A semiconductor chemical online dilution solution preparation control method and control system

Abstract

The present application relates to a kind of semiconductor chemicals online dilution liquid preparation control method and control system, the control method includes: selecting target chemical formula, load complete set of process control parameters bound with this formula, start liquid preparation flow;Real-time acquisition of raw conductivity and temperature data of liquid medicine, corresponding concentration value is obtained after temperature correction;Using the compound control architecture consisting of extended state observer, feedforward prediction controller and incremental PID controller, the output flow of stock solution metering unit is dynamically adjusted, and high-precision closed-loop regulation and control is realized;After liquid preparation is completed, concentration final inspection is carried out, when concentration check is qualified and dilution tank liquid level is up to standard, remove material lock and output qualified liquid medicine.The three-layer process depth coupling of disturbance estimation and equipment performance coupling, feedforward prediction and process timing coupling, control parameter and chemical formula coupling is effectively inhibited multi-source disturbance influence, meet the ultra-high-precision liquid preparation demand of advanced semiconductor process.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, specifically to an online dilution and preparation control method and control system for semiconductor chemicals. Background Technology

[0002] In the wet cleaning and etching processes of 12-inch advanced semiconductor manufacturing, a large amount of high-precision, high-stability process chemicals are required, typically including hydrofluoric acid, tetramethylammonium hydroxide, ammonia, and phosphoric acid. Even minute fluctuations in the concentration of these chemicals directly affect the wafer etching rate and cleaning uniformity, ultimately determining wafer yield. Advanced semiconductor manufacturing processes place extremely stringent technical requirements on the precision of solution concentration control, the continuity of solution supply, equipment safety interlock mechanisms, and fault tracing and localization capabilities. Currently, the industry commonly uses high-concentration chemical stock solutions mixed with pure water to complete the online dilution and preparation of the target process solutions on-site at the fab.

[0003] However, the following problems exist when using existing processes for online dilution: 1. Traditional PID control is a post-feedback mode with inherent lag, which is prone to overshoot and oscillation. The general active disturbance rejection algorithm has not been deeply modified to combine with the characteristics of the liquid preparation process, and cannot adapt to the superimposed disturbances from multiple sources such as pipeline pressure fluctuations, pump aging, and pipeline scaling. The concentration control accuracy of existing equipment is only ±0.003%, which cannot meet the ultra-high accuracy requirement of ±0.001% of advanced processes.

[0004] 2. Existing weighing-type liquid preparation equipment adopts a serial step-by-step feeding mode for raw liquid and pure water. It is necessary to mix and test after all materials have been fed, resulting in a long preparation cycle. In order to reserve the margin for feeding, mixing and replenishment correction, the dilution tank needs to reserve an extra-large volume redundancy, resulting in large equipment size, high cost of PFA anti-corrosion tank materials, and high cost of consumables. At the same time, the weighing sensor is susceptible to interference from workshop vibration and temperature drift, and the concentration cannot be monitored in real time during the feeding process, and real-time closed-loop correction cannot be achieved.

[0005] 3. Existing equipment uses a fixed parameter for the circulation mixing time, which is unrelated to the actual liquid level, chemical viscosity, and solution temperature of each batch. It cannot adaptively and dynamically adjust according to real-time dispensing conditions. Insufficient mixing is likely to occur under low liquid level and high viscosity conditions, while high liquid level and low viscosity conditions result in ineffective long circulation, energy waste, and slow dispensing cycle times. Furthermore, existing general-purpose dispensing equipment lacks automatic cleaning procedures and residue detection mechanisms when switching between different chemical types. Residual solutions in pipeline dead zones and circulation loops may cross-contaminate with new batches of chemicals, affecting solution purity and dispensing accuracy. This is especially problematic for process chemicals sensitive to impurities, such as hydrofluoric acid and phosphoric acid, where even trace amounts of cross-contamination can lead to a decrease in wafer yield. Summary of the Invention

[0006] The purpose of this invention is to provide an online dilution and preparation control method for semiconductor chemicals to solve the problems in the prior art.

[0007] A first aspect of the present invention provides a method for controlling the online dilution and preparation of semiconductor chemicals, comprising: S1. Formulation Adaptation and Pre-interlock: Select the target chemical formulation and load the complete set of process control parameters bound to the formulation; verify the pre-interlock conditions of the stock solution storage level and the dilution tank level, and start the solution preparation process after all conditions are met. S2. Synchronous parallel feeding and mixing: The raw liquid metering unit and the pure water inlet unit are turned on simultaneously to realize the synchronous parallel feeding of raw liquid and pure water, and the circulation mixing loop is started at the same time. S3. Real-time sampling and temperature correction: Real-time acquisition of the original conductivity and temperature data of the drug solution, and obtaining the corresponding concentration value after temperature correction; S4. Self-disturbance rejection composite closed-loop control: A composite control architecture consisting of an extended state observer, a feedforward predictive controller and an incremental PID controller is adopted to dynamically adjust the output flow of the raw liquid metering unit and achieve high-precision closed-loop control of concentration. Among them, the total disturbance estimate output by the expansion state observer is cross-validated with the pump body attenuation and pipeline flow resistance changes obtained online, and the control gain is dynamically corrected. The prediction coefficient of the feedforward prediction controller adaptively switches according to the liquid preparation process sequence to match the control requirements of different stages. The observer gain parameters of the expanded state observer and the prediction coefficients of the feedforward prediction controller are automatically matched and issued by the current chemical formula to adapt to the medium characteristics of different chemicals. S5. Concentration Verification and Transfer Interlock: After the solution is prepared, a final concentration check is performed. If the concentration is qualified and the liquid level in the dilution tank is up to standard, the transfer interlock is released and the qualified solution is output; if the concentration is not qualified, the transfer circuit is permanently locked.

[0008] In one possible implementation, the total disturbance estimate output by the expanded state observer is cross-validated with the pump body attenuation and pipeline flow resistance changes calculated online, and the control gain is dynamically corrected, including: Collect the output data of the PID controller, pump stroke data, pure water cumulative flow data, and final calibration concentration data for the current batch; The actual output coefficient of the metering pump is deduced by reverse calculation of material balance relationships; By comparing the actual output coefficient with the factory calibration parameters, the pump body attenuation and pipeline flow resistance change are calculated. The control gain of the expansion state observer is dynamically adjusted proportionally to the pump body attenuation, so that the disturbance compensation amount matches the current performance state of the equipment.

[0009] In one possible implementation, the feedforward prediction controller adaptively switches the prediction coefficients according to the process timing, including: The first and second rates of change of concentration deviation are calculated in real time. The first rate of change satisfies e'(k)=e(k)-e(k-1), and the second rate of change satisfies e''(k)=e(k)-2e(k-1)+e(k-2). e(k) is the deviation between the target concentration and the corrected concentration at the k-th sampling time. When the concentration deviation is greater than the preset concentration difference threshold, the first prediction coefficient is used to accelerate the concentration convergence speed. When the concentration deviation is less than or equal to the concentration difference threshold, the second prediction coefficient is used to suppress overshoot; During the concentration verification stage before material transfer, a zero prediction coefficient is used to avoid erroneous adjustments.

[0010] In one possible implementation, the observer gain parameter of the extended state observer and the prediction coefficient of the feedforward prediction controller are automatically matched and issued by the current chemical formulation to adapt to the media characteristics of different chemicals, including: Pre-store the bandwidth of the controlled object corresponding to multiple chemical formulas; Based on the currently selected chemical formulation, the corresponding controlled object bandwidth is retrieved, and the observer gain parameter adapted to the chemical is calculated according to the following relationship: β 01 =2ω c ,β 02 =ω c 2 In the formula, ω c， β is the bandwidth of the controlled object. 01 β 02 For observer gain parameters; The calculated observer gain parameters and corresponding prediction coefficients are synchronously sent to the extended state observer and the feedforward prediction controller.

[0011] In one possible implementation, step S3 includes: Pre-store calibration data tables linked to each chemical formulation; Based on the real-time collected drug solution temperature and original conductivity, the corrected concentration value at the standard temperature is obtained through bilinear interpolation. When the calibration data table is abnormal or the interpolation result exceeds the preset range, it automatically degrades to linear temperature compensation mode and triggers an alarm.

[0012] In one possible implementation, step S3 further includes a multi-sensor reliability fusion process, specifically comprising: The difference in concentration between the main and backup sensors after calibration is calculated in real time and compared with a set difference threshold to verify data consistency. The reliability weight of each sensor is calculated in real time based on multiple indicators such as sensor data stability, historical drift, self-verification consistency, and cumulative working time. The two sensors are weighted and fused according to their confidence levels to output the final concentration value. When the confidence weight of a single sensor falls below the set weight threshold, its weight is automatically reset to zero, switching to single-sensor operation mode and triggering an alarm.

[0013] In one possible implementation, step S4 is followed by: Once the concentration of the solution has been adjusted to a stable and converged state by the incremental PID controller, the self-verification process is triggered. The theoretical concentration value was calculated by taking the cumulative amount of the stock solution and the cumulative amount of pure water added in this batch and using the material balance relationship. Compare the theoretical concentration value with the actual measured and calibrated concentration value of the sensor to calculate the zero-point drift and range drift of the sensor. When the drift is less than the alarm threshold, drift compensation is performed; otherwise, an over-limit alarm is output.

[0014] In one possible implementation, step S2 includes: Real-time data collection of liquid level, drug viscosity, and medium temperature for the current batch; The optimal cyclic mixing time for this batch is dynamically calculated based on the collected liquid level, drug viscosity, and medium temperature.

[0015] A second aspect of the present invention provides an online dilution and preparation control system for semiconductor chemicals, comprising: The system includes a stock solution storage unit, a dilution tank unit, a dosing and metering unit, a pure water inlet unit, a circulation detection unit, a material transfer and conveying unit, and a PLC control unit. The PLC control unit integrates a synchronous parallel feeding control module, a multi-chemical formula adaptation module, a temperature dynamic compensation module, a self-disruption composite control module, and a material transfer interlock control module. The multi-chemical formulation adaptation module is used to store and distribute a complete set of process control parameters corresponding to different chemical formulations; The synchronous parallel feeding control module is used to synchronously start the feeding metering unit and the pure water inlet unit, so as to realize the synchronous parallel feeding of the raw liquid and pure water into the dilution tank unit, and at the same time start the circulation mixing loop of the circulation detection unit. The temperature dynamic compensation module is used to receive the original conductivity and temperature data of the drug solution collected by the cyclic detection unit, and output the corrected concentration value after temperature correction. The active disturbance rejection composite control module incorporates a composite control architecture consisting of an extended state observer, a feedforward predictive controller, and an incremental PID controller, used to dynamically adjust the output flow of the dosing metering unit. This composite control architecture employs a three-layer deep process coupling structure specifically designed for the multi-source disturbances unique to semiconductor solution preparation scenarios. The first layer is a disturbance estimation and equipment performance coupling structure, which is used to cross-validate the total disturbance estimate output by the extended state observer with the pump body attenuation and pipeline flow resistance changes obtained online, and dynamically correct the control gain. The second layer is a feedforward prediction and process timing coupling structure, which is used to enable the prediction coefficient of the feedforward prediction controller to adaptively switch with the liquid preparation process timing during the initial feeding stage, convergence stage, and concentration verification stage. The third layer is a coupling structure between control parameters and chemical formulation, which is used to automatically match and send the observer gain parameters of the extended state observer and the prediction coefficients of the feedforward prediction controller from the current chemical formulation. The material transfer interlock control module is configured to: release the material transfer lock of the material transfer unit when the liquid preparation is completed and the concentration verification is qualified, and the liquid level of the dilution tank unit reaches the standard; and permanently lock the material transfer circuit when the concentration is not qualified.

[0016] In one possible implementation, the raw temperature and concentration detection data output by the cyclic detection unit are synchronously distributed to the temperature dynamic compensation module, the active disturbance rejection composite control module, and the sensor online self-calibration module, respectively, for concentration correction, disturbance observation, and drift identification, ensuring the temporal consistency and medium consistency of the input data of each module. The estimated total disturbance of the system output by the self-disturbance rejection composite control module is used to generate a disturbance compensation amount in one direction to adjust the output flow of the metering unit; the other direction is output in reverse to the pump pipeline performance online evaluation and compensation module, which is cross-validated with the collected pure water flow and pump stroke measured data to separate and identify the rapid change in pipeline pressure disturbance and the slow change in equipment deterioration disturbance. The equipment performance degradation characterization parameters output by the pump pipeline performance online evaluation and compensation module are fed back into the active disturbance rejection composite control module to dynamically correct the internal control gain of the extended state observer, so as to realize the adaptive updating of the control model parameters with the actual operating state of the equipment. The formula configuration parameters output by the multi-chemical formula adaptation module are sent in parallel to the self-disturbance rejection composite control module, the temperature dynamic compensation module, and the abnormal operating condition graded handling module, so that the observer gain parameters, concentration interpolation parameters, and fault judgment thresholds are synchronously adapted and switched according to the target chemical formula.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. A composite control architecture combining extended state observer, feedforward prediction and incremental PID is adopted. Through three-layer deep process coupling, namely disturbance estimation and equipment performance coupling, feedforward prediction and process timing coupling, and control parameters and chemical formulation coupling, the influence of multi-source disturbances is effectively suppressed, meeting the ultra-high precision liquid preparation requirements of advanced semiconductor processes.

[0018] 2. The original solution and pure water are fed in a synchronous and parallel manner to replace the traditional serial step feeding process. The circulation mixing and concentration are started simultaneously and closed-loop control is implemented in real time to realize parallel operation of feeding, mixing and concentration correction. At the same time, the optimal mixing time is dynamically calculated according to the batch liquid level, drug viscosity and medium temperature to adapt to the mixing requirements under different working conditions.

[0019] 3. By dynamically adapting the cycle mixing time, it solves the problems of insufficient mixing in high-viscosity conditions, ineffective energy consumption in low-load conditions, and slow liquid preparation cycle caused by traditional fixed parameters; through a closed-loop control mechanism of automatic multi-level cleaning and residual concentration detection and confirmation when switching chemicals, it performs complete cycle cleaning and residual verification when switching between different types of chemicals, and performs rapid flushing optimization when switching between different concentrations of the same chemical, eliminating the risk of cross-contamination caused by pipeline residues from the root, ensuring the purity and proportioning accuracy of the solution, and adapting to the needs of multi-category corrosive chemical rotation production scenarios. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of the online dilution and preparation control method for semiconductor chemicals according to the present invention; Figure 2 This is a schematic diagram of the composite control architecture of the present invention; Figure 3 This is a schematic diagram of the architecture of the online dilution and preparation control system for semiconductor chemicals of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] It should be noted that the serial numbers assigned to the components in the embodiments of the present invention, such as "first" and "second", are only used to distinguish the described objects and have no sequential or technical meaning.

[0023] like Figure 3As shown, the first aspect of the present invention provides an online dilution and preparation control system for semiconductor chemicals. The system includes a human-machine operation unit, a stock solution storage unit, a dilution tank unit, a dosing and metering unit, a pure water inlet unit, a circulation detection unit, a material transfer and conveying unit, and a PLC control unit.

[0024] 1. Stock Solution Storage Unit: Equipped with a corrosion-resistant stock solution storage tank and a capacitive level sensor, suitable for storing various highly corrosive acid and alkali chemicals; the level signal is uploaded to the PLC control unit in real time for pre-dispensing interlock verification, setting the pre-dispensing interlock condition: the dilution process can only be started when the stock solution storage tank level is higher than the L threshold.

[0025] 2. Dilution Tank Unit: Equipped with a five-level capacitive liquid level detection device (HH, H, M, L, LL), corresponding to five liquid level thresholds. It has built-in liquid level interlocks for dilution start-up and material transfer start-up: dilution start-up requires the liquid level in the dilution tank to be lower than the M liquid level threshold, and material transfer start-up requires the liquid level in the dilution tank to be higher than the M liquid level threshold. If any liquid level condition is not met, the corresponding process interlock will be locked, realizing accurate monitoring of liquid level stratification and pre-process protection.

[0026] 3. Dosing and metering unit: including a raw material metering pump and a corrosion-resistant PFA check valve installed at the outlet end, which, together with the pure water pressure stabilizing pipeline and the small dead volume circulation loop, form a fluid disturbance suppression system to prevent flow fluctuations caused by drug backflow and pressure shock; a precise correspondence model is established between the raw material output flow rate and the pump stroke speed to provide a data basis for flow compensation and concentration control.

[0027] 4. Pure water inlet unit: The pipeline connects the main pressure regulator, flow controller, and high-precision flow meter in series to achieve stable pressure, stable flow, and accurate metering of pure water; real-time flow data is uploaded to the PLC control unit to provide a stable data source for pump performance evaluation, material balance, and concentration correction.

[0028] 5. Circulation Detection Unit: It adopts a short-distance, small dead volume closed-loop circulation sampling circuit structure, which connects the circulation pump, main concentration sensor, backup concentration sensor, and PT1000 temperature detection element in series before returning to the dilution tank inlet; the main concentration sensor and backup concentration sensor are arranged in series in the same flow channel, and the temperature detection element is installed in the same flow channel as the concentration sensor, which ensures the consistency of the sampling medium and the real-time sampling, and greatly reduces the pipeline lag volume.

[0029] 6. Material transfer and conveying unit: Equipped with a corrosion-resistant material transfer pump and a dual hardware and software locking mechanism, controlled by a dual condition interlock of concentration and liquid level. The lock can only be released when the concentration is verified to be qualified and the liquid level is up to standard.

[0030] 7. Human-Machine Operation Unit: Includes a touch screen and an integrated hardware button group. The hardware buttons include a power switch, a mode selection switch, a light inspection switch, an alarm reset switch, and a hardware emergency stop switch. The system is configured to forcibly trigger the whole machine light inspection process only when switching from STOP mode to AUTO mode. Only after the light inspection is passed can the system enter the liquid preparation standby state, thus preventing the operation of faulty equipment.

[0031] The PLC control unit integrates a synchronous parallel feeding control module, a multi-chemical formula adaptation module, a temperature dynamic compensation module, a self-disruption composite control module, a material transfer interlock control module, a pump pipeline performance online evaluation and compensation module, a sensor online self-calibration module, a multi-sensor redundancy verification and reliability fusion module, an adaptive mixing time adjustment module, an abnormal working condition graded handling module, a chemical switching anti-cross-contamination cleaning module, a fault source tracing and storage module, and a batch quality fingerprint traceability module.

[0032] A cross-module process data linkage and coupling mechanism is established between the various functional modules. The functional modules form a collaborative control system through targeted data interaction, rather than a simple superposition of independent functional units. The raw temperature and concentration data output by the cyclic detection unit are synchronously distributed to the temperature dynamic compensation module, the active disturbance rejection composite control module, and the sensor online self-calibration module, respectively, for concentration correction, disturbance observation, and drift identification, ensuring the temporal consistency and medium consistency of the input data of each module. The estimated total disturbance of the system output by the self-disturbance rejection composite control module is used to generate a disturbance compensation amount in one direction to adjust the output flow of the metering unit; the other direction is output in reverse to the pump pipeline performance online evaluation and compensation module, which is cross-validated with the collected pure water flow and pump stroke measured data to separate and identify the rapid change in pipeline pressure disturbance and the slow change in equipment deterioration disturbance. The equipment performance degradation characterization parameters output by the pump pipeline performance online evaluation and compensation module are fed back into the active disturbance rejection composite control module to dynamically correct the internal control gain of the extended state observer, so as to realize the adaptive updating of the control model parameters with the actual operating state of the equipment. The formula configuration parameters output by the multi-chemical formula adaptation module are sent in parallel to the self-disturbance rejection composite control module, the temperature dynamic compensation module, and the abnormal working condition graded handling module, so that the observer gain parameters, concentration interpolation parameters, and fault judgment thresholds are synchronously adapted and switched according to the target chemical formula. The fault tracing and storage module collects all module operation data and process parameters in real time and stores them by binding batch quality fingerprint codes; the abnormal operating condition classification and handling module receives fault signals from each module and triggers the corresponding level of handling logic.

[0033] The following is combined with Figure 1The control method and process for online dilution and solution preparation using the above system are described below: This control method is implemented based on the above system and specifically includes the following steps: S0, Forced inspection during mode switching: The system completes basic hardware self-checks upon power-up; it identifies the system's operating status switch. Only when the system switches from STOP mode to AUTO mode is a full-machine light inspection process forcibly triggered. The system can only enter the solution preparation standby state after all light inspections pass. If a light inspection fails, the hardware and software interlock prevents the system from entering automatic production mode, triggering an alarm and requiring manual troubleshooting and reset. This step is preparation work before the system performs solution preparation and dilution.

[0034] S1. Formula adaptation and pre-locking: The system selects the target chemical formula through the human-machine interface and determines whether it is a chemical type switching operation. If the new formula has different chemical types than the current formula, the chemical switching anti-cross-contamination cleaning procedure is forcibly started, and the process can only continue after the residue test is qualified. If it is a switch between different concentrations of the same chemical, only one rapid circulation rinse is performed, and the process can continue after the residue test is confirmed, thus balancing pollution prevention requirements and production efficiency.

[0035] The multi-chemical formulation adaptation module loads a complete set of process control parameters bound to the formulation, including PID parameters, active disturbance rejection observer parameters, feedforward prediction coefficients, liquid level interlock thresholds, mixing time base, concentration / conductivity / temperature three-dimensional LUT calibration data table and LUT version number, temperature operating range, sensor fusion weights, cleaning program parameters, and fault judgment thresholds.

[0036] The system performs a pre-lock check: confirming that the liquid level in the raw material storage unit is higher than the L liquid level threshold and the liquid level in the dilution tank unit is lower than the M liquid level threshold. The liquid preparation process can only be started after all conditions are met. If any condition is not met, the feeding process is prohibited.

[0037] The chemical switching and cross-contamination prevention cleaning procedure adopts a closed-loop mechanism of automatic cleaning - residual detection - confirmation and release: it sequentially performs a multi-stage cleaning process of emptying the dilution tank, injecting pure water to level M, starting the circulation pump for circulation cleaning, and emptying, with no less than 3 cleaning cycles for switching between strong acids and strong alkalis; after cleaning, the residual concentration of the final pure water cleaning solution is detected. If the residual concentration is lower than the formula threshold, the cleaning is deemed qualified, and the preparation of a new chemical solution is allowed; if the residual concentration exceeds the standard, an additional cleaning cycle is automatically added. If the residual concentration fails for 3 consecutive cycles, a moderate warning is triggered, prompting manual inspection of pipeline dead corners or scaling; the cleaning waste liquid is collected in a dedicated waste liquid collection tank through an independent drainage pipeline, and the cleaning record is stored in the fault traceability storage module.

[0038] S2. Synchronous and parallel feeding and mixing: The synchronous parallel flow feeding control module simultaneously activates the raw material metering pump and the pure water inlet unit of the metering unit, enabling the raw material and pure water to be fed into the dilution tank unit in a synchronous and parallel flow. At the same time, the circulation pump of the circulation detection unit is started to form a circulating mixing loop. This eliminates the traditional sequential and lagging process of feeding first, then mixing, and then detecting, and realizes parallel operation of feeding, mixing, and concentration adjustment.

[0039] The cycle mixing time adopts an adaptive adjustment method: real-time collection of liquid level, drug viscosity and medium temperature parameters of the current batch, and dynamic calculation of the optimal cycle mixing time for this batch based on the collected parameters; automatic extension of mixing time under high liquid level and high viscosity conditions to ensure mixing effect; automatic shortening of mixing time under low liquid level and low viscosity conditions to reduce ineffective energy consumption and slow liquid preparation cycle.

[0040] S3. Real-time sampling and temperature correction: The raw conductivity and temperature data of the drug solution are collected in real time, and the corresponding concentration value is obtained after temperature correction. The specific process is as follows: Pre-stored three-dimensional LUT calibration data tables for concentration / conductivity / temperature are bound to each chemical formulation, and the calibration data tables are managed according to the formulation version number; based on the real-time collected drug solution temperature and original conductivity, the corrected concentration value at a standard temperature of 25℃ is obtained through bilinear interpolation, which accurately adapts to the nonlinear characteristics of chemical concentration / conductivity / temperature; when the calibration data table data is abnormal or the interpolation result exceeds the preset reasonable range, it automatically degrades to linear temperature compensation mode and triggers a moderate warning.

[0041] This step also includes a multi-sensor credibility fusion process: Real-time calculation of the concentration difference ΔC = |C_c| between the primary and backup sensors after calibration. main -C backup The data consistency is verified by comparing it with a set difference threshold; based on multiple dimensions such as sensor data stability, historical drift, self-verification consistency, and cumulative working time, a weighted scoring method is used to calculate the reliability weight of each sensor in real time; the two sensors are weighted and fused according to their reliability weights to output the final concentration value. The fusion calculation satisfies the following: C fused = (ω main ×C main + ω backup ×C backup ) / (ω main + ω backup ), In the formula, C fused ω represents the concentration value after fusion. main ω backup The confidence weights of the primary and backup sensors are C and C, respectively. main C backupThese are the calibrated concentration values ​​of the main and backup sensors, respectively.

[0042] When the confidence weight of a single sensor is lower than the set weight threshold, its weight is automatically reset to zero, switching to single sensor operation mode and triggering a moderate warning.

[0043] S4, Active Disturbance Rejection Composite Closed-Loop Control: like Figure 2 As shown, a composite control architecture consisting of an extended state observer, a feedforward predictive controller, and an incremental PID controller is adopted to dynamically adjust the output flow of the dosing metering unit, achieving high-precision closed-loop control of concentration. This composite control architecture employs a three-layer deep process coupling to address the unique challenges of semiconductor solution preparation scenarios, including rapidly changing pipeline pressure disturbances, slowly changing pump aging and pipeline scaling disturbances, and nonlinear chemical disturbances, rather than a direct transplantation of a general active disturbance rejection control algorithm.

[0044] The final pump stroke control quantity is synthesized from three parts: PID feedback quantity, feedforward prediction quantity, and disturbance compensation quantity, satisfying the following: u(k) = u pid (k) + u ff (k)-z2(k) / b0 In the formula, u pid For incremental PID output, u ff z2(k) / b0 is the feedforward prediction output, and z2(k) / b0 is the disturbance compensation.

[0045] First layer: Disturbance estimation coupled with device performance.

[0046] The extended state observer expands the total system disturbance into independent state variables, estimating the total disturbance value online in real time. The output total disturbance estimate is cross-validated with the pump body attenuation and pipeline flow resistance changes calculated online, and the control gain is dynamically adjusted. The specific process is as follows: Collect the PID controller output data, pump stroke data, pure water cumulative flow data, and final correction concentration data for the current batch; deduce the actual output coefficient of the metering pump through material balance; compare the actual output coefficient with the factory calibration parameters to calculate the pump body attenuation and pipeline flow resistance change; dynamically adjust the control gain of the expansion state observer proportionally based on the pump body attenuation, ensuring the disturbance compensation matches the current performance state of the equipment, and the correction relationship satisfies: b0 = b 0_initial ×(1 - pump attenuation), In the formula, b0 is the corrected control gain, b 0_initial This is the initial calibration gain. When the attenuation exceeds the threshold, a graded alarm is automatically output, eliminating the need for offline calibration with system shutdown and additional weighing hardware.

[0047] The second layer: feedforward prediction coupled with process timing.

[0048] The prediction coefficient of the feedforward predictive controller adaptively switches according to the liquid preparation process sequence to match the control requirements of different stages. The specific process is as follows: The first and second rates of change of concentration deviation are calculated in real time. The first rate of change satisfies (e'(k)=e(k)-e(k-1), and the second rate of change satisfies e''(k)=e(k)-2e(k-1)+e(k-2). e(k) is the deviation between the target concentration and the corrected concentration at the k-th sampling time. Combining the feeding rate, liquid level change parameters and historical batch convergence curves, the trend of concentration deviation is predicted, and feedforward compensation is generated.

[0049] When the concentration deviation is greater than the preset concentration difference threshold (in the initial feeding stage), the first prediction coefficient is used to accelerate the concentration convergence speed; when the concentration deviation is less than or equal to the concentration difference threshold (convergence stage), the second prediction coefficient is used to suppress overshoot; in the concentration verification stage before material transfer (when the concentration difference is zero), the zero prediction coefficient is used to avoid erroneous adjustment.

[0050] Third layer: Coupling of control parameters and chemical formulation The observer gain parameters of the extended state observer and the prediction coefficients of the feedforward prediction controller are automatically matched and issued based on the current chemical formulation, adapting to the media characteristics of different chemicals. The specific process is as follows: Multiple chemical formulations are pre-stored, each corresponding to the bandwidth of the controlled object. Based on the currently selected chemical formulation, the corresponding bandwidth of the controlled object is retrieved, and the observer gain parameter adapted to the chemical is calculated according to the following relationship: β 01 =2ω c, β 02 =ω c 2 In the formula, ω c， β is the bandwidth of the controlled object. 01 β 02 The calculated observer gain parameters and corresponding prediction coefficients are synchronously sent to the extended state observer and the feedforward prediction controller, eliminating the need for manual adjustments when switching chemicals.

[0051] This step also includes an online self-calibration process for the sensor: after the concentration of the solution is adjusted to a stable convergence by the incremental PID controller, the self-calibration process is triggered; the cumulative amount of raw solution and pure water added in this batch is retrieved, and the theoretical concentration value is calculated through material balance; the theoretical concentration value is compared with the actual measured and calibrated concentration value of the sensor, and the zero-point drift and range drift of the sensor are calculated; when the drift is less than the alarm threshold, drift compensation is performed, otherwise an over-limit alarm is output, realizing online calibration of the sensor without disassembly.

[0052] Throughout the solution preparation process, the abnormal condition classification and handling module monitors the operating status signals of each module in real time and executes a three-level handling logic according to the severity of the fault: ① Minor alarms: Small pressure fluctuations, slight pump degradation, minor sensor drift, slight temperature exceedance, and slight decrease in sensor reliability. Only alarm information is recorded and audible and visual alerts are provided, and the equipment continues to operate. ②Medium warning: Pump attenuation is approaching the threshold, sensor drift is approaching the upper limit, LUT data is abnormal, temperature is approaching the threshold, sensor consistency deviation is increasing, cleaning residue is approaching the threshold. Allow the current batch of solution preparation and transfer to be completed, lock the automatic start of the next batch, and trigger the warning prompt for maintenance. ③ Serious faults: Concentration error, liquid level error, emergency stop, severe sensor drift, severe pump degradation, severe inconsistency between dual sensor data. Immediately lock the feeding circuit and the transfer circuit for emergency protection.

[0053] S5. Concentration verification and material transfer interlock.

[0054] After the solution is prepared, the final accuracy is verified by calibrating the fusion concentration. When the concentration verification is qualified and the liquid level in the dilution tank unit is higher than the M liquid level threshold, the transfer interlock control module releases the transfer lock of the transfer conveying unit and starts the transfer pump to deliver the qualified drug solution to the production line supply tank. If the concentration is not qualified, the transfer circuit is permanently locked, and the fault timing data and batch quality fingerprint are locked, waiting for troubleshooting and reset.

[0055] After the material transfer is completed, the batch quality fingerprint code is simultaneously uploaded to the production line MES system, or printed as a QR code label and affixed to the supply tank, realizing full-link tracking of the chemical solution from the dispensing equipment to the wafer process equipment.

[0056] The control method and control system of the present invention will be described below in the context of online liquid preparation, using specific scenarios as examples: Example 1: High-precision dilution and preparation of hydrofluoric acid (HF): The system performs a power-on self-test. When the target concentration formulation of HF 0.5% is selected on the HMI, the system determines that it matches the current formulation (no cleaning or switching is required) and automatically loads the corresponding complete set of parameters: PID parameter K. p =1.2, Ki=0.05, Kd=0.02; bandwidth ω of the controlled object c =50 rad / s, corresponding to the observer gain parameter β 01 =100, β 02 =2500; initial control gain b0=0.8; sampling period h=0.1s; HF dedicated 3D LUT calibration data table; five-level liquid level threshold; adaptive hybrid reference parameters.

[0057] Switching the system from STOP mode to AUTO mode forces the equipment to trigger a full-machine light inspection. After passing the light inspection, the system enters standby mode. The system automatically verifies the prerequisites that the liquid level in the raw material tank is >L and the liquid level in the dilution tank is <M. Once all conditions are met, the liquid preparation begins, generating the batch quality fingerprint code HF-20260831-001.

[0058] The stock solution and pure water flow synchronously into the dilution tank. Circulation and mixing are initiated instantly upon addition. The system dynamically calculates the optimal mixing time for this batch based on real-time liquid level and HF solution viscosity. The main and backup concentration sensors in the circulation loop collect the raw conductivity in real time. The multi-sensor fusion module performs consistency verification (ΔC=0.0002%, far below the threshold). After confidence weighting calculation, the main sensor has a weight of 0.55, and the backup sensor has a weight of 0.45. The weighted fusion outputs the fused concentration value. The PT1000 temperature element collects the solution temperature in real time and uses the HF 3D LUT calibration data table for interpolation to output the corrected concentration at a standard temperature of 25℃.

[0059] The active disturbance rejection composite control module starts working: the extended state observer estimates the total system disturbance in real time, tracks pure water pressure fluctuations (±0.015MPa) in the initial stage and actively compensates for them; the feedforward predictive controller predicts the deviation trend based on the concentration change rate and the second-order change rate, using a large predictive coefficient for rapid convergence in the initial feeding stage, and automatically switching to a small predictive coefficient to suppress overshoot during the convergence stage; the pump pipeline performance online evaluation module synchronously calculates the pump body attenuation and dynamically corrects the control gain b0. After PID convergence, the sensor online self-calibration is triggered, and the material balance shows that the deviation between the theoretical concentration and the measured concentration is 0.0008%, indicating that the sensor is normal.

[0060] After the solution preparation is completed, the final concentration deviation is 0.0008%, which is within the acceptable range of ±0.001%. Once the liquid level meets the standard, the transfer pump is unlocked to complete the supply of the drug solution. The batch quality fingerprint code is then uploaded to the MES system simultaneously.

[0061] Under this operating condition, the output is 1m 3 The process takes approximately 52 minutes to prepare the qualified solution, and the tank volume is 0.92m³. 3Compared with the traditional incremental PID scheme, the test results are as follows: convergence time was shortened from 62 min to 52 min, a reduction of 16.1%; maximum overshoot decreased from 0.0032% to 0.0011%, a reduction of 65.6%; steady-state fluctuation amplitude decreased from ±0.0025% to ±0.0008%, a reduction of 68.0%; after artificially introducing a ±0.02MPa pure water pressure step disturbance, the disturbance recovery time was shortened from 45s to 12s, a reduction of 73.3%; after 30 days of continuous operation, the accuracy decay decreased from 180% to 30%, and the equipment aging adaptability improved by 83.3%. Compared with the traditional serial weighing and feeding process, the liquid preparation time was shortened by about 30% from 85-100 min, and the dilution tank volume was reduced from 1.30-1.40 m³. 3 Reducing the size by 20%-35% significantly lowers the cost of tank consumables and the floor space required for cleanrooms.

[0062] Example 2: Tetramethylammonium hydroxide (TMAH) switching solution preparation.

[0063] Switching the TMAH formula directly on the human-machine interface, the system recognizes it as a chemical type switching condition (HF→TMAH) and forcibly starts the anti-cross-contamination cleaning procedure: performing 3 cycles of pure water injection-circulation cleaning-drain circulation, and finally the residual concentration of the cleaning solution is detected to be 0.0005%, which is lower than the TMAH formula residual threshold of 0.001%, and the cleaning is deemed qualified.

[0064] The multi-chemical formulation adaptation module automatically updates all control parameters with full coverage: TMAH controlled object bandwidth ω c =40 rad / s, corresponding to the observer gain parameter β 01 =80、β 02 =1600, synchronously update PID parameters, 3D LUT calibration data table, sensor fusion weights and fault thresholds, without the need for manual item-by-item debugging.

[0065] The subsequent process follows the same procedures as in Example 1: start / stop light inspection, synchronous parallel feeding, dual-sensor fusion, temperature nonlinearity correction, self-disturbance rejection composite control, adaptive mixing, online self-verification, double-locking material transfer, and batch quality fingerprint traceability. This achieves high-precision and stable dilution of the TMAH solution. The switching process takes approximately 8 minutes, significantly less than the 30 minutes or more required for manual cleaning, and the cleaning effect can be quantified and confirmed, eliminating the risk of cross-contamination caused by incomplete manual cleaning.

[0066] During solution preparation, the main concentration sensor experienced increased drift due to prolonged contact with the corrosive solution. Self-calibration detected a drift of 0.002%, close to the moderate threshold. The multi-sensor fusion module automatically reduced the main sensor's confidence weight to 0.2 and increased the backup sensor's weight to 0.8, maintaining accurate weighted fusion output.

[0067] The system triggers a moderate warning, indicating that the main sensor requires maintenance, but allows the current batch of solution preparation and material transfer to be completed, locking the automatic start of the next batch. After the current batch of solution preparation is completed, maintenance personnel replace the main sensor, and the system automatically performs sensor calibration and self-verification. After confirming that it is normal, it resumes dual-sensor fusion operation, avoiding the risk of batch scrapping and production line downtime caused by sensor failure in the traditional single-sensor architecture.

[0068] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional modules, sub-modules, and units as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for controlling online dilution and preparation of semiconductor chemicals, characterized in that, include: S1. Formulation Adaptation and Pre-locking: Select the target chemical formulation and load the complete set of process control parameters bound to the formulation; Verify the pre-interlock conditions between the stock solution storage level and the dilution tank level. Once all conditions are met, start the solution preparation process. S2. Synchronous parallel feeding and mixing: The raw liquid metering unit and the pure water inlet unit are turned on simultaneously to realize the synchronous parallel feeding of raw liquid and pure water, and the circulation mixing loop is started at the same time. S3. Real-time sampling and temperature correction: Real-time acquisition of the original conductivity and temperature data of the drug solution, and obtaining the corresponding concentration value after temperature correction; S4. Self-disturbance rejection composite closed-loop control: A composite control architecture consisting of an extended state observer, a feedforward predictive controller and an incremental PID controller is adopted to dynamically adjust the output flow of the raw liquid metering unit and achieve high-precision closed-loop control of concentration. Among them, the total disturbance estimate output by the expansion state observer is cross-validated with the pump body attenuation and pipeline flow resistance changes obtained online, and the control gain is dynamically corrected. The prediction coefficient of the feedforward prediction controller adaptively switches according to the liquid preparation process sequence to match the control requirements of different stages. The observer gain parameters of the expanded state observer and the prediction coefficients of the feedforward prediction controller are automatically matched and issued by the current chemical formula to adapt to the medium characteristics of different chemicals. S5. Concentration Verification and Transfer Interlock: After the solution is prepared, a final concentration check is performed. If the concentration is qualified and the liquid level in the dilution tank is up to standard, the transfer interlock is released and the qualified solution is output; if the concentration is not qualified, the transfer circuit is permanently locked.

2. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, The total disturbance estimate output by the expansion state observer is cross-validated with the pump body attenuation and pipeline flow resistance changes calculated online, and the control gain is dynamically corrected, including: Collect the output data of the PID controller, pump stroke data, pure water cumulative flow data, and final calibration concentration data for the current batch; The actual output coefficient of the metering pump is deduced by reverse calculation of material balance relationships; By comparing the actual output coefficient with the factory calibration parameters, the pump body attenuation and pipeline flow resistance change are calculated. The control gain of the expansion state observer is dynamically adjusted proportionally to the pump body attenuation, so that the disturbance compensation amount matches the current performance state of the equipment.

3. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, The feedforward prediction controller adaptively switches prediction coefficients according to the process timing, including: The first and second rates of change of concentration deviation are calculated in real time. The first rate of change satisfies e'(k)=e(k)-e(k-1), and the second rate of change satisfies e''(k)=e(k)-2e(k-1)+e(k-2). e(k) is the deviation between the target concentration and the corrected concentration at the k-th sampling time. When the concentration deviation is greater than the preset concentration difference threshold, the first prediction coefficient is used to accelerate the concentration convergence speed. When the concentration deviation is less than or equal to the concentration difference threshold, the second prediction coefficient is used to suppress overshoot; During the concentration verification stage before material transfer, a zero prediction coefficient is used to avoid erroneous adjustments.

4. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, The observer gain parameters of the expanded state observer and the prediction coefficients of the feedforward prediction controller are automatically matched and issued by the current chemical formulation to adapt to the media characteristics of different chemicals, including: Pre-store the bandwidth of the controlled object corresponding to multiple chemical formulas; Based on the currently selected chemical formulation, the corresponding controlled object bandwidth is retrieved, and the observer gain parameter adapted to the chemical is calculated according to the following relationship: b 01 =2h c ,b 02 =ω c 2 In the formula, ω c， β is the bandwidth of the controlled object. 01 β 02 For observer gain parameters; The calculated observer gain parameters and corresponding prediction coefficients are synchronously sent to the extended state observer and the feedforward prediction controller.

5. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, Step S3 includes: Pre-store calibration data tables linked to each chemical formulation; Based on the real-time collected drug solution temperature and original conductivity, the corrected concentration value at the standard temperature is obtained through bilinear interpolation. When the calibration data table is abnormal or the interpolation result exceeds the preset range, it automatically degrades to linear temperature compensation mode and triggers an alarm.

6. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, Step S3 further includes a multi-sensor reliability fusion process, specifically including: The difference in concentration between the main and backup sensors after calibration is calculated in real time and compared with a set difference threshold to verify data consistency. The reliability weight of each sensor is calculated in real time based on multiple indicators such as sensor data stability, historical drift, self-verification consistency, and cumulative working time. The two sensors are weighted and fused according to their confidence levels to output the final concentration value. When the confidence weight of a single sensor falls below the set weight threshold, its weight is automatically reset to zero, switching to single-sensor operation mode and triggering an alarm.

7. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, Following step S4, the following is also included: Once the concentration of the solution has been adjusted to a stable and converged state by the incremental PID controller, the self-verification process is triggered. The theoretical concentration value was calculated by taking the cumulative amount of the stock solution and the cumulative amount of pure water added in this batch and using the material balance relationship. Compare the theoretical concentration value with the actual measured and calibrated concentration value of the sensor to calculate the zero-point drift and range drift of the sensor. When the drift is less than the alarm threshold, drift compensation is performed; otherwise, an over-limit alarm is output.

8. The online dilution and preparation control method for semiconductor chemicals according to claim 1, characterized in that, Step S2 includes: Real-time data collection of liquid level, drug viscosity, and medium temperature for the current batch; The optimal cyclic mixing time for this batch is dynamically calculated based on the collected liquid level, drug viscosity, and medium temperature.

9. An online dilution and preparation control system for semiconductor chemicals, the system being used to execute the control method as described in claims 1-8, comprising a stock solution storage unit, a dilution tank unit, a dosing and metering unit, a pure water inlet unit, a circulation detection unit, a material transfer and conveying unit, and a PLC control unit; The PLC control unit integrates a synchronous parallel feeding control module, a multi-chemical formula adaptation module, a temperature dynamic compensation module, a self-disruption composite control module, and a material transfer interlock control module. The multi-chemical formulation adaptation module is used to store and distribute a complete set of process control parameters corresponding to different chemical formulations; The synchronous parallel feeding control module is used to synchronously start the feeding metering unit and the pure water inlet unit, so as to realize the synchronous parallel feeding of the raw liquid and pure water into the dilution tank unit, and at the same time start the circulation mixing loop of the circulation detection unit. The temperature dynamic compensation module is used to receive the original conductivity and temperature data of the drug solution collected by the cyclic detection unit, and output the corrected concentration value after temperature correction. The active disturbance rejection composite control module incorporates a composite control architecture consisting of an extended state observer, a feedforward predictive controller, and an incremental PID controller, used to dynamically adjust the output flow of the dosing metering unit. This composite control architecture employs a three-layer deep process coupling structure specifically designed for the multi-source disturbances unique to semiconductor solution preparation scenarios. The first layer is a disturbance estimation and equipment performance coupling structure, which is used to cross-validate the total disturbance estimate output by the extended state observer with the pump body attenuation and pipeline flow resistance changes obtained online, and dynamically correct the control gain. The second layer is a feedforward prediction and process timing coupling structure, which is used to enable the prediction coefficient of the feedforward prediction controller to adaptively switch with the liquid preparation process timing during the initial feeding stage, convergence stage, and concentration verification stage. The third layer is a coupling structure between control parameters and chemical formulation, which is used to automatically match and send the observer gain parameters of the extended state observer and the prediction coefficients of the feedforward prediction controller from the current chemical formulation. The material transfer interlock control module is configured to: release the material transfer lock of the material transfer unit when the liquid preparation is completed and the concentration verification is qualified, and the liquid level of the dilution tank unit reaches the standard; and permanently lock the material transfer circuit when the concentration is not qualified.

10. The online dilution and dispensing control system for semiconductor chemicals according to claim 9, characterized in that, include: The raw temperature and concentration data output by the cyclic detection unit are synchronously distributed to the temperature dynamic compensation module, the active disturbance rejection composite control module, and the sensor online self-calibration module, respectively, for concentration correction, disturbance observation, and drift identification, ensuring the temporal consistency and medium consistency of the input data of each module. The estimated total disturbance of the system output by the self-disturbance rejection composite control module is used to generate a disturbance compensation amount in one direction to adjust the output flow of the metering unit; the other direction is output in reverse to the pump pipeline performance online evaluation and compensation module, which is cross-validated with the collected pure water flow and pump stroke measured data to separate and identify the rapid change in pipeline pressure disturbance and the slow change in equipment deterioration disturbance. The equipment performance degradation characterization parameters output by the pump pipeline performance online evaluation and compensation module are fed back into the active disturbance rejection composite control module to dynamically correct the internal control gain of the extended state observer, so as to realize the adaptive updating of the control model parameters with the actual operating state of the equipment. The formula configuration parameters output by the multi-chemical formula adaptation module are sent in parallel to the self-disturbance rejection composite control module, the temperature dynamic compensation module, and the abnormal operating condition graded handling module, so that the observer gain parameters, concentration interpolation parameters, and fault judgment thresholds are synchronously adapted and switched according to the target chemical formula.

Images