Photoresist filling state monitoring system and method

Abstract

The invention relates to the technical field of industrial process state monitoring, in particular to a photoresist filling state monitoring system and method. The system is characterized in that a micro-pressure pulse control unit acquires a reflection pressure change signal, establishes a corresponding relation between liquid column length and pressure, and outputs a pressure fracture judgment result; the acoustic sensing and judging unit is used for collecting a liquid column fracture sound pressure signal, constructing a non-contact liquid column end point state recognition mechanism and outputting an acoustic fracture judging result; the multi-channel intelligent decision-making unit uses a double-state doubt-tolerant self-certification mechanism to judge whether a pressure fracture judgment result is true or false, and outputs a final liquid column fracture judgment signal in combination with an acoustic fracture judgment result; and the filling process data recording unit records end point judgment related parameters, fracture judgment results and end point control feedback information in each filling process. According to the invention, micropressure pulse application and reflection pressure signal analysis are combined, and acoustic determination is introduced, so that the terminal point state of the photoresist filling tail end liquid column is accurately monitored.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process status monitoring, and in particular to a photoresist filling status monitoring system and method. Background Art

[0002] Photoresist is a key photosensitive material widely used in precision processes such as semiconductor manufacturing and micro-nanofabrication. Its filling process has a direct impact on the quality of the finished product and production stability. Especially at the final stage of transporting the photoresist from the storage container through a pipeline to the dispensing container, the liquid column state changes sensitively and is affected by complex factors. The accuracy of endpoint determination not only affects the consistency of the single filling volume, but also directly affects the uniformity and stability of the subsequent coating process. Therefore, how to accurately identify the liquid column fracture state at the end of the filling process is a key technical link in photoresist filling quality control.

[0003] When the photoresist filling is nearing the end stage, there are still the following two problems that affect the precise filling of the photoresist: First, the photoresist liquid itself has a high viscosity and a low flow rate. When using pressure signal feedback analysis, the feedback of the pressure signal is easily interfered by non-fracture factors such as bubble retention, end liquid reflux, and micro-vibration of the inner wall of the pipe. These factors may form a fluctuation pattern similar to the real fracture in the pressure reflection signal, thereby causing misjudgment. At the same time, the pressure signal of the real fracture may not be reflected immediately due to the transmission delay, resulting in a missed judgment that the liquid column has broken but is still judged as not broken; second, in the endpoint judgment, there are differences in the response characteristics of the pressure signal and the acoustic signal to the fracture. When the two produce conflicting judgment results at the same time, if it is not possible to immediately confirm which result is more reliable, the wrong endpoint signal may be directly output, causing problems such as over-filling, under-filling or droplet residue, which will have an adverse effect on the quality of the photoresist and production stability. Summary of the Invention

[0004] The purpose of the present invention is to provide a photoresist filling status monitoring system and method to solve the two problems affecting the accurate filling of photoresist raised in the above background technology.

[0005] To achieve the above object, the present invention provides a photoresist filling status monitoring system, comprising:

[0006] The micro-pressure pulse control unit is used to apply micro-pressure pulses to the filling pipe, collect reflected pressure change signals, establish a corresponding relationship between liquid column length and pressure, and issue a fracture trigger instruction when the liquid column length approaches the critical length threshold interval, and output the pressure fracture judgment result of the liquid column fracture state;

[0007] The acoustic perception and discrimination unit is used to collect the sound pressure signal of the liquid column fracture, calculate the sound spectrum energy change, amplitude fluctuation rate and duration index to build a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state;

[0008] Multi-channel intelligent decision-making unit, which uses a dual-state doubt-tolerant self-verification mechanism to dynamically judge the authenticity of the pressure fracture judgment result and combines it with the acoustic fracture judgment result to output the final liquid column fracture judgment signal;

[0009] The filling process data recording unit is used to record the endpoint determination related parameters, fracture determination results and endpoint control feedback information during each filling process.

[0010] Preferably, the micro-pressure pulse is a periodic disturbance signal with an amplitude lower than the static pressure value of the liquid column and a pulse width less than 1 second, which is used to achieve liquid column state disturbance; the micro-pressure pulse is applied to the filling pipe according to the micro-pressure pulse frequency setting rules and pressure cycle adjustment strategy.

[0011] Preferably, the reflected pressure change signal includes an initial echo response value, a multi-order attenuation echo amplitude sequence and echo delay change data, which are used to determine whether the liquid column is close to fracture;

[0012] In the micro-pressure pulse control unit, a corresponding relationship between the liquid column length and the pressure is established. The specific method and steps are as follows:

[0013] The reflected pressure change signal is segmented according to the pulse number, and each segment of the reflected pressure change signal is mapped to an interval segment of the liquid column length range to form a pressure-length pre-calibration data set. The least squares method is used to establish a multivariate nonlinear mapping function between the liquid column length and the characteristic vector of the reflected pressure response signal, which is the corresponding relationship between the liquid column length and pressure.

[0014] Among them, the multivariable nonlinear mapping function is used to estimate the liquid column length in real time based on the reflected pressure change signal.

[0015] Preferably, when the liquid column length is close to being in the critical length threshold interval, the micro-pressure pulse control unit compares the real-time estimated liquid column length of the multivariable nonlinear mapping function with the critical length threshold interval, and issues a fracture trigger instruction when one of the fracture trigger conditions is met;

[0016] The fracture triggering conditions are: the length of the liquid column enters the critical length threshold range for the first time and shows no growth trend within two consecutive pulse cycles; the length of the liquid column fluctuates within the critical length threshold range and the initial echo amplitude of the reflected signal shows a continuous decrease exceeding a preset ratio;

[0017] The fracture trigger instruction is used to start the fracture state identification process and output the pressure fracture judgment result of the liquid column fracture state, and the pressure fracture judgment result includes the liquid column fracture state flag and the pulse number of the fracture occurrence time.

[0018] Preferably, the acoustic perception and discrimination unit processes the collected liquid column fracture sound pressure signal in segments, and calculates the sound spectrum energy change, amplitude fluctuation rate and duration index of the liquid column fracture sound pressure signal respectively. The specific calculation method is as follows:

[0019] A time-frequency analysis is performed on the liquid column fracture sound pressure signal within the target frequency band to extract the total frequency domain amplitude change trend, which is then differentiated with the static background sound spectrum to calculate the sound spectrum energy change. The envelope curve of the liquid column fracture sound pressure signal is extracted, and the number of amplitude changes per unit time is counted to obtain the amplitude fluctuation rate. Based on the start and end time intervals of continuous high-amplitude segments in the liquid column fracture sound pressure signal, the duration interval of the liquid column fracture event is determined to obtain the duration index.

[0020] Preferably, the non-contact liquid column endpoint state recognition mechanism is constructed based on the integration of acoustic spectrum energy change, amplitude fluctuation rate and duration indicators, and is used to identify whether the liquid column has reached a broken state, and to determine whether there is a retraction residue and an unbroken dragline phenomenon;

[0021] The specific construction method of the non-contact liquid column endpoint state recognition mechanism is as follows:

[0022] The acoustic spectrum energy changes, amplitude fluctuation rate and duration indicators of the acoustic pressure signals of the photoresist filling liquid column fracture are calculated multiple times, and the acoustic spectrum energy changes, amplitude fluctuation rate and duration indicators calculated each time are constructed into three-dimensional features; the fracture state of the photoresist filling liquid column each time is marked; a supervised multi-classification discrimination method is used to train the three-dimensional features and construct a state recognition and classification model; during the real-time filling process, the three-dimensional features of the real-time liquid column fracture acoustic pressure signal are input into the state recognition and classification model, and the real-time photoresist filling liquid column fracture state is output to obtain the acoustic fracture judgment result of the liquid column fracture state.

[0023] Preferably, in the multi-channel intelligent decision-making unit, a dual-state doubt-tolerant self-verification mechanism is used to evaluate the credibility of the pressure fracture judgment result and actively identify potential misjudgment patterns. When the pressure fracture judgment result is identified as unreliable, the liquid column fracture state is auxiliary judged and verified in combination with the acoustic fracture judgment result;

[0024] The two-state doubt-tolerant self-proving mechanism includes a misjudgment recognition model and an acoustic compensation module.

[0025] Preferably, the false positive identification model is constructed based on a multi-dimensional feature matching strategy, and is used to construct a doubt analysis vector based on the pressure rupture judgment result and the filling history data, and output a doubt score indicating the credibility of the pressure rupture judgment result;

[0026] The doubt analysis vector is a set of multidimensional anomaly feature combinations extracted from the pressure fracture judgment results and serves as the input of the misjudgment identification model. The multidimensional anomaly feature combination includes pressure fluctuation amplitude, disturbance peak response time, waveform oscillation symmetry, echo attenuation gradient, and disturbance response delay.

[0027] Among them, the doubt score is the nonlinear matching error between the doubt analysis vector and the characteristics of historical normal fracture samples, which is used to determine whether the pressure fracture judgment result is a misjudgment.

[0028] Preferably, the acoustic compensation module is used to analyze whether the doubt score exceeds a set threshold. When the doubt score exceeds the set threshold, the acoustic fracture judgment result is enabled to replace the pressure fracture judgment result, and the final liquid column fracture judgment signal is output.

[0029] On the other hand, the present invention provides a method for monitoring the state of photoresist filling, which is used in the above-mentioned system for monitoring the state of photoresist filling, comprising the following steps:

[0030] S10.1. Apply a micro-pressure pulse to the filling pipe and simultaneously collect the reflected pressure change signal to establish a correspondence between the liquid column length and pressure. When the liquid column length approaches the critical length threshold, a fracture trigger instruction is issued and a pressure fracture judgment result is output indicating the liquid column fracture state.

[0031] S10.2. Collect the acoustic pressure signal of the liquid column fracture, calculate the acoustic spectrum energy change, amplitude fluctuation rate, and duration indicators to establish a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state;

[0032] S10.3. Use a two-state doubt-tolerant self-verification mechanism to dynamically determine the authenticity of the pressure fracture determination result and combine it with the acoustic fracture determination result to output the final liquid column fracture determination signal;

[0033] S10.4. Record endpoint determination parameters, fracture determination results, and endpoint control feedback information during each filling process.

[0034] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0035] 1. In the present invention, based on the establishment of the corresponding relationship between the application of micro-pressure pulses and the reflected pressure signal, combined with the credibility analysis of the pressure fracture judgment results, it is possible to identify false fluctuations caused by bubble retention, end liquid backflow, and micro-vibration of the inner wall of the pipe at the end of the photoresist filling, as well as missed judgments caused by the delay in the transmission of the photoresist liquid column fracture signal;

[0036] 2. In the present invention, by introducing the acoustic fracture judgment result for cross-validation when there is an abnormality in the pressure fracture judgment result, combined with the complementarity of the two types of signals in the fracture response characteristics, dynamic correction of the endpoint judgment result is achieved, reducing the probability of over-filling, under-filling or droplet residue, and improving the photoresist filling quality and production consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A functional block diagram of an embodiment of the present invention;

[0038] Figure numerals: 1. Micro-pressure pulse control unit; 2. Acoustic perception and discrimination unit; 3. Multi-channel intelligent decision-making unit; 4. Filling process data recording unit. DETAILED DESCRIPTION

[0039] Example 1, as Figure 1 As shown, a photoresist filling state monitoring system is provided, including: a micro-pressure pulse control unit 1, an acoustic perception and discrimination unit 2, a multi-channel intelligent decision unit 3 and a filling process data recording unit 4.

[0040] In the micro-pressure pulse control unit 1 of this embodiment, the micro-pressure pulse control unit 1 is used to apply micro-pressure pulses to the filling pipe, and at the same time collect the reflected pressure change signal, establish a corresponding relationship between the liquid column length and the pressure, and issue a fracture trigger instruction when the liquid column length is close to the critical length threshold interval, and output the pressure fracture judgment result of the liquid column fracture state.

[0041] In this embodiment, the micro-pressure pulse is a periodic disturbance signal with an amplitude lower than the static pressure value of the liquid column and a pulse width less than 1 second, which is used to achieve liquid column state disturbance; the micro-pressure pulse is applied to the filling pipe according to the micro-pressure pulse frequency setting rules and pressure cycle adjustment strategy.

[0042] In this embodiment, the micro-pressure pulse is selected as a periodic disturbance signal with an amplitude lower than the static pressure value of the liquid column and a pulse width less than 1 second. The specific reasons are as follows:

[0043] If the applied micro-pressure pulse is greater than the static pressure value of the liquid column, it will cause the liquid column to move forward, withdraw or bubble, interfering with the actual filling process; the liquid column is inherently unstable at the end of filling, and excessively large pulses can easily lead to premature breakage or disturbed bubbles mixed into the photoresist, affecting product quality; disturbance signals below the static pressure are "metastable disturbances", which can be absorbed by the liquid column system without causing physical changes, and at the same time can produce perceptible micro-disturbances in the reflected signal; the photoresist filling path is short and the liquid column response time is very fast, and a micro-pressure pulse with a pulse width of 10ms to 200ms can reflect the interference response; if the pulse width is too long, the signal reflection will superimpose the responses of multiple state segments, resulting in blurred echo interference, making it difficult to identify whether the pressure reversal is caused by liquid column changes or long-term pressure interference.

[0044] In this embodiment, the micro-pressure pulse frequency setting rule is: based on the average velocity of the liquid column flow and the pipeline length parameters in the historical filling data, the optimal disturbance pressure frequency range is calculated and dynamically adjusted; the pressure cycle adjustment strategy is: identify whether the pressure reflection signal has a sudden change edge caused by the interruption of the liquid column continuity during the pressure cycle, and adjust the subsequent pressure interval according to the time position of the sudden change point within the cycle.

[0045] In this embodiment, the reflected pressure change signal includes an initial echo response value, a multi-order attenuation echo amplitude sequence, and echo delay change data, which are used to determine whether the liquid column is close to fracture;

[0046] In the micro-pressure pulse control unit 1, a corresponding relationship between the liquid column length and the pressure is established. The specific method and steps are as follows:

[0047] The reflected pressure change signal is segmented according to the pulse number, and each segment of the reflected pressure change signal is mapped to an interval segment of the liquid column length range to form a pressure-length pre-calibration data set. The least squares method is used to establish a multivariate nonlinear mapping function between the liquid column length and the characteristic vector of the reflected pressure response signal, which is the corresponding relationship between the liquid column length and pressure.

[0048] Among them, the multivariable nonlinear mapping function is used to estimate the liquid column length in real time based on the reflected pressure change signal.

[0049] In this embodiment, the initial echo response value refers to the maximum reflected pressure amplitude collected within the first time window after the micro-pressure pulse is applied, which can reflect the overall continuity of the liquid column; the multi-order attenuation echo amplitude sequence refers to the amplitude changes of the second-order, third-order and subsequent reflected waves collected in time sequence after the initial echo, which is used to characterize the absorption and reflection characteristics of the terminal state of the liquid column to the pressure wave; the echo delay change data refers to the time interval from the application of the micro-pressure pulse to the appearance of the peak value of each order of reflected waves, which is used to judge the change in the physical length of the liquid column.

[0050] The pulse number is a serial number automatically generated by the system according to the pressure application sequence during the micro-pressure pulse control process. Each applied pulse corresponds to a unique number, which is used to match the reflected pressure signal with the specific pressure event in the data processing stage; the pressure-length pre-calibration data set is a sample set formed by combining the reflected pressure change signal under the condition of known liquid column length and its corresponding length information, which serves as the training data source for the fitting model.

[0051] When the least squares method is used to establish the multivariate nonlinear mapping function between the liquid column length and the characteristic vector of the reflected pressure response signal, the reflection signal characteristics are first standardized, and then the regression coefficients of each characteristic and the liquid column length are calculated to minimize the square error between the predicted value and the measured length value.

[0052] In this embodiment, when the liquid column length is close to the critical length threshold interval, the micro-pressure pulse control unit 1 compares the real-time estimated liquid column length of the multivariable nonlinear mapping function with the critical length threshold interval, and issues a fracture trigger instruction when one of the fracture trigger conditions is met;

[0053] The fracture triggering conditions are: the length of the liquid column enters the critical length threshold range for the first time and shows no growth trend within two consecutive pulse cycles; the length of the liquid column fluctuates within the critical length threshold range and the initial echo amplitude of the reflected signal shows a continuous decrease exceeding a preset ratio;

[0054] The fracture trigger instruction is used to start the fracture state identification process and output the pressure fracture judgment result of the liquid column fracture state, and the pressure fracture judgment result includes the liquid column fracture state flag and the pulse number of the fracture occurrence time.

[0055] In this embodiment, the critical length threshold interval is a liquid column length range obtained based on the statistics of multiple batches of photoresist filling experimental data, which covers the boundary area between the stable length and the unstable length before the liquid column breaks; the upper limit of the critical length threshold interval corresponds to the critical state where the liquid column length is about to break, and the lower limit corresponds to the minimum length value that can be stably determined to be broken in most experiments.

[0056] The multivariable nonlinear mapping function takes the latest collected reflected pressure change signal as input during real-time operation, outputs an estimated value of the current length of the liquid column, and compares it with the critical length threshold interval; when judging whether the liquid column length is within this interval, the system simultaneously records the comparison result and the corresponding pulse number.

[0057] When judging that the length of the liquid column enters the critical length threshold interval for the first time, the system determines the "first entry" state by detecting whether the length value of the previous pulse cycle is outside the interval and the current length value is within the interval; no growth trend appears in two consecutive pulse cycles, which means that in the comparison of the length estimation results of the current pulse cycle and the previous pulse cycle, there is no situation where the length value is greater than the previous cycle, thereby confirming that the liquid column is in a stable or contraction stage; for the case where the liquid column length fluctuates within the critical length threshold interval, the system will continuously detect the ups and downs of the length estimation value within multiple pulse cycles, and when the change amplitude does not exceed the interval boundary, it is deemed to be a fluctuation within the interval; at the same time, the initial echo amplitude of the reflected signal is monitored, and the current amplitude is subtracted from the amplitude of the previous cycle and then divided by the amplitude of the previous cycle to obtain the decrease ratio. If this ratio continuously exceeds the preset ratio threshold, it is considered that the liquid column state is continuously attenuating, which is consistent with the fracture trend.

[0058] After the fracture trigger instruction is issued, the system will immediately call the fracture state identification process, start the end state judgment operation, and output the pressure fracture judgment result; in the pressure fracture judgment result, the liquid column fracture state flag is a binary flag, which is used to distinguish between fracture and non-fracture states; the pulse number at the time of fracture occurrence corresponds to the unique number of the micro-pressure pulse that triggered the fracture judgment in the pressure sequence, which is convenient for traceability and data archiving.

[0059] In the acoustic perception and discrimination unit 2 of this embodiment, the acoustic perception and discrimination unit 2 is used to collect the sound pressure signal of the liquid column fracture, and calculate the sound spectrum energy change, amplitude fluctuation rate and duration index to construct a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state.

[0060] In this embodiment, the acoustic perception and discrimination unit 2 performs segmented processing on the collected liquid column fracture sound pressure signal, and calculates the sound spectrum energy change, amplitude fluctuation rate and duration index of the liquid column fracture sound pressure signal respectively. The specific calculation method is as follows:

[0061] A time-frequency analysis is performed on the liquid column fracture sound pressure signal within the target frequency band to extract the total frequency domain amplitude change trend, which is then differentiated with the static background sound spectrum to calculate the sound spectrum energy change. The envelope curve of the liquid column fracture sound pressure signal is extracted, and the number of amplitude changes per unit time is counted to obtain the amplitude fluctuation rate. Based on the start and end time intervals of continuous high-amplitude segments in the liquid column fracture sound pressure signal, the duration interval of the liquid column fracture event is determined to obtain the duration index.

[0062] In this embodiment, the acquisition process of the liquid column fracture sound pressure signal is completed by a high-sensitivity acoustic sensor. The sampling frequency is selected to cover the main energy frequency band of the sound wave at the moment of liquid column fracture, and the sampling signal is preprocessed by an anti-aliasing filter to remove high-frequency noise and process environment background noise.

[0063] In the calculation of the sound spectrum energy change, the target frequency band is determined based on the analysis results of typical fracture sound spectra from a large number of filling tests, and the frequency band with significant difference in frequency distribution between the fracture event signal and the background noise is preferentially selected; the time-frequency analysis uses short-time Fourier transform to decompose the continuous time signal into a series of spectra of fixed time windows, calculate the sum of the squares of the frequency domain amplitudes of the target frequency band in each time window, and obtain an energy value sequence, and then subtract the energy mean value of the static background sound spectrum from this sequence to obtain the energy change value.

[0064] In the calculation of amplitude fluctuation rate, the envelope curve of the sound pressure signal is first extracted through envelope detection. The envelope curve reflects the amplitude change trend of the signal in the time domain. Then, within the specified analysis time window, the number of adjacent peak-to-valley changes on the envelope curve is counted and divided by the time window length to obtain the number of amplitude changes per unit time.

[0065] In the calculation of the duration indicator, the amplitude threshold is set as the background noise mean plus a multiple of the noise standard deviation. When the envelope curve is continuously higher than the amplitude threshold, its start and end time points are recorded, and the difference between the two is the duration of a liquid column rupture event. In order to avoid misjudging multiple independent high-amplitude disturbances as a continuous event, time interval judgment is introduced in the duration calculation process. If the interval between two high-amplitude signals exceeds the preset time threshold, they are recorded as two independent duration events.

[0066] In this embodiment, the non-contact liquid column endpoint state recognition mechanism is constructed based on the integration of acoustic spectrum energy changes, amplitude fluctuation rate, and duration indicators to identify whether the liquid column has reached a fracture state and whether there is a retraction residue or an unbroken dragline phenomenon.

[0067] The specific construction method of the non-contact liquid column endpoint state recognition mechanism is as follows:

[0068] The acoustic spectrum energy changes, amplitude fluctuation rate and duration indicators of the acoustic pressure signals of the photoresist filling liquid column fracture are calculated multiple times, and the acoustic spectrum energy changes, amplitude fluctuation rate and duration indicators calculated each time are constructed into three-dimensional features; the fracture state of the photoresist filling liquid column each time is marked; a supervised multi-classification discrimination method is used to train the three-dimensional features and construct a state recognition and classification model; during the real-time filling process, the three-dimensional features of the real-time liquid column fracture acoustic pressure signal are input into the state recognition and classification model, and the real-time photoresist filling liquid column fracture state is output to obtain the acoustic fracture judgment result of the liquid column fracture state.

[0069] In this embodiment, the marking process of the fracture state of the photoresist filling liquid column is completed manually by combining high-frame rate video recording and signal characteristic curves. The fracture state of the photoresist filling liquid column includes three states: "broken", "withdrawal residue" and "drag wire not broken"; the non-contact liquid column endpoint state recognition mechanism uses the liquid column fracture sound pressure signal obtained by acoustic sensing to analyze, without physical contact with the liquid column or the filling pipe, avoiding mechanical contact from interfering with the stability and cleanliness of the filling fluid.

[0070] In this embodiment, the training stage of the state recognition classification model adopts a supervised multi-classification discrimination method, uses a cross-validation strategy to evaluate the generalization ability of the model, optimizes the classification boundary parameters, and introduces a feature selection algorithm to eliminate redundant feature components with low correlation with state discrimination, thereby reducing the impact of acoustic environment noise on model accuracy.

[0071] In the multi-channel intelligent decision-making unit 3 of this embodiment, the multi-channel intelligent decision-making unit 3 uses a dual-state doubt-tolerant self-proving mechanism to dynamically judge the authenticity of the pressure fracture judgment result and combines it with the acoustic fracture judgment result to output the final liquid column fracture judgment signal.

[0072] In this embodiment, the dual-state doubt-tolerant self-verification mechanism in the multi-channel intelligent decision-making unit 3 is used to evaluate the credibility of the pressure fracture judgment result and actively identify potential misjudgment patterns. When the pressure fracture judgment result is identified as unreliable, the acoustic fracture judgment result is combined with the liquid column fracture state to perform auxiliary judgment verification;

[0073] The two-state doubt-tolerant self-proving mechanism includes a misjudgment recognition model and an acoustic compensation module.

[0074] In this embodiment, the two-state doubt-tolerant self-proving mechanism operates in the liquid column endpoint determination stage, with the purpose of verifying the credibility of the pressure fracture judgment results within the framework of the pressure judgment channel and the acoustic judgment channel, so as to avoid misjudgment caused by the characteristics of the pressure signal directly entering the final output.

[0075] In this embodiment, the false positive identification model is constructed based on a multi-dimensional feature matching strategy, and is used to construct a doubt analysis vector based on the pressure rupture judgment result and the filling history data, and output a doubt score indicating the credibility of the pressure rupture judgment result;

[0076] The doubt analysis vector is a set of multidimensional anomaly feature combinations extracted from the pressure fracture judgment results and serves as the input of the misjudgment identification model. The multidimensional anomaly feature combination includes pressure fluctuation amplitude, disturbance peak response time, waveform oscillation symmetry, echo attenuation gradient, and disturbance response delay.

[0077] Among them, the doubt score is the nonlinear matching error between the doubt analysis vector and the characteristics of historical normal fracture samples, which is used to determine whether the pressure fracture judgment result is a misjudgment.

[0078] In this embodiment, the filling history data includes the original waveform of the pressure reflection signal collected during the filling process of multiple batches of photoresist in the past, the key feature values ​​extracted after preprocessing, the actual observation status of the filling end point, and the corresponding filling environment parameters.

[0079] In this embodiment, the method for constructing a tolerance analysis vector based on the pressure fracture judgment result and the filling history data is as follows: five dimensions of abnormal features are extracted from the pressure reflection signal corresponding to the current pressure fracture judgment result, namely, pressure fluctuation amplitude, disturbance peak response time, waveform oscillation symmetry, echo attenuation gradient and disturbance response delay; these five features are normalized with similar features in the filling history data sample; these five features are combined in a fixed order to form a set of multi-dimensional abnormal feature combinations as the input of the misjudgment recognition model.

[0080] In this embodiment, the specific calculation method of the doubt score is as follows: a nonlinear matching metric based on a kernel function is used to map the current doubt analysis vector to the same feature space as the historical normal fracture sample, and its weighted Euclidean distance to the feature center of the normal fracture sample is calculated as the residual value; the residual value is mapped to a doubt score between 0 and 1 by a sigmoid function. The closer the value is to 1, the greater the difference from the normal fracture pattern, and the higher the possibility of misjudgment.

[0081] In this embodiment, the specific method for determining whether the pressure fracture judgment result is a misjudgment is as follows: when the doubt score is greater than the doubt threshold, the system determines that the current pressure fracture judgment result is a high-risk misjudgment and triggers the acoustic compensation module to perform cross-channel verification; when the doubt score is lower than the threshold, the current result is considered credible, and the pressure fracture judgment result is directly output as the final endpoint judgment signal; in order to avoid false triggering caused by transient anomalies, the system will perform two consecutive scoring confirmations when the doubt score exceeds the threshold, and the acoustic compensation process will only be executed when the threshold is exceeded twice in a row.

[0082] In this embodiment, the acoustic compensation module is used to analyze whether the doubt score exceeds the set threshold. When the doubt score exceeds the set threshold, the acoustic fracture judgment result is enabled to replace the pressure fracture judgment result, and the final liquid column fracture judgment signal is output.

[0083] In this embodiment, after receiving the doubt score output by the misjudgment identification model, the acoustic compensation module first compares it with a preset doubt threshold, which is obtained by statistical analysis of multiple batches of experiments to ensure that normal fluctuations can be distinguished from abnormal fluctuations under actual filling conditions; if the doubt score does not exceed the threshold, the system directly outputs the pressure fracture judgment result as the final judgment signal; if the doubt score exceeds the threshold, the acoustic compensation process is entered; in the acoustic compensation process, the system calls the real-time fracture judgment result of the acoustic perception path, and matches the acoustic signal judgment result with the corresponding pressure signal event in the time series to ensure that the two correspond to the same liquid column end event; in order to avoid misjudgment of the acoustic path under short-term strong noise interference, the acoustic compensation module will also perform a consistency verification on the acoustic fracture judgment result, including a rapid comparison of the similarity of the sound spectrum energy peak position and duration with the typical fracture template; only when the acoustic result passes the consistency verification will it replace the original pressure fracture judgment result and be output as the final liquid column fracture judgment signal.

[0084] In this embodiment, the final liquid column fracture determination signal output contains two types of information: one is a fracture status flag, which is used to indicate whether the liquid column has fractured; the other is an event timestamp and a pulse number, which are used to record the moment when the fracture determination is triggered, so that the filling control system can execute the filling stop action or enter the terminal finishing procedure.

[0085] In the filling process data recording unit 4 of this embodiment, the filling process data recording unit 4 is used to record endpoint determination related parameters, fracture determination results and endpoint control feedback information in each filling process.

[0086] In this embodiment, endpoint determination-related parameters include micro-pressure pulse application parameters, reflected pressure signal characteristic values, liquid column length estimation values, critical length threshold intervals, acoustic signal characteristic values, and tolerance analysis vectors; fracture determination results include pressure fracture determination results, acoustic fracture determination results, and the final endpoint determination signal; the final endpoint determination signal includes the stop-filling instruction trigger time, the actual stop-filling response time, abnormal event records, and data archiving information.

[0087] In a second embodiment, the present invention provides a method for monitoring the state of photoresist filling, which is used in a photoresist filling state monitoring system in the first embodiment, and includes the following steps:

[0088] S10.1. Apply a micro-pressure pulse to the filling pipe and simultaneously collect the reflected pressure change signal to establish a correspondence between the liquid column length and pressure. When the liquid column length approaches the critical length threshold, a fracture trigger instruction is issued and a pressure fracture judgment result is output indicating the liquid column fracture state.

[0089] S10.2. Collect the acoustic pressure signal of the liquid column fracture, calculate the acoustic spectrum energy change, amplitude fluctuation rate, and duration indicators to establish a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state;

[0090] S10.3. Use a two-state doubt-tolerant self-verification mechanism to dynamically determine the authenticity of the pressure fracture determination result and combine it with the acoustic fracture determination result to output the final liquid column fracture determination signal;

[0091] S10.4. Record endpoint determination parameters, fracture determination results, and endpoint control feedback information during each filling process.

[0092] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A photoresist filling status monitoring system, characterized in that: include: The micro-pressure pulse control unit (1) is used to apply micro-pressure pulses to the filling pipe, collect reflected pressure change signals, establish a corresponding relationship between liquid column length and pressure, and issue a fracture trigger instruction when the liquid column length is close to a critical length threshold interval, and output a pressure fracture judgment result of the liquid column fracture state; The acoustic sensing and distinguishing unit (2) is used to collect the liquid column fracture sound pressure signal, calculate the sound spectrum energy change, amplitude fluctuation rate and duration index to construct a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state; The multi-channel intelligent decision-making unit (3) uses a dual-state doubt-tolerant self-verification mechanism to dynamically judge the authenticity of the pressure fracture judgment result and combines it with the acoustic fracture judgment result to output a final liquid column fracture judgment signal; The filling process data recording unit (4) is used to record endpoint determination related parameters, fracture determination results and endpoint control feedback information in each filling process.

2. The photoresist filling status monitoring system according to claim 1, characterized in that: The micro-pressure pulse is a periodic disturbance signal with an amplitude lower than the static pressure value of the liquid column and a pulse width less than 1 second, which is used to achieve liquid column state disturbance; the micro-pressure pulse is applied to the filling pipe according to the micro-pressure pulse frequency setting rules and pressure cycle adjustment strategy.

3. The photoresist filling status monitoring system according to claim 2, characterized in that: The reflected pressure change signal includes an initial echo response value, a multi-order attenuation echo amplitude sequence, and echo delay change data, which is used to determine whether the liquid column is close to fracture; In the micro-pressure pulse control unit (1), a corresponding relationship between the liquid column length and the pressure is established, and the specific method steps are as follows: The reflected pressure change signal is segmented according to the pulse number, and each segment of the reflected pressure change signal is mapped to an interval segment of the liquid column length range to form a pressure-length pre-calibration data set. The least squares method is used to establish a multivariate nonlinear mapping function between the liquid column length and the characteristic vector of the reflected pressure response signal, which is the corresponding relationship between the liquid column length and pressure. Among them, the multivariable nonlinear mapping function is used to estimate the liquid column length in real time based on the reflected pressure change signal.

4. The photoresist filling status monitoring system according to claim 3, characterized in that: The micro-pressure pulse control unit (1) compares the real-time estimated liquid column length of the multivariable nonlinear mapping function with the critical length threshold interval when the liquid column length is close to the critical length threshold interval, and issues a fracture trigger instruction when one of the fracture trigger conditions is met; The fracture triggering conditions are: the length of the liquid column enters the critical length threshold range for the first time and shows no growth trend within two consecutive pulse cycles; the length of the liquid column fluctuates within the critical length threshold range and the initial echo amplitude of the reflected signal shows a continuous decrease exceeding a preset ratio; The fracture trigger instruction is used to start the fracture state identification process and output the pressure fracture judgment result of the liquid column fracture state, and the pressure fracture judgment result includes the liquid column fracture state flag and the pulse number of the fracture occurrence time.

5. The photoresist filling status monitoring system according to claim 4, characterized in that: The acoustic perception and discrimination unit (2) performs segmented processing on the collected liquid column fracture sound pressure signal, and respectively calculates the sound spectrum energy change, amplitude fluctuation rate and duration index of the liquid column fracture sound pressure signal. The specific calculation method is as follows: Perform time-frequency analysis on the sound pressure signal of the liquid column fracture within the target frequency band, extract the total frequency domain amplitude change trend, and perform differential analysis with the static background sound spectrum to calculate the sound spectrum energy change; The envelope curve of the liquid column fracture sound pressure signal is extracted, and the number of amplitude changes per unit time is counted to obtain the amplitude fluctuation rate; according to the start and end time intervals of continuous high-amplitude segments in the liquid column fracture sound pressure signal, the duration interval of the liquid column fracture event is determined to obtain the duration index.

6. The photoresist filling status monitoring system according to claim 5, characterized in that: The non-contact liquid column endpoint state recognition mechanism is constructed based on the integration of acoustic spectrum energy changes, amplitude fluctuation rate and duration indicators, and is used to identify whether the liquid column has reached a broken state and whether there is a retraction residue or an unbroken dragline phenomenon. The specific construction method of the non-contact liquid column endpoint state recognition mechanism is as follows: Calculate the acoustic spectrum energy change, amplitude fluctuation rate, and duration index of the acoustic pressure signal of the photoresist filling liquid column fracture multiple times, and construct the acoustic spectrum energy change, amplitude fluctuation rate, and duration index calculated each time into a three-dimensional feature; The fracture state of the liquid column during each photoresist filling process is marked. A supervised multi-classification discrimination method is used to train the three-dimensional features and construct a state recognition and classification model. During the real-time filling process, the three-dimensional features of the real-time liquid column fracture acoustic pressure signal are input into the state recognition and classification model, and the real-time photoresist filling liquid column fracture state is output to obtain the acoustic fracture judgment result of the liquid column fracture state.

7. The photoresist filling status monitoring system according to claim 6, characterized in that: In the multi-channel intelligent decision-making unit (3), a dual-state doubt-tolerant self-verification mechanism is used to evaluate the credibility of the pressure fracture judgment result and actively identify potential misjudgment patterns. When the pressure fracture judgment result is identified as unreliable, the liquid column fracture state is auxiliary judged and verified in combination with the acoustic fracture judgment result; The two-state doubt-tolerant self-proving mechanism includes a misjudgment recognition model and an acoustic compensation module.

8. The photoresist filling status monitoring system according to claim 7, characterized in that: The false positive identification model is constructed based on a multi-dimensional feature matching strategy, and is used to construct a doubt analysis vector based on the pressure rupture judgment result and the filling history data, and output a doubt score indicating the credibility of the pressure rupture judgment result; The doubt analysis vector is a set of multidimensional anomaly feature combinations extracted from the pressure fracture judgment results and serves as the input of the misjudgment identification model. The multidimensional anomaly feature combination includes pressure fluctuation amplitude, disturbance peak response time, waveform oscillation symmetry, echo attenuation gradient, and disturbance response delay. Among them, the doubt score is the nonlinear matching error between the doubt analysis vector and the characteristics of historical normal fracture samples, which is used to determine whether the pressure fracture judgment result is a misjudgment.

9. The photoresist filling status monitoring system according to claim 8, characterized in that: The acoustic compensation module is used to analyze whether the doubt score exceeds the set threshold. When the doubt score exceeds the set threshold, the acoustic fracture judgment result is enabled to replace the pressure fracture judgment result, and the final liquid column fracture judgment signal is output.

10. A method for monitoring the state of photoresist filling, used in a system for monitoring the state of photoresist filling according to any one of claims 1 to 9, characterized in that: The steps include: S10.

1. Apply a micro-pressure pulse to the filling pipe and simultaneously collect the reflected pressure change signal to establish a correspondence between the liquid column length and pressure. When the liquid column length approaches the critical length threshold, a fracture trigger instruction is issued and a pressure fracture judgment result is output indicating the liquid column fracture state. S10.

2. Collect the acoustic pressure signal of the liquid column fracture, calculate the acoustic spectrum energy change, amplitude fluctuation rate, and duration indicators to establish a non-contact liquid column endpoint state recognition mechanism, and output the acoustic fracture judgment result of the liquid column fracture state; S10.

3. Use a two-state doubt-tolerant self-verification mechanism to dynamically determine the authenticity of the pressure fracture determination result and combine it with the acoustic fracture determination result to output the final liquid column fracture determination signal; S10.

4. Record endpoint determination parameters, fracture determination results, and endpoint control feedback information during each filling process.

Images