Deep learning model for high-mix semiconductor manufacturing

Abstract

To solve the problem that adopting AI / ML technologies to semiconductor manufacturing has been difficult because building and deploying a deep learning model in a semiconductor manufacturing environment is not a simple, straightforward process but involves complex steps, especially in a high-mix manufacturing environment.SOLUTION: Disclosed are techniques for applying a neural network deep learning model in a fabrication strategy for high-mix semiconductor manufacturing, such as deposition, chemical-mechanical polishing (CMP), etching, photolithography, plating, etc. Training and normal operation modes of the fabrication strategy are described.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of priority to U.S. Provisional Patent Application No. 63 / 247,904, filed September 24, 2021, the contents of which are incorporated herein by reference in their entirety.

[0002] FIELD OF THE INVENTION The present disclosure generally relates to applying deep learning networks, such as recurrent neural network (RNN) deep learning models, in fabrication strategies for high-mix semiconductor manufacturing, such as deposition, chemical-mechanical polishing (CMP), etching, photolithography, and plating. [Background technology]

[0003] Semiconductor manufacturing typically utilizes linear process models for processing parameters such as calculating deposition time. For example, a linear process model of layer thickness versus process time can be used to calculate process time. However, linear process models can be limited in their application and use. Summary of the Invention [Problem to be solved by the invention]

[0004] Over the past decade, driven primarily by the tremendous expansion in the availability of data and computing power, artificial intelligence (AI) and machine learning (ML) technologies have found their way into many different domains, disrupting our lives and the way we solve problems. For example, these technologies refine online search and shopping results, customize advertisements, tailor news feeds, and even drive our cars. Recently, AI / ML's ability to autonomously learn from data and rapidly discover patterns and relationships has found application in metrology and inspection in the semiconductor manufacturing industry. However, adopting AI / ML technologies in semiconductor manufacturing has been challenging because building and deploying deep learning models in a semiconductor manufacturing environment is not a simple or straightforward process and involves complex steps, especially in high-mix manufacturing environments. [Means for solving the problem]

[0005] This disclosure describes techniques for applying recurrent neural network (RNN) deep learning models to fabrication strategies for high-mix semiconductor manufacturing, such as deposition, chemical mechanical polishing (CMP), etching, photolithography, and plating.

[0006] The present disclosure describes a method for setting at least one process parameter for manufacturing a semiconductor device, the method including receiving context information regarding the at least one process parameter, inputting the context information into a machine learning network, receiving a predicted value for the at least one process parameter from the machine learning network, setting the at least one process parameter based on the predicted value to manufacture the semiconductor device, receiving measurement results of the semiconductor device associated with the at least one process parameter, and feeding the measurement results back to the machine learning network.

[0007] The present disclosure also describes a control system including at least one hardware processor. The control system also includes at least one memory storing instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform operations including receiving context information regarding at least one process parameter, inputting the context information to a machine learning network, receiving a predicted value for the at least one process parameter from the machine learning network, setting the at least one process parameter based on the predicted value to manufacture a semiconductor device, receiving measurement results of the semiconductor device associated with the at least one process parameter, and feeding the measurement results back to the machine learning network.

[0008] The present disclosure further describes a machine storage medium having instructions embodying thereon that, when executed by the machine, cause the machine to perform operations including receiving context information regarding at least one process parameter, inputting the context information into a machine learning network, receiving a predicted value for the at least one process parameter from the machine learning network, setting the at least one process parameter based on the predicted value to manufacture a semiconductor device, receiving measurement results of the semiconductor device associated with the at least one process parameter, and feeding the measurement results back to the machine learning network.

[0009] The various figures of the accompanying drawings are merely illustrative of example implementations of the present disclosure and should not be considered as limiting its scope. [Brief explanation of the drawings]

[0010] [Figure 1] 1 illustrates example portions of a fabrication system, according to some examples of the present disclosure. [Figure 2]1 illustrates a framework for a run-to-run strategy based on an RNN deep learning model, in accordance with some examples of the present disclosure. [Figure 3] FIG. 1 is a flow diagram of a method for training an RNN model, according to some examples of the present disclosure. [Figure 4] 10 illustrates examples of selection features for a deposition rate model, according to some examples of the present disclosure. [Figure 5] 1 illustrates an example of one-hot encoding for transforming string data, according to some examples of the present disclosure. [Figure 6] 1 illustrates a flow diagram of a method for manufacturing a semiconductor device in a normal operating mode, according to some examples of the present disclosure. [Figure 7] 10 illustrates simulation results of a deposition controller, according to some examples of the present disclosure. [Figure 8] 1 illustrates an example block diagram comprising a machine upon which any one or more of the techniques (eg, methodologies) discussed herein may be implemented. DETAILED DESCRIPTION OF THE INVENTION

[0011] A recurrent neural network (RNN) deep learning model can be applied to fabrication strategies for high-mix semiconductor manufacturing, such as deposition, chemical mechanical polishing (CMP), etching, photolithography, and plating. The RNN model can be trained and then used at run time for the fabrication strategy. The training and normal operation modes of the fabrication strategy are described.

[0012] First, an RNN model is configured and trained to calculate process parameters during a training mode. In the training mode, relevant test data may be collected. The collected data may include different features (or model inputs). In some examples, the features may be independent of each other. The collected data may include sufficient variability based on the features for model extraction. The collected data may be preprocessed. For example, string data may be converted to numeric data. The collected data may also be scaled so that features are evaluated by their ratios rather than absolute values. Furthermore, the hyperparameters of the model may be tuned.

[0013] Second, in normal operation (runtime) mode, the trained RNN model can be deployed in a manufacturing process to calculate and set process parameters. Relevant data for each run can be collected. The data can be processed and filtered, and then fed back to the RNN model for updating. The updated RNN model can then predict process parameters of the semiconductor manufacturing process with improved accuracy. For example, machine settings for different processes, such as deposition time, exposure dose, and overlay settings, can be calculated using the RNN model techniques described herein.

[0014] 1 illustrates an example portion of a fabrication system 100 that may be used to implement one or more of the techniques shown and described herein. The fabrication system 100 may be used to manufacture and inspect substrates 102, such as semiconductor devices.

[0015] The fabrication system 100 may include one or more tools 110 (also referred to as machines) to perform processing steps on the substrate 102 for fabrication. In some examples, the tool 110 may include a chemical vapor deposition (CVD) machine for depositing one or more layers on the substrate 102. In some examples, the tool 110 may include a polishing tool for performing CMP, an illumination source for emitting a beam of radiation (e.g., electromagnetic waves) to project onto the substrate 102 for photolithography, a plating tool for plating the substrate, and / or other suitable tools for semiconductor fabrication.

[0016] The fabrication system 100 may include metrology tools 112 (also referred to as inspection tools) for measuring various features on or characteristics of the substrate 102. The various features and characteristics may include, for example, film thickness measurements of features formed on the substrate, critical-dimension (CD) measurements (x-dimension, y-dimension, and / or z-dimension), pitch of line-space features formed on the substrate, overlay offset from one layer to another on the substrate, or some other measurement or characteristic known to those skilled in the art. The inspection tools may be used to verify compliance of features formed on or otherwise found on the substrate. For example, inspection may be of the substrate on which the integrated circuit die was fabricated, the location of the die, the location of unwanted particulate matter, or other undesired or unintended features, etc.

[0017] The tool 110 and the metrology instrument 112 may be coupled to a control system 120. The control system 120 may include a computer processing unit (CPU) 122, a graphics processing unit (GPU) 124, a field programmable gate array (FPGA) 131 (or other suitable accelerators, such as a data processing unit (DPU), RNN, artificial neural network (ANN), etc.), memory 126, a display 128, input devices 130, and a communication interface 132 (e.g., a high performance network (HPC)). The control system 120 may also include front-end circuitry, such as a transmit signal chain, a receive signal chain, switch circuitry, digital and analog circuitry, etc. The transmit signal chain may provide control signals for the tool 110. The receive signal chain may receive process parameter measurements from the metrology instrument 112.

[0018] The front-end circuitry may be coupled to and controlled by one or more processor circuits, such as a CPU 122, a GPU 124, and an FPGA 131. The CPU 122 may be implemented as one or more multi-core processors. The GPU 124 and FPGA 131 may be used to accelerate the processing of data and the performance of the machine learning network, as described herein. The techniques shown and described herein may be performed by the CPU 122 working in conjunction with the GPU 124 for faster processing.

[0019] The CPU 122 and GPU 124 may be coupled to the memory 126 for executing instructions that cause the control system 120 to perform one or more of the storage of data related to manufacturing control, processing, or inspection, or otherwise perform the techniques shown and described herein. The control system 120 may be communicatively coupled to other portions of the system 100 using, for example, a wired or wireless communication interface 132.

[0020] For example, execution of one or more techniques shown and described herein may be accomplished on control system 120 or using other processing or storage facilities, such as computing facilities 140 (e.g., servers, cloud processing systems, data warehouses, general-purpose computing devices such as laptops, tablets, smartphones, desktop computers, etc.). For example, processing tasks that would be undesirably slow or exceed the capabilities of control system 120 if performed on control system 120 may be performed remotely (e.g., on a separate system), e.g., in response to a request from control system 120. Similarly, storage of inspection data or intermediate data may be achieved using remote facilities communicatively coupled to control system 120. Control system 120 may also include a display 128, e.g., for presentation of configuration information or results, and input devices 130, e.g., including one or more of a keyboard, trackball, function keys or softkeys, a mouse interface, a touch screen, a stylus, etc., for receiving operator commands, configuration information, or responses to queries.

[0021] The control system 120 may utilize the RNN model to predict and set the process parameters used by the tool 110. The RNN model may be provided via a computing facility 140 (e.g., a cloud processing system). Process results may be measured by metrology instruments 112 to update the RNN model. The RNN model may be built using run-to-run (R2R) algorithms and software scripts. For example, the RNN model may be built offline in a computer programming script, such as Python, to predict process parameters (e.g., deposition rate) used in the manufacturing process using historical production data. During runtime, a deposition control strategy (or other processing strategy) may invoke the RNN script to trigger model predictions, which are then used to calculate process time settings for each lot that will result in the desired target layer thickness value. The actual layer thickness of this lot may later be measured using metrology instruments 112 and fed back to the control system 120 to update the RNN model.

[0022] 2 shows a framework 200 for a R2R strategy based on an RNN deep learning model. The framework 200 may include a tool 110, a metrology instrument 112, a control system 120, and an RNN model 202. During runtime, the tool 110 may provide context information about a processing step performed by the tool 110, such as deposition, to the control system 120. The control system 120 may utilize the RNN model 202 to generate process parameter predictions (e.g., deposition rate predictions), which are then used to operate the tool 110 to perform the processing step (e.g., deposition). The metrology instrument 112 may measure results of the processing step (e.g., film thickness), which may then be used to update the RNN model 202.

[0023] As described herein, building and deploying a deep learning model is not a simple and straightforward process but involves complex steps, especially in a high-mix manufacturing environment. Therefore, subject matter expertise (SME) can be used to build a robust and effective deep learning model that can be used for the R2R control described herein. This is because, to build a working model, there must be correlations and meanings that can be extracted from the dataset, as well as sufficient, preferably independent, features (or inputs to the model) to be incorporated into the model. Feature selection may include manufacturing context, equipment hardware parameters, consumable usage, upstream parametric data, and the like, which may affect the model's output values ​​(or indicators). Manufacturing context may include the tool ID used to process the lot, the tier the lot is currently on, and the final product ID (e.g., memory, CPU, etc.), since the same process may generate variability depending on different combinations of the above manufacturing contexts. Equipment hardware parameters may include maintenance information (e.g., number of runs since last maintenance), machine / tool ​​configurations such as current and voltage settings. Upstream parametric data may include data from the same lot after it has been processed upstream, such as thickness before a deposition, etching, or polishing step.

[0024] Initially, an RNN model is constructed and trained before being deployed for use during runtime. FIG. 3 illustrates a flow diagram of a method 300 for training an RNN model. At operation 302, historical data may be acquired. In some examples, the historical data may be provided by an equipment automation program. In some examples, the historical data may be stored in a database and queried for acquisition by the R2R system. The database may be updated based on runtime operations, so data may be acquired prior to the runtime operations for updates.

[0025] However, simply throwing data into a deep learning algorithm without selecting appropriate features generally does not produce worthwhile results. A large dataset, in itself, does not guarantee a good ML model. Instead, selecting data with relevant features that contain enough variability for the model to extract, as described herein, produces better results.

[0026] In operation 304, a subset of the historical data may be selected. The selected data may be based on feature selection related to process parameters associated with the RNN model. Subject matter expert knowledge (SME) may be used to prune a large dataset to selected data to build a robust and effective deep learning model that can be used for R2R control. Correlated and meaningful features within the dataset may be extracted to build a reliable and accurate model. A minimum number of features (e.g., at least two) are extracted and incorporated into the model as inputs to the model. Features may be independent of each other. For example, routes and products may be considered non-independent. A route is a list of steps performed for many products, and routes are typically dedicated to a product line. Therefore, selecting routes and products as features may provide redundant context information for the model.

[0027] Feature selection may include manufacturing context, equipment hardware parameters, consumable usage, and upstream parametric data, etc., which may affect the model's output values. Manufacturing context may include the tool ID used to process the lot, the layer the lot is currently on, and the final product ID (e.g., memory, CPU, etc.), as the same process may generate variability depending on different combinations of the above manufacturing contexts. Equipment hardware parameters may include maintenance information (e.g., number of runs since last maintenance), machine / tool ​​configurations such as current and voltage settings. Upstream parametric data may include data from the same lot after upstream processing, such as thickness before deposition, etching, or polishing steps.

[0028] Feature selection may use data analysis for selection. For example, linear or nonlinear regression techniques may be used to identify the most relevant features for a specified processing parameter predicted by the RNN model. An analysis of variance (ANOVA) may be performed to estimate the variance of the processing parameter based on various features. Features that have a direct relationship with the variance may be identified and selected.

[0029] 4 illustrates an example of selection features for a deposition rate model. Here, features 402 may include machine / tool ​​402.1, layer 402.2, and product 402.3. Output (i.e., indicator 404) may also be included. Here, observed deposition rates are included as part of the selection data for training the model. The observed deposition rates may have been measured by a metrology tool in a previous run.

[0030] 3, in operation 306, a filtering operation may be performed on the dataset. The filtering operation may remove outliers in the dataset so that they do not adversely affect the model. The filtering operation may include a Tukey filter (also referred to as a Tukey window), a Grubb's filter, a basic constraints filter, and / or other suitable filters.

[0031] Preprocessing can be performed on the dataset. Selected data can be preprocessed before being used by the RNN model. For example, in a high-mix manufacturing environment (e.g., a foundry), the same process may be run on multiple tools to build different products with multiple layers, and each different combination of these contextual attributes, e.g., machine, product, layer, etc., can produce different results for the same process.

[0032] In addition to numeric data, categorical data, such as strings, may also be fed to a machine learning model. However, most machine learning algorithms typically process only numeric values, such as data matrices. In operation 308, the string data may be converted to numeric values. In some examples, one-hot encoding may be used to convert categorical data (e.g., string data) to numeric data. FIG. 5 illustrates an example of one-hot encoding of string data. Here, a table 502 of selected features containing categorical data is converted to a matrix 504 of numeric values. For example, machine type (CVD-01, CVD-02) is coded in columns 0 and 1 of matrix 504.

[0033] Returning to FIG. 3 , in operation 310, the data may be scaled. For example, the numerical values ​​in the matrix may be scaled or normalized so that values ​​are evaluated based on percentages rather than absolute values. Better performance of the machine learning model may be obtained with scaled data, so that features are evaluated by their percentages rather than by their absolute values ​​(e.g., size). For example, upstream parametric data may include multiple values ​​of different steps (e.g., thickness before the polishing step), and these numerical values ​​may be scaled or normalized. Scaling (or normalization) may ensure that the numerical values ​​of the feature are appropriately weighted and represented by their percentages. In some examples, the minimum and maximum values ​​of the feature may be determined. Based on the minimum and maximum values, the numerical values ​​for the feature may be normalized so that they are between 0 (lowest value) and 1 (highest value).

[0034] In operation 312, a set of variables known as hyperparameters of the RNN model may be adjusted. Hyperparameters may be configured in training mode, and their values ​​may be used to control the learning process and significantly affect model performance. Hyperparameters may include the number of hidden layers in the neural network, the number of neurons in each layer, batch size, epochs, dropout, etc. Batch size refers to the size of records fed to the model at one time (e.g., 100 records at a time). Epochs refer to the number of times the same data is fed to the model. Dropout refers to the percentage of data removed at each epoch. For example, if in the first epoch, 100 records are fed to the model using 10 batches of 10 records each, and the dropout rate is 20%, then in the second epoch, 80 of the 100 records are fed back to the model in 8 batches of 10 records each. The 20 records removed between the first and second epochs may be selected using a random selection technique.

[0035] In some examples, at least a portion of the hyperparameter tuning may be performed manually by an expert, e.g., a data scientist using heuristics. In some examples, the best values ​​of the hyperparameters for a given problem may not be known, and thus the technique may use rules of thumb, copy values ​​used for other problems, and / or search for the best values ​​by trial and error (e.g., an iterative process). For example, when designing an RNN model of deposition rate for a CVD process, the hyperparameters of a previous model designed for a similar process, such as physical vapor deposition (PVD), may be used. The PVD hyperparameters may be further tuned using the techniques described above. The hyperparameter tuning may determine optimal values ​​using automated techniques, such as grid search.

[0036] After the machine learning network completes the training process, the RNN model can be used in a normal operation (runtime) mode to set process parameters during manufacturing. Figure 6 illustrates a flow diagram of a method 600 for manufacturing a semiconductor device in a normal operation mode. In some examples, the method 600 can be performed by a fabrication system 100 (e.g., using the framework 200 and the method 300) having a trained RNN model as described herein.

[0037] In operation 602, the manufacturing tool / machine may provide context information to the control system regarding the processing steps performed by the manufacturing tool / machine. In a deposition example, the manufacturing tool may provide information regarding tool identification, product identification, layer number, etc. In operation 604, the control system may input the received context information into the trained RNN model. In operation 606, the RNN model may generate predicted process parameters (e.g., deposition rate predictions).

[0038] In operation 608, the control system may set process parameters (e.g., deposition time) of the manufacturing tool based on the predicted process parameters. In operation 610, the manufacturing tool may perform a processing step (e.g., deposition) using the set process parameters. In operation 612, a metrology tool may measure results of the processing step (e.g., film thickness). Then, in operation 614, the measurement results may be fed back to the RNN model to update the model. In some examples, the values ​​of the measurement results may be filtered and preprocessed as described herein (e.g., method 500). For example, filtering operations may be performed on the measurement results to remove outliers that may be generated by erroneous execution. In some examples, a detection system may be used to detect erroneous execution, and if erroneous execution is detected, results from the erroneous execution may be discarded and not fed back to the RNN model. For example, a fault detection and classification (FDC) system may be used to detect indicators of erroneous execution. For example, an FDC system may monitor tool temperature, gas flow, etc., which may be indicators of erroneous execution.

[0039] In some examples, measurements may be fed back to the RNN in batches. Measurements from a set of product lots may be saved and then fed back to the RNN as a batch to update the model. The batch size may be the same batch size adjusted in the hyperparameter tuning operation used to train the RNN model. In some examples, measurements may be monitored, and if the quality of the predicted processed parameters begins to deteriorate, updates to the RNN model may be stopped. The RNN model may then be reset.

[0040] Figure 7 shows simulation results of a deposition controller with an RNN deep learning model compared to a conventional linear model with an exponentially weighted moving average (EWMA) tuning algorithm. Simulation results on production data show that the deposition controller enhanced with an RNN deep learning model performs comparably (32% root mean squared error (RMSE) reduction) compared to a conventional R2R strategy with a linear model and an EWMA tuning algorithm (31% RMSE reduction). Therefore, in addition to being used in high-mix production environments, machine learning solutions can provide complementary capabilities in complex, time-sensitive situations where accurate physical or statistical models are not yet ready.

[0041] Some examples of using machine learning networks above have been described to predict deposition rates for CVD processes. For clarity, the techniques described herein using machine learning networks can be used for other semiconductor manufacturing processes, such as, but not limited to, etch rates for etching processes, polish rates for CMP, offset adjustment for photolithography, plating rates for plating, etc.

[0042] The techniques shown and described herein may be performed using part or all of a fabrication system 100 as shown in Figure 1, or alternatively using a machine 800 as discussed below in connection with Figure 8. Figure 8 illustrates an example block diagram comprising a machine 800 on which any one or more of the techniques (e.g., methodologies) discussed herein may be performed. In various examples, machine 800 may operate as a standalone device or may be connected (e.g., networked) to other machines.

[0043] In a networked deployment, machine 800 may operate as a server machine, a client machine, or both in a server-client network environment. In one example, machine 800 may function as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. Machine 800 may be a personal computer (PC), a tablet device, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a web appliance, a network router, switch, or bridge, or any machine capable of executing instructions (sequential or otherwise) that specify actions to be taken by that machine. Furthermore, while only a single machine is illustrated, the term “machine” is intended to include any collection of machines individually or collectively executing a set (or sets) of instructions to perform any one or more of the methodologies discussed herein, such as cloud computing, software as a service (SaaS), other computer cluster configurations, etc.

[0044] Examples described herein may include or operate by logic or multiple components or mechanisms. Circuitry is a collection of circuits implemented in tangible entities including hardware (e.g., simple circuits, gates, logic, etc.). Circuitry membership may be flexible over time and the variability of the underlying hardware. Circuitry includes elements that, alone or in combination, can perform specified operations when in operation. In one example, the hardware of a circuitry may be invariably designed (e.g., hardwired) to perform specific operations. In one example, hardware comprising circuitry may include variably connected physical components (e.g., execution units, transistors, simple circuits, etc.) including computer-readable media that are physically altered (e.g., magnetically, electrically, etc., via a change in physical state or a transformation of another physical property) to encode instructions for specific operations. When connecting the physical components, the underlying electrical properties of the hardware components may be altered, for example, from insulating to conductive properties or vice versa. The instructions enable embedded hardware (e.g., an execution unit or a load mechanism) to create, via variable connections, members of circuitry within the hardware to perform portions of specific operations during operation. Thus, the computer-readable medium is communicatively coupled to other components of the circuitry when the device is operating. In one example, any of the physical components may be used in more than one member of more than one circuitry. For example, during operation, an execution unit may be used in a first circuit of a first circuitry at one time and reused by a second circuit within the first circuitry or by a third circuit within the second circuitry at a different time.

[0045] The machine 800 (e.g., a computer system) may include a hardware-based processor 801 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), a hardware processor core, or any combination thereof), a main memory 803, and a static memory 805, some or all of which may communicate with each other via an interlink 830 (e.g., a bus). The machine 800 may further include a display device 809, an input device 811 (e.g., an alphanumeric keyboard), and a user interface (UI) navigation device 813 (e.g., a mouse). In one example, the display device 809, the input device 811, and the UI navigation device 813 may include at least a portion of a touchscreen display. Machine 800 may additionally include a storage device 820 (e.g., a drive unit), a signal generating device 817 (e.g., a speaker), a network interface device 850, and one or more sensors 815, such as a global positioning system (GPS) sensor, a compass, an accelerometer, or other sensor. Machine 800 may include an output controller 819, such as a serial controller or interface (e.g., universal serial bus (USB)), a parallel controller or interface, or other wired or wireless (e.g., infrared (IR) controller or interface, near field communication (NFC)), coupled to communicate with or control one or more peripheral devices (e.g., a printer, a card reader, etc.).

[0046] Storage device 820 may include a machine-readable medium on which is stored one or more sets of data structures or instructions 824 (e.g., software or firmware) that embody or are utilized by any one or more of the techniques or functions described herein. The instructions 824 may also reside, completely or at least partially, within main memory 803, static memory 805, mass storage device 807, or hardware-based processor 801 during execution thereof by machine 800. In one example, one of hardware-based processor 801, main memory 803, static memory 805, or storage device 820, or any combination thereof, may constitute a machine-readable medium.

[0047] Although the machine-readable medium is considered a single medium, the term "machine-readable medium" may include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) configured to store one or more instructions 824.

[0048] The term "machine-readable medium" may include any medium capable of storing, encoding, or carrying instructions for execution by machine 800 and causing machine 800 to perform any one or more of the techniques of this disclosure, or capable of storing, encoding, or carrying data structures used by or associated with such instructions. Non-limiting examples of machine-readable media may include solid-state memory, and optical and magnetic media. Thus, a machine-readable medium is not a transitory, propagating signal. Specific examples of large-scale machine-readable media may include semiconductor memory devices (e.g., Electrically Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM)) and non-volatile memory such as flash memory devices, magnetic or other phase-change or state-change memory circuits, magnetic disks such as internal hard disks and removable disks, magneto-optical disks, and CD-ROM and DVD-ROM disks.

[0049] The instructions 824 may further be transmitted or received over the communications network 821 using a transmission medium via a network interface device 850 utilizing any one of several transport protocols (e.g., frame relay, internet protocol (IP), transmission control protocol (TCP), user datagram protocol (UDP), hypertext transfer protocol (HTTP), etc.). Exemplary communication networks may include local area networks (LANs), wide area networks (WANs), packet data networks (e.g., the Internet), mobile phone networks (e.g., cellular networks), Plain Old Telephone (POTS) networks, and wireless data networks (e.g., the Institute of Electrical and Electronics Engineers (IEEE) 802.22 family of standards known as Wi-Fi®, the IEEE 802.26 family of standards known as WiMax®), the IEEE 802.27.4 family of standards, peer-to-peer (P2P) networks, among others. In one example, network interface device 850 may include one or more physical jacks (e.g., Ethernet, coaxial, or telephone jacks) or one or more antennas for connecting to communication network 821. In one example, the network interface device 850 may include multiple antennas for wireless communication using at least one of single-input multiple-output (SIMO), multiple-input multiple-output (MIMO), or multiple-input single-output (MISO) technologies.The term "transmission medium" is intended to include any intangible medium capable of storing, encoding, or carrying instructions for execution by machine 800, including digital or analog communication signals or other intangible media to facilitate communication of such software.

[0050] Various notes Each of the above non-limiting aspects can stand on its own or can be combined in various permutations or combinations with one or more of the other aspects or other subject matter described in this document.

[0051] The above detailed description includes references to the accompanying drawings, which form a part of the detailed description. The drawings show, by way of illustration, specific implementations in which the invention may be practiced. These implementations are also generally referred to as "examples." Such examples may include elements in addition to those shown or described. However, the inventors also contemplate examples in which only the elements shown or described are provided. Furthermore, the inventors also contemplate examples that use any combination or permutation of the elements (or one or more aspects thereof) shown or described either with respect to the particular example (or one or more aspects thereof) shown or described herein or with respect to other examples (or one or more aspects thereof).

[0052] In the event of a conflicting usage between this document and any document incorporated by reference, the usage in this document shall prevail.

[0053] As used herein, the terms "a" or "an" are used as is common in patent documents to include one or more, independent of any other instance or use of "at least one" or "one or more." The term "or" is used herein to refer to non-exclusion, so that, unless otherwise stated, "A or B" includes "A but not B," "B but not A," and "A and B." As used herein, the terms "including" and "in which" are used as the plain English equivalents of the respective terms "comprising" and "wherein." Also, in the following aspects, the terms "including" and "comprising" are open-ended, i.e., systems, devices, articles, compositions, formulations, or processes that include elements in addition to the elements listed after such terms in an aspect are still considered to be within the scope of that aspect. Furthermore, in the following embodiments, terms such as "first," "second," and "third" are used merely as labels and are not intended to impose numerical requirements on those objects.

[0054] The example methods described herein may be at least partially machine-implemented or computer-implemented. Some examples may include computer-readable or machine-readable media encoded with instructions operable to configure an electronic device to perform a method as described in the above examples. Implementations of such methods may include code, such as microcode, assembly language code, higher-level language code, etc. Such code may include computer-readable instructions for performing various methods. The code may form part of a computer program product. Further, in one example, the code may be tangibly stored, for example, during execution or at other times, on one or more volatile, non-transitory, or non-volatile tangible computer-readable media. Examples of these tangible computer-readable media may include, but are not limited to, hard disks, removable magnetic disks, removable optical disks (e.g., compact disks and digital video disks), magnetic cassettes, memory cards or sticks, random access memory (RAM), read-only memory (ROM), etc.

[0055] The above description is intended to be illustrative, not limiting. For example, the above examples (or one or more aspects thereof) may be used in combination with each other. Other implementations may be used, for example, by one of ordinary skill in the art, upon reviewing the above description. The Abstract is provided to enable the reader to quickly ascertain the nature of the present technical disclosure. The Abstract is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the aspects. Also, in the above Detailed Description, various features may be grouped together to streamline the disclosure. This should not be interpreted as intending that an unclaimed disclosed feature is essential to any claim. Rather, inventive subject matter may reside in fewer than all features of a particular disclosed implementation. Thus, the following aspects are incorporated into the Detailed Description as examples or implementations, and it is contemplated that each aspect stands on its own as a separate implementation, and that such implementations can be combined with each other in various combinations or permutations.

Claims

1. 1. A method for setting at least one process parameter for manufacturing a semiconductor device, the method comprising: receiving context information regarding a first manufacturing machine; inputting the context information about the first manufacturing machine into a machine learning network; receiving a predicted value for the at least one process parameter from the machine learning network for the first manufacturing machine; setting the at least one process parameter for the first manufacturing machine based on the predicted value to manufacture the semiconductor device; and receiving context information regarding a second manufacturing machine different from the first manufacturing machine; inputting the context information about the second manufacturing machine into the machine learning network; receiving a second predicted value for a second at least one process parameter from the machine learning network for the second manufacturing machine; receiving measurements of the semiconductor device associated with the at least one processing parameter for the first manufacturing machine; receiving second measurement results of a second semiconductor device associated with the second at least one processing parameter for the second manufacturing machine; and feeding back the measurement results to the machine learning network; the machine learning network is trained based on the identified features by performing a regression technique to estimate the variance of the at least one process parameter, and identifying the features as having a direct relationship with the variance; The method, wherein the plurality of features includes at least one manufacturing context feature, at least one equipment hardware parameter feature, and at least one upstream parametric data feature.

2. The method of claim 1 , wherein the machine learning network comprises a recurrent neural network (RNN) deep learning model.

3. the machine learning network: Obtaining data associated with a manufacturing history; selecting the subset of data based on a plurality of features associated with the at least one processing parameter, the plurality of features being independent; inputting the subset into the machine learning network for training.

4. The training process comprises: The method of claim 3 , further comprising filtering the subset of data to remove outliers.

5. The training process comprises: The method of claim 3 , further comprising converting string data in the subset of data to a numeric value.

6. The training process comprises: The method of claim 3 , further comprising scaling the subset of the data to generate normalized data.

7. The training process comprises: The method of claim 3 , further comprising tuning at least one hyperparameter associated with the machine learning network.

8. 1. A control system comprising: at least one hardware processor; at least one memory that, when executed by the at least one hardware processor, causes the at least one hardware processor to: receiving context information regarding a first manufacturing machine; inputting the context information about the first manufacturing machine into a machine learning network; receiving a predicted value for at least one process parameter from the machine learning network for the first manufacturing machine; setting the at least one process parameter for the first manufacturing machine based on the predicted value to manufacture a semiconductor device; and receiving context information regarding a second manufacturing machine different from the first manufacturing machine; inputting the context information about the second manufacturing machine into the machine learning network; receiving a second predicted value for a second at least one process parameter from the machine learning network for the second manufacturing machine; receiving measurements of the semiconductor device associated with the at least one processing parameter for the first manufacturing machine; receiving second measurement results of a second semiconductor device associated with the second at least one processing parameter for the second manufacturing machine; and feeding back the measurement results to the machine learning network; the machine learning network is trained based on the identified features by performing a regression technique to estimate the variance of the at least one process parameter, and identifying the features as having a direct relationship with the variance; The control system, wherein the plurality of features includes at least one manufacturing context feature, at least one equipment hardware parameter feature, and at least one upstream parametric data feature.

9. The control system of claim 8 , wherein the machine learning network comprises a recurrent neural network (RNN) deep learning model.

10. the machine learning network: Obtaining data associated with a manufacturing history; selecting the subset of data based on a plurality of features associated with the at least one processing parameter, the plurality of features being independent; inputting the subset into the machine learning network for training.

11. The training process comprises: The control system of claim 10 further comprising filtering the subset of data to remove outliers.

12. The training process comprises: The control system of claim 10 further comprising converting string data in the subset of data to a numeric value.

13. The training process comprises: The control system of claim 10 further comprising scaling the subset of the data to generate normalized data.

14. The training process comprises: The control system of claim 10 , further comprising tuning at least one hyperparameter associated with the machine learning network.

Images