Semiconductor boron trifluoride deep cooling precise control filling method and system

Abstract

The application discloses a boron trifluoride deep cooling precise control filling method and system, and belongs to the technical field of filling control. The method specifically comprises the following steps: collecting the pressure, temperature, trace moisture and heat exchange response parameters in a boron trifluoride to-be-filled steel cylinder, establishing a single-cylinder dynamic compensation model based on the collected parameters, implementing phased cooling of the steel cylinder to a target deep cooling temperature zone, adjusting the cooling process according to the temperature distribution and pressure change in the cooling process, forming a stable controlled gas-liquid condensation state, filling the boron trifluoride into the to-be-filled steel cylinder in the target deep cooling temperature zone by using a phased filling strategy, making the actual filling amount close to a target value according to the single-cylinder dynamic compensation model, stopping filling and sealing when the to-be-filled steel cylinder meets a preset consistency condition, and otherwise performing compensation filling or back-pumping correction. The application reduces filling deviation caused by uneven temperature distribution and pressure response deviation, and is suitable for the requirement of high-purity gas filling control in the semiconductor field.

Description

Technical Field

[0001] This invention belongs to the field of filling control technology, specifically to a method and system for the cryogenic precision control of boron trifluoride filling for semiconductors. Background Technology

[0002] Boron trifluoride, an important boron-containing gas, is widely used in semiconductor manufacturing processes such as ion implantation, etching, and doping, placing high demands on gas purity, filling accuracy, and supply stability. In practical applications, boron trifluoride is typically stored in high-pressure gas cylinders and released during use through a precision gas supply system. However, due to its strong chemical reactivity and sensitivity to moisture, it easily reacts with residual moisture or the inner wall of the cylinder during filling, thus affecting gas quality and the stability of subsequent processes.

[0003] In existing technologies, the filling of specialty gases such as boron trifluoride typically employs a constant-pressure filling method under ambient or cryogenic conditions, with the filling volume controlled by weighing or pressure detection. However, these methods often rely on a single parameter for control, failing to comprehensively consider individual differences in gas cylinders, adsorption behavior on the inner wall, and the hysteresis effect during gas-liquid phase transition. Under cryogenic conditions, the gas may experience localized condensation or uneven distribution within the cylinder, leading to phenomena such as false full filling and pressure deviation during temperature recovery. This results in deviations between the actual filling volume and the target value, making it difficult to meet the consistency requirements of semiconductor-grade applications. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method and system for precise control of cryogenic filling of boron trifluoride for semiconductors. This method can combine the individual characteristics of the gas cylinder to achieve phased regulation and has dynamic compensation capabilities. In particular, under cryogenic conditions, the gas distribution state can be accurately identified and controlled.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for cryogenic precision control of boron trifluoride filling for semiconductors, including: The pressure, temperature, trace moisture and heat exchange response parameters of the boron trifluoride cylinder to be filled were collected, and a single-cylinder dynamic compensation model was established based on the collected parameters. Based on the single-cylinder dynamic compensation model, the cylinder is cooled to the target cryogenic temperature range in stages, and the cooling process is adjusted according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state. Under the target cryogenic temperature range, a staged filling strategy is adopted to fill the steel cylinder to be filled with boron trifluoride, including initial constant pressure filling, middle pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model. When the cylinder to be filled meets the preset consistency conditions, filling is stopped and the cylinder is sealed; otherwise, compensation filling or backfilling correction is performed.

[0006] Specifically, the process of collecting pressure, temperature, trace moisture, and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and establishing a single-cylinder dynamic compensation model based on the collected parameters, includes: Vacuum treatment is performed on the steel cylinder to be filled and its connecting pipeline until the preset baseline pressure is reached. The steel cylinder is then sealed in an isolated state, and the initial pressure change trajectory and temperature stability range in the isolated state are recorded. A controlled disturbance is created by introducing a limited amount of boron trifluoride gas into the cylinder. During the introduction process, the transient change sequence of pressure, temperature and trace moisture inside the cylinder is collected simultaneously, and the heat exchange response between the gas and the inner wall of the cylinder is triggered. After the disturbance ends, the cylinder remains closed, and the pressure recovery process, temperature drop process, and trace moisture change process inside the cylinder are continuously collected and the corresponding time correlation sequences are recorded to form a multi-source coupled response dataset. The multi-source coupled response dataset is segmented according to a preset time window, and the correlation features of pressure change trend, temperature gradient evolution and moisture migration path in each segment are extracted. A mapping relationship between the gas distribution state inside the cylinder and the action state of the inner wall is established. Based on the aforementioned mapping relationship, the effective volume of the gas cylinder, the adsorption behavior of the inner wall, and the hysteresis characteristics of gas-liquid conversion are uniformly characterized to form a single-cylinder dynamic compensation model.

[0007] Specifically, the multi-source coupled response dataset is segmented according to a preset time window, and the correlation features of pressure change trends, temperature gradient evolution, and moisture migration paths within each segment are extracted. A mapping relationship is then established between the gas distribution state inside the cylinder and the action state of the inner wall, including: The continuous sampling sequences of pressure, temperature and trace moisture in the multi-source coupled response dataset are synchronously compared to identify the time of disturbance introduction, pressure inflection point, temperature transition and moisture fluctuation transition. The aforementioned times are used as time anchors to divide the multi-source coupled response dataset into the initial response segment, the transition coupling segment and the stabilization identification segment. The monotonic change trajectory of pressure, the gradient migration trajectory of temperature, and the delayed following trajectory of trace moisture were extracted for each segment. All trajectories were rearranged according to their sequential occurrence within the same segment to form the dominant response chain of the corresponding segment. The dominant response chains of each segment are connected in series according to the connection relationship between the beginning and end of adjacent segments. The leading, following and regressive relationships between pressure change trends, temperature gradient evolution and moisture migration paths are grouped to form a group of related features that characterize the local gas accumulation, interface expansion and inner wall contact migration sequence inside the cylinder. Based on the aforementioned associated feature group, the internal response process of the gas cylinder is divided into a free distribution state, a wall traction state, and a delayed release state. The order of occurrence and transition boundary of each state in each segment are mapped to the internal gas distribution state and the internal wall action state of the gas cylinder, forming a segmented mapping relationship for subsequent single-cylinder dynamic compensation model calls.

[0008] Specifically, based on the aforementioned mapping relationship, the effective volume of the gas cylinder, its inner wall adsorption behavior, and the gas-liquid conversion hysteresis characteristics are uniformly characterized to form a single-cylinder dynamic compensation model, including: The segmented mapping relationship is deconstructed, and the internal response process of the gas cylinder is segmented according to the correspondence between the gas distribution state and the internal wall action state in each segment, so as to obtain the start and end boundaries of each segment and their positional correlation in the time series. Based on the calibration results of the aforementioned sections, the response data within each section are separated and processed. The pressure response component related to the change in the effective volume of the gas cylinder, the adsorption response component related to the action of the inner wall, and the hysteresis response component related to the phase transition are extracted respectively, and independent response sequences within the corresponding sections are established. According to the chronological relationship of each segment in the time series, the independent response sequences are recombined across segments to connect the alternating influence relationship between the effective volume of the gas cylinder, the adsorption behavior of the inner wall and the gas-liquid conversion hysteresis characteristics, forming a multi-stage continuous response chain. Based on the multi-stage continuous response chain, the effective volume of the cylinder, the adsorption behavior of the inner wall and the gas-liquid conversion hysteresis characteristics are uniformly merged, and the boundary information of each section and its transfer order are embedded into it to generate a single-cylinder dynamic compensation model for dynamic calling during the filling process.

[0009] Specifically, based on the single-cylinder dynamic compensation model, the cylinder is cooled in stages to the target cryogenic temperature range, and the cooling process is adjusted according to temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state, including: Based on the single-cylinder dynamic compensation model, the effective volume characterization, inner wall adsorption behavior characterization, and gas-liquid conversion hysteresis characterization corresponding to the current cylinder to be filled are read. Based on the characterization results, the cylinder cooling process is divided into a pre-cooling transition zone, a relay acclimatization zone, and a cryogenic approach zone. The starting conditions, switching conditions, and dwell order of each zone are determined. The first stage of cooling is carried out on the cylinder to be filled according to the pre-cooling transition zone, so that the gas inside the cylinder enters the initial cooling state coordinated with the action of the inner wall. During the first stage of cooling, the temperature distribution and pressure change sequence at different positions on the outer wall of the cylinder to be filled are collected, and the collected results are compared with the corresponding section in the single-cylinder dynamic compensation model in the first round. When the results of the first round of comparison meet the preset switching conditions, the cylinder to be filled is transferred to the intermediate acclimatization zone. In this zone, the cold energy input is promoted in the order of slow cooling followed by fast cooling. Based on the migration direction of temperature distribution and the stage transition of pressure change, the correspondence between the gas distribution state inside the cylinder to be filled and the action state of the inner wall is calibrated for the second time. Based on the second calibration, the cylinder to be filled is transferred into the cryogenic approach zone. Using the single-cylinder dynamic compensation model as a reference, the local temperature difference shift, pressure response misalignment and condensation propulsion sequence changes that occur during the cooling process are identified segment by segment. When any segment is found to deviate from the preset trajectory, the cooling rhythm of the current zone, the zone dwell sequence or the connection time of adjacent zones is adjusted to correct the condensation propulsion path inside the cylinder to be filled. After completing the cryogenic approximation zone, a closed-loop check is performed on the cylinder to be filled. When the check result is consistent with the target state sequence corresponding to the single-cylinder dynamic compensation model, it is determined that the cylinder has entered the gas-liquid condensation state required for filling.

[0010] Specifically, when the first round of comparison results meets the preset switching conditions, the cylinder to be filled is transferred to the relay training interval, and the correspondence between the internal gas distribution state and the inner wall action state of the cylinder to be filled is calibrated a second time, including: When the results of the first round of comparison meet the preset switching conditions, the temperature distribution migration order, pressure change turning order and corresponding mapping state obtained in the interval are saved as the previous frozen segment, and the previous frozen segment is used as the starting reference of the relay acclimatization interval. Within the intermediate acclimatization interval, multiple rounds of cold energy delivery at different rhythms are applied to the cylinder to be filled according to a preset time sequence, so that the gas inside the cylinder alternates between continuous contraction and propulsion and intermittent stagnation. The pressure change sequence inside the cylinder and the temperature rearrangement sequence of the outer wall are collected synchronously between each round of cold energy delivery to form a progressive response group for the intermediate acclimatization stage. The progressive response group is connected end-to-end with the preceding freezing segment to track the leading, lagging, and reverting changes in the gas distribution state relative to the inner wall action state within the relay acclimatization interval, and rearranged into a state misalignment chain according to the order of occurrence of each type of change in the time series. Based on the aforementioned state misalignment chain, the corresponding boundaries, transfer order, and connection positions of the gas distribution state inside the cylinder to be filled and the action state of the inner wall within the relay training interval are recalibrated, and the recalibration results are incorporated into the segmented mapping relationship corresponding to the single-cylinder dynamic compensation model to complete the secondary calibration.

[0011] Specifically, after completing the cryogenic approximation zone, a closed-loop check is performed on the cylinder to be filled. When the check result matches the target state sequence corresponding to the single-cylinder dynamic compensation model, it is determined that the cylinder has entered the gas-liquid condensation state required for filling, including: After the cryogenic approach zone ends, the current cooling process is terminated, and the pressure sequence, temperature sequence, and corresponding state transition records formed by the cylinder to be filled at the end of this zone are sealed to form the final response group. Obtain the target state sequence corresponding to the current cylinder to be filled in the single-bottle dynamic compensation model, and align the final response group with the target state sequence segment by segment in chronological order to determine the corresponding starting point, connection order and termination boundary between each response segment and each target state node. Based on the segment-by-segment alignment results, the sequential offset, boundary misconnection and state return between each response segment in the final response group and the target state node are judged, and the gas distribution state and inner wall action state of the gas cylinder to be filled are checked in a closed loop according to the judgment results to form the current state sequence. The current state sequence is compared with the target state sequence. When the state order and transition boundary of the two are consistent, it is determined that the cylinder to be filled has reached the gas-liquid condensation state required for filling.

[0012] Specifically, the step of filling the cylinder with boron trifluoride using a staged filling strategy under the target cryogenic temperature range includes initial constant pressure filling, mid-stage pulse-type feeding, and final differential pressure adjustment, so that the actual filling amount approaches the target value according to the single-cylinder dynamic compensation model, including: Boron trifluoride was introduced into the cylinder to be filled under constant pressure in the target cryogenic temperature zone, and the mass change, internal pressure change and corresponding state transition records of the cylinder to be filled were collected. The collected results were then compared with the single-cylinder dynamic compensation model for interval mapping. When the preset interval boundary is reached, the system switches to pulse feeding to connect the quality changes and pressure recovery processes of adjacent pulse cycles, forming a continuous feeding sequence. The intermediate feeding sequence is connected with the initial filling sequence to identify the inter-segment offset of quality advancement and state transition, and the feeding rhythm and switching timing are adjusted based on the single-bottle dynamic compensation model. After completing the inter-segment correction, the remaining filling process is advanced by adjusting the pressure difference between the gas supply end and the cylinder end, so that the filling amount is close to the target value, and the corresponding quality change nodes are recorded until the filling is completed.

[0013] A cryogenic precision control filling system for boron trifluoride used in semiconductors, used to implement the aforementioned cryogenic precision control filling method for boron trifluoride used in semiconductors, includes: a parameter compensation module, a state adjustment module, a filling control module, and a filling determination module; The parameter compensation module is used to collect the pressure, temperature, trace moisture and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and to establish a single-cylinder dynamic compensation model based on the collected parameters. The state adjustment module is used to cool the cylinder to the target cryogenic temperature range in stages based on the single-cylinder dynamic compensation model, and to adjust the cooling process according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state. The filling control module is used to fill boron trifluoride into the cylinder to be filled in the target cryogenic temperature zone using a staged filling strategy, including initial constant pressure filling, middle pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model. The filling determination module is used to stop filling and seal the cylinder when it meets the preset consistency conditions; otherwise, it performs compensation filling or backfilling correction.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention proposes a method and system for precise cryogenic control of boron trifluoride filling for semiconductor applications. Based on a single-bottle dynamic compensation model, it implements staged cryogenic cooling control and combines a staged filling strategy of initial constant pressure filling, mid-stage pulsed feeding, and final differential pressure adjustment. This allows for full-process control of the gas distribution and the interaction with the inner wall within the cylinder. This enables simultaneous consideration of individual cylinder differences, inner wall adsorption behavior, and gas-liquid conversion hysteresis during the filling process, thereby improving the accuracy and consistency of filling quantity control, reducing filling deviations caused by uneven temperature distribution and pressure response offsets, and ensuring a stable gas-liquid state for the cylinder before it enters the service stage. This method is suitable for the high-purity gas filling control requirements in the semiconductor field. Attached Figure Description

[0015] Figure 1 Flowchart provided for this invention; Figure 2 The system architecture diagram provided for this invention; Figure 3 A schematic diagram of the electronic device provided by the present invention; Figure 4 A schematic diagram of the structure of a computer-readable storage medium provided by the present invention. Detailed Implementation

[0016] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0019] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0020] Example 1 Please see Figure 1 The present invention provides an embodiment of a method for cryogenic precision control of boron trifluoride filling for semiconductors, comprising the following specific steps: Step S1: Collect pressure, temperature, trace moisture and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and establish a single-cylinder dynamic compensation model based on the collected parameters.

[0021] The specific steps of step S1 are as follows: Step S101: Vacuum the gas cylinder to be filled and its connecting pipeline until the preset baseline pressure is reached, and keep the gas cylinder sealed in an isolated state. Record the initial pressure change trajectory and temperature stability range in the isolated state.

[0022] In this embodiment, a calibrated helium mass spectrometer leak detector is first used to pre-check the valve interface, welds, and threaded connections of the cylinder to be filled. After confirming that there is no obvious leakage, the cylinder valve is reliably connected to the vacuum pipeline system via a stainless steel compression fitting. The vacuum pipeline system is configured in series with a manual shut-off valve, an electromagnetic vacuum baffle valve, a capacitive thin-film vacuum gauge, a Roots vacuum pump unit, and a rotary vane backing pump. The ultimate vacuum of the Roots vacuum pump unit is not less than 5 × 10⁻⁶. -2Pa, pumping speed not less than 15L / s; then turn on the rotary vane back pump, Roots vacuum pump unit and each pipeline valve in sequence to perform staged vacuuming of the gas cylinder and connecting pipeline until the system pressure displayed by the capacitive thin film vacuum gauge stabilizes at the preset baseline pressure (in this embodiment the preset baseline pressure is 10Pa), and maintain the pumping at this pressure for 30 minutes to fully remove the gas and moisture adsorbed on the inner wall of the gas cylinder and the inner surface of the pipeline.

[0023] After the maintenance time is reached, a platinum resistance temperature sensor attached to the middle of the outer wall of the cylinder is used to collect the surface temperature of the cylinder in real time at a sampling frequency of 1 time / second. At the same time, a capacitive thin-film vacuum gauge is used to collect the internal pressure data of the cylinder simultaneously at a sampling frequency of 1 time / second. Data collection is carried out continuously for no less than 2 hours until the fluctuation range of the cylinder surface temperature does not exceed ±0.5℃ within 30 minutes. This is considered to be the temperature stabilization range. The upper and lower limits of the temperature stabilization range and the corresponding time range are recorded. At the same time, all pressure data points from the isolation time to the end of the temperature stabilization range are recorded in their entirety, and the initial pressure change trajectory under the isolation state is plotted.

[0024] Step S102: Introduce a limited amount of boron trifluoride gas into the cylinder to form a controlled disturbance, and simultaneously collect the transient change sequence of pressure, temperature and trace moisture inside the cylinder during the introduction process, while triggering the heat exchange response between the gas inside the cylinder and the inner wall.

[0025] In this embodiment, after step S101 is completed and the cylinder is within the temperature stable range, a stainless steel electropolished pipeline and diaphragm valve assembly, which have been pre-treated with 99.999% pure boron trifluoride standard gas through three vacuum-purging and replacement processes, are used to connect the cylinder valve to a 5mL quantitative loop injection system. After the quantitative loop injection system is pre-vacuumed to the same pressure as the cylinder baseline, boron trifluoride gas (purity ≥99.999%, moisture content ≤50ppb) is injected into the quantitative loop to 0.1MPa through a high-precision pressure regulating valve and then sealed and left to stand for 5 minutes to ensure uniform gas pressure within the quantitative loop.

[0026] Subsequently, the gas path was rapidly switched using a pneumatic three-way diaphragm valve with a response time ≤0.1s, allowing all boron trifluoride gas in the metering loop to be introduced into the cylinder. The amount of boron trifluoride gas introduced, converted to standard conditions, was 2.04 × 10⁻⁶. -4The initial partial pressure inside the cylinder is 0.5 kPa, corresponding to a mol pressure. This creates a controllable step gas disturbance. During this disturbance, the gas undergoes instantaneous heat absorption and cooling due to the Joule-Thomson effect, automatically triggering a heat exchange response between the gas inside the cylinder and the adsorption layer on the inner wall. Data is collected synchronously 10 seconds before the gas is introduced and continuously until 30 minutes after the gas is introduced, with all parameters fluctuating within ±0.1% for 10 minutes. The resulting transient pressure sequence includes the pressure step at the moment of introduction, pressure fluctuations during the heat exchange process, and the final stable value; the transient temperature sequence includes the instantaneous cooling, heat conduction recovery, and temperature equilibrium process; and the transient trace moisture sequence includes the baseline moisture concentration, the concentration change during the decomposition and adsorption of water on the inner wall, and the final equilibrium concentration. The evolution curve of the temperature difference between the inner wall of the cylinder and the internal gas during the heat exchange response is also recorded.

[0027] Step S103: After the disturbance ends, maintain the closed state, continuously collect data on the pressure recovery process, temperature drop process, and trace moisture change process inside the cylinder, and record the corresponding time correlation sequence to form a multi-source coupled response dataset.

[0028] In this embodiment, after the pneumatic three-way diaphragm valve is completely closed in step S102, and the cylinder is completely isolated from all external pipelines and gas source systems and maintained in a completely closed state, continuous data acquisition is immediately performed. During the acquisition process, the complete recovery process of the cylinder pressure gradually rising from the instantaneous low pressure value after the introduction of boron trifluoride to the dynamic equilibrium value is recorded in real time. This process is driven by the coupling of three factors: the thermal expansion effect of the gas as the temperature rises, the desorption effect of the gas physically adsorbed on the inner wall of the cylinder, and the gas molecule consumption effect caused by the irreversible hydrolysis reaction between boron trifluoride and trace moisture. Simultaneously, the complete drop process of the gas temperature in the cylinder gradually falling from the instantaneous low temperature value caused by the Joule-Thomson effect to the same as the ambient temperature is recorded. This process follows the unsteady-state heat conduction law and includes the rapid heat exchange stage between the inner wall of the cylinder and the internal gas, and the slow thermal equilibrium stage of natural convection and thermal radiation between the outer wall of the cylinder and the ambient air. At the same time, the concentration of trace moisture in the gas phase is continuously recorded.

[0029] During the data acquisition process, all data points were automatically stamped with a unified format timestamp. At the same time, key characteristic nodes in the process of parameter changes were automatically marked, including pressure recovery inflection point, temperature drop half-life time point, trace moisture concentration peak point, and hydrolysis reaction rate mutation point. Data acquisition was stopped when the pressure fluctuation inside the cylinder did not exceed ±0.05Pa, the temperature difference between the cylinder at three locations and the ambient temperature did not exceed ±0.2℃, and the hourly change rate of trace moisture concentration did not exceed ±0.02ppb / h. The gas-solid two-phase system inside the cylinder was determined to have reached a dynamic equilibrium state. Finally, a multi-source coupled response dataset containing high-precision timestamp sequence, multi-location cylinder temperature sequence, internal pressure sequence, gas phase trace moisture concentration sequence, and ambient temperature sequence was formed.

[0030] Step S104: The multi-source coupled response dataset is segmented according to a preset time window, and the correlation features of pressure change trend, temperature gradient evolution and moisture migration path in each segment are extracted. A mapping relationship between the gas distribution state inside the cylinder and the action state of the inner wall is established.

[0031] In this embodiment, the multi-source coupled response dataset obtained in step S103 is first preprocessed. A soft threshold denoising method combining a 50-point moving average filter with db4 wavelet basis 3-level decomposition is used to remove sensor electronic noise and environmental vibration interference, while retaining the trend and abrupt change characteristics of the data. Subsequently, synchronous differential operations and cross-comparisons are performed on the preprocessed pressure, temperature, and trace moisture continuous sampling sequences to identify key time anchor points. The disturbance introduction time is determined by the first abrupt change point where the absolute value of the first-order difference of the pressure sequence is greater than 0.2 Pa / s, and is also determined by the negative abrupt change point where the first-order difference of the temperature sequence is less than -0.5℃ / s and the first-order difference of the trace moisture sequence. Triple verification was performed on positive abrupt change points greater than 0.01 ppb / s to ensure that the time anchor error did not exceed 1 s; the pressure inflection point was determined by the zero-crossing point where the second-order difference of the pressure sequence turned from positive to negative, corresponding to the turning point where the pressure changed from rapid thermal expansion to slow increase through adsorption-desorption coupling; the temperature inflection point was determined by the moment when the average temperature change rate at the top, middle, and bottom of the cylinder was less than 0.01 ℃ / min, corresponding to the turning point where the heat exchange between the cylinder and the environment changed from unsteady-state rapid conduction to steady-state slow equilibrium; the moisture fluctuation transition point was determined by the zero-crossing point where the first-order difference of the trace moisture sequence turned from positive to negative, corresponding to the turning point where the rate of desorption of adsorbed water on the inner wall was less than the rate of hydrolysis and consumption of boron trifluoride.

[0032] Using the four time anchor points mentioned above as boundaries, the multi-source coupled response dataset is divided into an initial response segment (from the moment of disturbance introduction to the pressure inflection point, lasting approximately 15-30 minutes), a transition coupling segment (from the pressure inflection point to the temperature transition point, lasting approximately 2-4 hours), and a stabilization identification segment (from the temperature transition point to the end of the data collection, lasting approximately 18-22 hours). Least squares linear fitting and cubic spline interpolation methods are used to extract the characteristic trajectories within each segment: the initial response segment extracts the monotonically increasing pressure trajectory, the rapidly decaying trajectory of the cylinder's axial temperature gradient (characterized by the absolute value of the temperature difference between the upper, middle, and lower points), and the delayed following-increasing trajectory of trace moisture (characterized by the time difference between the increase in moisture concentration and the increase in pressure); the transition coupling segment extracts the logarithmically slow increasing trajectory of pressure and the temperature gradient... The exponential decay trajectory and the trajectory of trace moisture with initial peak and then slow decline; the stable fluctuation trajectory of extraction pressure (fluctuation amplitude ≤ ±0.05Pa), the homogenization trajectory of temperature (axial temperature difference ≤ 0.1℃), and the exponential decay trajectory of trace moisture in the stabilization identification segment; rearranged according to the order of appearance of each trajectory feature in the same segment to form the dominant response chain of each segment: the dominant response chain of the initial response segment is the instantaneous establishment and rapid decay of temperature gradient → monotonous thermal expansion of pressure → delayed desorption and rise of adsorbed moisture on the inner wall; the dominant response chain of the transition coupling segment is the continuous decay of temperature gradient → slow rise of pressure adsorption-desorption coupling → competitive balance between water desorption and hydrolysis; the dominant response chain of the stabilization identification segment is the complete homogenization of temperature field → dynamic equilibrium fluctuation of pressure → exponential decay dominated by water hydrolysis.

[0033] The three dominant response chains are connected sequentially in time. Partial least squares regression is used to quantify the leading coefficient, following coefficient, and retracement coefficient between pressure change trends, temperature gradient evolution, and moisture migration paths. The leading coefficient represents the degree of advance influence of a parameter change on other parameters, the following coefficient represents the degree of lag response of a parameter change to other parameters, and the retracement coefficient represents the reverse contribution of gas generated by water decomposition and adsorption to pressure changes. The above coefficients are grouped with the characteristic parameters of each trajectory (such as pressure rise rate, initial amplitude of temperature gradient, peak moisture concentration, and peak delay time) to form a correlation feature group containing leading feature group, following feature group, lag feature group, and retracement feature group. This feature group can quantitatively characterize the degree of local gas accumulation inside the cylinder, the adsorption and expansion rate of the gas-solid interface, and the contact and migration sequence of adsorbed water on the inner wall.

[0034] Based on the threshold range of the associated feature groups, the internal response process of the gas cylinder is divided into three states: the free distribution state corresponds to the initial response segment, with characteristic thresholds of axial temperature gradient > 2℃ / m, pressure rise rate > 0.1Pa / s, and moisture delay time > 5min. In this state, boron trifluoride gas mainly diffuses freely within the cylinder, with weak gas-solid interface interaction and axial non-uniform gas distribution. The wall traction state corresponds to the transition coupling segment, with characteristic thresholds of axial temperature gradient 0.1-2℃ / m, pressure rise rate 0.001-0.1Pa / s, and moisture delay time > 5min. When the concentration is in the peak range, a large number of gas molecules are adsorbed onto the inner wall of the cylinder. The traction effect of the wall on the gas molecules is dominant, the adsorption layer at the gas-solid interface gradually expands, and the adsorbed moisture on the inner wall begins to desorb in large quantities. The hysteresis release state corresponds to the stabilization identification section, whose characteristic thresholds are axial temperature gradient < 0.1℃ / m, pressure fluctuation amplitude < ±0.05Pa, and moisture concentration change rate < 0.02ppb / h. At this time, the adsorption at the gas-solid interface reaches dynamic equilibrium, the desorption rate is basically equal to the adsorption rate, the gas distribution is completely uniform, and the moisture change is dominated only by the irreversible hydrolysis reaction of boron trifluoride.

[0035] Finally, a one-to-one segmented mapping relationship is established between the characteristic threshold range, occurrence sequence, and state transition boundary conditions of each state (such as transitioning from a free distribution state to a wall traction state when the axial temperature gradient decays to 2℃ / m, and transitioning to a delayed release state when it decays to 0.1℃ / m) and physical quantities such as the uniformity of gas distribution inside the cylinder, the adsorption coverage of the inner wall, and the hydrolysis reaction activity, forming a mapping table containing state identifiers, characteristic thresholds, physical quantity mapping values, and corresponding compensation model call interfaces.

[0036] It should be noted that the determination of the free distribution state, wall traction state, and delayed release state is achieved through the joint use of multiple parameter thresholds, which include at least the axial temperature gradient, pressure change rate, and trace moisture change characteristics. When the above parameters fall into the corresponding threshold range, the corresponding state is determined. The transition between the states is achieved through threshold cross-triggered activation, thereby establishing a clear mapping relationship from data characteristics to physical states.

[0037] Step S105: Based on the mapping relationship, the effective volume of the gas cylinder, the adsorption behavior of the inner wall, and the gas-liquid conversion hysteresis characteristics are uniformly characterized to form a single-cylinder dynamic compensation model.

[0038] In this embodiment, the segmented mapping relationship generated in step S104, which includes state identifiers, feature thresholds, physical quantity mapping values, and state transition boundary conditions, is first deconstructed. The start and end time anchor points, state transition trigger thresholds, and mapping intervals of each physical quantity corresponding to the initial response segment, transition coupling segment, and stabilization identification segment are extracted. Based on the one-to-one correspondence between the gas distribution state and the inner wall action state, the complete response process inside the gas cylinder is labeled into three continuous and clearly defined independent segments: the free diffusion segment (corresponding to the free distribution state, accounting for 5%-10% of the total response time), the wall adsorption segment (corresponding to the wall traction state, accounting for 15%-20% of the total response time), and the dynamic equilibrium segment (corresponding to the delayed release state, accounting for 70%-80% of the total response time). At the same time, the relative position of each segment in the overall time series and the overlapping transition interval of adjacent segments (the width of the transition interval is 1-2 minutes) are recorded.

[0039] Subsequently, based on the segment calibration results, a multi-source coupled data component separation method combining Independent Component Analysis (ICA) and Partial Least Squares Path Modeling (PLS-PM) was employed to orthogonally decompose and extract independent components from the pressure, temperature, and trace moisture response data within each segment. In the free diffusion segment, since the gas-solid interface interaction is negligible, the pressure change is determined only by the gas thermal expansion effect and the effective volume of the cylinder. The amount of boron trifluoride standard material, the universal gas constant, and the measured average temperature of the cylinder were introduced as independent variables, and the measured pressure was the dependent variable. The pressure response component related to the effective volume of the cylinder was separated by linear regression using the ideal gas law, and the geometric effective volume of the cylinder was obtained through fitting. The volume V0 and the apparent volume deviation coefficient δv due to the temperature gradient are used. In the wall adsorption section, the pressure change is driven by the thermal expansion of the gas, the effective volume, and the physical adsorption of the inner wall. Therefore, the calibrated V0 and δv are substituted into the ideal gas law to calculate the theoretical pressure change caused by thermal expansion. The theoretical value is subtracted from the measured pressure sequence, and the remaining pressure deviation sequence is the adsorption response component related to the inner wall. The Langmuir adsorption isotherm that conforms to the monolayer adsorption characteristics of boron trifluoride on the inner wall of 316L stainless steel is used to fit this component to obtain the adsorption equilibrium constant Ka, saturated adsorption capacity qm, and adsorption rate constant kads, etc., which are characteristic parameters of the inner wall adsorption behavior.

[0040] In the dynamic equilibrium region, pressure changes are jointly dominated by the adsorption-desorption dynamic equilibrium and the metastable gas-liquid transition hysteresis. Therefore, by using the measured pressure-temperature hysteresis loop data in this region, and subtracting the pressure fluctuations caused by the adsorption-desorption equilibrium, the hysteresis response component related to phase transition is separated. The time constant τ for the pressure change lagging behind the temperature change, the hysteresis loop area A, and the critical phase transition pressure P are then fitted to obtain these components. cThe gas-liquid conversion hysteresis characteristic parameters are obtained. Then, according to the time sequence of free diffusion section → wall adsorption section → dynamic equilibrium section, the effective volume response sequence, adsorption response sequence, and hysteresis response sequence are recombined across sections. A cubic spline smoothing transition function is used to process the overlapping transition intervals of adjacent sections to eliminate parameter abrupt changes at the section boundaries. At the same time, the parameter transfer relationship between sections is established: V0 and δv output from the free diffusion section are used as correction parameters for calculating the adsorption response component of the wall adsorption section, and Ka and qm output from the wall adsorption section are used as background parameters for calculating the hysteresis response component of the dynamic equilibrium section. In this way, the alternating influence relationship between the effective volume of the gas cylinder, the inner wall adsorption behavior, and the gas-liquid conversion hysteresis characteristics is connected in series, forming a multi-stage continuous response chain with continuous parameter transfer and smooth boundary transition.

[0041] Finally, based on a multi-stage continuous response chain, the effective volume V0 of the gas cylinder, the apparent volume deviation coefficient δv, the adsorption characteristic parameters (ka, qm, kads), and the hysteresis characteristic parameters (τ, A, P) are analyzed. c As the core inherent parameters of the model, the start and end time anchor points, state transition trigger thresholds, and cubic spline transition function parameters of each section are used as the boundary control parameters of the model. A single-bottle dynamic compensation model is constructed in the form of a piecewise function. The model takes real-time pressure and temperature data during the filling process as input and the corrected actual filling mass as output. In the initial filling stage (corresponding to the free diffusion section), the effective volume parameter is used for basic mass calculation. In the middle filling stage (corresponding to the wall adsorption section), the adsorption response component is superimposed for first-order correction. In the later filling stage (corresponding to the dynamic equilibrium section), the hysteresis response component is further superimposed for second-order correction. At the same time, the boundary information and state transition sequence of each section are embedded into the automatic judgment logic of the model to realize the dynamic calling function of automatically switching the compensation formula according to the real-time working conditions during the filling process, thus generating a single-bottle dynamic compensation model.

[0042] It should be noted that the single-bottle dynamic compensation model is used to dynamically correct the actual filling mass of boron trifluoride in the cylinder during the filling process. It uses real-time pressure and temperature values ​​collected inside the cylinder as input variables, and combines these with the cylinder's effective volume parameters, inner wall adsorption parameters, and gas-liquid conversion hysteresis parameters to calculate the actual filling mass in stages. Specifically, in the initial filling stage, pressure and temperature are mapped to a base mass value based on ideal gas state relations. In the middle filling stage, an adsorption correction caused by inner wall adsorption behavior is superimposed on the base mass value. This adsorption correction is calculated based on the adsorption equilibrium constant and the relationship between saturated adsorption and temperature. In the later filling stage, a hysteresis correction caused by the gas-liquid conversion hysteresis effect is further superimposed. This hysteresis correction is determined based on the time delay of pressure change relative to temperature change and the hysteresis response characteristics. Through this staged superposition calculation, the output actual filling mass is made consistent with the actual mass of the contents inside the cylinder.

[0043] Step S2: Based on the single-cylinder dynamic compensation model, the cylinder is cooled to the target cryogenic temperature range in stages, and the cooling process is adjusted according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state.

[0044] The specific steps of step S2 are as follows: like Figure 2 As shown, step S201: Based on the single-bottle dynamic compensation model, read the effective volume characterization, inner wall adsorption behavior characterization and gas-liquid conversion hysteresis characterization corresponding to the current cylinder to be filled, and divide the cylinder cooling process into pre-cooling transition interval, intermediate acclimatization interval and deep cryogenic approximation interval according to the characterization results, and determine the starting conditions, switching conditions and dwell order of each interval.

[0045] In this embodiment, the first step is to read all the core characterization parameters of the single-bottle dynamic compensation model stored in the electronic tag of the cylinder to be filled using an RFID reader / writer. These parameters include effective volume characterization parameters (geometric effective volume V0, apparent volume deviation coefficient δv, volume temperature correction coefficient reference value fv), inner wall adsorption behavior characterization parameters (adsorption equilibrium constant ka0 at 25℃, saturated adsorption capacity qm, adsorption rate constant kads, adsorption enthalpy ΔHa), and gas-liquid conversion hysteresis characterization parameters (hysteresis time constant τ, hysteresis loop characteristic area A, critical phase change pressure P at -40℃). c The gas-liquid equilibrium enthalpy (ΔHv) was then used to divide the cylinder cooling process into intervals based on the influence weights of three types of characterization parameters. The pre-cooling transition interval corresponds to the stage dominated by the overall thermal inertia of the cylinder and is mainly controlled by the effective volume characterization parameter. The intermediate acclimatization interval corresponds to the stage dominated by the adsorption at the gas-solid interface on the inner wall and is mainly controlled by the adsorption behavior characterization parameter on the inner wall. The cryogenic approach interval corresponds to the stage dominated by the metastable gas-liquid transformation and is mainly controlled by the gas-liquid transformation lag characterization parameter. The irreversible residence order of the pre-cooling transition interval → intermediate acclimatization interval → cryogenic approach interval was determined, and jumping between intervals was prohibited.

[0046] Specifically, the initial conditions for the pre-cooling transition zone are set as follows: the initial average temperature of the cylinder ≥ 20℃ and the axial temperature gradient ≤ 5℃ / m. The switching conditions are determined by calculation using effective volume characterization parameters: when the average temperature of the cylinder drops to 15℃±1℃, the axial temperature gradient decays to ≤ 2℃ / m, and the volume temperature correction coefficient fv(T) = 1 + δv × (T - 298.15K) calculated based on δv fluctuates within ≤ ±0.0005 for 10 consecutive minutes, T represents the temperature, indicating that the pre-cooling transition phase is complete and the system switches to the next zone. The initial conditions for the relay acclimatization zone are that all switching conditions for the pre-cooling transition zone are met. The switching conditions are characterized by the adsorption behavior of the inner wall. Parameter calculation and determination: The van der Hoff equation was used to calculate the adsorption equilibrium constant at real-time temperature. When the average temperature of the cylinder drops to -20℃±2℃, the real-time adsorption amount q≥0.85qm, and the hourly change rate of adsorption amount within 15 minutes is ≤0.01qm / h, it is determined that the adsorption layer on the inner wall has reached a stable acclimatization state and can be switched to the next interval. The starting condition of the cryogenic approach interval is that all the switching conditions of the intermediate acclimatization interval are met, and the ending condition is determined by the calculation of gas-liquid conversion hysteresis characterization parameters: The Antoni equation was used to calculate the saturated vapor pressure Psat(T) of boron trifluoride at real-time temperature, and the gas-liquid conversion hysteresis pressure deviation ΔP=Psat(T)-P was calculated. c When the average temperature of the cylinder drops to the target filling temperature of -40℃±0.5℃, the axial temperature gradient further decreases to ≤0.5℃ / m, and the hysteresis pressure deviation ΔP≤0.01P c If the average temperature fluctuation of the cylinder is ≤ ±0.2℃ and the internal pressure fluctuation is ≤ ±0.001MPa within 30 minutes, the cryogenic approach stage is considered complete. If the switching conditions of any interval are not met within the preset maximum residence time (maximum residence time of 1h for precooling transition interval, 6h for intermediate acclimatization interval, and 12h for cryogenic approach interval), an abnormal alarm will be automatically triggered, indicating that the cylinder may have internal wall contamination, abnormal heat conduction, or gas-liquid conversion characteristics deviating from the normal range. Finally, a personalized cooling process control document for the single cylinder will be generated, which includes the precise temperature range of each interval, switching trigger threshold, maximum residence time, and abnormal handling rules.

[0047] Figure 2The specific implementation of the staged cryogenic cooling control process is illustrated. Specifically, the cylinder to be filled enters a pre-cooling transition zone during the initial cooling phase. Temperature and pressure are simultaneously collected by multiple temperature monitoring units (T1 to T5) positioned at different heights on the cylinder's outer wall and by internal pressure monitoring units. It can be observed that the temperature at each measuring point decreases rapidly over time with significant gradient differences, with the lower region cooling at a higher rate than the upper region, while the internal pressure decreases slowly. Once the temperature gradient becomes more gradual and meets preset switching conditions, the cylinder enters a relay acclimatization zone. In this phase, the cooling rate is reduced and intermittent cold energy input is introduced. The process begins by gradually converging the temperature curves at each measuring point, transforming the temperature distribution from a non-uniform state to a relatively balanced state, while simultaneously stabilizing the pressure change. This indicates a reorganization of the gas distribution and the state of the inner wall within the cylinder. Subsequently, when the temperature distribution further converges and the pressure response meets the switching conditions, the cylinder enters the cryogenic approach zone. During this stage, a small-scale, fine-control method is used to continue reducing the temperature. The temperature curves at each measuring point gradually approach the target cryogenic temperature, and the differences between them further decrease. The pressure inside the cylinder decreases synchronously and tends to stabilize, ultimately reaching the set cryogenic temperature zone at the target time point.

[0048] Step S202: Perform the first stage of cooling on the cylinder to be filled according to the pre-cooling transition zone, so that the gas inside the cylinder enters the initial cooling state coordinated with the action of the inner wall. During the first stage of cooling, collect the temperature distribution and pressure change sequence at different positions on the outer wall of the cylinder to be filled, and compare the collected results with the corresponding section in the single-cylinder dynamic compensation model for the first time.

[0049] In this embodiment, the cylinder to be filled is first vertically fixed in a 316L stainless steel jacketed low-temperature circulating bath, ensuring that the jacket completely encloses the cylinder body. The valve and neck connection are locally insulated with closed-cell polyurethane insulation cotton with a thickness of not less than 50mm to prevent localized overcooling and seal failure. The jacketed low-temperature circulating bath uses a 30% (v / v) ethylene glycol aqueous solution as the refrigerant, with a refrigerant circulation flow rate set to 15L / min. The temperature difference between the refrigerant inlet and outlet is controlled within ±0.3℃ to ensure temperature field uniformity. Then, the effective volume characterization parameters in the single-cylinder dynamic compensation model are read, and the corresponding initial cooling rate is matched according to the cylinder's geometric effective volume V0: when V0≤40L... The cooling rate is set to 1.5℃ / min; when 40L < V0 ≤ 80L, the cooling rate is set to 1.0℃ / min; when V0 > 80L, the cooling rate is set to 0.5℃ / min. The low-temperature circulating bath is started to cool the cylinder at a uniform rate according to the set rate. During the cooling process, the cylinder is kept in the same sealed and isolated state as in step S101 and is not connected to any external gas source or pipeline. During the first cooling stage, five platinum resistance temperature sensors arranged at equal intervals along the outer wall of the cylinder collect temperature distribution data at different locations in real time at a sampling frequency of 1Hz. At the same time, a capacitive thin-film vacuum gauge directly connected to the cylinder collects pressure change data inside the cylinder at a sampling frequency of 2Hz.

[0050] During the cooling process, the axial temperature gradient of the cylinder is calculated in real time (i.e., the difference between the maximum and minimum temperatures measured by the five temperature sensors divided by the effective height of the cylinder). When the axial temperature gradient exceeds 3℃ / m, the cooling rate is automatically reduced by 50% until the axial temperature gradient falls below 2℃ / m, at which point the original set rate is restored to prevent the cylinder from deforming due to excessive thermal stress or premature formation of an adsorption layer on the inner wall. When the arithmetic mean of the temperatures at the five locations on the cylinder drops to the target temperature of 15℃±1℃ in the pre-cooling transition zone, and the average temperature fluctuation within 10 minutes does not exceed ±0.2℃ and the axial temperature gradient does not exceed 2℃ / m, the first stage of cooling is stopped and the data comparison stage begins. The temperature distribution sequence and pressure change sequence of the entire first stage of cooling are automatically processed, and characteristic parameters such as the average cooling rate, maximum axial temperature gradient, pressure-temperature linear correlation coefficient, and average time difference between pressure change and temperature change are extracted. The measured pressure-temperature synchronous change curve is compared with the free diffusion section in the single-cylinder dynamic compensation model based on the effective volume. The theoretical pressure-temperature response curves calculated from V0 and the apparent volume deviation coefficient δv are compared point-by-point synchronously, using the root mean square error (RMSE) as the comparison criterion: if the RMSE ≤ 0.002 MPa, the cylinder's thermal inertia and effective volume characteristics are consistent with the model's characterization, the first round of comparison is qualified, and it can enter the intermediate training zone; if 0.002 MPa < RMSE ≤ 0.005 MPa, a slight systematic deviation is determined, and the apparent volume deviation coefficient δv in the model is automatically iteratively corrected based on the average deviation between the measured pressure and the theoretical pressure. After correction, the theoretical curve is recalculated and compared again until the RMSE meets the qualification requirements; if the RMSE > 0.005 MPa, a significant abnormality is determined in the cylinder, and a pre-cooling abnormality alarm is automatically triggered, the subsequent cooling process is stopped, and the cylinder is transferred to the re-inspection station to check for problems such as cylinder deformation, internal residual foreign matter, sensor installation deviation, or electronic tag parameter reading errors. All comparison results and correction parameters are automatically written into the cylinder's RFID electronic tag storage.

[0051] Step S203: When the first round of comparison results meet the preset switching conditions, the cylinder to be filled is transferred to the intermediate acclimatization zone. In this zone, the cold energy input is promoted in the order of slow cooling followed by fast cooling. Based on the migration direction of temperature distribution and the stage transition of pressure change, the correspondence between the gas distribution state inside the cylinder to be filled and the action state of the inner wall is calibrated for the second time.

[0052] In this embodiment, when the first round of comparison results in step S202 shows that the root mean square error of the pressure-temperature response curve is ≤0.002MPa or meets the requirement after iterative correction, the 5-point axial temperature distribution sequence, pressure change sequence, average cooling rate, maximum axial temperature gradient, corrected apparent volume deviation coefficient δv, and corresponding high-precision time axis information collected throughout the pre-cooling transition interval are automatically used as the preceding freezing segment and written into the temporary storage area of ​​the gas cylinder RFID electronic tag in binary format. The time end point of the freezing segment is the switching time of the pre-cooling transition interval, which serves as the absolute time zero point of the relay acclimatization interval.

[0053] Subsequently, multiple rounds of cold energy delivery were carried out in the relay acclimatization zone, employing an alternating rhythm of gradient cooling and isothermal residence. A total of three progressive cold energy delivery rounds were set: In the first round, the refrigerant temperature was reduced from 15℃ to 5℃ at a cooling rate of 0.8℃ / min, with the refrigerant circulation flow rate maintained at 15L / min, and the temperature was maintained at an isothermal residence for 30 minutes after reaching the target temperature; In the second round, the refrigerant temperature was reduced from 5℃ to -5℃ at a cooling rate of 0.5℃ / min, and the temperature was maintained at an isothermal residence for 60 minutes after reaching the target temperature; In the third round, the refrigerant temperature was reduced from -5℃ to -20℃ at a further cooling rate of 0.3℃ / min, and the temperature was maintained at an isothermal residence for 120 minutes after reaching the target temperature. This rhythm forced the gas inside the cylinder to alternate between continuous contraction and intermittent residence, allowing the adsorption layer on the inner wall to gradually nucleate and grow stably, while exposing the dynamic misalignment characteristics between gas distribution and inner wall adsorption.

[0054] During the cooling phase of each round of cold energy delivery, the sampling frequency of five platinum resistance temperature sensors was maintained at 1Hz, and the sampling frequency of the capacitive thin-film vacuum gauge was maintained at 2Hz. During the isothermal residence phase, the sampling frequency of both types of sensors was uniformly adjusted to 0.2Hz to capture the slow adsorption kinetics. The axial temperature gradient evolution curve, pressure change rate curve, and real-time adsorption rate curve derived based on the ideal gas law were calculated in real time for each round. The data from each cooling phase, residence phase, and corresponding characteristic parameters were integrated into an independent response subgroup. The three response subgroups were combined in chronological order to form a progressive response group for the relay acclimatization phase. The time of the progressive response group was then... The inter-axis is precisely aligned with the time endpoint of the preceding freezing phase, forming a continuous time series from the start of pre-cooling to the current moment of intermediate acclimatization. The cross-correlation analysis method is used to calculate the cross-correlation coefficient between the gas average temperature change and the outer wall average temperature change, and to determine the leading time difference of gas temperature relative to outer wall temperature (2-5s in this embodiment). The cross-correlation coefficient between pressure change and gas average temperature change is calculated, and to determine the lag time difference of pressure change relative to gas temperature (10-30s in this embodiment). The pressure return amplitude (i.e., the maximum value of pressure rising from the lowest point) within the first 10 minutes of each round of isothermal dwell phase is extracted as the characteristic parameter of the return change.

[0055] Based on the order of occurrence of various changes in the time series, the sequence is rearranged into a state misalignment chain: outer wall temperature decreases first → axial temperature gradient is established → internal gas temperature decreases with a lag → pressure contracts rapidly and decreases → cold energy delivery stops → outer wall temperature recovers slightly → internal gas temperature recovers with a lag → pressure recovers slightly → internal wall adsorption dominates and pressure slowly decays. This chain clearly characterizes the temporal misalignment relationship and causal transmission order between the gas distribution state and the internal wall action state. Based on the characteristic parameters of the state misalignment chain, a nonlinear least squares method is used to compare the measured three-round pressure change curves with those based on Langmuir adsorption. The theoretical isothermal curves were fitted to perform a secondary calibration of the original inner wall adsorption behavior characterization parameters in the single-bottle dynamic compensation model. The adsorption equilibrium constant Ka, adsorption rate constant kads, and adsorption enthalpy ΔHa were corrected. At the same time, based on the measured time point when the inner wall adsorption amount reached 0.85qm, the transition boundary between the wall adsorption section and the dynamic equilibrium section was recalibrated. The state transition threshold in the original segmented mapping relationship was adjusted to be consistent with the measured value. The recalibrated adsorption parameters and segmented boundary information were incorporated into the single-bottle dynamic compensation model, covering the original corresponding parameters to complete the secondary calibration.

[0056] When the third round of isothermal residence phase ends, if the measured adsorption amount is ≥0.85qm and the adsorption rate is ≤0.01qm / h for 15 consecutive minutes, the intermediate acclimatization is considered complete and the process can proceed to the cryogenic approach zone. If the above conditions are not met, the third round residence time will be automatically extended by 30 minutes, up to a maximum of 2 times. If the conditions are still not met, an adsorption anomaly alarm on the inner wall will be triggered.

[0057] Step S204: Based on the secondary calibration, the cylinder to be filled is transferred into the cryogenic approach zone. Using the single-cylinder dynamic compensation model as a reference, the local temperature difference shift, pressure response misalignment, and condensation propulsion sequence changes that occur during the cooling process are identified segment by segment. When any segment is found to deviate from the preset trajectory, the cooling rhythm of the current zone, the zone dwell sequence, or the connection point of adjacent zones is adjusted to correct the condensation propulsion path inside the cylinder to be filled.

[0058] In this embodiment, when step S203 completes the secondary calibration and the measured adsorption amount on the inner wall is ≥0.85qm and the adsorption rate for 15 minutes is ≤0.01qm / h, the cylinder to be filled is automatically transferred to the cryogenic approach zone. The refrigerant in the jacketed low-temperature circulating bath is replaced with a 50% volume fraction ethanol aqueous solution (freezing point ≤-60℃), the refrigerant circulation flow rate is adjusted to 12L / min, the temperature difference between the refrigerant inlet and outlet is controlled within ±0.2℃, and a segmented decreasing cooling program is started: the first segment lowers the refrigerant temperature from -20℃ to -30℃ at a rate of 0.2℃ / min, and then remains at a constant temperature after reaching the target temperature. The first cooling stage involves cooling to -35℃ at a rate of 0.1℃ / min for 30 minutes; the second stage involves cooling to -35℃ at a rate of 0.1℃ / min and holding at that temperature for 45 minutes; the third stage involves cooling to the target filling temperature of -40℃ at a rate of 0.05℃ / min and holding at that temperature for 60 minutes. This cooling rhythm gradually induces boron trifluoride gas to condense uniformly from the bottom of the cylinder upwards, avoiding gas-liquid two-phase stratification and sudden pressure changes caused by localized rapid condensation. Throughout the cryogenic approach range, five platinum resistance temperature sensors, evenly spaced along the axial direction of the cylinder's outer wall, collect real-time temperature data at the cylinder shoulder, upper cylinder, middle cylinder, lower cylinder, and bottom positions. The sampling frequency is 0.5Hz. A capacitive thin-film vacuum gauge collects gas phase pressure data inside the bottle at a sampling frequency of 1Hz. All data are synchronously compared point-by-point with the theoretical temperature distribution curve, theoretical pressure response curve, and gas-liquid conversion hysteresis curve of the dynamic equilibrium section in the single-bottle dynamic compensation model. The comparison time window is set to 5 minutes, and the sliding step size is 1 minute. Three typical deviations are identified in real time: the first is local temperature difference deviation, which is determined to be a local temperature difference deviation when the measured temperature difference between any two adjacent axial temperature sensors exceeds 1℃, or the axial temperature gradient exceeds 0.8℃ / m. This deviation is usually caused by uneven refrigerant flow field or uneven cylinder wall thickness; Second, pressure response misalignment. When the time difference between the measured pressure and the model-predicted pressure exceeds 120% of the lag time constant τ in the single-cylinder dynamic compensation model, or when the deviation rate between the measured pressure and the theoretical pressure exceeds 3%, it is judged as pressure response misalignment. This deviation is caused by individual differences in the gas-liquid conversion lag effect. Third, change in condensation propulsion sequence. When the temperature drop rate of the upper cylinder exceeds that of the lower cylinder by 0.1℃ / min, or when a condensation pressure inflection point appears at the cylinder shoulder before the cylinder bottom, it is judged as condensation propulsion sequence abnormality. This abnormality will cause the upper part to condense first to form a liquid seal, hindering the condensation of the lower gas and causing filling errors.

[0059] Differentiated correction strategies are implemented for different types of deviations: When a local temperature difference deviation is detected, the current cooling rate is immediately reduced by 50%, while the refrigerant circulation flow rate is increased by 20%, until the local temperature difference falls back to within 0.5℃, at which point the original cooling rhythm is resumed; when a pressure response misalignment is detected, cooling is immediately paused, the isothermal dwell time of the current stage is extended by 20 minutes, and the hysteresis time constant τ in the model is iteratively corrected based on the measured pressure deviation, with the correction range controlled within ±3%, until the pressure deviation rate drops below 1%, at which point cooling continues; when an abnormal condensation propagation sequence is detected, the temperature difference between the refrigerant inlet and outlet is immediately adjusted. Adjust the temperature to within 0.1℃, and temporarily wrap 30mm thick insulation cotton around the shoulder of the cylinder to reduce the upper cooling input rate. Force the condensation process to resume its normal upward sequence from the bottom of the cylinder. Once the lower cylinder temperature is at least 0.5℃ lower than the upper cylinder temperature and remains so for 10 minutes, remove the temporary insulation cotton and restore the original cooling parameters. After each correction, re-collect 10-minute data points for verification. If the deviation is eliminated, continue with the original cooling procedure. The cooling process continues when the arithmetic mean of the temperatures at the five locations on the cylinder stabilizes at -40℃ ± 0.5℃, the axial temperature gradient is ≤ 0.5℃ / m, and the internal pressure stabilizes at the critical phase transition pressure P specified in the model calibration. c When the pressure is ±0.001MPa and the fluctuation range of all parameters meets the above requirements within 30 minutes, the cryogenic approach range is considered complete, and all corrected gas-liquid conversion lag parameters and condensation propulsion path data are automatically incorporated into the single-bottle dynamic compensation model.

[0060] Step S205: After completing the cryogenic approximation zone, perform a closed-loop check on the cylinder to be filled. When the check result is consistent with the target state sequence corresponding to the single-cylinder dynamic compensation model, determine that the cylinder has entered the gas-liquid condensation state required for filling.

[0061] In this embodiment, when the cumulative isothermal dwell time of the third segment of the cryogenic approach interval reaches 60 minutes and all parameters meet the interval termination conditions, the current cooling process is terminated. All collected data within the last 60 minutes of the cryogenic approach interval are automatically extracted, including the platinum resistance temperature sequence at 5 positions along the cylinder axis, the gas phase pressure sequence inside the cylinder collected by the capacitive thin-film vacuum gauge, all cooling rhythm adjustment records within the interval, deviation identification and correction records, and state transition timestamps. After performing SHA-256 hash verification on the above data, it is sealed in encrypted binary format as the final response group and written into the storage area of ​​the cylinder's RFID electronic tag, while generating a unique data identifier.

[0062] Subsequently, the target state sequence, generated after secondary calibration, was extracted from the single-cylinder dynamic compensation model. This sequence contains four consecutive target state nodes: the end node of the first stage of cryogenic cooling (corresponding to refrigerant temperature -30℃, cylinder average temperature -29.5℃±0.5℃, axial temperature gradient ≤0.8℃ / m, pressure deviation ≤±0.002MPa), the end node of the second stage of cryogenic cooling (corresponding to refrigerant temperature -35℃, cylinder average temperature -34.5℃±0.5℃, axial temperature gradient ≤0.6℃ / m, pressure deviation ≤±0.0015MPa), and the end node of the third stage of cryogenic cooling (corresponding to refrigerant temperature -40℃, cylinder average temperature -39.5℃±0.5℃, axial temperature gradient ≤0.6℃ / m, pressure deviation ≤±0.0015MPa). The parameters are: ≤0.5℃ / m, pressure deviation ≤±0.001MPa), and final gas-liquid equilibrium node (corresponding to constant temperature for more than 30 minutes, average cylinder temperature fluctuation ≤±0.2℃, pressure fluctuation ≤±0.0005MPa, axial temperature gradient ≤0.3℃ / m). Each node is arranged in chronological order and is irreversible. The Dynamic Time Warping (DTW) algorithm is used to align the final response group with the target state sequence segment by segment. The global constraint window of DTW is set to 10 minutes, the local step size constraint is 1:2, and the matching similarity threshold is 0.95. The algorithm automatically aligns the corresponding start point, connection order, and termination boundary of each response segment with the target state node, eliminating the time axis scaling deviation caused by the adjustment of the cooling rhythm.

[0063] Based on the segment-by-segment alignment results, three typical deviations are identified: first, sequence offset, which is determined when the deviation between the measured arrival time and the theoretical arrival time of any target state node exceeds 15 minutes; second, boundary misconnection, which is determined when the measured parameter jump at the junction of two adjacent target state nodes exceeds 20% of the corresponding node threshold; and third, state return, which is determined when the average temperature of the cylinder rises by more than 0.5℃ or the gas phase pressure rises by more than 0.002MPa for more than 5 minutes at any given time. For state return; based on the return judgment results, combined with the pressure-temperature hysteresis curve and axial temperature distribution evolution curve in the final response group, Raoult's law combined with the gas-liquid two-phase equilibrium thermodynamic model is used to calculate the current volume ratio of liquid boron trifluoride in the cylinder (target value 5%-8%), the axial distribution uniformity of the liquid phase layer (target value ≥90%), and the proportion of remaining active sites in the inner wall adsorption layer (target value ≤10%). This is used to verify the current gas distribution state and the inner wall interaction state, forming a current state sequence containing 4 consecutive measured state nodes.

[0064] The current state sequence is compared with the target state sequence. The verification is considered successful when all of the following conditions are met: the state arrangement order is completely consistent with the target sequence, with no reversed or missing nodes; the temperature deviation of each measured state node is ≤ ±0.3℃, the pressure deviation is ≤ ±0.001MPa, and the axial temperature gradient deviation is ≤ ±0.2℃ / m; the duration deviation of each state node is ≤ ±10%; the liquid phase volume ratio, liquid phase layer uniformity, and the proportion of remaining active sites on the inner wall are all within the target range. If the verification is successful, the cylinder to be filled is determined to have reached the gas-liquid condensation state required for filling, and a filling permit is automatically generated and written to the electronic tag. If there is a slight deviation (such as the deviation of a single node parameter being between 1 and 1.5 times the threshold), the constant temperature residence time is automatically extended by 30 minutes and a closed-loop verification is performed again. If there is a serious deviation (such as sequence offset, boundary misconnection, or state reversion), an abnormal alarm before filling is triggered.

[0065] Step S3: Under the target cryogenic temperature range, a staged filling strategy is adopted to fill the cylinder with boron trifluoride, including initial constant pressure filling, mid-stage pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model.

[0066] like Figure 3 As shown, the specific steps of step S3 are as follows: Step S301: Introduce boron trifluoride into the cylinder to be filled under constant pressure in the target cryogenic temperature zone, and collect the mass change, internal pressure change and corresponding state transition records of the cylinder to be filled. Then, perform interval mapping between the collected results and the single-cylinder dynamic compensation model.

[0067] In this embodiment, after the cylinder to be filled has been processed and entered the target cryogenic temperature zone and maintained in a stable state, the cylinder is connected to the boron trifluoride gas supply source through a filling pipeline. The pressure at the gas supply end is adjusted to a preset constant pressure range through a pressure reducing valve or a mass flow control device, so that boron trifluoride is introduced into the cylinder at a basically constant pressure. During the introduction process, the cylinder is placed on a weighing device to acquire mass change data in real time. At the same time, the pressure change data inside the cylinder is acquired through a pressure detection unit set at the cylinder end. The collected data is synchronously recorded in combination with time series to form a continuous mass change sequence and pressure change sequence. Based on this, the collected data is divided according to a preset time window and compared segment by segment with the reference sequence of the corresponding initial filling interval in the single-cylinder dynamic compensation model. By identifying the mass growth rate change node and the pressure response inflection node, the current stage of the filling process is mapped to the corresponding interval position in the model, thereby obtaining an interval correspondence relationship that matches the current state of the cylinder.

[0068] Step S302: When the preset interval boundary is reached, switch to pulse feeding to connect the quality change and pressure recovery process of adjacent pulse cycles to form a continuous feeding sequence.

[0069] In this embodiment, when the current filling process reaches the preset interval boundary based on the rate of mass change and the trend of pressure change inside the cylinder, the original constant pressure continuous feeding method is switched to an intermittent pulse feeding method by controlling the solenoid valve or pneumatic valve on the filling pipeline. Each pulse cycle includes a feeding phase with the air inlet valve open and a pausing phase with the air inlet valve closed. During the feeding phase, boron trifluoride enters the cylinder under the pressure at the gas supply end, causing the cylinder mass to increase in a stepwise manner, and the pressure inside the cylinder rises accordingly. During the pausing phase, the air intake is stopped and the cylinder is kept sealed, allowing the pressure inside the cylinder to rise. The gas undergoes pressure drop and internal redistribution. Pressure recovery process data is continuously collected by a pressure sensor, and the mass change tends to stabilize during this stage by a weighing device. The mass increment data of the feeding stage in adjacent pulse cycles and the pressure recovery curve of the pausing stage are spliced ​​in chronological order to form a continuous time series of mass advancement and pressure return processes between each cycle. The spliced ​​sequence is then aligned with the reference response characteristics of the corresponding interval in the single-bottle dynamic compensation model to obtain a continuous feeding sequence that reflects the current feeding advancement path.

[0070] Step S303: Connect the mid-stage feeding sequence with the initial stage filling sequence, identify the inter-stage offset of quality advancement and state transition, and adjust the feeding rhythm and switching timing based on the single-bottle dynamic compensation model.

[0071] In this embodiment, after completing the mid-stage pulse feeding to form a continuous feeding sequence, the sequence is aligned with the mass change sequence and bottle pressure change sequence recorded during the initial constant pressure filling process. Specifically, the start time of the mid-stage feeding sequence is shifted and matched using the end time of the initial filling as the docking reference point. By comparing the mass increment change trend and pressure response transition characteristics of the two sequences in adjacent intervals, the sudden change nodes, hysteresis nodes, and abnormal fluctuation intervals in the pressure recovery process are identified, thereby determining the offset between segments in terms of mass advance rate and state transition sequence. Based on this, the reference response sequence of the corresponding mid-stage interval in the single-bottle dynamic compensation model is called, and the current actual sequence is aligned with the reference sequence segment by segment. According to the alignment results, the start duration, stop interval, and switching time between adjacent pulses of pulse feeding are adjusted so that the subsequent feeding process gradually approaches the reference trajectory of the corresponding interval in the model in terms of time rhythm and advance path, and provides a corrected state sequence basis for entering the final pressure differential adjustment stage.

[0072] Step S304: After completing the inter-segment correction, the remaining filling process is advanced by adjusting the pressure difference between the gas supply end and the cylinder end, so that the filling amount is close to the target value, and the corresponding quality change nodes are recorded until the filling is completed.

[0073] In this embodiment, after completing the inter-segment correction for mid-stage refueling, the filling process is switched to the final stage of differential pressure adjustment. By adjusting the pressure reducing device at the gas supply end and the opening of the filling valve, a gradually changing differential pressure sequence is formed between the gas supply end pressure and the pressure inside the cylinder, driving boron trifluoride to continue being introduced into the cylinder. Specifically, a small differential pressure is maintained in the initial stage to control the introduction rate. Subsequently, based on the rate of change of cylinder mass and the trend of pressure change inside the cylinder, the gas supply pressure is adjusted in stages or intermittently fine-tuned, so that the differential pressure changes in a step-like or gradually changing manner, thereby achieving the control of the remaining filling process. The process is refined and advanced. During this process, the mass change of the cylinder is continuously recorded by a weighing device, and the pressure data inside the cylinder is collected simultaneously. The mass increment node and pressure response node corresponding to each pressure difference adjustment are marked in time series form. At the same time, the node sequence is compared with the reference node of the final target interval in the single-cylinder dynamic compensation model. When it is identified that the mass increment is gradually decreasing and the pressure change tends to stabilize, the pressure difference adjustment amplitude is further reduced until the cylinder mass reaches the termination node corresponding to the target filling amount, and the termination determination of the filling process is completed accordingly.

[0074] Figure 3 This paper illustrates the specific implementation of a phased filling strategy and dynamic adjustment process. Specifically, after the cylinder to be filled completes cryogenic treatment and enters the filling state, it first enters the initial constant-pressure filling stage. By adjusting the gas supply pressure to a preset constant-pressure range, boron trifluoride is continuously introduced into the cylinder. At this time, the gas supply pressure is higher than the cylinder pressure, and the cylinder pressure gradually increases with filling. Simultaneously, the cylinder mass shows a continuous increasing trend, corresponding to a state sequence that gradually progresses from the initial state S0 to S2. When the preset range boundary is reached, the middle-stage pulse-type replenishment stage begins. In this stage, pulse gas supply is achieved by periodically opening and closing the inlet valve, causing the cylinder pressure to fluctuate periodically. That is, the pressure rises during the replenishment stage and decreases during the stop stage. During the stabilization phase, the pressure drops and internal redistribution occurs, while the cylinder mass gradually increases in a stepwise manner, corresponding to the state nodes progressing sequentially to S6. Subsequently, after the intermediate feeding advances to the target range, the final stage of differential pressure adjustment begins. By gradually reducing the pressure difference between the gas supply end and the cylinder end, the gas is introduced into the cylinder with a lower driving force, the pressure change inside the cylinder tends to be gradual, and the rate of cylinder mass increase gradually decreases and approaches the target filling amount. During this process, each state node (S7 to Sn) is continuously recorded until the termination node corresponding to the target filling amount is reached at the termination time, thus completing the entire phased filling and dynamic adjustment process.

[0075] Step S4: When the cylinder to be filled meets the preset consistency conditions, stop filling and seal it; otherwise, perform compensatory filling or backfilling for correction.

[0076] In this embodiment, after the final pressure differential adjustment is completed, the final value consistency judgment is performed on the cylinder to be filled. Specifically, the mass value of the cylinder at the moment the gas intake stops, the internal equilibrium pressure in the closed state, and the pressure change value after a preset time of rewarming are jointly collected and compared according to a preset judgment interval. The judgment interval can be set based on gas properties and filling standards, through experimental data or empirical values, for example, by establishing a reference window based on the correspondence between mass and pressure after multiple batches of cylinder filling. When the above three parameters fall within the corresponding window range, the intake valve and isolation valve are closed, the cylinder is sealed, and the final filling data is recorded. When any parameter does not fall within the corresponding range, compensation filling or back-pull correction is performed according to the direction of deviation. Compensation filling introduces a small amount of gas by briefly opening the intake valve to improve mass or pressure, while back-pull correction releases a small amount of gas by connecting the recovery pipeline to reduce the corresponding parameter value. After each adjustment, the joint collection and comparison of mass and pressure are repeated until the preset consistency condition is met, thereby completing the correction and confirmation of the final value of cylinder filling.

[0077] It should be noted that the preset consistency condition is a multi-parameter joint judgment standard used to determine whether the final filling value meets the requirements. It includes at least the correspondence between the mass value of the cylinder after filling, the internal equilibrium pressure of the cylinder in the closed state, and the pressure correction value after returning to the set environmental conditions. Specifically, based on the property change law of boron trifluoride in the target cryogenic temperature zone and during the reheating process, the mass and pressure data of the same type of cylinder under different filling batches are calibrated to form a correlation interval window between mass, pressure and reheating pressure. When the measured mass of the cylinder to be filled falls within the target mass range, and its corresponding internal equilibrium pressure and reheating pressure are both within this correlation interval and remain consistent with the state sequence corresponding to the single-cylinder dynamic compensation model, it is determined that the preset consistency condition is met.

[0078] Example 2 Please see Figure 2 Another embodiment of the present invention provides: a cryogenic precision control filling system for boron trifluoride for semiconductors, comprising: a parameter compensation module, a state adjustment module, a filling control module, and a filling determination module; The parameter compensation module is used to collect the pressure, temperature, trace moisture and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and to establish a single-cylinder dynamic compensation model based on the collected parameters. The state adjustment module is used to cool the cylinder to the target cryogenic temperature range in stages based on the single-cylinder dynamic compensation model, and to adjust the cooling process according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state. The filling control module is used to fill boron trifluoride into the cylinder to be filled in the target cryogenic temperature zone using a staged filling strategy, including initial constant pressure filling, middle pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model. The filling determination module is used to stop filling and seal the cylinder when it meets the preset consistency conditions; otherwise, it performs compensation filling or backfilling correction.

[0079] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0080] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for cryogenic precision control of boron trifluoride filling for semiconductors, characterized in that, include: The pressure, temperature, trace moisture and heat exchange response parameters of the boron trifluoride cylinder to be filled were collected, and a single-cylinder dynamic compensation model was established based on the collected parameters. Based on the single-cylinder dynamic compensation model, the cylinder is cooled to the target cryogenic temperature range in stages, and the cooling process is adjusted according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state. Under the target cryogenic temperature range, a staged filling strategy is adopted to fill the steel cylinder to be filled with boron trifluoride, including initial constant pressure filling, middle pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model. When the cylinder to be filled meets the preset consistency conditions, filling is stopped and the cylinder is sealed; otherwise, compensation filling or backfilling correction is performed.

2. The method for cryogenic precision control of boron trifluoride filling for semiconductors as described in claim 1, characterized in that, The process involves collecting pressure, temperature, trace moisture, and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and establishing a single-cylinder dynamic compensation model based on the collected parameters, including: Vacuum treatment is performed on the steel cylinder to be filled and its connecting pipeline until the preset baseline pressure is reached. The steel cylinder is then sealed in an isolated state, and the initial pressure change trajectory and temperature stability range in the isolated state are recorded. A controlled disturbance was created by introducing a limited amount of boron trifluoride gas into the cylinder, and transient change sequences of pressure, temperature and trace moisture inside the cylinder were collected simultaneously during the introduction process. After the disturbance ends, the cylinder remains closed, and the pressure recovery process, temperature drop process, and trace moisture change process inside the cylinder are continuously collected and the corresponding time correlation sequences are recorded to form a multi-source coupled response dataset. The multi-source coupled response dataset is segmented according to a preset time window, and the correlation features of pressure change trend, temperature gradient evolution and moisture migration path in each segment are extracted. A mapping relationship between the gas distribution state inside the cylinder and the action state of the inner wall is established. Based on the aforementioned mapping relationship, the effective volume of the gas cylinder, the adsorption behavior of the inner wall, and the hysteresis characteristics of gas-liquid conversion are uniformly characterized to form a single-cylinder dynamic compensation model.

3. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 2, characterized in that, The multi-source coupled response dataset is segmented according to a preset time window. Correlation features of pressure change trends, temperature gradient evolution, and moisture migration paths within each segment are extracted. A mapping relationship is established between the gas distribution state inside the cylinder and the state of action on the inner wall, including: The continuous sampling sequences of pressure, temperature and trace moisture in the multi-source coupled response dataset are synchronously compared, and the multi-source coupled response dataset is divided into the initial response segment, the transition coupling segment and the stabilization identification segment. The monotonic change trajectory of pressure, the gradient migration trajectory of temperature, and the delayed following trajectory of trace moisture were extracted for each segment. All trajectories were rearranged according to their sequential occurrence within the same segment to form the dominant response chain of the corresponding segment. The dominant response chains of each segment are connected in series according to the connection relationship between the beginning and end of adjacent segments to form a group of related features; Based on the aforementioned associated feature group, the internal response process of the gas cylinder is divided into a free distribution state, a wall traction state, and a delayed release state. The order of occurrence and the transition boundary of each state in each segment are mapped to the internal gas distribution state and the internal wall action state of the gas cylinder, forming a segmented mapping relationship.

4. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 3, characterized in that, Based on the aforementioned mapping relationship, the effective volume of the gas cylinder, its inner wall adsorption behavior, and the gas-liquid conversion hysteresis characteristics are uniformly characterized to form a single-cylinder dynamic compensation model, including: The segmented mapping relationship is deconstructed, and the internal response process of the gas cylinder is segmented according to the correspondence between the gas distribution state and the internal wall action state in each segment, so as to obtain the start and end boundaries of each segment and their positional correlation in the time series. Based on the calibration results of the aforementioned sections, the response data within each section are separated and processed. The pressure response component related to the change in the effective volume of the gas cylinder, the adsorption response component related to the action of the inner wall, and the hysteresis response component related to the phase transition are extracted respectively, and independent response sequences within the corresponding sections are established. According to the chronological relationship of each segment in the time series, the independent response sequences are recombined across segments to connect the alternating influence relationship between the effective volume of the gas cylinder, the adsorption behavior of the inner wall and the gas-liquid conversion hysteresis characteristics, forming a multi-stage continuous response chain. Based on the multi-stage continuous response chain, the effective volume of the cylinder, the adsorption behavior of the inner wall and the gas-liquid conversion hysteresis characteristics are uniformly merged, and the boundary information of each section and its transfer order are embedded into it to generate a single-cylinder dynamic compensation model for dynamic calling during the filling process.

5. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 1, characterized in that, Based on the single-cylinder dynamic compensation model, the cylinder is cooled in stages to the target cryogenic temperature range. During the cooling process, the cooling process is adjusted according to temperature distribution and pressure changes to form a stable and controlled gas-liquid condensation state, including: Based on the single-cylinder dynamic compensation model, the effective volume characterization, inner wall adsorption behavior characterization, and gas-liquid conversion hysteresis characterization corresponding to the current cylinder to be filled are read. Based on the characterization results, the cylinder cooling process is divided into a pre-cooling transition zone, a relay acclimatization zone, and a cryogenic approach zone. The starting conditions, switching conditions, and dwell order of each zone are determined. The first stage of cooling is carried out on the cylinder to be filled according to the pre-cooling transition zone, so that the gas inside the cylinder enters the initial cooling state coordinated with the action of the inner wall. During the first stage of cooling, the temperature distribution and pressure change sequence at different positions on the outer wall of the cylinder to be filled are collected, and the collected results are compared with the corresponding section in the single-cylinder dynamic compensation model in the first round. When the results of the first round of comparison meet the preset switching conditions, the cylinder to be filled is transferred to the intermediate acclimatization zone, and the correspondence between the gas distribution state inside the cylinder and the action state of the inner wall is calibrated for the second time. Based on the second calibration, the cylinder to be filled is transferred into the cryogenic approach zone. Local temperature difference deviations, pressure response misalignments, and changes in condensation propulsion sequence that occur during the cooling process are identified segment by segment. When any segment is found to deviate from the preset trajectory, the cooling rhythm of the current zone, the zone dwell sequence, or the connection point of adjacent zones is adjusted to correct the condensation propulsion path inside the cylinder to be filled. After completing the cryogenic approximation zone, a closed-loop check is performed on the cylinder to be filled. When the check result is consistent with the target state sequence corresponding to the single-cylinder dynamic compensation model, it is determined that the cylinder has entered the gas-liquid condensation state required for filling.

6. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 5, characterized in that, When the results of the first round of comparison meet the preset switching conditions, the cylinder to be filled is transferred to the intermediate training zone, and the correspondence between the internal gas distribution state and the internal wall action state of the cylinder to be filled is calibrated a second time, including: When the results of the first round of comparison meet the preset switching conditions, the temperature distribution migration order, pressure change turning order and corresponding mapping state obtained in the interval are saved as the previous frozen segment, and the previous frozen segment is used as the starting reference of the relay acclimatization interval. Within the intermediate acclimatization interval, multiple rounds of cold energy delivery at different rhythms are applied to the cylinder to be filled according to a preset time sequence. Between each round of cold energy delivery, the pressure change sequence inside the cylinder and the temperature rearrangement sequence of the outer wall are collected simultaneously to form a progressive response group for the intermediate acclimatization stage. The progressive response group is connected end-to-end with the preceding freezing segment to track the leading, lagging, and reverting changes in the gas distribution state relative to the inner wall action state within the relay acclimatization interval, and rearranged into a state misalignment chain according to the order of occurrence of each type of change in the time series. Based on the aforementioned state misalignment chain, the corresponding boundaries, transfer order, and connection positions of the gas distribution state inside the cylinder to be filled and the action state of the inner wall within the relay training interval are recalibrated, and the recalibration results are incorporated into the segmented mapping relationship corresponding to the single-cylinder dynamic compensation model to complete the secondary calibration.

7. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 6, characterized in that, After completing the cryogenic approximation zone, a closed-loop check is performed on the cylinder to be filled. When the check result matches the target state sequence corresponding to the single-cylinder dynamic compensation model, it is determined that the cylinder has entered the gas-liquid condensation state required for filling, including: After the cryogenic approach zone ends, the current cooling process is terminated, and the pressure sequence, temperature sequence, and corresponding state transition records formed by the cylinder to be filled at the end of this zone are sealed to form the final response group. Obtain the target state sequence corresponding to the current cylinder to be filled in the single-bottle dynamic compensation model, and align the final response group with the target state sequence segment by segment in chronological order to determine the corresponding starting point, connection order and termination boundary between each response segment and each target state node. Based on the segment-by-segment alignment results, the sequential offset, boundary misconnection and state return between each response segment in the final response group and the target state node are judged, and the gas distribution state and inner wall action state of the gas cylinder to be filled are checked in a closed loop according to the judgment results to form the current state sequence. The current state sequence is compared with the target state sequence. When the state order and transition boundary of the two are consistent, it is determined that the cylinder to be filled has reached the gas-liquid condensation state required for filling.

8. The method for cryogenic precision control filling of boron trifluoride for semiconductors as described in claim 1, characterized in that, The process of filling boron trifluoride into the cylinder under the target cryogenic temperature range using a staged filling strategy includes initial constant pressure filling, mid-stage pulse-type feeding, and final differential pressure adjustment, so that the actual filling amount approaches the target value according to the single-cylinder dynamic compensation model. Boron trifluoride was introduced into the cylinder to be filled under constant pressure in the target cryogenic temperature zone, and the mass change, internal pressure change and corresponding state transition records of the cylinder to be filled were collected. The collected results were then compared with the single-cylinder dynamic compensation model for interval mapping. When the preset interval boundary is reached, the system switches to pulse feeding to connect the quality changes and pressure recovery processes of adjacent pulse cycles, forming a continuous feeding sequence. The intermediate feeding sequence is connected with the initial filling sequence to identify the inter-segment offset of quality advancement and state transition, and the feeding rhythm and switching timing are adjusted based on the single-bottle dynamic compensation model. After completing the inter-segment correction, the remaining filling process is advanced by adjusting the pressure difference between the gas supply end and the cylinder end, so that the filling amount is close to the target value, and the corresponding quality change nodes are recorded until the filling is completed.

9. A cryogenic precision control filling system for boron trifluoride for semiconductors, used to implement the cryogenic precision control filling method for boron trifluoride for semiconductors according to any one of claims 1-8, characterized in that, include: Parameter compensation module, status adjustment module, filling control module, and filling determination module; The parameter compensation module is used to collect the pressure, temperature, trace moisture and heat exchange response parameters inside the boron trifluoride cylinder to be filled, and to establish a single-cylinder dynamic compensation model based on the collected parameters. The state adjustment module is used to cool the cylinder to the target cryogenic temperature range in stages based on the single-cylinder dynamic compensation model, and to adjust the cooling process according to the temperature distribution and pressure changes during the cooling process to form a stable and controlled gas-liquid condensation state. The filling control module is used to fill boron trifluoride into the cylinder to be filled in the target cryogenic temperature zone using a staged filling strategy, including initial constant pressure filling, middle pulse feeding and final differential pressure adjustment, so that the actual filling amount is close to the target value according to the single-cylinder dynamic compensation model. The filling determination module is used to stop filling and seal the cylinder when it meets the preset consistency conditions; otherwise, it performs compensation filling or backfilling correction.

Images