JP5599882B2 - Method and apparatus for predictive preventive maintenance of processing chambers - Google Patents

Description

  Advances in plasma processing have led to tremendous development in the semiconductor industry. A plasma processing system can be composed of many components. For ease of discussion, the term “component” is used to refer to an integral part or a multi-part assembly within a plasma processing system. Thus, the component may be a single item such as an edge ring or a complex item such as an entire process module. A multi-part component (such as a process module) may be formed from other multi-part components (such as a vacuum system, gas system, power system, etc.), and these multi-part components may also be It may be formed from part components or indivisible components.

  Over time, one or more components may wear out. Those skilled in the art are aware that if worn components are not repaired / replaced, the worn components can damage the chamber and / or the substrate. One method of identifying which components are to be replaced may include using a fixed schedule for component replacement. That is, the useful life of each component may be speculatively specified. Each component may be tracked for its use and replaced when it reaches the end of its useful life (predetermined by a fixed useful life schedule).

  Unfortunately, there are limitations on how to pre-determine the useful life of a component for replacement / repair purposes. First, the useful life of a component can vary depending on the environment surrounding the component. In one example, component 1 may be used in a process chamber that undergoes a process recipe or a mixture of multiple process recipes different from component 2. Thus, component 1 may wear out faster than component 2 even if component 1 and component 2 are the same type and model.

  Thus, a method that predetermines the useful life of a component requires unnecessary costs associated with taking the processing chamber offline and replacing the component, even if the useful life of the component has not actually expired. May occur. Also, the method of pre-determining the useful life cannot take into account the possibility that the component will wear out prematurely before its pre-determined theoretical useful life. Component degradation situations can often result in damage to the substrate and / or damage to the processing chamber and other components within the chamber.

  One method for determining component wear may involve tracking the evolution of one parameter, a so-called univariate mode. In one example, the health of a component may be monitored by tracking one parameter that can be measured by some sensor. For example, the RF bias voltage may be tracked. If the RF bias voltage is above a predetermined threshold, for example, the edge ring may be considered to have reached the end of its useful life.

  Unfortunately, univariate methods still have limitations. As described above, a given component is monitored by tracking a parameter. However, the parameter may be affected other than the status of the given component. In one example, the RF bias voltage may be monitored to monitor edge ring conditions. However, the value of the RF bias voltage may be influenced not only by the edge ring situation but also by other effects. For example, the RF bias voltage can also be affected by deposition on the chamber walls. Thus, when a high RF bias voltage is identified, the high RF bias voltage value does not necessarily indicate that there may be a problem with the edge ring. Rather, the problem will exist, but further analysis will be required before the cause of the problem is identified.

  Another problem with univariate methods is that univariate methods can be “go / no-go” methods. That is, univariate methods are typically used to identify when fault conditions can exist and allow component replacement. However, the univariate method may not help predict when component replacement may be necessary (not whether it is necessary). That is, in such a scenario, the univariate method could only be used to identify the problem, at best, not to predict when the problem (eg, the end of the useful life) could occur.

  As a result, when a component such as an edge ring actually wears out, a replacement component will not be immediately available. As a result, the processing chamber will need to stay offline until a new edge ring is obtained for replacement, for example. Of course, the manufacturer may choose to always have a replacement (such as an edge ring). This method of always providing spare components is because the manufacturer must allocate resources (funds and storage space) to keep the spare components available even when the components are still in normal operating condition. It can be uneconomical.

  Another method of identifying component wear may include using a monitoring patch. A monitoring patch is an item that can be mounted on a component. The monitoring patch may be mounted near the surface of the component or embedded in the component. A component may be considered to end its useful life, for example when the monitoring patch wears out. If a monitoring patch is embedded, the component is considered to end its useful life when, for example, the monitoring patch becomes visible.

  There are some limitations to the method of monitoring patches. First, a monitoring patch is necessary for each component to be monitored. Therefore, if 100 components need to be monitored, a monitoring patch must be mounted on each component. Monitoring patch methods can be very expensive and time consuming to implement and monitor depending on the number of components that can be monitored.

  Also, the use of monitoring patches will increase the risk of contamination. A monitoring patch is a foreign object that must enter the processing chamber. As noted above, in order to prevent damage to the processing chamber and / or damage to the substrate, the conditions of the processing chamber must be tightly controlled. The processing environment will change by introducing one or more monitoring patches into the processing chamber. Also, it may be unknown or difficult to measure how much the processing environment has changed due to the presence of the monitoring patch in the processing chamber.

  Another limitation of the monitoring patch method is that mounting a monitoring patch on a component can compromise the mechanical functionality of that component. That is, the mechanical behavior of the edge ring will change with the embedding of the monitoring pad in the ring. Unfortunately, the degree to which a patch can change the functionality of a component will vary because each component and / or each patch may be unique.

  Therefore, a non-invasive method for predicting component wear is desired.

  In the accompanying drawings, the present invention is shown by way of illustration and not limitation, and like reference numerals refer to like elements in the figures.

FIG. 6 illustrates a simplified flowchart of applying one multivariate prediction model to qualify components in one embodiment of the invention.

6 is a simplified flowchart illustrating a method for applying multiple prediction models in determining component health in an embodiment of the invention.

6 is a simplified flowchart illustrating a method of applying a multivariate prediction model with non-plasma test (NPT) in an embodiment of the invention.

6 is a simplified flowchart illustrating a method for qualifying components using a multivariate predictive model and data from non-plasma tests in an embodiment of the invention.

6 is a simplified flowchart illustrating a method of building a multivariate prediction model for qualifying components in an embodiment of the invention.

  The invention will now be described in detail with reference to a few embodiments as illustrated in the accompanying drawings. In the following description, numerous details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced without some or all of these details. In other instances, well known process steps and / or structures have not been described in detail in order not to unnecessarily obscure the present invention.

  Various embodiments, including methods and techniques, are described below. It should be borne in mind that the invention may also be directed to an article of manufacture that includes a computer readable medium having stored thereon computer readable instructions for implementing an embodiment of the inventive technique. Computer-readable media includes, for example, semiconductor, magnetic, magneto-optical, optical, or other forms of computer-readable media for storing computer-readable code. Furthermore, the invention can also be directed to an apparatus for carrying out an embodiment of the invention. Such an apparatus may include dedicated and / or programmable circuitry for carrying out tasks associated with embodiments of the invention. An example of such a device is a general purpose computer and / or dedicated computing device when properly programmed, and a dedicated / adapted computer / computing device and various tasks associated with embodiments of the invention. Combination with programmable circuit.

  To facilitate an understanding of the overall technology, some details of statistical models will be described, but the present invention relates to the prediction of part wear and how such predictions can be used in preventive maintenance. The specific model that can be used to perform the prediction may depend on the chamber or part involved. However, any model (eg, statistical model, electrical model, plasma model, hybrid model, combination of these models, etc.) may be used and a specific model for a specific chamber, a specific part, and / or a specific recipe The use of is within the purview of those skilled in the art.

  In an embodiment of the invention, a method for assessing the health of a chamber is provided. Embodiments of the invention include performing a chamber health index test. The chamber health index test discussed herein refers to a test that can determine and / or predict the health of a set of components (eg, a set of consumable parts) within a processing chamber. That is, the chamber health index test may be used to determine and / or predict when a component can reach the end of its useful life.

  In one embodiment of the invention, a method for assessing the health of a chamber is provided. In one example, one or more multivariate predictive models may be constructed to perform a chamber health index test in the processing chamber. Unlike the prior art, the prediction model (one or more) may be based on multiple parameters rather than a single parameter. As a result, the chamber health index test may be performed not only to identify when a component has worn out, but also to predict the remaining useful life of the component.

  In one embodiment, the construction of a set of robust prediction models (eg, a set of multivariate prediction models) may be based on data collected at various points in the preventive maintenance cycle. The preventive maintenance cycle is also referred to herein as a wet cleaning cycle. Data may be collected at least at the beginning and end of a wet cleaning cycle to eliminate noise in the data set that may be related to chamber conditions rather than the actual components themselves. In one embodiment of the invention, the construction of a set of robust prediction models may be based on data collected across multiple chambers. Data is collected across multiple chambers to eliminate noise that may be associated with chamber conditions that are specific to a particular chamber rather than component status.

  Once the set of multivariate predictive models has been constructed, the set of multivariate predictive models may be applied within the production environment to qualify the status of the chamber and its components. Embodiments of the invention provide a method for qualifying components using a single multivariate prediction model. That is, one multivariate prediction model may be used to qualify one component. In one embodiment of the invention, a method for applying two or more multivariate prediction models to predict wear of multiple parts and in determining a chamber health index is provided.

  In one embodiment, non-plasma testing (NPT) may be used in conjunction with chamber health index testing. In one embodiment, the NPT may be used to determine when a chamber health index test will need to be performed. The non-plasma test discussed herein refers to a test that can be performed by applying low power to the processing chamber at one or more predetermined frequencies to not ignite the plasma. During this test, no plasma is formed, but sufficient data is provided to approximate the status of the processing chamber. Since NPT is a relatively quick and inexpensive test to perform, NPT may be used as a preliminary test for the chamber health index test. In one embodiment of the invention, the NPT may be used to validate a chamber health index test. In one embodiment, data from the NPT may be associated with data collected during the chamber health index test for verification purposes.

  The features and advantages of the present invention may be better understood with reference to the drawings and discussions that follow.

  As described above, the health of the chamber may be used to determine when maintenance is required. In particular, a chamber health index test may be performed to assess the health of the chamber, such as consumable parts. The chamber health index test may be performed based on a model-based approach.

  1-4 illustrate various methods of applying one or more multivariate prediction models to qualify a set of components in one embodiment of the invention.

  FIG. 1 shows a simplified flowchart of applying one multivariate prediction model to qualify a set of components in one embodiment of the invention.

  In a first step 102, the recipe is executed in the processing chamber. The recipe may be either a client-specific recipe, a WAC recipe, or a non-client-specific recipe.

  The client-specific recipes discussed herein are recipes that can be adapted for a specific manufacturer. In one example, the client-specific recipe may be a process recipe for actual production that is unique to a particular manufacturer. In one embodiment, the ability to use client-specific recipes allows users (eg, process engineers) to build multivariate models that are specific to their needs.

  Non-client specific recipes discussed herein refer to recipes that can be designed to maximize the sensitivity to wear of a particular consumable part. In one example, the non-client specific recipe may be a recipe that may be configured to analyze a particular situation of the plasma processing system. Non-client specific recipes may be used by any manufacturer that will have a processing chamber with a similar chamber setup specified thereby.

  Another recipe that can be used is a waferless autoclean recipe. The WAC recipe may be performed after the production process. WAC recipes are usually not chamber specific.

  The type of recipe that can be performed may depend on the multivariate prediction model. In one example, if a multivariate prediction model was built using a client-specific recipe, the recipe that can be used to use the multivariate model may also be a client-specific recipe.

  In the next step 104, process data is collected by a set of sensors (eg, pressure sensor, temperature sensor, VI probe, OES, Langmuir probe, etc.).

  In the next step 106, the model is applied to the processed data. That is, the system may compare the processing data with a multivariate prediction model. In one embodiment, the prediction model is a statistical model. In another embodiment, the prediction model is an electrical model. In yet another embodiment, the prediction model is a plasma model. Furthermore, step 106 is not limited to the analysis of one component, but rather different multivariate prediction models may be applied to the same data set to analyze the health of the various components.

  In one embodiment, the multivariate prediction model may pull data from a library (108). The library may contain data (eg, constants) that can be used to support the model.

  After the analysis is performed, in a next step 110, the prediction model may output component wear data for each component being analyzed. That is, a data report detailing the sound state of each component being analyzed may be created.

  In the next step 112, the system may compare the output data to a useful lifetime threshold range. The useful life threshold for each component may be based on expertise, for example. In one embodiment, the useful life threshold may be user settable. Thus, the user may change the useful life threshold to adjust tool configuration, recipe settings, and the like.

  If the output data is outside the useful life threshold, a warning / error message may be provided at the next step 114. The warning / error message may identify the parameter that caused the useful life threshold to be breached. With data from the output report, the user (eg, a process engineer) will confidently proceed to determine a series of actions that may be needed to correct the problem. In one example, the component may require repair and / or replacement, for example.

  If the output data is within the useful life threshold, each component being analyzed is considered to be in good operating condition. In the next step 116, the system may wait for the time interval until the next measurement before performing the method as shown in FIG. 1 again.

  As can be seen from FIG. 1, by applying a multivariate model, the user will be able to determine the health of one or more components. Thus, a predictive model can not only identify when a component has deteriorated, but can also predict when a component may wear to an unacceptable level. In one example, the output report will not indicate that the edge ring has worn out because the edge ring is still operating correctly. However, the output report will indicate that approximately 75 percent of the useful life of the edge ring has been consumed and that replacement will soon be required. With this knowledge, the manufacturer may plan an upcoming repair.

  FIG. 2 is a simplified flowchart illustrating a method for applying multiple prediction models in determining component health in an embodiment of the invention.

  In a first step 202, the recipe is executed in the processing chamber. Similar to FIG. 1, the recipe may be either a client-specific recipe, a WAC recipe, or a non-client-specific recipe.

  In the next step 204, process data is collected by a set of sensors (eg, pressure sensor, temperature sensor, VI probe, OES, Langmuir probe, etc.).

  In the next step 206, a set of models is applied to the processing data. That is, two or more models (208, 210, 212) may be used to analyze the processing data. In one example, two models may be applied for component qualification. For example, model 210 may be an electrical model while model 208 is a multivariate statistical model.

  Similar to FIG. 1, step 206 allows different multivariate prediction models to be applied to the same data set to analyze the health of the various components.

  In one embodiment, the set of models may pull data from the library (214). The library may contain data (eg, constants) that can be used to support the model.

  After the analysis is performed, in the next step 216, the system may check to determine if the output from the model matches. In one example, if the output from the multivariate model 208 indicates 90 percent wear on the component while the output from the multivariate model 210 indicates 75 percent wear on the same component, then from these models The output data of do not match.

  If the output data from the models do not match, the next step 218 may apply a model with less noise to the data from the more robust model. In one example, the model 208 is an electrical model and will have a very high noise level although it is a more robust model. However, model 210 is a statistical model and is not very robust, but will have a lower noise level. The robust model discussed herein refers to a model that has few outliers (statistical outliers). In one embodiment, a less robust but lower noise model (model 210) may be applied to data from a more robust model (model 208 in this example) to qualify components. This method makes it possible to reduce noise while improving accuracy.

  Also, since the model outputs do not match, the system may report the differences and mark those models for update (step 220). That is, additional data may be collected to reconstruct one or more of the models. In one example, the method described in FIG. 5 (discussed later) can be applied to the electrical model (208) and the statistical model (210) (different / additional) to adjust the noise level and better relate the models May be implemented to rebuild (by data or a different modeling approach).

  However, if the data output from the models match, then in the next step 222, component wear data for the component being analyzed may be output from those predictive models.

  In the next step 224, the system may compare the output data with a useful lifetime threshold range.

  If the useful life threshold is exceeded, a warning / error message may be provided at the next step 226.

  If the output data is outside the useful life threshold, in the next step 228, the system may wait for the time interval until the next measurement.

  As can be seen from FIG. 2, the method described in FIG. 2 is somewhat similar to the method described in FIG. 1 except that it uses multiple prediction models rather than a single multivariate prediction model. . Verification can be provided by using more than one prediction model. Also, if one of the models is not very robust, additional models may be used to supplement the less robust model.

  FIG. 3 is a simplified flowchart illustrating a method of applying a multivariate predictive model with non-plasma test (NPT) in one embodiment of the invention.

  In a first step 302, a non-plasma test is performed. Non-plasma testing as discussed herein refers to a rapid test that can be performed by supplying low power to the processing chamber at a predetermined frequency. The power may not be strong enough to generate a plasma, but may be strong enough to provide electrical data about the chamber (eg, impedance, capacitance, etc.). In one embodiment, the non-plasma test may be a pre-test of the chamber health index test. That is, a chamber health index may be performed when the NPT indicates that a potential problem may exist in one of the components. Since NPT is a quick and fairly inexpensive test compared to the chamber health index test, the use of NPT as a prior test would help reduce cost of ownership.

  If the NPT indicates that a potential problem may exist, in a next step 304, a chamber health index test may be performed. To start the chamber health index test, a recipe may first be run.

  In the next step 306, data is acquired for analysis.

  In the next step 308, one or more models may be applied to the data for analysis (see discussion of FIGS. 1 and 2).

  Data may be extracted from the library 310 to perform the analysis. A library may contain data (eg, constants) that can be used to support a model (one or more).

  Once the data analysis is complete, an output report regarding the health of each component being analyzed may be provided (step 312).

  In the next step 314, the system may compare the output data with a useful lifetime threshold range. The useful life threshold for each component may be based on expertise, for example. In one embodiment, the useful life threshold may be user settable. Thus, the user may change the useful life threshold to adjust tool configuration, recipe settings, and the like.

  If the output data is outside the useful life threshold, a warning / error message may be provided at the next step 316. The warning / error message may identify the parameter that caused the useful life threshold to be breached. With data from the output report, the user (eg, a process engineer) will confidently proceed to determine a series of actions that may be needed to correct the problem. In one example, the component may require repair and / or replacement, for example.

  If the output data is within the useful life threshold, in the next step 318, the system may wait for the time interval until the next measurement.

  As can be seen from FIG. 3, the steps for qualifying components are somewhat similar to the steps described in FIG. 1 and / or FIG. Unlike FIG. 1 and / or FIG. 2, the method described in FIG. 3 relates to a quantitative method for identifying the time interval until the next measurement. That is, the NPT is used as an indicator as to when the next chamber health indicator test will need to be performed. As can be seen from the above, the steps described in FIG. 3 may be adjusted if more than one multivariate prediction model is used in the analysis of the data.

  FIG. 4 is a simplified flowchart illustrating a method for qualifying components using a multivariate predictive model and data from non-plasma tests in one embodiment of the invention. 3 and 4 both use a non-plasma test, but the method described in FIG. 4 is that NPT is used for verification rather than as a preliminary test of the chamber health index test. This is different from the method described in FIG.

  In the first step 402, the recipe is executed.

  In the next step 404, processing data is acquired during processing.

  In a next step 406, one or more models may be provided for the data for analysis. In one embodiment, data may be retrieved from library 408.

  The results from the analysis are provided as an output report (step 410).

  In the next step 412, the system may analyze the results to determine if additional verification is required. Additional verification may be needed when the results from step 410 are uncertain or too noisy. In one embodiment, the result may be compared to a predetermined noise level threshold range. If the result is outside the predetermined noise level threshold range, verification will be required.

  If additional verification is required, in the next step 414, NPT may be performed.

  During the NPT, data is collected (step 416).

  The data from the NPT and the data from the model analysis are correlated with each other by using the NPT data to verify the data collected during the model analysis (step 418).

  The consolidated data is output as a single report detailing the health of the component (s) being analyzed (step 420).

  Of course, if no additional verification is required (step 412), part wear information (from step 410) may be output as a single report (step 420).

  In a next step 422, the system may compare the output data with a useful lifetime threshold range. The useful life threshold for each component may be based on expertise, for example. In one embodiment, the useful life threshold may be user settable. Thus, the user may change the useful life threshold to adjust tool configuration, recipe settings, and the like.

  If the output data is outside the useful life threshold, a warning / error message may be provided at the next step 424. The warning / error message may identify the parameter that caused the useful life threshold to be breached. With data from the output report, the user (eg, a process engineer) will confidently proceed to determine a series of actions that may be needed to correct the problem. In one example, the component may require repair and / or replacement, for example.

  If the output data is within the useful life threshold, in the next step 426, the system may wait for the time interval until the next measurement.

  As can be seen from FIG. 4, a robust method for qualifying components is provided. The robust method not only provides two or more multivariate models to perform component analysis, but also includes NPT to validate the results of one or more multivariate predictive models. Yes.

  As can be seen from the above, the method for assessing the health of the chamber may be a model-based approach. The model-based approach may be based on an electrical model, a plasma model, a statistical model, or a hybrid model in one embodiment. To illustrate how a model can be built, FIG. 5 is a simplified flowchart illustrating a method for building a multivariate predictive model for qualifying components in one embodiment of the invention.

  In a first step 502, component life data is provided. Component life data may include functional and / or physical measurements for the component. In one example, if the component is new, the physical measurement is usually provided by the manufacturer. However, if the component is not new, the component lifetime may be determined by actually making a measurement on the component.

  In the next step 504, the recipe is executed. The recipe may be, for example, a client-specific recipe, a non-client-specific recipe, or a WAC (waferless automatic cleaning) recipe.

  Depending on the requirements set by the user (eg, process engineer), one or more of the above recipe types may provide the data required to create a model for qualifying the components of the processing chamber. May be executed to obtain.

  In the next step 506, process data is collected during the process. Process data may be captured by a set of sensors. As can be seen from the above, the number of sensors available, the type of sensors available, and the sensor specifications will affect the granularity of the processing data collected. Non-limiting examples of sensors that can be used in the processing chamber include, for example, pressure sensors, temperature sensors, voltage-current probes (VIPs), emission spectrometers, and the like. Specifically, data from VIP and OES may involve discretizing the wideband output at specific frequencies or their harmonic frequencies. Alternatively, full-spectrum broadband comparisons from these sensors may form the basis of the analysis.

  In one embodiment, steps 504 and 506 may be performed at various time intervals during the wet cleaning cycle to take into account potential drift that may occur during the wet cleaning cycle. For example, these steps may be performed when a component such as an edge ring has just been installed. These steps may be repeated during the middle and end of the wet cleaning cycle. The amount of data available to build a multivariate model depends on the number of times steps 504 and 506 are performed. As can be seen from the above, the number of times steps 504 and 506 can be performed may depend on the benefits that the user obtains from the additional data in building the model.

  Since parameters may change after performing a maintenance event (eg, wet cleaning), it may be necessary to identify and eliminate or adjust parameters affected by wet cleaning prior to building the multivariate model. In one example, the measured intensity (by OES) at several wavelengths may change after the transparent window of the processing chamber has been cleaned. In one embodiment, steps 504 and 506 may need to be performed multiple times during the wet cleaning cycle (508) to eliminate parameters that can be affected by wet cleaning.

  Additionally or alternatively, steps 502-506 may be performed across processing chambers (510). To create a multivariate model that can be applied across processing chambers (assuming chamber conditions are the same / similar), identify and eliminate or adjust parameters that can be affected by chamber-related conditions Would need to do. For example, data collected during the execution of standard qualification recipes performed in various process chambers can be developed to develop rules for converting sensor output from one chamber to another in both chambers. Can be used to match the output.

  In one embodiment of the invention, if the sensors are matched, steps 502-506 are performed during a wet cleaning cycle and / or to eliminate parameter changes that would not be related to component wear. Or it may not need to be repeated across processing chambers. A matched sensor should return the same absolute value for some defined chamber condition. If matched sensors are used, the measured chamber impedances for the same two chambers will be the same.

  Once sufficient data has been collected, in the next step 512, a multivariate model for the component may be created. Unlike the prior art, multivariate models are based on multiple parameters rather than one parameter. Because a large amount of data and / or granular data can be collected by a sensor, in one embodiment, a high-speed computing module may be used to handle data processing and analysis. In order to increase processing time, data may be sent directly from the sensor to the high-speed processing computing module without having to first go through the manufacturing facility host controller or eventually the processing module controller. Application No. 12 / 555,674, filed September 8, 2009 by Huang et al., Describes an example of a fast processing computing module suitable for handling data.

  However, not all collected data is related to the process of qualifying components. Filtering criteria may be applied to identify only those parameters that may be related to component wear. The filtering criteria, in one embodiment, may include parameters that are not related to the wet cleaning cycle. In another embodiment, the filtering criteria may also include chamber-dependent exclusion data. As an example, when a VI probe is used during wet cleaning to monitor a specific 27 MHz plasma, it often exceeds 100 MHz, even after closely tracking the wear of parts within the wet cleaning. The harmonic component of the part shifts.

  Once relevant parameters related to component wear are identified, one or more models may be built. In one embodiment, the processing data may be used to create a statistical model, an electrical model, and / or a plasma model. As an example, the same data may be used to build a wear information model for multiple chamber parts. Linear separation of part wear can be achieved by exchanging parts or by using a complex multi-stage plasma process so that only certain parts are monitored by each plasma.

  After each model is created, the system may check to determine if the noise level in the model is acceptable (step 514). The noise may come from, for example, physical measurements and / or sensor internal noise.

  In one example, noise will be present due to variations between components. That is, component characteristics will vary due to component size, component material composition, component configuration, and the like. In one example, the edge ring in processing chamber A and the edge ring in processing chamber B will look the same. However, the size of the edge ring in processing chamber A will be slightly larger than the edge ring in processing chamber B. Since variations between components can be specified, variations between components can be accounted for in the model. In one embodiment, an acceptable threshold range is defined. If the variation between components is outside the threshold range, the system will need to acquire additional data to build the model.

  Another source of noise may be due to component mounting. In one example, the placement of the edge ring may cause noise in the model. For example, there may be a gap between the edge ring and the electrostatic chuck. However, if this gap is different between processing chambers, the electrical characteristics of the processing chambers will also be different. Hence, a threshold range may be established. If the difference is outside the threshold range, additional data may be acquired to build the model.

  In the next step 516, the construction of the multivariate model is completed.

  Steps 502-516 may be performed to create one or more multivariate prediction models for one component. Of course, the same data file may also be used to create a multivariate prediction model for other components.

  As can be seen from FIG. 5, the described method accounts for component conditions beyond various environmental conditions, such as within one wet cleaning cycle, across multiple wet cleaning cycles, and within various processing chambers. Provide the creation of a multivariate prediction model to be taken into account. By collecting data in various situations, data not related to component wear can be eliminated. Thus, the predictive model allows the user to optimize part use, reduce wafer fragmentation, and plan for repairs that will predict upcoming replacement events, thereby reducing cost of ownership.

  As can be seen from the above, a method is provided for performing sanity checks on components in a processing chamber. By using one or more multivariate prediction models, a more accurate analysis for the component is provided. That is, by using data from multiple sensors, a chamber health index test can be used to estimate and / or predict the remaining useful life of a component without the need for external metrology measurements. A more granular and less invasive chamber health index check provides a more cost effective method for identifying component status. Thus, the cost of ownership is reduced with the reduction of resources and / or components that are wasted and / or damaged.

  Although the invention has been described in terms of several preferred embodiments, there are alternatives, substitutions, and equivalents within the scope of the invention. While various examples are provided herein, these examples are intended to be illustrative and not limiting of the invention.

Also, the name and summary are provided for convenience and should not be used to interpret the claims. Further, the abstract is provided in a highly abbreviated form and is provided for convenience and is therefore used to interpret and limit the overall invention as expressed in the claims. Should not be done. If the term “set” is used herein, such term should have the generally understood mathematical meaning of zero, one, or more components. Intended. It should also be noted that there are many alternative approaches for implementing the method and apparatus of the present invention. Accordingly, the following appended claims are construed to include all such alternatives, substitutions, and equivalents as included within the true spirit and scope of the invention.
Application Example 1: Method for assessing the health of a processing chamber, executing a recipe, receiving processing data from a set of sensors during execution of the recipe, and setting a multivariate prediction model Analyzing the processing data, generating a set of component wear data values, comparing the set of component wear data values to a set of useful life threshold ranges, and Generating a warning if the set is outside the useful life threshold range set.
Application Example 2: The method of Application Example 1, further comprising extracting data from a library to support the set of multivariate prediction models.
Application Example 3: The method according to Application Example 2, wherein the recipe includes one of a client-specific recipe, a non-client-specific recipe, and a waferless cleaning automatic cleaning recipe.
Application Example 4: The method according to Application Example 2, wherein the processing data is analyzed by using a first multivariate prediction model in the set of multivariate prediction models.
Application Example 5: The method according to Application Example 2, wherein the processing data is analyzed by using at least two multivariate prediction models in the set of multivariate prediction models.
Application Example 6: The method of Application Example 5, wherein the processing data uses a first multivariate prediction model to generate a first set of component wear data values and a second component wear. When there is a difference between using a second multivariate predictive model to generate a set of data values and the first set of component wear data values and the second set of component wear data values Applying the second multivariate predictive model to the first set of component wear data values, wherein the first set of component wear data values is the second component A method that is less noisy than a set of wear data values.
Application Example 7: The method according to Application Example 2, wherein the set of multivariate prediction models includes at least one of an electrical model, a statistical model, and a plasma model.
Application Example 8: The method according to Application Example 2, wherein the set of multivariate prediction models is configured to analyze two or more consumable parts, each consumable part is a set of the useful life threshold range. The method associated with one of the useful life threshold ranges.
Application Example 9: The method according to Application Example 2, wherein the set of useful lifetime threshold ranges is user-configurable.
Application Example 10: The method according to Application Example 2, further comprising performing the assessment of the healthy state of the processing chamber after a measurement interval, the measurement interval being one of a predetermined period and A method comprising performing an assessment, as determined by performing a non-plasma test.
Application Example 11: The method of Application Example 2, further comprising analyzing the set of component wear data values to determine whether verification is necessary, wherein the verification comprises the component Generate analysis when a set of wear data values is outside the noise level threshold range, perform non-plasma tests to validate the set of component wear data values, and generate an integrated set of component wear data values To associate the set of component wear data values with a set of non-plasma test data values, wherein the integrated set of component wear data values is compared with the set of useful life threshold ranges, and the warning is The integrated set of component wear data values is outside the useful life threshold range set. If that is generated, the method comprising the associate, a.
Application Example 12: An article of manufacture comprising a program storage medium implemented with computer readable code, wherein the computer readable code is configured to assess the health of the processing chamber and code for executing a recipe; To generate code for receiving process data from a set of sensors during execution of the recipe, code for analyzing the process data using a set of multivariate prediction models, and a set of component wear data values A code for comparing the set of component wear data values with a set of useful life threshold ranges, and generating a warning if the set of component wear data values is outside the set of useful life threshold ranges And a manufactured product comprising:
Application example 13: The product according to application example 12, further comprising a code for extracting data from a library to support the set of multivariate prediction models.
Application Example 14: The manufactured article according to Application Example 13, which is a code for performing the assessment of the healthy state of the processing chamber after a measurement interval, wherein the measurement interval is a predetermined period of time. Article of manufacture with code, determined by one and non-plasma test execution.
Application Example 15: The manufactured article according to Application Example 13 and further code for analyzing the set of component wear data values to determine whether verification is necessary, wherein the verification is A code that occurs when the set of component wear data values is outside a noise level threshold range; a code for performing a non-plasma test to verify the set of component wear data values; and a component wear data value Code for associating the set of component wear data values with a set of non-plasma test data values to generate a set of useful life threshold ranges. And the warning indicates that the integrated set of component wear data values is If it is set outside the useful life threshold value range, it is generated, the article of manufacture comprising a cord, a.
Application Example 16: The product according to Application Example 13, wherein the recipe includes one of a client-specific recipe, a non-client-specific recipe, and a waferless cleaning automatic cleaning recipe.
Application example 17: The manufactured article according to application example 13, wherein the code for analyzing the processing data is a code for using the first multivariate prediction model in the set of the multivariate prediction models. Including manufactured goods.
Application Example 18: The manufactured article according to Application Example 13, wherein the code for analyzing the processing data uses at least two multivariate prediction models in the set of multivariate prediction models. Including manufactured goods.
Application Example 19: The manufactured article according to Application Example 18, wherein the code for analyzing the processing data uses a first multivariate prediction model to generate a first set of component wear data values. A code for using a second multivariate predictive model to generate a second set of component wear data values, the first set of component wear data values and the second component wear Code for applying the second multivariate predictive model to the first set of component wear data values when there is a difference between the set of data values; and The multivariate prediction model is a manufactured product with less noise than the first multivariate prediction model.
Application example 20: The manufactured article according to application example 13, wherein the set of multivariate prediction models includes at least one of an electrical model, a statistical model, and a plasma model.

Claims (14)

A method for assessing the health of a processing chamber, comprising:
Running the recipe,
Receiving processing data captured by a set of sensors during execution of the recipe;
Extracting data from the library to support a set of multivariate predictive models;
And analyzing the processed data by using at least two multivariate predictive models within the set of multivariate prediction model,
Generating a set of component wear data values representing the health state of the component for a component that is an integral part or a multi-part assembly in the processing chamber ;
Comparing the set of component wear data values with a set of useful life threshold ranges;
And said component wear data values if it is outside the range of the useful life threshold value range, and generates an alert,
A method comprising: The method of claim 1 , comprising:
The method, wherein the recipe includes one of a client-specific recipe, a non-client-specific recipe, and a waferless automatic cleaning recipe. The method of claim 1 , comprising:
Said processing data, and the use of the first multivariate predictive models to generate a set of first said component wear data values, the second to generate a second set of said component wear data values The first set of component wear data values when there is a difference between using a multivariate predictive model and the first set of component wear data values and the second set of component wear data values. Applying the second multivariate prediction model to the method, wherein the first set of component wear data values is less noisy than the second set of component wear data values . The method of claim 1 , comprising:
The method, wherein the set of multivariate prediction models includes at least one of an electrical model, a statistical model, and a plasma model. The method of claim 1 , comprising:
The set of multivariate predictive models is configured to analyze two or more consumable parts, each consumable part being associated with one useful life threshold range of the set of useful life threshold ranges; Method. The method of claim 1 , comprising:
The method wherein the set of useful lifetime threshold ranges is user configurable. The method of claim 1 , further comprising:
Performing the assessment of the health of the processing chamber after a measurement interval, the measurement interval being determined by one of a predetermined period and performing a non-plasma test. The method of claim 1 , further comprising:
Analyzing the set of component wear data values to determine whether verification is required, wherein the verification occurs when the set of component wear data values is outside a noise level threshold range; Analysis and
Performing a non-plasma test to verify the set of component wear data values;
Associating the set of component wear data values with a set of non-plasma test data values to generate an integrated set of component wear data values, the integrated set of component wear data values comprising the useful life threshold range; is compared with the set, the warning, when integrated set of said component wear data value is outside the range of the useful life threshold value range, is generated, and associating,
A method comprising: A computer program in order to assess the health status of the processing chamber,
A function to execute the recipe,
A function for receiving processing data captured by a set of sensors during execution of the recipe;
The ability to pull data from a library to support a set of multivariate predictive models,
A function for analyzing the processed data by using at least two multivariate predictive models within the set of multivariate prediction model,
A function for generating a set of component wear data values representing the health state of the component for a component that is an integral part or an assembly of parts in the processing chamber ;
A function for comparing the set of component wear data values with a set of useful life threshold ranges;
If the component wear data value is outside the range of the useful life threshold value range, a function for generating an alert,
A computer program comprising: The computer program according to claim 9 , further comprising:
The assessment of the health of the processing chamber to a function for carrying out later measurement interval, the measurement interval is determined by the execution of one and non-plasma test default period, the computer program comprising functions . The computer program according to claim 9 , further comprising:
A function for analyzing the set of component wear data values to determine whether verification is required, wherein the verification is performed when the set of component wear data values is outside a noise level threshold range. The resulting function ,
A function for performing a non-plasma test to verify the set of component wear data values;
A function for associating the set of component wear data values with a set of non-plasma test data values to generate an integrated set of component wear data values, the integrated set of component wear data values comprising: is compared with a set threshold value range, the alerts, when integrated set of said component wear data value is outside the range of the useful life threshold value range, is generated, and functions,
A computer program comprising: A computer program according to claim 9 ,
The computer program includes one of a client-specific recipe, a non-client-specific recipe, and a waferless automatic cleaning recipe. A computer program according to claim 9 ,
Functions for analyzing the processed data has a function for using the first multivariate predictive models to generate a set of first said component wear data values, the second set of the components wear data values If there is a difference between the function for using the second multivariate predictive model to generate and the first set of component wear data values and the second set of component wear data values, And a function for applying the second multivariate prediction model to the first set of component wear data values, wherein the second multivariate prediction model is the first multivariate prediction model. Computer program with less noise than. A computer program according to claim 9 ,
The computer program , wherein the set of multivariate prediction models includes at least one of an electrical model, a statistical model, and a plasma model.

Images