Etching apparatus predictive maintenance method, device and related apparatus

Abstract

The application provides an etching equipment predictive maintenance method and device and related equipment. The method comprises: acquiring multi-dimensional data of a reaction chamber of an etching equipment; generating a plasma stability index according to plasma emission spectrum data; inputting a radio frequency power fluctuation value, a reaction gas concentration change rate and a static chuck center point temperature value into a feature fusion model to generate an etching uniformity prediction parameter; inputting the plasma stability index, the etching uniformity prediction parameter and a chamber pressure gradient into a time convolution network for collaborative analysis to generate decision encoding probability values corresponding to the chamber pressure gradient, the plasma stability and the etching uniformity respectively; determining a comprehensive health index of the etching equipment according to the decision encoding probability values; and generating a maintenance strategy according to a health index interval in which the comprehensive health index is located. The above scheme can quickly respond to real-time state changes of the etching equipment, generate a maintenance strategy and improve the operation stability of a production line.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a predictive maintenance method, apparatus and related equipment for etching equipment. Background Technology

[0002] In semiconductor manufacturing processes, etching equipment is a key piece of equipment that plays a crucial role in the quality and efficiency of chip manufacturing. The etching process involves transferring patterns onto the wafer surface to create tiny, precise structures that form the circuits and components in integrated circuits.

[0003] However, the long-term stability and reliability of etching equipment are crucial factors limiting manufacturing efficiency and cost. Currently, the operation monitoring of etching equipment mainly relies on periodic inspections and preventative maintenance plans. This approach presents several challenges: First, maintenance activities are often carried out only after equipment failure or performance degradation, and this reactive maintenance method may lead to production line downtime and reduced production efficiency. Second, maintenance methods typically rely solely on operator experience and periodic sensor data collection. This approach cannot comprehensively capture subtle changes or trends in equipment operation, nor can it achieve real-time and in-depth data analysis.

[0004] Based on this, how to quickly respond to real-time status changes of etching equipment, generate maintenance strategies, and improve the operational stability of the production line has become a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, this application provides a predictive maintenance method, apparatus and related equipment for etching equipment, which can quickly respond to real-time status changes of etching equipment, generate maintenance strategies and improve the operational stability of the production line.

[0006] In a first aspect, this application provides a predictive maintenance method for etching equipment, comprising:

[0007] Acquire multidimensional data of the reaction chamber of the etching equipment, wherein the multidimensional data includes one or more of the following: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value;

[0008] Based on the plasma emission spectrum data, a plasma stability index is generated;

[0009] The radio frequency power fluctuation value, the rate of change of the reactive gas concentration, and the temperature value of the center point of the electrostatic chuck are input into the feature fusion model to generate etching uniformity prediction parameters.

[0010] The plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient are input into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively.

[0011] The comprehensive health index of the etching equipment is determined based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity.

[0012] A maintenance strategy is generated based on the health index range in which the comprehensive health index falls.

[0013] Optionally, generating plasma stability indices based on the plasma emission spectrum data includes:

[0014] The plasma emission spectrum data is filtered using a Savitzky-Golay filter to obtain filtered spectrum data; wherein, the Savitzky-Golay filter is provided with a first preset time window, which represents the number of sampling points used in each filtering process;

[0015] The maximum and minimum values ​​of the spectral intensity of multiple characteristic wavelengths within a second preset time window, as well as a reference value, are obtained from the filtered spectral data, and a weighting coefficient is assigned to each characteristic wavelength; wherein each characteristic wavelength corresponds to a plasma, and the second preset time window is within the first preset time window;

[0016] The weighted volatility of each characteristic wavelength is determined based on the maximum and minimum values ​​of the spectral intensity of each characteristic wavelength within a second preset time window, as well as the reference value and the corresponding weighting coefficient.

[0017] The plasma stability index is determined based on the relationship between the weighted volatility of each characteristic wavelength and a preset volatility.

[0018] Optionally, the weighted volatility of each of the characteristic wavelengths can be determined using the following formula:

[0019]

[0020] Where, ΔI λ Indicates weighted volatility, w λ The weighting coefficients represent the characteristic wavelengths, and t-i1:t represents the second preset time window. Indicates the first preset time window, The maximum value of the spectral intensity The minimum value representing spectral intensity. This represents the baseline reference value, and i2 is greater than i1;

[0021] The reference value is determined by the ratio of the sum of the spectral intensities of each sampling point obtained within the first preset time window of the characteristic wavelength to the total number of sampling points.

[0022] The step of determining a plasma stability index based on the relationship between the weighted volatility of each of the characteristic wavelengths and a preset volatility includes: responding to the largest weighted volatility among the weighted volatility of the plurality of characteristic wavelengths; in response to the largest weighted volatility being lower than the preset volatility, the plasma stability index having a first identifier; and in response to determining that the largest weighted volatility exceeds the preset volatility, the plasma stability index having a second identifier.

[0023] Optionally, the step of inputting the RF power fluctuation value, the rate of change of the reactive gas concentration, and the temperature value of the electrostatic chuck center point into the feature fusion model to generate etching uniformity prediction parameters includes:

[0024] Based on the nonlinear processing unit in the feature fusion model, the rate of change of the concentration of the reactant gas and the fluctuation value of the radio frequency power are nonlinearly processed to generate a dynamic coupling coefficient.

[0025] When the dynamic coupling coefficient is greater than a preset coefficient, the etching uniformity prediction parameter is forcibly set to zero;

[0026] When the dynamic coupling coefficient is not greater than a preset coefficient, the dynamic coupling coefficient, the temperature value of the center point of the electrostatic chuck, and the standard deviation of the reaction gas concentration are input into the feature fusion model.

[0027] Matrix multiplication and addition operations are performed through the hidden layer in the feature fusion model to obtain multidimensional feature space parameters, and the multidimensional feature space parameters are enhanced by the GELU activation function.

[0028] The enhanced multidimensional feature space parameters are transformed into the etching uniformity prediction parameters using the Sigmoid function.

[0029] The standard deviation of the reactant gas concentration is obtained by normalizing the rate of change of the reactant gas concentration.

[0030] Optionally, the step of inputting the plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity includes:

[0031] A four-dimensional input vector is constructed, which includes the chamber pressure gradient, the plasma stability index, the etching uniformity prediction parameter, and the rate of change of the etching uniformity prediction parameter.

[0032] The four-dimensional input vector is processed using a five-layer residual dilated convolution, including: each layer performing sequential convolution operation, ReLU activation, residual summation on the four-dimensional input vector, and finally performing layer normalization and outputting a feature map; wherein, the kernel width of each convolution layer is 5, and the dilation factors are 1, 2, 4, 8, and 16 respectively.

[0033] The feature map is subjected to global average pooling using the Softmax function, and an original score is generated through a fully connected layer. Then, after normalization, the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity are output.

[0034] Optionally, the five-layer residual dilated convolution includes: a first layer of dilated convolution to cover plasma flicker; a second layer of dilated convolution to cover gas pressure fluctuations; a third layer of dilated convolution to cover radio frequency power drift; a fourth layer of dilated convolution to cover temperature gradient changes; and a fifth layer of dilated convolution to cover electrode aging accumulation.

[0035] Secondly, this application provides a predictive maintenance device for etching equipment, comprising:

[0036] The data acquisition unit is configured to acquire multidimensional data of the reaction chamber of the etching equipment, the multidimensional data including one or more of the following: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value;

[0037] The processing unit is configured to generate a plasma stability index based on the plasma emission spectrum data; and to input the radio frequency power fluctuation value, the reaction gas concentration change rate, and the electrostatic chuck center point temperature value into a feature fusion model to generate etching uniformity prediction parameters, and to input the plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively.

[0038] The maintenance unit is configured to determine the comprehensive health index of the etching equipment based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, and to generate a maintenance strategy based on the health index range in which the comprehensive health index is located.

[0039] Thirdly, this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the steps of the predictive maintenance method for etching equipment described in any of the preceding claims when running the computer program.

[0040] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the predictive maintenance method for etching equipment described in any of the preceding claims to be performed.

[0041] Fourthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the predictive maintenance method for etching equipment described in any of the preceding claims.

[0042] Compared with the prior art, the technical solution of this application has the following advantages:

[0043] The predictive maintenance method for etching equipment provided in this application can acquire multi-dimensional data related to the operating status of the etching equipment. Based on the plasma emission spectrum data from this multi-dimensional data, a plasma stability index characterizing plasma stability can be generated. Etching uniformity prediction parameters are generated by inputting the RF power fluctuation value, the rate of change of the reactive gas concentration, and the temperature value at the center point of the electrostatic chuck into a feature fusion model. By employing a temporal convolutional network to collaboratively analyze the plasma stability index, etching uniformity prediction parameters, and chamber pressure gradient, corresponding decision coding probability values ​​can be generated. Therefore, based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, the health index range of the etching equipment's comprehensive health index can be determined, and maintenance strategies can be selectively generated. Using this scheme, real-time status changes of the etching equipment can be responded to quickly, maintenance strategies can be generated, and the operational stability of the production line can be improved. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of this embodiment, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart of a predictive maintenance method for etching equipment according to an embodiment of this application is shown;

[0046] Figure 2 A flowchart of a method for determining plasma stability indicators according to an embodiment of this application is shown;

[0047] Figure 3 A schematic diagram of a predictive maintenance device for etching equipment according to an embodiment of this application is shown;

[0048] Figure 4 A schematic diagram of an optional hardware structure of an electronic device provided in one embodiment of this application is shown. Detailed Implementation

[0049] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0050] As described in the background section, relying on periodic inspections and preventative maintenance plans makes it impossible to effectively monitor the operating status of etching equipment.

[0051] To address these issues, predictive maintenance technology has been widely applied and researched in the semiconductor manufacturing industry in recent years. Predictive maintenance aims to predict potential equipment failures or performance degradation trends in advance through real-time data monitoring and analysis, thereby enabling timely implementation of necessary maintenance measures to avoid downtime and production losses caused by failures.

[0052] With the continuous advancement of sensor technology and data analysis capabilities, the methods for monitoring and maintaining etching equipment are also constantly evolving.

[0053] Here are some key technologies and trends:

[0054] A comprehensive sensor array: Modern etching equipment is equipped with a variety of sensors, including those for temperature, pressure, flow rate, and chemical concentration, forming a comprehensive sensor array. This array enables real-time monitoring of equipment operating parameters and surrounding environmental conditions, providing a solid data foundation for data-driven predictive maintenance.

[0055] Machine learning algorithms: Machine learning is particularly prominent in predictive maintenance. By monitoring and analyzing sensor data in real time, machine learning algorithms can identify abnormal patterns and potential fault signals in equipment operation.

[0056] Real-time data analytics platforms: To effectively implement predictive maintenance strategies, semiconductor manufacturers typically establish real-time data analytics platforms. These platforms integrate data acquisition, storage, processing, and analysis capabilities, enabling them to quickly respond to changes in equipment operating status and generate real-time predictions and alerts.

[0057] In one approach, the emission line spectrometer (OES) inherent in the etching equipment is used to detect the intensity changes of emitted spectral lines during the process. Based on the detection results, the process parameters are quantified. Statistical analysis is then performed on the process parameters that have been detected and quantified multiple times to determine the trend of their changes. Subsequently, control limits are set according to the requirements of the etching equipment for the process. When the trend of the process parameters exceeds the control limits, the etching equipment is maintained.

[0058] However, existing solutions mainly rely on traditional periodic maintenance and rule-based fault diagnosis systems. While these methods can ensure the operational stability of equipment to a certain extent, they also have obvious limitations and defects.

[0059] First, traditional periodic maintenance is often based on experience and time or lifespan, making it difficult to effectively adjust to the actual operating conditions and needs of the equipment. Such fixed maintenance plans can lead to unnecessary maintenance costs and production interruptions, especially when maintenance is performed when the equipment is not malfunctioning, wasting resources and time.

[0060] Secondly, while rule-based fault diagnosis systems can identify common fault modes, their diagnostic capabilities may be limited when faced with complex equipment operating conditions and changing production environments. Rule-based systems often require predefined fault modes and rule sets, making it difficult to flexibly respond to new fault types or equipment anomalies, resulting in low fault diagnosis accuracy or a high rate of missed diagnoses.

[0061] Furthermore, traditional methods of manual intervention and data analysis rely on the operator's experience and expertise. This approach is not only labor-intensive but also carries the risk of subjectivity and misjudgment, especially when faced with large amounts of complex data, where operators may be unable to promptly identify and analyze potential problems or anomalies in the equipment.

[0062] In summary, existing solutions have significant limitations in achieving intelligent equipment maintenance, cost efficiency, and production efficiency. To address these challenges, next-generation predictive maintenance systems need to combine advanced sensor technologies and machine learning algorithms to achieve real-time monitoring, accurate prediction, and intelligent maintenance decision-making for equipment operating status, thereby improving manufacturing efficiency and equipment utilization.

[0063] Specifically, this application acquires multidimensional data including RF power fluctuation values, reactant gas concentration change rates, chamber pressure gradients, plasma emission spectrum data, and electrostatic chuck center point temperature values. Through a feature fusion model and a temporal convolutional network, it can determine the comprehensive health index of the etching equipment. Based on the health index range of the etching equipment, a maintenance strategy can be generated. This enables rapid response to real-time changes in the etching equipment's status, generating maintenance strategies and improving the operational stability of the production line.

[0064] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the embodiments of this application are described below with reference to the accompanying drawings.

[0065] See Figure 1 The flowchart shown in one embodiment of this application illustrates a predictive maintenance method for etching equipment, as follows: Figure 1 As shown, steps S11 to S16 are included.

[0066] S11, acquire multidimensional data of the reaction chamber of the etching equipment, the multidimensional data including one or more of the following: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value.

[0067] In some embodiments, the reaction chamber of the etching apparatus provides an etching environment for the wafer. During the etching process, multidimensional data from the reaction chamber can be acquired in real time. This multidimensional data can reflect the real-time operating status of the etching apparatus.

[0068] In some embodiments, the radio frequency power fluctuation value refers to the instantaneous change in the output power of the radio frequency power supply, reflecting the stability of the plasma excitation energy, and can be used to characterize electrode aging or radio frequency power supply failure.

[0069] The RF power fluctuation value can be determined based on the set output power and the measured output power of the RF power supply.

[0070] For example, the radio frequency power fluctuation value ΔP rf The following formula can be used to determine it:

[0071]

[0072] Among them, P set Indicates the set output power, P inst This indicates the measured output power.

[0073] In some embodiments, the output power of the RF power supply is obtained by incorporating a power detector within the RF power supply.

[0074] The rate of change of reactant gas concentration refers to the rate of change of the concentration of reactant gas (e.g., Cl / BCl / CF, etc.) per unit time, and can be used to characterize gas nozzle clogging problems.

[0075] Specifically, the concentration of a specific gas is detected by mass spectrometry and infrared absorption spectrometry, thereby determining the rate of change of the concentration of the reactant gas.

[0076] The chamber pressure gradient is used to represent the non-uniformity of radial pressure distribution within the reaction chamber.

[0077] The pressure gradient of the reaction chamber can be determined by the pressure values ​​at multiple monitoring points at the edge of the chamber.

[0078] For example, the chamber pressure gradient G p The following formula can be used to determine it:

[0079]

[0080] Among them, P r1 P r2 ... P rn P is used to represent the pressure value at different monitoring points. avg P represents the average pressure. avg =(P r1 +P r2 +…+P rn ) / n, where n is an integer greater than 1.

[0081] In some embodiments, pressure values ​​can be acquired at three monitoring points: the top, sidewall, and bottom of the reaction chamber. Specifically, capacitive pressure sensors are installed at these three monitoring points, and the distribution of the capacitive pressure sensors satisfies the following condition: a 120° circular distribution.

[0082] Plasma emission spectroscopy data can be obtained using a multi-channel OES spectrometer, and the plasma emission spectra at the top / sidewalls / bottom of the reaction chamber can be acquired via fiber optic bundles.

[0083] The temperature value at the center point of the electrostatic chuck can be used to characterize the temperature difference on the wafer surface, reflecting the etching uniformity.

[0084] Among them, the electrostatic chuck serves as a carrier device, used to carry the wafer.

[0085] In some embodiments, the temperature value of the center point of the electrostatic chuck is obtained by placing a gallium nitride thin-film thermocouple on the surface of the electrostatic chuck.

[0086] By acquiring multidimensional data, the operating status of etching equipment can be analyzed from different perspectives, thereby improving the accuracy of the operating status.

[0087] S12, Based on the plasma emission spectrum data, generate plasma stability index.

[0088] In some embodiments, the plasma emission spectral data contains parameters for multiple plasmas. By analyzing the plasma emission spectral data, it is possible to determine whether each plasma is in a stable state, which is crucial for etching uniformity.

[0089] In some embodiments, plasma stability indices are determined after filtering the plasma emission spectrum data using a Savitzky-Golay filter.

[0090] See Figure 2 The flowchart shown in one embodiment of this application illustrates a method for determining plasma stability indicators, as follows: Figure 2 As shown, the following steps can be performed:

[0091] S21, the plasma emission spectrum data is filtered using a Savitzky-Golay filter to obtain filtered spectrum data; wherein, the Savitzky-Golay filter is provided with a first preset time window, which represents the number of sampling points used in each filtering process.

[0092] The Savitzky-Golay filter is a widely used digital signal processing tool. Its core function is to smooth the data (noise reduction) while preserving the important characteristics of the signal (especially peaks and inflection points), so as to remove noise while maintaining the original signal shape to the greatest extent.

[0093] In some embodiments, a first preset time window for the Savitzky-Golay filter is pre-configured, which represents the number of sampling points used in each filtering process of the plasma emission spectrum data. The number of sampling points includes: sampling points at the current time and sampling points at times before and after the current time.

[0094] For etching applications, the number of sampling points used in the first preset time window is 11. Accordingly, within this first preset time window, a third-order polynomial is used to perform filtering.

[0095] S22, obtain the maximum and minimum values ​​of the spectral intensity of multiple characteristic wavelengths within a second preset time window from the filtered spectral data, as well as the reference value, and assign a weighting coefficient to each characteristic wavelength; wherein, each characteristic wavelength corresponds to a plasma, and the second preset time window is within the first preset time window.

[0096] In some embodiments, multiple characteristic wavelengths can be obtained by analyzing the filtered spectral data. Each characteristic wavelength corresponds to a plasma used to perform etching.

[0097] For example, suppose an application scenario includes three types of plasma: Cl, with a characteristic wavelength of 308 nm; H, with a characteristic wavelength of 486 nm; and O, with a characteristic wavelength of 656 nm.

[0098] In some embodiments, after determining multiple characteristic wavelengths, the spectral intensity of each characteristic wavelength can be obtained within a second preset time window. This allows for the determination of the maximum and minimum values ​​of the spectral intensity of each characteristic wavelength.

[0099] Furthermore, after acquiring multiple characteristic wavelengths, weighting coefficients can be assigned to each characteristic wavelength in the current monitoring process according to the pre-configured weighting coefficients for the characteristic wavelengths.

[0100] In some embodiments, the weighting coefficients for each characteristic wavelength can be determined based on the differences in ionization energy of the reactant gases and the role played by the plasma.

[0101] For example, in this application, if the plasma corresponding to the characteristic wavelength is divided into Cl, H, and O, then the corresponding ionization energies of the reactive gases are 12.97 eV, 15.43 eV, and 12.06 eV, respectively. The lower the ionization energy of the reactive gases, the more sensitive the plasma stability is to the device state, and therefore requires a higher weighting. However, Cl, as the primary etching gas, will be given a higher weighting.

[0102] For example, the weighting coefficients for Cl, H, and O are 1.2, 0.8, and 1.0, respectively.

[0103] In some embodiments, the reference value is determined by the ratio of the sum of the spectral intensities of each sampling point obtained within a first preset time window of the characteristic wavelength to the total number of sampling points.

[0104] For example, suppose there are 20 sampling points corresponding to the characteristic wavelength within a first preset time window, thus obtaining 20 spectral intensities. By superimposing these 20 spectral intensities, the total intensity can be determined, and the ratio of the total intensity to 20 is the reference value.

[0105] It should be noted that the number of sampling points within the first preset time window varies for different characteristic wavelengths.

[0106] S23, determine the weighted volatility of each characteristic wavelength based on the maximum and minimum values ​​of the spectral intensity of each characteristic wavelength within the second preset time window, as well as the reference value and the corresponding weighting coefficient.

[0107] In some embodiments, the weighted volatility of each of the characteristic wavelengths is determined using the following formula:

[0108]

[0109] Where, ΔI λ Indicates weighted volatility, w λ The weighting coefficients represent the characteristic wavelengths, and t-i1:t represents the second preset time window. Indicates the first preset time window, The maximum value of the spectral intensity The minimum value representing spectral intensity. This represents the baseline reference value, and i2 is greater than i1.

[0110] In one specific embodiment, i1 takes the value 4. This represents the maximum spectral intensity within the last 5 sampling periods (t-4 to t). This represents the minimum spectral intensity within the last 5 sampling periods (t-4 to t).

[0111] i2 takes the value 10. This represents the average spectral intensity over the most recent 11 sampling periods (t-10 to t).

[0112] S24. Based on the relationship between the weighted volatility of each characteristic wavelength and a preset volatility, determine the plasma stability index.

[0113] The plasma stability index is used to characterize the stability of the plasma used by the etching equipment during operation. For example, a binary method can be used for labeling.

[0114] Specifically, the largest weighted volatility among the weighted volatility of the plurality of characteristic wavelengths is determined; in response to the largest weighted volatility being lower than the preset volatility, the plasma stability index has a first identifier; in response to the largest weighted volatility exceeding the preset volatility, the plasma stability index has a second identifier.

[0115] For example, the preset volatility can be 0.3. When the maximum weighted volatility is less than or equal to 0.3, the plasma instability flag Flag_plasma = 1, which is the first flag; when the maximum weighted volatility exceeds 0.3, the plasma instability flag Flag_plasma = 0, which is the second flag.

[0116] Thus, by processing plasma emission spectral data, the stability of the plasma used can be determined, thereby enabling the assessment of the stability of the etching gas during the etching process.

[0117] S13, input the RF power fluctuation value, the reaction gas concentration change rate and the electrostatic chuck center point temperature value into the feature fusion model to generate etching uniformity prediction parameters.

[0118] In some embodiments, the feature fusion model can fully explore the correlation between RF power fluctuations, reactive gas concentration changes, and electrostatic chuck center point temperature, and etching uniformity.

[0119] In some embodiments, the feature fusion model includes a nonlinear processing unit, a hidden layer, and an output layer, wherein the nonlinear processing unit is used to determine the dynamic coupling coefficient k. rf -gas.

[0120] Specifically, during actual monitoring, the inventors discovered that when the gas nozzle is severely clogged, the radio frequency power and gas concentration exhibit violent fluctuations (i.e., the dynamic coupling coefficient k). rf-gas (Sudden increase), this fault develops extremely quickly, and the feature fusion model may not be able to respond in time.

[0121] In this application, based on the determined dynamic coupling coefficient k rf-gas A forced triggering mechanism has been set up.

[0122] For example, step S13 may include:

[0123] Based on the nonlinear processing unit in the feature fusion model, the change rate of the reactive gas concentration and the radio frequency power fluctuation value are nonlinearly processed to generate a dynamic coupling coefficient.

[0124] In some embodiments, the dynamic coupling coefficient k rf-gas The determination process is as follows:

[0125]

[0126] Where, ΔC gas Indicates the rate of change of reactant gas concentration, ΔP rf Indicates the radio frequency power fluctuation value, W c It is a 2×32 dimensional parameter matrix, where W c The first row is the radio frequency feature extraction feature, W c The second line is the gas feature extraction feature; ReLU(x) = max(0, x), which realizes the suppression of negative interference caused by sensor noise and retains only the positive coupling relationship between radio frequency power and gas concentration.

[0127] For example, ΔC gas 5ppm, ΔP rf -10W, W c ·[ΔPrf ;ΔC gas If the value is less than 0, then ReLU = 0.

[0128] In other words, this application is approved by W. c The rate of change of reactant gas concentration and the fluctuation value of radio frequency power are mapped to a 32-dimensional vector. Then, the positive coupling relationship between radio frequency power and gas concentration is preserved by using the ReLU function. This allows the output of the enhanced dynamic coupling coefficient k. rf-gas .

[0129] When the dynamic coupling coefficient is greater than the preset coefficient, the etching uniformity prediction parameter is forcibly set to zero.

[0130] In some embodiments, if the dynamic coupling coefficient is greater than a preset coefficient (e.g., 2), it indicates that the etching equipment is in a nonlinear oscillation state, the radio frequency power and gas flow rate are in a chaotic relationship, and the confidence of the neural network prediction in the feature fusion model will drop sharply.

[0131] By forcing the etching uniformity prediction parameter to zero, the fault response speed will be improved, and the erroneous transmission of the etching uniformity prediction parameter to the temporal convolutional network will be avoided, which could lead to malfunctions.

[0132] In some embodiments, taking into account the effect of temperature changes, the dynamic coupling coefficient k can also be adjusted. rf-gas Corrections were made to improve the accuracy of the dynamic coupling coefficients.

[0133] For example, the temperature correction factor is: Among them, T base T represents the reference temperature. real This indicates the temperature value at the center point of the electrostatic chuck, where T represents a constant to prevent division by zero.

[0134] In one specific embodiment, T base You can choose 25 degrees Celsius, and T can be a smaller value, such as 10. -5 Celsius.

[0135] Accordingly, the corrected dynamic coupling coefficient is:

[0136]

[0137] When the dynamic coupling coefficient is not greater than a preset coefficient, the dynamic coupling coefficient, the temperature value of the center point of the electrostatic chuck, and the standard deviation of the reactive gas concentration are input into the feature fusion model to perform matrix multiplication and addition operations through the hidden layer in the feature fusion model to obtain multidimensional feature space parameters. Then, the multidimensional feature space parameters are enhanced by the GELU activation function, and then the enhanced multidimensional feature space parameters are transformed into the etching uniformity prediction parameters by the Sigmoid function of the output layer. The standard deviation of the reactive gas concentration is obtained by normalizing the rate of change of the reactive gas concentration.

[0138] In some embodiments, when the dynamic coupling coefficient is not greater than a preset coefficient, an input vector can be constructed. And optionally, for the input vector X in Standardize the process.

[0139] Then, after performing matrix multiplication and addition operations through a 64-node hidden layer (which can be understood as other values), the etching uniformity prediction parameter Q is obtained by passing it through the GELU activation function and then through the Sigmoid function of the output layer.

[0140] Q=σ(W q ·GELU(W h ·X in +b h ))

[0141] Among them, W h This represents the 64×3 weight matrix of the hidden layer, b h W represents a 64-dimensional bias matrix. q This represents the 1×64 weight vector of the output layer, and σ represents the Sigmoid function.

[0142] In some embodiments, Q is between 0 and 1, and a larger Q indicates better uniformity.

[0143] By configuring a prediction scheme with different relationships between the dynamic coupling coefficient and the preset coefficient, it is possible to respond to anomalies in a timely manner and obtain the RF power and gas concentration with positive coupling relationship, thereby improving the accuracy of etching uniformity prediction parameters.

[0144] S14, the plasma stability index, the etching uniformity prediction parameter, and the chamber pressure gradient are input into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively.

[0145] In some embodiments, by executing steps S11 to S13, plasma stability indices, etching uniformity prediction parameters, and chamber pressure gradients can be obtained, respectively. Thus, the temporal convolutional network can perform targeted collaborative analysis to output their respective decision coding probability values.

[0146] In some embodiments, step S14 may include: constructing a four-dimensional input vector, the four-dimensional input vector including the chamber pressure gradient, the plasma stability index, the etching uniformity prediction parameter, and the rate of change of the etching uniformity prediction parameter; processing the four-dimensional input vector using five layers of residual dilated convolution, including: performing convolution operation, ReLU activation, and residual summation on the four-dimensional input vector in each layer, and finally performing layer normalization and outputting a feature map; wherein the width of each convolution kernel is 5, and the dilation factors are 1, 2, 4, 8, and 16 respectively; performing global average pooling on the feature map using the Softmax function, generating an original score through a fully connected layer, and then outputting the decision encoding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity through normalization processing.

[0147] The decision encoding probability value of the chamber pressure gradient is P. c1 Based on G p Anomaly pattern generation; the decision coding probability value for plasma stability is P. c2 Generated based on plasma instability characteristics; the decision coding probability value for etching uniformity is P. c3 It is generated based on the trend of etching uniformity degradation.

[0148] In some embodiments, one of the decision coding probability values ​​can be selectively output based on the preset probability values ​​corresponding to plasma stability, chamber pressure gradient, and etching uniformity.

[0149] For example, P c1 When the value is greater than 0.8, the decision-encoded probability value of the chamber pressure gradient is used to execute the gas spray group cleaning operation; P c2 When the value is greater than 0.7, the decision-coded probability value of plasma stability is output to calibrate the radio frequency circuit; P c3 When the value is greater than 0.6, the decision code probability value of etching uniformity is output to initiate the bushing replacement operation.

[0150] In some embodiments, the five-layer residual dilated convolution includes: a first layer of dilated convolution to cover plasma flicker; a second layer of dilated convolution to cover gas pressure fluctuations; a third layer of dilated convolution to cover radio frequency power drift; a fourth layer of dilated convolution to cover temperature gradient changes; and a fifth layer of dilated convolution to cover electrode aging accumulation.

[0151] S15, determine the comprehensive health index of the etching equipment based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity.

[0152] The comprehensive health index (CHI) is used to represent the health level of the etching equipment after considering multiple parameters.

[0153] In one embodiment, the Comprehensive Health Index (CHI) of the etching equipment is determined as follows:

[0154] CHI=α(1-P c1 )+β(1-P c2 )+δ(1-P c3 )

[0155] Where α+β+δ=1 represents the weight coefficient of the abnormal cause corresponding to the respective decision coding probability value.

[0156] Among these, the weight coefficients of the abnormal causes corresponding to the decision coding probability values ​​can be determined based on maintenance costs.

[0157] For example, abnormal pressure will cause the entire batch of wafers to be scrapped, so it is given the highest weight; plasma instability can be recovered by restarting; while uniformity degradation develops slowly, allowing for delayed maintenance, so it is given the lowest weight.

[0158] For example, by establishing different weight combinations, a comprehensive health index for the etching equipment can be determined. Then, based on the comprehensive health index of the etching equipment and the quality of the actual etched products, a pre-configured weight combination can be selected.

[0159] S16, Generate a maintenance strategy based on the health index range in which the comprehensive health index is located.

[0160] In some embodiments, different maintenance strategies are adopted for different health index ranges.

[0161] For example, maintenance strategies include: normal production, activation of high-frequency monitoring, and triggering of preventative maintenance checks.

[0162] In one specific embodiment, if CHI ≥ 0.8, normal production is performed; if 0.6 ≤ CHI < 0.8, high-frequency monitoring is activated; if CHI < 0.6, preventive maintenance checks are triggered.

[0163] As can be seen from the above, the predictive maintenance method for etching equipment provided in this application can reduce maintenance costs, improve equipment utilization and production efficiency. Predictive maintenance not only improves the stability and reliability of the manufacturing process, but also promotes the industry's transformation towards intelligent manufacturing.

[0164] In the future, with the further development of artificial intelligence and big data analytics, predictive maintenance technology will become more widespread and mature. It is expected that in the semiconductor manufacturing and related industries, predictive maintenance will become a crucial means of ensuring equipment operational stability and production efficiency, providing strong support for companies to maintain a competitive edge in global competition.

[0165] The present invention also provides a predictive maintenance device for etching equipment corresponding to the above-described predictive maintenance method for etching equipment. The following detailed description is provided with reference to the accompanying drawings and specific embodiments.

[0166] See Figure 3 The diagram shown is a structural schematic of a predictive maintenance device for etching equipment according to an embodiment of this application. Figure 3 As shown, the predictive maintenance device 30 for etching equipment may include:

[0167] The data acquisition unit 31 is configured to acquire multidimensional data of the reaction chamber of the etching equipment. The multidimensional data includes one or more of the following: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value.

[0168] The processing unit 32 is configured to generate a plasma stability index based on the plasma emission spectrum data; and to input the radio frequency power fluctuation value, the reaction gas concentration change rate, and the electrostatic chuck center point temperature value into a feature fusion model to generate etching uniformity prediction parameters, and to input the plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively.

[0169] The maintenance unit 33 is configured to determine the comprehensive health index of the etching equipment based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, and to generate a maintenance strategy based on the health index range in which the comprehensive health index is located.

[0170] The specific working principles and processes of the data acquisition unit 31, processing unit 32 and maintenance unit 33 can be found in the aforementioned example.

[0171] It is understandable that the above division of units is only a logical functional division; in actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the above modules can be implemented by the processor calling software.

[0172] This embodiment also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, wherein when the processor runs the computer program, it performs the steps of the etching equipment predictive maintenance method described above.

[0173] See Figure 4 The diagram illustrates an optional hardware structure of an electronic device provided in one embodiment of this application.

[0174] The device of the present invention includes: at least one processor 41, at least one communication interface 42, at least one memory 43 and at least one communication bus 44.

[0175] In some embodiments, the number of processor 41, communication interface 42, memory 43 and communication bus 44 is at least one, and the processor 41, communication interface 42 and memory 43 communicate with each other through communication bus 44.

[0176] Communication interface 42 can be an interface for a communication module used for network communication, such as an interface for a GSM module.

[0177] The processor 41 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the defect detection method of this embodiment.

[0178] The memory 43 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0179] The memory 43 stores one or more computer instructions, which are executed by the processor 41 to implement the steps of the aforementioned predictive maintenance method for etching equipment.

[0180] It should be noted that the above-mentioned electronic device may also include other devices (not shown) that may not be essential to the content of this application; given that these other devices may not be essential to understanding the application content of the embodiments of this invention, this invention will not describe them one by one.

[0181] Accordingly, the present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, are used to implement the steps of a predictive maintenance method for etching equipment.

[0182] The present invention also provides a computer-readable storage medium storing one or more computer instructions for implementing the steps of a predictive maintenance method for etching equipment.

[0183] It should be understood that the memory in this embodiment can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be ROM, programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0184] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in this embodiment are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer program can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means.

[0185] It should be understood that in various embodiments of the present invention, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of this embodiment.

[0186] In the several embodiments provided by this invention, it should be understood that the disclosed methods, apparatuses, and systems can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and other division methods may exist in actual implementation; for example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0187] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0188] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can be physically included separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or in a combination of hardware and software functional units. For example, for various devices or products applied to or integrated into a chip, each module / unit can be implemented using hardware such as circuits, or at least some modules / units can be implemented using software programs running on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware such as circuits; for various devices or products applied to or integrated into a chip module, each module / unit can be implemented using hardware 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 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.

[0189] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, random access memory (RAM), magnetic disks, or optical disks.

[0190] While the embodiments disclosed in this application are as described above, the invention is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of this invention; therefore, the scope of protection of this invention should be determined by the scope defined in the claims.

Claims

1. A predictive maintenance method for etching equipment, characterized in that, include: Acquire multidimensional data of the reaction chamber of the etching equipment, including: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value; Based on the plasma emission spectrum data, a plasma stability index is generated; The radio frequency power fluctuation value, the rate of change of the reactive gas concentration, and the temperature value of the center point of the electrostatic chuck are input into the feature fusion model to generate etching uniformity prediction parameters. The plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient are input into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively. The comprehensive health index of the etching equipment is determined based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity. A maintenance strategy is generated based on the health index range in which the comprehensive health index falls.

2. The predictive maintenance method for etching equipment according to claim 1, characterized in that, The step of generating plasma stability indices based on the plasma emission spectrum data includes: The plasma emission spectrum data is filtered using a Savitzky-Golay filter to obtain filtered spectrum data; wherein, the Savitzky-Golay filter is provided with a first preset time window, which represents the number of sampling points used in each filtering process; The maximum and minimum values ​​of the spectral intensity of multiple characteristic wavelengths within a second preset time window, as well as a reference value, are obtained from the filtered spectral data, and a weighting coefficient is assigned to each characteristic wavelength; wherein each characteristic wavelength corresponds to a plasma, and the second preset time window is within the first preset time window; The weighted volatility of each characteristic wavelength is determined based on the maximum and minimum values ​​of the spectral intensity of each characteristic wavelength within a second preset time window, as well as the reference value and the corresponding weighting coefficient. The plasma stability index is determined based on the relationship between the weighted volatility of each characteristic wavelength and a preset volatility.

3. The predictive maintenance method for etching equipment according to claim 2, characterized in that, The weighted volatility of each of the characteristic wavelengths is determined using the following formula: in, Indicates weighted volatility, Weighting coefficients representing characteristic wavelengths Indicates the second preset time window, Indicates the first preset time window, The maximum value of the spectral intensity The minimum value representing spectral intensity. Indicates the benchmark reference value, and Greater than ; The reference value is determined by the ratio of the sum of the spectral intensities of each sampling point obtained within the first preset time window of the characteristic wavelength to the total number of sampling points. The step of determining a plasma stability index based on the relationship between the weighted volatility of each of the characteristic wavelengths and a preset volatility includes: responding to the largest weighted volatility among the weighted volatility of the plurality of characteristic wavelengths; in response to the largest weighted volatility being lower than the preset volatility, the plasma stability index having a first identifier; and in response to determining that the largest weighted volatility exceeds the preset volatility, the plasma stability index having a second identifier.

4. The predictive maintenance method for etching equipment according to claim 1, characterized in that, The step of inputting the RF power fluctuation value, the reaction gas concentration change rate, and the electrostatic chuck center point temperature value into the feature fusion model to generate etching uniformity prediction parameters includes: Based on the nonlinear processing unit in the feature fusion model, the rate of change of the concentration of the reactant gas and the fluctuation value of the radio frequency power are nonlinearly processed to generate a dynamic coupling coefficient. When the dynamic coupling coefficient is greater than a preset coefficient, the etching uniformity prediction parameter is forcibly set to zero; When the dynamic coupling coefficient is not greater than a preset coefficient, the dynamic coupling coefficient, the temperature value of the center point of the electrostatic chuck, and the standard deviation of the reaction gas concentration are input into the feature fusion model. Matrix multiplication and addition operations are performed through the hidden layer in the feature fusion model to obtain multidimensional feature space parameters, and the multidimensional feature space parameters are enhanced by the GELU activation function. The enhanced multidimensional feature space parameters are transformed into the etching uniformity prediction parameters using the Sigmoid function. The standard deviation of the reactant gas concentration is obtained by normalizing the rate of change of the reactant gas concentration.

5. The predictive maintenance method for etching equipment according to claim 1, characterized in that, The step of inputting the plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity includes: A four-dimensional input vector is constructed, which includes the chamber pressure gradient, the plasma stability index, the etching uniformity prediction parameter, and the rate of change of the etching uniformity prediction parameter. The four-dimensional input vector is processed using a five-layer residual dilated convolution, including: each layer performing sequential convolution operation, ReLU activation, residual summation on the four-dimensional input vector, and finally performing layer normalization and outputting a feature map; wherein, the kernel width of each convolution layer is 5, and the dilation factors are 1, 2, 4, 8, and 16 respectively. The feature map is subjected to global average pooling using the Softmax function, and an original score is generated through a fully connected layer. Then, after normalization, the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity are output.

6. The predictive maintenance method for etching equipment according to claim 5, characterized in that, The five-layer residual dilated convolution includes: a first layer of dilated convolution to cover plasma flicker; a second layer of dilated convolution to cover gas pressure fluctuations; a third layer of dilated convolution to cover radio frequency power drift; a fourth layer of dilated convolution to cover temperature gradient changes; and a fifth layer of dilated convolution to cover electrode aging accumulation.

7. A predictive maintenance device for etching equipment, characterized in that, include: The data acquisition unit is configured to acquire multidimensional data of the reaction chamber of the etching equipment, including: radio frequency power fluctuation value, reaction gas concentration change rate, chamber pressure gradient and plasma emission spectrum data, and electrostatic chuck center point temperature value. The processing unit is configured to generate a plasma stability index based on the plasma emission spectrum data; and to input the radio frequency power fluctuation value, the reaction gas concentration change rate, and the electrostatic chuck center point temperature value into a feature fusion model to generate etching uniformity prediction parameters, and to input the plasma stability index, the etching uniformity prediction parameters, and the chamber pressure gradient into a temporal convolutional network for collaborative analysis to generate decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, respectively. The maintenance unit is configured to determine the comprehensive health index of the etching equipment based on the decision coding probability values ​​corresponding to the chamber pressure gradient, plasma stability, and etching uniformity, and to generate a maintenance strategy based on the health index range in which the comprehensive health index is located.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the predictive maintenance method for etching equipment according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, the predictive maintenance method for the etching equipment according to any one of claims 1 to 6 is executed.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the predictive maintenance method for etching equipment as described in any one of claims 1 to 6.

Images