Method for thinning the surface layer of an SOI substrate

Abstract

The invention relates to a method for thinning a surface layer of an SOI substrate, the method comprising: a) a step of measuring the thickness of the surface layer in order to obtain its mean thickness (Emoy) and its thickness non-uniformity (U), followed by the steps, carried out in a single-wafer chemical treatment device, of: b) applying n cycles, where n is an integer greater than or equal to 1, of oxidation and etching to a front face of the surface layer, wherein each cycle involves dispensing ozonated water followed by dispensing hydrofluoric acid; c) performing inhomogeneous etching of the front face of the surface layer, involving dispensing a silicon etching solution in a central region for a time referred to as the non-uniform etching time (tnu); d) performing homogeneous etching of the front face of the surface layer involving dispensing a silicon etching solution in a sweeping motion from the edge to the centre of the surface layer and vice versa, for a time referred to as the uniform etching time (tu). Since the mean thickness (Emoy) and the thickness non-uniformity (U) are known, the number of cycles (n), the non-uniform etching time (tnu) and the uniform etching time (tu) are determined, prior to carrying out steps b), c) and d), on the basis of a model that relates them to an etched mean thickness of the surface layer and to a mean etching non-uniformity defined by the difference between an etched mean thickness in a central region and in a peripheral region of the surface layer.

Description

METHOD FOR THINNING THE SURFACE LAYER OF AN SOI SUBSTRATE FIELD OF THE INVENTION

[0001] The present invention relates to the field of semiconductors and in particular to SOI (Silicon on Insulator) substrates. The invention relates to a method for thinning the surface layer of an SOI substrate in a single-wafer chemical processing equipment, optimized to achieve a surface layer non-uniformity of less than or equal to + / -0.4nm.

[0002] TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0003] More and more applications based on SOI substrates require very good thickness uniformities of the silicon surface layer (also called active layer thereafter). For example, in digital applications, the active layer of FDSOI (fully depleted SOI) substrates must have very low thickness variations because these latter impact the threshold voltage of the transistors developed in and / or on the active layer; in photonic applications, the performances of devices such as filters or modulators are also strongly influenced by the thickness non-uniformities of the active layer of the SOI substrate.

[0004] Thickness and uniformity specifications therefore become very aggressive: for active layers with thicknesses typically less than 50nm, we expect non-uniformities within wafer (WiW) and between wafers (WtW) typically less than + / -0.4nm, i.e. all measured points of the layer are within + / -0.4nm of its average thickness, considering a peripheral exclusion zone of a few mm. Such uniformities are difficult to achieve because the succession of SOI substrate manufacturing steps piles up contributions to the non-uniformity of the surface layer.

[0005] A known solution for correcting the thickness non-uniformities of the active layer is to carry out a localized etching of said layer, by plasma etching processes ("plasma etch") as described for example in document US20140234992, or by ion beam etching processes ("cluster ion beam etch") as described in particular in document WO2013003745. This type of solution nevertheless has a drawback: the etching of the surface of the active layer creates a superficial layer of amorphous silicon likely to generate electrical problems and which must therefore be removed. The removal of the amorphous layer leads to an increase in the surface roughness, which degrades the performance of the device developed on the active layer.

[0006] Document WO2004015759 proposes an alternative solution, implementing localized sacrificial thermal oxidation, punctually consuming a more or less significant thickness of the active layer, so as to correct its thickness non-uniformities. The disadvantage of this approach is that a local temperature gradient is not easy to introduce into a silicon layer: the resolution of the non-uniformity correction can therefore be limited.

[0007] Document WO2013175278 also aims to improve the thickness uniformity of the surface layer of an SOI substrate, by implementing a SC1 type chemical etching solution: the quantity of solution dispensed, the application time and / or the temperature are adjusted according to the thickness of the layer to be etched.

[0008] Document EP3200219 describes a solution for thinning the active layer of an SOI substrate, in single-plate etching equipment, providing a non-uniformity on the plate of the order of + / -0.5nm.

[0009] It remains interesting to improve the thinning processes of the active layer of an SOI substrate, to ensure excellent thickness uniformities (better than + / -0.5nm, or even + / -0.4nm), for very low thicknesses and for SOI substrate diameters of at least 200mm, or even 300mm.

[0010] SUBJECT OF THE INVENTION

[0011] The present invention aims to improve the uniformity of the surface layer of an SOI substrate, without degrading the performance of said layer (roughness and defectivity), and to propose an industrial solution both in terms of reliability and costs. It relates in particular to a method for thinning the surface layer of an SOI substrate in single-plate chemical processing equipment, optimized to achieve a non-uniformity of the surface layer less than or equal to + / -0.4nm, with an efficiency at least equal to 90%.

[0012] BRIEF DESCRIPTION OF THE INVENTION

[0013] The invention relates to a method for thinning a surface layer of an SOI substrate comprising:

[0014] a) a step of measuring the thickness of the surface layer, to obtain an average thickness and a non-uniformity of thickness of said layer,

[0015] then the following steps, carried out in single-plate chemical processing equipment and during which the SOI substrate is rotated:

[0016] (b) the application of n cycle(s), with n an integer greater than or equal to 1, of oxidation and etching, to a front face of the surface layer, each cycle involving a dispensation of ozonated water followed by a dispensation of hydrofluoric acid,

[0017] c) carrying out an inhomogeneous etching of the front face of the surface layer, involving the dispensing of a silicon etching solution at a temperature between 20°C and 80°C, in a central region of the surface layer, for a given time, called the non-uniform etching time,

[0018] d) carrying out a homogeneous etching of the front face of the surface layer, involving dispensing a silicon etching solution in a sweeping movement from the edge to the center of the surface layer and vice versa, at a temperature between 20°C and 80°C, for a given time, called the uniform etching time.

[0019] Knowing the average thickness and the thickness non-uniformity of the surface layer,

[0020] the number of cycle(s), the non-uniform etching time and the uniform etching time are determined, prior to carrying out steps b), c) and d), from a model linking them to an average etched thickness of the surface layer and to an average etching non-uniformity defined by the difference between an average etched thickness in a central region of the surface layer and an average etched thickness in a peripheral region of the surface layer.

[0021] According to other advantageous and non-limiting characteristics of the invention, taken alone or in any technically feasible combination: the average etching non-uniformity is positive; the central region and the peripheral region correspond respectively to 30% and 70% of the area of ​​the surface layer; the average etched thickness and the average etching non-uniformity are each expressed in a quadratic form, as a function of the variables which are the number of cycle(s), the non-uniform etching time and the uniform etching time; the model for determining the number of cycle(s), the non-uniform etching time and the uniform etching time is a three-objective optimization model with lexicographic optimality; said objectives are:

[0022] (i) identifying a first set of solutions (non-uniform etching time; uniform etching time) making it possible to achieve equality between the average etched thickness of the surface layer and a target average thickness to be etched, for any number of cycle(s), said target average thickness to be etched being defined by the difference between the target average thickness of the surface layer and the average thickness measured in step a);

[0023] (ii) restricting the first set of solutions (non-uniform etching time; uniform etching time) identified in (i), to a second set of solutions (non-uniform etching time; uniform etching time) making it possible to minimize the difference between the average etching non-uniformity and a targeted average etching non-uniformity, for any number of cycle(s), the targeted average etching non-uniformity being defined by the difference between a targeted thickness non-uniformity of the surface layer and the thickness non-uniformity measured in step a),

[0024] (iii) selecting a recipe from the second set of solutions defined in (ii) allowing to maximize the number of cycle(s).steps b), c) and d) are carried out in order;the temperature during steps c) and d) is set at 65°C;in step c), the dispensing of the silicon etching solution is done only at the center, or according to a sweeping movement going from the center to 50% of the radius of the surface layer and vice versa;the average etching non-uniformity depends on an etching speed gradient defined between the center and the edge of the surface layer, said gradient being dependent on:

[0025] - a flow rate of the silicon etching solution in step c), and / or

[0026] - dispensing a solution on the rear face of the SOI substrate during all or part of step c), and / or

[0027] - the temperature applied in step c); the silicon etching solution, in step c) and / or in step d), is SC1. BRIEF DESCRIPTION OF THE FIGURES

[0028] Other characteristics and advantages of the invention will emerge from the detailed description of the invention which follows with reference to the appended figures in which:

[0029] Shows three examples of surface layer thickness maps, after conventional thinning sequences, before implementing a thinning method according to the present invention; the grayscale is in angstroms in the figure;

[0030]

[0031] Laet illustrate the non-uniformity of the surface layers of 25 SOI substrates, before (Av) and after (Ap) the implementation of a thinning method according to the present invention; lapresents the thickness deviations (expressed in Angstroms) measured with respect to the average thickness, as a function of the radius of the substrates, and lapresent the stacked maps of the surface layers of 25 SOI substrates, before (Av) and after (Ap) the performance of steps b), c), d) of the method according to the invention. The maps are derived from a thickness measurement at 625 points. DETAILED DESCRIPTION OF THE INVENTION

[0032] The present invention relates to a method for thinning the surface layer of an SOI substrate, formed from an active layer of high-quality monocrystalline silicon, arranged on a layer of silicon oxide, itself arranged on a support substrate, usually made of monocrystalline silicon.

[0033] There are several known methods of manufacturing such a substrate, including the Smart Cut process TM , based on an implantation of light species, to create a fragile plane buried in a donor substrate, and on direct bonding to assemble the donor substrate and a support substrate with a dielectric layer disposed between these two substrates. The transfer of a thin layer of the donor substrate onto the support substrate allows the SOI substrate to be obtained, the rest of the donor substrate being able to be reused for a plurality of subsequent transfers.

[0034] The donor and support substrates are each preferably in the form of a circular wafer with a diameter of 200mm or 300mm, or even 450mm, and a thickness typically between 500 and 900 microns.

[0035] Immediately after transfer, the free surface of the superficial thin layer exhibits a relatively high level of roughness (of the order of 4 to 8 nm RMS, measured with an atomic force microscope AFM, on a 30 x 30 micron scan), for a layer thickness, for example, of the order of 250nm. Surface finishing techniques must then be implemented to cure the layer and thin it down to a target thickness of less than or equal to 50nm, for example 20nm, 15nm or even 10nm. Among the finishing techniques, it is possible to implement sacrificial oxidation, etching and / or surface reconstruction (smoothing) heat treatments in a neutral or reducing atmosphere. These treatments are generally carried out at high temperatures, for example, oxidation can be carried out between approximately 750°C and 1100°C, and smoothing between 950°C and 1250°C.

[0036] A favorable finishing thinning sequence includes in particular: stabilization, involving oxidation of the surface of the superficial layer; the sacrificial oxide layer then being removed by chemical attack, thermal smoothing of the surface, under a neutral or reducing atmosphere, thinning by a new sacrificial oxidation of the surface of the superficial layer.

[0037] These steps are carried out in collective treatment furnaces and, although the processes are continuously improved, the various contributions of non-uniformities stack up, tending to generate active layers outside of specification.

[0038] The present invention thus proposes a thinning method which at least partially compensates for a non-uniformity in the thickness of the surface layer of an SOI substrate, an initial non-uniformity typically less than 15nm, or even less than 10nm. The overall thinning achieved by said method is low (in general, less than 5nm, or even less than 3nm): the proposed method is therefore advantageously applied after a finishing thinning sequence as described above.

[0039] The method comprises a first step a) of measuring the thickness of the surface layer to obtain an average thickness E moy and a thickness non-uniformity U of said layer. The thickness measurement is made at a plurality of points i distributed on the surface of the layer. The average thickness E moycorresponds to the average of the thicknesses measured at the plurality of points i; the thickness non-uniformity U reflects the distribution of the thicknesses of the layer at the plurality of points i. It is expressed from a thickness deviation value (at point i) relative to the average thickness E moy (for example, + / -0.4nm); in particular, a non-uniformity U of + / - 0.4nm means that the thicknesses of the surface layer at the plurality of points i are distributed in the range [E moy -0.4nm; E moy +0.4nm].

[0040] The number of measurement points i may be of the order of 20, 100, 500, or even 1000, and the measurement is preferably made up to 3 mm from a peripheral edge of the SOI substrate, knowing that the surface layer generally extends at least up to 0.5 mm from said edge. The points i may be distributed according to a radial geometry, a star, or according to a spiral geometry starting from the center of the layer, or according to any other geometry capable of effectively translating the non-uniformity of thickness of the measured layer.

[0041] Without being limiting, the thickness measurement of step a) can be carried out by known methods based on ellipsometry or reflectometry techniques, with or without the presence of a sacrificial oxide layer on the silicon surface layer.

[0042] Lamontre shows three examples of surface layer thickness maps, after conventional finishing thinning sequences, in different equipment. In general, there appears to be a radial non-uniformity, of more or less significant amplitude (gray level scale in angstroms in the figure), with a central excess thickness.

[0043] The following steps of the method according to the invention therefore advantageously make it possible to compensate for a layer profile having a central excess thickness. They are carried out in single-plate chemical processing equipment. Such equipment is well known and comprises in particular an arm for gripping the SOI substrate, one or more nozzles for dispensing different types of solutions (chemical or water) onto the front face of the substrate (on the surface layer side) and / or onto its rear face. During the solution dispensations, the substrate is driven by a rotational movement to distribute the solution over the entire face concerned (typically between 10 and 1000 revolutions per minute).

[0044] Steps b), c) and d) which will now be described, are carried out in order (preferential) or in disorder, but in all cases, one after the other, without leaving the equipment because the method according to the invention does not require intermediate thickness measurements, which is particularly advantageous.

[0045] Note that, if the SOI substrate comprises a sacrificial oxide layer on the surface silicon layer, this oxide layer is removed prior to the implementation of steps b), c) and d) of the method.

[0046] Step b) comprises the application of n oxidation and etching cycle(s) to the front (free) face of the surface layer, n being an integer greater than or equal to 1. Each cycle involves a dispensation of ozonated water followed by a dispensation of hydrofluoric acid (HF), so as to successively oxidize the surface layer and etch the oxide layer formed in each cycle. The thinning of the surface layer is of the order of 0.4 to 0.5 nm per cycle. Although the number of applicable cycles is not limited, it is preferable to apply between 1 and 10 cycles, or even advantageously between 1 and 5 cycles to remain industrially viable.

[0047] Such a cycle is known for substrate cleaning under the acronym "SCROD" (for "single-wafer spin cleaning with repetitive use of ozonated water and dilute HF" in English terminology).

[0048] Typically, the ozone concentration in ozonated water is between 20 and 40 ppm, the mass concentration of the HF solution is between 0.5% and 4%; the dispensing flow rate is 2 liters per minute. Step b) is preferably carried out at room temperature. Dispensing can be done via a fixed or mobile nozzle (i.e., one with a sweeping movement).

[0049] A rinsing step of the back side of the SOI substrate can be carried out in parallel with the application of the aforementioned cycles to the front side.

[0050] Step c) corresponds to the etching of the front face of the surface layer, involving the dispensing of a silicon etching solution (for example, SC1 (“standard clean 1” according to English terminology), TMAH, HF / H2O2, HF / O3, HF / HNO3+ CH3COOH, KOH, etc.) in a central region of the surface layer. The fact that the solution is only dispensed in the central region implies an inhomogeneous etching of the surface layer, resulting in a more significant etching in the central region than in the peripheral region.

[0051] An average engraving non-uniformity u grav is obtained at the end of steps b), c), d) of the process, defined by the difference between an average thickness etched in a central region of the surface layer e C grav and an average thickness etched in a peripheral region of the surface layer e P grav . This non-uniformity of engraving u gravis positive, so as to compensate for a central excess thickness of the surface layer. It is mainly linked to step c), even if the other steps are sometimes likely to add a slight contribution.

[0052] The dispensation in step c) can be made only in the center, or in a sweeping motion from the center to 30%, 50%, 80%, or even 90% of the radius of the surface layer and vice versa. In doing so, it is possible to adjust the profile of the etching non-uniformity between the central region and the peripheral region.

[0053] The average engraving non-uniformity u gravdepends on an etching speed gradient defined between the center and the edge of the surface layer or between the central region and the peripheral region. This gradient is dependent on the flow rate of the etching solution in step c). It can also be modulated by dispensing a solution on the back face of the SOI substrate during all or part of step c); this dispensation can in particular modify the temperature of the substrate locally. Finally, the etching speed gradient depends on the temperature applied in step c).

[0054] The dispensation of step c) is carried out at a temperature between 20°C and 80°C, for example 65°C, for a given time, called non-uniform etching time t nu . The non-uniform etching time t nu can vary from a few seconds to a few hundred seconds; in practice, it is less than 300s, 250s, or even 200s, to meet industrial efficiency requirements.

[0055] Step d) corresponds to carrying out an etching of the front face of the surface layer, involving dispensing a silicon etching solution (for example, SC1, TMAH, HF / H2O2, HF / O3, HF / HNO3+ CH3COOH, KOH, etc.) in a sweeping movement going from the edge to the center of the surface layer and vice versa; this sweeping ensures a certain homogeneity of etching over the entire surface of the layer.

[0056] The dispensation of step d) is carried out at a temperature between 20°C and 80°C, for example 65°C, for a given time, called uniform etching time t u . The uniform etching time t u can vary from a few seconds to a few hundred seconds; in practice, it is less than 400s, or even 300s, also for reasons of industrial efficiency.

[0057] The proportions (volume ratios) of ammonium hydroxide (NH4OH), hydrogen peroxide (H2O2) and deionized water of SC1, as silicon etching solution in step c) and step d), can be chosen from 1 / 1 / 40 to 1 / 1 / 10, for example. The dispensing flow rate is typically between 1 and 2 liters per minute.

[0058] In the method of the invention, the number of cycle(s) n, the non-uniform etching time t nu and the uniform etching time t u are determined, prior to carrying out steps b), c) and d), from an empirical model linking them to an average engraved thickness e grav of surface layer and average etching non-uniformity u grav .

[0059] For example, we can define that the central region (uniformly distributed around the center of the substrate) and the peripheral region (concentric, surrounding the central region) correspond respectively to 30% and 70% of the area of ​​the surface layer.

[0060] Thus, knowing the average thickness E moy and the thickness non-uniformity U of the surface layer, after step a), and knowing the target average thickness E visée and the targeted thickness non-uniformity (U visée ) of the surface layer, it is possible by using this model to determine the optimal values ​​for the number of cycle(s) n, the non-uniform etching time t nu and the uniform etching time t u .

[0061] It is advantageous to express, in the model, the average engraved thickness e grav and the average engraving non-uniformity u gravin a quadratic form, depending on the variables that are the number of cycle(s) n, the non-uniform etching time t nu and the uniform etching time t u , as presented below :

[0062]

[0063]

[0064] The quadratic form allows good precision to be achieved in determining the three variables, compared to a linear form.

[0065] The coefficients α1 to α9 and β1 to β9 are determined empirically from an experimental design and experimental measurements.

[0066] A three-objective optimization model with lexicographic optimality is particularly well suited to successively determine each of the variables. The objectives can be defined as follows: (i) identify a first set of solutions S n + (t nu , t u ) for the non-uniform etching time tnu and the uniform etching time t u , allowing equality to be achieved between the average engraved thickness e grav of surface layer and an average thickness to be etched aimed at e visée , for any number of cycle(s) n, said average thickness to be engraved referred to e visée being defined by the difference between the average thickness targeted E visée of the surface layer and the average thickness E moy measured in step a);

[0067] (ii) restrict the set of solutions S n + (t nu , t u ) identified with (i), to a second set of solutions S(t nu , t u ) for the non-uniform etching time t nu and the uniform etching time t u , allowing to minimize the difference between the average engraving non-uniformity u grav and an average non-uniformity of engraving aimed at u visée, for any number of cycle(s) n, the average etching non-uniformity targeted u visée being defined by the difference between a targeted thickness non-uniformity U visée of the surface layer and the thickness non-uniformity U measured in step a),

[0068] (iii) select a recipe from the second set of solutions S(t nu , t u ) defined in (ii), allowing to maximize the number of cycle(s) (n).

[0069] The first objective (i) aims to solve the second degree equation Equ3, resulting from the equality between the average engraved thickness e grav of surface layer and an average thickness to be etched aimed at e visée , said average thickness to be engraved referred to e visée being defined by the difference between the average thickness targeted E visée of the surface layer and the average thickness E moy measured in step a). Equation Equ3 is expressed with the uniform etching time t u(time applied to step d) as a variable:

[0070]

[0071] The uniform etching time t u being necessarily positive, the solution of this equation is expressed as a function of the non-uniform etching time t nu (step c) and the number of cycle(s) (step b) as follows:

[0072]

[0073] with

[0074]

[0075] At this stage, it is therefore possible to define a first set S n of possible solutions t u (uniform etching time) and t nu (non-uniform etching time), for each integer n (number of cycle(s)):

[0076]

[0077] The first set of possible solutions for t u and t nubeing further filtered on positive values ​​and depending on the technical feasibility of the equipment, namely a precision on times of 1s, thus giving the set of solutions S n + following at the end of the first stage (first objective):

[0078]

[0079] The second objective aims to minimize the difference between the average engraving non-uniformity u grav expressed in Equ2, and the average etching non-uniformity targeted u visée , defined by the difference between a targeted thickness non-uniformity U visée of the surface layer and the thickness non-uniformity U measured in step a).

[0080] Starting from the solutions of the set S n + retained at the first stage, only the recipes r are kept n to minimize the error on the engraving non-uniformity u grav :

[0081]

[0082] The second set of solutions S at the end of the second stage (second objective) therefore translates as:

[0083]

[0084] The third objective aims to minimize the cost of the process, which is equivalent to maximizing the number of cycles n of step b) of the process, because step b) is the most efficient in terms of thinning.

[0085] The recipe among the solutions in the second set of solutions S offering the largest number of cycles is therefore selected.

[0086] The trend curve Av of illustrates the initial non-uniformity (i.e. before the application of the method according to the invention) of surface layers of 25 SOI substrates. In particular, the deviations in thickness from the average thickness, at any point i measured, on the 25 surface layers, are compiled on the: we note a central excess thickness (radius 0) of 0.5 nm on average greater than the edge thickness.

[0087] The trend curve Ap illustrates the final non-uniformity of these same 25 surface layers, after the implementation of steps b), c), and d) of the method according to the invention. The deviations in thickness from the average thickness, at any point i measured, on the 25 surface layers, are compiled and an improved final non-uniformity appears with a deviation of less than 0.2 nm on average between center and edge. The vast majority of the measured points also fall within the targeted thickness non-uniformity requirement U visée of + / -0.4nm.

[0088] In this example, the average thickness is evaluated from 625 measurement points i. As shown in the stacked maps of the surface layers of the 25 SOI substrates (), the thickness non-uniformity profile is modified after the implementation of the thinning method according to the invention, and makes it possible to erase the central excess thickness and to improve the overall uniformity of the surface layers.

[0089] Of course, the invention is not limited to the embodiments and examples described, and variant embodiments may be made without departing from the scope of the invention as defined by the claims.

Claims

Method for thinning a surface layer of an SOI substrate comprising:a) a step of measuring the thickness of the surface layer, to obtain an average thickness (E moy ) and a thickness non-uniformity (U) of said layer, then the following steps, carried out in a single-plate chemical processing equipment and during which the SOI substrate is rotated: b) the application of n cycle(s), with n an integer greater than or equal to 1, of oxidation and etching, to a front face of the surface layer, each cycle involving a dispensation of ozonated water followed by a dispensation of hydrofluoric acid, c) carrying out an inhomogeneous etching of the front face of the surface layer, involving a dispensation of a silicon etching solution at a temperature between 20°C and 80°C, in a central region of the surface layer, for a given time, called non-uniform etching time (t nu),d) carrying out a homogeneous etching of the front face of the surface layer, involving dispensing a silicon etching solution in a sweeping movement going from the edge towards the center of the surface layer and vice versa, at a temperature between 20°C and 80°C, for a given time, called uniform etching time (t u ), in which, knowing the average thickness (E moy ) and the thickness non-uniformity (U) of the surface layer, the number of cycles (n), the non-uniform etching time (t nu ) and the uniform etching time (t u ) are determined, prior to carrying out steps b), c) and d), from a model linking them to an average engraved thickness (e grav ) of surface layer and an average etching non-uniformity (u grav) defined by the difference between an average thickness etched in a central region of the surface layer and an average thickness etched in a peripheral region of the surface layer. Thinning method according to the preceding claim, in which the average etching non-uniformity (u grav ) is positive. Thinning method according to one of the preceding claims, in which the central region and the peripheral region correspond respectively to 30% and 70% of the area of ​​the surface layer. Thinning method according to one of the preceding claims, in which the average etched thickness (e grav ) and the average engraving non-uniformity (u grav ) are each expressed in a quadratic form, as a function of the variables that are the number of cycle(s) (n), the non-uniform etching time (t nu ) and the uniform etching time (t u ). Thinning method according to the preceding claim, in which the model for determining the number (n) of cycle(s), the non-uniform etching time (t nu ) and the uniform etching time (t u ) is a three-objective optimization model with lexicographic optimality. Thinning method according to the preceding claim, wherein said objectives are: (i) identifying a first set of solutions (S n + (t nu , t u )) for the non-uniform etching time (t nu ) and the uniform etching time (t u ), allowing equality to be achieved between the average engraved thickness (e grav ) of surface layer and an average thickness to be etched targeted (e visée ), for any number of cycle(s) (n), said average thickness to be engraved targeted (e visée ) being defined by the difference between the target average thickness (E visée) of the surface layer and the average thickness (E moy ) measured in step a); (ii) restrict the first set of solutions (S n + (t nu , t u )) identified with (i), to a second set of solutions (S(t nu , t u )) for the non-uniform etching time (t nu ) and the uniform etching time (t u ), allowing to minimize the difference between the average engraving non-uniformity (u grav ) and an average non-uniformity of targeted engraving (u visée ), for any number of cycle(s) (n), the average non-uniformity of targeted etching (u visée ) being defined by the difference between a targeted thickness non-uniformity (U visée ) of the surface layer and the thickness non-uniformity (U) measured in step a), (iii) select a recipe from the second set of solutions (S(t nu , t u )) defined in (ii) allowing to maximize the number of cycle(s) (n). Thinning method according to one of the preceding claims, in which steps b), c) and d) are carried out in order. Thinning method according to one of the preceding claims, wherein the temperature during steps c) and d) is set at 65°C. Thinning method according to one of the preceding claims, in which, in step c), the dispensing of the silicon etching solution is done only in the center, or according to a scanning movement going from the center to 50% of the radius of the surface layer and vice versa. A thinning method according to one of the preceding claims, wherein the average etching non-uniformity (u grav) depends on an etching speed gradient defined between the center and the edge of the surface layer, said gradient being dependent on:- a flow rate of the silicon etching solution in step c), and / or- a dispensation of a solution on the rear face of the SOI substrate during all or part of step c), and / or- the temperature applied in step c). Thinning method according to one of the preceding claims, in which the silicon etching solution, in step c) and / or in step d), is SC1.

Images