Low-energy plasma-enhanced chemical vapor deposition: LEPECVD

Low-energy plasma-enhanced chemical vapour deposition: LEPECVD

What it is

The original LEPECVD: a 4-inch (100 mm) single-wafer silicon germanium growth system. The machine was developed at the ETH in Zürich. It is now at the L-NESS, Polo Regionale di Como. Details of publications are on the research page. LEPECVD has a page at lness.como.polimi.it/lepecvd.php.

LEPECVD Schematic

Schematic diagram of the low-energy plasma-enhanced chemical vapour deposition system.


The 4-inch LEPECVD system on which most of the research has been done so far.


The second-generation LEPECVD (LG2): a 200 mm single-wafer silicon germanium growth system.


What it does

These systems are used for the growth of epitaxial SiGe (see D. J. Paul Semicond. Sci. Technol. 19 (10) R75-R108 (2004)) films on silicon substrates, from the gaseous precursors silane (SiH4) and germane (GeH4). Additionally, the material can be doped by adding diborane (B2H6) or phosphene (PH3) to the mix, for p- or n-type material respectively.

Growth from gaseous precursors is generally known as chemical vapour deposition (CVD). The growth rates of CVD are

  1. exponentially dependent on the substrate temperature
  2. confusingly dependent on the mixture of Si and Ge in the gas and
  3. very low (maybe 10 nm min-1) unless you go to about 1000°C for Si.

Also, the Ge fraction in the gas is very different from the fraction you get in the alloy. For example, see J. M. Hartmann et al. Semicond. Sci. Technol. 15 (4) 362-369 (2000) or J. M. Hartmann et al. J. Crystal Growth 236 (1-3) 10-20 (2002). The slow growth rates and high temperatures often lead to serious roughening of the surface, and so chemical-mechanical polishing has to be employed... sometimes even twice (S. B. Samavedam et al. Appl. Phys. Lett. 73 [15] 2125-2127 [1998]).

Radio-frequency plasma-enhanced chemical vapour deposition (RF-PECVD) is used to create amorphous or microcrystalline SiGe films, and the growth rates can be around 10 nms-1. But these systems cannot be used for epitaxial single-crystal SiGe which is suitable for electronics applications.

The LEPECVD system gives the quality of conventional CVD (or better) at plasma-assisted rates (M. Rondanini et al. J Appl. Phys. 104 [01] 013304 [2008]; Surf. Coat. Technol. 201 [22-23] 8863-8867 [2007]). The 4-inch system can grow SiGe layers at 5-10 nms-1 at any substrate temperature. The growth rate has no strong dependence on substrate temperature or alloy composition. The Ge fraction of the alloy is almost exactly the same as of the gas mixture. By reducing the plasma efficiency, however, the growth rate can be reduced for the sake of control. This means we can grow strained-layer heterostructures and superlattices with sharp interfaces, as can be seen for example in M. Bollani et al. Mat. Sci. Eng. B 101 (1-3) 102-105 (2003).

What it's for

Relaxed graded silicon germanium buffer layers greatly extend the possibilities of the silicon germanium material system. Such a relaxed buffer layer, or virtual substrate, allows the growth of a tensile strained silicon quantum well for n-type conduction. Also, it is possible to grow compressively strained quantum wells for p-type conduction with any germanium fraction.

The traditional method of producing a virtual substrate is to linearly increase the germanium fraction x at around 10% per micron (E. A. Fitzgerald et al. Appl. Phys. Lett. 59 [7] 811-813 [1991]). LEPECVD is excellent for this purpose, since growth rates of 5-10 nms-1 can be reached. This means that a buffer which is graded at 7% per micron from pure Si to pure Ge, and capped with 2 microns of pure Ge, still takes less than one hour to grow. The threading dislocation density of such a buffer has been measured to be 1.5×105 cm-2 and the rms surface roughness is 3.2 nm (S. G. Thomas et al. J. Electron. Mater. 32 [9] 976-980 [2003]). For comparison, the threading dislocation density of a similar buffer grown by ultra-high vacuum chemical vapour deposition (UHV-CVD) is 2.1×106 cm-2 (and the rms roughness is 24 nm) but here the wafer has to be taken out of the growth system half-way through and subjected to chemical-mechanical polishing (CMP) (S. B. Samavedam et al. Appl. Phys. Lett. 73 [15] 2125-2127 [1998]).

Such structures can have excellent electrical properties (B. Rößner et al. Appl. Phys. Lett. 82 [5] 754-756 [2003], B. Rössner et al. Appl. Phys. Lett. 84 [16] 3058-3060 [2004]). These papers feature modulation-doped strained Ge channels on 70 % virtual substrates.

Standard p-type modulation-doped strained Ge channel heterostructure

Standard p-type modulation-doped strained Ge channel heterostructure

We have measured a 2 K mobility of 120 000 cm2V-1s-1 at a carrier density of 8.5×1011 cm-2. This is the highest p-type mobility ever reported in SiGe-based material. Material grown by molecular-beam epitaxy has reached 55 000 cm2V-1s-1 (at a carrier density of 5.5×1011 cm-2) at 4.2 K (Y. H. Xie et al. Appl. Phys. Lett. 63 [15] 2263-2264 [1993]). Room temperature channel mobilities of around 3000 cm2V-1s-1 can be extracted with mobility spectrum analysis (D. Chrastina et al. J. Appl. Phys. 94 [10] 6583-6590 [2003]).

Thin buffers

However, a typical graded virtual substrate is several microns thick and substantial reduction of buffer layer thickness is desirable. In fact, various methods have been developed over the past few years to arrive at highly relaxed SiGe layers as thin as a hundred nanometers (M. Luysberg et al. J. Appl. Phys. 92 [8] 4290-4295 [2002]). However, some of these methods require low temperature growth which cannot be realized by chemical vapour deposition (CVD) since growth rates decrease exponentially as the substrate temperature is reduced. Other methods employ ion implantation of H or He into a strained SiGe film, and subsequent thermal treatment (B. Holländer et al. Nucl. Instr. Meth. B 175-177 357-367 [2001]). This leads to bubble formation, facilitating dislocation loop nucleation close to the interface. Except for the extra processing steps this method appears very attractive but seems to be limited to virtual substrates with Ge content below 30 %. To reach 85 % relaxation of Si0.7Ge0.3, the buffers are approximately 200 nm thick with a threading dislocation density of 108 cm-2 (J. Cai et. al. J. Appl. Phys. 95 [10] 5347-5351 [2004]).

Thick and thin buffers with MBE overgrowth

Thick and thin buffers with MBE overgrowth

We have developed a method to grow thin relaxed SiGe buffers by LEPECVD (D. Chrastina et al. Thin Solid Films 459 [1-2] 37-40 [2004]). Field-effect transistors have been fabricated on both thick and thin virtual substrates at DaimlerChrysler. (As featured at PhysOrg.com.)

It was found that the thin buffer reduces the self-heating while the maximum oscillation and transit frequencies (fmax and ft) are not significantly impaired (T. Hackbarth et al. Appl. Phys. Lett. 83 [26] 5464-5466 [2003]). A review of the performance of strained-Si n-type FETs on different virtual substrates is given in M. Enciso Aguilar et al. Solid State Electron. 48 (8) 1443-1452 (2004).

Quantum Cascades

Emitters in the terahertz region of the electromagnetic spectrum (1-10 THz) may find a number of potential applications in the fields of medical and security imaging, pollution monitoring and bioweapons detection (P. H. Siegel. IEEE Trans. Microw. Theory 52 [10] 2438-2447 [2004]; Y. C. Shen et al. Appl. Phys. Lett. 86 [24] 241116 [2005]). GaAs quantum cascade lasers (QCL) emitting in the THz have been demonstrated (R. Köhler et al. Nature 417 [6885] 156-159 [2002]; B. Williams et al. Opt. Express 13 [9] 3331-3339 [2007]) but their applicability is hindered by the relatively low power delivered (∼100 mW) and low operating temperature (≤100 K) required. Such limitations mainly stem from the short non-radiative intersubband lifetimes which, in III-V cascades, are dominated by polar phonon scattering. The non-polar nature of SiGe alloys should result in a reduced temperature dependence of non-radiative lifetimes and allow high-temperature operation of SiGe QCLs. Other evident advantages over III-V structures would obviously come from a much easier integration with standard Si fabrication technology. Since its first demonstration (G. Dehlinger et al. Science 290 [5500] 2277-2280 [2000]) various groups have been able to obtain mid-infrared (L. Diehl et al Appl. Phys. Lett. 81 [25] 4700-4702 [2002]; I. Bormann et al. Appl. Phys. Lett. 80 [13] 2260-2262 [2002]) and THz (S. A. Lynch et al. Appl. Phys. Lett. 81 [9] 1543-1545 [2005]; R. Bates et al. Appl. Phys. Lett. 83 [20] 4092-4094 [2003]) emission from Si/SiGe quantum cascade structures, yet no laser action has been achieved so far. Actually, such a task is exceedingly demanding both from the point of view of the design and of the experimental realization. Due to the high electron effective mass in the (001) tunnel direction, valence band cascades appear to be more promising in SiGe, even though a detailed modeling of the complex heavy-hole (HH), light-hole (LH) and split-off (SO) bandstructure is required. Moreover, the reported Si/SiGe THz waveguide losses (A. De Rossi et al. IEEE J. Quantum Elect. 42 [12] 1233-1238 [2006]) and the limited gain attainable requires the deposition of relatively thick (≥10 μm) epitaxial heterostructures. We have obtained results on a bound-to-continuum cascade structure (J. Faist et al. Appl. Phys. Lett. 78 [2] 147-149 [2001]) grown by LEPECVD which employs a HH to HH intrawell transition to achieve emission in the THz (G. Isella et al. Proceedings of the IEEE Fifth International Conference on Group IV Photonics, 2008, p. 29; D. J. Paul et al. In SiGe, Ge, and Related Compounds, 2008, p. 865).

Integration of GaAs on Si

GaAs and pure Ge have very similar lattice constants, so GaAs structures are often deposited on Ge wafers. However, there would be the considerable cost and weight benefits were it possible to deposit GaAs on Si wafers: to realize this, a Ge buffer layer can be deposited on a Si wafer to form a virtual substrate. The buffer can of course be graded (S. G. Thomas et al. J. Electron. Mater. 32 [9] 976-980 [2003]), as employed successfully for an InGaAs laser structure (Y. Chriqui et al. Electron. Lett. 39 [23] 1658-1659 [2003], Y. Chriqui et al. Opt. Mater. 27 [5] 846-850 [2005]) but Ge can also be deposited directly on Si with no graded part. GaAs material deposited on such a Ge-on-Si VS tends to be under slight tensile strain due to the differences in thermal expansion between Si and Ge or GaAs (A. Rehman Khan et al. IEEE Conf. Emerging Technol. [2005]). Efficient solar cells have been demonstrated on such buffers (G. Gabetta et al. 31st IEEE PV Specialists Conference 850-853 [2005], R. Ginige et. al. Semicond. Sci. Technol. 21 [6] 775-780 [2006]).

Ge/Si photodiodes

Ge layers grown epitaxially on Si, with thicknesses of the order of 1-3 μm, can be used as photodiodes in their own right (G. Isella et. al. Semicond. Sci. Technol. 22 [1] S26-S28 [2007]; J. Osmond et al. Thin Solid Films 517 [1] 380-382 [2008]; J. Osmond et al. Proceedings of the IEEE Fifth International Conference on Group IV Photonics, 2008, p. 164). The diode structures usually take the p-i-n form where the n-type substrate is the bottom contact. The top contact is formed in situ by depositing heavily boron-doped Ge by LEPECVD - carrier concentrations of over 1018 cm-3 can be attained.

Quantum-confined Stark effect

Kuo et al. (Nature 437 [7063] 1334-1336 [2005], IEEE J. Sel. Topics Quantum Electron. 12 [6 Part 2] 1503-1513 [2006]) demonstrated the quantum-confined Stark effect (QCSE) in a strain-compensated SiGe/Ge superlattice grown on a virtual substrate. Similar structures have been grown by LEPECVD (Tsujino et al. Appl. Phys. Lett. 89 [26] 262119 [2006]), demonstrating QCSE, direct-gap photoluminescence (M. Bonfanti et al. Phys. Rev. B 78 [4] 041407 [2008]; D. Chrastina et al. Proceedings of the IEEE Fifth International Conference on Group IV Photonics, 2008, p. 194; M. Bonfanti et al. Physica E in press) and polarization-dependent absorption (M. Virgilio et al. Phys. Rev. B 79 [7] 075323 [2009]) at low temperatures. This is of technological interest since it may be a route towards a silicon-integrated electro-optical modulator. Since the active structure is strain-compensated (consisting of alternating tensile- and compressively-strained layers such that the total strain is zero with respect to the virtual substrate) there is essentially no limit to the thickness of the device - especially when grown by a high-efficiency gas-source processes such as LEPECVD.


A list of publications is given on the research page.

Google Searches

Copyright © 2004-2009 Danny Chrastina

Last updated: 4th March 2009

Email: danny at chrastina dot net

Valid XHTML 1.0!