Method and device for estimating available time of semiconductor equipment, storage medium and electronic equipment

Abstract

The invention discloses a semiconductor equipment available time estimation method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring product demand data of the semiconductor equipment; based on the obtained product demand data, determining an available time estimation model corresponding to the semiconductor device; inputting the obtained product demand data into an available time estimation model corresponding to the semiconductor equipment, and estimating the available time of the semiconductor equipment; wherein the available time estimation model is obtained by training an initial available time model by utilizing historical maintenance related data and historical downtime related data of the semiconductor equipment under the condition that the product demand data has the same product combination with the product demand data. By adopting the scheme, the available time of the semiconductor equipment can be accurately estimated.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and more specifically to a method, apparatus, storage medium, and electronic device for estimating the availability time of semiconductor devices. Background Technology

[0002] Semiconductor equipment is highly automated, and its availability directly impacts its output. In capacity calculations, assessments of semiconductor equipment output influence future investments in semiconductor equipment and whether customer demand can be met.

[0003] Current semiconductor equipment availability time calculations primarily rely on past performance as a reference for the future. However, in actual production, semiconductor equipment availability time is affected by many factors. Therefore, to more accurately calculate semiconductor equipment capacity for future product demand, a more precise estimate of semiconductor equipment availability time is needed. Summary of the Invention

[0004] The problem this invention aims to solve is: how to accurately estimate the availability time of semiconductor devices.

[0005] To address the above problems, embodiments of the present invention provide a method for estimating the availability time of a semiconductor device, the method comprising:

[0006] Obtain product demand data for semiconductor equipment;

[0007] Based on the acquired product demand data, determine the availability time prediction model for the semiconductor equipment.

[0008] The acquired product demand data is input into the availability time prediction model corresponding to the semiconductor equipment to predict the availability time of the semiconductor equipment.

[0009] The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

[0010] In one possible embodiment, the maintenance-related data includes: the number of preventive maintenance operations, the time required for preventive maintenance, and the maintenance time interval.

[0011] In one possible embodiment, the downtime-related data includes: the number of downtimes and the duration of downtime.

[0012] In one possible embodiment, the product demand data includes: product category identifiers, and key indicator values ​​for each product category.

[0013] In one possible embodiment, the product demand data may also include the output quantity of various types of products.

[0014] In one possible embodiment, the available time prediction model is a neural network model.

[0015] In one possible embodiment, the available time prediction model is trained using the following method:

[0016] Acquire several training samples regarding the semiconductor device, the training samples including: a total time interval, and product-related data, maintenance-related data, and downtime-related data within the total time interval;

[0017] The acquired training samples are input into the initial available time model to obtain the output error of the initial available time model;

[0018] When the output error does not meet the accuracy requirements, the average error of each layer of the initial available time model is calculated;

[0019] When the average error does not meet the accuracy requirements, the weights of each layer of the initial available time model are updated until the weights of each layer meet the accuracy requirements, thus obtaining the available time prediction model.

[0020] In one possible embodiment, before training the initial available time model using the acquired training samples, the method further includes:

[0021] The acquired training samples are normalized.

[0022] In one possible embodiment, the number of nodes in the neural network model is the same as the number of independent variables in the training samples.

[0023] This invention also provides a semiconductor device availability time estimation apparatus, the apparatus comprising:

[0024] Acquisition unit, suitable for acquiring product demand data for semiconductor equipment;

[0025] The determining unit is adapted to determine the availability time prediction model corresponding to the semiconductor device based on the acquired product demand data;

[0026] The estimation unit is adapted to input the acquired product demand data into the availability time estimation model corresponding to the semiconductor equipment, and to estimate the availability time of the semiconductor equipment.

[0027] The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

[0028] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the methods described above.

[0029] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the steps of any of the methods described above when running the computer program.

[0030] Compared with the prior art, the technical solution of the embodiments of the present invention has the following advantages:

[0031] The solution of this invention uses historical maintenance and downtime data of the semiconductor equipment under the same product combination as the product demand data to train the initial availability time model. Therefore, when product demand changes, different availability time prediction models can be used to predict the availability time of the semiconductor equipment. Each availability time prediction model is trained using historical maintenance and downtime data. Given the product combination, maintenance and downtime of the semiconductor equipment account for more than 80% of all unavailable time. Therefore, the solution of this invention can dynamically and more accurately predict the availability time of semiconductor production equipment based on product demand, which facilitates more accurate calculation of semiconductor equipment capacity. Attached Figure Description

[0032] Figure 1 This is a flowchart of a semiconductor device availability time estimation method according to an embodiment of the present invention;

[0033] Figure 2 This is a flowchart of a training method for an available time prediction model in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram showing the time distribution of a semiconductor device within a total time interval.

[0035] Figure 4 This is a schematic diagram of the training process of an available time prediction model in one embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram of a semiconductor device availability time prediction device according to an embodiment of the present invention. Detailed Implementation

[0037] In current semiconductor equipment availability calculations, past performance of semiconductor equipment is primarily relied upon as a reference for the future. Specifically, a rolling average is taken based on the performance of semiconductor equipment over the past year as a guide for the future.

[0038] However, in actual production, the inventors discovered that the product mix in semiconductor equipment manufacturing leads to variations in the maintenance frequency of semiconductor production equipment. Further research revealed that the uptime of semiconductor equipment is also affected by both scheduled maintenance and unscheduled downtime. Therefore, relying solely on past performance as a reference for the future is usually insufficient to accurately predict the uptime of semiconductor equipment, resulting in insufficient output and inability to meet order demands.

[0039] To address this problem, the present invention provides a method for predicting the availability time of semiconductor equipment. Using this method, given product demand, the availability time of semiconductor production equipment can be predicted more accurately using the availability time prediction model corresponding to the semiconductor equipment, which facilitates more accurate calculation of the semiconductor equipment's production capacity.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Reference Figure 1 This invention provides a method for predicting the availability time of a semiconductor device, the method comprising the following steps:

[0042] Step 11: Obtain product demand data for semiconductor equipment.

[0043] In some embodiments, the product demand data may include: product category identifiers, and key indicator values ​​for each product category.

[0044] In practice, the maintenance interval of semiconductor equipment is closely related to the type of products manufactured and their key performance indicators. For example, taking furnace tube area equipment as an example, the frequency of maintenance for furnace tube area equipment increases because a certain product requires a thicker oxide film to grow.

[0045] In practical implementation, products can be categorized according to their functions. For example, products can be divided into logic products, storage products, and computing products. Each product category is pre-coded to obtain numerical data for each category. For instance, when the product combination sequence is [1, 3, 4, 5] (i.e., product 1, product 3, product 4, and product 5), the corresponding key indicator sequence K is [K1, K3, K4, K5], where K1 is the key indicator value for product 1, K3 is the key indicator value for product 3, K4 is the key indicator value for product 4, and K5 is the key indicator value for product 5.

[0046] In practice, each type of product has key indicators that affect equipment uptime. Taking furnace tube equipment as an example, for oxide film growth equipment, the key indicator is the thickness of the grown oxide film. Each type of product may have only one key indicator affecting equipment uptime, or two or more may exist simultaneously. Each key indicator has a corresponding value.

[0047] In some embodiments, the product demand data may also include the sum of the output quantities of various product types. For example, when semiconductor equipment is produced continuously, the frequency of maintenance is higher, resulting in a reduction in the available time of the semiconductor equipment.

[0048] Step 12: Based on the acquired product demand data, determine the availability time prediction model corresponding to the semiconductor device.

[0049] The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

[0050] In practice, the same semiconductor device may have different availability time estimation models under different product combinations. In other words, a single semiconductor device may have multiple availability time estimation models, each corresponding one-to-one with a product combination. Here, "product combination" refers to the combination of product types in the product demand data. Once the product demand data is determined, the product combination is also determined, from which the availability time estimation model for that semiconductor device under the given product combination can be determined.

[0051] In some embodiments, the maintenance-related data may include: the number of preventive maintenance operations, the time required for each preventive maintenance operation, and the maintenance time interval.

[0052] Specifically, the maintenance-related data refers to maintenance-related data within the total time interval. The total time interval refers to the historical duration of the maintenance-related data; for example, the maintenance-related data may be within the last three months from the current date. Preventive maintenance refers to maintenance performed on semiconductor equipment in advance to avoid downtime; it is also called servicing and is typically performed periodically. The number of preventive maintenance operations refers to the number of preventive maintenance operations that occur within the total time interval. The time required for preventive maintenance refers to the average time required for each preventive maintenance operation within the total time interval. The maintenance time interval refers to the average time interval between two consecutive preventive maintenance operations within the total time interval.

[0053] In some embodiments, the downtime-related data includes: the number of downtimes and the duration of downtime.

[0054] Specifically, the downtime-related data refers to maintenance-related data within the total time interval. This total time interval is the same as the total time interval corresponding to the maintenance-related data; for example, both could be 3 months from the current date. The number of downtime occurrences refers to the number of times a downtime occurs within the total time interval. The downtime duration, including downtime repair time, refers to the average duration of each downtime occurrence within the total time interval.

[0055] Research has found that routine preventative maintenance and unpredictable downtime account for over 80% of all unavailability time for semiconductor equipment. Therefore, by combining maintenance and downtime data, the resulting availability time prediction model can analyze the impact of preventative maintenance and downtime on semiconductor equipment availability, thus enabling accurate predictions of equipment uptime.

[0056] Step 13: Input the obtained product demand data into the availability time prediction model corresponding to the semiconductor equipment to predict the availability time of the semiconductor equipment.

[0057] In practice, since the available time model is trained using maintenance-related data and downtime-related data as independent variables and available time as the dependent variable, the available time of the semiconductor equipment can be estimated by inputting the product demand data into the available time prediction model corresponding to the semiconductor equipment, thereby enabling a more accurate calculation of the semiconductor equipment's production capacity.

[0058] In specific implementations, the available time prediction model can be implemented using various machine learning models. There are no restrictions here, as long as maintenance-related data and downtime-related data are used as independent variables and available time is used as the dependent variable.

[0059] Figure 2 This is a flowchart illustrating a method for training a usable time prediction model according to an embodiment of the present invention. (Refer to...) Figure 2 The available time prediction model can be a neural network model. For a neural network model under a product portfolio, the training method can include the following steps:

[0060] Step 21: Obtain several training samples related to the semiconductor device.

[0061] The training samples include: a total time interval, and product-related data, maintenance-related data, and downtime-related data within the total time interval.

[0062] In practical implementation, assuming the available time of semiconductor equipment is U, the total time interval is O, the number of preventive maintenance is A, the maintenance time length is L, the number of downtimes is B, the downtime length is E, the maintenance time interval is D, the product portfolio sequence is C, the key indicator sequence is K, and the output quantity is Q, then the extracted mathematical model can be expressed as: U = (O, A, B, D, L, E, C, K, Q).

[0063] For example, the time distribution of a certain semiconductor device within the total time interval is as follows: Figure 3 As shown, refer to Figure 3 PM1, PM2, and PM3 represent preventative maintenance, and Down represents downtime. This means that within 108 days, there were three preventative maintenance operations and one downtime. Specifically, after two preventative maintenance operations (PM1 and PM2), a downtime occurred; after the downtime was repaired, a third preventative maintenance operation (PM3) was performed. The time interval between the first two preventative maintenance operations was 30 days, the time interval between the second preventative maintenance operation (PM2) and the downtime was 40 days, and the time interval between the downtime and the third preventative maintenance operation was 20 days.

[0064] Assuming the preventative maintenance period is 5 days, the downtime is 3 days, the product combination sequence C is [1, 3, 4, 5], the corresponding key indicator sequence K is [20, 20, 25, 15], and Q is 500, then based on the constructed mathematical model, the training sample can be extracted as: 83% = (108, 3, 1, 30, 5, 3, "1, 3, 4, 5", "20, 20, 25, 15", 500).

[0065] In specific implementation, refer to Figure 4 When obtaining training samples from a large amount of historical data, the specific steps may include the following:

[0066] 1) Extract feature variables;

[0067] Specifically, variables such as total time interval, number of preventive maintenance operations, maintenance duration, number of downtimes, downtime duration, maintenance time interval, product type identification, key indicator values ​​for various product types, and output quantity are extracted from a large amount of historical data.

[0068] 2) Collect basic data;

[0069] Specifically, the collected data is sorted according to the identification (ID) of the semiconductor device, and the selected data time interval (i.e., the total time interval) is determined. Then, the number of samples is selected according to the systematic sampling method. Non-numerical data in the samples are encoded to obtain the basic data.

[0070] Before encoding the selected samples, the samples can be cleaned and organized to ensure data integrity and consistency. Then, a label encoding method is used to convert non-numerical data (product types) into numerical data. For example, logical products can be encoded as "1", storage products as "2", and so on.

[0071] Step 22: Input the acquired training samples into the initial available time model to obtain the output error of the initial available time model.

[0072] In practice, a portion of the acquired samples can be used as training samples, and the remainder as test samples. For example, if 300 sets of data are collected, 200 sets can be used as training samples, and the remaining 100 sets as test samples. The test samples can be used to test the final available time prediction model to evaluate its performance.

[0073] In one embodiment, before training the initial available time model using the acquired training samples, the numerical data can be normalized to accelerate the training process and improve the model's accuracy. This data normalization can employ scaling normalization (also known as deviation normalization), which processes the numerical data into a reasonable range between 0 and 1. The normalized numerical data can be any numerical data in the training samples other than product category identifiers and key indicator values.

[0074] In specific implementation, refer to Figure 4 When constructing the initial available time model, it is necessary to set various parameters of the initial available time model.

[0075] Specifically, the initial available time model may include an input layer, a hidden layer, and an output layer. The number of nodes in the input layer of the initial available time model is the same as the number of independent variables in the training samples. For example, taking U = (O, A, B, D, L, E, C, K, Q) as an example, the initial available time model input layer has 9 nodes.

[0076] In practical implementation, the number of hidden layers in the initial available time model is crucial for the available time prediction model. Too few hidden layers will affect the accuracy of the available time prediction model, while too many hidden layers will increase the training time. In one embodiment, the initial available time model has one hidden layer with five nodes per layer, thus balancing the accuracy and training time of the available time prediction model.

[0077] In practice, the number of nodes in the initial available time model output layer can be set to 1, such as... Figure 4As shown, the output layer is used to output the device availability time U. The activation function can be a ReLU function, and the average error is used as the loss function for evaluation.

[0078] After constructing the network structure of the initial available time model, the initial available time model is initialized so that each layer in the initial available time model has initial weights and initial gradient vectors. Then, the acquired training samples are input into the initial available time model to obtain the prediction result of the initial available time model. The difference between the initial available time model and the actual available time of the semiconductor device is calculated to obtain the output error of the initial available time model.

[0079] Step 23: Determine whether the output error meets the accuracy requirements.

[0080] In specific implementation, when the output error of the initial available time model is greater than the preset first error threshold, it is determined that the output error of the initial available time model does not meet the accuracy requirements; otherwise, it is determined that the output error of the initial available time model meets the accuracy requirements.

[0081] If the output error does not meet the accuracy requirements, proceed to step 24.

[0082] Step 24: Calculate the average error of each layer of the initial available time model.

[0083] In practical implementation, when the output error does not meet the accuracy requirements, the predicted results of each layer of the initial available time model can be obtained, and the predicted results of each layer can be compared with the actual values ​​of that layer to obtain the error values ​​of each layer of the initial available time model. Then, the error values ​​of each layer of the initial available time model are averaged to obtain the average error of each layer of the initial available time model.

[0084] Step 25: Determine whether the average error meets the accuracy requirements.

[0085] In practice, the average error of each layer of the initial available time model can be compared with a preset second error threshold. If the average error of each layer of the initial available time model is greater than the preset second error threshold, it is determined that the average error of each layer of the initial available time model does not meet the accuracy requirements; otherwise, it is determined that the average error of each layer of the initial available time model meets the accuracy requirements.

[0086] If the average error of each layer of the initial available time model does not meet the accuracy requirements, proceed to step 26.

[0087] Step 26: Update the weights of each layer in the initial available time model.

[0088] In practice, the weights of each layer of the initial available time model are updated to obtain the updated available time model. The output error of the updated available time model is then recalculated with the updated weights, and it is determined whether the average error meets the accuracy requirements. If the accuracy requirements are not met, step 26 is executed until the average error of each layer of the updated available time model meets the accuracy requirements.

[0089] In practice, backpropagation and conjugate gradient methods can be used to update the weights of each layer in the neural network model, and the changes in the loss function can be monitored during training. Additionally, the available time of test samples can be used to estimate the model's performance, for example, by calculating metrics such as the accuracy and mean squared error of the prediction results, thus avoiding overfitting.

[0090] Taking a diffusion region oxide film growth device with oxide film thickness as an example, the present invention is used to obtain the available time prediction model of the diffusion region oxide film growth device under various product combinations. Then, referring to Table 1, 300 sets of historical data of the oxide film growth device are obtained and input into the corresponding available time prediction model to obtain the predicted output of the oxide film growth device.

[0091] Table 1

[0092]

[0093] The predicted output of the oxide film growth equipment was compared with the actual output, and the average error was calculated. As can be seen from Table 1, the average error between the predicted output and the actual output of the oxide film growth equipment obtained by using the available time prediction model of the present invention is small, indicating that the available time prediction model has high accuracy in predicting available time.

[0094] As can be seen from the above, the solution in this embodiment of the invention constructs an availability time prediction model based on maintenance-related data and downtime-related data. Under a given product combination, it predicts the future availability time range of semiconductor equipment. Based on this data, a capacity assessment is then performed, thereby enabling a more accurate assessment of the capacity requirements of semiconductor computing equipment.

[0095] To enable those skilled in the art to better understand and implement the present invention, the apparatus, testing system, electronic device and computer-readable storage medium corresponding to the above method are described in detail below.

[0096] Reference Figure 5 This invention also provides a semiconductor device availability time estimation device 50, which may include: an acquisition unit 51, a determination unit 52, and an estimation unit 53. Wherein:

[0097] The acquisition unit 51 is adapted to acquire product demand data for semiconductor equipment.

[0098] The determining unit 52 is adapted to determine the available time prediction model corresponding to the semiconductor device based on the acquired product demand data.

[0099] The estimation unit 53 is adapted to input the acquired product demand data into the availability time estimation model corresponding to the semiconductor equipment, and to estimate the availability time of the semiconductor equipment.

[0100] The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

[0101] The acquisition unit 51, determination unit 52 and estimation unit 53 can be implemented with reference to the description of the corresponding steps in the above-mentioned method for estimating the availability time of semiconductor devices, and will not be repeated here.

[0102] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the above methods.

[0103] In specific implementations, the computer-readable storage medium may include ROM, RAM, disk, or optical disk, etc.

[0104] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor runs the computer program, it performs the steps of any of the methods described above.

[0105] Regarding the modules / units included in the various devices and products described in the above embodiments, they can be software modules / units, hardware modules / units, or a combination of both. For example, for various devices and products applied to or integrated into a chip, all of their modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits; for various devices and products applied to or integrated into a chip module, all of their modules / units can be implemented using hardware methods such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The components can be implemented using software programs that run on the processor integrated within the chip module. The remaining (if any) modules / units can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into the terminal, each of its components / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or in different components within the terminal. Alternatively, at least some modules / units can be implemented using software programs that run on the processor integrated within the terminal, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits.

[0106] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for estimating the availability time of a semiconductor device, characterized in that, include: Obtain product demand data for semiconductor equipment; Based on the acquired product demand data, determine the availability time prediction model for the semiconductor equipment. The acquired product demand data is input into the availability time prediction model corresponding to the semiconductor equipment to predict the availability time of the semiconductor equipment. The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

2. The semiconductor device availability time estimation method as described in claim 1, characterized in that, The maintenance-related data includes: the number of preventive maintenance operations, the time required for preventive maintenance, and the maintenance interval.

3. The semiconductor device availability time estimation method as described in claim 1, characterized in that, The downtime-related data includes: the number of downtimes and the duration of downtime.

4. The semiconductor device availability time estimation method as described in claim 1, characterized in that, The product demand data includes: product category identifiers, and key indicator values ​​for each product category.

5. The semiconductor device availability time estimation method as described in claim 4, characterized in that, The product demand data also includes the output quantity of various types of products.

6. The semiconductor device availability time estimation method as described in claim 1, characterized in that, The available time prediction model is a neural network model.

7. The semiconductor device availability time estimation method as described in claim 6, characterized in that, The available time prediction model was trained using the following method: Acquire several training samples related to the semiconductor device, the training samples including: a total time interval, and product-related data, maintenance-related data, and downtime-related data within the total time interval; The acquired training samples are input into the initial available time model to obtain the output error of the initial available time model; When the output error does not meet the accuracy requirements, the average error of each layer of the initial available time model is calculated; When the average error does not meet the accuracy requirements, the weights of each layer of the initial available time model are updated until the weights of each layer meet the accuracy requirements, thus obtaining the available time prediction model.

8. The semiconductor device availability time estimation method as described in claim 7, characterized in that, Before training the initial available time model using the acquired training samples, the method further includes: The acquired training samples are normalized.

9. The semiconductor device availability time estimation method as described in claim 7, characterized in that, The number of nodes in the neural network model is the same as the number of independent variables in the training samples.

10. A semiconductor device availability time estimation device, characterized in that, include: Acquisition unit, suitable for acquiring product demand data for semiconductor equipment; The determining unit is adapted to determine the availability time prediction model corresponding to the semiconductor device based on the acquired product demand data; The estimation unit is adapted to input the acquired product demand data into the availability time estimation model corresponding to the semiconductor equipment, and to estimate the availability time of the semiconductor equipment. The availability time prediction model is obtained by training the initial availability time model using historical maintenance data and historical downtime data of the semiconductor equipment under the same product combination as the product demand data.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the steps of the method according to any one of claims 1 to 9.

12. An electronic device comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the method according to any one of claims 1 to 9.

Images