The invention relates to a reactor and process for epitaxial layer deposition. It applies especially to manufacturing equipment and process for the growth of epitaxial group IV semiconductor layers at substrate temperatures low in comparison to prior art. Examples are epitaxial silicon carbide layers on SiC and silicon substrates, epitaxial Si layers, epitaxial germanium layers and Si1-xGex alloy layers on silicon substrates, and epitaxial germanium layers on gallium arsenide substrates.
The field of lattice matched epitaxy encompasses homoepitaxy, where a single crystal layer is grown on a single crystalline substrate made from the identical material, or heteroepitaxy, where a single crystal layer is grown on a single crystalline substrate made from a different material but with identical lattice parameter.
The field of lattice mismatched heteroepitaxy concerns growth of a single crystalline film of material I on a single crystalline substrate of material II, whereby materials I and II differ in lattice parameter.
Epitaxial growth can be achieved by a large number of techniques, such as electrochemical deposition, liquid phase epitaxy (LPE), molecular beam epitaxy (MBE), magnetron sputter epitaxy (MSE), and various forms of chemical vapour deposition (CVD), including plasma assisted techniques, such as low-energy plasma-enhanced CVD (LEPECVD). Among these, thermal CVD techniques have so far proven to be most suitable for industrial production of epitaxial structures on a large scale.
In lattice matched epitaxy the main issues calling for low substrate temperatures are the requirements of interface abruptness and sharp doping profiles. The heteroepitaxy of lattice mismatched systems on the other hand involves a number of generic features which are not present in the lattice matched case:
Usually, island formation and dislocation nucleation are both thermally activated processes. They can therefore be influenced by the choice of substrate temperature during epitaxial growth. Irrespective of the mechanism by which islands are formed, their growth inevitably requires surface diffusion of atoms, which is always thermally activated. In view of these considerations, there is a clear need of lowering the substrate temperature during epitaxy if planar films are desired. Here, “low” means substrate temperatures well below those commonly used for epitaxial growth. This has the undesired consequence that the main method for epitaxial film production, i.e., CVD, becomes very slow, because the reaction rate in CVD is itself thermally activated at low substrate temperatures.
The present invention is of particular relevance for the deposition of group IV semiconductors, where in standard CVD very high substrate temperatures are used. For example in SiC homoepitaxy temperatures are usually above 1500° C. (see for example U.S. Pat. No. 6,641,938 to Landini et al., the entire disclosure of which is hereby incorporated by reference), where problems with susceptor stability arise (see for example U.S. Pat. No. 5,155,062 to Coleman, the entire disclosure of which is hereby incorporated by reference). In standard Si homoepitaxy substrate temperatures are above 900° C. (see for example German Pat. No. DE3525870 to Pozzetti et al., the entire disclosure of which is hereby incorporated by reference).
The heteroepitaxy of Germanium on Gallium Arsenide is an example in which problems arise at high substrate temperatures despite of a good match of the lattice parameters. Here, for temperatures above about 550° C. interfacial interdiffusion occurs, leading to undesirable doping of the Germanium layer. When depositing Ge on GaAs and especially on phosphides lattice matched to GaAs substrates it is also advantageous to initiate growth at low substrate temperature in order to avoid decomposition of the substrate because of its limited thermal stability in UHV. On the other hand, a Ge film may also be intentionally doped close to the interface by diffusion from the substrate upon raising the substrate temperature to above 600° C. during or after growth.
An important example of the lattice mismatched case is Silicon-Germanium heteroepitaxy. Germanium and silicon both crystallize with the diamond structure, and the lattice parameter of Ge exceeds that of Si by about 4%. The two elements are furthermore miscible in any proportion. The lattice misfit between a Si1-xGex alloy and Si can therefore be varied between 0 and 4% by choosing x. Strain relaxation in a Si1-xGex film grown epitaxially on a Si substrate has been found to change from plastic to elastic with increasing Ge content. The composition at which the change occurs thereby depends on substrate temperature and deposition rate.
There are several ways in which dislocations can be nucleated and propagate, once the thickness of a strained layer exceeds a certain critical value. In order to act as a misfit dislocation relieving strain, the Burgers vector of a dislocation needs to have an edge component lying in the interfacial plane. As long as the Ge content x is not too large, 60-degree dislocations nucleate at the surface of a Si1-xGex alloy, forming half-loops which expand by dislocation glide under the influence of the misfit strain. Upon reaching the interface, a half-loop forms a misfit segment in the interfacial plane. The two arms of the half-loop extending to the surface of the film, the so-called threading arms, are driven apart by the force exerted by the misfit strain, whereby the misfit segment elongates, relaxing an increasing portion of the misfit strain (see for example Blakeslee, Mat. Res. Soc. Symp. Proc. 148, 217 (1989), the entire disclosure of which is hereby incorporated by reference).
Many applications call for strain relaxed, flat layers, a requirement which in theory can be satisfied by a network of misfit dislocations. For such a network to form in practice, threading arms would have to move apart all the way to the edges of the substrate onto which the epitaxial film is grown, thus leaving behind pure misfit dislocations in the plane of the interface. Unfortunately, an ideal situation such as this never occurs, since threading arms tend to be blocked by dislocation interactions and other obstacles before the wafer edge is reached.
As a result of dislocation blocking, threading dislocations (TDs) appear at the surface of a relaxed epitaxial film. Since these TDs may affect film properties in a detrimental way, their density should be kept as low as possible. Ways must therefore be found to increase the length of misfit segments, for example by reducing TD blocking, or by avoiding films to relax by the mechanism of dislocation loop nucleation and dislocation glide.
One way to reduce dislocation interaction and thereby reduce TD densities was found to consist of grading the Ge content instead of growing an alloy layer at constant composition. Compositional grading has been shown to work for a number of deposition methods, differing vastly in deposition rate, such as UHV-CVD (see for example U.S. Pat. No. 5,659,187 to Legoues et al., the entire disclosure of which is hereby incorporated by reference), MBE (see for example U.S. Pat. No. 5,221,413 to Brasen et al., the entire disclosure of which is hereby incorporated by reference), and LEPECVD (see for example U.S. Pat. No. 7,115,895 to von Känel, the entire disclosure of which is hereby incorporated by reference).
All prior art techniques have certain disadvantages, however, especially when grading to high Ge contents is required. With the exception of LEPECVD, the throughputs of prior art techniques are very low because of low deposition rates, and necessary layer thicknesses of many micrometers. Moreover, again with the possible exception of LEPECVD, substantial surface roughening has been found, especially when grading to high Ge contents x. These surfaces exhibit a so-called cross-hatch formed by troughs and ridges arranged in a square pattern (see for example Lutz et al., Appl. Phys. Lett. 66, 724 (1995), the entire disclosure of which is hereby incorporated by reference). For example UHV-CVD grown buffer layers graded to pure Ge have exhibited rms surface roughness of 210 nm when grown on on-axis Si(001) substrates. Trenches on the cross-hatched surface were found to be as deep as 600 nm (see for example U.S. Pat. No. 6,039,803 to Fitzgerald et al., the entire disclosure of which is hereby incorporated by reference). The trenches were shown to be associated with pile-ups of threading dislocations because of increased dislocation interaction. Somewhat smoother surfaces and fewer pile-ups were observed on off-cut Si substrates. The rms roughness reached, however, 50 nm even in this case, with the deepest trenches still exceeding 400 nm (see for example U.S. Pat. No. 6,039,803 to Fitzgerald et al., the entire disclosure of which is hereby incorporated by reference). Surface roughening at high Ge contents is hard to control in CVD also because substrate temperatures above 700° C. are generally required to promote dislocation motion (and with it, strain relaxation) at the start of the grading, i.e., at low Ge content, whereas this is no longer the case at high Ge contents. Changing the substrate temperature during growth is, however, disadvantageous for most CVD techniques because of strong temperature dependence of deposition rate and Ge incorporation.
In order to improve the surface quality and lower the threading dislocation density, an intermediate chemical-mechanical polishing step after grading to x=0.5 was therefore found to be necessary (see for example U.S. Pat. No. 6,107,653 to Fitzgerald, the entire disclosure of which is hereby incorporated by reference).
One of the main problems of prior art approaches based on graded buffer layers is that the large layer thicknesses involved, together with different thermal expansion coefficients, favour crack formation upon cooling from the growth temperature to room temperature (see for example Yang et al., 3. Appl. Phys. 93, 3859 (2003), the entire disclosure of which is hereby incorporated by reference).
The dislocation structure of pure Ge layers deposited directly onto Si substrates has been found to be distinctly different from the one of Si1-xGex alloy with low Ge content. Here, a network of interfacial 90-degree dislocations was revealed in annealed films grown by hydrogen-surfactant MBE at substrate temperatures of 200° and 300° C. (see for example Sakai et al., Appl. Phys. Lett. 86, 221916 (2005), the entire disclosure of which is hereby incorporated by reference).
Using atmospheric pressure CVD to first deposit a Ge base layer at low substrate temperature (400-500° C.), and then a second Ge layer at higher temperature(750-850° C.), one micrometer thick Ge layers with surprisingly low defect densities could be grown (see for example U.S. Pat. No. 6,537,370 to Hernandez et al., the entire disclosure of which is hereby incorporated by reference). These layers were, however, again rough and needed chemical mechanical polishing. Moreover, as grown layers were found to be compressively strained and had to be annealed before chemical-mechanical polishing.
A closely related approach was described by Luan et al., using UHV-CVD at much lower growth rates (see Luan et al., Appl. Phys. Lett. 75, 2909 (1999), the entire disclosure of which is hereby incorporated by reference). In this case, a threading dislocation density of 2.3×107 cm−2 was observed on 1 μm thick Ge films after repeated temperature cycling. A similar, equally slow procedure with low-pressure CVD was shown to result in smooth surfaces, with rms roughness as low as 0.5 nm (see for example Colace et al., Appl. Phys. Lett. 72, 3175 (1998), the entire disclosure of which is hereby incorporated by reference).
In previous art, LEPECVD was shown to be superior to other techniques for epitaxial SiGe deposition in terms of speed and flexibility in the choice of substrate temperatures. LEPECVD has been proven to yield high-quality graded Si1-xGex layers (see for example U.S. Pat. No. 7,115,895 to von Känel, the entire disclosure of which is hereby incorporated by reference), and pure epitaxial Ge layers (see for example International Patent Application No. WO2005/108654 to von Känel, and von Känel et al., Jap. J. Appl. Phys. 39, 2050 (2000) the entire disclosure of which is hereby incorporated by reference). In contrast to other CVD techniques, neither the growth rate nor the alloy composition changes appreciably when the substrate temperature is varied in the range between 200 and 700° C.
In this previous art, the dense, low-energy plasma used in LEPECVD was generated by a low-voltage DC arc discharge as described for example in U.S. Pat. No. 6,454,855 to von Känel et al., the entire disclosure of which is hereby incorporated by reference.
Systems using a DC arc discharge for plasma generation are hard to scale to large substrate sizes of 300 mm and beyond, because a uniform plasma is very difficult to achieve over an area of this size. There have been attempts to improve plasma uniformity by replacing the original point source (see for example U.S. Pat. No. 6,454,855 of von Känel et al., the entire disclosure of which is hereby incorporated by reference) by a broad-area source as described for example in the International Patent Application No. WO2006/000846 to von Känel et al., the entire disclosure of which is hereby incorporated by reference. The use of a DC source does, however, have additional disadvantages. There are for example, metallic parts, such as a metallic cover of the plasma source, and thermionic emitters in direct contact with the plasma, and hence also with corrosive gases during a cleaning cycle. The same applies to the anode, in case that an anode is used, or to the metallic chamber walls in case of an anode-less design. These features are difficult or impossible to reconcile with in-situ plasma chamber cleaning for reasons of materials compatibility. They therefore increase hardware design complexity, thus raising costs, Hardware degradation affects process reproducibility, leads to chemical memory effects and finally to contamination during a deposition cycle. Chamber cleaning is, however, indispensable in order to avoid particulate contamination and undesirable doping of epitaxial films. Moreover, the thermionic emitters used to sustain the DC arc discharge are a potential source of metal contamination in the growing film.
The described disadvantages are absent when an RF plasma is used instead of a DC plasma. ICP sources have for example been shown to yield very uniform plasmas suitable for etching 300 mm wafers (see for example Collison et al., J. Vac. Sci. Technol. A 16, 100 (1998), the entire disclosure of which is hereby incorporated by reference). Furthermore, there are no metallic parts whatsoever in direct contact with the plasma. Reactor cleaning can therefore be achieved much more easily, in the absence of corrosion of any metal parts. This facilitates hardware design and reduces complexity and costs. Eliminating hardware degradation dramatically reduces cost of ownership of an RF plasma system.
For Si-based semiconductors it has been shown that ion energies have to be kept below about 15 eV when defect-free epitaxy is to be achieved on a wafer scale at low substrate temperatures (see for example U.S. Pat. No. 6,454,855 to von Känel et al., the entire disclosure of which is hereby incorporated by reference). One type of RF plasma source which can be operated in order to meet this requirement is manufactured by CCR Technology, Germany, and described for example in U.S. Pat. No. 6,936,144 to Weiler et al., the entire disclosure of which is hereby incorporated by reference. The corresponding matching network has been described for example in U.S. Pat. No. 7,276,816 to Weiler et al. the entire disclosure of which is hereby incorporated by reference. This kind of plasma source and matching network exhibits a feature of special relevance to the epitaxy of Si-based semiconductors. The RF-power applied to the source can be changed without affecting the ion energies and without readjustment of the matching network. This makes it possible to vary the flux of low-energy ions incident on the surface over a wide range (from tens of μA/cm2 to several mA/cm2). Intense bombardment by low-energy ions is expected to affect the morphology and strain state of an epitaxial film without leading to ion-induced damage. In particular, low-energy ion bombardment may enhance dislocation mobility and hence lead to a reduction of the threading dislocation density in relaxed epitaxial films. This suggests that the often desired reduction of the thickness of a relaxed film can be achieved by tuning the flux of low-energy ions impinging on the surface during growth.
It is one of the objects of the present invention to provide an epitaxy system and modes of operation of RF plasma sources such as those manufactured by CCR Technology in order to permit semiconductor epitaxy on a wafer scale with low defect density.
It is one of the objectives of the present invention to provide a manufacturing system for epitaxial semiconductors, which combines scalability with the ability to obtain high deposition rates at substrate temperatures low in comparison to those used in prior art. This simplifies equipment design, and allows better control over doping profiles, interface abruptness and planarity, as well as material stability, and any effects associated with lattice strain.
It is another object of the present invention to devise a production apparatus and process for epitaxial film deposition, which does not suffer from any growth rate limitations upon lowering the substrate temperature.
It is yet another object of the present invention to devise means for reducing thermal mismatch strain, thereby greatly enhancing the usable range of epitaxial layer thicknesses.
It is yet another object of the present invention to provide a system for Ge heteroepitaxy on Si or GaAs substrates, allowing epitaxial growth at very low substrate temperatures as low as 200° C., while maintaining high rates above 1 nm/s.
It is yet another object of the present invention to devise an epitaxial semiconductor manufacturing system for group IV semiconductors devoid of any metal parts in direct communication with the epitaxial reactor, which is scalable to large substrate sizes and compatible with in-situ plasma cleaning. A related manufacturing system, adapted to epitaxial deposition of compound semiconductor layers, has been described in International Patent Application No. WO2006/097804 to von Känel.
The present invention is applicable also to compound semiconductor epitaxy, using metal organic and hydride gases customary in metal organic chemical vapour deposition.
The present invention concerns an apparatus and process for low-energy plasma enhanced chemical vapour deposition using inductively coupled plasma sources providing ion energies below 20 eV at the substrate position. The epitaxial reactor is compatible with in-situ plasma cleaning by chlorine or fluorine containing gases. Apparatus and process apply especially to the epitaxial deposition of group IV layers and heterostructures, such as SiC homoepitaxy, SiC heteroepitaxy on Si, Si homoepitaxy, Si1-xGex heteroepitaxy on Si, and Ge heteroepitaxy on GaAs substrates. They can, however, be applied also to the deposition of compound semiconductor layers, when metal organic gases are used together with hydrides of group V elements.
The apparatus is designed such as to permit high ion densities above 1010 cm−3 at the substrate position, whereby process gases are highly excited and made more reactive, such as to allow substantial lowering of the substrate temperature during deposition with respect to those customary in pure thermal chemical vapour deposition.
The complete quartz enclosure 3 of the reactor requires only few seals at the gas injection points 13, 15, the pressure gauges (not shown), and the quartz to metal seals 14 close to the pump 9. The gas ring 12 separating the hot deposition region 4 from the cold reactor part 8 of the reactor also helps to throttle total reactor volume, thereby ensuring lower pressure gradients in the system.
An inductively coupled plasma (ICP) source is employed which may optionally comprise an assembly of coils 16 and/or a spiral antenna 17, both located in the vacuum chamber 2 outside the quartz enclosure 3. Gases for the plasma source are introduced through a gas inlet 15. For example He, Ar, H2 or mixtures of these gases may be used to feed the plasma source. The gas mixtures may be provided by a fast gas switching system 150 as shown on
The coils 16 and 17 may also be operated at two different frequencies, for example one of them at 13.56 MHz or above and the other at 2 MHz or below.
During plasma processing reactive gases are supplied through gas inlet 11 and distributed by a gas distribution ring 13, preferably made from fused quartz. The reactive gases can be hydrogen for wafer cleaning and as additive for layer deposition; silane and other Si containing gases for Si deposition; germane and other Ge containing gases for Ge deposition; mixtures of Si and Ge containing gases for Si1-xGex alloy deposition; methane and other C containing gases for diamond-like layer deposition; mixtures of Si and C containing gases for SiC deposition; and mixtures of Si, Ge and C containing gases for the deposition of Si1-x-yGexCy alloy layers. In addition, doping gases such as diborane, arsine, phosphine, and other gases containing elements suitable for doping group IV semiconductors, may be supplied through the gas inlet 11. The doping gases may be supplied in suitably diluted form, for example by diluting with H2, Ar or He. Preferably, the reactive gas mixtures are supplied by the fast gas switching system 150. This, together with an optional rapid change of the power supplied to the ICP source, has the advantage of allowing precise control over layer thicknesses at a monolayer scale.
Introducing the reactive gases through a gas distribution ring 13 close to the susceptor 5 has the advantage of improved gas utilization in comparison to supplying reactive gases through the gas inlet 15 of the plasma source.
It has the further advantage of lowering the ion energies because of thermalization by additional scattering in a region of locally increased pressure close to the substrate.
The application of the inventive scheme 100 can be extended to the epitaxial deposition of compound semiconductor layers, by using metal organic gases such as trimethyl gallium, trimethyl aluminium or trimethyl indium, along with nitrogen, arsine or phosphine.
Alternatively, cleaning gases such as Cl2, NF3, H2 and other chlorine or fluorine containing gases may be supplied through the gas distribution ring 13 during reactor cleaning in a low-energy plasma generated by the coils 16 and 17.
The quartz/ceramic enclosures 3, 7 may optionally be provided with additional heaters (not shown) in order to facilitate the desorption of impurities from their inner walls.
Optionally, an arrangement of coils may be provided outside the enclosures 1 or 3 for shaping the plasma generated by plasma sources 16, 17. An arrangement 200 which was found to be particularly adequate for correcting non-uniformities of the plasma in case of large substrate sizes is shown in
Even more degrees of freedom for shaping the magnetic field, and with it the plasma distribution, can be obtained by using elliptical instead of circular coils, and by tilting them with respect to the plane shown in
An inventive process 300 of semiconductor epitaxy using the system 100 of
The temperature of the wafer is then adjusted according to the appropriate epitaxy temperature T2 in step 320. In a low-energy plasma-enhanced chemical vapor deposition step 325 a semiconductor layer is grown epitaxially by supplying the appropriate gas mixture through the gas distribution ring 13. During epitaxial deposition the pressure in the deposition region 4 is preferably kept in the range of 10−4 to 10−1 mbar. The frequency and power supplied to ICP coils 16, 17 is preferably chosen such as to guarantee a density of low-energy ions of at least 1010 cm−3 at the wafer surface. Here, low-energy ions means ions with energies not exceeding 20 eV, with energies below 15 eV or even below 10 eV being more suitable for defect-free epitaxial growth, since surface bombardment by low-energy ions does not result in any damage. The lowest energies can be achieved by keeping the pressure in the substrate region above 10−2 mbar or preferably even 10−1 mbar, by introducing sufficient gas through the distribution ring 13. This allows the plasma source to be operated close to pure dissociation mode, where the fraction of ions becomes negligible with respect to neutral radicals. A high partial pressure of reactive gases in the deposition region 4 also has the advantage of allowing high deposition rates. For example a total pressure of about 3×10−2 mbar generated by a silane/germane mixture introduced through gas distribution ring 13 and pure argon or another noble gas through the source inlet 15 has been shown to be adequate for a deposition rate of 5-10 nm/s. Here, the pressure reading has been taken at a distance of about 30 cm from gas distribution ring 13, before turning on the plasma. For example for a reactor designed for the processing of 100 mm wafers, these values can be reached at a total reactive gas flow of 40 sccm and an argon flow of 50 sccm, respectively. For larger reactors correspondingly higher gas flows will have to be employed. In step 330 the decision is taken whether epitaxial layer growth is complete. If not, step 320 may be repeated by changing the substrate temperature to some value T3 suitable for further epitaxial growth 325. Alternatively, steps 320 and 325 may also be combined by changing the substrate temperature continuously or in a step-wise fashion during epitaxial growth.
Upon termination of epitaxial growth, the substrate temperature is changed to a value T4 suitable for wafer transfer, and the wafer is transferred to the load-lock through the gate valve 10 in step 335. In an optional cleaning step 340 the interior 4 of the enclosure 3 may then be exposed to a reactive plasma containing hydrogen, fluorine or chlorine species. The plasma cleaning step 340 is preferably carried out at chamber pressures in between 10−3 and 1 mbar, achieved by reactive gas flows 1 to 100 times as large as the flows suitable for epitaxial deposition. This has the advantage of fast etching of deposits from the walls of enclosure 3, the susceptor 5 and the gas distribution ring 13. After plasma cleaning with fluorine or chlorine containing species it is advisable to clean the epitaxial reactor additionally with a pure hydrogen plasma.
It is further advisable to pre-coat the interior of the enclosure 3 prior to the next deposition step in order to passivate O2, F and Cl impurities.
A schematic representation 500 of a third embodiment of the invention is shown in
A schematic representation 600 of a fourth embodiment of the invention is shown in
The present invention is applicable to any epitaxial layer/substrate combinations in which lowering the substrate temperature without loss of growth rate and/or decrease of layer quality is desirable, as is typically the case for chemical vapour deposition processes. Thereby, it should be understood that the term “low substrate temperature” depends on the material system, and therefore cannot be specified by a single number.
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/EP08/60704 | 8/14/2008 | WO | 00 | 7/30/2010 |
Number | Date | Country | |
---|---|---|---|
60956390 | Aug 2007 | US |