DE102024201726A1 - Method and system for producing a lithography system or one of its components and/or for adjusting a system parameter of the lithography system - Google Patents

Abstract

Process comprising:
- generating (S1) N copies (304) of a neural network (306);
- Providing (S2) T training data sets, which are generated by the steps, for i = 1 to T:
a) reading (S20) of parameters from an i _th lithography system (1, 300),
b) generating (S21) an i _th training data set depending on the read parameters;
- Training (S3) the N copies, comprising the steps, for j = 1 to N:
aa) training (S30) the j _th copy using the i _th training data set;
bb) determining (S31) a j _th change information of the trained j _th copy (304);
- generating (S4) anonymized change information (310);
- Adapting (S5) the neural network (306);
- Prediction (S6) of an aging effect using the adapted neural network (312); and
- producing (S7) a lithography system (1, 300) or one of its components (302) depending on the predicted aging effect and/or adapting a system parameter of a lithography system (1, 300) depending on the predicted aging effect.

Description

The present invention relates to a method and a system for manufacturing a lithography system or one of its components and/or for adapting a system parameter and/or for predicting a failure of the lithography system.

Microlithography is used to manufacture microstructured components, such as integrated circuits. The microlithography process is carried out using a lithography system equipped with an illumination system and a projection system. The image of a mask (reticle) illuminated by the illumination system is projected by the projection system onto a substrate coated with a light-sensitive layer (photoresist) and arranged in the image plane of the projection system, for example, a silicon wafer, in order to transfer the mask structure to the light-sensitive coating of the substrate.

Driven by the pursuit of ever smaller structures in the production of integrated circuits, EUV lithography systems are currently being developed that use light with a wavelength in the range of 0.1 nm to 30 nm, particularly 13.5 nm. Since most materials absorb light at this wavelength, such EUV lithography systems must use reflective optics, i.e., mirrors, instead of the previously used refractive optics, i.e., lenses.

Due to the high radiation exposure within lithography systems, various aging processes occur in optical elements, such as mirrors or lenses, over the system's lifetime. With ever-increasing laser power and innovative exposure systems, it is also expected that numerous new or unknown (aging) effects and effect progressions will occur in the future that are currently unknown.

An example of aging effects observed in numerous existing lithography systems is the degradation of optical elements. For example, UV radiation used in immersion lithography alters the optical properties, such as refractive index, density, absorption coefficient, and birefringence, of the optical materials used in the optical elements. Examples of such materials include optical glasses, which are predominantly used in i-line (365 nm) lithography systems, and synthetic quartz glass (fused silica), which is an essential component of the optical material in systems operating at a wavelength of 248 nm (KrF) or 193 nm (ArF).

In pulsed 193 nm laser radiation, as used in immersion lithography, the change in the refractive index of the synthetic quartz glass, which is often used as a lens material, is known as compaction (increase in refractive index) or rarefaction (decrease in refractive index) and is used, for example, in US$10,427,965 described.

Especially with new, high-precision immersion lenses, the requirements for the wavefront are so high that higher-order wavefront changes occurring over the operating life of the lens, even in the range of a few nm, sometimes even less than 1 nm, can render the lens unusable. To extend the service life, it is possible to make individual optical elements replaceable.

Since the production of such an optical replacement element takes a long time and a downtime of the lithography system results in a high loss of revenue, it is disadvantageous for the customer to start the production of such an optical element only when the lithography system is outside its specifications and such an optical replacement element is urgently needed.

For this reason, it is desirable to make changes in the optical properties of optical elements, in particular of projection lenses, as predictable as possible in order to be able to manufacture a required optical replacement element in advance in order to replace it during an already planned maintenance of the lithography system.

In addition, a precise prediction of the optical properties of an optical element at a time of a planned replacement makes it possible to optimize the replacement optical element by attaching individual aspheres such that the optical properties of the specific optical element are optimal after the replacement.

A further advantage of predicting changes in the optical properties of an optical element and knowing the effect of the lithography system settings on such an optical element is that a customer's usage behavior can be optimized. For example, different lighting settings can be suggested directly in the lithography system, which maintain production but affect the lithography components. phy system wears more evenly and thus ensures a longer service life.

The load and thus the wear of the optical elements are also influenced by their optical design and other operating parameters of the lithography system. These operating parameters include, among others, the wavelength, the number and/or duration of light pulses, the size of a field illuminated on a lithography mask, the distribution of light intensity in a pupil plane of the respective optical element, the polarization of the light, the transmission of the lithography mask, and the transmission distribution.

Many of these operating parameters are actively adjusted in the lithography system (e.g. size of the field illuminated on the mask, distribution of the intensity of the light in the pupil plane, ...) or are determined by the configuration (temporal length of the laser pulses).

The effects of the individual operating parameters on the optical elements of the lithography system are generally known and can be described using simulation calculations. However, since the exact machine usage data is known only to the customer or the user of the lithography system, and this data constitutes highly sensitive trade secrets that should not leave the customer's infrastructure, it is difficult and often requires expensive test setups to further investigate the known effects, improve their prediction accuracy, and adapt them to changing (operating and/or environmental) conditions.

Since the simulation-based calculation of the aging effects or the wear of the optical elements is also computationally very complex, a direct calculation of the effects of operating parameters in the lithography system itself requires too much computing capacity and computing time.

In summary, it can be stated that, in the current state of the art, an optimal and/or controlled prediction of aging effects of optical elements due to irradiation and/or effects of various operating parameters on such aging effects is only possible to a very limited extent.

Against this background, it is an object of the present invention to provide an improved method and/or system for manufacturing a lithography system or one of its components and/or for adapting a system parameter of the lithography system, in particular taking into account an anonymization of system data to ensure customer and/or system anonymity.

• - Erzeugen von N Kopien eines neuronalen Netzwerkes zur Vorhersage eines Alterungseffekts von optischen Elementen von Lithographieanlagen, wobei N > 1;
• - Bereitstellen von T Trainingsdatensätzen, welche erzeugt sind, durch die Schritte, für i = 1 bis T, wobei T ≥ N:
  1. a) Auslesen von Parametern aus einer i_ten Lithographieanlage,
  2. b) Erzeugen eines i_ten Trainingsdatensatzes in Abhängigkeit der ausgelesenen Paramater;
• - Trainieren der N Kopien, aufweisend die Schritte, für j = 1 bis N:
  • aa) Trainieren der j_ten Kopie mithilfe des i_ten Trainingsdatensatzes;
  • bb) Bestimmen einer j_ten Änderungsinformation der trainierten j_ten Kopie;
• - Erzeugen einer anonymisierten Änderungsinformation in Abhängigkeit von zumindest zwei der N Änderungsinformationen;
• - Anpassen des neuronalen Netzwerkes in Abhängigkeit der anonymisierten Änderungsinformation;
• - Vorhersage eines Alterungseffekt mithilfe des angepassten neuronalen Netzwerks; und weiter optional:
• - Herstellen einer Lithographieanlage oder einer ihrer Komponenten in Abhängigkeit des vorhergesagten Alterungseffekts und/oder Anpassen eines Anlagenparameters und/oder Prognose über einen Ausfall (insbesondere Prognose eines Ausfalls) einer Lithographieanlage in Abhängigkeit des vorhergesagten Alterungseffekts.

Accordingly, a method for manufacturing a lithography system or one of its components and/or for adjusting a system parameter and/or for predicting a failure of the lithography system is proposed. The method comprises the steps:

• - Generating N copies of a neural network for predicting an aging effect of optical elements of lithography systems, where N >1;
• - Providing T training data sets generated by the steps, for i = 1 to T, where T ≥ N:
  1. a) Reading parameters from an i _th lithography system,
  2. b) generating an i _th training data set depending on the read parameters;
• - Training the N copies, having the steps, for j = 1 to N:
  • aa) training the j _th copy using the i _th training dataset;
  • bb) determining a j _th change information of the trained j _th copy;
• - generating anonymized change information depending on at least two of the N change information;
• - Adapting the neural network depending on the anonymized change information;
• - Prediction of an aging effect using the adapted neural network; and further optionally:
• - Manufacturing a lithography system or one of its components depending on the predicted aging effect and/or adjusting a system parameter and/or predicting a failure (in particular predicting a failure) of a lithography system depending on the predicted aging effect.

According to the method, multiple copies (N copies) of a neural network are generated. Preferably, a number N of generated copies of the neural network corresponds to a number of lithography systems, each of which is assigned one of the copies. Furthermore, according to the method, at least T training data sets are provided, where T should be at least as large as N. The training data sets are preferably generated system-specifically for each of the lithography systems or at least for a subset of the lithography systems. Each of the generated training data sets is then preferably used to train the copy of the neural network assigned to the respective lithography system. In this training step, the respective copy of the neural network is therefore preferably trained locally on the respective lithography system. Since the neural network is preferably trained on each of the lithography systems based on system-specific training data, this respective training does not require a large amount of computing power. An evaluation device, preferably used in the respective lithography system, by which the training is carried out or on which the neural network is executed, can thus be designed with low computing power. This shows one of the advantages of the federated learning approach used here. Preferably, all training data from the lithography system is accessible on the lithography system, and the locally executed copy of the neural network can access this training data in order to be retrained locally using this training data, without the training data leaving the system operator's infrastructure. In this way, retraining does not pose any security risks.

After training the respective neural network, change information is determined for each lithography system, which in particular records one or more deviations between network parameters and/or hyperparameters of the copy of the neural network trained on the respective system and the incoming neural network. The respective system-specific change information is further processed in a subsequent step to generate anonymized, in particular summarized, change information. Anonymization is performed to prevent any conclusions about the respective lithography system from being traced back to the respective system. This is advantageous because lithography systems often use process-specific and/or user-specific settings and/or process parameters that may constitute trade secrets. In particular, once a predetermined number of change information items have been determined for a predetermined number of lithography systems, the respective change information can be uploaded to a secure server or a secure cloud. Preferably, no external access to the server or cloud is possible.

The initially copied neural network is then trained based on the anonymized change information to be optimized. This can improve the accuracy of predicting the aging effect.

In other words, the global neural network is updated based on the anonymized change information. The degree of change or optimization of the neural network can be determined preferably based on the anonymized change information. This prevents uncontrolled deterioration of the neural network despite the lack of direct insight into the respective plant-specific change information.

It is particularly preferred if the neural network trained on the basis of the anonymized change information is tested by comparing it with a simulation model for simulating the aging effect in order to evaluate the quality and/or credibility of the optimized neural network and/or the simulation model.

Once the updated neural network has been optimized compared to the previous neural network based on the performance data resulting from the comparison, the updated neural network can be made available to the lithography systems as a new release or software update. For this purpose, preferably N copies of the updated neural network are provided and transmitted to the respective lithography system. On the respective lithography system, the copy of the updated neural network can preferably be tested again by comparing it with a system-specific simulation model or using historical simulation data. Since simply testing or comparing the copy of the updated neural network with the respective simulation model requires comparatively little computing power compared to training a neural network, the local performance of the copy of the neural network can be easily tested on the respective lithography system. Such a verification or testing mechanism is advantageous to prevent uncontrolled deterioration (e.g., due to local overfitting) of the respective copy of the (updated) neural network or to identify local outliers.

In relation to the present method, the terminology “neural network” is synonymous with a machine learning model provided for the prediction of the aging effect.

The prediction or diagnosis of aging effects of an optical element using federated learning and the improved development opportunities for new lithography systems resulting from federated learning are not limited to degradation-based aging effects, but are equally applicable to other, possibly previously unknown aging effects. The method described here is therefore based on a "federated learning" approach, with which historical and/or currently existing domain knowledge about aging effects of optical elements can be comprehensively incorporated. Federated learning allows the calculations of aging effects to be efficiently carried out or executed on the respective participating lithography system. Federated learning also enables continuous optimization of the prediction accuracy of aging effects of optical elements of the individual lithography systems by leveraging knowledge transfer and/or knowledge acquisition from other lithography systems, without individual system information being disclosed or accessible. Thus, the present method enables an anonymized improvement in the prediction accuracy of aging effects of optical elements of the individual lithography systems, without critical information leaving the customer's infrastructure.

The present method can be used, on the one hand, to perform plant-specific monitoring of the aging effect, in particular by training and optimizing the respective copy of the neural network based on plant parameters. Furthermore, by retraining the input of the neural network used for copying, a global optimization of the neural network based on the anonymized change information can be performed. Such optimization of the neural network can improve global system knowledge about the aging effect. Furthermore, it can be preferred if, after retraining the neural network based on the anonymized change information, a comparison is carried out with simulation data from a simulation of the aging effect. If the comparison reveals differences, systematic and/or syntactical errors in the simulation model can be detected and, if necessary, corrected. The preferably continuous optimization of the neural network based on the iteratively determined anonymized change information improves the prediction accuracy of the aging effect. This knowledge is then used to redesign a lithography system or a system component, or to readjust a system parameter, in particular to minimize the aging effect. A further advantage of the present method is that by locally training the respective copies of the neural network on each of the lithography systems, an immediate prediction of the aging effect can be made for each lithography system. This provides a fast diagnostic path for the respective detection of the aging effect, resulting in better actionability. The present federated learning approach, in particular, enables a reliable future prediction of the aging effect, which can be advantageously used, preferably, in swap pool predictions. This can also improve strategic decisions regarding production and/or maintenance planning for each lithography system, for example, leading to an increase in system productivity and/or a reduction in maintenance-related system downtime. Furthermore, the indirect analysis of user behavior and/or optimization of the aging effect prediction based on the anonymized change information can provide an application and/or setting recommendation for a system user. Such an application and/or setting recommendation can, for example, consist of instructing the system user to change a system parameter to improve service life. Locally training the copies of the neural network on the respective lithography system using federated learning enables a direct aging-effect-related "cause to result" link, thus enabling efficient calculation of the aging effect.

This method also offers data protection and security-related advantages, as a system user no longer needs to send potentially confidential and/or sensitive data to a system manufacturer in order to receive suggestions for optimizing system parameters. This also eliminates the risk of unwanted data leaks or data misuse. Furthermore, the system manufacturer does not need to provide a complex security architecture for data storage, as the system operator preferably only receives the anonymized change information for retraining the neural network. This also reduces the amount of data traffic generated. The information contained in the anonymized change information is used "blindly" and cannot be manually reviewed. This significantly reduces the risk of data misuse.

According to one embodiment, the neural network for predicting the aging effect is used before generating the N copies based on historical and/or current and/or synthetic and/or pre-trained on the aging effect using simulation-generated training data.

"Pre-training" refers to a process or method step in which the neural network, especially an initially untrained one, is trained based on existing data and/or information about the aging effect in order to learn and/or improve its ability to predict this aging effect. The training data, especially initial data, preferably used for pre-training, can originate from various sources, including historical data (past data containing information about the aging effect), current data (current information about the aging effect), synthetically generated data (artificially generated data representing the aging effect), and simulatively generated data (data created through simulations and reflecting the aging effect). In other words, existing knowledge about the aging effect to be predicted is used to create a neural network. If real data from previous aging determinations is available, the neural network can be (pre-)trained based on this real data. This is particularly preferred because training based on real data allows a basic structure and/or hyperparameters of the neural network to be adapted to a structure of the real data. Alternatively or additionally, the neural network can also be trained based on artificially or synthetically generated data or based on simulation data, in particular to achieve faster network convergence, in particular by minimizing a loss function. In the present case, the neural network initially trained in this way is further improved, in particular continuously, using a federated learning approach by using current, plant-specific system knowledge to retrain the neural network. This allows, on the one hand, existing real data to be used to train an initial neural network. On the other hand, if real data is missing, the plant-specific system knowledge can be used.

According to one embodiment, at least some of the method steps are carried out iteratively in order to continuously and/or continuously improve a model performance of the neural network for predicting the aging effect.

“Iterative execution” means that the steps related to the neural network and its model performance for predicting the aging effect are performed repeatedly and step by step. Instead of training the neural network in a single step, the model is improved through a sequence of steps that are repeatedly executed. This allows the neural network to be continuously improved. The “model performance of the neural network” preferably refers to the ability or accuracy of the neural network to predict the aging effect. Model performance preferably describes how accurately and effectively the neural network makes these predictions. “Continuous improvement” in this case preferably means that the goal of the iterative process steps is to continuously increase the performance of the neural network over time. With each iteration, improvements are to be achieved in order to make the neural network more accurate and/or efficient at predicting the aging effect.

• Gewichten der zumindest zwei der N Änderungsinformationen; und
• Summieren der gewichteten zumindest zwei der N Änderungsinformationen zu der anonymisierten Änderungsinformation.

According to one embodiment, generating the anonymized change information as a function of at least two of the N change information comprises:

• Weights of at least two of the N change information; and
• Summing the weighted at least two of the N change information to the anonymized change information.

A single piece of anonymized change information is generated from the individual change information of the respective lithography system using data processing in order to retrain the neural network based on this anonymized change information. In this way, all lithography systems benefit from an improved neural network for predicting the aging effect, without the need to release or provide system-specific information. The anonymized change information is preferably based on a weighted sum of at least two of the system-specific change information items. The weighting is preferably random or dependent on the performance or performance distribution of the copies of the neural network. It is preferred if the total weights add up to 1. By aggregating the individual weights of many different lithography systems and/or system types, subsequent conclusions about individual applications and/or lithography systems are no longer possible. Thus, the anonymized change information no longer contains any security- or data protection-critical knowledge. It is particularly preferred if the anonymized change information is only generated when change information or related data for different lithography systems and/or lithography system types and/or system manufacturers are available. This is preferred because secure anonymization of change information is only possible once a certain amount of different change information has been collected. This also ensures that, based on the anonymized change information, neither system usage behavior for individual lithography systems nor system usage behavior for specific lithography system groups and/or users can be determined. For example, anonymized change information is only generated once there are at least 10 lithography systems, especially from different system operators and/or system types, since secure anonymization cannot be achieved beforehand.

One piece of change information is described above only as an example per lithography system or with the same total number as the lithography systems (N pieces of change information for N lithography systems). It should be understood that multiple pieces of change information for a lithography system only occur if, for example, the same network is used for individual components. However, these components can be considered separately. Therefore, in other embodiments, multiple pieces of change information per lithography system are also possible, but are not necessary for encryption or anonymization.

Each piece of change information preferably comprises many individual parameters (in the case of a neural network, in particular the weights and/or thresholds). These are preferably optimized and thereby modified during each training step, in particular locally. The change information of each i- _th machine therefore preferably consists of many individual parameters. A separate weight is preferably applied to each of these parameters for encryption or anonymization, with the sum of each of the parameters across all change information preferably being exactly one.

According to one embodiment, the weighting of at least two of the N change information items is performed randomly or based on a performance parameter distribution of the trained N copies of the neural network.

Preferably, multiple pieces of change information from multiple lithography systems are used to generate the weighted sum. The number of system-specific pieces of change information is preferably selected randomly to avoid any possibility of tracing the data back to individual lithography systems as early as the data selection step for generating the anonymized change information. However, when randomly selecting the system-specific change information, the respective performance of the respective copy of the neural network can preferably be taken into account, for example, to avoid considering change information originating from low-performing or poorly predictive copies of the neural network. Performance can be taken into account, for example, by a performance limit condition. A performance distribution based on the local change information can also be considered.

According to one embodiment, the aging effect comprises a radiation-induced degradation of optical elements and/or a thermal and/or mechanical aging effect and/or an aging effect due to material deposition and/or corrosion and/or chemical degradation.

Radiation-induced degradation specifically describes a process in which an optical element (such as a lens or a mirror) in the lithography system loses its efficiency over time due to continuous exposure to intense radiation, typically UV light or electron beams. This can occur in various ways. Firstly, a change in the optical properties of the optical element is possible, whereby, for example, the transmission, reflection, and/or refractive index of the materials of the optical element can change. Furthermore, radiation-induced damage to the surface of the optical element can occur. For example, cracks and/or discoloration can occur on the surface of the optical elements. Thermal effects can lead to the aging of the optical element through long-term exposure to high energy, which can lead to an increase in the temperature of the optical elements. This energy input can, in turn, change the shape and/or orientation of the optical element. Mechanical wear of the optical element can also lead to aging. In particular, moving parts in the lithography system can wear out through regular use, which can affect the alignment and/or focus of the optical elements. Deposition of materials can also cause aging of the optical elements. In an environment with dust and/or other particles, these can deposit on the optical elements and impair their performance. Such particles can also be generated by outgassing from components of the lithography system due to thermal energy input and can increase over the system's lifetime. Corrosion and/or chemical degradation can also influence aging. Certain environments Operating conditions and/or chemical exposure may lead to corrosion and/or other chemical damage to the optical elements.

According to one embodiment, the optical elements comprise lenses and/or mirrors and/or optical sensors.

Of course, the lithography system may also incorporate other optical elements not explicitly mentioned here. Therefore, the list provided here should not be considered limiting.

• Retrofit zumindest einer Komponente einer Lithographieanlage in Abhängigkeit des vorhergesagten Alterungseffekts, um den Alterungseffekt zu minimieren; und/oder
• schrittweises oder stufenweises Anpassen eines Anlagenparameters einer Lithographieanlage in Abhängigkeit des vorhergesagten Alterungseffekts, um den Alterungseffekt zu minimieren.

According to one embodiment, the production of a lithography system or one of its components depending on the predicted aging effect and/or the adaptation of a system parameter of a lithography system depending on the predicted aging effect comprises:

• Retrofitting at least one component of a lithography system depending on the predicted aging effect in order to minimize the aging effect; and/or
• Gradual or step-by-step adjustment of a system parameter of a lithography system depending on the predicted aging effect in order to minimize the aging effect.

A retrofit of at least one component of a lithography system based on the predicted aging effect preferably aims to modify or replace existing components of the lithography system to reduce the influence of aging. Such a retrofit can include updating or replacing parts such as lenses, mirrors, mechanical parts, or control elements that could be affected by aging effects such as wear, degradation, or loss of efficiency. Such a retrofit can be based on the aging effect predicted by the neural network and, if necessary, supplemented by further analyses, empirical data, and/or simulation-based forecasts. The gradual or step-by-step adjustment of a system parameter of a lithography system based on the predicted aging effect includes, for example, fine-tuning the system's operating parameters to compensate for or minimize the effects of aging. This can include a gradual or step-by-step adjustment of parameters such as light intensity, focusing, alignment, or temperature control. Proactive adjustment is also possible, preferably based on the prediction of the aging effect, in order to maintain the performance of the system and minimize potential downtime.

Through more precise predictions about the cause and/or type of component performance losses, e.g., due to degradation, it is conceivable, in addition to adjusting various known parameters or settings of the components, such as optimal rotation and/or mirroring, to generate new, improved lighting settings based on the new findings. On the one hand, these can optimize or improve the performance of the lithography system with already degraded components, which is tantamount to troubleshooting. On the other hand, they make it possible to extend the service life of the lithography system and/or the components in question. Furthermore, scenarios can be easily and accurately calculated, showing, for example, how a possible reduction in light source intensity could have a positive effect on the service life of individual components. The more precise prediction also makes it possible to determine when maintenance and/or replacement of a component in question becomes absolutely necessary in order to avoid downtime of the lithography system until replacement parts arrive.

The predictions made can also be used to improve the simulation of (aging) effects of potential new component developments. This can generate a better benefit assessment and/or improved target specifications. Furthermore, it is possible to plan early on which capacities individual components require in production planning and which components should be made interchangeable. Furthermore, the predictions made can enable specifications and/or new usage settings for components. For example, the prediction made here enables a rotation of the lighting settings of components, which can also be tested immediately.

According to one embodiment, minimizing the aging effect is based on solving an inverse problem related to the aging effect and/or solving a feedback problem and/or solving an optimization problem.

The term "solving an inverse problem related to the aging effect" preferably means determining the causes of the predicted aging effect. Here, this means inferring the underlying causes or mechanisms from the predicted aging effect. This can be done, for example, by analyzing the changes in the optical properties, such as a decrease in transparency or a change in the refractive index, in order to identify the specific degradation processes. The phrase “solving a feedback problem” means adapting the lithography system or one of its components or system parameters so that a desired state or performance is achieved. For example, the system parameters or components of the system can be adapted so that they achieve the desired performance despite the effects of aging. The phrase “solving an optimization problem” describes finding the best solution under given constraints, in particular taking several competing factors into account. In the present case, this preferably means finding the optimal operating conditions or component configurations to minimize the negative effects of aging, in particular taking into account factors such as availability and performance efficiency.

It is also conceivable to generate specific component improvements in a simulation based on simulation parameters that are adjusted based on aging effects. Furthermore, it may be possible to simulate swap options and/or redundancies based on the aging effects, attach these to implementation costs, and then simulate the overall costs in conjunction with the neural network, thus making decisions regarding the nature of future developments. In other words, it is possible to determine optical properties based on predicted aging effects in order to counteract these aging effects. These properties can be used as input for manipulators, for example, to at least partially automate simulations for the design of new components.

According to one embodiment, the generation of the anonymized change information as a function of at least two of the N change information items comprises, in particular, a respective local, unidirectional provision of the N change information items to a server or a cloud.

The anonymized change information is preferably generated on a server or cloud. Such a server or cloud is preferably protected from external access by appropriate security measures. For example, a server or cloud may have only an intranet connection and thus be isolated from access via the internet.

According to one embodiment, a layer structure and/or a layer-wise number of neurons of the neural network is selected depending on the aging effect.

In this context, the layered structure of the neural network preferably refers to the arrangement and structure of the various layers in the neural network, including the number of layers and the type of these layers (e.g., convolutional layers, pooling layers, fully connected layers). The layered structure is preferably selected to be specifically suited for the identification, analysis, and management of aging effects. The selection of the number of neurons per layer preferably refers to adjusting the number of neurons in each layer of the neural network. The number and distribution of neurons in the various layers can have a significant impact on the performance and specialization of the network. The number of neurons is preferably selected depending on the specific requirements of the aging effect. For example, a network designed to detect subtle, complex aging patterns might require a larger number of neurons in certain layers. The configuration of the neural network, both in terms of the layered structure and the number of neurons, is based on the specific properties and requirements of the aging effect to be detected. This means that the network is trained and/or configured to respond effectively to the specific patterns, indicators, or consequences of aging effects.

• • (insbesondere Beleuchtungs-)Settings der Lithographieanlage
• • Reticle Transmittion
• • Pulsdaten
• • Leistung der Quelle/des Lasers
• • Material der einzelnen Komponenten
• • Maschinentypen
• • Quellumgebung

The many relevant measurement data from the optics of a lithography system are, in particular, wavefront measurements that describe the aberrations of optical systems or components. To predict these for the future, it is preferable to include the type of use of the lithography system (possibly as a time series). For example, a convolutional neural network (CNN) with additional input layers can be integrated into the fully connected intermediate layer of the neural network. This preferably includes a large number of relevant usage data from the lithography system (especially including a data history). For example, the following usage data is included:

• • (especially lighting) settings of the lithography system
• • Reticle Transmission
• • Pulse data
• • Power of the source/laser
• • Material of the individual components
• • Machine types
• • Source environment

This usage data is preferably incorporated into the fully connected intermediate layer as feature inputs. The wavefront (Zernike) image to be considered is preferably incorporated into the convolutional neural network as input.

The size of the neural network is preferably flexible, and so is the layer structure. A simple convolutional network with three layers is conceivable, with usage data being incorporated as additional input data after layer 1. Depending on the desired complexity, a simple network with residual blocks, in which the usage conditions are incorporated within the residual block, is also conceivable.

However, a suitable network would in any case contain over 2000 trainable parameters, which would make efficient encryption possible.

According to one embodiment, the j _th change information each comprises aging-relevant information about plant-specific operating parameters and/or about a plant configuration and/or about a plant interface and/or about a plant history and/or about a plant make and/or about a plant type.

• eine, insbesondere zentrale, Recheneinrichtung,
• i Lithographieanlagen mit jeweils einer Auswerteeinrichtung,
• einen Server oder eine Cloud,
• wobei Recheneinrichtung dazu ausgebildet ist, N Kopien eines neuronalen Netzwerkes zur Vorhersage eines Alterungseffekts von optischen Elementen von Lithographieanlagen zu erzeugen, wobei N > 1, und eine i_te Kopie der N Kopien der Auswerteeinrichtung jeder i_ten Lithographieanlage bereitzustellen,
• wobei die Auswerteeinrichtung jeder der i_ten Lithographieanlage dazu ausgebildet ist, Parameter aus der i_ten Lithographieanlage auszulesen und einen i_ten Trainingsdatensatz in Abhängigkeit der ausgelesenen Parameter zu erzeugen, die i_te Kopie der N Kopien mithilfe des i_ten Trainingsdatensatzes zu trainieren, eine i_te Änderungsinformation der trainierten i_ten Kopie zu bestimmen, und die bestimmte i_te Änderungsinformation dem Server oder der Cloud bereitzustellen, so dass insgesamt N Änderungsinformationen bereitgestellt sind,
• wobei der Server oder die Cloud dazu ausgebildet ist, eine anonymisierte Änderungsinformation in Abhängigkeit von zumindest zwei der N Änderungsinformationen zu erzeugen, und die anonymisierte Änderungsinformation der Recheneinrichtung bereitzustellen,
• wobei die Recheneinrichtung und/oder die Auswerteeinrichtung dazu ausgebildet ist/sind, das neuronale Netzwerkes in Abhängigkeit der anonymisierten Änderungsinformation anzupassen, einen Alterungseffekt mithilfe des angepassten neuronalen Netzwerks vorherzusagen, und eine Steueranweisung zum Herstellen einer Lithographieanlage oder einer ihrer Komponenten in Abhängigkeit des vorhergesagten Alterungseffekts und/oder zum Anpassen und/oder zur Prognose über einen Ausfall eines Anlagenparameters einer Lithographieanlage in Abhängigkeit des vorhergesagten Alterungseffekts bereitzustellen.

Furthermore, a system for manufacturing a lithography system or one of its components and/or for adjusting a system parameter and/or for predicting a failure of the lithography system is proposed. The system comprises:

• a, in particular central, computing facility,
• i Lithography systems, each with an evaluation device,
• a server or a cloud,
• wherein the computing device is designed to generate N copies of a neural network for predicting an aging effect of optical elements of lithography systems, where N > 1, and to provide an i _th copy of the N copies to the evaluation device of each i _th lithography system,
• wherein the evaluation device of each of the i _th lithography systems is designed to read out parameters from the i _th lithography system and to generate an i _th training data set depending on the read-out parameters, to train the i _th copy of the N copies using the i _th training data set, to determine an i _th change information of the trained i _th copy, and to provide the determined i _th change information to the server or the cloud, so that a total of N change information is provided,
• wherein the server or the cloud is designed to generate anonymized change information depending on at least two of the N change information items, and to provide the anonymized change information to the computing device,
• wherein the computing device and/or the evaluation device is/are designed to adapt the neural network depending on the anonymized change information, to predict an aging effect using the adapted neural network, and to provide a control instruction for producing a lithography system or one of its components depending on the predicted aging effect and/or for adapting and/or predicting a failure of a system parameter of a lithography system depending on the predicted aging effect.

The aging effect can, for example, be determined locally in the evaluation device and used to output individual instructions, particularly those related to the individual lithography systems. The aging effect can also be determined in the computing device, for example, and used to generate a global update of lithography system software based on change parameters. Furthermore, it may be possible to make global changes to a system or product design, particularly by improving the simulation capabilities of individual components, and to compare this with other designs.

The lithography system preferably comprises projection optics. The lithography system may also have an illumination system. The lithography system or the projection exposure system may be an EUV lithography system. EUV stands for "Extreme Ultraviolet" and refers to a wavelength of the working light between 0.1 nm and 30 nm. The projection exposure system may also be a DUV lithography system. DUV stands for "Deep Ultraviolet" and refers to a wavelength of the working light between 30 nm and 250 nm.

"One" in this case is not necessarily limited to a single element. Rather, multiple elements, such as two, three, or more, may also be included. Any other counting term used here should not be understood as implying a limitation to the exact number of elements stated. Rather, numerical deviations upwards and downwards are possible, unless otherwise stated.

The embodiments and features described for the method apply accordingly to the proposed system and vice versa.

Further possible implementations of the invention also include combinations of features or embodiments described above or below with respect to the exemplary embodiments that are not explicitly mentioned. In this case, the person skilled in the art will also add individual aspects as improvements or additions to the respective basic form of the invention.

• 1 zeigt einen schematischen Meridionalschnitt einer Projektionsbelichtungsanlage für die EUV-Projektionslithographie;
• 2 zeigt ein schematisches Flussdiagramm eines Ausführungsbeispiels des vorliegenden Verfahrens zum Herstellen einer Lithographieanlage oder einer ihrer Komponenten und/oder zum Anpassen eines Anlagenparameters der Lithographieanlage; und
• 3 zeigt ein schematisches Blockschaltbild eines Ausführungsbeispiels des vorliegenden Verfahrens zum Herstellen einer Lithographieanlage oder einer ihrer Komponenten und/oder zum Anpassen eines Anlagenparameters der Lithographieanlage.

Further advantageous embodiments and aspects of the invention are the subject of the dependent claims and the exemplary embodiments of the invention described below. The invention will be explained in more detail below using preferred embodiments with reference to the accompanying figures.

• 1 shows a schematic meridional section of a projection exposure system for EUV projection lithography;
• 2 shows a schematic flow diagram of an embodiment of the present method for manufacturing a lithography system or one of its components and/or for adjusting a system parameter of the lithography system; and
• 3 shows a schematic block diagram of an embodiment of the present method for producing a lithography system or one of its components and/or for adapting a system parameter of the lithography system.

In the figures, identical or functionally equivalent elements are provided with the same reference numerals unless otherwise indicated. Furthermore, it should be noted that the representations in the figures are not necessarily to scale.

1 shows an embodiment of a projection exposure system 1 (lithography system), in particular an EUV lithography system. One embodiment of an illumination system 2 of the projection exposure system 1 has, in addition to a light or radiation source 3, an illumination optics 4 for illuminating an object field 5 in an object plane 6. In an alternative embodiment, the light source 3 can also be provided as a separate module from the remaining illumination system 2. In this case, the illumination system 2 does not include the light source 3.

A reticle 7 arranged in the object field 5 is exposed. The reticle 7 is held by a reticle holder 8. The reticle holder 8 can be displaced via a reticle displacement drive 9, in particular in a scanning direction.

In the 1 For explanation, a Cartesian coordinate system with an x-direction x, a y-direction y and a z-direction z is shown. The x-direction x runs perpendicular to the drawing plane. The y-direction y runs horizontally and the z-direction z runs vertically. The scanning direction runs in the 1 along the y-direction y. The z-direction z runs perpendicular to the object plane 6.

The projection exposure system 1 comprises a projection optics 10. The projection optics 10 serves to image the object field 5 into an image field 11 in an image plane 12. The image plane 12 runs parallel to the object plane 6. Alternatively, an angle other than 0° between the object plane 6 and the image plane 12 is also possible.

A structure on the reticle 7 is imaged onto a light-sensitive layer of a wafer 13 arranged in the image plane 12 in the region of the image field 11. The wafer 13 is held by a wafer holder 14. The wafer holder 14 can be displaced, in particular along the y-direction y, via a wafer displacement drive 15. The displacement of the reticle 7, on the one hand, via the reticle displacement drive 9, and the displacement of the wafer 13, on the other hand, via the wafer displacement drive 15, can be synchronized with each other.

The light source 3 is an EUV radiation source. The light source 3 emits, in particular, EUV radiation 16, which is also referred to below as useful radiation, illumination radiation or illumination light. The useful radiation 16 has, in particular, a wavelength in the range between 5 nm and 30 nm. The light source 3 can be a plasma source, for example an LPP source (Laser Produced Plasma, plasma generated with the aid of a laser) or a DPP source (Gas Discharged Produced Plasma, plasma generated by means of a gas discharge). It can also be a synchrotron-based radiation source. The light source 3 can be a free-electrode nen laser (English: Free Electron Laser, FEL).

The illumination radiation 16 emanating from the light source 3 is focused by a collector 17. The collector 17 can be a collector with one or more ellipsoidal and/or hyperboloidal reflection surfaces. The at least one reflection surface of the collector 17 can be exposed to the illumination radiation 16 at grazing incidence (GI), i.e., at angles of incidence greater than 45°, or at normal incidence (NI), i.e., at angles of incidence less than 45°. The collector 17 can be structured and/or coated, on the one hand, to optimize its reflectivity for the useful radiation and, on the other hand, to suppress stray light.

After the collector 17, the illumination radiation 16 propagates through an intermediate focus in an intermediate focal plane 18. The intermediate focal plane 18 can represent a separation between a radiation source module, comprising the light source 3 and the collector 17, and the illumination optics 4.

The illumination optics 4 comprises a deflecting mirror 19 and, downstream of this in the beam path, a first facet mirror 20. The deflecting mirror 19 can be a flat deflecting mirror or, alternatively, a mirror with a beam-influencing effect beyond the pure deflection effect. Alternatively or additionally, the deflecting mirror 19 can be designed as a spectral filter that separates a useful light wavelength of the illumination radiation 16 from stray light of a different wavelength. If the first facet mirror 20 is arranged in a plane of the illumination optics 4 that is optically conjugated to the object plane 6 as the field plane, it is also referred to as a field facet mirror. The first facet mirror 20 comprises a plurality of individual first facets 21, which can also be referred to as field facets. Of these first facets 21, 1 only a few examples are shown.

The first facets 21 can be designed as macroscopic facets, in particular as rectangular facets or as facets with an arcuate or partially circular edge contour. The first facets 21 can be designed as flat facets or, alternatively, as convexly or concavely curved facets.

As for example from the DE 10 2008 009 600 A1 As is known, the first facets 21 themselves can also be composed of a plurality of individual mirrors, in particular a plurality of micromirrors. The first facet mirror 20 can in particular be designed as a microelectromechanical system (MEMS system). For details, please refer to DE 10 2008 009 600 A1 referred to.

Between the collector 17 and the deflecting mirror 19, the illumination radiation 16 runs horizontally, i.e. along the y-direction y.

In the beam path of the illumination optics 4, a second facet mirror 22 is arranged downstream of the first facet mirror 20. If the second facet mirror 22 is arranged in a pupil plane of the illumination optics 4, it is also referred to as a pupil facet mirror. The second facet mirror 22 can also be arranged at a distance from a pupil plane of the illumination optics 4. In this case, the combination of the first facet mirror 20 and the second facet mirror 22 is also referred to as a specular reflector. Specular reflectors are known from US 2006/0132747 A1 , the EP 1 614 008 B1 and the US$6,573,978 .

The second facet mirror 22 comprises a plurality of second facets 23. In the case of a pupil facet mirror, the second facets 23 are also referred to as pupil facets.

The second facets 23 can also be macroscopic facets, which can be round, rectangular or hexagonal, or alternatively facets composed of micromirrors. In this regard, reference is also made to the DE 10 2008 009 600 A1 referred to.

The second facets 23 can have planar or alternatively convex or concave curved reflection surfaces.

The illumination optics 4 thus form a double-faceted system. This basic principle is also known as a fly's-eye integrator.

It may be advantageous not to arrange the second facet mirror 22 exactly in a plane that is optically conjugate to a pupil plane of the projection optics 10. In particular, the second facet mirror 22 may be arranged tilted relative to a pupil plane of the projection optics 10, as is shown, for example, in the DE 10 2017 220 586 A1 described.

With the help of the second facet mirror 22, the individual first facets 21 are imaged into the object field 5. The second facet mirror 22 is the last bundle-forming or actually The last mirror for the illumination radiation 16 in the beam path in front of the object field 5.

In a further embodiment of the illumination optics 4 (not shown), a transmission optics can be arranged in the beam path between the second facet mirror 22 and the object field 5, which transmission optics contributes in particular to the imaging of the first facets 21 into the object field 5. The transmission optics can have exactly one mirror, but alternatively also two or more mirrors, which are arranged one behind the other in the beam path of the illumination optics 4. The transmission optics can in particular comprise one or two mirrors for normal incidence (NI mirrors, normal incidence mirrors) and/or one or two mirrors for grazing incidence (GI mirrors, grazing incidence mirrors).

The illumination optics 4 has in the version shown in the 1 As shown, after the collector 17 there are exactly three mirrors, namely the deflection mirror 19, the first facet mirror 20 and the second facet mirror 22.

In a further embodiment of the illumination optics 4, the deflection mirror 19 can also be omitted, so that the illumination optics 4 can then have exactly two mirrors after the collector 17, namely the first facet mirror 20 and the second facet mirror 22.

The imaging of the first facets 21 by means of the second facets 23 or with the second facets 23 and a transmission optics into the object plane 6 is usually only an approximate imaging.

The projection optics 10 comprises a plurality of mirrors Mi, which are numbered according to their arrangement in the beam path of the projection exposure system 1.

In the 1 In the example shown, the projection optics 10 comprises six mirrors M1 to M6. Alternatives with four, eight, ten, twelve, or a different number of mirrors M1 are also possible. The projection optics 10 is a doubly obscured optics. The penultimate mirror M5 and the last mirror M6 each have a passage opening for the illumination radiation 16. The projection optics 10 has an image-side numerical aperture that is greater than 0.5 and can also be greater than 0.6, for example, 0.7 or 0.75.

Reflection surfaces of the mirrors Mi can be designed as freeform surfaces without a rotational symmetry axis. Alternatively, the reflection surfaces of the mirrors Mi can be designed as aspherical surfaces with exactly one rotational symmetry axis of the reflection surface shape. The mirrors Mi, like the mirrors of the illumination optics 4, can have highly reflective coatings for the illumination radiation 16. These coatings can be designed as multilayer coatings, in particular with alternating layers of molybdenum and silicon.

The projection optics 10 has a large object-image offset in the y-direction y between a y-coordinate of a center of the object field 5 and a y-coordinate of the center of the image field 11. This object-image offset in the y-direction y can be approximately as large as a z-distance between the object plane 6 and the image plane 12.

The projection optics 10 can, in particular, be anamorphic. It has, in particular, different magnifications βx, βy in the x and y directions x, y. The two magnifications βx, βy of the projection optics 10 are preferably (βx, βy) = (+/- 0.25, +/- 0.125). A positive magnification β means imaging without image inversion. A negative sign for the magnification β means imaging with image inversion.

The projection optics 10 thus leads to a reduction in the ratio 4:1 in the x-direction x, i.e. in the direction perpendicular to the scanning direction.

The projection optics 10 leads to a reduction of 8:1 in the y-direction y, i.e. in the scanning direction.

Other magnifications are also possible. Magnifications with the same sign and absolutely identical in the x and y directions (x, y), for example, with absolute values of 0.125 or 0.25, are also possible.

The number of intermediate image planes in the x- and y-directions x, y in the beam path between the object field 5 and the image field 11 can be the same or can be different, depending on the design of the projection optics 10. Examples of projection optics with different numbers of such intermediate images in the x- and y-directions x, y are known from US 2018/0074303 A1 .

Each of the second facets 23 is assigned to exactly one of the first facets 21 to form an illumination channel for illuminating the object field 5. This can result in particular in illumination according to the Köhler principle. The far field is divided into a plurality of object fields 5 by means of the first facets 21. The first facets 21 generate a plurality of images of the intermediate focus on the second facets assigned to them 23.

The first facets 21 are each imaged onto the reticle 7 by an associated second facet 23, superimposed on one another, to illuminate the object field 5. The illumination of the object field 5 is, in particular, as homogeneous as possible. It preferably has a uniformity error of less than 2%. Field uniformity can be achieved by superimposing different illumination channels.

By arranging the second facets 23, the illumination of the entrance pupil of the projection optics 10 can be geometrically defined. By selecting the illumination channels, in particular the subset of the second facets 23 that guide light, the intensity distribution in the entrance pupil of the projection optics 10 can be adjusted. This intensity distribution is also referred to as the illumination setting or illumination pupil fill.

A likewise preferred pupil uniformity in the area of defined illuminated sections of an illumination pupil of the illumination optics 4 can be achieved by redistributing the illumination channels.

Further aspects and details of the illumination of the object field 5 and in particular of the entrance pupil of the projection optics 10 are described below.

The projection optics 10 can, in particular, have a homocentric entrance pupil. This can be accessible. It can also be inaccessible.

The entrance pupil of the projection optics 10 cannot usually be precisely illuminated with the second facet mirror 22. When imaging the projection optics 10, which images the center of the second facet mirror 22 telecentrically onto the wafer 13, the aperture rays often do not intersect at a single point. However, a surface can be found in which the pairwise determined distance of the aperture rays is minimized. This surface represents the entrance pupil or a surface conjugate to it in spatial space. In particular, this surface exhibits a finite curvature.

It is possible that the projection optics 10 have different entrance pupil positions for the tangential and sagittal beam paths. In this case, an imaging element, in particular an optical component of the transmission optics, should be provided between the second facet mirror 22 and the reticle 7. With the help of this optical element, the different positions of the tangential entrance pupil and the sagittal entrance pupil can be taken into account.

In the 1 In the illustrated arrangement of the components of the illumination optics 4, the second facet mirror 22 is arranged in a surface conjugate to the entrance pupil of the projection optics 10. The first facet mirror 20 is arranged tilted relative to the object plane 6. The first facet mirror 20 is arranged tilted relative to an arrangement plane defined by the deflection mirror 19. The first facet mirror 20 is arranged tilted relative to an arrangement plane defined by the second facet mirror 22.

2 shows a schematic flow diagram of an embodiment of the present method for manufacturing a lithography system 300 or one of its components 302 and/or for adapting a system parameter of the lithography system 300. The method is preferably computer-implemented, i.e., at least partially executable on a computer. The aging effect may include a radiation-induced degradation of optical elements 100 and/or a thermal and/or mechanical aging effect and/or an aging effect due to material deposition and/or corrosion and/or chemical degradation.

In a step S1, N copies 304 of a neural network 306 are generated to predict an aging effect of optical elements 100 (see, for example, 1 ), in particular the mirrors M1-M6, of lithography systems, where N is greater than 1 and preferably an integer. The optical elements can also be lenses or sensors. The neural network 306 for predicting the aging effect is preferably (pre-)trained prior to the generation S1 of the N copies 304 on the basis of historical and/or current and/or synthetically and/or simulatively generated training data on the aging effect. A layer structure and/or a layer-by-layer number of neurons of the neural network 306 is preferably selected and/or defined depending on the aging effect to be predicted.

In a step S2, T training data sets are provided. The T training data sets are generated by the following steps, each of which is carried out for each lithography system 300. The T training data sets are generated by the following steps for i = 1 to T, where T is greater than or equal to N. In a step S20, parameters are read out from an i _th lithography system 300. In a step S21, of an i _th training data set of the T training data sets depending on the read parameters.

In a step S3, each of the N copies is trained. Training is performed by performing the following steps for j = 1 to N. In a step S30, the j _-th copy of the N copies 304 is trained using the i- _th training data set of the T training data sets. In a step S31, a j _-th change information item 308 of the trained j _-th copy 304 is determined.

In a step S4, anonymized change information 310 is generated as a function of at least two or any subset of the N pieces of change information 308. The generation S4 of the anonymized change information 310 as a function of at least two of the N pieces of change information 308 preferably comprises weighting the at least two of the N pieces of change information 308 and summing the weighted at least two of the N pieces of change information 308 to form the anonymized change information 310. The weighting of the at least two of the N pieces of change information 308 is preferably carried out randomly or based on a performance parameter distribution of the trained N copies 304 of the neural network 306. The j _th pieces of change information 308 preferably each comprise aging-relevant information about plant-specific operating parameters and/or about a plant configuration and/or about a plant interface and/or about a plant history and/or about a plant make and/or about a plant type.

In a step S5, the neural network 306 is adapted depending on the anonymized change information 310.

In a step S6, an aging effect is predicted using the adapted neural network 312.

In a step S7, a lithography system or one of its components is manufactured depending on the predicted aging effect and/or a system parameter of a lithography system is adapted depending on the predicted aging effect. Manufacturing S7 may include retrofitting at least one component 302 of a lithography system 300 depending on the predicted aging effect in order to minimize the aging effect. Alternatively or additionally, adaptation S7 may include a step-by-step or incremental adjustment of a system parameter of a lithography system 300 depending on the predicted aging effect in order to minimize the aging effect. Minimizing the aging effect is preferably based on solving an inverse problem related to the aging effect and/or solving a feedback problem and/or solving an optimization problem.

Particularly preferably, the method steps S1 to S5 are carried out iteratively in order to continuously improve the model performance of the neural network 306, 312 for predicting the aging effect. This is indicated by a dashed feedback arrow in 2 indicated.

3 shows a schematic block diagram of an embodiment of the present system 3000 for manufacturing a lithography system or one of its components and/or for adjusting a system parameter of the lithography system. The system is configured to carry out the present method.

The system 3000 comprises a preferably central computing device 3002. The system 3000 further comprises N lithography systems 1, 300, each with an evaluation device 3004. The system 3000 further comprises a server or cloud 3006.

The computing device 3002 is designed to generate the N copies 304 of the neural network 306 for predicting an aging effect of optical elements 100 of lithography systems 1, 300, where N > 1, and to provide an i _th copy 304 of the N copies 304 to the evaluation device 3004 of each i _th lithography system 1, 300.

The respective evaluation device 3004 of each of the i _th lithography system 1, 300 is designed to read out parameters from the i _th lithography system 1, 300 and to generate an i _th training data set of the T training data sets depending on the read out parameters, to train the i _th copy 304 of the N copies 304 using the i _th training data set of the T training data sets, to determine the respective i _th change information 308 of the trained i _th copy 304, and to provide the determined i _th change information 308, in particular unidirectionally, to the server or the cloud 3006, so that a total of N change information 308 is provided.

The server or cloud 3006 is configured to generate the anonymized change information 310 depending on at least two of the N change information items 308 and to provide the anonymized change information 310 to the computing device 3002.

The computing device 3002 is designed to adapt or retrain the neural network 306 depending on the anonymized change information 310, to predict an aging effect using the adapted neural network 306, and to provide a control instruction for producing a lithography system 1, 300 or one of its components 302 depending on the predicted aging effect and/or for adapting a system parameter of a lithography system 1, 300 depending on the predicted aging effect.

Although the present invention has been described using exemplary embodiments, it can be modified in many ways.

LIST OF REFERENCE SYMBOLS

1

Lithography system

2

lighting system

3

light source

4

Lighting optics

5

Object field

6

Object level

7

Reticle

8

Reticle holder

9

Reticle displacement drive

10

Projection optics

11

Image field

12

Image plane

13

wafers

14

Wafer holder

15

Wafer relocation drive

16

Illumination radiation

17

collector

18

Intermediate focal plane

19

Deflecting mirror

20

first faceted mirror

21

first facet

22

second facet mirror

23

second facet

100

optical element

300

Lithography system

302

Components

304

Copy of a neural network

306

neural network

308

Change information

310

anonymized change information

312

customized neural network

3000

system

3002

computing device

3004

Evaluation device

3006

Server or Cloud

S1

Process step

S2

Process step

S3

Process step

S4

Process step

S5

Process step

S6

Process step

S7

Process step

S20

Process step

S21

Process step

S30

Process step

S31

Process step

M1

Mirror

M2

Mirror

M3

Mirror

M4

Mirror

M5

Mirror

M6

Mirror

QUOTES CONTAINED IN THE DESCRIPTION

This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• US 10,427,965 [0006]
• DE 10 2008 009 600 A1 [0077, 0081]
• US 2006/0132747 A1 [0079]
• EP 1 614 008 B1 [0079]
• US 6,573,978 [0079]
• DE 10 2017 220 586 A1 [0084]
• US 2018/0074303 A1 [0098]

Claims (13)

Method for producing a lithography system (1, 300) or one of its components (302) and/or for adapting a system parameter and/or for predicting a failure of the lithography system (1, 300), the method comprising: - generating (S1) N copies (304) of a neural network (306) for predicting an aging effect of optical elements (100) of lithography systems (1, 300), where N >1; - providing (S2) T training data sets which are generated by the steps, for i = 1 to T, where T ≥ N: a) reading out (S20) parameters from an i _th lithography system (1, 300), b) generating (S21) an i _th training data set depending on the read-out parameters; - Training (S3) the N copies, comprising the steps, for j = 1 to N: aa) training (S30) the j _th copy using the i _th training data set; bb) determining (S31) a j _th piece of change information of the trained j _th copy (304); - generating (S4) anonymized change information (310) depending on at least two of the N pieces of change information (308); - adapting (S5) the neural network (306) depending on the anonymized change information; - predicting (S6) an aging effect using the adapted neural network (312); and - producing (S7) a lithography system (1, 300) or one of its components (302) depending on the predicted aging effect and/or adapting a system parameter and/or prognosticating a failure of a lithography system (1, 300) depending on the predicted aging effect. Procedure according to Claim 1 , wherein the neural network (306) for predicting the aging effect is pre-trained on the basis of historical and/or current and/or synthetically and/or simulatively generated training data about the aging effect before generating the N copies (304). Procedure according to Claim 1 or 2 , wherein at least the method steps S1 to S5 are carried out iteratively in order to improve a model performance of the neural network (306, 312) for predicting the aging effect. Method according to one of the Claims 1 - 3 , wherein generating (S4) the anonymized change information (310) as a function of at least two of the N pieces of change information (308) comprises: weighting the at least two of the N pieces of change information (308); and summing the weighted at least two of the N pieces of change information (308) to form the anonymized change information (310). Procedure according to Claim 4 , wherein the weighting of the at least two of the N change information items (308) is carried out randomly or on the basis of a performance parameter distribution of the trained N copies (304) of the neural network (306). Method according to one of the Claims 1 - 5 , wherein the aging effect comprises a radiation-induced degradation of optical elements (100) and/or a thermal and/or mechanical aging effect and/or an aging effect due to material deposition and/or corrosion and/or chemical degradation. Method according to one of the Claims 1 - 6 , wherein the optical elements (100) comprise lenses and/or mirrors (M1-M6) and/or optical sensors. Procedure according to Claim 1 , wherein the production (S7) of a lithography system (1, 300) or one of its components (302) depending on the predicted aging effect and/or the adaptation (S7) of a system parameter of a lithography system (1, 300) depending on the predicted aging effect comprises: retrofitting at least one component (302) of a lithography system (1, 300) depending on the predicted aging effect in order to minimize the aging effect; and/or gradually or step-by-step adaptation of a system parameter of a lithography system (1, 300) depending on the predicted aging effect in order to minimize the aging effect. Procedure according to Claim 8 , wherein the minimization of the aging effect is based on solving an inverse problem related to the aging effect and/or solving a feedback problem and/or solving an optimization problem. Method according to one of the Claims 1 - 9 , wherein the generation (S4) of the anonymized change information (310) as a function of at least two of the N change information items (308) comprises, in particular, a respective local, unidirectional provision of the N change information items (308) to a server or a cloud (3006). Method according to one of the Claims 1 - 10 , wherein a layer structure and/or a layer-wise number of neurons of the neural network (306) is selected depending on the aging effect. Method according to one of the Claims 1 - 11 , wherein the j _th change information (308) each comprises aging-relevant information about plant-specific operating parameters and/or about a plant configuration and/or about a plant interface and/or about a plant history and/or about a plant make and/or about a plant type. System (3000) for producing a lithography system (1, 300) or one of its components (302) and/or for adapting a system parameter of the lithography system (1, 300), the system (3000) comprising: a, in particular central, computing device (3002), i lithography systems (1, 300) each having an evaluation device (3004), a server or a cloud (3006), wherein the computing device (3002) is designed to generate N copies (304) of a neural network (306) for predicting an aging effect of optical elements (100) of lithography systems (1, 300), wherein N > 1, and to provide an i _th copy (304) of the N copies (304) to the evaluation device (3004) of each i _th lithography system (1, 300), wherein the evaluation device (3004) of each of the i _th lithography system (1, 300) is designed to read out parameters from the i _th lithography system (1, 300) and to generate an i _th training data set depending on the read out parameters, to train the i _th copy (304) of the N copies (304) using the i _th training data set, to determine an i _th item of change information (308) of the trained i _th copy (304), and to provide the determined i _th item of change information (308) to the server or the cloud (3006), so that a total of N items of change information (308) are provided, wherein the server or the cloud (3006) is designed to generate anonymized change information (310) depending on at least two of the N items of change information (308), and to provide the anonymized change information (310) to the computing device (3002), wherein the computing device (3002) is designed to train the neural network (306) depending on the anonymized change information (310), to predict an aging effect using the adapted neural network (306), and to provide a control instruction for producing a lithography system (1, 300) or one of its components (302) depending on the predicted aging effect and/or for adapting a system parameter and/or for predicting a failure of a lithography system (1, 300) depending on the predicted aging effect.