A method and system for evaluating the filtering effect of waste gas in semiconductor preparation

Abstract

The present invention discloses a method and system for evaluating the filtering effect of semiconductor manufacturing waste gas, specifically relating to the technical field of waste gas filtration; by synchronously collecting response data through multiple types of sensors, extracting interference features including abnormal spectrogram morphological features and time variation gradient features, and based on historical experimental data and gas interference mechanisms, constructing a machine learning model to predict the severity of errors, implementing a differential concentration correction strategy according to the error level, obtaining the corrected target gas concentration value, and finally using it to accurately evaluate the purification efficiency of the filtration system. The present invention realizes the intelligent identification and correction of cross-interference, significantly improving the accuracy of waste gas monitoring and the reliability of system evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of waste gas filtration, and particularly to a method and system for evaluating the filtration effect of waste gas in semiconductor manufacturing. Background Art

[0002] The evaluation of the filtration effect of waste gas in semiconductor manufacturing refers to the systematic analysis and evaluation of the purification efficiency and treatment effect of the filtration system through a series of methods and indicators after the generated waste gas is filtered during the semiconductor manufacturing process. This evaluation usually includes the detection and analysis of aspects such as the removal rate of harmful gases, the performance of filter materials, and the change in the concentration of discharged gas, aiming to ensure that the waste gas emissions meet environmental protection standards, guarantee production safety, and reduce the impact on the environment.

[0003] The following are the deficiencies of the existing technology:

[0004] During the semiconductor manufacturing process, the evaluation of the waste gas filtration effect is crucial. However, in the existing technology, when the continuous emission monitoring system (CEMS) detects the target gas, cross-interference errors often occur due to the interference of other components, resulting in inaccurate readings. For example, NDIR or electrochemical sensors may misjudge as , or the presence of interferes with the detection, resulting in distorted evaluation of pollutant concentration or purification efficiency. Such errors usually occur in the case of coexistence of multiple gases, similar concentrations, or inaccurate instrument calibration, posing challenges to high-precision evaluation. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for evaluating the filtration effect of waste gas in semiconductor manufacturing to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for evaluating the filtration effect of waste gas in semiconductor manufacturing, comprising:

[0007] Collecting response data obtained by synchronously detecting the same waste gas to be measured by multiple different types of gas sensors;

[0008] Extracting gas interference characteristics from the response data of different sensors, where the interference characteristics include abnormal spectral pattern characteristics and time variation gradient characteristics between sensors;

[0009] Based on historical experimental data and known interference gas influence mechanisms, training a cross-interference prediction model using a machine learning algorithm, and inputting the real-time collected gas interference characteristics into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data;

[0010] Select different gas concentration correction strategies according to the severity of the error, and perform differential correction on the original monitoring data of the target gas to obtain the target gas concentration value after cross-interference correction;

[0011] Evaluate the purification efficiency of the semiconductor waste gas filtration system based on the corrected target gas concentration value.

[0012] Preferably, the different sensors include NDIR infrared sensors, electrochemical gas sensors, Fourier transform infrared spectrometers, and mass spectrometers.

[0013] Preferably, the interference characteristics include abnormal spectral pattern characteristics between sensors. The extraction method is as follows: There are two sensors A and B. A total of n time points are sampled within the same time window, and two response sequences are obtained respectively: ; where: is the original response value of sensor A at the nth time point, is the original response value of sensor B at the nth time point. The two sequences are respectively normalized. After normalization, the Euclidean distance between the two sequences is calculated as the abnormal spectral pattern characteristic.

[0014] Preferably, the interference characteristics include the time change gradient characteristics between sensors. The extraction method is as follows: Let the discrete signal sequence of the sensor response changing with time be: ; The corresponding time points are: ; where represents the sensor response value at the mth time point, is the corresponding timestamp. Take the timestamp as the horizontal axis and the sensor response value as the vertical axis to construct a rectangular coordinate system, generate the sensor response curve S(t), and calculate its first derivative ; Select a sliding time window w with a fixed length. At each window , calculate the difference between the maximum change rate and the minimum change rate of the sensor response as the time change gradient characteristic.

[0015] Preferably, convert the abnormal spectral pattern characteristics and the time change gradient characteristics into a comprehensive feature vector. Use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes predicting the error severity value label caused by cross-interference in the current monitoring data for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the error severity value labels caused by cross-interference in all current monitoring data as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the error severity value caused by cross-interference in the current monitoring data according to the model output result. Among them, the machine learning model is a polynomial regression model.

[0016] Preferably, compare the severity value of the error caused by cross-interference in the currently obtained monitoring data with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the severity value of the error caused by cross-interference in the current monitoring data with the first standard threshold and the second standard threshold respectively;

[0017] If the severity value of the error is greater than the second standard threshold, it is determined that the current monitoring data is in a high-risk interference state, and the alarm mechanism needs to be triggered immediately or the redundant sensing data correction process needs to be started; if the severity value of the error is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it is determined as a medium-risk interference state; if the severity value of the error is less than the first standard threshold, it is determined as a low-risk interference state.

[0018] Preferably, correct the original concentration value Craw of the target gas at each moment according to the selected strategy, and calculate the corrected concentration value Ccorrected: ; where ΔC is the concentration error value caused by cross-interference, which is defined as the difference between the sensor output and the true concentration of the target gas;

[0019] Output the corrected target gas concentration Ccorrected as the true concentration estimate value at the current time point.

[0020] Preferably, the purification efficiency calculation formula of the semiconductor waste gas filtration system is: ; where η is the purification efficiency, is the concentration of the target gas before the waste gas enters the filtration system, is the concentration of the target gas after the waste gas is treated;

[0021] For each set of sampling data, calculate the purification efficiency value based on the corrected inlet and outlet concentrations;

[0022] Visualize the purification efficiency result. If the efficiency is lower than the set threshold, trigger a warning prompt to assist the maintenance personnel in making judgments and optimizing adjustments.

[0023] The present invention also provides a semiconductor manufacturing waste gas filtration effect evaluation system, including a data acquisition module, a feature extraction module, an error prediction module, a correction module, and a purification efficiency evaluation module;

[0024] Data acquisition module: Collect the response data obtained by synchronously detecting the same waste gas to be measured by multiple different types of gas sensors;

[0025] Feature extraction module: Extract gas interference features from the response data of the different sensors, where the interference features include abnormal spectral pattern features and time-varying gradient features between the sensors;

[0026] Error prediction module: Based on historical experimental data and known interference gas influence mechanisms, use a machine learning algorithm to train a cross-interference prediction model, and input the gas interference features collected in real time into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data;

[0027] Calibration module: According to the severity of the error, select different gas concentration calibration strategies to differentially calibrate the original monitoring data of the target gas, so as to obtain the target gas concentration value corrected for cross-interference;

[0028] Purification efficiency evaluation module: Evaluate the purification efficiency of the semiconductor waste gas filtration system based on the calibrated target gas concentration value.

[0029] In the above technical solution, the technical effects and advantages provided by the present invention:

[0030] 1. By integrating the response data of multiple types of sensors, the present invention extracts representative abnormal spectral pattern features and time-varying gradient features, uses a machine learning model to intelligently predict the severity of the error caused by cross-interference, and implements differential concentration calibration strategies according to different interference levels, thereby effectively improving the accuracy of the target gas concentration data. This method overcomes the identification error problems brought by factors such as the coexistence of multiple gases and poor sensor selectivity in traditional continuous emission monitoring systems, and is particularly suitable for high-precision dynamic monitoring of complex mixed waste gases in the semiconductor manufacturing process.

[0031] 2. Based on the calibrated target gas concentration value, the present invention constructs a purification efficiency evaluation model, which can realize scientific, real-time and visual evaluation of the purification effect of the semiconductor waste gas filtration system. When the purification efficiency is lower than the set threshold, the system can trigger an alarm in time, which helps operation and maintenance personnel discover problems such as filter material aging or system abnormalities, and realizes the whole-process closed-loop control. Overall, the present invention significantly improves the monitoring reliability, evaluation accuracy and intelligent level of the waste gas treatment system, and has good engineering practical value and promotion prospects. Description of the drawings

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0033] Figure 1 This is the flowchart of the method of the present invention.

[0034] Figure 2 This is the system module diagram of the present invention. Detailed implementation manners

[0035] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] Example 1. Please refer to Figure 1 As shown, a method for evaluating the filtering effect of waste gas in semiconductor manufacturing includes:

[0037] Collecting response data obtained by synchronously detecting the same waste gas to be measured by multiple different types of gas sensors;

[0038] Extracting gas interference characteristics from the response data of different sensors, where the interference characteristics include abnormal spectral pattern characteristics and time change gradient characteristics between sensors;

[0039] Based on historical experimental data and known interference gas influence mechanisms, training a cross-interference prediction model using a machine learning algorithm, and inputting the gas interference characteristics collected in real time into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data;

[0040] According to the severity of the error, selecting different gas concentration correction strategies to differentially correct the original monitoring data of the target gas to obtain the target gas concentration value corrected for cross-interference;

[0041] Evaluating the purification efficiency of the semiconductor waste gas filtering system based on the corrected target gas concentration value.

[0042] In a complex semiconductor waste gas environment, the chemical properties, molecular structures, spectral responses, etc. of different gases are different. Therefore, a single type of sensor is often difficult to comprehensively and accurately perceive. For this reason, the system is usually configured with the following types of sensors:

[0043] NDIR infrared sensor (Non-Dispersive Infrared): mainly used to detect gases with infrared absorption characteristics, such as , , etc. Such sensors deduce the gas concentration by measuring the absorption intensity of infrared light at specific wavelengths. It has a fast response and a long lifespan, but when multiple gases with similar absorption bands coexist, it is prone to interference and the accuracy decreases.

[0044] Electrochemical gas sensors: suitable for detecting , , , etc. gases. It generates a current signal through the electrochemical reaction of the detected gas on the electrode surface, and then calculates the concentration. Although it has high sensitivity and low power consumption, there are problems such as poor selectivity and sensitivity to temperature and humidity.

[0045] Fourier transform infrared spectrometer (FTIR): can simultaneously identify and quantitatively analyze multiple gas components. This technology is based on the absorption characteristics of molecules in the mid-infrared region and can provide a complete absorption spectrum. Although it is large in volume and high in cost, its resolution and anti-interference ability are extremely strong, and it is an important tool for the analysis of complex gas mixtures.

[0046] Mass spectrometer (such as gas chromatography-mass spectrometry GC-MS): can achieve qualitative and quantitative detection of trace components, is suitable for low-concentration and high-precision analysis, and has advantages especially in the monitoring and traceability analysis of trace impurities. It is usually used for offline verification or calibration of models and is not often used for real-time online, but its data is crucial for model training.

[0047] In actual deployment, these sensors are arranged at both the inlet and outlet ends of the exhaust gas filtration unit to achieve real-time monitoring of untreated and treated gases. The system uses a unified data acquisition module (such as an industrial Internet of Things acquisition card or an edge computing terminal) to synchronously trigger various sensors for data acquisition with millisecond-level precision, ensuring that the data of all sensors can correspond to the state of the exhaust gas at the same moment and avoiding misjudgment caused by time differences.

[0048] The collected data includes raw signals (such as absorption light intensity, current, voltage, spectra, etc.). The sampling frequency of each type of sensor is set according to its response speed. For example, the NDIR sensor can reach 1 Hz, the electrochemical sensor is usually 0.1 - 1 Hz, and the FTIR device usually scans several times per second.

[0049] The multi-dimensional data collected by multiple sensors not only complement each other in terms of accuracy, but more importantly, enhance each other in terms of information dimension: when the NDIR sensor makes a fuzzy judgment due to spectral overlap of certain gases, it can be corrected by the full-spectrum absorption data of FTIR; the electrochemical sensor is sensitive to certain small-molecule gases and can be used as an auxiliary judgment basis for the NDIR system; multiple signals perform "redundant verification" on the same gas, significantly improving the confidence of discrimination; when a certain sensor experiences short-term drift or data anomaly, the system can detect the anomaly and perform automatic substitution compensation by comparing the outputs of other sensors, enhancing the overall robustness. In addition, the system can also establish a cross-reference model or a joint feature vector based on the data of different sensors, providing more comprehensive and accurate data support for subsequent gas concentration inversion models or cross-interference correction models.

[0050] In this application, by deploying multiple types of sensors and performing synchronous detection, not only can the recognition accuracy of complex waste gas components be improved, but also the risk of misjudgment caused by cross-interference can be effectively addressed. This technical solution lays a solid foundation for constructing a highly reliable and intelligent waste gas filtration effect evaluation system, and also provides a rich and reliable data source for subsequent algorithm modeling and model correction. It reflects the current development trend of waste gas detection systems from "single-point perception" to "multi-source fusion intelligent recognition".

[0051] When the sensor responds to the target gas, it will output a curve or spectrum of the change of an electrical signal (such as voltage, current, infrared absorption intensity, etc.) over time. If the sensor is subject to cross-interference, its response curve will exhibit abnormal morphological characteristics, different from the characteristic profile under the standard state. The so-called abnormal spectral morphological characteristics refer to the inconsistent or unreasonable deviation of the response curves of multiple sensors to the same waste gas within the same time window in terms of shape, position, amplitude, etc. By extracting these abnormalities, it is possible to determine whether there is cross-interference.

[0052] The following methods can be used for the extraction and quantization of spectral features:

[0053] Waveform comparison difference: After normalizing the output curves of two or more sensors, calculate the Euclidean distance, dynamic time warping (DTW) distance or similarity measure (such as cosine similarity) between them;

[0054] Peak drift detection: Analyze whether the peak time position or peak amplitude in the response curve of each sensor has shifted. Typical features include main peak displacement (Peak Shift), waveform misalignment, appearance of abnormal double-peak structure, etc.;

[0055] Spectral profile anomaly detection: Use Fourier transform or wavelet transform to extract frequency domain features, compare the spectral distributions of different sensors. If there are inconsistent spectral energy distributions, missing frequency bands, or multi-frequency anomalies, they are regarded as morphological anomaly features.

[0056] Higher-order statistic analysis: Extract statistics such as the first derivative (slope), second derivative (curvature), skewness, and kurtosis of the response curve, and compare whether there are significant differences between different sensors.

[0057] The abnormal spectral pattern features can reveal the structural differences in the response behaviors of multiple sensors under non-ideal conditions, and are highly reliable feature bases for judging whether there is non-target gas interference. For machine learning modeling using spectral pattern features, they provide input variables with strong separability.

[0058] For example, the present invention can adopt the following method to extract abnormal spectral pattern features:

[0059] Suppose there are two sensors A and B, and a total of n time points are sampled within the same time window, obtaining two response sequences respectively: ; where: is the original response value of sensor A at the nth time point, is the original response value of sensor B at the nth time point. To eliminate the bias caused by amplitude differences, normalize the two sequences respectively (such as Z-score normalization or Min-Max normalization). Z-score normalization (standardization): ; where: is the mean of sequence A, is the standard deviation of sequence A. Similarly, 、 are the mean and standard deviation of B; 、 are the normalized response values.

[0060] After normalization, calculate the Euclidean distance between the two sequences, as the abnormal spectral pattern feature, representing the difference degree between the two response curves. The expression is: ; n is the total number of time points.

[0061] If ≈0, it means that the two curves are highly consistent and there is no obvious interference in the sensor response; if increases significantly, it means that there are significant differences between the sensor response curves, which may be caused by differences in target gas concentrations, cross-interference or faults in a certain sensor; the system can set an empirical threshold δ. If >δ, trigger the interference detection mechanism or start the correction model.

[0062] The time-varying gradient feature refers to the rate of change or trend of the sensor response value over time during continuous monitoring. If the exhaust gas components are pure and stable, the responses of each sensor should change slowly over time; however, when a certain interfering gas suddenly enters or its concentration changes abruptly, the sensor output will exhibit a drastic or non-linear change. This change gradient can reflect the instantaneous characteristics of cross-interference.

[0063] The extraction methods include:

[0064] First derivative (Δ value) analysis: Calculate the first derivative of the sensor response curve S(t), which is the amplitude of signal change per unit time. A sudden change or abnormal rise / fall indicates a possible unexpected gas perturbation.

[0065] Second derivative (acceleration) analysis: If the first derivative changes drastically, then calculate the second derivative to identify the inflection point of signal change, especially suitable for detecting the response rate inversion or non-linear steep increase caused by cross-interference.

[0066] Sliding window statistical features: Extract the mean, standard deviation, change amplitude, maximum jump rate, etc. within a fixed time window (such as 5 seconds) for feature input to the machine learning model.

[0067] Abnormal gradient threshold identification: Set an empirical threshold. If the change gradient in a certain time period exceeds the threshold, it is determined as a potential interference state.

[0068] The time-varying gradient feature reflects the immediate impact characteristics of the dynamic fluctuations of exhaust gas components on the sensor response; it is suitable for identifying cross-errors caused by transient interferences (such as gas pulses, valve switching, etc.); it is easy to calculate in real time and is suitable for deployment in embedded edge processing units to support rapid early warning.

[0069] For example, the present invention extracts the time-varying gradient feature in the following manner:

[0070] Let the discrete signal sequence of the sensor response changing with time be: ; the corresponding time points are: ; where represents the sensor response value at the m-th time point, is the corresponding timestamp (in seconds), usually with equally spaced sampling. Taking the timestamp as the horizontal axis and the sensor response value as the vertical axis, construct a rectangular coordinate system to generate the sensor response curve S(t), and calculate its first derivative, the expression is: ; represents the change rate of the sensor response within the i-th time interval, represents the sensor response value at the i-th time point, is the corresponding timestamp;

[0071] Select a sliding time window \(w\) of a fixed length (e.g., 5 seconds, corresponding to \(w\) sampling points), and in each window , calculate the difference between the maximum change rate and the minimum change rate of the sensor response as the time change gradient feature. The larger the time change gradient feature, the more significant the potential cross-interference situation.

[0072] Convert the abnormal spectrogram morphological feature and the time change gradient feature into a comprehensive feature vector. Use the comprehensive feature vector as the input of a machine learning model. The machine learning model takes predicting the error severity value label caused by cross-interference in the current monitoring data as the prediction target, and minimizing the sum of the prediction errors of the error severity value labels caused by cross-interference in all current monitoring data as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the error severity value caused by cross-interference in the current monitoring data according to the model output result. Among them, the machine learning model is a polynomial regression model.

[0073] Compare the error severity value caused by cross-interference in the obtained current monitoring data with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. The first standard threshold (lower, set as the slight error tolerance range); the second standard threshold (higher, set as the boundary of unacceptable errors); Compare the error severity value caused by cross-interference in the current monitoring data with the first standard threshold and the second standard threshold respectively;

[0074] If the error severity value is greater than the second standard threshold, it is determined that the current monitoring data is in a high-risk interference state, and the alarm mechanism needs to be triggered immediately or the redundant sensing data correction process needs to be started.

[0075] If the error severity value is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it is determined as a medium-risk interference state. The system records this state and can selectively perform soft correction or dynamic filtering, and prompt the user to pay attention.

[0076] If the error severity value is less than the first standard threshold, it is determined as a low-risk or no significant interference state, and the current monitoring data can be considered credible.

[0077] After obtaining the error severity value caused by cross-interference in the current monitoring data, based on this prediction result, this method further selects a corresponding gas concentration correction strategy to differentially process the original monitoring data of the target gas, so as to obtain the target gas concentration value after interference correction. The specific steps are as follows:

[0078] Preset multiple calibration strategies, each corresponding to an interference level (high, medium, low). Each strategy adopts different processing methods and model accuracies to match different levels of error risks and ensure a balance between performance and accuracy:

[0079] Low error level (the error severity value is less than the first standard threshold); Mild correction strategy, which can choose to use a fixed ratio correction factor or not correct to avoid excessive intervention;

[0080] Medium error level (between the first and second standard thresholds); Medium complexity calibration strategy, which is adjusted using empirical or parametric functions, such as local linear correction or bivariate interpolation;

[0081] High error level (the error severity value is greater than the second standard threshold); Strong correction strategy, which calls a trained machine learning regression model or finds similar patterns based on historical data for fitting and correction.

[0082] For the original concentration value Craw of the target gas at each moment, calibrate according to the selected strategy and calculate the corrected concentration value Ccorrected, and its expression form is as follows:

[0083] General expression: ; where ΔC is the concentration error value caused by cross-interference, defined as the difference between the sensor output and the true concentration of the target gas:

[0084] Example 1: Fixed ratio correction (mild correction): ; where α is a small interference ratio determined empirically (such as 1% - 2%).

[0085] Example 2: Correction based on model regression output (strong correction): ; where fmodel(X) is the error regression prediction value with the current feature vector as the input.

[0086] The differential calibration strategy has an adaptive ability and can dynamically adjust calibration parameters according to the statistical data during system operation, such as the size of the correction factor, classification threshold, strategy call order, etc., to adapt to the effects of sensor aging, gas component changes, environmental disturbances, etc.

[0087] Output the corrected target gas concentration Ccorrected as the true concentration estimate value at the current time point for use in evaluating the waste gas treatment effect, judging emission compliance, and subsequent control strategies.

[0088] The purification efficiency of the semiconductor waste gas filtration system usually refers to the removal ratio of the treatment unit for a specific target gas, reflecting the purification ability of the system for harmful or polluting components between the inlet and outlet. Its calculation formula is: ; where η is the purification efficiency (%), is the target gas concentration before the waste gas enters the filtration system (after interference correction), is the target gas concentration after the waste gas is treated (also after interference correction).

[0089] Sensor systems are arranged at the inlet end and the outlet end of the filtration unit respectively to monitor the target gas concentration in real time, and the aforementioned cross-interference correction algorithm is applied to all the original data to ensure the accuracy and reliability of the data.

[0090] For each set of sampled data, based on the corrected inlet and outlet concentrations, the purification efficiency value is calculated according to the above formula, and the dynamic purification efficiency evaluation on the continuous time series can be realized.

[0091] If combined with multiple gas components and different working conditions (flow rate, temperature, humidity, etc.), the change trend of the efficiency with the system state can also be analyzed and predicted by constructing a multivariable purification efficiency model.

[0092] The system can visually display the purification efficiency results. If the efficiency is lower than the set threshold (such as 85%), a warning prompt will be triggered, indicating possible problems such as saturated filter materials, equipment aging, and process out-of-control, to assist maintenance personnel in judgment and optimization adjustment.

[0093] Example 2, please refer to Figure 2 As shown, a waste gas filtration effect evaluation system for semiconductor preparation in this embodiment includes a data acquisition module, a feature extraction module, an error prediction module, a correction module, and a purification efficiency evaluation module;

[0094] Data acquisition module: Collect the response data obtained by synchronously detecting the same waste gas to be measured by multiple different types of gas sensors;

[0095] Feature extraction module: Extract gas interference features from the response data of the different sensors, and the interference features include abnormal spectrogram morphological features and time change gradient features between the sensors;

[0096] Error prediction module: Based on historical experimental data and the known interference gas influence mechanism, train a cross-interference prediction model using a machine learning algorithm, and input the gas interference features collected in real time into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data;

[0097] Correction module: According to the severity of the error, select different gas concentration correction strategies to differentially correct the original monitoring data of the target gas to obtain the target gas concentration value corrected by cross-interference;

[0098] Purification efficiency evaluation module: Based on the corrected target gas concentration value, evaluate the purification efficiency of the semiconductor waste gas filtration system.

[0099] All the above formulas are dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0100] It should be understood that the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context before and after.

[0101] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0102] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A method for evaluating the filtering effect of waste gas in semiconductor preparation, characterized in that: Including: Collecting response data obtained by synchronously detecting the same exhaust gas to be measured with multiple different types of gas sensors; Extracting gas interference characteristics from the response data of different sensors, where the interference characteristics include abnormal spectral pattern characteristics and time change gradient characteristics between sensors; The method for extracting the time-varying gradient feature is as follows: Let the discrete signal sequence of the sensor response varying with time be: S = [s1, s2,..., s m ; The corresponding time points are: T = [t1, t2,..., t m ; where s m represents the sensor response value at the m-th time point, and t m is the corresponding timestamp. Taking the timestamp t m as the horizontal axis and the sensor response value s m as the vertical axis, a rectangular coordinate system is constructed to generate the sensor response curve S(t), and its first derivative Δs i is calculated; A sliding time window w with a fixed length is selected. For each window W i = {Δs i , Δs i+1 ,..., Δs i+w-1}, the difference between the maximum change rate and the minimum change rate of the sensor response is calculated as the time-varying gradient feature; Based on historical experimental data and known interference gas influence mechanisms, using a machine learning algorithm to train a cross-interference prediction model, and inputting the gas interference characteristics collected in real time into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data; Specifically including: converting the abnormal spectral pattern characteristics and time change gradient characteristics into a comprehensive feature vector, using the comprehensive feature vector as the input of the machine learning model, the machine learning model takes predicting the severity value label of the error caused by cross-interference in the current monitoring data as the prediction target, and minimizing the sum of the prediction errors of the severity value labels of the errors caused by cross-interference in all current monitoring data as the training target, training the machine learning model until the sum of the prediction errors reaches convergence and then stopping the model training, and determining the severity value of the error caused by cross-interference in the current monitoring data according to the model output result, where the machine learning model is a polynomial regression model; According to the severity of the error, selecting different gas concentration correction strategies to differentially correct the original monitoring data of the target gas to obtain the target gas concentration value corrected for cross-interference; Specifically including: correcting the original concentration value Craw of the target gas at each moment according to the selected strategy, and calculating the corrected concentration value Ccorrected: Ccorrected = Craw - ΔC; where ΔC is the concentration error value caused by cross-interference, defined as the difference between the sensor output and the true target gas concentration; outputting the corrected target gas concentration Ccorrected as the true concentration estimate value at the current time point; Evaluate the purification efficiency of the semiconductor waste gas filtration system based on the corrected target gas concentration value; specifically including: The calculation formula for the purification efficiency of the semiconductor waste gas filtration system is: Among them, η is the purification efficiency, C in is the target gas concentration before the waste gas enters the filtration system, and C out is the target gas concentration after the waste gas is treated; for each set of sampling data, calculate the purification efficiency value based on the corrected inlet and outlet concentrations; visually display the purification efficiency results, and if the efficiency is lower than the set threshold, trigger a warning prompt to assist the maintenance personnel in judgment and optimization adjustment.

2. The evaluation method for the filtering effect of waste gas in semiconductor preparation according to claim 1, wherein: The different sensors include NDIR infrared sensors, electrochemical gas sensors, Fourier transform infrared spectrometers, and mass spectrometers.

3. A method for evaluating the filtering effect of waste gas in semiconductor preparation according to claim 1, characterized in that: The interference features include abnormal spectrogram morphological features between sensors, and the extraction method is as follows: Suppose there are two sensors A and B, and a total of n time points are sampled within the same time window, obtaining two response sequences respectively: A = [a1, a2,..., a n , B = [b1, b2,..., b n ; where: a n is the original response value of sensor A at the nth time point, b n is the original response value of sensor B at the nth time point. After normalizing the two sequences respectively, calculate the Euclidean distance between the two sequences as the abnormal spectrogram morphological feature.

4. A method for evaluating the filtering effect of waste gas in semiconductor preparation according to claim 3, characterized in that: Comparing the severity value of the error caused by cross-interference in the currently obtained monitoring data with the gradient standard thresholds, where the gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and comparing the severity value of the error caused by cross-interference in the current monitoring data with the first standard threshold and the second standard threshold respectively; If the severity value is greater than the second standard threshold, it is determined that the current monitoring data is in a high-risk interference state, and the alarm mechanism needs to be triggered immediately or the redundant sensing data correction process needs to be started; If the severity value is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it is determined as a medium-risk interference state; if the severity value is less than the first standard threshold, it is determined as a low-risk interference state.

5. A semiconductor manufacturing waste gas filtration effect evaluation system for implementing the semiconductor manufacturing waste gas filtration effect evaluation method according to any one of claims 1-4, characterized in that: Including a data acquisition module, a feature extraction module, an error prediction module, a correction module, and a purification efficiency evaluation module; Data acquisition module: Collect the response data obtained from the synchronous detection of the same exhaust gas to be measured by multiple gas sensors of different types; Feature extraction module: Extract gas interference features from the response data of different sensors. The interference features include abnormal spectral pattern features and time-varying gradient features between sensors; Error prediction module: Based on historical experimental data and known interference gas influence mechanisms, use machine learning algorithms to train a cross-interference prediction model, and input the gas interference features collected in real time into the cross-interference prediction model to predict the severity of the error caused by cross-interference in the current monitoring data; Calibration module: According to the severity of the error, select different gas concentration calibration strategies to differentially calibrate the original monitoring data of the target gas to obtain the target gas concentration value corrected for cross-interference; Purification efficiency evaluation module: Evaluate the purification efficiency of the semiconductor exhaust gas filtration system based on the calibrated target gas concentration value.

Images