JP7679160B2 - Semiconductor manufacturing equipment management system and method thereof - Google Patents

Description

The present invention relates to a management system and method for semiconductor manufacturing equipment.

For example, semiconductor power elements are required to have high breakdown voltage, low on-resistance, and low switching loss, but silicon (Si) power elements, which are currently the mainstream, are approaching their theoretical performance limits. Silicon carbide (SiC) has a dielectric breakdown field strength that is about one order of magnitude greater than that of Si. Therefore, by making the drift layer that maintains the breakdown voltage about one-tenth thinner and increasing the impurity concentration by about 100 times, it is theoretically possible to reduce the element resistance by more than three orders of magnitude. In addition, SiC has a band gap about three times larger than that of Si, which allows it to operate at high temperatures. SiC semiconductor elements are expected to exceed the performance of Si semiconductor elements, and the development of semiconductor manufacturing equipment for SiC is underway.

One type of semiconductor manufacturing equipment for SiC is the SiC epitaxial growth equipment. Epitaxial growth of SiC is a technology for forming a SiC film on an off-cut SiC substrate. Generally, SiC substrates have a high donor concentration, so the donor concentration and film thickness must be adjusted for each application of the withstand voltage, and epitaxial growth is performed to manufacture SiC devices. There are many requirements for epitaxial growth technology, such as larger diameter epitaxial growth in line with larger diameter substrates, ensuring uniformity of donor concentration, ensuring uniformity of epitaxial film thickness, high-speed growth, and low crystal defects.

To meet all of these requirements with precision, equipment with many control parameters (input parameters) is required. Accordingly, to fully utilize the performance of semiconductor manufacturing equipment, it is necessary to determine several to even several dozens of control parameters. Therefore, as equipment performance improves, the equipment structure becomes more complex, making it increasingly difficult to identify the combination of control parameters that will produce the desired film formation results. This lengthens device development time and causes development costs to increase.

Furthermore, in SiC epitaxial growth equipment, by-products derived from the material gases adhere firmly to components other than the SiC substrate (inner walls and susceptor) during SiC epitaxial growth. These by-products are exposed to high temperatures during epitaxial growth and evaporate, causing changes in the epitaxial results over time. In other CVD equipment, by-products can be easily removed by gas cleaning, but no effective gas cleaning method has yet been established for SiC. For this reason, frequent maintenance such as opening the chamber to remove the by-products is required, which increases development costs.

Therefore, in order to reduce development costs, there is a demand for a function or device that can semi-automatically search for optimal control parameters taking into account changes over time, easily bring out the best performance from the device, and even notify the user when maintenance is required. Methods for modifying process recipes taking into account changes over time are described in Patent Documents 1 and 2.

For example, Patent Document 1 is an example of a document disclosing a means for correcting deviations in film formation results due to changes over time. Patent Document 1 discloses a method for correcting film thickness deviations. Specifically, it discloses the following: "The control device is a control device that controls the operation of a substrate processing apparatus that forms a film on a substrate by atomic layer deposition, and has a recipe storage unit that stores film formation conditions according to the type of film, a model storage unit that stores a process model that represents the effect of the film formation conditions on the characteristics of the film, a log storage unit that stores actual measured values of the film formation conditions during film formation, and a control unit that calculates film formation conditions that satisfy the target characteristics of the film based on the measurement results of the characteristics of the film formed under the film formation conditions stored in the recipe storage unit, the process model stored in the model storage unit, and the actual measured values of the film formation conditions stored in the log storage unit" (Summary).

For example, Patent Document 2 is an example of a document that discloses a means for correcting differences due to assembly and dimensional variations between devices and chambers, i.e., machine differences and changes over time. Patent Document 2 discloses the following: "In a semiconductor manufacturing device that is equipped with multiple semiconductor manufacturing devices and a control device that controls each of these semiconductor manufacturing devices and controls the multiple semiconductor manufacturing devices according to a single supplied recipe to manufacture a common semiconductor device, a recipe correction amount is calculated by referring to recipe correction data stored in advance based on difference data between the performance of the device for which the recipe is used when the recipe is used and the device performance obtained by using a semiconductor manufacturing device to be used in the future out of the multiple semiconductor manufacturing devices, and the supplied recipe is corrected based on the calculated recipe correction amount and supplied to the semiconductor manufacturing device to be used in the future." (Summary).

JP 2017-174983 A JP 2013-135044 A

However, in the method of Patent Document 1, the trigger for correcting the recipe is deposition that exceeds the tolerance, and depending on the amount of deviation, it is not possible to proceed to the next process. In other words, this involves a risk of failure, which can lead to increased development costs. In addition, when increasing the number of items to be corrected, a new correlation model corresponding to the number of input parameters for the device must be created, and creating the correlation model requires a lot of man-hours.

In the method of Patent Document 2, correlation data of output results for each input parameter must also be prepared in advance as a means of correcting for machine differences and changes over time. When increasing the number of items to be corrected, a new correlation model corresponding to the number of input parameters for the device must be created, and creating the correlation model requires a lot of man-hours. In addition, in order to make the corrections, processing must be performed in advance after maintenance using a recipe for constructing performance history data, which can lead to increased development costs.

Among the inventions disclosed in this application, a brief summary of representative inventions is as follows.
One aspect of the present invention is a management system that determines a recipe for a semiconductor manufacturing equipment, the management system including one or more storage devices and one or more processors, wherein the one or more storage devices store history information of past processing of the semiconductor manufacturing equipment, the one or more processors acquire a target value of a specific object in a next processing by the semiconductor manufacturing equipment, and determine a recipe for the next processing of the semiconductor manufacturing equipment using one or more functions including an estimation model based on the history information and the target value, the input of the estimation model includes recipe candidates for the next processing of the semiconductor manufacturing equipment, and the output of the estimation model includes an estimated value of the specific object.

According to one aspect of the present invention, a recipe can be determined according to changes over time in semiconductor manufacturing equipment.

Problems, configurations and advantages other than those mentioned above will become clear from the explanation of the following examples.

In the first embodiment, an estimation of the film thickness of the Nth film forming process according to a comparative example is shown. In the first embodiment, the film thickness of the Nth film forming process is estimated. 13 shows an example of training data used to construct an estimation model after predicting an input that results in an output of X for the Nth time in the first embodiment. 1 shows the relationship between the processing order and the output (film thickness) in the training data in the first embodiment. 1 shows the relationship between the processing order in the training data and one of the input parameters (gas flow rate) in the first embodiment. In the first embodiment, a method for estimating a suitable input (recipe) using an estimation model is illustrated. In the first embodiment, it will be shown that the combination of input values and output values of the estimation model can change for each process. In the first embodiment, a configuration example of a semiconductor manufacturing management system is illustrated. 1 illustrates a schematic diagram of an example of a logical configuration of a semiconductor manufacturing management system in a first embodiment. 4A and 4B are schematic diagrams illustrating another example of a logical configuration of the semiconductor manufacturing control system according to the first embodiment. 1 shows an example of training data for updating an estimation model, which is stored in a training data database in the first embodiment. In the first embodiment, an example of a target value for the next film forming process will be shown. 4 shows an example of optimal input values (recipe) determined by a recipe search unit in the first embodiment. FIG. 2 is a logical configuration diagram for explaining the processing of the semiconductor manufacturing management system in the first embodiment. 4 shows a flowchart of a process executed by a semiconductor manufacturing management system in the first embodiment. In the first embodiment, an example of the configuration of a cluster device is shown. In the first embodiment, a configuration example of a SiC epi-cluster apparatus is shown. 13 is a graph showing an example of a change over time in output by an estimation model in the second embodiment. 13 is a graph showing an example of a change over time in output by an estimation model in the second embodiment. 13 is a graph showing an example of a change over time in output by an estimation model in the second embodiment. FIG. 11 is a conceptual diagram of training data in the second embodiment. In the second embodiment, a more specific example of training data will be given. In the second embodiment, an example of a target value for the next film forming process will be shown. 11 shows an example of optimal input values (recipe) determined by a recipe search unit in the second embodiment. FIG. 11 is a logical configuration diagram for explaining the processing of the semiconductor manufacturing management system according to the second embodiment. 11 shows a flowchart of a process executed by a semiconductor manufacturing control system in the second embodiment. 13 shows a flowchart of a process executed by a semiconductor manufacturing management system in the third embodiment. 13 shows an example of a recipe search function setting window in the third embodiment. 13 shows an example of a target value setting input window in the third embodiment. 13 shows an example of a board information input window in the third embodiment. 13 shows an example of an evaluation result input window in the third embodiment. In the third embodiment, an example of an optimal recipe output window is shown. 13 shows an example of a maintenance notification message box in the third embodiment. 13 shows an example of a maintenance evaluation result output window in the third embodiment. 13 shows an example of a history information output window in the third embodiment. In the fourth embodiment, an example of input data for training data including factors thought to affect changes over time is shown. In the fifth embodiment, an example of input data for training data including factors thought to affect changes over time is shown. In the sixth embodiment, an example of a graph of the function g_a(t) is shown. 13 shows an example of a graph of a time-varying function of recipe a in the sixth embodiment.

The following describes an embodiment of the present invention with reference to the drawings. Note that in the following embodiments, when necessary for convenience, the description will be divided into multiple sections or embodiments. However, unless otherwise specified, they are not unrelated to each other, and one is related to the other in terms of partial or complete modifications, details, supplementary explanations, etc.

In the following embodiments, when the number of elements (including the number, numerical value, amount, range, etc.) is mentioned, it is not limited to the specific number, and may be more than or less than the specific number, unless otherwise specified or clearly limited in principle. Furthermore, in the following embodiments, the components (including element steps, etc.) are not necessarily essential, unless otherwise specified or clearly considered essential in principle.
＜Overview＞

In this embodiment, the derivation of an appropriate recipe taking into account the aging of a semiconductor manufacturing device will be described.
The semiconductor manufacturing management system autonomously estimates changes over time in the chamber from historical information on past semiconductor manufacturing processes, and derives an appropriate recipe (processing conditions for semiconductor devices) suited to the equipment status.

In this embodiment, since the change over time is estimated from the processing history information, there is no need to deposit a film to check the change over time, and the labor costs can be reduced. In addition, in this embodiment, by constructing a machine learning model based on the processing history information, it is not necessary to prepare correlation data for each parameter that constitutes the recipe. Furthermore, since this embodiment can derive an appropriate recipe that takes into account the change over time, the number of maintenance operations can be reduced.

As described above, the semiconductor manufacturing management system of this embodiment uses a machine learning model (estimation model) to determine an appropriate recipe for the next process. The input of the estimation model includes a recipe, and the output includes an estimate of the value obtained as a result of processing by the semiconductor manufacturing equipment (estimate of a specific target).

An example of a film formation process is described below. An example of a film formation process is, for example, epitaxial film formation of Si or SiC. The input in the film formation process (estimated model) includes a recipe, and the output includes an estimated value of the characteristics of the film to be generated. The film characteristics include, for example, the film thickness or the impurity concentration profile. In the example described below, the film thickness is the output. Note that the features of this embodiment can be applied to semiconductor manufacturing processes other than the film formation process.

<Embodiment 1>
For the sake of simplicity, an example will be described in which the recipe of the film formation process that is repeatedly executed is the same. FIG. 1 shows an estimation of the film thickness of the Nth film formation process by a comparative example. The estimation model of the comparative example estimates the film thickness from the recipe without referring to the processing history information. In other words, the input of the estimation model is only the recipe of each film formation process. If there is a change over time in the film formation device, the film thickness (output) will vary even if the same recipe A (input A) is used. Therefore, the estimation model of the comparative example cannot accurately estimate the film thickness due to the Nth process. In other words, the comparative example cannot estimate the input A that becomes the target A (film thickness A).

Figure 2 shows the estimation of the film thickness of the Nth film formation process according to this embodiment. First, processing history information is added to the training data, and an estimation model is constructed using machine learning suitable for time series analysis (e.g., RNN (Recurrent Neural Network) or LSTM (Long-Short Term Memory)). Furthermore, an optimal solution search method is used to predict an input (input A) that causes the estimation model to output a value close to the target A of the Nth process. Here, processing history information is, for example, the number of processed substrates or the number of times of processing.

Next, consider an example where the output (film thickness) remains constant despite changes over time. The input (recipe) required to obtain a constant output (film thickness) changes due to the effects of changes over time. If we consider the amount of output fluctuation due to changes over time to be a function of the input, then it is believed that accurate predictions cannot be made even if the number of processed wafers or number of times processed is used as processing history information.

Therefore, by using recipe history as processing history information, it becomes possible to autonomously learn the feature quantities of changes over time. Figure 3 shows an example of training data 31 used to construct an estimation model after predicting an input that will result in an output of X in the Nth iteration. Figure 4 shows the relationship between the processing order and the output (film thickness) in the training data 31, and Figure 5 shows the relationship between the processing order and one of the input parameters (gas flow rate) in the training data 31.

The training data used to construct an estimation model for predicting an input that will result in an output of X at the Nth time is the data obtained by removing the Nth record from the training data 31 shown in FIG. 3. The training data 31 indicates the semiconductor manufacturing recipe, which is the input of the estimation model, the film thickness and concentration, their respective in-plane distribution, etc., which are the outputs, and recipe record information. The recipe history information of each record indicates the time series of the recipe for all past processing (after chamber maintenance). Note that all past processing may also mean all processing since chamber maintenance. The recipe history information "-" of the first record (No. 1 in the table) indicates that there is no processing (recipe) shown as history before the processing of that record.

Figure 6 shows a schematic diagram of a method for estimating an appropriate input (recipe) using an estimation model. The estimation model 23 outputs an output value 42 for an input value 41. The input value 41 and the output value 42 are represented, for example, by a vector. This method searches for an input value 41 that causes an error between a target value (also simply called a target value) 35 of a specific target in a semiconductor manufacturing device and an output value (estimated value) 42 to be within an acceptable range.

The input values 41 include input values (recipe candidates) for the semiconductor manufacturing equipment for the next process. By varying the recipe candidates in the input values 41, a recipe (input values 41) that obtains an output value 42 within an acceptable range is searched for. The method further adds the input values 41 to the training data 31 along with recipe history information. The method further updates the estimation model 23 using the updated training data 31.

In actual semiconductor manufacturing, the same target is not necessarily set every time. Therefore, as shown in Figure 7, the combination of input and output values of the estimation model may change for each process. By applying recipe history information as processing history information, it is possible to autonomously capture features associated with changes over time and build an accurate estimation model that includes changes over time.

A more specific example of the configuration of this embodiment will be described below. FIG. 8 shows a schematic diagram of an example of the configuration of a semiconductor manufacturing management system. In the example of FIG. 8, the semiconductor manufacturing management system 100 is composed of one computer. The semiconductor manufacturing management system 100 includes a processor 110, a memory 120, an auxiliary storage device 130, a network (NW) interface 140, an I/O interface 145, an input device 151, and an output device 152. The above components are connected to each other by a bus. The memory 120, the auxiliary storage device 130, or a combination of these, is a storage device.

Memory 120 is composed of, for example, a semiconductor memory, and is mainly used to temporarily store programs and data. Programs stored in memory 120 include an operating system (not shown), as well as an integrated management program 121, an apparatus control program 122, an estimation model program 123, a recipe search program 124, and an analysis and evaluation program 125.

The integrated management program 121 manages the other programs and mediates communication between them. The equipment control program 122 controls the semiconductor manufacturing equipment (e.g., a chamber). The estimation model program 123 is a model that takes a semiconductor manufacturing recipe as input and outputs an estimated value of a target object in semiconductor manufacturing (e.g., film characteristics in film formation), and any suitable model such as a neural network, support vector machine, or kernel is used. The recipe search program 124 searches for an appropriate recipe to achieve the target value of the target object in semiconductor manufacturing using the estimation model program 123. The analysis and evaluation program 125 analyzes and evaluates the actual measured value of the target object in semiconductor manufacturing.

The processor 110 executes various processes according to the programs stored in the memory 120. The processor 110 operates according to the programs to realize various functional units. For example, the processor 110 functions as an integrated management unit, an equipment control unit, an estimation model, a recipe search unit, and an analysis and evaluation unit according to each of the above programs.

The auxiliary storage device 130 stores a training data database 131. The training data database 131 stores data for training the estimation model. The auxiliary storage device 130 is composed of a large-capacity storage device such as a hard disk drive or a solid-state drive, and is used to store programs and data for long periods of time.

For ease of explanation, programs 121-125 are stored in memory 120, and training data database 131 is stored in auxiliary storage device 130, but the storage location of data in semiconductor manufacturing management system 100 is not limited. For example, programs and data stored in auxiliary storage device 130 are loaded into memory 120 at startup or when necessary, and various processes of semiconductor manufacturing management system 100 are performed by processor 110 executing the programs. Therefore, the processes executed by the functional units below are processes by processor 110 or semiconductor manufacturing management system 100 in accordance with the programs.

The network interface 140 is an interface for connecting to a network. The semiconductor manufacturing control system 100 communicates with other devices in the system or devices related to the system via the network interface 140. The input device 151 is a hardware device that allows the user to input instructions and information, and includes, for example, a keyboard and a pointing device. The output device 152 is a hardware device that displays various images for input and output, such as a display device.

The semiconductor manufacturing control system 100 includes one or more processors and one or more computing devices. Each processor may include a single or multiple computing units or processing cores. A processor may be implemented, for example, as a central processing unit, a microprocessor, a microcomputer, a microcontroller, a digital signal processor, a state machine, a logic circuit, a graphics processing unit, a system on a chip, and/or any device that manipulates signals based on control instructions.

The functions of the semiconductor manufacturing management system 100 may be implemented by distributed processing using a computer system including multiple computers. The multiple computers communicate with each other via a network to execute processing in a cooperative manner. Figures 9 and 10 show schematic diagrams of other logical configuration examples of a semiconductor manufacturing management system.

In the example of FIG. 9, the integrated management unit 21, the device control unit 22, the estimation model 23, the recipe search unit 24, and the analysis and evaluation unit 25 communicate with each other via a network (NW) 200. For example, the integrated management unit 21, the device control unit 22, the estimation model 23, the recipe search unit 24, and the analysis and evaluation unit 25 are each realized by a program implemented in a different computer. Rapid processing is achieved by distributed processing using multiple computers.

In the example of FIG. 10, the integrated management unit 21, the device control unit 22, and the analysis and evaluation unit 25 are implemented in one computer. The estimation model 23 and the recipe search unit 24 are implemented on the cloud. As will be described later, the processing by the recipe search unit 24 places the heaviest load, so by implementing the recipe search unit 24 on the cloud, rapid processing can be achieved. In addition, the recipe search unit 24 can be shared among multiple semiconductor manufacturing management systems.

Figure 11 shows an example of training data 31 for updating the estimation model 23, stored in the training data database 131. The training data 31 includes input data 311 input to the estimation model 23, and output data 312, which is target data to be output from the estimation model 23.

In the example of FIG. 11, each record of the input data 311 indicates input values for one film formation process that are input to the estimation model 23 during training of the estimation model 23. The input values are composed of multiple variable values (variable parameter values) that constitute the film formation recipe. In the example of FIG. 11, the input values are composed of the process time, process temperature, process pressure, gas flow rates of three different types of gas, and carrier gas flow rate, and are represented as vectors. Each input value is a recipe for an actually performed film formation.

11 illustrates three different types of gases, SiH ₄ , C ₃ H ₈ , and N ₂ , but other source gases containing Si, other source gases containing C, and other gases containing dopants (impurities) can be used. Other examples of other source gases containing Si are SiH ₄ and SiClH _3. The carrier gas is, for example, H _2. In addition to these gases, the input value may include an assist gas (for example, HCl).

Each record of the output data 312 indicates a target value for the output value output from the estimation model 23 for one film formation process during training of the estimation model 23. The output value is composed of multiple variable values indicating the characteristics of the formed film, and is expressed as a vector. In the example of FIG. 11, the output value is composed of the film thickness, impurity concentration, and crystal defect density. Each output value is an actual measured value of the result of an actually performed film formation process.

Records with the same process number in the input data 311 and output data 312 are associated with each other. In the example of FIG. 11, history information is not explicitly shown. The recipe history information for each record is the recipes for past processes (records) arranged in the order of processing.

Figure 12 shows an example of a target value 35 for the next film formation process. The target value 35 indicates the target value for the next process. In this example, it indicates the target value of the film characteristics to be obtained by the next film formation process. The elements of the target value 35 are the same as the elements of the output value of the estimation model 23. The recipe search unit 24 searches for an input value that will give an output value of the estimation model 23 that is close to the target value 35.

Figure 13 shows an example of optimal input values (recipe) 36 determined by the recipe search unit 24. The recipe search unit 24 sequentially inputs input values to the estimation model 23, and searches for input values for which the error between the output value and the target value 35 is within an acceptable range, as the optimal input values (recipe) 36.

The processing of the semiconductor manufacturing control system 100 will be described with reference to FIG. 14 and FIG. 15. FIG. 14 is a logical configuration diagram for explaining the processing of the semiconductor manufacturing control system 100. The recipe search unit 24 includes a model configuration unit 241, a recipe estimation unit 242, and a convergence determination unit 243. In order to obtain the optimal recipe for the next process, the user includes a target value and an error tolerance in the input data 51 from the input device 151. When the user analyzes the processing object obtained by the actual processing, the user further inputs the analysis evaluation result as the input data 51 from the input device 151. In the example of the film formation process, the analysis evaluation result is the output data of the training data 31 described with reference to FIG. 11. The analysis evaluation result is stored in the training data database 131.

The semiconductor manufacturing control system 100 outputs the optimal recipe for the next process as output data 52 from the output device 152. Furthermore, when the semiconductor manufacturing control system 100 analyzes the processing object obtained by the actual processing, it also outputs the analysis and evaluation results as output data 52 from the output device 152.

Figure 15 shows a flowchart of the process executed by the semiconductor manufacturing management system 100. First, the integrated management unit 21 acquires the target value and the error tolerance as input data 51 from the input device 151 (S101). The integrated management unit 21 passes the target value and the error tolerance to the recipe estimation unit 242.

The recipe estimation unit 242 uses the estimation model 23 to derive a recipe that achieves a value close to the target value and estimates the processing result (S102). For example, the recipe estimation unit 242 uses a random search method or a simulated annealing method to determine an input value that includes a recipe that is closest to the target value. The output value of the estimation model 23 for the input value is the estimated processing result.

The input values include the recipe for the current process and past recipe history. The recipe estimation unit 242 searches for input values that have a common recipe history but a different recipe for the current process, and for which the output value is within the allowable range. Each parameter value of the recipe has constraints (upper and lower limits) specific to the semiconductor manufacturing equipment. Candidate recipes are selected from within the range of the constraints. For example, upper and lower limits of the processing temperature and upper and lower limits of the pressure. In addition, a condition that clearly does not allow the target film formation may be set as the lower limit. For example, the processing temperature is set to a temperature at which epitaxial growth of SiC occurs, such as 1000°C or higher.

Next, the convergence determination unit 243 obtains the estimated processing result from the recipe estimation unit 242 and determines whether the error between the estimated processing result and the target value is within the error tolerance range specified by the user (S103). If the estimated processing result is not within the error tolerance range from the target value (S103: NO), the convergence determination unit 243 prompts the user to re-input the target value and the error tolerance range on the output device 152 via the integrated management unit 21. The flow returns to step S101. Note that the user can carry out processing even if the optimal recipe does not satisfy the tolerance range. This is similar to other embodiments.

If the estimated processing result is within the error tolerance from the target value (S103: YES), the device control unit 22 causes the process execution unit 27 of the processing chamber to execute processing (film formation processing in this example) using the determined optimal recipe, and further stores the optimal recipe in the training data database 131 (S104).

For example, the device control unit 22 acquires the optimal recipe determined by the recipe estimation unit 242 via the integrated management unit 21. The device control unit 22 specifies the optimal recipe and instructs the process execution unit 27 to execute processing, and adds the optimal recipe to the input data 311 of the training data 31 via the integrated management unit 21. The process execution unit 27 executes processing according to the specified optimal recipe.

The analysis and evaluation unit 25 analyzes and evaluates the processing results by the process execution unit 27 (S105). For example, the analysis and evaluation unit 25 controls an analysis and evaluation device (analysis and evaluation chamber) that contains and analyzes the processing object (in this example, the substrate on which a film has been formed) to perform analysis and evaluation of the processing object. In this example, the analysis and evaluation unit 25 analyzes and evaluates the characteristics of the formed film. The analysis and evaluation unit 25 stores the analysis and evaluation results in the training data database 131 via the integrated management unit 21, linking them to the corresponding optimal recipe (S106).

The integrated management unit 21 may obtain the analysis and evaluation results from the user via the input device 151. The integrated management unit 21 stores the analysis and evaluation results obtained from the user in the training data database 131 in association with the corresponding optimal recipe. The model construction unit 241 updates the estimation model 23 using the updated training data database 131 (S107).

The analysis and evaluation unit 25 compares the analysis and evaluation result with the target value taking into account the allowable error (S108). If the analysis and evaluation result is within the allowable error range from the target value (S108: YES), the analysis and evaluation unit 25 presents the comparison result to the user via the integrated management unit 21 and ends this flow. If the analysis and evaluation result is not within the allowable error range from the target value (S108: NO), the analysis and evaluation unit 25 presents the comparison result to the user via the integrated management unit 21 and asks the user whether or not re-deposition is necessary. If re-deposition is necessary, the integrated management unit 21 instructs the recipe estimation unit 242 to execute step S102. If re-deposition is not necessary, this flow ends.

As described above, the semiconductor manufacturing in this embodiment uses semiconductor manufacturing equipment and analysis and evaluation equipment. These equipment may be clustered. This allows for efficient semiconductor manufacturing. FIG. 16 shows an example of the configuration of a cluster equipment 400. The cluster equipment 400 includes multiple chambers. Specifically, the cluster equipment 400 includes a substrate analysis chamber 401, a cleaning chamber 402, an analysis and evaluation chamber 403, a processing chamber 404, a regeneration chamber 405, and a load lock chamber and a transfer chamber 406.

The substrate analysis chamber 401 is a chamber for analyzing and evaluating a substrate. It can be used to obtain substrate information described in embodiment 3. The substrate analysis chamber 401 can be used to measure substrate shapes such as warpage, edge shape, and plate thickness, or to evaluate dislocations, stacking faults, surface roughness (Ra, etc.), etc., using X-rays, PL light, laser light, a microscope, etc. The analysis and evaluation chamber 403 is a chamber for analyzing and evaluating a substrate processed in the processing chamber 404. The regeneration chamber 405 is a chamber for regenerating a substrate processed in the processing chamber 404. In the process described with reference to FIGS. 14 and 15, the process execution unit 27 is, for example, the processing chamber 404. The analysis and evaluation unit 25 or a user performs analysis and evaluation of the substrate processed by the analysis and evaluation chamber 403.

Figure 17 shows an example of the configuration of a SiC epi-cluster device 450. The SiC epi-cluster device 450 includes a substrate analysis chamber 451, an epi-analysis chamber 452, a defect analysis chamber 453, an epi-growth chamber 454, a CMP (Chemical Mechanical Polishing) chamber 455, and a load lock chamber and a transfer chamber 456.

The epitaxial growth chamber 454 is an example of a processing chamber, which epitaxially grows a SiC film on a substrate. The epitaxial analysis chamber 452 is an example of an analysis and evaluation chamber, which is used to analyze and evaluate the thickness and impurity concentration (dopant concentration) of the epitaxial film (i.e., SiC film) formed in the epitaxial growth chamber 454. The defect analysis chamber 453 is an example of an analysis and evaluation chamber, which is used to analyze and evaluate defects in the epitaxial film formed in the epitaxial growth chamber 454 using X-rays, PL light, laser light, and a microscope. The CMP chamber 455 is an example of a regeneration chamber 405, which is used to reuse the substrate if the formed epitaxial film does not have the desired properties.

In the process described with reference to Figures 14 and 15, the process execution unit 27 is, for example, the epitaxial growth chamber 454. The analysis and evaluation unit 25 or the user performs analysis and evaluation of the SiC film formed by the epitaxial analysis chamber 452 and the defect analysis chamber 453 to obtain values of film thickness, impurity concentration, and crystal defect density. If the SiC film does not have the desired characteristics, for example, the device control unit 22 or the user removes the SiC film from the substrate in the CMP chamber 455 and epitaxially grows the SiC film again on the substrate.

The determination of an appropriate recipe according to this embodiment can be applied to various semiconductor processes (equipment). For example, it can be applied to lithography equipment, film formation equipment, pattern processing equipment, ion implantation equipment, cleaning equipment, etc. Lithography equipment includes, for example, exposure equipment, electron beam lithography equipment, and X-ray lithography equipment. Film formation equipment includes, for example, CVD (Chemical Vapor Deposition) equipment, PVD (Physical Vapor Deposition) equipment, deposition equipment, sputtering equipment, and thermal oxidation equipment.

Pattern processing equipment includes, for example, wet etching equipment, dry etching equipment, electron beam processing equipment, and laser processing equipment. Ion implantation equipment includes, for example, plasma doping equipment and ion beam doping equipment. Cleaning equipment includes, for example, liquid cleaning equipment and ultrasonic cleaning equipment.

In optimizing the opening dimensions in an exposure tool, examples of input elements of the estimation model are the exposure dose, resist thickness, resist type, etc., and examples of output elements are the opening dimensions, deviation from the design dimensions, resist collapse, etc. In optimizing the film formation process of a CVD tool, examples of input elements are the gas flow rate, process temperature, applied bias, pressure, etc., and examples of output elements are film stress, density, particles inside the furnace, etc.

In optimizing the gate insulating film process for MOS and IGBT in CVD and thermal oxidation equipment, examples of input elements are gas flow rate, process temperature, furnace pressure, etc., and examples of output elements are interface state density, dielectric breakdown characteristics, channel mobility, PBTI & NBTI characteristics, etc. In optimizing the concentration profile in an ion implantation equipment, examples of input elements are implantation energy, dose amount, etc., and examples of output elements are peak concentration position, tail shape in the depth direction, etc.

In optimizing the etching shape in a pattern processing device, examples of input elements are gas flow rate, applied bias, etc., and examples of output elements are trench shape (taper angle, sub-trench, roughness of the trench bottom), etc. In optimizing the cleaning process in a cleaning device, examples of input elements are chemical type, chemical concentration, processing temperature, etc., and examples of output elements are particles, metal contamination, etching rate, actual concentration, etc. In optimizing the contact resistance in a laser annealing device, examples of input elements are wavelength, laser intensity, step, etc., and an example of an output element is resistance value.

As described above, in this embodiment, a recipe for the next process of the semiconductor manufacturing equipment is determined using one or more functions including the estimation model 23 based on historical information of past processes of the semiconductor manufacturing equipment and a specified target value. The one or more functions in this embodiment are configured by the estimation model 23. The input of the estimation model 23 further includes the recipe history of the semiconductor manufacturing equipment.

As described above, this embodiment uses a machine learning technique that uses past history information from semiconductor manufacturing equipment to autonomously capture feature quantities associated with changes over time, making it possible to build an accurate estimation model that also includes changes over time. In addition, because an optimal recipe that takes changes over time into account can be derived, the number of maintenance operations can be reduced. In addition, because a machine learning model is formed based on history information, there is no need to prepare correlation data.

<Embodiment 2>
In the second embodiment, the semiconductor manufacturing control system 100 determines whether or not maintenance of the semiconductor manufacturing equipment is necessary. This reduces the number of maintenance operations. For example, if the estimated result of the optimal recipe found by the search is not within an allowable range from the target value, the semiconductor manufacturing control system 100 determines that maintenance is necessary.

Figures 18A, 18B, and 18C are graphs showing an example of the change over time in output from an estimation model. In each graph, the horizontal axis indicates the gas flow rate as an input to the estimation model, and the vertical axis indicates the concentration distribution as an output of the estimation model. In each graph, a solid function curve 501 indicates the relationship between the input value and the output value of the estimation model. The semiconductor manufacturing equipment has a settable range for the input value. In addition, an allowable range is set based on a target value for the output value. In each graph, a dashed rectangle 502 indicates the area that satisfies both the settable range and the allowable range.

Figure 18A shows a graph in a state where the change over time is small. The appropriate gas flow rate that achieves an output within an acceptable range exists within the settable range. In this way, when the effect of change over time is small, an optimal input value that satisfies the target value exists, and the optimal recipe can be searched for. Figures 18B and 18C show graphs of an estimated model that is affected by change over time. The function curve 501 of the estimated model is expected to shift in the up, down, left, and right directions.

As shown in FIG. 18B, if function curve 501 deviates significantly upward, even the extreme value falls outside the tolerance range. As shown in FIG. 18C, if function curve 501 deviates significantly to the right, the input value corresponding to an output value within the tolerance range falls outside the settable range. In other words, if a recipe in which the output value falls within the tolerance range based on the target value cannot be found even after searching for the optimal recipe, it should be determined that maintenance of the semiconductor manufacturing equipment is necessary.

In this embodiment, an estimation model is constructed by including maintenance information in the training data. The estimation model makes it possible to determine whether maintenance is required and derive an optimal recipe. FIG. 19 is a conceptual diagram of training data 61. The difference from the training data in embodiment 1 is that the input value includes information on the maintenance performed, and further, the history information includes maintenance history in addition to recipe history.

For example, suppose there are three types of maintenance with different contents and effects (maintenance 1, maintenance 2, maintenance 3). The input value of the training data 61 includes information on the maintenance performed before processing. If maintenance was not performed, the value indicating maintenance is 0. If maintenance was performed, a numerical value corresponding to the type of maintenance is included in the input value. For example, the value indicating maintenance is 1 for maintenance 1, 2 for maintenance 2, and 3 for maintenance 3. Note that the maintenance identifier is arbitrary. By adding the contents of the maintenance performed to the input and history, the semiconductor manufacturing control system 100 can autonomously determine the necessary maintenance.

A specific example of this embodiment for determining whether maintenance is required will be described below. Differences from embodiment 1 will be mainly described. Figure 20 shows a more specific example of training data 61. In addition to the training data 31 in embodiment 1, the input data 611 has a pre-deposition maintenance field. The pre-deposition maintenance field indicates an identifier of maintenance performed before the deposition process, and "0" indicates that no maintenance was performed.

Figure 21 shows an example of a target value 65 for the next film formation process. Figure 22 shows an example of an optimal input value (recipe) 66 determined by the recipe search unit 24. The optimal input value (recipe) 66 has a cell for pre-film formation maintenance corresponding to the input data 611. The example in Figure 22 shows that maintenance 2 should be performed before performing the film formation process indicated by the optimal input value (recipe) 66.

The processing of the semiconductor manufacturing management system 100 will be described with reference to Figures 23 and 24. Figure 23 is a logical configuration diagram for explaining the processing of the semiconductor manufacturing management system 100 of this embodiment. The recipe search unit 24 includes a maintenance effect evaluation unit 245 in addition to the components of embodiment 1.

The user includes an identifier of the actual maintenance performed in addition to the information of embodiment 1 in the input data 51 from the input device 151. The maintenance identifier is stored in the training data database 131. The semiconductor manufacturing control system 100 outputs the necessary maintenance notification and the maintenance evaluation result in addition to the information of embodiment 1 as output data 52 from the output device 152.

Figure 24 shows a flowchart of the process executed by the semiconductor manufacturing management system 100 of this embodiment. Steps S121 to S128 correspond to steps S101 to S108 in the flowchart of Figure 15 of embodiment 1. The flowchart of Figure 24 differs from embodiment 1 in the step that is executed when the determination result in step S123 is NO.

In step S123, if the estimated processing result is not within the error tolerance from the target value (S123: NO), the recipe estimation unit 242 uses the estimation model 23 to derive a maintenance and recipe that achieves a value close to the target value, and estimates the processing result (S129). For example, the recipe estimation unit 242 uses a random search method or a simulated annealing method to determine an input value that is closest to the target value. The input value indicates any maintenance to be performed, and the value of the element corresponding to the maintenance is a value other than 0. The output value of the estimation model 23 for that input value is the estimated processing result.

The recipe estimation unit 242 notifies the user of the required maintenance on the output device 152 via the integrated management unit 21. For example, the recipe estimation unit 242 holds a message list associated with a maintenance identifier, and presents a message corresponding to the determined maintenance identifier on the output device 152. The user performs maintenance on the semiconductor manufacturing equipment in accordance with the maintenance notification, and inputs information about the performed maintenance from the input device 151. Note that the semiconductor manufacturing management system 100 may automatically perform the maintenance.

When the maintenance effect evaluation unit 245 receives maintenance information via the integrated management unit 21, it evaluates the effect of maintenance on the semiconductor manufacturing equipment. For example, the maintenance effect evaluation unit 245 uses a sensor implemented in the chamber to evaluate the thickness of deposits on the sidewall or the amount of reduction. The maintenance effect evaluation unit 245 presents the evaluation results on the output device 152 via the integrated management unit 21. By referring to the evaluation results, the user can check whether appropriate maintenance has been performed. Note that the user may perform the maintenance evaluation and omit the maintenance effect evaluation unit 245.

The device control unit 22 acquires the optimal recipe after maintenance determined by the recipe estimation unit 242 via the integrated management unit 21. The device control unit 22 specifies the optimal recipe and instructs the process execution unit 27 to execute processing, and adds information about the maintenance and optimal recipe to the input data 611 of the training data 61 via the integrated management unit 21. The process execution unit 27 executes processing according to the specified optimal recipe (S130).

The analysis and evaluation unit 25 analyzes and evaluates the processing results by the process execution unit 27 (S131). The analysis and evaluation unit 25 stores the analysis and evaluation results in the training data database 131 via the integrated management unit 21 in association with the corresponding optimal recipe (S132). The analysis and evaluation results may be input by the user. The analysis and evaluation unit 25 further presents the analysis and evaluation to the user on the output device 152 via the integrated management unit 21. The model construction unit 241 updates the estimation model 23 using the updated training data database 131 (S133). The flow then proceeds to step S128.

As described above, in this embodiment, the past processing history information of the semiconductor manufacturing equipment includes information about maintenance of the semiconductor manufacturing equipment. Furthermore, the input to the estimation model 23 includes maintenance that should be performed before the next processing. By adding the details of the maintenance to the processing history information, it is possible to autonomously estimate the changes over time of the semiconductor manufacturing equipment from the processing history information and determine whether maintenance is required.

Specifically, this embodiment can derive an optimal recipe tailored to the equipment status when maintenance is not required, and can present a maintenance method when maintenance is required. By presenting the required maintenance, the user can perform appropriate maintenance at the appropriate time. According to this embodiment, film deposition to check for changes over time in the semiconductor manufacturing equipment is no longer necessary, reducing the number of steps.

<Embodiment 3>
In the third embodiment, it is determined whether or not the substrate to be processed needs to be changed. As a result, a substrate that does not satisfy the required characteristics can be appropriately changed to a new substrate, and the number of unnecessary processes can be reduced. For example, when the estimated result of the optimal recipe found by the search is not within an allowable range from the target value, the semiconductor manufacturing control system 100 determines that the substrate needs to be changed.

For example, SiC substrates have more defects such as dislocations than Si substrates, and the dislocation density affects the defect density after epitaxial growth. For this reason, when the output of the estimation model 23 includes defect density, it is necessary to take into account defect information on the substrate. Furthermore, warping of the substrate also affects the epitaxial growth results. For this reason, by taking into account substrate information, it is possible to derive an optimal recipe with high accuracy.

The input of the estimation model 23 in this embodiment includes substrate information in addition to the input of embodiment 2. Therefore, the input data of the training data includes substrate information. When deriving an optimal recipe (input values that give an output value within an error tolerance range from a target value), the input substrate information needs to satisfy pre-specified substrate specifications. If the substrate has many defects, there is a possibility that the optimal recipe will not fall within the specified values (predetermined values).

For this reason, this embodiment uses the estimation model 23 to calculate how many defects are necessary to derive an optimal recipe. For example, one method is to gradually reduce the number of substrate defects in the substrate information input to the estimation model 23, calculate whether an optimal recipe can be derived under each defect number condition (whether it falls within an acceptable range), and notify the defect density at which the optimal recipe can be derived. The user is then asked to decide whether to change the substrate, and if the substrate is not to be changed, a decision is made as to whether maintenance is necessary.

The processing of the semiconductor manufacturing control system 100 will be described with reference to FIG. 25. In the following, differences from the second embodiment will be mainly described. The logical configuration diagram for explaining the processing of the semiconductor manufacturing control system 100 of this embodiment is substantially similar to FIG. 23 in the second embodiment. One difference from the second embodiment is that the input data 51 includes information on the substrate to be processed. As described above, the input data of the estimation model 23 and the training data includes substrate information indicating the substrate specifications.

In the flowchart of FIG. 25, steps S141 and S143 to S149 correspond to steps S121 to S128 in the flowchart of FIG. 24. In step S142, the integrated management unit 21 acquires the board information input by the user from the input device 151 and passes it to the recipe search unit 24.

In step S144, if the estimation result by the estimation model 23 is outside the allowable error range from the target value (S144: NO), the recipe estimation unit 242 uses the estimation model 23 to derive board specifications necessary to derive a recipe that achieves a value close to the target value (S150). Specifically, the recipe estimation unit 242 maintains the recipe and maintenance information in the input values to the estimation model 23, and changes only the board information. As described in the first embodiment, the recipe estimation unit 242 searches for board information in which the error between the output value of the estimation model 23 and the target value is within the allowable error range. The board specifications to be searched for are selected from a predetermined range.

If board information that can obtain an output value within the allowable range is found, the recipe estimation unit 242 determines to change the board (S151: YES). The recipe estimation unit 242 notifies the user of the board change and the specifications required for the new board on the output device 152 via the integrated management unit 21. The flow returns to step S142.

If no board information is found that can obtain an output value within the allowable range, the recipe estimation unit 242 determines not to change the board (S151: NO). The recipe estimation unit 242 then determines whether maintenance is required. Steps S152 to S156 correspond to steps S129 to S133 in the flowchart of FIG. 24.

In the above example, the substrate information is acquired from the user via the input device 151. In another example, the semiconductor manufacturing control system 100 may acquire the substrate information using a substrate evaluation device. Before performing the processing of the flowchart in FIG. 25, the semiconductor manufacturing control system 100 performs an evaluation of the substrate placed in the chamber. The maintenance processing of this embodiment may be omitted.

Below, we will explain examples of GUI images displayed on the output device 152 that can be used in some embodiments. The images explained below are merely examples, and any GUI image may be used as long as it is capable of inputting and outputting the necessary information. First, we will explain an example of a GUI image that allows the user to input data. Information entered in the input window is stored in the memory 120 or the auxiliary storage device 130 by the integrated management unit 21.

Figure 26 shows an example of a recipe exploration function setting window 531. In the recipe exploration function setting window 531, the user can enable or disable the recipe exploration function. When the recipe exploration function is enabled, each of the processes of the multiple embodiments in this specification is executed.

Figure 27 shows an example of a target value setting input window 532. The target value setting input window 532 accepts input of the target value and tolerance range of the estimation model 23 from the user. In the example of Figure 27, the target value and tolerance range are set for each parameter (element of the output value). Furthermore, the upper and lower limits of the tolerance range are set individually.

Figure 28 shows an example of a board information input window 533. The board information input window 533 accepts input of information about a board from the user. In the example of Figure 28, the board information input window 533 allows input of information about multiple boards, and the number of the slot into which the board is inserted, the board ID, and values of multiple attributes of the board are input. The input values to the estimation model 23 include attribute values of the board information.

Figure 29 shows an example of the evaluation result input window 534. The evaluation result input window 534 is a window in which the user inputs the evaluation results of an object processed by semiconductor manufacturing equipment. The evaluation result input window 534 has cells for inputting a process number for identifying the process and an evaluation value of the process result. In the above example of film formation, at least the values of film thickness, impurity concentration, and crystal defect density are input. This window is not used when the semiconductor manufacturing control system 100 automatically evaluates the processed object.

Next, an example of a GUI image for presenting information to the user will be described. FIG. 30 shows an example of an optimal recipe output window 535. The optimal recipe output window 535 is displayed on the output device 152 via the integrated management unit 21. It includes the recipe contents and the estimated results obtained from the recipe. For example, the recipe estimation unit 242 presents to the user, via the optimal recipe output window 535, a recipe that is estimated to produce an estimated result within the allowable range specified by the user, and the estimated result.

Figure 31 shows an example of a maintenance notification message box 536. The maintenance notification message box 536 presents the user with information about the maintenance required to obtain the desired processing result. As described above, information indicating the relationship between the maintenance and the message is set in advance, and the recipe estimation unit 242 refers to the information to obtain a message corresponding to the maintenance determined using the estimation model 23, and displays it in the maintenance notification message box 536.

Figure 32 shows an example of the maintenance evaluation result output window 537. The maintenance evaluation result output window 537 shows the maintenance content (identifier) and the evaluation result of that maintenance. This window 537 is used when the maintenance effect evaluation unit 245 evaluates the maintenance results. This window 537 is not used when the user evaluates the maintenance.

Figure 33 shows an example of the history information output window 538. The history information output window 538 shows basic information about the process, the process recipe, and the evaluation results of the processed object. The basic information about the process includes, for example, the process date and time and the identifier of the substrate to be processed. For example, the integrated management unit 21 displays the history information output window 538 in response to a request from a user. The basic information about the process is stored, for example, in the memory 120 or the auxiliary storage device 130, and is updated by the integrated management unit 21. The process recipe and the evaluation results of the processed object correspond to the input data and output data of the training data database 131.

<Embodiment 4>
The methods of the first, second and third embodiments use recipe history information, which requires a huge amount of calculation, and may result in a long calculation time and insufficient computer processing power. One possible solution to this problem is to reduce the number of variables (recipe elements), but this narrows the search range and may affect the derivation of an optimal solution. Therefore, the fourth, fifth and sixth embodiments explain methods for reducing the amount of calculation.

To obtain an optimal recipe (a recipe within an acceptable range) by calculation, it is preferable to autonomously acquire the feature quantities of the change over time from the history information, as described in the first, second, and third embodiments. However, as described above, the problem is that the amount of calculation increases. Therefore, in this embodiment, in order to reduce the amount of calculation, factors that are thought to affect the change over time are added to the input values of the estimation model.

34 shows an example of input data 711 of training data including factors thought to affect the change over time. In addition to the input data 311 of the training data 31 of the first embodiment, the input data 711 has columns for an integrated processing time, an integrated film thickness, and an integrated flow rate (SiH ₄ ). These indicate integrated values (cumulative values) from the last maintenance to the immediately preceding processing. The integrated values are an example of history information of past processing, similar to the recipe history. The output data of the training data of this embodiment is similar to the output data 312 of the first embodiment. Note that the integrated values input to the estimation model 23 simultaneously with the recipe may include target values of the recipe and processing.

FIG. 34 shows an example of a film formation process by epitaxial growth of SiC, where the integrated processing time, integrated film thickness, and integrated flow rate (SiH ₄ ) respectively indicate the integrated time of epitaxial growth time, the integrated value of epitaxial growth film thickness, and the integrated value of the flow rate of gas (SiH ₄ ) since the last maintenance. The integrated processing time and the integrated flow rate are the integrated values (cumulative values) of recipe elements of past processes by the semiconductor manufacturing equipment, and the integrated value of growth film thickness is the integrated value (cumulative value) of the process product. The process product is the product targeted by the film formation process, and does not include by-products. In this embodiment, the input value of the estimation model 23 includes the integrated value as history information instead of the recipe history described in the first, second, and third embodiments. This makes it possible to greatly reduce the variables of the input value compared to the input value including the recipe history.

The recipe estimation unit 242 uses the estimation model 23 to search for a recipe (optimum recipe) that gives an output value within an allowable range from the target value, as described in the other embodiments above. There are constraints of the semiconductor manufacturing equipment for each integrated value. For example, if the target film thickness is 10 μm, the integrated film thickness is the current value + 10 μm. The integrated processing time is the current value + processing time. Other processing temperatures and processing pressures also have upper limits determined by the equipment structure. The optimal recipe is searched for within the range of these constraints.

<Embodiment 5>
In the fifth embodiment, in order to reduce the amount of calculation, factors that are thought to affect the change over time are added to the input values of the estimation model. Fig. 35 shows an example of input data 811 of training data including factors that are thought to affect the change over time. The input data 811 has a column for by-product film thickness in addition to the input data 311 of the training data 31 of the first embodiment.

In the input data 811, the by-product film thickness column indicates the by-product film thickness value before processing is performed using the recipe of the same record. The by-product film thickness indicates the film thickness of the by-product deposited in the processing chamber, for example, the film thickness of the by-product deposited on the inner wall of the chamber or the film thickness of the by-product on the wafer susceptor. A by-product film is a film of any material that is not originally intended to be produced by processing. The by-product film thickness can be measured by the user or by a sensor in the chamber.

In this embodiment, the input values of the estimation model 23 include, as historical information, by-product film thickness values before processing, instead of the recipe history described in embodiments 1, 2, and 3. The by-product film thickness changes depending on past processing, and therefore indicates the processing history. By using the by-product film thickness as historical information, the variables of the input values can be significantly reduced compared to input values that include recipe history. In addition, the by-product film thickness more directly indicates the change over time of the semiconductor manufacturing equipment (chamber), and therefore the optimal recipe can be more appropriately estimated.

<Embodiment 6>
In the sixth embodiment, an estimation model (function) and a time-varying function are used to determine a recipe for the next process of a semiconductor manufacturing device. In this embodiment, the actual process result (training data output) is corrected by the time-varying function. The correction by the time-varying function removes components caused by the time-varying of the semiconductor manufacturing device. In other words, the corrected process result is a process result that is assumed to have no time-varying of the semiconductor manufacturing device.

In this embodiment, the input values to the estimation model 23 include recipe candidates for the next process, and do not include history information as in the other embodiments described above. The output data of the training data is a processing result that is assumed to be unchanged over time for the semiconductor manufacturing equipment, and is a value obtained by correcting the actual processing result with a time-change function. In comparing the output value (estimated value) of the estimation model 23 with the target value, the estimated value and/or the target value are compared after being corrected with the time-change function. In this way, the estimation model 23 outputs an estimated value of the processing result for a recipe for the semiconductor manufacturing equipment that is assumed to be unchanged over time.

As described above, since the estimation model 23 does not include historical information in the input values, the processing load for searching for an optimal recipe using the estimation model 23 can be reduced. Historical information of the semiconductor manufacturing equipment is used to create a time-varying function. Below, a method for deriving the time-varying function is explained using an example. The time-varying function is derived by regression analysis using the historical information.

As an example, the defect density D, which is one of the film formation results in a certain chamber state, is described by the following function g: The variable t of the function g represents the by-product film thickness in the chamber.
D = g(t)

When recipe a is executed when the by-product film thickness in the chamber is t1 or t2, defect densities D_t1 and D_t2 are expressed as follows, respectively.
D_t1=g_a(t1)
D_t2=g_a(t2)

Here, since the function g_a(t) is unknown, regression analysis is performed to derive the function g_a(t). Figure 36 shows an example graph of the function g_a(t). The black dots in the graph are the measured values, and the solid curve is a curve fitted to the measured values, which corresponds to the function g_a(t).

Next, consider the function g_a(t). The function g_a(t) is considered to be the sum of a factor caused only by the recipe a and a factor caused by changes over time. Therefore, the function g_a(t) can be rewritten as follows:
g_a(t)=D_a+f_a(t)
The function D_a is a constant that represents the defect density caused by recipe a regardless of aging. The function f_a(t) represents the defect density caused by aging in the processing of recipe a. The function f_a(t) is a aging function.

Next, consider the relationship between f(t) when recipe a and recipe b are executed. When the recipe is changed, the effect of aging may increase or decrease, but the physical mechanism by which the aging occurs does not change, so the following relationship holds:
f_b(t)=cf_a(t)
That is, gb(t) is g_b(t)=D_b+f_b(t)=D_b+cf_a(t)
Here, D_b is a constant representing the defect density caused by recipe b.

Since there are two unknowns, D_b and c, if there are two processing results using recipe b, the time-varying function for recipe b can be derived. As described above, the model construction unit 241 derives the time-varying function for each recipe, and further constructs an estimation model 23 that is not affected by changes over time. The input of the estimation model 23 does not include processing history information.

As described above, it is possible to use the time-change function to remove the effects of time-change from the output data of the training data, making it easier to construct an estimation model 23 that is not affected by time-change. In the above explanation, defect density was given as an example of an output value, but the output value may also be, for example, the concentration, average concentration, or in-plane concentration variation of an impurity at a certain coordinate on the wafer surface, the film thickness, average film thickness, or in-plane film thickness variation at a certain coordinate on the wafer surface, downfall defects, stacking faults, basal plane dislocations, etc.

Although the by-product thickness on the chamber wall has been given as an example of an input value, the input value may also be, for example, the by-product thickness on the susceptor or the variation in inner diameter caused by clogging of the injector. These require monitoring of the by-product thickness and inner diameter, but for simplicity, instead, the input value may be, for example, the accumulated film thickness, the accumulated flow rate of the material gas, the accumulated amount of heat given to the chamber, etc.

The time-change function can be used to determine whether maintenance is required. The recipe estimation unit 242 may determine whether maintenance is required based on the results of comparing the output of the time-change function with a threshold value. Figure 37 shows an example of a graph of the time-change function of recipe a. For example, when the value of the time-change function for a recipe determined to be the optimal recipe exceeds a specified value (a value determined in advance), the recipe estimation unit 242 may determine that maintenance is required.

The present invention is not limited to the above-described embodiments, but includes various modified examples. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. It is also possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. It is also possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

Furthermore, each of the above configurations, functions, processing units, etc. may be realized in hardware, for example by designing some or all of them as an integrated circuit. Furthermore, each of the above configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the programs, tables, files, etc. that realize each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card or SD card.

In addition, the control lines and information lines shown are those considered necessary for the explanation, and not all control lines and information lines on the product are necessarily shown. In reality, it can be assumed that almost all components are interconnected.

21 Integrated management unit, 22 Equipment control unit, 23 Estimation model, 24 Recipe search unit, 25 Analysis and evaluation unit, 27 Process execution unit, 31 Training data, 35 Target value, 41 Input value, 42 Output value, 51 Input data at input device, 52 Output data at output device, 61 Training data, 65 Target value, 66 Optimum input value, 100 Semiconductor manufacturing management system, 110 Processor, 120 Memory, 121 Integrated management program, 122 Equipment control program, 123 Estimation model program, 124 Recipe search program, 125 Analysis and evaluation program, 130 Auxiliary storage device, 131 Training data database, 140 Network interface, 145 I/O interface, 151 Input device, 152 Output device, 241 Model construction unit, 242 Recipe estimation unit, 243 Convergence judgment unit, 245 Maintenance effect evaluation unit, 311, 611, 711, 811 Input data of training data, 312, 612 Output data of training data, 400 Cluster device, 401 Substrate analysis chamber, 402 Cleaning chamber, 403 Analysis and evaluation chamber, 404 Processing chamber, 405 Regeneration chamber, 406 Load lock chamber and transfer chamber, 450 SiC epi cluster device, 451 Substrate analysis chamber, 452 Epi analysis chamber, 453 Defect analysis chamber, 454 Epi growth chamber, 455 CMP chamber, 456 Load lock chamber and transfer chamber, 501 Function curve, 502 Dashed rectangle, 531 Recipe search function setting window, 532 Target value setting input window, 533 Substrate information input window, 534 Evaluation result input window, 535 Optimal recipe output window, 536 Maintenance notification message box, 537 Maintenance evaluation result output window, 538 History information output window

Claims (3)

A management system that determines a recipe for a film formation process in a semiconductor manufacturing apparatus that forms a SiC film on a SiC substrate by epitaxial growth, comprising:
one or more storage devices;
one or more processors;
The semiconductor manufacturing equipment includes a means for monitoring a thickness of by-products on a chamber wall or a susceptor ,
the one or more storage devices store history information of past film formation processes of the semiconductor manufacturing equipment;
The one or more processors acquire target values of characteristics of a film to be formed in a next film formation process by the semiconductor manufacturing equipment;
determining a recipe for the next film formation process of the semiconductor manufacturing equipment using one or more functions including an estimation model based on the history information and the target value;
an input of the estimation model includes a recipe candidate for the next film formation process of the semiconductor manufacturing equipment;
an output of the estimation model includes an estimate of a property of the resulting film;
the one or more functions include a time-varying function indicating a time-varying change of the semiconductor manufacturing equipment,
the thickness of the by-product is utilized to derive the aging function;
the time-varying function is constructed based on the historical information including a thickness of the by-product ;
The one or more processors determine whether or not maintenance of the semiconductor manufacturing equipment is required based on a comparison result between the value of the time-varying function and a specified value. The management system according to claim 1,
A management system in which the one or more processors compare the target value and the estimated value after correcting the target value and/or the estimated value using the time-varying function, and determine the recipe based on the result of the comparison. A method for determining a recipe for a film formation process in a semiconductor manufacturing apparatus for forming a SiC film on a SiC substrate by epitaxial growth, comprising:
Monitoring the thickness of the by-products on the chamber walls or the susceptor ;
The management system stores history information of past film formation processes of the semiconductor manufacturing equipment,
The method comprises:
The management system acquires target values of characteristics of a film to be formed in a next film formation process by the semiconductor manufacturing equipment,
the management system determines a recipe for the next film formation process of the semiconductor manufacturing equipment using one or more functions including an estimation model based on the history information and the target value;
an input of the estimation model includes a recipe candidate for the next film formation process of the semiconductor manufacturing equipment;
an output of the estimation model includes an estimate of a property of the resulting film;
the one or more functions include a time-varying function indicating a time-varying change of the semiconductor manufacturing equipment,
the thickness of the by-product is utilized to derive the aging function;
the time-varying function is constructed based on the historical information including a thickness of the by-product ;
The method further comprises the step of: determining whether or not maintenance of the semiconductor manufacturing equipment is required based on a result of comparing the value of the time-varying function with a specified value.

Images