Methods for providing a prediction model, for ascertaining a temporal profile of a wavefront aberration and for operating a projection system
A neural network-based prediction model for wavefront aberrations in projection systems addresses inefficiencies in existing methods by providing continuous compensation, ensuring high imaging quality without operational interruptions.
Patent Information
- Application Number
- PCT/EP2026/062753
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-05
- Filing Date
- 2026-05-04
- Publication Date
- 2026-11-12
AI Technical Summary
Existing methods for correcting wavefront aberrations in projection systems of projection exposure apparatuses are inefficient due to the time-consuming nature of wavefront measurements and inaccuracies in simulation-based predictions, leading to interruptions in productive operation and suboptimal compensation of thermal deformations.
A computer-implemented method using a neural network-based prediction model to estimate a temporal profile of wavefront aberrations by training on datasets that combine thermal influencing variables with wavefront aberrations, allowing continuous compensation of thermal deformations through actuators.
Enables precise and continuous correction of wavefront aberrations without interrupting productive operation, improving imaging quality by accurately predicting and compensating for thermal deformations over extended periods.
Smart Images

Figure EP2026062753_12112026_PF_FP_ABST
Abstract
Description
04.05.2026 / BSMethods for providing a prediction model, for ascertaining a temporal profile of a wavefront aberration and for operating a projection system
[0001] This patent application claims the priority of German patent application DE 10 2025 116 980.2, filed on May 5, 2025, which is referred to and the content of which is incorporated herein in its entirety. The invention relates to a computer-implemented method for providing a prediction model designed to ascertain a temporal profile of a wavefront aberration of an optical system, wherein the optical system has at least one optical element. The invention furthermore relates to a method for ascertaining a temporal profile of a wavefront aberration of an optical system, in particular of a proj ection system of a proj ection exposure apparatus. Finally, the invention relates to a method for operating a proj ection system.
[0002] Microlithographic proj ection exposure apparatuses comprising an illumination system and a proj ection optical unit are used to produce microstructured components. The illumination system may be used to illuminate a reticle located in an obj ect plane of the proj ection optical unit. The projection optical unit serves to image a structure located on the reticle onto a wafer located in an image plane of the proj ection optical unit. In order to achieve a high imaging quality, the individual optical elements of the proj ection optical unit must be manufactured with very high accuracy and very low tolerances. Moreover, the temperature of the optical elements is regularly actively controlled in order to reduce changes in the imaging quality of the proj ection optical unit caused by thermal expansions of the optical elements. By way of example, channels for a thermal fluid may run through the optical elements. By virtue of a thermal fluid with a predefined volumetric flow and temperature being guided through these channels, the temperature of the optical elements is already able to bekept fairly constant. Although these measures make it possible to reduce deformations of the optical elements during ongoing operation, these cannot be completely prevented in many cases. In particular, in the case of both diffractive optical elements and reflective optical elements, a proportion of the imaging light that is used is converted into heat that, in many cases, cannot be dissipated quickly enough, potentially resulting in local heating of the optical elements. Other heat sources, for example electrical actuators, may also lead to undesired heating of optical elements.
[0003] In order to be able to correct thermal deformations caused thereby, the optical elements are usually mounted on a fixed supporting frame with the aid of actuators. The actuators may be used to readjust the pose of the individual optical elements in one or more of the six rigid body degrees of freedom relative to the supporting frame, such that it is possible to compensate at least partially for the thermal deformations. It is also possible to provide actuators for the purpose of deforming an optical surface, as is known for example from DE 10 2021 205 425 Al, in order thereby to counteract temperature-induced changes in shape.
[0004] In order for the abovementioned corrective measures to be able to be carried out with targeted accuracy during exposure operation, it is necessary to ascertain the present state of a wavefront of the imaging beam path generated by the projection system as accurately as possible. It is known in principle to record the state of the wavefront with the aid of a wavefront measurement. Wavefront aberrations identified by the measurement may then be compensated for by the above-described corrective measures, such that the proj ection system is able to operate with high accuracy in subsequent productive operation. By way of example, the wavefront measurement may be carried out at regular time intervals, for instance after eachexposure process in which the structure of a mask has been transferred to a wafer, or after a certain number of such exposure processes.
[0005] One drawback of this procedure is that carrying out the wavefront measurement takes a certain amount of time, which is not available for productive operation of the proj ection exposure apparatus. A further drawback is that, in this procedure, the wavefront aberration is detected only on a one-off basis. By contrast, wavefront aberrations that occur in the period between two wavefront measurements, in particular while carrying out an exposure process, cannot be ascertained and compensated for, or are able to be ascertained and compensated for only insufficiently.
[0006] A further possibility is to simulate the temperature-induced change in the wavefront over time proceeding from a known thermal initial state. For this purpose, it is possible to set up a simulation model into which the heat flow generated by the exposure radiation and by other heat sources may be input. The simulation model may for example comprise a finite element model. One problem with this procedure is that the prediction accuracy decreases over time due to inevitable imperfections of the simulation model. The wavefront aberrations predicted by the model are thus subj ect to increasing uncertainty over time, and compensation is no longer able to be carried out with the required accuracy. After a certain operating time, it is therefore necessary to return to a known thermal state. This usually requires an operating pause for the system during which the system is able to cool down or assume a predefined thermal state, meaning that productive operation has to be interrupted for a relatively long time.
[0007] The obj ect of the present invention is to provide a computer-implemented method for generating a prediction modelfor ascertaining a temporal profile of a wavefront aberration and a method for ascertaining a temporal profile of a wavefront aberration, in which method the abovementioned problems are avoided or at least reduced. This obj ect is achieved by the subj ect matter of the independent claims. Advantageous embodiments are described in the dependent claims.
[0008] Accordingly, the invention relates to a computer-implemented method for generating a prediction model designed to ascertain a temporal profile of a wavefront aberration of an optical system. The optical system has at least one optical element and may be designed in particular as a proj ection system of a proj ection exposure apparatus. The method comprises the following steps:- creating a wavefront model comprising a latent vector representing the wavefront aberration and a model function, wherein the model function is dependent on a thermal influencing variable acting on the at least one optical element and dependent on the latent vector, wherein a derivative of the latent vector with respect to time is described by the model function;- providing a plurality of training datasets, wherein a temporal profile of the thermal influencing variable is assigned to a temporal profile of the wavefront aberration of the optical system in each of the training datasets;- training the wavefront model with the aid of a neural network using the plurality of training datasets.
[0009] The optical elements may comprise both diffractive optical elements and reflective optical elements.
[0010] The wavefront aberration is usually observed in the image plane of the optical system, and may be described for example by virtue of the wavefront being represented by Zernike polynomials. In this case, the wavefront aberration may be described in particular with the aid of a set of Zernike coefficients (cf. EP 3 674 798 Al ). However, it is also possible to represent the wavefront in another way, in particular using another suitable base, and to describe it with the aid of associated coefficients. The base may in particular be orthogonal (that is to say formed by linearly independent base vectors). The base may also be based on polar coordinates.
[0011] The prediction model may be designed to output the temporal profile of the wavefront aberration on the basis of a temporal profile, input into the prediction model, of the thermal influencing variable and possibly additionally on the basis of an initial value of the wavefront aberration, which may be recorded for example by a measurement at a starting time.
[0012] The expression "temporal profile" (of the thermal influencing variable or the wavefront aberration) should be understood in the broad sense. In principle, the temporal profile may already be formed by two values at two different times. However, the temporal profile usually comprises a plurality of values of the respective variable at different times.
[0013] The temporal profile of the wavefront aberration may be described by a time-dependent wavefront vector z(t) the entries of which may be the respective parameters of the representation that is used (that is to say in particular the Zernike coefficients). The latent vector representing the wavefront aberration may be described by an assignment h(t) = Eθ1(z(t)), wherein Eg may be what is referred to as an encoder functiondefined by free parameters 01 (01 may in particular be a set of free parameters).
[0014] In one embodiment, the encoder function Eθ1is the identity function. In this case, the time-dependent wavefront vector z(t) corresponds to the latent vector h(t). As an alternative, the encoder function Eθ1may be designed, in a manner known in principle, to map the wavefront vector z(t) onto a latent vector h(t), wherein (t) in this case is an element of a latent space that usually has a lower dimensionality than the space of the wavefront vector z(t). In this case, the free parameters 01 of the encoder function Eg may be ascertained, in a manner known in principle, by machine learning with the aid of a neural network. A corresponding decoder function Dθ2may be ascertained for the encoder function Eθ1, using which decoder function the latent vector (t) is able to be transformed back into a wavefront vector z(t) as follows: z(t) = Dθ2(h(t)), wherein the free parameters θ2 of the decoder function Dθ2may likewise be ascertained, in a manner known in principle, by machine learning with the aid of a neural network (02 may in particular be a set of free parameters).
[0015] The derivative of the latent vector with respect to time may in particular be a first derivative with respect to time. Accordingly, the wavefront model may be based on a differential equation of the following form:wherein the expressionin the above representation is the model function dependent on the latent vector and the thermal influencing variable, wherein 03 denotes free parameters or a set of free parameters. The model function or itsproperties may be ascertained by the training method step. In particular, the method step of training the wavefront model may be used to ascertain the free parameters 03 of the model function.
[0016] After the wavefront model has been trained, a temporal profile of the thermal influencing variable and possibly an initial value of a wavefront aberration may be supplied thereto in order to obtain, as output variable, a temporal profile of the wavefront aberration that results from the temporal profile of the thermal influencing variable (and possibly from the initial value).
[0017] The invention is based on the finding that the dynamics of the latent vector — and thus also the dynamics of the wavefront aberration or of the wavefront vector — are able to be described by an ordinary differential equation that is able to be ascertained by the neural network. The use of ordinary differential equations, that is to say equations in which only derivatives with respect to a variable occur, is generally known for describing systems in nature, technology, biology and society. By way of example, " Newtonian mechanics", "the principles of psychophysics", the " RL circuit", or the "logistic function" may be described with the aid of ordinary differential equations. The above application cases have the common feature that a rate of change of a variable under consideration has a simple functional relationship with the variable under consideration itself. By contrast, the fact that the rate of change over time of the latent vector representing the wavefront vector is able to be described by a function (specifically the model function) of the latent vector and of the temporal profile of the thermal influencing variable is neither intuitively easy to understand nor predictable. This is because a wavefront aberration, in a system consisting of a plurality of optical elements, is predefined in a non-trivialmanner by all of the optical elements, with there being a complicated and multilayered relationship that is influenced significantly by thermal conduction within the optical elements and thermal expansion of the optical elements. Moreover, the thermal influencing variable itself may also assume a very complex form and be defined by the predefined illumination setting (predefined by the illumination system) and the reticle that is used. Against this background, it is surprising that it is possible to find a model equation that satisfies the ordinary differential equation of the wavefront model, and that a precise and reliable prediction of the wavefront aberration is possible over long periods with the aid of the trained wavefront model.
[0018] In one embodiment, at least some of the plurality of training datasets are generated by a simulation. Corresponding methods for ascertaining a wavefront aberration on the basis of a thermal influencing variable are known in principle (cf. WO 2020 / 182386 Al, WO 2022 / 167164 Al ). The simulation may in particular comprise an FEM simulation. It has been found that, by training the wavefront model with simulation-based training data, it is possible to obtain a wavefront model that is able to predict the wavefront aberration with high accuracy over long periods proceeding from a defined initial state. It has been found that the trained wavefront model is able to deliver a highly precise prediction over a long period even for combinations of an initial value of the wavefront aberration and a heat flow vector that were not present in the plurality of training datasets. It is thus possible to correctly describe unknown illumination settings and reticles even at the time when the model is created.
[0019] In one embodiment, at least some of the plurality of training datasets are generated by a measurement by virtue of the optical system being exposed to a defined temporal profileof the thermal influencing variable and a resulting temporal profile of the wavefront aberration of the optical system being measured. The wavefront model is able to be further improved through the generation of such training data. In particular, imperfections in the training data generated by a simulation that are based on an imperfection of the simulation are able to be compensated for by the generation of measured value-based training data. Moreover, temporal changes in the optical system that are not able to be captured, or not able to be captured completely, by a simulation, for example a degradation of the optical elements that takes place over time, are also able to be taken into account in the prediction model through the use of measured value-based training data.
[0020] The thermal influencing variable may comprise or be formed from a heat flow vector acting on the at least one optical element. The heat flow vector may be described for example by a time-dependent vector Q(t) the entries of which may define a heat flow incident on the at least one optical element. If the optical system has a plurality of optical elements, the heat flow vector Q(t) may have a corresponding number of entries for each of the optical elements. By way of example, entries 1-1000 of the heat flow vector may be assigned to a first of the optical elements and entries 1001-2000 may be assigned to a second of the optical elements, with this being able to be continued correspondingly for any further optical elements of the optical system that may be present. The thermal influencing variable may be defined by way of a vector description that relates to a suitable heat flow vector base. The heat flow vector base may be based on polar coordinates. By way of example, different entries of the vector may describe heat flows incident on the respective optical element in different directions and / or at different locations.
[0021] In one embodiment, the optical system is a proj ection system of a proj ection exposure apparatus. The proj ection system may have an obj ect field able to be illuminated by an illumination system of the proj ection exposure apparatus by way of a settable illumination setting. Furthermore, a reticle may be able to be positioned in the obj ect field, wherein the projection system may be designed to image at least a subregion of the reticle into an image plane. The thermal influencing variable may be determined taking into account the settable illumination setting and / or the selected reticle.
[0022] In one embodiment, it is additionally possible for the thermal influencing variable to be determined by ( further) taking into account a heat flow acting on the at least one optical element and generated by one or more active heating elements. Active heating elements may for example comprise preheaters and / or sector heaters.
[0023] The invention furthermore relates to a computer-implemented method for ascertaining a temporal profile of a wavefront aberration of an optical system having at least one optical element, comprising the following steps:- providing a trained prediction model (50), in particular with the aid of a method according to one of Claims 1 to 7, wherein the prediction model is based on a wavefront model (51 ), wherein the wavefront model comprises a latent vector representing the wavefront aberration and a model function, wherein the model function is dependent on a thermal influencing variable (56) acting on the at least one optical element (Ml - M6) and dependent on the latent vector, wherein a derivative of the latent vector with respect to time is described by the model function;inputting an initial state of the wavefront aberration into the prediction model;- inputting a temporal profile of the thermal influencing variable into the prediction model in order to obtain a temporal profile of the wavefront aberration on the basis of the initial state and the temporal profile of the thermal influencing variable.
[0024] The disclosure comprises further embodiments of the computer-implemented method for ascertaining a temporal profile of a wavefront aberration that are developed by further features that have already been described above in connection with the computer-implemented method for providing a prediction model.
[0025] In principle, it is possible to estimate the initial state of the wavefront aberration based on ambient conditions prevailing at the starting time under consideration. By way of example, the ambient conditions may comprise a temperature state and / or a previously known alignment state of the optical system. In one embodiment, the initial state of the wavefront aberration is ascertained with the aid of a wavefront measurement. The measured value thereby ascertained of the wavefront aberration may then be input into the prediction model, such that the latter is able to ascertain the temporal profile of the wavefront aberration proceeding from this initial state.
[0026] In one embodiment, ascertaining the temporal profile of the wavefront aberration serves the purpose of generating at least one control command for one or more actuators acting on the at least one optical element, wherein the control command is suitable for at least partially compensating for the wavefront aberration. In particular, a plurality of actuators assigned to different optical elements of the optical system maybe driven, by control commands, in such a way that the wavefront aberration is reduced by an overall optical effect generated jointly by the actuators. If the wavefront aberration or the temporal profile thereof is known, the actuators of the optical elements may be driven in a manner known in principle, which does not need to be explained in detail above. Within the scope of the methods described above, provision may be made for the at least one control command to be generated.Furthermore, provision may be made for the at least one control command to be sent to the actuators.
[0027] The invention furthermore relates to a method for operating a proj ection system of a proj ection exposure apparatus. The proj ection system comprises an obj ect field able to be illuminated by an illumination system of the proj ection exposure apparatus, optical elements for imaging the obj ect field into an image plane, at least one actuator for effecting a change in position and / or shape of at least one of the optical elements, and a control unit for driving the at least one actuator. The method comprises the following steps:- ascertaining a temporal profile of a wavefront aberration of the proj ection system with the aid of a method according to the invention;- generating, on the basis of the ascertained temporal profile of the wavefront aberration, at least one control command for one or more actuators acting on the at least one optical element, wherein the at least one control command is suitable for counteracting a wavefront aberration at at least one time of the temporal profile;operating the proj ection system proceeding from the initial state in such a way that the optical elements areexposed to the temporal profile of the thermal influencing variable of the wavefront aberration;- driving the at least one actuator during the operation of the proj ection system by way of the at least one control command.
[0028] The disclosure comprises further embodiments of the method for operating a proj ection system of a proj ection exposure apparatus that may be developed by further features that have already been described above in connection with the computer-implemented method for providing a prediction model or in connection with the computer-implemented method for ascertaining a temporal profile of a wavefront aberration.
[0029] The at least one control command may in particular be suitable for counteracting a wavefront aberration at a plurality of times of the temporal profile. In one embodiment, generating at least one control command comprises generating a control command temporal profile suitable for keeping a wavefront aberration of the proj ection system below a maximum tolerance value of the wavefront aberration during operation.
[0030] The invention will be described below by way of example on the basis of one advantageous embodiment with reference to the accompanying drawings, in which:Fig. 1 shows a schematic illustration of a photolithographic proj ection exposure apparatus;Fig. 2 shows a schematic illustration of a method for providing a prediction model and of a method for ascertaining a temporal profile of a wavefront aberration.
[0031] Figure 1 illustrates a schematic meridional section of a semiconductor lithography proj ection exposure apparatus 1. In this case, the proj ection exposure apparatus 1 comprises an illumination system 10 and a proj ection system 20. An obj ect field 11 in an obj ect plane or reticle plane 12 is illuminated with the aid of the illumination system 10. For this purpose, the illumination system 10 comprises an exposure radiation source 13, which, in the illustrated exemplary embodiment, emits illumination radiation at least comprising used light in the EUV range, that is to say with a wavelength of between 5 nm and 30 nm in particular. The exposure radiation source 13 may be a plasma source, for example an LPP (laser produced plasma) source or a GDPP (gas discharge produced plasma) source. It may also be a synchrotron-based radiation source. The exposure radiation source 13 may also be a free electron laser (FEL).
[0032] The illumination radiation emerging from the exposure radiation source 13 is initially focused in a collector 14. The collector 14 may be a collector with one or more ellipsoidal and / or hyperboloidal reflection surfaces. The illumination radiation may be incident on the at least one reflection surface of the collector 14 with grazing incidence (GI ), that is to say at angles of incidence of greater than 45°, or with normal incidence (NI ), that is to say at angles of incidence of less than 45°. The collector 14 may be structured and / or coated for the purpose of optimizing its reflectivity for the used radiation, on the one hand, and for the purpose of suppressing extraneous light, on the other hand.
[0033] Downstream of the collector 14, the illumination radiation propagates through an intermediate focus in an intermediate focal plane 15. If the illumination system 10 is constructed with a modular design, the intermediate focal plane 15 may be used, in principle, for the separation - includingthe structural separation - of the illumination system 10 into a radiation source module, comprising the exposure radiation source 13 and the collector 14, and the illumination optical unit 16 described below. In the case of a corresponding separation, the radiation source module and the illumination optical unit 16 then jointly form a modularly constructed illumination system 10.
[0034] The illumination optical unit 16 comprises a deflection mirror 17. The deflection mirror 17 may be a planar deflection mirror or alternatively a mirror with a beam-influencing effect going beyond the pure deflection effect. As an alternative or in addition, the deflection mirror 17 may be embodied as a spectral filter that separates a used light wavelength of the illumination radiation from extraneous light having a wavelength that deviates therefrom.
[0035] The deflection mirror 17 is used to deflect the radiation emanating from the exposure radiation source 13 to a first facet mirror 18. If - as in the present case - the first facet mirror 18 is arranged in a plane of the illumination optical unit 16 that is optically conjugate to the reticle plane 12 as a field plane, this facet mirror is also referred to as a field facet mirror.
[0036] The first facet mirror 18 comprises a multiplicity of micromirrors 18 ' able to be pivoted individually about two mutually perpendicular axes in each case, for the purpose of controllably forming facets that are each configured with an orientation sensor (not illustrated) for ascertaining the orientation of the micromirror 18 '. The first facet mirror 18 is thus a microelectromechanical system (MEMS system), as also described in DE 10 2008 009 600 Al, for example.
[0037] A second facet mirror 19 is arranged downstream of the first facet mirror 18 in the beam path of the illumination optical unit 16, thus resulting in a doubly faceted system, the fundamental principle of which is also referred to as a fly' s eye integrator. If the second facet mirror 19 - as in the illustrated exemplary embodiment - is arranged in a pupil plane of the illumination optical unit 16, it is also referred to as a pupil facet mirror. However, the second facet mirror 19 may also be arranged at a distance from a pupil plane of the illumination optical unit 16, as a result of which a specular reflector arises from the combination of the first and the second facet mirror 18, 19, as described in US 2006 / 0132747 Al, EP 1 614 008 Bl and US 6, 573, 978, for example.
[0038] The second facet mirror 19 need not in principle be constructed from pivotable micromirrors, but rather may comprise individual facets formed from one mirror or a manageable number of mirrors significantly larger than micromirrors, which facets are either stationary or are able to be tilted only between two defined end positions. It is however - as illustrated - also possible, in the second facet mirror 19, to provide a microelectromechanical system having a multiplicity of micromirrors 19 ' able to be pivoted individually about two mutually perpendicular axes in each case, each preferably comprising an orientation sensor.
[0039] The individual facets of the first facet mirror 18 are imaged into the obj ect field 11 with the aid of the second facet mirror 19, with this regularly only being approximate imaging. The second facet mirror 19 may be the last beam-shaping mirror or else actually the last mirror for the illumination radiation in the beam path upstream of the obj ect field 11.
[0040] In each case one of the facets of the second facet mirror 19 is assigned to exactly one of the facets of the first facet mirror 18 for the purpose of forming an illumination channel for illuminating the obj ect field 11. This may in particular result in illumination according to the Kohler principle.
[0041] The facets of the first facet mirror 18 are imaged overlaid on one another by way of a respective assigned facet of the second facet mirror 19, for the purpose of illuminating the obj ect field 11. Here, the illumination of the obj ect field 11 is as homogeneous as possible. It preferably has a uniformity error of less than 2%. Field uniformity may be achieved by overlaying different illumination channels.
[0042] By selecting the used illumination channels, which is possible without problems by way of a suitable setting of the micromirrors 18 ' of the first facet mirror 18, it is still possible to set the intensity distribution in the entrance pupil of the proj ection system 20 described below. This intensity distribution is also referred to as illumination setting. Incidentally, it may be advantageous here to arrange the second facet mirror 19 not exactly in a plane that is optically conjugate to a pupil plane of the proj ection system 20. In particular, the pupil facet mirror 19 may be arranged tilted relative to a pupil plane of the proj ection system 20, as described in DE 10 2017 220 586 Al, for example.
[0043] In the arrangement of the components of the illumination optical unit 16 as illustrated in Figure 1, however, the second facet mirror 19 is arranged in an area conjugate to the entrance pupil of the proj ection system 20. The deflection mirror 17 and the two facet mirrors 18, 19 are arranged tilted both relative to the obj ect plane 12 and relative to one another in each case.
[0044] In an alternative embodiment, not illustrated, of the illumination optical unit 16, a transfer optical unit comprising one or more mirrors may additionally be provided in the beam path between the second facet mirror 19 and the obj ect field 11. The transfer optical unit may in particular comprise one or two normal-incidence mirrors (NI mirrors) and / or one or two grazing-incidence mirrors (GI mirrors). Using an additional transfer optical unit, it is possible in particular to take account of different poses of the entrance pupil for the tangential and for the sagittal beam path of the projection system 20 described below.
[0045] It is alternatively possible for the deflection mirror 17 illustrated in Figure 1 to be dispensed with, for which purpose the facet mirrors 18, 19 should then be suitably arranged relative to the radiation source 13 and the collector 14.
[0046] The obj ect field 11 in the reticle plane 12 is transferred to the image field 21 in the image plane 22 with the aid of the proj ection system 20. For this purpose, the proj ection system 20 comprises a plurality of mirrors M, which are consecutively numbered in accordance with their arrangement in the beam path of the proj ection exposure apparatus 1. The mirrors Mi are optical elements 25.
[0047] In the example illustrated in Figure 1, the proj ection system 20 comprises six mirrors Mi to Me as optical elements 25. Alternatives with four, eight, ten, twelve or any other number of mirrors M± are likewise possible. The penultimate mirror Ms and the last mirror Me each have a passage opening for the illumination radiation, as a result of which the illustrated proj ection system 20 is a doubly obscured optical unit. The proj ection system 20 has an image-side numericalaperture that is greater than 0.3 and may also be greater than 0. 6 and may be for example 0.7 or 0.75.
[0048] The reflection surfaces of the mirrors M may be in the form of freeform surfaces without an axis of rotational symmetry. However, the reflection surfaces of the mirrors M may alternatively also be designed as aspherical surfaces with exactly one axis of rotational symmetry of the reflection surface shape. Just like the mirrors of the illumination optical unit 16, the mirrors M may have reflective coatings for the illumination radiation. These reflective coatings may be designed as multilayer coatings, in particular with alternating layers of molybdenum and silicon.
[0049] The proj ection system 20 has a large obj ect-image shift in the y-direction between a y-coordinate of a centre of the obj ect field 11 and a y-coordinate of the centre of the image field 21. This obj ect-image shift in the y-direction may be of approximately the same magnitude as a z-distance between the obj ect plane 12 and the image plane 22.
[0050] In particular, the proj ection system 20 may be designed to be anamorphic, that is to say it has different imaging scales px, py in the x-direction and y-direction in particular. The two imaging scales px, pyof the proj ection system 20 are preferably ( px, py) = (+ / - 0.25, + / - 0.125). An imaging scale p of 0.25 corresponds here to a reduction with a ratio of 4: 1, while an imaging scale p of 0.125 results in a reduction with a ratio of 8: 1. A positive sign in the case of the imaging scale p means imaging without image inversion; a negative sign means imaging with image inversion.
[0051] Other imaging scales are likewise possible. Imaging scales px, pywith the same sign and the same magnitude in the x-direction and y-direction are also possible.
[0052] The number of intermediate image planes in the x-direc-tion and in the y-direction in the beam path between the object field 11 and the image field 21 may be the same or different, depending on the embodiment of the proj ection system 20. Examples of proj ection systems 20 with different numbers of such intermediate images in the x-direction and y-direction are known from US 2018 / 0074303 Al.
[0053] In particular, the proj ection system 20 may comprise a homocentric entrance pupil. The latter may be accessible. However, it may also be inaccessible.
[0054] A reticle 30 (also referred to as mask) arranged in the obj ect field 11 is exposed by the illumination system 10 and transferred by the proj ection system 20 onto the image plane 21. The reticle 30 is held by a reticle holder 31. The reticle holder 31 is displaceable in particular in a scanning direction by way of a reticle displacement drive 32. In the exemplary embodiment illustrated, the scanning direction runs in the y-direction.
[0055] The reticle 30 may have an aspect ratio of between 1: 1 and 1: 3, preferably between 1: 1 and 1: 2, and particularly preferably of 1: 1 or 1: 2. The reticle 30 may be configured to be substantially rectangular and preferably has a length and a width of 5 to 7 inches ( 12.70 to 17.78 cm), more preferably a length and a width of 6 inches ( 15.24 cm). As an alternative thereto, the reticle 30 may have a length of 5 to 7 inches (12.70 cm to 17.78 cm) and a width of 10 to 14 inches (25.40 cm to 35.56 cm), preferably a length of 6 inches (15.24 cm) and a width of 12 inches (30.48 cm).
[0056] A structure on the reticle 30 is imaged onto a lightsensitive layer of a wafer 35 arranged in the region of the image field 21 in the image plane 22. The wafer 35 is held bya wafer holder 36. The wafer holder 36 is displaceable in particular in the y-direction by way of a wafer displacement drive 37. The displacement, firstly, of the reticle 30 by way of the reticle displacement drive 32 and, secondly, of the wafer 35 by way of the wafer displacement drive 37 may be synchronized with one another.
[0057] The proj ection exposure apparatus 1 illustrated in Figure 1 or the proj ection system 20 thereof comprises a supporting frame 41, illustrated only partially in Figure 1, to which the optical elements Ml - M6 are fastened with the aid of actuators 42. The actuators 42 allow the individual optical elements to move in a total of six rigid body degrees of freedom relative to the supporting frame. In Figure 1, this is illustrated just for the optical element M2 for the sake of a simpler overview. The optical elements additionally comprise additional actuators 43 by way of which it is possible to change a respective shape of the optical surface of the optical elements.
[0058] The actuators 42, 43 are connected to a control unit 44 that is designed to generate control commands for the actuators. A wavefront aberration of the proj ection system 20 is able to be at least partially compensated for by driving the actuators 42. The control commands are generated on the basis of a prediction model that is provided with the aid of a method according to the invention. This will be explained below with reference to Figure 2.
[0059] Figure 2 shows a schematic overview for illustrating a method according to the invention for providing a prediction model and of a method according to the invention for ascertaining a temporal profile of a wavefront aberration. An exemplary method for providing a prediction model will first be explained below.
[0060] In a first step, a plurality (j = 1,...,N) of training datasets 52 are provided, these being illustrated as follows in Figure 2:In this case,denotes a heat flow vector of the training dataset with the index j at the time t[ anddenotes a wavefront vector of the training dataset with the index j at the time t,. The entries of the wavefront vector are formed by the Zernike coefficients of the wavefront associated with the respective dataset. Each training dataset comprises heat flow vectors and wavefront vectors for a plurality of times t, = 1, dJ, wherein different training datasets may have a different number dJof times. The training datasets may be obtained with the aid of simulation methods known in principle to those skilled in the art.
[0061] As an alternative or in addition, training datasets may also be generated by a measurement in which a predefined temporal profile of a heat flow vector is applied to the proj ection system and corresponding wavefront vectors are measured at regular time intervals.
[0062] In a further step, a wavefront model 51 is created, said wavefront model being based on the following ordinary differential equation:In this case, fg3denotes a non-linear function defined by a set of 03 parameters.
[0063] The training datasets are then supplied to the wavefront model 51. For this purpose, the wavefront vectorsare first mapped into a latent space with the aid of an encoder function Eθ1,as follows:= (t), wherein (t) is the latent vector associated with the wavefront vector z(t). Onaccount of correlations within the entries of the wavefront vectors (the Zernike coefficients), the latent vector (t) has a lower dimensionality than the wavefront vector.
[0064] In one alternative embodiment, the encoder function may also correspond to the identity function, and so in this case the wavefront vectors and the latent vectors are identical.
[0065] Optionally, the training datasets may be scaled in a manner known in principle, for example to a value range between 0 and 1, in order to simplify subsequent calculations.
[0066] A solution to the ordinary differential equation is sought within the scope of the wavefront model. For this purpose, it is possible to use differential equation solvers known in principle to those skilled in the art, for example Runge-Kutta solvers. The wavefront model is additionally trained on the basis of the training datasets in a manner known in principle using a machine learning method with the aid of a neural network. As part of the training, the free parameters 01,02, 03 of the neural network are modified in such a way that the solution to the ordinary differential equation describes the training data as well as possible. Such a method is known in principle to those skilled in the art, and so a more detailed explanation may be dispensed with in the present case. By way of example, a stochastic gradient descent method, such as for example " ADAM" (cf. Kingma, Diederik; Ba, Jimmy (2014 ). " Adam: A Method for Stochastic Optimization". arXiv: 1412. 6980) may be used as part of the training. Such stochastic gradient descent methods require the gradients, resulting from the training steps ("rates of change" ), of the solution to the differential equation of the neural network, which result from the changes of the free parameters. These gradients may be ascertained with the aid of what are referred to as adjunct methods, as discussed in the context of neuralordinary differential equations, for example by Chen et. al " Neural Ordinary Differential Equations" (2018 ). In this case, use is preferably made of what are known as "discretize-then-optimize" methods, which are distinguished in that a back-propagation method is applied directly to the calculation steps of the differential equation solver.
[0067] A loss function used to train the wavefront model may be defined for example by way of a deviation between the wavefront aberrations predefined by the training data and the wavefront aberrations predicted by the wavefront model, for example as an RMS (root mean square) error averaged over time and all Zernike coefficients. A prediction model 50 is provided by training the wavefront model 51.
[0068] A method for ascertaining a temporal profile of a wavefront aberration and a method for operating a proj ection system will be explained below by way of example with reference to Figure 2. In order to ascertain a temporal profile of a wavefront aberration, an initial state 55 of a wavefront aberration of the optical system may first be determined. In the present case, for this purpose, a wavefront measurement is carried out at the time t = t0, using which a wavefront vector z(to) describing the initial state 55 is obtained. The initial state 55 is supplied to the prediction model 50.
[0069] A planned temporal profile 56 of a heat flow vector Q(t) is also determined. This is done taking into account a provided illumination setting of the illumination system 10. A reticle 30 used in the process and heating profiles acting on the optical elements with the aid of preheaters and / or sector heaters are additionally taken into account. For each of the optical elements Ml - M6, the heat flow vector Q(t)contains a suitable number of entries describing the planned thermal influence acting on the optical elements.
[0070] On the basis of the supplied initial state 55 and the supplied temporal profile of the thermal influencing variable 56 (or on the basis of the heat flow vector Q(t)), the prediction model 50 generates a latent vector (t) that is transformed into a wavefront vector z(t) within the scope of the prediction model by a decoder function(step 54 ). The wavefront vector z(t) represents a prediction for the temporal profile of the wavefront aberration 57, and is output for times t > t0.
[0071] The temporal profile of the wavefront aberration 57 is then supplied to the control unit 44. The control unit 44 is designed, in a manner known in principle, to generate control commands for the actuators 42, 43 on the basis of the supplied temporal profile of the wavefront aberration 57. The control commands are each control command temporal profiles using which the actuators 42, 43 are able to be actuated over time in such a way that the movements or deformations of the optical elements Ml - M6 generated thereby at least partially compensate for the predicted wavefront aberrations. The wavefront aberration is thus in particular able to be kept within a predefined tolerance range over the entire exposure process.
[0072] The proj ection exposure apparatus 1 is then operated in such a way that heat is applied to the optical elements Ml -M6 in accordance with the planned temporal profile 56 of the heat flow vector Q(t). At the same time, the control commands are sent to the respective actuators 42, 43 in accordance with the control command temporal profiles, such that the predicted wavefront aberrations (predicted in accordance with the temporal profile 57 ) are compensated for as explained above.
Claims
Patent claims1. Computer-implemented method for providing a prediction model (50) designed to ascertain a temporal profile of a wavefront aberration (57 ) of an optical system (20), wherein the optical system (20) has at least one optical element (Ml - M6), comprising the following steps:- creating a wavefront model (51 ) comprising a latent vector representing the wavefront aberration and a model function, wherein the model function is dependent on a thermal influencing variable (56) acting on the at least one optical element (Ml - M6) and dependent on the latent vector, wherein a derivative of the latent vector with respect to time is described by the model function;- providing a plurality of training datasets (53), wherein a temporal profile of the thermal influencing variable (56) is assigned to a temporal profile of the wavefront aberration of the optical system (20) in each of the training datasets (53);- training the wavefront model (51 ) with the aid of a neural network using the plurality of training datasets (53).
2. Method according to Claim 1, in which the model function is ascertained by training the wavefront model (51 ).
3. Method according to Claim 1 or 2, in which at least some of the plurality of training datasets (53) are generated by a simulation, preferably an FEM simulation.
4. Method according to one of Claims 1 to 3, in which at least some of the plurality of training datasets (53) are generated by a measurement by virtue of the at least one optical element (Ml - M6) being exposed to a defined temporal profile of the thermal influencing variable (56) and a resulting temporal profile of the wavefront aberration of the optical system (20) being measured.
5. Method according to one of Claims 1 to 4, in which the thermal influencing variable (56) comprises or is formed from a heat flow vector acting on the at least one optical element (Ml - M6).
6. Method according to one of Claims 1 to 5, in which the optical system (20) is a proj ection system (20) of a proj ection exposure apparatus ( 1 ), wherein the proj ection system (20) has an obj ect field able to be illuminated by an illumination system of the proj ection exposure apparatus by way of a settable illumination setting, wherein a reticle is able to be positioned in the obj ect field, wherein the projection system (20) is designed to image at least a subregion of the reticle into an image plane, wherein the thermal influencing variable (56) is determined taking into account the settable illumination setting and / or the reticle.
7. Method according to one of Claims 1 to 6, in which the thermal influencing variable (56) is determined taking into account a heat flow acting on the at least one optical element (Ml - M6) and generated by one or more active heating elements.
8. Computer-implemented method for ascertaining a temporal profile of a wavefront aberration (57 ) of an optical system (20) having at least one optical element (Ml - M6), comprising the following steps:- providing a trained prediction model (50), in particular with the aid of a method according to one of Claims 1 to 7, wherein the prediction model is based on a wavefront model (51 ), wherein the wavefront model comprises a latent vector representing the wavefront aberration and a model function, wherein the model function is dependent on a thermal influencing variable (56) acting on the at least one optical element (Ml - M6) and dependent on the latent vector, wherein a derivative of the latent vector with respect to time is described by the model function;- inputting an initial state (55) of the wavefront aberration into the prediction model (50);- inputting a temporal profile of the thermal influencing variable (56) into the prediction model (50) in order to obtain a temporal profile of the wavefront aberration (57 ) on the basis of the initial state and the temporal profile of the thermal influencing variable (56).
9. Method according to Claim 8, in which the initial state (55) is ascertained with the aid of a wavefront measurement.
10. Method according to one of Claims 1 to 9, in which ascertaining the temporal profile of the wavefront aberration (57 ) serves the purpose of generating at least one control command for one or more actuators (42, 43) acting on the at least one optical element (Ml - M6), wherein the at least one control command is suitable for at least partially compensating for the wavefront aberration.
11. Method for operating a proj ection system (20) of a proj ection exposure apparatus ( 1 ), wherein the proj ection system (20) has an obj ect field ( 11 ) able to be illuminated by an illumination system ( 10) of the proj ection exposure apparatus ( 1 ), optical elements (Ml - M6) for imaging the obj ect field ( 11 ) into an image plane (22 ), at least one actuator (42, 43) for effecting a change in position and / or shape of at least one of the optical elements (Ml - M6), and a control unit (44 ) for driving the at least one actuator (42, 43), wherein the method comprises the following steps:- ascertaining a temporal profile of a wavefront aberration (57 ) of the proj ection system (20) with the aid of a method according to one of Claims 8 to 10;- generating, on the basis of the ascertained temporal profile of the wavefront aberration (57 ), at least one control command for one or more actuators (42, 43) acting on the at least one optical element (Ml - M6), wherein the at least one control command is suitable for counteracting a wavefront aberration at at least one time of the temporal profile;- operating the proj ection system (20) proceeding from the initial state (55) in such a way that the optical elements (Ml - M6) are exposed to the temporal profile of the thermal influencing variable (56) of the wavefront aberration;- driving the at least one actuator (42, 43) during the operation of the proj ection system (20) by way of the at least one control command.
12. Method according to Claim 11, in which generating at least one control command comprises generating a control command temporal profile suitable for keeping a wavefront aberration of the projection system (20) below a maximum toler- ance value during operation.
