction as seen on RHEED pattern along the [110] azimuth (Fig. 3(b)). Finally, the XRDRSM around the (2 2 4) Bragg reflections (Fig. 3(c)) demonstrates that the grown GeSn layer is fully strained on Ge/Si (001) substrate.

Conclusion

The use of an improved delivery system for liquid CVD precursors allowed the
use of stannic chloride for the growth of GeSn. The new process developed by imec produces metastable GeSn with concentrations of substitutional tin of 13%.
TM Trademark of The Dow Chemical Company (“Dow”) or an affiliated company of Dow.

References

1. Control of vapor feed from liquid precursors to the OMVPE process, E. Woelk, R. DiCarlo, Journal of Crystal Growth, Available online 29 October 2013, In Press, Corrected Proof.
2. p and n-type germanium layers grown using iso-butyl germane in a III-V metal-organic vapor phase epitaxy reactor, R. Jakomin, G. Beaudoin, N. Gogneau, B. Lamare, L. Largeau, O. Mauguin, I. Sagnes, Thin Solid Films, 519, (2011), 4186–4191.
3. Crystalline Properties and Strain Relaxation Mechanism of CVD Grown GeSn, F. Gencarelli, B. Vincent, J. Demeule- meester, A. Vantomme, A. Moussa, A. Franquet, A. Kumar, H. Bender, J. Meersschaut, W. Vandervorst, R. Loo, M. Caymax, K. Temst, M. Heyns, ECS Trans. 50, (2013), 875-883.
4. Antimony surfactant for epitaxial growth of SiGe buffer layers at high deposition temperatures. Storck, P.; Vorder- westner, M.; Kondratyev, A.; Talalaev, R.; Amamchyan, A.; Woelk, E. Thin Solid Films vol. 518 issue 6 January 1, 2010. p. S23-S29.
5. M. V. Fischetti and S. E. Laux, Journal of Applied Physics 80, 2234 (1996).
6. D. W. Jenkins and J. D. Dow, Physical Review B, 36, 7994 (1987).
7. M. R. Bauer, J. Tolle, C. Bungay, A. V. G. Chizmeshya, D. J. Smith, J. Menéndez and J. Kouvetakis, Solid State Communication 127, 355 (2003).
8. A. Gassenq, F. Gencarelli, J. Van Campenhout, Y. Shimura, R. Loo, G. Narcy, B. Vincent, and G. Roelkens, OPTICS EXPRESS 20 (25) , 27297 (2012).
9. R. F. Spohn and C. B. Richenburg, ECS Transactions 50 (9), 921 (2012).
10. B. Vincent, F. Gencarelli, H. Bender, C. Merckling, B. Douhard, D. H. Petersen, O. Hansen, H. H. Henrichsen, J. Meersschaut, W. Vandervorst, M. Heyns, R. Loo, and M. Caymax, Appl. Phys. Lett., 99, 152103 (2011).
11. F. Gencarelli, B. Vincent, J. Demeulemeester, A. Vantomme, A. Moussa, A. Franquet, A. Kumar, H. Bender, J. Meerss- chaut, W. Vandervorst, R. Loo, M. Caymax, K. Temst, and M. Heyns ECS Journal of Solid State Science and Technology 2 (4), 134 (2013).
12. S. Gupta, B. Vincent, B. Yang, D. Lin, F. Gencarelli, J. Lin, R. Chen, O. Richard, H. Bender, B. Magyari-Koepe, M. Caymax, J. Dekoster, Y.; Nishi, and K. Saraswat, K. Extended Abstracts of the 2013 International Electronic Device Meeting (IEDM) (2012) p. 375.

EGBERT WOELK, PH.D., is director of marketing at Dow Electronic Materials, North Andover, MA. ROGER LOO, PH.D., is a principal scientist at imec, Leuven, Belgium.