The content of the German Patent Application DE 10 2020 204 508.9 is in corporated by reference herein.

The invention relates to a system for inspecting a mask for EUV lithogra phy. The invention moreover relates to a method for inspecting a mask for EUV lithography.

Lithography methods, for example, are used to produce micro-structured or nano-structured components, for example memory chips. In this context, structures are imaged from a mask onto a wafer with the aid of a projection exposure apparatus. To ensure that the masks are suitable for the envisaged purpose, they are tested with the aid of an inspection system, in particular an APMI (actinic patterned mask inspection) system, prior to their use. In particular, provision can also be made of an inspection of the substrate pro vided to produce a mask, which is also referred to as mask blank. To this end, use can be made of an ABI (actinic blank inspection) system.

By way of example, an ABI system is known from US 2017/0235031 Al. There is always the need to improve a system and a method for inspecting a mask for EUV lithography.

These objects are achieved by means of the features of the independent claims.

A core of the invention consists of utilizing a multi-stage method or a sys tem with a plurality of partial systems for the purposes of inspecting a li- thography mask. In this context, potential mask defects are initially identi fied and/or localized with the aid of a first partial system. Then, the poten tial mask defects are pre-classified, in particular assessed using a confi dence parameter, with the aid of a second partial system. Then, a subset of the potential mask defects is checked on the basis of the pre-classification, in particular on the basis of the confidence parameter, of the same.

According to the invention, it was recognized to be advantageous to sort out from the potential mask defects ascertained in the first partial step those which allow a decision to be made with a given reliability that these are in fact mask defects, without requiring further checks. It may likewise be ad vantageous to sort out those potential mask defects (so-called false posi tives) which allow a decision to be made with a given reliability that these are not defects or are defects that are not relevant to the further use of the mask, without requiring further checks.

In the checking step carried out by the third partial system, it is then only necessary to check a subset of the potential mask defects ascertained in the first method step. This leads to a considerable time saving.

The system and the method are equally suited to inspecting a mask and to inspecting a mask blank. Below, the term mask is understood to mean both actually structured masks and unstructured masks, i.e. mask blanks.

EUV radiation is understood to mean electromagnetic radiation at a wave length ranging from 5 nm to 30 nm. In particular, this can relate to radia tion at a wavelength of 13.5 nm or 7 nm. According to one aspect of the invention, the second partial system for pre- classification of the potential mask defects utilizes an automated image analysis method, in particular an automated pattern recognition method. In particular, this can be a fully automatic method. In particular, this can be a non-algorithmic method.

The second partial system can also include algorithmic pre- or post-treat ment steps, in particular filtering steps and/or transformation steps, for ex ample one or more Fourier transforms. In particular, the pre-classification can be carried out on the basis of data in the spatial domain or in the fre quency domain.

According to a further aspect of the invention, the second partial system for pre-classification of the potential mask defects utilizes machine learning. In particular, the pre-classification can be carried out in software-based fash ion.

In particular, the second partial system can comprise a data processing de vice. In particular, the data processing device serves to process the data provided by the first partial system to identify and/or localize potential mask defects.

The second partial system can be data-connected to the first partial system. It can also be embodied as a constituent part of the first partial system. It can also be embodied as a constituent part of the third partial system, as a constituent part of a common control device for the partial systems or as a separate partial system. The second partial system can comprise a separate optical system. In par ticular, it can comprise an optical system for imaging the mask. In particu lar, this can be an actinic system.

In this context, an actinic system is understood to be a system which, for imaging and/or testing, uses illumination radiation at a wavelength corre sponding to the wavelength that is provided for the subsequent use of the mask in a projection exposure apparatus for structuring a wafer.

In particular, the second partial system can comprise an EUV system.

According to a further aspect of the invention, the second partial system and the third partial system can use the same optical system for checking the mask. In particular, the second partial system and the third partial sys tem can comprise a common optical system. This reduces the structural outlay for the overall system.

In particular, the second partial system and the third partial system can be formed by a common optical system, in particular an identical optical sys tem, which is used with different measurement modes.

The second partial system and the third partial system can also each com prise independent optical systems. This allows the second and third inspec tion steps to be carried out in parallel. This can increase the overall throughput of the system.

According to a further aspect of the invention, the second partial system has a memory for storing a database with pre-classified mask defects. In particular, the second partial system can have a memory with a database with pre-classified mask defects. The database can also be interchangeable. In particular, the database can be updated and/or upgradable.

In particular, it is possible to compare the potential mask defects identified and/or localized by the first partial system with pre-classified mask defects from the database. It was found that this facilitates a fast yet reliable classi fication of mask defects.

In particular, the database can be upgradable. This improves the predicta bility of the classification of the mask defects.

The database can also be interchangeable. This renders it possible to keep available specific databases for different mask types, for example for masks with different structure elements. This can further improve the relia bility of the pre-classification.

According to a further aspect of the invention, the second partial system is embodied in such a way that it divides the potential mask defects into at least two classes, in particular into two, three or more classes.

In this context, one class contains the mask defects not requiring further checks. A class separate therefrom contains the mask defects that should be checked by the third partial system.

In particular, the defects reliably identified incorrectly (so-called false posi tives) and the defects reliably identified correctly are classified as the first class with the mask defects not requiring further checks. The remaining defects, which do not allow a statement with sufficient reli ability to be made, are classified as the further class.

The classification of the potential defects in relation to the different classes can be determined on the basis of one or more parameters. These can be continuous or discrete parameters.

It was found that it is usually possible to significantly reduce the number of mask defects to be checked in the third inspection step by way of the pre- classification carried out with the aid of the second partial system.

The ratio of the number of mask defects to be checked with the aid of the third partial system or in the third inspection step to the overall number of mask defects identified and/or localized by means of the first partial system or in the first inspection step is in particular no more than 20%, in particu lar no more than 10%, in particular no more than 5%, in particular no more than 3%, in particular no more than 2%, in particular no more than 1%.

According to a further aspect of the invention, the classification speed V2 of the second partial system is greater than the checking speed V ₃ of the third partial system, V2 > V3.

The classification speed V2 of the second partial system is in particular at least 500 classified defects per hour, in particular at least 750 classified de fects per hour, in particular at least 1000 classified defects per hour, in par ticular at least 1250 classified defects per hour, in particular at least 1500 classified defects per hour, in particular at least 2000 classified defects per hour. In particular, the following applies: V2/V3 > 1.1, in particular V2/V3 > 1.2, in particular V2/V3 > 1.5, in particular V2/V3 > 2, in particular V2/V3 > 2, in par ticular V2/V3 > 3, in particular V2/V3 > 5, in particular V2/V3 > 10. In particu lar, the following applies: V2/V3 > 1/(1 -a), where a specifies the expected ra tio of the number of potential mask defects to be checked by the third par tial system to the number of potential mask defects to be classified by the second partial system. The number of potential mask defects to be classi fied by the second partial system in this context just equals the number of potential mask defects identified and/or localized by the first partial sys tem. The ratio a can be determined experimentally on the basis of masks to be checked.

According to a further aspect of the invention, the first partial system for inspection of the mask uses illumination radiation at a wavelength of longer than 30 nm, in particular longer than 100 nm. In particular, the first partial system can comprise a non-actinic optical system. In particular, it can comprise a DUV system. In this respect, exemplary reference is made to US 8,103,086 B2.

In particular, the first partial system uses illumination radiation at a wave length of 193 nm or longer for the inspection of the mask.

According to a further aspect of the invention, the second partial system for pre-classification of the potential mask defects and/or the third partial sys tem for checking a subset of the potential mask defects comprises an opti cal system with illumination radiation at a wavelength in the EUV range.

In particular, the second partial system and/or the third partial system can comprise an actinic optical system. In particular, the second partial system and/or the third partial system can comprise an actinic aerial image system, i.e. an optical system with illumi nation radiation at a wavelength in the EUV range for producing an aerial image of the mask. In particular, this can be a scanning system. In respect of an actinic aerial image system, exemplary reference is made to DE 10 2010 029 049 Al.

According to a further aspect of the invention, the system comprises a memory unit which is data-connected to the first partial system. In particu lar, the memory unit serves to store a data record with data relating to po tential mask defects.

In particular, the memory unit can be data-connected or able to be data- connected to the second partial system.

The data in relation to potential mask defects ascertained by the first partial system can be stored, in particular, on a physical storage medium or in a virtual memory. In this context, a virtual memory can simplify the data transmission from the first partial system to the second partial system.

The data of the pre-classification of the potential mask defects ascertained by the second partial system can be stored on a physical storage medium or in a virtual memory. In this context, storage in a virtual memory simplifies the data transmission from the second partial system to the third partial sys tem.

According to a further aspect of the invention, the second inspection step comprises an automated image analysis method for pre-classifying the po tential mask defects. The image analysis method can comprise pre-processing steps, for example filter steps or transformation steps, in particular one or more Fourier trans forms.

The image analysis can be carried out in the spatial domain or in the fre quency domain.

According to one aspect of the invention, the second inspection step for pre-classifying the potential mask defects comprises a method based on machine learning. In particular, this can be a method based on supervised learning. This can also relate to reinforcement learning. In this context, the results from the third inspection step can be utilized for further training of the second inspection step. In this context, different types of masks, in par ticular, can be treated separately. This can lead to an improvement in the reliability of the pre-classification.

According to a further aspect of the invention, a distribution of one-dimen sional and/or two-dimensional structures in the image of the mask recorded in the first inspection step is analysed for the purposes of pre-classifying the potential mask defects.

These structures can be analysed either in the spatial domain or in the fre quency domain.

In particular, the one-dimensional structures can be critical dimensions (CDs). In particular, the two-dimensional structures can be contours of the images of the mask recorded in the first inspection step or a Fourier trans form of the same. It is also possible to analyse properties of these struc tures, for example the diameter thereof. According to a further aspect of the invention, an intensity distribution of an image of the mask recorded in the first inspection step is analysed for the purposes of pre-classifying the potential mask defects. In particular, the image of the mask can be analysed pixel-by-pixel. Upsampling or downsampling is possible.

One or more of the partial systems 2, 3, 4 can be imaging systems, which each record an overall image of the mask or of the mask blank in one expo- sure, or scanning systems, in which the images of the mask or of the mask blank are assembled from a plurality of exposures.

According to a further aspect of the invention, the second inspection step comprises one or more pre-processing steps, in particular one or more cor- rection steps for correcting one or more images recorded in the first inspec tion step in respect of photon noise.

According to a further aspect of the invention, the second inspection step comprises a comparison step for comparing one or more images of the mask recorded in the first inspection step with data from a database.

According to a further aspect of the invention, the third inspection step comprises an actinic method, in particular a method for ascertaining and/or analysing an aerial image, in particular using illumination radiation in the EUV range.

According to a further aspect of the invention, the first inspection step comprises a non-actinic method. In particular, the first inspection step can comprise a method for ascertaining an image of the mask using illumina tion radiation in the DUV wavelength range, in particular using illumina tion radiation at a wavelength of 193 nm.

In particular, the method according to the invention can comprise at least two optical inspection steps, in which images of the mask are ascertained and/or analysed using illumination radiation at different wavelengths. It was found that this firstly reduces the time required to inspect the mask and secondly can improve the reliability of the characterization of the mask de fects.

Further advantages, details and particulars of the invention are evident from the description of exemplary embodiments with reference to the fig ures. In detail:

Fig. 1 schematically shows a course of a method for inspecting a mask for EUV lithography and

Fig. 2 schematically shows constituent parts of a system for inspecting a mask for EUV lithography.

Fig. 2 schematically illustrates a system 1 for inspecting a mask for EUV lithography. The illustration should be understood to be exemplary.

The system 1 comprises a first partial system 2, a second partial system 3 and a third partial system 4. Preferably, so-called multilayer defects of the mask, in particular of the mask blank, are ascertained separately with the aid of a separate system, in particular an actinic blank inspection tool (ABI tool).

The first partial system 2 serves for optical inspection of the mask. In par ticular, it serves to identify and/or localize potential mask defects.

The first partial system 2 comprises, in particular, an inspection system for inspecting the mask using illumination radiation in the DUV range, in par ticular using illumination radiation at a wavelength of 193 nm. For details of such a system, reference should be made in exemplary and representa tive fashion to US 8, 103,086 B2, which is hereby incorporated in the pre sent application as part thereof.

The second partial system 3 serves, in particular, to pre-classify the poten tial mask defects identified and/or localized by means of the first partial system 2.

The second partial system 3 can comprise an optical system with a wave length in the EUV range.

In particular, the third partial system 4 serves to check the potential mask defects.

The third partial system 4 comprises, in particular, an optical system for checking the mask using a wavelength in the EUV range. For details of such a system, reference should be made in exemplary and representative fashion to DE 102010 029049 Al, which is hereby fully incorporated in the present application as part thereof. As is illustrated schematically in Figure 2, the first partial system 2 is data- connected to a first memory unit 5. The first memory unit 5, in turn, is data-connected to the second partial system 3.

The first memory unit 5 can be embodied as a separate memory unit, in particular as a separate storage medium, or as a virtual memory. It can also be embodied as a constituent part of the first partial system 2 or as a con stituent part of the second partial system 3.

The first memory unit 5 serves to store the data ascertained by the first par tial system 2 to identify and/or localize potential mask defects. The corre sponding data serve as input for the further inspection of the mask with the aid of the second partial system 3.

The second partial system 3 is data-connected to a second memory unit 6.

The second memory unit 6 can be embodied as a separate memory unit, in particular as a separate storage medium, or as a virtual memory. It can also be embodied as a constituent part of the second partial system 3 or as a constituent part of the third partial system 4.

The second partial system 3 can comprise an optical system for examining the mask, in particular for generating and analysing an aerial image of the mask. In particular, this can be an EUV system.

The third partial system 4 comprises an optical system for checking and an alysing potential mask defects. In particular, the optical system of the third partial system 4 is an actinic system. The second partial system 3 comprises a memory 7 for storing a database with pre-classified mask defects. The memory 7 can be embodied sepa rately from the second partial system 3.

Below, the basic course of the method for inspecting the mask is described with reference to Figure 1.

The method is a multi-stage method. In particular, the method comprises a first inspection step 8, a second inspection step 9 and a third inspection step 10.

The inspection steps 8, 9, 10 can in turn comprise one or more partial steps. The first inspection step 8 serves, in particular, to record an image of the provided mask and to identify and/or localize potential mask defects in the provided mask.

The second inspection step 9 serves, in particular, to pre-classify the poten- tial mask defects.

The third inspection step 10 serves, in particular, to check a subset of the potential mask defects. In the first inspection step 8, a structured lithography mask, in particular, is analysed with the aid of an optical system. In the first inspection step 8, a list of potential mask defects is created, in particular with the aid of a DUV system. The list of potential mask defects can be stored in the first memory unit 5. In the second inspection step 9, the mask is examined using an actinic opti cal system, in particular an EUV system. In this context, the potential de fects as per the list created in the first inspection step 8, in particular, are pre-classified.

In particular, provision is made for the potential defects ascertained in the first inspection step 8 to be divided into at least two classes, wherein the one class only contains potential defects requiring no further checks while the other class contains potential defects for which a final assessment as to whether this is in fact a defect, in particular a defect relevant to the envis aged use of the mask, cannot be made with sufficient reliability.

Classified as the first class are, in particular, potential defects that have a sufficient probability of not being actual defects (so-called false positives), in particular with a confidence of at least 95%, in particular at least 97%, in particular at least 99%. Moreover, classified as the first class are potential defects which are true defects, in particular defects relevant to the envis aged use of the mask, with a confidence of at least 95%, in particular at least 97%, in particular at least 99%.

The remaining potential defects, for which a final statement with a confi dence of at least 95% is not possible, are classified as the second class. The second inspection step 9 comprises, in particular, a fast disposition of the potential mask defects. The classification speed is at least 500 classified defects per hour, in particular at least 750 classified defects per hour, in particular at least 1000 classified defects per hour, in particular at least 1250 classified defects per hour, in particular at least 1500 classified de fects per hour, in particular at least 2000 classified defects per hour.

In particular, a method based on machine learning is used to pre-classify the potential mask defects in the second inspection step 9.

Different details and aspects of the method step for pre-classifying the po tential mask defects are described below.

The first partial system can have a fast disposition mode. In this mode of operation, the images in first inspection steps can be recorded particularly quickly. In this context, it is possible to dispense with the accuracy, in par ticular resolution, required for the mode used for the accurate, optical anal ysis of potential defects. Being able to delimit the images of the potential defects from photon noise with sufficient reliability may suffice.

The second inspection step 9 can comprise pattern recognition. The pattern recognition can comprise a threshold method (threshold setting) and an im age contour analysis.

One-dimensional structures and/or two-dimensional and/or logical struc tures can be analysed during the image contour analysis. In this context, logical structures are understood to be non-symmetrical or non-periodic structures.

In particular, provision can be made to set an automated threshold in order to measure a critical dimension (CD) for each structure element in the overall image of the mask. It was found that the analysis of the distribution of the values of the critical dimensions have clearly defined peaks in the frequency domain. Deviations from these peaks can indicate potential de fects. It was found that the power spectral density (PSD) of the critical di mensions in the aerial image has a continuum and peaks, wherein the prop erties of the peaks can be influenced by defects.

Further, provision can be made for properties of these peaks to be analysed, for example the full width at half maximum (FWHM) thereof. In this con text, photon noise can be taken into account.

Training of the system for machine learning can be provided in the spatial domain or in the frequency domain.

Further, provision can be made for a correction method for correcting pho ton noise to be carried out prior to the image analysis in the second inspec tion step 9.

According to one variant, provision can be made for variations of the im age contours or of a frequency distribution of measured structure parame ters of the possibly pre-processed image, in particular noise-corrected im age, to be analysed and to be compared to data in a database of defect-free structures and defects.

In particular, the second inspection step 9 can comprise a comparison step for comparing potential defects, in particular certain parameters of poten tial defects, to the corresponding data in a database.

The pre-classification can be implemented on the basis of a distribution of ascertained critical dimensions or a distribution of certain parameters of contours in the spatial domain or in the frequency domain. By way of ex ample, low-intensity peaks in the power spectral density (PSD) may indi cate defects. A pre-classification on this basis is possible after appropriate preceding training.

As an alternative to a pre-classification on the basis of critical dimensions or contours, provision can also be made for the intensity distribution or the frequency of changes of this distribution to be evaluated. In particular, the evaluation can be carried out pixel-by-pixel. Upsampling or downsampling is also possible. Once again, provision can be made in this context for a correction for taking account of photon noise. The distribution analysed in this way can be compared to data in a database which was used to train the system for a machine learning method. In this respect, it is noted that, in particular on account of the contours, defects have different frequencies to the usual mask structures, in particular different frequencies to the frequen cies of regular mask structures, in particular periodic mask structures.

Further processing steps, for example smoothing methods (image smooth ing) or low pass filtering steps, can be provided in all methods, in particular in order to filter out noise.

According to a further alternative, provision can be made for the images of the mask ascertained in the second inspection step 9, in particular images of the mask defects, to be compared directly to images in a database.

The database can be supplemented on an ongoing basis in all of the meth ods. As a result, it is possible to continually improve the pre-classification in the second inspection step 9, in particular render this more reliable. This leads as a result to the ratio of the potential mask defects to be checked in the third inspection step 10 to the number of potential mask defects ascer tained overall in the first inspection step 8 decreasing on average.

It was found that, with the aid of the pre-classification in the second in- spection step 9, it is possible to reduce the number of potential mask de fects to be checked in the third inspection step 10 from several thousand potential mask defects ascertained in the first inspection step 8 to less than 200, in particular less than 100. An actinic method, in particular an actinic aerial image method, is provided to check the potential mask defects in the third inspection step 10.

The overall result of the inspection method, in particular of the three in spection steps 8, 9 and 10, is a statement regarding the relevance of the mask defects for the envisaged uses of the mask. In particular, whether or not the mask satisfies certain specified quality criteria can be decided on the basis of this overall result.