Process recipe tuning method, apparatus, device, and medium

Abstract

The application provides a process recipe debugging method, device, equipment and medium. The method comprises the following steps: obtaining a sample data set; obtaining a mechanism relationship; training a target prediction model based on the sample data set and the mechanism relationship; inputting process pre-product information corresponding to a target product and a target process result into the target prediction model to obtain a debugging process recipe; obtaining a test process result obtained by testing and processing the target product by using the debugging process recipe, and verifying whether the test process result matches the target process result; if the test process result does not match the target process result, feeding back the test process result to the model, so that the model corrects the debugging process recipe according to the test process result, and then re-tests, verifies and corrects until the test process result matches the target process result, and the debugging process recipe used when the test process result matches the target process result is taken as the target process recipe. The application can avoid the problem that the process recipe debugging process is complex and inefficient due to the reliance on manual experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor manufacturing, and in particular to a process recipe debugging method, device, equipment and medium. BACKGROUND

[0002] In recent years, with the continuous progress of semiconductor manufacturing technology, the semiconductor manufacturing process is facing increasingly stringent requirements. In the semiconductor manufacturing process, the critical dimension and in-plane uniformity of the wafer have a greater impact on the electrical performance and yield of the chip, and therefore the process must be strictly controlled.

[0003] Taking a plasma etching process as an example, different products have different requirements for etching results (such as critical dimension, etching depth, sidewall angle, etc.). Therefore, before etching a batch of products, process verification needs to be performed in advance to determine a process recipe that can meet the process requirements (target process result). Among them, the process recipe is a combination of various process parameters under the working state of the process equipment, and its accuracy directly affects the process result.

[0004] At present, engineers usually perform cross tests by continuously adjusting parameters such as chamber pressure, etching gas flow, etching gas ratio, RF power, etching temperature, and etching mode of the etching process according to experience based on process requirements, and verify the test results through measurement, SEM (scanning electron microscope), TEM (transmission electron microscope), etc. However, during the debugging process, different process parameters may interact with each other or restrict each other, resulting in complex regulation and control. Moreover, due to the diversity of product requirements and the large number of process parameters involved, complex cross tests often need to be performed, resulting in low debugging efficiency and excessive consumption of human and material resources. SUMMARY

[0005] In order to solve the problem that the existing technology relies on manual experience to debug the process recipe, resulting in a complex and inefficient debugging process, the present application provides a process recipe debugging method, device, equipment and medium based on artificial intelligence.

[0006] In a first aspect, the present application provides a process recipe debugging method, comprising:

[0007] Obtaining a sample data set, the sample data set comprising process pre-product information, an actually applied process recipe, and a reached process result corresponding to a plurality of historical sample products, the process recipe comprising process parameters required for executing a target process;

[0008] Obtaining a plurality of mechanism relationships between the process parameters and the process result;

[0009] Training a target prediction model based on the sample data set and the mechanism relationships;

[0010] inputting process-pre-product information corresponding to a target product and a target process result into the target prediction model to predict a debugging process recipe corresponding to the target product;

[0011] obtaining a test process result obtained by testing and processing the target product by using the debugging process recipe, and verifying whether the test process result matches the target process result;

[0012] when the test process result does not match the target process result, feeding back the test process result to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then re-obtaining a test process result obtained by processing the target product by using the debugging process recipe until the test process result matches the target process result, and taking the debugging process recipe used when the test process result matches the target process result as a target process recipe.

[0013] Further, the obtaining of the plurality of mechanism relationships between the process parameters and the process result comprises: obtaining a plurality of mechanism relationships by using a statistical method, a machine learning method and / or a time series analysis method based on the sample data set;

[0014] The mechanism relationship comprises an influence trend and / or an influence degree of a single process parameter on the process result, or an influence trend and / or an influence degree of a plurality of process parameters on the process result.

[0015] Further, the training of the target prediction model based on the sample data set and the mechanism relationship comprises:

[0016] continuing pre-training of an initial large model in a general field based on the mechanism relationship to obtain a field enhanced model;

[0017] training the field enhanced model based on the sample data set to obtain the target prediction model.

[0018] Further, the training of the field enhanced model based on the sample data set comprises:

[0019] dividing the sample data set to obtain a training set and a test set;

[0020] training the field enhanced model based on the training set to obtain an initial prediction model;

[0021] testing the prediction accuracy of the initial prediction model based on the test set;

[0022] when the prediction accuracy meets a preset accuracy requirement, taking the initial prediction model as a target prediction model;

[0023] When the prediction accuracy does not meet the accuracy requirement, returning to the step of dividing the sample data set.

[0024] Further, the training of the domain enhanced model based on the training set comprises:

[0025] inputting the process-pre product information and process result corresponding to the historical sample product into the domain enhanced model to obtain a predicted process recipe corresponding to the historical sample product;

[0026] According to the predicted process recipe corresponding to the historical sample product and the actually applied process recipe, a model loss is obtained.

[0027] According to the model loss, the domain enhanced model is iteratively trained to obtain the initial prediction model.

[0028] Further, the process-pre product information, historical debugging process recipe and test process result reached by the historical debugging process recipe corresponding to the target product are collected into the sample data set.

[0029] Further, the target process is a plasma etching process.

[0030] The process-pre product information includes the type and thickness of the to-be-etched film layer.

[0031] The process parameters include chamber pressure, radio frequency power, radio frequency frequency, process gas type, process gas flow and / or process temperature.

[0032] The process result includes critical dimension, etching topography and / or process performance data.

[0033] In a second aspect, the present application provides a process recipe debugging device, comprising:

[0034] a sample acquisition module, configured to acquire a sample data set, the sample data set comprising process-pre product information corresponding to a plurality of historical sample products, actually applied process recipes and reached process results, the process recipe comprising process parameters required for executing a target process;

[0035] a mechanism acquisition module, configured to acquire a plurality of mechanism relationships between the process parameters and the process results;

[0036] a model training module, configured to train a target prediction model based on the sample data set and the mechanism relationships;

[0037] a recipe prediction module, configured to input process-pre product information corresponding to a target product and a target process result into the target prediction model to predict a debugging process recipe corresponding to the target product.

[0038] a recipe verification module configured to obtain a test process result obtained by using the debugging process recipe to test process the target product, and verify that the test process result does not match the target process result;

[0039] a recipe correction module configured to, when the test process result does not match the target process result, feed back the test process result to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then re-calls the verification module until the test process result matches the target process result, and uses the debugging process recipe used when the test process result matches the target process result as the target process recipe.

[0040] The third aspect of the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the process recipe debugging method as described above when executing the computer program.

[0041] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the process recipe debugging method as described above.

[0042] By using the above technical solution, the present application has the following beneficial effects compared with the prior art:

[0043] Improve the process recipe debugging efficiency: by training the target prediction model to process the process-before product information corresponding to the target product and the target process result, the corresponding debugging prediction recipe can be quickly generated, compared with the prior art which relies on manual experience to repeatedly debug the process recipe, the present application can reduce the number of test verifications based on artificial intelligence, and significantly improve the debugging efficiency of the process recipe.

[0044] Ensure the process recipe debugging accuracy: by using the feedback loop mechanism of prediction-verification-correction, that is, after the target prediction model outputs the debugging process recipe, the target product is tested and processed using the debugging process recipe to obtain a test process result, when the test process result does not match the target process result, the test process result is fed back to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then the test and correction are performed again for the new debugging process recipe, and such iterative cycle is performed until the test process result matches the target process result, and the debugging process recipe used when the test process result matches the target process result can accurately achieve the target process result, so it can be used as the target process recipe for subsequent batch processing products.

[0045] Effectively handle multi-parameter complex coupling problems: existing technology engineers are difficult to comprehensively and quantitatively analyze the complex relationship between the mutual coordination or restriction of numerous process parameters; the present application innovatively combines sample data set (historical experience) and mechanism relationship (physical law) to train the target prediction model, so that the model can not only learn historical data, but also integrate prior understanding of the process, so as to more scientifically and reliably analyze and predict the influence of multi-process parameter coupling on process results, and then output more reasonable and accurate process recipe.

[0046] Good versatility and scalability: sample data set and mechanism relationship can cover diversified product requirements, so the present application can be widely applied to various process scenarios in the field of semiconductor manufacturing. In addition, with the continuous accumulation of debugging data, by collecting the debugging data to the sample data set, the prediction ability and application range of the model will be further expanded, with long-term application value.

[0047] Significantly reduce resource consumption and production cost: by reducing the number of trials, the development cycle can be shortened and the consumption of high-cost trial-and-error materials can be reduced, thereby the labor and material costs of the process debugging stage can be directly reduced. At the same time, by quickly and accurately determining the optimal target process recipe, the overall efficiency of the production line and the product yield are improved. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The flowchart of the process recipe debugging method of embodiment 1 of the present application;

[0049] Figure 2 The structural block diagram of the process recipe debugging device of embodiment 2 of the present application;

[0050] Figure 3 The hardware architecture diagram of the electronic device of embodiment 3 of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0052] The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting thereof. The singular forms "a," "an," and "the" as used in the present disclosure and the appended claims are intended to include plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or," as used herein, refers to and encompasses any or all possible combinations of one or more of the associated listed items.

[0053] As mentioned before, the process recipe is very important in the process production, at present, engineers usually continuously debug process parameters according to manual experience and cross test, and verify the test results according to the process requirements. However, in the debugging process, mutual coordination or mutual restriction may occur between different process parameters, resulting in complex regulation and control. Moreover, due to the diversity of product requirements and the large number of process parameters involved, complex cross tests often need to be carried out, resulting in low efficiency of debugging and excessive consumption of human and material resources.

[0054] Therefore, the inventors have found that, by using artificial intelligence technology, a target prediction model for automatically predicting process recipes can be trained based on historical sample data. As long as the trained target prediction model is input with process pre-product information corresponding to a target product and expected target process results (i.e. process requirements), the corresponding process recipe can be intelligently output, thereby avoiding the problem of complex and inefficient debugging process caused by relying on manual experience and repeated trial and error to debug process recipes.

[0055] Meanwhile, considering that pure historical sample data is used for model training, there is a lack of physical constraints in the process, and thus the trained model may produce unreasonable (i.e. not in line with the physical laws of the process) predictions. On this basis, if the model training is combined with the mechanism relationship in the process and the historical sample data, the model not only learns historical data, but also incorporates prior understanding of the process, thereby being able to more scientifically and reliably analyze and predict the influence of multi-process parameter coupling on process results, and further being able to output more reasonable and accurate process recipes.

[0056] In addition, even if the model training is combined with the mechanism relationship in the process and the historical sample data, the obtained model may not be able to predict a process recipe that fully meets the process requirements at one time, and thus a feedback loop mechanism needs to be used to test and verify the process recipe output by the model and iteratively correct it until the latest process recipe can meet the process requirements, i.e. the process recipe can be used as the target process recipe for batch production of target products.

[0057] Based on the above technical ideas, the present disclosure provides a process recipe debugging method and device based on artificial intelligence, equipment and medium, which can effectively ensure the debugging efficiency of the process recipe and the rationality and accuracy of the debugging results.

[0058] Embodiment 1

[0059] The present application provides a process recipe debugging method, as shown in the figure, which specifically comprises the following steps: Figure 1

[0060] Step S1, obtaining a sample data set.

[0061] Specifically, the sample data set comprises process pre-product information corresponding to a plurality of historical sample products, an actually applied process recipe, and a process result achieved. The historical sample products can be derived from production test instances of a wafer foundry, pre-sale test instances of a manufacturer's equipment, and / or various experimental instances, etc. The process recipe comprises a series of process parameters required for executing a target process, and the process result comprises a result achieved by executing the target process according to the corresponding process parameters.

[0062] In an implementable manner, the aforementioned target process can be, for example, a plasma etching process. Accordingly, the sample product is a plasma etching product; the process pre-product information comprises the type (such as silicon, silicon carbide, etc.) and thickness of a film layer to be etched, and can also comprise information such as substrate type and photoresist type; the process parameters comprise parameters such as chamber pressure, radio frequency power, radio frequency frequency, process gas type, process gas flow, and / or process temperature; and the process result comprises parameters such as critical dimension, etching topography, and / or process performance data.

[0063] The critical dimension comprises characteristic dimensions of an etching pattern (such as a line, a hole, or a trench), such as line width, hole diameter, etc.; the etching topography comprises parameters such as etching pattern depth, sidewall topography, and / or sidewall roughness; and the process performance data comprises parameters such as etching rate and / or in-plane uniformity.

[0064] It should be understood that the present embodiment is not limited to application in the field of plasma etching processes, and can also be applied in the fields of processes such as deposition, chemical mechanical polishing (CMP), or ion implantation. That is, the aforementioned target process can also be a process such as deposition, chemical mechanical polishing (CMP), or ion implantation, and the present embodiment does not make any specific limitation in this regard.

[0065] In an implementable manner, after obtaining the sample data set, the present step further performs data cleaning on the sample data set to remove noisy data and repeated, erroneous, or incomplete sample data, thereby improving the accuracy of subsequent model training.

[0066] Step S2, obtaining a plurality of mechanism relationships between process parameters and process results in the target process.

[0067] ​From the perspective of acquisition approach, the mechanism relationship can include known and verified physical laws in the target process (for example, derived from professional knowledge and / or experience); some mechanism relationships can also be obtained by analyzing sample data sets based on statistical methods, machine learning methods and / or time series analysis methods.

[0068] From the physical meaning, the mechanism relationship includes the influence trend and / or influence degree of a single process parameter on the process result, or the influence trend and / or influence degree of multiple process parameters on the process result (cooperating or antagonizing each other). For example, the greater the chamber pressure, the worse the in-plane uniformity; the radio frequency power and the process gas flow are positively correlated with the etching rate, and so on. Other process results, such as critical dimension, side wall angle, etc., also have mechanism relationships related thereto, which are not described one by one here.

[0069] Step S3, based on the sample data set and the mechanism relationship, a target prediction model is trained, and the training is specifically performed through the following steps S31 and S32:

[0070] Step S31, based on the aforementioned mechanism relationship, the initial large model in the general field is continuously pre-trained to enhance the knowledge of the initial large model in the target process field, and a field enhanced model is obtained;

[0071] Step S32, based on the sample data set, the aforementioned field enhanced model is trained to obtain the target prediction model.

[0072] In an implementable manner, the initial large model includes but is not limited to the Deepseek series or the GPT series large model.

[0073] In an implementable manner, the continual pre-training in step S31 means that based on the mechanism relationship in the target process field, unsupervised learning is performed on the basis of the pre-trained initial large model to adjust the parameters of the model, so that the mechanism relationship in the target process field is internalized. After the continual pre-training based on the mechanism relationship, the obtained model can use these mechanism relationships for more reasonable reasoning.

[0074] In an implementable manner, step S32 is supervised training through the following steps:

[0075] Step S321, the sample data set is divided to obtain a training set and a test set.

[0076] For example, 70% of the sample data set is divided into a training set, and the remaining 30% of the sample data set is divided into a test set, but it should be understood that the embodiment does not make any specific limitation on the division ratio of the sample data set.

[0077] Step S322: Train the aforementioned domain enhancement model based on the training set to obtain the initial prediction model.

[0078] Specifically, firstly, the pre-process product information and process results corresponding to the historical sample products are input into the aforementioned domain enhancement model to obtain the predicted process formula corresponding to the historical sample products; then, based on the predicted process formula corresponding to the historical sample products and the actual applied process formula (labeled by the actual applied process formula), the model loss is obtained; finally, based on the model loss, the aforementioned domain enhancement model is iteratively trained using a preset optimization algorithm (such as gradient descent algorithm) until the preset training termination condition is met (such as model loss convergence) to obtain the initial prediction model.

[0079] Step S323: Test the prediction accuracy of the initial prediction model based on the test set, and determine whether the prediction accuracy meets the preset accuracy requirements.

[0080] Step S324: When the prediction accuracy meets the accuracy requirement, the initial prediction model is used as the target prediction model, and step S3 ends.

[0081] Step S325: If the prediction accuracy does not meet the accuracy requirement, return to step S321 to re-divide the sample dataset and repeat the above steps until the prediction accuracy of the initial prediction model meets the preset accuracy requirement.

[0082] Steps S1 to S3 above constitute the offline data processing process, through which a target prediction model for automatically predicting process formulations can be obtained.

[0083] Step S4: Input the pre-process product information and target process results (i.e. process requirements) corresponding to the target product into the aforementioned target prediction model to predict the process formula corresponding to the target product (for easy distinction, it is referred to as the debugging process formula).

[0084] Step S5: Obtain the experimental process results obtained when processing the target product using the debugged process formula, and determine whether the experimental process results match the input target process results.

[0085] Specifically, once the target prediction model predicts the corresponding debugging process formula for the target product, the engineer uses the debugging process formula to process the target product for experimental verification and obtains the corresponding process results (referred to as the experimental process results).

[0086] This step involves obtaining the corresponding experimental process results and matching them with the target process results to determine whether the adjusted process formulation meets the process requirements.

[0087] Step S6, when the test process result matches the target process result, it means that the debug process recipe output by the model meets the process requirements, and the debug process recipe is directly used as the target process recipe for subsequent batch processing of the target product.

[0088] Step S7, when the test process result does not match the target process result, it means that the debug process recipe output by the model meets the process requirements, and the test process result is fed back to the target prediction model to correct the debug process recipe according to the test process result, thereby obtaining a new debug process recipe; then return to step S6 to reacquire the test process result obtained by processing the target product with the debug process recipe for verification and correction until the test process result matches the target process result, that is, the debug process recipe used at the time of matching is used as the target process recipe for subsequent batch processing of the target product.

[0089] In this step, after feeding back the test process result to the target prediction model, the target prediction model will combine the pre-process product information corresponding to the target product, the target process result, the test process result, the deviation between the target process result and the test process result, and the aforementioned mechanism relationship for inference analysis, and generate a corrected debug process recipe according to the analysis result, so that the process result reached by the corrected debug process recipe is closer to the target process result.

[0090] Step S8, the pre-process product information corresponding to the target product, the historical debug process recipe, and the historical test process result corresponding to the historical debug process recipe are collected as new sample data to the sample data set, so as to use these new and high-quality sample data to train the target prediction model again in the future, so that the model can continuously adapt to new equipment states and product types, and improve the accuracy of model prediction.

[0091] The above steps S4-S8 are online data processing processes, and through the prediction-verification-correction closed loop optimization process, the target process recipe meeting the process requirements is finally obtained, so the target process recipe can be used for batch production of the target product.

[0092] Through the above steps, the embodiment brings the following advantages:

[0093] Improve the process recipe debugging efficiency: by training the target prediction model to process the pre-process product information corresponding to the target product and the target process result, the corresponding debug prediction recipe can be quickly generated, compared with the existing technology which relies on manual experience to repeatedly debug the process recipe, the embodiment based on artificial intelligence can reduce the number of test verification, and significantly improve the debugging efficiency of the process recipe.

[0094] Ensure the accuracy of the process recipe debugging: By adopting a feedback loop mechanism of prediction-verification-correction, that is, after the target prediction model outputs the debugging process recipe, the target product is processed by using the debugging process recipe to obtain the test process result, when the test process result does not match the target process result, the test process result is fed back to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then the new debugging process recipe is tested and corrected again, and the iteration is repeated until the test process result matches the target process result, then the debugging process recipe used at the time of matching can accurately achieve the target process result, so it can be used as the target process recipe for subsequent batch processing products.

[0095] Effectively handle the problem of complex coupling of multiple parameters: existing technology engineers are difficult to comprehensively and quantitatively analyze the complex relationship between the mutual coordination or restriction of numerous process parameters; the embodiment innovatively combines sample data set and mechanism relationship to train the target prediction model, so that the model not only learns historical data, but also integrates prior understanding of the process, thereby more scientifically and reliably analyzing and predicting the influence of multiple process parameter coupling on the process result, and then outputting a more reasonable and more accurate process recipe.

[0096] Good versatility and scalability: The sample data set and the mechanism relationship can cover diversified product requirements, and thus can be widely applied to various process scenarios in the semiconductor manufacturing field, including but not limited to chip manufacturing, display panel manufacturing, micro-nano optical lens manufacturing, etc. In addition, with the continuous accumulation of debugging data, the prediction ability and application range of the model will be further expanded by collecting the debugging data into the sample data set, and the model has long-term application value.

[0097] Significantly reduce resource consumption and production cost: by reducing the number of tests, the development cycle can be shortened and the consumption of high-cost trial-and-error materials can be reduced, thereby reducing the labor and material costs of the process debugging stage. At the same time, by quickly and accurately determining the optimal target process recipe, the overall efficiency of the production line and the product yield are improved.

[0098] Embodiment 2

[0099] The embodiment provides a process recipe debugging device, as shown in Figure 2 The device specifically comprises:

[0100] The sample acquisition module 11 is configured to acquire a sample data set, the sample data set comprising process pre-product information, an actually applied process recipe, and a reached process result corresponding to a plurality of historical sample products, and the process recipe comprising process parameters required for executing a target process;

[0101] The mechanism acquisition module 12 is configured to acquire a plurality of mechanism relationships between process parameters and process results.

[0102] a model training module 13 configured to train a target prediction model based on the sample data set and the mechanism relationship;

[0103] a recipe prediction module 14 configured to input the pre-process product information corresponding to the target product and the target process result into the target prediction model to predict a debug process recipe corresponding to the target product;

[0104] a verification module 15 configured to obtain a test process result obtained by testing and processing the target product by using the debug process recipe, and verify that the test process result does not match the target process result;

[0105] a correction module 16 configured to, when the test process result does not match the target process result, feed back the test process result to the target prediction model, so that the target prediction model corrects the debug process recipe according to the test process result, and then re-calls the verification module until the test process result matches the target process result, and takes the debug process recipe used when the test process result matches the target process result as the target process recipe;

[0106] a data collection module 17 configured to collect the pre-process product information corresponding to the target product, the historical debug process recipe, and the test process result reached by the historical debug process recipe into the sample data set.

[0107] In an implementable manner, the mechanism acquisition module can acquire known and verified physical laws (for example, derived from professional knowledge and / or experience) in the target process as the mechanism relationship; and can acquire a plurality of mechanism relationships based on the sample data set by using statistical methods, machine learning methods, and / or time series analysis methods.

[0108] The mechanism relationship includes an influence trend and / or an influence degree of a single process parameter on a process result, or an influence trend and / or an influence degree of a plurality of process parameters on a process result.

[0109] In an implementable manner, the model training module includes:

[0110] a first training unit configured to continue pre-training an initial large model in a general field based on the mechanism relationship to obtain a field enhanced model;

[0111] a second training unit configured to train the field enhanced model based on the sample data set to obtain the target prediction model.

[0112] In an implementable manner, the second training unit is specifically configured to:

[0113] divide the sample data set to obtain a training set and a test set;

[0114] The domain enhancement model is trained based on the training set to obtain an initial prediction model. The specific process is as follows: first, the process-pre product information and process results corresponding to the historical sample products are input into the domain enhancement model to obtain the predicted process recipe corresponding to the historical sample products; then, the model loss is obtained according to the predicted process recipe corresponding to the historical sample products and the actually applied process recipe; finally, the domain enhancement model is iteratively trained according to the model loss to obtain the initial prediction model.

[0115] The prediction accuracy of the initial prediction model is tested based on the test set.

[0116] When the prediction accuracy meets the preset accuracy requirement, the initial prediction model is taken as the target prediction model.

[0117] When the prediction accuracy does not meet the accuracy requirement, the step of dividing the sample data set is returned.

[0118] In an implementable manner, the target process is a plasma etching process.

[0119] The process-pre product information includes the type and thickness of the to-be-etched film layer.

[0120] The process parameters include chamber pressure, radio frequency power, radio frequency frequency, process gas type, process gas flow, and / or process temperature.

[0121] The process results include critical dimensions, etching topography, and / or process performance data.

[0122] The process recipe debugging device of the embodiment realizes the intelligentization and automation of process recipe debugging by training a target prediction model with causal reasoning ability in combination with historical sample data and domain mechanism relationships, can accurately and quickly iteratively optimize the recipe according to experimental feedback, significantly reduces the number of debugging experiments and resource consumption, and greatly improves the process development efficiency and success rate.

[0123] Embodiment 4

[0124] The embodiment provides an electronic device which can be expressed in the form of a computing device (for example, can be a server device) and includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor can implement the steps of the process recipe debugging method provided in Embodiment 1 when executing the computer program.

[0125] Figure 3 A hardware structure schematic diagram of the embodiment is shown, as shown in Figure 3 The electronic device 30 specifically includes:

[0126] At least one processor 31, at least one memory 32, and a bus 33 for connecting different system components, including the processor 31 and the memory 32, wherein:

[0127] The bus 33 includes a data bus, an address bus, and a control bus.

[0128] The memory 32 includes a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322, and can further include a non-volatile memory, such as a read-only memory (ROM) 323.

[0129] The memory 32 also includes a program / utility 325 having a set (at least one) of program modules 324, including but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which can include an implementation of a network environment, or a combination thereof.

[0130] The processor 31 performs various function applications and data processing by running the computer program stored in the memory 32, such as the steps of the process recipe debugging method provided by the embodiment 1 of the present application.

[0131] The electronic device 30 can further communicate with one or more external devices 34 (such as a keyboard, a pointing device, etc.) via an input / output (I / O) interface 35. Also, the electronic device 30 can communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. The network adapter 36 communicates with the other modules of the electronic device 30 via the bus 33. It should be appreciated that although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 30, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID (Redundant Array of Independent Disks) systems, tape drives, and data archival storage systems, etc.

[0132] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, such division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules.

[0133] Embodiment 4

[0134] The embodiment provides a computer readable storage medium, and a computer program is stored on the computer readable storage medium. The program is executed by a processor to implement the steps of the process recipe debugging method provided by the embodiment 1.

[0135] More specifically, the readable storage medium can include, but is not limited to, a portable disc, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0136] Although the specific embodiments of the present application are described above, it should be understood by those skilled in the art that this is merely an illustration, and the protection scope of the present application is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present application, and such changes and modifications fall within the protection scope of the present application.

Claims

1. A process recipe tuning method, characterized by, The method comprises the following steps: obtaining a sample data set, the sample data set comprising process-pre product information, actual application process recipes and achieved process results of a plurality of historical sample products, the process recipe comprising process parameters required for executing a target process; obtaining a plurality of mechanism relationships between the process parameters and the process results; based on the sample data set and the mechanism relationships, training a target prediction model by the following steps: based on the mechanism relationships, continuing pre-training an initial large model in a general field to obtain a field-enhanced model, training the field-enhanced model based on the sample data set to obtain the target prediction model; inputting process-pre product information corresponding to a target product and a target process result into the target prediction model to predict a debugging process recipe corresponding to the target product; obtaining a test process result obtained by testing and processing the target product by using the debugging process recipe, and verifying whether the test process result matches the target process result; when the test process result does not match the target process result, feeding back the test process result to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then re-obtaining a test process result obtained by processing the target product by using the debugging process recipe until the test process result matches the target process result, and taking the debugging process recipe used when the test process result matches the target process result as a target process recipe.

2. The process recipe debugging method of claim 1, wherein, The method comprises the following steps: obtaining a plurality of mechanism relationships between the process parameters and the process results by using statistical methods, machine learning methods and / or time series analysis methods based on the sample data set; 3. The process recipe debugging method of claim 1, wherein, the mechanism relationship comprises an influence trend and / or an influence degree of a single process parameter on the process result, or an influence trend and / or an influence degree of a plurality of process parameters on the process result. The method comprises the following steps: dividing the sample data set to obtain a training set and a test set; training the field-enhanced model based on the training set to obtain an initial prediction model; testing the prediction accuracy of the initial prediction model based on the test set; when the prediction accuracy meets a preset accuracy requirement, taking the initial prediction model as a target prediction model; 4. The process recipe debugging method of claim 3, wherein, when the prediction accuracy does not meet the accuracy requirement, returning to the step of dividing the sample data set. The method comprises the following steps: inputting process-pre product information and process results corresponding to the historical sample products into the field-enhanced model to obtain predicted process recipes corresponding to the historical sample products; obtaining a model loss according to the predicted process recipes corresponding to the historical sample products and the actual application process recipes; iteratively training the field-enhanced model according to the model loss to obtain the initial prediction model.

5. The process recipe debugging method according to any one of claims 1-4, wherein, The process before product information corresponding to the target product, the historical debugging process recipe, and the test process result reached by the historical debugging process recipe are collected to the sample data set.

6. The process recipe setup method according to any one of claims 1-4, wherein, The target process is a plasma etching process. The process before product information includes the type and thickness of the film layer to be etched. The process parameters include chamber pressure, radio frequency power, radio frequency frequency, process gas type, process gas flow, and / or process temperature. The process result includes critical dimension, etching topography, and / or process performance data.

7. A process recipe tuning apparatus, characterized by, Comprise: A sample acquisition module is configured to acquire a sample data set, wherein the sample data set comprises process before product information corresponding to a plurality of historical sample products, an actually applied process recipe, and a reached process result, and the process recipe comprises process parameters required for executing a target process; A mechanism acquisition module is configured to acquire a plurality of mechanism relationships between the process parameters and the process result; A model training module is configured to train a target prediction model based on the sample data set and the mechanism relationships by the following steps: based on the mechanism relationships, continuing pre-training an initial large model in a general field to obtain a field enhanced model, training the field enhanced model based on the sample data set to obtain the target prediction model; A recipe prediction module is configured to input process before product information corresponding to a target product and a target process result into the target prediction model to predict a debugging process recipe corresponding to the target product; A recipe verification module is configured to acquire a test process result obtained by testing and processing the target product using the debugging process recipe, and verify that the test process result does not match the target process result; A recipe correction module is configured to, when the test process result does not match the target process result, feed back the test process result to the target prediction model, so that the target prediction model corrects the debugging process recipe according to the test process result, and then re-calls the verification module until the test process result matches the target process result, and uses the debugging process recipe matched as a target process recipe.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and capable of running on the processor, characterized in that, The processor executes the computer program to realize the process recipe debugging method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the process recipe debugging method of any one of claims 1 to 6.

Images