Chamber temperature monitoring method and semiconductor processing apparatus

Abstract

This application relates to the field of semiconductor manufacturing technology, and provides a method for monitoring chamber temperature rise and semiconductor process equipment. The method includes: acquiring real-time data of relevant variables affecting the temperature rise process of the process chamber, and acquiring the real-time actual temperature value of the heating component in the process chamber, wherein the relevant variables include continuous variables and digital variables, and the digital variables include the water flow status and temperature control mode of the process chamber; inputting the real-time data of the relevant variables into a pre-trained temperature prediction model, and outputting the real-time temperature prediction value of the heating component through the temperature prediction model, wherein the temperature prediction model is trained based on the historical actual temperature values ​​of the heating component and the historical data of the relevant variables; and obtaining the real-time monitoring result of the process chamber temperature rise process by comparing the real-time actual temperature value of the heating component with the real-time temperature prediction value of the heating component, thereby improving the accuracy of the monitoring result of the process chamber temperature rise process.

Description

Technical Field

[0001] This specification relates to the field of semiconductor manufacturing technology, and in particular to a method for monitoring chamber temperature rise and semiconductor process equipment. Background Technology

[0002] In semiconductor manufacturing, when processing semiconductors in a process chamber, such as when processing wafers, it is often necessary to maintain the process chamber at a high temperature. Consequently, monitoring the temperature rise process within the process chamber is crucial. Currently, conventional process chamber temperature rise monitoring primarily relies on monitoring critical setpoints for heating parameters such as the power and current of the heating components. These critical setpoints are mainly manually set based on experience and experimental data, and different critical values ​​have varying impacts on chamber temperature monitoring. Monitoring the process chamber temperature rise process based on these critical setpoints can yield inaccurate results, affecting the normal operation of the process chamber and potentially leading to component damage. Summary of the Invention

[0003] To address the aforementioned technical problems, this application provides a method for monitoring chamber temperature rise and a semiconductor process equipment to improve the accuracy of monitoring results during the process chamber temperature rise process.

[0004] To achieve the above technical objectives, the embodiments of this application provide the following technical solutions:

[0005] In a first aspect, embodiments of this specification provide a method for monitoring the temperature rise of a chamber, applied to a process chamber, the process chamber including a heating component; the method includes: acquiring real-time data of relevant variables affecting the temperature rise process of the process chamber, and acquiring the real-time actual temperature value of the heating component in the process chamber, wherein the relevant variables include continuous variables and digital variables, the digital variables including the water flow status and temperature control mode of the process chamber; inputting the real-time data of the relevant variables into a pre-trained temperature prediction model, and outputting the real-time temperature prediction value of the heating component through the temperature prediction model, wherein the temperature prediction model is trained based on the historical actual temperature value of the heating component and the historical data of the relevant variables; and obtaining the real-time monitoring result of the temperature rise process of the process chamber by comparing the real-time actual temperature value of the heating component with the real-time temperature prediction value of the heating component.

[0006] According to the chamber temperature monitoring method provided in the first aspect, after obtaining the real-time monitoring result of the process chamber temperature rise process by comparing the real-time actual temperature value of the heating component with the real-time predicted temperature value of the heating component, the method further includes: if the real-time monitoring result indicates that the process chamber temperature rise is normal, updating the real-time data of the relevant variables to the historical data of the relevant variables, and updating the real-time actual temperature value of the heating component to the historical actual temperature value of the heating component; wherein, the updated historical actual temperature value of the heating component and the updated historical data of the relevant variables are used to retrain the temperature prediction model.

[0007] According to the chamber temperature monitoring method provided in the first aspect, before training the temperature prediction model, the historical data of the relevant variables and the actual historical temperature values ​​are divided into a training set and a validation set according to a preset ratio. The training process of the temperature prediction model is as follows: the historical data of the relevant variables in the training set are input into a preset original prediction model, and the training temperature prediction value of the heating component is output through the original prediction model, wherein the original prediction model is pre-configured based on a long short-term memory neural network; the training temperature prediction value of the heating component is compared with the actual historical temperature value of the heating component to see if it is less than the training threshold; if so, the parameters in the original prediction model are tested and optimized using the historical data of the relevant variables in the validation set, and the original prediction model is determined as the temperature prediction model; if not, the internal parameters of the original prediction model are adjusted, and the original prediction model is retrained until the temperature prediction model is obtained.

[0008] According to the chamber temperature monitoring method provided in the first aspect, the temperature prediction model includes an input layer, a first computation layer, a second computation layer, a fully connected layer, a random drop-out layer, and an output layer; the input layer, the first computation layer, the second computation layer, the fully connected layer, the random drop-out layer, and the output layer are connected in sequence.

[0009] According to the chamber temperature monitoring method provided in the first aspect, the step of inputting the real-time data of the relevant variables into a pre-trained temperature prediction model and outputting the real-time temperature prediction value of the heating component through the temperature prediction model includes: inputting the real-time data of the relevant variables into the input layer; inputting the intermediate data output by the input layer into the first computation layer, where the first computation layer performs a first data processing transformation; inputting the intermediate data output by the first computation layer into the second computation layer, where the second computation layer performs a second data processing transformation; inputting the intermediate data output by the second computation layer into the fully connected layer, where the fully connected layer performs regularization processing; inputting the intermediate data output by the fully connected layer into the random dropout layer, where neurons are randomly dropped according to a preset ratio during operation; and inputting the intermediate data output by the random dropout layer into the output layer, where the output layer performs regularization processing and outputs the real-time temperature prediction value of the heating component.

[0010] According to the chamber heating monitoring method provided in the first aspect, the step of acquiring real-time data of relevant variables during the heating process of the process chamber includes: collecting real-time data of relevant variables during the heating process of the process chamber; standardizing the real-time data of the continuous variables and performing unique thermal encoding on the real-time data of the numerical variables to obtain standardized data of the relevant variables; the step of inputting the real-time data into a pre-trained temperature prediction model and outputting the real-time temperature prediction value of the heating component through the temperature prediction model includes: inputting the standardized data of the relevant variables into the pre-trained temperature prediction model and outputting the real-time temperature prediction value of the heating component through the temperature prediction model.

[0011] According to the chamber temperature monitoring method provided in the first aspect, the step of obtaining a real-time monitoring result of the process chamber temperature rise process by comparing the actual real-time temperature value of the heating element with the predicted real-time temperature value of the heating element includes: calculating the difference between the actual real-time temperature value of the heating element and the predicted real-time temperature value of the heating element; if the difference is less than or equal to a preset temperature threshold, generating a real-time monitoring result indicating that the process chamber temperature rise is normal; if the difference is greater than the temperature threshold, generating a real-time monitoring result indicating that the process chamber temperature rise is abnormal.

[0012] According to the chamber temperature monitoring method provided in the first aspect, the historical data of the relevant variables are the historical time series data of the relevant variables, and the historical time series data include the historical data of the relevant variables corresponding to multiple consecutive historical moments during the historical normal temperature rise process of the process chamber.

[0013] According to the chamber temperature monitoring method provided in the first aspect, the continuous variables include one or more of the following: heating wire inner coil power, heating wire outer coil power, heating wire inner coil current, and heating wire outer coil current; the digital variables include the chamber water flow status and / or the chamber temperature control mode.

[0014] Secondly, embodiments of this specification provide a semiconductor process apparatus, comprising: a process chamber including a heating element; an upper electrode assembly located above the process chamber; a lower electrode assembly located within the process chamber for carrying a wafer and applying a bias voltage to the wafer; and a controller including at least one memory and at least one processor, the memory for storing a computer program; and the processor for implementing the chamber temperature monitoring method as described above by running the computer program stored in the memory.

[0015] According to the semiconductor process equipment provided in the second aspect, the process chamber is a physical vapor deposition process chamber, and the heating component includes an electrostatic chuck; the semiconductor process equipment further includes a power controller, which is used to determine a temperature control mode and transmit power control parameters under the temperature control mode to the power controller; the power controller is used to adjust the real-time temperature of the heater based on the power control parameters, and the temperature of the heater is conducted to the electrostatic chuck.

[0016] Thirdly, embodiments of this specification provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the chamber temperature monitoring method described above.

[0017] Fourthly, embodiments of this specification provide a computer program product or computer program, the computer program product including a computer program stored in a computer-readable storage medium; the processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program to implement the above-described chamber temperature monitoring method.

[0018] As can be seen from the above technical solutions, the embodiments of this application provide a chamber temperature rise monitoring method and semiconductor process equipment. The chamber temperature rise monitoring method inputs real-time data of relevant variables during the process chamber temperature rise process into a pre-trained temperature prediction model to obtain the real-time temperature prediction value of the heating component output by the temperature prediction model. Then, by comparing the actual real-time temperature value of the heating component with the real-time temperature prediction value, the real-time monitoring result of the process chamber temperature rise process is obtained. This temperature prediction model is trained based on the historical actual temperature values ​​of the heating component and historical data of relevant variables, realizing an intelligent monitoring process for chamber temperature rise, avoiding the uncertainty caused by manually setting threshold values, and improving the accuracy of the monitoring results for the process chamber temperature rise process. Furthermore, the relevant variables of the process chamber temperature rise process in this application include not only continuous variables used in traditional methods, but also digital variables such as the water flow status and temperature control mode of the process chamber. Digital variables are more stable and less susceptible to interference, improving the stability of the chamber temperature rise monitoring process. At the same time, the water flow status and temperature control mode can characterize the actual state of the process chamber itself, making the monitoring process of the process chamber more comprehensive and further improving the accuracy of the temperature rise process monitoring results. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 A schematic diagram of the PVD Al chamber structure provided for existing technology;

[0021] Figure 2 One of the schematic diagrams illustrating the changes in relevant variables during abnormal chamber temperature rise in existing technologies;

[0022] Figure 3 The second schematic diagram illustrating the changes in relevant variables during abnormal chamber temperature rise, provided for existing technologies;

[0023] Figure 4 A schematic diagram of the process steps for a chamber temperature monitoring method provided in one embodiment of this specification;

[0024] Figure 5 A schematic diagram of the temperature prediction model structure provided for one embodiment of this specification;

[0025] Figure 6 A schematic diagram illustrating the process from model training to model usage of the chamber temperature monitoring method provided in one embodiment of this specification;

[0026] Figure 7 This is a schematic diagram of a semiconductor process apparatus provided for one embodiment of this specification. Detailed Implementation

[0027] Unless otherwise defined, the technical or scientific terms used in the embodiments of this specification shall have the ordinary meaning understood by one of ordinary skill in the art to which this specification pertains. The terms "first," "second," and similar terms used in the embodiments of this specification do not indicate any order, quantity, or importance, but are merely used to avoid confusion of constituent elements.

[0028] Unless the context otherwise requires, throughout this specification, "a plurality of" means "at least two," and "including" is interpreted as open-ended or encompassing, that is, "including, but not limited to." In the description of this specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this specification. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example.

[0029] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0030] Overview

[0031] As described in the background section, during the fabrication of semiconductor devices, it is often necessary to maintain a high temperature within the process chamber when fabricating semiconductors. In particular, Physical Vapor Deposition (PVD) technology, primarily referring to Magnetron Sputtering, deposits thin films by bombarding a solid target with plasma. Primarily used for metal thin film deposition, it is a widely used thin film manufacturing process in the semiconductor industry. In the PVD process, metal thin film deposition is often accompanied by a certain high-temperature environment. The chamber uses a temperature controller to control the output of the heater, thereby achieving temperature control.

[0032] In the PVD process, the chamber heating process mainly involves two temperature control modes: Ramp mode and Servo mode. In Ramp mode, during the initial heating phase, when the actual temperature of the heating element deviates significantly from the setpoint, the temperature controller maintains a fixed heating rate (e.g., 3°C / min) to raise the temperature of the heating element. In Servo mode, during the heating process, when the actual temperature is about to reach the setpoint (e.g., a difference of 7°C), the controller switches from Ramp mode to Servo mode, and the temperature slowly approaches the setpoint. In the initial heating phase, when the setpoint and actual temperature differ significantly, the controller is in Ramp mode, controlling the temperature of the electrostatic chuck (ESC) in the process chamber at a fixed rate. As the actual ESC temperature gradually approaches the setpoint (i.e., the temperature difference is less than a certain fixed value), the controller switches to Servo mode, causing the ESC temperature to rise slowly.

[0033] Instance-like, such as Figure 1 Taking the PVD Al cavity shown as an example, the temperature controller determines the control parameters of the power controller under different temperature control modes based on the set temperature, the water flow status of the cavity, and the temperature difference between the actual temperature of the electrostatic chuck and the set value. This parameter acts on the inner and outer heating wires of the heater, causing them to heat up at the set power. The temperature is then conducted to the heating elements, including the ceramic disc ESC, via backblowing. The heat is then transferred to the silicon wafer (also known as a crystal wafer), achieving a fixed temperature increase per minute in the cavity. Furthermore, a water pipe connected to the outside is pre-laid in the heater, storing cooling water. The flow of this cooling water is controlled by valves. According to process requirements, when it is necessary to lower the temperature of the heater and electrostatic chuck, the flow of cooling water is controlled by the valves, thereby lowering the heater temperature. The high-temperature electrostatic chuck then conducts heat to the relatively low-temperature heater, cooling the electrostatic chuck and thus meeting the cooling requirements during wafer processing. The water flow status of the water pipe in the chamber can be obtained by monitoring the status of the water pipe control valve. When the valve is open, it indicates that the chamber is in a water-flow state, and when the valve is closed, it indicates that the chamber is not in a water-flow state.

[0034] During the heating process, because the ESC is made of ceramic and is relatively expensive, hardware failures in the heating circuit can easily lead to abnormal power output from the heating wire, causing a sudden rise or fall in the chamber temperature, which can result in the ceramic ESC disk shattering. Therefore, monitoring the status of the chamber heating process and promptly detecting any abnormalities is essential.

[0035] In existing technologies, the state monitoring during the chamber heating process mainly involves monitoring the interrelationships between several variables. Figure 1 Taking the PVD Al chamber as an example, during the heating process, the chamber mainly monitors variables such as the actual ESC temperature, the current and power of the inner heating wire, the current and power of the outer heating wire, etc. The temperature controller compares the actual ESC temperature with the set temperature value and determines the power controller setting value for the corresponding temperature control mode based on the water flow status of the chamber, thereby controlling the heating wire. When a fault occurs in the heating circuit, the output power and current of the inner and outer rings of the heater will deviate from the normal values, resulting in abnormal fluctuations.

[0036] like Figure 2 As shown, when a certain fault occurs, the power of both the inner and outer heating wires, displayed in the upper coordinate system, fluctuates between 40% and 60%, while the current of the outer heating wire, displayed in the lower coordinate system, suddenly drops from approximately 2.4A to approximately 0.2A. After the fault occurs, the actual temperature of the ESC drops abnormally. Based on this, the existing solution determines this type of abnormal temperature rise by monitoring whether a small current (less than 0.5A) appears when the inner or outer heating wire is at high power output (greater than or equal to 20%). 20% and 0.5A are pre-set threshold values.

[0037] like Figure 3 As shown, in another type of fault, the currents of the inner and outer heating wires, displayed in the lower coordinate system, first undergo a sudden change, with the currents frequently jumping between 0A and 5.6A within a short period. At this time, the power of the inner and outer heating wires, displayed in the upper coordinate system, has not yet begun to change. However, after a period of time, the power of the inner and outer heating wires rapidly decreases, and correspondingly, the actual temperature of the ESC drops abnormally. Therefore, by monitoring whether a small current (less than 0.5A) appears when the inner or outer heating wire is at high power output (greater than or equal to 20%), it can be determined whether the chamber temperature rise is abnormal. 20% and 0.5A are pre-set threshold values.

[0038] Based on the above, the conventional monitoring of the chamber heating process is determined by the relationship between the power and current of the inner and outer heating wires. When the power is maintained at a certain value, if the current is lower than a specific value, it is determined that the chamber heating is abnormal during that period, the machine alarms, and emergency cooling measures are taken.

[0039] However, conventional chamber heating process status monitoring relies on critical setpoints (e.g., 20% and 0.5A) for the power and current of the inner and outer heating wires to determine faults in the chamber heating circuit. These critical setpoints are primarily derived from experience and experimental data, and different values ​​have varying impacts on chamber status monitoring. If the critical setpoint is too small, the monitoring is overly sensitive, easily leading to false alarms and severely impacting chamber process efficiency. Conversely, if the critical setpoint is too large, the machine response is delayed when abnormal chamber heating occurs, potentially causing malfunctions such as ESC ceramic disc breakage. Furthermore, conventional chamber heating process status monitoring technology monitors current and power; when these signals are invalid or interfered with, it cannot detect abnormal temperature changes in the ESC when a fault occurs, resulting in low fault tolerance.

[0040] Based on the above analysis, this application provides a chamber temperature rise monitoring method and semiconductor process equipment. The chamber temperature rise monitoring method is based on a pre-trained temperature prediction model to achieve real-time monitoring of the chamber temperature rise process. It can train a temperature prediction model using historical actual temperatures of the heating element and historical data of relevant variables, and obtain a real-time temperature prediction value for the heating element based on real-time data of relevant variables. Then, by comparing the actual real-time temperature of the heating element with the predicted real-time temperature value, the real-time monitoring result of the process chamber temperature rise process is obtained. If the actual real-time temperature of the heating element deviates significantly from the predicted real-time temperature value, the real-time monitoring result indicates an abnormal temperature rise in the process chamber, indicating that the temperature of the heating element has seriously deviated from the original temperature trend. At this time, there may be a hardware fault, and the industrial control computer will issue an alarm and perform protective actions. If the actual real-time temperature of the heating element deviates slightly from the predicted real-time temperature value, the real-time monitoring result indicates that the temperature rise in the process chamber is normal, indicating that the real-time temperature value of the heating element is normal, and the process chamber can continue to the next step of operation.

[0041] Based on the above inventive concept, the chamber temperature monitoring method provided in the embodiments of this specification will be described exemplarily below.

[0042] Exemplary methods

[0043] This specification provides a method for monitoring the temperature rise of a process chamber. This method is applied to a process chamber to monitor the temperature rise process. The process chamber includes a heating element, which can be an electrostatic chuck within the chamber. A temperature controller configured in the process chamber can adjust the temperature control mode. Based on different temperature control models, the temperature controller adjusts the heating power of the inner and outer heating wires in the heater. The temperature of the heater changes accordingly, and the temperature conducted to the electrostatic chuck also changes, thereby adjusting the temperature of the entire process chamber. The chamber temperature rise monitoring method can be implemented within the temperature controller, enabling real-time monitoring of the process chamber temperature rise process.

[0044] like Figure 4 As shown, the methods for monitoring chamber temperature rise include:

[0045] Step 401: Obtain real-time data of relevant variables affecting the heating process of the process chamber, and obtain the real-time actual temperature value of the heating component in the process chamber. The relevant variables include continuous variables and digital variables. The digital variables include the water flow status and temperature control mode of the process chamber.

[0046] In this embodiment, when any variable changes during the heating process of the process chamber, it will affect the heating process. This variable is considered a relevant variable affecting the heating process, such as the power of the inner heating wire or the current of the outer heating wire. One or more relevant variables can be selected based on actual conditions and needs. When there are multiple relevant variables, each variable is collected in real-time using an appropriate method. For example, current is collected through a current sampling circuit, and power is collected using a power meter.

[0047] In this embodiment, the heating component in the process chamber refers to the main component that can raise the temperature inside the process chamber. For example, in the PVD process, the heating component can be such as... Figure 1 The electrostatic chuck described herein. The real-time actual temperature value of the heating components in the process chamber can also be collected in an appropriate manner according to the actual situation and needs; for example, it can be collected through a temperature sensor. Real-time data of relevant variables affecting the heating process of the process chamber, as well as the real-time actual temperature value of the heating components in the process chamber, are collected for subsequent processing.

[0048] In this embodiment, the process of collecting real-time data of relevant variables and the actual real-time temperature value of heating components can be continuously collected throughout the entire process of heating the process chamber according to a preset cycle. Each time data is collected, the following processing procedure is completed to obtain a real-time monitoring result, thereby realizing continuous monitoring of the heating process of the process chamber.

[0049] In this embodiment, to ensure a more comprehensive description of the process chamber heating process, the relevant variables include as many different types of variables as possible. These variables include continuous variables and numerical variables. Continuous variables refer to variables that change continuously over time, such as current, while numerical variables refer to variables that do not change continuously over time, such as the temperature control mode. The relevant variables for the process chamber heating process include not only continuous variables such as current and power used in traditional methods, but also numerical variables such as the water flow status and temperature control mode of the process chamber. Numerical variables are more stable and less susceptible to interference, improving the stability of the chamber heating monitoring process. Simultaneously, the water flow status and temperature control mode can characterize the actual state of the process chamber itself. The variety of variables, including continuous and numerical variables, increases the diversity of the basic data, enabling a more comprehensive characterization of the process chamber's heating state and resulting in more accurate heating process monitoring results.

[0050] Step 402: Input the real-time data of relevant variables into the pre-trained temperature prediction model, and output the real-time temperature prediction value of the heating component through the temperature prediction model. The temperature prediction model is trained based on the historical actual temperature value of the heating component and the historical data of relevant variables.

[0051] In this embodiment, real-time data of relevant variables are input into a pre-trained temperature prediction model. This model predicts the temperature of the heating element, outputting a real-time temperature prediction. The model is trained based on historical actual temperatures of the heating element and historical data of relevant variables. Using these historical data allows the model to more closely reflect the actual temperature changes of the heating element, improving the accuracy of the real-time temperature prediction and thus enhancing the accuracy of chamber temperature monitoring. Furthermore, by replacing manual setting of critical thresholds with a temperature prediction model, the intelligence of the chamber temperature monitoring process is improved, avoiding increased labor and time costs and increasing processing efficiency.

[0052] Step 403: By comparing the actual real-time temperature value of the heating component with the predicted real-time temperature value of the heating component, the real-time monitoring results of the heating process in the process chamber are obtained.

[0053] In this embodiment, after obtaining the real-time temperature prediction value of the heating component output by the temperature prediction model, the actual real-time temperature value of the heating component is compared with the real-time temperature prediction value to determine whether the heating process of the process chamber is normal at the current moment, thereby obtaining the real-time monitoring result of the heating process of the process chamber. Since the temperature prediction model is trained based on the historical actual temperature value of the heating component and historical data of related variables, it realizes the intelligent monitoring process of chamber heating, avoids the uncertainty caused by manually setting threshold values, and improves the accuracy of the monitoring results of the heating process of the process chamber.

[0054] In one embodiment, after obtaining the real-time monitoring result of the heating process chamber heating process by comparing the real-time actual temperature value of the heating component with the real-time predicted temperature value of the heating component, if the real-time monitoring result indicates that the heating of the process chamber is normal, the real-time data of the relevant variables are updated to the historical data of the relevant variables, and the real-time actual temperature value of the heating component is updated to the historical actual temperature value of the heating component; wherein, the updated historical actual temperature value of the heating component and the updated historical data of the relevant variables are used to retrain the temperature prediction model.

[0055] In this embodiment, to further improve monitoring accuracy, if it is determined that the process chamber temperature rise is normal based on the real-time data of the relevant variables and the actual real-time temperature of the heating component, the real-time data of the relevant variables can be updated to the historical data of the relevant variables, and the actual real-time temperature of the heating component can be updated to the historical actual temperature of the heating component. The temperature prediction model is then retrained using the updated historical actual temperature of the heating component and the updated historical data of the relevant variables. This allows the temperature prediction model to continuously learn as the process chamber temperature rises, making it more closely aligned with the actual situation of the current process chamber temperature rise. This further improves the accuracy of the real-time temperature prediction value of the heating component output by the temperature prediction model, thereby improving the accuracy of chamber temperature rise monitoring.

[0056] In this embodiment, if the real-time monitoring results indicate that the process chamber temperature rise is normal, the training data of the temperature prediction model can be updated using the newly obtained real-time data of the relevant variables and the actual real-time temperature value of the heating component. The updated historical actual temperature value of the heating component and the updated historical data of the relevant variables are then used to retrain the temperature prediction model, making the internal logic of the model more closely match the current process chamber temperature rise process. This significantly improves the accuracy of the real-time temperature prediction values ​​of the heating component output by the temperature prediction model, thereby improving the accuracy of the real-time monitoring results during the process chamber temperature rise process. Compared to the method of determining the chamber state by monitoring the relationship between the inner and outer circuits and power, monitoring the process chamber temperature rise using a continuously learned and trained temperature prediction model can quickly determine whether an abnormality has occurred during the process chamber temperature rise process, and it has strong anti-interference capabilities, making it less prone to false alarms.

[0057] In one embodiment, to facilitate the efficient operation of the temperature prediction model, preprocessing is performed on different types of variables to achieve data standardization. Specifically, real-time data of relevant variables during the heating process of the process chamber are acquired as follows: real-time data of relevant variables during the heating process of the process chamber are collected; real-time data of continuous variables are standardized, and real-time data of numerical variables are uniquely thermally encoded to obtain standardized data of relevant variables. When the real-time data is input into the pre-trained temperature prediction model, and the temperature prediction model outputs the real-time temperature prediction value of the heating component, the standardized data of relevant variables are input into the pre-trained temperature prediction model, and the temperature prediction model outputs the real-time temperature prediction value of the heating component.

[0058] In one embodiment, in the PVD process, preferably, continuous variables include one or more of the following: inner coil power of the heating wire, outer coil power of the heating wire, inner coil current of the heating wire, and outer coil current of the heating wire; digital variables include the chamber water flow status and / or the chamber temperature control mode. Of course, depending on the specific operating environment of the process chamber, the continuous and digital variables can also be selected from those more suitable for the actual operating environment.

[0059] In an exemplary embodiment, taking the PVD process as an example, the relevant variables include the inner coil power of the heating wire, the outer coil power of the heating wire, the inner coil current of the heating wire, the outer coil current of the heating wire, the water flow status of the chamber, and the chamber temperature control mode. Among them, the inner coil current (unit: A), the outer coil current (unit: A), the inner coil power (unit: W), and the outer coil power (unit: %) of the heating wire are continuous variables, while the water flow status of the chamber (0 indicates no water flow, 1 indicates water flow) and the chamber temperature control mode (0 indicates Servo mode, 1 indicates Ramp mode) are numerical variables.

[0060] In this embodiment, after real-time data of the aforementioned relevant variables are collected during the heating process of the process chamber, different preprocessing is performed on the real-time data according to the different data types.

[0061] For any continuous variable, to avoid overfitting the model, it is standardized using mean-variance (std). Specifically, based on historical data of the relevant variables used in model training, the minimum and maximum values ​​from the historical data are used to transform the real-time data into values ​​within the range of 0-1. The calculation formula is as follows:

[0062]

[0063] Where, x min x represents the minimum value in the historical data of the relevant variable. max represents the maximum value in the historical data of the relevant variable, x represents the real-time data of the relevant variable before standardization, and X represents the standardized data of the relevant variable after standardization.

[0064] For numerical variables (chamber water flow status and chamber temperature control mode), one-hot encoding is performed according to the combination method. One-hot encoding can solve the problem that the model cannot handle discrete data well, and it can also play a role in data augmentation features to a certain extent. The encoding rules are shown in Table 1 below:

[0065] Table 1. One-Hot Code Table Corresponding to Water Flow Status and Temperature Control Mode

[0066]

[0067] Based on the water flow status of the chamber and the real-time data (i.e., real-time status) of the chamber temperature control module, select the required One-Hot code.

[0068] By preprocessing the data, the difficulty of data processing for the subsequent temperature prediction model is reduced, and the data processing efficiency of the temperature prediction model is improved, thereby improving the processing efficiency of the entire chamber temperature monitoring process.

[0069] In one embodiment, the temperature prediction model is pre-trained. Before training the temperature prediction model, the historical data of relevant variables and the actual historical temperature values ​​are divided into a training set and a validation set according to a preset ratio. The training set is used to train the temperature prediction model, and the test set is used to test the generalization ability of the temperature prediction model and to test and optimize the trained temperature prediction model.

[0070] The training process of the temperature prediction model is as follows: historical data of relevant variables in the training set are input into the preset original prediction model, and the original prediction model outputs the training temperature prediction value of the heating component. The original prediction model is pre-configured based on a long short-term memory neural network. The training temperature prediction value of the heating component is compared with the historical actual temperature value of the heating component to see if it is less than the training threshold. If so, the parameters in the original prediction model are tested and optimized using historical data of relevant variables in the validation set, and the original prediction model is determined as the temperature prediction model. If not, the internal parameters of the original prediction model are adjusted, and the original prediction model is retrained until the temperature prediction model is obtained.

[0071] In this embodiment, the preset ratio for dividing the training set and the validation set is set according to the actual situation and needs. Preferably, the historical actual temperature values ​​of the heating component and the historical data of related variables are divided into the training set and the test set in a ratio of 8:2.

[0072] In this embodiment, since the heating process of the process chamber is a time-varying process and strongly correlated with time, the original prediction model is pre-configured based on a Long Short-Term Memory (LSTM) neural network. LSTM is a variation of Recurrent Neural Network (RNN), a type of neural network used to process sequential data. It borrows from the human brain's thought process of connecting preceding and following information during reasoning and computation, emphasizing the influence of memory on the neural network's input and output. LSTM, building upon RNN, introduces a gating mechanism, making it suitable for processing time-series data samples and more suitable for predicting real-time temperature during the heating process of the process chamber.

[0073] In one embodiment, in order to better train the temperature prediction model, the historical data of the relevant variables are the historical time series data of the relevant variables, which include the historical data of the relevant variables corresponding to multiple consecutive historical moments during the historical normal temperature rise process of the process chamber.

[0074] In this embodiment, the historical data of the relevant variables used to train the temperature prediction model are the historical time series data of the relevant variables. Each relevant variable is stored in chronological order (i.e., the order of multiple consecutive historical moments). The historical actual temperature value of the heating component is also stored in chronological order of the relevant variables, corresponding one-to-one with the historical data of the relevant variables. In other words, the historical time series data of the relevant variables collected at a certain historical moment corresponds to the historical actual temperature value of the heating component at that historical moment.

[0075] In this embodiment, to facilitate training, the historical data of the relevant variables can be preprocessed. The preprocessing process is similar to the real-time data preprocessing process for the relevant variables provided in the above embodiments. It should be noted that, to further improve training efficiency, the historical actual temperature values ​​of the heating component can be preprocessed as continuous variables.

[0076] In an exemplary embodiment, taking the PVD process as an example, the relevant variables include the inner coil power of the heating wire, the outer coil power of the heating wire, the inner coil current of the heating wire, the outer coil current of the heating wire, the water flow status of the chamber, and the chamber temperature control mode. Among them, the inner coil current (unit: A), the outer coil current (unit: A), the inner coil power (unit: %), and the outer coil power (unit: %) are continuous variables, while the water flow status of the chamber (0 indicates no water flow, 1 indicates water flow) and the chamber temperature control mode (0 indicates Servo mode, 1 indicates Ramp mode) are numerical variables.

[0077] In this embodiment, different preprocessing is performed on the historical time series data of the relevant variables according to different data types.

[0078] For any continuous variable or the historical actual temperature value of a heating component, to avoid model overfitting, it is standardized using mean-variance (std). Specifically, based on the historical data of the relevant variables used in model training, the minimum and maximum values ​​from the historical data are used to transform the real-time data into values ​​within the range of 0-1. The calculation formula is as follows:

[0079]

[0080] Where, x min This represents the minimum value among the historical data (or actual historical temperature values ​​of the heating element) of the relevant variable, x. max This represents the maximum value in the historical data of the relevant variable (or the actual historical temperature value of the heating element), x represents the historical data of the relevant variable (or the actual historical temperature value of the heating element) before standardization, and X represents the historical data of the relevant variable (or the actual historical temperature value of the heating element) after standardization.

[0081] For numerical variables (chamber water flow status and chamber temperature control mode), they are encoded using a combination method (One-Hot encoding). One-Hot encoding can solve the problem that the model cannot handle discrete data well, and it can also play a role in data augmentation features to a certain extent. The encoding rules are the same as those in Table 1 above.

[0082] In addition, to exclude outlier data based on trends, it may be necessary to randomly discard historical data at a certain percentage (e.g., 10%).

[0083] In one embodiment, to further improve the prediction accuracy of the temperature prediction model, the basic structure of the temperature prediction model is specifically adjusted based on the heating process of the process chamber. Specifically, such as... Figure 5 As shown, the temperature prediction model includes an input layer, a first operational layer (LSTM), a second operational layer (LSTM), a fully connected layer (Dense), a dropout layer, and an output layer; the input layer, first operational layer, second operational layer, fully connected layer, dropout layer, and output layer are connected sequentially. The output layer also adopts a fully connected layer structure.

[0084] In one embodiment, real-time data of relevant variables are input into a pre-trained temperature prediction model, and the model outputs a real-time temperature prediction value for the heating component. The specific processing steps are as follows: real-time data of relevant variables are input into an input layer; intermediate data output from the input layer is input into a first computational layer, where it undergoes a first data processing transformation; intermediate data output from the first computational layer is input into a second computational layer, where it undergoes a second data processing transformation; intermediate data output from the second computational layer is input into a fully connected layer, where it undergoes regularization; intermediate data output from the fully connected layer is input into a random dropout layer, where neurons are randomly dropped according to a preset ratio during operation; intermediate data output from the random dropout layer is input into an output layer, where it undergoes regularization and outputs the real-time temperature prediction value for the heating component.

[0085] In this embodiment, the construction and training of the temperature prediction model can utilize the Keras API. Keras is a simple and easy-to-use deep learning framework that provides several high-level APIs for conveniently building and training neural network models. A temperature prediction model containing a two-layer LSTM network is constructed, and historical actual temperatures of the input heating component and historical data of relevant variables are used to train and optimize the model. The structural distribution of the temperature prediction model is shown in Table 2 below.

[0086] Table 2. Structural Distribution of Temperature Prediction Models

[0087]

[0088]

[0089] As shown in Table 2 above, a total of 6 layers were used to construct the temperature prediction model. Each layer can realize the coupling transformation between the relevant input variables and related data. The number of layers is just the structure of the neural network used in the temperature prediction model. The relevant variables and related data provided in this application are preprocessed and then fed into the neural network through the input layer. After two LSTM layers and one Dense and Dropout layer, the corresponding data processing and transformation are performed, and the output is output through the Dense layer as the output layer.

[0090] Specifically, during the training phase, preprocessed historical data of relevant variables enters the input layer, then sequentially feeds into the first and second computational layers. After transformation operations by two LSTM layers, the data reaches the fully connected layer (Dense). In the fully connected layer, LTM is used... 1 -L 2 Hybrid regularization is used to prevent overfitting. The data is then fed into a Dropout layer, which randomly drops a certain percentage of neurons in the hidden layers during each training process. This effectively suppresses the interdependence between neurons and helps to learn more powerful features, while also preventing overfitting. Finally, the data is output through a fully connected layer (i.e., the output layer).

[0091] In this embodiment, regularization in machine learning is used to prevent overfitting and improve the model's generalization ability. 1 Regularization adds the sum of the absolute values ​​of the model coefficients to the loss function, while L... 2 Regularization adds the sum of squared model coefficients to the loss function to prevent the model from getting trapped in local optima during training and improve its generalization ability. 1 -L 2 Regularization combines the two regularization methods mentioned above, offering a more flexible approach. L is used in fully connected layers. 1 -L 2 Hybrid regularization, which involves adding an L to the objective function during optimization training in fully connected layers. 1 -L 2 The regularization term adds a regularization bias term to the training objective to prevent the model from getting stuck in local optima during the optimization process.

[0092] In this embodiment, the activation functions selected for each layer when constructing the above temperature prediction model are as follows:

[0093] The LSTM layer uses the sigmoid function as the gate function and the hyperbolic tangent tanh function as the activation function; the fully connected layer uses the ReLU linear rectified function as the activation function. Details are as follows:

[0094]

[0095]

[0096] ReLU; f(x) = max(x, 0)

[0097] Where x represents the data input to the function, f(x) represents the result output by the function, e is a constant, and max represents the maximum value.

[0098] In this embodiment, after constructing the original prediction model as described above, the Hyperopt toolkit can be used to optimize the model's hyperparameters. Hyperopt is a Python library primarily used for intelligent search of the optimal parameters for a given algorithm model. The optimal parameters for each layer are obtained by minimizing the loss function of the overall original prediction model, including the number of neurons, learning rate, and activation function selection. During training, the optimizer is used to determine the parameters by minimizing the mean squared error between the model output and the actual output.

[0099] In this embodiment, the BackPropagation Through Time (BPTT) algorithm is used when training the original prediction model. This algorithm, based on the chain rule, expands the LSTM network step by step and trains it using the forward propagation algorithm to optimize the weights and biases between adjacent neurons in the neural network, thereby reducing errors. Furthermore, during model training, the mean squared error (MSE) is selected as the loss function, and the mean absolute percentage error (MAPE) and mean absolute error (MAE) are used as metrics to evaluate the training effect. The specific calculation formulas are as follows:

[0100]

[0101]

[0102]

[0103] Among them, T act T represents the actual historical temperature value. pre This represents the predicted temperature value during training, and N represents the number of actual historical temperature values ​​in the test set.

[0104] In this context, the optimization loss function can be understood as the objective function. The goal of model optimization training is to make the predicted values ​​closer to the actual values. Therefore, the MSE loss function is used during the optimization process, aiming to minimize the squared difference between the predicted and actual values. After model training, the prediction results are evaluated using MAPE and MAE to assess the deviation between the predicted and actual values, thus evaluating the model training effect. Generally, the smaller the values ​​of these two metrics, the better the model training result and the higher the prediction accuracy. MAPE results are typically values ​​between 0 and 1, but the Tact value cannot be 0. Therefore, comparing the values ​​of MAPE and MAE together provides a better assessment of the model's training effect.

[0105] In one embodiment, the real-time monitoring result of the heating process chamber is obtained by comparing the actual real-time temperature value of the heating component with the predicted real-time temperature value of the heating component, as follows: calculate the difference between the actual real-time temperature value of the heating component and the predicted real-time temperature value of the heating component; if the difference is less than or equal to a preset temperature threshold, generate a real-time monitoring result indicating that the heating process chamber is heating normally; if the difference is greater than the temperature threshold, generate a real-time monitoring result indicating that the heating process chamber is heating abnormally.

[0106] In this embodiment, the temperature threshold is set according to the actual situation and needs. After the temperature prediction model outputs the real-time temperature prediction value of the heating component, the difference between the real-time actual temperature value of the heating component and the real-time temperature prediction value is calculated. If the difference is less than the set temperature threshold (preferably, the temperature threshold can be set to 5% of the real-time actual temperature value of the heating component), it is determined that the chamber temperature rise is normal at this time, and the real-time data of the relevant variables are updated to the historical data of the relevant variables, and the real-time actual temperature value of the heating component is updated to the historical actual temperature value of the heating component. The temperature prediction model is iteratively trained, and then the next sampling cycle is entered. If the difference is greater than the set temperature threshold, it is determined that the chamber temperature rise is abnormal and an alarm is triggered. The real-time data of the corresponding relevant variables and the real-time actual temperature value of the heating component are discarded.

[0107] In one exemplary embodiment, taking the PVD process flow as an example, as follows... Figure 6 As shown, the process of implementing the chamber temperature monitoring method from the model training stage to the model usage stage is as follows:

[0108] During the model training phase:

[0109] Obtain historical actual temperature values ​​of the heating element and historical data of related variables;

[0110] Data preprocessing is performed on the historical actual temperature values ​​of the heating components and the historical data of related variables to reduce the difficulty of data processing. Among them, continuous variables such as current, power and temperature are preprocessed using min-max standardization, and digital variables such as chamber water flow status and chamber temperature control mode are coded using One-Hot encoding.

[0111] The original prediction model is built based on the LSTM architecture;

[0112] Using the pre-processed historical actual temperature values ​​of the heating components and historical data of relevant variables, the original prediction model was trained, validated, and optimized. The final temperature prediction model was then compiled for subsequent use.

[0113] During the real-time use phase of the model:

[0114] Acquire real-time data of relevant variables and actual real-time temperature values ​​of heating components during the real-time operation of the process chamber;

[0115] Data preprocessing is performed on the real-time data of relevant variables and the actual real-time temperature values ​​of heating components. The preprocessing method used is consistent with the pre-training method used in the model training stage.

[0116] The preprocessed real-time data of relevant variables are input into the trained temperature prediction model;

[0117] The temperature prediction model outputs real-time temperature prediction values ​​for the heating components.

[0118] Calculate the difference between the actual real-time temperature of the heating element and the predicted real-time temperature of the heating element, and compare the difference with a preset temperature threshold.

[0119] If the difference is less than or equal to the preset temperature threshold, it is determined that the temperature rise of the process chamber is normal at this time. The real-time data of the relevant variables collected this time and the real-time actual temperature value of the heating component are updated to the training data of the temperature prediction model. The temperature prediction model is iteratively trained and then the next sampling cycle is entered.

[0120] If the difference is greater than the temperature threshold, it is determined that the temperature rise of the process chamber is abnormal, an alarm is triggered, and the real-time data of the relevant variables collected this time and the actual real-time temperature value of the heating component are discarded.

[0121] The chamber heating monitoring method provided in this application involves inputting real-time data of relevant variables during the heating process of the process chamber into a pre-trained temperature prediction model. This model outputs a real-time predicted temperature value for the heating component. The real-time monitoring result of the process chamber heating process is then obtained by comparing the actual real-time temperature value of the heating component with the predicted real-time temperature value. This temperature prediction model is trained based on historical actual temperatures of the heating component and historical data of relevant variables, enabling intelligent monitoring of the chamber heating process. This avoids uncertainties caused by manually setting threshold values ​​and improves the accuracy of the monitoring results. Furthermore, the relevant variables in this application's process chamber heating process include not only continuous variables used in traditional methods but also digital variables such as the water flow status and temperature control mode of the process chamber. Digital variables are more stable and less susceptible to interference, improving the stability of the chamber heating monitoring process. Simultaneously, the water flow status and temperature control mode characterize the actual state of the process chamber itself, making the monitoring process more comprehensive and further improving the accuracy of the heating process monitoring results.

[0122] Exemplary device

[0123] refer to Figure 7 This specification also provides a semiconductor process apparatus 100, including: a process chamber 20, an air intake assembly 20A, and an air extraction assembly (…). Figure 7 (not shown in the image), upper electrode assembly 20B, lower electrode assembly 20C, and controller ( Figure 7 (Not shown in the image), the process chamber includes heating components ( Figure 7 (Not shown in the diagram). The controller includes at least one processor and at least one memory, in which a computer program is stored. When executed by the processor, the computer program implements the chamber temperature monitoring method described in any of the above embodiments.

[0124] For example, the controller can be a host computer or a slave computer. The controller can open the valve of the air intake assembly 20A to introduce the corresponding process gas into the process chamber 20; the controller can also control the flow rate of the process gas by controlling the opening and closing degree of the valve of the air intake assembly 20A. The controller can also control the pressure inside the process chamber 20 by controlling the evacuation assembly to evacuate the air from the process chamber 20.

[0125] The upper electrode assembly 20B includes an RF coil group 21, an upper RF power supply 23, and an upper matching unit 25. The controller is also used to control the upper RF power supply 23 to provide upper electrode power to the RF coil group 21 through the upper matching unit 25, so that the RF coil group 21 excites the process gas inside the process chamber 20 to generate plasma.

[0126] The radio frequency coil assembly 21 may include an inner coil 21A and an outer coil 21B surrounding the inner coil 21A. In an optional embodiment, during an etching or deposition step, the current supplied to the inner coil 21A is less than the current supplied to the outer coil 21B. Optionally, the ratio of the current supplied to the outer coil 21B to the current supplied to the inner coil 21A is greater than 1 and less than or equal to 9.

[0127] The lower electrode assembly 20C includes a wafer carrier 22, a lower RF power supply 24, and a lower matching unit 26. The controller is also used to control the lower RF power supply 24 to provide lower electrode power to the lower electrode of the wafer carrier 22 through the lower matching unit 26, so that the lower electrode of the wafer carrier 22 is provided with RF bias, enabling the plasma to have bombardment capability.

[0128] The wafer carrier device 22 can be used to carry a wafer, which may include a tungsten-containing layer 10. The processor implements the chamber temperature monitoring method as described in any of the above embodiments by running a computer program stored in the memory, so as to perform plasma etching on the tungsten-containing layer 10 in the wafer to form an opening of a target depth.

[0129] The semiconductor process equipment 200 in this application embodiment can be either an inductively coupled plasma (ICP) etching apparatus or a capacitively coupled plasma (CCP) etching apparatus. This application embodiment does not limit the type of semiconductor process equipment 200.

[0130] In one embodiment, the process chamber is a physical vapor deposition process chamber, and the heating component includes an electrostatic chuck; the semiconductor process equipment also includes a power controller, which is used to determine the temperature control mode and transmit the power control parameters under the temperature control mode to the power controller; the power controller is used to adjust the real-time temperature of the heater based on the power control parameters, and the temperature of the heater is conducted to the electrostatic chuck.

[0131] In this embodiment, the electrostatic chuck is used to transfer the heat provided by the heater to the wafer.

[0132] For specific limitations and beneficial effects of the chamber temperature monitoring method, please refer to the relevant descriptions above. This manual will not repeat them here.

[0133] Exemplary computer program products and storage media

[0134] In addition to the methods and devices described above, the chamber temperature monitoring method provided in the embodiments of this specification can also be a computer program product, which includes computer program instructions that, when executed by a processor, cause the processor to perform the chamber temperature monitoring method according to the various embodiments of this specification described in the "Exemplary Methods" section above.

[0135] The computer program product described herein can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments described herein. These programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0136] Furthermore, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor of the chamber temperature monitoring methods according to various embodiments of this specification as described in the "Exemplary Methods" section above.

[0137] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this specification can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0138] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The embodiments described above are merely illustrative of several implementation methods outlined in this specification. While the descriptions are specific and detailed, they should not be construed as limiting the scope of the solutions provided in this specification. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this specification, and these all fall within the scope of protection of this specification. Therefore, the scope of protection for this patent should be determined by the appended claims.

Claims

1. A method for monitoring chamber temperature rise, characterized in that, Applied to a process chamber, the process chamber including a heating component; The method includes: Real-time data of relevant variables affecting the heating process of the process chamber are obtained, as well as the real-time actual temperature value of the heating component in the process chamber are obtained. The relevant variables include continuous variables and digital variables. The digital variables include the water flow status and temperature control mode of the process chamber. The continuous variables refer to variables that change continuously over time. The real-time data of the relevant variables are input into the pre-trained temperature prediction model, and the real-time temperature prediction value of the heating component is output through the temperature prediction model. The temperature prediction model is trained based on the historical actual temperature value of the heating component and the historical data of the relevant variables. By comparing the actual real-time temperature value of the heating element with the predicted real-time temperature value of the heating element, the real-time monitoring results of the heating process chamber are obtained. The temperature prediction model is pre-trained using the following process: historical data of the relevant variables in the training set are input into a preset original prediction model, and the original prediction model outputs the training temperature prediction value of the heating component. The original prediction model is pre-configured based on a long short-term memory neural network. The training set and validation set are obtained by dividing the historical data of the relevant variables and the actual historical temperature values ​​according to a preset ratio. Compare the predicted training temperature of the heating element with the actual historical temperature of the heating element to see if it is less than the training threshold; If so, the original prediction model is determined as the temperature prediction model after testing and optimizing the parameters in the original prediction model using historical data of the relevant variables in the validation set. If not, adjust the internal parameters of the original prediction model and retrain the original prediction model until the temperature prediction model is obtained.

2. The chamber temperature monitoring method according to claim 1, characterized in that, After obtaining the real-time monitoring result of the heating process of the process chamber by comparing the actual real-time temperature value of the heating element with the predicted real-time temperature value of the heating element, the method further includes: If the real-time monitoring results indicate that the temperature rise of the process chamber is normal, the real-time data of the relevant variables will be updated to the historical data of the relevant variables, and the real-time actual temperature value of the heating element will be updated to the historical actual temperature value of the heating element. The updated historical actual temperature values ​​of the heating component and the updated historical data of the relevant variables are used to retrain the temperature prediction model.

3. The chamber temperature monitoring method according to claim 1, characterized in that, The temperature prediction model includes an input layer, a first computation layer, a second computation layer, a fully connected layer, a random dropout layer, and an output layer. The input layer, the first operation layer, the second operation layer, the fully connected layer, the random drop layer, and the output layer are connected in sequence.

4. The chamber temperature monitoring method according to claim 3, characterized in that, The step of inputting real-time data of the relevant variables into a pre-trained temperature prediction model, and outputting real-time temperature prediction values ​​of the heating component through the temperature prediction model, includes: The real-time data of the relevant variables are input into the input layer; The intermediate data output from the input layer is input to the first computation layer, where the first computation layer performs the first data processing transformation. The intermediate data output from the first processing layer is input to the second processing layer, where the second processing layer performs a second data processing transformation. The intermediate data output from the second computation layer is input to the fully connected layer, where regularization is performed. The intermediate data output by the fully connected layer is input to the random dropout layer, and neurons are randomly dropped according to a preset ratio during the operation of the random dropout layer. The intermediate data output by the random discard layer is input to the output layer, which performs regularization processing and outputs the real-time temperature prediction value of the heating component.

5. The chamber temperature monitoring method according to claim 1, characterized in that, The acquisition of real-time data on relevant variables affecting the heating process of the process chamber includes: Real-time data of relevant variables are collected during the heating process of the process chamber; The real-time data of the continuous variables are standardized, and the real-time data of the numerical variables are one-hot encoded to obtain standardized data of the relevant variables. The step of inputting the real-time data into a pre-trained temperature prediction model and outputting the real-time temperature prediction value of the heating component through the temperature prediction model includes: The standardized data of the relevant variables are input into the pre-trained temperature prediction model, and the real-time temperature prediction value of the heating component is output through the temperature prediction model.

6. The chamber temperature monitoring method according to claim 1, characterized in that, The step of obtaining real-time monitoring results of the process chamber heating process by comparing the actual real-time temperature value of the heating element with the predicted real-time temperature value of the heating element includes: Calculate the difference between the actual real-time temperature of the heating element and the predicted real-time temperature of the heating element; If the difference is less than or equal to a preset temperature threshold, a real-time monitoring result indicating that the process chamber is heating normally is generated. If the difference is greater than the temperature threshold, a real-time monitoring result of the abnormal temperature rise in the process chamber is generated.

7. The chamber temperature monitoring method according to claim 1, characterized in that, The historical data of the relevant variables are the historical time-series data of the relevant variables, which include the historical data of the relevant variables corresponding to multiple consecutive historical moments during the historical normal heating process of the process chamber.

8. The chamber temperature monitoring method according to any one of claims 1-7, characterized in that, The continuous variables include one or more of the following: inner coil power of the heating wire, outer coil power of the heating wire, inner coil current of the heating wire, and outer coil current of the heating wire.

9. A semiconductor process apparatus, characterized in that, include: A process chamber, the process chamber including a heating element; The upper electrode assembly is located above the process chamber; The lower electrode assembly, located within the process chamber, is used to support the wafer and apply a bias voltage to the wafer; A controller, including at least one memory and at least one processor, the memory being used to store computer programs; The processor is configured to implement the chamber temperature monitoring method as described in any one of claims 1 to 8 by running a computer program stored in the memory.

10. The semiconductor process equipment according to claim 9, characterized in that, The process chamber is a physical vapor deposition process chamber, and the heating component includes an electrostatic chuck. The semiconductor process equipment also includes a power controller, which is used to determine a temperature control mode and transmit power control parameters under the temperature control mode to the power controller; the power controller is used to adjust the real-time temperature of the heater based on the power control parameters, and the temperature of the heater is conducted to the electrostatic chuck.

Images