The invention · relates to the field of producing semiconductor wafers or substrates comprising at least two superposed layers of microcomponents.

The stacking of several successive layers of components ("3D IC stacking") and the multilayer wafers obtained can notably be used in the fabrication of sensors for backside illumination image sensors, often called BSI sensors.

Such wafers are produced by bonding a donor substrate onto a receiver substrate which allows the transfer of a portion of wafer forming one layer of the donor substrate to the receiver substrate. The transferred layer may comprise electronic or optoelectronic microcomponents (that may be active or passive). The receiver substrate may also comprise such components.

It is often necessary to carry out treatments on the layer after its transfer, for example in order to form other microcomponents in order to provide surface access to the microcomponents that are already present or in order to achieve interconnections. These treatments are carried out essentially by photolithography.

In this context, the invention addresses the problem caused during such a subsequent treatment step by the uneven deformations that appear when the layer is transferred.

Such uneven deformations have actually been observed in a general manner during processes of three-dimensional integration of components, and in particular during the installation of components by photolithography.

Specifically it has been found that it is difficult to form, after the transfer of a layer, additional microcomponents that are satisfactorily aligned, parallel to the plane of the wafers, with the microcomponents present in the layer or in a lower layer.

This phenomenon of misalignment (or "overlay") is described with respect to Figures 1A to IE which illustrate an exemplary embodiment of a three-dimensional structure. This example comprises the transfer, onto a receiver substrate, of a layer of microcomponents formed on a donor , substrate, and the formation of an additional layer of microcomponents on the exposed face of the layer.

Figure 1A illustrates a donor substrate 10 on which is formed a first series of microcomponents 11. The microcomponents 11 are formed for example by photolithography using an exposure through a first mask. Figure IB, which is a section of the substrate of Figure 1A in the plane P, shows that the microcomponents 11 are formed in a thin surface layer of the substrate 10 and are flush with the surface of the latter.

As illustrated in Figure 1C, the face of the donor substrate 10 comprising the microcomponents 11 is then placed in contact with a face of a receiver substrate 20. The bonding between the substrate 10 and the substrate 20 is carried out for example by molecular adhesion. This gives a buried layer of microcomponents 11 at the bonding interface between the substrates. After the bonding and as shown in Figure ID, the donor substrate 10 is thinned in order to remove a portion of the material that is present above the layer of microcomponents 11, leaving on the substrate 20 only a thin layer 10a of the donor substrate 10.

As shown in Figure IE, a subsequent step consists in forming a second layer of microcomponents 12 at the exposed surface of the layer 10a, or in carrying out additional technological steps (replacement of the contacts, of interconnections, etc.) on this exposed surface, in alignment with the components included in the layer 10a.

In order to from the microcomponents 12 in alignment with the buried microcomponents 11 (or in order to carry out additional technological steps), a second photolithography mask obtained from that used to form the microcomponents 11 is used.

However, if the second mask has been defined directly on the base of the first mask, offsets in the plane of the substrate 20 and of the layer 10a are found between certain microcomponents 11 and 12 like the offsets Δ ΐ ΐ , Δ 22, Λ 33, Δ44, indicated in Figure IE (corresponding respectively to the offsets observed between the pairs of microcomponents lli/12i, 11 ₂ /12 _2/ 11 ₃ /12 ₃ and 11 ₄ /12 ₄ parallel to the plane of the layer). This misalignment ("overlay") phenomenon between the two layers of microcomponents 11 and 12 can be a source of short circuits, of losses of contact, of distortions in the stack or of connection faults between the microcomponents of the two layers.

The overall misalignment of the components 11 and 12 does not result from a simple transformation (obtained by the combination of a translation and of a rotation) which might be the result of an imprecise assembly of the substrates 10 and 20 during the bonding step. It results from local uneven deformations which appear in the layer 10a during the assembly of the donor substrate 10 to the receiver substrate 20.

Specifically, once bonded to the receiver substrate 20, the donor substrate 10 has a geometry (curvature and warp) that differs from that which it had at the beginning. This new geometry results in particular from the fact that the receiver substrate 20 also has a specific geometry (including curvature and warp) that differs from that of the donor substrate 10. Consequently, when the substrate 10 is placed in contact with the receiver substrate 20, the two substrates must adapt to the geometry of each other, which creates stress zones in each of the substrates.

Therefore there is a considerable risk, if no corrective measure is taken, that some of the microcomponents 12 formed on the exposed surface of the substrate after transfer (as shown in Figure IE) exhibit offsets of position with the corresponding microcomponents 11 of the order of several hundreds of nanometres, or even of a micron.

Thus, when the transferred microcomponents are imagers formed of pixels and the post-transfer treatment steps are designed to form colour filters on each of these pixels, a loss of the colouring function for certain pixels has been observed.

One solution for overcoming this difficulty is to use a misalignment correction algorithm making it possible to correct the exposure of the photolithography mask in order to carry out an appropriate etching of the series of the components 12.

Such a misalignment correction algorithm is based on the observation of the movement of the marks that are present on the layer 10a prior to its transfer. The practice is to choose these marks on a single given circle or contour around the centre of the wafer. These marks are either elements of microcomponents 11 of the first layer that are easy to identify, or indicators marked on the wafer specifically for the purpose of alignment.

Before carrying out the photolithographic exposure and the etching of the series of components 12, these marks are identified and the movements that they have sustained are determined. This set of movements is then transferred to a single mathematical model for the whole of the wafer.

An alignment is then applied in a systematic and identical manner for each component 12, using the single mathematical model to determine the aligned positions for the photolithography for the whole of the wafer. This makes it possible to dispense with positioning and calibrating the photolithography tool more than once per wafer to be treated. Specifically, one solution that would consist in realigning the photolithography tool for different zones of the wafer to be treated would not be satisfactory from an industrial point of view because it would impose constraints (time, manipulation) that would be too heavy for the user.

The single mathematical model comprises notably a correction function applied to coordinates (X, Y) of a position to be aligned, these coordinates usually having been the subject of a prior transformation in the context of the mathematical model. The correction determined by the function for a position to be aligned takes the form of a movement vector (Xc, Yc) which is added to the coordinates (X, Y). This correction usually is the last step of the alignment by the single mathematical model.

The practice is to use such a correction function which depends only on the distance of the position to be aligned to the centre, and is linear with respect to this distance. The value of the correction is moreover chosen to be zero at the centre of the wafer.

The components of the correction function on the two axes are predetermined by minimizing, on each of the axes, the average of the residual misalignments (that is to say after application of the correction function) of the series of marks chosen on a given circle around the centre.

Such an empirical method has given satisfactory results, but, with the technological advances, components such as the components 11 and 12 reduce in size, which requires ever finer alignments. Moreover, the proposed wafers increase in diameter. The result of this is an increased risk of the existence of different residual misalignments on the various zones of the wafers, notably the periphery, which are difficult to correct by a single mathematical model according to the prior art.

Finally, certain bonding methods, in particular bonding of the molecular bonding type induce particular deformations which are only partially corrected by the existing misalignment correction algorithms, as will be described in the rest of the description. The object of the invention is to solve these problems and thus improve the alignment of the microcomponents.

Summary of the Invention

In this context, one subject of the invention is a method for correcting misalignment of positions on a first wafer bonded onto a second wafer, the method comprising the application, to coordinates of each position, of a predetermined correction function for the said first wafer, the correction applied by the correction function being a function only of the distance of the position relative to the centre of the first wafer, characterized in that the applied correction varies, over the whole of the first wafer, in a non-linear manner with respect to the distance of the position relative to the centre.

By virtue of this method, the correction made is of better quality than that obtained with the prior methods, and allows a more precise alignment for a larger number of microcomponents while providing the fabricator with great speed in the alignment process, since no recalibration is necessary, the correction function being predetermined for the wafer.

According to one particularly advantageous embodiment, the misalignment correction method also comprises additionally a prior step of determination of the correction function at least as a function of misalignment values observed on the first wafer respectively for marks of a first series present at a first distance from the centre and marks of a second series present at a second distance from the centre.

This feature makes it possible to correlate the correction function with the misalignment values of marks at two different distances from the centre of the wafer and thus to obtain a good modelling of these misalignment values, by a non-linear function. In one embodiment, the correction function is linear by segments with respect to the distance of the position relative to the centre, a central zone defining a first segment and a peripheral zone defining a second segment. In this embodiment, two different linear functions on the two segments are chosen so as to faithfully model the misalignment values in order to correct them finely.

In another embodiment, the correction function is defined in a single manner over the whole of the first plate, in a non-linear manner with respect to the distance of the position relative to the centre. This choice of a function defined in a single manner unequally provides increased simplicity of the informatic implementation. It is possible to choose as a correction function an at least second-degree polynomial so as to faithfully model the misalignment values in order to correct them finely.

According to one feature of the invention, the applied correction increases as a function of the distance relative to the centre, which makes it possible to take account of the major misalignments occurring on the periphery of the wafers.

According to another feature of the invention, the applied correction increases to a greater degree as a function of the distance relative to the centre on a peripheral zone of the first wafer than on a central zone of the first wafer, which makes it possible to take yet further account of the major misalignments occurring on the periphery of the wafers.

The method is particularly advantageous when it is applied to a set of wafers or a first wafer with a diameter of more than 200 mm, for the reasons expressed in the introduction which mean that the misalignment is particularly great in the periphery of wafers of this dimension.

The method is also particularly advantageous if the bonding is a molecular bonding, because it makes it possible to correct the specific misalignments that appear on the structures that have been bonded in this manner.

It is specified that, according to the features of the invention, the correction function is added to coordinates of the position and that, moreover, it is applied to coordinates of the position that are obtained after application to initial coordinates of the position of a linear transformation with respect to the coordinates.

According to a second subject, the invention also relates to a method for producing a three-dimensional composite structure comprising a step of producing a first layer of microcomponents on one face of a first substrate, a step of bonding of the face of the first substrate comprising the microcomponent layer onto a second substrate and a step of producing a second layer of microcomponents on the face of the first substrate opposite to the face comprising the first layer of microcomponents, the method being characterized in that the step of producing the second layer is carried out while correcting misalignments of positions using a method as discussed above.

Another subject of the invention is a three-dimensional composite structure obtained according to an embodiment as has just been described above. The invention also relates to a backside illumination image sensor comprising such a three-dimensional composite structure.

Finally, the invention also relates to a device for correcting misalignment of positions on a first wafer bonded to a second wafer, the device comprising means for applying, to coordinates of each position, a predetermined correction function for the said first wafer, the correction applied by the correction function being a function only of the distance of the position relative to the centre of the first wafer, characterized in that the applied correction varies, over the whole of the first wafer, in a non-linear manner with respect to the distance of the position relative to the centre.

This device offers the advantages specified in relation to the misalignment correction method that is the subject of the invention.

The invention will now be described with respect to the figures. Brief Description of the Figures

Figures 1A to IE show the production of a three-dimensional structure of semiconductor components according to the prior art.

Figure 2 shows the experimental values of residual misalignment on a wafer after application of an alignment algorithm according to the prior art. Figures 3 and 4 show respectively the average values of radial misalignment and the vectors of misalignment for a wafer after bonding and transfer of a layer.

Figure 5 shows a simulation of the radial misalignment values according to a modelling of the effect of gravity.

Figure 6 shows a simulation of the final residual misalignment values according to the modelling and with an alignment algorithm according to the prior art.

Figure 7 shows a method for preparing a wafer according to the invention.

Figures 8 and 9 show two photolithography methods using a misalignment correction method according to the invention.

Figure 10 shows the experimental values of residual misalignment on a wafer after application of a misalignment correction algorithm according to the invention.

Figure 11 illustrates the misalignment correction algorithm according to a particular embodiment of the invention.

Figure 12 shows a method for producing a three-dimensional composite structure using a misalignment correction method according to the invention.

The invention will be described below with respect to an embodiment given as an illustration.

Detailed Description of the Invention

With reference to Figure 2, a semiconductor wafer 200 is shown in a view from above. This wafer results from a stacking and transfer process as described in the introduction. Of circular geometry, it is centred about a point X.

Figures 2 to 6 present data relating to this wafer and to the misalignment correction obtained with a known algorithm. These data will make it possible to discuss the advantages of the misalignment correction method according to the invention shown in Figures 7 to 12.

Alignment marks 201 to 204 are present on the surface of the wafer 200. In this instance, as an illustration, there are four of them. They are all at the same distance from the centre X. Other marks could have been shown. In practice, these marks may have been placed specifically for the purpose of serving as alignment indicators or might be elements of microcomponents that, can be seen individually on the wafer. The misalignment correction algorithm according to the prior art uses these marks placed at one and the same distance relative to the centre of the wafer.

In order to determine the quality of alignment obtained with a known misalignment correction algorithm, the residual movements have been determined for the wafer 200 point by point. "Residual movements" in this instance means the movements not taken into account by this algorithm, measured for a given position as the difference between the established real position of a mark and the position of the mark estimated by the alignment algorithm.

A central zone 210 can be seen that extends over two thirds of the diameter of the wafer 200 and for which the residual movement values are low. About this central zone 210 can be seen a circular zone 220 on which the residual movements are on the contrary very great.

To understand this distribution of the residual misalignment values, a detailed study has been carried out.

Shown in Figure 3 in the form of a graphic are the radial movement values measured experimentally, uncorrected, along a radius of a wafer similar to the wafer 200. In this instance it is a wafer with a 100 mm radius (or 200 mm in diameter).

Note that in a central zone the movement values are relatively constant and close to zero (in any case less than 50 nm or even 25 nm), but beyond a distance of 80 mm from the centre the movement values increase rapidly to reach values of the order of 100 nm.

With reference to Figure 4, shown in the form of vectors are the movements measured experimentally on the surface of a wafer 400 with a diameter of 300 mm. Note that, in this figure, there are vectors of movement of great intensity in a zone 420 close to the periphery of the wafer. Conversely, the vectors of movement in a central zone 410 of the wafer are of markedly shorter length.

With reference to Figure 5, this time there is a simulation of the absolute values of movement in the plane of the points of a wafer undergoing a bonding as a function of the position along a radius of the latter taking account of the effect of gravity during the bonding and the transfer of a thin layer (as shown in Figures IB to ID). The two wafers are poisoned horizontally during assembly, the horizontal position being important because of the determinant effect of gravity.

This simulation of the values of movement in the plane is shown in the form of the curve 510 which shows on the X axis the position along a diameter of the wafer and on the Y axis the intensity of the movement.

Note that, for positions moving away from the centre of the wafer, there are markedly higher values than for the positions at the centre of the wafer, with a sudden increase after an inflection point. The curve 520 also shows by comparison that a simple linear regression as a function of the distance relative to the centre of the wafer is incapable of taking account of the simulated phenomenon.

Figure 6 shows the simulated wafer specified in figure 5 (reference number 540), indicating the points for which the value of residual movement exceeds a certain threshold after the successive applications first of all of a linear transformation T with respect to the coordinates and then a linear correction C with respect to the distance relative to the centre of the wafer, according to the method of the prior art. Note that a central zone for which the values of residual movements are slight occupies the central two thirds of the wafer and that the points for which the values of residual movements are greater form a peripheral halo with a radius of approximately 12 cm.

Also shown in this figure are the values of residual movement simulated as a function of the distance relative to the centre for the wafer in the form of a curve 550. It is possible to recognize the central zone for which the values of residual movements are slight and a peripheral radial zone having high values.

Note that this distribution is totally the same as that found experimentally and shown in Figure 2. Thus, it appears that the model used to simulate the movements and based notably on the effect of gravity during the molecular bonding excellently reproduces the experimental data.

It should be specified that this result is obtained with two different simulations, one of them relating to the situation in which the donor substrate is beneath the receiver substrate during the bonding step and the other to the reverse situation. On the basis of these findings, an alignment algorithm according to the invention has been developed to obtain high-quality alignments for large-size wafers, having a layer transferred by virtue of a molecular bonding.

With reference to Figure 7, the determination of an alignment algorithm is carried out in the following manner. This method is applied to a circular semiconducting wafer having microcomponents on its surface and having to be the subject of a transfer.

Chosen first of all are two contours, for example circles centred about the centre X of the wafer and of respective diameter Dl and D2. Six to ten marks are chosen along each of these two circles during a step El. If the microcomponents that are present on the surface of the wafer are sufficiently visible, they may serve as marks. It is also possible to place marks specifically for the purpose of alignment.

During a step E2, which may be similar to the step of Figure 1C, the transfer of the layer is carried out. This step E2 may include steps of bonding, notably molecular bonding, and of thinning.

The following steps (E3 to E5) can be carried out following on from steps El and E2, in the same work unit, or conversely after delivery of the wafer to another work unit. These steps are consequently shown both in Figures 7 and in Figures 8 and 9, and are figured in dashed lines to indicate the alternative.

During a step E3, the movement of each of the marks is measured along two axes in the plane, in the form of a vector (X, Y).

A minimization algorithm is then used, during a step E4, so as to compute a transformation T comprising at least one translation and one rotation to be applied to each of the marks in order to minimize the misalignment on all of the marks, the marks of the circles with a diameter Dl and D2 being for example treated in the minimization with the same importance (the same weight).

It is also possible at this stage to compute a homothetic transformation and a vector of orthogonality which, combined with the translation and with the rotation, make it possible to obtain the best possible approximation by a single transformation T of all of the movements of the marks for the photolithography. The transformation T thus consisting of a translation, a rotation, a magnification and of the application of a vector of orthogonality, is a transformation acting in a linear manner on the coordinates of the positions. It is determined in order to be uniform at all the positions of the wafer.

During a step E5, for each of the marks of the circle with the diameter Dl, an intermediate residual movement is computed on the basis of the coordinates transformed by the transformation T determined during the step E4. This intermediate residual movement is thus defined as being the difference between the real movement and the correction made by the transformation T. It consists of a vector (x, y). Similarly, still during the step E5, for each of the marks of the circle with a diameter D2, an intermediate residual movement (x, y) is computed.

An average of the intermediate residual movements on the circle with the diameter Dl is then computed in the form of a vector of average residual movement (Xa, Ya) for all of the marks along the circle with the diameter Dl. An average value of the intermediate residual alignments is also computed for all of the marks along the circle with the diameter D2.

A correction function C is then determined in order to be applied to the coordinates of the positions originating from the transformation T. This correction function C depends only on the distance relative to the centre X of the position to be corrected. It is determined so as to minimize the averages of the residual movements obtained after its application, for the circle with the diameter Dl and for the circle with the diameter D2.

The correction value forms a vector (Xc, Yc) which is added to the coordinates originating from the previous step.

In a first variant, the correction function is a linear function with respect to the distance relative to the centre by segments, a first segment consisting of a central zone of the wafer, and a second segment consisting of a peripheral zone of the wafer. In this variant, the marks placed on the circle with the diameter Dl are included in the central zone, and the marks placed on the circle with the diameter D2 are included in the peripheral zone.

The correction function is then determined by choosing two affine functions (one affine function on each zone) which make it possible to cancel the average values of the intermediate residual movements along the circles of diameter Dl and D2. The function applied to the central zone of the wafer is moreover chosen to be zero at the centre of the wafer, while the second function is chosen to be equal to the first function at the meeting point between the two zones. The correction function is in most cases increasing on the two segments with a steeper slope on the second segment.

In another variant, the correction function is a function defined uniquely (without discontinuity or with no angular point) over the whole length of the radius of the wafer. It may thus be a polynomial of the second order or of a higher order, or a more complex function. It is then determined by choosing at least two parameters so as to minimize the average value of the intermediate residual movements computed in the previous step.

In one embodiment, the circles with the diameter Dl and D2 are treated with the same weight but it is also possible to allocate different weights to them.

The characteristics of the transformation T determined in the step E4 and of the correction C determined in the step E5 are retained, for example by saving them on a computer medium.

With reference to Figure 8, a photolithography method is put in place in a specialized work unit to which the wafer has been delivered or in the work unit where the preparation of the wafer took place.

Note at this stage that it is advantageous for the steps E3 to E5 to be carried out just before the exposure of the wafer by photolithography, as shown in Figure 8, even though, as mentioned above, they may be carried out while preparing the wafer.

During a step Fl, the parameters of the transformation T determined in the step E4 and of the correction C determined in the step E5 are placed in the memory of a photolithography device.

The transformation T and the correction C are then applied for each position that is to be the subject of a treatment in order to form a component on the surface of the wafer, according to an iterative process F2-F6, the step F2 being the beginning of the treatment of a position, and the step F5 the end of the treatment of a position. The step F3 is the application to the coordinates of the treated position of the transformation T. Step F4 is the subsequent application to the coordinates originating from the step F3 of the correction C as a function of the distance relative to the centre. The step F6 is then the actual photolithography carried out in the location of which the corrected coordinates have been obtained following the previous steps. The step F6 may also, as a variant, not be included in the iterative process.

The functions of transformation T and of correction C are determined in a general manner prior to the step of exposing the wafer by photolithography so as to carry out a single initial alignment. Thus, the photolithography device does not need to be moved for each zone of the slice to be treated. Moreover, the alignment obtained is particularly good, including for wafers of 300 mm bearing a layer transferred by a molecular bonding method.

Figure 9 shows an alternative of the method of Figure 7. After a step F'l of setting the parameters of the transformation determined in the step E4 and of the correction determined in the step E5 in the memory of a photolithography device, the transformation T is applied to all the positions to be treated during a step F'3, then, during a step F'4, the correction C is applied to the coordinates of all the positions originating from the step F'3. The step F'5 of photolithography is then carried out for all the positions.

Figure 10 shows a wafer 600 on which the marks 651, 652 and 661 to 664 have been shown, a first group of them (651 and 652) being situated on a circle with a radius Dl centred about the centre X, while the marks of a second group (661 to 664) are situated on a circle of radius D2 about the centre X. In the example shown, the distance D2 is approximately half of the distance Dl. The marks of the first group are situated in a peripheral zone 620 and the marks of the second group are situated in a central zone 610.

With reference to Figure 11, the correction C consists of a linear correction on the central zone 610 of the wafer, and of a different linear correction on the peripheral zone 620 of the wafer. Each of the corrections on these two zones is, in the embodiment shown, linear on the zone in question, but the whole of the correction C is non-linear on the whole of the wafer.

Thus, the correction function is linear by segments with respect to the distance relative to the centre, the central zone 610 defining a first segment and the peripheral zone 620 defining a second segment. The correction is greater on the peripheral zone. Reference 800 shows the difference between the correction on the peripheral zone and the central zone, and also the curve 550 of residual movement explained previously with reference to Figure 6. All of the linear corrections by segments as a function of the distance relative to the centre make it possible to satisfactorily approach the real movement and therefore to offer an alignment of great quality.

Alternatively, the correction function C could be defined in a single manner on all of the wafer but in a non-linear manner with respect to the distance relative to the centre. As mentioned previously, it could involve an at least second-degree polynomial, or a more complex function.

Moreover, the correction function according to a variant of the invention may be determined without making use of the marks placed on two different circles. Only one series of marks at one and the same distance may be used, with a non-linear function, for example a second-order polynomial, of which one of the parameters is determined on another criterion, or is chosen to be fixed. Thus, according to this variant, the correction is less closely adapted to the specific misalignments of the wafer, but all the same it makes it possible to take account of the non-linear character of the residual misalignment values.

Finally, Figure 12 shows a method for producing a three-dimensional composite structure using a misalignment-correction method according to the invention.

This method begins with a step Gl of forming a first series of microcomponents on an initial substrate, or donor substrate (as shown in Fig. 1A and IB). These microcomponents of the first series form a layer of microcomponents. This formation step is followed by a step G2 of bonding the donor substrate onto a receiver substrate (as shown in Fig. 1C), and, in certain variants, a thinning step G3 (Fig. ID).

The method continues with a step of determining the alignment parameters as shown in steps E3 to E5 in Figures 7 to 9. Thus, the parameters to be applied to determine the photolithography mask are known at this stage. These parameters notably define a linear transformation T with respect to the coordinates and a correction C that is a function of the distance relative to the centre, which varies, over the whole wafer, in a non-linear manner with respect to the distance relative to the centre.

Once these parameters are determined, the method continues with a photolithography step of a second series of components forming a component layer. This photolithography is carried out either by having determined a corrected mask for the whole wafer before beginning the exposure, as shown in Figure 9 (steps F'3 to F'5), or by correcting the positions individually just before carrying out the exposure, as shown in Figure 8 (steps F3 to F5). In both cases, the correction is made with the parameters determined during the steps E3 to E5.

The method for fabricating a three-dimensional composite structure may continue with the application of treatments or the formation of other component layers. The finished product may notably be a BSI sensor.

In summary, the invention greatly reduces the residual misalignments and improves the performance of the three-dimensional structures obtained. It is not limited to the embodiments described and extends to the alternatives that can be envisaged by those skilled in the art in the context of the scope of the claims.