A method for detecting semiconductor exhaust gas concentration based on noise cancellation

Abstract

This invention relates to the field of semiconductor exhaust gas monitoring technology, specifically a method for detecting semiconductor exhaust gas concentration based on noise cancellation. The method includes: assembling a primary infrared spectral sensor along the main exhaust gas emission path of a semiconductor process chamber; and placing a reference infrared spectral sensor with identical structural parameters nearby to monitor common-mode noise signals. The two sensors are synchronously triggered according to a preset sampling sequence to acquire the primary absorption spectrum signal and the reference background spectrum signal. The signals are processed using an improved adaptive filtering algorithm that adjusts the weighting coefficients based on a real-time noise correlation matrix. The characteristic absorption spectrum of the target gas after noise cancellation is extracted, and combined with a pre-stored standard database of gas absorption spectra, the concentration value of the specific gas in the exhaust gas is calculated. This method optimizes the noise cancellation effect of the spectral signal through dual-sensor noise matching and dynamic filtering adjustment, thereby improving the accuracy of semiconductor exhaust gas concentration detection.

Description

Technical Field

[0001] This invention relates to the field of semiconductor exhaust gas monitoring technology, and in particular to a method for detecting semiconductor exhaust gas concentration based on noise cancellation. Background Technology

[0002] The detection of exhaust gas concentration in semiconductor process chambers mostly relies on infrared spectroscopy detection technology. In the industry, it is common to deploy a single infrared spectral sensor in the main exhaust gas emission path to collect the absorption spectrum signal of the exhaust gas to be tested. Some detection schemes will add a reference sensor to collect the background signal. The two types of spectral signals are processed by conventional adaptive filtering algorithms, and then combined with the gas absorption spectrum standard database to complete the calculation of the concentration of a specific gas in the exhaust gas.

[0003] The spectral signal acquired by a single master sensor is mixed with environmental interference and common-mode noise generated by the sensor itself. Since conventional reference sensors and master sensors have different structural parameters, the acquired background signal cannot accurately match the noise signal characteristics of the master sensor. Conventional adaptive filtering algorithms use fixed filter weight coefficients, which cannot adjust parameters to keep up with real-time changes in noise signals. This results in limited noise cancellation effectiveness, and the extracted characteristic absorption spectra retain a significant amount of noise interference, ultimately leading to inaccuracies in the calculated gas concentration.

[0004] It is necessary to deploy a reference sensor with the same structural parameters as the main sensor to collect the common-mode noise signal corresponding to the main signal. It is also necessary to adjust the filtering algorithm parameters by calculating the noise-related characteristics in real time to improve the noise cancellation effect of the spectral signal, weaken the influence of noise on the characteristic absorption spectrum of the target gas, and make the spectral signal on which the exhaust gas concentration calculation is based more closely match the actual detection state. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a semiconductor exhaust gas concentration detection method based on noise cancellation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a semiconductor exhaust gas concentration detection method based on noise cancellation, comprising: A main infrared spectral sensor is installed on the main exhaust path of the semiconductor process chamber, and a reference infrared spectral sensor with the same structural parameters is installed in the vicinity of the main infrared spectral sensor. The reference infrared spectral sensor is used to monitor the common-mode noise signal introduced by the environment and the sensor itself. By using a preset sampling sequence, the main infrared spectral sensor and the reference infrared spectral sensor are synchronously triggered to collect the main absorption spectral signal and the reference background spectral signal of the exhaust gas to be tested, respectively. An improved adaptive filtering algorithm is used to process the main absorption spectrum signal and the reference background spectrum signal. The improved adaptive filtering algorithm adjusts its internal filter weight coefficients based on the noise correlation matrix calculated in real time. The noise-cancelled characteristic absorption spectrum of the target gas is extracted from the signal processed by the improved adaptive filtering algorithm. Based on the characteristic absorption spectrum of the target gas and combined with a pre-stored standard database of gas absorption spectra, the concentration of a specific gas in the exhaust gas emitted from the semiconductor process chamber is calculated.

[0007] As a further aspect of the present invention, the step of synchronously triggering the main infrared spectral sensor and the reference infrared spectral sensor through a preset sampling sequence to respectively acquire the main absorption spectral signal and the reference background spectral signal of the exhaust gas to be measured includes: Send start acquisition commands with the same timestamp to the main infrared spectral sensor and the reference infrared spectral sensor; Within the same integration time, the main infrared spectral sensor receives an infrared beam of the exhaust gas flowing through the main exhaust emission path and generates the main absorption spectral signal containing the characteristic absorption of the exhaust gas and background noise. Within the same integration time, the reference infrared spectral sensor receives a reference beam that does not pass through the flowing exhaust gas, originates from the same infrared source, but is attenuated by a neutral density filter, and generates the reference background spectral signal, which mainly contains background noise from the environment and the sensor itself. The main absorption spectrum signal and the reference background spectrum signal are aligned according to wavelength channels and stored in the data buffer corresponding to the same timestamp.

[0008] As a further aspect of the present invention, the improved adaptive filtering algorithm adjusts its internal filter weight coefficients based on the noise correlation matrix calculated in real time, including: From the reference background spectral signal, extract the spectral intensity sequence within a preset wavelength range as a reference noise vector; From the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range, the initial profile of the target gas absorption component is estimated, and the initial profile is subtracted from the main signal to obtain the main noise vector; Calculate the cross-correlation function between the reference noise vector and the main noise vector within a sliding time window, and construct the noise correlation matrix based on the cross-correlation function; Based on the eigenvalues ​​and eigenvectors of the noise correlation matrix, the tap weight coefficients of the finite impulse response filter in the improved adaptive filtering algorithm are dynamically adjusted so that the frequency response of the finite impulse response filter matches the noise statistical characteristics at the current moment. The reference background spectral signal is filtered using the finite impulse response filter with adjusted tap weights to generate an estimated noise signal. The estimated noise signal is subtracted from the main absorption spectrum signal to obtain the preliminary noise-cancelled spectrum signal.

[0009] As a further aspect of the present invention, estimating the initial profile of the target gas absorption component from the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range includes: Obtain the standard absorption spectral profile of the target gas within the preset wavelength range from the pre-stored gas absorption spectral standard database; Based on the overall intensity level of the main absorption spectral signal, the amplitude of the standard absorption spectral profile is scaled to generate a hypothetical absorption profile. The assumed absorption profile is subtracted point by point from the spectral intensity sequence of the main absorption spectral signal within the preset wavelength range, and the standard deviation of the remaining sequence is calculated. The optimal initial profile is obtained by iteratively adjusting the scaling factor of the hypothetical absorption profile to minimize the standard deviation of the remaining sequence.

[0010] As a further aspect of the present invention, after subtracting the estimated noise signal from the main absorption spectral signal to obtain the preliminary noise-cancelled spectral signal, the method further includes a step of suppressing residual noise in the preliminary noise-cancelled spectral signal: The spectral signal after preliminary noise cancellation is subjected to wavelet transform to decompose it into subband coefficients of different scales; Calculate the statistical variance of the sub-band coefficients at each scale, and set a threshold for the scale based on the statistical variance; A soft thresholding function is used to process the sub-band coefficients at each scale, shrinking coefficients with absolute values ​​below the threshold to zero, and retaining or reducing coefficients with absolute values ​​above the threshold. After thresholding, the sub-band coefficients at each scale are subjected to inverse wavelet transform to reconstruct a spectral signal with a cleaner spectrum after noise cancellation, which is used as the source signal for extracting the characteristic absorption spectrum of the target gas.

[0011] As a further aspect of the present invention, based on the characteristic absorption spectrum of the target gas and in conjunction with a pre-stored standard database of gas absorption spectra, the concentration value of a specific gas in the exhaust gas emitted from the semiconductor process chamber is calculated, including: From the purer, noise-cancelled spectral signal, locate the wavelength range in which the known characteristic absorption peaks of the specific gas are located; Within the wavelength range, the spectrum signal after noise cancellation is baseline corrected to eliminate background tilt caused by optical path scattering or non-specific absorption. Extract the integrated absorbance or peak height of the characteristic absorption peak from the baseline-corrected spectrum; According to Beer-Lambert's law, the concentration value of the specific gas is calculated by dividing the integrated absorbance or peak height of the characteristic absorption peak by the product of the molar absorptivity and optical path length of the specific gas stored in the gas absorption spectrum standard database.

[0012] As a further aspect of the present invention, the method further includes the step of performing online drift correction on the main infrared spectral sensor and the reference infrared spectral sensor: A standard gas calibration cell of known concentration is periodically inserted into the optical path between the main infrared spectral sensor and the reference infrared spectral sensor; With the standard gas calibration cell inserted, the standard response spectral signals output by the main infrared spectral sensor and the reference infrared spectral sensor are acquired respectively. The standard response spectral signal is compared point by point with the pre-stored ideal standard spectral signal for the gas in the standard gas calibration cell; Calculate the ratio of the responsivity of the standard response spectral signal to that of the ideal standard spectral signal at each wavelength point, and generate a wavelength-responsivity correction curve; In the subsequent exhaust gas concentration detection process, the main absorption spectrum signal and the reference background spectrum signal acquired in real time are divided by their corresponding wavelength-response correction curves to compensate for the sensitivity drift of the sensor over time and with the environment.

[0013] As a further aspect of the present invention, the ratio of the responsivity of the standard response spectral signal to that of the ideal standard spectral signal at each wavelength point is calculated to generate a wavelength-responsivity correction curve, including: For the main infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured main calibration spectrum; For the reference infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured reference calibration spectrum; From the pre-stored gas absorption spectrum standard database, the theoretical transmittance spectrum of the gas in the standard gas calibration cell under the same measurement conditions is read, and the theoretical transmittance spectrum is multiplied by the ideal light source spectrum to obtain the ideal standard spectrum signal. The wavelength-responsivity ratio sequence of the main sensor is obtained by dividing the measured main calibration spectrum by the ideal standard spectrum signal wavelength by wavelength. The wavelength-responsivity ratio sequence of the reference sensor is obtained by dividing the measured reference calibration spectrum by the ideal standard spectral signal wavelength by wavelength. The wavelength-responsivity ratio sequences of the main sensor and the reference sensor are respectively subjected to smoothing filtering to generate smooth and continuous wavelength-responsivity correction curves for the main sensor and the reference sensor.

[0014] As a further aspect of the present invention, the method further includes a feedback sampling parameter adaptive adjustment step based on the concentration calculation results: The time series of the concentration values ​​of the specific gas calculated in real time is monitored, and its short-term fluctuation variance is calculated. When the short-term fluctuation variance exceeds a preset stability threshold, it is determined that there is interference in the current detection. In response to the determination of interference, the sampling integration time of the main infrared spectral sensor and the reference infrared spectral sensor is dynamically adjusted, and the length of the sliding time window used to calculate the noise correlation matrix in the improved adaptive filtering algorithm is adjusted synchronously. Under the new sampling integration time and sliding time window length, the spectral signal is reacquired and processed, and the concentration value of the specific gas is recalculated until its short-term fluctuation variance is lower than the stability threshold.

[0015] As a further aspect of the present invention, the step of dynamically adjusting the sampling integration time of the main infrared spectral sensor and the reference infrared spectral sensor in response to determining the presence of interference includes: Based on the degree to which the short-term fluctuation variance exceeds the stability threshold, the recommended integral time adjustment amount is found from the preset integral time-variance lookup table; Increase or decrease the integration time of the main infrared spectral sensor and the reference infrared spectral sensor by the integration time adjustment amount, but ensure that the adjusted integration time remains within the sensor's allowable operating range; After adjusting the integration time, wait for the sensor output signal to stabilize and record the new sampling period; The number of sampling points included in the sliding time window in the improved adaptive filtering algorithm is scaled according to the ratio of the old and new integration times to keep the physical time length covered by the sliding time window basically constant, thereby ensuring the temporal representativeness of the statistical characteristics of the noise correlation matrix.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A reference infrared spectral sensor with identical structural parameters is assembled near the main infrared spectral sensor. The reference infrared spectral sensor can simultaneously capture noise signals generated by environmental factors and the sensor's own device characteristics. The acquired reference background spectral signal can completely correspond to the common-mode noise signal contained in the main absorption spectral signal. The noise source, fluctuation amplitude, and variation law of the main absorption spectral signal and the reference background spectral signal are highly consistent, and the degree of noise feature matching between the signals is improved. This avoids the problem of the reference signal and the actual noise signal not corresponding due to inconsistent sensor structural parameters, and provides a matching reference for the subsequent accurate processing of noise signals. This allows the signal processing stage to accurately correspond to the noise signal characteristics to be canceled, reduces the interference of irrelevant signals on the noise processing process, and maintains the consistency between the noise signal reference and the actual noise signal.

[0017] An improved adaptive filtering algorithm is used to process the main absorption spectrum signal and the reference background spectrum signal. The algorithm obtains the real-time correlation characteristics of the noise signal by calculating the noise correlation matrix in real time. Based on the calculated noise correlation matrix value, the internal weight coefficients of the filter are dynamically adjusted. The filter parameters can be synchronously adapted to the real-time changes of the noise signal, thus overcoming the limitation that fixed filter weight coefficients cannot adapt to real-time noise changes. During the filtering process, common-mode noise in the main absorption spectrum signal can be specifically canceled. The processed signal can completely retain the characteristic absorption spectrum information of the target gas, reduce the interference and coverage of the characteristic spectrum information by the noise signal, and improve the purity of the characteristic absorption spectrum signal. When performing concentration calculations in conjunction with a pre-stored gas absorption spectrum standard database, the spectral signal on which the calculation is based is closer to the real spectral state of the waste gas to be measured, making the calculated results of specific gas concentrations in the waste gas more consistent with the actual emission situation and reducing the deviation of the concentration values. Attached Figure Description

[0018] Figure 1 This is a flowchart of a semiconductor exhaust gas concentration detection method based on noise cancellation according to the present invention; Figure 2 The flowchart for synchronous triggering and spectral signal acquisition; Figure 3 A flowchart for estimating the initial profile of the target gas absorption components. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a method for detecting semiconductor exhaust gas concentration based on noise cancellation, the specific method including: A primary infrared spectral sensor is installed along the main exhaust path of the semiconductor process chamber, and a reference infrared spectral sensor with identical structural parameters is installed adjacent to the primary infrared spectral sensor. The reference infrared spectral sensor is used to monitor common-mode noise signals introduced by the environment and the sensor itself. The primary and reference infrared spectral sensors are synchronously triggered according to a preset sampling sequence to acquire the primary absorption spectrum signal and the reference background spectrum signal of the exhaust gas to be measured, respectively. An improved adaptive filtering algorithm is used to process the primary absorption spectrum signal and the reference background spectrum signal. The improved adaptive filtering algorithm adjusts its internal filter weight coefficients based on the real-time calculated noise correlation matrix. The noise-cancelled characteristic absorption spectrum of the target gas is extracted from the signal processed by the improved adaptive filtering algorithm. Based on the characteristic absorption spectrum of the target gas and combined with a pre-stored gas absorption spectrum standard database, the concentration value of a specific gas in the exhaust gas emitted from the semiconductor process chamber is calculated. The tap weight coefficients of the finite impulse response filter in the improved adaptive filtering algorithm are dynamically adjusted according to the eigenvalues ​​and eigenvectors of the noise correlation matrix. Specifically, the noise correlation matrix constructed within the sliding time window is... Its dimensions are , This represents the number of taps in the finite impulse response filter. Perform eigenvalue decomposition:

[0022] in: It is an eigenvalue diagonal matrix. For the first The eigenvalue reflects the difference between the main noise and the reference noise at the th . Correlation strength in each characteristic direction; For the corresponding eigenvector matrix, each column vector forms an orthogonal basis for the noise space. Based on the eigenvalues... The distribution determines the step size for adjusting the weight coefficients in each characteristic direction:

[0023] in: This is the global step size adjustment factor. This is a regularization parameter used to prevent step size divergence when eigenvalues ​​are too small. The physical meaning of this formula is: for eigendirections with strong correlation (large eigenvalues), a smaller adjustment step size is used to maintain stable convergence of the filter; for eigendirections with weak correlation (small eigenvalues), a larger adjustment step size is used to accelerate the filter's tracking speed of changes in noise statistical characteristics. The tap weight coefficient vector of a finite impulse response filter. Update as follows:

[0024] in: The error signal between the desired signal and the filter output. The input vector is the reference background spectral signal at the current time. The improved algorithm performs eigenvalue decomposition on the noise correlation matrix and assigns differentiated adjustment step sizes to different feature directions. This aligns the adjustment direction of the filter weight coefficients with the direction of the principal components of the noise statistical characteristics, thereby achieving better steady-state noise cancellation while maintaining fast tracking capability.

[0025] In one embodiment of the present invention, see [reference] Figure 2The system sends start acquisition commands with the same timestamp to both the main infrared spectral sensor and the reference infrared spectral sensor. Within the same integration time, the main infrared spectral sensor receives an infrared beam of exhaust gas flowing through the main exhaust emission path and generates a main absorption spectral signal containing exhaust gas characteristic absorption and background noise. Within the same integration time, the reference infrared spectral sensor receives a reference beam of exhaust gas that does not flow through the exhaust gas, originating from the same infrared source but attenuated by a neutral density filter, and generates a reference background spectral signal that mainly contains environmental and sensor background noise. The main absorption spectral signal and the reference background spectral signal are aligned according to wavelength channels and stored in a data buffer corresponding to the same timestamp. From the reference background spectral signal, a spectral intensity sequence within a preset wavelength range is extracted as a reference noise vector. From the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range, the initial contour of the target gas absorption component is estimated, and the initial contour is subtracted from the main signal to obtain the main noise vector. The cross-correlation function between the reference noise vector and the main noise vector within a sliding time window is calculated, and a noise correlation matrix is ​​constructed based on the cross-correlation function. According to the eigenvalues ​​and eigenvectors of the noise correlation matrix, the tap weight coefficients of the finite impulse response filter in the improved adaptive filtering algorithm are dynamically adjusted so that the frequency response of the finite impulse response filter matches the noise statistical characteristics at the current moment. The finite impulse response filter with adjusted tap weight coefficients is used to filter the reference background spectral signal to generate an estimated noise signal. The estimated noise signal is subtracted from the main absorption spectral signal to obtain the spectral signal after preliminary noise cancellation.

[0026] In practice, a start-up acquisition command with the same timestamp is sent to both the main infrared spectral sensor and the reference infrared spectral sensor. This command is issued by the central control unit, and the timestamp is accurate to the microsecond level to ensure synchronization. Within the same integration time, the main infrared spectral sensor receives an infrared beam of light passing through the main exhaust gas path. This infrared beam is generated by a broadband infrared light source. After penetrating the exhaust gas, it carries the characteristic absorption information of the target gas and background noise introduced by factors such as ambient temperature fluctuations and thermal radiation from optical elements, generating a main absorption spectrum signal containing the characteristic absorption of the exhaust gas and background noise. Within the same integration time, the reference infrared spectral sensor receives a reference beam from the same infrared light source that does not pass through the flowing exhaust gas but is attenuated by a neutral density filter. The intensity of the reference beam is adjusted to be similar to the intensity of the main beam after passing through the exhaust gas to avoid sensor saturation, generating a reference background spectrum signal that mainly contains background noise from the environment and the sensor itself. The main absorption spectrum signal and the reference background spectrum signal are aligned one-to-one according to wavelength channels, with wavelength alignment accuracy controlled within 0.1 nanometers, and stored in a data buffer corresponding to the same timestamp to provide matching data pairs for subsequent signal processing.

[0027] In some embodiments, a spectral intensity sequence within a preset wavelength range is extracted from the reference background spectral signal as a reference noise vector. This preset wavelength range is selected as the region where the characteristic absorption peaks of the target gas are weak, so as to reduce the interference of the target gas signal on noise estimation. An initial profile of the target gas absorption component is estimated from the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range. Specifically, a rough fit is performed based on the amplitude of the main absorption spectral signal and the known absorption line shape of the target gas, and the initial profile is subtracted from the main signal to remove most of the target gas components, thereby obtaining the main noise vector that mainly reflects the noise.

[0028] The cross-correlation function between the reference noise vector and the main noise vector within a sliding time window is calculated. The length of the sliding time window is set to cover at least several dozen sampling periods to capture the dynamic statistical characteristics of the noise. The calculation of the cross-correlation function reflects the similarity between the two noise signals under different time delays. Based on the cross-correlation function, a noise correlation matrix is ​​constructed, which describes the statistical correlation strength and time delay structure between the reference noise and the main noise. According to the eigenvalues ​​and eigenvectors of the noise correlation matrix, the tap weight coefficients of the finite impulse response filter in the improved self-adaptive filtering algorithm are dynamically adjusted so that the frequency response of the finite impulse response filter can track the noise statistical characteristics at the current moment, such as enhancing the suppression capability of high correlation frequency bands. The finite impulse response filter with adjusted tap weight coefficients is used to filter the reference background spectral signal to generate an estimated noise signal. This estimated noise signal is highly close to the common-mode noise mixed into the main absorption spectral signal in terms of waveform and spectrum. The estimated noise signal is subtracted from the main absorption spectral signal to cancel out most of the common-mode noise, resulting in a preliminary noise-cancelled spectral signal, in which the absorption characteristics of the target gas are highlighted.

[0029] In the example scenario, when the semiconductor process chamber is undergoing an etching process, the exhaust gas contains fluoride gas, whose infrared absorption peak is located in a specific band. The main absorption spectrum signal shows a significant absorption dip in this band, but at the same time, it is superimposed with slow drift noise caused by the periodic temperature change of the chamber exhaust pipe. The reference background spectrum signal mainly records this temperature drift noise. After the above adaptive filtering process, the temperature drift noise in the main absorption spectrum signal is significantly suppressed, and the outline of the fluoride absorption peak becomes clearer and sharper. Compared with the unprocessed original signal, the noise floor is reduced by about an order of magnitude, effectively improving the ability to identify weak absorption features.

[0030] It is understandable that in another scenario, if the process chamber is in standby mode and the exhaust gas flow rate is extremely low, both the main absorption spectrum signal and the reference background spectrum signal are mainly composed of random electrical noise and environmental background radiation, and the two are highly correlated. The adaptive filtering algorithm can accurately estimate and subtract this common-mode noise, making the processed spectrum signal more stable and reducing the probability of false alarms. Optionally, the noise correlation matrix can be constructed using the following discretization calculation formula:

[0031] in: Indicates time index and Location, time delay The elements of the cross-correlation matrix, The reference noise vector is at the th The amplitude of each sampling point The main noise vector is represented at the th The amplitude of each sampling point The summation operation represents the total number of sampling points contained within the sliding time window. This matrix reflects the statistical average within the finite time window and is used to guide the update direction and step size of the filter weight coefficients.

[0032] In one embodiment of the present invention, see [reference] Figure 3 The process involves: acquiring the standard absorption spectral profiles of the target gas within a preset wavelength range from a pre-stored gas absorption spectral standard database; scaling the amplitude of the standard absorption spectral profile based on the overall intensity level of the main absorption spectral signal to generate a hypothetical absorption profile; subtracting the hypothetical absorption profile point-by-point from the spectral intensity sequence of the main absorption spectral signal within the preset wavelength range and calculating the standard deviation of the remaining sequence; iteratively adjusting the scaling factor of the hypothetical absorption profile to minimize the standard deviation of the remaining sequence, thus obtaining the optimal initial profile; performing wavelet transform on the spectral signal after preliminary noise cancellation to decompose it into sub-band coefficients at different scales; calculating the statistical variance of the sub-band coefficients at each scale and setting a threshold for the scale based on the statistical variance; using a soft thresholding function to process the sub-band coefficients at each scale, shrinking coefficients with absolute values ​​below the threshold to zero and retaining or reducing coefficients with absolute values ​​above the threshold; performing inverse wavelet transform on the sub-band coefficients at each scale after thresholding to reconstruct a more spectrally pure noise-cancelled spectral signal, which serves as the source signal for extracting the characteristic absorption spectrum of the target gas.

[0033] In practice, the standard absorption spectral profile of the target gas within a preset wavelength range is obtained from a pre-stored gas absorption spectral standard database. This profile is stored in the form of high-resolution spectral data, containing the line intensity and line shape information of the target gas at specific temperatures and pressures. Based on the overall intensity level of the main absorption spectral signal, i.e., the average intensity of the signal across the entire wavelength range or the baseline intensity of a specific band, the amplitude of the standard absorption spectral profile is linearly scaled to generate a hypothetical absorption profile. This profile is consistent with the standard spectral line in shape but adapted to the current measurement conditions in amplitude. The hypothetical absorption profile is subtracted point by point from the spectral intensity sequence of the main absorption spectral signal within the preset wavelength range to obtain a set of residual sequences. Theoretically, this sequence should mainly contain noise and trace amounts of gas signals that have not been completely subtracted. The standard deviation of the residual sequences is calculated to measure the magnitude of the fitting residual between the hypothetical profile and the real signal. The scaling factor of the hypothetical absorption profile is iteratively adjusted, and the standard deviation of the residual sequences is recalculated after each adjustment. A numerical optimization method is used to find the scaling factor value that minimizes the standard deviation, thereby obtaining the optimal initial profile. This profile represents the best preliminary estimate of the target gas composition before the main noise vector is extracted.

[0034] In some embodiments, wavelet transform is performed on the spectral signal after preliminary noise cancellation. Compactly supported orthogonal wavelet basis functions are selected to decompose the signal into sub-band coefficients at different scales. Coarse-scale coefficients correspond to the slow variation trend and baseline components of the signal, while fine-scale coefficients correspond to high-frequency noise and residual interference. The statistical variance of the sub-band coefficients at each scale is calculated. The statistical variance reflects the degree of fluctuation in signal energy within that frequency band. A threshold for each scale is set based on the statistical variance. Higher thresholds are set for scales with larger variances to retain more signal details, while lower thresholds are set for scales with smaller variances to suppress weak noise. A soft... The threshold function processes the sub-band coefficients at each scale. The soft threshold function subtracts the threshold from the absolute value of the coefficient. If the result is positive, the sign is retained; otherwise, it is set to zero. This shrinks coefficients with absolute values ​​below the threshold to zero and retains or reduces coefficients with absolute values ​​above the threshold, effectively separating useful signal components from random noise. The sub-band coefficients at each scale after thresholding are subjected to inverse wavelet transform. The processed coefficients are used to reconstruct the signal, resulting in a more spectrally pure signal after noise cancellation. The high-frequency spikes and low-frequency drift remaining in this signal are further suppressed, serving as the source signal for extracting the characteristic absorption spectrum of the target gas.

[0035] In the example scenario, when processing the spectrum of exhaust gas containing trace amounts of silane, after initial noise cancellation following adaptive filtering, the spectral signal still exhibits slight oscillations near the characteristic absorption peak of silane. After wavelet transform decomposition to four scales, the fine-scale coefficients show high-frequency fluctuations with small variance. After applying a lower threshold, most of the fine-scale coefficients are set to zero, while the coarse-scale coefficients retain the overall envelope of the absorption peak. After soft thresholding and inverse transform reconstruction, the oscillation amplitude near the absorption peak is significantly reduced, and the absorption peak shape becomes smoother and more continuous, which is beneficial for subsequent accurate calculation of integrated absorbance.

[0036] It is understandable that if multiple interfering gases exist in the exhaust gas and their absorption spectra partially overlap with the target gas, the multi-scale analysis capability of wavelet transform can reveal subtle differences in the overlapping spectra at different scales. By setting differentiated thresholds for different scales, the characteristic structures of different gases can be relatively preserved while suppressing background fluctuations, thus improving the robustness of extracting target gas features in complex backgrounds. Optionally, the threshold for each scale... The coefficients of the sub-band at this scale can be determined using the following formula based on their statistical characteristics:

[0037] in: Indicates the first Threshold for each wavelet decomposition scale It is a modulating factor used to control the aggressiveness of noise suppression. It is the first Statistical standard deviation of scale subband coefficients It is the total number of sub-band coefficients at this scale, and the logarithmic term plays an adaptive adjustment role, so that the threshold changes dynamically with the signal length and noise level.

[0038] In one embodiment of the present invention, the wavelength range in which the known characteristic absorption peak of a specific gas is located is identified from the noise-cancelled spectral signal with a cleaner spectrum; within the wavelength range, baseline correction is performed on the noise-cancelled spectral signal with a cleaner spectrum to eliminate background tilt caused by optical path scattering or non-specific absorption; the integrated absorbance or peak height of the characteristic absorption peak is extracted from the baseline-corrected spectrum; according to the Beer-Lambert law, the integrated absorbance or peak height of the characteristic absorption peak is divided by the product of the molar absorptivity and optical path length of the specific gas stored in the gas absorption spectral standard database to calculate the concentration value of the specific gas.

[0039] In practice, from the noise-cancelled spectral signal with a cleaner spectrum, the wavelength range containing the known characteristic absorption peaks of a specific gas is located. This range is determined based on the center wavelength and full width at half maximum (FWHM) of the characteristic absorption band of the target gas in a pre-stored gas absorption spectrum standard database, covering the main contour range of the absorption peaks. Within this wavelength range, baseline correction is performed on the noise-cancelled spectral signal with a cleaner spectrum. A polynomial fitting method is used to simulate the background tilt caused by light path scattering or non-specific absorption, and the fitted baseline is subtracted from the original spectrum to eliminate low-frequency background shift, making the absorption peaks stand out relative to the flat baseline. In the baseline-corrected spectrum, the integrated absorbance of the characteristic absorption peak is extracted. This is obtained by numerically integrating the absorbance values ​​within the wavelength range of the absorption peak, or by extracting the peak height of the characteristic absorption peak, which is the difference between the absorbance at the peak and the baseline absorbance. According to Beer-Lambert's law, the integrated absorbance or peak height of the characteristic absorption peak is divided by the product of the molar absorptivity and optical path length of the specific gas stored in the gas absorption spectrum standard database to calculate the concentration value of the specific gas. The molar absorptivity is a constant calibrated in advance under standard conditions, and the optical path length is the physical length of the infrared beam passing through the exhaust gas path.

[0040] In some embodiments, when processing the characteristic absorption peaks of nitrogen oxide gases, a wavelength range of 5.2 to 5.4 micrometers is selected. This range covers the asymmetric stretching vibration absorption band of NO2 molecules. Baseline correction employs second-order polynomial fitting, effectively eliminating the background rise caused by scattering from dust particles. After extracting the integrated absorbance, the molar absorptivity of NO2 and the system-set optical path length of 0.5 meters are substituted into the calculation to obtain a stable concentration output. Referring to Table 1, a comparison is shown between the concentration values ​​calculated based on the integrated absorbance of the characteristic absorption peaks and the preset reference values ​​for three different concentration levels of hydrogen fluoride gas in the example scenario, reflecting the effectiveness of the calculation method.

[0041] Table 1: Calculation Results of Hydrogen Fluoride Gas Concentration 10.0 0.125 10.02 25.0 0.312 24.98 50.0 0.623 49.95 It is understandable that if the target gas is methane, whose characteristic absorption peak is located around 3.3 micrometers, after baseline correction, the peak height can be directly used for calculation. The resulting concentration value is consistent with the calculation result based on integrated absorbance within the error range, proving the interchangeability of the two extraction methods. Optionally, the Beer-Lambert law used in the concentration calculation can be expressed in the following form:

[0042] in: This represents the concentration of a specific gas to be determined. This represents the integrated absorbance or peak height of the characteristic absorption peak extracted from the baseline-corrected spectrum. This represents the molar absorptivity of a specific gas at a corresponding wavelength, stored in the gas absorption spectroscopy standard database. This indicates the effective optical path length of the infrared beam as it passes through the main path of the exhaust gas emission.

[0043] In one embodiment of the present invention, a standard gas calibration cell of known concentration is periodically inserted into the optical paths of the main infrared spectral sensor and the reference infrared spectral sensor. While the standard gas calibration cell is inserted, the standard response spectral signals output by the main infrared spectral sensor and the reference infrared spectral sensor are acquired. The standard response spectral signals are compared point-by-point with pre-stored ideal standard spectral signals for the gas in the standard gas calibration cell. The responsivity ratio of the standard response spectral signal to the ideal standard spectral signal at each wavelength point is calculated to generate a wavelength-responsivity correction curve. In subsequent exhaust gas concentration detection, the real-time acquired main absorption spectral signal and the reference background spectral signal are divided by their corresponding wavelength-responsivity correction curves to compensate for the sensitivity drift of the sensors over time and due to environmental factors. For the main infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured main calibration spectrum; for the reference infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured reference calibration spectrum. From the pre-stored gas absorption spectrum standard database, the theoretical transmittance spectrum of the gas in the standard gas calibration cell under the same measurement conditions is read. The theoretical transmittance spectrum is multiplied by the ideal light source spectrum to obtain the ideal standard spectral signal. The measured main calibration spectrum is divided by the ideal standard spectral signal wavelength by wavelength to obtain the wavelength-responsivity ratio sequence of the main sensor. The measured reference calibration spectrum is divided by the ideal standard spectral signal wavelength by wavelength to obtain the wavelength-responsivity ratio sequence of the reference sensor. Smoothing filtering is applied to the wavelength-responsivity ratio sequences of both the main sensor and the reference sensor to generate smooth and continuous wavelength-responsivity correction curves for both the main sensor and the reference sensor.

[0044] In practice, a standard gas calibration cell of known concentration is periodically inserted into the optical paths of the main infrared spectral sensor and the reference infrared spectral sensor. This calibration cell contains carbon dioxide gas of constant concentration to simulate a stable absorption reference. With the standard gas calibration cell inserted, the standard response spectral signals output by the main infrared spectral sensor and the reference infrared spectral sensor are acquired. At this time, the spectral signal reflects the comprehensive response characteristics of the sensor in the current state. The standard response spectral signal is compared point by point with the pre-stored ideal standard spectral signal for the gas in the standard gas calibration cell. The ideal standard spectral signal is a distortion-free reference generated based on theoretical transmittance and an ideal light source model. The responsivity ratio of the standard response spectral signal and the ideal standard spectral signal at each wavelength point is calculated to generate a wavelength-responsivity correction curve. This curve quantifies the sensitivity deviation of the sensor at each wavelength. In the subsequent exhaust gas concentration detection process, the real-time acquired main absorption spectral signal and the reference background spectral signal are divided by their corresponding wavelength-responsivity correction curves to achieve gain compensation for the signal at each wavelength point, thereby correcting the sensitivity decrease or shift caused by sensor aging or changes in ambient temperature and humidity.

[0045] In some embodiments, for the primary infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured primary calibration spectrum. This spectrum is the average of multiple samples to suppress random noise. For the reference infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured reference calibration spectrum. It is also averaged to improve the signal-to-noise ratio. The theoretical transmittance spectrum of the gas in the standard gas calibration cell under the same measurement conditions is read from a pre-stored gas absorption spectral standard database. This spectral data includes the absorption line parameters of carbon dioxide at standard temperature and pressure. The theoretical transmittance spectrum is multiplied by the ideal light source spectrum to obtain the ideal standard spectral signal, which represents the perfect sensing... The expected output under the conditions of the instrument and light source is obtained by dividing the measured main calibration spectrum with the ideal standard spectrum signal wavelength by wavelength to obtain the wavelength-responsivity ratio sequence of the main sensor. A ratio greater than 1 indicates that the sensor has a high responsivity at that wavelength, while a ratio less than 1 indicates that the responsivity is insufficient. The measured reference calibration spectrum is divided with the ideal standard spectrum signal wavelength by wavelength to obtain the wavelength-responsivity ratio sequence of the reference sensor. The wavelength-responsivity ratio sequences of the main sensor and the reference sensor are smoothed and filtered respectively. The moving average method is used to remove high-frequency fluctuations, generating smooth and continuous wavelength-responsivity correction curves for the main sensor and the reference sensor, ensuring that the corrected spectrum does not introduce additional high-frequency noise.

[0046] In the example scenario, after the system has run for 500 hours, an online drift correction is performed. A carbon dioxide standard gas calibration cell is inserted, and the response spectra of the main sensor and the reference sensor are collected and compared with the ideal spectrum to generate a correction curve. Referring to Table 2, the responsivity ratio at three typical wavelength points and the changes in spectral intensity before and after correction are listed. It can be seen that the correction effectively eliminates the response attenuation in the long-wavelength region caused by detector aging.

[0047] Table 2: Main Sensor Wavelength-Responsivity Correction Table 4.20 12.45 13.00 1.044 12.42 4.35 11.23 11.80 1.051 11.21 4.60 09.87 10.55 1.069 09.85 It is understandable that if the short-wavelength responsivity of the reference sensor decreases due to long-term exposure to harsh environments, generating a wavelength-responsivity correction curve specifically for the reference sensor using the same correction process, and compensating for the reference background spectral signal in subsequent processing, can ensure the authenticity of the reference signal, thereby improving the accuracy of noise estimation in adaptive filtering. Optionally, the smoothing of the wavelength-responsivity correction curve can be achieved using a weighted moving average formula:

[0048] in: Represents the smoothed first... The ratio of responsivity at each wavelength point Represents the original first The ratio of responsivity at each wavelength point They are symmetric weighting coefficients that satisfy the normalization condition. It is the half-width of the smooth window, which determines the degree of smoothing.

[0049] In one embodiment of the present invention, the time series of the concentration values ​​of a specific gas calculated in real time is monitored, and its short-term fluctuation variance is calculated; when the short-term fluctuation variance exceeds a preset stability threshold, it is determined that there is interference in the current detection; in response to the determination that there is interference, the sampling integration time of the main infrared spectral sensor and the reference infrared spectral sensor is dynamically adjusted, and the length of the sliding time window used to calculate the noise correlation matrix in the improved adaptive filtering algorithm is adjusted synchronously; under the new sampling integration time and sliding time window length, the spectral signal is reacquired and processed, and the concentration value of the specific gas is recalculated until its short-term fluctuation variance is lower than the stability threshold. Based on the degree to which the short-term fluctuation variance exceeds the stability threshold, the recommended integration time adjustment amount is found from the preset integration time-variance lookup table; the current integration time of the main infrared spectral sensor and the reference infrared spectral sensor is increased or decreased by the integration time adjustment amount, but it is ensured that the adjusted integration time remains within the sensor's allowable operating range; after adjusting the integration time, wait for the sensor output signal to stabilize and record the new sampling period; the number of sampling points included in the sliding time window in the improved adaptive filtering algorithm is scaled according to the ratio of the old and new integration times to keep the physical time length covered by the sliding time window basically constant, thereby ensuring the temporal representativeness of the statistical characteristics of the noise correlation matrix.

[0050] In practical implementation, the time series of concentration values ​​of a specific gas calculated in real time is monitored. The time series consists of concentration values ​​output from multiple consecutive sampling periods arranged in chronological order. Its short-term fluctuation variance is calculated, characterized by the mean of the sum of squared deviations of concentration values ​​over a fixed historical time period from their arithmetic mean. When the short-term fluctuation variance exceeds a preset stability threshold (set based on measurement accuracy requirements and variance statistics under historical stable operating conditions), interference is determined to exist. Interference may originate from sudden changes in exhaust gas flow, changes in background components, or optical window contamination. In response to the determination of interference, the sampling parameters of the main infrared spectral sensor and the reference infrared spectral sensor are dynamically adjusted. The sampling integration time directly affects the sensor's signal-to-noise ratio and response speed. Simultaneously, the length of the sliding time window used to calculate the noise correlation matrix in the improved adaptive filtering algorithm is adjusted. The length of the sliding time window determines the number of samples used for noise statistical analysis. Under the new sampling integration time and sliding time window length, the spectral signal is reacquired and processed, including triggering the sensor to acquire new data, running adaptive filtering and wavelet denoising steps, and recalculating the concentration value of the specific gas. The newly calculated concentration value is added to the time series, and the short-term fluctuation variance is recalculated. This process is repeated until the short-term fluctuation variance is lower than the stability threshold, indicating that the system has returned to a stable measurement state.

[0051] In some embodiments, the degree to which the short-term fluctuation variance exceeds the stability threshold is divided into several levels. The corresponding recommended integration time adjustment amount is found from a preset integration time-variance lookup table. The integration time-variance lookup table is established by testing the optimal integration time under different interference intensities during the system calibration phase. The integration time adjustment amount is increased or decreased for the current integration time of the main infrared spectral sensor and the reference infrared spectral sensor. The integration time adjustment follows the monotonicity principle, that is, the larger the variance, the larger the adjustment magnitude. However, it is ensured that the adjusted integration time is kept within the minimum and maximum integration time operating range allowed by the sensor to prevent sensor saturation or slow response. After adjusting the integration time, the sensor output signal is waited for to stabilize. The stabilization waiting time is at least several times the new sampling period to ensure the end of the transient process. The new sampling period is recorded for time alignment in subsequent data processing.

[0052] In some embodiments, the number of sampling points included in the sliding time window in the improved adaptive filtering algorithm is scaled according to the ratio of the old and new integration times. The scaling relationship is a linear ratio to keep the physical time length covered by the sliding time window basically constant, thereby ensuring the time representativeness of the statistical characteristics of the noise correlation matrix and avoiding distortion of the noise statistical characteristics due to changes in the sampling interval.

[0053] It is understandable that when the concentration of the target gas in the exhaust gas fluctuates rapidly, the short-term fluctuation variance will increase. The system improves the signal-to-noise ratio of a single measurement by extending the integration time, and at the same time expands the number of sampling points in the sliding window to maintain the statistical time span, which helps to accurately estimate the noise characteristics and effectively cancel them under dynamic operating conditions. Optionally, the scaling of the number of sampling points in the sliding time window is calculated using the following formula:

[0054] in: This indicates the number of sampling points that the adjusted sliding time window should include. This indicates the original number of sampling points in the sliding time window before adjustment. This indicates the original integration time of the primary infrared spectral sensor and the reference infrared spectral sensor before adjustment. This indicates the new integration time between the adjusted master infrared spectral sensor and the reference infrared spectral sensor, denoted by [symbol]. This indicates a floor operation, ensuring that the number of sampling points is an integer.

[0055] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for detecting semiconductor exhaust gas concentration based on noise cancellation, characterized in that, The method includes: A main infrared spectral sensor is installed on the main exhaust path of the semiconductor process chamber, and a reference infrared spectral sensor with the same structural parameters is installed in the vicinity of the main infrared spectral sensor. The reference infrared spectral sensor is used to monitor the common-mode noise signal introduced by the environment and the sensor itself. By using a preset sampling sequence, the main infrared spectral sensor and the reference infrared spectral sensor are synchronously triggered to collect the main absorption spectral signal and the reference background spectral signal of the exhaust gas to be tested, respectively. An improved adaptive filtering algorithm is used to process the main absorption spectrum signal and the reference background spectrum signal. The improved adaptive filtering algorithm adjusts its internal filter weight coefficients based on the noise correlation matrix calculated in real time. The noise-cancelled characteristic absorption spectrum of the target gas is extracted from the signal processed by the improved adaptive filtering algorithm. Based on the characteristic absorption spectrum of the target gas and combined with a pre-stored standard database of gas absorption spectra, the concentration of a specific gas in the exhaust gas emitted from the semiconductor process chamber is calculated.

2. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 1, characterized in that, The process involves synchronously triggering the main infrared spectral sensor and the reference infrared spectral sensor through a preset sampling sequence to acquire the main absorption spectral signal and the reference background spectral signal of the exhaust gas to be measured, respectively, including: Send start acquisition commands with the same timestamp to the main infrared spectral sensor and the reference infrared spectral sensor; Within the same integration time, the main infrared spectral sensor receives an infrared beam of the exhaust gas flowing through the main exhaust emission path and generates the main absorption spectral signal containing the characteristic absorption of the exhaust gas and background noise. Within the same integration time, the reference infrared spectral sensor receives a reference beam that does not pass through the flowing exhaust gas, originates from the same infrared source, but is attenuated by a neutral density filter, and generates the reference background spectral signal, which mainly contains background noise from the environment and the sensor itself. The main absorption spectrum signal and the reference background spectrum signal are aligned according to wavelength channels and stored in the data buffer corresponding to the same timestamp.

3. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 1, characterized in that, The improved adaptive filtering algorithm adjusts its internal filter weight coefficients based on the real-time calculated noise correlation matrix, including: From the reference background spectral signal, extract the spectral intensity sequence within a preset wavelength range as a reference noise vector; From the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range, the initial profile of the target gas absorption component is estimated, and the initial profile is subtracted from the main signal to obtain the main noise vector; Calculate the cross-correlation function between the reference noise vector and the main noise vector within a sliding time window, and construct the noise correlation matrix based on the cross-correlation function; Based on the eigenvalues ​​and eigenvectors of the noise correlation matrix, the tap weight coefficients of the finite impulse response filter in the improved adaptive filtering algorithm are dynamically adjusted so that the frequency response of the finite impulse response filter matches the noise statistical characteristics at the current moment. The reference background spectral signal is filtered using the finite impulse response filter with adjusted tap weights to generate an estimated noise signal. The estimated noise signal is subtracted from the main absorption spectrum signal to obtain the preliminary noise-cancelled spectrum signal.

4. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 3, characterized in that, Estimating the initial profile of the target gas absorption component from the spectral intensity sequence of the main absorption spectral signal within the same preset wavelength range includes: Obtain the standard absorption spectral profile of the target gas within the preset wavelength range from the pre-stored gas absorption spectral standard database; Based on the overall intensity level of the main absorption spectral signal, the amplitude of the standard absorption spectral profile is scaled to generate a hypothetical absorption profile. The assumed absorption profile is subtracted point by point from the spectral intensity sequence of the main absorption spectral signal within the preset wavelength range, and the standard deviation of the remaining sequence is calculated. The optimal initial profile is obtained by iteratively adjusting the scaling factor of the hypothetical absorption profile to minimize the standard deviation of the remaining sequence.

5. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 3, characterized in that, After subtracting the estimated noise signal from the main absorption spectral signal to obtain the preliminary noise-cancelled spectral signal, the method further includes a step of suppressing residual noise in the preliminary noise-cancelled spectral signal: The spectral signal after preliminary noise cancellation is subjected to wavelet transform to decompose it into subband coefficients of different scales; Calculate the statistical variance of the sub-band coefficients at each scale, and set a threshold for the scale based on the statistical variance; A soft thresholding function is used to process the sub-band coefficients at each scale, shrinking coefficients with absolute values ​​below the threshold to zero, and retaining or reducing coefficients with absolute values ​​above the threshold. After thresholding, the sub-band coefficients at each scale are subjected to inverse wavelet transform to reconstruct a spectral signal with a cleaner spectrum after noise cancellation, which is used as the source signal for extracting the characteristic absorption spectrum of the target gas.

6. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 5, characterized in that, Based on the characteristic absorption spectrum of the target gas and combined with a pre-stored standard database of gas absorption spectra, the concentration values ​​of specific gases in the exhaust gas emitted from the semiconductor process chamber are calculated, including: From the purer, noise-cancelled spectral signal, locate the wavelength range in which the known characteristic absorption peaks of the specific gas are located; Within the wavelength range, the spectrum signal after noise cancellation is baseline corrected to eliminate background tilt caused by optical path scattering or non-specific absorption. Extract the integrated absorbance or peak height of the characteristic absorption peak from the baseline-corrected spectrum; According to Beer-Lambert's law, the concentration value of the specific gas is calculated by dividing the integrated absorbance or peak height of the characteristic absorption peak by the product of the molar absorptivity and optical path length of the specific gas stored in the gas absorption spectrum standard database.

7. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 1, characterized in that, The method further includes a step of performing online drift correction on the master infrared spectral sensor and the reference infrared spectral sensor: A standard gas calibration cell of known concentration is periodically inserted into the optical path between the main infrared spectral sensor and the reference infrared spectral sensor; With the standard gas calibration cell inserted, the standard response spectral signals output by the main infrared spectral sensor and the reference infrared spectral sensor are acquired respectively. The standard response spectral signal is compared point by point with the pre-stored ideal standard spectral signal for the gas in the standard gas calibration cell; Calculate the ratio of the responsivity of the standard response spectral signal to that of the ideal standard spectral signal at each wavelength point, and generate a wavelength-responsivity correction curve; In the subsequent exhaust gas concentration detection process, the main absorption spectrum signal and the reference background spectrum signal acquired in real time are divided by their corresponding wavelength-response correction curves to compensate for the sensitivity drift of the sensor over time and with the environment.

8. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 7, characterized in that, Calculate the ratio of the responsivity of the standard response spectral signal to that of the ideal standard spectral signal at each wavelength point, and generate a wavelength-responsivity correction curve, including: For the main infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured main calibration spectrum; For the reference infrared spectral sensor, the standard response spectral signal measured under standard gas calibration conditions is recorded as the measured reference calibration spectrum; From the pre-stored gas absorption spectrum standard database, the theoretical transmittance spectrum of the gas in the standard gas calibration cell under the same measurement conditions is read, and the theoretical transmittance spectrum is multiplied by the ideal light source spectrum to obtain the ideal standard spectrum signal. The wavelength-responsivity ratio sequence of the main sensor is obtained by dividing the measured main calibration spectrum by the ideal standard spectrum signal wavelength by wavelength. The wavelength-responsivity ratio sequence of the reference sensor is obtained by dividing the measured reference calibration spectrum by the ideal standard spectral signal wavelength by wavelength. The wavelength-responsivity ratio sequences of the main sensor and the reference sensor are respectively subjected to smoothing filtering to generate smooth and continuous wavelength-responsivity correction curves for the main sensor and the reference sensor.

9. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 1, characterized in that, The method also includes a feedback-based adaptive adjustment step for sampling parameters based on concentration calculation results: The time series of the concentration values ​​of the specific gas calculated in real time is monitored, and its short-term fluctuation variance is calculated. When the short-term fluctuation variance exceeds a preset stability threshold, it is determined that there is interference in the current detection. In response to the determination of interference, the sampling integration time of the main infrared spectral sensor and the reference infrared spectral sensor is dynamically adjusted, and the length of the sliding time window used to calculate the noise correlation matrix in the improved adaptive filtering algorithm is adjusted synchronously. Under the new sampling integration time and sliding time window length, the spectral signal is reacquired and processed, and the concentration value of the specific gas is recalculated until its short-term fluctuation variance is lower than the stability threshold.

10. The method for detecting semiconductor exhaust gas concentration based on noise cancellation according to claim 9, characterized in that, The step of dynamically adjusting the sampling integration time of the main infrared spectral sensor and the reference infrared spectral sensor in response to the determination of interference includes: Based on the degree to which the short-term fluctuation variance exceeds the stability threshold, the recommended integral time adjustment amount is found from the preset integral time-variance lookup table; Increase or decrease the integration time of the main infrared spectral sensor and the reference infrared spectral sensor by the integration time adjustment amount, but ensure that the adjusted integration time remains within the sensor's allowable operating range; After adjusting the integration time, wait for the sensor output signal to stabilize and record the new sampling period; The number of sampling points included in the sliding time window in the improved adaptive filtering algorithm is scaled according to the ratio of the old and new integration times to keep the physical time length covered by the sliding time window basically constant, thereby ensuring the temporal representativeness of the statistical characteristics of the noise correlation matrix.

Images