Method and program for operating semiconductor device evaluating apparatus using character strings

Abstract

The objective of the present disclosure is to provide a technique with which it is possible to obtain instructions for an evaluating apparatus that evaluates semiconductor devices from a language model by means of character strings input into the language model. A method according to the present disclosure involves accepting specification data describing operational specifications of a semiconductor device evaluating apparatus from a language model by inputting instructions for the semiconductor device evaluating apparatus into the language model. The language model is fine-tuned in advance by means of at least one of domain knowledge relating to of the semiconductor device and domain knowledge relating to the evaluating apparatus.

Description

Method and program for operating semiconductor device evaluation equipment using character strings

[0001] The present disclosure relates to a technique for operating a semiconductor evaluation device using a character string.

[0002] In recent years, there has been active development of technology that uses a large language model (LLM) to receive answers to questions written in natural language from an artificial intelligence (AI) in natural language. Non-Patent Document 1 below describes the GPT model developed by OpenAI as one example.

[0003] Long Ouyang et al., “Training language models to follow instructions with human feedback”, 2022, URL: https: / / arxiv.org / abs / 2203.02155

[0004] A general-purpose large-scale language model such as that described in Non-Patent Document 1 learns the frequency of occurrence of character strings by, for example, repeatedly masking part of a character string and estimating the masked part for various document data. This allows knowledge in various fields (sometimes called domain knowledge) to be learned in advance using character strings, and when a question related to that field is given, an answer can be output based on the previously learned knowledge.

[0005] Although such large-scale language models can learn knowledge in various fields, they may not have sufficient knowledge in a specific field. Furthermore, while such large-scale language models are useful in the process of providing a question as input and obtaining an answer to that question, they are not necessarily optimized for applications such as obtaining operating instructions for a specific device. The technology described in Non-Patent Document 1 describes a method for training a large-scale language model, but does not particularly consider specific methods for such applications.

[0006] The present disclosure has been made in consideration of the above-described problems, and aims to provide a technology that enables instructions to an evaluation device that evaluates semiconductor devices to be obtained from a language model by inputting a string of characters into the language model.

[0007] In a method according to the present disclosure, specification data describing operational specifications of an evaluation device that evaluates a semiconductor device is received from a language model by inputting instructions for the evaluation device to the language model, and the language model is fine-tuned in advance using at least one of domain knowledge of semiconductor devices and domain knowledge of the evaluation device.

[0008] In another aspect of the present invention, specification data describing the operational specifications of a first evaluation apparatus for evaluating semiconductor devices and instructions for a second evaluation apparatus are input to a language model, whereby the specification data describing the operational specifications of the second evaluation apparatus are received from the language model, and the language model is fine-tuned by comparison information between output information of the first evaluation apparatus and output information of the second evaluation apparatus.

[0009] In still another aspect for achieving the above object, information on the layout of a semiconductor device and information instructing prediction of position information of defects occurring on the semiconductor device are input to a language model, and data on defect positions is received from the language model. The language model is fine-tuned by comparison information between the data on defect positions output from the language model and data on correct positions of the defects.

[0010] In another aspect of the present invention, an instruction to investigate an anomaly in a semiconductor device manufacturing process is input to a language model, whereby data on the cause of the anomaly is received. The language model is fine-tuned using sentence data on the cause of the anomaly and related data that serves as evidence of the cause of the anomaly.

[0011] According to the method of the present disclosure, instructions for an evaluation device that evaluates semiconductor devices can be obtained from a language model by inputting a string of characters into the language model. Other problems, configurations, advantages, etc. of the present disclosure will become clear from the description of the following embodiments.

[0012] 1 is a block diagram of a semiconductor evaluation system 100 according to a first embodiment. FIG. 2 is a diagram illustrating an example of a text creation system 200 that uses a general-purpose LLM 210 configured based on a large-scale language model. FIG. 3 is a diagram illustrating an example of domain knowledge learned by fine tuning. FIG. 4 is a diagram illustrating the distinction between unsupervised learning and supervised learning. FIG. 5 is a schematic diagram illustrating a process in which a learning module 113 performs unsupervised learning. FIG. 6 is a schematic diagram illustrating another process in which a learning module 113 performs unsupervised learning. FIG. 7 is a schematic diagram illustrating a process in which a learning module 113 performs supervised learning. FIG. 8 is a schematic diagram illustrating variations in combinations of input / output / correct answer / loss function in supervised learning. FIG. 9 is a schematic diagram illustrating a process of reinforcement learning performed by a reinforcement learning module 114. FIG. 10 is a schematic diagram illustrating an example of a re-learning procedure by the reinforcement learning module 114. FIG. 11 is a schematic diagram illustrating another example of a re-learning procedure by the reinforcement learning module 114. FIG. 12 is a schematic diagram illustrating another example of a re-learning procedure by the reinforcement learning module 114. FIG. 13 is a schematic diagram illustrating another example of a re-learning procedure by the reinforcement learning module 114. FIG. 14 is a schematic diagram illustrating an example of an input to an LLM and an output from an LLM. 1 is a schematic diagram showing the process of performing supervised learning based on a dataset including text data, SEM images, etc.;

[0023] FIG. 1 is a diagram showing the procedure for generating a recipe that reduces instrumental differences between multiple measurement tools;

[0024] FIG. 1 is a diagram showing the procedure for adjusting instrument conditions of another measurement tool so as to reduce instrumental differences between a reference instrument of the measurement tool and another measurement tool;

[0025] FIG. 1 is a diagram showing the procedure for adjusting instrument conditions of another measurement tool so as to reduce instrumental differences between a reference instrument of the measurement tool and another measurement tool (including manual operations);

[0026] FIG. 1 is a diagram showing an example of a GUI screen having a display field for displaying text to be input into the LLM and a display field for displaying adjustment details output by the LLM;

[0027] FIG. 1 is a diagram showing an example of a GUI screen having a display field for displaying text to be input into the LLM and a display field for displaying adjustment details (parameters and text) output by the LLM;

[0028] FIG. 2 is a diagram showing an overview of an LLM configured to output a recipe and measurement instructions for operating an instrument;

[0029] FIG. 2 is a diagram showing the procedure for predicting defect locations using an LLM;

[0029] FIG. 3 is a diagram showing the procedure for predicting defect locations using an LLM and generating a recipe based on input of the predicted defect locations into the LLM;

[0029] FIG. 3 is a diagram showing the procedure for deriving the cause of and evidence of an abnormal image using an LLM; A diagram showing the procedure for fine-tuning a model for identifying the cause of abnormal images.A diagram showing the procedure for fine-tuning the cause analysis model for abnormal images. A diagram showing the procedure for reinforcement learning of the cause identification model and cause analysis model for abnormal images.

[0013] 1 is a block diagram of a semiconductor evaluation system 100 according to a first embodiment of the present disclosure. The semiconductor evaluation system 100 is a system that performs evaluation processes such as inspection and measurement on semiconductor devices (samples). The semiconductor evaluation system 100 includes a computer system 101, a semiconductor evaluation tool 102, a design data storage medium 103, and a simulator 104, which are connected to each other by a communication bus 105.

[0014] The computer system 101 includes one or more processors 106 (including at least one of a CPU and a GPU), a memory 107, and a user interface 108.

[0015] The user interface 108 includes a display (not shown) and one or more input devices. The display can display image data output from the semiconductor evaluation tool 102, information output from the processor 106, and the like.

[0016] The semiconductor evaluation tool 102 is, for example, a CD-SEM (Critical Dimension-Scanning Electron Microscope), which is a device that generates an image and a luminance signal waveform based on the detection of secondary electrons and backscattered electrons emitted from a sample when the sample is irradiated with an electron beam. The CD-SEM is a device that measures the dimensions and two-dimensional shape of a pattern formed on a semiconductor device based on the luminance signal waveform. The semiconductor evaluation tool 102 may also be an inspection device that includes a charged particle beam, such as an SEM for defect review, or an optical inspection device.

[0017] The design data storage medium 103 stores design data that describes the physical circuit structure of a semiconductor device in, for example, a GDS II or OASIS file format.

[0018] The simulator 104 generates an observation image of the sample through simulation, or estimates a signal profile indicating the intensity distribution of secondary electrons and backscattered electrons. Specifically, the simulator 104 uses random numbers to reproduce the generation process of secondary electrons generated by a beam incident on the sample and backscattered electrons emitted from the sample, and performs repeated calculations to calculate the emission angle, energy, etc. of the secondary electrons and backscattered electrons. Monte Carlo simulation, etc., is used for the calculation.

[0019] The simulator 104 has a built-in processor (not shown) and a storage medium for storing data and programs required for the simulation. The simulator 104 is configured to read out necessary information (e.g., the pattern shape included in the semiconductor device, the material composition of the semiconductor device) from the design data storage medium 103, and to estimate an electron microscope image (when the semiconductor evaluation tool 102 is an electron microscope) based on previously stored structural information of the semiconductor evaluation tool 102, such as the position of a detector in the semiconductor evaluation tool 102 (the position relative to the beam irradiation location), and control conditions of the measurement device, including the beam irradiation conditions, which will be described later.

[0020] The simulation performed by the simulator 104 may derive images, brightness information, profiles, etc. using not only trajectory calculation methods such as Monte Carlo simulation, but also models that define the relationships between layout data (including three-dimensional data such as film thickness), coefficients that vary depending on the material, device conditions of the electron microscope, brightness information, etc. Furthermore, image data, etc. may be estimated by inputting known information into a learning device that has undergone learning using data sets of device conditions of the electron microscope, material information, 3D layout information, and brightness information (image data, profiles, etc.) as training data.

[0021] The memory 107 is configured with volatile and / or nonvolatile memory. The memory 107 has a non-transitory computer-readable storage medium. The memory 107 stores the following data: (a) an operating system 109 that executes various measurement or inspection-related processing steps (described later); (b) a learning module 110 that generates a learning model based on inputs such as a large-scale language model (described later), text data, and information output from the semiconductor evaluation tool 102; (c) a text creation module 111 that creates text data and the like using the model learned by the learning module 110; and (d) a measurement database 112 that stores measurement results and the like output from the semiconductor evaluation tool 102.

[0022] The learning module 110 has two modules: (a) a supervised / unsupervised learning module 113; and (b) a reinforcement learning module 114. The writing module 111 has a writing model 115. These sub-modules and models are described in more detail below.

[0023] In addition to the above, the memory 107 (e.g., a non-transitory computer-readable storage medium) may store, for example, the following: (a) an operating program executed by one or more processors included in the semiconductor evaluation tool 102; and (b) a recipe generation module that generates the operating program (recipe) based on parameters input from the user interface 108.

[0024] The recipe generation module is configured to execute recipe generation and correction processing of an already generated recipe, for example, based on the output of a text creation module 111 (described later). The recipe generation module is configured to generate a recipe based on, for example, design data, a simulation image generated by the simulator 104 based on input of the design data, parameters input from the user interface 108, etc.

[0025] FIG. 2 is a diagram showing an example of a writing system 200 that uses a general-purpose LLM 210 configured based on a large language model (LLM). FIG. 2 illustrates a writing system 200 that is primarily used for measuring or inspecting semiconductor devices. The writing system 200 generates an LLM suitable for measuring or inspecting semiconductor devices based on the general-purpose LLM 210. The LLM for measuring or inspecting semiconductor devices is based on the existence of a general-purpose LLM 210 that has been trained in advance using a large amount of language data.

[0026] The document creation system 200 includes: (a) an equipment supplier-side computer system 220 that performs fine tuning of the general-purpose LLM 210 mainly based on domain knowledge of the evaluation equipment; and (b) an equipment user-side computer system 230 that performs further fine tuning of the evaluation equipment LLM 222 provided by the equipment supplier based on domain knowledge of the semiconductor device (e.g., domain knowledge about manufacturing conditions in the manufacturing process of semiconductor wafers).

[0027] The equipment supplier-side computer system 220 includes an evaluation equipment LLM learning module 221. The evaluation equipment LLM learning module 221 performs fine tuning on the general-purpose LLM 210 mainly using domain knowledge of the evaluation equipment (semiconductor evaluation tool 102 in FIG. 1 ). As a result, the general-purpose LLM 210 further learns the domain knowledge of the evaluation equipment and is reconfigured as an evaluation equipment LLM 222.

[0028] The equipment user side computer system 230 includes a semiconductor LLM learning module 231. The semiconductor LLM learning module 231 performs further fine tuning using domain knowledge of semiconductor devices on the evaluation equipment LLM 222. As a result, the evaluation equipment LLM 222 further learns domain knowledge of semiconductor devices and is reconfigured as the semiconductor LLM 232.

[0029] The device user side computer system 230 further includes a reinforcement learning module 233. The reinforcement learning module 233 performs reinforcement learning, which will be described later, on the semiconductor LLM 232, thereby performing re-learning on the semiconductor LLM 232 so that the semiconductor LLM 232 can output an evaluation recipe that enables the evaluation device to achieve a desired target value. As a result, the semiconductor LLM 232 is reconfigured as a sentence creation model.

[0030] The device user side computer system 230 further includes a sentence creation module 234 including a sentence creation model that has been subjected to the reinforcement learning. The sentence creation module 234 uses the sentence creation model 115 to output an output string as a response to an input string.

[0031] In the system configuration illustrated in FIG. 2 , the equipment supplier's computer system 220 performs fine tuning using domain knowledge of an evaluation device such as an electron microscope, and the equipment user's computer system 230 performs fine tuning using domain knowledge of semiconductor devices. However, the computer systems that perform fine tuning are not limited to this. For example, the equipment supplier's computer system 220 may perform fine tuning using domain knowledge related to semiconductor devices, or the equipment user's computer system 230 may perform fine tuning using domain knowledge of the evaluation device. Similarly, reinforcement learning may be performed in either computer system. Furthermore, the equipment supplier's computer system 220 and the equipment user's computer system 230 may be configured as an integrated unit.

[0032] The equipment supplier computer system 220 and the equipment user computer system 230 can each be configured using the computer system 101 described in FIG. 1 . That is, (a) the evaluation equipment LLM learning module 221 and the semiconductor LLM learning module 231 can each be configured using the supervised / unsupervised learning module 113, (b) the reinforcement learning module 233 can be configured using the reinforcement learning module 114, and (c) the writing module 234 can be configured using the writing module 111 and the writing model 115. FIG. 1 shows an example in which these modules are integrated into a single computer system 101. <Embodiment 1: Fine Tuning> The general-purpose LLM 210 is configured in advance to learn various domain knowledge so that it can output an answer string that reflects that domain knowledge in response to an input question string. A wide range of domain knowledge can be learned. For example, if a specific book is learned, a summary of that book can be provided as an answer. Alternatively, if one has learned a specific programming language, one can describe a desired function in natural language and input it into the general-purpose LLM 210, which can output source code that implements that function.

[0033] Although the general-purpose LLM 210 has such versatility, it cannot output useful answers for knowledge in fields that it has not previously learned, and can only provide answers that are merely general. Therefore, in order to have the general-purpose LLM 210 learn knowledge in a specific field, tuning known as fine tuning must be performed. The present disclosure aims to input desired operations for a semiconductor device evaluation system (semiconductor evaluation tool 102 in FIG. 1 ) into the LLM and obtain an operational specification (evaluation recipe) that can realize the desired operation as a response from the LLM. In other words, the objective is to operate the semiconductor device evaluation system using natural language, etc. Therefore, in embodiment 1, the general-purpose LLM 210 is first fine-tuned to learn domain knowledge of the evaluation system and domain knowledge of semiconductor devices.

[0034] 3 shows an example of domain knowledge learned by fine tuning. The "Evaluation Device" row in the "Domain Knowledge" column in Fig. 3 is an example of domain knowledge related to the evaluation device learned by the evaluation device LLM learning module 221. The other rows in the "Domain Knowledge" column in Fig. 3 are examples of domain knowledge related to semiconductor devices learned by the semiconductor LLM learning module 231.

[0035] Fine tuning of domain knowledge related to the evaluation device can be performed by having the general-purpose LLM 210 learn document data such as the instruction manual for the evaluation device and the functional specifications of the evaluation device. This allows knowledge related to the operation (usage) of the evaluation device to be learned. Furthermore, the specifications of the image processing performed by the evaluation device or the results of the image processing can also be learned as domain knowledge of the evaluation device. The specifications of the image processing can be learned from document data such as specifications or textbooks that describe the contents of the image processing. The results of the image processing can be learned by studying the changes in the observed image before and after actually performing the image processing.

[0036] A recipe is data describing the operational specifications that define the evaluation process performed by the evaluation equipment. Depending on the user of the evaluation equipment, there may be parameters that are prohibited from being used in the recipe or that impose restrictions on the recipe. Such user-specific restrictions (parameter items that impose restrictions and the content of the restrictions) are called local rules. Because local rules are user-specific, it is more appropriate to learn them as domain knowledge specific to the semiconductor device that the user is evaluating, rather than learning them as general domain knowledge of the evaluation equipment. Therefore, we decided to learn local rules as domain knowledge of the semiconductor device.

[0037] Since recipes describe the instructions for the evaluation equipment, they appear to users as a list of parameters and are not necessarily easy to read. Measurement instructions are documents that describe the steps for creating a recipe in natural language, and in some cases, measurement instructions are easier for users to read. Therefore, we decided to learn measurement instructions as part of the domain knowledge of semiconductor devices.

[0038] The specifications of the image processing performed by the evaluation device can also be learned as domain knowledge of semiconductor devices. The learning method is the same as when learning as domain knowledge of the evaluation device.

[0039] A user who manufactures semiconductor devices may have a simulator that simulates the manufacturing process of semiconductor devices. This simulator uses design data that describes the three-dimensional shape, materials, and manufacturing process information (e.g., manufacturing conditions of semiconductor manufacturing equipment) of the sample to be measured (semiconductor device). In addition, the simulator may also use design data that describes the two-dimensional shape of the shape pattern formed on the semiconductor device. This design data can be learned as domain knowledge of semiconductor devices.

[0040] When an evaluation device performs evaluation by irradiating a sample (semiconductor device) with an electron beam (or any other charged particle beam), data describing the interaction between the electron beam and the sample material can be learned as domain knowledge of the semiconductor device.

[0041] Figure 4 shows the division between unsupervised learning and supervised learning. The instruction manual or functional specifications of the evaluation device can be learned by unsupervised learning. The image processing specifications can be learned by unsupervised learning. The image processing results can be learned by supervised learning. The measurement instructions can be learned using either unsupervised or supervised learning. The local rules can be learned by unsupervised learning. The design data can be learned by supervised learning. The interaction between the electron beam and the sample can be learned by supervised learning.

[0042] FIG. 5A is a schematic diagram showing the process by which the learning module 113 performs unsupervised learning. Unsupervised learning can be performed, for example, by Masked LM. Masked LM is a technique that learns the frequency of occurrence of character strings in a document by repeatedly masking a portion of the document and estimating the masked portion, thereby learning the knowledge described in the document. Learning progresses by feeding back the difference (loss) between the estimation result of the masked portion and the correct character string to the LLM. An example of a loss function is, for example, cross-entropy error, but is not limited to this.

[0043] FIG. 5B is a schematic diagram showing another process in which the learning module 113 performs unsupervised learning. Unsupervised learning can be performed by NSP (Next Sentence Prediction) in addition to Masked LM. Any other method may be used. NSP repeatedly predicts whether two sentences are semantically connected. This makes it possible to learn the correct connection between sentences. The loss function is the same as that of Masked LM.

[0044] FIG. 6 is a schematic diagram showing the process by which the learning module 113 performs supervised learning. Here, an example is shown in which a request for an evaluation device written in a natural language string and the design data described in FIG. 3 are input to the LLM. In this example, the LLM outputs a measurement instruction document describing the procedure for creating an evaluation recipe that implements the requested procedure. A measurement instruction document that can actually satisfy the request is provided as the correct answer data. Supervised learning can be performed by feeding back to the LLM the loss between the measurement instruction document output by the LLM and the correct measurement instruction document.

[0045] FIG. 7 shows variations in the combination of input / output / correct answer / loss function in supervised learning. When design data is used as input and measurement instructions are used as output and correct answer, the loss function is, for example, cross-entropy error. When measurement instructions are used as input and a recipe is used as output and correct answer, the loss function is, for example, cross-entropy error. When learning about the interaction between an electron beam and a sample, the input is the imaging conditions (voltages at each part, currents at each part, etc.) and 3D design data when the evaluation device captures an observed image of the semiconductor device, or the imaging conditions and information about the semiconductor device material, the output is the observed image, and the loss function is, for example, mean squared error. When learning about the results of image processing, the input is the recipe (particularly the filtering conditions for the image processing performed by the evaluation device), the output is the observed image, and the loss function is, for example, mean squared error.

[0046] The learning module 113 (or the learning module 221 for the evaluation equipment LLM and the learning module 231 for the semiconductor LLM) performs unsupervised learning / supervised learning according to the above procedure, thereby allowing the general-purpose LLM 210 to learn domain knowledge of the evaluation equipment and domain knowledge of the semiconductor device through fine tuning. As a result, when an input statement related to the evaluation equipment or the semiconductor device is given, the semiconductor LLM 232 can output a response reflecting this domain knowledge. <First Embodiment: Summary> The computer system 101 (or the equipment supplier-side computer system 220 and the equipment user-side computer system 230) according to the first embodiment allows the general-purpose LLM 210 to learn domain knowledge of the evaluation equipment and domain knowledge of the semiconductor device through fine tuning. As a result, when an input statement related to the evaluation equipment or the semiconductor device is given, the semiconductor LLM 232 can output a response reflecting this domain knowledge. Since the measurement instructions and recipes are learned in the fine tuning, the semiconductor LLM 232 can output these as output (see FIG. 7 ). That is, when a request for the evaluation device is input to the semiconductor LLM 232 using a character string, a response corresponding to the request can be obtained from the semiconductor LLM 232 in the form of a measurement instruction or recipe. This allows instructions to be given to the evaluation device using a character string via the LLM. In other words, the evaluation device can be operated using a character string. <Embodiment 2> In the first embodiment, an example configuration was described in which instructions to the evaluation device can be given using a character string via the LLM. According to the first embodiment, the objective of operating the evaluation device using a character string can be achieved to some extent. However, depending on the parameter items and their value configurations described in the evaluation device recipe, it is possible that the operation of the evaluation device will not necessarily be desirable (e.g., the desired target value cannot be achieved). Therefore, in the second embodiment of the present disclosure, based on the configuration of the first embodiment, an example configuration in which the LLM can output a more desirable recipe will be described.The configuration of the computer system 101 (or the equipment supplier's computer system 220 and the equipment user's computer system 230) is the same as that of the first embodiment.

[0047] 8 is a schematic diagram showing the process of reinforcement learning performed by the reinforcement learning module 114. The reinforcement learning module 114 (or the reinforcement learning module 233) retrains the LLM (semiconductor LLM 232 in embodiment 1) so as to output a recipe that can achieve the target value by feeding back to the LLM the results of evaluation of the semiconductor device by the evaluation device using the recipe (specification data in FIG. 8) output by the LLM, along with the target value. This is considered to enable the LLM to output a desirable recipe that can achieve the target value. The same can be considered when the LLM outputs a measurement instruction sheet.

[0048] The procedure for re-learning by the reinforcement learning module 114 is roughly as follows: Re-learning of the LLM is performed using at least one of the output data output by the evaluation device (or text data generated based on this; specific examples will be described later) as learning data. A string of characters describing instructions to the evaluation device in natural language sentences is input to the re-learned LLM. Data specifying the operation of the evaluation device (called specification data; for example, a recipe) is obtained as output from the LLM. The specification data is used to operate the evaluation device, and output data from the evaluation device is obtained. Re-learning is performed using the output data as learning data. Re-learning can be performed by repeating the above steps.

[0049] When relearning is performed, the target values ​​for each evaluation item are input to the LLM along with the evaluation results of the semiconductor device obtained by the evaluation device. Furthermore, the specification data used to obtain the evaluation results is also input. This allows the difference between the target values ​​and the evaluation results to be fed back, and also allows the specification data that caused the evaluation results to be learned. Therefore, relearning can be performed so that the LLM outputs specification data that brings the evaluation results closer to the target values.

[0050] The items evaluated by the evaluation device (i.e., items for which the evaluation results and target values ​​are targeted) can be, for example, the following: (a) throughput of the evaluation device; (b) values ​​measured by the evaluation device (e.g., pattern size); (c) feature quantities of observed images of semiconductor devices acquired by the evaluation device (e.g., noise amount, various histograms); and (d) probability of defect occurrence at each location on a semiconductor device. The evaluation device outputs an evaluation value for each of these evaluation items. These are fed back to the LLM together with the target values ​​for each evaluation item.

[0051] Examples of specification data that the LLM can output include: (a) data (i.e., recipe) that specifies the process that the evaluation device performs when evaluating a semiconductor device; (b) predicted defect positions in a semiconductor device (prediction of whether or not a defect exists at the measurement target position); (c) predicted causes of anomalies when an observed image of a semiconductor device acquired by the evaluation device is abnormal and measures to improve the abnormality; (d) an image quality improvement model that converts the observed image into an image of higher image quality; and (e) a pattern enhancement model that enhances specific shape patterns contained in the observed image. The LLM performs learning and re-learning in advance so as to output the above specification data.

[0052] When the LLM outputs the cause of the anomaly and its remedy, it may output, for example, a recipe modified to improve the anomaly and the cause of the anomaly along with the recipe before the improvement. This allows the user to understand which part of the recipe before the improvement had a problem. The image quality improvement model and the pattern enhancement model may be configured using data such as a mathematical model that implements these functions, or may be configured using any other data and / or program, such as artificial intelligence, that implements these functions.

[0053] FIG. 9 is a schematic diagram showing an example of a re-learning procedure by the reinforcement learning module 114. In this example, the LLM is pre-configured to receive as input a sentence describing instructions for the evaluation device and design data for a semiconductor device, and to output an operating recipe for the evaluation device as specification data. Since the design data may not necessarily be written in a format suitable for the LLM, a conversion module may be used to convert the design data into a format that is easy for the LLM to interpret. The input sentence may also be further converted appropriately (e.g., converting a natural language string into a numerical vector) before being input to the LLM.

[0054] The evaluation results of semiconductor devices output by the evaluation device may be fed back to the LLM as they are, or the evaluation results may be further processed to convert the output from the evaluation device into an evaluation value, which is then fed back to the LLM. One example of evaluation processing is for a user to manually evaluate the evaluation results output by the evaluation device and then generate an evaluation value by quantifying the results using rankings or the like. Any other evaluation processing may also be used.

[0055] Examples of items that may be considered (i.e., quantified) when further evaluating the evaluation results output by the evaluation device include the following: (a) throughput of the evaluation device (the number of semiconductor devices evaluated per unit time); (b) evaluation results (semiconductor device measurement success / failure); (c) the amount of noise in the observation image of the semiconductor device acquired by the evaluation device, various histograms, and measurement values ​​(e.g., pattern size, amount of deviation between upper and lower layers); (d) evaluation score indicating whether the measurement was successful or not, such as the image correlation value between an AP image (a unique circuit pattern image near the measurement position) and a captured image; (e) stability of the evaluation (e.g., matching success rate after long-term operation, variance of measurement values); (f) coverage of measurement positions for hotspots; (g) uniqueness of the observation image (AP image), etc.

[0056] Examples of algorithms that can be used to learn pairs of evaluation results (evaluation values) / target values / specification data include, but are not limited to, Temporal Difference Learning (TD learning) and Proximal Policy Optimization (PPO).

[0057] FIG. 10 is a schematic diagram showing another example of a relearning procedure by the reinforcement learning module 114. In the relearning procedure described in FIG. 9, when the target value and evaluation result are fed back to the LLM, they may be converted in advance into a format or value suitable for learning by the LLM. FIG. 10 shows one such example. Evaluation results may be fed back to the LLM without conversion and converted and then fed back to the LLM. FIG. 10 shows an example of such a combination. Alternatively, evaluation results from an evaluation device may be used in combination with evaluation values ​​obtained by further evaluating those evaluation results. Because the evaluation values ​​are quantified, they can be fed back to the LLM as is. However, because the evaluation results themselves have various formats and values, some conversion is desirable. The configuration of FIG. 10 is useful in such cases.

[0058] The target values ​​may be automatically extracted from an input statement input to the LLM. In the example shown in FIG. 9, these values ​​correspond to throughput and measurement accuracy. The extraction process may be performed by the user, or by inputting the input statement to an LLM (which may be the same as semiconductor LLM 232 or a separately configured LLM) that has been pre-configured to extract the target values ​​from the input statement.

[0059] FIG. 11 is a schematic diagram showing another example of a relearning procedure by the reinforcement learning module 114. In the relearning procedure described with reference to FIGS. 9 and 10 , the configuration of FIG. 11 is useful when verifying how the evaluation results of the evaluation device change when a recipe is changed. Specifically, the recipe output by the LLM and the evaluation results (and / or evaluation values) obtained when the evaluation device is operated using the recipe are fed back to the LLM. This allows learning of recipe / evaluation results (and / or evaluation values) / target values ​​as a pair. In other words, it is possible to learn a recipe that produced a certain evaluation result by associating it with that evaluation result. This allows learning of the relationship between the evaluation results and the recipe, thereby enabling the LLM to predict the evaluation results that will be obtained when the recipe is changed.

[0060] FIG. 12 is a schematic diagram showing another example of a relearning procedure by the reinforcement learning module 114. In the relearning procedure described in FIGS. 9 to 11, the LLM may output measurement instructions as an intermediate output. The measurement instructions are further input to the LLM to output a recipe. The LLM that outputs the measurement instructions (large-scale language model 1 in FIG. 12) and the LLM that outputs the recipe (large-scale language model 2 in FIG. 12) may be configured as separate LLMs, or they may be configured as an integrated whole. For convenience of description, these are represented as separate LLMs in FIG. 12.

[0061] Since a recipe is an instruction itself to an evaluation device, it may be written in a format that is not necessarily suitable for a user to understand. In contrast, a measurement instruction is a document that describes the procedure for creating a recipe using natural language strings, making it easy for a user to understand. The configuration shown in Figure 12 is useful in that it allows a user to understand what instructions were given to derive a recipe in the form of a measurement instruction.

[0062] As described with reference to Figures 9 to 12, the output data output by the evaluation device may be fed back to the LLM as is, or data generated by converting the output data (e.g., text data that is easy for the LLM to learn) may be fed back to the LLM. <Summary of Second Embodiment> The computer system 101 (or the equipment supplier computer system 220 and the equipment user computer system 230) according to the second embodiment performs reinforcement learning on the semiconductor LLM 232, thereby enabling the semiconductor LLM 232 (or the sentence creation model 115 after reinforcement learning) to output a recipe that enables the evaluation device to achieve a target value when evaluating a semiconductor device. This allows the evaluation device to be operated using text strings, and also allows recipes and other specification data that enable desired target values ​​to be obtained using text string instructions.

[0063] For example, if an abnormality is found in an observation image acquired by an evaluation device, it is necessary to revise the measurement instruction or recipe. However, creating the measurement instruction or recipe can be very costly and time-consuming for the user. According to the second embodiment, even in such a case, by instructing the LLM to create a recipe or measurement instruction using character strings, a recipe or measurement instruction that achieves the desired target value can be obtained. This reduces the burden on the user. <Third Embodiment> In the first and second embodiments, an example was described in which text or design data was input to the LLM and a recipe was obtained as output. The input to the LLM and the output from the LLM are not limited to these, and the LLM can be trained to achieve other inputs / outputs. In the third embodiment of the present disclosure, examples of these inputs / outputs are described. The configuration of the computer system 101 (or the equipment supplier's computer system 220 and the equipment user's computer system 230) is the same as that of the first embodiment.

[0064] 13 shows an example of input to and output from an LLM. The LLM here may be any of the general-purpose LLM 210, the evaluation device LLM 222, the semiconductor LLM 232, and the writing model 115.

[0065] When instructing the LLM to improve a program (such as source code, executable code, or machine learning model, the same applies below), a directive (sentence) requesting the LLM to do so is input, along with the program before the improvement. The LLM is pre-trained to output an improved program that incorporates the instructed improvements to the input program. The program output by the LLM can be used in an evaluation device. For example, the image quality improvement model and pattern enhancement model described in embodiment 1 can be obtained as output from the LLM.

[0066] The LLM can also be trained in advance to generate a program. In this case, a sentence specifying a desired function is input to the LLM, and the LLM outputs a program that implements that function. For example, the LLM can output the image quality improvement model or pattern enhancement model described in the first embodiment.

[0067] The LLM can be trained in advance to output the most suitable recipe for evaluating an input image (an observation image of a shape pattern formed on a semiconductor device). In this case, when the image is input to the LLM, a sentence instructing the LLM to output the most suitable recipe for evaluating the shape pattern of the image is also input. The LLM outputs the optimal recipe according to the instruction.

[0068] As described in the second embodiment, the LLM can be configured to input design data and instructions for a semiconductor device and output a recipe suitable for evaluating the semiconductor device of the design data. In this case, the LLM can output a measurement instruction as an intermediate output in addition to directly outputting the recipe.

[0069] As described in the second embodiment, the LLM can be configured to output remedial measures when an observed image is found to be abnormal. In this case, an observed image containing an abnormality is input to the LLM, and an instruction statement instructing the LLM to output a recipe capable of correcting the abnormality is also input. The LLM is pre-trained to output a corrected recipe capable of improving the observed image of the portion (circuit pattern) where the abnormality occurred. Furthermore, the cause of the abnormality and the corrections in the recipe may also be output. While the cause of the abnormality and the corrections in the recipe are not themselves involved in the automatic operation of the evaluation device, this information can be used when explaining the significance and cause of the recipe correction to the user. <Fourth Embodiment> In the fourth embodiment of the present disclosure, yet another example of fine-tuning of the LLM illustrated in FIG. 6 and other figures is described. FIG. 14 illustrates a process in which information related to requests to the evaluation device (inspection device) and information related to the semiconductor manufacturing process other than design data information are associated with the information related to the requests to the evaluation device to generate learning data for the LLM. Fine-tuning is preferably performed by associating text data with at least one other data.

[0070] Information related to requirements for the evaluation device is, for example, text data. Information related to the semiconductor manufacturing process other than design data information includes, for example, measurement tools such as CD-SEM and SEM images obtained with SEM-type defect inspection tools. It may also be images obtained from an optical inspection device. This may be tuned to create a recipe for the optical inspection device, or may be used to set SEM device conditions to approximate images obtained with the optical inspection device.

[0071] SEM images and the like are acquired in the semiconductor manufacturing process to evaluate the performance of devices. For example, image data may be input along with text data such as "Please set the imaging conditions so that the contrast between the pattern and the substrate is greater than in the input image." In this case, the LLM may be tuned by associating it with information about the operating program (recipe) of the evaluation device that acquired the image.

[0072] Another type of information related to semiconductor processes is process information. Semiconductor devices are manufactured through multiple manufacturing processes, but the measurement conditions for devices manufactured using the same manufacturing process are often similar, even if the generation or type of device changes. Therefore, it is possible to input text data such as, "Please create a measurement instruction sheet for the imaging conditions commonly used in the input process." In this case, if the LLM has been previously trained or tuned using recipe information for past generations, etc., for each manufacturing process, it can receive measurement instructions, etc., including imaging conditions (device conditions for the measurement device) suitable for that manufacturing process.

[0073] Furthermore, other information related to semiconductor processes includes device information, such as the device name and device identification information. If the device type is the same, the measurement conditions for measuring the device are often the same or similar. Therefore, by performing fine tuning based on the device information, it is possible to receive measurement instructions and the like that include imaging conditions suitable for the device.

[0074] As explained above, even for devices of different generations, if the manufacturing process and type are the same, the measurement recipe settings used are often similar. Therefore, an LLM can be trained and tuned using a dataset of process information, device information, and / or recipe information, as well as text data. A loss function is calculated between the measurement instructions output by the tuned LLM and the correct measurement instructions, and supervised learning can be performed on the LLM using backpropagation to minimize the difference. Fifth Embodiment In a fifth embodiment of the present disclosure, a recipe setting method for reducing inter-machine variance between multiple measurement tools (semiconductor evaluation tools), such as CD-SEMs, is described. Modern semiconductor manufacturing processes involve hundreds of steps, and multiple manufacturing lines may be used for this purpose. In such cases, manufacturing conditions must be controlled to eliminate manufacturing errors between lines and devices. It is also necessary to minimize inter-machine variance between measurement tools. More specifically, if measurement tool A (reference tool) measures a pattern dimension of 20 nm, it is desirable to adjust the tool conditions so that measurement tool B measuring the same pattern also produces the same measurement result. With the recent trend toward finer semiconductor device patterns, the required performance values ​​for machine differences are becoming smaller every year.

[0075] In embodiment 5, a method and system that can create a recipe that reduces machine differences, modify a recipe, or propose equipment adjustment conditions using LLM will be described. By using LLM, it is possible to reduce or eliminate the adjustment work that has been required to reduce machine differences, and as a result, it is possible to reduce the cost and time required for the work.

[0076] 15 is a diagram illustrating an example of an LLM operation method suitable for suppressing differences in recipes between reference machine A (CD-SEM_A) (first evaluation machine), measurement machine B (CD-SEM_B) (second evaluation machine), and measurement machine C (CD-SEM_C). In Fig. 15, a sentence including required specifications is input to the LLM, and a recipe (that satisfies the required specifications) is output.

[0077] FIG. 15 shows an example of inputting a required specification into the LLM, such as "Please create a recipe with the following specifications. However, please adjust the existing recipe so that the measurement error is XX nm. Required specifications: Measurement value difference between devices is within XX nm, throughput is within XX." Also, in the example of FIG. 15, the recipe of the reference machine A is input into the LLM together with the text data, but if there is no existing recipe, this does not have to be input. Furthermore, the comparison results (e.g., comparison results of images obtained by the reference machine and other devices) may be input into the LLM together with the text data, and the LLM may be tuned.

[0078] Next, the recipe output from the LLM is read into multiple devices (reference device A, measurement device B, measurement device C) and measurements are performed. If the multiple devices are CD-SEMs, each CD-SEM can output image data, pattern dimension measurement results, etc. The computer system 101 or other calculation device compares the CD-SEM output results and outputs the results. The computer system 101, etc. may be configured to be able to send and receive information via a communication medium, etc., so that it can receive the output of each CD-SEM.

[0079] The recipe is an operating program for operating the CD-SEM, and includes settings such as the device conditions for the electron microscope and the image processing conditions for the output image. It is desirable to set this recipe so that the differences between CD-SEMs are minimized. To achieve this, in the fifth embodiment, the recipe is optimized using the following procedure.

[0080] First, learning is performed by inputting text data such as "Create a recipe with the following specifications. However, please adjust the existing recipe so that the measurement error is XX nm. Required specifications: Measurement value difference between devices within XX nm, throughput within XX" into the LLM along with recipe information for reference machine A. Based on this input, the LLM outputs the recipe or setting specifications for the device. When the setting specifications are output, the recipe is generated by manually or automatically entering the specifications as the setting conditions for the recipe.

[0081] Furthermore, input information for the LLM includes semiconductor manufacturing process information (the final manufacturing process before measurement of the semiconductor device to be measured), device name, design drawing, material information, and a group of output images from reference machine A. If such additional information is present, it is advisable to explain in the text data how to handle that data. Furthermore, other information regarding reference machine A, device B, and device C may be input. This other information includes device information such as device setting parameters, detailed information on the components that make up the electron microscope, and adjustment details for the components.

[0082] In the fifth embodiment, the recipes generated as described above are used to operate the reference machine A, machine B, and machine C to measure the dimensions of the patterns formed on the semiconductor devices. The output of the LLM may be a recipe common to each machine, or may be a recipe specific to each machine that has been adjusted to reduce machine differences.

[0083] The computer system 101 etc. receives output images and pattern measurement values ​​of the reference machines A, B, and C controlled by the LLM output recipe. The computer system 101 etc. obtains index values ​​for comparing the output of each machine, such as PSNR (Peak Signal Noise Ratio), contrast, and sharpness, from each image, or compares the machines based on the output of a computer installed in the measurement machine. Image correlation may also be performed. Furthermore, comparison of measurement values ​​and area values ​​of pattern portions may also be performed. Other index values ​​indicating the degree of deviation between the index values ​​of each machine may also be used. Deep learning models, evaluation models, mathematical formulas, etc. may be used to compare images and evaluate the comparison results.

[0084] The computer system 101 etc. functions as a preprocessing device for training the LLM based on output information from the device. The computer system 101 etc. functions as an input device for inputting the comparison results together with text data etc. The computer system 101 etc. repeatedly inputs the recipes generated by the LLM and the comparison results back into the LLM, thereby enabling recipe generation for multiple devices according to various variable factors.

[0085] FIG. 16 is a diagram showing the recipe generation process for tool B, which is primarily intended to reduce the machine difference between reference tool A and tool B. One difference from FIG. 15 is that recipe information for tool B, etc., is input to the LLM. Information on reference tool A and tool B includes, for example, tool parameters, detailed information on components used in the electron microscope, and component adjustment details. Information input to the LLM also includes images of reference tool A, measurement results from reference tool A, semiconductor manufacturing process information, device names, design drawings, and material information. When the tool (measuring tool) is an overlay error measuring tool, the appearance of the underlying pattern differs depending on the depth to the underlying pattern and the material. Therefore, adjustments are made to change the energy of the electron beam depending on the material and depth.

[0086] In the example of Figure 16, a sentence such as "We want to keep the measurement value difference between the devices within XX nm compared to the reference device. Based on the current device information, what part of device B should be adjusted? If automatic adjustment is possible, please adjust it" is input to the LLM. Since the LLM is fine-tuned based on information from at least one of the reference device A and device B, the adjustment details can be output by inputting the above required specifications. The adjustment details can be, for example, a sentence describing various setting values ​​for internal device parameters, text information written in a format readable by the device, or both.

[0087] The computer system 101 or the like controls the apparatus B based on the recipe output from the LLM or the recipe adjusted based on the parameters output from the LLM, etc. The computer that controls the apparatus B and the computer that generates the recipe may be one or more computers.

[0088] The computer system 101 or the like receives the output data of device B and compares it with the output data of reference device A to generate input data for the LLM. This input data is the comparison result between reference device A and device B, or data indicating the discrepancy between the outputs of the two devices, and the LLM estimates adjustment parameters according to the current degree of discrepancy based on this input. Note that image data of reference device A or the like may be stored in advance in a storage medium as a reference image.

[0089] By performing the above process or repeating it two or more times, it is possible to estimate parameters of device B that suppress the machine difference between reference device A and device B.

[0090] FIG. 17 shows an example of a recipe generation process for device B to reduce the machine difference between reference device A and device B, in which the recipe is manually adjusted based on the output of the LLM. In the example of FIG. 17, adjustment details are output by inputting text including required specifications and device information into the LLM. The device is manually adjusted based on these adjustment details. Text indicating that the adjustment has been completed is again input into the LLM, and a verification recipe related to the machine difference is output. The verification recipe is read into the device, device B is automatically operated, and an image is captured. The captured image is then compared with an image captured by the reference device, an evaluation value is calculated, and the evaluation result is fed back to the LLM.

[0091] Manual tasks include working with the adjustment details output from the LLM, and subsequent input actions to the LLM such as command input, GUI operation, and character input. Information on reference machine A and equipment B includes parameters within the equipment, information on parts used in the equipment, and the adjustment history of the parts. In addition, the adjustment details output from the LLM include corrections to parameters and recipes within the equipment.

[0092] 18 is a diagram showing an example of a GUI screen having a display field for a sentence (prompt) to be input into the first LLM and a display field for displaying the adjustment content output by the LLM. The sentence input into the prompt field is the same as that input into the LLM shown in FIG. 16. FIG. 18 shows an example in which the LLM outputs the adjustment content in the comment field, saying, "Manual work will be required. Please adjust part XX. Please let us know when the adjustment is complete. We will perform a machine difference evaluation."

[0093] 19 also illustrates an example of a GUI screen that includes a prompt field, a comment field displaying the output of the LLM, and a parameter adjustment field. The computer system 101, etc., is configured to display parameters registered in the recipe for apparatus B in the "current value" field of the parameter adjustment field illustrated in FIG. 19, and parameters recommended by the LLM in the "estimated value" field. The parameters displayed in the "estimated value" field are, for example, searched for and extracted from text data output by the LLM to display the corresponding parameters. The computer system 101, etc., is configured to apply parameters selected in check boxes provided for each "estimated value" field as update data for the recipe of apparatus B.

[0094] As shown in the example of FIG. 17, by creating a message prompting the operator to make an adjustment and using the LLM as an application for creating a recipe that reflects the adjustment conditions set by the operator, the update conditions for the recipe recommended by the LLM can be selected based on the operator's judgment based on their own experience.

[0095] FIG. 20 illustrates a processing procedure when the adjustments output by the LLM are divided into those requiring manual operation (adjustment) and those requiring a recipe update by the LLM. The LLM is configured to output a recipe for tool B and measurement instructions required for manual operation. The computer system 101 compares the output of a tool operating according to the recipe output by the LLM with that of a tool operating according to a recipe manually set based on the measurement instructions output by the LLM, and feeds back the comparison results, thereby enabling tuning of the LLM in accordance with the operator's intentions. <Embodiment 6> Recent advances in semiconductor device pattern miniaturization and complexity have increased the difficulty of manufacturing. Due to the increasing complexity of semiconductor devices, both the number of manufacturing processes and the number of inspection points are on the rise. The increase in the number of inspections and measurements also tends to increase costs.

[0096] In a sixth embodiment of the present disclosure, a method for predicting hot spots in a semiconductor device will be described. A hot spot in the sixth embodiment refers to a location that is likely to become a defect in a semiconductor device, such as a broken or short circuited wiring. Defects are likely to occur in locations where the process margin of a semiconductor device is small, and can also be defined as layout locations that are at high risk of becoming defective, such as where a pattern is broken, where patterns come into contact, or where desired dimensional conditions are not met.

[0097] In embodiment 6, a method for predicting hot spots using LLM will be described. By selecting appropriate hot spots, it is possible to selectively measure and inspect areas with a high defect rate, thereby reducing production costs.

[0098] FIG. 21 is a diagram illustrating a hot spot prediction method using an LLM. The LLM stored in one or more computer storage media is configured to output predicted defect locations based on input of text data and design data. More specifically, one or more computer systems input text and design drawings (regardless of whether they are 2D or 3D) to the LLM, thereby outputting predicted defect locations. Based on this output result and the correct defect occurrence location, a loss value is calculated and fed back to the LLM. The loss value is calculated using a loss function such as cross-entropy error.

[0099] It is desirable that material information, process information (information on the exposure tool and etching tool), process window (focus and dose margin), SEM images, and defect types can also be input as input to the LLM. It is also desirable that defect position information occurring in older versions of the device can also be input. While the sixth embodiment describes an example in which design data is input as layout data for a semiconductor device, SEM images showing defects and defect position information may alternatively be input to the LLM.

[0100] Defects during semiconductor processing can occur when the focus or dose of an exposure tool deviates from the process window. Defects can also occur due to overetching by an etching tool. Furthermore, defects can also occur based on events that occurred in a process prior to the manufacturing process being measured, so that information (such as internal parameters of the manufacturing tool in the previous process and irradiation time) can also be used as input information.

[0101] Furthermore, by inputting information about materials used in the current and previous processes (resist agents and other semiconductor materials), the accuracy of the estimated information can be improved. Regarding the design data input to the LLM, by inputting layout information from not only the current process but also the previous process, the hot spot prediction performance can be improved. Furthermore, output information from a measurement device such as a CD-SEM (SEM images and measurement results of patterns where defects occur), position information on wafers where defects occur (defect position information identified by a defect inspection device, etc.), and material information on defects identified by an X-ray inspection device or the like built into the defect inspection device may also be input.

[0102] The learning model is minimized by minimizing the magnitude of the deviation between the predicted defect position estimated by the LLM and the correct value. For example, the output of a process simulator may be used as the initial correct data, and the defect position may be predicted using a model optimized to minimize the loss value.

[0103] 21 shows an example of inputting a sentence such as "Please predict the defect location for the XX process. However, please keep it within XX points. Also, I would like the throughput to be within XX." into the LLM. FIG. 21 also shows an example of inputting design data in addition to the sentence data. According to the LLM shown in FIG. 21, it is possible to derive a number of defect location candidates appropriate for measurement and inspection by referring to the design data.

[0104] 22 is a diagram showing a procedure for predicting hotspot location information using a first LLM and generating a recipe based on the input of the prediction results to a second LLM. Although the diagram shows multiple LLMs, the process illustrated in FIG. 22 may be performed using a single LLM. The input text data shown in FIG. 21 is an example of the same case. According to the method illustrated in FIG. 22, recipe update conditions recommended by a single LLM can be selected based on the operator's empirical judgment, and then a recipe can be generated.

[0105] Furthermore, by feeding back the output results from the measurement / inspection device to the LLM, the prediction accuracy of the LLM can be improved. The output of the measurement / inspection device includes defect presence / absence information (such as the probability of presence / absence) at the predicted defect position, defect classification information, etc. By inputting defect classification information, etc., it is possible to obtain an output corresponding to an instruction for predicting the position of a specific defect type.

[0106] The recipe output from the LLM can also be provided to multiple measurement devices. The computer system 101 receives output from defect inspection devices, such as defect location inspection results and evaluation values ​​such as device throughput, and generates feedback information for the LLM. The text input to the LLM includes evaluation results for required performance values, and evaluation can be performed using an AI model for evaluation, manual judgment, or software incorporating algorithms. Furthermore, material information, process information (exposure tool and etching tool information), process window (focus and dose margin), SEM images, and defect types can also be input to the LLM. Defect location information occurring in older device versions can also be input. <Embodiment 7> Today's large-scale semiconductor manufacturing processes involve hundreds of processes on a single line. Measurement and inspection are performed to evaluate the device's performance at each process and to inspect for defects and foreign particles. For example, electron microscopes and optical inspection devices for semiconductor evaluation and inspection perform measurements and inspections based on acquired images and signal waveforms.

[0107] However, during the measurement and inspection process, malfunctions in the measurement and inspection equipment or malfunctions in the semiconductor manufacturing equipment (process abnormalities) can cause abnormal images (excessive noise, low contrast, pattern defects), resulting in measurement or inspection failures. Since stopping the production line for a long time to identify the cause of such abnormalities reduces manufacturing efficiency, it is desirable to identify the cause as soon as possible.

[0108] In a seventh embodiment of the present disclosure, an LLM capable of identifying the cause of an anomaly in a semiconductor manufacturing process and presenting possible evidence is described. Fig. 23 is a diagram illustrating a procedure for outputting the cause of the anomaly and its evidence based on instructions (text input) to the LLM. Other input data to the LLM include SEM images, measurement values ​​output by a measurement device, design data, and / or device information (e.g., error messages and electron microscope probe current values).

[0109] The LLM shown in FIG. 23 includes three models: a cause identification model, a cause analysis model, and a summary model. The computer system 101 or the like is configured to receive data related to problems in the semiconductor manufacturing process from the LLM. The computer system 101 or the like inputs the received text data into the LLM as a cause identification model, thereby receiving data related to the cause of the abnormality from the LLM. The cause analysis model is configured to output recipes and equipment information (electron microscope equipment parameters such as probe current) as evidence. The recipe information includes images obtained as a result of automatic equipment photography, measurement values, and equipment conditions at that time.

[0110] The summary model shown in Fig. 23 is configured to take causes and evidence as input and output text and images. The summary model calculates an evaluation value for the output results using a machine learning model or manual evaluation, and feeds it back to the LLM.

[0111] FIG. 24 is a diagram showing an overview of the LLM for cause identification. It is desirable for the LLM for cause identification to learn by backpropagation based on loss calculations between the output of the LLM and the correct sentence regarding the cause. More specifically, a sentence describing the problem that occurred is input to the LLM, and the cause is output as a sentence (multiple outputs are possible). The error between the correct cause and the output cause is calculated as a loss value, and the result is fed back to the LLM. The sentence may be manually input via the computer system 101 or the like, or automatically input from related data in a database. The loss value may be calculated using any error function, such as an AI model for evaluation or cross-entropy error.

[0112] Figure 25 shows an overview of the LLM for cause analysis. It is desirable for the LLM for cause analysis to learn by backpropagation based on loss calculations between the output of the LLM and related data such as the inspection equipment. More specifically, a sentence describing the cause is input into the LLM, and related data for the inspection equipment used to prove the cause is output (multiple outputs are possible). The correct related data for the inspection equipment is compared with the output data, a loss value is calculated, and the result is fed back to the LLM. The related data for the inspection equipment includes the internal parameters of the inspection equipment, such as the recipe, probe voltage, and current, as well as data related to equipment information.

[0113] In the procedure (system) shown in Fig. 23, tuning is performed to optimize the relationship between a query about an abnormality that can be confirmed in an image and the cause of the abnormality, as well as the relationship between the cause and evidence (the second cause that led to the cause (first cause)). By inputting a query about an abnormality into one or more models tuned in this way, it is possible to obtain a response that includes evidence.

[0114] 24 and 25 may be the same model or different models. As will be described in the seventh embodiment, tuning is performed using both the text data and the related information on the output of the measurement and inspection device (information on the evidence that contributed to the cause), thereby enabling appropriate learning by experts who manage the semiconductor process from a bird's-eye view.

[0115] 26 is a diagram showing the reinforcement learning procedure for the cause identification model and the cause analysis model. Specifically, the cause identification model is configured to input a sentence describing the problem that has occurred and output a sentence describing the cause. The cause analysis model inputs the sentence output from the cause identification model, outputs inspection-related data, and automatically operates the inspection equipment based on the data.

[0116] The computer system 101 or the like evaluates data output from an inspection device or the like and provides feedback to the cause identification model and the cause analysis model. Data output from the inspection device includes images and internal device information. In the case of images, index values ​​such as the measurement value and sharpness of the circuit pattern contained in the image, and the noise level of the image are used as evaluation values. Internal device information may include information such as probe current values ​​and voltages, and these information may be used as evaluation values. Reinforcement learning using the evaluation results as rewards enables optimization of each model.

[0117] In the fine-tuning of the LLM and the reinforcement learning phase using the actual equipment, SEM images, measurement results, equipment information, material information, process information, process window (focus and dose margins), past abnormality cases, wafer coordinates (positions), design drawings, etc. may be input.

[0118] For example, text data such as "An abnormal image (entered along with the text data) has been found. Please tell me the cause." can be input to the LLM. By creating a composite model that includes a cause identification model and a cause analysis model and can output causes and evidence, it is possible to obtain an answer such as "The cause is 'anomaly in the previous process.' We took photos of the area around the location where the abnormal image occurred, and found the following abnormal images frequently... Based on the abnormal results, we believe..." <Regarding Modifications of the Present Disclosure> The present disclosure is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present disclosure, and it is not necessary to include all of the described configurations. Furthermore, parts of one embodiment can be replaced with the configurations of other embodiments. Furthermore, parts of one embodiment can be added to the configurations of other embodiments. Furthermore, parts of each embodiment can be added, deleted, or replaced with parts of other embodiments.

[0119] In the above embodiments, the learning module 113, the reinforcement learning module 114, and the sentence creation module 111 (or the learning module 221 for the evaluation device LLM, the learning module 231 for the semiconductor LLM, the reinforcement learning module 233, and the sentence creation module 234) can be configured by hardware such as a circuit device that implements these functions, or by software that implements these functions being executed by an arithmetic device such as a CPU (Central Processing Unit).

[0120] In the above embodiments, all or some of the learning module 113, reinforcement learning module 114, and sentence creation module 111 (or the learning module 221 for the evaluation device LLM, the learning module 231 for the semiconductor LLM, the reinforcement learning module 233, and the sentence creation module 234) may be configured as an integrated unit, or these modules may be configured separately and then configured to operate in conjunction with each other. In either case, these modules implement a method of operating the evaluation device with character strings via the LLM.

[0121] 101: Computer system 102: Semiconductor evaluation tool 111: Text creation module 113: Supervised / unsupervised learning module 114: Reinforcement learning module

Claims

1. A method executed by a computer, comprising a step of receiving specification data describing operational specifications of an evaluation device from a pre-trained language model by inputting instructions for the evaluation device to evaluate a semiconductor device, the language model having been fine-tuned in advance using at least one of domain knowledge of the semiconductor device or domain knowledge of the evaluation device.

2. The method according to claim 1, wherein the step of receiving the specification data comprises the steps of: inputting an input string describing instructions for the evaluation device to the language model; and receiving a response to the input string from the language model, wherein the language model has been fine-tuned in advance so that, by learning data describing technical knowledge about the semiconductor device and data describing technical knowledge about the evaluation device, the language model receives a request regarding the semiconductor device or the evaluation device as the input string and outputs an answer to the request as the response, and wherein, in the step of inputting the input string, a string requesting the language model to output data describing instructions for the evaluation device is input as the input string, and, in the step of receiving the response, the language model receives the response based on a result of the fine-tuning, thereby receiving from the language model the response reflecting technical knowledge regarding each of the semiconductor device and the evaluation device as the instruction.

3. The method according to claim 1, further comprising a step of performing the fine-tuning on the language model, wherein in the step of performing the fine-tuning, the language model is made to learn at least one of the following technical knowledge regarding the evaluation device by unsupervised learning: an instruction manual for the evaluation device; and a functional specification of the evaluation device.

4. The method according to claim 1, further comprising a step of performing the fine-tuning on the language model, wherein in the step of performing the fine-tuning, data describing the specifications of the image processing performed by the evaluation device is trained in the language model by unsupervised learning, or data describing the results of the image processing performed by the evaluation device is trained in the language model by supervised learning, as technical knowledge related to the evaluation device.

5. The method according to claim 1, further comprising a step of performing the fine-tuning on the language model, wherein in the step of performing the fine-tuning, the language model is made to learn, by unsupervised learning, at least one of the following technical knowledge relating to the semiconductor device: data describing parameter items that impose constraints in a recipe that specifies the operation of the evaluation apparatus and the contents of those constraints; data describing the specifications of the image processing performed by the evaluation apparatus; and measurement instructions describing the procedure for creating an evaluation recipe that defines the processing to be performed by the evaluation apparatus when evaluating the semiconductor device.

6. The method according to claim 1, further comprising a step of performing the fine tuning on the language model, wherein in the step of performing the fine tuning, the language model is made to learn, by supervised learning, at least one of the following technical knowledge regarding the semiconductor device: a measurement instruction document describing a procedure for creating an evaluation recipe defining a process to be performed by the evaluation equipment when evaluating the semiconductor device; data describing at least one of the shape, material, or manufacturing conditions of the semiconductor device; data describing a layout of a pattern shape formed on the semiconductor device; and data describing an interaction between a material of the semiconductor device and a charged particle beam irradiated to the semiconductor device by the evaluation equipment when evaluating the semiconductor device.

7. The method according to claim 6, wherein the supervised learning is performed by at least one of the following: design data of the semiconductor device is used as an input to the language model, and the measurement instructions are used as an output from the language model; the measurement instructions are used as an input to the language model, and the evaluation recipe is used as an output from the language model; imaging conditions when the evaluation device captures an observation image of the semiconductor device and three-dimensional design data of the semiconductor device, or information on the imaging conditions and a material of the semiconductor device, are used as inputs to the language model, and the observation image of the semiconductor device captured by the evaluation device is used as an output from the language model; and filtering conditions for image processing of the observation image of the semiconductor device acquired by the evaluation device is used as an input to the language model, and the observation image is used as an output from the language model.

8. The method according to claim 1, wherein the language model has been previously trained to receive instructions for the evaluation device as a character string and output specification data describing operational specifications of the evaluation device, and the method further comprises the steps of: re-training the language model using at least one of output data output by the evaluation device or text data generated based on the output data as training data; inputting, to the re-trained language model, a character string in a natural language that instructs the evaluation device to evaluate the semiconductor device; receiving the specification data as output from the language model; operating the evaluation device using the specification data received from the language model to evaluate the semiconductor device; and supplying, to the re-training, the output data or the text data obtained by the step of operating the evaluation device.

9. The method according to claim 8, wherein in the step of performing re-learning, the following are input to the language model as the learning data: a target value of the evaluation result obtained by the evaluation device evaluating the semiconductor device, the specification data, and the evaluation result, and the re-learning is performed so that the language model outputs the specification data that brings the evaluation result closer to the target value.

10. The method according to claim 9, wherein in the step of performing re-learning, at least one of the following is input as the target value or evaluation result: a target value for the throughput of the evaluation device or its evaluation result; a target value for a measurement value obtained by the evaluation device measuring the semiconductor device or its evaluation result; a target value for a feature quantity of an observation image of the semiconductor device acquired by the evaluation device or its evaluation result; or a target value for a defect occurrence probability in the semiconductor device or its evaluation result.

11. The method according to claim 9, wherein in the step of performing the relearning, at least one of the following is input as the specification data: an evaluation recipe that defines the processing to be performed by the evaluation device when evaluating the semiconductor device; a prediction result of the defect position in the semiconductor device; a cause of an abnormality and a remedy for the abnormality in the case where the observation image of the semiconductor device acquired by the evaluation device is abnormal; an image quality improvement model that converts the observation image of the semiconductor device output by the evaluation device into an observation image of higher image quality; and a pattern enhancement model that emphasizes a specific shape pattern contained in the observation image of the semiconductor device output by the evaluation device.

12. The method of claim 9, wherein the language model is configured to output, as the specification data, an evaluation recipe that defines the processing to be performed by the evaluation device when evaluating the semiconductor device, and in the step of performing re-learning, the evaluation recipe output by the language model is input as the specification data, and the target value and the evaluation result used as input when the language model outputs the evaluation recipe are input, thereby causing the language model to learn how the evaluation result changes when the evaluation recipe is changed.

13. The method of claim 9, wherein the language model is configured to output, as an intermediate output, measurement instructions describing procedures for creating an evaluation recipe that defines the processing to be performed by the evaluation equipment when evaluating the semiconductor device, and in the step of performing re-learning, the measurement instructions output by the language model as the intermediate output are input again to the language model, thereby performing the re-learning so that the language model outputs the specification data.

14. The method of claim 2, wherein the language model is configured to learn the frequency of occurrence of character strings in each of the data describing technical knowledge about the semiconductor device and the data describing technical knowledge about the evaluation device, and thereby, when the input character string related to either the technical knowledge about the semiconductor device or the technical knowledge about the evaluation device is input, to reflect at least either the technical knowledge about the semiconductor device or the technical knowledge about the evaluation device in the response.

15. The method according to claim 1, wherein the language model is configured to perform at least one of the following: receive a natural language string and a program as input, and output a result of improving the program in accordance with conditions specified by the natural language string; receive a natural language string as input, and output a program that performs processing specified by the natural language string; receive a natural language string and an image of the semiconductor device as input, and output parameters of the evaluation device suitable for the evaluation device to evaluate the shape pattern of the semiconductor device represented by the image while satisfying the conditions specified by the natural language string; receive a natural language string and design data of the semiconductor device as input, and output parameters of the evaluation device suitable for the evaluation device to evaluate the semiconductor device represented by the design data while satisfying the conditions specified by the natural language string, or measurement instructions describing procedures for creating these parameters; receive a natural language string and an image of an abnormal shape pattern of the semiconductor device as input, and output parameters of the evaluation device suitable for the evaluation device to correct the abnormal shape pattern represented by the image while satisfying the conditions specified by the natural language string.

16. A program defining a method executed by a computer, the program being configured to cause the computer to execute a step of receiving specification data describing operational specifications of an evaluation device from a pre-trained language model by inputting instructions for the evaluation device that evaluates a semiconductor device into the language model, the language model having been fine-tuned in advance using at least one of domain knowledge of the semiconductor device and domain knowledge of the evaluation device.

17. A method executed by a computer, comprising a step of receiving specification data describing operational specifications of a first evaluation device for evaluating semiconductor devices and instructions for a second evaluation device from a pre-trained language model, the specification data describing the operational specifications of the second evaluation device being input to the language model, wherein the language model is fine-tuned by comparison information between output information of the first evaluation device and output information of the second evaluation device.

18. A method executed by a computer, comprising: receiving data on defect locations from a pre-trained language model by inputting information on the layout of a semiconductor device and information instructing prediction of location information of defects occurring on the semiconductor device to the language model; and the language model is fine-tuned by comparison information between the data on defect locations output from the language model and data on correct locations of the defects.

19. A method implemented by a computer, comprising the step of receiving data on the cause of an anomaly by inputting instructions to investigate an anomaly in a semiconductor device manufacturing process to one or more pre-trained language models, the language models being fine-tuned with sentence data on the cause of the anomaly and associated data that serves as evidence of the cause.

Images