Learning model generation method, information processing method, computer program, and information processing device

Abstract

Provided are a learning model generation method, an information processing method, a computer program, and an information processing apparatus for realizing prediction of structural parameters based on a reflected light spectrum acquired at the time of substrate processing. In a learning model generation method according to an embodiment, an information processing device performs: a process for acquiring a structure parameter and a reflected light spectrum before and after a change in a state of a substrate caused by a substrate process for changing the state of the substrate; calculating a structure parameter and a reflected light spectrum at a predetermined timing in a change interval on the basis of the structure parameter and the reflected light spectrum before and after the change; and generating, by machine learning using the learning data set, a learning model that receives the reflected light spectrum as an input and outputs a predicted value of the structure parameter, the learning data set includes the structure parameter and the reflected light spectrum before and after the change, and the calculated structure parameter and the reflected light spectrum at the predetermined timing.

Description

Technical Field

[0001] This disclosure relates to methods for generating learning models, information processing methods, computer programs, and information processing apparatus. Background Technology

[0002] Patent Document 1 discloses an etching monitoring device comprising a continuous-wave broadband light source, an illumination system that modulates incident light from the light source by a shutter, and a collection system that collects reflected light reflected from an illumination area on a substrate. The device determines characteristic values ​​based on the processing light that suppresses background light by processing the reflected light, and controls the etching process based on the determined characteristic values.

[0003] Patent Document 1: Japanese Patent Publication No. 2020-517093 Summary of the Invention

[0004] This disclosure provides a method for generating a learning model based on the prediction of structural parameters obtained during substrate processing, an information processing method, a computer program, and an information processing apparatus.

[0005] In one embodiment of the method for generating a learning model, an information processing apparatus performs the following processing: acquiring structural parameters and reflected light spectra of the substrate before and after a change in the state of the substrate caused by substrate processing that changes the state of the substrate; calculating structural parameters and reflected light spectra at a predetermined time within a change range based on the acquired structural parameters and reflected light spectra before and after the change; and generating a learning model that takes the reflected light spectra as input and outputs predicted values ​​of structural parameters by using machine learning on a learning dataset, wherein the learning dataset includes the structural parameters and reflected light spectra before and after the change, and the calculated structural parameters and reflected light spectra at the predetermined time.

[0006] According to this disclosure, it is expected that predictions based on structural parameters obtained during substrate processing using reflected light spectra can be achieved. Attached Figure Description

[0007] Figure 1 This is a schematic diagram used to illustrate the general outline of the information processing system involved in this embodiment.

[0008] Figure 2 This is a schematic diagram used to illustrate the general structure of the substrate processing apparatus according to this embodiment.

[0009] Figure 3 This is a schematic diagram used to illustrate an example of structural parameters measured by a structural parameter measuring device.

[0010] Figure 4This is a schematic diagram illustrating an example of a reflected light spectrum measured by a spectroreflectometer of a substrate processing apparatus.

[0011] Figure 5 This is a block diagram illustrating an example of the hardware structure of the information processing apparatus according to this embodiment.

[0012] Figure 6 This is a block diagram illustrating an example of the functional structure of the information processing apparatus according to this embodiment.

[0013] Figure 7 This is a schematic diagram illustrating specific examples of intermediate spectrum synthesis and intermediate spectrum selection processes.

[0014] Figure 8 This is a schematic diagram illustrating a specific example of the generative processing of the learning model.

[0015] Figure 9 This is a schematic diagram illustrating a specific example of predictive processing using learning model 5.

[0016] Figure 10 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus according to this embodiment during the learning phase.

[0017] Figure 11 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus according to this embodiment during the prediction phase.

[0018] Figure 12 This is a block diagram illustrating an example of the functional structure of the information processing apparatus according to Embodiment 2.

[0019] Figure 13 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus according to Embodiment 2 during the learning phase.

[0020] Figure 14 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus according to Embodiment 2 during the prediction phase. Detailed Implementation

[0021] Specific examples of the information processing system according to embodiments of this disclosure will be described below with reference to the accompanying drawings. Furthermore, this disclosure is not limited to these examples, as indicated by the claims, and is intended to include all modifications of the same meaning and scope as the claims.

[0022] [Implementation Method 1]

[0023] <System Overview>

[0024] Figure 1This is a schematic diagram used to explain the overview of the information processing system according to this embodiment. The information processing system according to this embodiment is configured to include a substrate processing apparatus 100 and an information processing apparatus 1. The substrate processing apparatus 100 is an apparatus for performing various substrate processing on substrates such as semiconductor wafers, such as CVD (Chemical Vapor Deposition), sputtering, or etching. The information processing apparatus 1 is an apparatus for monitoring and controlling the operation of the substrate processing apparatus 100. For example, the information processing apparatus 1 acquires information measured by measuring devices or sensors provided with the substrate processing apparatus 100, and controls the operation of the substrate processing apparatus 100 based on the acquired information, thereby enabling the substrate processing apparatus 100 to perform various substrate processing.

[0025] In the information processing system of this embodiment, the information processing device 1 uses a learning model 5, or so-called AI (Artificial Intelligence), obtained through pre-processed machine learning, to monitor and control the substrate processing device 100. Therefore, the processing performed by the information processing system of this embodiment is roughly divided into two stages: a learning stage, in which information for machine learning is collected and a learning model 5 is generated; and a prediction stage, in which the substrate processing device 100 is monitored and controlled based on predictions made using the generated learning model 5. Figure 1 The preceding paragraph briefly illustrates the structure of the information processing system in the learning phase of generative learning model 5. Figure 1 The following paragraph briefly illustrates the structure of the information processing system in the prediction phase using learning model 5.

[0026] In the learning phase of generative learning model 5, a structural parameter measuring device 140 is used in the information processing system according to this embodiment. The structural parameter measuring device 140 is a device for measuring the structural parameters of a substrate processed by the substrate processing apparatus 100. The structural parameter measuring device 140 measures the structural parameters of the substrate before substrate processing by the substrate processing apparatus 100 and the structural parameters of the substrate after substrate processing by the substrate processing apparatus 100. The structural parameters of the substrate measured by the structural parameter measuring device 140 are index values ​​that quantify the state of the substrate, and may include, for example, the etching depth and the thickness of the etched film.

[0027] Furthermore, in the information processing system according to this embodiment, the substrate processing apparatus 100 irradiates the substrate to be processed with light from a light source and measures the reflected light spectrum obtained by spectral analysis of the reflected light. The substrate processing apparatus 100 can measure the reflected light spectrum at any time during substrate processing, before and after the start of substrate processing, and continuously and repeatedly measure the reflected light spectrum during substrate processing.

[0028] The information processing device 1 acquires the reflected light spectrum repeatedly measured by the substrate processing device 100, and acquires the structural parameters of the substrate measured by the structural parameter measuring device 140 before and after substrate processing. It then stores this information in the substrate processing DB (database) in a corresponding manner, such as the ID attached to the substrate and the date and time of substrate processing.

[0029] In the field of substrate manufacturing, substrate monitoring techniques are known that monitor the state of the substrate during substrate processing, for example, by measuring the reflected light spectrum using a spectrophotometer. According to this substrate monitoring technique, for example, by pre-learning the reflected light spectrum representing the state of the substrate at the end of substrate processing, the end timing of substrate processing can be determined based on the reflected light spectrum. On the other hand, if it is possible not only to predict the state of the substrate at the end of substrate processing, but also to predict the state of the substrate at any time between the start and end of substrate processing, it is expected that substrate processing can be controlled in real time according to the state of the substrate. However, the state of the substrate does not necessarily change at the same rate; in order to predict the state of the substrate at any time, it is necessary to pre-learn the relationship between the reflected light spectrum and the state of the substrate at each time point.

[0030] To perform such learning, operations such as stopping substrate processing at various time points and removing the substrates from the substrate processing apparatus 100 are required. Furthermore, the structural parameters of each removed substrate need to be measured by the structural parameter measuring device 140. Therefore, in generating a learning model to predict the state of a substrate at any given time point, it has been difficult to collect a sufficient amount of training data, thus hindering the achievement of high-accuracy learning models.

[0031] Therefore, in the information processing system according to this embodiment, the information processing device 1 collects the reflected light spectrum of the substrate continuously measured by the substrate processing device 100 during substrate processing, and the structural parameters measured by the structural parameter measuring device 140 before and after substrate processing, and calculates the structural parameters at any time during substrate processing based on the collected information. The information processing device 1 generates a learning dataset that establishes a correspondence between the reflected light spectrum and the structural parameters at any time during substrate processing. The information processing device 1 uses the generated learning dataset to perform machine learning processing, generating a learning model 5 that takes the reflected light spectrum at any time as input and predicts the structural parameters of the substrate to be processed.

[0032] During the prediction phase using the generated learning model 5, the information processing system may or may not use the structural parameter measuring device 140 (however, it may also use the structural parameter measuring device 140). The information processing device 1 acquires the reflected light spectrum continuously measured by the substrate processing device 100, inputs the acquired reflected light spectrum into the learned learning model 5, and acquires the predicted values ​​of the structural parameters output by the learning model 5. Based on the structural parameters predicted by the learning model 5, the information processing device 1 can perform actions such as changing the processing conditions (formula) of the substrate processing device 100 or stopping processing due to abnormalities.

[0033] Furthermore, in this embodiment, the information processing device 1 that performs the processing of generating the learning model 5 through machine learning in the learning phase and the information processing device 1 that performs the processing of predicting structural parameters at arbitrary timings during substrate processing using the learned learning model 5 in the prediction phase will be described as the same device, but it is not limited to this. The information processing device 1 that performs the processing in the learning phase and the information processing device 1 that performs the processing in the prediction phase can also be different devices. For example, the information processing device 1 in the learning phase can be a server device with high computing power, and the information processing device 1 in the prediction phase can be a control device arranged inside or near the substrate processing device 100.

[0034] Furthermore, during the learning phase, the substrate processing device 100 and the structural parameter measuring device 140, which are the objects of the information processing device 1 for collecting information such as reflection spectra and structural parameters, may not be a single device. The information processing device 1 may also collect information from multiple substrate processing devices 100 and structural parameter measuring devices 140. Additionally, during the prediction phase, multiple information processing devices 1 may be used to predict structural parameters. For example, a control device capable of distributing the learning model 5 generated by the information processing device 1 to multiple substrate processing devices 100 may be used, and the multiple control devices may use the learning model 5 to control the substrate processing devices 100 respectively.

[0035] Figure 2 This is a schematic diagram used to explain the general structure of the substrate processing apparatus 100 according to this embodiment. The substrate processing apparatus 100 according to this embodiment is configured to include a spectrophotometer 110, a plasma processing chamber 120, and a control device 130, etc.

[0036] The spectrophotometer 110 is, for example, an apparatus that measures the reflected light from a substrate 160 during plasma etching in a plasma processing chamber 120 by irradiating the substrate 160 with light. The spectrophotometer 110 includes a light source 111, a shutter 112, an irradiation device 113, a light receiving device 114, a beam splitter 115, and an irradiation control device 116. The light source 111 emits light to form an incident ray 117. The shutter 112 modulates the light emitted from the light source 111. The irradiation device 113 forms the incident ray 117 by irradiating the substrate 160 with the light modulated by the shutter 112 through an optical window 121. The light irradiated onto the substrate 160 is reflected on the substrate 160 to form a reflected ray 118. Furthermore, the irradiation device 113 also transmits a portion of the light modulated by the shutter 112 to the beam splitter 115.

[0037] The light-receiving device 114 receives the reflected light 118 formed via the optical window 122. The reflected light 118 received by the light-receiving device 114 is transmitted to the beam-splitting device 115. The beam-splitting device 115 splits the reflected light 118 and measures the reflected light spectrum (the intensity of light at each wavelength). The beam-splitting device 115 outputs the measured reflected light spectrum to the information processing device 1. In addition, the beam-splitting device 115 instructs the illumination control device 116 to increase or decrease the light intensity so that the intensity of the light transmitted from the illumination device 113 reaches a predetermined intensity. The illumination control device 116 controls the operation of the light source 111 and the shutter 112. Furthermore, the illumination control device 116 controls the intensity of the light emitted from the light source 111 based on the instruction from the beam-splitting device 115.

[0038] In the plasma processing chamber 120, substrate processing such as plasma etching is performed on the substrate 160 under specified processing conditions (recipe). The control device 130 controls various operating terminals of the plasma processing chamber 120 based on preset processing conditions (recipe) and commands provided from the information processing device 1, thereby controlling the substrate processing performed in the plasma processing chamber 120.

[0039] <Structural parameters and reflected light spectrum>

[0040] Figure 3 This is a schematic diagram used to illustrate an example of structural parameters measured by the structural parameter measuring device 140. Figure 3 The left side shows an example of the cross-sectional shape and structural parameters of the substrate 160 before substrate processing begins. Figure 3 The right side shows an example of the cross-sectional shape and structural parameters of the substrate 160 after plasma etching. In this embodiment, the structural parameter measuring device 140 measures structural parameters such as the etching depth of the substrate, mask CD (critical dimensions), mask thickness, and the thickness of the etched film.

[0041] In the illustrated example, the etching depth, mask CD, mask thickness, and etched film thickness are measured by the structural parameter measuring device 140 before the substrate processing begins, as follows.

[0042] • Etching depth = 0,

[0043] • Mask CD = CDmask - in,

[0044] • Mask thickness = dmask - in,

[0045] • Etched film thickness = d1

[0046] Similarly, the etching depth, mask CD, mask thickness, and etched film thickness measured by the structural parameter measuring device 140 after the substrate processing are as follows.

[0047] • Etching depth = dout

[0048] • Mask CD = CDmask - out

[0049] • Mask thickness = dmask - out

[0050] • Etched film thickness = d1 - dout

[0051] Figure 4 This is a schematic diagram illustrating an example of a reflected light spectrum measured by a spectroreflectometer 110 of a substrate processing apparatus 100. In this example, a reflected light spectrum measured by the spectroreflectometer 110 and output from the spectrometer 115, and acquired by the information processing apparatus 1, is shown during the measurement period from before the start of substrate processing to after the end of substrate processing. Figure 4 Chart 220 shown on the left is a graph with wavelength on the horizontal axis and substrate processing time (etching processing time) on the vertical axis. Different colors in Chart 220 represent different light intensities at different wavelengths at different times. Additionally, Figure 4 The graph 230 shown on the right is a graph with wavelength as the horizontal axis and light intensity as the vertical axis. The reflected light spectrum 231 before the substrate processing begins and the reflected light spectrum 232 after the substrate processing ends, shown in the graph 220, are represented by continuous curves.

[0052] like Figure 3 and Figure 4 As shown, the information processing apparatus 1 according to this embodiment can collect the following information as a learning dataset:

[0053] • Structural parameters and reflected light spectrum of substrate 160 before substrate processing begins; and

[0054] • Structural parameters and reflected light spectrum of substrate 160 after substrate processing.

[0055] On the other hand, while reflected light spectra can be collected during the measurement interval from before the start of substrate processing to after the end of substrate processing, it is difficult to collect structural parameters. In order to collect structural parameters at each time point during the measurement interval, it is necessary to stop the substrate processing at each time point, remove the substrate 160 from the plasma processing chamber 120, and measure the structural parameters of the removed substrate 160.

[0056] Therefore, the information processing apparatus 1 according to this embodiment uses the structural parameters and reflected light spectra before and after substrate processing to calculate the structural parameters and reflected light spectra at any time within the measurement interval. Furthermore, the information processing apparatus 1 generates a learning dataset and enables the learning model 5 to learn the relationship between the structural parameters and the reflected light spectrum. This learning dataset includes the structural parameters and reflected light spectra before, at any time within the measurement interval, and after the substrate processing. In this way, by calculating the structural parameters and reflected light spectra at any time within the measurement interval, the information processing apparatus 1 can collect a sufficient amount of learning dataset for machine learning of the learning model 5. As a result, the information processing apparatus 1 can predict the structural parameters of the substrate at any time with high accuracy.

[0057] In this embodiment, during the learning phase of generating the learning model 5, the information processing device 1 generates a learning dataset based on the structural parameters measured by the structural parameter measuring device 140 and the reflected light spectrum measured by the spectrometer 115, and generates the learning model 5 by performing machine learning using the generated learning dataset. Furthermore, during the prediction phase utilizing the learning model 5, the information processing device 1 predicts the structural parameters of the substrate 160 of the monitored object using the learned learning model 5, based on the reflected light spectrum measured by the spectrometer 115. Based on the structural parameters predicted by the learning model 5, the information processing device 1, for example, notifies the control device 130 of a stop instruction to stop the plasma etching process. Alternatively, based on the structural parameters predicted by the learning model 5, the information processing device 1 notifies the control device 130 of a change instruction to change the formula.

[0058] <Device Structure>

[0059] Figure 5 This is a block diagram illustrating an example of the hardware structure of the information processing apparatus 1 according to this embodiment. The information processing apparatus 1 according to this embodiment is configured to include a processor 301, a memory 302, an auxiliary storage device 303, an I / F (Interface) device 304, a communication device 305, and a drive device 306, etc. These hardware components of the information processing apparatus 1 are interconnected via a bus 307.

[0060] Processor 301 has various computing devices such as CPU (Central Processing Unit) or GPU (Graphics Processing Unit). Processor 301 reads and executes various programs (e.g., the board monitoring program described later) from memory 302. Memory 302 has main storage devices such as ROM (Read Only Memory) and RAM (Random Access Memory). Processor 301 and memory 302 form a so-called computer, whereby processor 301 executes various programs read from memory 302, thereby enabling the computer to perform various functions.

[0061] Auxiliary storage device 303 stores various programs and various data (e.g., learning datasets) used when the processor 301 executes these programs. I / F device 304 is a connection device that connects the operation device 310 and display device 320 to the information processing device 1. Communication device 305 is a communication device for communicating with the beam splitter 115, control device 130, and structural parameter measuring device 140. Drive device 306 is a device for setting the recording medium 330. The recording medium 330 mentioned here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, floppy disks, and magneto-optical disks. Alternatively, the recording medium 330 may also include semiconductor memories that record information electrically, such as ROMs and flash memory.

[0062] Furthermore, various programs installed on the auxiliary storage device 303 can be installed, for example, by setting the recording medium 330 to be distributed on the drive device 306, and having the drive device 306 read the various programs recorded on the recording medium 330. Alternatively, various programs installed on the auxiliary storage device 303 can also be installed by downloading them from a network via the communication device 305.

[0063] Figure 6 This is a block diagram illustrating an example of the functional structure of the information processing apparatus 1 according to this embodiment. Furthermore, the following will predict the future development of the information processing apparatus 1. Figure 3 The etching depth in the illustrated structural parameters will be explained. As described above, a substrate monitoring program is installed in the information processing device 1, and the processor 301 executes the program, so that the information processing device 1 functions as an intermediate etching depth setting unit 410, an intermediate spectrum synthesis unit 420, an intermediate spectrum selection unit 430, and a learning unit 440 during the learning phase.

[0064] Furthermore, during the learning phase, when the aforementioned functional units are operating, the learning dataset storage unit 470 stores the reflected light spectrum measured by the spectrometer 115 during the measurement interval from before the start to after the plasma etching process (substrate processing). Figure 4 (See Figure 220). Additionally, in the learning dataset storage unit 470, the etching depth measured by the structural parameter measuring device 140 before and after the start of the plasma etching process, and the reflected light spectrum measured by the spectrometer 115 before and after the start of the plasma etching process (see Figure 220). Figure 4 The curves 231 and 232 in Figure 230 are established to correspond to each other and stored as a learning dataset.

[0065] The intermediate etching depth setting unit 410 reads the etching depth before the start of the plasma etching process and the etching depth after the end of the plasma etching process, stored in the learning dataset storage unit 470. The intermediate etching depth setting unit 410 calculates the etching depth at the midpoint between the start and end of the plasma etching process, i.e., the etching depth at the midpoint of the plasma etching process. For example, the etching depth before the start of the plasma etching process is "0", and the etching depth after the end of the plasma etching process is "d". out In the case of "", the intermediate etching depth setting unit 410 calculates the intermediate etching depth using the following formula. The intermediate etching depth setting unit 410 stores the calculated intermediate etching depth as the positive resolution data for the intermediate reflected light spectrum in the learning dataset.

[0066] The etch depth in the middle = dou / 2

[0067] The intermediate spectrum synthesis unit 420, an example of a computing unit, reads the reflected light spectrum 231 before the start of the plasma etching process and the reflected light spectrum 232 after the end of the plasma etching process, which are stored in the learning dataset storage unit 470. Based on the read reflected light spectra 231 and 232, the intermediate spectrum synthesis unit 420 calculates candidates for intermediate reflected light spectra before and after the start and end of the plasma etching process.

[0068] Specifically, the intermediate spectrum synthesis unit 420 first performs scale transformation and parallel shift processing on the reflected light spectrum 231 before the plasma etching process begins, in either or both of the wavelength axis and light intensity axis directions. Similarly, the intermediate spectrum synthesis unit 420 performs scale transformation and parallel shift processing on the reflected light spectrum 232 after the plasma etching process ends, in either or both of the wavelength axis and light intensity axis directions. At this time, the intermediate spectrum synthesis unit 420 calculates the similarity between the two reflected light spectra obtained through the scale transformation and parallel shift processing, and performs scale transformation and parallel shift processing on the two reflected light spectra in opposite directions to maximize the similarity. Furthermore, the intermediate spectrum synthesis unit 420 can calculate, for example, cosine similarity as the similarity between the two reflected light spectra, but is not limited to this; the similarity can be calculated through any calculation.

[0069] Next, the intermediate spectrum synthesis unit 420 calculates the average intensity of the light for each wavelength for the two reflected light spectra obtained through scale transformation and parallel shift processing. The intermediate spectrum synthesis unit 420 then uses the average value of the reflected light spectra obtained therefrom as a candidate for an intermediate reflected light spectrum and notifies the intermediate spectrum selection unit 430 of this value.

[0070] The intermediate spectrum selection unit 430 reads the reflected light spectrum (reference) stored in the learning dataset storage unit 470 during the measurement interval from before the start of the plasma etching process to after the end of the plasma etching process. Figure 4 (See Figure 220). The intermediate spectrum selection unit 430 calculates the correlation or error between multiple reflected light spectra in the read measurement interval and the candidate intermediate reflected light spectrum notified from the intermediate spectrum synthesis unit 420, and selects a reflected light spectrum similar to the candidate intermediate reflected light spectrum from the multiple reflected light spectra in the measurement interval. The intermediate spectrum selection unit 430 uses the selected reflected light spectrum as input data for the intermediate etching depth calculated by the intermediate etching depth setting unit 410 and stores it in the learning dataset.

[0071] The learning unit 440 uses the learning dataset updated by the intermediate etching depth setting unit 410 and the intermediate spectrum selection unit 430 to perform machine learning on the learning model 5. Furthermore, the learning model 5, having completed machine learning through the learning unit 440, is notified to the prediction unit 450.

[0072] On the other hand, the information processing device 1 functions as a prediction unit 450 and a determination unit 460 during the prediction phase. Furthermore, when the aforementioned functional units of the information processing device 1 are activated during the prediction phase, a learning model 5 that has completed learning through the learning unit 440 is set in the prediction unit 450.

[0073] During the substrate manufacturing process, the prediction unit 450 acquires the reflected light spectrum from the spectrometer 115 at predetermined intervals for the substrate under monitoring. The acquired reflected light spectra are sequentially input into the learned model 5, which has already been trained, thereby predicting the etching depth corresponding to each reflected light spectrum. The prediction unit 450 then sequentially inputs the predicted etching depths into the determination unit 460.

[0074] The determination unit 460 determines whether to stop the plasma etching process based on the etching depth input by the prediction unit 450. A stop condition for stopping the plasma etching process is preset in the determination unit 460. The determination unit 460 determines whether the input etching depth meets the stop condition. Furthermore, the stop condition may include a determination timing and a determination item. For example, the determination timing may be set to "the entire range of the measurement interval," and the determination item may be set to "the target value of the etching depth." Additionally, if the determination unit 460 determines that the stop condition is met, it notifies the control device 130 of a stop instruction for the plasma etching process. Thus, the control device 130 can stop the plasma etching process at an appropriate timing based on the etching depth.

[0075] Furthermore, if the information processing device 1 performs the generation process of the learning model 5 but does not perform the prediction process using the learning model 5, the information processing device 1 may not include the prediction unit 450 and the determination unit 460. Similarly, if the information processing device 1 does not perform the generation process of the learning model 5 but performs the prediction process using the learning model 5, the information processing device 1 may not include the intermediate etching depth setting unit 410, the intermediate spectrum synthesis unit 420, the intermediate spectrum selection unit 430, the learning unit 440, and the learning dataset storage unit 470. Information related to the learning model 5 generated by one information processing device 1 (such as the structure and internal parameters of the learning model 5) is provided to other information processing devices 1 via a communication or recording medium. Other information processing devices 1 can reproduce the learning model 5 based on the information provided from one information processing device 1 and perform processes such as prediction and control using the learning model 5.

[0076] <Processing of the learning and prediction phases>

[0077] (1) Intermediate spectrum synthesis and intermediate spectrum selection processing

[0078] Figure 7 This is a schematic diagram illustrating specific examples of intermediate spectrum synthesis and intermediate spectrum selection processes. Figure 7 The first chart shown from top to bottom is... Figure 4The same as Figure 230 shows the reflected light spectrum 231 before the plasma etching process begins and the reflected light spectrum 232 after the plasma etching process ends. The intermediate spectrum synthesis unit 420 reads out these two reflected light spectra 231 and 232.

[0079] The intermediate spectrum synthesis unit 420 performs scale transformation and parallel shift processing on the two read-out reflected light spectra 231 and 232 based on the following equation (1). Furthermore, in equation (1), I is the light intensity, λ is the wavelength, and the reflected light spectrum 231 before the plasma etching process begins is set to (I... incoming , λ incoming The reflected light spectrum 232 after the plasma etching process is set to (I) post - etch , λ post - etch Furthermore, α, β, γ, and δ are coefficients for the specified scale transformation and parallel shift. Additionally, the reflected light spectrum 511 obtained by performing scale transformation and parallel shift processing on the reflected light spectrum 231 before the plasma etching process begins is set as (I' incoming ,λ'incoming), the reflected light spectrum 512 obtained by performing scale transformation and parallel shift processing on the reflected light spectrum 232 after plasma etching is set as (I' post - etch ,λ' post - etch ).

[0080] [Formula 1]

[0081]

[0082] The intermediate spectrum synthesis unit 420 appropriately changes the coefficients α, β, γ and δ in equation (1) and searches for the combination of coefficients α, β, γ and δ that have the highest similarity between the two reflected light spectra 511 and 512 obtained based on equation (1). Figure 7 The second chart from the top shows the case where the similarity of the two reflected light spectra 511 and 512 obtained through scale transformation and parallel shift processing is the greatest.

[0083] Next, the intermediate spectrum synthesis unit 420 calculates the average value of the two reflected light spectra 511 and 512 that have been scaled and paralleled to achieve the highest similarity based on the following equation (2). The intermediate spectrum synthesis unit 420 uses the calculated average value as a candidate 513 for the intermediate reflected light spectrum and notifies the intermediate spectrum selection unit 430. Figure 7 The third chart from the top shows candidate 513 of the middle reflected light spectrum.

[0084] [Formula 2]

[0085]

[0086] The intermediate spectrum selection unit 430 reads multiple reflected light spectra from the learning dataset storage unit 470 during the measurement interval from before the start of the plasma etching process to after the end of the plasma etching process. The intermediate spectrum selection unit 430 calculates the correlation or error between the multiple reflected light spectra read in the measurement interval and the candidate 513 of the intermediate reflected light spectrum notified by the intermediate spectrum synthesis unit 420, and selects a reflected light spectrum 520 similar to the candidate 513 from the multiple reflected light spectra in the measurement interval. Figure 7 The fourth chart from the top shows the candidate intermediate reflected light spectrum 513 and the similar reflected light spectrum 520. The intermediate spectrum selection unit 430 uses the selected reflected light spectrum 520 as input data for the intermediate etching depth calculated by the intermediate etching depth setting unit 410 and stores it in the learning dataset storage unit 470.

[0087] Furthermore, in this embodiment, the case in which data about intermediate time points of the plasma etching process are included in the learning dataset has been described, but it is not limited thereto, and data about arbitrary timing of the plasma etching process can be included in the learning dataset.

[0088] The reflected light spectrum at any given time is obtained by performing a scale transformation and parallel shift process on the reflected light spectrum 231 before the start of the plasma etching process and the reflected light spectrum 232 after the end of the plasma etching process, based on the following equation (3). Furthermore, equation (3) is an extension of equation (1) in a manner corresponding to points other than the intermediate point. By appropriately adjusting the newly introduced variable w, a scale transformation and parallel shift process can be performed to calculate the candidate reflected light spectrum at any given time. Equation (3) becomes equation (1) when the variable w = 0, and the candidate reflected light spectrum at the intermediate point can be calculated. When the variable w > 0, the candidate reflected light spectrum 231 close to the start of the plasma etching process is calculated, and when the variable < 0, the candidate reflected light spectrum 232 close to the end of the plasma etching process is calculated.

[0089] [Formula 3]

[0090]

[0091] Furthermore, for example, if the etching depth before the plasma etching process begins is set to d0, and the etching depth after the plasma etching process ends is set to d1, the etching depth d for any given time can be calculated based on the following formula.

[0092] d = (d1 - d0) × (1 - w) / 2

[0093] Furthermore, the value of variable w should be appropriately set within the range of -1 < w < 1. The closer the value of w is to 1, the shallower the etching depth d; the closer the value of w is to -1, the deeper the etching depth d.

[0094] (2) Generation and processing of learning models

[0095] Figure 8 This is a schematic diagram illustrating a specific example of the generation process of the learning model. The learning dataset 600 shown is stored in the learning dataset storage unit 470 of the information processing device 1. The learning dataset 600 stores "reflected light spectrum before plasma etching process begins," "selected intermediate reflected light spectrum," and "reflected light spectrum after plasma etching process ends" as input data. Additionally, the learning dataset 600 stores "etching depth before plasma etching process begins," "intermediate etching depth," and "etching depth after plasma etching process ends" as forward etch data. "Reflected light spectrum before plasma etching process begins," "etching depth before plasma etching process begins," "selected intermediate reflected light spectrum," "intermediate etching depth," and "reflected light spectrum after plasma etching process ends," and "etching depth after plasma etching process ends," respectively, correspond to the input and output.

[0096] The learning unit 440 has a learning model 5 whose internal model parameters are set to appropriate initial values. The learning model is, for example, a neural network or an SVM (Support Vector Machine) structure, which accepts the reflected light spectrum as input and outputs a predicted value of the etching depth. Furthermore, the learning unit 440 has a comparison / modification unit 602. The comparison / modification unit 602 compares the output data of the learning model 5 based on the reflected light spectrum input with the positive resolution data of the learning dataset 600 corresponding to the input reflected light spectrum, and updates the model parameters of the learning model 5 based on the error between the two sets of data. Thus, the learning unit 440 can perform so-called supervised machine learning using the learning dataset 600 and generate the learning model 5.

[0097] (3) Predictive processing

[0098] Figure 9This is a schematic diagram illustrating a specific example of predictive processing using learning model 5. The prediction unit 450 of the information processing apparatus 1 has a learning model 5 generated by the learning unit 440 through machine learning. For the substrate of the monitored object, the prediction unit 450 acquires the reflected light spectrum from the spectrometer 115 at predetermined intervals and sequentially inputs it into the learning model 5, thereby obtaining a predicted value for the etching depth corresponding to each reflected light spectrum. The prediction unit 450 outputs the predicted etching depth value to the determination unit 460.

[0099] <Flowchart>

[0100] Figure 10 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus 1 according to this embodiment during the learning phase. First, the information processing apparatus 1 according to this embodiment acquires a measurement value of the etching depth before the start of the plasma etching process from the structural parameter measuring device 140 (step S1). Next, the information processing apparatus 1 begins acquiring the reflected light spectrum measured by the spectrometer 115 of the substrate processing apparatus 100 (step S2). Then, the information processing apparatus 1 performs a plasma etching process using the substrate processing apparatus 100 (step S3), during which the acquisition of the reflected light spectrum continues. After the plasma etching process is completed, the information processing apparatus 1 stops acquiring the reflected light spectrum (step S4). Through steps S2 to S4, the information processing apparatus 1 can acquire the reflected light spectrum before the start of the plasma etching process, during the plasma etching process, and after the plasma etching process is completed, and store it in the learning dataset storage unit 470. Next, the information processing apparatus 1 acquires a measurement value of the etching depth after the completion of the plasma etching process from the structural parameter measuring device 140 (step S5). The information processing device 1, which acquires the etching depth and reflected light spectrum before the start of the plasma etching process, the reflected light spectrum during the plasma etching process, and the etching depth and reflected light spectrum after the end of the plasma etching process through steps S1 to S5, stores this information as a learning dataset in the learning dataset storage unit 470 (step S6).

[0101] Next, the intermediate etching depth setting unit 410 of the information processing device 1 calculates the intermediate etching depth based on the etching depth before the plasma etching process begins and the etching depth after the plasma etching process ends (step S7). Additionally, the intermediate spectrum synthesis unit 420 of the information processing device 1 performs scale transformation and parallel shift processing on the reflected light spectrum before the plasma etching process begins and the reflected light spectrum after the plasma etching process ends, calculates the average value of the two reflected light spectra with the highest similarity, and thereby calculates a candidate intermediate reflected light spectrum (step S8). The intermediate spectrum selection unit 430 of the information processing device 1 selects the intermediate reflected light spectrum by choosing a reflected light spectrum similar to the candidate intermediate reflected light spectrum calculated in step S8 from among the multiple reflected light spectra measured during the plasma etching process (step S9). The intermediate spectrum selection unit 430 appends the selected intermediate reflected light spectrum to the learning dataset stored in step S6 (step S10).

[0102] After collecting a sufficient amount of learning dataset, the learning unit 440 of the information processing device 1 performs supervised machine learning using multiple learning datasets stored in the learning dataset storage unit 470 (step S11), and determines the model parameters of the learning model 5, thereby generating the learning model 5. The learning unit 440 stores information such as model parameters related to the generated learning model 5 (step S12) and ends the processing.

[0103] Figure 11 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus 1 according to this embodiment during the prediction phase. First, the determination unit 460 of the information processing apparatus 1 according to this embodiment reads from a memory or the like and sets a pre-stored stop condition (step S21). The prediction unit 450 of the information processing apparatus 1 reads the learning model 5 generated through machine learning (step S22).

[0104] Information processing device 1 begins acquiring the reflected light spectrum from substrate processing apparatus 100 that performs plasma etching processing on the substrate of the monitored object (step S23). Then, information processing device 1 begins plasma etching processing using substrate processing apparatus 100 (step S24). Then, information processing device 1 continues to acquire the measurement results of the reflected light spectrum during plasma etching processing.

[0105] The prediction unit 450 of the information processing device 1 inputs the reflected light spectrum acquired from the substrate processing device 100 to the learning model 5, and acquires the predicted value of the etching depth output by the learning model 5, thereby predicting the etching depth for the reflected light spectrum (step S25). The determination unit 460 of the information processing device 1 determines whether the etching depth predicted in step S25 satisfies the stop condition set in step S21 (step S26). If the stop condition is not met (S26: No), the determination unit 460 returns the processing to step S25. If the stop condition is met (S26: Yes), the determination unit 460 stops the plasma etching process of the substrate processing device 100 (step S27). Then, the information processing device 1 ends the acquisition of the reflected light spectrum from the substrate processing device 100 (step S28) and ends the processing.

[0106] <Summary>

[0107] In the information processing system of this embodiment with the above structure, the information processing device 1 acquires the structural parameters (etching depth) and reflected light spectrum of the substrate before and after the change in the state of the substrate caused by the substrate processing (plasma etching processing) that changes the state of the substrate. Based on the acquired structural parameters and reflected light spectrum before and after the change, the device calculates the structural parameters and reflected light spectrum at a predetermined time within the change range. The device generates a learning model 5 by using machine learning on a learning dataset that takes the reflected light spectrum as input and outputs predicted values ​​of the structural parameters. The learning dataset includes the structural parameters and reflected light spectrum before and after the change, and the calculated structural parameters and reflected light spectrum at the predetermined time.

[0108] In addition, in the information processing system of this embodiment, the information processing device 1 obtains the reflected light spectrum of the substrate of the monitored object from the substrate processing device 100, inputs the obtained reflected light spectrum into the learning model 5 pre-generated by machine learning, and obtains the predicted value of the structural parameters output by the learning model 9, thereby predicting the structural parameters of the target substrate at the time when the reflected light spectrum is measured.

[0109] Therefore, the information processing system according to this embodiment can collect a sufficient amount of learning dataset and perform machine learning on learning model 5, and can be expected to generate learning model 5 with high prediction accuracy. Therefore, it is expected that the information processing system can predict structural parameters (etching depth) at arbitrary timing during substrate processing (plasma etching process).

[0110] [Implementation Method 2]

[0111] In Embodiment 1 described above, the case where the predicted etching depth is used as a structural parameter was explained. However, the structural parameter that can be predicted at any given time is not limited to the etching depth; for example, it could also be the mask CD. Therefore, in Embodiment 2, the case where the predicted mask CD is used as a structural parameter is explained. Furthermore, the following description will focus on the differences from Embodiment 1 described above.

[0112] Figure 12 This is a block diagram illustrating an example of the functional structure of the information processing apparatus 1 according to Embodiment 2. During the learning phase, the information processing apparatus 1 according to Embodiment 2 functions as an intermediate mask CD setting unit 910, an intermediate spectrum synthesis unit 420, an intermediate spectrum selection unit 430, and a learning unit 440. Furthermore, during the learning phase, when the aforementioned functional units are operating, the learning dataset storage unit 970 according to Embodiment 2 stores the reflected light spectrum measured by the spectrometer 115 during the measurement interval from before the start to after the plasma etching process (substrate processing). Additionally, the learning dataset storage unit 470 establishes a correspondence between the mask CD measured by the structural parameter measuring device 140 before and after the start and end of the plasma etching process and the reflected light spectrum measured by the spectrometer 115 before and after the start and end of the plasma etching process, and stores this as a learning dataset.

[0113] The intermediate mask CD setting unit 910 reads the mask CD before the start of the plasma etching process and the mask CD after the end of the plasma etching process, stored in the learning dataset storage unit 970. The intermediate mask CD setting unit 910 calculates the intermediate mask CD before the start of the plasma etching process and after the end of the plasma etching process. For example, if the mask CD before the start of the plasma etching process is "CDmask-in" and the mask CD after the end of the plasma etching process is "CDmask-out", the intermediate mask CD setting unit 910 calculates the intermediate mask CD using the following formula. The intermediate mask CD setting unit 910 stores the calculated intermediate mask CD as the forward resolution data for the intermediate reflected light spectrum in the learning dataset.

[0114] The middle mask CD = (CDmask - in + CDmask - out) / 2

[0115] also, Figure 12 The intermediate spectrum synthesis unit 420, intermediate spectrum selection unit 430, and learning unit 440 used in Embodiment 1 Figure 6 The intermediate spectrum synthesis unit 420, intermediate spectrum selection unit 430, and learning unit 440 described herein are the same, so their descriptions are omitted here.

[0116] On the other hand, the information processing device 1 according to Embodiment 2 functions as a prediction unit 450 and a determination unit 960 during the prediction stage. The function of the prediction unit 450 is the same as that used in Embodiment 1. Figure 6 The prediction section 450 is the same as described above, therefore the description is omitted here. However, Figure 12 The prediction unit 450 shown predicts the mask CD for the substrate of the monitored object by receiving the reflected light spectrum at a predetermined period, and then sequentially inputs the predicted mask CD to the determination unit 960.

[0117] The determination unit 960 determines whether the change conditions for modifying the plasma etching process formula are met based on the mask CD predicted by the prediction unit 450. The information processing device 1 has pre-set change conditions for modifying the plasma etching process formula, and the determination unit 960 determines whether these change conditions are met. Furthermore, the change conditions include a determination timing and a determination item; for example, the determination timing is set to "when the plasma processing starts and immediately after the plasma processing starts," and the determination item is set to "a value outside the allowable range of the mask CD." Additionally, if the determination unit 460 determines that the change conditions are met (for example, if the mask CD input by the prediction unit 450 exceeds the allowable value and becomes outside the allowable range), it outputs an alarm and outputs a change instruction to the control device 130 of the substrate processing apparatus 100 to modify the plasma etching process formula. Thus, the control device 130 can modify the formula in real time during the plasma etching process.

[0118] Figure 13This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus 1 according to Embodiment 2 during the learning phase. First, the information processing apparatus 1 according to Embodiment 1 acquires the measurement value of the mask CD before the start of the plasma etching process from the structural parameter measuring device 140 (step S41). Next, the information processing apparatus 1 begins acquiring the reflected light spectrum measured by the spectrometer 115 of the substrate processing apparatus 100 (step S42). Then, the information processing apparatus 1 performs a plasma etching process using the substrate processing apparatus 100 (step S43), during which the acquisition of the reflected light spectrum continues. After the plasma etching process is completed, the information processing apparatus 1 ends the acquisition of the reflected light spectrum (step S44). Through steps S42 to S44, the information processing apparatus 1 can acquire the reflected light spectrum before the start of the plasma etching process, during the plasma etching process, and after the plasma etching process is completed, and store it in the learning dataset storage unit 470. Next, the information processing apparatus 1 acquires the measurement value of the mask CD after the plasma etching process is completed from the structural parameter measuring device 140 (step S45). The information processing device 1, which acquires the mask CD and reflected light spectrum before the start of the plasma etching process, the reflected light spectrum during the plasma etching process, and the mask CD and reflected light spectrum after the end of the plasma etching process through steps S41 to S45, stores this information as a learning dataset in the learning dataset storage unit 470 (step S46).

[0119] Next, the intermediate mask CD setting unit 910 of the information processing device 1 calculates the intermediate mask CD based on the mask CD before the plasma etching process begins and the mask CD after the plasma etching process ends (step S47). Additionally, the intermediate spectrum synthesis unit 420 of the information processing device 1 performs scale transformation and parallel shift processing on the reflected light spectrum before the plasma etching process begins and the reflected light spectrum after the plasma etching process ends, and calculates the average value of the two reflected light spectra with the highest similarity, thereby calculating a candidate intermediate reflected light spectrum (step S48). The intermediate spectrum selection unit 430 of the information processing device 1 selects an intermediate reflected light spectrum by selecting a reflected light spectrum similar to the candidate intermediate reflected light spectrum calculated in step S48 from multiple reflected light spectra measured during the plasma etching process (step S49). The intermediate spectrum selection unit 430 appends the selected intermediate reflected light spectrum to the learning dataset stored in step S46 (step S50).

[0120] After collecting a sufficient amount of learning dataset, the learning unit 440 of the information processing device 1 performs supervised machine learning (step S51) using multiple learning datasets stored in the learning dataset storage unit 470, and determines the model parameters of the learning model 5, thereby generating the learning model 5. The learning unit 440 stores information such as model parameters related to the generated learning model 5 (step S52) and ends the processing.

[0121] Figure 14 This is a flowchart illustrating an example of the processing steps performed by the information processing apparatus 1 according to Embodiment 2 during the prediction phase. First, the determination unit 460 of the information processing apparatus 1 according to Embodiment 2 reads and sets pre-stored change conditions from a memory or the like (step S61). The prediction unit 450 of the information processing apparatus 1 reads the learned model 5 generated through machine learning (step S62). The information processing apparatus 1 begins acquiring the reflected light spectrum from the substrate processing apparatus 100 that performs plasma etching processing on the substrate of the monitored object (step S63). Then, the information processing apparatus 1 begins plasma etching processing using the substrate processing apparatus 100 (step S64). Then, the information processing apparatus 1 continues to acquire the measurement results of the reflected light spectrum during the plasma etching process.

[0122] The prediction unit 450 of the information processing device 1 inputs the reflected light spectrum obtained from the substrate processing device 100 to the learning model 5, and obtains the predicted value of the mask CD output by the learning model 5, thereby predicting the mask CD for the reflected light spectrum (step S65). The determination unit 460 of the information processing device 1 determines whether the mask CD predicted in step S65 satisfies the change conditions set in step S61 (step S66). Specifically, the information processing device 1 determines whether the mask CD predicted based on the reflected light spectrum measured at the start of the plasma etching process and immediately after the start of the plasma etching process exceeds the allowable value and becomes a value outside the allowable range.

[0123] If the change conditions are met (S66: Yes), the determination unit 460 outputs an alarm and changes the formula related to the plasma etching process performed by the substrate processing apparatus 100 (step S67), causing the process to proceed to step S68. Thereafter, in the substrate processing apparatus 100, plasma etching of the substrate 160 is performed using the changed formula. If the change conditions are not met (S26: No), the determination unit 460 does not change the formula and causes the process to proceed to step S68. After the plasma etching process on the target substrate is completed, the information processing apparatus 1 stops the plasma etching process in the substrate processing apparatus 100 (step S68). Next, the information processing apparatus 1 stops acquiring the reflected light spectrum from the substrate processing apparatus 100 (step S69) and ends the process.

[0124] In the information processing system according to Embodiment 2 of the above structure, the information processing device 1 acquires the mask CD and reflected light spectrum before and after the change in the state of the substrate caused by the plasma etching process that changes the state of the substrate. Based on the acquired mask CD and reflected light spectrum before and after the change, the mask CD and reflected light spectrum at a predetermined time in the change range are calculated. A learning model 5 is generated by using machine learning on a learning dataset that takes the reflected light spectrum as input and outputs a predicted value of the mask CD. The learning dataset includes the mask CD and reflected light spectrum before and after the change, and the calculated mask CD and reflected light spectrum at the predetermined time.

[0125] In addition, in the information processing system according to Embodiment 2, the information processing device 1 obtains the reflected light spectrum of the substrate of the monitored object from the substrate processing device 100, inputs the obtained reflected light spectrum into the learning model 5 pre-generated by machine learning, and obtains the predicted value of the mask CD output by the learning model 9, thereby predicting the mask CD of the target substrate at the time when the reflected light spectrum is measured.

[0126] Therefore, the information processing system according to Embodiment 2 can collect a sufficient amount of learning dataset and perform machine learning on the learning model 5, and can be expected to generate a learning model 5 with high prediction accuracy. Thus, it is expected that the information processing system can predict the mask CD at any timing during the plasma etching process.

[0127] [Other Implementation Methods]

[0128] In the above-described Embodiment 1 or Embodiment 2, the case of predicting the etch depth or mask CD as a structural parameter was described, but the predicted structural parameters are not limited to these. The learning model 5 can also predict structural parameters other than the etch depth or mask CD.

[0129] Furthermore, in Embodiment 1 or Embodiment 2 described above, the calculation of intermediate structural parameters was explained in order to collect a sufficient amount of learning data. However, the newly calculated structural parameters are not limited to intermediate structural parameters; the information processing device 1 can also calculate structural parameters at predetermined timings other than intermediate ones and use them as learning data. Specifically, it is also possible to calculate the structural parameters and reflected light spectrum at a timing that reaches, for example, 1 / 4 × the amount of change, within the range of changes between the state of the substrate before the start of the plasma etching process and the state of the substrate after the end of the plasma etching process. Alternatively, it is also possible to calculate the structural parameters and reflected light spectrum at a timing that reaches, for example, 3 / 4 × the amount of change, within the range of changes between the state of the substrate before the start of the plasma etching process and the state of the substrate after the end of the plasma etching process. That is, it is also possible to calculate structural parameters and reflected light spectra at any timing within the range of changes to generate a learning data set.

[0130] Furthermore, in Embodiment 1 or Embodiment 2 described above, a learning dataset was generated based on the structural parameters and reflected light spectrum before the plasma etching process began, and the structural parameters and reflected light spectrum after the plasma etching process ended. However, the method for generating the learning dataset is not limited to this; the learning dataset can also be generated based on the structural parameters and reflected light spectrum before and after the change in the state of the substrate.

[0131] The terms "before" and "after" in the context of substrate state changes include, for example, "immediately before and immediately after the end of substrate processing" or "before and after the point of change in the multi-layer structure during substrate processing." However, regardless of the specific case, it is predicated on measuring the structural parameters and reflected light spectrum before and after the substrate state change. In this case, the structural parameters and reflected light spectrum at arbitrary timings within the change range are calculated to generate a learning dataset. In the prediction unit 450, the structural parameters are predicted based on the reflected light spectrum measured during substrate processing.

[0132] Furthermore, in Embodiment 1 or Embodiment 2 described above, the cases of stopping the plasma etching process and changing the formula by using the structural parameters predicted by the prediction unit 450 were explained. However, the method of utilizing the structural parameters predicted by the prediction unit 450 is not limited to this, and the information processing device 1 may also use the predicted structural parameters for other control processes. In this case, the determination unit is set with conditions (determination timing, determination item) corresponding to the control process used.

[0133] Furthermore, in Embodiment 1 or Embodiment 2 described above, the calculation method for the correlation or error between the intermediate spectrum selection unit 430 and the candidate intermediate reflected light spectrum was not mentioned, but the calculation method for the correlation or error is arbitrary. For example, the correlation can be calculated using correlation coefficients such as Pearson, Spearman, or Kendall. Additionally, the error can be calculated using index values ​​such as MSE (Mean Squared Error) or MAE (Mean Absolute Error).

[0134] Furthermore, in Embodiment 1 or Embodiment 2 described above, it is explained that the information processing device 1 processes the reflected light spectrum measured by the spectrometer 115. However, the information processing device 1 may also process the reflected light spectrum measured by the spectrometer 115 after processing it. The processing mentioned here includes, for example, standardizing the reflected light spectrum, calculating the difference with the reflected light spectrum used as a reference, reducing the intensity of light at a specific wavelength to zero, and quantizing the characteristics of the reflected light spectrum.

[0135] Furthermore, in the above-described implementation method 1 or implementation method 2, no specific example of learning model 5 is mentioned, but learning model 5 may be a model that uses machine learning algorithms such as principal component regression, partial least squares regression, neural network, support vector machine, random forest regression or gradient boosting regression to perform actions.

[0136] Furthermore, in the above-described embodiment 1 or embodiment 2, it was explained that the information processing system has Figure 1 The system structure shown is shown. However, the system structure of the information processing system is not limited to this. For example, the substrate processing apparatus 100 and the information processing apparatus 1 do not need to be separate devices; the information processing apparatus 1 can also be implemented as a function of the substrate processing apparatus 100, which includes the spectrophotometer 110, the plasma processing chamber 120, and the control device 130. In this case, each functional unit of the information processing apparatus 1 can be implemented in the control device 130.

[0137] Furthermore, in Embodiment 1 or Embodiment 2 described above, the application of the information processing device 1 to the substrate processing apparatus 100 performing plasma etching was explained. However, the application of the information processing device 1 is not limited to the substrate processing apparatus 100 performing plasma etching; it can also be used in substrate processing apparatuses performing substrate processing other than plasma etching. Examples of substrate processing apparatuses performing substrate processing other than plasma etching include substrate processing apparatuses performing film deposition or CMP (Chemical Mechanical Polishing) processes. Furthermore, when the information processing device 1 is applied to a substrate processing apparatus performing film deposition, for example, the film thickness is predicted as a structural parameter. However, the film thickness referred to here can be the thickness of a single-layer film or the thickness of a multilayer film. Additionally, it can be the film thickness when the film is deposited on a substrate (the thickness of a solid film) or the film thickness when the film is deposited on a patterned structure.

[0138] Furthermore, the substrate processed by the substrate processing apparatus 100 described above can contain any structure (pattern). Examples include structures with openings (holes) in the insulating film, structures with grooves (channels), and structures that combine holes and grooves.

[0139] The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of this disclosure is defined by the claims, not by the foregoing, and is intended to include all modifications within the equivalent meaning and scope of the claims.

[0140] The items described in each embodiment can be combined with each other. Furthermore, the independent and dependent claims described in the claims can be combined with each other in all combinations, regardless of their referencing form. Moreover, the claims can be described in a form that refers to claims that refer to two or more other claims (multiple claim form), but are not limited to this. It is also possible to use a form that describes multiple claims that refer to at least one multiple claim (multiple-referencing-multiple-claims).

[0141] Explanation of reference numerals in the attached figures

[0142] 1…Information processing device; 5…Learning model; 100…Substrate processing device; 110…Spectroreflectometer; 111…Light source; 112…Shutter; 113…Illumination device; 114…Light receiving device; 115…Spectroreflection device; 116…Illumination control device; 117…Incident light; 118…Reflected light; 120…Plasma processing chamber; 121, 122…Optical windows; 130…Control device; 140…Structural parameter measuring device; 160…Substrate; 220, 230…Graphics; 301…Processor; 302…Memory ; 303…Auxiliary storage device; 304…I / F device; 305…Communication device; 306…Drive device; 310…Operating device; 320…Display device; 330…Recording medium; 410…Intermediate etching depth setting unit; 420…Intermediate spectrum synthesis unit; 430…Intermediate spectrum selection unit; 440…Learning unit; 450…Prediction unit; 460…Decision unit; 470…Learning dataset storage unit; 600…Learning dataset; 602…Comparison / modification unit; 910…Intermediate mask CD setting unit; 960…Decision unit.

Claims

1. A method for generating a learning model, wherein an information processing device performs the following processing: Obtain structural parameters and reflected light spectrum of the substrate before and after the state change caused by substrate processing that alters the state of the substrate. Based on the obtained structural parameters and reflected light spectra before and after the change, the structural parameters and reflected light spectra at a specified timing within the change range are calculated; and A learning model is generated by using machine learning on a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The learning dataset includes the structural parameters and reflected light spectra before and after the aforementioned changes, as well as the calculated structural parameters and reflected light spectra at the aforementioned specified timing.

2. The method for generating a learning model according to claim 1, wherein, Based on the structural parameters before and after the changes, the structural parameters at the specified timing are calculated.

3. The method for generating a learning model according to claim 1, wherein, Based on the reflected light spectra before and after the aforementioned changes, the candidate reflected light spectra at the specified timing are calculated. The reflected light spectrum at the specified timing is calculated by selecting a reflected light spectrum similar to the calculated candidate spectrum from a plurality of reflected light spectra pre-measured for the substrate processing described above.

4. The method for generating a learning model according to claim 3, wherein, The above-mentioned candidates are calculated by scaling or parallel shifting the reflected light spectrum before and after the above changes relative to any one or both of the wavelength axis and the light intensity axis.

5. The method for generating a learning model according to claim 3, characterized in that, By calculating the correlation or error between the aforementioned multiple reflected light spectra and the reflected light spectra at the aforementioned specified time, a reflected light spectrum similar to the aforementioned candidate spectrum is selected from the aforementioned multiple reflected light spectra.

6. The method for generating a learning model according to claim 1, wherein, The terms "before" and "after" include: before the start of the substrate processing and after the end of the substrate processing, immediately before and immediately after the end of the substrate processing, or before and after the point of change of the multi-layer film in the substrate processing.

7. An information processing method, wherein an information processing device performs the following processing: Obtain the reflected light spectrum of the target substrate; The obtained reflected light spectrum of the aforementioned object substrate is input into a learning model generated by machine learning using a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The dataset for this study includes: structural parameters and reflected light spectra of the substrate before and after the state change caused by substrate processing that changes the state of the substrate, and structural parameters and reflected light spectra at a specified timing within the change range calculated based on the above structural parameters and reflected light spectra before and after the change. as well as By obtaining the predicted values ​​of the structural parameters output by the above learning model, the structural parameters of the above-mentioned object substrate at the time when the reflected light spectrum is measured can be predicted.

8. The information processing method according to claim 7, wherein, Determine whether the predicted structural parameters of the target substrate meet the stopping conditions that would halt the substrate processing. If the above-mentioned stopping conditions are determined to be met, the substrate processing of the target substrate is stopped.

9. The information processing method according to claim 7, wherein, Determine whether the predicted structural parameters of the target substrate meet the change conditions for substrate modification. If the above-mentioned change conditions are met, the substrate processing of the target substrate is changed.

10. A computer program that causes a computer to perform the following processes: Obtain structural parameters and reflected light spectrum of the substrate before and after the state change caused by substrate processing that alters the state of the substrate. Based on the obtained structural parameters and reflected light spectra before and after the change, the structural parameters and reflected light spectra at a specified timing within the change range are calculated; and A learning model is generated by using machine learning on a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The learning dataset includes the structural parameters and reflected light spectra before and after the aforementioned changes, as well as the calculated structural parameters and reflected light spectra at the aforementioned specified timing.

11. A computer program that causes a computer to perform the following processes: Obtain the reflected light spectrum of the target substrate; The obtained reflected light spectrum of the aforementioned object substrate is input into a learning model generated by machine learning using a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The dataset for this study includes: structural parameters and reflected light spectra of the substrate before and after the state change caused by substrate processing that changes the state of the substrate, and structural parameters and reflected light spectra at a specified timing within the change range calculated based on the above structural parameters and reflected light spectra before and after the change. as well as By obtaining the predicted values ​​of the structural parameters output by the above learning model, the structural parameters of the above-mentioned object substrate at the time when the reflected light spectrum is measured can be predicted.

12. An information processing apparatus comprising a processing unit, The above-mentioned processing department performs the following processing: Obtain structural parameters and reflected light spectrum of the substrate before and after the state change caused by substrate processing that alters the state of the substrate. Based on the obtained structural parameters and reflected light spectra before and after the change, the structural parameters and reflected light spectra at a specified timing within the change range are calculated; and A learning model is generated by using machine learning on a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The learning dataset includes the structural parameters and reflected light spectra before and after the aforementioned changes, as well as the calculated structural parameters and reflected light spectra at the aforementioned specified timing.

13. An information processing apparatus comprising a processing unit, The above-mentioned processing department performs the following processing: Obtain the reflected light spectrum of the target substrate; The obtained reflected light spectrum of the aforementioned object substrate is input into a learning model generated by machine learning using a learning dataset, which takes the reflected light spectrum as input and outputs predicted values ​​of structural parameters. The dataset for this study includes: structural parameters and reflected light spectra of the substrate before and after the state change caused by substrate processing that changes the state of the substrate, and structural parameters and reflected light spectra at a specified timing within the change range calculated based on the above structural parameters and reflected light spectra before and after the change. as well as By obtaining the predicted values ​​of the structural parameters output by the above learning model, the structural parameters of the above-mentioned object substrate at the time when the reflected light spectrum is measured can be predicted.

Images