FR2935057A1 - Diffracting pattern i.e. diffraction grating, physical or geometric parameter real-time-monitoring method for semiconductor medium, involves determining current parameter vector from parameter vector associated to selected signature vector - Google Patents

Abstract

The method involves bearing a scalar processor (25) to be selected among K-signature vectors, where the signature vectors minimize weighted sum of distance between a current signature vector and one of the signature vectors, and weighted sum of a criterion implementing a parameter vector associated to the former signature vector and another parameter vector associated to the selected signature vector at previous iteration. The current parameter vector is determined from the latter parameter vector associated to the selected signature vector. An independent claim is also included for a device for real-time monitoring evolution of a physical or geometric parameter of a diffracting pattern during chemical and/or physical treatment.

Description

B8829 1 METHOD OF TRACKING REAL TIME OF THE EVOLUTION OF A DIFFRACTANT PATTERN

Field of the Invention The present invention relates to a device and method for detecting the shape of a diffracting pattern and, more particularly, to a device and a method for real-time tracking of geometric and / or physical parameters of a diffracting pattern. a diffracting pattern undergoing physical and / or chemical treatment. DISCUSSION OF THE PRIOR ART Various techniques make it possible to determine the shape of a diffractive pattern. One of them, called scattometry, takes advantage of the correlation between the optical properties of a beam diffracted by the pattern and the geometrical and / or physical parameters of the pattern. A scalerometry method can implement different measurement techniques including the reflectometry, or the spectroscopic ellipsometry that determines, as a function of the wavelength, the polarization changes between a beam illuminating the pattern and the beam diffracted by the pattern. Such polarization changes are correlated with the geometric and / or physical parameters of the pattern.

Subsequently, we will call "pattern" any diffracting pattern whose characteristic dimensions are of the order of B8829

2 a few nanometers to a few micrometers. These are, for example, diffraction gratings formed on the surface of a semiconductor medium. FIG. 1 represents a support 1 comprising a face 2 on which a pattern 3 is formed, for example a diffraction grating comprising a periodic succession of lines 13. A polarizer 5, placed in a direction oblique with respect to the face 2, transmits a light beam 7 of known polarization towards the pattern 3. The beam 7 is diffracted by the pattern 3 to form a light beam 9 which is received by an analyzer 11. The analyzer 11 makes it possible to draw curves representative of the change of polarization between the beams 7 and 9, as a function of the wavelength of the beam 7. Such curves are called the scalerometric signature of the diffracting element. FIG. 2 shows an example of a crenel pattern 13 of a line of a diffraction grating formed on the support 1. In this example, the shape of the slot is defined by three parameters: the height h of the slot, the width half-height CD of the crenel, and the angle 0 that forms each of the side walls of the crenel with the perpendicular to the face 2 of the support 1. Figure 3 represents the scalerometric signature of a diffraction grating consisting of a succession of lines each having the shape of the line 13 of Figure 2, the signature being obtained by spectroscopic ellipsometry. The scalerometric signature comprises two curves 15 and 17 for the evolution of two quantities representative of the norm and phase changes of the two orthogonal projections of the polarization of the incident light beam 7 (scatto-ellipsometry). We are interested here in methods for determining the geometric and / or physical parameters of a pattern from its scalerometric signature. To make this determination, a first method 35 is to theoretically model the shape of the pattern and its B8829

3 diffraction behavior using unknowns then to determine the unknowns of the modeling by comparison with the acquired scattérometric signature. However, pattern diffraction modeling and methods of obtaining unknowns by comparison with curves are relatively complicated. It is also possible to compare an acquired scalerometric signature with signatures previously stored in a library, each signature of the library being associated with a particular pattern. The signature of the library closest to the one acquired is then chosen, and the geometrical and / or physical parameters associated with the chosen signature are considered to be the parameters of the pattern. The larger the signature library, the greater the accuracy of the determination. However, if one wants to achieve an accurate determination of geometric and / or physical parameters, the library must include a large number of signatures and the search in the library becomes long.

The time required for carrying out the above methods is notably incompatible with the real-time monitoring of the evolution of the shape of a pattern which varies over time, for example when the pattern undergoes a chemical treatment and / or physical.

Summary There is therefore a need for a device and a method for real-time monitoring of the evolution of the shape of a pattern subjected to a chemical and / or physical treatment. Thus, an embodiment of the present invention provides a method of real-time monitoring of the evolution of geometric and / or physical parameters of a diffractive pattern during a physical and / or chemical treatment, comprising: before the start of processing: store, in a vector memory associated with a vector processor, a signature vector library, B8829

4 the coordinates of each signature vector corresponding to measurement signals associated with a reference pattern among reference patterns; storing, in a scalar memory associated with a scalar processor, parameter vectors, the coordinates of each parameter vector corresponding to geometric and / or physical parameters of a reference pattern, each parameter vector being associated with a signature vector, and During the course of the processing, iteratively determines a current parameter vector by repeating the following steps: determining a current signature vector from measurements made on the pattern; causing the vector processor to determine, in the library, the K signature vectors closest to the current signature vector, K being an integer greater than or equal to 2; causing the scalar processor to select, from among the K signature vectors, the signature vector which minimizes the weighted sum of a distance between the current signature vector and said signature vector and a criterion implementing the parameter vector associated with said signature vector and at least the parameter vector associated with at least one of the signature vectors selected at least one preceding iteration; and determining the current parameter vector from the parameter vector associated with the selected signature vector. According to one embodiment of the present invention, the current parameter vector is determined from the parameter vector associated with the selected signature vector and from at least one current parameter vector determined at a previous iteration. According to an embodiment of the present invention, the current parameter vector is determined, in addition, from at least one parameter vector associated with a selected signature vector at a next iteration.

B8829

According to one embodiment of the present invention, the method further comprises an initial step of determining, before the start of treatment, an initial parameter vector of the diffracting pattern. According to one embodiment of the present invention, the method further comprises deciding whether to stop or modify the patterned processing from the current parameter vector. According to one embodiment of the present invention, the determination of the current signature vector is carried out by spectroscopic ellipsometry. There is also provided a device for monitoring in real time the evolution of geometrical and / or physical parameters of a diffractive pattern during a physical and / or chemical treatment, comprising the following elements: a vector memory containing a signature vector library, the coordinates of each signature vector corresponding to measurement signals associated with a reference pattern among reference patterns; A scalar memory containing a set of parameter vectors, the coordinates of each parameter vector corresponding to geometric and / or physical parameters of a reference pattern, each parameter vector being associated with a signature vector; Means for iteratively determining a current signature vector from measurements made on the pattern; a vector processor associated with the vector memory and adapted to iteratively determine, in the library, the K signature vectors closest to the current signature vector, K being an integer greater than or equal to 2; and a scalar processor associated with the scalar memory and adapted to iteratively select, from among the K signature vectors, the signature vector which minimizes the sum B8829

6 weighted by a distance between the current signature vector and said signature vector and a criterion implementing the parameter vector associated with said signature vector and at least the parameter vector associated with at least one of the selected signature vectors at least a previous iteration, and to determine a current parameter vector from the parameter vector associated with the selected signature vector. According to one embodiment of the present invention, the means for determining the current signature vector is adapted to perform spectroscopic ellipsometry measurements. According to one embodiment of the present invention, the vector processor is a graphics processor. It is also intended to use the above device for tracking geometric and / or physical parameters of a diffraction grating. BRIEF DESCRIPTION OF THE DRAWINGS These and other objects, features, and advantages will be set forth in detail in the following description of particular non-limiting embodiments in connection with the accompanying figures, in which: FIG. described, illustrates a device for acquiring a scalerometric signature; Figure 2, previously described, illustrates an example of a pattern; Figure 3, previously described, illustrates an example of a scalerometric signature; Fig. 4 is a block diagram of a real-time tracking device according to one embodiment; Figure 5 illustrates the formation of a vector associated with a scalerometric signature; Fig. 6 is a flowchart of a real-time tracking method according to one embodiment; Figs. 7A-7C illustrate a library of scalerometric signatures according to one embodiment; B8829

FIGS. 8, 9 and 10A to 10D illustrate steps of searching for K vectors closest to a vector in the library of FIGS. 7A to 7C; and Figs. 11A-11D illustrate further steps of the method of Fig. 6. For the sake of clarity, like elements have been designated by like references in the various figures. Detailed Description In the remainder of the description, a signature 10 is a set of values measured by an acquisition device. In addition, the last signature provided by the signature acquisition device is called the current signature. In order to carry out real-time monitoring of the evolution of geometrical and / or physical parameters of a pattern, the requestor has defined a method and a device allowing a fast search, in a library of signatures, of the signature. closer to a current signature. To do this, the applicant has taken advantage of the parallel processing properties of vector processors such as graphics processing units (GPUs), for example graphics processors marketed by Nvidia, ATI or Intel, in association with processors. classic scalars (for example any usual processor of a computer). Moreover, in order to follow the evolution of the geometrical and / or physical parameters of the pattern, the Applicant proposes to carry out signature measurements as often as possible but with a reduced resolution to reduce the acquisition time of each signature. For example, in the case of acquisitions made by spectroscopic ellipsometry, the Applicant 30 plans to make measurements for a reduced number of wavelengths. The values measured at the different wavelengths are assembled in the form of a vector which is subsequently called a signature vector. By way of example, in the case of the signature of FIG. 3, 4, 8, 16 or 32 measurement points corresponding to 2, 4, 8 or 16 wavelengths B8829 can be taken.

8 to form a signature vector having 4, 8, 16 or 32 coordinates. The low resolution of each acquisition is offset by a high acquisition frequency. Each signature vector of the library is associated with a parameter vector whose coordinates correspond to the different geometrical and / or physical parameters of the considered pattern. To perform the different measurements in real time, a measurement system such as that of FIG. 1 is used. It is possible to use a single pair of polarizers / analyzers and to scan a range of wavelengths or to provide several pairs of polarizers / analyzers, each torque being dedicated to one of the wavelengths selected for acquisition. Preferably, a single pair polarizer / analyzer associated with several photomultipliers placed in parallel is used. Figure 4 is a block diagram of a device 20 for determining geometric and / or physical parameters of a pattern 21. The shape and / or structure of the pattern 21 is likely to change over time. For example, the optical index of the material forming the pattern may be one of the physical parameters that is monitored. The evolution of the pattern may be due to physical and / or chemical treatments applied to the pattern, for example a heating operation, etching, etc. The device 20 comprises an acquisition device 23 such as the device of FIG. 1 which measures the change in the polarization state of a beam reflected by the pattern and which provides the coordinates of the signature vector associated with the pattern. The acquisition device 23 communicates with a scalar processor 25. The device 20 also comprises a vector processor 27 which communicates with the scalar processor 25. The scalar processor 25 is associated with a scalar memory 29 and the vector processor 27 is associated with a vector memory 31.

B8829

In vector memory 31, a library of signature vectors Xi corresponding to different patterns j is stored, j being an integer between 1 and N, N being an integer corresponding to the number of signature vectors of the library. The N signature vectors of the library correspond to different values of different geometrical and / or physical parameters of known patterns. In scalar memory 29, geometric and / or physical parameters of known patterns (e.g. parameters h, CD, and 0) are stored. We call Ti the set of geometric and / or physical parameters associated with a pattern j. Subsequently, a set of parameters Ti will be called parameter vector. For example, the scalar processor can store data in the following way: j = 1, T1 = (h1, CD1, 01) j = 2, T2 = (h2, CD2, 02)

j = N, TN = (hN, CDN, 0N). Thus, in the scalar memory 29, there are stored parameter vectors Ti while in the vector memory, a library of corresponding signature vectors Xi is formed. The index j associated with the signature and parameter vectors allows the link between the two sets of vectors. FIG. 5 illustrates an example of a signature vector 25 stored in the vector memory 31. FIG. 5 illustrates a scalerometric signature consisting of two curves 15 and 17. The signature vector Xi associated with the signature of FIG. 5 comprises M coordinates x1'7 , i varying between 1 and M. As an example, in FIG. 5, M = 8 is taken. In the example shown, the coordinates of the signature vector Xi are associated with 4 acquisition wavelengths. X1, X2, X3 and X4. For each of these wavelengths, the point of intersection with the curves 15 and 17 is determined. For example, the coordinates x1 7, x2, j, x3 'and x4,7 of the vector X1 can be equal to the values on curve 15, resp.

10, for the wavelengths X1, X2, X3 and X4 and the coordinates x5j, x67, x7,7 and x8, j can be equal to the values on the curve 17, respectively, for the wavelengths X1, X2, X3 and X4. It will be noted that the coordinates of the signature vectors Xi may be defined in an order different from that presented here. Figure 6 is a flowchart illustrating a method implemented here. The initial step 41 of the method consists in forming a library of signature vectors in the vector memory 31, each signature vector of the library being associated with a pattern j, and storing parameter vectors Ti associated with the different patterns j in the scalar memory . The next step 43 is to perform an initial acquisition, prior to the start of pattern editing, to determine the initial geometric and / or physical parameters of the static pattern. The initial acquisition is performed by any known method, for example by plotting the complete scalerometric signature of the pattern and then comparing this signature to complete signatures stored in a library to obtain the signature closest to that of the pattern and the geometric parameters and / or physical associated with it. The initial acquisition can also be performed by an AFM (Atomic Force Microscope) measurement which is slow but accurate. This step makes it possible to know precisely the geometrical and / or physical parameters of the initial pattern, for example h, CD and e. In step 45, the processing modifying the shape and / or structure of the pattern begins. This treatment can be an etching, a heating operation causing the creep of the pattern, or any other chemical and / or physical treatment that is to be followed in real time. In step 47, an acquisition of the current signature vector is performed. The current signature vector includes the same B8829

11 number of coordinates that the signature vectors of the library 31, these coordinates being obtained, for example, as described in relation to Figure 5. In step 49, a search is performed in the signature vector library 31 , K signature vectors closest to the current signature vector. This search is performed by the vector processor 27, as we will see later in connection with Figures 9 and 10A to 10D.

In step 51, one or more time filtering operations are carried out making it possible to choose, from among the K signature vectors, the signature vector closest to the current signature vector. At least one of these filtering operations is performed taking into account, on the one hand, the proximity of the K signature vectors and the current vector and, on the other hand, the evolution of the parameters over time. In step 53, it is decided to continue, stop or modify the treatment applied to the pattern based on the results obtained in the previous step. If the processing is to continue, return to step 47 where a new acquisition of a current vector is performed. If it is decided to stop or modify the treatment applied to the pattern, the process proceeds to step 53 in which the treatment is stopped or modified. If it is decided, in step 53, to modify the treatment, then one can return to step 47 to continue monitoring pattern evolution. The different steps of the method of FIG. 6 are described in more detail below, step 41 in relation to FIGS. 7A to 7C, step 49 in relation with FIGS. 8, 9 and 10A to 10D and FIG. step 51 in connection with FIGS. 11A to 11D. FIGS. 7A to 7C illustrate the storage of a library of signature vectors in the vector memory 31 and, more precisely, in a vector memory associated with a graphics processor.

B8829

The data stored in a vector memory associated with a graphics processor are divided into textures. Each texture has four layers, a red layer R, a green layer G, a blue layer B and a transparency layer oc. Each layer comprises a matrix of memory cells. The layers R, G, B and I are shown in the figures, one above the other. Each memory cell of a texture is associated with coordinates p, q, r, p and q being, respectively, the number of the row and the column of the memory cell on a layer of the texture and r being the number of the layer of the memory cell, r varying between 1 and 4. For example, we will take r = 1 for the layer R, r = 2 for the layer G, r = 3 for the layer B and r = 4 for the oc layer. Thus, an element whose index is p, q, r is stored in the memory cell of the p-th line, q-th column and r-th layer. In the remainder of this description, a layer R, G, B, a of a given texture will be designated by adding, in index of this layer, the name of the associated texture (for example RIND for the layer R of a texture called IND).

In addition, the association of the four cells formed in identical lines and columns of the four layers is called "texel" (abbreviation of the English expression "texture element"). Subsequently, a texel will be identified by adding, as an index, its associated row and column, p and q.

According to one embodiment, each x1'7 coordinate of a signature vector Xi of the library is stored in a different texture but in identical coordinates cells p, q, r. Thus, a first texture receives the first coordinates x1 7 of each vector Xi, a second texture receives the second coordinates x2,7 of each vector Xj, and so on. FIG. 7A illustrates the memorization of the first coordinate of each signature vector of the library in a first texture called C1 and FIG. 7B illustrates, by way of example, the memorization of an Mth and last coordinate.

13 data of each vector of the library in an Mth texture called CM. In the example shown, a coordinate xlpq, is the i-th coordinate of the vector Xi, stored in the memory cell coordinates p, q, r. Thus, the texture of FIG. 7A includes all the coordinates xpq ,,, of the vectors Xi, and the texture of FIG. 7B includes all the coordinates x P, q ,,, of the vectors Xi. FIG. 7C illustrates an additional IND index texture formed in the graphics memory 31, this texture comprising the indices j associated with each signature vector Xi. This texture makes it possible to link the signature vectors of the vector memory and the parameter vectors of the scalar memory 29. The index texture comprises, in the memory cell associated with the vector Xi, the index j. Thus, the coordinates of each signature vector Xi are placed in the same memory cell of the textures C1 to CM and the index j associated with the signature vector Xi is placed in the same memory cell of the index texture IND.

For example, in the textures, the vectors Xi were distributed first along the same texel, then in line and column by column in the layers. Note that the placement order of the signature vectors in the textures does not matter.

FIG. 8 illustrates the result of a preliminary step in search of K vectors closest to a current vector in a library such as that of FIGS. 7A to 7C, K being a positive integer greater than 1. A current signature vector Y is obtained using the acquisition device 23 and is stored in the scalar memory 29 of the scalar processor 25. Using two initially empty working textures, calculations are performed to obtain a texture DIS comprising the distances between each of the vectors Xi of the library and the current vector Y. We note here y ', i varying from 1 to m, the coordinates of the vector B8829

For example, in the case where the distance considered is associated with the norm 1 (Manhattan distance), in the coordinate cell p, q, r of the texture DIS (FIG. ), the term: ap, 9, r = xp 1j 2, j, 9, r Y1 + xp, 9, r Y2 + ... + xB, J r _Y8 To obtain the term ap, q, r, we can work alternately on the two initially empty working textures. Indeed, a graphics processor can realize, in a clock cycle, only operations on a limited number of memory cell elements. Thus, to obtain the terms ap, q, r, a number of intermediate steps may be necessary, depending on the number of coordinates M of the signature vectors. The intermediate steps will not be described in more detail here. In the figures, the textures 15 obtained during the different steps and not the intermediate work textures obtained during the intermediate steps are shown. It will be understood that the standard used to form the elements of the DIS texture may be any other known standard adapted to the intended application, for example, Standard 2 (Euclidean standard). Thus, as illustrated in FIG. 8, a texture DIS is obtained comprising, at each location p, q, r, the distance ap, q, r between the signature vector Xi associated with the location p, q, r and the current signature vector Y. The use of a vector processor to perform this step allows the calculation of all distances ap, q, r of the texture substantially simultaneously. Once the texture DIS has been obtained, the search for the K vectors closest to the current vector Y is equivalent to the search for the K smallest distances stored in the texture DIS. To obtain the K smallest distances of the texture DIS, a first step is carried out which makes it possible to form a texture MIN in which distances resulting from a first selection are stored in the layers RMIN and GMIN5 B8829

15 and in which the indices associated with these distances are stored in the layers BMIN and IMIN. In the case where it is desired to obtain K vectors closest to the current vector, this first minima search step makes it possible to pass the texture. DIS to the MIN texture is performed on groups comprising A texels. On the distances contained in these A texels of the DIS texture, the smaller K's are stored in selected cells of the RMIN and GMIN layers of the corresponding A texels of the MIN texture. The indices associated with these distances are stored in the corresponding cells of the layers BMIN and MIN • By working on initial A texels of the texture DIS, up to K = 2A smaller distances can be stored in the texture MIN. It is also possible to store a number K <2A distances in the texture MIN, not completing all the cells of this texture. In the case where K> 2, calculations similar to those presented below in relation with FIGS. 10A to 10D (where K = 2) are performed on the distances of the RMIN and GMIN layers • FIG. 9 illustrates the result of such a first step in the case where it is desired to find K = 2 vectors closest to the current vector Y. This first step of the search consists in working texel by texel (A = 1) on the texture DIS. On the four distances stored in each texel p, q of the texture DIS, we look for the two smaller ones that will be stored in the RMIN and GMIN layers of the texel p, q. This halves the number of distances from the DIS texture to the MIN texture.

More precisely, in each texel p, q of the texture MIN, the layer RMIN receives the smallest distance of the texel p, q corresponding to the texture DIS, the layer GMIN receives the second smallest distance of the corresponding texel from the texture DIS, the BMIN layer receives the index (IND texture) associated with the first smallest distance and the aMIN layer receives B8829

16 index (IND texture) associated with the second smallest distance of the texel considered. Thus, on the RMIN layer, a matrix of first minima corresponding to the first minima of each texel of the DIS texture is obtained. The first minima are called (3p (RMIN layer), where p and q are, respectively, the line and the column of the texel under consideration.The GMIN layer receives the second minima of each texel, called yp, q. (3p, q are called ip, q (BMIN layer) and the indices associated with the second minima yp, q are called i'p, q (OEMIN layer) Figures 10A to 10D illustrate steps of finding the two smallest distances Figure 10A shows the memory cells of the RMIN layer which contain the distances 13p, q and the memory cells of the layer GMIN which contain the distances yp, q, p and q being, respectively, the line numbers and of column of the considered cells, p and q varying from 1 to s = 2s, s being a positive integer In the example represented, p and q vary from 1 to 8 (s = 3).

The proposed method of searching for the two smaller minima in the RMIN and GMIN layers is to reduce the number of distances stored in these layers by a factor of 4 at each iteration and then repeat this operation as many times as necessary. In the example of FIG. 10A, to divide the number of minima by a factor of 4, the layers RMIN and GMIN are virtually divided into four neighborhoods QI to Q4, the neighborhood QI corresponding to the cells 13p, q and yp, q with 1 <_ p <_ S / 2 and 1 <_ q <_ S / 2, the quarter Q2 at cells 13p, q and Yp, q with S / 2 <_ p <_ S and 1 <_ q <_ S / 2, the neighborhood Q3 at cells 13p, q and yp, q with 1≤p S / 2 and S / 2≤ q S, and the neighborhood Q4 at cells 13p, q and yp, q with S / 2 <_ p <_ S and S / 2 <_ q <_ S. Cell elements from the same locations in the four neighborhoods of the RMIN and GMIN layers are then compared. Figure 10B illustrates RMIN2 and GMIN2 layers of a MIN2 texture obtained after a first iteration. In B8829 17 each cell of the neighborhood QI of the layer RMIN2, the first minimum 5'p, q of each of the distances (3p, q and yp, q located in the cells of the same locations of each of the neighborhoods of the RMIN and GMIN • In each cell of the neighborhood QI of the layer GMIN2, we memorize the second minimum y'p, q of each of the distances (3p, q and Yp, q located in the cells of the same locations of each of the neighborhoods of the layers RMIN and GMIN • Thus, the distances 5'p, q and y'p, q which are situated in line p and in column q (p and q ranging from 1 to S / 2), respectively, RMIN2 and GMIN2 of FIG. 10B are: R p, q - min 1 jp, q '13p, q' Yp + S ~, q'Yp, q 2 Y p, q = min21 [3p, q 'R p +%, q 'Rp, q + ~' p + ~, q'Yp, q'Yp + ~, q + ~ the operators "min" and "min2" being, respectively, operators returning the first and second minima of a set of values. , the RMIN2 and GMIN2 layers contain four times less distance than the RMI layers N and GMIN • The RMIN2 and GMIN2 layers are obtained using the parallel computing capabilities of the vector processor. This step has the advantage of only lasting a few clock cycles. At each iteration, the indices i and i 'corresponding to each distance 5'p, q and y'p, q are stored in the texel p, q, respectively in the layers BMIN2 and aMIN2. Then, the operation allowing to pass The RMIN and GMIN layers at the RMIN2 and GMIN2 layers of Figure 10B are repeated at the lower left-hand side of the RMIN2 and GMIN2 layers of Figure 10B. Thus, the number of distances is divided again by four. Figures 10C and 10D illustrate the following two iterations. In FIG. 10C, the elements R "p, q of a layer RMIN3 and y" p, q of a layer GMIN3 are defined by (p and q ranging from 1 to S / 4). p, q - mini {p, q'R p + /, q'R p, q, Y p, q'Y p 4} B8829

Y p, 9 = min 2 R p, 4 'R p 4' R p, 4 + / '13'p' yp, 4 'y p + ~, 4 yp, 9'yp and, in Figure 10D, the cells 13 p, q of a layer RMIN4 of a layer GMIN4 are defined by: and Yp, qp, q "'p, q"' R "p, q" 'R p, 4 + 8' Y "p, Q ' p, qp, q, 7 + In the example shown, where s = 3 and thus S = 8, there are, in the layers RMIN4 and GMIN4, unique elements which are, respectively: P1,1 = Min 1 "1, 1 "," 2, "2.11", 2.21f and yl, 1 = min 2 "1.1, R" 1.2, R "2.1, R" 2.2 The distance (311 is the smallest distance memorized in the RMIN and GMIN layers of the MIN texture and the distance y11 is the second smallest distance stored in the RMIN and GMIN layers of the MIN texture, and the i and i 'indices associated with the distances (311 and yll are stored in the first texel (first line and first column) of the final texture, respectively, in the layers BMIN4 and GMIN4. Thus, by carrying out the operations defined in relation to FIGS. 10A to 10D on FIGS. In the case of the first line and the first column, two vectors closest to the current vector Y and the indices associated with these two closest vectors are obtained in the texel of the first line and the first column. Using the graphics processor for these steps provides the two closest vectors quickly. It should be noted that similar operations on the MIN texture can be performed to obtain more than two smaller distances. FIGS. 11A to 11D illustrate, in greater detail, the results obtained during steps 49 and 51 of the method of FIG. 6 in the case where K = 4 (search for 4 vectors closest to each current vector Y). FIGS. 11A to 11D show curves having as ordinates one of the coordinates of the parameter vectors Ti associated with the vectors Xi, that is to say B8829

19 a geometrical and / or physical parameter that one wishes to follow in real time, and whose abscissa is time. By way of example, the parameter represented corresponds to the half-height width CD of the pattern 13 of FIG.

FIG. 11A illustrates the result obtained after searching for four signature vectors Xi closest to the current signature vector in the library, as a function of time. The four values of the parameter CD, some of which are possibly equal, are obtained at each instant of acquisition of a current vector Y by associating the four vectors Xi closest to the vector Y to the associated parameter vectors Ti stored in the memory 29, via the corresponding indices j. In this figure, we find that the search for the 4 closest vectors has aberrant variations.

At the step illustrated in FIG. 11B, a first temporal filtering was carried out which makes it possible to choose, from the four vectors Xi closest to the current vector Y, the vector closest to the current vector in the sense of a criterion which takes in account the distance between the vectors (for example associated with the norm 1) but also a temporal parameter. Thus, for example, one of the four vectors Xi, the vector which minimizes, for the current vector Y acquired at time t, the criterion C given by the following formula: (t) τT (t00t) P being a weighting parameter chosen in a suitable manner, where At is the time separating two acquisitions, Ti being the parameter vector associated with the signature vector Xi and T (t-At) being the parameter vector associated with the closest signature vector of the current vector of the vector previous acquisition, at t-At. Thus, the vector closest to the current vector Y acquired at time t is determined by minimizing the sum of a first term that takes into account the proximity between the vectors Xi and Y and a second term that takes into account the evolution of parameters over time. This step is performed by the B8829 scalar processor

25. Thus, the curve 61 shown in solid lines is obtained in FIG. 11B. The initial measurement of step 43, which makes it possible to completely define the initial structure of the pattern, makes it possible to know an exact value of the geometrical and / or physical parameters of the initial pattern in order to start the iterative process correctly. FIG. 11C illustrates the result obtained (curve 63) after smoothing the curve of FIG. 11B.

Unlike temporal filtering to obtain the curve 61, the smoothing making it possible to obtain the curve 63 is carried out not on the vectors but on each of the geometrical and / or physical parameters of the parameter vectors, for example on the parameters h, CD and O. For example, the curve 63 can be obtained from the curve 61 of FIG. 11D by working on the first or second derivatives of the curve 61. Thus, the smoothing can be shifted in time by a few iterations. For example, the smoothing can be obtained by the method of Tikhonov.

FIG. 11D represents a comparison between the curve 63 (in solid lines) and a theoretical curve 65 (in dashed line) of evolution of the parameter CD. Note that the differences between curves 63 and 65 are minimal. Thus, the approximations made during the search for the K nearest vectors and during successive filtering involve little difference from the theory. Thus, the device and method presented herein provide for the use of a vector processor and a scalar processor for the purpose of combining the capabilities of these processors for fast and improved reconstruction of geometric and / or physical parameter variations. of a pattern during an iterative process of monitoring the evolution of the pattern in real time. The vector processor makes it possible to quickly find the K vectors closest to the current vector in the signature vector library since it produces B8829

21 calculations (search for distances) simultaneously on the N signature vectors of the library. The scalar processor is then used to process a limited number of vectors, K in the present case, especially during time filtering.

Thus, the search for K nearest vectors and successive filtering can be performed in real time. The method and the device according to one embodiment allow the real-time monitoring of the evolution of a reason subject to a treatment and, if necessary, the modification or the stop of the treatment. By way of example, the inventors have shown that the use of a graphic processor for the search of K vectors closest to a current vector is 15 times faster than the same search carried out by a conventional scalar processor for important vector libraries (vector memory with a capacity of 400 to 600 megabytes, for example). Particular embodiments of the present invention have been presented. Several variations and modifications will occur to those skilled in the art. In particular, the considered patterns may have any periodic shape that can be defined using a finite number of geometric and / or physical parameters. It will also be possible to follow the evolution of patterns comprising a single isolated form on the surface of a support.

By way of example, the method according to one embodiment may be used in the manufacture of image sensors for monitoring in real time the creep of resin patterns forming lenses above the elements of the image sensors. It can also be used to monitor the fabrication of nanoscale diffraction gratings. The method described herein may also be used to monitor the fabrication of circuits having a periodic structure, the beam 7 then pointing directly to the circuit. It can also be used to monitor a control zone on the edge of the circuit (cutting line) which contains a periodic pattern of which B8829

22 the variation makes it possible to follow the manufacturing process of the chip. In addition, each signature vector may have a shape different from that presented here and the scalerometric signatures may be obtained by scattometry methods other than spectroscopic ellipsometry. Note also that it was considered here, by way of example, that the vector processor was a graphics processor. As a variant, this processor may be any other type of vector processor associated with a suitable memory. For example, we can use a STI Cell processor (Sony Computer Entertainment, Toshiba and IBM), or a dedicated FPGA (field-programmable gate array).

Claims (10)

REVENDICATIONS1. A method for real-time monitoring of the evolution of geometrical and / or physical parameters of a diffractive pattern (21) during a physical and / or chemical treatment, comprising: before the start of processing: storing in a memory vector (31) associated with a vector processor (27), a signature vector library (Xi), the coordinates of each signature vector corresponding to measurement signals associated with a reference pattern among reference patterns; storing, in a scalar memory (29) associated with a scalar processor (25), parameter vectors (Ti), the coordinates of each parameter vector corresponding to geometric and / or physical parameters of a reference pattern, each parameter vector being associated with a signature vector, and during processing, iteratively determining a current parameter vector by repeating the following steps: determining a current signature vector from measurements made on the pattern (21); causing the vector processor to determine, in the library, the K signature vectors closest to the current signature vector, K being an integer greater than or equal to 2; causing the scalar processor to select, from among the K signature vectors, the signature vector which minimizes the weighted sum of a distance between the current signature vector and said signature vector and a criterion implementing the parameter vector associated with said signature vector and at least the parameter vector associated with at least one of the signature vectors selected at least one preceding iteration; and determining the current parameter vector from the parameter vector associated with the selected signature vector. 2. Method according to claim 1, wherein the current parameter vector is determined from the parameter vector associated with the selected signature vector and from at least one current parameter vector determined at a previous iteration. The method of claim 2, wherein the current parameter vector is further determined from at least one parameter vector associated with a selected signature vector at a next iteration. The method according to any one of claims 1 to 3, further comprising an initial step of determining, prior to the start of treatment, an initial parameter vector of the diffractive pattern. The method of any one of claims 1 to 4, further comprising deciding whether to stop or modify the patterned processing from the current parameter vector. The method of any one of claims 1 to 5, wherein the determination of the current signature vector is performed by spectroscopic ellipsometry. 7. Apparatus for real-time monitoring of the evolution of geometric and / or physical parameters of a diffractive pattern (21) during a physical and / or chemical treatment, comprising the following elements: a vector memory (31) ) containing a signature vector library (Xi), the coordinates of each signature vector corresponding to measurement signals associated with a reference pattern among reference patterns; a scalar memory (23) containing a set of parameter vectors (Ti), the coordinates of each para-meter vector corresponding to geometric and / or physical parameters of a reference pattern, each parameter vector being associated with a signature vector B8829 means for iteratively determining a current signature vector (23) from measurements made on the pattern; a vector processor (27) associated with the vector memory (31) and adapted to iteratively determine, in the library, the K signature vectors closest to the current signature vector, K being an integer greater than or equal to 2; and a scalar processor (25) associated with the scalar memory (29) and adapted to iteratively select among the K signature vectors the signature vector which minimizes the weighted sum of a distance between the current signature vector and said vector signature and a criterion implementing the parameter vector associated with said signature vector and at least the parameter vector associated with at least one of the signature vectors selected at at least one preceding iteration, and determining a current parameter vector from the parameter vector associated with the selected signature vector. 8. Device according to claim 7, wherein the means for determining the current signature vector is adapted to perform spectroscopic ellipsometry measurements. 9. Device according to claim 7 or 8, wherein the vector processor is a graphics processor. 10. Use of the device according to any one of claims 7 to 9 for tracking geometric and / or physical parameters of a diffraction grating.