Method and system for dynamically adjusting silicon carbide CMP polishing solution based on multi-parameter feedback control

Abstract

The invention belongs to the technical field of semiconductor manufacturing, and discloses a silicon carbide CMP polishing solution dynamic adjustment method and system based on multi-parameter feedback control, and the method comprises the steps: collecting the oxidation-reduction potential, pH value and turbidity data of a polishing solution in real time; inputting sensor data into an edge computing platform for filtering and standardization processing; establishing a dynamic adjustment model based on a fuzzy PID control algorithm, and taking a process state index as a control parameter switching basis; and a high-precision executing mechanism is driven to finish dynamic replenishing of the oxidizing agent, the pH regulator and the dispersing agent, and closed-loop control over the components of the polishing solution is achieved. The state of the polishing solution can be monitored and automatically adjusted in real time, the problems of parameter drift and manual intervention lag in a traditional open-loop control mode are effectively solved, the stability and consistency of the silicon carbide polishing process are guaranteed, the polishing quality is improved, and meanwhile the waste of the polishing solution and the maintenance cost are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor manufacturing, in particular to a silicon carbide CMP polishing liquid dynamic adjustment method and system based on multi-parameter feedback control. BACKGROUND

[0002] In the manufacturing process of high-hardness and high-brittle materials, chemical mechanical polishing (CMP) is a key process to achieve ultra-smooth surface, and the polishing liquid is the core consumable material. Material removal rate (MRR) and surface roughness (Ra) are usually used as core indicators to evaluate the performance of the polishing liquid. In the polishing process of high-hardness and high-brittle materials, as the polishing time increases, the chemical properties (such as oxidation-reduction potential and pH value) and physical state (such as turbidity) of the polishing liquid will change, resulting in a decrease in MRR and an increase in Ra. When the MRR is lower than the process threshold or the Ra exceeds the acceptable range due to the deterioration of the polishing liquid performance, the polishing liquid needs to be replaced to ensure the processing quality. In current production, the state of the polishing liquid is judged by manual offline detection (such as sampling analysis of MRR at regular intervals) or static rule control (such as pre-setting the amount of liquid supplement according to the number of polishing discs), which has defects such as response lag, insufficient precision, high maintenance and operation cost, and is easy to cause waste of polishing liquid or affect the processing quality. Therefore, it is crucial to develop a method for realizing real-time closed-loop adjustment of the composition of the polishing liquid to maintain the stability of the silicon carbide CMP process and reduce the cost of consumables. SUMMARY

[0003] The present application aims to provide a silicon carbide CMP polishing liquid dynamic adjustment method and system based on multi-parameter feedback control to solve the problems raised in the background.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] A silicon carbide CMP polishing liquid multi-parameter online regulation method, comprising the following steps:

[0006] Step 1, preloading the baseline value of the polishing liquid parameter in the historical process database;

[0007] Step 2, real-time collection of oxidation-reduction potential, pH value and turbidity data of the polishing liquid, and polishing head pressure and rotation speed parameters of the polishing platform by multi-source sensors;

[0008] Step 3, time stamp alignment and sliding window filtering processing of the multi-source sensor data by the edge computing platform, and standardization to a unified dimension to obtain standardized data;

[0009] Step 4, comparison and calculation of the process state deviation index PDI between the pre-processed multi-source sensor data and the baseline value;

[0010] Step 5, outputting the control instruction according to the comparison between the real-time process state deviation index PDI and the preset threshold value.

[0011] Further, in the step 2, the multi-source sensor comprises:

[0012] a redox potential sensor for detecting a redox potential of the polishing liquid;

[0013] a pH sensor for detecting a pH value of the polishing liquid;

[0014] a laser turbidity sensor for detecting a dispersion state of the abrasive in the polishing liquid;

[0015] a pressure sensor for detecting a pressure value between the polishing head and the wafer;

[0016] a polishing head rotation speed sensor for detecting a rotation speed of the polishing head;

[0017] a polishing pad rotation speed sensor for detecting a rotation speed of the polishing pad.

[0018] Further, the step 3 comprises: the edge computing platform performing timestamp synchronization on the sensor data, and then eliminating noise of the redox potential, the pH value, the rotation speed of the polishing head, the rotation speed of the polishing pad, and the pressure value between the polishing head and the wafer through recursive average filtering, and performing normalization processing to obtain standardized data.

[0019] Further, the step 4 specifically comprises:

[0020] a) comparing the standardized data outputted in the step 3 with a reference value to generate the following characteristic deviation values:

[0021] an oxidation decay rate deviation based on a percentage deviation amount of the real-time oxidation decay rate from a rated oxidation decay rate;

[0022] a pH shift amount an absolute difference value between the real-time pH and a target pH;

[0023] an agglomeration index deviation a positive deviation amount of the real-time from a rated agglomeration index;

[0024] b) calculating a process state deviation index PDI through a dynamic weighting formula:

[0025]

[0026] wherein, PDI is the process state deviation index, is the oxidation decay rate deviation, is the pH shift amount, is the agglomeration index deviation, is an oxidant activity deviation weight coefficient, is a pH stability weight coefficient, is a suspended particle stability weight coefficient, the weight coefficient satisfies: , and decreases over time, increases over time.

[0027] Further, the step 5 specifically comprises: comparing the process state deviation index PDI with a threshold value:

[0028] ① When the process state deviation index PDI is greater than the threshold value, the fuzzy PID controller generates an oxidant, pH adjuster and dispersant compensation control instruction based on the process state deviation index PDI and its change rate to control the actuator to adjust the polishing liquid parameters;

[0029] ② When the process state deviation index PDI is less than or equal to the threshold value, maintain the current output value of all actuators constant;

[0030] ③ When the fluctuation amplitude of the process state deviation index PDI in a plurality of consecutive sampling periods exceeds the platform warning baseline value, trigger the alarm unit to prompt the polishing liquid to be replaced, and the setting range of the warning baseline value is: , wherein is the exponential mean value of the device steady-state operation stage, and k is a process sensitivity coefficient factor.

[0031] 6. The silicon carbide CMP polishing liquid multi-parameter online control method according to claim 5, characterized in that the fuzzy PID controller adopts a three-level architecture, including a fuzzification layer, a fuzzy reasoning and decision-making layer, and a parameter self-tuning layer.

[0032] The fuzzification layer is responsible for receiving the process state deviation index PDI and its first derivative, and converting them into fuzzy language variables through a triangular membership function.

[0033] Input variable definition:

[0034] The deviation amount E = PDI - threshold value, and the fuzzy language variable set is wherein, NB, NS, ZO, PS, and PB represent the deviation degree of the process state: severe recession, mild recession, stable state, mild over-standard, and severe over-standard.

[0035] The deviation change rate , and the fuzzy language variable set is ;

[0036] The triangular membership function is used to realize the mapping from the continuous domain to the discrete language variable, and the expression of the triangular membership function is:

[0037]

[0038] wherein, is the input variable actual value is the membership degree of a specific fuzzy linguistic variable, parameter is the left boundary point of the membership function, defining the starting point of the effective range of the linguistic variable; parameter is the right boundary point of the membership function, defining the ending point of the effective range of the linguistic variable; parameter is the peak point of the membership function, indicating the core area of full membership of the linguistic variable;

[0039] The fuzzy reasoning decision layer stores a two-dimensional fuzzy rule base based on double input variables, and performs logical reasoning according to a predefined decision matrix; the fuzzy reasoning decision layer performs fuzzy reasoning operation, and outputs the adjustment amount of the PID controller parameters, and the core rule structure paradigm is:

[0040] If E is A and EC is B, then

[0041] wherein, the number of fuzzy rules is determined by the number of linguistic values of the input variables E and EC, E represents the deviation amount, EC represents the deviation rate of change, and usually constitutes a order decision matrix, wherein n and m represent the number of linguistic variables of E and EC, respectively; respectively, are the proportional, integral, and differential parameter adjustment amounts; A is the linguistic value of the deviation amount E, and B is the linguistic value of the deviation rate of change EC; represents the control action vector for a specific fuzzy condition;

[0042] The parameter self-tuning layer is composed of a defuzzification unit, a decoupling operation unit, and a dynamic correction unit, and converts the fuzzy output into an accurate compensation instruction through multi-stage processing;

[0043] The defuzzification unit adopts the gravity center method:

[0044]

[0045] wherein, is the accurate adjustment amount of the PID parameters to be solved, respectively, correspond to the proportional term p, the integral term i, and the differential term d; represents the total number of effective fuzzy rules actually triggered at this moment; refers to the rule number currently processed; represents the activation strength of the ith rule; represents the original fuzzy adjustment component of the j-type parameter preset by the ith rule;

[0046] The decoupling operation unit realizes multi-compensation amount cooperation:

[0047]

[0048] wherein, , , are output compensation vectors, corresponding to oxidant, pH value, and agglomeration degree compensation respectively; K is a decoupling matrix, acting on the PID parameter correction vector to realize cross compensation, and satisfying: is a matrix , the element in the i-th row and j-th column, is a predefined coupling coefficient;

[0049] The dynamic correction unit adds historical trend adjustment:

[0050] Based on the process state deviation index sequence of the last control cycle, the historical trend factor is calculated.

[0051] The historical trend analysis is independently performed on each compensation, and finally three corrected compensation quantities are output :

[0052]

[0053] wherein, , , are the final compensation quantities after correction, corresponding to the final compensation quantities of oxidant, pH value, and agglomeration degree respectively; , , correspond to the preconfigured adjustment sensitivity coefficients of oxidant, pH value, and agglomeration degree respectively; , , correspond to the historical trend factors of oxidant, pH value, and agglomeration degree respectively.

[0054] Further, the fuzzy reasoning decision layer has a dynamic knowledge base updating mechanism:

[0055] According to the time series variance of the system energy gradient and the process state deviation index , the rule weight matching the real-time working condition is established The rule applicability is quantified, and g(·) refers to the weight function. When the same group of rules is activated continuously for N times , the reorganization operator is triggered; the activated reorganization operator automatically updates the rules according to the following feedback mechanism:

[0056]

[0057] wherein, is a configurable learning rate, is a boundary saturation function, is an actual compensation effect vector, is a rule prediction compensation vector, is an updated rule set;

[0058] By comparing the difference between the actual compensation effect and the rule prediction output, a closed-loop feedback loop is formed to realize online adaptive correction of the rule base parameters.

[0059] Further, the historical trend factor is constructed based on a sliding window:

[0060]

[0061] wherein, i is a traversal variable within the window, t is a current time reference point, is a trend balance coefficient, is a sign function, n is a dynamic sliding window width, is a current PDI value, is a historical reference value, is a PDI window mean value.

[0062] The application also provides a silicon carbide CMP polishing liquid adaptive regulation system based on multi-source data fusion, which is used to realize the control method as described above, and characterized in that it comprises:

[0063] A data acquisition module, comprising a redox potential sensor, a pH sensor, a laser turbidity sensor, a pressure sensor, a polishing head speed sensor, and a polishing pad speed sensor, wherein the redox potential sensor detects the redox potential of the polishing liquid, the pH sensor detects the pH value of the polishing liquid, the laser turbidity sensor detects the dispersion state of the abrasive in the polishing liquid, the pressure sensor detects the pressure value between the polishing head and the wafer, the polishing head speed sensor detects the speed of the polishing head, and the polishing pad speed sensor detects the speed of the polishing pad.

[0064] An edge computing platform, which is in communication connection with the data acquisition module, performs time alignment and recursive average filtering processing on the original data collected by the data acquisition module to eliminate high-frequency noise, performs normalization processing on each parameter data, and compares and calculates the process state deviation index by comparing the preprocessed data with the baseline value.

[0065] An actuator module, comprising an oxidizing agent metering pump, a dispersant feeding device, and a pH adjuster feeding device, which is used to automatically adjust the composition of the polishing liquid according to the control instruction to realize dynamic component control of the polishing liquid.

[0066] The upper computer system comprises an adaptive control module adopting a three-layer fuzzy controller architecture to process the pre-processed process state deviation index PDI, generate a polishing liquid state evaluation result and a regulation instruction through threshold comparison, and feed back to the operator through a human-computer interface.

[0067] Compared with the prior art, the beneficial effects of the present application are that the traditional silicon carbide polishing liquid control mode is limited to offline detection and empirical replacement strategy, and has inherent defects such as response lag, dynamic condition oversight and resource waste; the present application innovatively integrates multi-source data such as process dynamic characteristics, carries a fuzzy PID control algorithm model, and realizes real-time capture and component adaptation of the polishing liquid state. At the same time, the dynamic regulation of the concentration of key additives is carried out before the critical point of abrasive performance attenuation to avoid the risk of surface defects, improve the effective utilization rate of the polishing liquid, and simultaneously reduce the single piece processing cost. The present application can avoid the process instability and excessive resource consumption caused by response lag and extensive management in the traditional technology, thereby maintaining the stability of the silicon carbide CMP process and reducing the production cost. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 is a flow chart of a silicon carbide CMP polishing liquid dynamic regulation method based on multi-parameter feedback control of the present application.

[0069] Figure 2 is a structural schematic diagram of a polishing platform and a polishing liquid circulation system in the present application.

[0070] Figure 3 is a structural schematic diagram of a silicon carbide CMP polishing liquid dynamic regulation system based on multi-parameter feedback control of the present application.

[0071] Label description: 1, polishing head speed sensor, 2, polishing head pressure sensor, 3, polishing pad speed sensor, 4, polishing liquid circulation pipeline, 5, primary recovery bucket, 6, first centrifugal pump, 7, recovery filter, 8, liquid supply bucket, 9, second centrifugal pump, 10, third centrifugal pump, 11, liquid supply filter, 12, multi-parameter sensing unit (pH sensor, laser turbidity sensor, oxidation-reduction potential sensor), 13, fourth centrifugal pump, 14, wastewater bucket, 15, communication line, 16, three-channel actuator, 17, fifth centrifugal pump, 18, electric valve. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0073] As Figures 1-2 shown, the present application provides a multi-parameter feedback control-based dynamic adjustment method for silicon carbide CMP polishing liquid, comprising the following steps:

[0074] Step 1, preloading the baseline values of the polishing liquid parameters in the historical process database, including: rated oxidation recession rate, target pH value, rated agglomeration index, automatically loaded into the host computer memory at system startup as a benchmark reference for real-time control.

[0075] Step 2, real-time collection of multi-source data information of the polishing liquid and the polishing platform during the processing process through multi-source sensors. Specifically, it includes:

[0076] First, a polishing liquid supply end detection subsystem is arranged in the polishing liquid circulating pipeline, which includes a multi-parameter sensor unit 12 (pH sensor, laser turbidity sensor, oxidation-reduction potential sensor), wherein the laser turbidity sensor is used to detect the dispersion state of the abrasive in the polishing liquid, the pH sensor is used to detect the pH value of the polishing liquid, and the oxidation-reduction potential sensor is used to detect the oxidation-reduction potential of the polishing liquid. When detecting, the second centrifugal pump 9 supplies liquid to the multi-parameter sensor unit, and the fourth centrifugal pump 13 transports the detected waste liquid to the waste bucket 14. Then, a polishing machine end detection subsystem is arranged, which includes a polishing head speed sensor 1, a polishing head pressure sensor 2 and a polishing pad speed sensor 3. In the polishing machine end detection subsystem, the polishing head speed sensor 1 is used to detect the polishing head speed, the polishing head pressure sensor 2 is used to detect the polishing head pressure value, and the polishing pad speed sensor 3 is used to detect the polishing pad speed.

[0077] Step 3, the detection subsystem transmits the collected data to the edge computing platform, and the edge computing platform performs data preprocessing work, including:

[0078] Step 3.1, uniform time label, time stamp alignment, taking the sensor with the lowest sampling frequency as the time reference, and uniform time label for sensor data with different sampling frequencies.

[0079] Step 3.2, denoising, using sliding window filtering method for denoising, smoothing the input data curve.

[0080] Step 3.3, standardization, normalizing all sensor data to the range of 0-1 to facilitate model input processing.

[0081] Step 4, then the edge computing platform transmits the preprocessed data to the host computer control system through wireless communication, and the host computer executes the following PDI calculation process:

[0082] Step 4.1, feature deviation calculation, the input layer receives the normalized sensor data stream (oxidative decay rate / pH / agglomeration index), and generates three key deviation quantities (oxidative decay rate deviation, pH shift, and agglomeration index deviation) by comparing with the baseline values in real time.

[0083] Step 4.2, dynamic weighted PDI synthesis, the process state deviation index is calculated by the following dynamic weighting formula:

[0084]

[0085] wherein, PDI is the process state deviation index, is the oxidative decay rate deviation, is the pH shift, is the agglomeration index deviation, is the oxidant activity deviation weight coefficient, is the pH stability weight coefficient, is the suspended particle stability weight coefficient, and the weight coefficients satisfy: , and decreases over time, increases over time.

[0086] The timing control mechanism is: the oxidative decay weight exponentially decays with polishing time; the agglomeration control weight increases over time; and the pH shift weight remains constant.

[0087] Step 5, according to the comparison of the real-time process state deviation index PDI and the preset threshold value, the control instructions are output, including:

[0088] The process state deviation index PDI is compared with the threshold value:

[0089] ① When the process state deviation index PDI is greater than the threshold value, the fuzzy PID controller generates control instructions for oxidant, pH adjuster, and dispersant compensation based on the process state deviation index PDI and its change rate to control the actuator to adjust the polishing liquid parameters;

[0090] ② When the process state deviation index PDI is less than or equal to the threshold value, the current output values of all actuators are maintained constant;

[0091] ③ When the fluctuation amplitude of the process state deviation index PDI in consecutive multiple sampling periods exceeds the platform warning baseline value, the warning unit is triggered to prompt the polishing liquid replacement, and the setting range of the warning baseline value is: , wherein is the exponential mean value in the steady-state operation stage of the equipment, and k is the process sensitivity coefficient factor.

[0092] The fuzzy PID controller is implemented in a three-level architecture, including a fuzzification layer, a fuzzy reasoning and decision-making layer, and a parameter self-tuning layer.

[0093] In the fuzzification layer of the fuzzy PID controller, the system receives the process state deviation index PDI and its first-order derivative in real time, i.e., the deviation E (defined as the difference between PDI and a threshold value) and the deviation change rate EC (defined as the rate of change of PDI over time). Among them, the fuzzy language variable set of the deviation E is {NB, NS, ZO, PS, PB}, corresponding to the state changes of the process state from severe recession, mild recession, stable state, mild over-standard, and severe over-standard; the fuzzy language variable set of the deviation change rate EC is {NB, ZO, PB}, representing the change trend of rapid deterioration, stable state, and rapid improvement.

[0094] And a triangular membership function is used to realize the mapping from the continuous domain to the discrete language variable, and the parameters are reasonably configured according to the process tolerance range to ensure that the fluctuation range of the normal working condition is covered. The triangular membership function is as follows:

[0095] wherein, is the actual value of the input variable The membership degree of a specific fuzzy language variable is parameterized is the left boundary point of the membership function, defining the starting point of the effective action range of the language variable; the parameter is the right boundary point of the membership function, defining the end point of the effective action range of the language variable; the parameter is the peak point of the membership function, indicating the core area of complete membership of the language variable.

[0096] The fuzzy reasoning and decision-making layer of the control model stores a two-dimensional fuzzy rule base based on double-input variables, and performs logical reasoning according to a pre-defined 5x3 order decision matrix to output the adjustment amount of the PID controller parameters, and each rule follows a unified logical structure: "If E is A and EC is B, then ", wherein the number of fuzzy rules is determined by the number of language values of the input variables E and EC, E represents the deviation, and EC represents the deviation change rate, usually constituting a order decision matrix, wherein n and m represent the number of language variables of E and EC, respectively; are the proportional, integral, and differential parameter adjustment amounts, respectively; A is the language value of the deviation E, and B is the language value of the deviation change rate EC; indicates the control action vector for a specific fuzzy condition. At the same time, the system has online learning capability, and through the dynamic knowledge base updating mechanism in the fuzzy reasoning and decision-making layer, the system energy gradient and the time series variance of the process state deviation index Establishing rule weight To quantify the rule applicability, and trigger the reorganization operator when the same group of rules is activated for 10 consecutive times. The activated reorganization operator is automatically updated according to the following feedback mechanism:

[0097] Wherein, is a configurable learning rate, is a boundary saturation function, is the actual compensation effect vector, is the rule predicted compensation vector, is the updated rule set.

[0098] By comparing the difference between the actual compensation effect and the rule predicted output, a closed-loop feedback loop is formed to realize online adaptive correction of the rule library parameters, ensuring that the control rules are continuously close to the actual process characteristics.

[0099] Finally, the parameter self-tuning layer is composed of a defuzzification unit, a decoupling operation unit and a dynamic correction unit. After entering the parameter self-tuning top layer, the defuzzification unit first uses the gravity method for defuzzification processing:

[0100] Wherein, is the accurate adjustment amount of the PID parameters that need to be solved, corresponding to the proportional term p, the integral term i and the differential term d respectively; represents the total number of effective fuzzy rules actually triggered at this moment; refers to the rule number currently processed; represents the activation strength (membership calculation result) of the i th rule; represents the original fuzzy adjustment component of the j type parameter preset for the i th rule.

[0101] This method calculates the accurate parameter adjustment amount by synthesizing the contributions of all activated rules.

[0102] Secondly, the decoupling operation unit configures a decoupling conversion matrix K to realize multi-compensation amount cooperation to output a compensation vector according to the coupling characteristics of the three components of the polishing liquid:

[0103] Wherein, , , is the output compensation vector, corresponding to the oxidant, pH value and agglomeration degree compensation amount respectively; K is the decoupling matrix, acting on the PID parameter correction amount vector to realize cross compensation and meet: , is the element in the i th row and j th column of the matrix , is a predefined coupling coefficient.

[0104] Finally, the historical trend adjustment is added to the dynamic correction unit, and the historical trend factor is calculated based on the sliding window

[0105] where i is the traversal variable within the window, t is the current time reference point, is the trend balance coefficient, is the sign function, n is the dynamic sliding window width, is the current PDI value, is the historical reference value, is the PDI window mean value.

[0106] Then, the historical trend analysis is performed on each compensation amount independently, and finally three corrected compensation amounts are output

[0107] wherein, is the final compensation amount after correction, corresponding to the final compensation amount of oxidant, pH value, and agglomeration degree respectively; corresponding to the pre-configured adjustment sensitivity coefficient of oxidant, pH value, and agglomeration degree respectively; corresponding to the historical trend factor of oxidant, pH value, and agglomeration degree respectively.

[0108] After the model training is completed, the final compensation instruction is transmitted to the three-channel execution mechanism 16 through the communication bus, and the modulated polishing liquid is injected into the circulating pipeline through the electric valve 18 to the polishing machine, forming a complete closed-loop regulation loop.

[0109] Step 6, repeat the polishing process, and use the multi-parameter feedback control silicon carbide CMP polishing liquid dynamic adjustment system to perform real-time closed-loop adjustment on the polishing liquid components to maintain the stability of the silicon carbide CMP process. For example, Figure 2 ​​​​​​​​As shown, a polishing head speed sensor 1 and a polishing head pressure sensor 2 are installed on the polishing head of the polishing machine, a polishing pad speed sensor 3 is installed on the bottom plate of the polishing machine, the polishing machine is connected with a polishing liquid circulating system, the polishing liquid enters a primary recovery barrel 5 through a circulating pipeline 4, is transported to a recovery filter 7 by a first centrifugal pump 6, and then enters a liquid supply barrel 8 after treatment; then, the polishing liquid is divided into two branches: one branch is transported to a multi-parameter sensing unit 12 for component detection by a second centrifugal pump 9, detection data are transmitted to an edge computing platform through a communication line 15, and finally, the polishing liquid is sent into a wastewater barrel 14 by a fourth centrifugal pump 13; the other branch is returned to the polishing machine through a third centrifugal pump 10, a recovery filter 11 and an electric valve 18. The host computer receives data of the edge computing platform through the communication line 15 and issues instructions to a three-channel actuator 16, and the actuator is connected to the circulating loop through a fifth centrifugal pump 17.

[0110] Reference Figure 3 As shown, a silicon carbide CMP polishing liquid dynamic adjustment system based on multi-parameter feedback control is used to realize the detection method described above, and includes a data acquisition module, an edge computing platform, an actuator module and a host computer system. The data acquisition module includes a pH sensor, a laser turbidity sensor, a redox potential sensor, a polishing head speed sensor, a polishing head pressure sensor and a polishing pad speed sensor, wherein the laser turbidity sensor is used to detect the dispersion state of the abrasive in the polishing liquid, the pH sensor is used to detect the pH value of the polishing liquid, the redox potential sensor is used to detect the redox potential of the polishing liquid, the polishing head speed sensor is used to detect the polishing head speed, the polishing head pressure sensor is used to detect the polishing head pressure value, and the polishing pad speed sensor is used to detect the polishing pad speed. The edge computing platform pre-processes the original data collected by the multi-source sensors. The host computer system is used to process the pre-processed multi-source data to generate PDI, and performs closed-loop feedback control on the components of the polishing liquid through a loaded fuzzy PID controller, and finally sends compensation instructions to a three-channel actuator for compensation operation. The system man-machine interface displays the process curve and alarm log for the operator.

[0111] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for on-line multi-parameter regulation of a silicon carbide CMP polishing solution, characterized in that, The method comprises the following steps: Step 1, preloading the polishing liquid parameter baseline value in the historical process database; Step 2, collecting the redox potential, pH value and turbidity data of the polishing liquid, and the polishing head pressure and rotation speed parameters of the polishing platform in real time through the multi-source sensor; Step 3, the edge computing platform performs timestamp alignment and sliding window filtering processing on the multi-source sensor data, and standardizes the data to a unified dimension to obtain standardized data; Step 4, comparing the preprocessed multi-source sensor data with the baseline value to calculate the process state deviation index PDI; Step 5, comparing the real-time process state deviation index PDI with the preset threshold value to output the control instruction.

2. The method according to claim 1, wherein the method is characterized by, In step 2, the multi-source sensor comprises: a redox potential sensor for detecting the redox potential of the polishing liquid; a pH sensor for detecting the pH value of the polishing liquid; a laser turbidity sensor for detecting the dispersion state of the abrasive in the polishing liquid; a pressure sensor for detecting the pressure value between the polishing head and the wafer; a polishing head rotation speed sensor for detecting the rotation speed of the polishing head; a polishing pad rotation speed sensor for detecting the rotation speed of the polishing pad.

3. The method according to claim 1, wherein the method is characterized by, Step 3 comprises: the edge computing platform performs timestamp synchronization on each sensor data, and then eliminates the noise of the redox potential, pH value, rotation speed of the polishing head, rotation speed of the polishing pad and pressure value between the polishing head and the wafer through recursive average filtering, and performs normalization processing to obtain standardized data.

4. The method according to claim 1, wherein the method is characterized by, Step 4 specifically comprises: a) comparing the standardized data output in step 3 with the reference value to generate the following characteristic deviation values: Oxidation degradation rate bias based on a percentage deviation from a real-time to rated oxidation degradation rate; pH offset the absolute difference between the real-time pH and the target pH; Agglomeration index deviation , real time Positive deviation amount with respect to the rated agglomeration index; b) calculating the process state deviation index PDI through a dynamic weighting formula: ， wherein, is a process state deviation index, is an oxidation decay rate deviation, is a pH shift, is an agglomeration index deviation, is an oxidizing agent activity deviation weight coefficient, is a pH stability weight coefficient, is a suspended particle stability weight coefficient, the weight coefficients satisfy: , and decreases over time, increases over time.

5. The method for online control of multiple parameters of silicon carbide CMP polishing slurry according to claim 1, characterized in that, Step 5 specifically comprises: comparing the process state deviation index PDI with the threshold value: ① When the process state deviation index PDI is greater than the threshold value, the fuzzy PID controller generates the control instruction for compensating the oxidant, pH adjuster and dispersant based on the process state deviation index PDI and its change rate to control the actuator to adjust the polishing liquid parameters; ② When the process state deviation index PDI is less than or equal to the threshold value, the current output values of all actuators are maintained constant; ③ When the fluctuation range of the process state deviation index PDI in continuous multiple sampling periods exceeds the platform warning baseline value, the warning unit is triggered to prompt the polishing liquid replacement, and the setting range of the warning baseline value is: , , wherein is the index mean value of the device steady-state operation stage, and k is the process sensitive coefficient factor.

6. The method according to claim 5, wherein the method is characterized by, The fuzzy PID controller is implemented in a three-level architecture, including a fuzzification layer, a fuzzy reasoning and decision-making layer, and a parameter self-tuning layer; The fuzzification layer is responsible for receiving the process state deviation index PDI and its first derivative, and converting them into fuzzy language variables through a triangular membership function; The input variable definition: The deviation amount E = PDI - threshold, the fuzzy language variable set is Wherein, NB, NS, ZO, PS, PB represent the deviation degree of process state respectively: severe recession, light recession, stable state, light overproof, severe overproof. Bias change rate , the fuzzy language variable set is ; The triangular membership function is used to realize the mapping from the continuous domain to the discrete language variable, and the expression of the triangular membership function is: ， wherein, is the input variable actual value is the membership degree of a specific fuzzy linguistic variable, parameter is the left border point of the membership function, defining the starting point of the effective range of the linguistic variable; parameter is the right border point of the membership function, defining the ending point of the effective range of the linguistic variable; parameter is the peak point of the membership function, identifying the core region of full membership of the linguistic variable; The fuzzy reasoning and decision-making layer stores a two-dimensional fuzzy rule base based on double input variables, and performs logical reasoning according to a pre-defined decision matrix; the fuzzy reasoning and decision-making layer performs fuzzy reasoning operation and outputs the adjustment amount of the PID controller parameters, and the core rule structure paradigm is: If E is A and EC is B, then , wherein the number of fuzzy rules is determined by the number of linguistic values of the input variables E and EC, E represents the deviation amount, EC represents the deviation change rate, and usually constitutes a decision matrix of order n*m, where n and m represent the number of linguistic variables of E and EC, respectively; respectively represent the proportional, integral, and derivative parameter adjustment amounts; A represents the linguistic value of the deviation amount E, and B represents the linguistic value of the deviation change rate EC; represents the control action vector for a specific fuzzy condition. The parameter self-tuning layer is composed of a defuzzification unit, a decoupling operation unit and a dynamic correction unit, which converts the fuzzy output into accurate compensation instructions through multi-level processing; The defuzzification unit adopts the gravity method: ， wherein, is the PID parameter adjustment amount to be solved, respectively correspond to the proportional term p, the integral term i, and the derivative term d; represents the total number of effective fuzzy rules actually triggered at this moment; refers to the rule number currently processed; represents the activation strength of the i-th rule; represents the original fuzzy adjustment component of the j-type parameter preset for the i-th rule; The decoupling operation unit realizes the cooperation of multiple compensation amounts: ， Wherein, , , is an output compensation vector, corresponding to the oxidant, pH value, and agglomeration degree compensation amount respectively; K is a decoupling matrix, acting on the PID parameter correction vector Cross compensation is realized, and the following is satisfied: is the element in the i-th row and the j-th column of the matrix , is a predefined coupling coefficient; The dynamic correction unit adjusts the historical trend: Based on a recent sequence of process state deviation indices for the control period, compute a history trend factor ; The historical trend analysis is independently performed for each compensation amount, and finally three corrected compensation amounts are output : ， Wherein, , , is the final compensation amount after correction, corresponding to the final compensation amount of oxidant, pH value, and agglomeration degree, respectively; , , correspond to the adjustment sensitivity coefficients of the pre-configured oxidant, pH value, and agglomeration degree, respectively; , , correspond to the historical trend factors of oxidant, pH value, and agglomeration degree, respectively.

7. The method for online control of multiple parameters of silicon carbide CMP polishing slurry according to claim 6, characterized in that, The fuzzy reasoning and decision-making layer has a dynamic knowledge base updating mechanism: System energy gradient module Temporal variance of process state deviation index Establishing rule weight matching real-time working condition Quantifying rule applicability, g(·) is a reference weight function, when the same group of rules is activated for consecutive N times Triggering a reorganizing operator; the activated reorganizing operator automatically updates the rules according to the following feedback mechanism: ， wherein, is a configurable learning rate, is a boundary saturation function, is an actual compensation effect vector, is a regular prediction compensation vector, is an updated rule set; By comparing the difference between the actual compensation effect and the rule prediction output, a closed-loop feedback loop is formed to realize online adaptive correction of the rule base parameters.

8. The method of claim 6, wherein the multiple parameters are selected from the group consisting of pH, temperature, and concentration of the abrasive particles. the historical trend factor based on a sliding window construction: ， where i is the iteration variable within the window, t is the current time reference point, is the trend balancing coefficient, is the sign function, n is the dynamic sliding window width, is the current PDI value, is the historical reference value, is the PDI window mean.

9. A silicon carbide CMP polishing liquid adaptive regulation system based on multi-source data fusion, for realizing the control method of any one of claims 1-8, characterized in that, Comprise: The data acquisition module comprises a redox potential sensor, a pH sensor, a laser turbidity sensor, a pressure sensor, a polishing head rotation speed sensor and a polishing pad rotation speed sensor, the redox potential sensor detects the redox potential of the polishing liquid; The pH sensor detects the pH value of the polishing liquid; the laser turbidity sensor detects the dispersion state of the abrasive in the polishing liquid; the pressure sensor detects the pressure value between the polishing head and the wafer; the polishing head rotation speed sensor detects the rotation speed of the polishing head; the polishing pad rotation speed sensor detects the rotation speed of the polishing pad; The edge computing platform is communicatively connected to the data acquisition module, which performs time alignment and recursive average filtering on the raw data collected by the data acquisition module to eliminate high-frequency noise; The parameter data is normalized; The preprocessed data is compared with the baseline value to calculate the process state deviation index; The actuator module comprises an oxidant metering pump, a dispersant feeding device and a pH adjusting agent feeding device, which is used to automatically adjust the composition of the polishing liquid according to the control instruction, so as to realize the dynamic component control of the polishing liquid; The host computer system contains an adaptive control module using a three-layer fuzzy controller architecture to process the process state deviation index PDI obtained by preprocessing, generate polishing liquid state evaluation results and control instructions through threshold comparison, and feed back to the operator through the human-machine interface.

Images