Ignition control method of remote plasma source

Abstract

The invention discloses an ignition control method of a remote plasma source in the technical field of remote plasma sources, which comprises the following steps: S1, setting a gas current excitation threshold value, and introducing excitation gas; s2, judging whether an ignition signal is received or not, and if yes, closing a relay and entering a constant current stage; s3, whether system ignition is normal or not is judged, and if not, the ignition voltage is adjusted through a self-learning type ignition threshold value calibration method; s4, the output power is improved, and the ignition voltage is adjusted through an ignition threshold calibration method during the period; s5, entering a constant power stage when the output power reaches the full-load power; according to the self-learning type ignition threshold value calibration method, gas flow, gas pressure and cavity temperature are collected in real time, a current load impedance value is predicted through an LSF-BP model, and an ignition voltage threshold value is updated according to the current load impedance value. According to the method, the ignition threshold is dynamically adjusted according to the operation condition, the ignition stability can be effectively improved, and equipment damage and maintenance cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of remote plasma sources, and particularly to a method for controlling the ignition of a remote plasma source. Background Art

[0002] As an efficient and controllable plasma generation technology, the remote plasma source (RPS) has a wide range of applications in the fields of semiconductor manufacturing, material surface treatment, thin film deposition, and pollutant degradation. Its core advantage lies in separating the plasma generation area from the process area, avoiding the direct bombardment of high-energy particles on sensitive substrates, thereby significantly improving the process uniformity and device reliability. However, the reliable ignition and stable maintenance of the plasma are the key prerequisites for realizing the above advantages and also the core challenges in technology research and development. Too low ignition voltage is likely to cause failure to ignite, and too high ignition voltage is likely to cause damage to the plasma chamber.

[0003] The operation of a remote plasma source generally first introduces argon or helium and applies a high voltage for pre-ignition, and then introduces a cleaning gas such as NF3 for ionization to generate atoms for chamber cleaning. The pressure in the existing CVD / PVD machine chambers is relatively high. In a vacuum, air molecules, water, fluoride gases, and precursors diffused in the machine chamber can be adsorbed on the surface of the remote plasma source, so there will be a situation of poor ignition of the remote plasma source.

[0004] The ignition voltage in the ignition process of a traditional remote plasma source is usually based on a preset threshold. Under different environmental conditions, such as different gas pressures, gas flows, and chamber temperatures, etc., it may change. The load impedance will change accordingly, and the voltage required to maintain the ionization process will also mutate, resulting in a low ignition success rate and even damage to the equipment. The prior art cannot adaptively adjust the ignition voltage. In case of poor ignition, only the reaction chamber of the remote plasma source can be replaced, resulting in increased costs. Summary of the Invention

[0005] In order to solve the problem that the ignition voltage of the traditional ignition method cannot be adaptively adjusted, the present application provides a method for controlling the ignition of a remote plasma source. Through a self-learning ignition threshold calibration method, the system can predict the load impedance value under different gas flows, gas types, and chamber temperature changes, automatically adjust the ignition voltage threshold according to the impedance value, accurately adjust the ignition voltage, reduce the risk of equipment damage, greatly improve the ignition success rate and system stability, and significantly improve the adaptability and reliability of the equipment.

[0006] The embodiment of the present application provides a method for controlling the ignition of a remote plasma source, including the following steps: S1: Set the gas current excitation threshold imin , introduce the excitation gas, and calculate the default ignition voltage V based on a preset formula break ; S2: Determine whether an ignition signal is received. If received, the relay is energized and enters the constant current stage, and the PWM drive unit is modulated by the current error; S3: Determine whether the system ignition is normal. If it is determined to be abnormal, the ignition voltage is adjusted by the self-learning ignition threshold calibration method, and then re-determined. If it is determined to be normal, go to step S4; S4: Increase the output power, and during this period, adjust the ignition voltage by the self-learning ignition threshold calibration method; S5: When the output power reaches the full load power, enter the constant power stage, and control the output power by modulating the PWM drive unit with the power error; Wherein, the self-learning ignition threshold calibration method collects the gas flow Q, gas pressure P, and cavity temperature T in real time, and uses the LSF-BP model combining the least square polynomial fitting and the BP neural network to predict the load impedance value F, and updates the ignition voltage threshold accordingly.

[0007] The beneficial effects of the above embodiments are as follows: The self-learning ignition threshold self-calibration technology dynamically adjusts the ignition threshold according to the operating conditions of the device by monitoring the ignition state in real time, combining with the LSF-BP load impedance prediction model to iteratively obtain the optimal weight λ and output the predicted value F of the load impedance. This method can effectively improve the ignition stability, reduce equipment damage, and reduce the maintenance cost of the equipment.

[0008] Based on the above embodiments, the present application can be further improved as follows: In one embodiment of the present application, in step S1, before the ignition operation, the resonant output current reference value is instantaneously reduced to the current reference value . By actively reducing the resonant output current reference value 50 ms before ignition, the impact of the surge current on the relay S is effectively suppressed, avoiding the ablation of the relay contacts caused by excessive instantaneous current, significantly extending the service life of the relay, and ensuring the stability of the ignition circuit at the same time.

[0009] In one embodiment of the present application, in step S2, after the relay is energized, the resonant output current needs to be collected and judged whether it is within the preset range and maintained for a set time. If so, go to step S3. If not, it is considered that the ignition fails, the system reports an error and disconnects the relay. By monitoring the current range and duration in real time, ignition anomalies (such as unstable arcs or gas ionization failures) can be quickly identified, the relay can be disconnected and reported in time, preventing damage to internal components of the cavity (such as electrodes or insulating materials) caused by continuous abnormal current, and reducing the equipment maintenance cost.

[0010] In one embodiment of the present application, the constant current stage control is as follows: the resonant output current reference value is compared with the current value sampled by the power circuit i pri to output a current error . After current PI modulation, a duty cycle signal is generated. By adopting current closed-loop feedback control, the duty cycle is dynamically adjusted to maintain a constant current output, ensuring the accuracy and stability of the current during the ignition stage, avoiding current fluctuations caused by sudden changes in load impedance, and improving the success rate of plasma breakdown.

[0011] In one embodiment of the present application, the constant power stage control is as follows: the output power reference value P ref is compared with the actual power P o to output a power error e 0. After power PI modulation, a resonant output current reference value i ref is obtained. The current reference value i ref is compared with the current value sampled by the power circuit i pri to output a current error e 1. The duty cycle d is output through a current PI modulator; the current error e 1 is passed through a PI modulator and inverted to obtain the switching frequency f s . After the plasma is stabilized, power closed-loop control is introduced to adaptively adjust the output power and switching frequency, effectively coping with the real-time changes in load impedance (such as sudden changes in gas flow or pressure), ensuring the stability of continuous plasma discharge, and avoiding the problem of arc extinction caused by insufficient power.

[0012] In one embodiment of the present application, the LSF-BP model includes: An input layer that receives the normalized data of Q, P, and T; A hidden layer that performs nonlinear transformation using the ReLU activation function; An output layer that outputs the load impedance value F by linearly combining the least squares fitting result and the neural network prediction value. By combining the linear fitting ability of the least squares method and the nonlinear feature learning ability of the BP neural network, the model can simultaneously capture the complex relationship between the load impedance and gas parameters (Q / P / T), significantly improving the impedance prediction accuracy, and further optimizing the real-time performance and accuracy of ignition voltage adjustment.

[0013] In one embodiment of the present application, the LSF-BP model adjusts the weights of the least squares fitting result and the neural network prediction value through the backpropagation algorithm until the total error E is not greater than the set threshold, so that the fitting result approaches t the actual value at a certain moment, thereby realizing the adaptive adjustment of the control parameters. Based on the online learning mechanism, the model weights are dynamically optimized, enabling the prediction results to quickly adapt to the long-term drift of environmental parameters (such as cavity aging or gas composition changes), ensuring that the system maintains high-precision prediction capabilities throughout its entire life cycle and reducing the frequency of manual calibration.

[0014] In one embodiment of the present application, the ignition control method is implemented based on a remote plasma source power circuit, and the remote plasma source power circuit includes: An adjustable turns ratio coupling transformer, whose secondary coil is respectively connected to the ignition circuit and the maintenance circuit; A sampling unit that collects the output voltage, current of the resonant converter, and the current of the ignition circuit; A control unit that dynamically adjusts the turns ratio of the adjustable turns ratio coupling transformer according to the predicted value of the load impedance. By adjusting the turns ratio of the secondary coil of the transformer, the voltage / current ratio of the ignition circuit and the maintenance circuit is directly changed, realizing impedance matching at the hardware level, solving the problem that traditional fixed-parameter transformers cannot adaptively adjust during load mutation, and reducing energy loss.

[0015] In one embodiment of the present application, the maintenance circuit maintains the stability of the bus voltage through constant power control after the relay is disconnected, ensuring the continuous operation during the plasma ionization stage. After the plasma enters the steady state, the maintenance circuit stabilizes the bus voltage through closed-loop control, avoiding voltage dips caused by power supply interruptions or load mutations, ensuring the energy supply required for continuous plasma ionization, and improving the uniformity and consistency of the process (such as thin film deposition). BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 It is a step flow chart of an ignition control method for a remote plasma source in an embodiment of the present application; Figure 2 It is a control block diagram of an ignition control method for a remote plasma source in an embodiment of the present application; Figure 3 It is a control block diagram of a self-learning ignition threshold calibration method in an embodiment of the present application; Figure 4 Flow chart of least - squares polynomial fitting of load impedance in the embodiment of the present application; Figure 5 Flow chart of LSF - BP load impedance prediction model algorithm in the embodiment of the present application; Figure 6 Structural diagram of each layer of the LSF - BP load impedance prediction model in the embodiment of the present application. Detailed implementation manners

[0018] The following further clarifies the present invention in conjunction with the detailed implementation manners. It should be understood that these implementation manners are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent modifications made by those skilled in the art to the present invention all fall within the scope defined by the appended claims of this application.

[0019] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0020] In the description of the present invention, it should be noted that unless otherwise clearly defined and limited, the terms "set", "installed", "connected", "coupled" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0021] In the description of the present invention, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine different embodiments or examples described in the present invention and the features of different embodiments or examples.

[0022] The embodiment of the present application provides a method for controlling the ignition of a remote plasma source, solves the problem that the ignition voltage of the traditional ignition method cannot be adaptively adjusted, realizes automatically adjusting the ignition voltage threshold according to the impedance value, precisely adjusts the ignition voltage, reduces the risk of equipment damage, and greatly improves the ignition success rate and system stability.

[0023] The technical solution in the embodiment of the present application for solving the above problems has the following general idea: Embodiment: As Figure 1 shown, a method for controlling the ignition of a remote plasma source includes the following steps: Step S1: After the remote plasma source is turned on and running, first, the control system sets the gas current excitation threshold i min , and introduces the excitation gas, and calculates the default ignition voltage according to the following formula (14) V break . Through the phase adjustment unit, the phase adjustment unit is used to instantaneously reduce the resonant output current reference value to the current reference value 50 ms before the ignition operation, which can effectively avoid the surge current and play a role in protecting the relay S.

[0024] Step S2: Judge whether an ignition signal is received. If an ignition command is received, drive and output to close the relay S. Otherwise, the system continues to introduce the excitation gas; after the relay S is closed, it enters the constant current stage, that is, compare the resonant output current reference value with the current value sampled by the power circuit i pri to output the current error . After current PI modulation, start the PWM drive unit. Then judge whether it satisfies i min ≤ i pri ≤ i max and maintain it for more than 500 ms. If not satisfied, it is considered that the ignition fails, the system reports an error and disconnects the relay S. If satisfied, jump to Step S3.

[0025] Step S3: Judge whether the system ignition is normal; if it is determined that the ignition is abnormal, adjust the ignition voltage through the self-learning ignition threshold calibration method, and use the LSF-BP load impedance prediction model to iterate out the optimal weight , and output the predicted value of the load impedance , and then update the self-calibrated ignition voltage according to the following formula (17) V break , perform ignition, and then judge whether the ignition is normal. If it is abnormal, perform cyclic ignition processing, and the upper limit of the cycle count is N times to improve the ignition success rate; if the ignition is still not successful after exceeding the cycle upper limit, it is determined that the ignition fails, the system reports an error and disconnects the relay S; during Step S3, when it is judged that the ignition is normal, it is considered that the ignition is successful, disconnect the relay and jump to Step S4.

[0026] Judge whether the ignition is normal, mainly by judging the resonant current value i priWhether the actual value is greater than or equal to the set ignition current threshold, such as 30 A. The set ignition current threshold is generally located i min and i max between.

[0027] Step S4: Increase the output power and adjust the ignition voltage through a self - learning ignition threshold calibration method; in the actual working condition, after the plasma ignition is successful, it is necessary to increase the gas flow or change the gas pressure, increase the load impedance to increase the output power of the power supply; this leads to a sudden change in the plasma impedance when the load suddenly increases, and the required sustaining voltage V o also increases correspondingly to maintain the discharge, that is, the ignition voltage V break also needs to be increased, otherwise it will cause the plasma ignition to go out; this embodiment uses a self - learning ignition threshold calibration method to predict the load impedance in real time and adjusts the plasma ignition voltage according to the load impedance, ensuring the reliability of the remote plasma source operation.

[0028] Step S5: The current system is still in the constant - current stage. Continuously increase the output power of the power supply. When reaching the full - load power, the remote plasma source enters the constant - power stage control; the constant - power stage control includes: comparing the output power error P ref obtained by multiplying the output power reference value V loop and the current i pri with the actual power P o to output the power error e 0. After power PI modulation, obtain the resonant output current reference value i ref , compare the current reference value i ref with the current value sampled by the power circuit i pri to output the current error e 1. Output the duty cycle d through the current PI modulator; take the current error e 1 through the PI modulator and invert it to obtain the switching frequency f s , and finally output the PWM drive signal D 1~ D 4 of the inverter switching tube through the PWM drive unit; thus realizing the reliable operation of the remote plasma source.

[0029] The above self-learning ignition control method for the remote plasma source is implemented based on the remote plasma source power circuit and the corresponding control unit.

[0030] Figure 2 The control block diagram of the above self-learning ignition voltage control method for the remote plasma source is shown, including a remote plasma source power circuit, a constant current stage control unit, and a constant power stage control unit; The remote plasma source power circuit includes a control unit (not shown in the figure), a PWM drive unit (not shown in the figure), a DC / AC resonant converter, an ignition circuit, a maintenance circuit, a sampling unit, and a reaction cavity.

[0031] The control unit (including the constant current stage control unit and the constant power stage control unit) is used to control the PWM drive unit to output a pulse signal; the PWM drive unit controls the output current of the resonant converter; the output current of the resonant converter affects the input current of the ignition circuit; the resonant converter converts the pulse signal output by the PWM drive unit into an AC source and is used to provide an input AC source for the ignition circuit; the ignition circuit generates a high-voltage ignition signal to cause the plasma load to oscillate and ionize at a high frequency, and the resonant converter is used to provide the input AC source. The resonant converter includes an adjustable turns ratio coupling transformer T1, and the ignition circuit is connected to the secondary coil N3 of the adjustable turns ratio coupling transformer T1; the maintenance circuit is connected in series with the secondary coil N2 of the adjustable turns ratio coupling transformer T, and the vacuum reaction cavity is used as the secondary side of the high-frequency transformer to only provide i o to meet the low-voltage energy input in the ionization maintenance stage; among them, the turns ratios of the secondary coils N2 and N3 can both be modulated; The maintenance circuit is used to maintain the stability of the bus voltage after entering the constant power stage (ionization stage) from the constant current stage (ignition stage) when the relay is disconnected; The sampling unit is used to collect the ignition current of the ignition circuit, and the output voltage and output current of the resonant converter; the relay S is controlled by the control unit to be attracted or disconnected.

[0032] Figure 3 The control block diagram of the self-learning ignition threshold calibration method in the above self-learning ignition control method for the remote plasma source is shown, including: the aforementioned remote plasma source power circuit, a multi-sensor module, an LSF-BP load impedance prediction model, and an ignition voltage adaptive update module; The multi-sensor module includes a gas flow meter, a pressure sensor, and a temperature sensor to monitor the gas flow rate of the reaction cavity Q , gas pressure P and the temperature of the reaction chamber T ; The LSF-BP load impedance prediction model is used to iterate the optimal weight and output the predicted value of the load impedance ; The ignition voltage adaptive update module calculates the ignition voltage required by the system at the current moment through the following formula (17). The adjustable turn ratio coupling transformer T1 adaptively adjusts the ignition voltage required by the system at the current moment and the sustaining voltage according to the LSF-BP load impedance prediction model and the ignition voltage adaptive update module.

[0033] The LSF-BP load impedance prediction model is further described as follows: First, the least squares method fitting prediction output load impedance is described. Figure 4 Shows the flowchart of the least squares polynomial fitting load impedance; Combining Figure 4 , the least squares polynomial fitting LSF algorithm is as follows: In the remote plasma system, the gas flow rate Q , the gas pressure P and the reaction chamber temperature T 's influence on the load impedance can be described by the following ideal gas model: When the pumping speed S is fixed, the pressure P is proportional to the flow rate P = Q / S , the load impedance Z is proportional to the flow rate and inversely proportional to the temperature T . The influence of the gas flow rate Q on the load impedance Z can be expressed by Equation (1): (1); Where k Q is the proportionality constant, related to system parameters such as electrode area, collision cross-section, etc.

[0034] At a constant temperature, the load impedance is proportional to the pressure: (2); Where k P is the proportionality constant, related to plasma density and collision frequency.

[0035] At a constant pressure, due to the decrease in gas density, the load impedance is inversely proportional to the reaction chamber temperature: (3); Where k T is the proportionality constant.

[0036] P1: Define the linear model; Combining Equation (1), Equation (2) and Equation (3), the load impedance Z and Q ,P , T approximate linear relationship: (4); where k 1, k 2, k 3 are variable coefficients, k 0 is the constant term; assuming there are n groups of observation data ( Q i , P i , t i , Z i ), where ; Use the least squares method to fit the variable coefficients and the constant term.

[0037] P2: Construct the data matrix and the observation vector; Construct the design matrix X and the observation vector Z: (5); P3: Establish the normal equations; The goal of the least squares method is to minimize the sum of squared residuals: (6); Take the partial derivative with respect to the variable vector to be solved and set the derivative to zero to obtain the normal equations: (7); P4: Solve for the fitting coefficients; The optimal solution of the parameters is: (8); P5: Output the predicted value of the current load impedance y 1; Substitute the optimal solution of the solved parameters into Equation (4) to obtain the predicted value of the current load impedance, denoted as y 1.

[0038] The BP neural network is a multi-layer feedforward network, consisting of three layers: the input layer, the hidden layer, and the output layer. During implementation, data is transmitted from the input layer to the output layer. The hidden layer mainly non-linearly transforms the data transmitted by the input layer into a feature matrix. Then, the output layer calculates the error with the expected data and propagates the error forward from the output layer. The connection weights between each layer of the network and the neuron biases of each layer are adjusted through the backpropagated error to optimize the entire neural network and achieve the purpose of learning data features. Its working process is divided into two parts: the forward transmission of the system and the backpropagation of the error. The process of forward transmission is mainly divided into receiving information, processing information, and outputting information. When the difference between the output and the expected value is too large, the system will start to backpropagate the error, and each weight value will be changed and adjusted according to the actual situation of the error until the output meets the requirements or reaches the set number of iterations, and then the training of the system will end.

[0039] Construct an adaptive algorithm: Use the least squares polynomial fitting (LSF) and the BP neural network to form a prediction model to adjust the ignition threshold. The final predicted threshold of the ignition voltage is combined by the prediction results of two parts with a certain weight; utilize the self-learning ability of the BP neural network to online adjust the weight value of the value function and preprocess to obtain the nearest weight control parameter. In view of the overfitting or underfitting error existing in the least squares method, optimize it using the LSF-BP load impedance prediction model algorithm.

[0040] Figure 5 The flowchart of the LSF-BP load impedance prediction model algorithm of the self-learning ignition threshold calibration method provided by this embodiment is shown; Figure 6 The network structure diagram of each layer of the LSF-BP load impedance prediction model provided by this embodiment is shown; combined Figure 5 、 Figure 6 , a self-learning ignition threshold calibration method, includes the following steps: Step N1: First, determine the structure of the BP network, adopting a structure including an input layer, a hidden layer, and an output layer; where i represents the input layer nodes, j represents the hidden layer nodes, l represents the output layer nodes, and determine the number of input nodes m and the number of hidden layer nodes q , with 3 output layer nodes; and initialize the initial values of the weighted coefficients from the input layer to the hidden layer and from the hidden layer to the output layer 、 and as well as the bias amount; The calculation formula for each layer is: (9); Among them, T ( x) represents the activation function, represents the weight coefficient, b represents the activation threshold; Step N2: Construct a sample data set according to the test data: When the remote plasma source is working, use multiple sensors to measure the gas flow Q , gas pressure P and chamber temperature T to construct a sample data set {( x , y , z )}, where x represents the one-dimensional set vector Q of the gas flow , y represents the one-dimensional set vector P of the gas pressure , z represents the one-dimensional set vector T of the chamber temperature ; Step N3: Combine x , y , and z in the sample data set {( x , y and z to form a two-dimensional input matrix, and then perform data normalization operation and update on the two-dimensional input matrix; at the same time, establish an output sample data set f {( x , y , z )} where f is used as the label vector corresponding to the network predicted value of the load impedance; Standardize all eigenvalue of the sample data set {( x , y , z )}: The normalization process is to uniformly transform each resonant current eigenvalue sequence into the interval, and the calculation formula is formula (10); (10); where, in an input eigenvalue sequence, M i is the input eigenvalue after normalization, N i is the input eigenvalue before normalization, N max is the maximum input eigenvalue, N min is the minimum input eigenvalue.

[0041] Step N4: Use the two-dimensional input matrix updated by the normalization operation to train the BP model and output the network predicted value f of the load impedance , while initializing the weight set vector on the two-dimensional output nodes of the BP neural network model , the biases of each layer of neurons in the BP neural network model, and then calculate the error for the label vector f and the network predicted output value ; For a multi-layer network, the calculation is performed in a forward propagation manner, that is, each layer is calculated according to Equation (9) until the last output layer; During the forward propagation process, after the input data is calculated by the perceptron node, it passes through the activation function T ( x ) to obtain the output result; the activation function T ( x ) of the hidden layer is selected as the ReLU function, and the activation function l of the BP neural network is: (11); Then the output by forward propagation is: (12); Among them, W ij represents the weight coefficient connecting the i-th neuron in the previous layer and the j-th neuron in the current layer, and b represents the bias of the j-th neuron in the current layer.

[0042] Step N5: Construct an LSF-BP load impedance prediction model based on weight update combination, and its expression is as follows: (13); In Equation (13), , λ 1 and λ 2 are the predicted load impedance value of the combined model, the LSF weight, and the BP neural network weight respectively, y 2 is the forward propagation output value ; Step N6: Calculate the overall error E for the entire sample set, and judge whether the error E meets the design requirements, that is, the error , N is the error threshold, where E is calculated from Equation (13); if it is satisfied, the neural network training is ended; if it is not satisfied, the neural network learning is performed using backpropagation, the output result is compared with the expected structure, and the weights λ 1 and λ 2 are adjusted and updated online through multiple iterations, so that the fitting result approaches the t at the f ( k ) value, thereby realizing the adaptive adjustment of the control parameters; (13); Step N7: Iterate the optimal weights using the LSF - BP load impedance prediction model and output the predicted load impedance value .

[0043] The ignition voltage adaptive update module is further described as follows: During the remote plasma ignition process, the load impedance Z and the ignition voltage V ignition relationship can be divided into two stages: before breakdown and after breakdown. Among them, the stage before breakdown is the ignition critical stage, and the stage after breakdown is the plasma steady state stage. The specific formulas are as follows: 1. Gas is not ionized, before breakdown stage; When the gas is not ionized, the system is approximately open - circuit and shows high impedance. Sufficient breakdown voltage needs to be applied to trigger ionization. The breakdown voltage is described by Paschen's law: (14); p is the gas pressure; d is between the electrodes; A , B is a constant related to the gas type (for example, argon: A ≈13.6cm − 1 Torr −1 , B ≈176V·cm −1 Torr −1 ); γ is the secondary electron emission coefficient.

[0044] At this time, the load impedance is extremely large and approximately in an open - circuit state: (15); The voltage needs to reach V break to trigger the discharge. There is no direct proportional relationship between impedance and voltage, and only the breakdown condition needs to be met.

[0045] 2. Plasma is formed, after breakdown stage; Once the gas is ionized to form plasma, the impedance drops suddenly and enters a steady state. At this time, the relationship between the load impedance and the voltage follows Ohm's law: (16); Among them, there is a relationship of , where K is an empirical coefficient, , V o is the sustaining voltage; I oFor maintaining current; Z o is the plasma maintenance resistance; when the load suddenly increases, such as changes in gas flow, pressure or temperature, the plasma impedance increases, and the maintenance voltage needs to be increased V o to maintain the discharge, that is, the ignition voltage V break also needs to be increased; The relationship between plasma impedance and voltage is affected by electron density n e , conductivity σ e , electron mobility μ e and so on, and can be further expressed as: (17); where the load impedance Z 0 is the predicted load impedance value output by the aforementioned LSF-BP load impedance prediction model .

[0046] For low-pressure plasma, the conductivity σ e can be approximated as: (18); where ν en is the electron-neutral particle collision frequency, e is the electron charge, m e is the electron mass.

[0047] The above self-learning ignition threshold calibration method based on load impedance prediction enables the system to predict the load impedance value under different gas flows, gas types and chamber temperature changes through the self-learning ignition threshold self-calibration technology, automatically adjust the ignition voltage threshold according to the impedance value, precisely regulate the ignition voltage, reduce the risk of equipment damage, greatly improve the ignition success rate and system stability, and significantly improve the adaptability and reliability of the equipment.

[0048] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for igniting and controlling a remote plasma source, characterized in that, It includes the following steps: S1: Set the gas current excitation threshold, introduce the excitation gas, and calculate the default ignition voltage based on a preset formula; S2: Determine whether an ignition signal is received. If received, the relay is energized and enters the constant current stage, and the PWM drive unit is modulated by the current error; S3: Determine whether the system ignition is normal. If it is determined to be abnormal, the ignition voltage is adjusted by the self-learning ignition threshold calibration method and re-determined. If it is determined to be normal, proceed to step S4; S4: Increase the output power, and during this period, adjust the ignition voltage by the self-learning ignition threshold calibration method; S5: When the output power reaches the full load power, enter the constant power stage, and control the output power by modulating the PWM drive unit with the power error; Among them, the self-learning ignition threshold calibration method collects the gas flow rate Q, gas pressure P, and chamber temperature T in real time, and uses the LSF-BP model combined with the least square polynomial fitting and BP neural network to predict the load impedance value F, and updates the ignition voltage threshold accordingly.

2. The ignition control method according to claim 1, wherein: In the step S1, before the ignition operation, the resonant output current reference value is instantaneously reduced to the current reference value .

3. The ignition control method according to claim 1, characterized in that: In step S2, after the relay is energized, the resonant output current needs to be collected and judged whether it is within the preset range and maintained for a set time. If so, proceed to step S3. If not, it is considered that the ignition fails, the system reports an error and disconnects the relay.

4. The ignition control method according to claim 1, wherein: The constant current stage control is as follows: taking the resonant output current reference value and comparing it with the current value sampled by the power circuit i pri to output a current error , and generating a duty cycle signal after current PI modulation.

5. The ignition control method according to claim 1, characterized in that: The constant power stage control is as follows: the output power reference value P ref is compared with the actual power P o to output the power error e 0. After power PI modulation, the resonant output current reference value i ref is obtained. The current reference value i ref is compared with the current value sampled by the power circuit i pri to output the current error e 1. The duty cycle is output through the current PI modulator d ; The current error e 1 is passed through a PI modulator and inverted to obtain the switching frequency f s .

6. The ignition control method according to claim 1, characterized in that: The LSF-BP model includes: The input layer, which receives the normalized data of Q, P, and T; The hidden layer, which performs nonlinear transformation using the ReLU activation function; The output layer, which outputs the load impedance prediction value F by linearly combining the least square fitting result and the neural network prediction value.

7. The ignition control method according to claim 6, characterized in that: The LSF-BP model adjusts the weights of the least square fitting result and the neural network prediction value through the backpropagation algorithm until the total error E is not greater than the set threshold.

8. The ignition control method according to claim 1, wherein: The ignition control method is implemented based on a remote plasma source power circuit, and the remote plasma source power circuit includes: An adjustable turns ratio coupling transformer, whose secondary coil is respectively connected to the ignition circuit and the maintenance circuit; A sampling unit, which samples the output voltage, current of the resonant converter, and the current of the ignition circuit; A control unit, which dynamically adjusts the turns ratio of the adjustable turns ratio coupling transformer according to the load impedance value F.

9. The ignition control method according to claim 8, wherein: The maintenance circuit maintains the stability of the bus voltage through constant power control after the relay is disconnected.

Images