Predictive maintenance method and predictive maintenance device

Abstract

An example of a predictive maintenance method includes determining whether simulation data measured in a substrate processing that has used a recipe exceeds an allowable threshold value corresponding to the recipe and that has been predetermined, and in a case where it is determined that the simulation data exceeds the allowable threshold value in the determination, notifying a user that a related module that has been predetermined in advance in relation to the simulation data has deteriorated.

Description

Technical Field

[0001] Examples related to predictive maintenance methods and predictive maintenance equipment are described. Background Technology

[0002] For example, in semiconductor manufacturing facilities, various modules are used to control or monitor processes. Examples of such modules include mass flow controllers (MFCs), automatic pressure controllers (APCs), RF generators, photodetectors, and temperature measuring devices. RF generators can produce high-frequency power with predetermined waveforms and can also measure plasma emission intensity or plasma emission time.

[0003] Due to prolonged use, some modules will degrade over time. If a module fails, production will cease, making it necessary to detect module degradation before failure occurs. However, simply repairing or replacing modules to prevent degradation will result in the same degradation process being repeated multiple times. Summary of the Invention

[0004] Some examples described in this article can address the problems mentioned above. These examples can provide predictive maintenance methods and equipment that can improve systems and formulations.

[0005] In some examples, a predictive maintenance method includes: determining whether simulated data measured in a substrate process using a given formulation exceeds an allowable threshold corresponding to the formulation and predetermined; and, if the simulation data exceeds the allowable threshold, notifying the user that the relevant module associated with the simulation data has degraded. Attached Figure Description

[0006] Figure 1 A view showing an example configuration of the system is displayed;

[0007] Figure 2 This is a flowchart of the learning phase;

[0008] Figure 3 This is a flowchart of the monitoring phase;

[0009] Figure 4 This is a view that shows an example of hardware configuration;

[0010] Figure 5 This is a view showing another configuration example of the hardware; and

[0011] Figure 6 This is a view that shows an example of data stored in a storage medium. Detailed Implementation

[0012] Figure 1This is a view illustrating a configuration example of a system including a predictive maintenance device. The system includes a unique platform controller (UPC) 19, a process module controller (PMC) 20, and a storage medium 21. According to one example, UPC 19, PMC 20, and storage medium 21 are used as predictive maintenance devices.

[0013] The system includes a chamber 10; and a platform 12 and spray heads 14 disposed within the chamber 10. The platform 12 and spray heads 14 provide a parallel plate structure. Gas, with a flow rate controlled by a mass flow controller 50, is supplied from a gas source 52 through slits in the spray heads 14 to the space between the platform 12 and the spray heads 14. Gas, with a flow rate controlled by a mass flow controller 54, is supplied from a gas source 56 through slits in the spray heads 14 to the space between the platform 12 and the spray heads 14. According to one example, these gases are used to treat a substrate disposed on the platform 12. According to another example, a different gas may be used.

[0014] Based on commands from PMC 20, RF generator 60 applies high-frequency power to spray head 14 via, for example, an RF sensor and a matching box. Photodetector 30 converts the plasma light generated in the space between platform 12 and spray head 14 into a voltage and outputs that voltage. Automatic pressure controller (APC) 34 vents gases already used for substrate processing or chamber cleaning to the outside of chamber 10. Substrate processing using this system includes, for example, plasma-based film formation, plasma-based etching, or plasma-based film modification. According to one example, the system is configured as a PEALD device or a pulsed CVD device.

[0015] According to one example, the module for processing the substrate is controlled by a PMC20. According to another example, multiple recipes are stored in the PMC20, and the PMC20 controls the module for substrate processing according to the recipes. Figure 1 In the example, MFC50, MFC54, APC34, RF generator 60, photodetector 30, and temperature measuring device 32 are modules controlled by PMC20. According to another example, another module can be controlled.

[0016] For example, PMC20 is a microcomputer. According to one example, PMC20 is used as an anomaly detection controller. PMC20 may include a computing unit, a storage unit, an alarm determination unit, and a sensor monitoring unit. According to one example, UPC19 receives alarm signals from PMC20 and displays or records such alarms.

[0017] Storage medium 21 is connected to PMC 20 and UPC 19. Storage medium 21 is the portion of which necessary data for the operation of the substrate processing apparatus is stored, for example, on a hard disk.

[0018] In this way, UPC19, PMC20, and storage medium 21 serve as controllers for substrate processing and also as predictive maintenance devices. The operation of the predictive maintenance devices is broadly divided into a learning phase as a preparation phase and a monitoring phase for performing predictive maintenance operations.

[0019] Figure 2 This is a flowchart illustrating an example of the learning phase. In the learning phase, the controller determines the number of times to learn in the first step S1. For example, it should be determined here that learning is performed five times. Next, in step S2, the controller processes the virtual wafer according to a specific formula and determines whether the simulated data obtained through processing triggers a limit alarm. The limit alarm is set to cause the controller to detect simulated data that deviates from the expected range due to anomalies, etc.

[0020] When a limit alarm is issued in step S2, the simulated data is an outlier; therefore, the controller excludes the simulated value from the learning object in step S5 and proceeds to step S6.

[0021] The simulation data obtained in step S2 without triggering a limit alarm is stored as data in step S3. Next, the controller increments the count by 1 in step S4, and in step S6, determines whether the process has completed the predetermined number of learning iterations. Currently, one learning iteration has been completed, so the process returns to step S2 for a second learning iteration. In this example, the controller processes multiple virtual wafers according to the same formula until five simulation data from five learning iterations are accumulated through the process in step S3.

[0022] The controller performs multiple learning processes in this manner. In step S3, by processing multiple virtual wafers according to the same specific recipe, the controller can obtain the average value of multiple measured analog data. For example, based on the average value, the controller determines an allowable threshold. If the analog data does not exceed the allowable threshold, the module will not fail, but if the analog data exceeds the allowable threshold, there is concern that the module may fail. According to one example, the allowable threshold is a more stringent standard than the aforementioned limit alarm. In other words, in some cases, the analog data output from a module that has no problems under normal use but has deteriorated to some extent due to aging does not meet the allowable threshold. Therefore, when the analog data has exceeded the allowable threshold, it does not necessarily mean that the module has failed, but it can be said that the module is in a state where failure may occur sooner or later.

[0023] The controller then performs this learning process on other recipes and determines an allowable threshold for each recipe. The controller then stores multiple recipes and corresponding allowable thresholds in a storage medium. According to one example, the controller acquires one set of simulation data from the processing based on a recipe and determines an allowable threshold for the simulation data. According to another example, it is acceptable for the controller to acquire multiple sets of simulation data from different modules in the processing based on a recipe and determine an allowable threshold for each set of simulation data.

[0024] Furthermore, separate from the learning phase, the controller correlates the analog data with relevant modules, which are modules associated with the analog data, and stores the results in a storage medium as needed. Degradation of a relevant module affects the analog data associated with that module. In other words, the controller correlates a module with analog data affected by module degradation. This determines the correspondence between the analog data and the relevant modules. According to one example, the analog data is the flow output signal of MFC50 and MFC54, and the relevant modules are MFC50 and MFC54. According to another example, the analog data is the pressure signal of APC34, and the relevant module is APC34. According to yet another example, the analog data is a signal measured by RF generator 60, a signal detected by photodetector 30, and a signal measured by temperature measuring device 32, and the corresponding relevant modules are RF generator 60, photodetector 30, and temperature measuring device 32.

[0025] According to another example, it is acceptable for the controller to omit such association operations and identify the module providing the simulation data as the associated module. According to yet another example, it is acceptable to associate one set of simulation data with multiple modules.

[0026] Figure 3 This is a flowchart illustrating an example of the monitoring phase. The monitoring phase is when the controller performs predictive maintenance on the module, either during or after the product substrate is processed. Due to the processing of the product substrate, the module degrades over time. According to one example, the controller specifies a formulation and then monitors this degradation over time. Figure 3 In step SA, the controller determines whether the simulated data measured during substrate processing using a specific formulation exceeds a pre-determined allowable threshold corresponding to that formulation. If the simulated data exceeds the allowable threshold, in step SC, the controller notifies the user that the "relevant module" associated with the simulated data has degraded. Upon receiving the notification, the user checks the target module in step SD and takes necessary measures.

[0027] In a specific example, if the MFC50 and MFC54 have degraded due to aging from processing on a large number of product substrates, and the simulated data measured by the MFC has exceeded an acceptable threshold, this means that signs of failure have been detected in the MFC50 and MFC54; and the controller issues an alert for predictive maintenance. According to one example, the user specifies which recipe provided the simulated data that triggered the alert, thus revealing that the processing performed with that recipe has placed a heavy load on the relevant module.

[0028] On the other hand, if step SA determines that the simulation data does not exceed the allowable threshold, this means there are no signs of failure in the module, and the controller determines the result is acceptable in step SB. In response to the acceptance determination, the controller returns the process to step SA, allowing the next substrate to be processed, and again compares the simulation data to the allowable threshold. When using the same formulation as the previous formulation, the same allowable threshold is used; when using a formulation different from the previous formulation, the allowable threshold corresponding to the changed formulation is used.

[0029] Therefore, the controller compares the simulation data to the allowable threshold for each recipe, allowing the user to discover that the load on the module is small in one recipe but large in another. In other words, the controller determines whether the simulation data has exceeded the allowable threshold for each substrate treatment by using multiple recipes with different contents, and then the user discovers that a particular recipe is particularly degrading the module. For example, the user finds that the MFC50 degrades significantly in a recipe with a specific gas flow. In this case, as part of predictive maintenance, the user has the opportunity to change the recipe to one that applies a smaller load to the module, or to adjust the recipe to reduce the load on the module.

[0030] According to one example, the controller can calculate a safety level, which indicates how much the simulated data obtained from multiple recipes deviates from multiple pre-determined allowable thresholds for the recipes. The higher the safety level, the closer the simulated data is to the average value obtained during the monitoring phase. In other words, a recipe providing high safety is considered a recipe that applies a low load to the module. Then, according to another example, the controller can designate a recipe that gives a particularly low safety level as a "high-load recipe."

[0031] According to one example, the controller automatically or manually changes the high-load recipe. According to another example, the controller automatically or manually reduces the frequency of use of the high-load recipe. According to yet another example, if a high-load recipe has been found, the controller can change a module that has been pre-associated with simulation data and provides the high-load recipe to a module resistant to the high-load recipe.

[0032] The predictive maintenance described above can be performed on a module associated with a single compartment, or it can be applied to multiple compartment modules, such as the four compartment modules of a four-compartment module (QCM). When the predictive maintenance is applied to a module associated with a QCM, the controller can eliminate or reduce the differences between the compartments of the QCM.

[0033] Next, an example will be described. Figure 4 This is a view illustrating a configuration example of a predictive maintenance device. The predictive maintenance device includes processing circuitry 70b for performing all or at least a portion of the processes described above. Processing circuitry 70b can perform at least the learning and monitoring phases described above. The processing circuitry can be dedicated hardware, or it can be a CPU (also known as a central processing unit, central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP) that executes programs stored in memory.

[0034] Figure 4 This is a block diagram of a predictive maintenance device assuming the processing circuit 70b is dedicated hardware. The predictive maintenance device includes a receiver 70a, processing circuit 70b, and output device 70c. Receiver 70a receives analog data measured through a substrate processed using the prepared formula. Processing circuit 70b corresponds to, for example, a single circuit, a composite circuit, a programming processor, a parallel programming processor, an ASIC, an FPGA, or a combination thereof. The functionality of the predictive maintenance device can be implemented by the respective processing circuitry, or the functionality can be implemented jointly by the processing circuitry. According to one example, the processing circuitry acts as a controller, determining whether the analog data exceeds a pre-determined permissible threshold for the formula. Output device 70c sends voice, image, alarm signals, etc., to another device. Output device 70c is a display that notifies the user that a relevant module associated with the analog data has deteriorated, for example, if it has been determined during the monitoring phase that the analog data exceeds a permissible threshold.

[0035] Figure 5 This is a block diagram illustrating an example configuration of a predictive maintenance device where the processing circuitry is a CPU. In this case, the aforementioned series of processes is controlled by a program. In other words, it is executed automatically. Figure 2 and 3 The process. When the processing circuit 80b is as follows: Figure 5 As shown in the CPU diagram, each function of the predictive maintenance device is implemented through software, firmware, or a combination of both. The software or firmware is described as a program and stored in a computer-readable storage medium 80c. Information regarding permissible thresholds and relevant modules is also stored in the storage medium 80c. In short, the program instructs the computer to specify a recipe, then compares simulated data with permissible thresholds and issues an alarm when the simulated data exceeds the permissible thresholds.

[0036] exist Figure 4 and5 In any configuration, the processing circuitry used as the controller can calculate a safety factor, indicating how much difference exists between multiple simulation data obtained from processing multiple recipes and multiple pre-determined allowable thresholds for the multiple recipes. It can also specify a high-load recipe, which is a recipe that gives a lower safety factor than the standard safety factor. The controller then notifies the user that the high-load recipe will be changed; notifies the user that the frequency of use of the high-load recipe will be suppressed; or notifies the user that modules for which the high-load recipe has been given and pre-associated with simulation data will be modified to resist the high-load recipe. The notification is executed by the output device.

[0037] Figure 6 This is a table showing an example. Figure 6 The diagram shows the controller monitoring four simulation data points for each of recipes R1, R2, and R3. Allowable thresholds and related modules are set for all simulation data to be monitored. Furthermore, for all simulation data, a safety level is displayed, indicating how much the most recent simulation data deviates from the allowable threshold. In this example, regarding the safety level, in the column examining the simulation data for high-load recipes, the simulation data gives a safety level lower than the predetermined standard. Figure 6 As the example shows, setting the security level in a list format improves user convenience.

[0038] The methods or apparatus exemplified so far involve monitoring simulation data related to the formulation to detect module degradation due to aging. Therefore, compared to simply maintaining or periodically inspecting the modules, these methods and apparatuses can provide a wider range of information for improving the system or refining the formulation.

Claims

1. A predictive maintenance method, comprising: A system having multiple modules is provided, the system being configured to run multiple recipes, wherein each of the multiple modules has different output simulation data, wherein each of the multiple recipes has multiple output simulation data corresponding to the different output simulation data of each of the multiple modules, wherein the system is configured to measure and record multiple output simulation data from the multiple modules of the system while running each of the multiple recipes; Determine multiple permissible thresholds for each output simulation data point of multiple output simulation data points for each of multiple formulations, wherein determining the permissible thresholds for each output simulation data point of multiple output simulation data points includes: Obtain the average value of each output analog data point for each module in processing multiple virtual wafers; and Based on this average value, an allowable threshold for each of the multiple output simulation data points is determined. After determining multiple permissible thresholds for each of the multiple output simulation data, a substrate processing is performed using a first recipe for a first substrate, and a first set of data is measured during the substrate processing, wherein the first set of data includes multiple output simulation data for each of the multiple modules measured while running the first recipe. Determine whether any data in the first set of data measured during substrate processing using the first formulation exceeds the permissible threshold for each of the multiple output simulation data for the first formulation; Calculate the safety factor for each of the multiple output simulation data for the first formulation, wherein calculating the safety factor includes calculating the margin between the measured data and the allowable threshold; If any data in the first set of data measured in the substrate processing that has used the first formulation exceeds an allowable threshold for each of the multiple output simulation data for the first formulation, the user is notified that the module associated with the data that has exceeded the allowable threshold has degraded. Specify whether the first formulation is a high-load formulation for the system, wherein the high-load formulation is a formulation that provides a lower safety level than a predetermined standard; and Change the high-load formula, suppress the frequency of high-load formula use, or limit the system's use of high-load formula.

2. The predictive maintenance method according to claim 1, wherein, The first set of data includes the flow output signal of the mass flow controller, and The multiple modules include a mass flow controller.

3. The predictive maintenance method according to claim 1, wherein, The first set of data includes the pressure signal from the automatic pressure controller, and The multiple modules include an automatic pressure controller.

4. The predictive maintenance method according to claim 1, wherein, The first set of data consists of signals measured by an RF generator, signals detected by a photodetector, or signals measured by a temperature measuring device.

5. The predictive maintenance method according to claim 1, wherein, The system includes multiple reaction chambers, and the method is performed for each of the multiple reaction chambers.

6. A predictive maintenance device, comprising: A receiver configured to receive analog data measured during substrate processing using the first formulation; A controller configured to determine whether output simulation data exceeds an allowable threshold, wherein the allowable threshold has been pre-calculated for a first formulation by processing multiple substrates using a first formulation, wherein the controller calculates the allowable threshold for each of the multiple output simulation data for the first formulation based on the average value across the multiple substrates for each of the multiple output simulation data for the first formulation, and wherein the controller is configured to calculate each safety factor, wherein calculating the safety factor includes calculating a margin between the simulation data and the pre-calculated threshold; and The display is configured to notify the user that, if it is determined in the process of determining whether the output analog data exceeds an allowable threshold, the relevant module previously associated with the analog data has degraded. The controller is further configured to specify a high-load recipe, wherein the high-load recipe is a recipe that provides a safety level below a predetermined threshold, and to suppress the frequency of use of the high-load recipe, limit the use of the high-load recipe, or change the high-load recipe.

7. The predictive maintenance device according to claim 6, wherein, The controller includes a storage medium configured to store multiple recipes and multiple permissible thresholds corresponding to the multiple recipes.

8. The predictive maintenance device according to claim 6, wherein, The received analog data includes the flow output signal from the mass flow controller.

9. The predictive maintenance device according to claim 8, wherein, The controller is configured to notify the user that the high-load recipe will be changed.

10. The predictive maintenance device according to claim 8, wherein, The controller is configured to notify the user that the frequency of use of the high-load recipe will be suppressed.

11. The predictive maintenance device according to claim 8, wherein, If the high-load recipe has been found, the controller is configured to notify the user that the module that has been given the high-load recipe and has been pre-associated with the simulation data will be changed to a module resistant to the high-load recipe.

Images