Information processing apparatus, program, and process condition search method

Abstract

The present application provides an information processing device, a program, and a process condition search method that use a machine learning model of a semiconductor manufacturing device to search for a process condition that achieves a target process result. To solve the above problem, the information processing device generates a machine learning model of a semiconductor manufacturing device that performs processing in accordance with a process condition, searches for a process condition that achieves a target process result using the machine learning model, and includes: a module that selects a machine learning model suitable for a data set from among a plurality of machine learning models generated using a plurality of regression methods; a module that performs optimization calculation using the selected machine learning model, calculates a plurality of process conditions, a predicted value of a process result, and a reliability of the predicted value; a module that selects one or more process conditions in accordance with the predicted value of the process result and the reliability of the predicted value; and a module that displays the selected process conditions, the predicted value of the process result, and the reliability of the predicted value.

Description

TECHNICAL FIELD

[0001] The present application relates to an information processing apparatus, a program, and a process condition search method. BACKGROUND

[0002] For example, a technique of calculating a physical quantity at the time of repeatedly performing simulation of etching processing while changing a parameter has been known (for example, refer to Patent Literature 1).

[0003] PRIOR ART DOCUMENTS

[0004] PATENT LITERATURE

[0005] Patent Literature 1: International Publication No. 2016 / 132759 SUMMARY

[0006] PROBLEMS TO BE SOLVED BY THE INVENTION

[0007] The present application provides a technique of searching for a process condition capable of achieving a target process result using a machine learning model of a semiconductor manufacturing apparatus.

[0008] TECHNICAL SOLUTION FOR SOLVING THE PROBLEMS

[0009] One embodiment of the present application is an information processing apparatus that generates a machine learning model of a semiconductor manufacturing apparatus that performs processing in accordance with a process condition, searches for the process condition capable of achieving a target process result using the machine learning model, and includes: a machine learning model selection module that selects, from among a plurality of machine learning models generated using a plurality of regression methods, a machine learning model suitable for a data set used in learning of the machine learning model; a calculation module that performs optimization calculation using the selected machine learning model, calculates a plurality of process conditions that achieve the target process result, a predicted value of the process result corresponding to the process condition, and a reliability of the predicted value; a process condition selection module that selects one or more process conditions from among the plurality of process conditions that achieve the target process result, based on the predicted value of the process result and the reliability of the predicted value; and a display control module that displays the selected process condition, the predicted value of the process result corresponding to the selected process condition, and the reliability of the predicted value.

[0010] EFFECTS OF THE INVENTION

[0011] According to the present application, it is possible to search for a process condition capable of achieving a target process result using a machine learning model of a semiconductor manufacturing apparatus. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1This is a structural diagram of an example of the information processing system involved in this embodiment.

[0013] Figure 2 This is a hardware structure diagram of an example computer.

[0014] Figure 3 This is a functional block diagram of an example of a parsing server involved in this embodiment.

[0015] Figure 4 This is a flowchart illustrating an example of the processing of the information processing system involved in this embodiment.

[0016] Figure 5 This is an image of an example of an input screen.

[0017] Figure 6 This is an example image of the output screen.

[0018] Figure 7 This is an example representation of the result of a multi-objective optimization calculation.

[0019] Figure 8 This is an example of how reliable a prediction can be.

[0020] Explanation of reference numerals in the attached figures

[0021] 1. Information Processing System

[0022] 10 Semiconductor manufacturing equipment

[0023] 20 Device Controller

[0024] 22 Main Computers

[0025] 24 External measuring instruments

[0026] 26. Resolution Server

[0027] 40 Network

[0028] 50 Machine Learning Model Generation Department

[0029] 52 Machine Learning Model Selection Department

[0030] 54 Computing Department

[0031] 56. Process Condition Selection Department

[0032] 58 Display Control Department

[0033] 60 Process Results Receiving Department

[0034] 62 Judgment Department

[0035] 64 Feedback Department

[0036] 66. Dataset Storage Department. Detailed Implementation

[0037] Hereinafter, with reference to the accompanying drawings, the methods for carrying out the present invention will be described.

[0038] <System Architecture>

[0039] Figure 1 This is a structural diagram of an example of the information processing system involved in this embodiment. Figure 1 The information processing system 1 shown includes a semiconductor manufacturing apparatus 10, an apparatus controller 20, a host computer 22, an external measuring instrument 24, and a parsing server 26.

[0040] The semiconductor manufacturing apparatus 10, apparatus controller 20, host computer 22, external measuring instrument 24, and parsing server 26 are communicatively connected via a network 40 such as a LAN (Local Area Network).

[0041] Semiconductor manufacturing apparatus 10, for example, is a thermal processing film deposition apparatus. It executes processes corresponding to each step of the semiconductor manufacturing process (e.g., film deposition, etching, ashing, cleaning, etc.) according to control commands (process conditions) output from apparatus controller 20. Process conditions are the conditions of the semiconductor manufacturing process, which are combinations of parameters used to control (adjust) the control components (control knobs) of the semiconductor manufacturing apparatus 10. The combinations of parameters used to adjust the control knobs are diverse.

[0042] The device controller 20 is a controller having a computer structure for controlling the semiconductor manufacturing apparatus 10. The device controller 20 outputs the process conditions, calculated and selected as described later, as parameters for adjusting the control knob of the semiconductor manufacturing apparatus 10 to the semiconductor manufacturing apparatus 10.

[0043] The host computer 22 is an example of a human-machine interface (MMI) that receives instructions for the semiconductor manufacturing apparatus 10 from the operator and provides information related to the semiconductor manufacturing apparatus 10 to the operator through a display or the like.

[0044] External measuring instrument 24 is a film thickness measuring instrument, thin film resistance measuring instrument, particle measuring instrument, etc., used to measure the result (process result) after processing according to process conditions. External measuring instrument 24 measures the measured values ​​of indicators used to judge the quality (pass or fail, right or wrong) of the process result. There are various indicators for judging the quality of the process result. For example, external measuring instrument 24 measures the adhesion of the film on a wafer, such as a monitoring wafer (film formation result), as an example of a measured value of the process result.

[0045] As described below, the parsing server 26 generates multiple machine learning models using various regression methods, and uses these generated machine learning models to calculate the process conditions that can achieve the target process result. Furthermore, as described below, the parsing server 26 selects the process conditions to be proposed to operators, etc., from the multiple process conditions that can achieve the target process result, based on the predicted value of the process result corresponding to each process condition and the reliability of that predicted value.

[0046] The operator causes the semiconductor manufacturing apparatus 10 to perform processing according to the process conditions selected by the analysis server 26. After the processing is performed according to the process conditions, the operator uses an external measuring instrument 24 to measure the measured values ​​of the process results. The process conditions and the process results (the process results corresponding to the process conditions) when the semiconductor manufacturing apparatus 10 is processed according to the process conditions are fed back to the analysis server 26, thereby improving the accuracy of the process conditions calculated in subsequent calculations.

[0047] When the measured value of the process result reaches the target value of the process result, the operator ends the search for process conditions that can achieve the target process result. When the measured value of the process result does not reach the target value of the process result, the operator continues to search for process conditions that can achieve the target process result.

[0048] in addition, Figure 1 The information processing system 1 is just one example; of course, various system structures exist depending on their uses and purposes. For example... Figure 1 The distinction between devices such as semiconductor manufacturing apparatus 10, device controller 20, host computer 22, external measuring device 24, and parsing server 26 is just one example.

[0049] For example, the information processing system 1 can be a structure that integrates two or more of the following components: semiconductor manufacturing apparatus 10, apparatus controller 20, main computer 22, external measuring device 24, and parsing server 26, or a structure that is further divided into various other structures.

[0050] <Hardware Structure>

[0051] The device controller 20, main computer 22, and parsing server 26 of the information processing system 1 are, for example, composed of... Figure 2 The computer implementation of the hardware structure. Figure 2 This is a hardware structure diagram of an example computer.

[0052] Figure 2The computer 500 includes an input device 501, an output device 502, an external I / F (interface) 503, RAM (Random Access Memory) 504, ROM (Read Only Memory) 505, a CPU (Central Processing Unit) 506, a communication I / F 507, and an HDD (Hard Disk Drive) 508, which are interconnected via bus B. Additionally, the input device 501 and the output device 502 can be connected and used when necessary.

[0053] Input device 501 includes a keyboard, mouse, panel, etc., for operators to input various operation signals. Output device 502 includes a monitor, etc., for displaying the processing results of computer 500. Communication I / O 507 is an interface for connecting computer 500 to network 40. HDD 508 is an example of a non-volatile storage device for storing programs and data.

[0054] External I / F 503 is an interface to external devices. Computer 500 can read and / or write to storage media 503a, such as SD (Secure Digital) memory cards, via external I / F 503. ROM 505 is an example of a non-volatile semiconductor memory (storage device) storing programs and data. RAM 504 is an example of a volatile semiconductor memory (storage device) temporarily storing programs and data.

[0055] CPU506 is a computing device that reads programs and data from storage devices such as ROM505 and HDD508 into RAM504 and executes them, thereby realizing the overall control and functions of computer 500.

[0056] Figure 1 The device controller 20, host computer 22, and parsing server 26 shown can, for example, be connected via a... Figure 2 The computer 500, with its hardware structure, executes programs to implement the various functions described later.

[0057] <Functional Structure>

[0058] The parsing server 26 of the information processing system 1 involved in this embodiment, for example, is through... Figure 3 The functional modules shown are implemented. Figure 3 This is a functional block diagram of an example of a parsing server involved in this embodiment. Additionally, Figure 3 The functional block diagram is omitted, and the structures not required in the description of this embodiment are not shown.

[0059] The parsing server 26 executes the program used by the parsing server 26 to realize the machine learning model generation unit 50, machine learning model selection unit 52, calculation unit 54, process condition selection unit 56, display control unit 58, process result receiving unit 60, judgment unit 62, feedback unit 64 and dataset storage unit 66.

[0060] The dataset storage unit 66 stores the datasets used by the machine learning model generation unit 50 and the machine learning model selection unit 52. The datasets stored in the dataset storage unit 66 associate the parameter combinations (process conditions) of the adjustment control knob with the measured values ​​of the process results corresponding to those process conditions. For example, the datasets stored in the dataset storage unit 66 are prepared by collecting data on the measured values ​​of the process results when the semiconductor manufacturing apparatus 10 performs processing under multiple process conditions.

[0061] The dataset stored in the dataset storage unit 66 includes learning data used by the machine learning model generation unit 50 and evaluation data used by the machine learning model selection unit 52. For example, in a statistical method called cross-validation for evaluating generalization performance, the data is split into k parts, with one part used as evaluation data and the remaining k-1 parts used as learning data.

[0062] The machine learning model generation unit 50 uses the learning data of the dataset stored in the dataset storage unit 66 to generate machine learning models using multiple regression methods such as linear regression and nonlinear regression.

[0063] The machine learning model selection unit 52 selects one or more machine learning models suitable for the prepared dataset by using the evaluation data of the dataset stored in the dataset storage unit 66 from among the multiple machine learning models generated by the machine learning model generation unit 50.

[0064] The calculation unit 54 can, for example, acquire the process result set as a target by the operator. The calculation unit 54 can acquire the process result as a target stored in the dataset storage unit 66, etc., or the process result as a target input by the operator to the input screen, etc.

[0065] The calculation unit 54 uses one or more machine learning models selected by the machine learning model selection unit 52 to perform multi-objective optimization calculations, calculating multiple process conditions that achieve the target process result. Since there are multiple indicators of the target process result, the calculation unit 54 performs multi-objective optimization to calculate multiple process conditions that achieve all the indicators of the target process result, or process conditions that are close to the target process result.

[0066] Furthermore, there are machine learning models capable of calculating the reliability of predicted values. Therefore, the calculation unit 54 calculates the calculated process conditions, the predicted value of the process result corresponding to those process conditions, and the reliability of the predicted value. The reliability of the predicted value from a machine learning model that cannot calculate the reliability of the predicted value is calculated using another machine learning model capable of calculating the reliability of the predicted value. The reliability of the predicted value becomes one of the factors used to determine which process conditions the semiconductor manufacturing apparatus should follow, in other words, for program selection of process conditions.

[0067] The process condition selection unit 56 selects one or more process conditions from a plurality of process conditions calculated by the calculation unit 54. For example, the process condition selection unit 56 selects a process condition with good predicted process performance and high reliability based on the predicted values ​​of the process results for each of the plurality of process conditions calculated by the calculation unit 54 and the reliability of those predicted values. The predicted process performance can be determined based on the degree to which the predicted value achieves the target of the process result. The reliability of the process condition can be determined based on the reliability of the predicted value of the process result.

[0068] The display control unit 58 displays, on the output screen or the like, one or more process conditions selected by the process condition selection unit 56, the achievement rate of the predicted value of the process result corresponding to the process condition, and the reliability of the predicted value. Therefore, the operator can confirm that the machine learning model can achieve one or more process conditions predicted as the target process result. Furthermore, the display control unit 58 displays, on the input screen or the like, the input received from the operator as the target process result, and the results obtained from searching for the aforementioned process conditions to determine whether to continue or end.

[0069] Subsequently, the operator performs processing on the semiconductor manufacturing apparatus 10 according to one or more process conditions selected by the process condition selection unit 56. Furthermore, the operator uses an external measuring instrument 24 to measure the actual values ​​of the process results after processing according to the process conditions. The operator provides the measured values ​​of the process results obtained using the external measuring instrument 24 to the analysis server 26. This process of providing the measured values ​​of the process results to the analysis server 26 can be performed either from the external measuring instrument 24 via a network 40 or via a storage medium such as a USB (Universal Serial Bus) memory.

[0070] The process result receiving unit 60 receives measured values ​​of process results corresponding to one or more process conditions. The judgment unit 62 compares the target values ​​of the process results for each of the one or more process conditions selected by the process condition selection unit 56 with the measured values ​​of the process results received by the process result receiving unit 60, and determines whether to continue or end the search for process conditions based on the degree to which the measured values ​​achieve the target values ​​of the process results. For example, the judgment unit 62 causes the display control unit 58 to display the judgment result indicating whether to continue or end the search for process conditions.

[0071] The feedback unit 64 adds the process conditions and the measured values ​​of the process results corresponding to those process conditions to the dataset stored in the dataset storage unit 66, and then feeds back the process results. Thus, the parsing server 26 in this embodiment, by feeding back the measured values ​​of the process results corresponding to the process conditions, activates its learning function, improving the accuracy of subsequent calculations.

[0072] <Processing>

[0073] The following is an example of how operators use machine learning models to search for process conditions that can achieve the desired film formation result. The film formation result is an example of a process outcome.

[0074] Figure 4 This is a flowchart illustrating an example of the processing of the information processing system according to this embodiment. The information processing system 1 according to this embodiment needs to prepare a dataset for machine learning. In step S10, for example, an operator performs a film deposition process on the semiconductor manufacturing apparatus 10 while varying the process conditions, and collects measured values ​​of the film deposition results when performing the film deposition process according to each process condition, thereby preparing a dataset. Furthermore, the preparation of the dataset does not necessarily have to be done manually by an operator. For example, the dataset can also be prepared by automatically executing an experimental planning method. The prepared dataset is stored in the dataset storage unit 66.

[0075] The operator begins the search for process conditions that can achieve the target film formation result, for example, by setting... Figure 5 The required information is displayed on the input screen 1000 as shown. Press the calculation execution button 1008 to start the search for process conditions.

[0076] Figure 5 This is an image of an example of an input screen. Figure 5 The input screen 1000 includes a target value input field 1002, a control knob range limit input field 1004, a display field 1006 showing a list of available models, a calculation execution button 1008, and a film formation result input field 1010.

[0077] The target value input field 1002 is used to set the indicators (target values) for judging the quality of the film deposition result. Target values ​​for the film deposition result may include multiple indicators such as film thickness, in-plane uniformity, refractive index, and wet etching rate (WER). Furthermore, the priority of each target value can be set in the target value input field 1002.

[0078] The control knob variation range limit input field 1004 is used to set the upper and lower limits of the variation range of the parameter adjusted by the control knob (temperature, pressure, gas, etc.) for each control knob. Furthermore, the control knob variation range limit input field 1004 can also be configured with a usage priority for each control knob. For example, when using N2 gas, the upper and lower limits of the N2 gas flow rate can be set.

[0079] Display column 1006, which represents a list of available models, displays one or more machine learning models suitable for the prepared dataset as a list of available machine learning models. Display column 1006 can also receive changes to the list of available machine learning models from the operator.

[0080] The calculation execution button 1008 is an example of a start instruction receiving unit that receives a start instruction for searching process conditions from the operator. The parsing server 26, upon receiving the press operation of the calculation execution button 1008 from the operator, starts, for example, a multi-objective optimization calculation using a machine learning model, which is represented as a usable model in the display bar 1006.

[0081] The film deposition result input field 1010 is a field where the measured value of the film deposition result measured by the external measuring instrument 24 is entered after the semiconductor manufacturing apparatus 10 performs film deposition processing according to one or more process conditions selected by the process condition selection unit 56.

[0082] return Figure 4 In step S12, the machine learning model generation unit 50 of server 26 parses the learning data of the dataset stored in dataset storage unit 66 to generate machine learning models using multiple regression methods such as linear regression and nonlinear regression. Furthermore, the machine learning model selection unit 52 selects one or more machine learning models from the multiple machine learning models generated by the machine learning model generation unit 50 that are suitable for the prepared dataset by using evaluation data of the dataset stored in dataset storage unit 66.

[0083] In step S14, the calculation unit 54, for example, obtains the target value of the film formation result set in the target value input field 1002. Furthermore, the calculation unit 54 performs multi-objective optimization calculations using one or more machine learning models selected by the machine learning model selection unit 52, calculating for each process condition the process conditions for achieving the target value of the film formation result, the predicted value of the film formation result corresponding to that process condition, and the reliability of that predicted value.

[0084] In step S16, the process condition selection unit 56 selects one or more process conditions from the plurality of process conditions calculated by the calculation unit 54. The process condition selection unit 56 selects the process condition with good predicted process performance and high reliability based on the predicted film formation results of each of the plurality of process conditions calculated by the calculation unit 54 and the reliability of the predicted values. The process condition selection unit 56 may also receive the process condition selection operation from the operator.

[0085] In step S18, the control unit 58 indicates the degree of achievement of the predicted value of the film formation result corresponding to the process condition selected by the process condition selection unit 56, and the reliability of the predicted value, for example, in the case of... Figure 6 The output screen shown is 1100, etc.

[0086] Figure 6 This is an example image of the output screen. Figure 6 The output screen 1100 includes a display panel 1102 showing the results of multi-objective optimization, a display panel 1104 showing the reliability of the prediction, a display panel 1106 showing the predicted film formation results, and a display panel 1108 showing a list of proposed process conditions.

[0087] The result display column 1102 for multi-objective optimization shows the results of multi-objective optimization calculations performed by the computation unit 54 using one or more machine learning models. For example, in the result display column 1102 for multi-objective optimization, the results are represented by... Figure 7 The chart shown represents the results of multi-objective optimization calculations.

[0088] Figure 7 This is an example representation of the result of a multi-objective optimization calculation. Figure 7 The results of the multi-objective optimization, column 1102, is a graph showing the degree of achievement of the predicted value relative to the first objective value of the film formation result on the horizontal axis and the degree of achievement of the predicted value relative to the second objective value of the film formation result on the vertical axis. It is an example of process conditions representing the results of the multi-objective optimization calculation. Figure 7 In the chart, each scatter plot represents the result of multi-objective optimization calculation, i.e., multiple process conditions.

[0089] The column 1104 representing the reliability of the prediction indicates the reliability of the prediction obtained from the results of multi-objective optimization calculations using more than one machine learning model. For example, in the column 1104 representing the reliability of the prediction, the reliability is expressed as follows: Figure 8 The chart shown indicates the reliability of the prediction.

[0090] Figure 8 This is an example of how reliable a prediction can be. Figure 8 The column 1104, which represents the reliability of the prediction, is a graph with the degree of achievement of the predicted value relative to the first target value of the film formation result on the horizontal axis and the reliability of the predicted value on the vertical axis. It is an example of how the predicted value achieves the target value relative to each process condition and how reliable the predicted value is. Figure 8 In the chart, each scatter point represents the result of multi-objective optimization calculation, i.e., multiple process conditions.

[0091] The prediction film deposition result display column 1108 represents a list of process conditions selected from multiple process conditions proposed for performing film deposition in the semiconductor manufacturing apparatus 10, based on the results of multi-objective optimization calculations. Furthermore, the prediction film deposition result display column 1108 can also receive changes to the list of process conditions for performing film deposition in the semiconductor manufacturing apparatus 10 from operators. The prediction film deposition result display column 1106 represents the predicted value of the film deposition result for each process condition proposed in the prediction film deposition result display column 1108.

[0092] In addition, Figure 6 The first and second objective values ​​shown in the output screen 1100, specifically in column 1102 representing the results of multi-objective optimization and column 1104 representing the reliability of the prediction, are merely examples, representing the preferred methods that operators can select. Operators should refer to... Figure 6 The output screen 1100 determines one or more process conditions for the film formation process performed in the semiconductor manufacturing apparatus 10.

[0093] The operator performs a film deposition process on the semiconductor manufacturing apparatus 10 under one or more predetermined process conditions. When the film deposition process is complete, the operator uses an external measuring instrument 24 to measure the actual values ​​of the film deposition results under each process condition. The operator inputs the measured values ​​of the film deposition results obtained using the external measuring instrument 24 into, for example... Figure 5 The film formation result input field 1010 is provided to the parsing server 26. The process result receiving unit 60 receives the measured values ​​of the film formation results corresponding to one or more process conditions.

[0094] When the process result receiving unit 60 receives the measured values ​​of film formation results corresponding to one or more process conditions, the judgment unit 62 of the parsing server 26 proceeds from the processing in step S20 to the processing in step S22.

[0095] In step S22, the determination unit 62 compares the target value of the film deposition result of each process condition in the semiconductor manufacturing apparatus 10 that performs the film deposition process with the measured value of the film deposition result, and determines whether to continue or end the search for process conditions based on the degree to which the measured value is achieved relative to the target value of the film deposition result.

[0096] When it is determined that the search for process conditions should continue, the process proceeds from step S24 to step S26. The feedback unit 64 adds the process conditions for the film deposition process performed in the semiconductor manufacturing apparatus 10 and the measured values ​​of the corresponding process results to the dataset stored in the dataset storage unit 66, and feeds back the process results. The parsing server 26 returns from step S26 to step S12. When it is determined that the search for process conditions should end, the parsing server 26 terminates the process. Figure 4 The processing of flowcharts.

[0097] Furthermore, the above description illustrates an example of selecting process conditions in step S16 based on the predicted values ​​of film formation results for each of the multiple process conditions calculated by the calculation unit 54 and the reliability of those predicted values; however, this is not limited to this example. Alternatively, in step S16, process conditions may be selected based on the priority of the target value of the film formation result, the usage priority of each control knob, the similarity of the process conditions calculated by each machine learning model, etc.

[0098] In step S16, the process condition selection unit 56 selects one or more process conditions from the multiple process conditions calculated by the calculation unit 54 through multi-objective optimization calculation, for example, as follows.

[0099] For example, in the processing of step S16, the process condition that maximizes the sum Ji of the two objective functions fj(xi) is selected. The process condition xi is calculated by performing multi-objective optimization. The objective function is fj(xi). In the example below, J1 > J2 > J3, when there is only one proposed process condition, the process condition selection unit 56 proposes process condition x1 to the operator.

[0100] J1 = f1(x1) + f2(x1)

[0101] J2 = f1(x2) + f2(x2)

[0102] J3 = f1(x3) + f2(x3)

[0103] Furthermore, as mentioned above, when a priority is set for the target value of the film formation result and when a priority is set for the use of the control knob, it is possible to address this by assigning weights to the above formulas corresponding to the priorities.

[0104] Alternatively, multiple regression methods can be used differently depending on the purpose in step S12. For example, multiple regression methods can enable the differentiated use of machine learning models corresponding to the linear / nonlinear characteristics of each control knob. For the process conditions of the control knob that are fully represented by a linear model, a linear regression model is used. Furthermore, for the process conditions of the control knob with strong nonlinearity, a nonlinear model is used. The selection of the machine learning model is performed by the machine learning model selection unit 52, but it can also be performed, for example, under the operation of the operator.

[0105] Alternatively, in step S12, a switch can be made from a nonlinear model to a local linear regression model. In regions with a sufficiently large amount of data, switching from a nonlinear model to a local linear regression model allows for optimization using a more accurate machine learning model.

[0106] Furthermore, this embodiment describes an example where the film formation result includes multiple target values ​​and the process conditions include multiple parameters that are adjusted by multiple control knobs. However, it can also be applied to cases where the film formation result includes a single target value and the process conditions include a parameter that is adjusted by a single control knob.

[0107] According to this embodiment, the target value for judging the quality of the process result is diverse, and the target value of the process result is different for each customer of the semiconductor manufacturing apparatus 10. Therefore, the problem of spending time searching for process conditions to achieve the target value of the customer's process result can be solved.

[0108] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments, and various modifications and substitutions can be made to the above embodiments without departing from the scope of the present invention. For example, in this embodiment, a heat treatment film formation apparatus has been described as an example, but it can also be applied to batch film formation apparatuses of CVD (chemical vapor deposition), thermal oxidation, ALD (atomic layer deposition), etc.

Claims

1. An information processing device, characterized in that: The information processing device generates a machine learning model of a semiconductor manufacturing apparatus that performs processing according to process conditions, uses the machine learning model to search for process conditions that can achieve the target process result, and includes: A machine learning model selection module selects a machine learning model from a plurality of machine learning models generated using multiple regression methods that is suitable for use on the dataset in learning the machine learning model. The calculation module uses the selected machine learning model to perform optimization calculations, calculating multiple process conditions for achieving the target process result, predicted values ​​of the process results corresponding to the process conditions, and the reliability of the predicted values; A process condition selection module, based on the predicted value of the process result and the reliability of the predicted value, selects one or more process conditions from a plurality of process conditions that achieve the target process result; and The display control module displays the selected process conditions, the predicted value of the process result corresponding to the selected process conditions, and the reliability of the predicted value. The process condition selection module further selects one or more process conditions from the plurality of process conditions that achieve the target process result based on the priority assigned to the plurality of target values ​​included in the target process result, the priority of the plurality of adjustment objects included in the process conditions that are adjusted in order to achieve the target process result, and the degree of achievement of the target value relative to the predicted value of the process result.

2. The information processing apparatus as described in claim 1, characterized in that: Based on the comparison between the measured value of the process result obtained by the semiconductor manufacturing apparatus performing the process according to the selected process conditions and the target value of the process result, it is determined whether to continue or end the search for the process conditions.

3. The information processing apparatus as described in claim 1 or 2, characterized in that: It also includes a feedback unit, which feeds back the selected process conditions and the measured values ​​of the process results obtained by the semiconductor manufacturing apparatus in accordance with the process conditions to the machine learning model selection module.

4. The information processing apparatus as described in claim 1, characterized in that: The display control module displays an input field for assigning priorities to the multiple target values, an input field for assigning priorities to the multiple adjustment objects, and the machine learning model available for use, and displays a screen for receiving changes from the machine learning model available for use.

5. A computer program product, characterized in that: The computer program product includes programs. When the program is executed by the processor, the information processing device performs multiple steps. The information processing device generates a machine learning model of a semiconductor manufacturing apparatus that performs processing according to process conditions, and uses the machine learning model to search for process conditions that can achieve the target process result. The multiple steps include: The machine learning model selection step selects the machine learning model from among a plurality of machine learning models generated using multiple regression methods that is suitable for the dataset used in learning the machine learning model; The calculation step uses the selected machine learning model to perform optimization calculations, calculating multiple process conditions that achieve the target process result, predicted values ​​of the process results corresponding to the process conditions, and the reliability of the predicted values; The process condition selection step involves selecting one or more process conditions from a plurality of process conditions that achieve the target process result, based on the predicted value of the process result and the reliability of the predicted value. The display control steps show the selected process conditions, the predicted value of the process result corresponding to the selected process conditions, and the reliability of the predicted value. The process condition selection step further selects one or more process conditions from the plurality of process conditions that achieve the target process result based on the priority assigned to the plurality of target values ​​included in the target process result, the priority of the plurality of adjustment objects included in the process conditions that are adjusted in order to achieve the target process result, and the degree of achievement of the target value relative to the predicted value of the process result.

6. A method for searching process conditions, characterized in that: The process condition search method described above is a process condition search method for information processing devices. The information processing device generates a machine learning model of a semiconductor manufacturing apparatus that performs processing according to process conditions, and uses the machine learning model to search for process conditions that can achieve the target process result. The process condition search method includes: The machine learning model selection step involves selecting, from among a plurality of machine learning models generated using multiple regression methods, the machine learning model suitable for the dataset used in learning the machine learning model: The calculation step involves using the selected machine learning model to perform optimization calculations, calculating multiple process conditions for achieving the target process result, predicted values ​​of the process results corresponding to the process conditions, and the reliability of the predicted values: The process condition selection step, based on the predicted value of the process result and the reliability of the predicted value, selects one or more process conditions from a plurality of process conditions that achieve the target process result: and The display control steps show the selected process conditions, the predicted value of the process result corresponding to the selected process conditions, and the reliability of the predicted value. The process condition selection step further selects one or more process conditions from the plurality of process conditions that achieve the target process result based on the priority assigned to the plurality of target values ​​included in the target process result, the priority of the plurality of adjustment objects included in the process conditions that are adjusted in order to achieve the target process result, and the degree of achievement of the target value relative to the predicted value of the process result.

Images