Olivier Kermarrec, Yves Campidelli, Daniel Bensahel, STMicroelectronics, Crolles, France
Chun-Hao Ly, Patrick Mauvais, Air Liquide, Jouy en Josas, France
Si1-xGex-based devices are competing with III-V materials for high-speed applications. To meet these device specifications, higher germanium content, 50 to 100%, will be needed. This means growing crystals at lower temperatures to ensure the desired Ge concentration and to minimize strain-related defectivity. Low temperature chemical vapor deposition of these films is increasingly sensitive to the quality of process gases, however, particularly with respect to moisture. It is impotant to quantify the influence of moisture in the carrier gas in terms of strain in epitaxial Si1-xGex thin films.
Silicon germanium is a promising technology for high-speed applications such as telecommunications . Competing with some III-V high-speed circuit technologies, such as InP, it offers additional performance higher speed, reduced electronic noise, lower supply power at a lower cost. Internal structural problems have slowed its progress, however, so that only recently has SiGe been integrated into silicon chip manufacturing processes. A 4.2% mismatch between Si and SiGe lattice parameters presents a challenge to the growth of SiGe layers on Si substrates. Nevertheless, for thin (nanometers to hundreds of nanometers) SiGe layers, SiGe can adapt to the smaller Si lattice parameter, accumulating compressive strain energy in the layer [2, 3]. The layers can then be integrated in devices.
Under certain conditions for example, if the layer is too thick or if it is improperly grown the layer may relieve its strain and recover its bulk lattice parameter. This relaxation is usually mediated by the emission of misfit dislocations at the SiGe-Si interface , leading to poor crystallographic quality. Consequently, the growth of SiGe layers, usually by chemical vapor deposition (CVD), is very sensitive to growth conditions. Moreover, impurity phase segregation strongly affects the film quality. Hence, high quality process gases, precursors, and carrier gases, and precise control of the growth process are required.
Unlike silicon epitaxial layers, which are typically grown at temperatures >1000°C, SiGe thin films must be grown at temperatures below 750°C to obtain higher germanium concentrations and prevent lattice relaxation. Moreover, with the diverse applications of SiGe technology [5-7], demands for increased germanium content to improve mobility, band gap offsets, optical properties, and strain are stronger than ever. To increase germanium content and maintain good control of the growth rate, though, such layers require growth temperatures as low as 600-700°C [8, 9]. At these temperatures, gaseous contaminants like O2 and H2O have higher equilibrium concentrations in SiGe , such that detrimental quantities of oxygen in SiGe have been detected. When oxygen is present in SiGe, H2O plays an important role. This study evaluates the impact of moisture on the quality of SiGe thin films.
Figure 1. The experiment is based on a patented, nonintrusive, optimised sampling method, designed by Air Liquide.
Experiment: Hygrometry control
The experimental set-up consists of two main parts (Fig. 1): a humidifier and a diode laser hygrometer (DLH) attached to a gas-sampling system. The humidifier injects controlled amounts of moisture into the process gas mixture, which consists of silane (SiH4) and germane (GeH4), with hydrogen (H2) as the carrier gas. The DLH measures the moisture level in the mixture before it enters the process chamber. Both systems were integrated upstream in an industrial 200mm CVD reactor (ASM Epsilon One) for low-temperature epitaxy.
Hydrogen and process gases coming from the cylinder supply were purified upstream from the installation to ensure good reproducibility of the moisture level prior to humidification. For gas purification, the H2 flow rate could be very high, ranging from 10 to 50 standard liters/min (slm) Therefore, the humidifier was specifically designed to achieve moisture levels ranging from hundreds of parts per billion (ppb) to units of parts per million (ppm) in the final process mixture. Traditional permeation systems were inappropriate because of their shorter lifetimes. Finally, in the absence of humidification, the low moisture level, a few hundreds of ppb, was used as a reference level, [H2O]ref. This value, held confidential, is under strict electronic grade regulation, as defined by semiconductor manufacturers.
The moisture addition system
The humidifier is composed of two buffers: a saturator and a condenser. A hydrogen carrier gas is run through de-ionised (DI) water in the saturator, where the temperature is held at 2°C, and is saturated with moisture to 6580 ppm. The condenser, also maintained at 2°C, retains the water in the hydrogen. A back-pressure regulator controls the pressure in the buffer and also allows continuous flow. Finally, a few standard cubic centimeters/min (sccm) of moisturized hydrogen can be introduced into the process gas line. The amount of added moisture depends on bath temperature, pressure, and flow-rate introduced into the process. Once all the parameters are fixed, the moisture level is held constant.
The gas-sampling system
When measuring process parameters, in-line and off-line sensors must be installed at several locations in the equipment. For in-line sensors, the analysis must take into account possible changes in the process pressure, which can alter optimal conditions for sensitivity and reliability. In the case of off-line sensors, the gas must be sampled and fed to an off-line sensor, outside the process gas mixture. An external vacuum pump is then needed, particularly in low-pressure processes, and a pressure regulating system is required in order to operate the sensor in its optimal pressure range. In addition, gases must be safely disposed of or evacuated, particularly when analyzing toxic gases.
The sampling system used here is a semi-off line configuration: instead of using its own external pump, the sensor uses the dry pump of the Epsilon One. The gas sampling system consists of a sampling line installed between a sampling port located at the inlet or eventually at the outlet of the process chamber, and at the inlet of the sensor. This sampling line is completed with a vent line "exhaust" (Fig. 1), positioned at the outlet of the sensor and connected to a port located downstream of the process pressure regulating throttle valve. Calibrated orifices are located on the sampling and vent lines in order to adjust the pressure in the sensor between the process pressure and the lowest available pressure in the exhaust system.
In the case where the process pressure is changing, e.g., at different process steps, several orifices can be placed in parallel on the sampling line, each used sequentially during each process step to maintain a constant flow rate and pressure through the sensor. If one of the parallel lines is by-passed, the sensor pressure becomes the process pressure and the flow rate is limited only by the orifice located on the vent line. The benefit of such a system is that optimum conditions for the moisture sensor, in terms of sensitivity and reliability, can be maintained without imposing process modifications. This system also allows continuous sampling without modifying the analyzer to adapt to process conditions, such as pressure and gas type. One disadvantage is the requirement of external piping.
Figure 2. The measurement principle of a diode laser hygrometer a) and the signal as a function of the level of moisture b).
Diode laser hygrometer
The H2O analyzer [11, 12], developed by Air Liquide, was designed to measure the moisture in a variety of configurations and processes where moisture is a contaminant. The detection limits range from 20 to 50 ppb, making this tool preferable to Fourier transform infrared analysis (FTIR) or intracavity laser spectroscopy (ILS) for the continuous analysis of H2O levels in electronic specialty gases (ESGs). The hygrometer (Fig. 2) uses a tunable InGaAs/InP diode laser light source emitting at 1.368μm. The emitted light is split equally into two beams, reference and probing. The analyzed gas is sampled from the process line, then passes through a cell where the probing laser beam is focused. The cell is an enclosed volume composed of two windows and two mirrors. The probing beam enters the cell through one of the windows, undergoes several reflections between the mirrors (to lengthen the absorption path), exits the cell through the other window, and is focused onto an InGaAs infrared (IR) detector.
The signal generated by the detector is then subtracted from that of the reference light. This allows measurement of the light intensity absorbed by moisture and derivation of its concentration. The emitted light bandwidth is narrower than that of the moisture absorption band, so the diode laser wavelength must be swept in order to cover the selected moisture absorption band. The laser wavelength depends on both temperature and current, though laser current is easier and more accurately controlled. When the temperature stabilizes, the diode laser is rapidly swept over the moisture absorption band by modulating the diode current with a "sawtooth" function. The averaged value of the current is defined for the diode laser emission at 1.368μm. The total volume of the cell is one liter. The cell itself is made of electropolished 316L stainless steel. This ensures the compatibility with most ESGs, in particular HBr  and HCl, which are known to be very corrosive, especially in the presence of moisture. The diode laser hygrometer was calibrated with nitrogen and in parallel with an AMETEK hygrometer.
To study and quantify the influence of moisture on epitaxy quality, several SiGe layers were grown on 200mm Si wafers in the presence of different moisture levels, ranging from hundreds of ppb to units of ppm. The wafers were first heated to ~1000°C to remove native oxide from the surface, prior to layer growth in a hydrogen atmosphere, at a reduced pressure (a few torr). The hydrogen used during the bake was the same as used during the SiGe process, and therefore contained the same level of moisture. The temperature was then lowered to the growth temperature, below 750°C. Once this temperature was reached, silane and germane were introduced into the process line where they were mixed before entering the growth chamber. Fully strained layers, 80nm thick with 10% germanium, were grown.
Between each wafer processed at a given moisture level, the humidifier was turned off and the process line was purged with purified hydrogen to remove the remaining moisture. Then, as usually performed in epitaxy, the chamber was cleaned with an HCl/pure H2-based process. This also eliminates the impact of moisture accumulation in the chamber. There was therefore no memory effect.
Figure 3. The surface of the SiGe layers at 25,000¥ magnification shows the emergence of dislocation defects as moisture is added to the process gases during crystal growth.
Quantifying the effects of moisture
Optical inspection of the wafers under Normarski light revealed smooth, mirror-like surfaces when moisture was 13¥[H2O]ref ppb. Above this moisture level, the surface appeared hazy. Subsequent scanning electron microscopy (SEM) revealed defect-free surfaces when the purest gases were used, and the formation of large dome-shaped defects, about 250nm wide, as moisture was added (Fig. 3). The density of these defects increased with increasing moisture levels. For 21¥[H2O]ref ppb, the density measured by atomic force microscopy (AFM) was ~2 ¥ 108 domes/cm2. Also from AFM measurements, surface roughness root mean square (rms) and maximum peak to valley distance (Rmax) as a function of the moisture level during growth were tracked as moisture was added. Both rms and Rmax suddenly increased when the moisture concentration reached 13¥[H2O]ref ppb (Fig. 4). Between [H2O]ref and 13¥[H2O]ref ppb, Rmax values increased, but at a significantly slower rate. These observations confirm that the appearance and the number of dome-like defects is strongly related to the moisture concentration in the process gases.
Figure 4. Atomic force measurements of a SiGe layer indicate a rapid increase in surface roughness and amplitude as the moisture content in the process gases increases.
What is the origin of the defects? As revealed by a Secco etch, they are caused by emerging dislocations in the SiGe layer. The Secco etch is a very useful chemical method that reveals emerging surface dislocations; particularly in SiGe layers where x is <20%.
After the etch, no dislocations were detected for layers with no moisture added, but reached 1.4 ¥ 109 dislocations/cm2 for 21 ¥ [H2O]ref ppb of H2O. The impact of moisture on dislocation nucleations in SiGe epitaxial layers is significant.
A critical moisture level beyond which moisture is disruptive to high-quality epitaxial growth is 13¥[H2O]ref ppb. At this level, moisture in process gases causes dislocation nucleation in the SiGe layer, which degrades the crystallographic quality of the layer and, implicitly, its electrical and optical properties. This moisture concentration is critical for Ge content of 10% and a layer thickness of 80nm. With increased Ge content and layer thickness, this critical value will most likely decrease.
Moisture level/oxygen correlation
Deleterious effects of the presence of moisture in process gases are due to the incorporation of oxygen atoms in the epitaxial layer during growth. To quantify the impact of the presence of H2O molecules in hydrogen on atomic oxygen concentration and distribution within the layer, the wafers were analyzed by secondary ion mass spectrometry (SIMS). For accurate atomic concentrations, the relation between the number of atoms/sec detected and the real atomic concentration must be accurately determined. For atomic oxygen, however, this relation was not well known. Therefore, these results are in terms of counts/sec, instead of atoms/cm3, indicating relative concentrations. The oxygen profiles obtained for the layers exposed to [H2O]ref ppb, 13¥[H2O]ref ppb, and 21¥[H2O]ref ppb of moisture are reported. The samples exhibited a higher local oxygen peak at the SiGe/Sisubstrate interface due to the adsorption and accumulation of oxygen atoms at the surface of the substrate during the high-temperature bake in a hydrogen atmosphere, prior to the SiGe epitaxy. The adsorption of oxygen does not result in oxide layer formation. If oxide were formed, the SiGe would then be polycrystalline, which, in spite of the high dislocations concentration measured, is not the case.
Figure 5. SIMS profiles show an increased presence of atomic oxygen in SiGe layers as moisture is added to the process gases.
The oxygen concentrations at the SiGe/Sisubstrate interface and in the SiGe layer were correlated with the moisture level in the process mixture (Figs. 5 and 6). For the purest gas mixture, no interfacial peak was observed and 9 ¥ 103 counts/sec of oxygen were detected. The oxygen detected reached 3.9 ¥ 104 counts/sec in the layer and 1.9 ¥ 105 counts/sec at the SiGe/Sisubstrate interface for 21¥[H2O]ref ppb of added moisture. From the maximum moisture level to ensure good structural quality of SiGe layers, 13¥[H2O]ref ppb, the critical oxygen concentration is ~11 ¥ 103 counts/sec.
Figure 6. The SIMS measurement illustrates the evolution of atomic oxygen in the SiGe layer and at the SiGe/Si interface with the addition of moisture to the process gases.
A direct correlation between moisture impurity in process gases and atomic oxygen present in epitaxial SiGe layers was demonstrated, both qualitatively and quantitatively. The resulting incorporation of oxygen atoms can induce dislocations into the strained layers, which may degrade device performance and, subsequently, reliability. The present study determined stuctural changes noted by alterations in the surface morphology. But the incorporation of oxygen atoms in the lattice at concentrations lower than those degrading the surface can also negatively affect electrical and optical properties. This requires further study, which is currently being addressed. Moisture impact on quality as a function of the process temperature, germanium content, and strained/relaxed nature of the layers will also be investigated in future studies.
The authors wish to thank Gérard De Luca, Jean-Marie Friedt, Jean-Marc Girard, Benjamin Jurcik, Olivier Letessier, all from Air Liquide, for their participation in this work.
1. B. Martinet, et al., "100 GHz SiGe:C HBTs Using Nonselective Base Epitaxy", ESSDERC 2001, invited paper, accepted.
2. H. Gao, W.D. Nix, "Surface Roughening of Heteroepitaxial Thin Films," Annu. Rev. Mater. Sci., 29, pp. 173-209, 1999.
3. R. People, J.C. Bean, "Calculation of Critical Layer Thickness vs. Lattice Mismatch for GexSi1-x/Si Strained-Layer Heterostructures," Applied Physics Letters, Vol. 47, n3, pp. 322-324, 1985.
4. H.P. Strunk, et al., "Relaxation of Misfit-Induced Strain in Semiconductor Heterostructures," Phys. Stat. Sol. (a), Vol. 171, pp. 215-225, 1999.
5. H. Presting, "Near and Mid Infrared Silicon/Germanium-Based Photodetection", Thin Solid Films, Vol. 321, pp. 186-195, 1998.
6. L. Colace, G. Masini, G. Assanto, "Ge-on-Si Approaches to the Detection of Near-Infrared Light," IEEE Journal of Quantum Electronics, 35 (12), pp. 1843-1852, 1999.
7. D.J. Paul, "Silicon-Germanium Strained Layer Materials in Microelectronics," Advanced Materials, Vol. 11 (3), pp. 191-204, 1999.
8. V. Loup, et al., "CVD Epitaxial Growth," Solid State Technology, 46 (7), pp. 91-102, 2001.
9. D.W. Greve, "Growth of Epitaxial Germanium-Silicon Heterostructures by Chemical Vapour Deposition," Materials Science and Engineering, B18, pp. 22-51, 1993.
10. J. MacDraw, "Progress in In Situ Contamination Control," Semiconductor International, 21 (5), pp. 71-78, 1998.
11. J.J.F. McAndrew, et al., "Increasing Equipment Up-time Through In Situ Moisture Monitoring," Solid State Technology, 41 (8), p. 61, 1998.
12. J. McAndrew, et al., "In-Situ and On-Line Moisture Monitoring in Reactive Gas Environments", MICRO, 18 (2), p. 39, 2000.
13. A. Zdunek, et al., "Onsite Process Gas Measurement of Moisture and Particles In Corrosive Distribution Lines," Symposium on Contamination-Free-Manufacturing (CFM) for Semiconductor Processing, Semicon West, 2000.
Olivier Kermarrec received an engineering degree in device physics from the Physics Engineering School of the Technical University of Grenoble (INPG). He is a PhD student at STMicroelectronics, within the advanced front-end materials team, working on epitaxial SiGe and SiGe:C materials. STMicroelectronics, Zone Industrielle Pre Roux, BP 16-38190 Crolles, France; ph 33/4-7676-4403, fax 33/4-7690-3443, email@example.com.
Yves Campidelli has spent more than 25 years working on MBE, CBE, and CVD systems in semiconductors, starting his career at France Telecom R&D. He joined STMicroelectronics in 1999 as a process R&D engineer.
Daniel Bensahel earned a Doctorat d'Etat in solid state physics in 1979 at Grenoble University, before joining France Telecom R&D in Meylan, where he worked on SOI and Si and SiGe materials. In 2000, he joined STMicroelectronics at Crolles as advanced front-end materials manager.
Chun-Hao Ly is an electronics engineer in R&D at Air Liquide, where he is currently in charge of R&D projects with ST Microelectronics. He graduated from Ecole Nationale Supérieure des Procédés Electroniques et Optiques at Orléans in 1996 and earned a masters degree in management from the University of Orléans in 1997.
Patrick Mauvais is a research technician in gas applications for the electronics and laboratory group of Air Liquide Corporate R&D center. He has expertise in trace analysis in electronics specialty gases.