Modeling method of material removal rate prediction model, related equipment and storage medium

Abstract

The invention provides a modeling method of a material removal rate prediction model, related equipment and a storage medium, the model is suitable for material removal rate prediction of a chemical mechanical polishing process, and the method comprises the following steps: establishing a first sample set based on first production line data of a first semiconductor device processing production line and a corresponding material removal rate; inputting at least part of samples in the first sample set into a to-be-trained neural network model for training to obtain a trained first model; establishing a second sample set based on second production line data of a second semiconductor device processing production line and a corresponding material removal rate; fixing at least part of parameters of the first model, inputting at least part of samples in the second sample set into the first model for training to obtain a trained second model, and taking the trained second model as a material removal rate prediction model of a second semiconductor device processing production line; the chemical mechanical polishing technology processing equipment of the first semiconductor device processing production line and the chemical mechanical polishing technology processing equipment of the second semiconductor device processing production line are different.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of semiconductor device processing technology, and in particular to a modeling method, related equipment, and storage medium for a material removal rate prediction model. Background Art

[0002] The material removal rate of the chemical mechanical polishing (CMP) process is affected by multi-source data from semiconductor device manufacturing lines. Uncovering the influence of multi-source data from semiconductor device manufacturing lines on the material removal rate of the CMP process is of great significance for CMP process optimization.

[0003] However, existing material removal rate prediction models usually require a large amount of processing line data to train them. Therefore, for processing lines with scarce data, it is usually difficult to establish a material removal rate prediction model with high prediction accuracy.

[0004] In addition, since there are many similar devices in the integrated circuit processing production line with the same working principles but different structures, the material removal rate prediction model trained based on the data of a specific processing production line has low prediction accuracy for other production lines.

[0005] Therefore, how to establish a material removal rate prediction model with high prediction accuracy based on a small amount of processing line data becomes a challenge. Summary of the Invention

[0006] In response to the above technical problems, the embodiments of the present disclosure provide a modeling method, related equipment and storage medium for a material removal rate prediction model, which can establish a material removal rate prediction model with high prediction accuracy based on a small amount of processing production line data in the absence of data.

[0007] In a first aspect, an embodiment of the present disclosure provides a modeling method for a material removal rate prediction model, wherein the material removal rate prediction model is applicable to material removal rate prediction in a chemical mechanical polishing process, and the modeling method includes:

[0008] Establishing a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate;

[0009] Inputting at least part of the samples in the first sample set into a neural network model to be trained to obtain a trained first model;

[0010] Establishing a second sample set based on second production line data of a second semiconductor device processing production line and its corresponding material removal rate;

[0011] Fixing at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training to obtain a trained second model, and using the second model as a material removal rate prediction model for the second semiconductor device processing line;

[0012] The chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

[0013] Optionally, the number of samples in the second sample set is smaller than the number of samples in the first sample set.

[0014] Optionally, fixing at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training, and obtaining a trained second model includes:

[0015] adjusting at least some parameters of the first model;

[0016] At least part of the samples in the second sample set are input into the first model with adjusted parameters to train, to obtain a trained second model.

[0017] Optionally, the first production line data includes: first type characteristic parameters, the first type characteristic parameters being suitable for characterizing a processing environment when a chemical mechanical polishing process is performed on the first production line;

[0018] The second production line data includes: first-type characteristic parameters, where the first-type characteristic parameters are suitable for characterizing a processing environment when a chemical mechanical polishing process is performed on the second production line.

[0019] Optionally, the first type of characteristic parameters includes at least one of the following:

[0020] The pressure applied by the polishing head on different areas of the semiconductor device, the retainer pressure, the polishing pad speed, the polishing head speed, the polishing head swing trajectory, the dresser swing trajectory, the dresser speed, the dresser downforce, the polishing fluid flow rate, and the incoming semiconductor device information.

[0021] Optionally, the first production line data further includes:

[0022] The second type of characteristic parameters is suitable for characterizing the usage of the polishing consumables when the chemical mechanical polishing process is performed on the first production line; and / or

[0023] The third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the first production line;

[0024] The second production line data also includes:

[0025] The second type of characteristic parameters is suitable for characterizing the usage of the polishing consumables when the chemical mechanical polishing process is performed on the second production line; and / or

[0026] The third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the second production line.

[0027] Optionally, establishing the first sample set based on first production line data of a first semiconductor device processing line and its corresponding material removal rate includes:

[0028] labeling the first type of feature parameters with a first label;

[0029] labeling the second type of feature parameters with a second label; and / or

[0030] labeling the third type of feature parameters with a third label;

[0031] The establishing of a second sample set based on the second production line data of the second semiconductor device processing production line and the corresponding material removal rate includes:

[0032] labeling the first type of feature parameters with a first label;

[0033] labeling the second type of feature parameters with a second label; and / or

[0034] A third label is attached to the third type of feature parameters.

[0035] Optionally, inputting at least part of the samples in the first sample set into a neural network model to be trained for training to obtain a trained first model includes:

[0036] assigning a first static weight to the first type of feature parameter according to the first label;

[0037] assigning a second static weight to the second type of feature parameter according to the second label; and / or

[0038] assigning a third static weight to the third type of feature parameter according to the third label;

[0039] Passing the first static weight, the second static weight and / or the third static weight to the loss function of the neural network model;

[0040] The fixing of at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training, and obtaining a trained second model includes:

[0041] assigning a first static weight to the first type of feature parameter according to the first label;

[0042] assigning a second static weight to the second type of feature parameter according to the second label; and / or

[0043] assigning a third static weight to the third type of feature parameter according to the third label;

[0044] Passing the first static weight, the second static weight and / or the third static weight to the loss function of the first model;

[0045] The second static weight and the third static weight are both greater than the first static weight.

[0046] Optionally, the second type of characteristic parameters includes at least one of the following:

[0047] The number of times the polishing pad is used, the number of times the dresser is used, the number of times the retainer is used, the number of times the adsorption film is used, the use time of the polishing fluid, and the use time of the filter element;

[0048] The third type of characteristic parameters includes at least one of the following:

[0049] Polishing pad porosity, polishing pad hardness, polishing pad density, polishing pad pore size, polishing pad elastic modulus, slurry abrasive particle size, slurry abrasive particle volume fraction and the hardness of the removed material.

[0050] In a second aspect, an embodiment of the present disclosure provides a modeling device for a material removal rate prediction model, comprising:

[0051] A first sample set establishing module configured to establish a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate;

[0052] A first model training module is configured to input at least part of the samples in the first sample set into the neural network model to be trained to obtain a trained first model;

[0053] A second sample set establishing module configured to establish a second sample set based on second production line data of a second semiconductor device processing production line and its corresponding material removal rate;

[0054] a second model training module configured to fix at least some parameters of the first model, input at least some samples in the second sample set into the first model for training, and obtain a trained second model, wherein the second model serves as a material removal rate prediction model for the second semiconductor device processing line;

[0055] The chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

[0056] In a third aspect, an embodiment of the present disclosure provides a storage medium having a computer program stored thereon, which, when executed by a processor, executes the steps of the modeling method of the material removal rate prediction model described in any of the above embodiments.

[0057] In a fourth aspect, an embodiment of the present disclosure provides a computer program product, comprising a computer program, which, when running, executes the steps of the modeling method of the material removal rate prediction model described in any of the above embodiments.

[0058] Using the modeling method of the material removal rate prediction model provided by the embodiment of the present disclosure, a first sample set is established based on the first production line data of a first semiconductor device processing line and its corresponding material removal rate, and at least part of the samples in the first sample set are input into the neural network model to be trained for training, thereby obtaining a trained first model. A second sample set is established based on the second production line data of a second semiconductor device processing line and its corresponding material removal rate, and at least part of the weight parameters of the first model are fixed, and at least part of the samples in the second sample set are input into the first model for training, thereby obtaining a trained second model, which can serve as a material removal rate prediction model for the second semiconductor device processing line. Since the second model is trained on the basis of fixing at least part of the parameters of the first model, even if the second production line data of the second semiconductor device processing line is scarce, the second model can achieve a high material removal rate prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0060] Figure 1 A flowchart illustrating an example method for modeling a material removal rate prediction model consistent with some embodiments of the present disclosure is shown.

[0061] Figure 2 A structural example diagram of a modeling device for a material removal rate prediction model consistent with some embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0062] Existing material removal rate prediction models usually require a large amount of processing line data to train them. Therefore, for processing lines with scarce data, it is usually difficult to establish a material removal rate prediction model with high prediction accuracy.

[0063] For ease of understanding, the following briefly introduces the chemical mechanical polishing (CMP) process.

[0064] CMP technology combines the action of chemical reagents with mechanical action. On the basis of mechanical polishing, corresponding abrasive slurry is added according to the semiconductor device material and surface requirements to achieve enhanced polishing and selective polishing effects.

[0065] The equipment involved in the CMP process includes a rotating table, polishing pad, polishing head, polishing slurry dispenser, holder, and dresser. During the CMP process, a semiconductor device (e.g., a wafer) is mounted beneath the polishing head, held in close contact with the polishing pad under pressure and rotating with the head. Simultaneously, a polishing slurry composed of abrasives and chemicals is delivered to the polishing pad via the polishing slurry dispenser. As the rotating table drives the polishing pad and the polishing head oscillates, the polishing slurry enters the space between the semiconductor device and the polishing pad, chemically reacting with the lower surface of the semiconductor device, forming a very thin chemical reaction layer. Micromechanical friction between the abrasive particles then removes the products of the chemical reaction from the semiconductor device's surface. This alternating process of chemical and mechanical action achieves surface flattening of the semiconductor device. After polishing, the polishing pad can be trimmed with a dresser to maintain its performance.

[0066] The material removal rate (MRR) in the CMP process refers to the thickness of the workpiece erased per unit time. It has a significant impact on the polishing rate, polishing accuracy and surface quality of semiconductor devices, and is a key indicator for evaluating the CMP process.

[0067] The CMP process must not only effectively remove material but also precisely control the MRR (Magnetic Reduction Rate) to achieve specific surface topography requirements. MRR can be used to further analyze and calculate thickness variations on the surface of semiconductor devices, providing real-time profiles and features. This can provide decision-making guidance for application processes such as layout design and electrical characteristic analysis.

[0068] From the above, we can see that establishing a material removal rate prediction model with high prediction accuracy and revealing the influence of semiconductor device processing production line data on MRR are of great significance for optimizing the CMP process.

[0069] In response to the above problems, the present disclosure provides some modeling methods for material removal rate prediction models. These methods establish a first sample set based on the first production line data of the first semiconductor device processing line and its corresponding material removal rate, and input at least part of the samples in the first sample set into the neural network model to be trained for training, so as to obtain a trained first model. A second sample set is established based on the second production line data of the second semiconductor device processing line and its corresponding material removal rate, and at least part of the parameters of the first model are fixed, and at least part of the samples in the second sample set are input into the first model for training, so as to obtain a trained second model, which can be used as a material removal rate prediction model for the second semiconductor device processing line. Since the second model is trained on the basis of fixing at least part of the weight parameters of the first model, even if the second production line data of the second semiconductor device processing line is scarce, the second model can achieve a higher material removal rate prediction accuracy.

[0070] In order to enable those skilled in the art to better understand and implement the embodiments of the present disclosure, the concepts, schemes, principles and advantages of the embodiments of the present disclosure are described in detail below with reference to the accompanying drawings and through specific application examples.

[0071] Figure 1 A flow chart illustrating a method for modeling a material removal rate prediction model consistent with some embodiments of the present disclosure is shown. The method for modeling a material removal rate prediction model can be executed by a processor or a first device. For example, the processor can be a central processing unit (CPU), a microprocessor (microprocessor), or a field programmable gate array (FPGA). For example, the first device can be a modeling device for a material removal rate prediction model. In some embodiments, referring to Figure 2 The structure example diagram of a modeling device for a material removal rate prediction model consistent with some embodiments of the present disclosure is shown, and the first device can be Figure 2 In one application scenario, referring to Figure 2 The modeling device T may include a first sample set building module T1, a first model training module T2, a second sample set building module T3 and a second model training module T4.

[0072] Reference Figure 1 In some embodiments, the modeling method of the material removal rate prediction model may include steps A, B, C, and D. It is understandable that the modeling method of the material removal rate prediction model may include more or fewer steps, and the order of the steps may also be the same or different.

[0073] Step A: establishing a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate.

[0074] The first semiconductor device processing production line refers to a semiconductor device processing production line that can perform chemical mechanical polishing process.

[0075] In some embodiments, the first production line data may include actual operating data of processing equipment on the first semiconductor device processing line in a stable operating state. Equipment of the same type with the same operating principle but different structures can be used as processing equipment on the first semiconductor device processing line.

[0076] In some embodiments, the first production line data may include simulation operation data obtained by simulating processing equipment on a first semiconductor device processing production line.

[0077] For example, by obtaining simulation data on process conditions and material removal rate of chemical mechanical polishing process based on material removal mechanism model of solid-solid contact mechanics, fluid dynamics, chemical reaction kinetics and wear theory, the influence of processing parameters such as processing equipment speed, pressure, grinding fluid flow rate on removal rate is studied.

[0078] In some embodiments, simulation may be performed using finite element analysis software, kinematics analysis software, mathematical calculation software, etc.

[0079] In some embodiments, the semiconductor device may be a wafer.

[0080] In some embodiments, the first production line data may include first-type characteristic parameters, wherein the first-type characteristic parameters may characterize a processing environment when a chemical mechanical polishing process is performed on the first production line.

[0081] In some embodiments, the processing environment may include incoming semiconductor device information.

[0082] For example, when the semiconductor device is a wafer, the first type of characteristic parameters may represent the incoming semiconductor device information. For example, the first type of characteristic parameters may include mask information parameters, height difference parameters between different structural layers on the wafer surface, etc.

[0083] In some embodiments, the processing environment may include process parameters.

[0084] For example, the first type of characteristic parameters may characterize process parameters of a semiconductor device. For example, the first type of characteristic parameters may include pressure parameters applied by the polishing head to different regions of the semiconductor device, retainer pressure parameters, polishing pad speed parameters, polishing head speed parameters, polishing head swing trajectory parameters, dresser swing trajectory parameters, dresser speed parameters, dresser downforce parameters, and polishing liquid flow parameters.

[0085] Verification of semiconductor device processing production lines shows that the first type of characteristic parameters are related to the material removal rate, which is a nonlinear relationship.

[0086] For example, increasing the polishing head speed increases the relative motion between the polishing head and the semiconductor device, increasing the number of contacts between the abrasive particles and the semiconductor device per unit time, thereby improving the material removal rate. However, excessively high polishing head speeds can increase instability during the chemical mechanical polishing process, such as excessive vibration and heat generation, which can reduce the material removal rate. Therefore, the relationship between polishing head speed and material removal rate is nonlinear.

[0087] In some embodiments, the processing environment may include semiconductor device incoming material information and process parameters.

[0088] In some embodiments, the first production line data may include one of the above-mentioned first type of characteristic parameters.

[0089] In some embodiments, the first production line data may include multiple types of the above-mentioned first type of characteristic parameters.

[0090] In some embodiments, the first production line data covers a larger range.

[0091] For example, the polishing head rotation speed can cover the interval [30 rpm, 120 rpm]; the polishing pad rotation speed can cover the interval [30 rpm, 120 rpm]; and the polishing liquid flow rate can cover the interval [20 ml / min, 400 ml / min].

[0092] It is understood that the embodiments of the present disclosure do not impose any specific restrictions on the first type of characteristic parameters. The above embodiments are merely exemplary.

[0093] Step B: input at least part of the samples in the first sample set into the neural network model to be trained to obtain a trained first model.

[0094] In some embodiments, the neural network model to be trained may include a deep convolutional neural network, such as VGGNet.

[0095] In some embodiments, the neural network model to be trained may include a recurrent neural network.

[0096] In some embodiments, the first sample set may be pre-processed.

[0097] For example, abnormal samples in the first sample set can be removed. If there is production line data in the first sample set with a material removal rate significantly higher than other samples, the sample can be removed to eliminate the impact of the abnormal data on the training process.

[0098] In some embodiments, the first sample set may be divided into a training sample set, a validation sample set, and a test sample set. The training sample set is input into a neural network model to be trained to obtain a trained first model.

[0099] For example, the first sample set may be divided in proportion, for example, 80% of the sample set may be used as a training sample set, 10% as a validation sample set, and 10% as a test sample set.

[0100] It is understandable that the embodiment of the present disclosure does not impose any specific limitation on the number of samples input into the neural network model to be trained. Those skilled in the art can divide the first sample set according to the actual collected processing line data.

[0101] Step C: establishing a second sample set based on second production line data of a second semiconductor device processing line and its corresponding material removal rate.

[0102] The second semiconductor device processing production line refers to a semiconductor device processing production line that can perform chemical mechanical polishing process.

[0103] In some embodiments, the range covered by the second production line data is smaller than the range covered by the first production line data.

[0104] In some embodiments, the range covered by the first production line data includes the range covered by the second production line data.

[0105] For example, in the data for the first production line, the polishing head speed can cover the range [30 rpm, 120 rpm]; the polishing pad speed can cover the range [30 rpm, 120 rpm]; and the polishing slurry flow rate can cover the range [20 ml / min, 400 ml / min]. In the data for the second production line, the polishing head speed can cover the range [50 rpm, 80 rpm]; the polishing pad speed can cover the range [50 rpm, 80 rpm]; and the polishing slurry flow rate can cover the range [30 ml / min, 100 ml / min].

[0106] In some embodiments, the chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

[0107] For example, the equipment involved in the CMP process includes: a rotary table, a polishing pad, a polishing head, a polishing liquid dispenser, a retainer, a dresser, etc. In some embodiments, the chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line is different, which may mean that at least one of the equipment is different.

[0108] In some embodiments, the presence of different devices may include different models.

[0109] In some embodiments, the presence of differences in the devices may include being capable of performing the same semiconductor device process but having different structures.

[0110] In some embodiments, the second production line data may include first-type characteristic parameters, wherein the first-type characteristic parameters may characterize a processing environment when a chemical mechanical polishing process is performed on the second production line.

[0111] More embodiments of the first type of characteristic parameters can refer to the aforementioned embodiments and will not be described in detail here.

[0112] In some embodiments, the second production line data may include one of the aforementioned first type of characteristic parameters.

[0113] In some embodiments, the second production line data may include multiple types of the aforementioned first type of characteristic parameters.

[0114] Step D: fix at least part of the parameters of the first model, input at least part of the samples in the second sample set into the first model for training, and obtain a trained second model, which serves as the material removal rate prediction model for the second semiconductor device processing line.

[0115] For example, the neural network model to be trained adopts the VGGNet model, which may include multiple convolutional layers and multiple fully connected layers.

[0116] In some embodiments, the parameters of all convolutional layers in the trained first model may be fixed.

[0117] In some embodiments, parameters of some convolutional layers in the trained first model may be fixed.

[0118] For example, the parameters of some convolutional layers near the input layer in the trained first model can be fixed. Because the chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different, although both can perform the chemical mechanical polishing process, the relationship between the data of the first production line and its corresponding material removal rate is not exactly the same as the relationship between the data of the second production line and its corresponding material removal rate.

[0119] By adopting the above embodiment, by fixing the weight parameters of some convolutional layers close to the input layer in the first model after training, the general features of the relationship between the first production line data and its corresponding material removal rate learned by the first model can be fixed, and then the some convolutional layers close to the output layer with unfixed weights can be allowed to learn new features of the relationship between the second production line data and its corresponding material removal rate, thereby further improving the material removal rate prediction accuracy of the second model.

[0120] In some embodiments, the second sample set may be pre-processed.

[0121] More embodiments of preprocessing the second sample set may refer to the aforementioned embodiment of preprocessing the first sample set, and will not be described in detail here.

[0122] Using the above embodiment, a first sample set is established based on first line data of a first semiconductor device processing line and its corresponding material removal rate, and at least a portion of the samples in the first sample set are input into a neural network model to be trained for training, thereby obtaining a trained first model. A second sample set is established based on second line data of a second semiconductor device processing line and its corresponding material removal rate, and at least a portion of the parameters of the first model are fixed, and at least a portion of the samples in the second sample set are input into the first model for training, thereby obtaining a trained second model. The second model can serve as a material removal rate prediction model for the second semiconductor device processing line. Because the second model is trained based on at least a portion of the parameters of the first model, even if the second line data of the second semiconductor device processing line is scarce, the second model can achieve a high material removal rate prediction accuracy.

[0123] In some embodiments, the number of samples in the second sample set is smaller than the number of samples in the first sample set.

[0124] Using the above embodiment, since the number of samples in the first sample set is large, the prediction accuracy of the first model obtained by training based on at least part of the samples in the first sample set is high, so that the second model obtained by training based on the second sample set with a smaller number of samples can also achieve high prediction accuracy.

[0125] In some embodiments, step D may include the following steps:

[0126] Step D1: Adjust at least some parameters of the first model.

[0127] In some embodiments, the weight parameters of all fully connected layers in the first model can be adjusted.

[0128] For example, the weight parameters of all fully connected layers in the first model are initialized to random values.

[0129] Step D2: Input at least part of the samples in the second sample set into the first model after structural adjustment for training to obtain a trained second model.

[0130] By adopting the above embodiment, by initializing the weight parameters of all fully connected layers in the first model to random values, the negative impact of the first model on the training of the second model can be avoided, thereby further improving the material removal rate prediction accuracy of the second model.

[0131] In some embodiments, the first production line data further includes a second type of characteristic parameters, wherein the second type of characteristic parameters are suitable for characterizing the usage of polishing consumables when performing a chemical mechanical polishing process on the first production line. The second production line data further includes a second type of characteristic parameters, wherein the second type of characteristic parameters are suitable for characterizing the usage of polishing consumables when performing a chemical mechanical polishing process on the second production line.

[0132] For example, the second type of characteristic parameters may include the number of times the polishing pad is used, the number of times the dresser is used, the number of times the retainer is used, the number of times the adsorption film is used, the use time of the grinding liquid, and the use time of the filter element.

[0133] Verification of semiconductor device processing production lines shows that the second type of characteristic parameters are related to the material removal rate.

[0134] Specifically, polishing consumables will experience varying degrees of wear during use, which will affect the material removal rate.

[0135] For example, as the polishing pad wears during polishing, the stress at the edge of the semiconductor device is higher than that at the center, resulting in a decrease in material removal rate.

[0136] For another example, as the use time increases, the abrasive particles in the grinding fluid may gradually wear out or aggregate, resulting in a decrease in material removal rate. In some embodiments, the first production line data and the second production line data may include one of the second type of characteristic parameters.

[0137] In some embodiments, the first production line data and the second production line data may include multiple types of the above-mentioned second type of characteristic parameters.

[0138] It is understood that the embodiments of the present disclosure do not impose any specific restrictions on the second type of characteristic parameters. The above embodiments are merely exemplary.

[0139] Since the chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different, although both can perform the chemical mechanical polishing process, the relationship between the first production line data and its corresponding material removal rate and the relationship between the second production line data and its corresponding material removal rate are not exactly the same, so the prediction accuracy of the first model and the second model is also inconsistent.

[0140] By adopting the above embodiment, by adding the second type of characteristic parameters on the basis of the first type of characteristic parameters, the applicability of the parameters in the first model in other semiconductor device processing lines that have different chemical mechanical polishing process equipment from the first semiconductor device processing line can be improved, thereby avoiding a significant decrease in the prediction accuracy of the second model obtained by training based on the first model.

[0141] In some embodiments, the first production line data further includes a third type of characteristic parameters, wherein the third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the first production line. The second production line data further includes a third type of characteristic parameters, wherein the third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the second production line.

[0142] For example, the third type of characteristic parameters may include polishing pad porosity, polishing pad hardness, polishing pad density, polishing pad pore size, polishing pad elastic modulus, grinding liquid abrasive particle size, grinding liquid abrasive particle volume fraction and removed material hardness.

[0143] Verification of semiconductor device processing production lines shows that the third type of characteristic parameters are related to the material removal rate, and the relationship is nonlinear.

[0144] For example, a higher hardness polishing pad can more effectively apply pressure, ensuring closer contact between the abrasive particles and the semiconductor device surface, thereby improving material removal efficiency. However, if the polishing pad hardness is too high, it can cause the semiconductor device surface to wear too quickly, making it impossible to effectively remove surface material, thereby reducing material removal efficiency.

[0145] In some embodiments, the first production line data and the second production line data may include one of the third type of characteristic parameters mentioned above.

[0146] In some embodiments, the first production line data and the second production line data may include multiple types of the third type of characteristic parameters mentioned above.

[0147] It is understood that the embodiments of the present disclosure do not impose any specific restrictions on the third type of characteristic parameters. The above embodiments are merely exemplary.

[0148] By adopting the above embodiment, by adding a third type of characteristic parameters on the basis of the first type of characteristic parameters, the applicability of the parameters in the first model in other semiconductor device processing lines that have different chemical mechanical polishing process equipment from the first semiconductor device processing line can be improved, thereby avoiding a significant decrease in the prediction accuracy of the second model obtained by training based on the first model.

[0149] In some embodiments, the first production line data further includes a second type of characteristic parameters and a third type of characteristic parameters. The second production line data further includes a second type of characteristic parameters and a third type of characteristic parameters.

[0150] By adopting the above embodiment, by adding the second type of characteristic parameters and the third type of characteristic parameters on the basis of the first type of characteristic parameters, the applicability of the parameters in the first model in other semiconductor device processing lines that have different chemical mechanical polishing process equipment from the first semiconductor device processing line can be further improved, thereby avoiding a significant decrease in the prediction accuracy of the second model obtained by training based on the first model.

[0151] In some embodiments, when the first production line data includes first-type characteristic parameters and second-type characteristic parameters, step A may include the following steps:

[0152] Step A1: attach a first label to the first type of feature parameters.

[0153] For example, the first label may be the number “0.” In the first sample set, each first-category feature parameter is labeled with the first label “0.”

[0154] Step A2: attaching a second label to the second type of feature parameters.

[0155] For example, the second label may be the number "1". In the first sample set, each second-category feature parameter is labeled with the first label "1". In some embodiments, step C may include the following steps:

[0156] Step C1: attaching a first label to the first type of feature parameters.

[0157] For example, the first label may be the number “0.” In the second sample set, each first-category feature parameter is labeled with the first label “0.”

[0158] Step C2: attaching a second label to the second type of feature parameters.

[0159] For example, the second label may be the number “1.” In the second sample set, each second-category feature parameter is labeled with the first label “1.”

[0160] By adopting the above embodiment, on the one hand, the probability of negative impact on model training due to too many characteristic parameters can be reduced; on the other hand, by affixing labels representing different categories to multiple characteristic parameters, the relationship between each type of characteristic parameter and material removal rate can be clarified, thereby improving the applicability of the weight parameters in the first model in other semiconductor device processing lines that have different chemical mechanical polishing process equipment from the first semiconductor device processing line, thereby avoiding a significant decrease in the prediction accuracy of the second model obtained based on the first model training.

[0161] In some embodiments, when the first production line data includes first-category characteristic parameters, second-category characteristic parameters, and third-category characteristic parameters, step A may include the following steps:

[0162] Step A1: attach a first label to the first type of feature parameters.

[0163] For example, the first label may be the number “0.” In the first sample set, each first-category feature parameter is labeled with the first label “0.”

[0164] Step A2: attaching a second label to the second type of feature parameters.

[0165] For example, the second label may be the number “1.” In the first sample set, each second-category feature parameter is labeled with the first label “1.”

[0166] Step A3: attach a third label to the third type of feature parameters.

[0167] For example, the third label may be the number “2.” In the first sample set, each third-category feature parameter is labeled with the first label “2.”

[0168] Step C may include the following steps:

[0169] Step C1: attaching a first label to the first type of feature parameters.

[0170] For example, the first label may be the number “0.” In the second sample set, each first-category feature parameter is labeled with the first label “0.”

[0171] Step C2: attaching a second label to the second type of feature parameters.

[0172] For example, the second label may be the number “1.” In the second sample set, each second-category feature parameter is labeled with the first label “1.”

[0173] Step C3: attaching a third label to the third type of feature parameters.

[0174] For example, the third label may be the number “2.” In the second sample set, each third-category feature parameter is labeled with the first label “2.”

[0175] In some embodiments, when the first production line data includes first-category characteristic parameters and third-category characteristic parameters, step A may include the following steps:

[0176] Step A1: attach a first label to the first type of feature parameters.

[0177] For example, the first label may be the number “0.” In the first sample set, each first-category feature parameter is labeled with the first label “0.”

[0178] Step A3: attach a third label to the third type of feature parameters.

[0179] For example, the third label may be the number “2.” In the first sample set, each third-category feature parameter is labeled with the first label “2.”

[0180] Step C may include the following steps:

[0181] Step C1: attaching a first label to the first type of feature parameters.

[0182] For example, the first label may be the number “0.” In the second sample set, each first-category feature parameter is labeled with the first label “0.”

[0183] Step C3: attaching a third label to the third type of feature parameters.

[0184] For example, the third label may be the number “2.” In the second sample set, each third-category feature parameter is labeled with the first label “2.”

[0185] It is understood that the embodiments of the present disclosure do not impose any specific restrictions on the first tag, the second tag, and the third tag. The above embodiments are merely illustrative.

[0186] In some embodiments, step B may include the following steps:

[0187] Step B1: assigning a first static weight to the first type of feature parameter according to the first label.

[0188] The static weight refers to a parameter assigned to the feature parameter that remains unchanged throughout the entire training process of the neural network model.

[0189] The first static weight may be a specific constant value.

[0190] Step B2: assigning a second static weight to the second type of feature parameter according to the second label.

[0191] The second static weight may be a specific constant value.

[0192] Step B3: assigning a third static weight to the third type of feature parameters according to the third label.

[0193] The third static weight may be a specific constant value.

[0194] Step B4: passing the first static weight, the second static weight, and the third static weight to the loss function of the neural network model.

[0195] In some embodiments, the second static weight and the third static weight are both greater than the first static weight.

[0196] For example, the first static weight may be set to a constant "2", the second static weight may be set to a constant "4", and the third static weight may be set to a constant "4".

[0197] Step D may include the following steps:

[0198] Step D1: assigning a first static weight to the first type of feature parameter according to the first label.

[0199] The static weight refers to a parameter assigned to the feature parameter that remains unchanged throughout the entire training process of the neural network model.

[0200] The first static weight may be a specific constant value.

[0201] Step D2: assigning a second static weight to the second type of feature parameter according to the second label.

[0202] The second static weight may be a specific constant value.

[0203] Step D3: assigning a third static weight to the third type of feature parameters according to the third label.

[0204] The third static weight may be a specific constant value.

[0205] Step D4: passing the first static weight, the second static weight and / or the third static weight to the loss function of the first model.

[0206] It should be understood that the embodiments of the present disclosure do not impose specific limitations on the first static weight, the second static weight, and the third static weight, and the above embodiments are merely illustrative. Furthermore, the second static weight and the third static weight may be the same or different, as long as both the second static weight and the third static weight are greater than the first static weight.

[0207] By adopting the above embodiment, by transferring the first static weight, the second static weight and the third static weight to the loss function of the neural network model, and by setting the second static weight and the third static weight to be greater than the first static weight, the contribution of the second type of characteristic parameters and the third type of characteristic parameters in the loss function is greater than that of the first type of characteristic parameters, thereby reducing the coupling degree between the first type of characteristic parameters and the material removal rate, thereby avoiding a significant decrease in the prediction accuracy of the second model obtained by training based on the first model.

[0208] The embodiment of the present disclosure also provides a modeling device for a material removal rate prediction model. Figure 2 The structure of a modeling device for a material removal rate prediction model consistent with some embodiments of the present disclosure is shown in FIG. In some embodiments, the modeling device T may include:

[0209] The first sample set establishing module T1 is configured to establish a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate.

[0210] In some embodiments of the present disclosure, the first sample set establishment module may be implemented by an integrated circuit. In some embodiments, the first sample set establishment module may be implemented by a processor, such as a central processing unit (CPU), a microprocessor (microprocessor), or a field programmable gate array (FPGA). In some embodiments, the first sample set establishment module may be implemented by a combination of an integrated circuit and a processor.

[0211] The first model training module T2 is configured to input at least part of the samples in the first sample set into the neural network model to be trained to obtain a trained first model.

[0212] In some embodiments of the present disclosure, the first model training module may be implemented by an integrated circuit. In some embodiments, the first model training module may be implemented by a processor, such as a central processing unit (CPU), a microprocessor (microprocessor), or a field programmable gate array (FPGA). In some embodiments, the first model training module may be implemented by a combination of an integrated circuit and a processor.

[0213] The second sample set establishing module T3 is configured to establish a second sample set based on the second production line data of the second semiconductor device processing production line and the corresponding material removal rate.

[0214] In some embodiments of the present disclosure, the second sample set establishment module may be implemented by an integrated circuit. In some embodiments, the second sample set establishment module may be implemented by a processor, such as a central processing unit (CPU), a microprocessor (microprocessor), or a field programmable gate array (FPGA). In some embodiments, the second sample set establishment module may be implemented by a combination of an integrated circuit and a processor.

[0215] The second model training module T4 is configured to fix at least part of the parameters of the first model, input at least part of the samples in the second sample set into the first model for training, and obtain a trained second model. The second model serves as the material removal rate prediction model of the second semiconductor device processing production line.

[0216] In some embodiments of the present disclosure, the second model training module may be implemented by an integrated circuit. In some embodiments, the second model training module may be implemented by a processor, such as a central processing unit (CPU), a microprocessor (microprocessor), or a field programmable gate array (FPGA). In some embodiments, the second model training module may be implemented by a combination of an integrated circuit and a processor.

[0217] In some embodiments, the chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

[0218] Using the above embodiment, a first sample set is established based on the first production line data of a first semiconductor device processing line and its corresponding material removal rate through a first sample set establishment module, and at least a portion of the samples in the first sample set are input into the neural network model to be trained through a first model training module for training, thereby obtaining a trained first model. A second sample set is established based on the second production line data of a second semiconductor device processing line and its corresponding material removal rate through a second sample set establishment module, and at least a portion of the weight parameters of the first model are fixed through a second model training module, and at least a portion of the samples in the second sample set are input into the first model for training, thereby obtaining a trained second model. The second model can serve as a material removal rate prediction model for the second semiconductor device processing line. Since the second model is trained based on at least a portion of the parameters of the fixed first model, even if the second production line data of the second semiconductor device processing line is scarce, the second model obtained by this device can achieve a high material removal rate prediction accuracy.

[0219] In the embodiment of the present disclosure, the modeling device of the material removal rate prediction model can adopt the modeling method of the material removal rate prediction model described in any of the aforementioned embodiments to establish a material removal rate prediction model. The specific steps can be referred to the aforementioned embodiments and will not be repeated here.

[0220] The present disclosure also provides a storage medium storing a computer program that, when executed by a processor, executes the steps of the method for modeling a material removal rate prediction model described in any of the above embodiments. The specific steps can be found in the above embodiments and will not be repeated here.

[0221] In some embodiments of the present disclosure, the storage medium may be any suitable readable storage medium such as an optical disc, a mechanical hard disk, or a solid-state drive.

[0222] The present disclosure also provides a computer program product, including a computer program, which, when executed, executes the steps of the method for modeling a material removal rate prediction model described in any of the above embodiments. The specific steps can be found in the above embodiments and will not be repeated here.

[0223] It should be noted that the modules in the embodiments of the present disclosure may be composed of discrete components or implemented by a single electrical chip.

[0224] In this disclosure, unless otherwise expressly provided or limited, ordinal numbers, such as "first" and "second," are used only to distinguish and describe related objects and are not to be understood as indicating or implying the relative importance or order of the related objects. In addition, ordinal numbers do not represent the number of related objects.

[0225] The terms "or" and "and / or" in this disclosure are used to describe the relationship between associated objects, which indicates non-exclusive inclusion. For example, "A and / or B" and "A or B" can both include: "A alone", "B alone", or "A and B", where "A" and "B" can include a single object or multiple objects. For another example, "A, B and / or C", "A, B or C" and "A, B and C" can both include: "A alone", "B alone", "C alone", "A and B", "A and C", "B and C", or "A, B and C", where "A", "B" and "C" can include a single object or multiple objects. In addition, " / " in this disclosure is used to indicate the "or" relationship between the preceding and following associated objects. In this disclosure, "at least one of A or B" and "one or more of A and B" have the same meaning as "A or B" above, and "one or more of A, B and C" and "at least one of A, B or C" have the same meaning as "A, B or C" above. "One or more of A, B and C" have the same meaning as "A, B or C" above.

[0226] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described or recorded in detail in a particular embodiment, reference can be made to the relevant descriptions of other embodiments. In addition, the above embodiments can be freely combined as needed. Although the embodiments of the present disclosure are disclosed above, the present disclosure is not limited thereto. Any person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present disclosure.

Claims

1. A modeling method for a material removal rate prediction model, characterized in that: The material removal rate prediction model is applicable to the material removal rate prediction of the chemical mechanical polishing process, and the modeling method includes: Establishing a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate; Inputting at least part of the samples in the first sample set into a neural network model to be trained to obtain a trained first model; Establishing a second sample set based on second production line data of a second semiconductor device processing production line and its corresponding material removal rate; Fixing at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training to obtain a trained second model, and using the second model as a material removal rate prediction model for the second semiconductor device processing line; The chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

2. The method according to claim 1, characterized in that The number of samples in the second sample set is smaller than the number of samples in the first sample set.

3. The method according to claim 2, characterized in that The fixing of at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training, and obtaining a trained second model includes: adjusting at least some parameters of the first model; At least part of the samples in the second sample set are input into the first model with adjusted parameters to train, to obtain a trained second model.

4. The method according to claim 1, wherein The first production line data includes: first-type characteristic parameters, the first-type characteristic parameters being suitable for characterizing a processing environment when a chemical mechanical polishing process is performed on the first production line; The second production line data includes: first-type characteristic parameters, where the first-type characteristic parameters are suitable for characterizing a processing environment when a chemical mechanical polishing process is performed on the second production line.

5. The method according to claim 4, characterized in that The first type of characteristic parameters includes at least one of the following: The pressure applied by the polishing head on different areas of the semiconductor device, the retainer pressure, the polishing pad speed, the polishing head speed, the polishing head swing trajectory, the dresser swing trajectory, the dresser speed, the dresser downforce, the polishing fluid flow rate, and the incoming semiconductor device information.

6. The method according to claim 4, characterized in that The first production line data also includes: The second type of characteristic parameters is suitable for characterizing the usage of the polishing consumables when the chemical mechanical polishing process is performed on the first production line; and / or The third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the first production line; The second production line data also includes: The second type of characteristic parameters is suitable for characterizing the usage of the polishing consumables when the chemical mechanical polishing process is performed on the second production line; and / or The third type of characteristic parameters is suitable for characterizing the material properties of the polishing consumables when the chemical mechanical polishing process is performed on the second production line.

7. The method according to claim 6, characterized in that The step of establishing a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate includes: labeling the first type of feature parameters with a first label; labeling the second type of feature parameters with a second label; and / or labeling the third type of feature parameters with a third label; The establishing of a second sample set based on the second production line data of the second semiconductor device processing production line and the corresponding material removal rate includes: labeling the first type of feature parameters with a first label; labeling the second type of feature parameters with a second label; and / or A third label is attached to the third type of feature parameters.

8. The method according to claim 7, characterized in that Inputting at least part of the samples in the first sample set into the neural network model to be trained for training to obtain a trained first model includes: assigning a first static weight to the first type of feature parameter according to the first label; assigning a second static weight to the second type of feature parameter according to the second label; and / or assigning a third static weight to the third type of feature parameter according to the third label; Passing the first static weight, the second static weight and / or the third static weight to the loss function of the neural network model; The fixing of at least some parameters of the first model, inputting at least some samples in the second sample set into the first model for training, and obtaining a trained second model includes: assigning a first static weight to the first type of feature parameter according to the first label; assigning a second static weight to the second type of feature parameter according to the second label; and / or assigning a third static weight to the third type of feature parameter according to the third label; Passing the first static weight, the second static weight and / or the third static weight to the loss function of the first model; The second static weight and the third static weight are both greater than the first static weight.

9. The method according to claim 7, characterized in that The second type of characteristic parameters includes at least one of the following: The number of times the polishing pad is used, the number of times the dresser is used, the number of times the retainer is used, the number of times the adsorption film is used, the use time of the polishing fluid, and the use time of the filter element; The third type of characteristic parameters includes at least one of the following: Polishing pad porosity, polishing pad hardness, polishing pad density, polishing pad pore size, polishing pad elastic modulus, slurry abrasive particle size, slurry abrasive particle volume fraction and the hardness of the removed material.

10. A modeling device for a material removal rate prediction model, characterized in that: include: A first sample set establishing module configured to establish a first sample set based on first production line data of a first semiconductor device processing production line and its corresponding material removal rate; A first model training module is configured to input at least part of the samples in the first sample set into the neural network model to be trained to obtain a trained first model; A second sample set establishing module configured to establish a second sample set based on second production line data of a second semiconductor device processing production line and its corresponding material removal rate; a second model training module configured to fix at least some parameters of the first model, input at least some samples in the second sample set into the first model for training, and obtain a trained second model, wherein the second model serves as a material removal rate prediction model for the second semiconductor device processing line; The chemical mechanical polishing process equipment of the first semiconductor device processing line and the second semiconductor device processing line are different.

11. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for modeling a material removal rate prediction model according to any one of claims 1 to 9 are executed.

12. A computer program product comprising a computer program, characterized in that When the computer program is running, the steps of the method for modeling a material removal rate prediction model according to any one of claims 1 to 9 are executed.

Images