Charged particle irradiation device, system, method and program

Abstract

Provided is a charged particle irradiation device that can adapt to changes in charged objects, environments, etc. as quickly as possible and adaptively irradiate charged particles even during operation. Provided is a charged particle irradiation device, comprising: a pre-irradiation state detection unit that detects the charged state of the charged object before irradiation with charged particles, i.e., the pre-irradiation charged state; an irradiation time generation unit that generates an irradiation time based on past irradiation times of charged particles and the charged state of the charged object after the irradiation; an irradiation unit that irradiates the charged object in the pre-irradiation charged state based on the generated irradiation time; a post-irradiation state detection unit that detects the charged state of the charged object after irradiation with charged particles, i.e., the post-irradiation charged state; a machine learning unit that causes a machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state, and the generated irradiation time; and an end determination unit that causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, and the learning unit to repeat operations until the machine learning model meets the learning end condition.

Description

Technical Field

[0001] The present invention relates to, for example, an apparatus for irradiating charged particles such as ions. Background Art

[0002] In recent years, the technology of irradiating charged particles such as ions has become popular in various fields. As an example, there is a static electricity removal device. A static electricity removal device refers to a device that uses corona discharge or the like to irradiate positively or negatively charged ions on a charged object to neutralize the charged charge (for example, Patent Document 1). Static electricity removal devices are used in the manufacturing process of various products such as electronic devices, liquid crystals or semiconductors, and help improve the yield of the target products. In addition, the technology of irradiating charged particles is used in various occasions such as semiconductor processes and thin film processes in addition to static electricity removal devices.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2009-205815 Summary of the invention

[0006] Technical issues to be solved by the invention

[0007] However, in the past, in such devices using the technology of irradiating charged particles, parameters such as the irradiation time of charged particles were fixed during the operation of the device. Therefore, even if the charged state of the object or the environment changes during operation, it is impossible to cope with it, and as a result, it is possible that the object cannot be properly converted to the target charged state. As a result, for example, it is possible that the yield rate is reduced.

[0008] The present invention has been made to solve the above-mentioned technical problems, and its object is to provide a charged particle irradiation device that can adapt to changes in charged objects, environment, etc. as quickly as possible and irradiate charged particles adaptively even during the operation of the device.

[0009] Technical solutions for solving technical problems

[0010] The above technical problems can be solved by a charged particle irradiation device having the following structure.

[0011] That is, the charged particle irradiation device of the present invention irradiates charged particles to a charged object with a predetermined irradiation time, and the charged particle irradiation device comprises: a pre-irradiation state detection unit, which detects the charged state of the charged object before irradiation with the charged particles, i.e., the pre-irradiation charged state; an irradiation time generation unit, which generates an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit, which irradiates the charged object in the pre-irradiation charged state with charged particles based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit, which detects the charged state of the charged object after being irradiated with the charged particles, i.e., the post-irradiation charged state; a machine learning unit, which enables a predetermined machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state and the irradiation time generated by the irradiation time generation unit; and an end determination unit, which enables the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit and the learning unit to repeat the operation until the machine learning model satisfies a predetermined learning end condition.

[0012] According to such a structure, by irradiating charged particles a sufficient number of times for machine learning while learning one by one, a learning model for estimating the irradiation time can be generated quickly and adaptively. That is, a charged particle irradiation device can be provided that can adapt to changes in charged objects, environments, etc. as quickly as possible and irradiate charged particles adaptively. Accordingly, for example, for charged particle irradiation operations, it can be expected to improve efficiency and yield.

[0013] The charged particle irradiation device may further include: an estimation unit, which generates an estimated irradiation time of the charged particles based on the charged state before irradiation, the target charged state after irradiation, and the machine learning model learned by the learning unit; an estimated irradiation unit, which irradiates the charged object with charged particles based on the estimated irradiation time; and a switching unit, which uses the estimated irradiation unit instead of the irradiation unit to irradiate the charged object with charged particles when the learning end condition is met in the end determination unit.

[0014] The irradiation time generation unit may calculate the irradiation time of the charged particles to the charged object by performing an approximation process based on past irradiation times of the charged particles and a charged state of the charged object after the irradiation.

[0015] The approximation process can be performed by first-order approximation or straight-line approximation to the target charging state.

[0016] The approximation process may be performed using a value obtained by biasing the approximation process result regarding the irradiation time.

[0017] The learning end condition may be that the error between the irradiation time as the output of the machine learning model and the irradiation time when the charged state after irradiation becomes the target state or a state close to it is less than a predetermined value or a predetermined ratio.

[0018] The charged particle irradiation device may further include: a change detection unit for detecting changes in the charged object or the environment; and a work start judgment unit for operating the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, the learning unit and the end judgment unit when a predetermined change is detected in the change detection unit.

[0019] The charged state of the charged object may be the potential of the charged object.

[0020] The charged particle irradiation device may perform a predetermined error process when it is determined that the charged state after irradiation is equal to or greater than a predetermined ratio with respect to the charged state before irradiation.

[0021] The error handling may include device stop processing, which is to add points according to the ratio when it is determined that the charged state after irradiation is greater than a predetermined ratio with respect to the charged state before irradiation, and stop the charged particle irradiation device when the points become greater than a predetermined value.

[0022] The error processing may include a process of excluding each of the determined charged states from a target of machine learning in the machine learning unit when it is determined that the charged state after irradiation is equal to or greater than a predetermined ratio with respect to the charged state before irradiation.

[0023] The machine learning model may be a learning model utilizing a tree structure.

[0024] The present invention can be considered as a static electricity removing device. That is, the static electricity removal device of the present invention is a static electricity removal device that irradiates charged particles to a charged object with a predetermined irradiation time, and the static electricity removal device comprises: a pre-irradiation state detection unit that detects the charged state of the charged object before irradiation with the charged particles, that is, the pre-irradiation charged state; an irradiation time generation unit that generates an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit that irradiates the charged object in the pre-irradiation charged state based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit that detects the charged state of the charged object after being irradiated with the charged particles, that is, the post-irradiation charged state; a machine learning unit that causes a predetermined machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state and the irradiation time generated by the irradiation time generation unit; and an end determination unit that causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit and the learning unit to repeat the operation until the machine learning model meets the predetermined learning end condition.

[0025] The present invention can be considered as a charged particle irradiation system. That is, the charged particle irradiation system of the present invention is a charged particle irradiation system for irradiating charged particles to a charged object with a predetermined irradiation time, and the charged particle irradiation system comprises: a pre-irradiation state detection unit, which detects the charged state of the charged object before irradiation with the charged particles, i.e., the pre-irradiation charged state; an irradiation time generation unit, which generates an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit, which irradiates the charged object in the pre-irradiation charged state based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit, which detects the charged state of the charged object after being irradiated with the charged particles, i.e., the post-irradiation charged state; a machine learning unit, which causes a predetermined machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state, and the irradiation time generated by the irradiation time generation unit; and an end determination unit, which causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, and the learning unit to repeat the operation until the machine learning model satisfies a predetermined learning end condition.

[0026] The present invention can also be considered as a charged particle irradiation method. That is, the charged particle irradiation method of the present invention is a charged particle irradiation method for irradiating a charged object with charged particles for a predetermined irradiation time, and comprises: a pre-irradiation state detection step, detecting the charged state of the charged object before irradiation with the charged particles, i.e., the pre-irradiation charged state; an irradiation time generation step, generating the irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation step, irradiating the charged object in the pre-irradiation charged state based on the irradiation time generated by the irradiation time generation step; a post-irradiation state detection step, detecting the charged state of the charged object after being irradiated with the charged particles, i.e., the post-irradiation charged state; a machine learning step, causing a predetermined machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state and the irradiation time generated by the irradiation time generation step; and an end determination step, causing the pre-irradiation state detection step, the irradiation time generation step, the irradiation step, the post-irradiation state detection step and the learning step to repeat until the machine learning model meets the predetermined learning end condition.

[0027] The present invention can be considered as a computer program. That is, the computer program of the present invention enables a computer to function as a charged particle irradiation device, which irradiates charged particles to a charged object for a predetermined irradiation time, and the charged particle irradiation device comprises: a pre-irradiation state detection unit, which detects the charged state of the charged object before irradiation with the charged particles, i.e., the pre-irradiation charged state; an irradiation time generation unit, which generates an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; and an irradiation unit, which generates an irradiation time of the charged object based on the irradiation time generated by the irradiation time generation unit. The charged object in the state is irradiated with charged particles; a post-irradiation state detection unit detects the charged state of the charged object after being irradiated with the charged particles, i.e., the post-irradiation charged state; a machine learning unit causes a predetermined machine learning model to learn the correspondence between the pre-irradiation charged state, the post-irradiation charged state and the irradiation time generated by the irradiation time generation unit; and an end determination unit causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit and the learning unit to repeat the operation until the machine learning model satisfies a predetermined learning end condition.

[0028] Effects of the Invention

[0029] According to the present invention, it is possible to provide a charged particle irradiation device that can adapt to changes in charged objects, environment, etc. as quickly as possible and irradiate charged particles adaptively even during operation of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a block diagram of a static electricity removal device.

[0031] Figure 2 This is a general flow chart of the operation of the static electricity removal device.

[0032] Figure 3 Detailed flowchart for adaptive learning.

[0033] Figure 4 A detailed flowchart of the learning process is shown below.

[0034] Figure 5 This is a conceptual diagram for explaining approximate calculation.

[0035] Figure 6 Detailed flowchart of the error rate calculation process.

[0036] Figure 7 This is an explanatory diagram showing an example of the process of the adaptive learning process.

[0037] Figure 8 This is a flowchart of the static electricity removal process based on the learned model.

[0038] Fig. 9 This is an explanatory diagram showing the amount of data required until learning converges.

[0039] Fig.10 This is an explanatory diagram showing the relationship between the input data to be learned initially and the number of data until learning converges.

[0040] Fig.11 This is an explanatory diagram showing the relationship between the switching of the initial potential and the transition of the error rate.

[0041] Reference numerals

[0042] 1: control unit; 2: power control unit; 3: high voltage power supply; 4: abnormal output unit; 5: compressed air nozzle; 6: corona discharge electrode; 7: surface potentiometer; 8: surface potentiometer; 9: static electricity removal object; 100: static electricity removal device DETAILED DESCRIPTION

[0043] The following refers to the attached Figures 1 to 11 A preferred embodiment of the present invention will be described in detail.

[0044] (1. First Implementation Method)

[0045] In the present embodiment, as an example of a charged particle irradiation device, a static electricity removal device 100 for removing static electricity from an object will be described.

[0046] (1.1 Structure of static electricity removal device)

[0047] Reference Figure 1 The structure of the static electricity removal device 100 according to this embodiment will be described. Figure 1 The block diagram of the static electricity removal device 100 is shown in the figure. As can be seen from the figure, the static electricity removal device 100 uses compressed air from the compressed air nozzle 5 to spray the static electricity removal object 9 and irradiate the positive or negative charged particles, i.e., ions, generated by the corona discharge electrode (needle electrode) 6, thereby electrically neutralizing the static electricity removal object 9 and removing static electricity. In addition, the static electricity removal object is transported at a constant speed or intermittently by a belt conveyor or the like and the static electricity removal process is performed sequentially.

[0048] The static electricity removal device 100 includes a control unit 1, a power supply control unit 2, a high-voltage power supply 3, an abnormal output 4, a compressed air nozzle 5, a corona discharge electrode 6, a surface potentiometer 7 for measuring an initial potential, and a surface potentiometer 8 for measuring a residual potential after static electricity removal treatment.

[0049] The control unit 1 is composed of a CPU and the like, and executes a program to realize the following operation. The control unit 1 generates information about the applied voltage provided to the power supply control unit 2, such as the duty ratio of the voltage, based on the sensor output from the surface potentiometers 7 and 8. In addition, when the control unit 1 detects a predetermined abnormality based on the output of the surface potentiometers 7 and 8, it outputs to the abnormality output unit 4.

[0050] The power control unit 2 controls the high-voltage power supply 3 in accordance with the instruction from the control unit 1. The high-voltage power supply 3 applies voltage to the corona discharge electrode 6 in accordance with the applied voltage generated in the control unit 1. A compressed air nozzle 5 is provided near the corona discharge electrode 6, and compressed air from the compressed air nozzle 5 is ejected toward the static electricity removal object 9, thereby providing the positive or negative charged particles generated at the corona discharge electrode (needle electrode) 6 to the static electricity removal object 9.

[0051] In addition, the structure of the static electricity removal device 100 is not limited to the description of this embodiment. Therefore, for example, each structure can be connected via a network, or multiple structures can be separated or combined into a predetermined device. In addition, part of the work performed by the control unit 1, such as machine learning processing, can also be executed in other devices.

[0052] (1.2 Operation of the static electricity removal device)

[0053] Next, refer to Figure 2 to Figure 8 The operation of the static electricity removal device 100 will be described.

[0054] Figure 2This is a general flow chart of the operation of the static electricity removal device 100. As can be seen from the figure, when the process starts, the static electricity removal device 100 performs a standby process until the static electricity removal object 9 is definitely detected in the charged particle irradiation area of ​​the static electricity removal device 100 (S1 No). In this state, when the static electricity removal object 9 is detected in the charged particle irradiation area (S1 Yes), a determination process is performed to determine whether adaptive learning is required (S2).

[0055] The determination process (S2) of whether adaptive learning is required is a process for determining whether the static electricity removal object 9 or the environment has changed. In the present embodiment, the process is to measure the potential of the static electricity removal object 9 and determine whether the initial potential is the initial potential that has been learned or a value close to it.

[0056] If the result of the determination process of whether adaptive learning is required is that it is determined that adaptive learning is not required (S3: No), the irradiation process of charged particles based on the existing learned model is performed (S7). On the other hand, if it is determined that adaptive learning is required (S3: Yes), after the adaptive learning described later is performed (S5), the irradiation process of charged particles based on the obtained learned model is performed (S7). After these processes, the series of processes (S1 to S7) are repeated again.

[0057] (1.2.1 Adaptive Learning Processing)

[0058] Reference Figure 3 to Figure 7 The details of the adaptive learning (S5) are described.

[0059] Figure 3 Detailed flowchart of adaptive learning. As can be seen from the figure, when the process starts, the static electricity removal device 100 performs a standby process until the static electricity removal object 9 is accurately detected in the charged particle irradiation area of ​​the static electricity removal device 100 (S51 No). In this state, when the static electricity removal object 9 is detected in the charged particle irradiation area (S51 Yes), the process of detecting the initial potential of the static electricity removal object 9 using the surface potentiometer 7 is performed (S52).

[0060] After the initial potential is detected, the following processing is performed (S53): first, the charged particles are irradiated to the static electricity removal object 9 for a predetermined appropriate irradiation time, and then, the charged particles are irradiated to the static electricity removal object 9 for an irradiation time calculated based on the approximate calculation (S68). After the irradiation processing, the residual potential is detected using the surface potentiometer 8 (S55).

[0061] After the residual potential detection process, it is determined whether the residual potential is a normal value (S57). More specifically, when the residual potential is 20% or more of the initial potential (S57: No), it is determined that it is not a normal value.

[0062] When the residual potential is not a normal value (S57: No), a processing is performed to add a number of attention points indicating the possibility of abnormality such as mechanical abnormality (S64). More specifically, when the residual potential is 20% or more and less than 35% of the initial potential, 1 is added to the number of attention points. When the residual potential is 35% or more and less than 50% of the initial potential, 2 is added to the number of attention points. When the residual potential is 50% or more of the initial potential, 3 is added to the number of attention points.

[0063] After that, it is determined whether the number of attention points is greater than a predetermined value, for example, greater than 5 points (S65). If the number of attention points is not greater than the predetermined value (S65 No), no learning process is performed, and the system returns to the start state (S51) to perform a series of processes. On the other hand, if the number of attention points is greater than the predetermined value (S65 Yes), the control of the static electricity removal device 100 is stopped as an abnormal state, and an error signal is sent to the abnormal output unit 4, and the abnormal output unit 4 outputs the meaning of the abnormality to a connected device not shown in the figure, such as a display device, a voice output device, etc. (S66).

[0064] On the other hand, when the residual potential is a normal value (S57: Yes), a learning process is performed to obtain a learned model (S59) that takes the initial potential and the residual potential as inputs and outputs an estimated irradiation time of charged particles.

[0065] Reference Figure 4 Details of the learning process performed in the static electricity removal device 100 will be described.

[0066] Figure 4 Detailed flowchart of the learning process (S59) performed in the control unit 1. As can be seen from the figure, when the learning process starts, a process (S591) of reading parameters of the learning model from the storage unit of the control unit 1 is performed. Here, in this embodiment, a learning tree (refer to Japanese Patent Laid-Open No. 2016-173686 as an example) is used as a machine learning algorithm.

[0067] When learning is performed using a learning tree, input data is accumulated in correspondence with each state space divided hierarchically according to branch conditions. As an estimated output, the output value or output vector corresponding to each data contained in each state space after learning is calculated by taking the arithmetic mean. With such a structure, the learning tree is suitable for sequential learning (online learning).

[0068] When the process of reading the parameters of the learning model such as the number of layers, the number of dimensions or the number of divisions is completed, the process of inputting the initial potential and the residual potential into the learning model is performed (S592). When these inputs are performed, the following process is performed: based on the initial potential and the residual potential, the corresponding nodes are determined from the start node to the end node and the path is determined, and each node on the path is associated to store the input and output data sets (S593). After that, the learning process is completed.

[0069] Back to Figure 3 When the learning process (S59) is completed, it is determined whether the residual potential is the target potential (0 [V]) or is sufficiently close to the target potential (S60). In the case where it is determined that the residual potential is not the target potential or is not sufficiently close to the target potential (S60 No), a process of calculating the irradiation time used in the next time step using a predetermined approximate calculation is performed (S68).

[0070] Figure 5 t_1 is a conceptual diagram for explaining approximate calculation. As can be seen from the figure, when the static electricity removal device 100 starts the process, it irradiates the charged particles for an appropriately determined irradiation time t_1 (S53). Assuming that the result of the irradiation is a predetermined potential V resid When _1 is measured as a residual potential (S55), the residual potential V is calculated using a first-order approximation. resid _1 Approximate irradiation time t until the target potential reaches 0 [V] app _1.

[0071] The control unit 1 calculates the approximate irradiation time t app _1 plus a deviation of about ±20% is calculated as t_2 in the next time step (S68). By repeating such processing (S51 to S60, S68), learning can be quickly converged with a small amount of learning object data as described later.

[0072] At the end of repeating a series of processes including the above learning process (S59) and approximate calculation process (S68), when the residual potential reaches a state consistent with the target value or sufficiently close to the target value (S60 Yes), the error rate calculation process (S62) is performed next.

[0073] Figure 6 Detailed flowchart of the error rate calculation process (S62). As can be seen from the figure, when the process starts, the process of reading the latest learned model is performed (S621). After the reading process, the process of inputting the initial potential and the target (residual) potential to the learned model is performed (S622), and the process of determining the path composed of a series of nodes corresponding to the initial potential and the target potential is performed (S624).

[0074] After that, the control unit 1 performs a process of determining an output node for output from a series of nodes on the path (S625), and performs an estimation process of the irradiation time (S627). Based on the estimated irradiation time and the correct irradiation time when the residual potential is the target value, the error rate [%] is calculated. That is, the error rate is calculated by (estimated irradiation time - correct irradiation time) / correct irradiation time × 100. After calculating the error rate, the process ends.

[0075] Back to Figure 3 If the error rate is greater than a predetermined value, for example, greater than 10% (S63 No), a series of processes including the learning process (S59) and the approximate calculation process (S68) are repeated again. On the other hand, if the error rate is less than a predetermined value (S63 Yes), for example, less than 10%, the process ends.

[0076] Figure 7 1 is an explanatory diagram showing an example of the process of the series of adaptive learning processes described above. In the example of this diagram, the purpose is to make the static electricity removal target object 9 with an initial potential of 20 [V] change to a target potential of 0 [V].

[0077] First, in time step 1, the static electricity removal device 100 irradiates the static electricity removal object 9 with an appropriate irradiation time (= 0.274 [s]). Because the residual potential at this time is 0.8 [V], the first-order approximation using the least squares method is performed based on the residual potential to calculate the approximate irradiation time (= 0.3 [s]). The learning model is machine-learned using the values ​​of the initial potential, residual potential, and irradiation time at this time. After that, when the initial potential is set to 20 [V] and the target (residual) potential is set to 0 [V] to make the learned model perform an estimation process, the estimated irradiation time is calculated to be 0.274.

[0078] Next, in time step 2, the static electricity removal device 100 irradiates the static electricity removal object 9 with a new irradiation time (=0.36[s]) obtained by adding a deviation value of about ±20% to the approximate irradiation time of step 1 (=0.3), that is, adding a value of 20% (=0.06) in this example. Because the residual potential at this time is -0.5[V], the first-order approximation using the least squares method is performed based on the residual potential to calculate the approximate irradiation time (=0.28[s]). The learning model is machine-learned using the initial potential, residual potential, and irradiation time values ​​at this time. When the learned model is estimated by setting the initial potential to 20[V] and the target potential to 0[V], the estimated irradiation time is calculated to be 0.337.

[0079] Furthermore, in time step 3, the static electricity removal device 100 irradiates the static electricity removal object 9 with a new irradiation time (=0.24 [s]) obtained by adding a value of about -14% (=-0.04) to the approximate irradiation time (=0.28) of step 2. Because the residual potential at this time is 0 [V], it is determined that the target potential has been reached. The learning model is machine-learned using the values ​​of the initial potential, residual potential, and irradiation time at this time. When the learned model is then estimated by setting the initial potential to 20 [V] and the target potential to 0 [V], the estimated irradiation time is calculated to be 0.25. According to (0.25-0.24) / 0.24×100, the error rate at this time is calculated to be 4.2%, that is, the learning converges in time step 3.

[0080] According to the above configuration, since learning data is collected using the approximate calculation results, the search efficiency can be improved, thereby enabling rapid learning convergence to be expected.

[0081] Furthermore, according to the above configuration, since learning is performed based on data acquired with a predetermined bias during the approximation process, the search efficiency is further improved, thereby enabling faster learning convergence to be expected.

[0082] (1.2.2 Anti-static treatment based on the learned model)

[0083] Reference Figure 8 The charged particle irradiation process ( S7 ) based on the learned model will be described. Figure 8 This is a flowchart of the static electricity removal process based on the learned model.

[0084] When the process starts, the learning model is read (S71). After the reading, the initial potential of the static electricity removal object 9 is measured (S72). After the measurement, the measured initial potential and target potential are input into the learning model (S73). Based on this, a path consisting of a series of corresponding nodes is determined (S74).

[0085] The process of determining the output node that generates the output from the series of nodes constituting the path is performed (S75), and the irradiation time is estimated based on the output node (S77). Thereafter, the irradiation of charged particles based on the estimated irradiation time is performed (S79), and the process ends.

[0086] According to the above configuration, it is possible to provide a charged particle irradiation device that can adapt to changes in charged objects, environment, etc. as quickly as possible and irradiate charged particles adaptively even during operation of the device.

[0087] (1.3 Experimental results)

[0088] Next, refer to Figures 9 to 11 Various experimental results using the static electricity removal device 100 of this embodiment will be described.

[0089] Fig. 9 This is an explanatory diagram showing the amount of data required until learning converges. Fig. 9 It is shown as follows: the number of data until learning convergence in the learning model when only the usual linear approximation is used and the number of data until learning convergence in the learning model when the linear approximation result has a deviation of ±20% are divided into the case where the initial potential is changed multiple times for learning and the case where the initial potential is fixed at 20 [V] for learning.

[0090] As can be seen from the figure, when the initial potential is changed multiple times, 99 data are required until learning converges when only linear approximation is used, while only 42 data are required when the linear approximation result is allowed to have a deviation of ±20%. Similarly, when the initial potential is set to 20 [V], 43 data are required until learning converges when only linear approximation is used, while only 14 data are required when the linear approximation result is allowed to have a deviation of ±20%.

[0091] That is, it was confirmed that in all cases, when the linear approximation result was allowed to have a deviation of ±20%, learning converged with a smaller amount of data than when only the normal linear approximation was performed.

[0092] In addition, the deviation value is not limited to ±20%, and can be adjusted to approximately ±10% or ±5%.

[0093] Fig.10 This is an explanatory diagram showing the relationship between the input data to be learned initially and the number of data until learning converges when the target potential is set to 0 [V]. According to the figure, when the initial potential is set to 20 [V], the number of data until convergence is 26 in the case of data whose initial learning residual potential changes to 15 [V] in the learning process, the number of data until convergence is 14 in the case of data whose initial learning residual potential changes to 10 [V] in the learning process, the number of data until convergence is 12 in the case of data whose initial learning residual potential changes to 5 [V] in the learning process, and the number of data until convergence is 3 in the case of data whose initial learning residual potential changes to 1 [V] in the learning process.

[0094] That is, from this result, it was confirmed that when data of a residual potential close to the target potential is learned from the beginning of learning, the number of data required until learning converges is small, that is, learning converges quickly.

[0095] Fig.11This is an explanatory diagram showing the relationship between the switching of the initial potential and the change of the error rate. According to the figure, when learning is initially performed so that the initial potential 20 [V] becomes the target potential 0 [V], the error rate decreases as the learning proceeds. Next, when the initial potential is switched to 10 [V] near the time step 28, although the error rate temporarily increases to about 40%, the error rate decreases again as the learning proceeds. Furthermore, when the initial potential is switched to 5 [V] near the time step 65, although the error rate temporarily increases to about 65%, the error rate decreases sharply as the learning proceeds.

[0096] That is, the results confirm that learning is performed to rapidly reduce the error rate even when the potential of the static electricity removal target object 9 changes during the operation of the static electricity removal device 100. In addition, by performing learning once, rapid learning can be expected in subsequent additional learning.

[0097] (2. Modifications)

[0098] In the above embodiment, a learning tree is used as a machine learning model, but the present invention is not limited to such a structure. Therefore, other learning models such as a neural network and a support vector machine (SVM) may also be used.

[0099] In the above embodiment, the first-order approximation is used as an example of the approximation process, but the present invention is not limited to such a structure. Therefore, for example, an exponential function or the like may be used for approximation.

[0100] Industrial Applicability

[0101] The present invention can be utilized at least in the industry of manufacturing charged particle irradiation devices and the like.

Claims

1. A charged particle irradiation device for irradiating a charged object with charged particles for a predetermined irradiation time, the charged particle irradiation device comprising: a pre-irradiation state detection unit for detecting a charged state of the charged object before irradiation with charged particles, i.e., a pre-irradiation charged state; an irradiation time generating unit for generating an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit that irradiates the charged object in the charged state before the irradiation with charged particles based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit for detecting a charged state of the charged object after being irradiated with the charged particles, that is, a post-irradiation charged state; a machine learning unit that causes a predetermined machine learning model to learn the correspondence between the charged state before irradiation, the charged state after irradiation, and the irradiation time generated by the irradiation time generation unit; and The end determination unit causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, and the machine learning unit to repeatedly operate until the machine learning model satisfies a predetermined learning end condition, wherein: The learning end condition is that the error between the irradiation time as the output of the machine learning model and the irradiation time when the charged state after irradiation becomes the target state or a state close to it is less than a predetermined value or a predetermined ratio.

2. The charged particle irradiation device according to claim 1, wherein: The charged particle irradiation device further comprises: an estimating unit that generates an estimated irradiation time of charged particles based on the pre-irradiation charged state, the target post-irradiation charged state, and the machine learning model learned by the machine learning unit; an estimated irradiation unit that irradiates the charged object with charged particles based on the estimated irradiation time; and The switching unit irradiates the charged object with charged particles using the estimated irradiation unit instead of the irradiation unit when the learning end condition is satisfied in the end determination unit.

3. The charged particle irradiation device according to claim 1, wherein: In the irradiation time generating unit, The irradiation time of the charged particles to the charged object is calculated by performing an approximate process based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation.

4. The charged particle irradiation device according to claim 3, wherein: The approximation process is performed by first-order approximation or straight-line approximation to the target charging state.

5. The charged particle irradiation device according to claim 3, wherein: The approximation process is performed using a value obtained by offsetting the approximation process result regarding the irradiation time.

6. The charged particle irradiation device according to claim 1, wherein: The charged particle irradiation device further comprises: a change detection unit for detecting a change in the charged object or the environment; and The operation start determination unit operates the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, the machine learning unit, and the end determination unit when a predetermined change is detected in the change detection unit.

7. The charged particle irradiation device according to claim 1, wherein: The charged state of the charged object is the electric potential of the charged object.

8. The charged particle irradiation device according to claim 1, wherein: The charged particle irradiation device performs a predetermined error process when it is determined that the charged state after irradiation is equal to or greater than a predetermined ratio with respect to the charged state before irradiation.

9. The charged particle irradiation device according to claim 8, wherein: The error handling includes device stop processing, which is to add points according to the ratio when it is determined that the charged state after irradiation is greater than a predetermined ratio with respect to the charged state before irradiation, and stop the charged particle irradiation device when the points become greater than a predetermined value.

10. The charged particle irradiation device according to claim 8, wherein: The error processing includes a process of excluding each of the determined charged states from the machine learning target in the machine learning unit when it is determined that the charged state after irradiation is equal to or greater than a predetermined ratio with respect to the charged state before irradiation.

11. The charged particle irradiation device according to claim 1, wherein: The machine learning model is a learning model using a tree structure.

12. A static electricity removal device for irradiating a charged object with charged particles for a predetermined irradiation time, the static electricity removal device comprising: a pre-irradiation state detection unit for detecting a charged state of the charged object before irradiation with charged particles, i.e., a pre-irradiation charged state; an irradiation time generating unit for generating an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit that irradiates the charged object in the charged state before the irradiation with charged particles based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit for detecting a charged state of the charged object after being irradiated with the charged particles, that is, a post-irradiation charged state; a machine learning unit that causes a predetermined machine learning model to learn the correspondence between the charged state before irradiation, the charged state after irradiation, and the irradiation time generated by the irradiation time generation unit; and The end determination unit causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, and the machine learning unit to repeatedly operate until the machine learning model satisfies a predetermined learning end condition, wherein: The learning end condition is that the error between the irradiation time as the output of the machine learning model and the irradiation time when the charged state after irradiation becomes the target state or a state close to it is less than a predetermined value or a predetermined ratio.

13. A charged particle irradiation method, comprising irradiating a charged object with charged particles for a predetermined irradiation time, the charged particle irradiation method comprising: a pre-irradiation state detection step, detecting the charged state of the charged object before irradiation with charged particles, i.e., the pre-irradiation charged state; An irradiation time generating step of generating an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation step of irradiating the charged object in the charged state before the irradiation with charged particles based on the irradiation time generated by the irradiation time generation step; a post-irradiation state detection step, detecting the charged state of the charged object after being irradiated with the charged particles, i.e., the charged state after irradiation; a machine learning step, causing a predetermined machine learning model to learn the correspondence between the charged state before irradiation, the charged state after irradiation, and the irradiation time generated by the irradiation time generating step; and The end determination step causes the pre-irradiation state detection step, the irradiation time generation step, the irradiation step, the post-irradiation state detection step, and the machine learning step to repeat until the machine learning model satisfies a predetermined learning end condition, wherein: The learning end condition is that the error between the irradiation time as the output of the machine learning model and the irradiation time when the charged state after irradiation becomes the target state or a state close to it is less than a predetermined value or a predetermined ratio.

14. A computer program product comprising a computer program for causing a computer to function as a charged particle irradiation device, wherein: The charged particle irradiation device irradiates charged particles to a charged object for a predetermined irradiation time, and the charged particle irradiation device comprises: a pre-irradiation state detection unit for detecting a charged state of the charged object before irradiation with charged particles, i.e., a pre-irradiation charged state; an irradiation time generating unit for generating an irradiation time of the charged particles to the charged object based on the past irradiation time of the charged particles and the charged state of the charged object after the irradiation; an irradiation unit that irradiates the charged object in the charged state before the irradiation with charged particles based on the irradiation time generated by the irradiation time generation unit; a post-irradiation state detection unit for detecting a charged state of the charged object after being irradiated with the charged particles, that is, a post-irradiation charged state; a machine learning unit that causes a predetermined machine learning model to learn the correspondence between the charged state before irradiation, the charged state after irradiation, and the irradiation time generated by the irradiation time generation unit; and The end determination unit causes the pre-irradiation state detection unit, the irradiation time generation unit, the irradiation unit, the post-irradiation state detection unit, and the machine learning unit to repeatedly operate until the machine learning model satisfies a predetermined learning end condition, wherein: The learning end condition is that the error between the irradiation time as the output of the machine learning model and the irradiation time when the charged state after irradiation becomes the target state or a state close to it is less than a predetermined value or a predetermined ratio.

Images