Method for predicting characteristics, method for manufacturing a semiconductor device, method for predicting characteristics, method for predicting characteristics, method for generating a trained model

Abstract

This invention provides a characteristic prediction method that can accurately predict the characteristics of a product. [Solution] The characteristic prediction method includes the steps of: obtaining a trained model that defines the relationship between a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, a sequence of wafers in a predetermined arrangement order within the processing apparatus that perform the first process simultaneously in each of the plurality of first processes, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and inputting the first sequence of wafers in the first process performed in the processing apparatus and the measured first characteristics corresponding to the first sequence into the trained model to predict the second characteristics corresponding to the second sequence after the first sequence, wherein the trained model includes a time-series model with the sequence in a time series.

Description

Technical Field

[0004] ,

[0006] , , , , , ,

[0005] , , , , , ,

[0001] The present disclosure relates to a characteristic prediction method, a method for manufacturing a semiconductor device, a characteristic prediction program, a characteristic prediction device, and a learned model generation method.

Background Art

[0002] In a manufacturing system, a system is known that stores stored data in time series and predicts data missing in time series from the stored data (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] After processing a wafer, another process may be performed to measure the characteristics of the wafer. In such a case, in a processing apparatus that processes a plurality of wafers simultaneously, simply arranging the plurality of wafers in the time series of processing may not be able to appropriately estimate the characteristics.

[0005] An object of the present disclosure is to provide a characteristic prediction method, a method for manufacturing a semiconductor device, a characteristic prediction program, a characteristic prediction device, and a learned model generation method capable of appropriately predicting characteristics.

Means for Solving the Problems

[0006] Embodiments of the present disclosure are characteristic prediction methods comprising: obtaining a trained model that defines the relationship between a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, a sequence of wafers in a predetermined arrangement order within the processing apparatus that perform the first process simultaneously in each of the plurality of first processes, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and inputting the first sequence of wafers in the first process performed in the processing apparatus and the measured first characteristics corresponding to the first sequence into the trained model, wherein the trained model includes a time-series model with the sequence in a time series.

[0007] Embodiments of the present disclosure are methods for manufacturing a semiconductor device, comprising the steps of: performing the above-described characteristic prediction method; changing the conditions of the first or second process based on the second characteristic; and performing the first or second process on a wafer corresponding to the second serial number using the changed conditions.

[0008] Embodiments of the present disclosure are methods for generating a trained model, comprising the steps of: obtaining training data relating a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, arranged in a time series, and relating a sequence number obtained by arranging the plurality of wafers that perform the first process simultaneously in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and generating a trained model for predicting a second characteristic corresponding to a second sequence number after the first sequence number by using machine learning on the training data, inputting a first sequence number of the plurality of wafers in the first process performed in the processing apparatus and a first characteristic corresponding to the measured first sequence number, wherein the trained model includes a time series model with the sequence numbers as a time series.

[0009] This disclosure can be implemented not only as a characteristic characteristic prediction method and a trained model generation method, but also as a characteristic prediction program, characteristic prediction device, trained model generation program, and trained model generation device that process such characteristic steps. Furthermore, it can be implemented as a semiconductor integrated circuit that implements part or all of the characteristic prediction device and trained model generation device, or as an estimation system including an estimation device. [Effects of the Invention]

[0010] According to this disclosure, the characteristics can be appropriately predicted. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 is a flowchart of the process for predicting characteristics in the first embodiment. [Figure 2] Figure 2 is a cross-sectional view showing a method for manufacturing a nitride semiconductor device. [Figure 3] Figure 3 is a cross-sectional view showing a method for manufacturing a nitride semiconductor device. [Figure 4] Figure 4 is a plan view illustrating the batch processing in the first embodiment. [Figure 5] Figure 5 is a block diagram of a system including a characteristic prediction device and a trained model generation device according to the first embodiment. [Figure 6] Figure 6 is a block diagram of the information processing device in the first embodiment. [Figure 7] Figure 7 is a functional block diagram of the trained model generation device according to the first embodiment. [Figure 8] Figure 8 shows the data array of the training data in the first embodiment. [Figure 9] Figure 9 is a functional block diagram of the characteristic prediction device according to the first embodiment. [Figure 10] Figure 10 shows the data array of the input data in the first embodiment. [Figure 11]FIG. 11 is a diagram showing a data array of prediction data in the first embodiment. [Figure 12] FIG. 12 is a diagram showing the arrangement order of wafers in a prediction example. [Figure 13] FIG. 13 is a diagram showing a prediction model used in a prediction example. [Figure 14] FIG. 14 is a diagram showing the processing of a prediction model in time series in a prediction example. [Figure 15] FIG. 15 is a diagram showing the processing of a prediction model in time series in a prediction example. [Figure 16] FIG. 16 is a diagram showing a leakage current with respect to a serial number in a prediction example. [Figure 17] FIG. 17 is a diagram showing a leakage current with respect to a serial number in a prediction example. [Figure 18] FIG. 18 is a diagram showing another example of the arrangement order of wafers. [Figure 19] FIG. 19 is a flowchart showing a method for manufacturing a semiconductor device according to the second embodiment.

Embodiments for Carrying Out the Invention

[0012] [Description of Embodiments of the Present Disclosure] First, the embodiments of the present disclosure will be listed and described.

[0013] (1) Embodiments of the present disclosure are characteristic prediction methods comprising: obtaining a trained model that defines the relationship between a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, a sequence of wafers in a predetermined arrangement order within the processing apparatus that perform the first process simultaneously in each of the plurality of first processes, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and inputting the first sequence of wafers in the first process performed in the processing apparatus and the measured first characteristics corresponding to the first sequence into the trained model to predict the second characteristics corresponding to the second sequence after the first sequence, wherein the trained model includes a time-series model with the sequence in a time series. This makes it possible to appropriately predict the characteristics. (2) In (1) above, the time series model may include an LSTM, a Transformer, or a GRU. This can improve the accuracy of predicting the characteristics. (3) In (1) or (2) above, the first sequence number, the second sequence number, the first characteristic, and the second characteristic may be input to the trained model to predict the third characteristic corresponding to the third sequence number after the second sequence number. This makes it possible to predict the third characteristic of the third sequence number that follows the second sequence number. (4) In any of (1) to (3) above, the trained model may output a characteristic corresponding to a sequence number two or more predetermined numbers later for a sequence number corresponding to an input characteristic. This improves the accuracy of characteristic prediction. (5) In any of (1) to (4) above, the first sequential number may include a number of numbers equal to or greater than the number of wafers processed simultaneously by the processing device. This improves the accuracy of predicting the characteristics. (6) In any of (1) to (5) above, the processing apparatus may be a semiconductor device manufacturing apparatus. This allows for accurate prediction of the characteristics of the semiconductor device. (7) In any of (1) to (5) above, the processing apparatus is an epitaxial growth apparatus, the first process is a step of forming a semiconductor epitaxial layer on the substrate, the second process includes a step of forming electrodes on the semiconductor epitaxial layer, and the characteristics may be electrical characteristics measured using the electrodes. This makes it possible to predict the characteristics of a semiconductor device with high accuracy. (8) In (7) above, the semiconductor epitaxial layer may include a nitride semiconductor layer. This allows for accurate prediction of the characteristics of the nitride semiconductor device. (9) Embodiments of the present disclosure are methods for manufacturing a semiconductor device, comprising the steps of: performing a characteristic prediction method in any of (1) to (8); changing the conditions of the first or second process based on the second characteristic; and performing the first or second process on a wafer corresponding to the second serial number using the changed conditions. This makes it possible to improve the characteristics of the semiconductor device. (10) Embodiments of the present disclosure include a computer that functions as a characteristic prediction program that includes a time-series model of the sequence numbers obtained by arranging a plurality of first processes in a processing apparatus that arranges a plurality of wafers and simultaneously performs a first process, and arranging the plurality of wafers that simultaneously perform the first process in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process, and the computer that functions as a character prediction program that inputs the first sequence numbers of the plurality of wafers in the first process performed in the processing apparatus and the measured first characteristics corresponding to the first sequence numbers into the character prediction program, and the character prediction program that includes a time-series model of the sequence numbers in time series. This makes it possible to appropriately predict the characteristics. (11) Embodiments of the present disclosure are characteristic prediction devices comprising: an acquisition unit that acquires a learned model that defines the relationship between a plurality of first processes performed in a processing device that arranges a plurality of wafers and performs a first process simultaneously, a plurality of first processes performed in a plurality of first processes arranged in a predetermined arrangement order within the processing device, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and a prediction unit that inputs the first sequential numbers of the plurality of wafers in the first process performed in the processing device and the measured first characteristics corresponding to the first sequential numbers into the learned model, wherein the learned model includes a time-series model with the sequential numbers in a time series. This enables appropriate prediction of characteristics. (12) Embodiments of the present disclosure are methods for generating a trained model, comprising: a step of obtaining training data relating a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, arranged in a time series, and a sequence number obtained by arranging the plurality of wafers that perform the first process simultaneously in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process; and a step of generating a trained model for predicting a second characteristic corresponding to a second sequence number after the first sequence number by using machine learning on the training data, inputting a first sequence number of the plurality of wafers in the first process performed in the processing apparatus and a first characteristic corresponding to the measured first sequence number, wherein the trained model includes a time series model with the sequence numbers as a time series. This enables appropriate prediction of characteristics. (13) Embodiments of the present disclosure include a memory and a processor which acquires a trained model that defines the relationship between a plurality of first processes performed in a processing apparatus which arranges a plurality of wafers and performs a first process simultaneously, a sequence of wafers in which the plurality of wafers that perform the first process simultaneously in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the plurality of wafers measured after another second process has been performed on the plurality of wafers that have performed the first process, and inputs the first sequence of wafers in the first process performed in the processing apparatus and the measured first characteristics corresponding to the first sequence to the trained model to predict a second characteristic corresponding to a second sequence after the first sequence, wherein the trained model includes a time series model in which the sequence is arranged in time series, thereby enabling appropriate prediction of characteristics.

[0014] [Details of the embodiments of this disclosure] Specific examples of the characteristic prediction method, semiconductor device manufacturing method, characteristic prediction program, characteristic prediction apparatus, and trained model generation method according to embodiments of this disclosure will be described below with reference to the drawings. However, this disclosure is not limited to these examples and is intended to include all modifications within the meaning and scope of the claims as indicated by the claims.

[0015] At least some of the embodiments described below may be combined in any way. The characteristic prediction device is configured to include a computer, and each function of the characteristic prediction device is performed by the execution of a computer program stored in the computer's storage device by the computer's CPU (Central Processing Unit). The computer program can be stored on a storage medium such as a CD-ROM (Compact Disc Read Only Memory) or a DVD (Digital Versatile Disc).

[0016] (First Embodiment) (Explanation of the process) In the first embodiment, the processing of a wafer for predicting its characteristics will be described. The wafer is, for example, a semiconductor wafer, and the processing is, for example, a process in the manufacturing process of a semiconductor device. Figure 1 is a flowchart of the processing for predicting characteristics in the first embodiment. As shown in Figure 1, the wafer is prepared (step S10). The wafer is a wafer that has completed the processing up to the step before the first processing.

[0017] Next, the first process is performed (step S11). The first process is performed on multiple wafers simultaneously. The first process is, for example, a process using a batch-type processing apparatus, and is, for example, a film deposition process to grow a film on a wafer, an etching process to etch a part of the wafer, or a surface treatment to treat the surface of the wafer. For the film deposition process, a film deposition apparatus such as a CVD (Chemical Vapor Deposition) apparatus or a PVD (Physical Vapor Deposition) apparatus is used. For the etching process, an etching apparatus such as a dry etching apparatus or a wet etching apparatus is used. For the surface treatment, a plasma surface treatment apparatus using plasma, or a surface treatment apparatus using wet treatment is used.

[0018] Next, a second process is performed (step S12). The second process involves a film deposition process, etching process, or surface treatment on the wafer that has undergone the first process. The second process may consist of multiple processes. For example, the second process may include a process for manufacturing a semiconductor device.

[0019] Next, the wafer characteristics are measured (step S13). The wafer characteristics are electrical characteristics. For example, if the second process includes a process of forming electrodes, the wafer characteristics may also be electrical characteristics measured electrically using the electrodes. The wafer characteristics may also be physical characteristics of the wafer, such as the width of the pattern, the depth of the pattern, and the thickness of the film. The wafer characteristics may also be optical characteristics, such as the refractive index. The wafer characteristics may be one type of characteristic or multiple types of characteristics. Then the process ends. After completion, other processes may be performed on the wafer.

[0020] (Examples of nitride semiconductor manufacturing) As an example of wafer processing, the manufacturing of nitride semiconductor devices will be described. Figures 2 and 3 are cross-sectional views showing the manufacturing method of nitride semiconductor devices. As shown in Figure 2, a substrate 10 is prepared as the wafer in step S10. The substrate 10 is, for example, a silicon carbide substrate, a sapphire substrate, or a diamond substrate.

[0021] As the first process in step S11, a semiconductor layer 12 is deposited on the substrate 10 using the MOCVD (Metal Organic CVD) method. The semiconductor layer 12 is, for example, a nucleation layer 12A, an electron transport layer 12B, and an electron supply layer 12C. The nucleation layer 12A is, for example, an aluminum nitride (AlN) layer. The electron transport layer 12B is, for example, a gallium nitride (GaN) layer. The electron supply layer 12C is, for example, an aluminum gallium nitride (AlGaN) layer. The gas used when growing the gallium nitride layer is, for example, trimethylgallium (TMG) gas and ammonia gas. The gas used when growing the aluminum nitride layer is, for example, trimethylaluminum (TMA) gas and ammonia gas. The gas used when growing the aluminum gallium nitride layer is, for example, TMA gas, TMG gas, and ammonia gas. Triethylaluminum gas and triethylgallium gas may be used instead of TMA gas and TMG gas, respectively.

[0022] Next, as shown in Figure 3, as the second process of step S12, an insulating layer 17 is formed on the semiconductor layer 12. The insulating layer 17 is, for example, a silicon nitride layer and is formed using CVD. Openings 14A and 15A are formed in the insulating layer 17. Openings 14A and 15A are formed, for example, using photolithography and etching. Source electrodes 14 and drain electrodes 15 are formed within openings 14A and 15A, respectively. The source electrodes 14 and drain electrodes 15 are, for example, a titanium layer and an aluminum layer from the semiconductor layer 12 side and are formed using vacuum deposition and lift-off methods. An opening 16A is formed in the insulating layer 17 between the source electrode 14 and the drain electrode 15. The opening 16A is formed, for example, using photolithography and etching. A gate electrode 16 is formed within opening 16A. The gate electrode 16 is, for example, a nickel layer and a gold layer from the semiconductor layer 12 side and is formed using vacuum deposition and lift-off methods.

[0023] Based on the above, a GaN HEMT (Gallium Nitraide High Electron Mobility Transistor) is manufactured as the semiconductor device 18. In step S13, the electrical characteristics of the semiconductor device 18 are measured. The electrical characteristic can be leakage current. The leakage current is measured, for example, as follows: A negative voltage is applied to the gate electrode 16 to deplete the upper parts of the electron supply layer 12C and the electron transport layer 12B. A voltage (e.g., 100V) is applied between the source electrode 14 and the drain electrode 15 to measure the leakage current flowing between the source electrode 14 and the drain electrode 15.

[0024] As shown in Figure 3, if the number of steps in the second process is large, it may take more than a month from the first process to the measurement of characteristics. In order to stably manufacture the semiconductor device 18, it may be necessary to predict the characteristics before measuring the characteristics in step S13. For example, if the predicted characteristics are not the desired characteristics, the actual characteristics can be made to the desired characteristics by changing the conditions of the first or second process.

[0025] The electrical properties of a GaN HEMT are influenced by the growth of the semiconductor layer 12 shown in Figure 2. For example, leakage current is the current flowing in the region of the semiconductor layer 12 close to the substrate 10. Therefore, the film quality of the semiconductor layer 12 affects the leakage current. Thus, the leakage current is predicted before or immediately after the deposition of the semiconductor layer 12.

[0026] In processing equipment, the processing order can affect the characteristics. For example, in MOCVD equipment, a wafer 40 is introduced into a chamber and a raw material gas is supplied to deposit a semiconductor layer 12 on the wafer 40. At this time, the film quality of the semiconductor layer 12 may change depending on the conditions inside the chamber. For example, when the semiconductor layer 12 is deposited, product deposits adhere to the inside of the chamber. If the film quality of the semiconductor layer 12 changes depending on the amount of product deposits, the film quality of the semiconductor layer 12 will change with each processing cycle, and the leakage current of the GaN HEMT will change. When the inside of the chamber is cleaned, the product inside the chamber is removed, so the film quality of the semiconductor layer 12 is reset, and the leakage current is also reset.

[0027] Therefore, it was considered possible to predict unknown characteristics from the processing order based on information relating the processing order and characteristics of past processing devices. As described later, when we used machine learning to predict characteristics from the processing order based on information relating the processing order and characteristics, the results differed significantly from the actual characteristics. As a reason for this, we focused on the arrangement order in batch-type processing devices.

[0028] Figure 4 is a plan view illustrating batch processing in the first embodiment. Figure 4 shows the arrangement of wafers in a MOCVD apparatus as a processing apparatus for performing the first processing. Multiple wafers 40 are arranged within a circular susceptor 42. The wafers 40 are arranged concentrically. Six wafers 40 are arranged circumferentially on the inside, and twelve wafers 40 are arranged circumferentially on the outside of the six. Figure 4 is an example of batch processing, and the number and arrangement method of wafers 40 will differ depending on the processing apparatus.

[0029] In MOCVD equipment, the film quality of the semiconductor layer 12 differs depending on the placement of the wafer 40 due to factors such as the temperature distribution of the susceptor 42, the location where the raw material gas is introduced into the chamber, and the location where the gas is exhausted from the chamber. For this reason, we considered that the placement order of the wafer 40, as well as the processing order, is important. As described later, by using machine learning on processing placement information that correlates the processing order with the placement order and characteristics, we were able to predict characteristics that closely matched the actual characteristics based on the processing order.

[0030] The first embodiment can also be applied to batch-type processing equipment such as film deposition equipment, etching equipment, or surface treatment equipment other than MOCVD equipment, if the subsequent characteristics depend on the placement position of the wafer 40, etc.

[0031] (Example of a system) Figure 5 is a block diagram of a system including a characteristic prediction device and a trained model generation device according to the first embodiment. As shown in Figure 5, the characteristic estimation system of the first embodiment includes one or more information processing devices 20 to 23. The information processing devices 20 to 23 are, for example, computers, and may be portable or stationary. The information processing devices 20 to 23 are connected to a network 25. The network 25 is, for example, a wireless or wired LAN (Local Area Network). Information processing device 20 is a terminal used by the user to predict characteristics. Information processing device 21 is a terminal to which a storage device 24 that stores data and trained models is connected. The storage device 24 is, for example, a semiconductor storage device, an optical storage device, or a magnetic storage device. Information processing device 22 is a terminal that inputs processing information such as the processing order and arrangement order of the processing device 27. The processing information may be automatically sent from the processing device 27 to the information processing device 22, or the user may input it into the information processing device 22. Information processing device 23 is a terminal that inputs characteristic information measured by the measuring device 28. The characteristic information may be automatically sent from the measuring device 28 to the information processing device 23, or it may be entered by the user into the information processing device 23.

[0032] At least two of the information processing devices 20 to 23 may be operated by a single information processing device. Alternatively, one information processing device may operate on all three of the information processing devices 20 to 23.

[0033] (Computer block diagram) Figure 6 is a block diagram of the information processing device in the first embodiment. The computer 30, which is the information processing device 20 to 23, includes a processor 32, memory 34, input / output device 36, and internal bus 38. The processor 32 is, for example, a CPU (Central Processing Unit) and executes a characteristic prediction program, a trained model generation program, a characteristic prediction method, and a trained model generation method (hereinafter also simply referred to as the program and method). The memory 34 is, for example, volatile memory or non-volatile memory and stores data used by the processor 32 when executing the program and method. The memory 34 may also store the program executed by the processor 32. The input / output device 36 inputs data acquired by the processor 32 from an external device and outputs data output by the processor 32 to the external device. The external device is another computer or another program within the same computer. The internal bus 38 connects the processor 32, memory 34, and input / output device 36 and transmits data. The program is stored in the storage medium 35. The storage medium 35 is, for example, a tangible medium that is not temporary, such as a CD-ROM or DVD.

[0034] (Generating pre-trained models) Figure 7 is a functional block diagram of the trained model generation device according to the first embodiment. As shown in Figure 7, the trained model generation device 50 includes an acquisition unit 51, a training data generation unit 52, a model generation unit 53, and an output unit 54. The information processing device 21 works in cooperation with the program to function as the acquisition unit 51, the training data generation unit 52, the model generation unit 53, and the output unit 54.

[0035] The acquisition unit 51 acquires the processing order, placement order, and characteristics. The processing order is information indicating the chronological order in which the processing device 27 processed the wafer in step S11 of Figure 1. The placement order is information indicating the position in which the wafer was placed within the processing device 27 in step S11. The placement order is, for example, information on the position within the susceptor 42 where the wafer 40 was placed in Figure 4. The characteristics are information on the characteristics of the wafer measured in step S13 after the second processing in step S12 of Figure 1.

[0036] The training data generation unit 52 generates training data that associates sequential numbers with characteristics based on the processing order, arrangement order, and characteristics. Figure 8 shows the data arrangement of the training data in the first embodiment. "Processing order" is the order of processing in the processing device 27, where "1" indicates the first processing, "2" indicates the second processing, and "M" indicates the Mth processing. "Arrangement order" is the arrangement position of the wafers in the processing device 27, where "1" indicates the first position, "2" indicates the second position, and "N" indicates the Nth position. "Sequential number" is the number obtained by arranging wafers that have undergone the same processing in arrangement order. The training data generation unit 52 sets the "sequential numbers" of wafers with processing order "1" and arrangement order "1", "2", and "N" to "1", "2", and "N", respectively. The training data generation unit 52 assigns the "sequence numbers" of wafers with processing order "M" and placement order "1", "2", and "N" to "(M-1)N+1", "(M-1)N+2", and "MN", respectively. "Characteristics" are characteristic information of the wafer. The "characteristics" range from X(1) to X(MN) depending on the "processing order" and "placement order". The training data generation unit 52 generates training data relating the "sequence numbers" from 1 to MN to the "characteristics" from A(1) to A(MN). The training data generation unit 52 may generate training data each time it acquires the processing order, placement order, and characteristics, or it may generate training data at regular intervals. The training data generation unit 52 stores the generated training data in the memory 34 or storage device 24.

[0037] Returning to Figure 7, the model generation unit 53 acquires training data. The model generation unit 53 generates a trained model by machine learning the training data. The model generation unit 53 generates a trained model using a time-series model, treating the sequential numbers in the training data as a time series. The time-series model is, for example, an RNN (Recurrent Neural Network). As a time-series model, a model capable of long-term memory may be used, such as an LSTM (Long Short Term Memory) model, a Transformer model, or a GRU (Gated Recurring Unit) model. The model generation unit 53 may generate a trained model each time the training data generation unit 52 generates training data, or it may generate a trained model at predetermined intervals. The model generation unit 53 functions, for example, in the information processing device 21 in Figure 5.

[0038] The output unit 54 outputs the trained model generated by the model generation unit to the memory 34 or storage device 24.

[0039] (Characteristic prediction) Figure 9 is a functional block diagram of the characteristic prediction device according to the first embodiment. As shown in Figure 9, the characteristic prediction device 55 includes acquisition units 56A and 56B, an input data generation unit 57, a prediction unit 58, and an output unit 59. The information processing device 20 functions in cooperation with the program as the acquisition units 56A and 56B, the input data generation unit 57, the prediction unit 58, and the output unit 59.

[0040] The acquisition unit 56A acquires a first processing sequence, a first arrangement sequence, and a first characteristic that is different from the processing sequence, arrangement sequence, and characteristics acquired by the acquisition unit 51 of the trained model generation device 50. The first processing sequence includes processing sequences that follow the processing sequence in Figure 7. Part of the first processing sequence may overlap with the processing sequence acquired by the acquisition unit 51. The input data generation unit 57 generates input data using the same method as the training data generation unit 52 generates training data.

[0041] Figure 10 shows the data array of the input data in the first embodiment. The "first processing order" is from "1" to "L". The "first arrangement order" is from "1" to "N". The "first sequence number" is from "1" to "LN". The "first characteristic" is from X(1) to X(LN) according to the "first processing order" and the "first arrangement order". The input data generation unit 57 generates input data relating the "first sequence number" from 1 to LN and the "first characteristic" from A(1) to A(LN).

[0042] Returning to Figure 9, the acquisition unit 56B acquires the trained model from memory 34 or storage device 24. The prediction unit 58 acquires the input data and the trained model. The prediction unit 58 predicts the second characteristic for the second sequence number by inputting the input data into the trained model.

[0043] Figure 11 shows the data array of the prediction data in the first embodiment. The "second processing order" is from "L+1" to "L+K". The "second arrangement order" is from "1" to "N". The "second sequence number" is from "LN+1" to "(L+K)N". The "second characteristic" is from X(LN+1) to X((L+K)N)) according to the "second processing order" and the "second arrangement order". In this way, the prediction unit 58 predicts the second characteristic X(LN+1) to X((L+K)N) corresponding to the second sequence number LN+1 to (L+K)N in the processing orders L+1 to L+K (second processing order) from the last processing order L of the first processing order. The prediction unit 58 may also expand the second sequence number to the second processing order and the second arrangement order.

[0044] Returning to Figure 9, the output unit 59 outputs the second characteristics predicted by the prediction unit 58 to an external device or storage device 24, corresponding to the second processing order and the second arrangement order.

[0045] (Example of prediction) We will explain a prediction example where the processing apparatus is an MOCVD apparatus for depositing the semiconductor layer 12 shown in Figure 2, and its characteristics are those of the GaN-HMET leakage current shown in Figure 3.

[0046] In Figure 2, the deposited semiconductor layer 12 consists of an aluminum nitride nucleation layer 12A, a gallium nitride electron transport layer 12B, and an aluminum gallium nitride electron supply layer 12C. The source gases used were TMA gas, TMG gas, and ammonia gas.

[0047] Figure 12 shows the wafer arrangement order in a prediction example. As shown in Figure 12, in the susceptor 42, 6 wafers 40 are arranged on the inner circumference and 12 wafers are arranged on the outer circumference. The arrangement order is numbered counterclockwise starting from "1" on the inner circumference, with the wafers outside "1" being numbered "7" and so on, and so on, counterclockwise.

[0048] (Predictive model) Figure 13 shows the prediction model used in the prediction example. As shown in Figure 13, the prediction model 60 comprises an input layer 61, a hidden layer 62, and an output layer 63. The prediction model 60 includes an LSTM model and is a trained model generated using a trained model generation device. The trained model was generated using training data that correlates sequential numbers and characteristics on 21,963 wafers.

[0049] Figures 14 and 15 show the time-series processing of the prediction model in a prediction example. This example uses input data from 72 wafers. As shown in Figure 14, input data 64 is input to the input layer 61 in time series. Input data X1 to X72 are characteristics corresponding to sequence numbers 1 to 72. When input data X70 to X72 are input, the data 65A output to the output layer 63 are denoted as X73 to X75. Data X73 to X75 are predicted data for the characteristics of sequence numbers 73 to 75.

[0050] The prediction model 60 is machine-trained so that the characteristic corresponding to the third sequential number following the characteristic input to the input layer 61 is output to the output layer 63.

[0051] As shown in Figure 15, after inputting the characteristics of input data 64 with sequence number 72 into the input layer 61, X73 of data 65A is input into the input layer 61, followed by X74 and X75 being input into the input layer 61 sequentially. When data 65A is input into the input layer 61, the data 65B output from the output layer 63 are X76 to X78, respectively. Subsequently, by inputting data 65B into the input layer 61, X79 to X81 are output to the output layer 63. By repeating this process sequentially, the characteristics from X73 onwards can be predicted.

[0052] (Number of input data) Leakage current was predicted using 72 and 9 data points as input data (number of wafers). Figure 16 shows the leakage current for each sequence number in the prediction example. In Figure 16, the measured data are the measured values ​​of the leakage current for each sequence number. The 72 data points are data in which leakage currents from sequence numbers 1 to 72 were input as input data, and the leakage currents from sequence number 73 onwards were predicted. The 9 data points are data in which leakage currents from sequence numbers 64 to 72 were input as input data, and the leakage currents from sequence number 73 onwards were predicted.

[0053] As shown in Figure 16, the data for 9 points did not predict the actual measured data. For 72 points, the data for the sequence numbers up to 140th position, as well as the positions of the peaks and bottoms relative to the sequence numbers, were in relatively good agreement with the actual measured data.

[0054] As described above, by setting the number of input data points to at least equal to the number of wafers processed in a batch, the predicted data will be in relatively good agreement with the measured data.

[0055] (Order of arrangement) Leakage current was predicted for two cases: when the arrangement order within the same process for sequentially numbered data points was as shown in Figure 12, and when the arrangement order was randomly assigned. Figure 17 shows the leakage current for sequentially numbered data points in the prediction example. In Figure 17, "Measured Data" represents the measured leakage current for sequentially numbered data points. "Arranged Order" refers to the arrangement order as shown in Figure 12, while "Random" refers to the arrangement order being random. The number of input data points is 72 in both cases.

[0056] As shown in Figure 17, with random selection, leakage currents for sequential numbers 73 and above could not be predicted. With sequential selection, the peak and bottom positions for sequential numbers up to around 240 were relatively consistent with the measured data.

[0057] As described above, when there is a set arrangement order, the predicted data matches the actual data better than when the arrangement order is random.

[0058] Figure 18 shows another example of wafer arrangement order. As shown in Figure 18, the arrangement order of the wafers 40 in the susceptor 42 is such that wafer 40 adjacent to wafer 40 "1" on the outer perimeter is numbered "2", wafer 40 adjacent to wafer 40 "2" on the outer perimeter is numbered "3", and so on. In this way, sequential numbers may be assigned in order of angle with respect to the center of the susceptor 42. Sequential numbers may also be assigned sequentially to adjacent wafers 40 among a plurality of adjacent wafers 40.

[0059] In the first embodiment, as shown in Figures 7 and 8, the training data generation unit 52 arranges the first processes in chronological order as sequential numbers, and arranges multiple wafers that simultaneously perform the first process in a predetermined arrangement order within the processing unit. The training data generation unit 52 provides training data that relates the sequential numbers to the characteristics of each of the multiple wafers measured after another second process has been performed on the multiple wafers that have undergone the first process. The model generation unit 53 acquires the training data and generates a trained model that defines the relationship between the sequential numbers and characteristics by using machine learning on the training data.

[0060] As shown in Figures 9 to 11, the input data generation unit 57 takes the first sequence numbers of multiple wafers in the first processing performed in the processing device and the measured first characteristics corresponding to the first sequence numbers as input data. The prediction unit 58 acquires the input data, inputs the input data into a trained model, and predicts the second characteristics corresponding to the second sequence number following the first sequence number. The trained model includes a time series model in which the sequence numbers are arranged in time series. In this way, by inputting the input data, which includes the sequence numbers including the arrangement order in processing order, into the time series model, the characteristics can be appropriately predicted.

[0061] The time series model may include an LSTM, Transformer, or GRU capable of long-term memory. This allows for the storage of outputs from several processing steps prior, thereby improving the accuracy of predicting characteristics.

[0062] As shown in Figure 15, in addition to the first sequence numbers (1 to 72) and first characteristics (X1 to X72), the second sequence numbers (73 to 75) and second characteristics (X73 to X75) are input into the trained model to predict the third characteristics (X76 and later) corresponding to the third sequence numbers (76 and later) that follow the second sequence number. This makes it possible to predict the third characteristics of the third sequence numbers that follow the second sequence number.

[0063] As shown in Figures 14 and 15, the trained model outputs a characteristic corresponding to a sequence number two or more predetermined numbers later (three in Figures 14 and 15) for each sequence number corresponding to the input characteristic. This improves the accuracy of characteristic prediction. The predetermined number can be appropriately selected to improve the accuracy of characteristic prediction.

[0064] As shown in Figure 16, the first sequence number (1 to 72) includes a number of numbers equal to or greater than the number of wafers processed simultaneously by the processing unit (18). This improves the accuracy of characteristic prediction. From the viewpoint of improving prediction accuracy, the first sequence number can be more than twice the number of wafers processed simultaneously, or more than three times the number of wafers processed simultaneously.

[0065] The processing device is a semiconductor device manufacturing device. In batch-type semiconductor device manufacturing devices, the characteristics may depend on the wafer placement position. Therefore, by predicting the characteristics using the first embodiment, the characteristics of the semiconductor device can be predicted with high accuracy. In addition, in the semiconductor device manufacturing process, many steps are performed from the time the wafer is processed using the manufacturing device until the characteristics are measured, and it takes a long time before the characteristics can be measured. Therefore, by predicting the characteristics using the first embodiment, the occurrence of defective products before the characteristics can be measured can be reduced.

[0066] The processing apparatus is an epitaxial growth apparatus. The first process is the process of forming a semiconductor layer 12 (semiconductor epitaxial layer) on a substrate 10. The second process includes the process of forming electrodes on the semiconductor layer 12. The characteristics are electrical characteristics measured using the electrodes. In an epitaxial growth apparatus, the film quality of the semiconductor epitaxial layer depends on the wafer placement position. As a result, the electrical characteristics depend on the wafer placement order in the epitaxial growth apparatus. Therefore, by predicting the electrical characteristics using the first embodiment, the characteristics of the semiconductor device can be predicted with high accuracy. In addition, many processes are performed from epitaxial growth to measurement of electrical characteristics, and it takes a long time before the electrical characteristics can be measured. Therefore, by predicting the characteristics using the first embodiment, the occurrence of defective products before the electrical characteristics can be measured can be reduced.

[0067] When a semiconductor epitaxial layer includes a nitride semiconductor layer, the electrical properties depend on the wafer arrangement order in the epitaxial growth apparatus. Therefore, by predicting the electrical properties using the first embodiment, the properties of nitride semiconductor devices can be predicted with high accuracy. .

[0068] In the first embodiment, the first sequence number and the first characteristic are input to a trained model in which the relationship between sequence numbers and characteristics is defined, and the second characteristic corresponding to the second sequence number is predicted. In addition to the sequence number, the processing conditions of the first process may also be used as an explanatory function. That is, the first sequence number, the first processing conditions, and the first characteristic may be input to a trained model in which the relationship between sequence numbers, processing conditions, and characteristics is defined, and the second characteristic corresponding to the second sequence number and the second processing conditions may be predicted.

[0069] (Second Embodiment) The second embodiment is an example of a method for manufacturing a semiconductor device that includes the characteristic prediction method of the first embodiment. Figure 19 is a flowchart of the method for manufacturing a semiconductor device according to the second embodiment. As shown in Figure 19, the second characteristic is predicted using the characteristic prediction method of the first embodiment (step S20). It is determined whether the second characteristic is within the target range (step S21). If yes, the first process (step S11) to measurement (step S13) is performed without changing the conditions of the first or second process.

[0070] In step S21, if the result is No, the conditions for the first or second process are changed according to the second characteristic (step S22). The conditions include substrate temperature, gas flow rate, and vacuum level. Subsequently, the first or second process is performed using the changed conditions. The details from the first process to measurement are the same as in Figure 1 and are therefore omitted from the explanation.

[0071] According to the second embodiment, as shown in step S20 of Figure 19, the characteristic prediction method of the first embodiment is performed to predict the second characteristic corresponding to the second serial number. As shown in steps S21 and S22, the conditions for the first or second process are changed based on the predicted second characteristic. In step S11 or S12, the first process or process is performed on the wafer corresponding to the second serial number using the changed conditions. This makes it possible to reduce the likelihood of the characteristics falling outside the target range by changing the conditions for the first or second process if it is predicted that the characteristics will fall outside the target range before performing the first or second process. Thus, the characteristics of the semiconductor device can be improved.

[0072] If the conditions for the second process are to be changed, steps S21 and S22 may be executed between steps S11 and S12.

[0073] If the first process is the process of depositing the semiconductor layer 12 shown in Figure 2, the conditions of the first process may be changed by, for example, performing the first process after cleaning the chamber of the MOCVD apparatus. This removes the product in the chamber, thereby initializing the film quality of the semiconductor layer 12. Thus, the conditions of the first process may also be the conditions of the inner surface of the chamber.

[0074] The above-mentioned processor may be any type of processor suitable for computer control, such as a CPU, GPU (Graphics Processing Unit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), or ASIC (Application Specification Integrated Circuit). Furthermore, multiple physically separated processors may cooperate with each other to perform the above-mentioned processes. For example, processors installed in multiple physically separated computers may cooperate with each other via a network such as a LAN (Local Area Network), WAN (Wide Area Network), or the Internet to perform the above-mentioned processes.

[0075] The above program may be installed into the memory via the network from an external server device, or it may be distributed on a recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory, and then installed into the memory from that recording medium.

[0076] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this disclosure is indicated by the claims, not in the sense described above, and all modifications in the sense and scope equivalent to the claims are intended. [Explanation of symbols]

[0077] 10 circuit boards 12 Semiconductor Layers 12A Nucleation layer 12B Electron Transport Layer 12C electron supply layer 14 Source electrodes 14A, 15A, 16A aperture 15 Drain electrode 16 Guard Station 17. Insulating layer 18 Semiconductor Equipment 20, 21, 22, 23 Information Processing Devices 24 Storage device 25 Networks 27 Processing Unit 28 Measuring device 30 Computers 32 processors 34 memory 35 Storage medium 36 Input / Output Devices 38 Internal bus 40 wafers 42 Susceptors 50 Pre-trained model generator 51, 56A, 56B Acquisition Department 52 Training Data Generation Unit 53 Model Generation Unit 55 Characteristic prediction device 57 Input Data Generation Unit 58 Prediction Section 59 Output section 60 Predictive Models 61 Input Layer 62 Hidden Layers 63 Output Layer 64 Input Data 65A, 65B Data

Claims

1. Steps include obtaining a trained model that defines the relationship between a sequence of wafers arranged in a processing apparatus that simultaneously performs a first process on multiple wafers, arranged in chronological order, and the wafers in which the first process is performed simultaneously in each of the multiple first processes, arranged in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the multiple wafers measured after another second process is performed on the wafers on which the first process has been performed; The steps include inputting the first sequence numbers of a plurality of wafers in the first processing performed in the apparatus and the measured first characteristics corresponding to the first sequence numbers into the trained model to predict the second characteristics corresponding to the second sequence numbers following the first sequence numbers, Includes, The aforementioned trained model includes a time series model in which the sequence numbers are used as a time series, and is a method for predicting characteristics.

2. The characteristic prediction method according to claim 1, wherein the time series model includes LSTM, Transformer, or GRU.

3. A characteristic prediction method according to claim 1 or 2, comprising inputting the first sequence number, the second sequence number, the first characteristic, and the second characteristic into the trained model to predict the third characteristic corresponding to the third sequence number after the second sequence number.

4. The characteristic prediction method according to claim 1 or claim 2, wherein the trained model outputs a characteristic corresponding to a sequence number two or more predetermined numbers after a sequence number corresponding to an input characteristic.

5. The characteristic prediction method according to claim 1 or 2, wherein the first sequential number includes a number of numbers equal to or greater than the number of wafers processed simultaneously by the processing device.

6. The characteristic prediction method according to claim 1 or 2, wherein the processing apparatus is a semiconductor device manufacturing apparatus.

7. The aforementioned processing apparatus is an epitaxial growth apparatus, The first process is a step of forming a semiconductor epitaxial layer on the substrate, The second process includes the step of forming an electrode on the semiconductor epitaxial layer, The characteristic prediction method according to claim 1 or claim 2, wherein the characteristic is an electrical characteristic measured using the electrode.

8. The characteristic prediction method according to claim 7, wherein the semiconductor epitaxial layer includes a nitride semiconductor layer.

9. By performing the characteristic prediction method described in claim 1 or claim 2, A step of changing the conditions of the first or second process based on the second characteristic, The steps include: performing the first or second process on the wafer corresponding to the second serial number using the modified conditions; A method for manufacturing a semiconductor device containing [a specific component].

10. Computers, An acquisition unit acquires a trained model that defines the relationship between a sequence of wafers arranged in a processing apparatus that arranges multiple wafers and performs a first process simultaneously, the sequence of wafers that perform the first process simultaneously in each of the multiple first processes arranged in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the multiple wafers measured after another second process has been performed on the wafers that have performed the first process. A prediction unit inputs the first sequence numbers of a plurality of wafers in the first processing performed in the processing apparatus and the measured first characteristics corresponding to the first sequence numbers into the learned model to predict the second characteristics corresponding to the second sequence numbers following the first sequence numbers, To make it function as, The aforementioned trained model is a characteristic prediction program that includes a time series model in which the aforementioned sequential numbers are used as a time series.

11. An acquisition unit acquires a trained model that defines the relationship between a processing apparatus that arranges multiple wafers and simultaneously performs a first process, the multiple first processes performed in a time series, and the sequential numbers obtained by arranging the multiple wafers that simultaneously perform the first process in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the multiple wafers measured after another second process is performed on the multiple wafers that have performed the first process. A prediction unit inputs the first sequence numbers of a plurality of wafers in the first processing performed in the processing apparatus and the measured first characteristics corresponding to the first sequence numbers into the learned model to predict the second characteristics corresponding to the second sequence numbers following the first sequence numbers, Equipped with, The aforementioned trained model includes a time-series model in which the sequential numbers are arranged as a time series, and is a characteristic prediction device.

12. A step of acquiring training data relating a plurality of first processes performed in a processing apparatus that arranges a plurality of wafers and performs a first process simultaneously, arranged in chronological order, and a sequence number obtained by arranging the plurality of wafers that perform the first process simultaneously in each of the plurality of first processes in a predetermined arrangement order within the processing apparatus, and the characteristics of each of the plurality of wafers measured after another second process is performed on the plurality of wafers that have performed the first process, The steps include: generating a trained model for predicting a second characteristic corresponding to a second sequential number following the first sequential number by using machine learning on the aforementioned training data, inputting the first sequential number of a plurality of wafers in a first process performed in the processing apparatus and the first characteristic corresponding to the measured first sequential number; Includes, A method for generating a trained model, wherein the trained model includes a time-series model in which the sequential numbers are used as a time series.

Images