Oxygen concentration prediction system and oxygen concentration control system

Abstract

To improve control precision of oxygen concentration of a silicon ingot.SOLUTION: An oxygen concentration prediction system has manufacture actual result data consisting of sets of control parameter groups which can be operated during crystal growth, fixed parameter groups whose values are known before the start of the crystal growth, monitor parameter groups on which in-furnace conditions in the crystal growth are reflected, and oxygen concentrations of silicon ingots, and uses a control parameter group, a fixed monitor parameter group, and a monitor parameter group of reference data selected from the manufacture actual result data, and a control parameter group and a fixed parameter group for a target silicon ingot to predict an oxygen concentration of the target silicon ingot.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an oxygen concentration prediction system and an oxygen concentration control system. [Background technology]

[0002] The Czochralski method (hereafter referred to as the CZ method) is a typical method for manufacturing silicon single crystals (so-called silicon ingots) for semiconductor devices. When growing silicon ingots using this CZ method, oxygen inevitably dissolves from the quartz crucible into the silicon melt, resulting in oxygen contamination of the silicon ingot. Since oxygen in silicon ingots affects device characteristics when oxygen precipitates are formed during heat treatment processes during semiconductor device manufacturing, device manufacturers have various requirements for oxygen concentration.

[0003] However, even at present, it is extremely difficult to accurately predict and freely control the oxygen concentration in a silicon ingot over the entire length of the ingot. When growing silicon ingots using the CZ method, oxygen dissolves at the contact interface between the quartz crucible and the silicon melt, and the dissolved oxygen is transported within the melt according to the behavior of thermal convection (natural convection, forced convection) in the silicon melt. In recent years, the mainstream method for pulling large-diameter silicon ingots is one in which a magnetic field is applied to the melt to precisely control the convection of the melt, and it is known that the behavior of oxygen in the melt is greatly affected by the type and strength of the magnetic field, the amount of remaining melt, and other factors.

[0004] In addition, it is known that oxygen evaporation from the melt occurs very actively on the free surface, and the oxygen concentration in the melt (especially near the free surface) is greatly affected by the flow rate of argon (Ar) gas on the free surface. In other words, the oxygen concentration incorporated into the crystal is determined by the complex interaction of many factors, such as dissolution from the quartz crucible into the melt, transport in the melt by convection, and evaporation from the free surface. In addition, in actual manufacturing sites, there are various manufacturing variations due to factors such as changes over time in the materials used and variations in the furnace environment caused by variations in the settings of the materials inside the furnace between lots, variations between materials, machine differences between devices, and adjustment variations by operators, so that even if each lot is grown under the same operating conditions, it is not necessarily the case that silicon ingots with the same oxygen concentration profile can be obtained. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 5-262593 [Patent Document 2] Japanese Patent Application Publication No. 10-291891 Summary of the Invention [Problem to be solved by the invention]

[0006] Incidentally, the oxygen concentration in a silicon ingot is measured by the following procedure.

[0007] (1) After the single crystal has been pulled, an evaluation sample is cut from a designated portion of the single crystal ingot. (2) The evaluation sample is polished, etched, and cleaned. (3) The evaluation sample is measured by Fourier transform infrared spectroscopy (FT-IR).

[0008] Thus, there is a time lag between the completion of pulling the silicon ingot and the measurement of the oxygen concentration. This means that it takes time for the actually measured oxygen concentration to be reflected in the control of the subsequent silicon ingot manufacturing process. Naturally, this time lag also has a large effect on the accuracy of the control of the oxygen concentration.

[0009] Therefore, various methods have been proposed for predicting the oxygen concentration from the manufacturing conditions, rather than obtaining a sample from a silicon ingot and directly measuring the oxygen concentration (see, for example, Patent Document 1 or Patent Document 2). However, as described above, there are a wide variety of factors that affect the oxygen concentration in a silicon ingot, and it is still not easy to predict the oxygen concentration with a sufficiently high degree of accuracy. In particular, it is difficult to use monitor parameters that reflect the state inside the furnace during crystal growth to predict the oxygen concentration of a silicon ingot that is about to be manufactured, making it difficult to determine a group of control parameters for the silicon ingot that is about to be manufactured.

[0010] An object of the present invention is to provide an oxygen concentration prediction system and an oxygen concentration control system that improve the control accuracy of the oxygen concentration in a silicon ingot. [Means for solving the problem]

[0011] In order to solve the above problems, an oxygen concentration prediction system according to one embodiment of the present invention is an oxygen concentration prediction system for predicting an oxygen concentration of a target silicon ingot, comprising manufacturing history data constituted by a set of a control parameter group that can be manipulated during crystal growth, a fixed parameter group whose values ​​are known before the start of crystal growth, a monitor parameter group that reflects the state inside a furnace during crystal growth, and an oxygen concentration of the silicon ingot, a first prediction model that inputs the control parameter group, the fixed parameter group, and the monitor parameter group and outputs an oxygen concentration, a second prediction model that inputs the control parameter group and the fixed parameter group and outputs an oxygen concentration, and a reference data is selected from the manufacturing history data and the reference data is input to the first prediction model. the control parameter group, the fixed parameter group, and the monitor parameter group of the reference data into the first prediction model to predict a first oxygen concentration predicted value; a second prediction means for inputting the control parameter group and the fixed parameter group of the reference data into the second prediction model to predict a second oxygen concentration predicted value; a third prediction means for inputting a control parameter group and a fixed parameter group for the target silicon ingot into the third prediction model to predict a third oxygen concentration predicted value; and a correction means for correcting the third oxygen concentration predicted value using a difference between the first oxygen concentration predicted value and the second oxygen concentration predicted value, and setting the corrected third oxygen concentration predicted value as the oxygen concentration of the target silicon ingot.

[0012] In this way, the oxygen concentration prediction system of one embodiment of the present invention does not use a group of monitor parameters reflecting the conditions inside the furnace during crystal growth when predicting the oxygen concentration of a target silicon ingot, but improves the prediction accuracy of the oxygen concentration of the target silicon ingot by correcting the oxygen concentration using a first oxygen concentration prediction value using reference data that includes the group of monitor parameters and a second oxygen concentration prediction value using reference data that does not include the group of monitor parameters.

[0013] In addition, the first prediction model in the oxygen concentration prediction system of one embodiment of the present invention is trained using three parameter groups, the control parameter group, the fixed parameter group, and the monitor parameter group, in the manufacturing history data, and the oxygen concentration of a silicon ingot manufactured using the three parameter groups, as teacher data, and the second prediction model is trained using two parameter groups, the control parameter group and the fixed parameter group, in the manufacturing history data, and the oxygen concentration of a silicon ingot manufactured using the two parameter groups, as teacher data. The oxygen concentration prediction system of one embodiment of the present invention uses a first prediction model and a second prediction model, and each is trained in advance using teacher data.

[0014] Moreover, an oxygen concentration prediction system according to one embodiment of the present invention is an oxygen concentration prediction system for predicting an oxygen concentration of a target silicon ingot, the system comprising: manufacturing history data constituted by a set of a control parameter group that can be operated during crystal growth, a fixed parameter group whose values ​​are known before the start of crystal growth, a monitor parameter group reflecting a furnace condition during crystal growth, and an oxygen concentration of the silicon ingot; a third prediction model that inputs a first control parameter group, a first fixed parameter group, a first monitor parameter group, a second control parameter group, and a second fixed parameter group and outputs an oxygen concentration; and a fourth prediction means that selects reference data from the manufacturing history data, and sets the control parameter group of the reference data as the first control parameter group, the fixed parameter group of the reference data as the first fixed parameter group, the monitor parameter group of the reference data as the first monitor parameter group, a control parameter group for the target silicon ingot as the second control parameter group, and a fixed parameter group for the target silicon ingot as the second fixed parameter group, and inputs these into the third prediction model to predict the oxygen concentration of the target silicon ingot.

[0015] In this way, the oxygen concentration prediction system of one embodiment of the present invention does not use a group of monitor parameters reflecting the conditions inside the furnace during crystal growth when predicting the oxygen concentration of a target silicon ingot, but improves the prediction accuracy of the oxygen concentration of the target silicon ingot by making predictions using reference data that includes the group of monitor parameters.

[0016] The third prediction model in the oxygen concentration prediction system of one embodiment of the present invention is trained using teacher data that combines a first silicon ingot in the manufacturing performance data, a second silicon ingot for the first silicon ingot, the control parameter group, the fixed parameter group, and the monitor parameter group in the first silicon ingot, the control parameter group and the fixed parameter group in the second silicon ingot, and the oxygen concentration in the second silicon ingot. The third prediction model in the oxygen concentration prediction system of one embodiment of the present invention is trained using teacher data that combines reference data including a monitor parameter group and peripheral data not including a monitor parameter group.

[0017] Furthermore, the training data preferably includes a data set related to a combination of first silicon ingots common to a plurality of second silicon ingots, in order to learn the influence of perturbation on prediction without using the monitor parameter group from prediction with the monitor parameter group.

[0018] It is preferable to select the reference data from production performance data of a silicon ingot that is close to the production conditions of the target silicon ingot, because the reference data is considered to include a group of monitor parameters that are useful in predicting the oxygen concentration of the target silicon ingot.

[0019] The control parameters preferably include the crucible rotation speed, the magnetic field strength, and the furnace pressure. The fixed parameters preferably include the heater usage count, the crucible usage count, and the quartz crucible thickness. The monitor parameters preferably include the crystal diameter, the crown shape, the total current flow time, the integrated power consumption, and the temperature measurement values ​​at each point in the furnace.

[0020] Furthermore, an oxygen concentration control system of one embodiment of the present invention comprises the oxygen concentration prediction system described above, and a single crystal pulling apparatus that controls the group of control parameters so that the oxygen concentration of the target silicon ingot predicted by the oxygen concentration prediction system falls within a predetermined range.

[0021] The oxygen concentration prediction system described above does not use a group of monitor parameters reflecting the conditions inside the furnace during crystal growth when predicting the oxygen concentration of a target silicon ingot, and therefore can bring the oxygen concentration of the target silicon ingot within a specified range by adjusting a group of control parameters. Effect of the Invention

[0022] According to the present invention, it is possible to provide an oxygen concentration prediction system and an oxygen concentration control system that improve the control accuracy of the oxygen concentration in a silicon ingot. [Brief description of the drawings]

[0023] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of an oxygen concentration control system. [Diagram 2] FIG. 2 is a diagram showing a schematic configuration of a neural network in the first prediction model. [Diagram 3] FIG. 3 is a diagram showing a schematic configuration of a neural network in the second prediction model. [Figure 4] FIG. 4 is a diagram illustrating learning in the first prediction model. [Diagram 5] FIG. 5 is a diagram illustrating learning in the second prediction model. [Figure 6] FIG. 6 is a diagram for explaining a prediction method in the oxygen concentration prediction system. [Figure 7] FIG. 7 is a diagram showing a schematic configuration of a neural network in the third prediction model. [Figure 8] FIG. 8 is a diagram illustrating learning in the third prediction model. [Figure 9] FIG. 9 is a diagram for explaining a prediction method in the oxygen concentration prediction system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. However, the present invention is not limited to the embodiment described below. In addition, in each drawing, the same or corresponding elements are appropriately assigned the same reference numerals. Furthermore, it should be noted that the drawings are schematic, and the dimensional relationship of each element, the ratio of each element, etc. may differ from the actual ones. There may also be parts in which the dimensional relationship and ratio differ between the drawings.

[0025] [Oxygen concentration control system] 1 is a diagram showing a schematic configuration of an oxygen concentration control system according to an embodiment of the present invention. As shown in the figure, the oxygen concentration control system 100 includes an oxygen concentration prediction system 10 and a single crystal pulling apparatus 20.

[0026] The single crystal pulling apparatus 20 is an apparatus for manufacturing a silicon ingot S by using the CZ method. The single crystal pulling apparatus 20 includes a chamber furnace 21, a heater 22, a crucible 23, and a motor 24. The single crystal pulling apparatus 20 manufactures a silicon ingot S made of a silicon single crystal by pulling up the silicon melt in the crucible 23 with the motor 24 while crystallizing the silicon melt. As described above, in recent large-diameter silicon ingots, it is common to apply a magnetic field to the melt in order to precisely control the convection of the melt.

[0027] The single crystal pulling apparatus 20 is provided with various sensors, and production performance data D is acquired for each production of silicon ingot S. The production performance data D is broadly divided into three types of parameter groups: a control parameter group, a fixed parameter group, and a monitor parameter group. The production performance data D also includes measured values ​​of oxygen concentration in silicon ingots S produced using the three types of parameter groups.

[0028] The control parameters are parameters that can be manipulated during crystal growth, and include, for example, the crucible rotation speed, magnetic field strength, and furnace pressure. The fixed parameters are parameters whose values ​​are known before the start of crystal growth, and include, for example, the number of heater uses, the number of crucible uses, and the thickness of the quartz crucible. The monitor parameters are parameters that reflect the conditions inside the furnace during crystal growth, and include, for example, the crystal diameter, the shape of the crown, the total power supply time, the accumulated power consumption, and the temperature measurements at each point inside the furnace.

[0029] On the other hand, the oxygen concentration prediction system 10 is a system that inputs a group of control parameters and a group of fixed parameters for a target silicon ingot, and predicts the oxygen concentration of the target silicon ingot to be manufactured.

[0030] The oxygen concentration prediction system 10 refers to the manufacturing history data D to predict the oxygen concentration of the target silicon ingot S. The data referred to in predicting the oxygen concentration of the target silicon ingot S is manufacturing history data of a silicon ingot (reference ingot) whose characteristics and manufacturing conditions are similar to those of the target silicon ingot. The reference data includes a group of control parameters that can be manipulated during crystal growth, a group of fixed parameters whose values ​​are known before the start of crystal growth, and a group of monitor parameters whose values ​​reflect the state inside the furnace during crystal growth.

[0031] Moreover, the oxygen concentration prediction system 10 includes a neural network, and the neural network is trained using the manufacturing performance data D. Note that the prediction and training methods in the oxygen concentration prediction system 10 will be described in detail later.

[0032] The oxygen concentration control system 100 including the oxygen concentration prediction system 10 and the single crystal pulling apparatus 20 as described above controls a group of control parameters so that the oxygen concentration of the target silicon ingot S predicted by the oxygen concentration prediction system 10 falls within a predetermined range. That is, when the oxygen concentration of the target silicon ingot S predicted by the oxygen concentration prediction system 10 falls within the predetermined range, the target silicon ingot S is manufactured using the group of control parameters input to the oxygen concentration prediction system 10. On the other hand, when the oxygen concentration of the target silicon ingot S to be predicted falls outside the predetermined range, the group of control parameters is adjusted, and the oxygen concentration is predicted again using the adjusted group of control parameters. Then, the adjustment of the group of control parameters is repeated until the oxygen concentration of the target silicon ingot S to be predicted falls within the predetermined range.

[0033] In this way, the oxygen concentration control system 100 includes the oxygen concentration prediction system 10 and the single crystal pulling apparatus 20, and controls the control parameters so that the oxygen concentration of the target silicon ingot S predicted by the oxygen concentration prediction system 10 falls within a predetermined range, thereby improving the control accuracy of the oxygen concentration of the target silicon ingot S. Conventionally, it has been difficult to use the monitor parameters reflecting the state inside the furnace during crystal growth to predict the oxygen concentration of the target silicon ingot S, but the oxygen concentration control system 100 can control the oxygen concentration of the target silicon ingot S to fall within a predetermined range simply by adjusting the control parameters that can be manipulated during crystal growth by referring to the monitor parameters of the reference data.

[0034] [Oxygen Concentration Prediction System According to the First Embodiment] The oxygen concentration prediction system according to the first embodiment is configured using an artificial intelligence technique including a neural network. The oxygen concentration prediction system according to the first embodiment includes a first prediction model and a second prediction model. Fig. 2 is a diagram showing a schematic configuration of the neural network in the first prediction model, and Fig. 3 is a diagram showing a schematic configuration of the neural network in the second prediction model.

[0035] As shown in Fig. 2, the neural network 11 in the first prediction model inputs a control parameter group, a fixed parameter group, and a monitor parameter group, and outputs the oxygen concentration of a silicon ingot. Note that the neural network 11 shown in Fig. 2 is illustrated as having three nodes in the input layer, but the control parameter group, the fixed parameter group, and the monitor parameter group each include a plurality of parameters, and the number of nodes in the input layer corresponds to the number of inputs of these multiple parameters. In addition, the configuration of the nodes in the intermediate layer is also omitted for convenience, and does not show the actual configuration of the nodes.

[0036] As shown in Fig. 3, the neural network 12 in the second prediction model receives the control parameters and the fixed parameters and outputs the oxygen concentration of the silicon ingot. Similarly, the node configuration in the neural network 12 shown in Fig. 2 is omitted for convenience and does not represent the actual node configuration.

[0037] (Learning Phase) FIG. 4 is a diagram illustrating learning in the first prediction model, and FIG. 5 is a diagram illustrating learning in the second prediction model.

[0038] As shown in FIG. 4, the first prediction model is trained using three parameter groups, namely, a control parameter group, a fixed parameter group, and a monitor parameter group, in the manufacturing performance data and the oxygen concentration of a silicon ingot manufactured using the three parameter groups as training data T1.

[0039] For example, when learning using manufacturing performance data for an ingot ID of X, a control parameter group: a, a fixed parameter group: b, and a monitor parameter group: c are input to the input layer of the neural network 11 in the first prediction model, and the difference between the value output from the output layer and the oxygen concentration: d is evaluated. This evaluation is performed for all of the training data T1 recorded in the manufacturing performance data. Optimization is performed so that the difference between the value output from the output layer and the oxygen concentration: d is minimized.

[0040] On the other hand, as shown in Figure 5, the second prediction model learns using two parameter groups, a control parameter group and a fixed parameter group, in the manufacturing performance data and the oxygen concentration of a silicon ingot manufactured using the two parameter groups as training data T1.

[0041] For example, when learning using manufacturing performance data for an ingot ID of X, the control parameter group: a and the fixed parameter group: b are input to the input layer of the neural network 12 in the second prediction model, and the difference between the value output from the output layer and the oxygen concentration: d is evaluated. This evaluation is performed for all of the training data T1 recorded in the manufacturing performance data, and optimization is performed so that the difference between the value output from the output layer and the oxygen concentration: d is minimized.

[0042] It should be noted here that the first prediction model is trained using the monitor parameter group, but the second prediction model is trained without using the monitor parameter group. In other words, the oxygen concentration prediction system according to the first embodiment includes a first prediction model trained using the monitor parameter group and a second prediction model trained without using the monitor parameter group.

[0043] (Prediction phase) Fig. 6 is a diagram for explaining a prediction method in an oxygen concentration prediction system. The oxygen concentration prediction system according to the first embodiment includes a first prediction model trained using a group of monitor parameters and a second prediction model trained without using the group of monitor parameters. However, for ease of explanation, Fig. 6 illustrates a neural network 11 in one first prediction model and neural networks 12 in two second prediction models. In an actual configuration, it is not necessary to provide duplicate neural networks 12.

[0044] As shown in FIG. 6, in the prediction phase, the oxygen concentration prediction system of the first embodiment predicts a first oxygen concentration prediction value, a second oxygen concentration prediction value, and a third oxygen concentration prediction value, corrects the third oxygen concentration prediction value using the difference between the first oxygen concentration prediction value and the second oxygen concentration prediction value, and sets the corrected third oxygen concentration prediction value as the oxygen concentration of the target silicon ingot.

[0045] The calculation method for the correction is not limited to a specific method, but for example, the correction formula can be the following calculation method. Target silicon ingot oxygen concentration = First oxygen concentration predicted value - Second oxygen concentration predicted value + Third oxygen concentration predicted value

[0046] The first oxygen concentration predicted value is a predicted value output from the output layer of the neural network 11 by inputting the control parameter group: a, the fixed parameter group: b, and the monitor parameter group: c of the reference ingot into the input layer of the neural network 11.

[0047] The second oxygen concentration predicted value is a predicted value output from the output layer of the neural network 12 by inputting the control parameter group: a and the fixed parameter group: b of the reference ingot into the input layer of the neural network 12 .

[0048] On the other hand, the third oxygen concentration predicted value is a predicted value output from the output layer of the neural network 12 by inputting the control parameter group A and the fixed parameter group B for the target ingot into the input layer of the neural network 12 .

[0049] As can be seen from the definitions of the first oxygen concentration prediction value, the second oxygen concentration prediction value, and the third oxygen concentration prediction value as described above, the difference between the first oxygen concentration prediction value and the second oxygen concentration prediction value is the difference between whether or not the set of monitor parameters is used in the prediction. Therefore, by correcting the third oxygen concentration prediction value using the difference between the first oxygen concentration prediction value and the second oxygen concentration prediction value, it is possible to offset the error caused by not using the set of monitor parameters in the prediction.

[0050] In this way, the oxygen concentration prediction system of the first embodiment does not use the monitor parameters of the target ingot in the prediction phase, but improves the prediction accuracy of the oxygen concentration of the target silicon ingot by correcting the oxygen concentration using a first oxygen concentration prediction value using the monitor parameters of the reference ingot and a second oxygen concentration prediction value that does not use the monitor parameters of the reference ingot.

[0051] Although the control parameter group: a and the fixed parameter group: b of the reference ingot do not necessarily have to match the control parameter group: A and the fixed parameter group: B of the target ingot, it is preferable to select a silicon ingot close to the manufacturing conditions of the target silicon ingot from the manufacturing performance data. This is because the closer the control parameter group: a and the fixed parameter group: b of the reference ingot are to the control parameter group: A and the fixed parameter group: B of the target ingot, the better the error caused by not using the monitor parameter group for prediction can be offset.

[0052] [Oxygen Concentration Prediction System According to the Second Embodiment] The oxygen concentration prediction system according to the second embodiment is configured using an artificial intelligence technique equipped with a neural network. The oxygen concentration prediction system according to the second embodiment includes a third prediction model, and FIG. 7 is a diagram showing a schematic configuration of a neural network in the third prediction model.

[0053] As shown in Fig. 7, the neural network 13 in the third prediction model inputs the first control parameter group, the first fixed parameter group, and the first monitor parameter group, and the second control parameter group and the second fixed parameter group, and outputs the oxygen concentration of the silicon ingot. Note that the node configuration in the neural network 12 shown in Fig. 7 is omitted for convenience, and does not represent the actual node configuration.

[0054] (Learning Phase) FIG. 8 is a diagram illustrating learning in the third prediction model.

[0055] As shown in FIG. 8, the third prediction model is trained using training data T2 that combines the control parameter group, fixed parameter group, and monitor parameter group for the first silicon ingot, the control parameter group and fixed parameter group for the second silicon ingot, and the oxygen concentration in the second silicon ingot.

[0056] The first silicon ingot is selected from the manufacturing history data as one corresponding to the reference ingot, and the second silicon ingot is selected from the manufacturing history data as one corresponding to a peripheral ingot with characteristics and manufacturing conditions similar to those of the reference ingot. Therefore, for one first silicon ingot (reference ingot), there may be a plurality of second silicon ingots (peripheral ingots) having properties similar to the first silicon ingot (reference ingot). In other words, the teacher data T2 includes a data set regarding a combination of first silicon ingots that are common to a plurality of second silicon ingots. It is preferable to select the reference ingot itself as the peripheral ingot. This is because when the reference ingot itself is selected as the peripheral ingot, it is possible to train the model to return the output of the oxygen concentration of the reference ingot.

[0057] For example, as shown in Fig. 8, when learning is performed using a set of manufacturing performance data with ingot ID X and ingot ID Y1, the control parameter group: a is input as the first control parameter group, the fixed parameter group: b is input as the first fixed parameter group, the monitor parameter group: c is input as the first monitor parameter group, the control parameter group: e1 is input as the second control parameter group, and the fixed parameter group: f1 is input as the second fixed parameter group to the input layer of the neural network 13 in the third prediction model, and the difference between the value output from the output layer and the oxygen concentration: d1 is evaluated. Such an evaluation is performed for all of the teacher data T2 recorded in the manufacturing performance data. Optimization is performed so that the difference between the value output from the output layer and the oxygen concentration: di is minimized.

[0058] By performing the optimization as described above, when the neural network 13 receives the control parameters, fixed parameters, and monitor parameters for the first silicon ingot (reference ingot) and the control parameters and fixed parameters for the second silicon ingot (peripheral ingot), it outputs an estimated value of the oxygen concentration in the second silicon ingot (peripheral ingot).

[0059] 8, the data of the first silicon ingot (reference ingot) does not include the oxygen concentration, and the data of the second silicon ingot (peripheral ingot) does not include the monitor parameters, but there is no need to separately prepare manufacturing performance data that does not include the oxygen concentration or the monitor parameters. It is only necessary not to use them as the teacher data T2 when learning the neural network 13 in the third prediction model.

[0060] (Prediction phase) Fig. 9 is a diagram for explaining a prediction method in the oxygen concentration prediction system. As shown in Fig. 9, in the prediction phase, the oxygen concentration prediction system according to the second embodiment inputs a control parameter group a of a reference ingot as a first control parameter group, b a fixed parameter group b of the reference ingot as a first fixed parameter group, c a monitor parameter group c of the reference ingot as a first monitor parameter group, A a control parameter group A for a target silicon ingot as a second control parameter group, and B a fixed parameter group B for the target silicon ingot as a second fixed parameter group to the input layer of the neural network 13, and predicts the oxygen concentration of the target silicon ingot.

[0061] In this way, the oxygen concentration prediction system of the second embodiment does not use the monitor parameters of the target ingot in the prediction phase, but considering the method of optimizing the neural network 13, the influence of excluding the monitor parameters in the reference data is taken into account, thereby improving the prediction accuracy of the oxygen concentration of the target silicon ingot.

[0062] In addition, as in the second embodiment, the control parameter group: a and the fixed parameter group: b of the reference ingot do not necessarily have to be the same as the control parameter group: A and the fixed parameter group: B of the target ingot, but it is preferable to select a silicon ingot that is close to the manufacturing conditions of the target silicon ingot from the manufacturing performance data. This is because the closer the control parameter group: a and the fixed parameter group: b of the reference ingot are to the control parameter group: A and the fixed parameter group: B of the target ingot, the better the error caused by not using the monitor parameter group for prediction can be offset.

[0063] Although the present invention has been described above based on the embodiments, the present invention is not limited to the above-described embodiments. [Explanation of symbols]

[0064] 100 Oxygen concentration control system 10. Oxygen Concentration Prediction System 11, 12, 13 Neural Networks 20 Single crystal pulling device 21 Chamber furnace 22 Heater 23 Crucible 24 Motor S Silicon ingot D. Manufacturing performance data T1, T2 training data

Claims

1. An oxygen concentration prediction system for predicting an oxygen concentration of a target silicon ingot, comprising: manufacturing performance data consisting of a set of control parameters that can be manipulated during crystal growth, including the crucible rotation speed and magnetic field strength, fixed parameters whose values ​​are known before the start of crystal growth, including the number of times the crucible has been used, monitor parameters that reflect the conditions inside the furnace during crystal growth, including the crystal diameter and temperature measurement values ​​at various points inside the furnace, and the oxygen concentration of the silicon ingot; a first prediction model that is trained using three parameter groups, i.e., the control parameter group, the fixed parameter group, and the monitor parameter group in the manufacturing performance data, and an oxygen concentration of a silicon ingot manufactured using the three parameter groups, as teacher data, and that inputs the control parameter group, the fixed parameter group, and the monitor parameter group to output an oxygen concentration; a second prediction model that is trained using two parameter groups, the control parameter group and the fixed parameter group, in the manufacturing performance data and an oxygen concentration of a silicon ingot manufactured using the two parameter groups as teacher data, and that inputs the control parameter group and the fixed parameter group to output an oxygen concentration; a first prediction means for selecting reference data having values ​​of the control parameter group and the fixed parameter group close to those of the target silicon ingot from the manufacturing performance data, and inputting the values ​​of the control parameter group, the fixed parameter group, and the monitor parameter group of the reference data into the first prediction model to predict a first oxygen concentration predicted value; a second prediction means for predicting a second oxygen concentration prediction value by inputting values ​​of the control parameters and the fixed parameters of the reference data into the second prediction model; a third prediction means for inputting values ​​of a group of control parameters and a group of fixed parameters for the target silicon ingot into the second prediction model to predict a third oxygen concentration prediction value; a correction means for correcting the third oxygen concentration predicted value using a difference between the first oxygen concentration predicted value and the second oxygen concentration predicted value, and setting the corrected third oxygen concentration predicted value as the oxygen concentration of the target silicon ingot; An oxygen concentration prediction system comprising:

2. An oxygen concentration prediction system for predicting an oxygen concentration of a target silicon ingot, comprising: manufacturing performance data consisting of a set of control parameters that can be manipulated during crystal growth, including the crucible rotation speed and magnetic field strength, fixed parameters whose values ​​are known before the start of crystal growth, including the number of times the crucible has been used, monitor parameters that reflect the conditions inside the furnace during crystal growth, including the crystal diameter and temperature measurement values ​​at various points inside the furnace, and the oxygen concentration of the silicon ingot; a third prediction model configured to receive a first control parameter group, a first fixed parameter group, a first monitor parameter group, a second control parameter group, and a second fixed parameter group and output an oxygen concentration, the third prediction model being trained using teacher data that combines a first silicon ingot in the manufacturing history data, a second silicon ingot having values ​​of the control parameter group and the fixed parameter group that are close to those of the first silicon ingot, and the control parameter group, the fixed parameter group, and the monitor parameter group in the first silicon ingot, the control parameter group and the fixed parameter group in the second silicon ingot, and the oxygen concentration in the second silicon ingot; a fourth prediction means for selecting, from the manufacturing performance data, reference data having values ​​similar to those of the target silicon ingot, the control parameter group of the reference data as the first control parameter group, the fixed parameter group of the reference data as the first fixed parameter group, the monitor parameter group of the reference data as the first monitor parameter group, the control parameter group for the target silicon ingot as the second control parameter group, and the fixed parameter group for the target silicon ingot as the second fixed parameter group, and inputting these into the third prediction model to predict the oxygen concentration of the target silicon ingot; An oxygen concentration prediction system comprising:

3. The oxygen concentration prediction system according to claim 2 , wherein the training data includes a data set relating to a combination of first silicon ingots that is common to a plurality of second silicon ingots.

4. 4. The oxygen concentration prediction system according to claim 1, wherein the group of control parameters further includes a furnace pressure, the group of fixed parameters further includes a number of times the heater is used, a number of times the crucible is used, and a thickness of the quartz crucible, and the group of monitor parameters further includes a total power supply time, an integrated power consumption, and a shape of the crown portion.

5. The oxygen concentration prediction system according to any one of claims 1 to 4, a single crystal pulling apparatus that controls the group of control parameters so that the oxygen concentration of the target silicon ingot predicted by the oxygen concentration prediction system falls within a predetermined range; An oxygen concentration control system comprising:

Images