Alignment method of charged particle beam device

Abstract

To provide a method with which alignment of a charged particle beam apparatus is improved, in particular in terms of ease of use, accuracy and / or a speed.SOLUTION: A method of aligning a charged particle beam apparatus M includes the steps of: providing a charged particle beam apparatus M in a first alignment state; using an alignment algorithm, by a processing unit P, for effectively achieving alignment transition from the first alignment state towards a second alignment state of the charged particle beam apparatus M; and providing data related to the alignment transition to a modification algorithm for modifying the alignment algorithm, in order to effectively achieve the modified alignment transition.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a method for aligning a charged particle beam device and a charged particle beam device having a processing unit for performing such a method.

Background Art

[0002] A charged particle beam device may in principle be any device that generates a beam, for example for illuminating an object of interest, using charged particles such as electrons, protons, and / or ions. The charged particle beam can be used for the study of samples, the inspection of objects, and / or the partial removal of the above-mentioned objects. Examples of applications include transmission electron microscopes, scanning electron microscopes, (plasma) focused ion beam technology, and lithography technology.

[0003] Many of these charged particle beam devices require alignment of the charged particle beam. For example, in a transmission electron microscope (TEM), the electron beam needs to be focused so that a parallel beam irradiates the object of interest and then the transmitted beam is focused onto a detector. In a scanning electron microscope, the electron beam is focused onto a small point on the object, the beam is scanned over the object, and the resulting secondary emission is detected to inspect the object of interest. Alignment of the charged particle beam device is necessary to obtain high-quality results.

Summary of the Invention

[0004] The alignment of charged particle beam devices is very difficult, and it has been found that advanced expertise of the operator is required to achieve it in a quick and reliable way. For example, in the case of a TEM, alignment involves setting up the gun and condenser system, selecting the aperture of the condenser, the height of the sample (the height of the eucentricity), accurately obtaining the shift and tilt rotation of the beam, accurately tilting the beam under the optical axis of the objective lens (the center of rotation and the aperture of the objective lens if necessary), and narrowing the focus of the diffraction lens, appropriately correcting the spherical aberration of each lens at each point, and one or more of these parameters and operations are included. Alignment in a TEM may require several repetitions to obtain the desired final result, which takes a lot of time and is prone to errors.

[0005] For this reason, an object of the present invention is to provide a method by which the alignment of a charged particle beam device is improved, particularly with respect to ease of use, accuracy, and / or speed. Further, an object of the present invention is to provide a charged particle beam device having the above-described quality.

[0006] For this reason, the present disclosure provides a method for aligning a charged particle beam device as defined in claim 1.

[0007] The method defined herein includes the step of providing a charged particle beam device in a first alignment state. The above-described first alignment state generally means, for example, a state in which a charged particle beam device such as a charged particle microscope is not aligned, that is, a state in which it is not (optimally) aligned to perform the intended use of the charged particle beam device.

[0008] The method defined herein includes the step of using an alignment algorithm to effectively achieve a transfer of alignment. The above-mentioned alignment algorithm may be executed by a processing unit which may be part of the charged particle beam apparatus or externally connected thereto. The alignment algorithm guides the charged particle beam apparatus from the above-mentioned first alignment state to a second alignment state. Generally, the second alignment state is an improved alignment state, i.e., a more (optimally) aligned state for carrying out the purpose of use of the above-mentioned charged particle beam apparatus, but this is not necessarily the case. A state with less alignment may also be considered. The transfer from the above-mentioned first alignment state to the above-mentioned second alignment state is defined herein as a transfer of alignment.

[0009] In the method defined herein, data related to the transfer of alignment may be provided to a correction algorithm. The above data includes, for example, settings, speeds, timings, and / or calibration results of the charged particle beam apparatus, but other data related to any alignment state and / or configuration of the charged particle beam apparatus is also conceivable. The data includes alignment states, alignment operations, and corresponding quality parameters.

[0010] The correction algorithm defined herein is adjusted to correct the above-mentioned alignment algorithm. The corrected alignment algorithm is adjusted to perform a corrected transfer of alignment starting from the above-mentioned first alignment state. The corrected transfer of alignment results in a second alignment state that is substantially the same or at least very similar as the alignment endpoint, but may differ in the intermediate steps until that second alignment state is reached. Alternatively, the corrected transfer of alignment may result in a corrected second alignment state that is completely different from the above-mentioned initial second alignment state. In the latter case, the correction algorithm is adjusted to provide a corrected alignment algorithm that results in a corrected second alignment that is improved with respect to the above-mentioned initial second alignment.

[0011] In this way, based on the data obtained during the alignment transition from the first alignment state to the second alignment state, by modifying the alignment algorithm using a correction algorithm, the alignment of the charged particle beam device can be performed in an improved manner, and as a result, an improved alignment algorithm can be obtained that has an improved second alignment state or enables a charged particle beam device with a more rapid alignment transition. Thereby, the objectives defined herein are achieved.

[0012] Advantageous embodiments of the present disclosure will be described below.

[0013] In one embodiment, the alignment algorithm is a trainable decision-making algorithm arranged to infer a series of alignment operations. The trainable decision-making algorithm, also referred to as an agent, can be used by a processing unit to infer the above-described series of alignment operations. In this embodiment, a series of inferred alignment operations are performed to bring the charged particle beam device into a second alignment state. The data related to the above-described alignment transition, i.e., the change in the alignment state, can be stored for further use, for example, by modifying at least one of the agents.

[0014] The alignment transition may include a series of individual alignment time steps. Data corresponding to a single alignment time step, such as at least one of the alignment state, alignment operation, and / or quality parameter, may be stored for further use of the stored data for the purpose of modifying the alignment algorithm.

[0015] In one embodiment, the above-described method steps correspond to a single iteration that is repeated at least once to align the charged particle beam device.

[0016] This method may include the step of modifying the sorting algorithm. The modification of the sorting algorithm may include using the data related to the above-mentioned sorting transition, that is, the data related to the sorted state. In addition, additional data can be used for the modification of the sorting algorithm.

[0017] The modification in this case may be understood as changing at least one of the hidden state, activation, trainable parameters, or the structure of the machine learning model that underlies the decision-making mechanism of the sorting algorithm. In the first two cases, it means that the agent can infer the sorting operation using an iterative or autoregressive model. Such models may be used to perform sorting where the inference of the operation requires not a single observation but a history of past observations. However, the modification of the trainable weights and the structure of the machine learning model can be performed offline, that is, it means that the sorting is performed by the agent and the modification is made after the completion of the sorting rather than during the sorting.

[0018] In addition, modifying the sorting algorithm or the agent may include the step of providing a sorting algorithm such as an agent, the step of providing at least the data of the sorting transition executed by the above-mentioned sorting algorithm, the step of determining the necessary modification of the above-mentioned sorting algorithm, using the data of the sorting transition, and the step of modifying the sorting algorithm based on the determined modification.

[0019] In one embodiment, the method includes the further step of using the modified alignment algorithm described above to effectively achieve subsequent alignment transitions. The subsequent alignment transitions may be used to bring the charged particle beam apparatus from the second alignment state described above to a third alignment state. The third alignment state may have improved alignment characteristics compared to the second alignment state described above. In practice, the method defined herein is at least partially repeated to bring the charged particle beam apparatus to an even more improved alignment state. To effectively achieve further modified alignment transitions, the data related to the subsequent alignment transitions described above may be provided to the modification algorithm described above to modify the modified alignment algorithm described above. Thereby, if desired, the alignment algorithm can be checked and improved as necessary after each alignment procedure.

[0020] The alignment transitions may include a plurality of individual alignment operations. This generally applies to any alignment transition, such as the alignment transitions, modified alignment transitions, subsequent alignment transitions, etc. referred to herein. Each of these alignment transitions may include a series of individual alignment steps referred to herein as individual alignment operations. These individual alignment operations may include, for example, setting the intensity of components of the charged particle beam apparatus, setting the position of the object of interest, setting beam characteristics such as beam shift and / or beam tilt, setting the focus of one or more lenses, correcting the spherical aberration of one or more lenses, and including one or more of the selection of the order of these individual operations and settings.

[0021] In one embodiment, the method includes the step of determining a quality parameter. The quality parameter may relate to a final alignment state, such as the second alignment state, the third alignment state, etc. The quality parameter may be related to an alignment transition or, alternatively, to one or more of the individual alignment operations therein. It is contemplated that one or more quality parameters are determined for a single alignment transition. Further, it is also contemplated that one or more quality parameters are determined for a single individual alignment operation.

[0022] In one embodiment, the above-mentioned step of determining the above-mentioned quality parameters includes evaluating one or more of the above-mentioned plurality of individual alignment operations. Thereby, it becomes possible to evaluate individual alignment operations and associate respective quality parameters with them. The correction algorithm may consider these individual quality parameters when correcting the alignment algorithm.

[0023] The above-mentioned evaluation may include identifying individual alignment operations that result in desirable and / or undesirable alignment states. For example, a reward may be given in return for an individual alignment operation that led to an improvement in alignment and it may be promoted with the corrected alignment algorithm. However, an individual alignment operation that caused a deterioration in alignment may be penalized and blocked with the corrected alignment algorithm. By evaluating individual alignment operations, for example, operations that deviate from a desirable alignment or slow operations can be identified and prevented from being used with the corrected alignment algorithm. Therefore, the method may include the step of correcting the above-mentioned alignment algorithm so that undesirable alignment operations are prevented during the use of the above-mentioned corrected alignment algorithm.

[0024] In one embodiment, the alignment algorithm includes at least one neural network. As is known to those skilled in the art, a neural network (NN), also called an artificial neural network (ANN) or a simulated neural network (SNN), is a group of interconnected artificial neurons that uses a mathematical or computational model for information processing based on a connectionist approach to computing. The artificial neural network as defined herein is an adaptive system that changes its structure based on external or internal information flowing through the network. In more practical terms, a neural network is a non-linear statistical data modeling and decision-making tool that can be used to model complex relationships between inputs and outputs or to find patterns in data. Learning with a neural network is particularly effective in application methods where it is impossible to manually design functions due to the complexity of the data and tasks. Thus, the application of neural networks is advantageous for optimizing the alignment procedures of any charged particle beam device.

[0025] In particular, when the above-described machine learning model is a neural network, the above-described procedure for determining the correction may be performed by calculating a loss function and subsequently calculating the gradient of the loss function with respect to the trainable weights of the neural network.

[0026] The above-described correction of the agent may be performed by updating the trainable weights of the neural network using a traditional optimizer (such as ADAM) based on the calculated gradient.

[0027] Alternatively, the correction rules used in the above-described correction procedure may not be given in terms of mathematical formulas (as used in traditional neural network optimizers), but may be learned from the data of the metaloop of the training procedure.

[0028] In an embodiment, the method may include providing stored data obtained during the transfer of alignment. This enables, for example, the use of on-policy, off-policy, and offline reinforcement learning algorithms.

[0029] The modified algorithm defined herein may be a training algorithm for training a neural network. In this sense, the modified algorithm may be part of the neural network or an external algorithm that provides the inputs necessary for training the neural network. Data related to the transfer of alignment may be used as an input for training the neural network. Here, quality parameters related to the transfer of alignment and / or individual alignment operations may be used as a penalty / reward mechanism for training the neural network. Retraining at least one of the neural networks described above may use one or more of the undesirable alignment states described above as input parameters. In this regard, the neural network may be trained by deep reinforcement learning, which focuses on balancing exploration (of unknown regions) and exploitation (of current knowledge). Unlike other types of learning, such as supervised learning and unsupervised learning, there is no need to present pairs of labeled input / output, nor is there a need to explicitly correct sub-optimal behavior.

[0030] Thus, in one embodiment, the above step of modifying the above algorithm includes the step of training or retraining at least one of the above neural networks, particularly using deep reinforcement learning.

[0031] The charged particle beam apparatus is set in an undesirable alignment state, and it is conceivable to use a modified alignment algorithm to effectively achieve further transitions. By effectively achieving further transitions, it can then be mainly used to collect additional data, and that data can be used to modify the (previously modified) alignment algorithm again. Further, the undesirable alignment state may also be related to an alignment state in which it has been found that the alignment algorithm is performed sub-optimally.

[0032] In one embodiment, a training algorithm is provided, and the training algorithm is used to identify and / or set an undesirable alignment state. With this training algorithm, it is possible to quickly and effectively identify situations and alignment states that the alignment algorithm has difficulty with, and use those difficult situations as they are to test the alignment algorithm, collect data related to alignment transitions, or use a modified algorithm to improve the alignment algorithm. That is, a two-step approach is adopted in which a training algorithm is used to identify a difficult alignment state, and an alignment algorithm is used to effectively align the charged particle beam apparatus starting from that difficult alignment state. When using a neural network for training and a neural network for alignment, by using them to reinforce each other, an accurate and fast alignment neural network can be obtained quickly and effectively.

[0033] According to one aspect, a method for training an alignment algorithm for aligning a charged particle beam apparatus is provided, providing an alignment algorithm to be trained; providing data related to the transition of the alignment of the charged particle beam apparatus, wherein the transition of the alignment extends from a first alignment state to a second alignment state; A step of providing the data to a correction algorithm for correcting the above-described sorting algorithm, including a step of using a processing unit to provide the data.

[0034] The above method of training the above sorting algorithm may be executed, for example, inside the charged particle beam device, such as inside the processing unit of the charged particle beam device. Additionally, or alternatively, at least a part of the above training method may be performed externally. For example, the method of training the sorting algorithm may be executed on the cloud or on a stand-alone computer device. In that case, the charged particle beam device may transfer the above data related to the alignment transfer to the cloud or the computer, and at that time, the next step of providing the data to the correction algorithm may be performed.

[0035] In an embodiment, the method may include a step of providing stored data acquired during the alignment transfer. Thereby, for example, on-policy, off-policy, and offline reinforcement learning algorithms can be used.

[0036] The method of training the sorting algorithm may include one or more embodiments as described herein, where relevant.

[0037] The method may include, for example, a step of determining at least one quality parameter of the above-described stored data, and correcting the above-described sorting algorithm based on the above at least one quality parameter.

[0038] In one embodiment, the method may include providing stored data related to the transfer of multiple alignments of multiple charged particle beam devices. This enables data to be collected from multiple charged particle beam devices and used to modify a single alignment algorithm based on that data. This embodiment is particularly effective when performing the method of training the alignment algorithm described above on a cloud or stand-alone computer. This allows a vast amount of data to be provided to a single training algorithm, enabling effective training of the algorithm.

[0039] In an embodiment where the method is executed outside the charged particle beam device, the modified alignment algorithm may be provided back to each charged particle beam device as an update to the alignment algorithm. The modified alignment algorithm can also be provided to multiple charged particle beam devices as an update to each alignment algorithm. The modified alignment algorithm may include modifications specific to one of the multiple charged particle beam devices. Thus, the method defined herein may include providing multiple modified alignment algorithms.

[0040] Note that the charged particle beam device may be a physical device or a virtual device. For example, a virtual device such as a so-called digital twin can be advantageously used to test the modified alignment algorithm and / or generate a vast amount of alignment transfer data that can be used to modify the alignment algorithm, for example, by training the alignment neural network described above.

[0041] According to one aspect, there is provided a charged particle beam device including a processing unit arranged to execute at least a part of the method defined herein.

[0042] According to a further aspect, a misalignment method for a charged particle beam device is provided, similar to the methods as disclosed herein. According to this aspect, a charged particle beam device is provided in a first alignment state, and a processing unit uses a misalignment algorithm to effectively achieve a transition of alignment of the charged particle beam device from the first alignment state to a second alignment state, and the second alignment state is less optimal compared to the first alignment state.

[0043] In one embodiment, the alignment method defined herein can be used to make the charged particle beam device in a more aligned state.

[0044] According to one aspect, the above alignment algorithm and the above misalignment algorithm are used for so-called curriculum training to improve at least one of the above agents, i.e., at least one of the alignment algorithm and the misalignment algorithm. According to this aspect, a method for curriculum training is provided, as follows providing at least one charged particle beam device; using a misalignment method, such as defined herein, to misalign the at least one charged particle beam device, using at least a first agent; using an alignment method, such as defined herein, to align the at least one charged particle beam device, using at least a second agent; determining at least one quality parameter related to at least a part of the alignment transition obtained during the misalignment and / or alignment transition; modifying at least one of the first agent or the second agent based on the determined at least one quality parameter; corresponds to a single iteration that is executed at least once to improve at least one of the agents involved in the above curriculum training.

[0045] In the above step of modifying the above-described first agent or second agent, additional data may be used.

[0046] In the curriculum training method, two populations (also referred to as teams) of agents (algorithms) may be maintained. The first team of agents (also referred to as the alignment team) receives training for the alignment of the charged particle beam device, and the second team of agents (also referred to as the misalignment team) receives training for the misalignment of the charged particle beam device. The curriculum training method is a zero-sum game. In each iteration, the agents of the alignment team perform the alignment of the charged particle beam device. A quality parameter related to the performed alignment is calculated. The agents of the misalignment team put the charged particle beam device in a state where it is difficult for the agents of the alignment team to perform the alignment. The agents may be modified according to the following rules. The value of the utility function of the agents of the alignment team is calculated based on the quality parameter related to the performed alignment. The value of the utility function of the agents of the misalignment team is the negative value of the utility function of the agents of the alignment team. Thus, the agents are trained in an adversarial manner. By competing between the two agent teams, the quality of the actions performed by the agents is gradually improved.

[0047] The procedure for modifying the agents may be formulated as a reinforcement learning task. In this case, the quality parameter calculated in relation to the performed alignment corresponds to the reward provided to the agents. The trainable parameters of the neural network of the agents are updated such that the probability of a qualitatively excellent (high-reward) action is increased and the probability of a defective (low-reward) action is decreased.

[0048] The value of the utility function of the agents of the misalignment team may be calculated based on the quality parameter related to the misalignment action performed by the agents. In this case, the curriculum training method is no longer a zero-sum game.

[0049] The misaligned team of agents may be substituted with a hard-coded algorithm that does not include a trainable machine learning model. In this case, the curriculum training method is reduced to a single-team population-based training task. If the population includes a single agent, the curriculum training method is further reduced to the traditional reinforcement learning formulation of the training procedure for a single agent.

[0050] The population may include agents with the same neural network architecture and agents with different architectures. Each agent has its own set of trainable weights. Thus, each agent devises its own strategy (sequence of actions) to align or misalign the charged particle beam device.

[0051] The best-performing trained agent can be selected and further utilized at the end of the curriculum training procedure to align the user's charged particle beam device.

[0052] Embodiments of the present disclosure provide a method for aligning a charged particle beam device using a trainable sequential decision-making algorithm (agent). The population of agents described above is trained according to an iterative curriculum, gradually improving the quality of the actions inferred by the agents to perform the alignment. The best-performing trained agents enter the introduction stage. Further, the agents can initiate a retraining procedure to adapt to the particularities of the user's charged particle beam device.

[0053] As described herein, the data may include the provided data and the stored data.

[0054] The data provided to the above-mentioned correction algorithm may include alignment trajectories depicted by other agents different from the agent provided to the correction algorithm, which were depicted by an initial version of the provided agent (i.e., algorithm), or were depicted by a demonstrator algorithm. The alignment trajectories may be depicted by interactions with a plurality of different charged particle beam devices. The above-mentioned previous version of the agent is the state before one or more corrections are executed on the agent by the correction algorithm. The above-mentioned demonstrator algorithm is an algorithm that can infer a (potentially good) alignment operation different from the alignment operation inferred by the current version of the provided agent. The above-mentioned demonstrator algorithm may be an agent that utilizes a machine learning model for operation inference, or alternatively, may be a hard-coded algorithm that utilizes the knowledge of a knowledgeable person to perform alignment.

[0055] The data provided may include demonstrations by human experts. The above-mentioned human experts are those who perform the alignment of the charged particle beam device and record and store its operation. In this case, the alignment trajectories of the human experts can be utilized in the training process to incorporate human knowledge into the machine learning model that forms the basis of the agent's inference mechanism.

[0056] Also, the data provided to the above-mentioned correction algorithm may further include data generated by an external algorithm for artificial data generation. The above-mentioned external algorithm for artificial data generation is an algorithm that can generate data without physical interaction with the charged particle beam device.

[0057] The data provided may further include quality parameters corresponding to the alignment state and / or alignment operation. In particular, the above-mentioned quality parameters may be treated as a reward provided to the agent in response to the alignment operation inferred by the agent during alignment.

[0058] The structure of the provided data can determine the structure of the correction algorithm. The correction algorithm is a supervised learning task if the provided data is labeled, an unsupervised learning task if there is no label corresponding to the data, an active learning task if the provided data is partially labeled (i.e., the agent requires the human to label only the samples that cause uncertainty during action inference rather than all data samples), and may be organized as a self-supervised learning task if the agent is trained on a downstream task. If the provided data includes quality parameters, the correction algorithm may be organized as a reinforcement learning task. The above quality parameters may be treated as a reward provided to the agent according to the alignment operation. If the provided data includes the alignment trajectory depicted by the current version of the agent, an on-policy reinforcement learning algorithm may be used to train the agent. An off-policy reinforcement learning algorithm may be used if the provided alignment trajectory was depicted by a previous version of the agent. If the provided dataset is static (i.e., the agent is not allowed to interact with the environment to collect more training data), the agent may be trained by offline reinforcement learning.

[0059] Next, the present invention will be explained in more detail based on exemplary embodiments and the accompanying schematic drawings.

Brief Description of the Drawings

[0060]

Figure 1

Figure 2

Figure 3a

Figure 3b

Figure 3c

Figure 4

DETAILED DESCRIPTION OF THE INVENTION

[0061] FIG. 1 (not to scale) is a high-level schematic diagram of an embodiment of a charged particle microscope M according to an embodiment of the present invention. More specifically, it shows an embodiment of a transmission microscope M, in this case a TEM / STEM (however, in the context of the present invention it is equally effective for, for example, an SEM (see FIG. 2) or an ion-based microscope). In FIG. 1, an electron source 4 within a vacuum enclosure 2 propagates along an electron optical axis B', traverses an electron optical illuminator 6, and generates an electron beam B that serves to direct / focus electrons towards a selected portion of a sample S (which can be, for example, (locally) thinned or planarized). Also depicted is a deflector 8 that can be used (inter alia) to effect a scanning operation of the beam B.

[0062] The sample S is held in a sample holder H that is positionable in multiple degrees of freedom by a positioning device / stage A that moves a cradle A' to which a holder H is (removably) fixed. For example, the sample holder H may include (inter alia) fingers that are movable (in the XY plane, in the illustrated Cartesian coordinate system, typically also with possible tilting motions parallel to Z and centered on X / Y). Such motions enable different portions of the sample S to be irradiated / imaged / inspected by an electron beam B that moves along the axis B' (Z direction) (and / or enable a scanning operation to be performed instead of beam scanning). If desired, an optional cooling device (not shown) can be brought into close thermal contact with the sample holder H to maintain it (and the sample S thereon) at, for example, cryogenic temperatures.

[0063] The electron beam B interacts with the sample S in such a way as to diverge various types of "stimulated" radiation from the sample S, such as (for example) secondary electrons, backscattered electrons, X-rays, optical radiation (cathodoluminescence). If desired, one or more of these radiation types can be detected using an analysis device 22, which can be, for example, a coupled scintillator / photomultiplier tube, or an EDX or EDS (energy-dispersive X-ray spectroscopy) module, and in such cases, an image can be constructed using basically the same principle as in an SEM. However, alternatively or additionally, electrons that cross (pass through) the sample S and emerge from it and continue to propagate (substantially along axis B', albeit with some deflection / scattering) can be studied. Such a transmitted electron beam generally enters an imaging system (projection lens) 24 that includes various electrostatic / magnetic lenses, deflectors, correctors (such as aberration correctors). In the normal (non-scanning) TEM mode, this imaging system 24 can focus the transmitted electron beam onto a fluorescent screen 26, which can be retracted / stored (as schematically indicated by arrow 26') so as not to obstruct axis B' if desired. An (partial) image (or diffraction pattern) of the sample S is formed on the screen 26 by the imaging system 24 and can be viewed through a viewing port 28 located in an appropriate part of the wall of the housing 2. The retraction mechanism of the screen 26 can be, for example, mechanical and / or electrical in nature and is not depicted here.

[0064] Instead of viewing the image on the screen 26, the fact that the depth of focus of the electron beam emerging from the imaging system 24 is generally very large (for example, on the order of 1 meter) can be utilized. As a result, various other types of analytical devices can be used downstream of the screen 26, as follows. TEM camera 30. In camera 30, the electron beam can be processed by controller / processor 20 to form a still image (or diffraction pattern) that can be displayed on a display device 14 such as a flat panel display. When not needed, camera 30 can be stored / retracted (as schematically shown by arrow 30') so as not to interfere with axis B'. STEM camera 32. The output from camera 32 can be recorded as a function of the (X, Y) scan position of beam B on sample S, and an image can be constructed that "maps" the output from camera 32 as a function of X and Y. Camera 32 can include, for example, a single pixel that is 20 mm in diameter, as opposed to the matrix of pixels typically found in camera 30, although camera 32 can also be an electron microscope pixel array detector (EMPAD). Further, camera 32 generally has a much higher acquisition rate (e.g., 10 2 points / second) than camera 30 (e.g., 10 6 images / second). Again, when not needed, camera 32 can be stored / retracted (as schematically shown by arrow 32') so as not to interfere with axis B' (although such storage may not be necessary in the case of, for example, a doughnut-shaped annular dark field camera 32, in which a central hole allows the beam to pass through when the camera is not in use). As an alternative to imaging using camera 30 or 32, a spectroscopic device 34, which can be, for example, an EELS module, can also be invoked.

[0065] Note that the order and position of items 30, 32, 34 are not strict, and various possible variations are conceivable. For example, spectroscopic device 34 can also be incorporated into imaging system 24.

[0066] In the illustrated embodiment, microscope M further includes a stowable X-ray computed tomography (CT) module, indicated generally by reference numeral 40. In computed tomography (also referred to as tomography), a sample is penetrated with different lines of sight using a radiation source and a (opposite) detector, and the sample is observed transmissively from various angles.

[0067] Note that the controller (computer processor) 20 is connected to various illustrated components via a control line (bus) 20'. This controller 20 can provide various functions such as synchronization of operations, provision of set points, signal processing, execution of calculations, and display of messages / information to a display device (not shown). Needless to say, the (schematically drawn) controller 20 may be inside or outside the housing 2 (partially), and may have an integrated or composite structure as desired. The controller includes a data processing device P arranged to perform the methods defined herein, as shown in this embodiment.

[0068] Those skilled in the art will understand that it is not necessary to keep the inside of the housing 2 in a strict vacuum. For example, in so-called "environmental TEM / STEM", a background atmosphere of a predetermined gas can be intentionally introduced / maintained inside the housing 2. Also, those skilled in the art will understand that in practice, the volume of the housing 2 takes the form of a small tube (e.g., about 1 cm in diameter) essentially along the axis B' as much as possible, but it is advantageous to limit it to expand to accommodate structures such as the line source 4, the sample holder H, the screen 26, the camera 30, the camera 32, and the spectrometer 34.

[0069] Next, referring to FIG. 2, another embodiment of the device according to the present invention is shown. FIG. 2 (not to scale) is a highly schematic diagram of a charged particle microscope M according to the present invention, and more specifically, shows an embodiment of a non-transmission type microscope M, in this case, an SEM (however, in the context of the present invention, for example, an ion-based microscope is equally effective). In the figure, parts corresponding to the items in FIG. 1 are denoted by the same reference numerals and will not be described separately here. Added to FIG. 1 are (inter alia) the following parts. 2a: A vacuum port that is opened to introduce / remove an item (component, sample) into / from the interior of the vacuum chamber 2, or, for example, an auxiliary device / module is attached thereto. Note that the microscope M may include a plurality of such ports 2a if desired. 10a, 10b: Schematic depictions of the lens / optical element of the illuminator 6. 12: A voltage source for biasing (floating) the sample holder H, or at least the sample S, to the potential with respect to the ground if desired. 14: A display such as an FPD or a CRT. 22a, 22b: A segmented electron detector 22a including a plurality of independent detection segments (e.g., quadrants) arranged around a central opening 22b through which the beam B can pass. Such a detector can be used, for example, to examine the output (secondary or backscattered) electron beam (angle dependence) coming out of the sample S.

[0070] Here too, there is a controller 20. The controller is connected to the display 14, and the display 14 may be connectable to a data processing device P arranged to perform the method defined in this specification. In the illustrated embodiment, the data processing device P has a different structure that does not form part of the controller and does not constitute part of the microscope P. The data processing device P may be local or in the cloud and, in principle, location is not chosen. Note that in all embodiments described in this specification, the data processing unit P can be part of a charged particle beam device such as a charged particle microscope or can be arranged externally.

[0071] Referring now to FIG. 3a, an embodiment of the method defined herein is schematically shown. The method includes step 101 of providing a charged particle beam device in a first alignment state, a step of using an alignment algorithm 102 by a processing unit P to effectively achieve a transition of alignment from the first alignment state to a second alignment state of the charged particle beam device, and step 103 of providing data related to the transition of alignment to a correction algorithm to effectively achieve a corrected transition of alignment. In the illustrated embodiment, the method includes step 104 of modifying the alignment algorithm by the processing unit P to obtain a corrected alignment algorithm.

[0072] In one embodiment, step 102 of using the alignment algorithm by the processing unit P can be performed by the processing unit P which is part of the charged particle beam device. Step 103 of providing data related to the above-described transition of alignment to the correction algorithm may include steps of being performed at least partly within the charged particle beam device and providing data outside the charged particle beam device. Step 104 of modifying the above-described alignment algorithm may be performed within the charged particle beam device, or may be performed on a cloud or an external computer device.

[0073] Referring now to FIG. 3b, a further embodiment of the method is shown. This method is equivalent to the embodiment shown in FIG. 3a, but here, step 105 of feeding back the corrected alignment algorithm (obtained in step 104) to the charged particle beam device and step 102 of using the (corrected) alignment algorithm to establish a further transition of alignment are added.

[0074] Next, referring to FIG. 3c, a further embodiment of this method is shown. This method is equivalent to the embodiment shown in FIG. 3a. However, here it is shown that step 102 of using the alignment algorithm includes sub-steps 102a-102C that provide a plurality of individual alignment operations. Here, a total of three individual alignment operations 102a-102C are shown, but it will be apparent to those skilled in the art that in principle any number of individual alignment operations are conceivable.

[0075] In all embodiments shown in FIGS. 3a-3c, the method may include a step of determining a quality parameter. The above-mentioned step of determining the above-mentioned quality parameter includes, in the embodiment shown in FIG. 3c, a step of evaluating one or more of the above-mentioned plurality of individual alignment operations. The above-mentioned evaluation may include identifying individual alignment operations that result in an undesirable alignment state. The method may include a step of modifying the above-mentioned alignment algorithm so that undesirable alignment operations are prevented during the use of the above-mentioned modified alignment algorithm.

[0076] As described above, the alignment algorithm may include at least one neural network, and step 104 of modifying the above-mentioned algorithm includes a step of retraining the at least one neural network. This may include a step of retraining the at least one neural network using one or more of the above-mentioned undesirable alignment states.

[0077] In the embodiment shown, the step of providing a charged particle beam device may include a step of setting the charged particle beam device in an undesirable alignment state. Thereafter, an alignment algorithm (whether modified or not) may be used to effectively achieve an alignment transition or a further alignment transition. It is conceivable to use a training algorithm to identify and / or set an undesirable alignment state. The training algorithm may include a neural network.

[0078] Referring to FIG. 4 here, the system 201 as defined in this specification is shown. The system 201 includes a charged particle beam apparatus CPBA such as a charged particle microscope M, and has an alignment algorithm AA for the alignment of the CPBA. The CPBA is connected to a correction algorithm MA. This connection may be local like the device itself or external like a wireless connection to a cloud-based correction algorithm MA. The correction algorithm may correct the alignment algorithm AA of the CPBA based on the data received by the CPBA and return the corrected algorithm to the CPBA. The training algorithm TA may be connected to the correction algorithm MA and / or the CPBA. Based on the received data, the training algorithm TA is designed to select and / or set an alignment state that is relatively difficult for the alignment algorithm to solve. Then, further alignment transitions may be performed, and subsequent data may be used again to check whether further corrections are needed.

[0079] The CPBA as shown in FIG. 4 may be a system that combines multiple CPBAs. It is conceivable that multiple CPBAs, each having an (arbitrary) alignment algorithm AA, are used to provide data to the correction algorithm MA. That data can be used to update or correct the existing alignment algorithm AA for one or more CPBAs.

[0080] The CPBA shown in FIG. 4 may be a physical device or a virtual device such as a digital twin. The CPBA as shown in FIG. 4 may include at least one physical CPBA and at least one virtual device. The at least one virtual device may be used for the rapid training and / or correction of the alignment algorithm. The alignment algorithm and / or the training algorithm may be a neural network. In particular, the alignment algorithm is corrected based on deep reinforcement learning.

[0081] From the description of the above figures, the method defined in this specification relates to training an alignment algorithm for the alignment of a charged particle beam apparatus, providing an alignment algorithm AA to be trained; providing data related to the transition of the alignment of the charged particle beam apparatus, the transition of the alignment extending from a first alignment state to a second alignment state; providing the data to a correction algorithm MA for correcting the above alignment algorithm, the processing unit P being used, is clearly included.

[0082] From the description of the above figures, it is clear that a charged particle beam apparatus such as a charged particle microscope is arranged to perform part of the method described in this specification.

[0083] Desired protection is conferred by the appended claims.

Claims

A method for aligning a charged particle beam device, comprising: - providing a charged particle beam device in a first alignment state; - using, by a processing unit, an alignment algorithm to effectively achieve a transition of alignment from the first alignment state to a second alignment state of the charged particle beam device, wherein the first alignment state and the second alignment state describe a plurality of parameters of a plurality of components including a plurality of lenses of the charged particle beam device, and the alignment algorithm includes at least one neural network; - providing data related to the alignment transition to a correction algorithm for correcting the alignment algorithm to effectively achieve a corrected alignment transition, wherein correcting the alignment algorithm comprises retraining the at least one neural network using one or more undesirable alignment states; A method as described above. The method according to claim 1, further comprising setting the charged particle beam device to an undesirable alignment state and using the corrected alignment algorithm to effectively achieve a further transition.

3. The method according to claim 1 or 2, further comprising using the corrected alignment algorithm to effectively achieve a subsequent alignment transition.

4. The method according to any one of claims 1 to 3, wherein the alignment transition includes a plurality of individual alignment operations.

5. The method according to any one of claims 1 to 4, further comprising determining a quality parameter.

6. The method according to claim 5, wherein the step of determining the quality parameter includes evaluating one or more of the plurality of individual alignment operations included in the alignment transition.

7. The method according to claim 6, wherein the evaluation includes identifying an individual alignment operation that results in an undesirable alignment state.

8. The method according to claim 7, further comprising correcting the alignment algorithm such that an undesirable alignment operation is prevented during use of the corrected alignment algorithm.

9. The method according to any one of claims 1 to 8, using a training algorithm for identifying and / or setting an undesirable alignment state.

10. The method according to claim 9, wherein the training algorithm includes a neural network.

11. A method for training an alignment algorithm for the alignment of a charged particle beam device, comprising: - providing an alignment algorithm to be trained, wherein the alignment algorithm includes at least one neural network; - providing data related to the transition of the alignment of the charged particle beam device, wherein the transition of the alignment extends from a first alignment state to a second alignment state, and the first alignment state and the second alignment state describe a plurality of parameters of a plurality of components including a plurality of lenses of the charged particle beam device; - providing the data to a correction algorithm for correcting the alignment algorithm, wherein correcting the alignment algorithm comprises retraining the at least one neural network using one or more undesirable alignment states, and using a processing unit to provide the data; A method comprising:

12. The method according to claim 11, comprising setting the charged particle beam device to an undesirable alignment state and using the corrected alignment algorithm to effectively achieve further transitions.

13. The method according to claim 11, comprising determining at least one quality parameter of the stored data and correcting the alignment algorithm based on the at least one quality parameter.

14. The method according to claim 11 or 13, comprising providing stored data related to the transitions of the alignments of a plurality of charged particle beam devices.

15. The method according to any one of claims 1 to 14, wherein the charged particle beam device is a virtual device.

16. A charged particle beam device comprising a processing unit arranged to execute the method according to any one of claims 1 to 15.

Images