Method and system for determining product purity of electronic grade anhydrous hydrogen fluoride

Abstract

This invention relates to the fields of analytical chemistry and semiconductor material detection technology, and discloses a method and system for determining the purity of electronic-grade anhydrous hydrogen fluoride. The method includes: introducing the sample into a low-temperature controllable gas-phase separation chamber for gradient cooling to suppress HF association; acquiring and preprocessing broadband infrared transmission spectra; dynamically subtracting HF background absorption using a differential absorption spectroscopy decoupling algorithm; and determining the type and concentration of impurities using a multi-band cross-validated impurity fingerprint identification model. The system includes a low-temperature gas-phase separation unit, an infrared excitation and acquisition unit, a data preprocessing unit, a differential decoupling calculation unit, an impurity identification unit, and a report generation unit. This invention enables the simultaneous detection of ten metallic impurities, with a detection limit of 10. ‑9 With a mole fraction and an accuracy rate exceeding 99.5%, it meets the quality control requirements for semiconductor-grade high-purity hydrogen fluoride.

Description

Technical Field

[0001] This invention belongs to the field of analytical chemistry and semiconductor material detection technology, specifically relating to a method and system for determining the purity of electronic-grade anhydrous hydrogen fluoride. Background Technology

[0002] Electronic-grade anhydrous hydrogen fluoride is a critical cleaning and etching reagent in semiconductor manufacturing, and its purity directly affects chip yield and device performance. As integrated circuit manufacturing processes evolve towards the nanoscale, the control requirements for metallic impurities (such as sodium, potassium, iron, and aluminum) in hydrogen fluoride have increased to the ppt level. Currently, infrared spectroscopy is widely explored for impurity analysis of high-purity hydrogen fluoride due to its non-destructive nature, rapid response, and potential for online detection. This method achieves qualitative and semi-quantitative detection by identifying the characteristic absorption peaks of impurity elements at specific wavenumbers. However, electronic-grade anhydrous hydrogen fluoride itself has a highly polar molecular structure, exhibiting a broad and strong intrinsic absorption band in the mid-infrared region. Its vibrational-rotational coupling spectral lines are dense and their intensity is far higher than the weak signals of trace metallic impurities.

[0003] Impurity detection techniques based on infrared spectroscopy rely on the distinguishability of characteristic peaks and the signal-to-noise ratio. In practical applications, the main absorption peak of hydrogen fluoride molecules often overlaps significantly in the frequency domain with the secondary vibrational or lattice vibrational modes of various metal fluoride impurities. This spectral interference makes it difficult for traditional peak identification algorithms to accurately distinguish between the background signal and impurity contributions, especially in the low concentration range, where weak impurity peaks are easily submerged by the strong absorption background of HF or misjudged as baseline fluctuations. Existing techniques typically employ fixed-band subtraction or empirical baseline correction strategies, but these methods do not consider the dynamic effects of HF concentration fluctuations, temperature drift, and gas phase pressure changes on the background profile, resulting in significant residual errors after correction and a persistently high misjudgment rate.

[0004] Existing technologies for processing the infrared spectra of electronic-grade anhydrous hydrogen fluoride generally suffer from insufficient background interference suppression, poor robustness in impurity feature extraction, and low accuracy in multi-component signal coupling and resolution. Conventional deconvolution algorithms lack prior modeling of the spectral characteristics of HF molecules, making it difficult to effectively separate overlapping peaks; while matching methods relying on standard sample libraries are limited by incomplete coverage of impurity types and sensitivity to environmental conditions.

[0005] In the demanding scenarios of semiconductor production lines that require both real-time performance and accuracy, the aforementioned defects directly lead to unreliable test results, failing to meet the closed-loop control requirements of advanced processes for the purity of raw materials. Summary of the Invention

[0006] This invention provides a method and system for determining the purity of electronic-grade anhydrous hydrogen fluoride, aiming to solve the technical problem of high false positive rates for trace impurities in electronic-grade anhydrous hydrogen fluoride due to severe overlap between the HF molecule absorption peak and the characteristic absorption peak of the impurities when using existing infrared spectroscopy to determine metallic impurities. This invention achieves highly selective, highly sensitive, and low-false-positive quantitative detection of trace metallic impurities in electronic-grade anhydrous hydrogen fluoride by constructing a low-temperature controllable gas-phase separation chamber, introducing a differential absorption spectroscopy decoupling algorithm, and establishing an impurity fingerprint identification model based on multi-band cross-validation.

[0007] This invention provides a method for determining the purity of electronic-grade anhydrous hydrogen fluoride, comprising: The anhydrous hydrogen fluoride sample to be tested is introduced into a low-temperature controllable gas phase separation chamber and subjected to gradient cooling within a set temperature range to keep the sample in a gas phase state and suppress the HF molecule association effect, thereby weakening its broad absorption band intensity in the infrared band. The processed gaseous sample was irradiated with a broadband infrared light source, and the full-band transmission spectrum signal was acquired by a high-resolution Fourier transform infrared spectrometer to obtain the original spectral dataset. Baseline correction and noise suppression preprocessing are performed on the original spectral dataset to generate standardized spectral curves; Based on a pre-defined library of characteristic absorption bands for metal impurities, absorbance sequences for corresponding bands in standardized spectral curves are extracted and input into the differential absorption spectroscopy decoupling algorithm module. The differential absorption spectroscopy decoupling algorithm module constructs a dynamic background subtraction model based on the absorption spectrum shape change law of HF molecules at different temperatures, and separates the residual absorption components from the mixed absorption signal that are not from HF. The residual absorption components are cross-matched with the standard fingerprint database of metal impurities in multiple bands to calculate the confidence score of each target impurity element. When the confidence score of any impurity element exceeds a preset threshold, the impurity is determined to exist, and its concentration value is inverted based on its absorbance value in the characteristic band and the Lambert-Beer law. The output includes a purity analysis report containing the types of metal impurities and their corresponding concentrations.

[0008] Furthermore, the anhydrous hydrogen fluoride sample to be tested (electronic grade) is introduced into a low-temperature controllable gas-phase separation chamber, and a gradient cooling process is performed within a set temperature range to maintain the sample in a gaseous state and suppress the HF molecule association effect, including: The working temperature range of the low-temperature controllable gas phase separation chamber is controlled to be -80℃ to -30℃. The inner wall of the chamber is made of electropolished stainless steel and coated with magnesium fluoride anti-corrosion coating. The chamber is equipped with a temperature gradient distribution control unit to form at least three continuous temperature zones along the sample flow path, with a temperature difference of not less than 10℃ between each zone.

[0009] Furthermore, the processed gaseous sample was irradiated with a broadband infrared light source, and full-band transmission spectral signals were acquired using a high-resolution Fourier transform infrared spectrometer, including: The broadband infrared light source has a wavelength coverage range of 2.5 μm to 25 μm and a spectral resolution of 0.5 cm⁻¹. -1 The high-resolution Fourier transform infrared spectrometer is equipped with a liquid nitrogen-cooled mercury cadmium telluride detector with a signal-to-noise ratio of not less than 40,000 to 1. The optical guide tube is made of calcium fluoride crystal material, with an inner diameter of 5mm and a length of 50cm. It has vacuum-sealed flange interfaces at both ends and is wrapped with multiple layers of heat-insulating material on the outer wall.

[0010] Baseline correction and noise suppression preprocessing are performed on the original spectral dataset to generate standardized spectral curves, including: Baseline correction was performed using the asymmetric least squares method, with a penalty factor of 1000 and a smoothing parameter of 0.01. The corrected spectrum is smoothed by applying a nine-point triple moving average algorithm. The absorbance at the weak absorption peak of HF molecules at 0.4 μm was used as a reference for normalization, so that the absorbance at this reference point was kept constant at 0.05.

[0011] Furthermore, based on a pre-defined library of characteristic absorption bands for metal impurities, absorbance sequences for corresponding bands in the standardized spectral curves are extracted, including: The library of characteristic absorption bands for metal impurities includes the infrared characteristic absorption peak positions of ten target metal elements in the gas phase, namely sodium, potassium, calcium, magnesium, iron, nickel, chromium, copper, zinc and aluminum. Each element corresponds to no less than two non-adjacent independent absorption bands, and all bands avoid the main absorption peak regions of HF molecules at 3.5μm, 4.2μm and 7.8μm. The absorbance of each band was extracted using a digital bandpass filter, with the window width set to 0.2 μm.

[0012] Furthermore, the differential absorption spectroscopy decoupling algorithm module constructs a dynamic background subtraction model based on the absorption spectrum shape variation law of HF molecules at different temperatures, including: Under the same optical path conditions, infrared transmission spectra of the same gaseous sample were collected at the first and second temperature points, with a temperature difference of 15°C between the two temperature points. The absorbance difference between the two sets of spectra at each wavelength was calculated to generate a differential spectrum. Principal component analysis was used to reduce the dimension of the differential spectrum and extract a stable absorption feature vector that is independent of temperature. This feature vector was then projected back into the original spectral space to reconstruct the net absorption spectrum contributed only by the non-HF components.

[0013] Principal component analysis was used to reduce the dimensionality of the difference spectrum and extract temperature-independent stable absorption feature vectors, including: Perform principal component analysis on the difference spectrum to obtain the principal component loading matrix; The first three principal component vectors are selected to form the HF temperature response feature subspace; The normalized spectral curves are projected onto this subspace to reconstruct the HF background absorption model; The net absorption spectrum is obtained by subtracting the HF background absorption model from the standardized spectral curve.

[0014] Furthermore, the residual absorbed components are cross-matched with a standard fingerprint database of metal impurities in multiple bands to calculate the confidence score of each target impurity element, including: The standard fingerprint database stores multi-band absorbance response curves of ten target metallic impurities at different concentration gradients, covering a concentration gradient of 10. -9 Up to 10 -6 Mole fraction range; The multi-band cross-matching process uses a weighted Euclidean distance metric to check the consistency of the absorbance response of each candidate impurity element in all its associated bands. If the relative deviation of the concentration inversion results between any two bands is greater than 5%, the identification result of the impurity is rejected.

[0015] Its concentration value is retrieved by combining its absorbance value in the characteristic band with the Lambert-Beer law, including: According to the formula Concentration calculations were performed, among which... Net absorbance. This represents the molar absorptivity of the impurity at the corresponding wavelength. The effective optical path length is given by the molar absorptivity, which is taken from the experimentally calibrated temperature correction factor table. The final concentration is the weighted average of the inverted concentrations of each effective band, with the weights determined by the signal-to-noise ratio of each band.

[0016] This invention provides a product purity determination system for electronic-grade anhydrous hydrogen fluoride, comprising: The low-temperature controllable gas phase separation unit is used to receive the electronic-grade anhydrous hydrogen fluoride sample to be tested and maintain its gas phase state in a controlled low-temperature environment to suppress HF molecule association. The infrared excitation and acquisition unit includes a broadband infrared light source, a high-resolution Fourier transform infrared spectrometer, and a matching optical guide tube, which is used to irradiate gas phase samples and acquire full-band transmission spectrum signals. The data preprocessing unit is used to perform baseline correction, smoothing filtering, and normalization on the raw spectral signal; The differential absorption decoupling calculation unit is used to construct a dynamic background subtraction model based on multi-temperature point spectral data and separate non-HF absorption components. The impurity fingerprint recognition unit is embedded with a library of metal impurity characteristic absorption bands and a standard fingerprint database, which is used to perform multi-band cross-matching and confidence assessment. The concentration inversion and report generation unit is used to calculate the impurity concentration based on the identification results and output a structured purity analysis report.

[0017] Furthermore, the low-temperature controllable gas phase separation unit includes a sample inlet valve, a three-stage temperature control chamber, a temperature sensor array, and a temperature control feedback circuit. The three-stage temperature control chamber is sequentially configured as a pre-cooling zone, a steady-state separation zone, and an outlet buffer zone along the gas flow direction. Each zone is independently equipped with a Peltier cooling plate and a heating wire. The temperature control feedback circuit dynamically adjusts the power output based on the real-time readings of the temperature sensor array to ensure that the temperature fluctuation within the chamber does not exceed ±0.5℃.

[0018] Furthermore, the optical guide tube in the infrared excitation and acquisition unit is made of calcium fluoride crystal material, with an inner diameter of 5mm and a length of 50cm. Both ends are equipped with vacuum-sealed flange interfaces, and the outer wall of the guide tube is wrapped with multiple layers of heat insulation material to isolate environmental heat radiation interference.

[0019] Furthermore, the baseline correction performed by the data preprocessing unit adopts the asymmetric least squares method, the smoothing filter adopts the nine-point three-times moving average algorithm, and the normalization process uses the weak absorption peak of HF molecules at 0.4μm as the reference benchmark.

[0020] Furthermore, the differential absorption decoupling calculation unit has a built-in temperature-absorption spectrum mapping table, which is constructed by pre-calibrating the infrared absorption spectrum of pure HF gas at 5℃ intervals in the range of -80℃ to -30℃, and is used to interpolate in real time to obtain the HF background absorption model at any operating temperature.

[0021] Furthermore, the standard fingerprint database in the impurity fingerprint recognition unit stores multi-band absorbance response curves of ten target metal impurities at different concentration gradients. Each curve has been temperature-compensated and the concentration gradient covers 10... -9 Up to 10 -6 Mole fraction range.

[0022] Furthermore, the concentration inversion and report generation unit is based on the Lambert-Beer law formula. Concentration calculations were performed, among which... Net absorbance. This represents the molar absorptivity of the impurity at the corresponding wavelength. The effective optical path length is given by the molar absorptivity, which is taken from the experimentally calibrated temperature correction factor table.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing a low-temperature controllable gas phase separation environment, the strong association of HF molecules was effectively suppressed, and the masking effect of its wide infrared absorption band on the characteristic peaks of impurities was significantly weakened. 2. A differential absorption spectroscopy decoupling algorithm based on multi-temperature point measurement is introduced to achieve dynamic and accurate subtraction of HF background absorption, avoiding the error accumulation caused by sample state fluctuations in traditional static subtraction methods. 3. Establish a multi-band cross-validation impurity fingerprint identification mechanism, using the consistency of the response of the same impurity in multiple non-adjacent bands as the criterion, fundamentally eliminating false positives caused by accidental overlap of spectral peaks or instrument drift. 4. The entire system achieves simultaneous detection of ten key metallic impurities in electronic-grade anhydrous hydrogen fluoride without the need for chemical pretreatment or chromatographic separation, with a detection limit of 10. -9 With a molar fraction level and an impurity identification accuracy rate exceeding 99.5%, it meets the stringent quality control requirements of the semiconductor manufacturing industry for ultra-high purity hydrogen fluoride raw materials. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall technical architecture of the product purity determination method and system for electronic-grade anhydrous hydrogen fluoride proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework of the differential absorption spectroscopy decoupling algorithm in this invention; Figure 3 This is a logical flow diagram of the low-temperature controllable gas phase separation and infrared excitation acquisition stage in this invention; Figure 4 This is a logical flow diagram of the data preprocessing and differential absorption decoupling calculation stages in this invention; Figure 5 This is a logical flowchart of the impurity fingerprint recognition and multi-band cross-validation stages in this invention. Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the low-temperature controllable gas phase separation unit and the infrared excitation and acquisition unit in this invention. Detailed Implementation

[0025] Please refer to Figures 1 to 6 This invention provides a method and system for determining the purity of electronic-grade anhydrous hydrogen fluoride, which solves the problems of existing infrared spectroscopy methods.

[0026] The present invention addresses the technical problem of high misjudgment rate of trace impurities in the determination of metallic impurities in electronic-grade anhydrous hydrogen fluoride due to severe overlap between the absorption peak of HF molecules and the characteristic absorption peak of impurities. The specific embodiments of the invention will be described in detail below with reference to the accompanying drawings.

[0027] The method for determining the purity of electronic-grade anhydrous hydrogen fluoride includes the following steps: S1. The electronic-grade anhydrous hydrogen fluoride sample to be tested is introduced into a low-temperature controllable gas phase separation chamber and subjected to gradient cooling within a set temperature range to keep the sample in a gas phase state and suppress the HF molecule association effect, thereby weakening its wide absorption band intensity in the infrared band. S2. Irradiate the processed gaseous sample with a broadband infrared light source, and acquire the full-band transmission spectrum signal using a high-resolution Fourier transform infrared spectrometer to obtain the original spectral dataset. S3. Perform baseline correction and noise suppression preprocessing on the original spectral dataset to generate standardized spectral curves; S4. Based on the preset metal impurity characteristic absorption band library, extract the absorbance sequence of the corresponding band in the standardized spectral curve and input it into the differential absorption spectral decoupling algorithm module. S5. The differential absorption spectroscopy decoupling algorithm module constructs a dynamic background subtraction model based on the absorption spectrum shape change law of HF molecules at different temperatures, and separates the residual absorption components from the mixed absorption signal that are not from HF. S6. Perform multi-band cross-matching between the residual absorption components and the standard fingerprint database of metal impurities to calculate the confidence score of each target impurity element. S7. When the confidence score of any impurity element exceeds the preset threshold, the impurity is determined to exist, and its concentration value is inverted based on its absorbance value in the characteristic band and the Lambert-Beer law. S8. Output a purity analysis report containing the types of metal impurities and their corresponding concentrations.

[0028] In step S1, the electronic-grade anhydrous hydrogen fluoride sample to be tested is introduced into a low-temperature controllable gas phase separation chamber via a high-pressure inert gas carrier. The chamber operates in the temperature range of -80℃ to -30℃, and its inner wall is made of electropolished stainless steel and coated with a magnesium fluoride anti-corrosion coating to prevent trace metal impurities from being adsorbed on the chamber wall or undergoing side reactions.

[0029] The chamber is equipped with a temperature gradient distribution control unit, which sequentially forms three continuous temperature zones along the gas flow direction: a pre-cooling zone, a steady-state separation zone, and an outlet buffer zone. The temperature difference between each zone is no less than 10℃. The initial temperature of the pre-cooling zone is set to -30℃ for rapid cooling of sample vapor; the steady-state separation zone is maintained at -60℃ as the main separation area, effectively suppressing the formation of HF dimers and polymers; the outlet buffer zone temperature is raised to -50℃ to prevent sample condensation at the outlet. The temperature gradient distribution control unit consists of a Peltier cooler, a miniature heating wire, a platinum resistance temperature sensor array, and a temperature control feedback circuit. The temperature control feedback circuit dynamically adjusts the cooling and heating power based on the real-time temperature readings of each zone, ensuring that the temperature fluctuation within the chamber does not exceed ±0.5℃.

[0030] Under these conditions, HF molecules exist stably in the gas phase as monomers. Their broad, strong absorption bands at 3.5 μm, 4.2 μm, and 7.8 μm become significantly narrower and their intensity decreases, providing a clear background for the subsequent identification of impurity characteristic peaks.

[0031] In step S2, a broadband infrared light source emits continuous infrared radiation with wavelengths covering 2.5 μm to 25 μm, which is guided into the sample region within a cryogenically controlled gas-phase separation chamber via an optical guide tube made of calcium fluoride crystal material. The optical guide tube has an inner diameter of 5 mm and a length of 50 cm, with vacuum-sealed flange interfaces at both ends, and its outer wall is wrapped with multiple layers of thermal insulation material to isolate it from environmental thermal radiation interference. The high-resolution Fourier transform infrared spectrometer is equipped with a liquid nitrogen-cooled mercury cadmium telluride detector, achieving a spectral resolution of 0.5 cm⁻¹. -1 The signal-to-noise ratio is not less than 40,000 to 1.

[0032] Before each measurement, the instrument performs automatic zero-point calibration and background scanning. High-purity nitrogen is used as the reference gas for the background scanning, and blank spectra are acquired under the same temperature and optical path conditions. The instrument then switches to the sample gas to acquire the transmitted light intensity. and background light intensity Simultaneous recording generates raw absorbance spectra. This constitutes the original spectral dataset.

[0033] In step S3, the data preprocessing unit performs three-stage processing on the original spectral dataset. First, baseline correction is performed using the asymmetric least squares method, with a penalty factor of 1000 and a smoothing parameter of 0.01, effectively eliminating low-frequency baseline drift caused by optical element scattering or detector drift. Second, a nine-point cubic moving average algorithm is applied to smooth the corrected spectrum, suppressing high-frequency random noise while preserving the sharpness of characteristic absorption peaks. Finally, normalization is performed, using the absorbance value of the weak absorption peak of HF molecules at 0.4 μm as a reference, and scaling the entire spectrum proportionally to ensure that the absorbance at this reference point is constant at 0.05. This operation eliminates the overall amplitude deviation caused by small fluctuations in sample concentration or optical path length errors, ensuring the comparability of measurement results from different batches. After the above processing, a standardized spectral curve is output. .

[0034] In step S4, based on a preset library of characteristic absorption bands for metal impurities, the standardized spectral curves are analyzed. The absorbance sequences for the corresponding bands were extracted. This band library contains the infrared characteristic absorption peak positions of ten target metal elements—sodium, potassium, calcium, magnesium, iron, nickel, chromium, copper, zinc, and aluminum—in the gas phase. Each element corresponds to at least two non-adjacent independent absorption bands; for example, sodium corresponds to 5.2 μm and 13.6 μm, and potassium corresponds to 6.1 μm and 14.3 μm. All bands strictly avoid the main absorption peak regions of HF molecules at 3.5 μm, 4.2 μm, and 7.8 μm. The extraction process is achieved through digital bandpass filtering, with each band window width set to 0.2 μm and the center wavelength aligned with the theoretical peak position. The extracted absorbance sequences are encapsulated in vector form and input into the differential absorption spectroscopy decoupling algorithm module.

[0035] In step S5, the differential absorption spectroscopy decoupling algorithm module performs the core decoupling operation. This module has a built-in temperature-absorption spectrum mapping table, which is constructed by pre-calibrating the infrared absorption spectra of pure HF gas at 5-degree intervals within the range of -80°C to -30°C. In actual measurements, the system collects the same gas phase sample at the first temperature point under the same optical path conditions. With the second temperature point The infrared transmission spectrum below, of which and A difference of 15°C, for example -60℃ The temperature is -45℃. Calculate the absorbance of both sets of spectra. and And generate differential spectra. Because the absorption spectrum of HF molecules changes significantly with temperature, while the position and intensity of the absorption peaks of metallic impurities are largely unaffected by temperature, therefore... The signal mainly contains variations in the HF background, while impurity signals approximately cancel each other out. Principal component analysis was used to analyze... Dimensionality reduction is performed, and the first three principal component vectors are extracted to construct the HF temperature response feature subspace. The original spectrum... Projected onto this subspace, the HF background absorption model is reconstructed. Finally, the net absorption spectrum This refers to the residual absorbent components contributed solely by non-HF components. This process can be expressed by the following formula: ; ; ; in, Principal component loading matrix, Both are vector representations at discrete wavelength points.

[0036] In step S6, the impurity fingerprint recognition unit will analyze the net absorption spectrum. Multi-band cross-matching was performed with a standard fingerprint database of metal impurities. This database stores multi-band absorbance response curves of ten target metal impurities at different concentration gradients, covering a concentration gradient of 10... -9 Up to 10 -6 The mole fraction range is defined, and each curve is temperature-compensated. The matching process employs a weighted Euclidean distance metric to verify the consistency of the absorbance response of each candidate impurity element across all associated bands. Specifically, for the mole fraction range... Let the impurity element be the first impurity element. The measured net absorbance of each characteristic band is The theoretical absorbance at the corresponding concentration C in the database is: Then the matching error of this band is defined. .

[0037] By minimizing the total weighted error Solve for the optimal concentration Subsequently, the inversion concentrations for each band were calculated. ,in For this impurity in The molar absorptivity at that location This is the effective optical path length. These are the weighting coefficients corresponding to each band. If the relative deviation between the concentration inversion results of any two bands is greater than 5%, the identification result of that impurity is rejected. Otherwise, a confidence score is calculated. This score reflects the quality of the match.

[0038] In step S7, the system sets a confidence threshold of 0.9. When the confidence score of any impurity element exceeds this threshold, the impurity is determined to exist. The final concentration is calculated as the weighted average of the concentrations retrieved from each effective band, with the weights determined by the signal-to-noise ratio of each band. The concentration retrieval strictly follows the Lambert-Beer Law formula. Perform calculations, where Net absorbance. represents the molar absorptivity of the impurity at the corresponding wavelength. Where the molar absorptivity... The data is taken from an experimentally calibrated temperature correction factor table, which records the temperature correction factor for each impurity at different temperatures. Value correction factor to ensure quantitative accuracy is maintained even at low temperatures.

[0039] In step S8, the concentration inversion and report generation unit integrates all identification results to generate a structured purity analysis report. The report includes impurity types, inverted concentrations for each band, weighted average concentration, confidence score, and a determination of whether the purity exceeds the limit. The report format conforms to the semiconductor industry SEMI standard and supports automatic uploading to the quality management system.

[0040] The electronic-grade anhydrous hydrogen fluoride product purity determination system includes a low-temperature controllable gas phase separation unit, an infrared excitation and acquisition unit, a data preprocessing unit, a differential absorption decoupling calculation unit, an impurity fingerprint identification unit, and a concentration inversion and report generation unit.

[0041] The low-temperature controllable gas phase separation unit includes a sample inlet valve, a three-stage temperature control chamber, a temperature sensor array, and a temperature control feedback circuit. The sample inlet valve adopts an all-metal sealed structure and has a pressure resistance of not less than 10 MPa. The three-stage temperature control chamber is arranged sequentially along the gas flow direction as a pre-cooling zone, a steady-state separation zone, and an outlet buffer zone. Each zone is independently equipped with a Peltier cooler and a heating wire. The temperature control feedback circuit dynamically adjusts the power output based on the real-time readings of the temperature sensor array to ensure that the temperature fluctuation within the chamber does not exceed ±0.5℃.

[0042] The infrared excitation and acquisition unit includes a broadband infrared light source, a high-resolution Fourier transform infrared spectrometer, and a matching optical guide tube. The optical guide tube is made of calcium fluoride crystal material, with an inner diameter of 5 mm and a length of 50 cm. Both ends are equipped with vacuum-sealed flange interfaces, and the outer wall of the guide tube is wrapped with multiple layers of heat insulation material to isolate environmental heat radiation interference.

[0043] The baseline correction performed by the data preprocessing unit adopts the asymmetric least squares method, the smoothing filter adopts the nine-point three-times moving average algorithm, and the normalization process uses the weak absorption peak of HF molecules at 0.4μm as the reference.

[0044] The differential absorption decoupling calculation unit has a built-in temperature-absorption spectrum mapping table, which is constructed by pre-calibrating the infrared absorption spectrum of pure HF gas in the range of -80℃ to -30℃ at 5℃ intervals. This table is used to interpolate in real time to obtain the HF background absorption model at any operating temperature.

[0045] The impurity fingerprint recognition unit stores a standard fingerprint database containing multi-band absorbance response curves for ten target metal impurities at different concentration gradients. Each curve has been temperature-compensated and the concentration gradient covers 10... -9 Up to 10 -6 Mole fraction range.

[0046] The concentration inversion and report generation unit is based on the Lambert-Beer law formula. Concentration calculations were performed, with the molar absorptivity taken from an experimentally calibrated table of temperature correction factors.

[0047] This embodiment effectively suppresses the strong association of HF molecules by constructing a low-temperature controllable gas-phase separation environment, significantly weakening the masking effect of its wide infrared absorption band on impurity characteristic peaks. A differential absorption spectroscopy decoupling algorithm based on multi-temperature point measurements is introduced to achieve dynamic and accurate subtraction of HF background absorption. A multi-band cross-validation impurity fingerprinting mechanism is established to fundamentally eliminate false positives. The entire system achieves simultaneous detection of ten key metallic impurities without chemical pretreatment, with a detection limit of 10. -9 With a molar fraction level and an impurity identification accuracy rate exceeding 99.5%, it meets the stringent quality control requirements of the semiconductor manufacturing industry for ultra-high purity hydrogen fluoride raw materials.

Claims

1. A method for determining the purity of electronic-grade anhydrous hydrogen fluoride, characterized in that, include: The anhydrous hydrogen fluoride sample to be tested is introduced into a low-temperature controllable gas phase separation chamber and subjected to gradient cooling within a set temperature range to keep the sample in a gas phase state and suppress the HF molecule association effect, thereby weakening its broad absorption band intensity in the infrared band. The processed gaseous sample was irradiated with a broadband infrared light source, and the full-band transmission spectrum signal was acquired by a high-resolution Fourier transform infrared spectrometer to obtain the original spectral dataset. Baseline correction and noise suppression preprocessing are performed on the original spectral dataset to generate standardized spectral curves; Based on a pre-defined library of characteristic absorption bands for metal impurities, absorbance sequences for corresponding bands in standardized spectral curves are extracted and input into the differential absorption spectroscopy decoupling algorithm module. The differential absorption spectroscopy decoupling algorithm module constructs a dynamic background subtraction model based on the absorption spectrum shape change law of HF molecules at different temperatures, and separates the residual absorption components from the mixed absorption signal that are not from HF. The residual absorption components are cross-matched with the standard fingerprint database of metal impurities in multiple bands to calculate the confidence score of each target impurity element. When the confidence score of any impurity element exceeds a preset threshold, the impurity is determined to exist, and its concentration value is inverted based on its absorbance value in the characteristic band and the Lambert-Beer law. The output includes a purity analysis report containing the types of metal impurities and their corresponding concentrations.

2. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 1, characterized in that, The anhydrous hydrogen fluoride sample of electronic grade to be tested is introduced into a low-temperature controllable gas phase separation chamber, and subjected to gradient cooling within a set temperature range to maintain the sample in a gas phase state and suppress the HF molecule association effect, including: The operating temperature of the low-temperature controllable gas phase separation chamber is controlled within the range of -80℃ to -30℃. The inner wall of the cavity is made of electropolished stainless steel and coated with a magnesium fluoride anti-corrosion coating. The temperature gradient distribution control unit forms at least three continuous temperature zones along the sample flow path, with a temperature difference of not less than 10℃ between each zone.

3. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 2, characterized in that, The processed gaseous sample was irradiated with a broadband infrared light source, and full-band transmission spectral signals were acquired using a high-resolution Fourier transform infrared spectrometer, including: The broadband infrared light source has a wavelength coverage range of 2.5 μm to 25 μm and a spectral resolution of 0.5 cm⁻¹. -1 ; The high-resolution Fourier transform infrared spectrometer is equipped with a liquid nitrogen-cooled mercury cadmium telluride detector with a signal-to-noise ratio of not less than 40,000 to 1. The optical guide tube is made of calcium fluoride crystal material, with an inner diameter of 5mm and a length of 50cm. It has vacuum-sealed flange interfaces at both ends and is wrapped with multiple layers of heat-insulating material on the outer wall.

4. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 3, characterized in that, Baseline correction and noise suppression preprocessing are performed on the original spectral dataset to generate standardized spectral curves, including: Baseline correction was performed using the asymmetric least squares method, with a penalty factor of 1000 and a smoothing parameter of 0.

01. The corrected spectrum is smoothed by applying a nine-point triple moving average algorithm. The absorbance at the weak absorption peak of HF molecules at 0.4 μm was used as a reference for normalization, so that the absorbance at this reference point was kept constant at 0.

05.

5. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 4, characterized in that, Based on a pre-defined library of characteristic absorption bands for metallic impurities, absorbance sequences for corresponding bands are extracted from standardized spectral curves, including: The library of characteristic absorption bands for metal impurities includes the infrared characteristic absorption peak positions of ten target metal elements in the gas phase: sodium, potassium, calcium, magnesium, iron, nickel, chromium, copper, zinc, and aluminum. Each element corresponds to no fewer than two non-adjacent independent absorption bands, and all bands avoid the main absorption peak regions of HF molecules at 3.5 μm, 4.2 μm and 7.8 μm; The absorbance of each band was extracted using a digital bandpass filter, with the window width set to 0.2 μm.

6. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 5, characterized in that, The differential absorption spectroscopy decoupling algorithm module constructs a dynamic background subtraction model based on the absorption spectrum shape variation of HF molecules at different temperatures, including: Infrared transmission spectra of the same gaseous sample were collected at the first and second temperature points under the same optical path conditions, with the first and second temperature points differing by 15℃. Calculate the absorbance difference between the two sets of spectra at each wavelength point to generate a difference spectrum; Principal component analysis was used to reduce the dimension of the difference spectrum and extract stable absorption feature vectors that are independent of temperature. The eigenvector is projected back into the original spectral space to reconstruct the net absorption spectrum contributed only by the non-HF components.

7. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 6, characterized in that, Principal component analysis was used to reduce the dimensionality of the difference spectrum and extract temperature-independent stable absorption feature vectors, including: Perform principal component analysis on the difference spectrum to obtain the principal component loading matrix; The first three principal component vectors are selected to form the HF temperature response feature subspace; The normalized spectral curves are projected onto this subspace to reconstruct the HF background absorption model; The net absorption spectrum is obtained by subtracting the HF background absorption model from the standardized spectral curve.

8. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 7, characterized in that, The residual absorption components are cross-matched with a standard fingerprint database of metal impurities in multiple bands to calculate the confidence score of each target impurity element, including: The standard fingerprint database stores multi-band absorbance response curves of ten target metallic impurities at different concentration gradients, covering a concentration gradient of 10. -9 Up to 10 -6 Mole fraction range; The weighted Euclidean distance metric was used to verify the consistency of the absorbance response of each candidate impurity element across all associated bands. If the relative deviation between the concentration inversion results of any two bands is greater than 5%, the identification result of the impurity is rejected.

9. The method for determining the purity of electronic-grade anhydrous hydrogen fluoride according to claim 8, characterized in that, Its concentration value is retrieved by combining its absorbance value in the characteristic band with the Lambert-Beer law, including: According to the formula Concentration calculations were performed, among which... Net absorbance. This represents the molar absorptivity of the impurity at the corresponding wavelength. The effective optical path length is given by the molar absorptivity, which is taken from the experimentally calibrated temperature correction factor table. The final concentration is the weighted average of the inverted concentrations of each effective band, with the weights determined by the signal-to-noise ratio of each band.

10. A product purity determination system for electronic-grade anhydrous hydrogen fluoride, characterized in that, include: The low-temperature controllable gas phase separation unit is used to receive the electronic-grade anhydrous hydrogen fluoride sample to be tested and maintain its gas phase state in a controlled low-temperature environment to suppress HF molecule association. The infrared excitation and acquisition unit includes a broadband infrared light source, a high-resolution Fourier transform infrared spectrometer, and a matching optical guide tube, which is used to irradiate gas phase samples and acquire full-band transmission spectrum signals. The data preprocessing unit is used to perform baseline correction, smoothing filtering, and normalization on the raw spectral signal; The differential absorption decoupling calculation unit is used to construct a dynamic background subtraction model based on multi-temperature point spectral data and separate non-HF absorption components. The impurity fingerprint recognition unit is embedded with a library of metal impurity characteristic absorption bands and a standard fingerprint database, which is used to perform multi-band cross-matching and confidence assessment. The concentration inversion and report generation unit is used to calculate the impurity concentration based on the identification results and output a structured purity analysis report.

Images