Prediction methods, learning methods, computer programs, information processing devices, processing systems, and methods for manufacturing articles.

Abstract

This technology provides advantages in predicting potential abnormalities in the processing results obtained by processing equipment used to process components. [Solution] A prediction method for predicting that an abnormality may occur in the processing result of a processing device that processes a material includes: a first determination step of determining whether an abnormality may occur in the processing result using a pre-prepared regression model based on device data which is data indicating the state of the processing device that may affect the processing result; a second determination step of determining whether an abnormality may occur in the processing result using a pre-prepared classification model based on the device data; and a third determination step of determining whether an abnormality may occur in the processing result based on the determination result in the first determination step and the determination result in the second determination step.

Description

Technical Field

[0001] The present invention relates to a prediction method, a learning method, a computer program, an information processing apparatus, a processing system, and an article manufacturing method.

Background Art

[0002] In the manufacturing process of articles such as semiconductor devices, requirements for various processes are becoming stricter. For example, in a lithography apparatus, requirements for overlay error are becoming stricter. In order to ensure meeting the requirements for overlay error, after the lithography process, the overlay error can be inspected using an overlay inspection apparatus. If the requirements for overlay error are not met, subsequent processes may not be executed. Alternatively, rework may be performed in which the formed pattern is removed and then the same process is re-executed. When inspecting the overlay error using an overlay inspection apparatus, corresponding time and cost are required therefor. Therefore, it has become mainstream to minimize the inspection target of the overlay error. In recent years, it has also been considered to reduce the time and cost for inspection by predicting the overlay error using machine learning.

[0003] The overlay error can be predicted, for example, using a regression model generated by supervised learning using machine learning. Supervised learning has two phases: a learning phase and a prediction phase. The target variable to be predicted can be the overlay error, and data indicating the state of the lithography apparatus that caused the overlay error can be used as the explanatory variable. In the learning phase, a regression model representing the relationship between the apparatus data and the overlay error is generated, and in the prediction phase, the overlay error is predicted according to the apparatus data and the regression model. The regression model may be a polynomial function or may be SVM, GBDT, CNN, RNN, or the like.

[0004] When predicting overlay error using a regression model, numerical data such as 1.2 nm or 3 nm is predicted. Therefore, in order to determine an anomaly when the overlay error is above a certain value, a threshold of, for example, 10 nm can be set, and an anomaly can be determined if the numerical data exceeds that threshold. However, prediction errors always occur when predicting overlay error using a regression model. Therefore, as a method for setting a threshold that includes prediction errors, there was a method of drawing an ROC curve based on the obtained overlay error value and setting a threshold based on the AUC value to predict anomalies (Patent Document 1). [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] U.S. Patent No. 11687007 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] The method of determining the threshold from the ROC curve can work well when the prediction error of the overlay error is not large. However, this method has the problem that the results are poor when the prediction error of the overlay error is large, because the calculation is based on the predicted value of the overlay error. One example of a situation where the prediction error is large is when the number of data points with large overlay errors is insufficient compared to the number of data points with small overlay errors, such as during initial studies. In such cases, the following problems may arise. In machine learning that generates regression models, the regression model is generated to minimize the difference between the predicted value and the actual value. Therefore, a regression model optimized for actual values ​​with a large number of data points and small overlay errors is generated. Also, if training is performed using actual values ​​with a small number of data points and large overlay errors, the number of data points for training becomes small, making it difficult to improve the accuracy of the regression model. For these reasons, when the number of data points is small, such as during initial studies, even when a large overlay error should be output as the predicted value, a small overlay error may often be output as the predicted value. As a result, the accuracy of anomaly detection decreases, and there is a tendency for anomalies to be missed.

[0007] If many anomalies are missed, unnecessary processing will be performed on the substrate, potentially increasing manufacturing costs. Therefore, even with a small amount of training data, it is necessary to predict the occurrence of anomalies with higher accuracy. Here, we have focused on overlay error as an example of processing results from a lithography device. However, it would also be useful to predict anomalies based on other processing results, such as focus error, CD uniformity (line width uniformity), or placement error of formed features. Furthermore, it would also be useful to predict anomalies that may occur in other processing devices, such as imprint devices and droplet placement devices (printers).

[0008] The present invention aims to provide a technique advantageous for predicting that abnormalities may occur in the processing results of a processing apparatus that processes materials. [Means for solving the problem]

[0009] One aspect of the present invention relates to a prediction method for predicting that an abnormality may occur in the processing result of an apparatus for processing a material, the prediction method comprising: a first determination step of determining whether an abnormality may occur in the processing result using a pre-prepared regression model based on apparatus data which is data indicating the state of the apparatus that may affect the processing result; a second determination step of determining whether an abnormality may occur in the processing result using a pre-prepared classification model based on the apparatus data; and a third determination step of determining whether an abnormality may occur in the processing result based on the determination result in the first determination step and the determination result in the second determination step. [Effects of the Invention]

[0010] According to the present invention, an advantageous technique is provided for predicting that abnormalities may occur in the processing results of a processing apparatus that processes a component. [Brief explanation of the drawing]

[0011] [Figure 1] A diagram illustrating a prediction method for predicting potential abnormalities in the processing results of a processing device that processes components. [Figure 2] This diagram illustrates the learning phase (learning method) used to generate the regression and classification models in the prediction method (prediction phase) shown in Figure 1. [Figure 3] A conceptual diagram to explain overlay errors. [Figure 4] A conceptual diagram to explain focus error. [Figure 5] A diagram illustrating binary classification. [Figure 6] A diagram illustrating multi-class classification. [Figure 7] Figure 1 shows a first modified example of the learning phase (learning method) used to generate the regression and classification models in the prediction method (prediction phase). [Figure 8]A diagram showing an example configuration of a processing system having an information processing system that performs a learning phase and a prediction phase. [Figure 9] A schematic diagram showing an example of the configuration of an exposure apparatus as an example of a processing device. [Figure 10] A schematic diagram showing an example configuration of an imprinting device, another example of a processing device. [Figure 11] A schematic diagram showing an example configuration of a droplet dispensing device, which is yet another example of a processing device. [Modes for carrying out the invention]

[0012] The embodiments will be described in detail below with reference to the attached drawings. Note that the following embodiments do not limit the invention as defined in the claims. While the embodiments describe multiple features, not all of these features are essential to the invention, and the features may be combined in any way. Furthermore, in the attached drawings, identical or similar configurations are given the same reference numerals, and redundant descriptions are omitted.

[0013] The embodiments disclosed below relate to a prediction method for predicting that an abnormality may occur in the processing result by a processing apparatus for processing a component, a computer program for executing the prediction method, and an information processing apparatus, etc. The component may be, for example, a substrate such as a wafer. The material of the component may be, for example, a semiconductor or glass, but may be other materials. The component may have one or more layers on a base material (e.g., a semiconductor substrate or a glass substrate), or it may not have such layers. The one or more layers may include a photoresist film. The processing apparatus may be, for example, an exposure apparatus, an imprint apparatus, a liquid placement apparatus, etc., for manufacturing articles such as semiconductor devices or electronic devices, but may be an apparatus used for other purposes.

[0014] First, an example of applying the prediction method of this embodiment to an exposure apparatus as an example of a processing apparatus will be described. FIG. 9 schematically shows a configuration example of an exposure apparatus 900. Here, the exposure apparatus 900 will be described as a step-and-scan exposure apparatus (scanner) that exposes a substrate 302 while synchronously scanning a reticle stage 902 and a substrate stage 301. However, the exposure apparatus 900 may be configured as a step-and-repeat exposure apparatus (stepper) that exposes the substrate 302 with the reticle stage 902 and the substrate stage 301 stationary.

[0015] The exposure apparatus 900 may include, for example, a light source 907, an illumination optical system 908, a reticle stage 902, a projection optical system 404, a substrate stage 301, a substrate chuck 407, and a control unit 916. The exposure apparatus 900 may also include a laser interferometer system 909 (measuring instrument), a laser interferometer system 307 (measuring instrument), a focus sensor 405, a substrate transfer unit 912, a reticle transfer unit 914, and an alignment scope 915. In FIG. 9, the direction parallel to the optical axis of the projection optical system 404 is defined as the Z-axis direction, and two directions orthogonal to each other within a plane perpendicular to the Z-axis direction are defined as the X-axis direction and the Y-axis direction. The light source 907 may be, for example, any one of a high-pressure mercury lamp, an ArF excimer laser, a KrF excimer laser, and an EUV light source. The light source 907 may be disposed outside a chamber that houses components other than the light source 907 in the exposure apparatus 900, or may be housed in one chamber together with components other than the light source 907. The light emitted from the light source 907 illuminates the reticle 406 via the illumination optical system 908. The reticle 406 may also be referred to as a reticle or a mask. The reticle 406 has a pattern to be transferred onto a substrate 302 coated with a photoresist and is mounted on the reticle stage 902. The pattern of the reticle 406 usually includes a plurality of features (e.g., lines, holes, etc.). The reticle stage 902 holds the reticle 406 via a reticle chuck and is driven by a reticle drive mechanism such as a linear motor.

[0016] The projection optical system 404 exposes the substrate 302 by projecting the pattern of the original plate 406 onto the substrate 302 placed on the substrate chuck 407. Thereby, the pattern of the original plate 406 is transferred to the photoresist applied to the substrate 302. The pattern of the original plate 406 is reduced and projected onto the substrate 302 according to, for example, the projection magnification (e.g., 1 / 4) of the projection optical system 404. The pattern of the original plate 406 is sequentially projected onto a plurality of shot regions of the substrate 302, whereby the plurality of shot regions are sequentially exposed. The substrate stage 301 is movable in the X direction and the Y direction by being driven by a substrate drive mechanism such as a linear motor. The substrate chuck 407 is mounted on the substrate stage 301 and holds the substrate 302. The substrate stage 301 may be configured to position the substrate chuck 407 in the Z-axis direction, θz direction, θx direction, and θy direction. The θz direction is rotation around the Z axis, the θx direction is rotation around the X axis, and the θy direction is rotation around the Y axis. The substrate 302 held by the substrate chuck 407 is driven and positioned by driving the substrate stage 301 and the substrate chuck 407.

[0017] The laser interferometer system 909 may include a plurality of laser interferometers for measuring the positions in the X-axis direction and Y-axis direction of the original plate stage 902 and the attitude (θx, θy, θz) of the original plate stage 902. The laser interferometer system 307 may include a plurality of laser interferometers for measuring the positions in the X-axis direction and Y-axis direction of the substrate stage 301 that holds the substrate 302 and the attitude (θx, θy, θz) of the substrate stage 301. The original plate stage 902 and the substrate stage 301 are controlled in position and attitude by the control unit 916 based on the positions and attitudes measured by the laser interferometer system 909 and the laser interferometer system 307.

[0018] The focus sensor 405 may include a light projection system 405a that projects multiple beams onto the substrate 302 at an oblique incidence, and a light receiving system 405b that receives the multiple beams reflected by the substrate 302. The light receiving system 405b may include a detection unit that detects the multiple beams incident on the light receiving system 405b and provides the detected signals to the control unit 916. The light projection system 405a and the light receiving system 405b may be installed so as to straddle the optical axis of the projection optical system 404. The control unit 916 detects the position of the substrate 302 in the Z-axis direction based on the detection signals detected by the focus sensor 405 and controls the movement of the substrate 302 by the substrate stage 301.

[0019] The substrate transport unit 912 transports the substrate 302. The substrate transport unit 912 can transport the substrate 302 from, for example, a substrate storage container containing the substrate 302 to the substrate stage 301. The substrate transport unit 912 can also transport the substrate 302 from the substrate stage 301 to the substrate storage container. The master plate transport unit 914 transports the master plate 406. The master plate transport unit 914 can transport the master plate 406 from, for example, a master plate storage container containing the master plate 406 to the master plate stage 902. The master plate transport unit 914 can also transport the master plate 406 from the master plate stage 902 to the master plate storage container.

[0020] The alignment scope 915 captures images of marks on the substrate 302 and acquires digital image signals in order to position (align) the substrate 302 held by the substrate chuck 407. The alignment scope 915 may include an image sensor that outputs an image signal corresponding to the light intensity distribution formed by the reflected light from the substrate 302, and an A / D converter that converts the image signal obtained from the image sensor into a digital image signal. The control unit 916 detects the position of the marks on the substrate 302 based on the digital image signals provided by the alignment scope 915, and positions the substrate 302 by controlling the position of the substrate stage 301 based on the result.

[0021] The control unit 916 can control the exposure process of the substrate 302 by controlling each component of the exposure apparatus 900. The control unit 916 is an information processing device that may be composed of, for example, a PLD (Programmable Logic Device) such as an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), a computer with a program installed, or a combination of all or some of these. The control unit 916 may be composed of multiple processors such as CPUs. Furthermore, the control unit 916 may be arranged inside the housing together with the other components of the exposure apparatus 900, or it may be arranged separately from the other components of the exposure apparatus 900.

[0022] The control unit 916 may be configured to perform exposure processing (lithography processing) of the substrate 302 according to processing conditions obtained from a storage device or the like (not shown). Processing conditions applied to the exposure processing may include, for example, transport conditions that define the speed and path when the substrate transport unit 912 transports the substrate 302 to the substrate stage 301. Processing conditions applied to the exposure processing may also include, for example, positioning conditions that define the tolerance when the substrate stage 301 positions the substrate 302. Processing conditions applied to the exposure processing may also include, for example, measurement conditions that define the irradiation time and irradiation timing of multiple beams in the measurement by the focus sensor 405. Processing conditions applied to the exposure processing may also include, for example, exposure conditions that define the identifier of the master plate 406 when the substrate 302 is exposed, the layout of multiple shot areas on the substrate 302, and the illumination mode.

[0023] Figure 1 schematically shows the flow of a prediction method 100 that predicts that an abnormality may occur in the processing result by an exposure apparatus 900, which is a processing apparatus for processing a material. The prediction method 100 may be executed by a control unit 916. The control unit 916 may constitute an information processing apparatus or prediction apparatus that executes the prediction method 100. The prediction method 100 may also be executed by a computer program.

[0024] The prediction method 100 may include a first determination step S10, a second determination step S20, and a third determination step S30. In the first determination step S10, the control unit 916 determines, using a pre-prepared regression model, whether an abnormality may occur in the processing result, based on device data DD, which is data indicating the state of the exposure apparatus 900 that may affect the processing result of the exposure apparatus 900 as a processing device. The explanatory variable of the regression model is the device data DD, and the dependent variable of the regression model is the processing result of the exposure apparatus 900. In the second determination step S20, the control unit 916 determines, using a pre-prepared classification model, whether an abnormality may occur in the processing result of the exposure apparatus 900, based on the device data DD. The explanatory variable of the classification model is the device data DD, and the dependent variable of the classification model is the processing result of the exposure apparatus 900. In the third determination step S30, the control unit 916 determines, based on the determination result in the first determination step S10 and the determination result in the second determination step S20, whether an abnormality may occur in the processing result of the exposure apparatus 900. In the third determination step S30, the control unit 916 may output a determination result PF. The determination result PF may include information indicating that an abnormality may occur in the processing result by the exposure device 900, or it may include information indicating that no abnormality will occur in the processing result by the exposure device 900.

[0025] The first determination step S10 may include a step S11 in which a regression model is used to obtain a predicted value indicating the processing result by the exposure device 900 based on the device data DD. The first determination step S10 may also include a step S12 in which an abnormality may occur in the processing result by the exposure device 900 based on the predicted value and a preset tolerance range. The regression model outputs a predicted value of the processing result determined based on the input device data DD. The tolerance range may be expressed by information such as a threshold indicating the upper limit of the tolerance range, or by information such as a threshold indicating the lower limit of the tolerance range. Alternatively, the tolerance range may be expressed by information such as a first threshold indicating the lower limit of the tolerance range and a second threshold indicating the upper limit of the tolerance range.

[0026] In the third determination step S30, the control unit 916 determines, based on the determination result in the first determination step S10 and the determination result in the second determination step S20, that an abnormality may occur in the processing result by the exposure device 900. Specifically, the control unit 916 may determine that an abnormality may occur in the processing result if at least one of the determination results in the first determination step S10 and the determination result in the second determination step S20 indicates that an abnormality may occur in the processing result by the exposure device 900. The third determination step S30 can also be understood as a step that outputs a logical OR of the determination result in the first determination step S10 and the determination result in the second determination step S20.

[0027] The following will explain more specific examples based on the example where the items to be evaluated as processing results by the exposure apparatus 900 are overlay error and focus error. First, the overlay error will be explained with reference to Figure 3. When a substrate 302 is placed on a substrate stage 301, the actual transfer position 304 of a feature may have errors (shifts) ΔX306 and ΔY305 relative to a target position 303 where a certain feature should be formed. If the target position 303 is a target position based on the feature of the lower layer, ΔX306 and ΔY305 can be understood as alignment errors. If the target position 303 is a target position not based on the feature of the lower layer, ΔX306 and ΔY305 can be understood as placement errors of the transferred feature. ΔX306 is an error in the X-axis direction, and ΔY305 is an error in the Y-axis direction. If the overlay error exceeds the acceptable range (or threshold), the connection between the features of the layer on which the pattern was formed using the exposure device 900 (the exposed layer) and the features of the layer below may become poor, potentially resulting in connection failures.

[0028] The position of the substrate stage 301 is measured using the laser interferometer system 307 (laser interferometer 307x and laser interferometer 307y). Measuring the transfer position 304 in the exposure apparatus 900 is difficult due to various error factors, such as the accuracy of the X-axis bar mirror 308 and Y-axis bar mirror 310 mounted on the substrate stage 301, rotation control errors of the substrate stage 301, and changes in the wavelength of light used in the laser interferometer system 307 due to temperature changes. Therefore, in the learning phase, it is advisable to measure the overlay error after exposure using an overlay inspection device and then use machine learning to determine the relationship between the measured overlay error and the device data. Then, in the prediction phase, it is advisable to predict the overlay error using the device data and its relationship.

[0029] Instrument data that affects overlay error includes, for example, Position data of the original stage 902 and the circuit board stage 301, Control deviation data for the original stage 902 and the substrate stage 301 Synchronization error data for the original stage 902 and the circuit board stage 301. Data indicating the environment in which the exposure apparatus 900 is located (e.g., temperature, humidity, and atmospheric pressure data), Measurement data of marks placed on the circuit board (measured values, mark images, waveforms, waveform evaluation values, etc.), Data indicating the environment in which the exposure device 900 is located during alignment measurement (measurement of the position of marks placed on the substrate) (e.g., temperature, humidity, atmospheric pressure data), or control values ​​in the exposure device 900, or vibrations from the floor, It may include at least one of the following.

[0030] The focus error will be explained below with reference to Figure 4. Figure 4 is a schematic side view showing the components related to the focus of the exposure apparatus 900. The distance between the image plane of the projection optical system 404 and the top surface of the substrate 302 is the focus error ΔZ403. In the exposure apparatus 900, focus control can be performed so that the focus error ΔZ403 falls within the depth of field. The focus error ΔZ403 can be evaluated by observing the substrate 302 using an SEM or the like after exposure.

[0031] Instrument data that affects the focus error ΔZ403 includes, for example, Data showing measurement conditions (light irradiation time, irradiation timing, etc.) from the focus sensor 405. Control deviation of board stage 301, Data indicating the environment in which the exposure device 900 is located during focus measurement (e.g., temperature, humidity, and atmospheric pressure data), Data showing the aberrations of the projection optical system 404 during focus measurement. It may include at least one of the following.

[0032] The items to be evaluated as processing results from the exposure apparatus 900 may include overlay error and focus error, as well as CD uniformity and at least one of feature placement error. These processing results may depend on process information as well as apparatus data. Therefore, a prediction method (prediction phase) may be implemented based on process information as well as apparatus data.

[0033] The following describes an example in which the prediction method of this embodiment is applied to imprinting, which is another example of a processing apparatus. Figure 10 schematically shows an example of the configuration of an imprinting apparatus NIL. The imprinting apparatus NIL is a device that transfers the pattern of a mold M to an imprinting material IM on a substrate S. The imprinting material IM uses a curable composition (sometimes called an uncured resin) that hardens when curing energy is applied. As curing energy, electromagnetic waves, heat, etc., are used. As electromagnetic waves, for example, light such as infrared rays, visible light, ultraviolet rays, etc., whose wavelength is selected from the range of 10 nm to 1 mm. The curable composition is a composition that hardens by irradiation with light or by heating. Of these, the photocurable composition that hardens with light contains at least a polymerizable compound and a photopolymerization initiator, and may optionally contain a non-polymerizable compound or a solvent. The non-polymerizable compound is at least one selected from the group of sensitizers, hydrogen donors, internally added mold release agents, surfactants, antioxidants, polymer components, etc. The imprint material is applied to the substrate in a film-like manner by a spin coater or slit coater. Alternatively, it may be applied to the substrate in droplet form, or in island-like or film-like forms formed by multiple connected droplets, by a liquid spray head. The viscosity of the imprint material (viscosity at 25°C) is, for example, 1 mPa·s to 100 mPa·s.

[0034] The imprint apparatus NIL may include a substrate stage SS including a substrate chuck SC for holding a substrate S, and a substrate drive mechanism SSD for driving the substrate stage SS. The imprint apparatus NIL may also include a mold drive mechanism MD for holding and driving a mold M. The substrate drive mechanism SD and the mold drive mechanism MD constitute a relative drive mechanism that drives at least one of the substrate SD and the mold MD so that the relative position of the substrate S and the mold M is adjusted. The adjustment of the relative position by the relative drive mechanism includes driving the mold M to contact the imprint material IM on the substrate S, and separating the mold IM from the cured imprint material (cured pattern). The adjustment of the relative position by the relative drive mechanism also includes aligning the shot area of ​​the substrate S and the pattern area PR of the mold M. The substrate drive mechanism SSD may be configured to drive the substrate S around a plurality of axes (e.g., three axes: X, Y, and θZ; preferably six axes: X, Y, Z, θX, θY, and θZ). The mold may include a mold deformation mechanism DM that deforms the two-dimensional shape of the pattern region PR of the mold M. The mold deformation mechanism DM can deform the pattern region PR of the mold M by, for example, applying force to the side of the mold M. The mold drive mechanism MD may be configured to drive the mold M along multiple axes (for example, three axes: Z-axis, θX-axis, and θY-axis; preferably six axes: X-axis, Y-axis, Z-axis, θX-axis, θY-axis, and θZ-axis). The imprint apparatus NIL may include a pressure controller CPC that controls the three-dimensional shape of the pattern region PR of the mold M by adjusting the pressure in a sealed space SP formed on the back of the mold M. By adjusting the pressure in the sealed space SP, the pressure controller CPC can deform the pattern region PR of the mold M into a downward convex shape or flatten it.

[0035] The imprint apparatus NIL may include one or more alignment scopes AS for measuring the alignment error between the shot area of ​​the substrate S and the pattern area PR of the mold M. The imprint apparatus NIL may include a curing unit CU for curing the imprint material IM by irradiating it with curing energy through the mold M to form a cured pattern. The imprint apparatus NIL may include a dispenser DP for applying or placing the imprint material IM onto the substrate S. The imprint apparatus NIL may include an off-axis scope OAS for detecting the position of alignment marks on the substrate S. The imprint apparatus NIL may include a control unit CNT for controlling each component of the imprint apparatus NIL. The control unit CNT may consist of, for example, a PLD such as an FPGA, or an ASIC, or a general-purpose or dedicated computer with a program installed, or a combination of all or part of these.

[0036] For cured patterns formed by the NIL imprinting device, reducing the overlay error between the cured pattern and the underlying pattern is also important.

[0037] The device data DD that affects the overlay error is, for example, The alignment error between the shot area of ​​the substrate S and the pattern area PR of the mold M, measured using an alignment scope AS. Data indicating the respective positions of substrate S and mold M. Control data for the pressure controller CPC that controls the three-dimensional shape of the pattern region PR of mold M. Control data for the mold deformation mechanism DM that controls the two-dimensional shape of the pattern region PR of mold M. Control data for the mold drive mechanism MD (e.g., control data related to position and force), Control data for the circuit board chuck SC, Data indicating the environment in which the imprint device NIL is located (e.g., temperature, humidity, and pressure data), It may include at least one of the following.

[0038] The prediction method 100 may include a first determination step S10, a second determination step S20, and a third determination step S30. In the first determination step S10, the control unit CNT may determine, using a pre-prepared regression model, that an abnormality may occur in the processing result, based on device data DD indicating the state of the imprint device NIL that may affect the processing result of the imprint device NIL. The explanatory variable of the regression model is the device data DD, and the dependent variable of the regression model is the processing result of the imprint device NIL. In the second determination step S20, the control unit CNT may determine, using a pre-prepared classification model, that an abnormality may occur in the processing result of the imprint device NIL, based on the device data DD. The explanatory variable of the classification model is the device data DD, and the dependent variable of the classification model is the processing result of the imprint device NIL. In the third determination step S30, the control unit CNT may determine, based on the determination result in the first determination step S10 and the determination result in the second determination step S20, that an abnormality may occur in the processing result of the imprint device NIL. In the third determination step S30, the control unit CNT may output a determination result PF. The determination result PF may include information indicating that an abnormality may occur in the processing result by the imprint device NIL, or it may include information indicating that no abnormality occurs in the processing result by the imprint device NIL.

[0039] The first determination step S10 may include a step S11 that uses a regression model to obtain a predicted value indicating the processing result by the imprint device NIL. The first determination step S10 may also include a step S12 that determines whether an abnormality may occur in the processing result by the imprint device NIL based on the predicted value and a preset tolerance range. The regression model outputs a predicted value of the processing result determined based on the input device data DD. The tolerance range may be represented by information such as a threshold indicating the upper limit of the tolerance range, or by information such as a threshold indicating the lower limit of the tolerance range. Alternatively, the tolerance range may be represented by information such as a first threshold indicating the lower limit of the tolerance range and a second threshold indicating the upper limit of the tolerance range.

[0040] In the third determination step S30, the control unit CNT may determine, based on the determination results in the first determination step S10 and the determination results in the second determination step S20, that an abnormality may occur in the processing result by the imprint device NIL. Specifically, the control unit CNT may determine that an abnormality may occur in the processing result if at least one of the determination results in the first determination step S10 and the determination results in the second determination step S20 indicates that an abnormality may occur in the processing result by the imprint device NIL. The third determination step S30 can also be understood as a step that outputs a logical OR of the determination results in the first determination step S10 and the determination results in the second determination step S20.

[0041] The following describes an example of applying the prediction method of this embodiment to a droplet placement device (printer), which is yet another example of a processing device. Figure 11 schematically shows an example configuration of a droplet placement device PRT. The droplet placement device PRT may be configured to place a functional material on a substrate S by ejecting droplets INK of the functional material from an ejection nozzle NZ. The droplet placement device PRT may include a substrate stage SS including a substrate chuck SC that holds the substrate S, and a substrate drive mechanism SSD that drives the substrate stage SS. The droplet placement device PRT may also include an ejection head DH that ejects droplets INK of the functional material from an ejection nozzle NZ. The ejection head DH may have an element, such as a piezoelectric element PE, that provides energy to the functional material to eject droplets INK from the ejection nozzle NZ. The ejection head DH may also include a detection unit DC that detects the ejection state of droplet INK based on residual vibrations that occur in the ejection nozzle NZ after a drive signal for ejecting droplet INK is given to the piezoelectric element PE (residual vibrations in the ejection nozzle NZ after droplet ejection). A signal indicating residual vibration may be output from the piezoelectric element PE.

[0042] In a droplet placement apparatus (PRT), it is crucial that droplets of a target volume (INK) are placed at the target position on the substrate S. If droplets are placed at a position deviated from the target position on the substrate S, if the volume of the droplets deviates from the target volume, or if droplets are not placed at all, it indicates an abnormal processing result by the PRT. For example, if the control deviation of the substrate stage SS exceeds an acceptable value, or if the discharge timing of the droplets from the discharge nozzle NZ is not synchronized with the control of the substrate S (substrate stage SS), droplets may be placed at a position deviated from the target position on the substrate S. The control deviation of the substrate stage SS can be affected, for example, by the environment in which the droplet placement apparatus (PRT) is installed (e.g., temperature, humidity, atmospheric pressure).

[0043] The device data DD that affects the processing results by the droplet placement device PRT may include, for example, data indicating the detection results by the detection unit DC (residual vibration in the discharge nozzle NZ after droplet discharge). Alternatively, the device data DD that affects the processing results by the droplet placement device PRT may include, for example, Data indicating the position of substrate S (substrate stage SS), Control data for the substrate chuck SC, and, Data indicating the environment in which the droplet placement device (PRT) is installed (e.g., temperature, humidity, and pressure data), It may include at least one of the following.

[0044] The processing device can be various devices other than those described above. This embodiment is particularly useful for processing devices where there is very little abnormal data that can be included in the training data, resulting in data imbalance. Regarding the drive system of the processing device, for example, Measured value of the position of the driven object, Values ​​or control values ​​related to the state of the measuring device, Values ​​related to holding, which are elements that change the shape of the driven object. When multiple objects come into contact with the driven object, contact pressure, If the position of the object being measured is a relative position, the initial position measurement value and, Environmental variables such as temperature, humidity, atmospheric pressure, and vibration affect the state of the measuring device. At least one data point can be used as an explanatory variable, and the processing result as the dependent variable. The processing result could be, for example, positional displacement, or, in the case of a machining tool, the amount of material removed.

[0045] The learning phase is described below. Figure 2 illustrates the learning phase (learning method) that generates the regression and classification models used in the prediction method (prediction phase) shown in Figure 1. The learning phase may also be understood as preparation for executing the prediction method shown in Figure 1, in which case the processing in the learning phase may be understood as constituting a part of the prediction method.

[0046] In the learning phase, first, device data 201 (equivalent to the device data DD mentioned above) and processing result data 202 representing the processing results from the corresponding processing unit (e.g., overlay error, focus error, etc.) are prepared as training data. In the learning phase, a first generation step may be performed in which regression learning 203 is carried out to generate a regression model 205 using device data 201 as explanatory variables and processing result data 202 as the target variable (training data). In the learning phase, a second generation step may be performed in which classification learning 206 is carried out to generate a classification model 207 using device data 201 as explanatory variables and processing result data 202 as the target variable (training data). Here, processing result data 202 may be provided to classification learning 206 through classification target variable conversion 204. Classification target variable conversion 204 is a process that converts processing result data 202, which may have continuous values, into a discrete target variable.

[0047] As an example, in the case of binary classification, there can be two types: within the acceptable range and outside the acceptable range. Figure 5 visually illustrates binary classification. The horizontal axis represents, for example, the overlay error or focus error as a result of processing by the exposure device 900. The acceptable range 510 is the range specified by the lower limit 501=t1 and the upper limit 502=t2. The outside the acceptable range 511 is all areas excluding the acceptable range 510. In this case, if the value of the processing result data 202 in Figure 2 is x, the classification target variable 204 outputs within the acceptable range 510 if t1≦x≦t2, and outside the acceptable range 511 otherwise. Note that in the above, "≦" can also be replaced with "<", so they can be used interchangeably depending on the situation. The same applies to the following notation. The relationship between the device data 201 and the binarized target variable output from the classification target variable 204 is learned in classification learning 206, and a classification model 207 is generated.

[0048] The following provides a specific example of the prediction phase shown in Figure 1, following the example above. In the first determination step S10, the control unit 916 can determine whether an abnormality may occur in the processing result by the exposure device 900, based on the device data DD and using the regression model generated in the learning phase. More specifically, in step S11, the control unit 916 obtains a predicted value indicating the processing result by the exposure device 900, based on the device data DD and using the regression model generated in the learning phase. Then, in step S12, the control unit 916 performs binary classification based on the predicted value and the acceptable range (threshold). Specifically, the control unit 916 determines that if t1 ≤ x ≤ t2, it is within the acceptable range (510), and otherwise it is outside the acceptable range (511).

[0049] On the other hand, in the second determination step S20, the control unit 916 may determine, based on the device data DD and using the classification model generated in the learning phase, that an abnormality may occur in the processing result by the exposure device 900. The classification model outputs information indicating that it is within the acceptable range 510 or outside the acceptable range 511. In the third determination step S30, the control unit 916 may determine that an abnormality has occurred in the processing result by the exposure device 900, based on the determination result in the first determination step S10 and the determination result in the second determination step S20. Specifically, the control unit 916 may determine that an abnormality has occurred in the processing result if at least one of the determination results in the first determination step S10 and the determination result in the second determination step S20 indicates that an abnormality may occur in the processing result by the exposure device 900.

[0050] Classification and regression models may be generated by machine learning methods such as SVM, GBDT, CNN, or RNN.

[0051] Next, we will discuss the tolerance range for anomaly detection. The tolerance range for overlay error can be determined as a threshold for determining whether to proceed to the next step in the manufacturing process. For example, exceeding the threshold means that the misalignment between the lower layer and the layer above it (the current layer) is too large. As a result, the area of ​​the connection becomes smaller, the resistance increases, and when current flows, it may exceed the allowable heat limit. The tolerance range for focus error can be set to within the depth of focus. Exceeding the depth of focus means that the photoresist on the substrate does not react sufficiently, which may cause the dimensions of the transferred features to deviate from the target dimensions, or the features may not be transferred at all. Therefore, by setting the tolerance range as described above, anomaly detection can be performed appropriately.

[0052] Here, regression models are generated so that the output values ​​are as close to the true values ​​as possible, while classification models are generated so that the result of whether or not the value falls within an acceptable range is the best possible. Therefore, the logic by which the models are optimized differs between the two. Thus, by using both regression and classification models in combination to determine anomalies in the processing results, the accuracy of the determination can be improved.

[0053] The following describes the first modified example of the learning phase illustrated in Figure 2. Figure 7 shows the first modified example of the learning phase (learning method) for generating regression and classification models. In the first modified example, step 701 is added to the learning phase illustrated in Figure 2. Step 701 provides the first generation step (regression learning 203) with a second dataset, which is obtained by removing data pairs that do not belong to a predetermined tolerance range from a first dataset containing multiple data pairs, each composed of device data 201 and corresponding processing result data 202.

[0054] In the prediction method shown in Figure 1, process S12 makes a decision based on the predicted value predicted using the regression model and the acceptable range (threshold). Therefore, if the prediction accuracy of the regression model 205 improves, the accuracy of the decision in process S12 also improves. Thus, we consider improving the accuracy of the regression model 205 within the acceptable range. As a way to improve the accuracy when generating a regression model using machine learning, if the target variable becomes too large, the measurement accuracy of the measuring device deteriorates. For example, if the operator that performs A / D conversion only has an 8-bit resolution for the measurement range, the measurement value that can be measured with 1 bit becomes large. Therefore, by setting the measurement standard within the acceptable range, minimizing the measurement value that can be measured, and creating the regression model 205 using data within the acceptable range with improved measurement accuracy, it is possible to use a target variable with high accuracy and generate a regression model with high prediction accuracy.

[0055] Below, we will explain an example of performing multi-class classification instead of binary classification as a second variation of the learning phase shown in Figure 2. Figure 6 illustrates the concept of multi-class classification. First, as in the example in Figure 5, we consider the case where t11≦x≦t21 is within the acceptable range. In this case, the prediction accuracy of the prediction result can be calculated in the learning phase. This prediction accuracy is thought to have an error equal to the prediction accuracy. If this prediction accuracy is Δpred, then the interval between t12 and t11, and / or between t21 and t22, is increased by Δpred. Although this interval is outside the acceptable range, in practice it may be appropriate to consider it as within the acceptable range. Therefore, by setting t12≦x≦t22 as the second acceptable range and widening the initial acceptable range by the error Δpred, it is possible to correct the over-detection of anomalies. Conversely, if there are many missed anomaly detections, if the original acceptable range is from t12 to t22, the acceptable range may be changed to from t11 to t21 as the second acceptable range. In this way, by generating a multi-class classification model and increasing the classification range, it is possible to properly manage false positives and false negatives.

[0056] The following describes an example of a variation of the prediction phase in which the output of the regression model is corrected based on the output of the classification model. When the predicted values ​​from the regression model are used for purposes other than anomaly detection, for example, when feedforward is performed on other devices based on the predicted values, it may be necessary to improve the accuracy of the predicted values ​​from the regression model itself.

[0057] Consider a case where a classification model performs multi-class classification as illustrated in Figure 6, and the predicted value from the regression model is classified by a classification threshold, but the result differs from the prediction by the classification model. For example, if the classification result is judged to be between t12 and t11, and the predicted value from the regression model is judged to be within the range of t11 and t21, then the midpoint value between t12 and t11 ((t12+t11) / 2) may be calculated and used as the predicted value from the regression model to correct it. In this way, by correcting the predicted value from the regression model based on the prediction result from the classification model, the prediction accuracy based on the regression model can be improved.

[0058] Figure 8 shows an example configuration of a processing system 800 having an information processing system 820 that performs the learning and prediction phases described above. The information processing system 820 may have a configuration in which an information device 821 that communicates with a processing device 810 (e.g., an exposure device 900, an imprint device NIL), a judgment device 822 having an anomaly detection function, and an external device 823 are connected via a network 824. The information device 821, the judgment device 822, and the external device 823 can communicate with each other via the network 824. The judgment device 822 may include a computer that performs calculations for generating regression models and classification models, and for performing anomaly detection, a storage device that stores the regression models and classification models, and a storage device that can store the results of the anomaly detection and electronic data of the intermediate calculations up to the anomaly detection. The judgment device 822 may consist of one computer or multiple computers. If the amount of information provided from the information device 821 to the judgment device 822 is large, the information device 821 and the judgment device 822 may be connected by a dedicated communication channel separate from the network 824, which can help suppress the increase in network 824 traffic.

[0059] In the learning phase, device data is provided from the information device 821 to the judgment device 822, and processing results are provided from the external device 823 to the judgment device 822. The judgment device 822 then performs learning and generates a classification model and a regression model. The classification model and regression model can be stored, for example, in a storage unit provided by the judgment device 822 or in an external storage unit. Subsequently, device data to be subjected to abnormality determination is provided from the information device 821 to the judgment device 822, and the judgment device 822 performs abnormality determination using the device data and the classification model and regression model stored in the storage unit. After that, the judgment device 822 provides the determination result to the external device 823 via the network 824, and the external device 823 can decide what to do with the component determined to be abnormal (e.g., discard, rework, etc.) based on the determination result.

[0060] The following describes an article manufacturing method according to one embodiment. The article manufacturing method may be, for example, a manufacturing method for articles such as semiconductor devices, display devices, or MEMS. The article manufacturing method may include a first step of processing a component using a processing system 800, and a second step of further processing the component that has gone through the first step to obtain an article, if the information processing system 820 does not determine that an abnormality has occurred in the first processing step. In the first step, a pattern or film may be formed on a component such as a substrate. In the second step, the component on which the pattern or film is formed may be processed (e.g., etching, dicing, sealing).

[0061] This specification and accompanying drawings include the following disclosures: (Item 1) A prediction method for predicting that an abnormality may occur in the processing result of a processing device that processes a component, A first determination step involves determining whether an abnormality may occur in the processing result, using a pre-prepared regression model based on device data, which is data indicating the state of the processing apparatus that may affect the processing result. A second determination step, based on the aforementioned device data, uses a pre-prepared classification model to determine if an abnormality may occur in the processing result. A prediction method characterized by including a third determination step of determining whether an abnormality may occur in the processing result based on the determination result in the first determination step and the determination result in the second determination step. (Item 2) The first determination step described above is: A step of obtaining a predicted value representing the processing result using the regression model based on the aforementioned device data, A step of determining whether an abnormality may occur in the processing result based on the predicted value and a preset tolerance range, The prediction method according to item 1, characterized by including the following. (Item 3) In the third determination step, if at least one of the determination results in the first determination step and the determination results in the second determination step indicates that an abnormality may occur in the processing result, it is determined that an abnormality may occur in the processing result. The prediction method described in item 1, characterized by the following: (Item 4) A first generation step for generating the regression model, A second generation step for generating the aforementioned classification model, A prediction method according to any one of items 1 to 3, further comprising the following: (Item 5) The first generation step further includes providing the first generation step with a dataset obtained by removing data pairs that do not belong to a predetermined tolerance range from a dataset containing a plurality of data pairs, each consisting of data indicating the state of the processing device and data indicating the processing results by the processing device. The prediction method described in item 4, characterized by the following: (Item 6) The process further includes correcting the output of the regression model based on the output of the classification model. The prediction method described in item 4, characterized by the following: (Item 7) The process further includes determining the acceptable range of the classification model based on the prediction accuracy of the prediction results generated by the classification model in the second generation step, The prediction method described in item 4, characterized by the following: (Item 8) The aforementioned classification model is a multi-class classification model. A prediction method according to any one of items 1 to 7, characterized by the following: (Item 9) The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. In the first and second determination steps, the processing results of the processing apparatus determine that at least one anomaly of overlay error, focus error, CD uniformity, and feature placement error is present. A prediction method according to any one of items 1 to 8, characterized by the following: (Item 10) The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. The data from the aforementioned device is Position data of the master plate stage and the circuit board stage, Control deviation data of the original plate stage and the substrate stage Synchronization error data of the original plate stage and the substrate stage, Temperature, humidity, and atmospheric pressure data of the environment in which the exposure apparatus is located, Measurement data of the marks placed on the substrate, and Data indicating the environment in which the exposure apparatus is located when measuring the position of the marks placed on the substrate, control values ​​in the exposure apparatus, or vibrations from the floor, including at least one of the following: A prediction method according to any one of items 1 to 9, characterized by the following: (Item 11) The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. The data from the aforementioned device is Data showing measurement conditions using the focus sensor. Data showing the control deviation of the substrate stage. Data indicating the environment in which the exposure device was positioned during focus measurement, and Data showing aberrations of the projection optical system during focus measurement. including at least one of the following: A prediction method according to any one of items 1 to 9, characterized by the following: (Item 12) The aforementioned processing apparatus is an imprint apparatus that transfers the pattern of the pattern area of ​​the mold to the imprint material on the substrate. The data from the aforementioned device is Data indicating the respective positions of the substrate and the mold, Control data for controlling the three-dimensional shape of the pattern region of the mold, Control data for controlling the two-dimensional shape of the pattern region of the mold, Control data for the mold drive mechanism that drives the mold, and Control data for the substrate chuck that holds the substrate, including at least one of the following: A prediction method according to any one of items 1 to 9, characterized by the following: (Item 13) The aforementioned processing apparatus is a droplet placement apparatus that places the functional material on a substrate by discharging droplets of the functional material from a discharge nozzle. The aforementioned apparatus data includes data showing residual vibration in the discharge nozzle after droplet discharge. A prediction method according to any one of items 1 to 9, characterized by the following: (Item 14) The prediction method described in any one of items 1 to 13 includes a generation step of generating regression models and classification models by learning, In the generation process, learning is performed using data indicating the state of the processing apparatus for processing the material as the explanatory variable and data indicating the processing result by the processing apparatus as the objective variable. A learning method characterized by the following: (Item 15) A computer program for causing a computer to perform the prediction method described in any one of items 1 through 13. (Item 16) A computer program that causes a computer to execute the learning method described in item 14. (Item 17) An information processing device that controls a processing device for processing components, Based on device data, which is data indicating the state of the processing apparatus that may affect the processing results of the processing apparatus, a first determination is made using a pre-prepared regression model to determine whether an abnormality may occur in the processing results. A second determination is made using a pre-prepared classification model based on the aforementioned device data to determine if an abnormality may occur in the processing result. A third determination is made to determine whether an abnormality may occur in the processing result, based on the results of the first determination and the second determination. An information processing device characterized by performing the following actions. (Item 18) The first determination includes obtaining a predicted value representing the processing result using the regression model, and determining whether an abnormality may occur in the processing result based on the predicted value and a predetermined tolerance range. The information processing device described in item 17, characterized by the features described herein. (Item 19) In the third determination, if at least one of the results of the first determination and the second determination indicates that an abnormality may occur in the processing result, it is determined that an abnormality has occurred in the processing result. The information processing device described in item 18, characterized by the features described herein. (Item 20) A processing system comprising a processing device for processing components and an information processing system for controlling the processing device, The information processing system uses a regression model and a classification model to determine if an abnormality may occur in the processing result, based on device data which is data indicating the state of the processing device that may affect the processing result of the processing device, and controls the processing device based on the result of the determination. A processing system characterized by the following: (Item 21) A method for manufacturing articles, The first step involves processing the components using the processing system described in item 20, If the information processing system does not determine that an abnormality has occurred in the first step, a second step is performed to further process the member that has gone through the first step to obtain an article, A method for manufacturing articles, characterized by including the following: (others) The invention is not limited to the embodiments described above, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, claims are attached to disclose the scope of the invention.

Claims

1. A prediction method for predicting that an abnormality may occur in the processing result of a processing device that processes a component, A first determination step involves determining whether an abnormality may occur in the processing result, using a pre-prepared regression model based on device data, which is data indicating the state of the processing apparatus that may affect the processing result. A second determination step, based on the aforementioned device data, uses a pre-prepared classification model to determine if an abnormality may occur in the processing result. A prediction method characterized by including a third determination step of determining whether an abnormality may occur in the processing result based on the determination result in the first determination step and the determination result in the second determination step.

2. The first determination step is, A step of obtaining a predicted value representing the processing result using the regression model based on the aforementioned device data, A step of determining whether an abnormality may occur in the processing result based on the predicted value and a preset tolerance range, The prediction method according to claim 1, characterized by including the following:

3. In the third determination step, if at least one of the determination results in the first determination step and the determination results in the second determination step indicates that an abnormality may occur in the processing result, it is determined that an abnormality may occur in the processing result. The prediction method according to feature 1.

4. A first generation step for generating the regression model, A second generation step for generating the aforementioned classification model, The prediction method according to claim 1, further comprising the following:

5. The first generation step further includes providing the first generation step with a dataset obtained by removing data pairs that do not belong to a predetermined tolerance range from a dataset containing a plurality of data pairs, each consisting of data indicating the state of the processing device and data indicating the processing results by the processing device. The prediction method according to feature 4.

6. The process further includes correcting the output of the regression model based on the output of the classification model. The prediction method according to feature 4.

7. The process further includes determining the acceptable range of the classification model based on the prediction accuracy of the prediction results generated by the classification model in the second generation step, The prediction method according to feature 4.

8. The aforementioned classification model is a multi-class classification model. The prediction method according to feature 1.

9. The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. In the first and second determination steps, the processing results of the processing apparatus determine at least one anomaly of overlay error, focus error, CD uniformity, and feature placement error. The prediction method according to any one of claims 1 to 8.

10. The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. The data from the aforementioned device is Position data of the master plate stage and the circuit board stage, Control deviation data of the original plate stage and the substrate stage Synchronization error data of the original plate stage and the substrate stage, Temperature, humidity, and atmospheric pressure data of the environment in which the exposure apparatus is located, Measurement data of the marks placed on the substrate, and Data indicating the environment in which the exposure apparatus is located when measuring the position of the marks placed on the substrate, control values ​​in the exposure apparatus, or vibrations from the floor, including at least one of the following: The prediction method according to any one of claims 1 to 8.

11. The aforementioned processing apparatus is an exposure apparatus that exposes a substrate by projecting the pattern of a master plate onto the substrate using a projection optical system. The data from the aforementioned device is Data showing measurement conditions using the focus sensor. Data showing the control deviation of the substrate stage. Data indicating the environment in which the exposure device was positioned during focus measurement, and Data showing aberrations of the projection optical system during focus measurement. including at least one of the following: The prediction method according to any one of claims 1 to 8.

12. The aforementioned processing apparatus is an imprint apparatus that transfers the pattern of the pattern area of ​​the mold to the imprint material on the substrate. The data from the aforementioned device is Data indicating the respective positions of the substrate and the mold, Control data for controlling the three-dimensional shape of the pattern region of the mold, Control data for controlling the two-dimensional shape of the pattern region of the mold, Control data for the mold drive mechanism that drives the mold, and Control data for the substrate chuck that holds the substrate, including at least one of the following: The prediction method according to any one of claims 1 to 8.

13. The aforementioned processing apparatus is a droplet placement apparatus that places the functional material on a substrate by discharging droplets of the functional material from a discharge nozzle. The aforementioned apparatus data includes data showing residual vibration in the discharge nozzle after droplet discharge. The prediction method according to any one of claims 1 to 8.

14. The prediction method according to any one of claims 1 to 8 includes a generation step of generating a regression model and a classification model used in the prediction method by learning, In the generation process, learning is performed using data indicating the state of the processing apparatus for processing the material as the explanatory variable and data indicating the processing result by the processing apparatus as the objective variable. A learning method characterized by the following:

15. A computer program for causing a computer to execute the prediction method described in any one of claims 1 to 8.

16. A computer program for causing a computer to execute the learning method described in claim 14.

17. An information processing device that controls a processing device for processing components, Based on device data, which is data indicating the state of the processing apparatus that may affect the processing results of the processing apparatus, a first determination is made using a pre-prepared regression model to determine whether an abnormality may occur in the processing results. A second determination is made based on the aforementioned device data, using a pre-prepared classification model, to determine if an abnormality may occur in the processing result. A third determination is made to determine whether an abnormality may occur in the processing result, based on the results of the first determination and the second determination. An information processing device characterized by performing the following actions.

18. The first determination includes obtaining a predicted value representing the processing result using the regression model, and determining whether an abnormality may occur in the processing result based on the predicted value and a predetermined tolerance range. The information processing apparatus according to feature 17.

19. In the third determination, if at least one of the results of the first determination and the second determination indicates that an abnormality may occur in the processing result, it is determined that an abnormality has occurred in the processing result. The information processing apparatus according to feature 18.

20. A processing system comprising a processing device for processing components and an information processing system for controlling the processing device, The information processing system uses a regression model and a classification model to determine if an abnormality may occur in the processing result, based on device data which is data indicating the state of the processing device that may affect the processing result of the processing device, and controls the processing device based on the result of the determination. A processing system characterized by the following:

21. A method for manufacturing articles, A first step of processing a member using the processing system described in claim 20, If the information processing system does not determine that an abnormality has occurred in the first step, a second step is performed to further process the member that has gone through the first step to obtain an article, A method for manufacturing articles, characterized by including the following:

Images