Universal and multifunctional polycrystalline diaphragm thickness accurate control system

Abstract

The invention, which relates to the thin film thickness control field, discloses a universal and multifunctional polycrystalline diaphragm thickness accurate control system comprising the following components: a data acquisition layer, a data processing layer, a control strategy layer and a man-machine interaction layer. According to the invention, by adopting a plurality of multi-crystal oscillator detection modules with different ranges and frequencies and cooperating with an advanced data acquisition and processing algorithm, high-precision detection of film thickness is realized, and continuity and accuracy of film thickness data in a range switching process are ensured by an intelligent range switching and self-adaptive compensation mechanism, so that the film thickness detection precision is improved. The problems of insufficient precision and data interruption possibly occurring in a traditional film thickness detection method are avoided, meanwhile, the control strategy layer automatically adjusts coating process parameters according to real-time data, PID parameters are continuously optimized through the self-learning function, and the stability and precision of film thickness control are further improved.

Description

Technical Field

[0001] The invention relates to the technical field of film thickness control, and in particular to a universal and multifunctional polycrystalline oscillator film thickness precision control system. Background Art

[0002] As an important field in modern materials science, thin film deposition technology is widely used in many industries such as semiconductor manufacturing, optical thin films, coating technology, etc. Its core is to form one or more layers of thin films with specific functions and properties on the substrate material through physical or chemical methods. In the thin film deposition process, precise control of film thickness is crucial to ensure the uniformity, density and performance of the film. In recent years, with the advancement of science and technology and the continuous improvement of industrialization needs, the requirements for the accuracy and efficiency of thin film deposition technology are becoming higher and higher, prompting the relevant control systems to continue to develop in the direction of high precision and intelligence.

[0003] Traditional film thickness detection and control methods have exposed a series of shortcomings when facing high-precision and diversified production needs. In terms of accuracy, traditional methods often rely on a single detection method with limited range and are easily disturbed by environmental factors, resulting in insufficient accuracy of film thickness control. Secondly, in terms of adaptability, traditional systems can usually only be set for specific types of thin film materials and processes, lacking flexibility and universality, and are difficult to adapt to complex and changing process scenarios. In addition, traditional systems also have obvious shortcomings in the degree of automation and human-computer interaction experience. The operation is complicated and the data display is not intuitive, making it difficult to meet the requirements of modern industry for intelligence and convenience. Especially in terms of range switching and data processing, traditional systems often lack intelligent switching mechanisms and advanced algorithm support, and are prone to data interruptions or errors, affecting the continuity and stability of thin film deposition.

[0004] In view of the above problems, it is necessary to optimize the existing poly-crystal oscillator film thickness precision control system. By adopting a layered architecture design, through multiple poly-crystal oscillator detection modules with different ranges and frequencies, and in combination with advanced data acquisition and processing algorithms, high-precision detection of film thickness is achieved. Therefore, it is of great significance to develop a universal and multi-functional poly-crystal oscillator film thickness precision control system that can comprehensively realize the above characteristics. Summary of the invention

[0005] The purpose of the present invention is to make up for the deficiencies of the prior art and to provide a universal and multi-functional multi-crystal oscillator film thickness precision control system, which can adopt a layered architecture design, including a data acquisition layer, a data processing layer, a control strategy layer and a human-computer interaction layer, and realize high-precision detection of film thickness through multiple multi-crystal oscillator detection modules with different ranges and frequencies and advanced data acquisition and processing algorithms. At the same time, the system has innovative functions such as integration of intelligent range switching and adaptive compensation, data purification and calibration of multi-algorithm fusion, intelligent PID parameter recommendation and correction of material adaptation, and continuous optimization of PID parameters based on self-learning of historical data, which significantly improves the accuracy and stability of film thickness control. In addition, the intuitive and convenient user interface provides operators with a good human-computer interaction experience, facilitates monitoring and operation of the entire system, and further improves production efficiency and the convenience of process optimization.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a universal and multifunctional polycrystalline oscillating film thickness precision control system, which includes the following components: a data acquisition layer, a data processing layer, a control strategy layer and a human-computer interaction layer;

[0007] The data acquisition layer selects a multi-crystal oscillator detection module with a suitable range and frequency according to the coverage film thickness range and accuracy requirements and installs it at the corresponding position of the coating equipment. The signal acquisition channel is constructed by connecting the data acquisition module circuit through a signal line. The built-in signal acquisition circuit is used to continuously collect the frequency and Q value signals of each multi-crystal oscillator module, and converts them into film thickness data based on the mathematical model algorithm of the quantitative relationship between the crystal oscillator frequency change and the film thickness, and transmits them to the data processing layer according to the set format and transmission protocol;

[0008] The data processing layer is provided with a range monitoring module to compare the film thickness data with the current crystal oscillator range in real time. When the range is exceeded, the next appropriate range crystal oscillator is selected according to a preset rule and a switching instruction is sent through a control circuit to realize switching. At the same time, the film thickness data is compensated in real time by an adaptive compensation algorithm module based on the frequency-film thickness conversion relationship and Q value characteristic difference of the previous and next crystal oscillators, and the digital filtering algorithm module is used to filter the original data according to the set filtering parameters using a filtering algorithm to remove abnormal components. After processing, the crystal oscillator frequency and Q value data are calibrated by the calibration algorithm module according to the calibration coefficient table to make them meet the unified system standard for subsequent processing and transmission;

[0009] The control strategy layer recommends and corrects the material-adaptive intelligent PID parameters. After receiving the information of the thin film material to be deposited, the control strategy layer searches for the corresponding initial PID parameter combination recommendation output through the query matching algorithm through the built-in material property database. At the same time, an interactive control is set in the human-computer interaction interface for the operator to input the correction value according to experience and needs. The PID parameters are updated for control through the parameter update algorithm. After each coating is completed, historical data is collected through the data acquisition interface and stored in the historical database. The machine learning algorithm module is used to mine the rules, and the parameter optimization algorithm module is used to update and replace the existing PID parameters according to the rules and store them for backup.

[0010] The human-computer interaction layer uses a graphical interface design tool to divide the user interface into a real-time data display area and an operation area according to the principle of functional zoning. In the real-time data display area, the current crystal oscillator number, crystal oscillator frequency, Q value life, recent deposition thickness, cumulative thickness, and cumulative film thickness value data of multiple crystal oscillators are displayed at a set update frequency through text boxes and chart components, and a film thickness change curve chart is drawn over time. In the operation area, an input box control is created to receive the parameter modification value entered by the operator, and a button control is set to bind a function code to trigger a parameter update or history record viewing operation, and the modified value is updated to the corresponding parameter storage area according to the process or the history record is retrieved for viewing and analysis.

[0011] Furthermore, the data acquisition layer converts the change in crystal frequency into film thickness data based on a mathematical model algorithm of the quantitative relationship between the change in crystal frequency and the film thickness, and the algorithm formula is: Where d represents the calculated film thickness, k 1 is the proportionality coefficient related to the crystal material and structural characteristics, Δf is the frequency change of the crystal, and f 0 is the initial reference frequency of the crystal oscillator, ρ is the density parameter of the thin film material, k 2 is a correction factor related to the crystal oscillator quality factor Q value, Q is the real-time quality factor of the crystal oscillator, d 0 Initial film thickness offset.

[0012] Furthermore, the data processing layer is provided with a range monitoring module to compare the film thickness data with the current crystal range in real time. When the range is exceeded, the next suitable range crystal is selected according to a preset judgment rule and a switching instruction is sent through the control circuit to realize the switching. The preset judgment rule is: Among them, S switch It is the range switching flag, which is used to determine whether the crystal oscillator range switching operation is required. switch When the value is 1, it means that the current film thickness d exceeds the maximum range d of the jth crystal oscillator being used. max (j), range switching action needs to be performed, when S switchWhen the value of is 0, it means that the film thickness is still within the current crystal oscillator range. No need to switch, keep the current maximum oscillator to continue data collection and detection. d is the current film thickness value, d max (j) is the maximum range of the jth crystal oscillator.

[0013] Furthermore, the data processing layer uses an adaptive compensation algorithm module to compensate the film thickness data in real time based on the frequency-film thickness conversion relationship and Q value characteristic difference of the front and rear crystal oscillators. The compensation algorithm formula is: Where Δf compensated : The frequency change after compensation, Δf is the original frequency change, k f is the frequency compensation coefficient, d is the current film thickness value, d th (j) is the switching threshold film thickness of the jth crystal oscillator, d max (j+1) is the maximum range of the j+1th crystal oscillator, f 0 (j+1) is the initial reference frequency of the j+1th crystal oscillator, f 0 (j) is the initial reference frequency of the jth crystal oscillator.

[0014] Furthermore, the data processing layer uses a digital filtering algorithm module to filter the original data according to the set filtering parameters using a filtering algorithm to remove abnormal components. The algorithm formula is: in, is the prior state estimate at time k, F k is the state transition matrix, is the posterior state estimate at time k-1, B k is the control input matrix, u k is the control input vector, Q k is the process noise covariance matrix, K k is the Kalman gain matrix, H k is the observation matrix, R k is the observation noise covariance matrix, z k is the observation vector at time k. After processing, the crystal frequency and Q value data are calibrated through the calibration algorithm module according to the calibration coefficient table to make them meet the unified standard of the system for subsequent processing and transmission. The calculation formula of the calibration algorithm is: Among them, y calibrated is the crystal oscillator parameter value after calibration, y filtered is the crystal oscillator parameter value after filtering, a 0 , and b 0 Calibration factor.

[0015] Furthermore, the control strategy layer recommends and modifies the material-adaptive intelligent PID parameters. After receiving the information of the thin film material to be deposited, the control strategy layer searches for the corresponding initial PID parameter combination recommendation output through the query matching algorithm through the built-in material property database. The algorithm formula is: Among them, K p is the recommended proportionality factor, ρ m is the density of the thin film material to be deposited, ρ t is the density of the reference full film material, β is the density influence index, E m is the elastic modulus of the thin film material to be deposited, E r is the elastic modulus of the reference film material, γ is the elastic modulus influence index, is the base scale factor.

[0016] Furthermore, the control strategy layer human-machine interaction interface is provided with interactive controls for operators to input correction values ​​according to experience and needs, and the PID parameters are updated for control through the parameter update algorithm, and the algorithm formula is: in, is the corrected proportionality factor, is the proportionality coefficient before correction, K p Is the basic data for correction operation, ΔK p is the proportional coefficient correction

[0017] Furthermore, the control strategy layer uses the machine learning algorithm module to mine the rules, and updates and replaces the existing PID parameters according to the rules through the parameter optimization algorithm module and stores them for backup. Specifically, for each neuron i in the input layer, the connection weight between it and the neuron j in the hidden layer is w ij , the input net of hidden layer neuron j j The calculation formula is: net j =∑ i w ij × i +b j , where x i is the input value of neuron i in the input layer, b j is the bias term of hidden layer neuron j, and the output a of hidden layer neuron j j It is calculated by the activation function f(·), and the formula is: The connection weight between hidden layer neuron j and output layer neuron k is w jk , where k corresponds to the three parameters K of PID p , K i , K d , the input net of output layer neuron k k The calculation formula is: net k =∑ j w jk ×a j +b k , the output y of the output layer neuron k k Calculated by activation function: y k =netk , then the optimized PID parameter calculation formula is: in is the optimized proportionality coefficient, is the scaling factor currently in use, is the K corresponding to the output layer of the neural network p The adjustment amount of the neuron output can be calculated in the same way. and Among them, w ij represents the connection weight between input layer neuron i and hidden layer neuron j, w jk represents the connection weight between hidden layer neuron j and output layer neuron k, b j is the bias term of hidden layer neuron j, b k is the bias term of neuron k in the output layer, x i is the input value of the input layer, a j is the output value of the hidden layer neuron, net j and net k are the weighted sum of the inputs of hidden layer neuron j and output layer neuron k, y k The output value of the output layer neuron, that is, the PID parameter adjustment amount, and are the optimized PID parameters.

[0018] Compared with the prior art, this universal and multifunctional polycrystalline oscillator film thickness precision control system has the following beneficial effects:

[0019] 1. The present invention realizes high-precision detection of film thickness by adopting multiple multi-crystal oscillator detection modules with different ranges and frequencies, and cooperating with advanced data acquisition and processing algorithms. The intelligent range switching and adaptive compensation mechanism ensure the continuity and accuracy of film thickness data during the range switching process, avoiding the problems of insufficient accuracy and data interruption that may occur in traditional film thickness detection methods. At the same time, the control strategy layer automatically adjusts the coating process parameters according to real-time data, and continuously optimizes the PID parameters through the self-learning function, further improving the stability and accuracy of film thickness control.

[0020] 2. The present invention is compatible with a variety of coating equipment software and shows good software environment adaptability in practical applications. In addition, the system also has intelligent features, such as material-adaptive intelligent PID parameter recommendation and correction, continuous optimization of PID parameters based on self-learning of historical data, etc., which enables the system to automatically adjust and optimize control parameters according to the characteristics of thin film materials and historical coating data, thereby improving process adaptability and automation.

[0021] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 It is a universal and multifunctional polycrystalline diaphragm thickness precision control system process operation diagram;

[0024] Figure 2 The flowchart is a universal and multifunctional polycrystalline diaphragm thickness precision control system. DETAILED DESCRIPTION

[0025] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0026] Embodiment 1

[0027] This embodiment describes in detail the specific application of a universal and multifunctional polycrystalline oscillator film thickness precision control system in the field of optical thin film coating. Through the present invention, the film thickness can be accurately controlled, the yield rate and optical performance consistency of optical thin film products can be effectively improved, and the needs of high-precision optical thin film manufacturing can be met.

[0028] The thickness range of optical thin films usually spans a large range, ranging from tens of nanometers (such as some high-precision filter films) to several microns (such as thick film layers in some multi-layer optical thin film structures), and different optical thin film application scenarios have different requirements for film thickness accuracy, generally requiring accuracy in the range of several nanometers to tens of nanometers. Based on this, multiple crystal oscillator detection modules with different ranges and frequencies are selected to cover the required film thickness detection range. For example, three different specifications of quartz crystal oscillator detection modules are selected, with ranges of 100nm-1μm, 1μm-5μm and 5μm-10μm, and frequency ranges from several megahertz to tens of megahertz to meet the detection needs in the deposition process of optical thin films of different thicknesses. These crystal oscillator modules are carefully installed in the vacuum chamber of the optical coating equipment, and the installation position is close to the substrate material (such as glass lens substrate, etc.), and the environment is ensured to be stable to avoid unnecessary external vibration interference and interference with other components in the coating process, while ensuring that the frequency and Q value change signals caused by the deposition of thin films on the crystal oscillator surface can be accurately collected.

[0029] Start the optical coating equipment and start depositing thin films (assuming that the film deposited this time is titanium dioxide TiO for telescope lenses). 2 After the anti-reflection film is applied, the data acquisition module continuously monitors the frequency and Q value of each multi-crystal oscillator detection module in real time at a sampling frequency of multiple times per second through the built-in high-precision signal acquisition circuit, and uses the core algorithm formula for film thickness calculation Data conversion is performed to calculate the film thickness data. For the titanium dioxide film being deposited, the system obtains its density ρ of about 4.23 g / cm from the pre-established material property database. 3 , k 1 As a proportionality factor related to the crystal material (quartz) and structural characteristics (such as electrode shape, crystal cutting method, etc.), it is determined by multiple measurements and calibrations of the selected specific multi-crystal oscillator detection module under standard, known film thickness deposition experimental conditions. It is assumed that its value is 0.001 (the unit is determined by the overall dimension coordination of the formula), k 2 It is a correction factor related to the crystal quality factor Q value. After a lot of experimental tests and data analysis, its value is determined to be 0.05 for this system and the common optical thin film deposition process range. The initial reference frequency f of the crystal oscillator is 0 , when the system starts up, a calibration measurement is performed for each crystal oscillator module and recorded. For example, the f 0 The frequency change Δf monitored in real time is the difference between the current frequency f and f 0 For example, if f is currently monitored to be 9.99MHz, then Δf=9.99-10=-0.01MHz. The Q value is measured in real time by a dedicated circuit module. Assuming that the current Q value is 1000, in addition, d 0As the initial film thickness offset, taking into account the slight deviations such as the extremely thin background contamination film layer that may exist on the surface of the crystal oscillator, the value is determined to be 5nm (which can be adjusted according to the actual calibration situation) by calibrating and measuring the system under standard experimental conditions and analyzing and evaluating the common initial states. By substituting these parameter values ​​into the formula, the real-time film thickness data can be calculated, and then the film thickness data is transmitted to the data processing layer according to the set data transmission protocol.

[0030] During the deposition of titanium dioxide thin films, as the film thickness gradually increases, the data processing layer uses the range switching decision formula Real-time judgment of whether the film thickness exceeds the current crystal oscillator range, where d is the current film thickness value calculated by the data acquisition layer, d max (j) represents the maximum range of the jth crystal oscillator. For example, in the early stage of deposition, a crystal oscillator with a range of 100nm-1μm is used for detection. When the film thickness d exceeds 1μm (assuming that the current film thickness reaches 1.2μm), S switch The value of becomes 1, indicating that the film thickness exceeds the current crystal oscillator range. At this time, it is necessary to automatically switch to the next appropriate range (i.e., a crystal oscillator with a range of 1μm-5μm) according to the preset rules to continue testing. After the range is switched, in order to ensure that the film thickness data remains continuous and accurate at the moment of switching and in the subsequent process, the adaptive compensation formula is used. Compensate the film thickness data, where Δf compensated is the frequency change after compensation, Δf is the original frequency change (i.e., the frequency change obtained from the data acquisition layer before switching), k f is the frequency compensation coefficient. Through a large number of experimental tests and theoretical analysis, its value is determined to be 0.8 for different crystal oscillator combinations in this system. th (j) is set to a film thickness value close to the maximum range of the jth crystal oscillator. Here, for a crystal oscillator with a range of 100nm-1μm, d th (j) is set to 0.9 μm, which is a key parameter for defining the range switching transition interval. max (j+1) is the maximum range of the j+1th crystal oscillator (the crystal oscillator to be switched to, here is the crystal oscillator with a range of 1μm-5μm), that is, 5μm, f 0 (j+1) and f 0 (j) are the initial reference frequencies of the j+1th crystal oscillator and the jth crystal oscillator, respectively. Assume that f 0 (j+1) is 8MHz, f 0 (j) is 10MHz. By substituting these parameters into calculations, the compensated frequency change is obtained, and then the film thickness is accurately calculated based on the compensated frequency change and other information to avoid affecting the accuracy of film thickness monitoring due to range switching, thereby achieving seamless connection over the range.

[0031] After receiving the collected data, the Kalman filter algorithm is first used to filter the raw data to remove high-frequency noise and abnormal signals caused by factors such as equipment electrical noise and environmental interference, making the data smoother and more stable. Kalman filtering involves multiple equations. Specifically, the state prediction equation middle, Represents the prior state estimate (prediction value) at time k. Here, it is used to represent the estimated value of the crystal oscillator frequency or Q value at time k obtained based on the state estimate at the previous time (time k-1) and the system dynamic model prediction. For example, it predicts the change of the crystal oscillator frequency at the next time. k is the state transfer matrix, which is determined according to the physical model of the crystal oscillator and the dynamic characteristics of the influence of the optical thin film deposition process on the crystal oscillator. For the state variable of the crystal oscillator frequency, if its simple linear change relationship at adjacent moments and the influence of factors such as deposition rate are considered, its element values ​​are determined through experimental analysis and theoretical modeling. Assume that F k is a 1×1 matrix with an element value of 0.9 (indicating an attenuation trend of frequency change, just for example). is the a posteriori state estimate (updated value) at time k-1, that is, the accurate crystal frequency or Q value and other parameter values ​​obtained after filtering at the previous moment, B k is the control input matrix. In the process of optical thin film deposition, although there are relatively few externally controllable factors affecting the crystal state, considering that the deposition power may have a certain indirect effect on the crystal, it is assumed here that B k is a 1×1 matrix, and the element value is determined to be 0.01 (for example only) based on the experimental relationship between deposition power and frequency change. k is the control input vector, corresponding to the vector of adjustable parameter values ​​such as deposition power. If the current deposition power is set to 100W, then u k The value of is 100 (the unit is determined according to the actual situation, and is only indicated here), and the covariance prediction equation inside, is the prior estimation error covariance matrix at time k, which is used to measure the covariance of the error between the predicted value and the true value when predicting the state at time k. Assuming the initial The diagonal elements of the matrix (representing the frequency prediction error variance) are set to 0.1 based on previous experience and calibration (the unit is determined by the overall formula coordination), Q k is the process noise covariance matrix, which reflects the covariance of the noise introduced by various unpredictable interference factors (such as small environmental fluctuations, small random vibration changes of the crystal oscillator itself, etc.) during the state transition from time k-1 to time k. Through the analysis and experimental measurement of interference factors in the operation of optical coating equipment, it is assumed that Q k The matrix diagonal elements are set to 0.01, and the Kalman gain calculation equation In, K k is the Kalman gain matrix, which is dynamically calculated based on the prior estimation error covariance, observation model, and observation noise covariance, and is used to weigh the weights of the predicted value and the observed value when updating the state estimate. k is the observation matrix, which establishes the connection between the system state variables (such as crystal oscillator frequency) and the actual observable data (the original crystal oscillator frequency data collected by the data acquisition layer). Assume that H k is a 1×1 matrix with element value 1 (indicating a direct observation relationship, just a simple example), R k is the observation noise covariance matrix, which describes the covariance of the noise in the observation data caused by factors such as sensor accuracy limitation and electromagnetic interference during the observation process (i.e., data collection process). Through the noise analysis of sensor performance test and actual collected data, assuming that R k The matrix diagonal elements are set to 0.05 (just for example), and the state update equation middle, is the posterior state estimate (updated value) at time k, that is, after the Kalman filter algorithm, using the observation data z at time k k Prior state estimates The final state estimate obtained after correction, that is, the accurate crystal frequency or Q value and other parameter values ​​at time k obtained after filtering, will be used as input data for calibration and other data processing operations in the future. Using the pre-stored calibration coefficient a 0 、b 0 Calibrate the filtered crystal frequency and Q value data to convert them into accurate values ​​that meet the unified standards within the system. 0 and b 0 It is a coefficient determined in advance by conducting a large number of calibration experiments on the multi-crystal oscillator detection module under standard, known precise film thickness and ideal environmental conditions. For example, for the frequency calibration of a selected crystal oscillator, the frequency is measured multiple times under known film thickness deposition conditions and compared with the standard value to determine a 0 is 1.05 (used to scale the filtered data to make it conform to the system standard range or proportional relationship), b 0 is 0.1 (used to correct the offset and eliminate possible system fixed deviations, etc.), the filtered crystal oscillator frequency and Q value are substituted into the formula to obtain the accurate value after calibration, and the processed accurate film thickness data is transmitted to the control strategy layer.

[0032] For the titanium dioxide thin film material to be deposited, the control strategy layer is based on the PID parameter initial value recommendation formula based on material characteristics (with the proportional coefficient K p For example, ) automatically recommends the corresponding PID parameters and obtains the elastic modulus E of titanium dioxide from the material properties database m About 200 GPa and other parameters, select a commonly used reference film material (such as silicon dioxide (SiO 2 ), its density ρ r About 2.2g / cm 3 , elastic modulus E r It is about 70GPa. α is used as the comprehensive adjustment coefficient. After a large number of deposition experiments on different optical thin film materials and the analysis of the corresponding film thickness control effects, its value is determined to be 0.8 (the value range is determined according to the overall design and actual debugging experience of the system, and is used to adjust the K calculated based on material properties. p The overall scaling adjustment is performed), β is the density influence index, and its value is 0.6 (used to describe the film material density to the proportional coefficient K) after fitting and analyzing a large number of experimental data. p γ is the elastic modulus influence index, and its value is determined to be 0.4 through experimental analysis (reflecting the change of elastic modulus when K p The sensitivity of the corresponding changes), It is the initial value of the proportionality coefficient determined for the reference thin film material silicon dioxide under the standard, fully debugged and optimized film thickness control process conditions, assuming it is 0.5 (dimensionless, reflecting the basic response strength setting of the system to film thickness control under the benchmark condition). Substituting these parameter values ​​into the formula, the recommended K is calculated. p Similarly, K can be calculated i (integral coefficient) and K d The recommended value of (differential coefficient) can be input by the operator through the human-computer interaction interface based on the past experience of producing similar optical films and the actual observations in the current coating process, and the PID parameter correction formula can be used to input the correction value. For example, if the operator finds that the film thickness growth rate in the current deposition process is slightly faster than expected, the operator can manually modify the recommended PID parameters through the corresponding K p Enter the correction value ΔK in the input box p =-0.1 (indicates reducing the proportional coefficient to slow down the film thickness growth rate), the system uses the parameter update algorithm to manually correct K p The values ​​are updated and used in the control algorithm to ensure that the parameters are more in line with the actual process requirements.

[0033] After each coating process, the system automatically collects and organizes the historical data related to this process, such as the material information of the deposited titanium dioxide film (including material name, density, elastic modulus, etc.), the actual PID parameters used (K p , K i , K dThese data are stored in the historical database according to specific data structures and classification rules to facilitate subsequent query and analysis. For example, the coating batch number is used as the primary index, and each batch records the above-mentioned parameter information corresponding to each stage of the coating in detail, and sets the corresponding field format for different data types for storage to ensure the integrity and standardization of the data. First, the film thickness deviation data is normalized, and then the normalized historical data is deeply mined and analyzed through the PID parameter optimization formula based on the neural network (involving the forward propagation of the neural network and other calculation processes). It is assumed that a simple three-layer neural network structure (which can be more complex and have more layers in actual applications to adapt to more complex data relationships) is used for analysis and mining. In the input layer, the normalized film deviation, deposition time, material density, elastic modulus and other historical data features are used as input values ​​of the input neurons. For example, there are 5 input neurons, corresponding to the above different feature data, and the connection weight of each neuron and hidden neuron j is w ij , the input net of hidden layer neuron j j By calculating the formula net j =∑ i w ij × i +b j Calculated, where b j is the bias term of the hidden layer neuron j, and its initial value can be set randomly (such as 0.1, etc.), then the output a of the hidden layer neuron j j Through the activation function f(·) (here the commonly used Sigmoid function For example, it can perform non-central transformation and feature capture on the input data. Between the hidden layer and the output layer, the hidden layer neuron j and the output layer neuron k (here corresponds to the three parameters K of PID p , K i , K d , assuming that k represents the output neurons corresponding to these three parameters) the connection weight is w jk , the input net of output layer neuron k k The calculation formula is net k =∑ j w jk ×a j +b k , where b k is the bias term of the output layer neuron k, which can also be initially set to a value (such as 02, etc.). The output y of the output layer neuron k is k(i.e., the corresponding PID parameter adjustment amount) is calculated by the activation function (which can be a linear function here, because the final output PIC parameter adjustment amount is in the real number range) to obtain y k =net k For example, after the calculation of the neural network, the output layer corresponds to K p The amount of adjustment of the neuron output is 0.05, if the current proportional coefficient is is 04, then the final optimized proportional affinity Similarly, we can calculate and By continuously performing such neural network learning and analysis on a large amount of historical data, we can dig out better PID parameter combination rules under different process conditions and different film thickness stages. Based on the better PID parameter combination rules obtained through the mining and analysis, the existing PID parameters are updated and replaced through the parameter optimization algorithm module, and the updated parameters are stored in the parameter library, and the corresponding applicable conditions (such as specific titanium dioxide film thickness range, deposition rate range, etc.) are marked so that they can be directly called and used in subsequent similar coating scenarios. For example, after learning historical data accumulated through multiple coating processes, it is found that for titanium dioxide films with a thickness in the range of 200nm-500nm and a deposition rate of 1nm / s-2nm / s, the new PID parameter combination can significantly improve the film thickness control accuracy. In subsequent coating, when the corresponding conditions are met, the system will automatically apply the optimized parameters, thereby improving the film thickness control accuracy and making the optical performance of the manufactured optical film more stable and in line with standard requirements.

[0034] During the optical thin film coating process, the operator monitors various data in real time through the interface of the human-computer interaction layer. The interface adopts a graphical design and divides different display areas. For example, in a main real-time data display area, clear text boxes are used to display the current crystal oscillator number (such as showing that the current working one is crystal oscillator No. 2 with a range of 1μm-5μm), crystal oscillator frequency (displaying the accurate frequency value currently monitored, such as 8.5MHz), Q value life (displaying the remaining available Q value in percentage form, assuming that the current display is 80%, indicating that the Q value is still in a good state and can continue to work reliably), and the latest deposition thickness (displaying the thickness of the film deposited in the last data acquisition cycle, such as showing that 5n was deposited in the last 10 seconds). m), cumulative thickness (displaying the total film deposition thickness from the start of coating to the current moment, for example, the current cumulative deposition has reached 300nm) and cumulative film thickness values ​​of multiple crystal oscillators (if multiple crystal oscillators are involved in detecting the film thickness at different stages, the cumulative film thickness corresponding to each crystal oscillator will be summarized and displayed here to facilitate comparison and comprehensive understanding of the film thickness information), etc., to intuitively understand the film thickness growth and the working status of the crystal oscillator. At the same time, it will also display the trend of film thickness change over time in real time by drawing curve charts. The horizontal axis is the time axis (in seconds) and the vertical axis is the film thickness (in nanometers). The operator can intuitively see whether the film thickness growth curve is stable and in line with expectations, so as to promptly discover potential coating abnormalities. The operator can Through the interactive controls such as input boxes and buttons on the interface, according to the design requirements of different optical thin film products, the parameters such as the crystal oscillator frequency and Q value loss corresponding to the set thickness of any thin film material can be customized. For example, for the anti-reflection film design of a new telescope lens, the target thickness is changed from the original 500nm to 600nm. The operator only needs to enter 600 in the corresponding "target film thickness" input box. The system will automatically associate and calculate and update the corresponding crystal oscillator frequency and the expected Q value loss and other parameter reference values ​​according to the built-in algorithm (these reference values ​​will be based on the previously accumulated coating data of the same type of materials and the algorithm model estimation), and display them in real time on the interface for the operator to further confirm and fine-tune. In addition, if the operator finds that The Q value of the front crystal oscillator decreases quickly, which may affect the accuracy of subsequent film thickness detection. The specially set "Q value compensation adjustment" button can also be used to trigger the system to implement corresponding Q value compensation strategies (such as appropriately adjusting the data acquisition frequency or enabling the backup calibration algorithm, etc., which are specifically implemented according to the system's preset compensation mechanism) to ensure the accuracy of the data during the entire coating process. Operators can view historical records at any time. The interface provides a convenient query and filtering function, which can be filtered and queried according to the coating product name, coating date, thin film material and other conditions. For example, if you want to view all the coating records of titanium dioxide anti-reflection films in the past month, after selecting the corresponding filtering conditions, the system will list the coating batches that meet the requirements. Click on the specific batch record.You can view in detail the various process data of the coating process, including film thickness at different times, PID parameter changes, crystal oscillator working status and other information, compare the process data of different batches of the same type of optical film, and analyze the film thickness control effect. For example, through comparison, it is found that in two coatings, although the PID parameters used are the same, the film thickness deviation is different. After further checking other related parameters, it is found that this is caused by a slight difference in the substrate temperature during the two coatings. This can provide a basis for further optimizing the process (such as more accurate control of the substrate temperature, etc.), and the data of the entire deposition process will be saved in real time to meet the requirements of optical film production for process traceability and quality control.

[0035] In summary, the present invention can accurately control the film thickness, effectively improve the yield rate and optical performance consistency of optical thin film products, and meet the needs of high-precision optical thin film manufacturing. Whether it is for the production of common optical lens anti-reflection films, filters and other products, or for the manufacturing of some special optical thin film application scenarios, it can provide reliable and accurate film thickness control guarantees, helping the optical thin film industry to improve product quality and production efficiency.

[0036] Embodiment 2

[0037] This embodiment describes in detail the application of a universal and multifunctional polycrystalline oscillator film thickness precision control system for thin film deposition in semiconductor chip manufacturing. Through the present invention, the accuracy and stability of the thin film deposition process in the chip manufacturing process can be effectively improved, the high performance and high reliability of the chip can be guaranteed, and the strict requirements of the semiconductor industry for advanced manufacturing technology can be met.

[0038] According to the high-precision film thickness range requirements of semiconductor thin film deposition, select multi-crystal oscillator detection modules with high resolution, small range and high-precision frequency detection capabilities. For example, select a special crystal oscillator module with a range between 1nm-100nm and high frequency stability, and carefully install it in the appropriate position in the reaction chamber of the semiconductor thin film deposition equipment to ensure that the frequency and Q value signals of the crystal oscillator can be accurately obtained, and the signal acquisition process will not interfere with the clean environment and process of semiconductor manufacturing.

[0039] When copper metal film is used as the chip interconnection circuit layer, the data acquisition module continuously monitors the crystal oscillator related parameters in real time, uses the film thickness calculation algorithm formula, and combines the copper film density (about 8.96g / cm 3 ) and the corresponding coefficients of the crystal oscillator, real-time frequency, and Q value changes, accurately calculate the film thickness data and transmit it to the data processing layer.

[0040] During the deposition of metal thin films, due to the extremely high requirements for film thickness accuracy, the data processing layer closely monitors the relationship between film thickness and crystal oscillator range. When the film thickness approaches the upper limit of the current crystal oscillator range (for example, the film thickness reaches the current crystal oscillator range of 90nm, close to the maximum range of 100nm), the range switching decision formula is used to quickly judge and switch to the next appropriate range of crystal oscillator for continued detection. At the same time, the film thickness data is accurately compensated by the adaptive compensation algorithm to ensure the accuracy and continuity of the film thickness data during the range switching process, and to avoid the accumulation of very small film thickness errors affecting the electrical performance of the chip. For example, the adaptive compensation formula is used to accurately adjust the frequency change compensation amount caused by the crystal oscillator switching, so that the subsequent film thickness calculation is not affected. After receiving the collected data, the Kalman filter and other digital filtering algorithms are also used to remove the noise interference introduced by the complex electromagnetic environment and small process fluctuations inside the semiconductor manufacturing equipment, so that the crystal oscillator frequency and Q value data are purer and more stable. Then, according to the calibration algorithm, the filtered data is calibrated using the coefficients pre-calibrated for the semiconductor thin film deposition process, and converted into accurate film thickness data that meets the high-precision requirements of semiconductor chip manufacturing, and transmitted to the control strategy layer.

[0041] For depositing copper metal thin films, the control strategy layer automatically recommends appropriate PID parameters based on material properties. The PID parameter initial value recommendation formula based on material properties is used to compare the material property parameters of copper (such as density, elastic modulus, etc.) with the system benchmark reference material. The initial PID parameters are generated by combining the corresponding adjustment coefficient and influence index. The operator can adjust the PID parameters by inputting correction values ​​through the human-computer interaction interface according to the chip manufacturing process specifications and real-time observation of the current deposition process to ensure that the film thickness control can accurately meet the design requirements of each layer of the chip film and ensure the consistency and reliability of the chip electrical performance. After the film deposition process in each chip manufacturing, The system collects and organizes detailed historical data related to copper thin film deposition, including material parameters, actual PID parameters used, film thickness deviation, etc. After normalizing the film thickness deviation and other data using the data normalization formula, it uses the neural network-based PID parameter optimization algorithm to mine better PID parameter combination rules from a large amount of historical data. For example, with the continuous accumulation of historical data on copper thin film deposition in the chip manufacturing process, the system learns more appropriate PID parameter adjustment methods under different process temperatures, deposition rates and other conditions. The optimized parameters are applied in subsequent chip manufacturing to further improve the film thickness control accuracy, which helps to improve the overall performance and yield of the chip.

[0042] In the thin film deposition process of semiconductor chip manufacturing, operators can view various key data in real time through the human-computer interaction interface, such as the current working status of the crystal oscillator (frequency, Q value, etc.), real-time film thickness, cumulative film thickness and other information, so as to timely grasp the progress and quality of thin film deposition. At the same time, according to the changes in chip design or the process adjustment requirements of different batches of chip manufacturing, the relevant parameters corresponding to the set thickness of the deposited thin film material can be conveniently modified through the interactive controls on the interface, and historical records can be viewed at any time to compare the process data differences of thin film deposition of different batches of chips, analyze the film thickness control effect and the impact on chip performance, and provide strong data support for optimizing the chip manufacturing process. In addition, the data of the entire deposition process will be strictly saved in real time to meet the data traceability and quality control requirements of the semiconductor manufacturing industry.

[0043] In summary, the present invention can effectively improve the accuracy and stability of the thin film deposition process in the chip manufacturing process, ensure the high performance and high reliability of the chip, and meet the semiconductor industry's strict requirements for advanced manufacturing technology.

[0044] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A universal and multifunctional polycrystalline diaphragm thickness precision control system, characterized in that: The system includes the following components: data acquisition layer, data processing layer, control strategy layer and human-computer interaction layer; The data acquisition layer selects a multi-crystal oscillator detection module with a suitable range and frequency according to the coverage film thickness range and accuracy requirements and installs it at the corresponding position of the coating equipment. The signal acquisition channel is constructed by connecting the data acquisition module circuit through a signal line. The built-in signal acquisition circuit is used to continuously collect the frequency and Q value signals of each multi-crystal oscillator module, and converts them into film thickness data based on the mathematical model algorithm of the quantitative relationship between the crystal oscillator frequency change and the film thickness, and transmits them to the data processing layer according to the set format and transmission protocol; The data processing layer is provided with a range monitoring module to compare the film thickness data with the current crystal oscillator range in real time. When the range is exceeded, the next appropriate range crystal oscillator is selected according to a preset rule and a switching instruction is sent through a control circuit to realize switching. At the same time, the film thickness data is compensated in real time by an adaptive compensation algorithm module based on the frequency-film thickness conversion relationship and Q value characteristic difference of the previous and next crystal oscillators, and the digital filtering algorithm module is used to filter the original data according to the set filtering parameters using a filtering algorithm to remove abnormal components. After processing, the crystal oscillator frequency and Q value data are calibrated by the calibration algorithm module according to the calibration coefficient table to make them meet the unified system standard for subsequent processing and transmission; The control strategy layer recommends and corrects the material-adaptive intelligent PID parameters. After receiving the information of the thin film material to be deposited, the control strategy layer searches for the corresponding initial PID parameter combination recommendation output through the query matching algorithm through the built-in material property database. At the same time, an interactive control is set in the human-computer interaction interface for the operator to input the correction value according to experience and needs. The PID parameters are updated for control through the parameter update algorithm. After each coating is completed, historical data is collected through the data acquisition interface and stored in the historical database. The machine learning algorithm module is used to mine the rules, and the parameter optimization algorithm module is used to update and replace the existing PID parameters according to the rules and store them for backup. The human-computer interaction layer uses a graphical interface design tool to divide the user interface into a real-time data display area and an operation area according to the principle of functional zoning. In the real-time data display area, the current crystal oscillator number, crystal oscillator frequency, Q value life, recent deposition thickness, cumulative thickness, and cumulative film thickness value data of multiple crystal oscillators are displayed at a set update frequency through text boxes and chart components, and a film thickness change curve chart is drawn over time. In the operation area, an input box control is created to receive the parameter modification value entered by the operator, and a button control is set to bind a function code to trigger a parameter update or history record viewing operation, and the modified value is updated to the corresponding parameter storage area according to the process or the history record is retrieved for viewing and analysis.

2. A universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1, characterized in that: The data acquisition layer converts the change in crystal frequency into film thickness data based on a mathematical model algorithm of the quantitative relationship between the change in crystal frequency and the film thickness. The algorithm formula is: Among them, d represents the calculated film thickness value, k1 is the proportional coefficient related to the crystal material and structural characteristics, Δf is the frequency change of the crystal, f0 is the initial reference frequency of the crystal, ρ is the density parameter of the thin film material, k2 is the correction coefficient related to the crystal quality factor Q value, Q is the real-time quality factor of the crystal, and d0 is the initial film thickness offset.

3. The universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1 is characterized in that: The data processing layer is provided with a range monitoring module to compare the film thickness data with the current crystal range in real time. When the range is exceeded, the next appropriate range crystal is selected according to the preset judgment rule, and a switching instruction is sent through the control circuit to realize the switching. The preset judgment rule is: Among them, S switch It is the range switching flag, which is used to determine whether the crystal oscillator range switching operation is required. switch When the value is 1, it means that the current film thickness d exceeds the maximum range d of the jth crystal oscillator being used. max (j), range switching action needs to be performed, when S switch When the value of is 0, it means that the film thickness is still within the current crystal oscillator range. No need to switch, keep the current maximum oscillator to continue data collection and detection. d is the current film thickness value, d max (j) is the maximum range of the jth crystal oscillator.

4. The universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1 is characterized in that: The data processing layer uses an adaptive compensation algorithm module to compensate the film thickness data in real time based on the frequency-film thickness conversion relationship of the front and rear crystal oscillators and the Q value characteristic difference. The compensation algorithm formula is: Where Δf compensated : The frequency change after compensation, Δf is the original frequency change, k f is the frequency compensation coefficient, d is the current film thickness value, d th (j) is the switching threshold film thickness of the jth crystal oscillator, d max (j+1) is the maximum range of the j+1th crystal oscillator, f0(j+1) is the initial reference frequency of the j+1th crystal oscillator, and f0(j) is the initial reference frequency of the jth crystal oscillator.

5. The universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1 is characterized in that: The data processing layer uses a digital filtering algorithm module to filter the original data and remove abnormal components according to the set filtering parameters using a filtering algorithm. The algorithm formula is: in, is the prior state estimate at time k, F k is the state transition matrix, is the posterior state estimate at time k-1, B k is the control input matrix, u k is the control input vector, Q k is the process noise covariance matrix, K k is the Kalman gain matrix, H k is the observation matrix, R k is the observation noise covariance matrix, z k is the observation vector at time k. After processing, the crystal frequency and Q value data are calibrated through the calibration algorithm module according to the calibration coefficient table to make them meet the unified standard of the system for subsequent processing and transmission. The calculation formula of the calibration algorithm is: Among them, y calibrated is the crystal oscillator parameter value after calibration, y filtered are the filtered crystal parameter values, a0, and b0 calibration coefficients.

6. The universal and multifunctional polycrystalline oscillator film thickness precision control system according to claim 1 is characterized in that: The control strategy layer recommends and modifies the material-adaptive intelligent PID parameters. After receiving the thin film material information to be deposited, the control strategy layer searches for the corresponding initial PID parameter combination recommendation output through the query matching algorithm through the built-in material property database. The algorithm formula is: Among them, K p is the recommended proportionality factor, ρ m is the density of the thin film material to be deposited, ρ t is the density of the reference full film material, β is the density influence index, E m is the elastic modulus of the thin film material to be deposited, E r is the elastic modulus of the reference film material, γ is the elastic modulus influence index, is the base scale factor.

7. The universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1 is characterized in that: The control strategy layer human-machine interaction interface is provided with interactive controls for operators to input correction values ​​according to experience and needs, and the PID parameters are updated for control through the parameter update algorithm, and the algorithm formula is: in, is the corrected proportionality factor, is the proportionality coefficient before correction, K p Is the basic data for correction operation, ΔK p is the proportional coefficient correction.

8. The universal and multifunctional polycrystalline diaphragm thickness precision control system according to claim 1 is characterized in that: The control strategy layer uses the machine learning algorithm module to mine the rules, and updates and replaces the existing PID parameters according to the rules through the parameter optimization algorithm module and stores them for backup. Specifically, for each neuron i in the input layer, the connection weight between it and the neuron j in the hidden layer is w ij , the input net of hidden layer neuron j j The calculation formula is: net j =∑ i w ij × i +b j , where x i is the input value of neuron i in the input layer, b j is the bias term of hidden layer neuron j, and the output a of hidden layer neuron j j It is calculated by the activation function f(·), and the formula is: The connection weight between hidden layer neuron j and output layer neuron k is w jk , where k corresponds to the three parameters K of PID p , K i , K d , the input net of output layer neuron k k The calculation formula is: net k =∑ j w jk ×a j +b k , the output y of neuron k in the output layer k Calculated by activation function: y k =net k , then the optimized PID parameter calculation formula is: in is the optimized proportionality coefficient, is the scaling factor currently in use, is the K corresponding to the output layer of the neural network p The adjustment amount of the neuron output can be calculated in the same way. and Among them, w ij represents the connection weight between input layer neuron i and hidden layer neuron j, w jk represents the connection weight between hidden layer neuron j and output layer neuron k, b j is the bias term of hidden layer neuron j, b k is the bias term of neuron k in the output layer, x i is the input value of the input layer, a j is the output value of the hidden layer neuron, net j and net k are the weighted sum of the inputs of hidden layer neuron j and output layer neuron k, y k The output value of the output layer neuron, that is, the PID parameter adjustment amount, and are the optimized PID parameters.

Images