Hybrid physics / machine learning modeling of processes

Abstract

To provide a method and a semiconductor processing tool for generating a hybrid model for modeling a process in a semiconductor processing apparatus.SOLUTION: A method of creating a hybrid machine learning model includes identifying a first set of cases over a first range of process and / or hardware parameters, conducting experiments in a laboratory for the first set of cases, collecting experimental output from the experiments, and running a physics-based simulation for the first set of cases. The method further includes collecting model outputs from the simulation and correlating the model outputs with the experimental outputs using a machine learning algorithm to provide a hybrid machine learning model.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Patent Application No. 17 / 166,965, filed February 3, 2021, the entire contents of which are incorporated herein by reference.

[0002] FIELD Embodiments of the present disclosure relate to the field of semiconductor processing, and more particularly to the use of hybrid modeling and virtual sensors for processes within semiconductor processing tools. [Background technology]

[0003] Semiconductor substrate processing is becoming increasingly complex as feature sizes in semiconductor devices continue to shrink. A given process can involve many different process parameters (i.e., knobs) that can be individually controlled to provide a desired result on the wafer. For example, the desired result on the wafer can refer to feature profile, layer thickness, layer chemical composition, etc. As the number of knobs increases, the theoretical process space available for process tuning and optimization becomes exponentially larger.

[0004] When hardware changes are made to a semiconductor processing tool, the knobs must be changed to account for the new hardware settings. Because hardware changes are costly to implement, it is beneficial to be able to predict or estimate the performance of the new hardware before physically building it. Traditional approaches use intuition and trial-and-error (both of which can be subjective) to gain qualitative understanding from past experiments with similar hardware and estimate the performance of the new hardware and / or identify new process parameters. Depending on the application, insights from physics models are also used. However, physics-based approaches are often incomplete or inherently disparate (e.g., separate models for temperature, plasma, and flow). This means that no existing approach provides a quantitative and objective path for tuning processes for new hardware. Summary of the Invention

[0005]

[0003] Embodiments described herein include a process for generating a hybrid model for modeling a process in a semiconductor processing equipment. In a particular embodiment, a method for creating a hybrid machine learning model includes identifying a first set of cases spanning a first range of process parameters and / or hardware parameters and conducting laboratory experiments for the first set of cases. The method can further include collecting experimental outputs from the experiments and performing physics-based simulations for the first set of cases. In one embodiment, the method can further include collecting model outputs from the simulations and correlating the model outputs with the experimental outputs using a machine learning algorithm to provide a hybrid machine learning model.

[0006] Additional embodiments may include a semiconductor processing tool with a virtual sensor. In one embodiment, the semiconductor processing tool comprises a chamber and a controller for varying a control variable of the semiconductor processing tool. In one embodiment, the controller receives as an input a difference between a measured output variable from the chamber and an output variable setpoint. In one embodiment, the semiconductor processing tool further comprises a virtual sensor for generating an estimated system state variable used to determine the output variable setpoint.

[0007] An additional embodiment may include a method for creating a hybrid machine learning model. In one embodiment, the method includes identifying a first set of cases spanning a first range of process and / or hardware parameters and performing a physics-based simulation for the first set of cases. In one embodiment, the method further includes collecting output from the physics-based simulation and generating a reduced-order physics simulation model using a first machine learning algorithm. In one embodiment, the method may further include identifying a second set of cases spanning a second range of process and / or hardware parameters, the second set of cases being smaller than the first set of cases, and conducting laboratory experiments on the second set of cases. In one embodiment, the method may further include collecting experimental output from the experiments and performing a physics-based simulation for the second set of cases, the physics-based simulation using the reduced-order physics simulation model. In one embodiment, the method may further include collecting model output from the simulation and correlating the model output with experimental output using a second machine learning algorithm to provide a hybrid machine learning model. [Brief explanation of the drawings]

[0008] [Figure 1A]FIG. 1 is a process flow diagram illustrating a process for creating a reduced-order physics simulation model, according to one embodiment. [Figure 1B] FIG. 1 is a process flow diagram illustrating a process for creating a hybrid machine learning model, according to one embodiment. [Figure 1C] FIG. 1 is a process flow diagram illustrating a process for deploying a hybrid machine learning model under new process and / or hardware conditions, according to one embodiment. [Figure 2] FIG. 1 is a perspective view of a radical oxidation tool according to one embodiment. [Figure 3] FIG. 1 illustrates the use of a hybrid model in a radical oxidation tool, according to one embodiment. [Figure 4A] 10 is a graph illustrating hybrid model predictions compared to actual results, according to various embodiments. [Figure 4B] 10 is a graph illustrating hybrid model predictions compared to actual results, according to various embodiments. [Figure 4C] 10 is a graph illustrating hybrid model predictions compared to actual results, according to various embodiments. [Figure 4D] 10 is a graph illustrating hybrid model predictions compared to actual results, according to various embodiments. [Figure 5A] 1 is a control architecture illustrating the use of virtual sensors, according to one embodiment. [Figure 5B] 1 is a control architecture incorporating virtual sensors, according to one embodiment. [Figure 6] FIG. 1 is a more detailed diagram of a control architecture incorporating a virtual sensor and a loop for providing updates to a model that generates the virtual sensor readings, according to one embodiment. [Figure 7A] 1 is a control architecture comprising a virtual sensor and a controller for updating parameters in a model to generate virtual sensor readings, according to one embodiment. [Figure 7B]1 is a control architecture with a virtual sensor and a controller that utilizes a Kalman filter, according to one embodiment. [Figure 8] FIG. 1 is a block diagram of an exemplary computer system according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] Methods for modeling process conditions within a semiconductor processing tool and the use of virtual sensors are described herein. In the following description, numerous specific details are set forth in order to provide a thorough understanding of embodiments of the present disclosure. It will be apparent to those skilled in the art that embodiments of the present disclosure may be practiced without these specific details. In other instances, well-known aspects have not been described in detail in order to not unnecessarily obscure embodiments of the present disclosure. Furthermore, it should be understood that the various embodiments shown in the figures are illustrative representations and have not necessarily been drawn to scale.

[0010] As mentioned above, there is no quantitative and objective approach for estimating the performance of a new hardware configuration or for providing new process parameters after hardware changes. Therefore, complex and subjective process design methods are currently used. This leads to expensive process design and may not identify the optimal process parameters for a given hardware configuration. Furthermore, in a high-volume manufacturing (HVM) environment, multiple tools may be used in parallel to perform a desired process on a substrate. The process parameters for each tool may need to vary. Therefore, each tool must undergo expensive process optimization.

[0011] Accordingly, embodiments disclosed herein include machine learning models that use features extracted from one or more physics-based models of a system. The methods described herein include extracting features from the physics-based models and training a machine learning algorithm using experimental data obtained from processing physical substrates. Specifically, the methods disclosed herein can include generating a physics-based simulation reduced order model (ROM) and using the ROM in combination with the experimental data to generate a hybrid machine learning model. The hybrid machine learning model can then be deployed to predict on-wafer results for new process conditions, new hardware, or even different processing tools.

[0012] The hybrid machine learning model can be generated for any semiconductor processing tool. For example, the hybrid machine learning model may be used for a deposition tool or an etch tool. In a particular embodiment, the hybrid machine learning model is generated for a radical oxidation tool.

[0013] 1A, a process flow diagram illustrating a process 110 for forming a reduced-order physics simulation model, according to one embodiment, is shown. In one embodiment, process 110 begins with step 111, which involves identifying a set of cases spanning a wide range of process and / or hardware parameters. Because the process and / or hardware parameters are computationally modeled, a wide range of process and / or hardware parameters is possible. The computational cost is significantly less than the cost required to perform physical experiments with various process and / or hardware parameters.

[0014] In one embodiment, process 110 continues at step 112 with performing a physics-based simulation for the set of cases. The physics-based simulation is calculated to determine an output based on how process parameters and / or hardware parameters interact according to natural physical laws. The physics-based simulation is performed computationally; that is, it is not necessary to actually process a substrate to determine the results of the physics-based simulation.

[0015] In one embodiment, process 110 continues at step 113 with collecting output from the physics-based simulation. The output may be referred to as simulation output because it is the result of a simulation rather than the processing of an actual substrate.

[0016] In one embodiment, the process 110 continues at step 114 with applying the simulation output to a machine learning algorithm. The machine learning algorithm correlates the process and / or hardware parameters with the simulation output to generate a reduced-order physics-based simulation model 115. The machine learning algorithm includes a mathematical model that correlates the simulation output with the process and / or hardware parameters. The model can include one or more of singular value decomposition (SVD), proper orthogonal decomposition (POD), Gaussian process regression, other kernel-based regression, response surface-based regression, neural network models, regression using radial basis functions, and regression models that consider spatial connectivity. In one embodiment, the machine learning model typically has model parameters that need to be determined. One of the primary tasks involved in forming the reduced-order model is selecting the combination of mathematical model and model parameters that results in the best fit of the simulation output to the process and / or hardware parameters. The reduced-order simulation model 115 allows subsequent process and / or hardware parameters to be explored in less time than would be required to perform a full physics-based simulation.

[0017] 1B, a process 120 for creating a hybrid machine learning model is shown, according to one embodiment. As described in more detail below, a hybrid machine learning model allows for computationally predicting outcomes on a substrate based on a given set of process and / or hardware parameters. A hybrid machine learning model can be applied to changes in a single tool or even to changes in various instances of a tool.

[0018] In one embodiment, process 120 may begin with step 121, which involves identifying a set of cases spanning a range of process and / or hardware parameters. The range of cases in step 121 may be smaller than the range of cases in step 111 because the range of cases is explored using a physical substrate, which is more time-consuming and cost-intensive than running physics-based simulations alone.

[0019] In one embodiment, process 120 can follow a pair of branches that can be executed in parallel (although not necessarily in all embodiments). The first branch begins with step 122, which involves conducting laboratory experiments for the set of cases identified in step 121. The experiments involve physically processing substrates according to selected process and / or hardware parameters. In one embodiment, the first branch can follow step 123, which involves collecting output from the experiments. The output from the experiments can include, for example, output on the substrate, such as deposition thickness, etch rate, composition, uniformity, etc.

[0020] In one embodiment, the second branch begins with step 124, which involves running a physics-based simulation for the set of selected cases. In some embodiments, the physics-based simulation is the same simulation used in step 112. In other embodiments, the physics-based simulation may utilize a reduced-order physics simulation model developed in process 110. If a reduced-order physics simulation model is used in step 124, the time and computational resources required to run the simulation may be reduced. In one embodiment, the second branch may continue with collecting output from the physics-based simulation.

[0021] In one embodiment, the first and second branches converge again at step 126, which involves using a machine learning algorithm to correlate the collected experimental outputs with the collected physics-based simulation outputs. The machine learning algorithm includes a mathematical model that correlates the collected experimental outputs with the collected physics-based simulation outputs. The model can include one or more of singular value decomposition (SVD), proper orthogonal decomposition (POD), Gaussian process regression, other kernel-based regression, response surface-based regression, neural network model, regression using radial basis function, and regression model that considers spatial connectivity. The machine learning algorithm determines the selection of the mathematical model and corresponding model parameters to minimize the error between the predicted on-substrate properties and the experimentally measured on-substrate properties. The machine learning algorithm outputs a hybrid machine learning model 127 that can take process parameters and / or hardware parameters as inputs and output on-substrate outputs such as deposition thickness, etch rate, composition, uniformity, etc.

[0022] 1C , a process 130 for deploying the hybrid machine learning model 127 is shown, according to one embodiment. In one embodiment, the process 130 begins with selecting new process and / or hardware conditions. The new process and / or hardware conditions may be any process and / or hardware conditions, including those that are different from or outside the range of process and / or hardware conditions explored in steps 111 and 121. In some embodiments, the process and / or hardware conditions may even be on a different tool instance than the tool explored in process 120. That is, once the hybrid machine learning model is developed, it has the flexibility to be deployed to similar processing tools throughout a manufacturing facility, even in the absence of experimental data.

[0023] In one embodiment, process 130 may continue to step 132, which includes evaluating the physics simulation using the reduced order physics simulation model developed in step 115 (provided that hardware parameters were included in the formation of the model developed in step 115) or by running a physics simulation. The reduced order physics simulation or the output of the physics simulation may then be fed to a hybrid machine learning model in step 133. The reduced order physics simulation model enables process conditions and / or hardware conditions to be mapped to physical space for use by the hybrid machine learning model in step 133.

[0024] Step 133 can include evaluating the hybrid machine learning model developed in step 127 above. The hybrid machine learning model can output on-substrate results at 134. That is, new process and / or hardware conditions can be directly mapped to on-substrate results, such as deposition thickness, etch rate, composition, uniformity, etc. This is a significant improvement over existing processes that require physical testing of the substrate to obtain on-substrate results.

[0025] 2, a perspective view of a semiconductor processing tool 240 is shown in accordance with one embodiment. While a particular semiconductor processing tool 240 is shown, it will be understood that the semiconductor processing tool 240 may be any processing tool typical in semiconductor manufacturing, such as a deposition tool, an etch tool, etc. In the particular embodiment shown in FIG. 2, the semiconductor processing tool is a radical oxidation tool.

[0026] In one embodiment, the semiconductor processing tool 240 can include a gas inlet 241. Gas can enter the gas inlet 241 and enter the chamber 245 through a tunnel 242. The top of the chamber 245 can be sealed with a quartz plate 243. A heating element (not shown) can be disposed on the quartz plate 243 for rapid thermal control within the chamber 245. In one embodiment, by-products and excess reactants can be removed from the chamber 245 by an exhaust 244. The exhaust 244 can be in fluid communication with a vacuum pump (not shown), or the like.

[0027] 3, a diagram 350 is shown, by way of example, illustrating how the hybrid model can be used in a radical oxidation tool. As shown, a set of process inputs is provided in block 351. The process inputs can include process parameters used in the radical oxidation process, such as, but not limited to, soak time, temperature, pressure, total gas flow, H2 side flow, H2 %, etc. In one embodiment, the process inputs can also include hardware configurations, such as, but not limited to, the geometry of various parts of the tool (e.g., injection cartridge), the spacing between the substrate and the quartz plate 343, etc.

[0028] In one embodiment, the process inputs in block 351 are provided to a physics-based model or a reduced-order physics-based model in block 352. The model can provide outputs based on physical equations. For example, on-wafer outputs can include pressure, deposition rate, and mole fraction, while off-wafer outputs can include temperature.

[0029] In one embodiment, the process inputs of block 351 and the model outputs of block 353 may be fed to a hybrid model 354. The hybrid model 354 may be substantially similar to any of the hybrid models described in more detail above. The hybrid model processes the incoming data from the process inputs of block 351 and the model outputs of block 353 and provides an output of a predicted deposition on the wafer at block 355.

[0030] The hybrid model was shown to provide an accurate mapping of expected output on a substrate. For example, Figures 4A-4D are plots of normalized deposition rates across a substrate for various process parameters. In Figures 4A-4D, a hybrid model of a radical oxidation process was generated using a process similar to that described above and deployed to tools with widely varying injection cartridge geometries. This hybrid model was used to predict deposition rates across the substrate surface, and experimental data was then acquired to confirm the accuracy of the hybrid model. In Figures 4A-4D, the hybrid model predictions closely matched the experimental data. For example, the average error was less than 9% across a range of processing conditions.

[0031] In yet another embodiment disclosed herein, physics-based models and machine learning can be used to provide virtual sensors within semiconductor processing tools. This is particularly useful for determining process conditions that cannot be easily measured (or measured at all) with traditional physical sensors. Installing physical sensors within a processing tool is expensive and cumbersome. However, process control is effective when process conditions (especially on a substrate) are known. Physics-based models can address this issue by providing virtual sensors that provide details of on-substrate characteristics without the need to use physical sensors. Physics-based models can also be useful for controller testing and virtual experimentation for controller development.

[0032] Virtual sensors may be used to help control a process. Similar to physical sensors, virtual sensor outputs may be compared by a controller against set points to determine whether changes need to be made to a process. Additionally, embodiments disclosed herein may utilize machine learning or artificial intelligence to continuously update physics-based models to improve the accuracy of virtual sensor outputs.

[0033] 5A, a simplified diagram of a control architecture 560 for a processing tool is shown, according to one embodiment. As shown, a chamber 561 can include physical sensors 562 that feed a controller 565. The controller sends control signals back to the chamber 561 to adjust one or more processing conditions. In another loop, a model 563 (e.g., a physics-based model) is connected to a virtual sensor 564. The virtual sensor 564 outputs a value to the controller 565. A more detailed description of the virtual sensor 564 is provided below.

[0034] 5B, a more detailed diagram of a control architecture 560 is shown according to one embodiment. In one embodiment, an output variable (or vector) y is fed to a virtual sensor 564. The virtual sensor outputs a virtual sensor variable (or vector) y1. The desired virtual sensor variable y1 des The setpoint 566 of is compared against the output variable y by the controller 565. A control signal u is provided to the chamber 561 to vary the output variable y depending on the calculated difference.

[0035] 6, a diagram of a tool's control architecture 670 is shown that includes a virtual sensor 676 coupled to an updatable model 673, according to one embodiment. In one embodiment, the control architecture 670 begins with a chamber 671. The chamber 671 may refer to any part of a semiconductor processing tool. In one embodiment, an output variable y (or vector) is generated by a first controller 672 to generate a desired output variable y des The first controller 672 returns the input variable u (or vector) to the chamber 671. The input variable u is also fed to a model 673, which is described in more detail below. The desired output variable y des is generated by the second controller 678 utilizing the virtual sensor data.

[0036] In one embodiment, model 673 is a physics-based model, i.e., model 673 is a model that generates estimates of system state variables (or vectors). Calculate the reactions in chamber 671 from a physics-based perspective to provide TIFF2026021345000002.tif4170. Estimated system state variables TIFF2026021345000003.tif4170 can be a virtual sensor value. The measurements in TIFF2026021345000004.tif4170 are values ​​that are desired but typically unknown or unmeasured. For example, estimated state variables TIFF2026021345000005.tif4170 may be the wafer temperature in some embodiments. However, other estimated state variables may be used. TIFF2026021345000006.tif4170, or multiple different estimated state variables It should be understood that even TIFF2026021345000007.tif4170 can be provided by model 673.

[0037] In one embodiment, the estimated state variables TIFF2026021345000008.tif4170 is provided to the virtual sensor 676 where it can be accessed by the system. In a particular embodiment, the virtual sensor 676 receives the estimated state variables TIFF2026021345000009.tif4170 is fed to a second controller 678, which in turn generates the estimated state variables Set TIFF2026021345000010.tif4170 to the value of the state variable x des Compare with. TIFF2026021345000011.tif4170 and x des Depending on the difference between y des to the first controller.

[0038] In one embodiment, the model 673 can be continuously updated through a machine learning or artificial intelligence block 675. In particular, the estimated state variables TIFF2026021345000012.tif4170 is also fed to a second model 674. The second model generates the estimated output variables Output TIFF2026021345000013.tif5170 (or a vector). Estimated output variables TIFF2026021345000014.tif5170 is compared with the output variable y from chamber 671. Next, machine learning block 675 calculates the estimated output variable The first model 673 can be modified and refined (e.g., using the state-space matrices A, B, C, and / or D) to approximate the estimated state variables y. This will also lead to more accurate predictions of TIFF2026021345000016.tif4170.

[0039] 7A, a diagram of a control architecture 780 with a virtual sensor 785 is shown, according to one embodiment. During an experiment 781 in a chamber, an output variable y is provided to a controller 784. The controller converts the output variable y into estimated output variables y, which are generated using various physics models 783 and 782. Compare with TIFF2026021345000017.tif5170. In one embodiment, the model 783 for the state estimates is controlled by Equation 1, and the model 782 for the output variables is controlled by Equation 2. TIFF2026021345000018.tif14170

[0040] In Equation 1 and Equation 2, matrices A, B, C, and D are functions of the parameters of the experiment 781 and can be determined using a physics-based model or a system model. If a statistical model is used, matrices A, B, C, D may not have a physical basis, and varying A, B, C, D will not correlate with the physical parameters. Furthermore, it will be understood that A, B, C, D may be functions of time and x and y.

[0041] In one embodiment, the assumptions of the control architecture 780 are the measured output y and the predicted output y The error between the model and the actual state is due to uncertain parameters in the system, and the physics is correct. That is, the model 783 for the state estimate is not modified due to physics. Noise in the system is not taken into account. In other words, noise in the system is offset by changing the parameter values ​​A, B, C, or D. Changing the model parameters can be done through optimization and / or inverse techniques, as long as the controller 784 has appropriate hypotheses to begin with. Furthermore, it should be understood that the computational complexity depends on the matrices A, B, C, and D. With today's computing power, this computational complexity is well within the realm of real-time computation. In this way, a real-time virtual sensor 785 is possible.

[0042] 7B, a diagram of a control architecture 780 with a virtual sensor 785 is shown, according to one embodiment. During an experiment 781 in a chamber, an output variable y is provided to a controller 786. The controller 786 converts the output variable y into estimated output variables y, which are generated using various physics models 783 and 782. Compare to TIFF2026021345000020.tif5170. In one embodiment, model 783 for the state estimates is controlled by Equation 1, and model 782 for the output variables is controlled by Equation 2. In contrast to the embodiment of FIG. 7A, controller 786 may apply a Kalman filter with gain L.

[0043] In Equation 1 and Equation 2, matrices A, B, C, and D are functions of the parameters of the experiment 781 and can be determined using a physics-based model, a system model, or a statistical model. It will be further understood that A, B, C, and D can be functions of time and x and y.

[0044] In one embodiment, the assumptions of the control architecture 780 are the measured output y and the predicted output y Any errors between the measured and predicted outputs are due to error sources, and the physics and parameters are correct. This means that the model 783 for the state estimate is corrected to account for errors, rather than being modified due to the physics. Noise in the system is also accounted for. This model framework can be used to predict the state estimate, enabling real-time virtual sensors 785. Furthermore, the model automatically corrects for any errors between the measured and predicted outputs by modifying the parameters of the model 783 and / or 782.

[0045] In one embodiment, the controller architecture with virtual sensor capabilities described herein can be tested in various ways. In one embodiment, the controller architecture can be tested in an operational chamber or system. That is, physical substrate processing can be used to test the controller architecture. Tool time and other resources are required to perform this process. In another embodiment, the controller architecture with virtual sensor capabilities can be tested through software simulation. For example, a virtual chamber modeled with a physics-based model and / or a hybrid model can be used to test the controller architecture. In such an embodiment, only computational resources are required, saving valuable tool time, substrates, and other physical resources.

[0046] FIG. 8 shows a diagrammatic representation of a machine in the exemplary form of a computer system 800, within which a set of instructions may be executed to cause the machine to perform any one or more of the methodologies described herein. In alternative embodiments, the machine may be connected (e.g., networked) to other machines in a local area network (LAN), an intranet, an extranet, or the Internet. The machine may operate as a server or a client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine may be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a web appliance, a server, a network router, switch, or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by the machine. Furthermore, while only a single machine is illustrated, the term “machine” shall be taken to include any collection of machines (e.g., computers) that individually or jointly execute a set of instructions (or multiple sets) to perform any one or more of the methodologies described herein.

[0047] The exemplary computer system 800 includes a processor 802, a main memory 804 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc.), a static memory 806 (e.g., flash memory, static random access memory (SRAM), MRAM, etc.), and a secondary memory 818 (e.g., a data storage device), which communicate with each other via a bus 830.

[0048] Processor 802 represents one or more general-purpose processing devices, such as a microprocessor, a central processing unit, or the like. More specifically, processor 802 may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. Processor 802 may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. Processor 802 is configured to execute processing logic 826 for performing the operations described herein.

[0049] Computer system 800 may further include a network interface device 808. Computer system 800 may also include a video display device 810 (e.g., a liquid crystal display (LCD), a light emitting diode display (LED), or a cathode ray tube (CRT)), an alphanumeric input device 812 (e.g., a keyboard), a cursor control device 814 (e.g., a mouse), and a signal generation device 816 (e.g., a speaker).

[0050] The secondary memory 818 may include a machine-accessible storage medium (or more specifically, a computer-readable storage medium) 832 having stored thereon one or more sets of instructions (e.g., software 822) that embody any one or more of the methods or functions described herein. The software 822 may also reside, completely or at least partially, within the main memory 804 and / or within the processor 802 during its execution by the computer system 800, with the main memory 804 and the processor 802 also constituting machine-readable storage media. The software 822 may further be transmitted or received over the network 820 via the network interface device 808.

[0051] Although machine-accessible storage medium 832 is shown in the exemplary embodiment to be a single medium, the term "machine-readable storage medium" shall be interpreted to include a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) that store one or more sets of instructions. The term "machine-readable storage medium" shall also be interpreted to include any medium that can store or encode a set of instructions that can be executed by a machine to cause the machine to perform any one or more of the methods of the present disclosure. The term "machine-readable storage medium" shall therefore be interpreted to include, but is not limited to, solid-state memory, and optical and magnetic media.

[0052] According to one embodiment of the present disclosure, a machine-accessible storage medium stores instructions for causing a data processing system to perform a method for creating a hybrid machine learning model.

[0053] Thus, a method for generating a hybrid machine learning model has been disclosed.

Claims

1. It is a semiconductor processing tool, Chamber, A controller for changing the control variable of the semiconductor processing tool, the controller which receives the difference between the output variable measured from the chamber and the output variable set value as input, and A virtual sensor for generating estimated system state variables used to determine the output variable setting value, A semiconductor processing tool equipped with the following features.

2. The semiconductor processing tool according to claim 1, further comprising a second controller for changing the output variable setting value, the second controller receiving the difference between the estimated system state variable and the system state variable setting value as input.

3. The semiconductor processing tool according to claim 2, further comprising a first model, the first model receiving the control variable as input and outputting the estimated system state variable provided to the virtual sensor.

4. The semiconductor processing tool according to claim 3, further comprising a second model, the second model receiving the estimated system state variables as input and outputting estimated values ​​of the output variables.

5. The semiconductor processing tool according to claim 4, further comprising a machine learning algorithm, wherein the machine learning algorithm receives the difference between the output variable and the estimated value of the output variable as input, and the machine learning algorithm updates the first model.

6. The semiconductor processing tool according to claim 5, wherein the machine learning algorithm utilizes a Kalman filter.

7. The semiconductor processing tool according to claim 2, wherein the estimated system state variable is wafer temperature.

8. The semiconductor processing tool according to claim 7, wherein the semiconductor processing tool is a radical oxidation tool.

Images