Machine parameter control method, equipment and storage medium

Abstract

The present disclosure provides a machine parameter control method, device, equipment and storage medium, which belongs to the field of semiconductor manufacturing technology. The method includes: obtaining the current wafer condition factor data of the current batch of wafers on the previous machine and the current machine condition factor data of the current machine when the current batch of wafers reaches the current machine; obtaining n preset machine control parameters when the current batch of wafers is produced on the current machine, where n is a positive integer greater than 1; processing the current wafer condition factor data, the current machine condition factor data and the n preset machine control parameters through a trained machine learning model to obtain n current predicted measurement results of the current batch of wafers; based on the n current predicted measurement results and the target measurement results of the current batch of wafers, determining the target machine control parameters of the current batch of wafers when produced on the current machine from the n preset machine control parameters. The present disclosure can improve the control effect.

Description

Technical Field

[0001] The present disclosure relates to the field of semiconductor manufacturing technology, and more particularly to a method and apparatus for controlling machine parameters, a computer device, and a computer-readable storage medium. Background Art

[0002] In semiconductor manufacturing, batch control (also known as run-to-run control or R2R control) is a type of feedback control used in semiconductor manufacturing. Run-to-Run (RtR, or R2R) control is an optimization control method for intermittent processes. In related technologies, R2R control uses a single linear algorithm for control. However, as semiconductor critical dimensions continue to shrink, the effectiveness of this single linear algorithm needs to be improved due to reduced measurement sampling rates and increased process complexity. Summary of the Invention

[0003] The purpose of the present disclosure is to provide a machine parameter control method, device, equipment and storage medium, which can improve the control effect.

[0004] An embodiment of the present disclosure provides a machine parameter control method, which includes: obtaining current wafer condition factor data of a current batch of wafers on a previous machine and current machine condition factor data of the current machine when the current batch of wafers reaches the current machine; obtaining n preset machine control parameters for the current batch of wafers when they are produced on the current machine, where n is a positive integer greater than 1; processing the current wafer condition factor data, the current machine condition factor data and the n preset machine control parameters through a trained machine learning model to obtain n current predicted measurement results of the current batch of wafers; and determining, based on the n current predicted measurement results and the target measurement results of the current batch of wafers, the target machine control parameters for the current batch of wafers when they are produced on the current machine from the n preset machine control parameters.

[0005] According to another aspect of the present disclosure, a machine parameter control device is provided, which includes: an acquisition module for obtaining current wafer condition factor data of a current batch of wafers on a previous machine and current machine condition factor data of the current machine when the current batch of wafers reaches the current machine; the acquisition module is also used to obtain n preset machine control parameters of the current batch of wafers when they are produced on the current machine, where n is a positive integer greater than 1; the acquisition module is also used to process the current wafer condition factor data, the current machine condition factor data and n preset machine control parameters through a trained machine learning model to obtain n current predicted measurement results of the current batch of wafers; a determination module is used to determine, from the n preset machine control parameters, the target machine control parameters of the current batch of wafers when they are produced on the current machine based on the n current predicted measurement results and the target measurement results of the current batch of wafers.

[0006] According to another aspect of the present disclosure, a computer device is provided, comprising one or more processors; and a memory configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the computer device implements the machine parameter control method in any embodiment of the present disclosure.

[0007] According to another aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the machine parameter control method in any embodiment of the present disclosure.

[0008] According to another aspect of the present disclosure, a computer program product is provided. When the computer program is executed by a processor, the machine parameter control method in any embodiment of the present disclosure is implemented.

[0009] The machine parameter control method, device, equipment and storage medium provided in some embodiments of the present disclosure obtain the current wafer condition factor data of the current batch of wafers on the previous machine, the current machine condition factor data of the current machine when the current batch of wafers reaches the current machine, and n preset machine control parameters of the current batch of wafers when being produced on the current machine, where n is a positive integer greater than 1, and process the current wafer condition factor data, the current machine condition factor data and the n preset machine control parameters through a trained machine learning model to obtain n current predicted measurement results of the current batch of wafers, and then determine the target machine control parameters of the current batch of wafers when being produced on the current machine from the n preset machine control parameters based on the n current predicted measurement results and the target measurement results of the current batch of wafers, thereby improving the control effect and producing high-quality wafer products. In addition, the present disclosure does not need to rely on the experience of engineers to control the machine, saving the time required for control. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic structural diagram of a wafer provided by an embodiment of the present disclosure is shown.

[0011] Figure 2 The R2R process control flow chart provided in the related art is shown.

[0012] Figure 3 Schematic diagram of the etching process of controlling film thickness by the R2R system in related technology.

[0013] Figure 4 Schematic diagram of the regression equation in related technology.

[0014] Figure 5 Schematic diagram of the machine control parameter regulation process in related technology.

[0015] Figure 6 The machine parameter control system provided by the embodiment of the present disclosure is shown.

[0016] Figure 7 A flow chart of a machine parameter control method according to an embodiment of the present disclosure is shown.

[0017] Figure 8 A schematic diagram of establishing a machine learning model in one embodiment of the present disclosure is shown.

[0018] Figure 9 A schematic diagram of a machine parameter control process in an embodiment of the present disclosure is shown.

[0019] Figure 10 A flow chart of a machine parameter control method in another embodiment of the present disclosure is shown.

[0020] Figure 11 A schematic diagram showing key data indicators corresponding to wafers provided by an R2R system according to an embodiment of the present disclosure is shown.

[0021] Figure 12 A schematic diagram showing relevant information of FDC (Fault Detection and Classification) and an MM reaction chamber according to an embodiment of the present disclosure.

[0022] Figure 13 A schematic diagram of AI data flow according to an embodiment of the present disclosure is shown.

[0023] Figure 14 A schematic diagram of an autoencoding model according to an embodiment of the present disclosure is shown.

[0024] Figure 15 The LightGBM model architecture of an embodiment of the present disclosure is shown.

[0025] Figure 16The structure of the CART tree according to an embodiment of the present disclosure is shown.

[0026] Figure 17 The trained machine learning model architecture according to an embodiment of the present disclosure is shown.

[0027] Figure 18 A schematic diagram illustrating experimental results of measuring the final CD result according to an embodiment of the present disclosure.

[0028] Figure 19 A schematic diagram of a machine parameter control device in an embodiment of the present disclosure is shown.

[0029] Figure 20 A schematic structural diagram of a computer device in an embodiment of the present disclosure is shown.

[0030] Figure 21 A schematic diagram of a computer-readable storage medium in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0031] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0032] The semiconductor manufacturing process can be divided into the front-end process and the back-end process. Specifically, the front-end process is the process for the entire wafer (Wafer), which may include wafer manufacturing and wafer testing (the test object is each Die (crystal) in the entire wafer, the purpose is to ensure that each Die in the entire wafer can basically meet the device characteristics or design specifications, which may include voltage, current, timing and function verification). Among them, Figure 1 FIG. 1 shows a schematic structural diagram of a wafer provided by an embodiment of the present disclosure. Figure 1 As shown, a wafer 10 may include multiple dies 11. The back-end process is the various processes that are carried out on the individual chips after they are packaged by the wafer dicing sheet, which may include packaging, final testing, and finished product storage.

[0033] Semiconductor manufacturing is inseparable from machine processing. Machines are equipment for processing or testing wafer products. When processing wafer products, the machine control parameters need to be adjusted. The machine control parameters determine the quality of the wafer products. Good machine control parameters are conducive to controlling the machine to produce high-quality wafer products.

[0034] In a related technology, R2R estimates the production recipe parameters (i.e., machine control parameters, which refer to the parameters configured during wafer production, such as oxygen concentration, production time, etc.) using a single linear algorithm. Currently, the most widely used method is to use a linear regression model to obtain the machine control parameters. Figure 2 As shown, the lot (the same batch of wafers) pre-measurement data (i.e. wafer pre-measurement data, excluding the front-station FDC, where the front-station FDC refers to the wafer front-station sensor record) obtained through pre-measurement, and the historical feedback measurement data (i.e. wafer measurement results) obtained through post-measurement, the R2R system performs linear calculations on the wafer pre-measurement data and wafer measurement results to obtain the machine control parameters. The machine operation is controlled by the machine control parameters. For example, as Figure 3 As shown, in an application scenario where the R2R system controls the film thickness by controlling the time parameter (a machine control parameter), the exposure equipment etches the film layer 302 on the wafer 301 according to the controlled time parameter, and removes the photoresist (i.e., removes the photoresist 303) after the etching is completed.

[0035] The basis of linear regression model application is to approximate the data corresponding to the input variables and output variables in the physical or chemical reactions of each process step into a linear function equation. Figure 4 As shown in the figure, the horizontal axis is the machine control parameters, and the vertical axis is the measurement results. By fitting the machine control parameters and the measurement results, the linear function equations h1(x) and h2(x) are obtained. Figure 5 As shown, the R2R system adjusts the machine control parameters and configures the machine control parameters between the upper control line (UCL) and the lower control line (LCL) to generate wafer products with better quality. Figure 5 It can be seen that when the R2R system in the relevant technology is used to regulate the machine control parameters, many wafer measurement results exceed the upper and lower limits, indicating that the regulation effect is poor.

[0036] Through research, the inventors discovered that as semiconductor critical dimensions continue to shrink, the effectiveness of existing single-site linear control systems needs to be improved as measurement sampling rates decrease and process complexity increases. Furthermore, in actual control, the interrelationships between individual processes and the influence of internal interference factors within the tool make single-site measurement feedback mechanisms inadequate for meeting process requirements.

[0037] In another related technology, engineers rely on experience to adjust tool control parameters. They study each process station and tool individually to develop a calculation formula. However, the production process for semiconductor products like dynamic random access memory (DRAM) involves numerous process stations, typically over 1,000, making manual processing time-consuming.

[0038] Based on this, the embodiments of the present disclosure provide a machine parameter control method, device, equipment and storage medium, which can be applied to semiconductor manufacturing scenarios. For example, it can be applied to specific scenarios of wafer etching, or it can also be wafer lithography and film forming processes, which are not specifically limited. In the embodiments of the present disclosure, the present disclosure obtains n current predicted measurement results by using the current wafer condition factor data of the previous machine, the current machine condition factor data of the current machine when the current batch of wafers reaches the current machine, and the n preset machine control parameters (n is a positive integer greater than 1) of the current batch of wafers when they are produced on the current machine. According to the n current predicted measurement results and the target measurement results of the current batch of wafers, the optimal control value is determined from the n preset machine control parameters as the target machine control parameter of the current batch of wafers when they are produced on the current machine. The current batch of wafers is controlled by the target machine control parameter when they are produced on the current machine, which can improve the control effect and improve the control efficiency.

[0039] In order to facilitate an overall understanding of the technical solution provided by the embodiment of the present disclosure, the machine parameter control system provided by the embodiment of the present disclosure will be described below.

[0040] like Figure 6As shown, the condition factors (including current wafer condition factor data and current machine condition factor data) and n preset machine control parameters (e.g., preset machine control parameter 1, preset machine control parameter 2, ..., preset machine control parameter n) are input into the machine parameter control system 60 to obtain n current predicted measurement results for the current batch of wafers (e.g., current predicted measurement result 1, current predicted measurement result 2, ..., current predicted measurement result n). The target measurement results of the current batch of wafers are input into the machine parameter control system 60, and the n current predicted measurement results are compared with the target measurement results of the current batch of wafers. The preset machine control parameters corresponding to the current predicted measurement results that are closest to the target measurement results are the derived optimal parameters and serve as the target machine control parameters. For example, assuming that among the n current predicted measurement results, the difference between the current predicted measurement result 2 and the target measurement result is the smallest, then the preset machine control parameter 2 corresponding to the current predicted measurement result 2 is determined to be the target machine control parameter. The machine parameter control system 60 can be deployed with a trained machine learning model. This trained machine learning model processes current wafer condition factor data, current machine condition factor data, and n preset machine control parameters to obtain n current predicted measurement results for the current batch of wafers. The machine parameter control system 60 controls the current machine to produce the current batch of wafers based on the target machine control parameters.

[0041] It should be noted that the machine parameter control system 60 may be the R2R system provided in the embodiment of the present disclosure, or may be other control systems that can achieve the above functions.

[0042] First, embodiments of the present disclosure provide a method for controlling machine parameters. This method can be performed by any electronic device equipped with a machine parameter control system. For example, the method can be performed by semiconductor manufacturing equipment such as an exposure machine, or by other processing equipment communicatively connected to the semiconductor manufacturing equipment, without specific limitation.

[0043] Figure 7 A flow chart of a machine parameter control method according to an embodiment of the present disclosure is shown. Figure 7 As shown, the machine parameter control method provided in the embodiment of the present disclosure may include the following S701 to S704.

[0044] S701 , obtaining the current wafer condition factor data of the current batch of wafers on the previous tool and the current tool condition factor data of the current tool when the current batch of wafers arrives at the current tool.

[0045] In the embodiments of the present disclosure, the current batch of wafers refers to a batch of wafers that are currently being produced in the process. The preceding machine is at least one machine used by the current batch of wafers to perform a preceding process before the current process. The current machine is the machine used by the current batch of wafers to perform the current process. For example, assuming that each machine is numbered in sequence according to the process sequence of the current batch of wafers, the current machine is the t-th machine, and the preceding machine may include at least one of the 1st to t-1th machines, where t is a positive integer greater than 1. In the following examples, the preceding machines include the 1st to t-1th machines for example, but the present disclosure is not limited to this.

[0046] In an embodiment of the present disclosure, the current wafer condition factor data of the prior machine may include any relevant information of the current batch of wafers when the wafers are on the prior machine. For example, the current wafer condition factor data may include at least one of the wafer measurement results of the current batch of wafers after being manufactured by the prior machine, the process waiting time of the current batch of wafers between adjacent prior machines, and the prior machine condition factor data of the current batch of wafers before, during, and after production using the prior machine. Among them, the prior machine condition factor data refers to any information related to the prior machine, for example, the prior machine condition factor data may include at least one of the model, manufacturer, maintenance cycle, and measurement statistics within a predetermined time period of the prior machine. The predetermined time period may be a period of time closest to the current time, such as the last 7 days or the last month, and the present disclosure does not limit this. The measurement statistics within the predetermined time period may, for example, include the average, maximum, minimum, etc. of the line width and / or film thickness of the wafers generated using the machine in the last 7 days or the last month. For example, assuming that the current machine is the t-th machine, the current wafer condition factor data of the previous machine in the present disclosure may include the wafer measurement results (e.g., line width, film thickness, etc.) of the current batch of wafers after being processed by the 1st to t-1th machines in sequence, the process waiting time (the process waiting time from the previous previous machine to the next previous machine, and / or the process waiting time from the previous machine to the current machine), and the previous machine condition factor data of the 1st to t-1th machines. In some embodiments, the previous machine condition factor data may also include environmental information monitored by sensors in the FDC of the 1st to t-1th machines (e.g., at least one of temperature, humidity, chemical concentration, etc., but the present disclosure is not limited thereto). The previous machine condition factor data of the 1st to t-1th machines can also be used to reflect the condition of the current batch of wafers, that is, the embodiment of the present disclosure takes into account the impact of all previous machines before the t-th machine on the current machine.

[0047] In the embodiment of the present disclosure, the current machine status factor data of the current machine when the current batch of wafers arrives at the current machine may include any relevant information reflecting the status of the current machine when the current batch of wafers arrives at the current machine. The current machine status factor data can be used to evaluate the recent working conditions, losses and accuracy of the current machine. Assuming that the current machine is the t-th machine, the current machine status factor refers to the status factor of the t-th machine. The current machine status factor may include at least one of the model, manufacturer, maintenance cycle, and measurement statistics within a predetermined time period of the current machine. The measurement statistics within a predetermined time period may include, for example, the mean, maximum, minimum, etc. of the line width and / or film thickness of the wafer generated by the current machine using the current machine in the past 7 days or the past month. The current machine status factor takes into account the influence of the internal interference factors of the t-th machine itself.

[0048] S702 , obtaining n preset machine control parameters for a current batch of wafers produced on a current machine, where n is a positive integer greater than 1.

[0049] In the embodiment of the present disclosure, the preset machine control parameters of the current batch of wafers when the current machine is producing refer to the parameters configured to control the current machine to produce the current batch of wafers. As to what the preset machine control parameters are specifically, they can be determined based on the current batch of wafers and / or the current machine, that is, when the current batch of wafers and / or the current machine are different, the content of the preset machine control parameters can be adaptively set. In the embodiment of the present disclosure, the preset machine control parameters can be parameters arbitrarily set between the upper and lower limits of the machine control parameters of the current machine. The upper and lower limits of the machine control parameters of the current machine include the upper limit of the machine control parameters and the lower limit of the machine control parameters. The maximum value of the preset machine control parameters of the current machine cannot exceed the upper limit of the machine control parameters, and the minimum value cannot exceed the lower limit of the machine control parameters. In some embodiments, the preset machine control parameters can also be set based on experience and process requirements.

[0050] Exemplarily, obtaining n preset tool control parameters for a current batch of wafers when manufactured on a current tool may include: obtaining upper and lower limits for the tool control parameters for the current batch of wafers when manufactured on the current tool; and uniformly generating n preset tool control parameters within the upper and lower limits using an arithmetic progression. For example, assuming n = 6, the upper limit of the tool control parameter is 50, the lower limit of the tool control parameter is 45, and the difference between two adjacent preset tool control parameters is 1. The six preset tool control parameters uniformly generated using the arithmetic progression are 45, 46, 47, 48, 49, and 50, respectively.

[0051] S703, processing the current wafer condition factor data, the current machine condition factor data and n preset machine control parameters through the trained machine learning model to obtain n current predicted measurement results of the current batch of wafers.

[0052] The machine learning model is a pre-trained machine learning model. Current wafer condition factor data, current tool condition factor data, and n preset tool control parameters are input into the pre-trained machine learning model, and the machine learning model outputs n current predicted measurement results. In the embodiments of the present disclosure, the machine learning model can be any artificial intelligence (AI) model, and the present disclosure does not limit the specific nature of the machine learning model.

[0053] S704 , determining target tool control parameters for the current batch of wafers when manufactured on the current tool from n preset tool control parameters based on the n current predicted measurement results and the target measurement results of the current batch of wafers.

[0054] In the disclosed embodiment, a current predicted measurement result closest to a target measurement result of a current batch of wafers can be determined from a plurality of current predicted measurement results, and the determined closest current predicted measurement result can be used as a target tool control parameter.

[0055] The machine parameter control method provided by the disclosed embodiments, when determining target machine control parameters, firstly, takes into account the interrelationships between various processes by using the current wafer condition factor data obtained from the previous machine; secondly, it also takes into account the interference factors within the current machine through the current machine condition factor data, thereby improving the accuracy of the determined target machine control parameters, thereby enhancing the control effect and producing high-quality wafer products. Furthermore, there is no need to rely on the engineer's experience to control the machine, which saves the time required for control and improves the control efficiency.

[0056] In an exemplary embodiment, the trained machine learning model is used to process the current wafer condition factor data, the current tool condition factor data, and n preset tool control parameters to obtain n current predicted measurement results for the current batch of wafers. This may include: combining the i-th preset tool control parameter with the current wafer condition factor data and the current tool condition factor data into an i-th input vector, where i is a positive integer greater than or equal to 1 and less than or equal to n; and inputting the i-th input vector into the machine learning model to obtain the i-th current predicted measurement result for the current batch of wafers.

[0057] like Figure 8As shown, the embodiment of the present disclosure establishes a machine learning model based on the wafer dimension and the machine dimension through a machine learning model, so as to achieve the simulation of the current predicted measurement results. Among them, the wafer dimension may include the current wafer condition factor data (which includes the front-end process measurement results, that is, the wafer measurement results of the current batch of wafers after being processed by the previous machine); the machine dimension may include the current machine condition factor data. The current wafer condition factor data and the current machine condition factor data can respectively include data of multiple dimensions and multiple factors. Therefore, the machine learning model provided by the embodiment of the present disclosure is a high-dimensional multi-factor influence model. The formula of the machine learning model is shown in the following formula 1:

[0058]

[0059] in is a machine learning model, (x,y,z i ) is the i-th input vector composed of the i-th preset machine control parameter, the current wafer condition factor data, and the current machine condition factor data, x is the current wafer condition factor data of the previous machine, y is the current machine condition factor data, z i is the i-th preset machine control parameter, where i is a positive integer greater than or equal to 1 and less than or equal to n.

[0060] For example, the i-th input vector is input into the machine learning model, and the i-th current predicted measurement result of the current batch of wafers is obtained by the following formula 2:

[0061]

[0062] in, is the i-th current prediction measurement result.

[0063] The n current prediction measurement results can be obtained by formula 2, that is, the set of n current prediction measurement results is

[0064] In the disclosed embodiments, each preset machine control parameter may include one or more machine control parameters. The number of upper and lower limits for the machine control parameters is determined by the number of machine control parameters included in each preset machine control parameter. For example, assuming that each preset machine control parameter includes two machine control parameters, namely oxygen concentration and production time, the upper and lower limits for the oxygen concentration and the upper and lower limits for the production time are set accordingly. When n preset machine control parameters are uniformly generated using an arithmetic progression, n preset oxygen concentrations within the upper and lower limits of the oxygen concentration and n preset production times within the upper and lower limits of the production time are generated respectively.

[0065] The machine learning model simulates the physical / chemical reaction conditions within the actual machine, i.e., the prediction process using the trained machine learning model. The simulation results (i.e., the current predicted measurement results) are then used to determine the optimal machine parameters (i.e., target machine control parameters). The disclosed embodiments can perform machine manufacturing simulations by varying z when x and y are known, and obtain the optimal machine parameter (target machine control parameter) configuration by observing the current predicted measurement results. The disclosed embodiments are not limited to this.

[0066] For example, the disclosed embodiments can provide effective optimal machine control parameters (target machine control parameters) under different wafer conditions, that is, the current wafer condition factor data of the previous machine can be changed, the current machine condition factor data can be fixed, and the optimal machine control parameters (target machine control parameters) under different wafer conditions can be obtained.

[0067] For example, the disclosed embodiments can provide effective optimal control parameters for the machine under different machine conditions, that is, the current machine condition factor data can be changed, the current wafer condition factor data of the previous machine can be fixed, and the optimal control parameters for the machine under different wafer conditions can be obtained.

[0068] The embodiments of the present disclosure can be universally extended across different processes, which can increase the applicability of the present disclosure. In addition, the embodiments of the present disclosure do not require excessive manpower and professional domain knowledge, and are implemented at low cost and high speed.

[0069] In an exemplary embodiment, based on n current predicted measurement results and target measurement results of the current batch of wafers, determining the target tool control parameters of the current batch of wafers when manufactured on the current tool from n preset tool control parameters can include: calculating the absolute difference between each current predicted measurement result and the target measurement result in the n current predicted measurement results; and determining the preset tool control parameter corresponding to the current predicted measurement result with the smallest absolute difference as the target tool control parameter.

[0070] For example, the current predicted measurement result with the smallest absolute difference is found by the following formula 3:

[0071]

[0072] in, is the i-th current prediction measurement result, Target MET t is the target measurement result, Abs() is for finding the absolute difference, and Min() is for finding the minimum value.

[0073] The absolute difference between the current predicted measurement result and the target measurement result is used to indicate the similarity between the current predicted measurement result and the target measurement result. If the absolute difference is closer to 0, it means that the current predicted measurement result is closer to the target measurement result (the consistency is better). On the contrary, if the absolute difference is larger, it means that the current predicted measurement result is less similar to the target measurement result (the consistency is worse).

[0074] The embodiment of the present disclosure determines the current predicted measurement result that is closest to the target measurement result among multiple current predicted measurement results, and determines the preset machine control parameters corresponding to the closest current predicted measurement result as the target machine control parameters. The method provided by the embodiment of the present disclosure has a good control effect, and the current machine can produce high-quality wafer products by configuring the target machine control parameters. In addition, the embodiment of the present disclosure simulates wafer manufacturing (i.e., obtains the current predicted measurement result through a machine learning model), which can improve the credibility of the machine learning model.

[0075] In an exemplary embodiment, the machine parameter control method may further include the following steps A1 to A5.

[0076] Step A1: obtaining batch control parameter values ​​of a current batch of wafers produced on a current machine.

[0077] Among them, the batch control parameter value is the R2R calculated value of the current batch of wafers when produced on the current machine, that is, Figure 2 The R2R system shown obtains the tool control parameters of the current batch of wafers when being manufactured on the current tool.

[0078] Step A2: obtaining a floating threshold value of a batch control parameter value.

[0079] The floating threshold can be set according to experience and process requirements, for example, it can be 0.5, but this disclosure does not limit this.

[0080] Step A3: Determine the batch control range according to the batch control parameter value and the floating threshold.

[0081] In the disclosed embodiment, the upper limit of the batch control range can be the sum of the batch control parameter value and the floating threshold, and the lower limit of the batch control range can be the difference between the batch control parameter value and the floating threshold. For example, if the batch control parameter value is 10 and the floating threshold is 0.5, the batch control range is 9.5-10.5.

[0082] Step A4: Determine whether the target machine control parameter is within the batch control range.

[0083] Determine whether the target machine control parameters determined by the machine learning model are within the batch control range. That is, the embodiment of the present disclosure can avoid large differences in AI values ​​(i.e., target machine control parameters obtained using the machine learning model) by using the card control logic of the original R2R system (i.e., the batch control range determined based on the R2R calculated value).

[0084] In step A5 , if the target tool control parameters are within the batch control range, the target tool control parameters are transmitted to the production system, so that the production system controls the current tool to manufacture the current batch of wafers according to the target tool control parameters.

[0085] In the disclosed embodiment, if the target machine control parameter is within the batch control range, it can be determined that the difference between the AI ​​value and the R2R calculated value is small, and the target machine control parameter can be transmitted to the production and manufacturing system so that the production and manufacturing system controls the current machine to produce the current batch of wafers according to the target machine control parameter. If the target machine control parameter is not within the batch control range, it can be determined that the difference between the AI ​​value and the R2R calculated value is large. In this case, the R2R calculated value can be used to transmit to the production and manufacturing system, and the production and manufacturing system can control the current machine to produce the current batch of wafers according to the R2R calculated value. Alternatively, the machine learning model can be used to re-determine new target machine control parameters.

[0086] The disclosed embodiment determines whether the target machine control parameters are within the batch control range. If so, the target machine control parameters are transmitted to the production manufacturing system, so that the production manufacturing system controls the current machine to produce the current batch of wafers according to the target machine control parameters. This allows the use of the original R2R card control logic to avoid large differences in AI values ​​and achieve better control effects.

[0087] In an exemplary embodiment, determining whether the target machine control parameter is within the batch control range may include: determining whether the target machine control parameter is within a preset reasonable range; if the target machine control parameter is within the preset reasonable range, obtaining the previous machine control parameter of the previous batch of wafers of the current batch of wafers when produced on the current machine; determining whether the adjustment amplitude of the target machine control parameter compared to the previous machine control parameter is greater than an amplitude threshold; if the adjustment amplitude is less than or equal to the amplitude threshold, determining whether the target machine control parameter is within the batch control range.

[0088] In an embodiment of the present disclosure, the preset reasonable range may be a preset range configured in advance for the current machine, that is, the current machine needs to be within the preset reasonable range to work normally. The preset reasonable range can be set based on experience and process requirements, and the present disclosure is not limited thereto. The previous batch of wafers of the current batch of wafers is a batch of wafers that are processed by the current machine and are located before the current batch of wafers. The method for determining the control parameters of the previous machine of the previous batch of wafers may be the R2R calculated value of the previous batch of wafers when the current wafers were produced, or it may be the target machine control parameters of the previous batch of wafers when the current wafers were produced, which are determined by the aforementioned trained machine learning model. The present disclosure does not limit this.

[0089] In the embodiment of the present disclosure, the amplitude threshold can be set according to actual needs, and the present disclosure does not limit its specific value.

[0090] For example, assuming that the machine learning model is an AI model, such as Figure 9 As shown, the machine parameter control process may include the following S901 to S908.

[0091] S901, obtaining an R2R calculated value through the R2R system.

[0092] S902: Determine whether to enable the AI ​​model, that is, whether to use the target machine control parameters determined by the AI ​​model to replace the R2R calculated values. Whether to enable this function can be determined based on whether the AI ​​model's training accuracy reaches a threshold. If so, the AI ​​model is enabled and S903 is executed; if not, the AI ​​model is disabled and S904 is executed.

[0093] S903: When it is determined to start the AI ​​model, the AI ​​computing module (which includes a pre-trained AI model) is called through an electronic device (which can provide web services), and the reservation lot information (i.e., the current wafer condition factor data, the current machine condition factor data, and n current predicted measurement results) is input into the AI ​​computing module to calculate and predict the optimal machine parameters (target machine control parameters, i.e., the current lot value in the figure), and then execute S905.

[0094] S904, when it is determined that the AI ​​model is not to be turned on, R2R will reply to MM (i.e. MES (Manufacturing Execution System)) as the unloading value. MES will pass the unloading value to EAP (Equipment Automation Program) and replace the unloading value with the recipe corresponding to the current machine through the relevant unloading instructions to perform the current Run behavior; when AI control is turned on, the R2R system will actively request AI to obtain the unloading value corresponding to the current Lot (batch or batch) of wafers (called AI unloading value or AI calculated value) when calculating the unloading value, and perform basic checks on the AI ​​unloading value (for example: the single adjustment range of the card control and the unloading value range), and reply to the MES system.

[0095] At step S905, after the AI ​​model is enabled and the AI ​​calculation module is called to obtain the current AI batch order value, the current AI batch order value is used as the AI ​​calculated value (i.e., the target machine control parameters determined by the AI ​​model). The AI ​​return value key parameters are checked to see if they meet the requirements. The AI ​​return value key parameters are the optimal machine parameters (target machine control parameters) predicted by the AI ​​model. This check can be performed to see if the AI ​​return value key parameters are within a preset reasonable range. If so, the process proceeds to step S906. If not, the process returns to step S901 above and returns the R2R calculated value to the MM as the order value.

[0096] S906, when the AI ​​return value key parameters are within the preset reasonable range, continue to check the single adjustment range. Checking the single adjustment range means that the adjustment range of the machine control parameters of two adjacent batches of wafers (the previous batch of wafers and the current batch of wafers of the current batch of wafers) should not exceed a reasonable range, that is, the adjustment range will not be too large. In other words, whether the adjustment range of the target machine control parameter compared with the previous machine control parameter is less than or equal to the amplitude threshold. If it is less than or equal to the amplitude threshold, execute S907. If it is greater than the amplitude threshold, return to the above S901 and use the R2R calculated value as the MM order value.

[0097] S907: When the adjustment amplitude of the target machine control parameter compared to the previous machine control parameter is less than or equal to the amplitude threshold, continue checking the tuning spec. Checking the tuning spec means that the optimal machine parameter predicted by the AI ​​model should be within the reasonable threshold range of the R2R calculated value (i.e., the batch control range mentioned above), for example, within the range of plus or minus 0.5 of the R2R calculated value (i.e., after the AI ​​calculation module calculates the value, the original R2R card control logic is used to avoid large differences in the AI ​​value). If it meets the requirements, execute S908. If it does not meet the requirements, return to S901 above and use the R2R calculated value as the MM order value.

[0098] S908, use the AI ​​shipment value (i.e. the aforementioned current AI batch shipment value) or the AI ​​calculated value as the MM shipment value in S904.

[0099] It should be noted that the embodiment of the present disclosure establishes a model by tool (By layer), and each tool completes one layer of the semiconductor device.

[0100] For example, the AI ​​model can be installed as a plug-in in the R2R system. The AI ​​model in the disclosed embodiments can be installed as a plug-in in the R2R system, achieving lightweight, low-cost, and high reliability. This also enables low-cost transformation and rapid rollout. Furthermore, by calibrating the AI ​​calculation results, reliability can be effectively improved, ensuring stable production line operation.

[0101] The following describes how to train a machine learning model.

[0102] In an exemplary embodiment, the machine control parameter adjustment method may further include the following steps S1001 to S1006 .

[0103] S1001, obtaining historical wafer condition factor data of a historical batch of wafers on a previous machine.

[0104] In the embodiment of the present disclosure, the historical batch of wafers is relative to the current batch of wafers, that is, it may include any batch of wafers manufactured on the previous machine before the current batch of wafers. The historical wafer status factor data may include any relevant information of the historical batch of wafers when it was on the previous machine. For example, the historical wafer status factor data may include at least one of the wafer measurement results of the historical batch of wafers after being manufactured by the previous machine (which may be referred to as historical measurement information), the process waiting time of the historical batch of wafers between adjacent previous machines, and the status factor data of the previous machine before, during, and after the production of the historical batch of wafers using the previous machine (the process waiting time and the status factor data of the previous machine may also be referred to as historical shipping information).

[0105] The embodiment of the present disclosure establishes a general data integration framework by collecting wafer and tool condition factor data. The general data integration framework includes Figure 11 The key data indicators shown and Figure 12 The machine-related information shown.

[0106] like Figure 11As shown, the key data indicators corresponding to each wafer provided by the R2R system may include shipping information and measurement information. For the current batch of wafers, this is called current shipping information and current measurement information; for historical batches of wafers, this is called historical shipping information and historical measurement information. Among them, the shipping information may include historical shipping information for the 1st to tth machines, for example, it may include process site, LotID (i.e., batch identification of the same batch of wafers, ID is short for Identity) and wafer ID (Wafer ID) and recipe shipping value (which can be the target machine control parameters of the historical batch of wafers on the previous machine obtained by using the AI ​​model prediction, or the R2R calculated value of the historical batch of wafers on the previous machine obtained by using the R2R system). The measurement information includes historical measurement information for the 1st to tth tools, and may include, for example, at least one of the following: the measurement site number, the measurement site name, the measurement target value (i.e., the target measurement result for the historical batch of wafers), and the measurement average value (i.e., the average of the wafer measurement results for multiple historical batches of wafers on the previous tool). The historical unloading information and historical measurement information for the 1st to t-1th tools may serve as historical wafer condition factor data for the previous tool in the training sample. The unloading value of the recipe for the tth tool, i.e., the current tool, serves as the historical tool control parameter for the historical batch of wafers produced on the current tool in the training sample, and the historical measurement results of the historical batch of wafers after production on the tth tool serve as the label for the training sample.

[0107] like Figure 12 As shown, the production maintenance (PM) information of a tool may include the tool, reaction chamber, PM time, and PM items. Figure 12 The MM in this context refers to the Manufacturing Execution System (MES). The MM system stores all PM information, with some stored in the FDC system. The front-end FDC can include at least one or more of the following: machine temperature, humidity, and chemical concentration. Electronic equipment can store PM information in the R2R system.

[0108] It should be noted that the universal data integration framework can also be referred to as a universal data table. Each piece of data in the universal data table is a record in the universal data table, and each test site needs to record these records. The disclosed embodiments utilize machine learning to automatically filter and model the records in the universal data table, eliminating the need for manual intervention and achieving the goal of adapting to different process sites. The disclosed embodiments can be a universal control method that can create a corresponding model for each test site to predict the optimal machine parameters for that site.

[0109] The disclosed embodiments can also be applied in Advanced Process Control (APC) to realize the automated determination and correction of overlay error values. For example, in the production process of semiconductor components, defects at the edge of the wafer (that is, the crystal edge) have a great impact on the process and product yield. In the related art, the image of the crystal edge taken by the measuring machine after the process treatment (such as after photolithography treatment, etching treatment, etc.) is usually used as the measurement image, and manual observation is performed based on the measured measurement image to determine whether there are defects in the crystal edge. However, defect detection through manual observation has the problem of high labor costs, and is prone to missed detection or false detection. In addition, due to the manual observation method, the accuracy of defect judgment needs to be fed back in the yield test stage, which usually requires a lag of 2 weeks, resulting in untimely defect discovery. For example Figure 13 As shown, the AI ​​data analysis platform obtains historical APC data from the production process of semiconductor components and calculates machine control parameter values ​​through an AI calculation module. The machine then controls production based on these machine control parameter values. The disclosed embodiments, when applied to APC, can automatically determine and correct overlay errors without the need for manual observation, enabling timely detection of defects, saving costs, and reducing the risk of missed or false detections.

[0110] In an exemplary embodiment, obtaining historical wafer condition factor data of a historical batch of wafers on a previous machine may include the following steps B1 to B3.

[0111] Step B1, obtaining the initial wafer front-end measurement results of a historical batch of wafers on a previous machine.

[0112] The initial wafer front-end measurement results represent the measurement results of various front-end parameters of a historical batch of wafers after being processed at the previous previous station before entering the next previous machine for processing.

[0113] Step B2: Calculate the correlation parameter between the initial wafer front-end measurement result and the historical measurement result.

[0114] The historical measurement results represent measurement result vectors of a historical batch of wafers (a historical batch of wafers corresponding to the measurement results of the initial wafer front-end station) after being processed by the next preceding tool.

[0115] In the embodiment of the present disclosure, the correlation parameter can be obtained by calculating the Pearson correlation coefficient, but the present disclosure is not limited thereto.

[0116] For example, the correlation parameter can be obtained by the following formula 4.

[0117]

[0118] Among them, ρjk is the correlation parameter, J is the initial wafer front-end measurement result vector corresponding to the front-end parameter, K is the measurement result vector of the corresponding historical batch wafer on the next preceding machine, σ j is the standard deviation of J, σ k is the standard deviation of K. It should be noted that the larger the absolute value of the correlation parameter is, the higher the correlation is.

[0119] Step B3: determining the front-end measurement parameters corresponding to the initial wafer front-end measurement results whose correlation parameters meet the preset conditions as target front-end measurement parameters.

[0120] The preset conditions can be set based on process requirements or experience. For example, the pre-measurement parameters corresponding to the initial wafer front-end measurement results with a correlation parameter greater than 0.9 can be determined as the target pre-measurement parameters, or the pre-measurement parameters with correlation parameters in the top 10% can be taken as the target pre-measurement parameters, and a parameter name list corresponding to the target pre-measurement parameters can be saved.

[0121] It should be noted that the training samples determine which pre-measurement parameters are retained (the retained pre-measurement parameters are called target pre-measurement parameters). The parameters corresponding to the historical measurement results can be the same as or different from the parameters of the initial wafer front-end measurement results. For example, the parameter line width corresponding to the initial wafer front-end measurement results can affect the parameter film thickness corresponding to the historical measurement results.

[0122] In the embodiment of the present disclosure, historical wafer condition factor data may include measurement results of target pre-measurement parameters of historical batches of wafers after being produced on previous machines; current wafer condition factor data may include measurement results of target pre-measurement parameters of current batches of wafers after being produced on previous machines.

[0123] Exemplarily, the historical wafer status factor data may also include the process waiting time between adjacent preceding machines for historical batches of wafers.

[0124] The present disclosure filters multiple front-end measurement parameters corresponding to the initial wafer front-end measurement results through correlation parameters, which can reduce the amount of calculation, save the training time of the machine learning model, and also reduce the memory occupied by the training samples.

[0125] In another exemplary embodiment, obtaining historical wafer condition factor data for a previous wafer batch at a previous machine may include: obtaining initial wafer front-end station sensor data recorded for the previous wafer batch at the previous machine; and using an autoencoder model to reduce the dimensionality of the initial wafer front-end station sensor data to determine target wafer front-end station sensor data. The historical wafer condition factor data includes the target wafer front-end station sensor data.

[0126] As to what the autoencoding model is specifically, the embodiment of the present disclosure does not impose any restrictions, as long as it is a model that can achieve dimensionality reduction of the initial wafer front-end sensor recorded data. For example, the autoencoding model can be an autoencoder (Autoencoder) model, and the Autoencoder model reduces the dimensionality of the initial wafer front-end sensor recorded data, inputs the initial wafer front-end sensor recorded data in both the input (Input) and the output (Output), learns self-features in the network, sets the dimension of the Code (coding) layer to 100, and extracts the calculation result of the Code layer, which is the feature result after dimensionality reduction. The embodiment of the present disclosure reduces the dimensionality of the initial wafer front-end sensor recorded data through the autoencoding model, which is conducive to the rapid implementation of the present disclosure between different processes.

[0127] S1002: Obtain historical tool condition factor data of the current tool when a historical batch of wafers arrives at the current tool. The historical tool condition factor data may include at least one of the current tool's model, manufacturer, maintenance cycle, and measurement statistics within a predetermined time period.

[0128] S1003: Obtain historical machine control parameters for a historical batch of wafers produced on the current machine. Historical machine control parameters are parameters that control the machine's operation, and the machine produces wafer products based on these historical machine control parameters. For example, the historical machine control parameters may be film thickness.

[0129] S1004: Obtain historical measurement results of historical batches of wafers produced on the current machine. The historical measurement results are measurement results obtained in actual production.

[0130] S1005, processing historical wafer condition factor data, historical tool condition factor data, and historical tool control parameters through a machine learning model to obtain historical predicted measurement results of historical batches of wafers after being manufactured on the current tool.

[0131] Historical wafer condition factor data, historical tool condition factor data, and historical tool control parameters are input into the machine learning model to obtain historical predicted measurement results. The historical predicted measurement results and historical measurement results are used as training samples to train the machine learning model.

[0132] S1006: Training a machine learning model based on historical measurement results and historical predicted measurement results.

[0133] The present disclosure does not limit the specific machine learning model. For example, the machine learning model is a LightGBM (Light Gradient Boosting Machine) model. The LightGBM model is a framework for implementing the GBDT algorithm, supports efficient parallel training, and has advantages such as faster training speed, lower memory consumption, better accuracy, and support for distributed processing of massive data. For another example, the machine learning model is a support vector machine (SVM) model, which is a binary classification model.

[0134] For example, Figure 15 As shown in FIG, 12 months of wafer manufacturing data collected from the tool is used as the training sample set {X, Y}, where X is the historical predicted measurement result and Y is the historical measurement result. The training process is as follows: Steps C1 to C4.

[0135] Step C1, obtain a sample set {X, Y}.

[0136] Step C2, according to the distribution weight vector D t Get the training subset S t .

[0137] Step C3: training multiple learners based on different subsets.

[0138] Step C4, obtain the joint learner F(X).

[0139] For example, Figure 16 As shown, the learner is a Classification and Regression Tree (CART), and the training of the CART tree may include the following steps D1 to D3.

[0140] Step D1, select the best feature X at each layer (a) Perform tree splitting.

[0141] In step D2, only one branch is selected for new feature segmentation.

[0142] Step D3, on the leaf node The mean of is taken as the prediction result.

[0143] For example, a trained machine learning model is Figure 17 As shown, current wafer condition factor data, current machine condition factor data and multiple preset machine control parameters are input into the input layer, and multiple current predicted measurement results of the current batch of wafers are obtained in the output layer.

[0144] Among all the influencing factors, the wafer front-end measurement results (referred to as "front measurement") and wafer front-end sensor records (referred to as "front-end FDC") have many features. If they are not processed, they will interfere with other factors. The embodiment of the present disclosure can reduce the interference between features through the above-mentioned automated factor screening method and automated feature dimensionality reduction method, reduce the amount of data processing, and speed up processing efficiency.

[0145] The following description takes the ETCH CAP (etched capacitor) process as an example.

[0146] Experimental group 1 uses a combination of R2R system and AI fine-tuning for regulation, experimental group 2 uses AI for regulation, and the control group uses R2R system for regulation. Figure 18 The final experimental results of measuring CD (Critical Dimension) are shown in Table 1 below.

[0147] Table 1

[0148] CAP Final CD Experimental Group 1 Experimental Group 2 control group Average measurement value 52.41338 52.45059 52.19179 Absolute error from target 0.087 0.049 0.308

[0149] It can be seen from Table 1 that the parameter adjustment effect of the present disclosure in the ETCH CAP (etching capacitor) process section is 84% ​​better than that of the control group (R2R).

[0150] Based on the same inventive concept, the present disclosure also provides a machine parameter control device, as described in the following embodiments. Since the principles of the device embodiment are similar to those of the above-mentioned method embodiment, the implementation of the device embodiment can refer to the implementation of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0151] Figure 19 A schematic diagram of a machine parameter control device according to an embodiment of the present disclosure is shown. Figure 19 As shown, the device may include an acquisition module 191 and a determination module 192. The acquisition module 191 may be used to obtain the current wafer condition factor data of the current batch of wafers on the previous machine and the current machine condition factor data of the current machine when the current batch of wafers reaches the current machine. The acquisition module 191 may also be used to obtain n preset machine control parameters for the current batch of wafers when they are produced on the current machine, where n is a positive integer greater than 1. The acquisition module 191 may also be used to process the current wafer condition factor data, the current machine condition factor data and the n preset machine control parameters through a trained machine learning model to obtain n current predicted measurement results for the current batch of wafers. The determination module 192 may be used to determine the target machine control parameters for the current batch of wafers when they are produced on the current machine from the n preset machine control parameters based on the n current predicted measurement results and the target measurement results for the current batch of wafers.

[0152] In one embodiment, the current machine is the t-th machine, the previous machines include the 1st to t-1th machines, and t is a positive integer greater than 1.

[0153] In one embodiment, the acquisition module 191 can also be used to obtain the upper and lower limits of the machine control parameters of the current batch of wafers when manufactured on the current machine; within the upper and lower limits of the machine control parameters, n preset machine control parameters are evenly generated using an arithmetic progression.

[0154] In one embodiment, the acquisition module 191 can also be used to combine the i-th preset machine control parameter with the current wafer condition factor data and the current machine condition factor data into the i-th input vector, where i is a positive integer greater than or equal to 1 and less than or equal to n; the i-th input vector is input into the machine learning model to obtain the i-th current predicted measurement result of the current batch of wafers.

[0155] In one embodiment, the determination module 192 may further be configured to calculate the absolute difference between each of the n current predicted measurement results and the target measurement result; and determine the preset machine control parameter corresponding to the current predicted measurement result with the smallest absolute difference as the target machine control parameter.

[0156] In one embodiment, the acquisition module 191 can also be used to obtain the batch control parameter value of the current batch of wafers when produced on the current machine; obtain the floating threshold of the batch control parameter value; determine the batch control range based on the batch control parameter value and the floating threshold; determine whether the target machine control parameter is within the batch control range; if the target machine control parameter is within the batch control range, transmit the target machine control parameter to the production manufacturing system, so that the production manufacturing system controls the current machine to produce the current batch of wafers according to the target machine control parameter.

[0157] In one embodiment, the acquisition module 191 can also be used to determine whether the target machine control parameter is within a preset reasonable range; if the target machine control parameter is within the preset reasonable range, obtain the previous machine control parameter of the previous batch of wafers when the current batch of wafers was produced on the current machine; determine whether the adjustment amplitude of the target machine control parameter compared with the previous machine control parameter is greater than the amplitude threshold; if the adjustment amplitude is less than or equal to the amplitude threshold, determine whether the target machine control parameter is within the batch control range.

[0158] In one embodiment, the acquisition module 191 can also be used to obtain historical wafer condition factor data of a historical batch of wafers on a previous machine; obtain historical machine condition factor data of the current machine when the historical batch of wafers reaches the current machine; obtain historical machine control parameters of the historical batch of wafers when produced on the current machine; obtain historical measurement results of the historical batch of wafers after being produced on the current machine; process the historical wafer condition factor data, historical machine condition factor data and historical machine control parameters through a machine learning model to obtain historical predicted measurement results of the historical batch of wafers after being produced on the current machine; and train the machine learning model based on the historical measurement results and the historical predicted measurement results.

[0159] In one embodiment, the acquisition module 191 can also be used to obtain the initial wafer front-end measurement results of a historical batch of wafers on a previous machine; calculate the correlation parameters between the initial wafer front-end measurement results and the historical measurement results; determine the front-end measurement parameters corresponding to the initial wafer front-end measurement results whose correlation parameters meet preset conditions as target front-end measurement parameters; wherein the historical wafer condition factor data includes the measurement results of the historical batch of wafers after being produced on a previous machine for the target front-end measurement parameters; the current wafer condition factor data includes the measurement results of the current batch of wafers after being produced on a previous machine for the target front-end measurement parameters.

[0160] In one embodiment, the acquisition module 191 can also be used to obtain the initial wafer front-end sensor recording data of a historical batch of wafers on a previous machine; use the autoencoding model to reduce the dimension of the initial wafer front-end sensor recording data to determine the target wafer front-end sensor recording data; wherein the historical wafer condition factor data includes the target wafer front-end sensor recording data.

[0161] In one embodiment, the historical wafer status factor data further includes the process waiting time between adjacent preceding tools for historical batches of wafers.

[0162] In one embodiment, the historical machine condition factor data includes at least one of the model, manufacturer, maintenance cycle, and measurement statistics within a predetermined period of time of the current machine.

[0163] See also Figure 20 , Figure 20 This is a schematic diagram of the structure of a computer device provided by an embodiment of the present disclosure. Figure 20As shown, the computer device in the embodiment of the present disclosure may include: one or more processors 2001, a memory 2002, and an input / output interface 2003. The processor 2001, the memory 2002, and the input / output interface 2003 are connected via a bus 2004. The memory 2002 is used to store computer programs, which include program instructions. The input / output interface 2003 is used to receive and output data, such as for data exchange between a host computer and the computer device, or for data exchange between virtual machines in the host computer. The processor 2001 is used to execute the program instructions stored in the memory 2002.

[0164] Among them, the processor 2001 can perform the following operations: obtain the current wafer condition factor data of the current batch of wafers on the previous machine and the current machine condition factor data of the current machine when the current batch of wafers reaches the current machine; obtain n preset machine control parameters when the current batch of wafers is produced on the current machine, where n is a positive integer greater than 1; process the current wafer condition factor data, the current machine condition factor data and the n preset machine control parameters through the trained machine learning model to obtain n current predicted measurement results of the current batch of wafers; determine the target machine control parameters of the current batch of wafers when produced on the current machine from the n preset machine control parameters based on the n current predicted measurement results and the target measurement results of the current batch of wafers.

[0165] The memory 2002 may include a read-only memory and a random access memory, and provides instructions and data to the processor 2001 and the input / output interface 2003. A portion of the memory 2002 may also include a non-volatile random access memory. In a specific implementation, the computer device may execute the implementation methods provided in each step of any of the above method embodiments through its built-in functional modules. For details, please refer to the implementation methods provided in each step in the figures shown in the above method embodiments, and will not be repeated here.

[0166] The embodiments of the present disclosure provide a computer device including a processor, an input / output interface, and a memory. The processor obtains a computer program in the memory to execute the steps of the method shown in any of the above embodiments.

[0167] The present disclosure also provides a computer-readable storage medium storing a computer program. Figure 21 A schematic diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown. Figure 21 As shown, the computer readable storage medium 2100 stores a program product capable of implementing the above method of the present disclosure. The computer program is suitable for being loaded by the processor and executing the machine parameter control method provided in each step of any of the above embodiments.

[0168] The present disclosure also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method provided in any of the optional embodiments described above.

Claims

1. A machine parameter control method, characterized in that: include: Obtaining current wafer condition factor data of a current batch of wafers on a previous machine and current machine condition factor data of the current machine when the current batch of wafers arrives at the current machine; Obtaining n preset machine control parameters for the current batch of wafers when manufactured on the current machine, where n is a positive integer greater than 1; Processing the current wafer condition factor data, the current tool condition factor data, and n preset tool control parameters through a trained machine learning model to obtain n current predicted measurement results of the current batch of wafers; According to n current predicted measurement results and target measurement results of the current batch of wafers, target tool control parameters for the current batch of wafers when manufactured on the current tool are determined from n preset tool control parameters.

2. The method according to claim 1, wherein The current machine is the t-th machine, the previous machines include the 1st to t-1th machines, and t is a positive integer greater than 1.

3. The method according to claim 1, wherein Obtaining n preset machine control parameters for the current batch of wafers when manufactured on the current machine, including: Obtain upper and lower limits of machine control parameters for the current batch of wafers when manufactured on the current machine; Within the upper and lower limits of the machine control parameters, n preset machine control parameters are uniformly generated using an arithmetic progression.

4. The method according to claim 1, wherein The trained machine learning model processes the current wafer condition factor data, the current tool condition factor data, and n preset tool control parameters to obtain n current predicted measurement results for the current batch of wafers, including: Combining the i-th preset machine control parameter with the current wafer condition factor data and the current machine condition factor data into an i-th input vector, where i is a positive integer greater than or equal to 1 and less than or equal to n; The i-th input vector is input into the machine learning model to obtain the i-th current predicted measurement result of the current batch of wafers.

5. The method according to claim 1, wherein Determining target tool control parameters for the current batch of wafers when manufactured on the current tool from n preset tool control parameters based on n current predicted measurement results and target measurement results of the current batch of wafers, including: Calculating the absolute difference between each of the n current predicted measurement results and the target measurement result; The preset machine control parameter corresponding to the current predicted measurement result with the smallest absolute difference is determined as the target machine control parameter.

6. The method according to claim 1, wherein Also includes: Obtaining batch control parameter values ​​of the current batch of wafers when manufactured on the current machine; Obtaining a floating threshold value of the batch control parameter value; determining a batch control range according to the batch control parameter value and the floating threshold; Determining whether the target machine control parameter is within the batch control range; If the target tool control parameter is within the batch control range, transmitting the target tool control parameter to the production manufacturing system, so that the production manufacturing system controls the current tool to manufacture the current batch of wafers according to the target tool control parameter; The step of determining whether the target machine control parameter is within the batch control range includes: Determining whether the target machine control parameter is within a preset reasonable range; If the target machine control parameter is within the preset reasonable range, obtaining the previous machine control parameter of the previous batch of wafers of the current batch when produced on the current machine; determining whether an adjustment amplitude of the target machine control parameter compared to the previous machine control parameter is greater than an amplitude threshold; If the adjustment amplitude is less than or equal to the amplitude threshold, it is determined whether the target machine control parameter is within the batch control range.

7. The method according to claim 1, wherein Also includes: Obtain historical wafer condition factor data of a historical batch of wafers on the preceding machine; Obtaining historical machine status factor data of the current machine when the historical batch of wafers arrives at the current machine; Obtaining historical machine control parameters of the historical batch of wafers when manufactured on the current machine; Obtaining historical measurement results of the historical batch of wafers after being manufactured on the current tool; Processing the historical wafer condition factor data, the historical tool condition factor data, and the historical tool control parameters through the machine learning model to obtain historical predicted measurement results of the historical batch of wafers after being manufactured on the current tool; The machine learning model is trained based on the historical measurement results and the historical predicted measurement results.

8. The method according to claim 7, wherein Obtain historical wafer condition factor data for the historical batch of wafers on the preceding machine, including: Obtaining initial wafer front-end measurement results of the historical batch of wafers on the prior machine; Calculating a correlation parameter between the initial wafer front-end measurement result and the historical measurement result; Determine the front-end measurement parameters corresponding to the initial wafer front-end measurement results whose correlation parameters meet the preset conditions as the target front-end measurement parameters; The historical wafer condition factor data includes the measurement results of the target pre-measurement parameters of the historical batch of wafers after being produced by the previous machine; the current wafer condition factor data includes the measurement results of the target pre-measurement parameters of the current batch of wafers after being produced by the previous machine.

9. The method according to claim 7 or 8, wherein Obtain historical wafer condition factor data for the historical batch of wafers on the preceding machine, including: Obtaining data recorded by sensors at an initial wafer front station of the historical batch of wafers on the previous machine; Using an autoencoder model to reduce the dimension of the initial wafer front-end sensor recorded data to determine target wafer front-end sensor recorded data; The historical wafer condition factor data includes the target wafer front-end sensor recorded data.

10. A computer device, characterized in that: include: one or more processors; The memory is configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the computer device implements the method according to any one of claims 1 to 9.

11. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is run on a computer, the computer is caused to perform the method according to any one of claims 1 to 9.

Images