A cleaning system and method for real-time adjustment of cleaning process parameters

Abstract

The disclosure provides a cleaning system and method for real-time adjustment of cleaning process parameters, belonging to the technical field of semiconductor manufacturing equipment. The system comprises a feeding unit, a cleaning treatment unit, a discharging unit, a transmission unit and a central control module. The feeding unit is provided with a first laser detection assembly for initial wafer surface scanning; the transmission unit is integrated with a follow-up detection assembly on the mechanical hand for real-time detection of the wafer surface during the process gap; the cleaning treatment unit is provided with a double-point concentration sensor on the chemical tank for monitoring the stratification of the chemical liquid; the discharging unit is provided with a third laser detection assembly for final surface detection. The central control module adjusts the cleaning parameters in real time based on the detection data. The disclosure realizes full-process surface state monitoring and dynamic parameter adjustment, solves the problems of fixed cleaning parameters, lack of process feedback and stratification of tank concentration in the prior art, and improves the cleaning uniformity and yield.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor manufacturing equipment technology, and in particular to a cleaning system and method for real-time adjustment of cleaning processing parameters. Background Technology

[0002] In the manufacturing process of very large-scale integrated circuits (VLSI), wet cleaning is one of the most crucial and repetitive processes, accounting for over 30% of the entire chip manufacturing process. As feature sizes continue to shrink to the nanometer level, the requirements for wafer surface cleanliness increase exponentially. Chemical mechanical polishing (CMP) is widely used to achieve global planarization of the wafer surface, but it introduces a large number of contaminants, including abrasive particles, organic residues, and metal ions. These contaminants adhere tightly to the wafer surface through van der Waals forces or electrostatic adsorption. If cleaning is incomplete, it can lead to short circuits, open circuits, or electrical performance degradation, severely reducing chip yield. Classical cleaning processes have been the cornerstone of wet cleaning since their inception, but their fixed formulations and static control modes can no longer meet the high-precision requirements of advanced processes.

[0003] Currently, mass-production wet cleaning equipment is mainly divided into two types: single-wafer type and tank type. Tank type cleaning machines are still widely used in the upstream process due to their high throughput. In a typical tank type cleaning machine, a batch of wafers is loaded into a wafer cassette and then sequentially immersed in multiple chemical solution tanks by an automated robotic arm according to a preset formula. Existing equipment is usually equipped with a certain degree of process control functions, such as monitoring the concentration of the chemical solution through concentration sensors installed on the chemical delivery pipeline and maintaining a stable concentration through an automatic replenishment system. In addition, run-to-run control strategies are used to adjust the process parameters of the next batch based on the feedback of the detection results of the previous batch of wafers on offline metrology equipment. However, these control methods are inherently lagging and static, and cannot intervene in real time during the cleaning process.

[0004] The existing operating methods have three main problems. First, there is a lack of real-time feedback and remedial mechanisms between processes. In the current quality control model, wafers must complete all cleaning steps and leave the cleaning machine before they can be inspected, resulting in delayed feedback. If the first cleaning step fails to meet the standard, the system cannot detect it and can only continue to execute subsequent steps, causing defects to be masked or solidified, ultimately leading to the scrapping of the entire batch of wafers. Second, rigid parameter settings lead to over-cleaning or under-cleaning. Existing equipment executes a fixed formula, but the condition of incoming wafers fluctuates greatly. Uniform high-intensity cleaning can cause over-cleaning of clean wafers, increasing surface roughness and wasting chemicals, while insufficient cleaning intensity may lead to under-cleaning of dirty wafers. Third, there is a risk of concentration stratification in large-volume tanks. Due to the different specific gravities of the components in the solution and temperature gradients, fluid dead zones and concentration stratification are easily formed at the bottom of the tank. Existing single-point sensors cannot detect local high-concentration deposits, resulting in extremely poor cleaning uniformity within the batch and within the wafer. Summary of the Invention

[0005] This disclosure aims to overcome several inherent defects in existing semiconductor wet cleaning technologies, mainly including lag in cleaning parameter adjustment due to the lack of a real-time feedback mechanism between processes, over-cleaning or under-cleaning phenomena caused by fixed process formulations, and batch-to-batch non-uniformity caused by stratification of the cleaning solution concentration in large-volume cleaning tanks. To solve these technical problems, this disclosure provides a real-time adjustable cleaning system and method based on full-process surface condition monitoring. This system integrates a servo detection component into a multi-axis robotic arm, implements dual-point anti-stratification monitoring on the cleaning solution tank, and incorporates advanced feedforward and feedback control algorithms, ultimately achieving intelligent, adaptive closed-loop control of the cleaning process.

[0006] To achieve the aforementioned objectives, the core technical solution adopted in this disclosure is as follows: The system includes a loading unit, a cleaning unit, and an unloading unit arranged sequentially along the wafer process flow, a transfer unit responsible for moving wafers between units, and a central control module for unified scheduling and intelligent decision-making of the entire system. The loading unit is equipped with a first laser detection component for non-contact, rapid line scanning of the wafer before it enters the cleaning process to obtain its initial surface condition data. The central control module then rates the degree of contamination of the wafer based on this initial data and generates a predicted cleaning plan for that wafer. This plan includes the baseline process parameters for subsequent cleaning processes, thus establishing the basis for the system's feedforward control.

[0007] The core innovation of this disclosure lies in the innovative design of the transmission unit. This unit comprises at least one multi-axis robotic arm, crucially integrating a servo detection component. This component is embedded in the wrist of the robotic arm, and its detection optical path points to the surface of the wafer being transported through a light-transmitting window flush with the outer surface of the robotic arm. This integrated design ensures the stability and reliability of the detection component during the high-speed movement of the robotic arm, while avoiding interference with external clean airflow and potential chemical residue. Power and signal cables are routed through a hollow wiring channel inside the robotic arm, effectively preventing particulate contamination caused by cable wear. Its function is to simultaneously perform real-time, non-contact scanning and detection of the wafer surface condition during the process intervals when the robotic arm grasps the wafer and transports it between different cleaning tanks. The acquired real-time data is immediately transmitted to the central control module. Based on this real-time data and the initial contamination rating, the central control module dynamically adjusts the process parameters of subsequent cleaning processes using feedforward control logic. For example, if the system detects that the previous cleaning process is not up to standard, it will automatically adjust the concentration, temperature, or cleaning time of the next cleaning solution tank before the wafer enters the next process, thereby achieving true real-time feedback and process remediation between processes.

[0008] To address the issue of drug concentration stratification within the tank, the cleaning unit disclosed herein features specialized optimizations in fluid dynamics and monitoring layout. Each drug tank comprises an inner tank and an outer tank surrounding it. At least two concentration monitoring points are vertically spaced along the side wall of the outer tank, each equipped with a central concentration sensor and a bottom concentration sensor, used to monitor drug concentration at different liquid levels. The central control module continuously compares the readings from these two sensors. Once the concentration difference exceeds a preset safety threshold, stratification is detected, and the circulation pump is immediately switched to a high-turbulence mixing mode to forcibly mix the drug until the concentration returns to uniformity. Furthermore, the bottom of the outer tank is designed with a sloped surface at a specific angle, which helps guide sediment towards the drain outlet using gravity, thereby further eliminating dead zones in the fluid flow from a physical structural perspective.

[0009] A third laser inspection component is installed at the unloading unit to perform final surface condition inspection on the wafers after all cleaning and drying. This final inspection data is not only used to determine whether the product is qualified, but more importantly, it serves as "true value" feedback to the central control module. The central control module utilizes this long-term "initial state-adjustment action-final result" data chain, through built-in adaptive learning algorithms such as weighted moving averages or neural networks, to continuously verify, evaluate, and optimize its process parameter adjustment model, enabling the entire system to continuously self-improve and evolve its process formulation.

[0010] In summary, this disclosure, through deep integration and innovation of the aforementioned hardware structure and intelligent upgrade of the software control logic, successfully constructs an intelligent system capable of "sensing" the real-time status of the wafer, "deciding" on optimal process parameters, and "executing" precise cleaning. The method encompasses a series of steps, from initial scanning and rating using the first laser detection component of the loading unit, to real-time inter-process detection during transport using a servo detection component on the robotic arm, to real-time adjustment of subsequent cleaning parameters based on this detection data, to ensuring the uniformity of the cleaning solution using a dual-point sensor in the tank, and finally, to closed-loop optimization using unloading detection. This system and method work synergistically to effectively solve the long-standing core pain points in semiconductor wet cleaning, such as "blind cleaning" and "uneven mixing," significantly improving process window control capabilities, product yield, and production efficiency. Attached Figure Description

[0011] Figure 1 This is a top view of the overall structure of the cleaning system provided in this embodiment, showing the layout of each unit and the direction of material flow.

[0012] Figure 2 This is an enlarged schematic diagram of a robotic arm structure integrating a follow-up detection component in an embodiment of this disclosure, which focuses on showing the embedded installation of the detection component on the robotic arm and the optical path.

[0013] Figure 3 This is a cross-sectional structural diagram of the drug solution tank in this embodiment, which focuses on showing the spatial positions of the concentration sensors in the inner tank, outer tank, and the middle and bottom of the outer tank sidewall.

[0014] Figure 4 This is a schematic flowchart of a cleaning method for real-time adjustment of cleaning processing parameters provided in an embodiment of this disclosure. Detailed Implementation

[0015] The technical solutions in this disclosure will now be clearly and completely described with reference to the accompanying drawings.

[0016] See Figure 1The cleaning system 100 adopts a modular layout, with the entire system housed within a microenvironmentally sealed cavity equipped with a fan and filter unit to maintain a high level of cleanliness. The loading unit 10, located at the system's input, typically includes a loading port for placing wafer cassettes and an indexing robot for retrieving wafers. At the loading unit 10, a first laser detection component 11 is installed. This component 11 quantifies the initial contamination level (e.g., haze value or particle density) of the wafer by emitting a laser of a specific wavelength (preferably 532 nm) onto the wafer surface and receiving the scattered light signal. The central control module 50 receives this data, uses a preset grading algorithm to initially grade the wafers (e.g., extremely clean, normal, dirty, extremely dirty), and generates a "predicted cleaning plan" for the batch or wafer, setting the reference time and concentration for each cleaning tank level. This constitutes the system's feedforward control input. Specifically, the first laser inspection assembly 11 is mounted above the wafer loading port and suspended by an inverted L-shaped bracket. It is configured to perform a non-contact, rapid line scan of the wafer surface to obtain an initial contamination rating during the process of the indexing robot removing the wafer from the wafer cassette. The first laser inspection assembly 11 includes a solid-state laser emitting a wavelength of 532 nm. This wavelength is suitable for effectively generating Rayleigh scattering to detect particles smaller than 100 nm, while avoiding the high damage and cost to optical components caused by shorter wavelengths. The laser illuminates the wafer surface at a grazing angle, for example, an incident angle of 60 degrees, and a photomultiplier tube receives the scattered light in the vertical direction. To accommodate wafers of different sizes, the first laser inspection assembly 11 is mounted on a motorized guide rail that can slide along the X-axis to achieve full-surface coverage scanning. To prevent acidic gases in the cleaning area from corroding the precision optical lens, the assembly is encased in a corrosion-resistant polytetrafluoroethylene shell, and a nitrogen curtain generator is installed in front of the lens to continuously spray clean nitrogen to block volatile chemicals. In addition, the lens surface is coated with a hydrophobic coating and is equipped with a ring-shaped nitrogen purging nozzle to form an air curtain to prevent dust from settling.

[0017] The transfer unit 20 is responsible for moving wafers between units; its core component is a multi-jointed robotic arm. (See also...) Figure 2The robotic arm includes a robotic arm body 21 and an end effector 23. The end effector 23 is made of corrosion-resistant polyetheretherketone (PEEK) material and is shaped like a U- or Y-shaped fork, securing the wafer via a Bernoulli suction cup structure. The Bernoulli suction cup includes a gas flow channel and an adsorption surface. The gas flow channel is configured to spray clean gas onto the wafer surface to generate negative pressure levitation force. The adsorption surface has limiting pins to prevent lateral slippage of the wafer. The limiting pins are made of an electrostatic conductive material to eliminate triboelectric static electricity. A key innovation of this disclosure is the integration of a follow-up detection component. A second laser detection component 22 is embedded and integrated at the wrist position of the robotic arm body 21. The surface of the robotic arm has a light-transmitting window made of high-purity quartz glass resistant to hydrofluoric acid corrosion, and the window surface is strictly flush with the outer skin of the robotic arm to avoid the formation of liquid accumulation pits or turbulence protrusions. A nitrogen purging nozzle is provided on the outer side of the light-transmitting window, which is configured to spray dry nitrogen tangentially along the surface of the light-transmitting window. The second laser detection component 22 includes a laser emitter and a photodetector. The laser emitter is configured to emit a 532nm diode-state laser beam, and the photodetector is configured as a photomultiplier tube or charge-coupled device. The incident angle between the laser emitter and the wafer surface is set between 45 and 70 degrees to optimize the detection sensitivity of the scattering cross-section for submicron particles. To accommodate the high-speed movement of the robotic arm, the second laser detection component 22 is encapsulated in a streamlined housing embedded within a structural recess of the robotic arm. For cable management, the power supply and signal lines of the sensor are connected to the base through a hollow cable routing channel inside the robotic arm. The channel has a sheath structure to prevent cable abrasion and avoid particle generation by the external cable chain during movement. During the process of the robotic arm removing the wafer from the previous cleaning tank and transferring it to the next cleaning tank, the robotic arm performs a rapid scanning motion, using the second laser detection component 22 to perform a line scan of the wafer surface to obtain a residual distribution map. This process does not consume additional process time and achieves real-time feedback between processes. Specifically, during the "air transport" stage where the robotic arm removes the wafer from the previous cleaning tank (e.g., SC-1 cleaning) and moves it to the next cleaning tank (e.g., SC-2 cleaning), the servo detection component moves along with the robotic arm to dynamically scan the surface of the adsorbed wafer. This process does not consume additional process time. The detection data is transmitted back to the central control module 50 in real time. If the residue after SC-1 cleaning is found to be higher than expected, the system will immediately trigger the process remediation mechanism, automatically modifying the parameters of the upcoming SC-2 process (e.g., temporarily extending the soaking time from 300 seconds to 450 seconds, or instructing the automatic replenishment system to increase the HCl concentration), thereby achieving a "better late than never" real-time closed-loop control.

[0018] exist Figure 2In this embodiment, the transfer unit 20 is shown as a robot arm. In other embodiments, the transfer unit 20 may be embodied as a linear transfer module, in which the wafer carrier moves linearly along a fixed guide rail, and the detection components are mounted on a slide that moves with the guide rail, maintaining relative stillness or controlled scanning motion with respect to the wafer. Regardless of the mechanical form employed, as long as the transfer unit has the capability to hold the wafer and move it from one process station to the next in the 'in-air transfer' stage, and to perform follow-up scanning of the wafer using the mechanically coupled detection components, it falls within the protection scope of this disclosure.

[0019] The cleaning unit includes several chemical solution tanks 30 for chemical treatment. See also... Figure 3 The drug solution tank 30 adopts a double-sleeve design, including an inner tank 31 and an outer tank 32. The top edge of the inner tank 31 has a serrated overflow weir 35 to disrupt surface tension and ensure continuous overflow. The outer tank 32 surrounds the inner tank and has a sloping bottom with an inclination angle of 3 to 5 degrees, with the lowest point connected to the circulation pump inlet to eliminate dead zones. To address concentration stratification, two concentration monitoring points are vertically spaced on the sidewall of the outer tank 32, equipped with a central concentration sensor 33 and a bottom concentration sensor 34. The central concentration sensor 33 is located at 50% to 60% of the height of the outer tank sidewall and monitors the average concentration in the middle of the drug solution tank 30; the bottom concentration sensor 34 is located 50 mm to 100 mm from the bottom of the outer tank sidewall and monitors the concentration in the bottom sedimentation zone. The preferred sensor is an online near-infrared spectroscopy analyzer, configured to measure the absorption spectrum of the drug solution in the 900nm to 2500nm wavelength range via a fiber optic probe, and to analyze the independent concentrations of various chemical components in the drug solution using chemometric algorithms. The end face of the sensor probe is flush with the inner wall of the outer tank via a flange structure to avoid the generation of eddy current dead zones. The central control module 50 compares the readings of the middle concentration sensor 33 and the bottom concentration sensor 34 in real time. When the difference between the two exceeds a preset threshold, it controls the circulation pump of the drug solution tank to increase its speed to enter a turbulent stirring mode. The turbulence generated by the high flow rate stirs up the bottom sediment until the readings tend to be consistent.

[0020] The unloading unit 40 is located at the system's output end and is responsible for returning the dried wafers to the wafer cassette. A third laser detection component 41 is installed at the unloading unit 40 to perform final surface condition detection on the cleaned and dried wafers. The structure of the third laser detection component 41 is similar to that of the first laser detection component 11, including a laser emitter and a photodetector, and is installed above the unloading channel. The detection data is not only used to determine whether the product is qualified, but also fed back as a truth value to the central control module 50 to verify the effectiveness of the preceding adjustment strategy. By accumulating data pairs of "initial state - adjustment strategy - final result" over a long period, the system continuously optimizes the process parameter adjustment model, or in other words, corrects the process parameter adjustment model, using machine learning algorithms (such as neural networks or weighted moving averages), thus achieving the self-evolution of the process formulation.

[0021] The central control module 50 is the intelligent core of the system, typically implemented using a PLC or industrial PC. The module communicates with each detection component and sensor via high-speed I / O cards, executing the following detailed logic: For data acquisition, it reads the scattered light intensity values ​​from the three laser detection components and the concentration values ​​from the two concentration sensors in real time. For the grading algorithm, it classifies the wafer contamination level into multiple grades based on the scattered light intensity values, such as extremely clean, normal, dirty, and extremely dirty. For parameter adjustment, based on feedforward and feedback control logic, it performs multi-dimensional adjustments to the cleaning time, solution concentration, solution temperature, and circulation flow rate. For example, when the initial rating is "very poor," the cleaning time is increased by 30 seconds and the solution concentration by 0.1%; when the rating is "good," the cleaning time is decreased by 30 seconds. More specifically, regarding feedforward control, based on data from the first laser detection component 11 at the feeding end, if the incoming wafer has extremely low scattered light intensity (clean), the central control module 50 automatically calls the "energy-saving mode" formula. For example, it reduces the H2O2 consumption of SC-1 by 20% and shortens the cleaning time by 30 seconds. This not only saves costs but also avoids excessive roughening of the clean surface. Regarding feedback control, based on data from the robotic arm follow-up detection component, if particle residue is found to be still at the "warning" level after SC-1, the central control module 50 automatically modifies the formula of the subsequent SC-2. For example, it activates the megasonic function of the SC-2 tank (megasonic is not commonly used in SC-2, but it can be activated in emergency situations to assist in removing stubborn particles), or adds an extra rinsing cycle after SC-2. In terms of batch-to-batch learning, the central control module 50 records the input (initial contamination level), process variables (actual cleaning parameters), and output (final detection results at the feeding end) for each batch. It monitors equipment state drift using a weighted moving average algorithm. If it detects that the compensation required to achieve the same cleaning effect increases with the increase in the usage time of the cleaning solution, the system will automatically calculate the end of the cleaning solution's lifespan and provide an early warning for solution replacement. In other words, the central control module 50 receives detection data from the follow-up detection component and, based on this data and the initial contamination rating, uses feedforward control logic to adjust the process parameters of subsequent cleaning processes in real time. It also corrects the process parameter adjustment model based on feedback data from the third laser detection component 41.

[0022] More specifically, the central control module 50 executes a two-layer control strategy. The first layer is "wafer-level feedforward control" based on data from the second laser detection component 22, which is a real-time response mechanism, such as extending the cleaning time of the next tank immediately upon detecting residue. The second layer is "batch-level feedback control" based on data from the third laser detection component 41. This layer is used to correct the process parameter adjustment model itself. To avoid frequent model oscillations caused by random fluctuations in individual wafers, this disclosure employs a periodic correction strategy or a threshold-triggered correction strategy, i.e., correction is performed when the statistical deviation between the feedback data and the expected value exceeds a preset threshold. Specifically, the central control module 50 internally runs an exponentially weighted moving average algorithm. The system calculates the moving average of the final surface quality deviation of the most recent batch (e.g., 25 wafers). Only when this moving average exceeds a preset statistical control limit does the system determine that a systematic drift has occurred (such as chemical aging or filter clogging) and trigger an update of the model parameters.

[0023] In addition, the central control module 50 interacts with the factory host system through a communication protocol to achieve batch-to-batch control optimization.

[0024] Accordingly, see Figure 4 This disclosure also provides a cleaning method for real-time adjustment of cleaning processing parameters. The method is executed using the system described in the foregoing embodiments and includes the following steps S401, S402, S403, S404 and S405.

[0025] S401: At the feeding unit, the first laser detection component is used to perform an initial surface scan on the wafer entering the system to obtain an initial contamination rating.

[0026] S402: The central control module generates a projected cleaning plan for the wafer based on the initial contamination rating. The projected cleaning plan includes baseline process parameters for subsequent cleaning processes.

[0027] S403: The wafer is transported between the loading unit, the various chemical tanks of the cleaning unit, and the unloading unit via the transmission unit.

[0028] S404: During the process gap when the transmission unit holds and moves the wafer, the surface condition of the wafer is scanned and detected in real time using the detection component provided on the transmission unit.

[0029] S405: The central control module receives the detection data from the detection component and, based on the detection data and the initial contamination rating, adjusts the process parameters of subsequent cleaning processes in real time using feedforward control logic.

[0030] This disclosure allows for multi-dimensional adjustment of process parameters, including but not limited to adjusting chemical concentration by fine-tuning the opening time of the replenishment valve, adjusting cleaning time by adjusting the dwell time of the robot in the tank, adjusting temperature by adjusting the set point of the online heater, and adjusting sound energy intensity according to particle removal requirements.

[0031] In its implementation, the system also incorporates environmental adaptability design. For example, inside the wet cleaning machine, where acidic volatile gases and water vapor are abundant, all optical components are protected with nitrogen purging and hydrophobic coatings. The robotic arm's motion trajectory has been optimized to ensure that scanning and transfer actions are synchronized without increasing cycle time. The circulating pump in the chemical tank utilizes magnetic levitation technology to achieve precise speed control and rapid response. Experimental data shows that applying this system in the post-cleaning process of 12-inch wafer CMP significantly improves the metal ion removal rate, reducing copper residue from 5e10 atoms / cm³ in the control group. 2 Reduced to 1e9 atoms / cm 2 The following results demonstrate that, by avoiding tank delamination and enabling timely remediation, edge yield was improved by approximately 1.5%; and by cleaning clean incoming materials on demand, monthly chemical consumption was reduced by 18%. These effects validate the technological advancements and practical value of this disclosure.

[0032] It should be noted that the technical solutions described in this disclosure can be combined arbitrarily as long as they do not conflict.

[0033] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A cleaning system for real-time adjustment of cleaning processing parameters, characterized in that, include: A loading unit is used to provide wafers to be cleaned. The loading unit is equipped with a first laser detection component for performing an initial surface scan to obtain an initial contamination rating before the wafers enter the cleaning process. The cleaning unit includes at least one solution tank for containing cleaning fluid and immersing the wafer. The unloading unit is used to receive the cleaned and dried wafers; A transfer unit is configured to transport the wafer between the loading unit, the cleaning unit, and the unloading unit; as well as The central control module is communicatively connected to the feeding unit, the cleaning unit, and the transmission unit. The transmission unit is equipped with a detection component, which is configured to perform real-time non-contact scanning detection of the surface condition of the wafer during the process gap when the transmission unit holds and moves the wafer. The central control module is configured to receive the detection data from the detection component and, based on the detection data and the initial contamination rating, adjust the process parameters of subsequent cleaning processes in real time using feedforward control logic.

2. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 1, characterized in that, The transmission unit includes a robotic arm, and the detection component is a follow-up detection component integrated on the robotic arm body of the robotic arm.

3. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 2, characterized in that, The follow-up detection component is embedded in the wrist of the robotic arm body of the robotic hand. The robotic arm body has a light-transmitting window on the sensing path of the follow-up detection component. The surface of the light-transmitting window is flush with the outer surface of the robotic arm body. The power supply cable and signal cable of the follow-up detection component pass through the hollow wiring channel inside the robotic arm body and are connected to the base of the robotic hand.

4. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 3, characterized in that, A nitrogen purging nozzle is provided on the outer side of the light-transmitting window, and the nitrogen purging nozzle is configured to spray dry nitrogen tangentially along the surface of the light-transmitting window.

5. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 1, characterized in that, The unloading unit is equipped with a third laser detection component, which is used to perform final surface scanning after the wafer has been cleaned; The central control module is also configured to receive feedback data from the third laser detection component and correct the preset process parameter adjustment model based on the feedback data.

6. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 5, characterized in that, The central control module is configured to perform periodic corrections on the process parameter adjustment model, or to perform corrections when the statistical deviation between the feedback data and the expected value exceeds a preset threshold.

7. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 1, characterized in that, The liquid tank includes an inner tank and an outer tank surrounding the inner tank. The top edge of the inner tank is provided with a serrated overflow weir, and the bottom of the outer tank is provided with a sloping bottom surface. The inclination angle of the sloping bottom surface is 3 to 5 degrees, and the lowest point is connected to the inlet of the circulation pump.

8. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 7, characterized in that, At least two concentration monitoring points are arranged vertically on the side wall of the outer tank, and a middle concentration sensor and a bottom concentration sensor are installed thereon. The middle concentration sensor is located at 50% to 60% of the height of the side wall of the outer tank, and the bottom concentration sensor is located at 50 mm to 100 mm from the bottom of the side wall of the outer tank. The central control module is configured to compare the measured values ​​of the middle concentration sensor and the bottom concentration sensor in real time. When the difference between the two exceeds a preset threshold, the module controls the circulation pump of the drug tank to increase its speed to enter a turbulent stirring mode.

9. The cleaning system for real-time adjustment of cleaning processing parameters according to claim 8, characterized in that, The middle concentration sensor and the bottom concentration sensor are configured to measure the absorption spectrum of the drug solution in the 900nm to 2500nm wavelength range using fiber optic probes. The end face of the sensor probe is flush with the inner wall of the outer tank via a flange structure.

10. A cleaning method for real-time adjustment of cleaning processing parameters, the method being executed using the system described in any one of claims 1 to 9, characterized in that, The method includes: At the feeding unit, the first laser detection component is used to perform an initial surface scan on the wafer entering the system to obtain an initial contamination rating; Based on the initial contamination rating, the central control module generates a projected cleaning plan for the wafer, which includes baseline process parameters for subsequent cleaning processes. The wafer is transported between the loading unit, the various chemical tanks of the cleaning unit, and the unloading unit via the transmission unit. During the process gaps when the transmission unit holds and moves the wafer, the surface condition of the wafer is scanned and detected in real time using the detection component provided on the transmission unit. The central control module receives the detection data from the detection component and, based on the detection data and the initial contamination rating, adjusts the process parameters of subsequent cleaning processes in real time using feedforward control logic.

Images