PROCESS FOR OBTAINING A NITRIDE LAYER

Abstract

A process for obtaining a nitride (N) layer preferably obtained from at least one of gallium (Ga), indium (In) and aluminium (Al), may include: on a stack including a substrate and at least the following layers successively disposed from the substrate: a creep layer having a glass transition temperature, T.sub.glass transition, and a crystalline layer, forming pads by etching the stack so that each pad includes at least a creep segment formed by at least a portion of the creep layer, and a crystalline segment formed by the crystalline layer; and growing by epitaxy a crystallite on each of the pads and continuing the epitaxial growth of the crystallites so as to form the nitride layer. The epitaxial growth may be carried out at a temperature T.sub.epitaxy, such that T.sub.epitaxy≥k1×T.sub.glass transition, with k1 being 0.8.

Claims

1. A process for obtaining a nitride (N) layer optionally obtained from gallium (Ga), indium (In), and/or aluminium (Al), the process comprising: on a stack comprising a substrate and, successively disposed from the substrate a first layer, as a creep layer comprising a material having a glass transition temperature, T.sub.glass transition, and a second layer, as a crystalline layer, which is crystalline and different from the creep layer, forming pads by etching at least the crystalline layer and at least a portion of the creep layer so that: (i) each pad (1000a-1000e) comprises: (i-a) a first segment, as a creep segment, formed by at least a portion of the creep layer, ; and (i-b) a second crystalline segment, as a crystalline segment, formed by the crystalline layer and surmounting the creep segment, ; and each pad comprises a section whose maximum dimension is in a range of from 10 to 500 nm; and growing by epitaxy a crystallite on at least some of the pads and continuing the epitaxial growth of crystallites at least until coalescence of the crystallites carried by two adjacent pads, so as to form the nitride layer, wherein the growing by epitaxy is carried out at a temperature T.sub.epitaxy, corresponding to formula (1):
T.sub.epitaxy≥k1×T.sub.glass transition   (1), wherein k1≥0.8

2. The process of claim 1, wherein the creep layer comprises: a silicon oxide Si.sub.xO.sub.y, x and y being integers; a glass; a borosilicate glass; or a borophosphosilicate glass (BPSG).

3. The process of claim 1, wherein k1≥1.

4. The process of claim 1, wherein
T.sub.epitaxy≤k2×T.sub.melting min   (2), wherein T.sub.melting min is a lowest melting temperature among melting temperatures of the segments forming the pad, and k2≤0.9.

5. The process of claim 1, wherein the pads comprise a buffer layer surmounting the crystalline layer, the buffer layer comprising a different material from that of the nitride layer.

6. The process of claim 5, wherein the buffer layer is formed by epitaxy deposition on top of the crystalline layer, before the forming of the pads by etching.

7. The process of claim 5, wherein the nitride layer is a gallium nitride layer (GaN), and wherein the pads comprise, before the growing by epitaxy of the nitride layer, a priming layer, surmounting the buffer layer and comprising gallium nitride (GaN).

8. The process of claim 1, wherein the stack comprises, before the forming of the pads by etching, a priming layer, surmounting the crystalline layer, wherein the priming layer comprises the same material as the nitride layer.

9. The process of claim 1, wherein, prior to the forming, the stack comprises an elaborate substrate of silicon on insulator (SOI) type comprising a base substrate surmounted successively by an oxide layer forming the creep layer and a semiconductor layer forming the crystalline layer.

10. The process of claim 1, wherein the creep segment has a height e.sub.220 conforming to formula (3)
e.sub.220≤0.1×d.sub.pad   (3), wherein dpad is a diameter of the pad or more generally a distance edge to edge of the pad taken, at the creep segment and in a direction parallel to a main plane wherein the substrate extends.

11. The process of claim 1, wherein the pads have a height H.sub.pad, and wherein two adjacent pads are spaced by a distance D, conforming to formula (4):
H.sub.pad/D<2   (4)

12. The process of claim 1, wherein the crystalline layer comprises silicon.

13. The process of claim 1, wherein the crystalline layer comprises SiC or Al.sub.2O.sub.3.

14. The process of claim 1, wherein the forming of the pads comprises etching the crystalline layer and etching only a first portion of the creep layer so as to keep a second portion of the creep layer between the pads.

15. The process of claim 1, wherein the forming of the pads is carried out so that
d.sub.crystallite/d.sub.pad≥k3   (5), wherein k3=3, d.sub.pad is a maximum dimension of the section of the pad taken in a direction parallel to a main plane wherein the substrate extends, d.sub.crystallite is a dimension of the crystallite measured in the same direction as d.sub.pad at a time of coalescence of the crystallites.

16. The process of claim 1, wherein each pad has an upper face, and wherein the growing by epitaxy of the crystallites takes place at least in part from the upper face.

17. A microelectronic device, comprising: a continuous nitride (N) layer; pads; and a substrate surmounted by the plurality of pads, wherein each pad comprises: a first segment, as a creep segment, having a glass transition temperature, T.sub.glass, transition, and a second segment, as a crystalline segment formed of a crystalline material, wherein the creep segment and the crystalline segment are successively disposed from the substrate, wherein the continuous nitride (N) layer is entirely supported by the pads, and wherein a creep segment material and a nitride layer material are selected so that:
T.sub.epitaxy≥k1×Tglass transition, wherein k1=0.8, T.sub.epitaxy is a minimum temperature allowing formation of the nitride layer by epitaxy.

18. A light-emitting diode, comprising: the device of claim 17.

19. The process of claim 1, wherein the creep layer comprises SiO.sub.2.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0096] The purposes, objects, as well as the features and advantages of the invention will emerge better from the detailed description of embodiments of the latter which is illustrated by the following appended drawings wherein:

[0097] FIG. 1, composed of FIGS. 1a to 1e, illustrates steps of an exemplary process according to the present invention.

[0098] FIG. 2a schematically illustrates, in top view, the coalescence of two crystallites in the context of a conventional pendeo-epitaxy process.

[0099] FIG. 2b schematically illustrates, in top view, the coalescence of two crystallites in the context of an exemplary process according to the present invention.

[0100] FIG. 3, composed of FIGS. 3a and 3b schematically illustrate, in side view, the coalescence of two crystallites in the context of an exemplary process according to the present invention.

[0101] FIG. 4 is a photo illustrating a pad surmounted by a crystallite, obtained by implementing an exemplary process according to the present invention.

[0102] FIG. 5 is composed of FIGS. 5a and 5b which are photos showing a layer obtained by epitaxy after implementing the process according to the present invention.

[0103] FIG. 6 is composed of FIGS. 6a to 6c which are photos illustrating the result of the implementation of an exemplary process according to the present invention.

[0104] FIG. 7 is composed of FIGS. 7a to 7c which are photos illustrating the result of the implementation of an exemplary process according to the present invention.

[0105] FIG. 8 is composed of FIGS. 8a to 8b which are photos illustrating respectively, in sectional side view and in top view, the result of the implementation of an exemplary process according to the present invention.

[0106] The figures are given by way of examples and are not limitative of the invention. They are principle schematic representations intended to facilitate the understanding of the invention and are therefore not necessarily on the same scale as the practical applications. Particularly, the relative thicknesses of the different layers, segments and crystallites are not representative of reality.

DETAILED DESCRIPTION OF THE INVENTION

[0107] It is specified that in the context of the present invention, the terms “on”, “surmounts”, “covers” or “underlying” or their equivalents do not mean “in contact with”. Thus, for example, “depositing a first layer on a second layer” does not necessarily mean that the two layers are in direct contact with each other, but this means that the first layer at least partially covers the second layer by being either directly in contact therewith or by being separated therefrom by at least one other layer or at least one other element including air.

[0108] In the following description, the thickness or the height is taken in a direction perpendicular to the main faces of the different layers. In the figures, the thickness or the height is taken vertically.

[0109] Similarly, when indicating that an element is located in line with another element, this means that these two elements are both located on the same line perpendicular to the main plane of the substrate, that is to say on the same line oriented vertically on the figures.

[0110] A layer, “based” on a material A, means a layer comprising only this material A or comprising this material A and possibly other materials, for example doping elements

[0111] In the following description, the terms crystals and crystallites will be considered as equivalent.

[0112] The general principle of the process according to the invention will now be described with reference to FIGS. 1a to 1e.

[0113] As illustrated in FIG. 1a, a stack is provided comprising at least one substrate 100, successively surmounted by a creep layer 200 and a crystalline layer 300. Thus the creep layer 200 is disposed between the substrate 100 and the crystalline layer 300.

[0114] According to an exemplary embodiment, the substrate 100 is based on amorphous or crystalline silicon. It ensures the mechanical strength of the stack.

[0115] The crystalline layer 300 has a lower face opposite the creep layer 200 and an upper face whose function is to serve as a base layer to epitaxially grow the nitride layer desired to be obtained in the end. For example, the layer desired to be obtained in the end is a layer 550 of gallium nitride GaN. According to an exemplary embodiment, the crystalline layer 300 is based on monocrystalline silicon.

[0116] Preferably, the creep layer 200 is made of a viscous material. It is made of an amorphous material such as an oxide, preferably a silicon oxide SixOy, such as SiO2. The role of this layer will be explained in the following description.

[0117] The creep layer 200 has a glass transition temperature. Therefore, it has a glass transition and has the behaviour of glass transition materials. Like all materials with a glass transition temperature, the creep layer 200 under the effect of a rise in temperature, deforms without breaking and without returning to its initial position after a drop in temperature. On the contrary, the crystalline layer 300 naturally does not have a glass transition. When a temperature threshold is reached, the crystalline layer deforms, then dislocates and can break.

[0118] In an advantageous but non-limiting manner, this stack constitutes a substrate of the semiconductor on insulator type, preferably silicon on insulator (SOI). In this case, the creep layer 200 is formed by the buried oxide layer (BOX) of the SOI substrate.

[0119] According to an advantageous exemplary embodiment illustrated in FIG. 1b, a buffer layer 400 is deposited by epitaxy on the upper face of the crystalline layer 300. When the layer 550 desired to be obtained in the end is a GaN layer and the crystalline layer 300 is a silicon-based layer, this buffer layer 400 is typically made of aluminium nitride (AlN). This allows to avoid the phenomenon called “Melt-back etching”, generated by the very high reactivity between silicon and gallium at the usual epitaxy temperatures (1000/1100° C.) and which leads to very strongly degrading the GaN layer.

[0120] Typically, the thickness of the AlN layer is comprised between 50 and 200 nanometres (10.sup.−9 metres).

[0121] Consequently, the creep layer 200 and the crystalline layer 300 are different. The creep layer 200 is not crystalline.

[0122] As illustrated in FIG. 1b, a primer layer 500 can also be deposited by epitaxy, on the upper face of the buffer layer 400. This primer layer 500 has the function of facilitating the resumption of growth of the crystallites 510 during the following steps. In this case, it is from an upper face of the primer layer 500 that the growth of the crystallite (510a-510e) by epitaxy at least partially occurs. This primer layer 500 is preferably made of the same material as the layer 550 desired to be obtained in the end. Typically, when the layer desired to be obtained in the end is a layer 550 of gallium nitride GaN, the primer layer 500 is also made of GaN. Typically, it has a thickness comprised between 50 and 200 nanometres.

[0123] It will be noted that layers 400 and 500 are only optional.

[0124] According to embodiments not illustrated in FIG. 1, it is possible to provide only the buffer layer 400 or only the primer layer 500, or even none of these two layers 400 and 500.

[0125] Regardless of the embodiment selected, that is to say with or without primer layer 400 and with or without buffer layer 500, the growth by epitaxy of the crystallite 510a-510e, takes place at least in part or only from the upper face of the pad 1000a-1000e. Thus, this upper face is formed either by the crystalline segment 300a-300b, or by the segment formed by the primer layer 400a-400b, or by the segment formed by the buffer layer. This allows in particular to rapidly obtain crystallites of significant thickness.

[0126] It will be noted that the upper faces of the buffer layer 400 and the primer layer 500, that is to say the faces opposite the nitride layer 550 desired to be grown, have polarities of Gallium (Ga), and not nitrogen (N) type, which considerably facilitates the production of a high-quality epitaxial nitride layer 550.

[0127] As illustrated in FIG. 1c, pads 1000a-1000e are formed from the stack. These pads are obtained by etching the stack into the creep layer 200.

[0128] To form the pads by etching, numerous etching techniques known to the person skilled in the art can be used. In particular, conventional lithography techniques can be used, such as photolithography techniques comprising the formation of a mask, for example made of resin, and then the transfer of the patterns of the mask into the stack. E-beam lithography techniques or nanometric printing techniques can also be used.

[0129] These pads 1000a-1000e are small and can be called nano-pads. Typically, the maximum dimension of the section of the pads is comprised between a few tens and a few hundred nanometres, more precisely between 10 and 500 nanometres and preferably 100 nm. This maximum dimension of the section of the pads is referenced d.sub.pad in FIG. 1c. If the pads are of circular sections, this maximum dimension d.sub.pad corresponds to the diameter of the pads. If the pads are of hexagonal section, this maximum dimension d.sub.pad corresponds to the diagonal or to the diameter of the circle passing through the angles of the hexagon.

[0130] According to an exemplary embodiment, the segments have a height e220 and two adjacent pads are spaced by a distance D, such that:

e220/D<1, and preferably e220/D<1.5. Preferably e220/D<2.

[0131] According to an exemplary embodiment, the pads have a height H.sub.pad and two adjacent pads are spaced by a distance D, such that:

H.sub.pad/D<2, and preferably H.sub.pad/D<1.5. Preferably H.sub.pad/D≤1.

[0132] The distance D and the height H.sub.pad are referenced in FIG. 1d.

[0133] As illustrated in FIG. 1c, the pads are etched through the entire primer layer 500, the entire buffer layer 400 (when the latter are present), the entire crystalline layer 300. Preferably, only a portion 220 of the creep layer 200 is etched. This embodiment has the advantage of avoiding that during epitaxy the nitride layer 550 develops on the creep segments 220. This selectivity of epitaxy is encountered in particular when the nitride layer 550 which is grown by epitaxy is GaN and the creep segments are in SiO2. On the contrary, if, with these same materials, the creep layer 200 is etched over its entire thickness, then during epitaxy, the nitride layer develops from the upper face of the substrate 100, usually formed of silicon. This situation is obviously not desirable

[0134] Furthermore, it has been observed that the fact of keeping an unetched portion 210 of the creep layer 200 allows to facilitate the creeping of the segment 220, particularly when the crystallites are disoriented in twist, that is to say in the plane of the nitride layer 550 desired to be obtained.

[0135] Preferably, the etched thickness e220, and therefore forming the height of the creep segment 220, is equal to half the thickness of the creep layer 200. This allows to have a very good reorientation of the crystallites during the formation of grain joints.

[0136] FIG. 1d illustrates the formation of crystallites 510a-510e by epitaxial growth from the primer layer 500 (or from the upper face of the crystalline layer 300 when the layers 400 and 500 are absent).

[0137] As illustrated in this FIG. 1d, the pads 1000a-1000e each comprise a crystallite 510a-510e carried by a stack of segments 400a-400e, 300a-300e, 220a-220e. The segments extend in the main direction of extension of the pad, that is to say vertically in FIGS. 1a to 1e.

[0138] The segments form discs if the section of the pads is mainly circular. If the section of the pads 1000a-1000e is polygonal, for example hexagonal, the segments then form cylinders of hexagonal section. Preferably, the segments are full. In other words, and as clearly illustrated in the figures, the section of the pads is taken along a plane parallel to the planes wherein the creep layer 220 and the crystalline layer 300 mainly extend. Thus, this section is taken along a plane perpendicular to the plane of the sheet comprising FIGS. 1a to 1e and along a plane parallel to the plane of the sheet comprising FIGS. 2a and 2b.

[0139] The growth continues and extends laterally. The crystallites 510a-510e develop until they coalesce and form a layer 550 as illustrated in FIG. 1e.

[0140] In other words, and as clearly appears from the figures, the layer 550 extends between several pads. It is continuous.

[0141] This growth of crystallites 510a-510e does not extend downward. Moreover, this growth is selective in that it does not take place on the creep layer 200 typically made of an oxide. In this sense, the growth of the crystallites 510a-510e takes place according to the principle of pendeo-epitaxy.

[0142] It will be noted that it is particularly advantageous to etch the pads 1000a-1000e after formation by epitaxy of the buffer layer 400 and the primer layer 500 (when these layers are present). Indeed, if one of these layers 400, 500 was deposited after etching, it would partly be formed at least between the pads 1000a-1000e on the upper face of the creep layer 200. In the case where the nitride layer 550 is GaN, the creep layer 200 is SiO.sub.2, then, at the epitaxy deposition temperature, the epitaxial growth of the layer 550 would not take place selectively but would on the contrary also take place between the pads 1000a-1000e, which of course is not desirable.

[0143] In a particularly advantageous manner, the temperature T.sub.epitaxy at which the epitaxy is carried out is greater than or of the order of the glass transition temperature T.sub.glass transition of the creep layer 200. Thus, during epitaxy, the creep segments 220a-220e are brought to a temperature which allows them to deform.

[0144] Consequently, if the crystallites 510a-510b carried by two adjacent pads 1000a-1000b are disoriented relative to each other, during the coalescence of these two crystallites, the joint 560 formed at their interface, usually designated grain joint or coalescence joint, will be formed without dislocation to make up for these disorientations. The deformation of the creep segments 220a-220e thus allows to make up for these disorientations.

[0145] As illustrated in FIGS. 2a and 2b, the disorientation of the two crystallites 510a-510b can result from a misalignment in the plane (that is to say around axes contained in the plane of FIGS. 1a-1e), this is referred to as twist.

[0146] FIG. 2a, which shows a top view of the crystallites 510a-510b, illustrates this type of misalignment with the presence of an angle a between the crystallites. This FIG. 2a schematically illustrates that, with the conventional processes, the joint 560 which is formed during the coalescence creates dislocations 561 to compensate for the misalignments.

[0147] FIG. 2b very schematically illustrates the formation of the coalescence joint 560 by implementing the process according to the invention. This FIG. 2b shows the twist deformation 570a, 570b of the creep segments 220a, 220b allowing to form a coalescence joint 560 without dislocation. In this case, making up for this deformation by the creep segments 220a-220b leads to a twist of the creep segments 220a-220b around their main axis. Each segment rotates in an opposite direction as illustrated in the right view of FIG. 2b.

[0148] As illustrated in FIGS. 3a and 3b, the defect in orientation of the two crystallites can also result from a misalignment out of the plane (that is to say around axes perpendicular to the plane of FIGS. 1a-1e), this is referred to as tilt. In this case, making up for this deformation by the creep segments 220a-220e leads to a tilting of the pads.

[0149] FIG. 3a illustrates by the angle β the defect of angular orientation between the crystalline planes of the crystallites 510a and 510b before coalescence. FIG. 3b very schematically illustrates the deformation of the creep segment 220 allowing to form a coalescence joint 560 without dislocation. It has been assumed in this diagram that the disorientation is distributed equally between the two crystallites 510a and 510b, that is to say β/2. The diagram is also applicable in the case where the crystalline disorientations are different in angle in the two adjacent crystallites and whose sum is equal to β.

[0150] FIGS. 2 and 3 are not intended to give a visual representation close to reality concerning the deformation of the creep segments. These figures are intended to give a schematic representation to easily understand the operating principle of the invention.

[0151] The following paragraphs are intended to provide additional explanations and details regarding the principle of operation of the invention.

[0152] Modelling and Explanation of the Principle of the Invention

[0153] The paragraphs below aim at explaining more precisely the phenomena at the origin of the invention and which allow to obtain an epitaxial layer without or with very little dislocation.

[0154] When two grains or crystallites are rigidly linked to the substrate on which they have been deposited and are slightly disoriented relative to each other, the energy of formation of their coalescence joint is high (the energy can be fairly easily calculated since it is a first approximation of a network of interfacial dislocations. For this purpose, reference can be made to the following publication: J. P. Hirth and J. Lothe “Theory of dislocations”, John Wiley and sons (1982).

[0155] On the other hand, if the crystallites 510 rest on pads 1000a-1000e as is the case in pendeo-epitaxy, and these pads 1000a-1000e can be deformed, by twisting or tilting, the disorientation of the crystallites 510 (for example GaN) relative to each other is “transmitted” to the pads carrying these crystallites 510. In the case of a simple twist, as illustrated in FIG. 2b, the phenomenon can be modelled by saying that the attractive forces which cause the crystallites to rotate to join and form an “internal” surface result in a twisting torque at the nano-pads 1000a-1000e. This twisting torque will be even greater as the ratio between the size of the pads 1000a-1000e and the size of the crystallites is small. If the deformation of the pads 1000a-1000e is easy, the coalescence joint 560 then simply transforms into an internal surface for which the “perfect” atomic bonds will be formed. Energetically speaking, this will only be possible if the energy released during coalescence by the formation of an “internal surface” is greater than the energy that must be spent so that the pads 1000a-1000e themselves deform to absorb the initial disorientations between the adjacent crystallites 510a-510b. The fact that the underlying pad is a pedestal made of SiO2 is favourable since SiO2 creeps at the growth temperature. SiO2 is a material which is effectively well adapted for this process, but this does not exclude the use of other materials, such as glasses which can creep at equivalent or lower temperatures, such as borosilicate glasses or borophosphosilicate glasses (BPSG), also used for the manufacture of SOI.

[0156] In the case of GaN crystallites of hypothetically cubic shape and of 500 nm side, carried by pads of circular section of 100 nm diameter, the energy released during coalescence can be expressed as the difference between the internal cohesion energy and the surface energies of the two surfaces which will be contacted. The surface energy values are of the order of 0.1 to 0.2 eV/Angström.sup.2 (for this purpose reference can be made to the following publication: C. E. Dreyer, A. Janotti, and C. G. Van de Walle “Absolute surface energies of polar and nonpolar planes of GaN”, PHYSICAL REVIEW B 89, 081305(R) 2014). The cohesion energies are, in turn, in the order of 11 eV (Ga—N bond).

[0157] As a first approximation, starting from a hexagonal surface (type 0001) a value of 36 eV per unit cell (three Ga-N bonds per unit cell) is obtained, namely, for the hexagonal unit cell considered (surface 26 A.sup.2), an energy of cohesion per surface of 1.35 eV/Angstrom.sup.2. The approach was simplified by considering surfaces to be contacted of type c (0001), but the corrections related to the crystal orientation of these surfaces (number of bonds per unit cell and surface energies) are of the second order and only slightly modify this numerical evaluation.

[0158] Then a “released” energy (final state−initial state) upon contacting by forming covalent bonds of 1.35−0.3=1.05, or about 1 eV per Angström.sup.2, is obtained. This is obviously independent of the disorientation between the crystallites 510 to be contacted. If crystallites of a cubic shape (to simplify the estimation as in the diagrams in FIG. 2b) and of 500 nm side are considered, then a released energy of 25 10.sup.E6×1.6 10.sup.E-19, or of the order of 4 10.sup.E-12 Joules, is obtained. This is to be compared to the energy required to deform the pads 1000a-1000e made of SiO2 under the crystallites as will be detailed below.

[0159] At the GaN epitaxy temperature (about 1100° C.), the silicon oxide has a visco-plastic behaviour with a viscosity which depends on both the temperature and the mechanical stress undergone by the material.

[0160] As the temperature increases, the viscosity of SiO2 decreases, like any material having a glass transition. Above a certain stress, the viscosity of SiO2 also decreases when the mechanical stress increases. In other words, unlike a conventional fluid wherein the plastic deformation rate is proportional to the shear stresses (viscosity independent of the stress), beyond a threshold value called breaking stress, the SiO2 deformation rate increases exponentially with stress. In practice, at 1100° C., the SiO2 breaking stress is around 500 MPa: at this temperature, the viscosity of SiO2 is therefore independent of the stress up to stresses of 500 MPa. Beyond this, the viscosity drops by several orders of magnitude and a very rapid plastic deformation then gives a residual stress close to the ultimate stress. This residual stress then decreases much more slowly by conventional creep (several minutes to several hours).

[0161] Before alignment, the GaN crystallites 510 have a disorientation relative to each other of an average of 1 to 6°. Before the neighbouring crystallites are bonded together to form covalent atomic bonds, the interatomic forces acting between the crystallites are the Van der Waals forces which are attractive at very short range. This results in the existence of a rotational torque applied to the contact point to align them. After a very slight rotation of a fraction of a degree, the breaking stress of 500 MPa is reached in the creep segment 220a-220e under the pads 1000a-1000e. Creep then becomes almost instantaneous in the oxide and the rotational torque required for rotation is constant during the end of the alignment phase.

[0162] If the geometry of FIG. 2b is used again, i.e. a crystallite 510a-510b of cubic shape of 500 nm side and centred on a pad 1000a-1000b of smaller circular section of 100 nm side. The complete coalescence of the crystallites 510 is carried out in typically a few hundred seconds. If 1 s is taken as the characteristic time for contacting the crystallites 510, a numerical model shows that to turn the pad by 2° in about one second, a mechanical work of the order of 0.5.sup.E-20 to 1.sup.E-20 Joules must be provided, taking into account the failure of the oxide at high temperature. The energy released by contacting the two surfaces, of the order of 10.sup.E-11 to 10.sup.E-13 Joules, is therefore very much greater than the mechanical energy of deformation of the underlying pads 1000a-1000e. This energy is also of several orders of magnitude smaller than that which would have been necessary to align the pads constructed on solid silicon (bulk) or to align the pads formed by a material having no glass transition phase around the temperature of epitaxial growth of crystallites 510.

[0163] The considerations in the above paragraphs show that to obtain coalescence of the crystallites without dislocation, the following parameters can be adjusted: [0164] The “mechanical fracture” properties of the material forming the creep segment at high temperature under relatively low stresses of 500 MPa. [0165] The sufficiently small size of the support pad 1000a-1000b compared to the distance D between the pads, allows to create a stress in the creep segment which is, for a given rotational torque, greater than the ultimate stress.

[0166] Moreover, as indicated above, it will be ensured that the epitaxy temperature T.sub.epitaxy makes the creep of the creep segment 220a-220b possible. In practice, T.sub.epitaxy≥600° C. (in the context of molecular beam epitaxy), T.sub.epitaxy≥900° C. and preferably T.sub.epitaxy≥1000° C. and preferably T.sub.epitaxy≥1100° C. These values allow to reduce in a particularly effective manner the defects in the epitaxial layer when the creep layer is made of SiO2. In practice, T.sub.epitaxy≤1500° C.

[0167] In order to facilitate the formation of coalescence joints 560 without dislocation, it will be preferable to apply the following conditions:

T.sub.epitaxy≥k1.Math.T.sub.glass transition, with k1=0.8, preferably k1=1 and preferably k1=1.5.

[0168] According to an exemplary embodiment, T.sub.epitaxy≤k2.Math.T.sub.melting min, T.sub.melting min being the lowest melting temperature among the melting temperatures of the segments forming the pad. This is mainly the crystalline segment and the creep segment. According to an exemplary embodiment, k2=0.9. This allows to avoid a diffusion of the species of material with the lowest melting temperature.

[0169] Thus, in the case where the pad is formed of SiO2 creep segments and of silicon crystalline segments, T.sub.epitaxy≤1296° C. Indeed, T.sub.melting min is equal to the melting temperature of silicon since the melting temperature of silicon is equal to 1440° and the melting temperature of SiO2 is equal to 1970° C.

[0170] According to an exemplary embodiment, the height e.sub.220 of the creep segment is such that e220≥0.017 d.sub.pad. Preferably, e220≥0.05 d.sub.pad. Preferably, e220≥0.1 d.sub.pad. Preferably, e220≥1 d.sub.pad. These values, allow to obtain sufficient deformation to reduce the stresses at the grain joint.

[0171] The pads 1000a-1000e have a height H.sub.pad, and two adjacent pads 1000a-1000e are spaced by a distance D, such that: H.sub.pad/D<2 and preferably H.sub.pad/D≤1.

[0172] Advantageously, the step of forming the pads 1000a-1000e is carried out so that d.sub.crystallite/d.sub.pad≥k3, d.sub.pad being the maximum dimension of the section of the pad 1000a-1000e taken in a direction parallel to a main plane wherein the substrate 100 extends. d.sub.crystallite corresponding to the dimension of the crystallite measured in the same direction as d.sub.pad at the time of the coalescence of the crystallites 510a-510e.

[0173] According to an example 100≥k3≥1.1. Preferably, 50≤k3≥1.5. Preferably, 5≥k3≥2.

[0174] According to an example k3≥3, preferably 100≥k3≥3. Preferably 50≥k3≥3. Preferably 5≥k3≥3.

[0175] This feature allows the creep segments to deform in order to absorb in a particularly effective manner the mechanical stresses which arise when two adjacent crystallites begin to coalesce. Thus, this feature effectively contributes to reducing the density of defects within the nitride layer obtained in the end.

Exemplary Embodiments

[0176] FIG. 4 is a photo taken with a scanning electron microscope (SEM). It allows to see a stack comprising successively the substrate 100, the unetched portion 210 of the creep layer, the creep segment 220, the crystalline layer 300 as well as a crystallite 510, here made of GaN. This crystallite 510 is pyramidal in shape due to the lower growth rate of the semi-polar planes which form the facets of the pyramid. In this example, the size of the pad is 100 nm.

[0177] Empty spaces 230 can be seen on either side of the creep segment 220.

[0178] To ensure the coalescence of the pyramids and obtain two-dimensional growth, it may be advantageous to proceed as follows. In a first step, pyramids are made on each of the nano-pads 1000a-1000e using given growth conditions until the pyramids adjoin. In a second step, these growth conditions are modified to induce lateral growth. This two-step growth procedure is for example described in the following publication: Shields and al. 2011, Nanopendeo coalescence overgrowth of GaN on etched nanorod array Phys. Status Solidi C 8, No. 7-8, 2334-2336 (2011).

[0179] FIGS. 5a and 5b give examples of a layer 550 obtained according to this process for a matrix of 40×50 pads. In this example, the size of the pads is of 500 nm and their pitch is of 1 μm. This pad dimension allows to very significantly reduce the number of dislocations appearing on the crystallite of a pad. Moreover, this pad size allows to optimise the deformation and the transmission of the torque during coalescence.

[0180] In a particularly advantageous manner, it is clearly observed that the layer 550 has separated from the underlying substrate 100. Delamination of the layer 550 desired to be obtained in the end is therefore particularly easy. This automatic delamination can be complete or partial. It will be described in more detail below.

[0181] It is thus possible to produce self-supported GaN plates (or templates or platforms), in this example of size 40×50 μm.

[0182] It is in particular possible to adapt the number of pads and the period so that the size of these plates is compatible with the manufacture of diodes with specific dimensions and which can be very easily transferred to any type of support (electrical/thermal conductor/optical reflector . . . ).

[0183] In these FIGS. 5a and 5b, it is also noted that the coalescence joints 560 form without dislocation and that no dislocation is present at the crystallites 510.

[0184] FIGS. 6a to 6c and 7a to 7c are photos showing matrices 580 of GaN crystallites 510 forming pyramids and obtained after growth on the pads. The pads were produced by electron beam lithography (e-beam) from a stack comprising the following layers: [0185] creep layer 200: SiO2 [0186] crystalline layer 300: Si [0187] buffer layer 400: AlN [0188] primer layer 500: GaN

[0189] In the example illustrated in FIGS. 6a to 6c, the pads have a diameter of 200 nm and are spaced by 1 μm apart. In the example illustrated in FIGS. 7a to 7c, the pads have a diameter of 500 nm and are spaced by 2 μm apart. The top of the crystallites 510 is flat in this example. This top is referenced 515 in the figures.

[0190] Characterisations of Coalescence Defects

[0191] FIG. 8a is a photo taken with a transmission electron microscope (TEM) of a section of GaN crystals 510 having coalesced. In this figure, first of all it is noticed that most of the dislocations 561 resulting from re-growth are aligned in the growth plan at the start of growth.

[0192] Moreover, it clearly appears that the coalescence of the crystallites 510 coming from adjacent pads is done with very few coalescence defects at the coalescence joint 560. This is to be compared with the conventional methods for which the coalescence generates threading defects, that these conventional methods are based on a lateral overgrowth illustrated in FIG. 9 and as described in the publication E B Yakimov and A. Y Polyakov, 2015 Ebic investigation of dislocations in FLOG GaN, Phys. Status Solidi, or that they are based on a pendeo type overgrowth from one-dimensional structures as described in the publication mentioned above Shields and Al. 2011, Nanopendeo coalescence overgrowth of GaN on etched nanorod array Phys. Status Solidi C 8, No. 7-8, 2334-2336 (2011).

[0193] It is also noted in FIG. 8a that the pads 1000a-1000e, at half their height, are split, which will allow easy delamination of the epitaxial layer 550 on the matrices 580 of 1000a-1000e, to make templates or pseudo-substrates for the subsequent overgrowth of active layers, as explained above.

[0194] One possible interpretation is that this cracking of the pads 1000a-1000e is due to plastic deformation (under the effect of the twisting stresses of the pads 1000a-1000e) when the crystallites 510 coalesce at the epitaxial growth temperature. As indicated above, this decoupling of the epitaxial layer 550 relative to the substrate 100 during growth is a considerable advantage since no curvature of the plates thus formed is possible when cooling.

[0195] FIG. 8b confirms the absence of defects in the coalescence between two crystals, for example of pyramidal shape, using the cathodoluminescence technique. This technique allows performing spatial mapping of the emission of photons by the considered material. The image of panchromatic cathodoluminescence in FIG. 8b shows that, apart from the emission variations related to the extraction of light from different facets (or ridges), at the connection between two crystals, there is no non-radiative re-combinations, which demonstrates the absence of structural defects in the region of the coalescence joint.

[0196] In view of the above description, it clearly appears that the present invention provides a particularly effective solution for reducing the density of defects in a layer of epitaxial nitride, in particular from pads etched in a stack of SOI type.

[0197] The invention is not limited to the embodiments described above and extends to all the embodiments covered by the claims.