A closed loop feedback and model adaptive system and method for polishing fluid delivery

Abstract

The application discloses a kind of closed-loop feedback and model adaptive system and method of polishing liquid supply, belong to chemical mechanical polishing technical field.The system includes swing arm base, at least two independently movable polishing liquid supply pipelines and its driving mechanism, controller and at least one sensor.Driving mechanism includes motor and ball screw, for driving supply port reciprocating movement along the length direction of swing arm;Sensor is used to monitor polishing liquid distribution uniformity parameter and / or wafer removal rate distribution parameter in real time.Controller obtains sensor feedback data in the process of initial scanning according to scanning recipe, compares and calculates deviation with target surface shape, maps deviation as scanning parameter adjustment amount based on pre-established polishing liquid flow field distribution mathematical model, dynamically corrects the control instruction of driving mechanism, realizes adaptive closed-loop adjustment of supply port moving track.The application can effectively improve the controllability of polishing liquid distribution uniformity on polishing pad and wafer surface shape.

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing technology, and in particular to a closed-loop feedback and model adaptive system and method for polishing slurry supply. Background Technology

[0002] Chemical mechanical polishing (CMP) is currently the core process for achieving global planarization of wafer surfaces in semiconductor manufacturing. During CMP, a slurry is delivered to a rotating polishing pad via a slurry arm, where the wafer material is removed through the combined action of chemical etching and mechanical grinding. The uniformity of the slurry distribution on the polishing pad directly affects the wafer profile and final planarity; therefore, the design of the slurry supply device has a crucial impact on CMP process performance.

[0003] In existing CMP equipment, the polishing slurry supply arm typically uses a fixed supply port or a manually adjustable supply port. Once a fixed supply port is installed, the position of the polishing slurry droplet on the polishing pad remains constant and cannot be dynamically adjusted during the process. As the polishing pad is used continuously, its surface condition and abrasive carrying capacity will change, and the wafer surface shape also has different requirements at different process stages. A fixed polishing slurry droplet makes it difficult to achieve uniform coverage of the polishing pad, especially in terms of flow field control in the wafer edge and center regions.

[0004] To address these issues, some improvement solutions propose a manually adjustable supply port position, where the polishing slurry application point is changed by tightening or loosening the screws on the swing arm base, followed by calibration using a fixture. However, this adjustment method is cumbersome, its accuracy is highly dependent on the operator's experience, and recalibration is required after each equipment maintenance, making it difficult to guarantee positional consistency. Furthermore, manual adjustment is a static adjustment, unable to be dynamically optimized based on real-time process conditions during polishing, thus failing to meet the high requirements for surface flatness and stability in advanced CMP processes.

[0005] To address the aforementioned shortcomings, several movable supply port solutions have been proposed. These solutions use a drive mechanism to move the supply port along the swing arm, achieving a sweeping motion at the polishing slurry landing point. However, most existing solutions employ open-loop control, meaning they move along a preset trajectory, lacking real-time monitoring and closed-loop adjustment capabilities for the actual distribution of the polishing slurry and the wafer surface shape. Furthermore, existing technologies lack systematic mathematical models and standardized mapping rules for quickly and accurately determining the optimal sweeping parameters based on different surface shape deviations.

[0006] Therefore, developing a polishing slurry supply system that can achieve adaptive closed-loop regulation and intelligently optimize sweeping parameters based on a flow field distribution model is of great significance for improving the uniformity and controllability of the CMP process. Summary of the Invention

[0007] In view of this, embodiments of this application provide a closed-loop feedback and model adaptation system and method for polishing fluid supply, so as to at least partially solve the above problems.

[0008] According to a first aspect of the embodiments of this application, a closed-loop feedback and model adaptation system for polishing slurry supply is provided, comprising:

[0009] Swing arm base;

[0010] At least two independently movable polishing slurry supply lines, each with a polishing slurry supply port at the front end;

[0011] A drive mechanism connected to each pipeline, the drive mechanism including a motor and a ball screw, is used to drive the supply port to reciprocate along the length of the swing arm;

[0012] The controller, electrically connected to the motor, is used to control the drive mechanism to move the supply port according to the preset sweeping formula;

[0013] At least one sensor, positioned above the polishing pad or near the polishing head, is used to monitor in real time the uniformity of the polishing slurry distribution on the polishing pad and / or the wafer removal rate distribution parameters.

[0014] The controller is configured to:

[0015] During the initial sweeping process according to the sweeping formula, feedback data from the sensors is acquired in real time.

[0016] The feedback data is compared with the target surface shape, and the deviation is calculated;

[0017] Based on the pre-established mathematical model of the polishing fluid flow field distribution, the deviation is mapped to the adjustment amount of the sweeping parameters, which include the moving stroke of the supply port, the moving speed curve, the endpoint dwell time, and the relative motion phase of the two supply ports.

[0018] The control commands of the drive mechanism are dynamically corrected according to the adjustment amount to achieve adaptive closed-loop adjustment of the supply port movement trajectory.

[0019] In some embodiments, the sensor includes at least one of the following: an optical sensor for online monitoring of the polishing slurry film thickness and radial distribution on the surface of the polishing pad; an electrochemical sensor for detecting the polishing slurry composition or pH value; an eddy current or optical film thickness measurement unit for measuring the removal rate distribution within the wafer surface; and a high-speed image acquisition unit for capturing the flow pattern of the polishing slurry; the sensor transmits the monitoring data to the controller in real time via wired or wireless means.

[0020] In some embodiments, the mathematical model for the polishing slurry flow field distribution is established by dividing the polishing pad into N concentric annular regions, establishing a mapping relationship between the supply port movement trajectory parameters and the cumulative supply amount of polishing slurry in each annular region, and determining this mapping relationship through fluid dynamics simulation and CMP process test data fitting to form a radial concentration distribution function C(r)=F(tra p The parameters of the supply port movement trajectory include the starting position, the ending position, the movement speed as a function of time v(t), and the relative motion curves of the two supply ports.

[0021] In some embodiments, the controller has a built-in standardized mapping rule library, which records the correspondence between different types of surface deviations and the optimal sweep parameter adjustment amount; the mapping rule library is implemented using a lookup table, decision tree, or a pre-trained neural network model; after the controller calculates the real-time surface deviation, it matches or deduces the corresponding sweep parameter adjustment amount from the rule library and outputs it to the drive mechanism.

[0022] In some embodiments, the drive mechanism further includes a position sensor and a limit switch for limiting the movement range of the supply port; the motor is a servo motor or a stepper motor, and the ball screw is a miniature precision ball screw; each polishing fluid supply line adopts a spiral hose, the tail end of the spiral hose is fixed to the swing arm base, and the head end is fixedly connected to the nut seat of the ball screw through a connector.

[0023] In some embodiments, the system further includes a fixed polishing fluid supply port located in the middle of the swing arm or near the center of the polishing head, for providing a basic polishing fluid flow rate; the two movable supply ports are located on the left and right sides of the fixed supply port, respectively, and the center line connecting the three supply ports is parallel to the length direction of the swing arm; the controller selects to enable the combination mode of the fixed supply port and the movable supply port, or to enable only the movable supply port, according to process requirements.

[0024] In some embodiments, when the controller performs adaptive closed-loop regulation, it adopts independent PID control loops for the two supply ports respectively, or adopts a coupled control strategy to make the movement of the two supply ports meet the preset relative motion relationship; when the sensor detects that the uniformity index of polishing fluid distribution exceeds the threshold, the controller prioritizes adjusting the moving speed curve and the endpoint dwell time, and adjusts the moving stroke when the uniformity index continues to deviate, so as to avoid mechanical wear caused by frequent reversal of the drive mechanism.

[0025] In some embodiments, the swing arm base is also integrated with a high-pressure water nozzle for cleaning the movable supply port and the fixed supply port during the polishing gap; when the controller executes the cleaning program, it drives the movable supply port to perform a full-stroke reciprocating motion at a speed 1.5 to 3 times higher than the sweeping speed during polishing, and at the same time, it links the high-pressure water nozzle to perform pulse spraying to remove the polishing liquid deposits remaining on the inner wall of the pipeline and nozzle.

[0026] According to a second aspect of the embodiments of this application, a closed-loop feedback and model adaptive control method for polishing slurry supply is provided, applied to the system described in any of the above claims, comprising the following steps:

[0027] Step S1: Establish a mathematical model of the polishing fluid flow field distribution in advance, and construct a standardized mapping rule library between surface deviation and sweeping parameter adjustment amount;

[0028] Step S2: Select the initial sweeping formula according to the initial surface shape of the target wafer, and control the movable feed ports on both sides to perform open-loop sweeping motion according to the initial stroke, speed and dwell time.

[0029] Step S3: During the polishing process, the uniformity parameters of the polishing slurry distribution or the wafer removal rate distribution parameters are acquired in real time through sensors to generate real-time surface shape data.

[0030] Step S4: Compare the real-time surface shape data with the target surface shape point by point and calculate the deviation vector;

[0031] Step S5: Input the deviation vector into the mathematical model of the flow field distribution, and obtain the adjustment amount of the sweep parameters through the inverse solution of the mapping rule base;

[0032] Step S6: The adjustment amount is sent to the drive mechanism to dynamically correct the movement trajectory parameters of the supply port;

[0033] Step S7: Repeat steps S3 to S6 until polishing is complete or the surface deviation converges to within the allowable range.

[0034] In some embodiments, the specific method for establishing the mapping rule base in step S1 is as follows: using a multi-factor orthogonal experimental method, different combinations of sweeping parameters are traversed on the test wafer, the corresponding polished surface shape is measured, and a reverse mapping model from surface shape deviation to sweeping parameter adjustment is established using partial least squares method or support vector regression, and the model is solidified in the controller in the form of a coefficient matrix or rule table.

[0035] In step S5, when the deviation vector indicates that the removal rate of the wafer edge region is low, the adjustment amount output by the mapping rule library includes: extending the stroke endpoint of the movable supply port located near the edge of the polishing pad outward by 2mm to 5mm, while extending the dwell time of the endpoint by 0.5s to 2s, and reducing the moving speed of the supply port on the other side to increase the local concentration; when the deviation vector indicates that the removal rate of the wafer center region is low, the adjustment amount output by the mapping rule library is: shifting the moving stroke of both supply ports towards the center of the polishing pad, and adopting a co-directional polymerization motion mode.

[0036] In some embodiments, the method further includes an anomaly diagnosis step: the controller monitors the residual between the sensor feedback data and the theoretical distribution predicted based on the flow field mathematical model in real time. When the residual exceeds a preset threshold and continues for multiple sampling cycles, it is determined to be an unexpected anomaly caused by polishing pad wear, polishing fluid deterioration, or supply port blockage. The controller issues an alarm signal and automatically calls the self-cleaning program or suggests replacing the polishing pad / polishing fluid. The self-cleaning program includes starting the high-pressure water nozzle and driving the movable supply port to perform a full-stroke reciprocating motion.

[0037] In some embodiments, the controller supports running two or more closed-loop control threads simultaneously, each corresponding to a different polishing stage: the coarse polishing stage uses a larger stroke range and a faster sweeping speed to quickly cover a large area, while the fine polishing stage uses a smaller stroke range and a slower sweeping speed combined with fine feedback adjustment to optimize local flatness; the mapping rule base parameters for different stages are stored separately and automatically switched by the controller according to polishing time or cumulative removal amount.

[0038] The beneficial effects of this invention include:

[0039] a. Significantly improves the uniformity of polishing slurry distribution and the controllability of wafer surface shape.

[0040] This invention, by setting at least two independently movable polishing slurry supply ports and achieving closed-loop adaptive adjustment based on real-time sensor feedback and a mathematical model of flow field distribution, can dynamically correct sweeping parameters such as the movement stroke, speed curve, and endpoint dwell time of the supply ports. Compared with existing open-loop sweeping or manual adjustment schemes, this invention can effectively compensate for flow field distribution deviations caused by factors such as polishing pad wear and fluctuations in polishing slurry properties, making the cumulative supply of polishing slurry in each annular region of the polishing pad closer to the target distribution. This significantly improves the flatness of the wafer center, middle, and edge regions, and enhances the stability of the surface shape and process repeatability.

[0041] b. Implement model-driven intelligent sweep parameter optimization to reduce process development difficulty.

[0042] This invention pre-establishes a mathematical model of the polishing slurry flow field distribution and constructs a standardized mapping rule library between surface shape deviation and sweep parameter adjustment through multi-factor orthogonal experiments. When the sensor detects a deviation between the real-time surface shape and the target value, the controller can quickly solve for the optimal sweep parameter adjustment based on the mapping rule library, eliminating the need for repeated trial and error based on operator experience. This approach not only significantly shortens the process formulation development cycle but also makes sweep parameters portable and reproducible across different processes and wafer models, facilitating the standardization and intelligent management of CMP processes.

[0043] c. Improve equipment operational safety and reduce maintenance costs

[0044] This invention eliminates the risk of human error and collisions between the polishing head / trimming head and the supply arm by fixing the position of the swing arm base, thus avoiding the need for manual calibration of the polishing slurry application point. Simultaneously, the controller's built-in anomaly diagnosis module monitors the residual between sensor feedback data and flow field mathematical model predictions in real time. When the residual exceeds a threshold, it automatically identifies abnormal conditions such as polishing pad wear or pipeline blockage and issues an alarm or initiates a self-cleaning program. This mechanism effectively prevents wafer scratches or polishing pad damage caused by supply port blockage or abnormal polishing slurry supply, improving the long-term reliability and safety of the equipment. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0046] Figure 1 This is a schematic diagram of a chemical mechanical polishing apparatus provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of a polishing fluid supply device provided in an embodiment of the present invention;

[0048] Figure 3 This is a radial partition diagram of the mathematical model of polishing fluid flow field distribution provided in an embodiment of the present invention;

[0049] Figure 4 This is a flowchart illustrating the construction of a standardized mapping rule base according to an embodiment of the present invention;

[0050] Figure 5 This is a flowchart illustrating the application of a standardized mapping rule base according to an embodiment of the present invention;

[0051] Figure 6 This is a flowchart of a closed-loop feedback and model adaptive control method for polishing fluid supply provided in an embodiment of the present invention;

[0052] Figure 7 This is a surface shape curve diagram corresponding to an abnormal wafer edge removal rate provided in an embodiment of the present invention;

[0053] Figure 8 It is a surface profile curve of the wafer after closed-loop feedback and model adaptive control of the polishing slurry supply as described in this application. Detailed Implementation

[0054] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.

[0055] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0056] It should be understood that although the terms "first," "second," "third," etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0057] This application first provides a closed-loop feedback and model adaptive system for polishing fluid supply. Figure 1 This illustrates a specific structure of the system when applied to a chemical mechanical polishing (CMP) device. For example... Figure 1 As shown, the chemical mechanical polishing equipment includes: a polishing disc 100, a polishing pad 200 disposed on the polishing disc 100, a polishing head 300 for clamping the wafer and applying pressure, a dresser 400 for dressing the surface of the polishing pad 200, and a polishing slurry supply device 500.

[0058] The polishing disc 100 is driven to rotate by a spindle, and the polishing pad 200 is attached to the upper surface of the polishing disc 100. The polishing head 300 is located above the polishing pad 200 and is used to press the wafer against the polishing pad 200 and to rotate the wafer during the polishing process. The dresser 400 is disposed in the edge area of ​​the polishing pad 200 and is used to dress the surface of the polishing pad 200 during the polishing gap or during the polishing process, removing the glaze layer on the surface and restoring the roughness and abrasive carrying capacity of the polishing pad 200.

[0059] Figure 2 This is a schematic diagram of a polishing slurry supply device 500 provided in one embodiment of this application. It may include a swing arm base 510, a swing arm 520, a drive mechanism 530, and at least two independently movable polishing slurry supply pipes. The swing arm base 510 is fixed to the frame of the chemical mechanical polishing equipment, and its position remains fixed after the equipment is installed, requiring no further manual adjustment. The swing arm 520 extends horizontally from the swing arm base 510 above the polishing pad 200. The length of the swing arm 520 is approximately aligned with the radial direction of the polishing pad 200. Figure 1 As shown.

[0060] In this embodiment, there are two independently movable polishing slurry supply lines: a first movable supply line 541 and a second movable supply line 542. Each line has a first supply port 541a and a second supply port 542a at its front end. The two movable supply lines are arranged parallel to each other and independently along the length of the swing arm 520. Each line is connected to an independent slurry source (such as a polishing slurry storage tank and a pumping system), and its flow rate and injection pressure can be independently controlled.

[0061] Each movable supply tube is connected to a set of drive mechanisms 530. The drive mechanism 530 includes a motor 531 and a ball screw 532. The motor 531 is preferably a servo motor or a stepper motor, characterized by fast response and high positioning accuracy. The ball screw 532 is a miniature precision ball screw with a lead of 1mm to 5mm and a positioning accuracy within ±0.05mm. The output shaft of the motor 531 is connected to one end of the ball screw 532 via a coupling, and the nut seat of the ball screw 532 is fixedly connected to the movable supply tube via a connector. When the motor 531 rotates, the ball screw 532 converts the rotational motion into linear motion of the nut seat, thereby driving the supply port to reciprocate along the length of the swing arm 520. Each movable supply tube is driven by an independent motor 531, therefore the movement stroke, speed, and direction of the supply ports on both sides can be independently controlled.

[0062] To achieve precise control of the supply port movement, the drive mechanism 530 also includes a position sensor and limit switches to limit the range of supply port movement. The position sensor can be a magnetic sensor or a photoelectric sensor to detect the current position of the nut seat or supply port in real time and feed the position signal back to the controller, forming a position closed loop. The limit switches are located at the extreme positions at both ends of the swing arm 520, triggering an emergency stop when the supply port movement exceeds the safe range to prevent mechanical collision.

[0063] The rear end of each polishing fluid supply line uses a spiral flexible hose (545) as the connection section, such as... Figure 2 As shown. The tail end of the spiral hose 545 is fixed to the swing arm base 510 and connected to the liquid supply source; the head end is fixedly connected to the nut seat of the ball screw 532 through a connector and moves with the nut seat. The spiral hose 545 is made of corrosion-resistant material, preferably polytetrafluoroethylene or perfluoroalkoxy resin, with an inner diameter of 3mm to 12mm and 3 to 8 spiral turns. The structural design of the spiral hose 545 allows it to freely expand and contract in the length direction, which not only moves with the supply port but also avoids the bending fatigue and wear problems caused by rigid pipelines in reciprocating motion.

[0064] Figure 1In the illustrated embodiment, the closed-loop feedback and model adaptation system for the polishing slurry supply also includes a controller. The controller can be a standalone hardware module, such as a PLC, embedded industrial computer, or industrial PC, or it can be integrated into the main control system of the CMP equipment. The controller is electrically connected to the motor 531 and is used to control the drive mechanism 530 to move the supply port according to the preset sweeping formula.

[0065] To achieve real-time monitoring of the polishing slurry distribution and wafer surface shape, the closed-loop feedback and model adaptive system for polishing slurry supply is equipped with at least one sensor. In this embodiment, the sensor includes an optical sensor 601 and an eddy current film thickness measurement unit 602. The optical sensor 601 is disposed above the polishing pad 200, for example, mounted on the swing arm 520 or a separate bracket, for online monitoring of the polishing slurry film thickness and its radial distribution on the surface of the polishing pad 200. The optical sensor 601 can be a laser reflective film thickness gauge or a spectral reflective film thickness gauge, which calculates the film thickness by measuring the reflection interference signal of the polishing slurry film layer to the incident light. The eddy current film thickness measurement unit 602 is disposed on the top surface of the polishing disk 100 to measure the metal film thickness on the wafer surface in real time through a detection window matched on the polishing pad 200, thereby calculating the removal rate distribution within the wafer surface. Both the optical sensor 601 and the eddy current film thickness measurement unit 602 transmit the monitoring data to the controller in real time via wired or wireless means.

[0066] The controller is configured to perform the following operations: (a) during the initial sweeping process according to the sweeping formula, acquire feedback data from the optical sensor 601 and / or the eddy current film thickness measurement unit 602 in real time; (b) compare the feedback data with the target surface shape (i.e., the desired wafer surface shape) and calculate the deviation; (c) based on a pre-established mathematical model of the polishing slurry flow field distribution, map the deviation to an adjustment amount of the sweeping parameters, wherein the mathematical model of the flow field distribution specifically involves dividing the polishing pad radially into N concentric annular regions and establishing the supply port movement trajectory parameter tra. p The mapping relationship between C(r) and the cumulative supply of polishing fluid in each annular region is C(r) = F(tra) p (a) The adjustment amount of the sweep parameters is obtained by inverse mapping F⁻¹ of the mapping relationship or by standardizing the mapping rule base, where N is an integer greater than or equal to 2. The sweep parameters include the moving stroke of the supply port, the moving speed curve, the endpoint dwell time, and the relative motion phase of the two supply ports; (d) The control command of the drive mechanism 530 is dynamically corrected according to the adjustment amount to realize the adaptive closed-loop adjustment of the supply port moving trajectory; (e) The above operations (a) to (d) constitute a complete closed-loop technical link from "data acquisition" to "control execution". Among them, tra p This is short for trajectory_params.

[0067] Specifically, the liquid film thickness distribution data on the polishing pad surface collected by the optical sensor 601 is input to the controller as a continuous function with radial position r as the independent variable and liquid film thickness h(r) as the dependent variable; the metal film thickness data on the wafer surface collected by the eddy current film thickness measurement unit 602 is simultaneously input to the controller with radial position r as the independent variable and film thickness T(r) as the dependent variable. The controller performs spatiotemporal registration and fusion processing on the two sets of data to generate real-time surface shape data P. actual (r). This real-time surface shape data and the target surface shape P pre-stored in the controller. target (r) performs point-by-point interpolation to obtain the surface deviation vector ΔP(r) = P actual (r)-P target (r).

[0068] Subsequently, this deviation vector is substituted as an input variable into the pre-established radial concentration distribution function C(r)=F(tra p The inverse function F -1 (tra p In this process, the required sweep parameter adjustment Δtra is obtained by iterative solution or table lookup interpolation. p Specifically, the inverse mapping process includes the following steps: First, the controller divides the polishing pad into N concentric annular regions, each annular region corresponding to a component Δp in the surface deviation vector ΔP. i Then, according to the mapping relationship C(r)=F(tra) p Determine the desired cumulative polishing slurry supply correction value ΔC within each annular region. i This allows the corrected concentration distribution to compensate for the corresponding deviation components; finally, through the inverse mapping F⁻¹ of the mapping relationship or by querying the standardized mapping rule base, the cumulative supply correction values ​​{ΔC1,ΔC2,…,ΔC} of each annular region are adjusted. N The inverse solution is the adjustment amount Δtra of the supply port movement trajectory parameter. p The inverse mapping is solved numerically using Newton's iteration method or the conjugate gradient method, and the mapping rule base is implemented using a lookup table, decision tree, or a pre-trained neural network model.

[0069] This adjustment includes, but is not limited to: the correction amount Δr for the endpoint position of the supply port travel. start and Δr end The scaling factor k of the moving speed curve v Increment of endpoint dwell time Δt stayThe controller converts the above adjustment amounts into pulse control signals for the motor 531, including pulse frequency, pulse quantity, and direction signals. These signals are amplified by the drive circuit and output to the motor 531 to achieve precise displacement control of the ball screw 532. The specific technical relationship between the above data flow and control flow is a complete mapping from the "raw sensor signal" to the "surface deviation vector," then to the "sweep parameter adjustment amount," and finally to the "motor drive command."

[0070] Through the closed-loop feedback control described above, the closed-loop feedback and model adaptive system of the polishing slurry supply can compensate in real time for the flow field distribution deviation caused by factors such as polishing pad wear, changes in polishing slurry properties, and fluctuations in ambient temperature during the polishing process, thereby significantly improving the uniformity of polishing slurry coverage and the controllability of wafer surface shape.

[0071] Figure 2 In the illustrated embodiment, the swing arm 520 extends horizontally in a long strip shape, and two independent drive mechanisms 530 are installed parallel to each other inside or on its surface. A first ball screw and a second ball screw are arranged parallel to each other along the length of the swing arm 520 and are driven by a first motor and a second motor, respectively. A first movable supply pipe 541 and a second movable supply pipe 542 are fixed to their respective ball screw nut seats via connectors. The front end of each movable supply pipe extends from the front end of the swing arm 520, with the supply port facing the surface of the polishing pad 200. The two movable supply ports are arranged side-by-side in the width direction of the swing arm 520, and their landing points are substantially aligned in the radial direction of the polishing pad 200.

[0072] Furthermore, the polishing slurry supply device 500 also includes a fixed-position polishing slurry supply port 543a. The fixed supply port 543a is located at the front end of the fixed supply pipe 543, which is positioned at the middle of the swing arm 520, and is used to provide a basic polishing slurry flow rate. A first movable supply port 541a and a second movable supply port 542a are located on the left and right sides of the fixed supply port 543a, respectively, and the line connecting their centers is parallel to the length direction of the swing arm 520. The position of the fixed supply port 543a is determined during the equipment design phase and remains stationary during use. The controller can select different supply modes according to process requirements: for example, in the rough polishing stage, only two movable supply ports can be used for wide-area sweeping to achieve broad coverage of the polishing slurry on the polishing pad; in the fine polishing stage, both the fixed supply port and the movable supply ports can be used simultaneously, with the fixed supply port providing a stable basic flow rate in the central area and the movable supply ports adjusting the concentration in the edge or local areas.

[0073] Furthermore, the swing arm 520 also integrates a high-pressure water nozzle 550, located at the front end of the swing arm 520 or near the supply port, for cleaning the movable and fixed supply ports during the polishing interval. The high-pressure water nozzle 550 is connected to a deionized water high-pressure supply system, with a spray pressure reaching several megapascals. When the controller executes the cleaning program, it drives the movable supply port to perform a full-stroke reciprocating motion at a speed 1.5 to 3 times higher than the sweeping speed during polishing, while simultaneously coordinating with the high-pressure water nozzle 550 to perform pulsed spraying. Utilizing the impact force of the high-speed water flow and the sweeping effect generated by the rapid movement of the supply port, it effectively removes polishing fluid deposits remaining on the pipes and nozzle inner walls, preventing abrasive agglomeration and clogging.

[0074] Combined with the preceding text Figure 1 and Figure 2 The system hardware structure of the embodiments of this application has been described in detail. The following will be combined with... Figures 3 to 5 The mathematical model of the polishing fluid flow field distribution and the mapping rule base on which the system relies to achieve closed-loop feedback regulation are further elaborated.

[0075] To achieve the aforementioned closed-loop feedback regulation, this application pre-established a mathematical model of the polishing fluid flow field distribution. For example... Figure 3 As shown, the polishing pad 200 is divided into N concentric annular regions from the center to the edge. For example, the figure shows six regions, labeled as regions I, II, III, IV, V, and VI, where region I is the central region and region VI is the outermost edge region. When the movable supply port sweeps along the length of the swing arm 520, the landing position of the supply port on the polishing pad 200 changes over time. The time the supply port stays within each annular region, the sweeping speed, and the jet flow rate of the polishing slurry collectively determine the cumulative supply of polishing slurry within that region.

[0076] Let the trajectory parameter of the supply port be tra p This parameter set includes: starting position r start That is, the position of the feed port in the radial direction of the polishing pad when it begins to sweep; the termination position r endThis refers to the position of the supply port when it finishes sweeping; the moving speed as a function of time, v(t), i.e., the speed curve, which can be uniform, variable, or segmented variable; and the relative motion curves of the two supply ports, such as reverse symmetrical motion, synchronous motion in the same direction, or differential motion. Specifically, the reverse symmetrical motion mode refers to the first moving supply port 541a and the second moving supply port 542a moving synchronously back and forth in opposite directions with the same stroke range and speed, so that when one supply port moves towards the center of the polishing pad, the other supply port moves towards the edge of the polishing pad at the same time, thereby achieving a balanced redistribution of polishing fluid in the radial direction; the synchronous motion mode refers to the two supply ports moving synchronously with the same speed and direction, so that the two polishing fluids are superimposed at the same radial position, which is suitable for increasing the concentration in a local area; the differential motion mode refers to the two supply ports sweeping with different speeds or different stroke ranges, so that the polishing fluid forms a differentiated supply distribution in different radial areas of the polishing pad, which is suitable for compensating for multiple types of surface shape deviations at the same time. The controller automatically selects the most suitable mode from the three relative motion modes mentioned above, or dynamically switches between them, based on the type and distribution characteristics of the real-time surface deviation. The cumulative supply of polishing slurry at each position radially on the polishing pad can then be expressed as the radial concentration distribution function C(r) = F(tra). p ), where r is the radial position coordinate.

[0077] Specifically, the mapping relationship C(r) = F(tra) p The process of establishing the polishing pad includes: First, dividing the polishing pad radially into N concentric annular regions, each annular region corresponding to a radial position interval [r]. i ,r i+1 Then, for a given sweep parameter tra p The cumulative residence time of the supply port in each annular region is calculated, and combined with the jet flow rate of the supply port, the cumulative supply amount C of polishing fluid in each annular region is obtained. i Finally, the cumulative supply of the N annular regions is combined into a radial concentration distribution vector C=[C1,C2,…,C…]. N ] T That is, C(r) = F(tra) p The inverse mapping F⁻¹ is achieved as follows: given a target concentration distribution C target (r), which can be solved by iterative solution or table lookup interpolation to obtain the solution satisfying C. target (r)=F(tra p The sweeping parameter tra p The table lookup interpolation method relies on a standardized mapping rule base, and the iterative solution uses Newton's iteration method or the conjugate gradient method for numerical solution.

[0078] The mapping relationship F was determined jointly through fluid dynamics simulation and CMP process test data fitting. Specifically, firstly, a three-dimensional simulation model including polishing pad rotation, polishing slurry injection, and flow was established using computational fluid dynamics (CFD) software. Different combinations of sweeping parameters were input to simulate the cumulative supply of polishing slurry in each annular region. Then, a process verification experiment was conducted on a CMP device to measure the actual wafer surface shape after polishing. The simulation results were compared and corrected with the actual measurement data. Through multiple rounds of iterative fitting, a high-precision mathematical model was finally obtained, which can accurately predict the flow field distribution on the polishing pad under any given sweeping parameters.

[0079] Furthermore, the controller has a built-in standardized mapping rule base. This rule base records the correspondence between different types of surface shape deviations and the optimal sweep parameter adjustment amount. Types of surface shape deviations include, but are not limited to: center bulge (the thickness of the wafer's center region is higher than that of the edge region, indicating a low removal rate in the center region); edge sag (the thickness of the wafer's edge region is higher than that of the center region, indicating a low removal rate in the edge region); center depression (abnormal thickness in the center region of the wafer); and local high points. The rule base can be implemented using lookup tables, decision trees, or pre-trained neural network models. After the controller calculates the real-time surface shape deviation, it matches or deduces the corresponding sweep parameter adjustment amount from the rule base and outputs it to the drive mechanism 530.

[0080] Figure 4 This paper demonstrates the offline construction process of a standardized mapping rule base. First, a multi-factor orthogonal experimental scheme is designed to determine the sweep parameter combinations to be traversed, including the stroke ratio (the ratio of the stroke lengths of the two feed ports); the speed ratio (the ratio of the moving speeds of the two feed ports); the phase difference (the difference in the starting times of the two feed ports); and the endpoint dwell time (the duration for which the feed port stops moving after reaching the endpoint and continues spraying polishing slurry). Then, polishing experiments are conducted on a test wafer according to each set of parameters, and the wafer surface shape after polishing is measured using an eddy current film thickness measurement unit 602. Next, multivariate statistical analysis methods such as partial least squares or support vector regression are used to establish an inverse mapping model from surface shape deviation to sweep parameter adjustment. This model is embedded in the controller in the form of a coefficient matrix or rule table for use in online applications.

[0081] Figure 5 The online application process of the rule base is demonstrated: the controller receives real-time feedback data from the sensor, calculates the real-time surface deviation vector, and then inputs the deviation vector into the mapping rule base for querying or inference. The rule base matches the corresponding sweep parameter adjustment amount, and the controller converts the adjustment amount into a drive command and outputs it to the motor 531.

[0082] It should be noted that the rule base can be periodically updated based on historical process data. The controller is equipped with a self-learning module, which records the causal relationship between the surface deviation change trajectory and the applied sweep parameter adjustment during each closed-loop adjustment process. It continuously optimizes the coefficients of the mathematical model and the accuracy of the mapping rule base through reinforcement learning or linear regression methods, enabling the system to continuously improve its intelligence level during long-term operation.

[0083] Specifically, the data accumulation method of the self-learning module is as follows: After each polishing process, the controller will accumulate the "initial surface deviation vector ΔP" for the entire polishing process. initial "The applied sweep parameter adjustment sequence {Δtra p 1, Δtra p 2, ..., Δtra p m}” and “final surface deviation vector ΔP” final Each training sample is stored in the database as a complete training sample. Each training sample also includes process condition tags, such as polishing pad usage time, polishing fluid batch number, and wafer type, to facilitate subsequent modeling based on process conditions. The database uses a time-series structure, indexed by timestamps, and supports data retrieval and filtering based on process conditions.

[0084] The self-learning module's model update strategy employs a hybrid approach of "primarily batch updates, supplemented by online fine-tuning": after accumulating 50 new training samples, a batch retraining is triggered, refitting the model parameters of the mapping rule base using all historical data; between two batch updates, the controller uses an online learning algorithm to fine-tune the model parameters in real time to adapt to gradually changing factors such as progressive wear of the polishing pad. Batch updates are executed during equipment standby or preventative maintenance to avoid impacting normal polishing operations.

[0085] The self-learning module also includes a confidence evaluation mechanism. Specifically, when the controller calls the mapping rule base for inference, it simultaneously calculates the similarity between the current input bias vector ΔP and the bias vectors of each training sample in the rule base, using Euclidean distance as the similarity metric. If the Euclidean distance to the nearest neighbor sample is less than a preset threshold d... th (d in this embodiment) th If the Euclidean distance is less than or equal to 0.15, it is considered a high confidence level, and the rule base inference result is directly output; if the Euclidean distance is between d and d, it is considered a high confidence level. th With 2D th If the distance is between 2d and 3d, it is determined to be of medium confidence, and a conservative coefficient (0.7 in this example) is added to the rule base inference result to avoid excessive adjustment; if the Euclidean distance is greater than 2d... thIf the confidence level is low, it indicates that the current surface deviation pattern has never appeared in the historical data. The controller triggers a new orthogonal test to supplement the data, and at the same time, it suspends closed-loop adjustment and only operates in open-loop sweep mode until the test is completed.

[0086] To verify the actual effectiveness of the self-learning module, the applicant tracked the changes in the prediction accuracy of the rule base during 200 consecutive polishing operations. Prediction accuracy is defined as: the adjustment amount Δtra of the rule base output. p_predicted Compared with the actual optimal adjustment amount Δtra p_optimal The normalized root mean square error (NRMSE) between the two is calculated. Initially (after the first polishing), the prediction NRMSE of the rule base is 12.5%. After 50 polishing data accumulations and one batch update, the NRMSE decreases to 7.8%. After 100 polishings, the NRMSE further decreases to 5.2%. After 200 polishings, the NRMSE stabilizes at around 4.1%, an improvement of 67.2% compared to the initial state. The above data shows that the self-learning module can significantly improve the prediction accuracy and process adaptability of the mapping rule base through continuous data accumulation and model optimization.

[0087] After completing the description of the system hardware structure, flow field distribution mathematical model, and mapping rule base, the following section combines... Figure 6 This application provides a detailed description of the specific process of the closed-loop feedback and model adaptive control method for polishing slurry supply provided in the embodiments of this application.

[0088] Based on the aforementioned closed-loop feedback and model adaptive system and mathematical model for polishing slurry supply, this application provides a closed-loop feedback and model adaptive control method for polishing slurry supply. For example... Figure 6 As shown, the method includes the following steps:

[0089] Step S1 involves pre-establishing a mathematical model of the polishing slurry flow field distribution and constructing a standardized mapping rule library between surface deviation and sweeping parameter adjustment amounts. For details on the construction method, please refer to [link to documentation]. Figure 4 The process and related descriptions shown are not repeated here.

[0090] Step S2: Select an initial sweep recipe based on the initial surface shape of the target wafer, for example, by selecting an initial sweep recipe based on a wafer thickness distribution map obtained through offline measurement. The initial sweep recipe can be from a historically successful recipe in the process database that is closest to the current process conditions, or it can be a starting value set by the process engineer based on experience. The controller controls the movable feed ports on both sides to perform an open-loop sweep motion according to the initial stroke, speed, and dwell time.

[0091] In step S3, during the polishing process, the radial distribution data of the polishing liquid film thickness on the polishing pad surface is acquired in real time by the optical sensor 601, and the change data of the metal film thickness on the wafer surface is acquired in real time by the eddy current film thickness measurement unit 602. The controller fuses these two types of data to generate real-time wafer surface shape data.

[0092] Specifically, the optical sensor 601 collects the liquid film thickness value h at N sampling points (N=50 in this embodiment, and the sampling interval is 3mm) at equal intervals along the radial direction of the polishing pad. i (i=1,2,…,N), forming a liquid film thickness distribution vector H=[h1,h2,…,h N ] T The eddy current film thickness measurement unit 602 simultaneously acquires the metal film thickness value T at the corresponding radial position on the wafer surface. i This forms the film thickness distribution vector T=[T1,T2,…,T N ] T The controller inputs the H and T vectors into a pre-trained data fusion model, which uses a Kalman filter algorithm to optimally estimate the data from the two different types of heterogeneous sensors, and outputs a real-time surface shape vector P. actual =[p1,p2,…,p N ] T , where p i This represents the normalized material removal amount or remaining film thickness at the i-th radial position. The above data fusion process takes into account the differences in sampling frequency and spatial resolution between the optical sensor and the eddy current sensor, and achieves precise alignment of the two types of data in the spatiotemporal dimensions through timestamp synchronization and spatial interpolation algorithms.

[0093] Step S4: Compare the real-time surface shape data with the target surface shape point by point, and calculate the deviation vector. Target surface shape vector P target =[q1,q2,…,q N ] T Pre-stored in the controller's memory unit, it represents the ideal surface shape expected to be achieved in this process. Deviation vector ΔP=P actual -P target =[Δp1,Δp2,…,Δp N ] T , where Δp i =p i -q i When ||ΔP||∞ is less than the preset threshold ε (ε=5% in this embodiment), the surface shape is determined to meet the process requirements and no adjustment is needed; when ||ΔP||∞≥ε, the controller starts the closed-loop adjustment program. Here, ||ΔP||∞ is the maximum absolute value of the deviation at each point; in this embodiment, the preset threshold ε is 5%.

[0094] Step S5: Input the deviation vector into the mathematical model of the flow field distribution, and obtain the adjustment amount of the sweep parameters through the inverse solution of the mapping rule base.

[0095] Specifically, the deviation vector ΔP is first input into the positive model F of the mathematical model of the flow field distribution. This model divides the polishing pad into N concentric annular regions (corresponding one-to-one with the sampling points) and establishes the supply port movement trajectory parameters tra. p = (r start r end The mapping relationship between v(t), φ(t)) and the cumulative supply of polishing fluid in each annular region is calculated by the controller. p_current Below, the theoretical surface shape P predicted by the forward model F. predicted =F(tra p_current Then calculate the prediction bias ΔP. predicted =P actual -F(tra p_current If ΔP predicted If the difference between the model and ΔP exceeds a preset threshold, it indicates that the current model accuracy is insufficient. In this case, the online model correction subroutine is triggered, and the model coefficients are fine-tuned using the recursive least squares method.

[0096] After the model correction is completed, the controller inputs the deviation vector ΔP into the inverse mapping F⁻¹ of the flow field distribution mathematical model or the standardized mapping rule base for inverse solution. Specifically, the inverse solution process includes: first, inputting each component Δp of the surface deviation vector ΔP into the inverse mapping F⁻¹ of the flow field distribution mathematical model for inverse solution. i Mapped to the correction value ΔC of the cumulative supply of polishing slurry within each concentric annular region. i This allows the corrected concentration distribution to compensate for the corresponding surface shape deviation; then, the cumulative supply correction values ​​{ΔC1,ΔC2,…,ΔC} of the N annular regions are adjusted using the inverse mapping F⁻¹. N The inverse solution is the adjustment amount Δtra of the supply port movement trajectory parameter. p The inverse mapping is solved numerically using Newton's iteration method or the conjugate gradient method. When the iteration of the inverse mapping fails to converge or the convergence speed does not meet the real-time control requirements, the controller switches to the standardized mapping rule base and directly outputs the adjustment amount Δtra through table lookup matching or neural network inference. p This is to ensure the real-time performance of closed-loop control.

[0097] The mapping rule base stores an inverse mapping model G, which is pre-established through multi-factor orthogonal experiments. This model takes the bias vector ΔP as input and the sweep parameter adjustment Δtra as input. p For the output, i.e., Δtra p=G(ΔP). The mapping rule base is implemented using a pre-trained neural network model. In this embodiment, a three-layer BP neural network is used, with 20 hidden layer nodes and ReLU activation function. This network model has been trained to convergence through a large amount of experimental data and can complete the nonlinear mapping inference from the bias vector to the adjustment amount within 10ms of feedforward. The adjustment amount Δtra of the inference output is... p =(Δr start , Δr end Δk v , Δt stay Δφ) is decomposed into specific driving parameters: stroke end position correction Δr start and Δr end (Unit: mm), velocity curve scaling factor Δk v (Dimensionless, range 0.5~2.0), Endpoint dwell time increment Δt stay (Unit: s), and the offset Δφ of the relative motion phase between the two supply ports (unit: rad).

[0098] Step S6: The adjustment amount is sent to the drive mechanism 530 to dynamically correct the movement trajectory parameters of the supply port. That is, the updated sweep parameters are tra p_new =tra p_current +Δtra p The controller will tra p_new The parameters in the controller are converted into pulse control commands for the motor 531. Specifically, the controller converts the adjusted stroke endpoint coordinates, speed curve function, dwell time, and other parameters into pulse control commands for the motor 531, which precisely drive the ball screw 532 through the PID control loop, so that the supply port moves according to the new trajectory. Specifically, when the controller corrects the relative motion phase of the two supply ports according to the adjustment amount Δφ output by the mapping rule library, if the absolute value of Δφ exceeds a preset threshold (π / 3 in this embodiment), the controller automatically switches the relative motion mode of the two supply ports: when Δφ is close to π, the controller switches the two supply ports to a reverse symmetrical motion mode, so that the two supply ports move synchronously in opposite directions, realizing a balanced redistribution of polishing fluid in the radial direction; when Δφ is close to 0, the controller switches the two supply ports to a synchronous motion mode in the same direction, so that the two supply ports move synchronously at the same speed and direction, forming a polishing fluid supply superposition effect in the target area; when Δφ is between 0 and π, the controller adopts a differential motion mode, so that the two supply ports sweep at different speeds or different stroke ranges, forming a differentiated concentration distribution in different radial areas of the polishing pad. Through the above phase adaptive switching mechanism, it is ensured that the flow field distribution after phase adjustment matches the correction requirements of the target surface deviation, realizing accurate compensation for various surface deviation types.

[0099] Step S7: Repeat steps S3 to S6 until polishing is complete or the surface deviation converges to within the allowable range. Figure 6 As shown, the feedback loop returns from step S7 to step S3, forming a continuous closed-loop feedback process. After each adjustment, the sensor will monitor again, and the controller will correct again based on the new deviation until the deviation is less than the threshold or the polishing time ends.

[0100] It should be emphasized that when the controller performs adaptive closed-loop regulation, it can use independent PID control loops for the two supply ports respectively, or it can use a coupling control strategy to make the movement of the two supply ports meet the preset relative motion relationship, such as always maintaining symmetrical reverse movement, or moving in the same direction according to a fixed ratio. The coupling control strategy includes the following three relative motion modes: (1) Reverse symmetrical mode - the two supply ports move synchronously in opposite directions with the same stroke range and speed. When one supply port moves towards the center of the polishing pad, the other side moves towards the edge at the same time, realizing the balanced redistribution of polishing liquid in the radial direction, which is suitable for compensating for the overall surface tilt deviation; (2) Same direction synchronous mode - the two supply ports move synchronously with the same speed and direction, so that the two polishing liquids are superimposed at the same radial position, which is suitable for increasing the concentration of local areas to compensate for local depressions or low removal rates; (3) Differential mode - the two supply ports sweep at different speeds or different stroke ranges, forming a differentiated concentration distribution in different radial areas of the polishing pad, which is suitable for compensating for multiple types of composite surface deviations at the same time. The controller automatically selects the most suitable mode from the three relative motion modes mentioned above, or dynamically switches between them, based on the type, location, and magnitude of the surface deviation, in order to achieve accurate compensation for the surface deviation.

[0101] When the optical sensor 601 detects that the uniformity index of the polishing slurry distribution exceeds the threshold, the controller prioritizes adjusting the moving speed curve and the endpoint dwell time, because these two parameters have less impact on the mechanical system and respond faster. When the uniformity index continues to deviate, the moving stroke is then adjusted to avoid frequent reversals of the drive mechanism that could cause mechanical wear. In this application, the so-called coupled control strategy refers to a preset functional relationship between the moving parameters (such as speed, stroke, and phase) of the two supply ports, rather than completely independent control. For example, when the moving speed of one supply port changes, the moving speed of the other supply port changes synchronously according to a preset ratio.

[0102] To more intuitively illustrate the application effects of the aforementioned closed-loop feedback and model adaptive control methods in actual processes, the following will combine... Figure 7 and Figure 8 The specific adjustment examples shown will be used to illustrate this.

[0103] Figure 7This illustrates a typical anomaly encountered in CMP processes: a low material removal rate in the wafer edge region. For example... Figure 7 As shown, the horizontal axis represents the radial position of the wafer (from the center to the edge), and the vertical axis represents the wafer surface film thickness or material removal amount. The solid lines in the figure represent the measured surface shape. It can be clearly seen that in the wafer edge region, the measured curve is significantly higher than in other regions, indicating that the material removal amount in this region is insufficient.

[0104] Once the controller identifies the aforementioned surface shape deviation, it performs a lookup or inference using the mapping rule base. If the deviation vector indicates a low removal rate in the wafer edge region, the adjustment output by the mapping rule base includes: extending the travel endpoint of the movable supply port located near the edge of the polishing pad outward by 2mm to 5mm, while simultaneously extending the dwell time of this endpoint by 0.5s to 2s, and reducing the moving speed of the supply port on the other side to increase the local polishing slurry concentration. The physical principle behind these adjustments is that the landing point of the edge-side supply port is further moved towards the edge of the polishing pad, and its dwell time in that area is increased, allowing more polishing slurry to be delivered to the polishing pad region corresponding to the wafer edge; at the same time, the moving speed of the center-side supply port is reduced, slightly decreasing the amount of polishing slurry supplied to the center region, thereby achieving a redistribution of polishing slurry from the center region to the edge region.

[0105] If the deviation vector indicates that the removal rate in the center region of the wafer is too low, the output will be the opposite adjustment amount: the movement stroke of both supply ports will be shifted towards the center of the polishing pad, and a co-directional aggregation motion mode will be adopted, that is, both supply ports will move towards the center at the same time and converge near the center, so as to increase the residence time and supply amount of polishing fluid in the center region.

[0106] Figure 8 The surface profile curves of the same batch of wafers are shown after applying the closed-loop feedback and model adaptive control of this invention. From Figure 8 It can be seen that after dynamically adjusting the sweeping parameters, the measured surface shape curve of the wafer edge region is basically consistent with that of the middle region. The problem of low edge removal rate has been effectively compensated, the overall in-plane non-uniformity of the wafer has been significantly reduced, and the flatness has met the process requirements.

[0107] The system and method provided in this application can be further expanded with multi-stage control and anomaly diagnosis functions to better adapt to complex process requirements and improve operational safety.

[0108] In actual CMP processes, different polishing stages have different requirements for surface shape adjustment. The controller of this invention supports the simultaneous operation of two or more closed-loop control threads, each corresponding to a different polishing stage. For example, the rough polishing stage uses a larger stroke range and a faster sweeping speed to quickly cover a large polishing pad; the fine polishing stage uses a smaller stroke range and a slower sweeping speed, combined with fine feedback adjustment to optimize local flatness. The mapping rule base parameters for different stages are stored in the controller, which automatically switches between them based on polishing time or cumulative removal amount. For example, when the cumulative removal amount reaches 80% of the total removal amount, the controller automatically switches from the rough polishing stage to the fine polishing stage, while simultaneously adjusting the sweeping parameters and PID control parameters accordingly.

[0109] To improve the safety and reliability of the closed-loop feedback and model-adaptive system for polishing slurry supply, this method also includes an anomaly diagnosis step. The controller monitors the residual between the sensor feedback data and the theoretical distribution predicted based on the flow field mathematical model in real time. Under normal circumstances, the feedback data and theoretical predictions should be basically consistent, with a small residual. When the residual exceeds a preset threshold (e.g., 20%) and persists for multiple sampling cycles (e.g., 5 consecutive cycles), the controller determines that an unexpected anomaly has occurred. Possible causes of the anomaly include: excessive wear of the polishing pad leading to changes in surface morphology, deterioration of the polishing slurry (e.g., abrasive agglomeration, pH drift), and partial blockage of the supply port leading to a decrease in flow rate. The controller then issues an alarm signal (e.g., audible and visual alarm or sending an error code to the main control system) and automatically invokes the self-cleaning program (starting the high-pressure water nozzle 550 and rapid reciprocating motion) to attempt to clear the blockage; if the residual still does not recover after self-cleaning, the controller prompts the operator to check and replace the polishing pad or polishing slurry.

[0110] To effectively distinguish between the various potential causes of anomalies, the controller incorporates an anomaly diagnosis decision tree. By fusing multi-dimensional monitoring data from multiple sensors, it achieves automatic classification of anomaly types. These multiple sensors may include an optical sensor 601, an eddy current film thickness measurement unit 602, and flow and pressure sensors installed on the liquid supply line, etc., to fully acquire monitoring data.

[0111] Specifically, the judgment logic of this anomaly diagnosis decision tree is as follows:

[0112] First, the controller reads the actual flow rate Q of the polishing fluid detected by the flow sensor. actual With the set flow rate Q set The deviation between them is ΔQ=Q actual -Q set If ΔQ is less than -15% (i.e., the actual flow rate drops by more than 15% compared to the set value), and the supply line pressure P... supplyIf the flow rate increases by more than 10%, it is determined that the supply port is partially blocked. This is because the reduced cross-sectional area of ​​the flow channel leads to a decrease in flow rate and an increase in pressure. At this time, the controller automatically calls the self-cleaning program to try to clear the blockage. If the flow rate recovers after self-cleaning, it is confirmed that the blockage has been cleared. If the flow rate does not recover after self-cleaning, it is determined that the supply port is completely blocked by hard particles (such as solidified abrasive), and manual cleaning or replacement of the supply port component is required.

[0113] If ΔQ is within ±15% (i.e., the flow rate is normal), the controller further compares the wafer material removal rate (MRR) fed back by the eddy current film thickness measurement unit 602. actual MRR with target removal rate target The deviation between them is ΔMRR = MRR actual -MRR target If ΔMRR is less than -20% (the actual removal rate decreases by more than 20% compared to the target value), and the radial distribution of the polishing liquid film thickness detected by the optical sensor 601 is normal (i.e., the liquid film thickness at each radial position is not obviously abnormal), then it is determined that the polishing pad is excessively worn. The microporous structure on the surface of the polishing pad tends to be smooth due to long-term use, resulting in a decrease in mechanical grinding ability. In this case, the controller recommends replacing the polishing pad.

[0114] If ΔQ and ΔMRR are normal or have only slight deviations, but the radial distribution pattern of the polishing fluid film thickness detected by the optical sensor 601 shows abnormal changes (e.g., a local sharp drop in film thickness at a certain radial position, or a significant reduction in film coverage), the controller further reads the monitoring data from the pH sensor or electrochemical sensor installed in the polishing fluid storage tank. If the pH value deviates from the set range (in this embodiment, the set pH range is 10.2 to 11.0; a deviation exceeding ±0.3 is considered abnormal), it is determined that the polishing fluid has deteriorated, and the controller recommends replacing the polishing fluid; if the pH value is normal, it is determined to be an abnormality of other unknown causes, and the controller issues a general alarm signal and records the abnormal data for subsequent analysis.

[0115] The judgment process of the above-mentioned anomaly diagnosis decision tree is executed in real time in the controller, with a response time of less than 100ms from sensor data update to anomaly type determination output. When the anomaly is determined to be a supply port blockage, the controller prioritizes the self-cleaning procedure; when the anomaly is determined to be polishing pad wear or polishing fluid deterioration, the controller does not execute the self-cleaning procedure (because self-cleaning is ineffective for such anomalies), but instead directly sends a maintenance request to the equipment's main control system. This mechanism effectively avoids invalid self-cleaning operations due to misjudgment, improving equipment maintenance efficiency.

[0116] The above embodiment is illustrated using a configuration of two movable supply ports plus a single fixed supply port as an example, but this application is not limited thereto. Several alternative variations are further described below.

[0117] In some simplified embodiments, only two movable supply ports may be provided instead of a fixed supply port. In this case, both supply ports participate in the sweeping motion, and the controller coordinates their relative motion phases to achieve different flow field distributions. On larger polishing pads, three or more movable supply ports can be provided, each controlled by an independent drive mechanism. The controller coordinates the movement of multiple supply ports simultaneously based on the flow field mathematical model to further optimize the uniformity of polishing fluid coverage.

[0118] In addition, alternatives to spiral hoses include corrugated pipes or expansion sleeves, but spiral hoses have advantages such as small bending radius, wear resistance and easy installation, making them the preferred option.

[0119] As mentioned earlier, the swing arm 520 integrates a high-pressure water nozzle 550. In the specific cleaning mode, the controller first stops the polishing fluid supply and then starts the high-pressure water supply system. Next, the controller drives the two movable supply ports to perform a full-stroke reciprocating motion at a high speed (e.g., 40mm / s to 60mm / s, 1.5 to 3 times the polishing speed), while simultaneously controlling the high-pressure water nozzle 550 to spray deionized water in a pulsed manner (e.g., once every 0.5s, each pulse lasting 0.2s). The rapid movement of the supply ports combined with the high-pressure water pulse spray effectively flushes the supply ports across the entire width of the polishing pad, removing polishing fluid residue adhering to the inner wall of the nozzle and the pipe outlet. The cleaning program automatically stops after 10 to 30 seconds, and the system returns to standby mode.

[0120] To verify the technical effectiveness of this application, the applicant conducted comparative experiments on the same CMP machine (300mm wafer oxide CMP process), using the following three polishing schemes: Scheme A is the traditional fixed feed port scheme (the landing point is fixed at the midpoint of the polishing pad radius); Scheme B is the single-pipe open-loop sweeping scheme (a single movable feed port reciprocates according to a fixed sweeping formula, without feedback adjustment); Scheme C is the closed-loop feedback and model adaptive scheme of this application (dual movable feed ports + real-time sensor feedback + flow field mathematical model + closed-loop adjustment of mapping rule base). Each experiment used the same wafer, polishing pad, polishing slurry, and process parameters (polishing time 120s, polishing pad speed 90rpm, polishing head pressure 3psi), only the polishing slurry supply method was different. Each experiment was repeated 10 times, and the average value was taken as the final result.

[0121]

[0122] The above experimental data show that:

[0123] (1) Regarding in-plane non-uniformity (WIWNU), the proposed solution C achieves 2.1%, which is a 75.9% improvement over the conventional fixed solution A (8.7%) and a 59.6% improvement over the open-loop sweep solution B (5.2%). This improvement far exceeds the reasonable expectations of those skilled in the art—under the same process conditions, in-plane non-uniformity can be reduced to about one-quarter of that of the conventional solution simply by changing the polishing slurry supply method.

[0124] (2) Regarding the removal rate deviation in the edge and center regions, Scheme C has a deviation of only +1.8% and -1.2%, respectively, while Scheme A has a deviation of +15.3% and -9.4%, and Scheme B has a deviation of +8.1% and -4.6%, respectively. This indicates that this application can accurately distribute the polishing slurry to the region with the most severe surface deviation, and achieve fine control of the removal rate in each region of the wafer's radial direction. This is a technical effect that cannot be achieved by traditional fixed supply and open-loop sweeping schemes.

[0125] (3) In terms of polishing slurry utilization, scheme C reached 89%, which is 111.9% higher than scheme A (42%) and 41.3% higher than scheme B (63%). Polishing slurry is one of the most expensive consumables in CMP process. The significant improvement in utilization directly reduces process cost and reduces waste liquid discharge, which has significant economic benefits and environmental value.

[0126] (4) In terms of batch consistency (characterized by surface shape standard deviation), the standard deviation of scheme C is 0.6%, which is much lower than that of scheme A (3.2%) and scheme B (2.1%). This shows that this application not only improves the surface quality of a single batch, but also significantly improves the process repeatability and stability between different batches, and eliminates process fluctuations caused by factors such as manual calibration error and batch differences of polishing pads.

[0127] (5) To further verify the accuracy of the mathematical model of the flow field distribution, the applicant conducted field measurements of the flow field distribution under different combinations of sweeping parameters, and compared the radial concentration distribution C predicted by the mathematical model with the actual flow field distribution. predicted (r) and the measured concentration distribution C measured (r) Comparison was performed. In all 24 sets of experiments, the average absolute error between the model predictions and the measured values ​​was 3.2%, the maximum absolute error was 5.8%, and the root mean square error was 4.1%. The above accuracy indicators show that the mathematical model of flow field distribution established in this application can accurately describe the distribution law of polishing fluid under different sweeping parameters, providing a reliable decision basis for closed-loop feedback control.

[0128] As can be seen, this application, through a combination of technologies including "dual movable supply ports + real-time sensor feedback + flow field distribution mathematical model + standardized mapping rule base," achieves significantly improved performance compared to existing technologies in multiple dimensions, such as wafer surface uniformity, controllable edge / center region removal rate, polishing slurry utilization, and batch-to-batch consistency. These improvements are not simply the sum of individual technical features, but rather stem from the synergistic effect of the specific technical correlation between sensor data and the mathematical model—real-time sensor data corrects model prediction biases, model prediction results guide control command output, and the new state after control execution is recaptured by the sensor, forming a continuously optimizing positive feedback loop. The overall technical effect produced by this integrated closed-loop architecture of "perception-modeling-decision-execution" is significantly better than the sum of the effects of using each feature individually, constituting the high level of inventiveness of this application.

[0129] In summary, the closed-loop feedback and model-adaptive system and method for polishing slurry supply provided by this invention, through the setting of at least two independently movable polishing slurry supply ports, combined with real-time monitoring by the optical sensor 601 and the eddy current film thickness measurement unit 602, a flow field distribution mathematical model, and a standardized mapping rule library, achieves dynamic closed-loop adjustment of the polishing slurry landing point. Compared with the prior art, this invention significantly improves the uniformity of polishing slurry coverage on the polishing pad and the controllability of the wafer surface shape, reduces the difficulty of process development, and improves the safety and convenience of equipment operation and maintenance. The above embodiments are only preferred embodiments of this invention and are not intended to limit the scope of protection of this invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0130] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application.

[0131] The above embodiments are only used to illustrate the embodiments of this application, and are not intended to limit the embodiments of this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this application. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of this application, and the patent protection scope of the embodiments of this application should be defined by the claims.

Claims

1. A closed-loop feedback and model adaptive system for polishing slurry supply, characterized in that, include: Swing arm base; At least two independently movable polishing slurry supply lines, each with a polishing slurry supply port at the front end; A drive mechanism connected to each pipeline, the drive mechanism including a motor and a ball screw, is used to drive the supply port to reciprocate along the length of the swing arm; The controller, electrically connected to the motor, is used to control the drive mechanism to move the supply port according to the preset sweeping formula; At least one sensor, positioned above the polishing pad or near the polishing head, is used to monitor in real time the uniformity of the polishing slurry distribution on the polishing pad and / or the wafer removal rate distribution parameters. The controller is configured to: During the initial sweeping process according to the sweeping formula, feedback data from the sensors is acquired in real time. The feedback data is compared with the target surface shape, and the deviation is calculated; Based on the pre-established mathematical model of the polishing fluid flow field distribution, the deviation is mapped to the adjustment amount of the sweeping parameters, which include the moving stroke of the supply port, the moving speed curve, the endpoint dwell time, and the relative motion phase of the two supply ports. The control commands of the drive mechanism are dynamically corrected according to the adjustment amount to achieve adaptive closed-loop adjustment of the supply port movement trajectory.

2. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, The sensor includes at least one of the following: an optical sensor for online monitoring of the polishing fluid film thickness and radial distribution on the surface of the polishing pad; an electrochemical sensor for detecting the polishing fluid composition or pH value; an eddy current or optical film thickness measurement unit for measuring the removal rate distribution within the wafer surface; and a high-speed image acquisition unit for capturing the flow pattern of the polishing fluid; the sensor transmits the monitoring data to the controller in real time via wired or wireless means.

3. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, The mathematical model for the polishing slurry flow field distribution is established as follows: the polishing pad is divided into N concentric annular regions, and a mapping relationship is established between the supply port movement trajectory parameters and the cumulative supply of polishing slurry in each annular region. This mapping relationship is determined jointly by fluid dynamics simulation and CMP process test data fitting, forming a radial concentration distribution function C(r)=F(tra p The parameters of the supply port movement trajectory include the starting position, the ending position, the movement speed as a function of time v(t), and the relative motion curves of the two supply ports.

4. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 3, characterized in that, The controller has a built-in standardized mapping rule library, which records the correspondence between different types of surface deviations and the optimal sweep parameter adjustment amount. The mapping rule library is implemented using a lookup table, decision tree, or a pre-trained neural network model. After the controller calculates the real-time surface deviation, it matches or deduces the corresponding sweep parameter adjustment amount from the rule library and outputs it to the drive mechanism.

5. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, The drive mechanism also includes a position sensor and a limit switch for limiting the movement range of the supply port; the motor is a servo motor or a stepper motor, and the ball screw is a miniature precision ball screw; each polishing liquid supply line adopts a spiral hose, the tail end of the spiral hose is fixed to the swing arm base, and the head end is fixedly connected to the nut seat of the ball screw through a connector.

6. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, The system also includes a fixed polishing fluid supply port, located in the middle of the swing arm or near the center of the polishing head, to provide basic polishing fluid flow; two movable supply ports are located on the left and right sides of the fixed supply port, and the center line connecting the three supply ports is parallel to the length direction of the swing arm; the controller selects to enable the combination mode of the fixed supply port and the movable supply port, or to enable only the movable supply port, according to process requirements.

7. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, When the controller performs adaptive closed-loop regulation, it adopts independent PID control loops for the two supply ports, or adopts a coupled control strategy to make the movement of the two supply ports meet the preset relative motion relationship. When the sensor detects that the uniformity index of polishing fluid distribution exceeds the threshold, the controller prioritizes adjusting the moving speed curve and the endpoint dwell time. When the uniformity index continues to deviate, the moving stroke is adjusted to avoid mechanical wear caused by frequent reversal of the drive mechanism.

8. The closed-loop feedback and model adaptive system for polishing slurry supply according to claim 1, characterized in that, The swing arm base is also integrated with a high-pressure water nozzle for cleaning the movable supply port and the fixed supply port during the polishing interval. When the controller executes the cleaning program, it drives the movable supply port to perform a full-stroke reciprocating motion at a speed 1.5 to 3 times higher than the sweeping speed during polishing, and at the same time, it links the high-pressure water nozzle to perform pulse spraying to remove the polishing liquid deposits remaining on the inner wall of the pipeline and nozzle.

9. A closed-loop feedback and model adaptive control method for polishing slurry supply, applied to the system described in any one of claims 1 to 8, characterized in that, Includes the following steps: Step S1: Establish a mathematical model of the polishing fluid flow field distribution in advance, and construct a standardized mapping rule library between surface deviation and sweeping parameter adjustment amount; Step S2: Select the initial sweeping formula according to the initial surface shape of the target wafer, and control the movable feed ports on both sides to perform open-loop sweeping motion according to the initial stroke, speed and dwell time. Step S3: During the polishing process, the uniformity parameters of the polishing slurry distribution or the wafer removal rate distribution parameters are acquired in real time through sensors to generate real-time surface shape data. Step S4: Compare the real-time surface shape data with the target surface shape point by point and calculate the deviation vector; Step S5: Input the deviation vector into the mathematical model of the flow field distribution, and obtain the adjustment amount of the sweep parameters through the inverse solution of the mapping rule base; Step S6: The adjustment amount is sent to the drive mechanism to dynamically correct the movement trajectory parameters of the supply port; Step S7: Repeat steps S3 to S6 until polishing is complete or the surface deviation converges to within the allowable range.

10. The closed-loop feedback and model adaptive control method for polishing slurry supply according to claim 9, characterized in that, The specific method for establishing the mapping rule base in step S1 is as follows: using a multi-factor orthogonal experimental method, different combinations of sweeping parameters are traversed on the test wafer, and the corresponding polished surface shape is measured. Partial least squares method or support vector regression is used to establish an inverse mapping model from surface shape deviation to sweeping parameter adjustment amount, and the model is solidified in the controller in the form of coefficient matrix or rule table.

11. The closed-loop feedback and model adaptive control method for polishing slurry supply according to claim 9, characterized in that, In step S5, when the deviation vector indicates that the removal rate of the wafer edge region is low, the adjustment amount output by the mapping rule library includes: extending the stroke endpoint of the movable supply port located near the edge of the polishing pad outward by 2mm to 5mm, while extending the dwell time of the endpoint by 0.5s to 2s, and reducing the moving speed of the supply port on the other side to increase the local concentration; when the deviation vector indicates that the removal rate of the wafer center region is low, the adjustment amount output by the mapping rule library is: shifting the moving stroke of both supply ports towards the center of the polishing pad, and adopting a co-directional polymerization motion mode.

12. The closed-loop feedback and model adaptive control method for polishing slurry supply according to claim 9, characterized in that, The method also includes an anomaly diagnosis step: the controller monitors the residual between the sensor feedback data and the theoretical distribution predicted based on the flow field mathematical model in real time. When the residual exceeds the preset threshold and continues for multiple sampling cycles, it is determined to be an unexpected anomaly caused by polishing pad wear, polishing fluid deterioration, or supply port blockage. The controller issues an alarm signal and automatically calls the self-cleaning program or suggests replacing the polishing pad / polishing fluid. The self-cleaning program includes starting the high-pressure water nozzle and driving the movable supply port to perform a full-stroke reciprocating motion.

13. The closed-loop feedback and model adaptive control method for polishing slurry supply according to claim 9, characterized in that, The controller supports running two or more closed-loop control threads simultaneously, each corresponding to a different polishing stage: the coarse polishing stage uses a larger stroke range and a faster sweeping speed to quickly cover a large area, while the fine polishing stage uses a smaller stroke range and a slower sweeping speed with fine feedback adjustment to optimize local flatness; the mapping rule base parameters for different stages are stored separately and automatically switched by the controller based on polishing time or cumulative removal amount.

Images