US 6,521,041 B2Grant

Etch stop layer system

Issue Date:2003-02-18

•10 Claims

•10 Drawing Sheets

Abstract

A SiGe monocrystalline etch-stop material system on a monocrystalline silicon substrate. The etch-stop material system can vary in exact composition, but is a doped or undoped Si_1−xGe_xalloy with x generally between 0.2 and 0.5. Across its thickness, the etch-stop material itself is uniform in composition. The etch stop is used for micromachining by aqueous anisotropic etchants of silicon such as potassium hydroxide, sodium hydroxide, lithium hydroxide, ethylenediamine/pyrocatechol/pyrazine (EDP), TMAH, and hydrazine. For example, a cantilever can be made of this etch-stop material system, then released from its substrate and surrounding material, i.e., “micromachined”, by exposure to one of these etchants. These solutions generally etch any silicon containing less than 7×10¹⁹cm⁻³of boron or undoped Si_1−xGe_xalloys with x less than approximately 18. Alloying silicon with moderate concentrations of germanium leads to excellent etch selectivities, i.e., differences in etch rate versus pure undoped silicon. This is attributed to the change in energy band structure by the addition of germanium. Furthermore, the nondegenerate doping in the Si_1−xGe_xalloy should not affect the etch-stop behavior. The etch-stop of the invention includes the use of a graded-composition buffer between the silicon substrate and the SiGe etch-stop material. Nominally, the buffer has a linearly-changing composition with respect to thickness, from pure silicon at the substrate/buffer interface to a composition of germanium, and dopant if also present, at the buffer/etch-stop interface which can still be etched at an appreciable rate. Here, there is a strategic jump in germanium and concentration from the buffer side of the interface to the etch-stop material, such that the etch-stop layer is considerably more resistant to the etchant.

Metadata

Assignee

Massachusetts Institute of Technology

Inventors

Kenneth C. Wu
Eugene A. Fitzgerald
Jeffrey T. Borenstein

Application Information

Application Number:US 09/289,514

Filing Date:1999-04-09

Priority Date:1998-04-10

Art Unit:7

Classifications

IPC:

C30B 2502

Field of Search:

117 94117 95117 97117915117939148 331

Patent Drawings (10 sheets)

Description

Priority Information

[0002] This application claims priority from provisional application Ser. No. 60/081,301 filed Apr. 10, 1999.

Background of the Invention

[0003] The invention relates to the field of etch-stop material systems on monocrystalline silicon.

[0004] Microelectromechanical systems (MEMS) form the bridge between conventional microelectronics and the physical world. They serve the entire spectrum of possible applications. MEMS include such varied devices as sensors, actuators, chemical reactors, drug delivery systems, turbines, and display technologies. At the heart of any MEMS is a physical structure (a membrane, cantilever beam, bridge, arm, channel, or grating) that is “micromachined” from silicon or some other electronic material. Since MEMS are of about the same size scale and, ideally, fully integrated with associated microelectronics, naturally they should capitalize on the same materials, processes, equipment, and technologies as those of the microelectronics industry. Because the process technology for silicon is already extensively developed for VLSI electronics, silicon is the dominant material for micromachining. Silicon is also mechanically superior to compound semiconductor materials and, by far, no other electronic material has been as thoroughly studied.

[0005] A wide array of micromachined silicon devices are fabricated using a high boron concentration “etch-stop” layer in combination with anisotropic wet etchants such as ethylenediamine and pyrocatechol aqueous solution (EDP), potassium hydroxide aqueous solution (KOH), or hydrazine (N₂H₂). Etch selectivity is defined as the preferential etching of one material faster than another and quantified as the ratio of the faster rate to the slower rate. Selectivity is realized for boron levels above 10¹⁹cm⁻³, and improves as boron content increases.

[0006] It should be noted that etch stops are also used in bond and etch-back silicon on insulator (BESOI) processing for SOI microelectronics. The etch-stop requirements differ somewhat from those of micromachining, e.g., physical dimensions and defects, but the fundamentals are the same. Hence, learning and development in one area of application can and should be leveraged in the other. In particular, advances in relaxed SiGe alloys as substrates for high speed electronics suggests that a bond-and-etch scheme for creating SiGe-on-insulator would be a desirable process for creating high speed and wireless communications systems.

Summary of the Invention

[0007] Accordingly, the invention provides a SiGe monocrystalline etch-stop material system on a monocrystalline silicon substrate. The etch-stop material system can vary in exact composition, but is a doped or undoped Si_1−xGe_xalloy with x generally between 0.2 and 0.5. Across its thickness, the etch-stop material itself is uniform in composition. The etch stop is used for micromachining by aqueous anisotropic etchants of silicon such as potassium hydroxide, sodium hydroxide, lithium hydroxide, ethylenediamine/pyrocatechol/pyrazine (EDP), TMAH, and hydrazine. For example, a cantilever can be made of this etch-stop material system, then released from its substrate and surrounding material, i.e., “micromachined”, by exposure to one of these etchants. These solutions generally etch any silicon containing less than 7×10¹⁹cm⁻³of boron or undoped Si_1−xGe_xalloys with x less than approximately 18.

[0008] Alloying silicon with moderate concentrations of germanium leads to excellent etch selectivities, i.e., differences in etch rate versus pure undoped silicon. This is attributed to the change in energy band structure by the addition of germanium. Furthermore, the nondegenerate doping in the Si_1−xGe_xalloy should not affect the etch-stop behavior.

[0009] The etch-stop of the invention includes the use of a graded-composition buffer between the silicon substrate and the SiGe etch-stop material. Nominally, the buffer has a linearly-changing composition with respect to thickness, from pure silicon at the substrate/buffer interface to a composition of germanium, and dopant if also present, at the buffer/etch-stop interface which can still be etched at an appreciable rate. Here, there is a strategic jump in germanium and concentration from the buffer side of the interface to the etch-stop material, such that the etch-stop layer is considerably more resistant to the etchant.

[0010] In accordance with one embodiment of the invention, there is provided a monocrystalline etch-stop layer system for use on a monocrystalline silicon substrate, the system comprising a graded layer of Si_1−xGe_xand a uniform etch-stop layer of Si_y−1Ge_y. In a particular embodiment, the buffer layer is graded up to approximately Si_0.8Ge_0.2and a uniform etch-stop layer of approximately Si_0.7Ge_0.3.

[0011] In another embodiment of the invention, there is provided a method of fabricating a monocrystalline etch-stop layer on a silicon substrate comprising depositing a graded buffer layer of Si_1−xGe_xon the silicon substrate; and depositing a uniform etch-stop layer of Si_1−yG_yon the graded buffer layer. In a particular embodiment, the buffer layer is graded up to approximately Si_0.8Ge_0.2on the silicon substrate; and the uniform etch-stop layer of Si_0.7Ge_0.3is deposited on the graded buffer layer.

[0012] In yet another embodiment of the invention, there is provided a method of micromachining an integrated device comprising providing a silicon substrate; depositing a graded buffer layer of Si_1−xGe_xon the silicon substrate; depositing a uniform etch-stop layer of Si_1−yGe_yon the graded buffer layer; etching portions of the silicon substrate and the graded buffer layer in order to release the etch-stop layer; and processing the released etch-stop layer.

Brief Description of the Drawings

[0013] FIGS. 1A-1D are functional block diagrams of exemplary epitaxial SiGe etch stop structures configured on a silicon substrate in accordance with the invention;

[0014] FIG. 2 is a cross-sectional TEM micrograph of the structure of FIG. 1B;

[0015] FIG. 3 is a cross-sectional TEM micrograph of the structure of FIG. 1C;

[0016] FIG. 4 is graph of dopant concentrations of the structure of FIG. 1A;

[0017] FIG. 5 is a graph of dopant concentrations of the structure of FIG. 1D;

[0018] FIG. 6A is a graph showing the cylindrical etch results of the structure of FIG. 1A; FIG. 6B is graph showing a magnification of the left side of FIG. 6A;

[0019] FIG. 7 is a graph showing the cylindrical etch results of the structure of FIG. 1D;

[0020] FIG. 8 is a graph showing the etch rates for <100> intrinsic silicon in 34% KOH at 60° C. normalized by 18.29 μm/hr of the structures of FIGS. 1A-1D;

[0021] FIG. 9 is a photograph of a top view of a micromachined proof mass; and

[0022] FIG. 10 is a block diagram of a process for fabricating an SiGe-on-insulator structure.

Detailed Description of the Invention

[0023] In the traditional method of forming etch stops in Si micromachining or in certain SOI processes, good etch-stop results are only obtained at very high concentrations of boron, and the dopant's effect on the silicon crystal structure becomes vitally important. Substitution of a silicon atom site with boron, a smaller atom than silicon, contracts the silicon lattice. As the doped lattice remains coherent with the lattice of the undoped substrate, a biaxial “lattice mismatch” stress is generated in the plane of the substrate. This stress biaxially elongates, i.e., elastically strains, the doped material in the same plane. As the base of a unit cell is strained, so is the height via Poisson distortion. Therefore, the Si:B lattice is vertically contracted as it is horizontally expanded, leading to a smaller vertical lattice constant than the equilibrium value. For thin layers of Si:B, it is energetically favorable for the material to be elastically strained like this, i.e., “pseudomorphic”.

[0024] As the thickness of the doped layer grows, however, the total strain energy per unit area of film increases proportionally, until the layer surpasses a “critical thickness” when it is energetically favorable to introduce dislocations instead of elastically straining the film. Dislocation loops are heterogeneously nucleated at the film surface or film edges and grow larger, gliding towards the substrate-film interface. When a loop meets the interface, the two ends (now called “threading” dislocations because they traverse the thickness of the film) continue to travel away from each other, trailing a line defect at the interface known as a “misfit” dislocation. The misfit dislocations accommodate the lattice-mismatch stress, relieving the horizontal and vertical strains and restoring the in-plane and perpendicular lattice constants to the equilibrium value, i.e., “relaxing” the material. For a low-mismatched lightly strained epitaxial film on a diamond cubic or zincblende substrate, a mesh of orthogonal <110> misfit dislocations is the most likely configuration because of the {111} <110> easy slip system for these crystal structures at elevated temperatures, such as those involved in diffusion and most CVD processes.

[0025] At high enough quantities, the effects of any dissimilar-sized substitutional atom on the silicon microstructure are the same as those of boron. Of course, the impact depends on the relative size and concentration of the substitutional species. Also, incorporation of a larger atom than silicon, e.g., germanium, would result in compressive stress and strain rather than a tensile situation like Si:B.

[0026] In the conventional etch stop process, extremely high concentrations of boron are needed to achieve a high etch rate selectivity. These very high boron concentrations lead to dislocation introduction in the thick films that are desired in many MEMS applications. Since the p++ process is created usually through a diffusion process, there is a gradient in dislocation density and a gradient in the boron concentration. Because the etch stops in the boron concentration gradient, the thin film part typically possesses large curvature, which is compensated for by an annealing treatment. In addition, the etch stop selectivity is extremely sensitive to the boron concentration. If the concentration falls below the critical 7×10¹⁹cm⁻³, the selectivity is drastically different. Thus, since this boron concentration is near the solubility limit, dopant concentration fluctuations in the vertical and lateral dimensions produce low yields in MEMS processes. The SiGe etch stop breaks the link between dopant concentration and etch selectivity. Also, since the SiGe alloy is a miscible alloy system, there is continuous complete solubility between Si and Ge.

[0027] The theory of anisotropic etching of silicon as described by Seidel et al., J. Electrochem. Soc. 137, pp. 3626-31 (1990), incorporated herein by reference, is widely considered the appropriate model. Although specifics like absolute etch rate and dissolution products may differ, the general concept is valid for all anisotropic etchants, as they are all aqueous alkaline solutions and the contribution of the etchant is modeled as nothing more specific than H₂O and OH⁻. Indeed, the existing literature shows consistent behavior among the etchants.

[0028] Early work on etch rate reduction in p++ Si:B presented no hypotheses beyond empirical data. Two possible explanations for the etch-stop phenomenon were proposed: stronger bonding from the high boron concentration or the formation of a boron-based passivation layer. As research accumulated, the etch-stop theories aligned along two credible approaches. The electronic models assign etch-stop behavior to the action of carriers while the passivation models directly attribute it to the formation of a passivating oxide-based layer on the silicon surface.

[0029] Others concluded that the etch-rate decrease is sensitive to hole concentration and not to atomic concentration of boron or stress. They observed an etch rate drop that was proportional to the fourth power of the increase in boron concentration beyond about 3×10¹⁹cm⁻³. Four electrons are required by a red-ox etching process they described, leading them to explain the etch-stop effect in p++ material as an increased probability that the electrons are lost to Auger recombination because of the higher hole concentrations.

[0030] Seidel et al. agreed with the electron recombination hypothesis. They saw the etch rate begin to fall around 2-3×10¹⁹cm⁻³, which agrees with the doping level for the onset of degeneracy, 2.2×10¹⁹cm⁻³. At degeneracy, the Fermi level drops into the valence band and the band-bending is confined to a thickness on the order of one atomic layer. The injected electrons needed for etching are able to tunnel through such a narrow potential well and recombine in the p++ bulk crystal, which halts the etching reaction. The remnant etch rate in the etch-stop regime is attributed to the conduction band electrons, whose quantity is inversely proportional to the hole, i.e. boron, concentration. Four electrons are required to etch one silicon atom, which explains the dependence of the remnant etch rate on the fourth power of the boron concentration.

[0031] It was observed that the formation of an SiO_xpassivation layer on p++ Si:B(2×10²⁰cm⁻³) in aqueous KOH by in situ ellipsometric measurements. In the case of p⁺ −Si, a large number of holes at the surface causes spontaneous passivation with a thin oxide-like layer. The layer is not completely networked like thermal oxide, so it is etched faster and there is still transport of reactants and etch products across the layer, leading to some finite overall etch rate. The lattice strain induced by a high dopant concentration could enhance the layer's growth. Furthermore, the etch rate reduction is not a Fermi-level effect since the phenomenon is exhibited by both heavily doped p- and n-silicon.

[0032] Chen et al., J. Electrochem. Soc. 142, p.172 (1995), assimilated the observations and hypotheses above and their own findings into a composite electrochemical model, where etch stopping is attributed to the enhancement of the oxide film growth rate under high carrier concentration. The key process is hole-driven oxidation at the interface, which inhibits etching by competing with a reaction for Si-Si bonds and hydroxyl radicals, but more importantly, by building the SiO_xbarrier. In p++ silicon, a sufficient quantity of holes for etch-stop behavior is supplied as the converse of the electron action outlined by Seidel et al. That is, instead of electrons thermally escaping the potential well or tunneling through into the bulk crystal, holes from the bulk crystal thermally overcome or tunnel through the potential barrier to the interface. It will be appreciated that this etch-stop process is dynamic, i.e., it is a continuous competition of silicon dissolution and formation/dissolution of the oxide-like layer, whose net result is a nonzero etch rate.

[0033] Germanium is appealing as an etch-resistant additive because it is isoelectronic to, and perfectly miscible in, silicon and diffuses much less readily than dopants and impurities in silicon. Furthermore, the epitaxy of silicon-germanium alloys is selective with respect to silicon oxide, facilitating patterning and structuring, and even affords higher carrier mobilities to electronics monolithically integrated with MEMS.

[0034] Existing germanium-based etch-stop systems are, at best, only marginally suitable for silicon micromachining. In spite of the aforementioned advantages to using germanium, currently there is an inadequate understanding of the etch-stop effect in silicon-germanium materials and no information on anisotropic etching of high germanium content solid solutions.

[0035] Many isotropic etchants for pure germanium exist. Common to all of these is an oxidizer, such as HNO₃or H₂O₂, and a complexing agent to remove the oxide, like HF or H₃PO₄. Early studies were made on isotropic germanium etching by solutions such as “Superoxol”, a commercially available H₂O₂-HF recipe. More recently, investigations have been made on various combinations of HNO₃, HNO₂, HF, H₂SO₄, H₂SO₂, CH₃COOH, H₂O₂, and H₂O.

[0036] In fact, some of these compositions selectively etch germanium or silicon-germanium alloys over silicon, because of differences in the relative oxidation or oxide dissolution rates, but only one etchant exhibits the inverse preference relevant to this project: 100% NH₄OH at 75° C. directly attacks polysilicon at 2.5 μm/hr but polygermanium at only 660 Å/hr. Unfortunately, the selectivity is only about 37, the etch rate for silicon is impracticably slow, and the etch is isotropic, which limits its usefulness in micromachining.

[0037] Previous results with heavy concentrations of germanium in silicon are likewise discouraging with respect to silicon micromachining. Shang et al., J. Electrochem. Soc. 141, p. 507 (1994), incorporated herein by reference, obtained a selectivity of 6 for relaxed, dislocated Si_0.7Ge_0.3:B(10¹⁹cm⁻³) in a KOH-propanol-K₂Cr₂O₇aqueous solution. Yi et al., Mat. Res. Soc. Symp. Proc. 3779, p. 91 (1995), developed a novel NH₄NO₃-NH₄OH etchant with selectivities better than 1000 at 70° C. for 10% and higher germanium alloys. The mixture does not etch pure germanium, but etches pure silicon at 5.67 μm/hr, a weak pace for micromachining purposes. Both systems are isotropic.

[0038] By holding the Si_0.7Ge_0.3:B film under the critical thickness, Shang's team improved the selectivity in the same KOH-propanol-K₂Cr₂O₇solution to about 40. Narozny et al., IEEE IEDM (1988) 563, were the first to use such a “strain-selective” recipe, but only realized a selectivity of 20 (for 30% germanium doped with 10¹⁸cm⁻³boron) and a sluggish etch rate of 1.5 μm/hr at room temperature for pure silicon.²⁶Although the results of Shang et al. and Narozny et al. might have simply been from the well-established etch-stop ability of boron, Godbey et al., Appl. Phys. Lett. 56, p. 374 (1990), achieved a selectivity of 17 with undoped Si_0.7Ge_0.3. None of the articles on strain-selective etchants offer an explanation for the selectivity.

[0039] The anemic etch rate is a grave disadvantage because many MEMS structures can be fairly large compared to typical VLSI dimensions. Moreover, MEMS structures subjected to strain-selective etchants would have to be thinner than the critical thickness. However, as a pseudomorphic structure is released and its strain relieved, the selectivity would deteriorate. A sacrificial strained etch-stop layer could be used, imposing additional process steps and design constraints, but would at least provide advantages over current oxide/nitride sacrificial layers: monocrystallinity can continue above the layer and silicon-gernanium's growth selectivity with respect to oxide adds design/ patterning freedom.

[0040] The consensus of the research community has been that low concentrations of germanium have little or no effect on etch stopping in anisotropic etchants like KOH and EDP. Up to 12% germanium, Seidel et al. detected no significant suppression of etch rate. p++ layers strain-compensated with 2% germanium showed no remarkable differences from those without germanium. By implanting germanium, Feijóo et al., J. Electrochem. Soc. 139, pp. 2312-13 (1992), attained a maximum selectivity of 12 to 24 in EDP at 80° C., corresponding to a dose with a peak concentration of about 0.6%.

[0041] Finne et al., J. Electrochem. Soc. 114, p.969 (1967), however, observed that Si_1−xGe_xsolid solutions with x as small as 0.05 did not etch in an ethylenediamine-pyrocatechol-water (EPW) solution. This discrepancy may be attributed to the use of {111 } wafers, where accurate measurements are difficult because etching in the <111> direction is very slow. No other information has been reported on germanium-rich alloys in anisotropic media.

[0042] Corresponding to the ostensible ineffectiveness of germanium as an etch-stop agent in most publications, there has been little discussion of the source of the limited selectivity that has been detected. Seidel et al. cautioned that their model for heavily-doped boron etch stops is not applicable to germanium because the element is isoelectronic to silicon. They assumed instead that the small reduction of the etch rate is either due to changes in the energy band structure, or else a consequence of the extremely high concentration of lattice defects such as misfit dislocations which could act as recombination centers.

[0043] The invention provides a SiGe monocrystalline etch-stop material system on a monocrystalline silicon substrate. The etch-stop material system can vary in exact composition, but is a doped or undoped Si_1−xGe_xalloy with x generally between 0.2 and 0.5. Across its thickness, the etch-stop material itself is uniform in composition. The etch stop is used for micromachining by aqueous anisotropic etchants of silicon such as potassium hydroxide, sodium hydroxide, lithium hydroxide, ethylenediamine/pyrocatechol/pyrazine (EDP), TMAH, and hydrazine. For example, a cantilever can be made of this etch-stop material system, then released from its substrate and surrounding material, i.e., “micromachined”, by exposure to one of these etchants. These solutions generally etch any silicon containing less than 7×10¹⁹cm⁻³of boron or undoped Si_1−xGe_xalloys with x less than approximately 18.

[0044] Thus, it has been determined that alloying silicon with moderate concentrations of germanium leads to excellent etch selectivities, i.e., differences in etch rate versus pure undoped silicon. This is attributed to the change in energy band structure by the addition of germanium. Furthermore, the nondegenerate doping in the Si_1−xGe_xalloy should not affect the etch-stop behavior.

[0045] The etch-stop of the invention includes the use of a graded-composition buffer between the silicon substrate and the SiGe etch-stop material. Nominally, the buffer has a linearly-changing composition with respect to thickness, from pure silicon at the substrate/buffer interface to a composition of germanium, and dopant if also present, at the buffer/etch-stop interface which can still be etched at an appreciable rate. Here, there is a strategic jump in germanium and concentration from the buffer side of the interface to the etch-stop material, such that the etch-stop layer is considerably more resistant to the etchant. For example, the buffer could grade up to Si_0.85Ge_0.15, then jump to a uniform etch-stop layer of Si_0.7Ge_0.3. Nominally, the composition gradient in the buffer is 5-10% Ge/micron, and the jump in Ge concentration is 5-15 relative atomic percent Ge. The buffer and etch-stop materials are deposited epitaxially on a standard silicon substrate, such as by chemical vapor deposition (CVD) or molecular beam epitaxy (MBE). Note in the above example that the germanium concentration leads to etch stop behavior, and therefore doping concentrations in the etch stop can be varied independently, without affecting etch selectivity.

[0046] With respect to the effect of crystalline defects on the etch-stop behavior, in accordance with the invention using Si_1−xGe_xalloys, the influence of defects is minimal. The use of a graded buffer suppresses the threading dislocation density (TDD) in the top etch-stop layer, which leads to a uniform, nearly defect-free Si_1−xGe_xetch stop.

[0047] The significance of the jump in concentration(s) at the end of the graded region is that the part must be well defined and dimensions well controlled. Thus, a high selectivity should exist between the top etch-stop layer and the end of the graded region for abrupt, predictable etch-stop behavior. A smooth compositional transition from buffer to etch-stop layer would lead to curved edges and greater dimensional variability in the micromachined part, whereas compositional jumps would yield clean, sharp edges and precise dimensions in the released structure. However, if the jump is too large, e.g., greater than ˜20-25 atomic % Ge, the corresponding change in lattice constant, i.e., the lattice mismatch, would create defects.

[0048] The Si_1−xGe_xetch-stop material system, which can be substituted for heavily boron-diffused layers, broadens the spectrum of available etch-stop materials, including undoped (isoelectronic) materials, thus improving the design flexibility for micromachined structures. For example, standard micromachining processes limit the dimensions of silicon sensor structures to a single uniform thickness. Resonant devices for inertial sensing would benefit considerably from more flexible design in which the resonators are thinner than the seismic mass. The invention provides an enabling technology for such a multi-thickness structure. Such a fundamental advantage makes the novel technology widely applicable to the fabrication of MEMS by silicon micromachining.

[0049] A tremendously significant application is the ability to integrate mechanical and electronic devices on the same material. Replacement of the heavily boron-doped etch stop, which is incompatible with integrated circuit (IC) requirements, by isoelectronic and/or moderately-doped etch stops of device quality allows concurrent processing of mechanical devices and associated electronics on the same wafer. Germanium is perfectly miscible in silicon and diffuses much less readily than dopants and impurities. Alloying with germanium also affords higher carrier mobilities to the electronic devices.

[0050] Furthermore, epitaxy of Si_1−xGe_xalloys is selective with respective to silicon oxide, which facilitates patterning and structuring. In addition, defects do not seem to affect the etch-stop efficacy of these materials.

[0051] In developing the germanium-based etch stops of the invention, standard 3″ or 4″ phosphorous-doped (2-4 Ω•cm) or boron-doped (7-10.2 Ω•cm) (001) silicon substrates were used. The wafers were cleaned for 10 minutes in a piranha bath (3:1 95% H₂SO₄in water: 30% H₂O₂in water) and 10 seconds in 4.4% HF and DI water. The substrates were then left in the load lock (˜10⁻⁸Torr) of the vertical UHVCVD reactor overnight. On the following day, the substrates were raised to the lip of the reactor chamber for about two hours to drive off any volatiles, organics, and water. Then the wafers were desorbed of whatever oxide remained by raising them into the 850-900° C. reactor chamber for several minutes. A silicon buffer layer on the order of 1 μm was deposited with SiH₄while the reactor was brought to process temperature. Following this preparation procedure each time, the epitaxial structures were grown in the temperature range 750-900° C. using SiH₄, GeH₄, 1% B₂H₆in H₂, and 1% PH₃in H₂.

[0052] KOH and EDP were used in the etching. KOH is a commonly studied etchant, the simplest and easiest to consider, and relatively easy and safe to use. Although details of absolute etch rate differ, various anisotropic silicon etchants have behaved consistently. Seidel et al.'s well-subscribed theory of anisotropic etching is explicitly etchant-nonspecific. Results, discussions, and conclusions regarding anisotropic etching and etch-stopping of silicon are widely considered to be valid for any anisotropic etchant. Cylindrical etching and patterned oxide masks were both used to determine the efficacy of Ge concentration on etch rate.

[0053] To test the utility of the relaxed epitaxial SiGe etch stops, epitaxial structures were fabricated: WU_—2, WU_—3, WU_—4, and UHV_—17 as shown in FIGS. 1A-1D. FIG. 1A is a functional block diagram of an epitaxial SiGe etch stop structure 100 (WU_—2) configured on a silicon substrate 102. The structure includes a plurality of relaxed graded layers 104 that vary from Si_0.98Ge_0.02, 5×10²⁰cm⁻³B at the substrate surface, to the top surface layer of Si_0.74Ge_0.26, 10¹⁸cm³P. The thickness of each layer are provided in μm.

[0054] FIG. 1B is a functional block diagram of an epitaxial SiGe etch stop structure 110 (WU_—3) configured on a silicon substrate 112. The structure includes a plurality of relaxed graded layers 114 that vary from Si_0.99Ge_0.01at the substrate surface, to the top surface layer of Si_0.84Ge_0.16.

[0055] FIG. 1C is a functional block diagram of an epitaxial SiGe etch stop structure 120 (WU_—4) configured on a silicon substrate 122. The structure includes a relaxed graded layer 124 of Si_0.66Ge_0.34.

[0056] FIG. 1D is a functional block diagram of an epitaxial SiGe etch stop structure 130 (WU_—4) configured on a silicon substrate 132. The structure includes a plurality of relaxed graded layers 134 that vary from Si_0.97Ge_0.03, 3×10¹⁵cm⁻³B at the substrate surface, to the top surface layer of Si_0.66Ge₃₄, 4×10¹⁶cm⁻³B.

[0057] The compositional grading is known to considerably relax the superficial epitaxial layer while effectively suppressing the TDD. The slow grading rate and generous thickness of these epistructures assure a well-relaxed top film. Thus, the graded buffer enables etching experiments on relaxed, high quality, high germanium content alloys, an etching regime that has never been accessible before. As discussed heretofore, prior research dealt with pseudomorphic Si_1−xGe_xlayers or low concentrations of germanium to minimize dislocations, or heavy germanium alloys saturated with threading dislocations. Hence, the grading technique permits one to use the intrinsic etch-stop properties of Si_1−xGe_xsolid solutions.

[0058] Based on the approximate volume of a cross-sectional TEM sample, a single threading dislocation in a TEM sample represents a TDD of about 10⁸cm⁻². FIG. 2 is a cross-sectional TEM micrograph of structure 110 (WU_—3). The top surface is in the upper right direction. The parallel lines (misfit dislocations) define the graded buffer. No threading dislocations can be found, which confirms high crystalline quality. The blurred vertical bands are “bend contours”, an artifact of TEM, not threading dislocations.

[0059] The absence of threading dislocations in FIG. 2 confirms that structures 110 (WU_—2), 120 (WU_—3), and 130 (UHV_—17), which were processed in virtually identical fashion, contain very few defects. TDDs in such relaxed, graded structures have been shown to be in the range of 10⁵-10⁶cm⁻². By omitting the graded buffer, structure 120 (WU_—4) was intentionally processed to be significantly imperfect, as verified by FIG. 3. FIG. 3 is a cross-sectional TEM micrograph of structure 120 (WU_—4). The top surface is to the right. In contrast to FIG. 2, this film is saturated with threading dislocations, which confirms poor crystalline quality. The misfit dislocations in all four of these samples are buried under such a thick overlayer that they cannot possibly affect etching from the top surface.

[0060] Dopant concentrations of structures 100 (WU_—2) and 130 (UHV_—17) are shown in the graphs of FIGS. 4 and 5 respectively. The dopant concentrations were calculated from the mobilities of pure silicon and pure germanium, as indicated. Since structure 130 (UHV_—17) contains 30% germanium, the true boron content lies somewhere in between, closer to the pure silicon line. Regardless, it is clear that the boron doping does not approach the levels needed for etch stopping. Structure 130 was doped p-type to investigate potential interactions or synergies with germanium that were suppressed in structure 100 by intentional background n-doping.

[0061] The characteristics of these materials (top layer) that are relevant to etching are summarized in the following table.


sample	avg % Ge (EDX)	doping [cm	defect density (TEM)

WU_2	26	10	Low
WU_3	17	None	Low
WU_4	34	None	High
UHV_17	30	4 × 10	Low

[0062] Structure 100 (WU_—2) was used to identify the critical germanium concentration by cylindrically etching and to obtain etch rate values by etching from the top surface.

[0063] The cylindrical etch results of structure 100 (WU_—2), as shown in the graph of FIG. 6A, confirm the etch-stop behavior of germanium and narrowed the threshold germanium concentration to the range of 16-22%. It was ensured that there were no effects from boron by doping the film n-type. The terraces on the left of the graph, defined by the round dots, represent the layers in the epistructure. The left scale reflects the depth of each layer while the right scale relates the nominal germanium concentration of each layer. The arc is the initial groove surface, and the square dots trace the etched surface.

[0064] FIG. 6B is a magnification of the left side of FIG. 6A. It is clear that the etch rate increases dramatically somewhere around 18-20% germanium, suggesting that the critical germanium concentration is in that vicinity.

[0065] The cylindrical etch results of structure 130 (UHV_—17), as shown in the graph of FIG. 7, show the etch accelerating dramatically around 4.8-5 μm depth. The 5% Ge/μm grading rate reasonably assures that the threshold germanium concentration is near 20% germanium. The profiles of each side of the groove are shown. The lower bar marks where the profile begins to deviate from the initial grooved shape. The depth of this point appears to be 4.8-5.0 μm below the top surface.

[0066] The results of the etch rate tests using oxide windows are presented in the following table.


wafer	at % Ge	concentration Ge [cm	etch rate [μm/hr]

WU_2	25.6	1.28 × 10	0.070
WU_3	16.9	8.45 × 10	0.234
WU_4	34.0	1.70 × 10	0.040
UHV_17	30.0	1.50 × 10	0.133

[0067] The etch rate for <100> intrinsic silicon in 34% KOH at 60° C. was taken as 18.29 μm/hr from Seidel et al The experimental data for structures 100 (WU_—2), 110 (WU_—3), 120 (WU_—4), and 130 (UHV_—17) are shown in the table. Normalized by 18.29 μm/hr, they are plotted in the graph of FIG. 8 along with Seidel et al.'s points.

[0068] Some features in FIG. 8 should be emphasized. First, there was appreciably greater variability, both up and down, in the individual etch rates of “good” structure 120 (WU_—4) pieces than of the other good samples, hence the error bar. A comparison of all the data supports the belief that the considerable surface roughness of structure 120 (WU_—4), from lattice-mismatch stress and the high TDD, is probably to blame. Thus, the graded layer has already proven its efficacy since the graded layer samples did not display this problem.

[0069] The shape of the new curve very closely resembles that of EDP-boron curve, adding confidence in the new data as well as implying the existence of a universal etch-stop model. In addition, KOH, a more environmentally friendly etch stop than EDP, appears to be a better etch stop with SiGe alloy than EDP with the conventional p++ etch stop.

[0070] Despite the popular sentiment in the literature, it is indisputable that silicon-germanium alloys with sufficient germanium are exceptional etch stops that rival the most heavily boron-doped materials. Three different etching techniques and two etchant systems, KOH and EDP, conclusively show this. The intersection of the steep portion of the KOH-germanium curve with unity relative etch rate, the so-called “critical concentration” as defined by Seidel et al., appears to be 2×10²¹cm⁻³, i.e., 4%, for germanium. Although this value is about 100 times greater than their “critical concentration” for boron, higher selectivities can theoretically be attained with germanium because there are neither solid solubility nor electrical activity limits.

[0071] The substantial selectivities obtained from the well-relaxed, low-defect sample structures 100 (WU_—2), 110 (WU_—3), and 130 (UHV_—17) indicate that strain, induced by defects or dissimilar atomic radii, is not principally responsible for etch-stop behavior.

[0072] Defects do not play a central role in etch resistance. The excellent results from WU_—2, WU_—3, and UHV_—17, relaxed materials with low TDDs, controvert the speculation that lattice defects serving as recombination centers cause the etch stop behavior with germanium or isoelectronic additives, respectively. Furthermore, a comparison of the etch rate of structure 120 (WU_—4) to the KOH-germanium trendline indicates that even a high TDD does not influence etch stopping dramatically (if at all), nor in a predictable fashion.

[0073] The immediately attractive explanation for germanium's newfound etch-stop potency in silicon is the mechanism outlined by R. Leancu, Sensors and Actuators,A 46-47 (1995) 35-37, incorporated herein by reference. For 15-30% germanium, it seems more logical to interpolate from the bulk properties of pure germanium than to postulate only how germanium influences the properties of otherwise pure silicon. That is, one should give the germanium atom just as much credit as the silicon atom, since it is no longer a dopant, but rather an alloying species in the truest sense. Thus, the silicon-germanium alloys in question should show a palpable influence from the etching characteristics of pure germanium, which include a slow rate in KOH.

[0074] Keeping this simple chemistry approach in mind, a completely miscible binary system like silicon-germanium would display a linear dependence of etch rate versus alloy composition. Even without etch rate data at high germanium concentrations, including pure germanium, FIG. 8 plainly illustrates that such is not the case. Along the same lines, it is unclear why there would be some critical concentration of germanium for an etch-stop effect if the etch rate is simply a consequence of chemical structure, i.e., the proportion of each element. In fact, a nonlinear plot and a critical concentration imply that the etch rate is a function of energy band structure rather than chemical structure.

[0075] On a related note, FIG. 8 shows that the germanium-KOH curve is remarkably similar in shape, but not necessarily slope, to the boron-EDP curve, which ascribes its shape to the electronic etch-stop theory. It is difficult to imagine that the germanium-KOH data would just happen to resemble the boron-EDP data, based on a completely different model that warns of no applicability to germanium. That is, it is highly improbable that the true etch-stop mechanism for germanium is entirely unrelated to the true mechanism for boron when the shapes agree so well.

[0076] There are reasons to consider an energy band model to account for the etch-stop behavior in silicon-germanium solid solutions. First, the Si_1−xGe_xdata resemble the p++ Si:B data, including the critical concentration and power-law dependence of the remnant etch rate, and the p++ Si:B data is explained well by energy band effects. At these quantities, germanium is known to markedly change the band structure of silicon. Furthermore, two possible mechanisms for the etch stop effect of germanium were defects and energy bands. Defect enhanced recombination can be eliminated due to our graded layer approach. Energy band structure is the only other possibility.

[0077] Pure bulk germanium has an energy bandgap, E_g, of 0.66 eV at room temperature, compared to 1.12 eV for pure bulk silicon. Hence, the addition of germanium to silicon reduces the bandgap: unstrained Si_0.7Ge_0.3, the situation for samples WU_—2, WU_—3, WU_—4, and UHV_—17, has an energy gap of approximately 1.04 eV. Germanium also has a smaller electron affinity, χ, than silicon, 4.00 eV versus 4.05 eV. Thus, the incorporation of germanium decreases the electron affinity as well. As germanium is added, the shrinking bandgap and electron affinity reduce the band-bending, the potential well in the conduction band, and the potential barrier in the valence band.

[0078] The height of the potential barrier in the valence band, b, is given by:

[0079] b=(χ−d)+½E_g [1]

[0080] for a generic intrinsic semiconductor, where d is the distance of the Fermi level from E=0, the reference vacuum level. It is understood that the bandgap of Si_1−xGe_xdoes not change perfectly linearly with germanium concentration, but it is not known how electron affinity decreases with increasing germanium content. Nevertheless, if the two functions are approximated as linear, then b is also roughly linearly dependent on germanium concentration.

[0081] Adding germanium to intrinsic silicon also increases the amount of equilibrium electrons and holes, n_iand P_i, respectively, via the decreasing bandgap:

\begin{matrix} n_{i} = p_{i} = {(N_{c} N_{v})}^{\frac{1}{2}} \exp (- \frac{E_{g}}{2 kT}) & [2] \end{matrix}

[0082] where N_cand N_vare the effective density of states in the conduction and valence bands, respectively, k is Boltzmann's constant, and T is temperature. To simplify the description, N_cand N_vwill be assumed to be constant and equal to the values for pure silicon. Again, if E_g's dependence on germanium concentration is considered linear, then p_iis exponentially related to germanium concentration.

[0083] The increased p_iincreases the passivation reaction. For the intrinsic situation, it is assumed that the well/barrier is not sharp enough to allow tunneling. This is especially true for Si_1−xGe_x, with the shallower barrier. Furthermore, the inversion layer at the surface is n-type. Then the supply of holes to the passivation reaction is h, the amount of holes from the bulk that overcome the potential barrier thermally. Thus, h is a Boltzmann activated process:

\begin{matrix} h = p_{i} \exp (- \frac{b}{kT}) & [3] \end{matrix}

[0084] Since p_iis exponentially dependent on germanium content while b is linearly related, h is overall exponentially related to germanium concentration. This can easily be seen by substituting expressions [1] and [2] into [3], yielding:

\begin{matrix} h = {(N_{c} N_{v})}^{\frac{1}{2}} \exp (\frac{- E_{g} - χ + d}{kT}) & [4] \end{matrix}

[0085] where E_gand χ are linearly dependent on germanium content. If a critical hole concentration exists for interrupting the etch process, then a critical germanium concentration will be observed.

[0086] The potential barrier in the valence band increases as the Fermi level moves closer to the valence band, but the hole concentration is significantly increased by p-doping. In fact, the two effects exactly offset each other. In the extrinsic case, the equilibrium hole concentration, p, is defined as:

\begin{matrix} p = n_{i} \exp (\frac{\frac{E_{g}}{2} - E_{F}}{kT}) & [5] \end{matrix}

[0087] E_g/2−E_Fis precisely the change in b when the material is doped. Then, when expression [5] is substituted for p_iin equation [3], E_g/2−E_Fexactly cancels the change in b in expression [3]. Thus, with nondegenerate doping, the value of h never changes from:

\begin{matrix} h = n_{i} \exp (- \frac{b_{i}}{kT}) & [6] \end{matrix}

[0088] where b_iis the height of the barrier in the intrinsic material. Thus, a great advantage of the SiGe etch stop is that the etch selectivity depends only on Ge concentration.

[0089] Test structures of structure 110 (WU_—3), completely undoped material, were fabricated and probed. The structure 110 (WU_—3) did not provide the 'hardest' etch stop available with SiGe alloys because the germanium concentration (15-17%) was near the concentration when etch stop selectivity starts to decrease. The results were very promising as shown in FIG. 9. FIG. 9 is a photograph of a top view of a micromachined proof mass 900. Even at these low Ge concentrations, etched parts like the proof mass in FIG. 9 are possible. Higher Ge concentrations in the uniform layer ( 30%) result in extremely hard etch stops, with selectivities approaching 1000:1.

[0090] It is apparent from cylindrical and top surface etching with EDP and KOH and actual structures micromachined in EDP that relaxed silicon-germanium alloys with sufficient germanium are exceptional etch stops. Selectivities as high as 1000, corresponding to 34% germanium, have been obtained in KOH for the <100> direction. Neither strain nor defects are responsible for these results. High defect density does not influence the etch rate of Si_1−xGe_xdramatically. A plot of relative etch rate versus germanium concentration follows the same shape as p++ Si:B data, including a critical concentration and a power-law dependence of the remnant rate. The etch stop behavior in relaxed SiGe alloys is correlated to changes in band structure, which are solely connected to Ge concentration.

[0091] The extremely high etch selectivities achieved with the SiGe etch stop material system of the invention have immediate applications in forming semiconductor/oxide structures. One method of forming silicon on insulator is to bond a Si wafer to another Si wafer that is coated with silicon dioxide. If one of the wafers is thinned, then a thin layer of Si on silicon dioxide/Si is created. Such structures are useful in low power electronics and high speed electronics since the Si active layer is isolated from a bulk Si substrate via the silicon dioxide layer.

[0092] The main disadvantage of this process is the difficulty in thinning one side of the silicon substrate-silicon dioxide-silicon substrate sandwich. In order to have high reproducibility and high yield, the entire wafer must be thinned uniformly and very accurately. Buried etch stops have been used with little success. Even buried, thin layers of strained SiGe have been used, but as mentioned earlier these etch demonstrate etch selectivities <<100, and therefore are not sufficient. The relaxed SiGe alloys of the invention are ideally suited for this type of etch stop. By bonding a structure 1000 of a graded SiGe layer 1004 and a uniform composition layer 1006 on a silicon wafer 1002 to a structure 1008 having a silicon wafer 1010 coated with silicon dioxide 1012, the etch-stop of the invention can be used to create a very uniform relaxed SiGe alloy on silicon dioxide, which in turn is on a silicon wafer. This process is shown schematically in FIG. 10.

[0093] Once the structures are bonded through, for example, annealing, the silicon substrate 1002 and graded layer 1004 are selectively etched away. The finished structure 1014 is a SiGe-on-insulator substrate. It will be appreciated that the structure 1008 can also be a bulk insulating material, such as glass or a glass ceramic.

[0094] An entire new materials system from which to make highly effective etch stops has been developed. The new system offers many advantages over current technologies. Germanium is isoelectronic to and perfectly soluble in silicon, and hardly diffuses in it. The deposition of silicon-germanium is selective with respect to oxide. Defects do not weaken the etch-stop efficacy. The etch-stop material can be completely undoped, and according to the proposed band structure model, nondegenerate doping does not influence the etch-stop behavior. This affords incredible utility and design flexibility, especially to integration with microelectronics. To this end, germanium would even afford higher carrier mobilities.

[0095] Although the present invention has been shown and described with respect to several preferred embodiments thereof, various changes, omissions and additions to the form and detail thereof, may be made therein, without departing from the spirit and scope of the invention.

Claims

What is claimed is:

1. A monocrystalline etch-stop layer system for use on a monocrystalline silicon substrate, said system comprising a graded layer of Si_1−xGe_xand a uniform etch-stop layer of Si_1−yGe_y, wherein said uniform etch-stop layer of Si_1−yGe_yhas a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³.

2. A monocrystalline etch-stop layer system for use on a monocrystalline silicon substrate, said system comprising a buffer layer graded up to approximately Si_0.83Ge_0.17and a uniform etch-stop layer of approximately Si_0.7Ge_0.3, wherein said uniform etch-stop layer of Si_0.7Ge_0.3has a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³.

3. A method of fabricating a monocrystalline etch-stop layer on a silicon substrate comprising:

depositing a graded buffer layer of Si_1−xGe_xon said silicon substrate; and

depositing a uniform etch-stop layer of Si_1−yGe_yon said graded buffer layer, wherein said uniform etch-stop layer of Si_1−yGe_yhas a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³.

4. A method of fabricating a monocrystalline etch-stop layer on a silicon substrate comprising:

depositing a buffer layer graded up to approximately Si_0.83Ge_0.17on said silicon substrate; and

depositing a uniform etch-stop layer of Si_0.7Ge_0.3on said graded buffer layer, wherein said uniform etch-stop layer of Si_0.7Ge_0.3has a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³.

5. A method of micromachining an integrated device comprising:

depositing a graded buffer layer of Si_1−xGe_xon a silicon substrate;

etching portions of said silicon substrate and said graded buffer layer in order to release said etch-stop layer; and

processing said released etch-stop layer.

6. A method of fabricating a semiconductor structure comprising:

depositing a graded buffer layer of Si_1−xGe_xon a first substrate;

bonding said uniform etch-stop layer to a silicon dioxide film provided on a second substrate; and

removing said first substrate.

7. A monocrystalline etch-stop layer system for use on a monocrystalline silicon substrate, said system comprising an etch-stop layer of Si_1−xGe_x, wherein said etch-stop layer of Si_−xGe_xhas a relative etch rate which is less than approximately the relative etch rate of Si doped with _7×10¹⁹boron atoms/cm³.

8. A method of fabricating a monocrystalline etch-stop layer on a silicon substrate comprising:

depositing an etch-stop layer of Si_1−yGe_yon said silicon substrate, wherein said etch-stop layer of Si_1−yGe_yhas a relatives etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³.

9. A method of micromachining an integrated device comprising:

depositing a etch-stop layer of Si_1−yGe_yon said silicon substrate, wherein said etch-stop layer of Si_1−yGe_yhas a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³;

etching portions of said silicon substrate in order to release said etch-stop layer; and

processing said released etch-stop layer.

10. A method of fabricating a semiconductor structure comprising:

depositing an etch-stop layer of Si_1−yGe_yon a first substrate, wherein said etch-stop layer of Si_1−yGe_yhas a relative etch rate which is less than approximately the relative etch rate of Si doped with 7×10¹⁹boron atoms/cm³;

bonding said etch-stop layer to a silicon dioxide film provided on a second substrate; and

removing said first substrate.

Patent Citations (6)

Patent	Date	Inventor	Cited By
US5013681(A)	1991-05-01	Godbey et al.	Applicant
US5413679(A)	1995-05-01	Godbey	Applicant
US5906951(A)	1999-05-01	Chu et al.	Examiner
US6059895(A)	2000-05-01	Chu et al.	Applicant
EP828296	1998-03-01		Applicant
WO99/53539	1999-10-01		Applicant

Non-Patent Literature (6)

Borenstein, J.T., N.D. Gerrish, M.T. Currie and E.A. Fitzgerald, “A New Ultra-Hard Etch-Stop Layer for High Precision Micromachining”, Proceedings of the 1999 12th IEEE International Conference on Micro Electro Mechanical Systems (MEMs), Jan. 17-21, 1999, pp. 205-210.Applicant
Feijoo, D., J.C. Bean, L.J. Peticolas, L.C. Feldman and W.C. Liang, “Epitaxial Si-Ge Etch Stop Layers with Ethylene Diamine Pyrocatechol for Bonded and Etchback Silicon-on-Insulator”, Journal of Electronic Materials, vol. 23, No. 6, Jun. 1994, pp. 493-496.Applicant
Fitzgerald, E.A., Y.H. Xie, M.L. Green, D. Brasen, A.R. Kortan, J. Michel, Y.J. Mii and B.E. Weir, “Totally Relaxed GexSi1-x Layeres with Low Threading Dislocation Densities Grown on Si Substrates”, Applied Physics Letters, vol. 59, No. 7, Aug. 12, 1991, pp. 811-813.Applicant
“Selective Etching of SiGe/Si Heterostructures,” Chang et al. Journal of the Electrochemical Society. Jan. 1991, No. 1.Applicant
“Silicon-On-Insulator by Wafer Bonding: A Review,” W.P. Mazara. Journal of the Electrochemical Society. Jan. 1991, No. 1.Applicant
“Si/SiGe High-Speed Field-Effect Transistors,” K. Ismail. Electron Devices Meeting, Washington, D.C. Dec. 10, 1995.Applicant

Abstract