WO2006106556A1

WO2006106556A1 - Method and system for facilitating the determination of the end point in plasma etching processes

Info

Publication number: WO2006106556A1
Application number: PCT/IT2006/000223
Authority: WO
Inventors: Diego Zappa; Giuseppe Fazio; Roberto Colombo
Original assignee: Stmicroelectronics S.R.L.
Priority date: 2005-04-05
Filing date: 2006-04-04
Publication date: 2006-10-12

Abstract

A method for facilitating the determination of the end point of a dry plasma etching process of a material, is proposed. The method includes performing an analysis of the whole spectrum of a radiation generated during the plasma etching process of the material, the analysis comprising evaluating the time trend of a plurality of spectral components of the radiation, each spectral component indicating the time trend of the radiation intensity of a corresponding wavelengths interval of the radiation. The method further includes, on the basis of such analysis, selecting at least one of the spectral components, wherein the at least one spectral component has a time trend indicative of the evolution of the etching process of the material. The performing of the spectral analysis comprises performing a statistical analysis of the time trend of the whole spectrum of the radiation and, on the basis of the results of the statistical analysis, selecting the at least one spectral component.

Description

METHOD AND SYSTEM FOR FACILITATING THE DETERMINATION OF THE END POINT FOR PLASMA ETCHING PROCESSES

The present invention generally refers to the field of the manufacturing processes for semiconductor devices; in particular, the invention regards dry etching processes, particularly, dry plasma etching processes, and specifically a method for detecting the end point of the etching process, i.e., for ascertaining the completion of the etching process.

In the field of the manufacturing processes for microelectronic technologies, detecting the achievement of the "Etching Process" (EP) of a dry plasma etching process for correctly defining the process parameters is a really critical aspect.

Particularly, referring to Figure 1, three generic steps Ia, Ib, Ic of an exemplary dry plasma etching process are shown, which correspond to three subsequent time instants ta, tb and tc, respectively. A semiconductor wafer 100 comprising a superficial layer 110 of a material (e.g., silicon) to be removed is placed inside a reaction chamber for dry plasma etching processes, schematically illustrated in Figure 1 with the reference numeral 120. When the etching process is started at the time t=ta (step Ia), the wafer 100 is subjected to a plasma flux (not shown in the Figure) for the selective removal of the silicon forming the superficial layer 110. The achievement of the EP can be observed by analyzing the time trend of the radiation spectrum present within the reaction chamber 120 in which the etching process is performed. Particularly, it may be useful performing such analysis in connection with particular selected wavelengths, having, for example, some correlation with the characteristic wavelengths of the emission spectrum of the material of the superficial layer 110 to be removed (in this case, silicon), also in relation to the etching chemistry that is used. By way of example, Figure 1 includes a time diagram 130 showing the trend of the radiation intensity IwI of one of such selected wavelengths depending on time: as long as the superficial layer 110 to be removed has silicon molecules, the etching process continues to trigger radiation emissions within the reaction chamber 120, corresponding to such selected wavelength. Later, at a subsequent time instant t=tb (step Ib)₅ it can be observed that the superficial layer 110 is almost entirely removed; at this point, the radiation intensity IwI corresponding to the wavelength tends to decrease in a relatively fast way. In other words, the EP is almost reached. Lastly, at the time t=tc (step Ic) the superficial layer 110 is completely removed, and the radiation intensity at the selected wavelength is substantially zero. In the practice, the behavior of the intensity of the radiation emission at the selected wavelength is much more complex; indeed, such behavior does not exhibit a simple monotonic trend as shown in the time diagram 130, and the indication of the EP is less evident. As a consequence, the EP will be generally made to correspond to the time instant at which a strong (or fast) variation in the radiation emission at the selected wavelength is observed. In other words, once a wavelength has been selected, the time instants which are candidate to correspond to the EP are those at which the corresponding trend shows a high non-linearity.

The optical spectrometry technology is usually exploited in order to perform a temporal observation of the spectrum within the reaction chamber, which is becoming the best and the most flexible operative control instrument for dry plasma etching processes. However, the analysis and the observation of the radiation spectrum detectable within the reaction chamber is a really complex operation, because such radiation comprises several components, not all of which being easily identifiable and forecasted. In fact, the detectable spectrum comprises, in addition to the (desired) components due to the emission of the material under etching, all the components due to the material layers below and/or above the layer to be removed. Moreover, a further aspect strongly affecting the total spectrum is the radiation spectrum of the plasma itself, which can vary in a non-negligible way according to the interaction between the plasma radiation and the radiation (of variable intensity) emitted by the material of the layer under etching. Another aspect which is not to be neglected for observing the detectable spectrum is the type of reaction chamber used during the process, because each reaction chamber may behave in a different way.

It is of the utmost importance having an accurate estimation of the achievement of the EP in all the cases in which the etching process is not higly selective. In fact, an inaccurate estimation of the achievement of the EP would involve the execution of over dimensioned or under dimensioned etching of the material layer to be removed which, even if of reduced entity, would provoke the changing of the desired performances of the semiconductor device achievable by means of the wafer processing.

Making reference to Figure 2, the main steps of a known method for detecting the EP are illustrated in a more detailed way.

After having placed a silicon wafer into a reaction chamber in which a proper plasma has been supplied, the first step consists of acquiring the spectrum of the radiation present within the reaction chamber. Such spectrum, which is variable in the time depending on the progress of the etching process, is acquired by means of known optical-spectrometry techniques. The result of such spectrum acquisition is for example provided by the spectrometer in form of a numerical matrix, comprising the substantial data describing the time trend of the acquired spectrum. For example, a typical matrix of this type is composed by a number of columns depending on the width of the spectrum interval under examination. In particular, a particular wavelength (or an interval of wavelengths centered around a central wavelength) corresponds to each column. The elements of each column represent the radiation intensities detected at the corresponding wavelength, as a function of the time instant at which the measurement has taken place: in fact, each row of the matrix corresponds to a particular time instant. In this way, a column of the numerical matrix, i.e., a particular wavelength of the radiation spectrum, is selected, and then the time trend of the radiation intensity is observed by scanning the elements of the column row by row. The three subsequent steps are manually performed by a process engineer, based on his/her previous experiences and on the literature.

In particular, the second step consists of empirically and manually searching a best wavelength (or "optimal" wavelength) among those composing the detected spectrum. A wavelength is defined as optimal when it is possible to deduce a clear indication of the EP by observing the time trend of the radiation intensity corresponding to such wavelength. Said best wavelength is in particular deduced by analyzing the temporal evolution of the detected spectrum during the entire etching process; once the best wavelength is selected (thus, once the corresponding column is selected within the numerical matrix), the operations of the following steps are performed on said spectrum component only. This step may require an excessive time (even of the order of several days) if the type of etching is particularly difficult, as in the case in which the zone to be etched on the wafer is poorly exposed, or has a reduced superficial area.

The third step consists of removing the noise capable to distort the time trend of the radiation intensity at the selected best wavelength (in short, the selected trend) by means of a proper filtering operation.

The fourth step consists of searching the "best" algorithm to be used for detecting the EP. For this purpose, a proper sequence of conditions is defined that must be respected by the selected time trend for providing indications useful for detecting the desired EP. Said conditions may depend on features that are intrinsic to the process, like for example the plasma type, the material type, the surface dimension and thickness of the layer to be etched, or depend on features depending on the type of reaction chamber, such as for example the etching speed of the reaction chamber. Once the algorithm has been selected, an EP is consequently deduced. Said

EP is a candidate to be the best EP for the considered etching process. Subsequently, a high number (for example, one hundred) of wafer are tested; said wafers are subjected to the etching process using the EP previously calculated by means of the selected algorithm. By observing the wafers processed in this way, it is possible to know if the etching process has been successful or not, that is, the correctness of the calculated EP. If the etching process has not been successful (for example, the layer has not been entirely removed, or a lower layer, which has not to be etched, has been notched), it is not easy to know the reason. In fact, the causes producing an erroneous etching process may be multiple; for example, the wavelength which has been selected may not be really the best for the etching process, or the algorithm for the search of the EP may include an error or the reaction chamber may had a malfunctioning.

In the case in which the results of the previous step has given satisfactory results, the last step consists in a pre-production step of a further high number of processed wafers, for the purpose of consolidating and reconfirming the data obtained during the test step. The need of executing a such high number of test and pre-production processes is mainly due to the empiricism of the system used for searching the best wavelength. In fact, a process engineer has to conduct a careful experimental analysis on an high number of wafers, before being sure that the found EP is effectively adapted to be used for the particular used etching process, for the purpose of discarding possible positive results due to fortuitous results not dependent to the optimal selection of the EP. In fact, an experimental verify on a small number of wafer may also provide inspected results because of a temporary and not observable alteration of the reaction chamber conditions. As a consequence, using the known method for detecting the EP, the complete transfer of the process from the development phase to the production phase is often affected by significant delays.

In order to improve said method, and in particular to reduce the setting times of the etching process, the Applicant has found an alternative methodology, as indicated in the first claim.

Particularly, a method for facilitating the determination of the end point of a dry plasma etching process of- a material is provided. Such method includes performing an analysis of the whole spectrum of a radiation generated during the plasma etching process of the material. The analysis comprises evaluating the time trend of a plurality of spectral components of the radiation. Each spectral component indicates the time trend of the radiation intensity of a corresponding wavelengths interval of the radiation. The method further includes, on the basis of such analysis, selecting at least one of said spectral components, wherein said at least one spectral component has a time trend indicative of the evolution of the etching process of said material. Said performing the spectral analysis comprises performing a statistical analysis of the time trend of the whole spectrum of the radiation and, on the basis of the results of the statistical analysis, selecting the at least one spectral component.

Further features and advantages of the solution according the present invention will be best understood by reference to the following detailed description, given purely by way of a non-restrictive indication, to be read in conjunction with the accompanying drawings:

Figure 1 illustrates in a schematic way a wafer subjected to an etching process in a reaction chamber for dry plasma etching process, and the time evolution of the radiation intensity corresponding to a particular wavelength; Figure 2 shows the process steps for detecting the EP of a dry plasma etching process according to a methodology known in the art;

Figure 3 shows a scheme of a system adapted to implement the method for searching the optimal wavelengths for detecting the EP of a dry plasma etching process, according to an embodiment of the invention; Figure 4 schematically illustrates the structure of a computer which may be included into the elaboration system of Figure 3;

Figure 5 illustrates a three-dimensional diagram of the acquisition of a generic radiation spectrum performed by the elaboration system of Figure 3;

Figure 6 shows a three-dimensional diagram of the radiation spectrum of Figure 5 after having been subjected to a smoothing operation;

Figure 7 shows a three-dimensional diagram of the residuals of the smoothed spectrum of Figure 6;

Figure 8 shows a three-dimensional diagram of the modified series obtained from the residual illustrated in Figure 7; Figure 9 comprises two diagrams showing the growths of the cumulated and retrocumulated sums of two modified series of Figure 8;

Figure 10 illustrates two different methods for estimating the convexity of the cumulated sums intensities according to two different embodiments of the present invention; and Figure 11 shows a functional modules diagram of a computer program capable of performing the procedures of the method for searching the optimal wavelengths for detecting the EP of a dry plasma etching process, according to an embodiment of the invention.

By using statistical techniques commonly not used in the field of the controls for the semiconductor processes, the Applicant has found a method for detecting one or more wavelengths among those forming the radiation spectrum in, e.g., a reaction chamber for a dry plasma etching process, adapted to be advantageously used for individuating an optimal EP of the etching process itself.

Such searching method is much faster and efficient with respect to the known methodologies normally used, and allows to obtain much more reliable results.

Moreover, thanks to the reliability of the results obtainable by means of such method, it is possible to reduce the duration of the pre-production phases (i.e., to reduce the number of wafer on which the etching process is applied for the verify purpose).

Particularly, the number of phases will depend on the intensity of the random noise of the etching process under evaluation.

Making reference to Figure 3, a scheme of a system adapted to implement the method for searching the optimal wavelengths for detecting the EP of a dry plasma etching process is illustrated, according to an embodiment of the invention.

A spectrum acquisition system 300 is connected to a reaction chamber 310 in which a semiconductor material wafer 320 is subjected to a dry plasma etching process. In the same way as for the case descripted in the introduction, the spectrum acquisition system 300 performs a temporal analysis of the radiation within the reaction chamber 310, providing as an output a corresponding data set, for example in the form of a matrix 330 including data regarding the acquired spectrum. The matrix 330 includes a plurality of columns c(i) (i=l,2,....,n) and a plurality of rows r(j) (j=l,2,...,m). Each column c(i) corresponds to a wavelength wl(i) (or to an interval of wavelength centered therein) of the acquired spectrum, while each row r(j) corresponds to a time instant t(j). The value stored in the generic element of the matrix 330 at the row r(j) and at the column c(i) represents the measure of the radiation intensity performed by the spectrum acquisition system 300 corresponding to the wavelength wl(i) at the time t(j).

Once the data matrix 330 has been acquired, the choice of the optimal waΛ'elength(s) is not performed in a manual way by means of a process engineer according to his/her experience, but instead it is automatically managed by a processing system 340 implementing a plurality of filtering and statistical processing.

For example, the processing system 340 may include a computer 400, schematically having the structure illustrated in Figure 4. The computer 400 generally includes several functional units connected in parallel to a data communication bus 403 (for example a PCI bus). In particular, a Central Processing Unit (CPU) 405, typically comprising a microprocessor, controls the operation of the computer. A working memory 407, typically a RAM (Random Access Memory) is directly exploited by the CPU 405 for the execution of programs and for temporary storage of data, and a Read Only Memory (ROM) 409 is used for the non-volatile storage of data, and stores for example a basic program for the bootstrap of the computer. The computer 400 comprises several peripheral units, connected to the bus 403 by means of respective interfaces. Particularly, peripheral units that allow the interaction with a human user (for example, the process engineer) are provided, such as a display device 411, a keyboard 413 and a pointing device 415 (for example a mouse). The computer 400 also includes peripheral units for local mass-storage of programs (operating system, application programs) and data, such as one or more magnetic Hard-Disk Drivers (HDD), globally indicated as 417, driving magnetic hard disks, a CD-ROM/DVD driver 419, or a CD-ROM/DVD juke-box, for reading/writing CD-ROMs/DVDs. Other peripheral units may be present, such as a floppy-disk driver for reading/writing floppy disks, a memory card reader for reading/writing memory cards, a Universal Serial Bus (USB) adapter with one or more USB ports, printers and the like. The computer 400 is further equipped with communications peripherals, globally identified by 421; for example, the communications peripherals may include a GSM modem, an RS232 interface, an Ethernet interface. The processing system 340 outputs one or more ordering tables 350 indicating the wavelengths wl(i) of the spectrum that are mere meaningful for detecting the EP. In particular, each ordering table 350 classifies the wavelengths wl(i) reputed to be significant for detecting the EP according to an importance order based on a particular observation method (for example, an ordering table may contain the wavelengths corresponding to the most noticeable temporal variations of the spectrum, or those having the most non-linear behavior), as will be described in detail in the following of the present description.

At this point, the process engineer may perform the choice of the optimal wavelength wl(i) by means of the examination of such ordering tables 350. It has to be noted that the examination job that has to be executed by the process engineer is in this case strongly reduced with respect to the case of the known methodologies, because the ordering tables 350 include all the indications needed for choosing the optimal wavelength wl(i).

The operations of the processing system 340 may be set by regulating several processing parameters PE which are provided thereto. For example, depending on the processing parameters PE, it is possible to set the time interval IT in which the radiation spectrum is analyzed and the number of wavelengths wl(i) that are desired to be visualized in the ordering table 350.

Once the data matrix 330 has been received, the processing system 340 acquires the data included therein and, for example, displays them as a three- dimensional diagram. Particularly, making reference to Figure 5, a diagram 500 exemplificative of the acquisition of a generic radiation spectrum is illustrated; a first reference axis of the diagram 500 (corresponding to the rows of the matrix 330), denoted in the Figure as "t", is used for discriminating the time; a second reference axis (corresponding to the rows of the matrix 330), denoted as "wl", is used for discriminating the wavelengths; a third reference axis, denoted as "IwI", is used for discriminating the radiation intensity.

Selecting a specific value wl(i) among those indicated on the reference axis wl, means selecting the corresponding column c(i) in the data matrix 330. hi particular, the diagram obtained from a such selection, denoted in Figure 5 as 510, represents the time variation of the radiation intensity emitted within the reaction chamber at the wavelength wl(i).

At this point, the processing system 340 carries out a plurality of steps for processing the data acquired by means of the matrix 330. During a first step, the processing system 340 divides the matrix 330 into various sub-blocks, wherein each sub-block is composed by an odd number k (for example, 5) of adjacent columns c(i) (i.e., of adjacent wavelengths wl(i)). For each sub-block, the processing system 340 computes the median of the intensity values corresponding to all the wavelengths wl(i) of the sub-block for each time instant t(j). Particularly, making reference to the matrix 330, the processing system 340 generates a column corresponding to the wavelength wl(i) of the central column c(i) of the sub-block at the place of the selected sub-block. Such column will be designated as "median wl(i)". The median wl(i) is a column in which the element corresponding to the row r(j) is the median value among the radiation intensity values corresponding to the wavelengths of the sub-block at the time t(j). Once the processing system 340 has generated the median wl(i), corresponding to the sub- block formed by k columns having as central column the column c(i), the subsequent median wl(i+l) is generated, which correspond to the sub-block formed by k columns having as central column the subsequent column c(i+l). Therefore, it has to be noted that adjacent sub-blocks include common columns. By joining all the medians wl(i) a median matrix, denoted with 330m, is generated.

During a second step, the processing system 340 executes a "smoothing" action on the values of the median matrix 330m, for example according to a Multivariate Exponentially Weighted Moving Average (MEWMA) statistical model. In this way, a "smoothed" matrix 3301 is obtained, which corresponding graphical depiction is illustrated in Figure 6 by means of the diagram 600.

During a third step, the processing system 340 computes the elements of the smoothed matrix 3301 in such a way to calculate the residual thereof. For this purpose, at the value of each element of each column c(i) corresponding to the row r(j) (and, thus, to the time t(j)) the value of the element in the same column c(i) corresponding to the row r(j-l) (and, thus, to the preceding time t(j-l)) is subtracted. In this way, a residual matrix 33Or is obtained, which three-dimensional picture is illustrated in Figure 7 by means of the diagram 700. The meaning of the residual operation is to provide an indication of the variation speed of the radiation intensity per each wavelength wl(i) during all the considered time interval. For example. Figure 7 illustrates the time trends of the residuals corresponding to two wavelengths wl(i) (diagram 710) and wl(i+h) (diagram 720). According to the diagrams 710 and 720, the residual corresponding to the wavelength wl(i) comprises evident pikes, each one corresponding to a fast variation of the radiation intensity, while the residual corresponding to the wavelength wl(i+h) has a more regular trend.

In a fourth step, the processing system 340 executes a filtering operation on the residuals that have been previously calculated, for the purpose of discarding the residuals contributing to the temporal variations of the radiation spectrum in a non- determinant way, and thus discarding the residuals that are not relevant for identifying the EP of the etching process. The processing system 340 calculates a discrimination threshold sd, settable by the process engineer, such that the percentage of variability explained by the residual exceeding sd is equal to a desired value (for example, the 90%). The aim is to eliminate the values contained into the residual that are interpreted as noise. For each wavelength wl(i), the processing system 340 controls if the corresponding residual reaches (even in correspondence of a single time instant) or exceeds the value of the discrimination threshold sd. hi the affirmative case, the residual (and the corresponding wavelength) is maintained, otherwise it is discarded (i.e., the corresponding column c(i) is removed from the matrix). Referring for example to the two residuals illustrated by means of the diagrams 710 and 720 of Figure 7, it is possible to observe that the residual corresponding to the wavelength wl(i+h) has much more possibilities of being discarded with respect the residual corresponding to the wavelength wl(i), because the latter residual has pikes of much lower intensity. The matrix thus obtained, called "modified" matrix 330t, which columns are referred as "modified series", is depicted in a three-dimensional way in the diagram 800 of Figure 8. For example, Figure 8 shows the time trends of two modified series corresponding to two wavelengths wl(i) (diagram 810) and wl(i+h) (diagram 820).

In a fifth step, the processing system 340 calculates (in absolute value) the cumulated and retrocumulated sums of the modified series corresponding to each wavelength wl(i) for prefixed temporal windows. For this purpose, for the calculation of the cumulated sums, a first time interval ITl-ITl ' (ITl < ITl ") is defined, and for the calculation of the retrocumulated sums a second time interval IT2-IT2' (IT2 < IT2', IT2 > ITl) is defined.

Examining the cumulated sums, for each wavelength wl(i) (i.e., for each selected column c(i) in the modified matrix 330t), the processing system 340, in correspondence of each time instant t(j), calculates the sum of the absolute values of the elements of the selected column c(i) starting from the element corresponding to the time ITl till the element corresponding to the time t(j) (execution order of the sum in accordance to the sign of the direction of the time axis t). Vice versa, the retrocumulated sum is performed by the processing system

340 for each wavelength wl(i) and in correspondence to each time instant t(j), by summing the absolute values of the elements of the selected column c(i) starting from the element corresponding to the time IT2' till the element corresponding to the time t(j) (execution order of the sum having an opposite sign with respect to the sign of the direction of the time axis t).

Figure 9 shows a first diagram 910 illustrating the growth of the cumulated sums of the modified series depicted in Figure 8 by means of the diagrams 810 and 820, and a second diagram 920 in which the grc^vth of the retrocumulated sums of the same modified series are illustrated. At this point, once all the cumulated and retrocumulated sums are calculated for each wavelength, the processing system 340 is capable of generating two different ordering tables 350, the first table including the wavelengths wl(i) of the spectrum classified according to a decreasing order of cumulated sum growth, the second table including the wavelengths wl(i) of the spectrum classified according to a decreasing order of retrocumulated sum growth. In this way, the process engineer is capable of recognizing instantaneously at which wavelengths wl(i) there correspond cumulated (or retrocuinulated) sums having the highest cumulated (or retrocumulated) growths. Being capable of obtaining said information in a fast and automatic way is of the utmost importance within the field of the optimal wavelength searching for detecting the EP of the etching process. Li fact, if the EP is located in one of the two temporal windows ITl- IT I' or IT2-IT2', the EP can be detected with a high probability by observing the time trend of the radiation intensity corresponding to one wavelength wl(i) among the wavelengths wl(i) having the highest cumulated (or retrocumulated) growths. As pointed out in a previous part of the present document, having information regarding the presence of non-linearity within the time trends of the radiation intensity corresponding to the various wavelengths is really useful. For this purpose, the processing system 340 is capable of accomplishing a further operation, adapted to estimate the non-linearity (particularly, by means of a convexity estimation) of the growth of the cumulated and retrocumulated sums previously calculated. Particularly, Figure 10 illustrates how the convexity estimation is calculated on the growths of the cumulated sums previously illustrated in the diagram 910. Each curve is interpolated by a single interpolating straight line 1005 passing through the origin of the system (in this case, corresponding to the time ITl; in the case of retrocumulated sums, corresponding to the time IT2'). At this point, the convexity of each curve can be calculated in two different ways: either taking into account the vertical distance between a plurality of points on the curve (measuring points) and the corresponding points on the interpolating straight line 1005 (Figure 10, square A), or taking into account the orthogonal distance between points of the interpolating straight line 1005 and points of the curve (Figure 10, square B).

Once all the convexity estimations have been calculated, the processing system 340 is capable of generating four further ordering tables 350, the first two tables including the wavelengths wli(i) of the spectrum classified according to a decreasing order of convexity (of the growth of the cumulated and retrocumulated sums, respectively) calculated exploiting the vertical distance, the second two tables including the wavelengths wl(i) of the spectrum classified according to a decreasing order of convexity (of the growth of the cumulated and retrocumulated sums, respectively) calculated exploiting the orthogonal distance.

Since the equation needed for interpolating the measuring points is not known a priori, the calculation of the convexity estimation may become very complex. Consequently, it is useful to compute the convexity using both methods, and then comparing the obtained results with those obtained by the growth of the cumulated and retrocumulated sums tables. If the results are confirmed by one or more of said tables, i.e., if some wavelengths wl(i) are classified in the same position in more than one table, it means that the procedure has given a good outcome, and the wavelength wl(i) placed in the head of most ordering tables 350 is almost certainly the optimal wavelength wl(i) for observing the EP of the etching process.

At this point, the process engineer is capable of deducing a good candidate for the EP of the etching process by observing the trend of the radiation intensity corresponding to the optimal wavelength wl(i). Once such EP has been deduced, it is tested on a number of wafers to be processed substantially lower (e.g., ten wafers) with respect to the case of the known methodologies. In fact, the method just described, in addition to be faster, guarantees a high reliability, because almost all the wavelengths wl(i) of the emission spectrum of the radiation in the reaction chamber are processed and analyzed, and not only one, manually chosen by the process engineer based on empirical considerations. In other words, the analysis performed by the method just described takes into account the correlations among the various wavelengths of the spectrum.

Turning now to Figure 11 , a graphical illustration by means of functional modules of a computer program 1100 is shown, the computer program 1100 being for example adapted to be executed by means of the computer 400 included into the processing system 340, capable of performing the procedures of the searching method of the optimal wavelengths previously described.

The computer program 1100 receives (module 1105) the data matrix 330, for example acquired by means of the spectrum acquisition system 300, and visualizes it by means of a tree-dimensional diagram on the computer display, for example, by means of a Graphical User Interface (GUI) 1110. Subsequently, (module 1115) the median matrix 330m is calculated from the data matrix 330; the median matrix 330m is then subjected to the smoothing operation (module 1120) by means of the MEWMA process, generating the smoothed matrix 3301. The smoothed matrix 3301 is then processed for generating the residual matrix 330r (module 1 125); the latter matrix is then modified (module 1130) for obtaining the modified matrix 330t. At this point, starting from the modified matrix 330t, the computer program 1100 computes the cumulated and retrocumulated sums (module 1135) and estimates the convexity thereof (module 1140). Finally, using both the results obtained by the computation of the cumulated and retrocumulated sums and by the convexity estimations, the computer program 1100 generates (module 1145) the ordering tables 350 classifying the wavelengths reputed to be significant for detecting the EP, and visualizes them on the computer display by means of the GUI. Moreover, by means of the GUI, the computer program 1100 allows to graphically visualize the "intermediate" matrixes (330m, 3301, 330r, 330t) which are generated during the various steps of the procedure previously described.

It has to be observed that the procedure previously described is based on the hypothesis of a process capable of exploiting a statistical model of the MEWMA type. If is not possible using a such procedure, the previous steps may be repeated by identifying the exact stochastic process within the multivariate family ARIMA(p,d,q) (wherein ARIMA states for AutoRegressive Integrated Moving Average)

The methodology previously described has been developed on a dry etching process for Shallow Trench Isolation (STI) and for Self- Aligned Gate (SAGate), and has been tested on apparatuses for dry etching on silicon manufactured by LAM (particularly, the model LAM TCP 9400DFM), which uses the Verity Spectraview OES software for detecting the EP. The results of such approach, obtained in relation with three different etching processes, are shown in the following table:

In particular, the algorithm provides the three optimal wavelengths wl(i) for each different etching process. Moreover, the last row of the table shows the real optimal wavelength wl(i) for each process obtained by means of the known methodologies. It has to be observed that such real optimal wavelength is always included within the best three wavelengths provided by the proposed algorithm in all the etching processes taken into account.

Concluding, with respect to the known methodologies, it is possible to improve the algorithms for detecting the EP₅ both in terms of time (spending hours, instead of days), both in terms of definition and quality, thanks to the better choice conditions.

The transfer from the development phase to the production phase will be thus supported by the consistence of the data on the EP and, consequently, by the estimation of the critical parameters of the variation which depend on the apparatus, the wafer, and so on.

Claims

1. A method for facilitating the determination of the end point of a dry plasma etching process of a material, comprising: performing an analysis of the whole spectrum of a radiation generated during the plasma etching process of the material, said analysis comprising evaluating the time trend of a plurality of spectral components of the radiation, each spectral component indicating the time trend of the radiation intensity of a corresponding wavelengths interval of the radiation; and on the basis of such analysis, selecting at least one of said spectral components, wherein said at least one spectral component has a time trend indicative of the evolution of the etching process of said material, characterized in that said performing the spectral analysis comprises performing a statistical analysis of the time trend of the whole spectrum of the radiation and, on the basis of the results of the statistical analysis, selecting the at least one spectral component.

2. The method of claim 1, wherein said performing the statistical analysis of the time trend of the whole spectrum of the radiation includes: individuating a plurality of spectrum sections, each section comprising a plurality of spectral components corresponding to adjacent intervals of wavelengths; and for each spectrum section, calculating a corresponding median component whose time trend is given by the median value of the radiation intensities of the spectral components forming the spectrum section.

3. The method of claim 2, wherein said performing the statistical analysis of the time trend of the whole spectrum of the radiation further includes varying the time trend of the intensity of each median component by means of the application of a Multivariate Exponentially Weighted Moving Average (MEWMA) statistical model.

4. The method of claim 3, wherein said performing the statistical analysis of the time trend of the whole spectrum of the radiation further includes calculating the 5 residuals of each varied median component, the value of each residual in the time being obtained by subtracting from the intensity of the corresponding varied median component at a certain time instant the intensity of the same median component at a previous time instant.

10 5. The method of claim 4, wherein said performing the statistical analysis of the time trend of the whole spectrum of the radiation further includes selecting the residuals whose values reach a predetermined threshold value.

6. The method of claim 5, wherein said performing the statistical analysis of

15 the time trend of the whole spectrum of the radiation further includes calculating, for each selected residual, cumulated and retrocumulated sums of the time trend of the residual value, said cumulated sum being a function of the time whose value is calculated starting from a preselected first time instant and gradually summing the residual values proceeding toward a second preselected time instant subsequent to

20 the first time instant, and said retrocumulated sum being a function of the time which value is calculated starting from a third time instant and gradually summing the residual values proceeding toward a fourth preselected time instant preceding the third time instant.

O 5 7. The method of claim 6, wherein said performing the statistical analysis of the time trend of the whole spectrum of the radiation further includes calculating a convexity degree of the time trend of the cumulated and retrocumulated sums.

8. The method of claim 7, wherein said selecting the at least one spectral 30 component includes evaluating the cumulated and retrocumulated sums and the corresponding convexity degree.

9. A data processing system which comprises means adapted to execute the method according to any one among the previous claims.

10. A computer program which comprises instructions for executing the method according to any one among the claims from 1 to 8 when said computer program is executed by means of a computer.