Cleaning method and control system for semiconductor wafer cleaning system

Abstract

The present invention discloses a cleaning method and a control system for a semiconductor wafer cleaning system in the field of information technology. The control method for the semiconductor wafer cleaning system comprises the following steps: acquiring a wafer surface image, processing the wafer surface image using an image segmentation algorithm, identifying contaminated areas and non-contaminated areas in the wafer surface image, and judging the degree of contamination and the type of contamination on the wafer surface according to the area, morphology and grayscale characteristics of the contaminated areas; recording cleaning process data for each batch of wafers, the cleaning process data including cleaning formula, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate and surface damage, applying big data analysis technology to explore the optimal cleaning process for different types of wafers, and dynamically updating and optimizing the cleaning formula database to improve cleaning efficiency and yield.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a cleaning method and a control system for a semiconductor wafer cleaning system. Background Art

[0002] Semiconductor wafer cleaning is a critical step in the semiconductor manufacturing process, and its cleaning quality directly affects the yield of subsequent processes and device performance. However, as integrated circuit feature sizes continue to shrink and wafer surface structures become increasingly complex, higher requirements are placed on the cleaning process.

[0003] Traditional cleaning methods are difficult to completely remove various pollutants on the wafer surface, such as particles, organic matter, metal ions, etc., and the cleaning effect is easily interfered with by human factors, and the stability and consistency are poor. At the same time, the degree and type of surface contamination of wafers of different batches and models vary greatly. It is difficult to achieve the optimal cleaning effect using fixed cleaning formulas and parameters. Excessive cleaning will damage the surface structure of the wafer, while insufficient cleaning will not meet the requirements of subsequent processes, resulting in device performance degradation or even failure. Therefore, there is an urgent need for an intelligent wafer cleaning control system that can monitor the wafer surface status and cleaning environment parameters in real time, adaptively adjust the cleaning process, and ensure thorough cleaning while maximizing wafer protection, improving cleaning efficiency and yield, and reducing energy consumption and costs. This is of great significance for the research and development and application of advanced process technologies. Summary of the Invention

[0004] The present invention provides a cleaning method for a semiconductor wafer cleaning system, the method comprising the following steps:

[0005] Step S101, obtaining a wafer surface image, processing the wafer surface image using an image segmentation algorithm, identifying contaminated areas and non-contaminated areas in the wafer surface image, and determining the degree of contamination and contamination type of the wafer surface based on the area, morphology, and grayscale characteristics of the contaminated areas;

[0006] Step S102, selecting an optimal cleaning recipe from a preset cleaning recipe database based on the contamination level and the contamination type, and dynamically adjusting cleaning time, cleaning agent concentration, and cleaning temperature parameters in the optimal cleaning recipe based on the structural complexity of the wafer surface, the structural complexity including line width, aspect ratio, and pattern density, to obtain a dynamically adjusted cleaning recipe;

[0007] Step S103: During the wafer cleaning process, the ambient temperature, humidity, and pressure parameters within the cleaning equipment are collected in real time by sensors, and the ambient temperature, humidity, and pressure parameters are compared with preset optimal ranges, which are determined based on historical cleaning data and wafer material properties. If the deviation of the comparison result exceeds a preset threshold, an alarm is triggered, and the process parameters of the cleaning equipment are automatically adjusted to maintain the stability of the cleaning environment.

[0008] Step S104: After cleaning is completed, the wafer surface is inspected again using machine vision, the inspection image after cleaning is compared with the image before cleaning, the contamination removal rate is calculated, and the contamination removal rate is compared with a preset cleaning qualification standard. If the cleaning qualification standard is not met, the process returns to step S102 and the cleaning process is repeated until the cleaning qualification standard is met.

[0009] Step S105: After the cleaning reaches the qualified standard, the surface of the wafer is tested using an atomic force microscope to obtain the roughness and defect density parameters of the wafer surface, and the roughness and defect density parameters are compared with the preset wafer surface quality standards to determine whether the cleaning process has caused damage to the surface structure of the wafer. If damage to the surface structure is detected, the type and degree of damage are recorded, and the parameters of the cleaning recipe are adjusted to reduce similar damage.

[0010] Step S106: Recording cleaning process data for each batch of wafers, including cleaning recipes, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate, and surface damage. Applying big data analysis technology to discover optimal cleaning processes for different wafer types, dynamically updating and optimizing the cleaning recipe database to improve cleaning efficiency and yield.

[0011] In step S107, based on the wafer production plan, feedback information from the equipment management system, historical cleaning effect data and energy consumption data, the energy consumption and material consumption of the cleaning equipment are predicted and optimized. An intelligent scheduling algorithm based on multi-objective optimization is applied to comprehensively consider factors such as cleaning effect, energy consumption, material consumption and equipment utilization, reasonably arrange cleaning tasks, minimize cleaning costs, take into account production efficiency and delivery time, and realize intelligent management and control of the cleaning process.

[0012] The present invention provides a semiconductor wafer cleaning system and control system, which mainly include:

[0013] Wafer surface contamination detection module, used to identify contaminated areas and determine the degree and type of contamination;

[0014] Cleaning recipe optimization module, used to select the optimal cleaning recipe and dynamically adjust parameters according to the wafer structure;

[0015] Cleaning process monitoring module, used to monitor cleaning environment parameters in real time and automatically adjust to maintain stability;

[0016] Cleaning effect evaluation module, used to detect the wafer surface after cleaning and determine whether it meets the cleaning standards;

[0017] Data analysis and optimization module, used to record cleaning data, analyze optimal processes and update the cleaning formula database;

[0018] Intelligent scheduling and control module is used to optimize the energy and material consumption of cleaning equipment and reasonably arrange cleaning tasks.

[0019] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0020] The present invention discloses a cleaning method for a semiconductor wafer cleaning system. The method uses machine vision and image segmentation algorithms to identify the contamination of the wafer surface, automatically selects the optimal cleaning formula based on the contamination characteristics, and dynamically adjusts the cleaning parameters in combination with the complexity of the wafer structure. During the cleaning process, environmental parameters are monitored in real time and automatically adjusted to maintain stability. After cleaning, the wafer is retested, and if the standards are not met, the cleaning is repeated. An atomic force microscope is used to detect the surface morphology and evaluate the impact of cleaning on the wafer structure. Cleaning data is recorded and big data analysis is applied to continuously optimize the cleaning process. The present invention also combines production plans and historical data to optimize energy and material consumption and intelligently schedule cleaning tasks. Through the above method, intelligent and refined control of the wafer cleaning process is achieved, cleaning efficiency and yield are improved, costs are reduced, and production efficiency is taken into account while ensuring cleaning quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a flow chart of the cleaning method of the semiconductor wafer cleaning system of the present invention.

[0022] Figure 2 It is a structural schematic diagram of the control system of the semiconductor wafer cleaning system of the present invention. DETAILED DESCRIPTION

[0023] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0024] like Figure 1-2 The semiconductor wafer cleaning system and control system of this embodiment may specifically include:

[0025] Step S101, obtaining a wafer surface image, processing the wafer surface image using an image segmentation algorithm, identifying contaminated areas and non-contaminated areas in the wafer surface image; and judging the degree of contamination and contamination type of the wafer surface based on the area, morphology and grayscale characteristics of the contaminated areas.

[0026] Step S101 also includes the following steps:

[0027] a. Acquire a high-definition image of the wafer surface and use an image preprocessing algorithm to denoise and enhance the image to improve image quality. b. Based on the wafer surface features, use an image segmentation algorithm to divide the image into different regions to distinguish between contaminated and uncontaminated areas. c. Extract the area, morphology, and grayscale characteristic parameters of the contaminated regions to construct a contamination feature vector. d. Input the contamination feature vector into a pre-trained contamination classification model to determine the contamination level on the wafer surface and obtain the contamination grade. e. Input the contamination feature vector into a pre-trained contamination type classification model to determine the contamination type on the wafer surface and obtain the specific contamination type. f. Based on the contamination grade and type, retrieve the corresponding cleaning solution formula and cleaning parameters from the cleaning process database. g. Input the cleaning solution formula and cleaning parameters into the cleaning equipment control module to automatically configure the cleaning equipment process parameters. h. Start the cleaning equipment and clean the wafer according to the configured process parameters. After cleaning is complete, acquire another image of the wafer surface and repeat steps b and c to determine the post-cleaning contamination level and type. If the cleaning requirements are not met, return to step f and adjust the cleaning process until the cleaning process is qualified.

[0028] For example, to obtain high-definition images of the wafer surface, a high-resolution industrial camera, such as one with 10 megapixels or more, combined with a specific lighting source, such as coaxial illumination or darkfield illumination, can produce clear images of the wafer surface. Choosing the appropriate lighting method can highlight surface defects; for example, darkfield illumination can highlight small scratches and particles. Suppose a high-definition image of the wafer surface is obtained using an industrial camera with a resolution of 2048x2048 pixels and a coaxial illumination source. Image preprocessing algorithms are used to remove image noise and enhance image features. Common denoising algorithms include median filtering and Gaussian filtering. For example, median filtering can effectively remove salt-and-pepper noise while preserving image edge information. Assuming that an image contains some random bright and dark noise, a 3x3 median filter can effectively remove this noise by replacing each pixel value with the median value of the nine surrounding pixels. Image enhancement algorithms can use histogram equalization to enhance image contrast and make contaminated areas more visible. Assuming that the grayscale range of the original image is narrow, the grayscale range can be expanded to 0-255 through histogram equalization, making the image clearer. The goal of the image segmentation algorithm is to divide the image into different regions. Commonly used image segmentation algorithms include threshold segmentation, edge detection, region growing, etc. Assuming that the contaminated area on the surface of the wafer presents a darker grayscale value, the threshold segmentation algorithm can be used to divide the pixels with grayscale values ​​below a certain threshold into the contaminated area, and the pixels above the threshold into the non-contaminated area. Assuming that the threshold is determined to be 100 by analyzing the grayscale histogram of the image, then the pixels with grayscale values ​​below 100 are considered to be contaminated areas. The characteristic parameters of the area, morphology and grayscale of the contaminated area are extracted to construct a contamination feature vector. For example, the proportion of the area of ​​the contaminated area to the entire wafer area, the morphological features such as the perimeter and circularity of the contaminated area, and the grayscale features such as the average grayscale value and grayscale variance of the contaminated area can be calculated. Assuming that the area of ​​a contaminated area is 100 pixels, the perimeter is 40 pixels, and the average grayscale value is 50, a feature vector can be constructed.

[0029] [01,40,50], where 01 represents the proportion of the contaminated area. The contamination degree classification model can be trained using machine learning algorithms such as support vector machines (SVM) or neural networks. Assuming that a large number of wafer surface images are collected and the corresponding contamination degree levels (such as mild, moderate, and severe) are labeled, these data can be used to train an SVM classification model. The extracted contamination feature vector is input into the trained SVM model to obtain the contamination degree level. For example, the feature vector [01,40,50] is input and the model outputs "mild contamination". The contamination type classification model can also be trained using a similar method. Assuming that wafer image data of different types of contamination, such as particle contamination, organic contamination, and metal contamination, are collected and labeled, a neural network classification model can be used to train this data. The extracted contamination feature vector is input into the trained neural network model to obtain the contamination type. For example, the feature vector [01,40,50] is input and the model outputs "particle contamination". The cleaning process database stores the cleaning solution formulas and cleaning parameters corresponding to different contamination types and contamination degrees. For example, for mild particulate contamination, a diluted ammonia solution can be used for cleaning, with a cleaning time of 5 minutes and a cleaning temperature of 60°C. For severe organic contamination, a more concentrated organic solvent can be used for cleaning, with a cleaning time of 10 minutes and a cleaning temperature of 80°C. Assuming the contamination level is "mild" and the contamination type is "particulate contamination," the corresponding cleaning solution formula retrieved from the database is "10% ammonia solution," with cleaning parameters of "cleaning time of 5 minutes and cleaning temperature of 60°C." The cleaning equipment control module automatically configures the cleaning equipment based on the obtained cleaning solution formula and cleaning parameters. For example, the control module controls the valve to open, injecting 10% ammonia solution into the cleaning tank, controls the heater to raise the cleaning tank temperature to 60°C, and sets the cleaning time to 5 minutes. After cleaning is complete, the wafer surface image is captured again, and the previous steps are repeated to determine the post-cleaning contamination level and type. Assuming the post-cleaning wafer surface image analysis results indicate a contamination level of "no contamination," the cleaning process is considered satisfactory. If contamination persists after cleaning, adjust the cleaning process parameters, such as increasing the cleaning solution concentration, extending the cleaning time, or raising the cleaning temperature, and repeat the cleaning process until the cleaning requirements are met. Assuming the contamination level after cleaning is "mild," the cleaning time can be extended to 10 minutes and the cleaning process repeated.

[0030] Step S102, based on the judgment results of the contamination degree and the contamination type, select the optimal cleaning formula from the preset cleaning formula database; based on the structural complexity of the wafer surface, the structural complexity includes line width, aspect ratio and graphic density, dynamically adjust the cleaning time, cleaning agent concentration and cleaning temperature parameters in the optimal cleaning formula to obtain a dynamically adjusted cleaning formula.

[0031] Acquire contamination level data and contamination type data of the wafer surface; select an optimal cleaning formula from a preset cleaning formula database based on the contamination level data and contamination type data; acquire the structural complexity of the wafer surface, wherein the structural complexity includes line width, aspect ratio and pattern density; dynamically adjust the cleaning time, cleaning agent concentration and cleaning temperature parameters in the optimal cleaning formula based on the structural complexity to obtain a dynamically adjusted cleaning formula; acquire wafer surface data after cleaning; determine whether the wafer surface data after cleaning reaches a preset threshold; if not, trigger a cleaning formula optimization process.

[0032] For example, the sensor module can monitor the contamination level and type of contamination on the wafer surface in real time using various methods. For example, an optical sensor, such as a spectroscopic reflectometer, can be used to determine the type and concentration of contaminants by analyzing the wavelength and intensity of light reflected from the wafer surface. If the wafer surface is contaminated with organic matter, the reflection spectrum will exhibit an absorption peak within a specific wavelength range. By analyzing the intensity of the absorption peak, the concentration of the organic contamination can be determined. Alternatively, an electrochemical sensor can be used to determine the type and concentration of contaminants by measuring changes in the electrochemical signal when the wafer surface comes into contact with a specific solution. For example, if the wafer surface is contaminated with metal ions, contact with a specific solution will produce a specific electrochemical reaction, thereby changing the electrochemical signal. By analyzing the changes in the electrochemical signal, the type and concentration of the metal ions can be determined. The acquired contamination level data can be quantitative, such as the contaminant concentration value, or qualitative, such as light, moderate, or heavy contamination. The contamination type data can be specific contaminant names, such as particulate matter, organic matter, or metal ions. A pre-set cleaning recipe database stores cleaning recipes for different contamination types and levels. For example, for particulate contamination, ultrasonic cleaning with a specific cleaning agent can be used; for organic contamination, an organic solvent can be used; and for metal ion contamination, a specific chemical reagent can be used. Each cleaning agent recipe includes parameters such as cleaning agent type, concentration, cleaning time, and cleaning temperature. If light particulate contamination is detected on the wafer surface, an ultrasonic cleaning recipe for this condition from the database can be selected. Machine vision technology uses a high-resolution camera to capture images of the wafer surface and then uses image processing algorithms to analyze the structural complexity of the wafer surface. For example, edge detection algorithms can identify lines and patterns on the wafer surface and measure parameters such as line width and aspect ratio. For example, if a wafer surface contains dense lines and patterns with a line width of 10 nanometers and an aspect ratio of 5, the wafer surface is considered to have a high level of structural complexity. Based on the structural complexity of the wafer surface, the parameters in the cleaning recipe need to be dynamically adjusted. For example, if the wafer surface is highly complex, the cleaning time may need to be appropriately extended to ensure that the cleaning solution can fully penetrate into small crevices and completely remove contaminants. At the same time, the cleaning agent concentration may need to be reduced to avoid damage to delicate structures. If the wafer surface structure is highly complex, the cleaning time can be extended from 5 minutes to 10 minutes, and the cleaning agent concentration can be reduced from 10% to 5%. The cleaning temperature also needs to be adjusted based on the specific situation. For example, for certain temperature-sensitive materials, the cleaning temperature needs to be kept within a lower range. The actuator module controls the cleaning equipment to perform cleaning operations based on the dynamically adjusted cleaning recipe. For example, the control module can control the valve opening and closing to adjust the flow and concentration of the cleaning solution; control the heater to adjust the cleaning solution temperature; and control the robotic arm to place the wafer into the cleaning tank and control the cleaning time.During the cleaning process, sensors monitor real-time parameters such as the cleaning fluid's temperature, pressure, and flow rate. For example, a temperature sensor measures the cleaning fluid's temperature in real time and transmits this data to the control system. If the cleaning fluid temperature falls below a preset value, the control system automatically controls the heater to raise the cleaning fluid temperature. Similarly, pressure and flow sensors monitor the cleaning fluid's pressure and flow rate in real time and automatically adjust them as needed. After cleaning is complete, the sensor module again detects the degree and type of contamination on the wafer surface and compares it with the pre-clean contamination data to determine whether the cleaning effect meets the required standards. For example, if the pre-clean contamination level is heavy and the post-clean contamination level is light, the cleaning effect is considered to meet the required standards. If the cleaning effect does not meet the required standards, the failure data is fed back to the intelligent control system. The intelligent control system can use machine learning algorithms, such as reinforcement learning, to automatically adjust cleaning recipe parameters such as cleaning time, detergent concentration, and cleaning temperature based on the failure data and historical cleaning data. If some organic contamination still remains after cleaning, the intelligent control system can appropriately increase the organic solvent concentration or extend the cleaning time based on historical data. The optimized cleaning recipe is stored in the cleaning recipe database for reference in subsequent cleaning operations. By continuously accumulating and improving the cleaning formula library, we can improve cleaning efficiency and quality, and reduce resource consumption during the cleaning process. For example, by optimizing the cleaning formula, we can shorten the cleaning time and reduce the concentration of the cleaning agent, thereby saving energy and reducing waste liquid emissions.

[0033] Step S103: During the wafer cleaning process, the ambient temperature, humidity and pressure parameters in the cleaning equipment are collected in real time through sensors, and the ambient temperature, humidity and pressure parameters are compared with the preset optimal range, which is determined based on historical cleaning data and wafer material characteristics. If the deviation of the comparison result exceeds the preset threshold, an alarm is triggered, and the process parameters of the cleaning equipment are automatically adjusted to maintain the stability of the cleaning environment.

[0034] Acquire the ambient temperature, humidity and pressure parameters in the wafer cleaning equipment collected in real time by the sensor module; obtain the optimal environmental parameter range that matches the current wafer material characteristics from the historical cleaning database; compare the ambient temperature, humidity and pressure parameters collected in real time with the optimal environmental parameter range to obtain a deviation value; if the deviation value exceeds a preset threshold, trigger an alarm, and determine the cleaning equipment process parameters that need to be adjusted and the adjustment range; adjust the corresponding process parameters of the cleaning equipment according to the process parameter adjustment instruction to obtain an adjustment result; if the deviation value still exceeds the preset threshold after multiple consecutive adjustments, it is determined that there is a cleaning equipment failure and the equipment self-test program is triggered.

[0035] For example, during the wafer cleaning process, controlling ambient temperature, humidity, and pressure is crucial. For example, if the ambient temperature is too high, it can accelerate the evaporation of the cleaning fluid, thus affecting cleaning effectiveness. Suppose historical data shows that for a specific wafer material, the optimal cleaning ambient temperature should be controlled between 23°C and 25°C. The ambient temperature monitored in real time by the sensor module is 27°C, exceeding the set optimal range. The central processing unit (CPU) calculates a temperature deviation of 2°C and determines that this deviation exceeds the allowable threshold of 1°C. The CPU triggers an alarm and, based on the direction and magnitude of the deviation, determines the need to lower the ambient temperature. This can involve increasing the cooling power of the air conditioner or adjusting the workshop's ventilation system to lower the ambient temperature. Upon receiving the adjustment command, the actuator module adjusts the corresponding device parameters, such as setting the air conditioner temperature to 23°C. After the adjustment is complete, the actuator module feeds the adjustment results back to the CPU, which then continues to monitor the ambient temperature to ensure it remains within the optimal range. During continuous monitoring, if the ambient parameters fail to stabilize within the preset threshold after repeated adjustments, the CPU considers a possible device failure. At this point, a self-test program is triggered to check the temperature control module, sensors, and other components of the cleaning equipment. If the self-test program detects that the air conditioning system is running low on refrigerant, the central processing unit retrieves instructions for refrigerant replenishment from the knowledge base and executes a self-repair program to replenish refrigerant to ensure the proper functioning of the air conditioning system. Furthermore, all environmental parameter data and equipment adjustment data are recorded and added to a historical cleaning database. This data will be used to analyze and optimize subsequent cleaning processes. For example, if data analysis reveals frequent cooling needs within a certain time period, improvements to the overall heat dissipation design of the workshop may be recommended to reduce reliance on air conditioning and improve energy efficiency and cleaning efficiency. This continuous optimization can significantly improve the quality and efficiency of wafer cleaning, ensuring a more stable and reliable wafer cleaning process.

[0036] In step S104, after cleaning is completed, the wafer surface is inspected again using machine vision, and the inspection image after cleaning is compared with the image before cleaning to calculate the contamination removal rate; the contamination removal rate is compared with the preset cleaning qualification standard. If the cleaning qualification standard is not met, return to step S102 and re-execute cleaning until the cleaning qualification standard is met.

[0037] Acquire a wafer surface image, the image containing information on the degree of contamination on the wafer surface; analyze and process the image to obtain a contamination parameter characterizing the degree of contamination on the wafer surface; determine, based on the contamination parameter, a cleaning solution formula and cleaning process parameters required for the current cleaning from a preset cleaning solution formula library; clean the wafer using a multi-stage cleaning strategy, the multi-stage cleaning including at least a first-stage cleaning and a second-stage cleaning, the first-stage cleaning and the second-stage cleaning using different cleaning solutions and cleaning process parameters; during the wafer cleaning process, acquire real-time monitoring data characterizing the cleaning process, the monitoring data including cleaning solution temperature, cleaning solution concentration, and cleaning solution flow rate; determine whether the cleaning solution formula and cleaning process parameters need to be adjusted based on the real-time monitoring data, and if so, dynamically adjust the cleaning solution formula and cleaning process parameters; after completing the wafer cleaning, acquire an image of the wafer surface after cleaning, compare and analyze the post-cleaning image with the pre-cleaning image to obtain a contamination removal rate; determine whether the contamination removal rate reaches a preset threshold, and if not, return to re-execute the cleaning process until the contamination removal rate reaches the preset threshold.

[0038] For example, the application of machine vision technology is crucial in the wafer manufacturing process. For example, high-resolution cameras capture images of the wafer surface, accurately identifying tiny surface contaminants such as dust or chemical residue. These images are analyzed using image processing algorithms to assess the wafer's contamination level. Based on this information, the cleaning solution's chemical formula can be precisely adjusted, such as increasing the proportion of a specific solvent to specifically remove specific types of contaminants. A multi-stage cleaning strategy is common during the cleaning process. For example, a milder cleaning solution may be used initially to remove large areas of contamination, followed by a more aggressive solution for localized treatment of stubborn contaminants. This phased approach effectively improves cleaning efficiency while minimizing potential damage to the wafer material. Real-time monitoring of cleaning solution parameters such as temperature, concentration, and flow rate is crucial to ensuring effective cleaning. For example, if sensors detect that the cleaning solution's temperature falls below a set point, the intelligent control system automatically heats the solution to maintain cleaning effectiveness. Similarly, if the concentration is low, the system automatically adds more cleaning agent. After cleaning is complete, machine vision technology is used to inspect the wafer surface again. By comparing the images before and after cleaning, the contamination removal rate can be calculated. For example, image analysis software can identify changes in the wafer surface before and after cleaning, such as the reduction in contaminants, and calculate the contamination removal rate. If the contamination removal rate does not meet the preset standard, the system automatically repeats the cleaning process until satisfactory cleaning results are achieved. This highly automated and precisely controlled cleaning process not only improves production efficiency but also ensures wafer quality and reliability, playing a vital role in the semiconductor manufacturing industry.

[0039] Step S105, after the cleaning meets the qualified standards, the surface of the wafer is morphologically detected using an atomic force microscope to obtain the roughness and defect density parameters of the wafer surface; the roughness and defect density parameters are compared with the preset wafer surface quality standards to determine whether the cleaning process causes damage to the surface structure of the wafer; if it is detected that the surface structure is damaged, the damage type and degree are recorded, and the parameters of the cleaning formula are adjusted to reduce similar damage.

[0040] A preset wafer surface quality standard is obtained, comprising a target roughness and a target defect density. Real-time monitoring data is obtained from a sensor module, comprising the degree of contamination on the wafer surface, the temperature of the cleaning solution, the concentration of the cleaning solution, and the flow rate of the cleaning solution. Based on the monitoring data, a cleaning solution formulation and cleaning process parameters are determined. The wafer is cleaned using a multi-stage cleaning strategy, comprising the use of different cleaning solutions and cleaning methods at different stages. Based on the monitoring data and historical cleaning data, a machine learning algorithm is used to optimize the cleaning process to obtain adjusted cleaning process parameters. The wafer is cleaned again based on the adjusted cleaning process parameters. An atomic force microscope is used to perform topographical inspection on the cleaned wafer surface to obtain surface roughness and defect density parameters. The obtained roughness and defect density parameters are compared with the preset wafer surface quality standard to determine whether the cleaning process has caused damage to the wafer surface structure. If damage is determined to have occurred, the type and extent of the damage are recorded. Based on the damage type and damage severity, the cleaning solution formulation parameters are adjusted to obtain a new cleaning solution formulation. The wafer is cleaned again using the new cleaning solution formulation. The surface of the cleaned wafer is again topographically inspected using an atomic force microscope to obtain new roughness and defect density parameters. The new roughness and defect density parameters are compared with the preset wafer surface quality standards to determine whether the new cleaning solution formulation effectively reduces the damage.

[0041] For example, using a sensor module to monitor the contamination level on the wafer surface and cleaning fluid parameters, such as temperature, concentration, and flow rate, in real time is crucial for ensuring effective cleaning. For example, a high-precision temperature sensor monitors the cleaning fluid temperature in real time, maintaining it within a range of 50±2°C to ensure optimal chemical activity. A concentration sensor monitors the reactive oxygen species concentration in the cleaning fluid, maintaining it within 0±1 mol / L to ensure sufficient oxidizing power to remove contaminants. A flow rate sensor monitors the flow rate of the cleaning fluid, maintaining it at 5±2 L / min to ensure uniform coverage of the wafer surface. Based on these monitoring results, an intelligent control system automatically adjusts the cleaning fluid formulation and process parameters. For example, if the temperature sensor indicates a low temperature, the system automatically heats the cleaning fluid to within the set range; if the concentration is low, the system automatically replenishes the reactive oxygen species. This intelligent adjustment mechanism, based on a pre-set algorithm model, rapidly responds to parameter changes, ensuring the cleaning process is always optimal. A multi-stage cleaning strategy utilizes different cleaning fluids and cleaning methods at different stages. For example, the first stage uses an alkaline cleaning solution to remove organic contaminants, the second stage uses an acidic cleaning solution to remove metal ions, and the third stage uses deionized water for rinsing. The time and temperature of each stage are strictly controlled. For example, the first stage is kept at 60°C for 10 minutes; the second stage is kept at 40°C for 8 minutes; and the third stage is kept at room temperature for 5 minutes. Machine learning algorithms are used to optimize the cleaning process, adjusting control parameters through the continuous accumulation of cleaning data. For example, the system records the temperature, concentration, flow rate, and corresponding contaminant removal rate for each cleaning step, and uses this data to train a machine learning model. The model analyzes the data and finds that the contaminant removal rate is highest at 55°C, so the system sets this temperature as the optimal setting. After cleaning, the wafer surface is topographically inspected using an atomic force microscope to obtain surface roughness and defect density parameters. For example, AFM scans show a surface roughness of 5nm and a defect density of 1×10^6 defects / cm after cleaning. 2 These parameters are compared with the preset wafer surface quality standards, which are roughness not exceeding 1nm and defect density not exceeding 2×10^6 / cm 2Based on the comparison results, the system determines whether the cleaning process has caused damage to the wafer surface structure. If surface damage is detected, the type and extent of the damage are recorded. For example, if microcracks are detected on the surface, the length and depth of the cracks are recorded, such as a crack length of 10μm and a depth of 2μm. Based on the type and extent of damage, the cleaning recipe parameters are adjusted to reduce the occurrence of similar damage. For example, if a large number of microcracks are present, this may be caused by the cleaning solution being too acidic. In this case, the system will adjust the concentration of the acidic cleaning solution, for example, from 2mol / L to 5mol / L. The wafer is then cleaned again using the adjusted cleaning recipe. The adjusted wafer surface is then repeatedly inspected using an atomic force microscope to confirm whether the new cleaning parameters have effectively reduced damage. For example, upon re-inspection of the cleaned wafer after adjustment, the crack length is reduced to 5μm and the depth is reduced to 1μm, indicating that the adjustment was effective. This multi-parameter monitoring and intelligent adjustment mechanism can significantly improve cleaning performance and reduce damage to the wafer surface. Through real-time monitoring and intelligent control, the cleaning process is always optimized, improving wafer yield. High-precision atomic force microscopy provides a reliable basis for evaluating cleaning effectiveness, ensuring that wafer surface quality meets production requirements. The application of machine learning algorithms continuously optimizes the cleaning process, improving cleaning efficiency and wafer quality. This comprehensive cleaning strategy not only improves cleaning results but also continuously optimizes the cleaning process through a data-driven approach, reducing resource waste and environmental pollution, and possesses significant technical and economic significance.

[0042] Step S106, record the cleaning process data of each batch of wafers, the cleaning process data including cleaning formula, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate and surface damage; apply big data analysis technology to explore the optimal cleaning process for different types of wafers, and dynamically update and optimize the cleaning formula database to improve cleaning efficiency and yield.

[0043] The sensor module monitors the wafer surface contamination level and cleaning fluid parameters in real time, including temperature, concentration, and flow rate. The central processing unit analyzes and processes the monitoring data to obtain cleaning process data for each batch of wafers, including cleaning formula, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate, and surface damage. The cleaning process is optimized based on a machine learning algorithm, and control parameters are adjusted through the continuously accumulated cleaning data.

[0044] For example, a sensor module monitors the contamination level on the wafer surface and cleaning fluid parameters, such as temperature, concentration, and flow rate, in real time. For example, the sensor module can include optical sensors and electrochemical sensors. The optical sensor detects surface contamination by measuring changes in reflected light intensity, while the electrochemical sensor monitors the cleaning fluid's concentration by measuring changes in conductivity. Suppose a monitoring result shows that the cleaning fluid temperature is 60°C, the concentration is 5%, the flow rate is 2 liters / minute, and the wafer surface contamination level is moderate. Based on the monitoring results, the intelligent control system automatically adjusts the cleaning fluid formula and cleaning process parameters. For example, if the system detects that excessively high temperature may cause damage to the wafer surface, it will automatically lower the temperature to 55°C and increase the cleaning fluid concentration to 6% based on the contamination level to enhance cleaning effectiveness. The intelligent control system typically adjusts parameters based on preset algorithms and thresholds to ensure optimal cleaning results. A central processing unit analyzes and processes the collected data to generate cleaning process data for each batch of wafers, including cleaning recipe, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate, and surface damage. For example, cleaning process data for a batch of wafers showed a 95% contamination removal rate, a 30-minute cleaning time, and a 10-liter cleaning volume. The central processing unit (CPU) analyzes this data in multiple dimensions, revealing potential issues during the cleaning process, such as excessive cleaning time, which may lead to increased surface damage. A human-machine interface displays various cleaning process parameters and status in real time, providing parameter settings and fault alarms. For example, the operator can view the current cleaning solution temperature, concentration, and flow rate, as well as the degree of contamination on the wafer surface. If the system detects an abnormally high cleaning solution temperature, the interface immediately displays a fault alarm, prompting the operator to intervene. The user-friendly interface facilitates quick understanding and response, improving production efficiency. Machine learning algorithms are used to optimize the cleaning process, adjusting control parameters based on accumulated cleaning data. For example, the system analyzes historical data and finds that a 6% cleaning solution concentration yields the highest contamination removal rate and lowest surface damage rate, setting these parameters to the optimal value. The application of machine learning algorithms makes the cleaning process more intelligent, dynamically adjusting parameters based on actual conditions to improve cleaning results. Cleaning parameters are automatically adjusted based on the degree of wafer contamination, reducing cleaning fluid waste. For example, if sensors detect light contamination on the wafer surface, the system automatically reduces the concentration and amount of cleaning fluid to avoid over-cleaning, saving costs and protecting the wafer surface. This dynamic adjustment mechanism not only improves resource utilization but also reduces environmental pollution. Big data analysis technology is used to discover the optimal cleaning process for different wafer types. For example, by analyzing large amounts of cleaning data, the system discovers that a specific cleaning recipe and process parameters for a certain wafer type produce the best cleaning results. This recipe is then stored in the database for future production reference.The application of big data analysis technology enables cleaning process optimization to no longer rely on empirical evidence, but rather on scientific data analysis, improving process reliability and consistency. The cleaning recipe database is dynamically updated and optimized to enhance cleaning efficiency and yield. For example, the system evaluates cleaning results after each cleaning cycle. If a new cleaning recipe is found to be more effective, the database is updated to ensure that the cleaning recipe in the database is always optimized. This dynamic update mechanism ensures continuous improvement of the cleaning process, enhancing production efficiency and product quality. The system also features self-diagnosis and self-repair capabilities, enabling timely detection and resolution of system failures, enhancing system reliability and stability. For example, the system uses a built-in diagnostic program to regularly check the operating status of each module. If a sensor failure is detected, it automatically switches to a backup sensor and prompts maintenance personnel to perform repairs. The application of self-diagnosis and self-repair capabilities significantly reduces the impact of system failures on production and improves system operational efficiency. Through the integrated application of these technologies and measures, the wafer cleaning process has achieved a high degree of automation and intelligence, improving cleaning results and product quality while also reducing production costs and environmental pollution, bringing significant economic and social benefits to the enterprise.

[0045] In step S107, the energy consumption and material consumption of the cleaning equipment are predicted and optimized based on the wafer production plan, feedback information from the equipment management system, historical cleaning effect data, and energy consumption data. An intelligent scheduling algorithm based on multi-objective optimization is applied to comprehensively consider factors such as cleaning effect, energy consumption, material consumption, and equipment utilization, so as to reasonably arrange cleaning tasks, minimize cleaning costs, and take into account both production efficiency and delivery time, thereby realizing intelligent management and control of the cleaning process.

[0046] Data from wafer production planning and equipment management systems, as well as historical cleaning performance and energy consumption data, are acquired. Based on this data, a machine learning algorithm is used to predict the energy and material consumption of the cleaning equipment. Based on the predicted results, a multi-objective optimization algorithm is applied to construct a function that minimizes cleaning costs. This function combines cleaning performance, predicted energy and material consumption, equipment utilization, production efficiency, and delivery time. Based on this objective function and constraints, the multi-objective optimization problem is solved to obtain a preliminary cleaning task scheduling plan. The sensor module monitors the wafer surface contamination level and cleaning fluid parameters, including temperature, concentration, and flow rate, in real time. Based on the wafer surface contamination level, the cleaning fluid concentration and flow rate are dynamically adjusted to optimize cleaning intensity and resource utilization. After determining the final cleaning parameters, the cleaning fluid formula and cleaning process parameters are adjusted to generate a final cleaning task scheduling plan. This cleaning task scheduling plan is distributed to the actuator module to automatically execute the cleaning process. After the cleaning process is completed, cleaning performance data, actual energy consumption data, and material consumption data are collected and stored in a historical database for subsequent prediction and optimization.

[0047] For example, during the wafer cleaning process, feedback from the wafer production plan and equipment management system is first needed. This data can help predict the energy and material consumption of the cleaning equipment. For example, if the production plan indicates that a large number of high-precision wafers will be processed in the coming week, the system can predict that the energy consumption of the cleaning equipment will increase during this period due to the need for a higher-standard cleaning process. Next, by applying machine learning algorithms, the system can analyze historical data and current equipment status to predict the energy and material consumption under different cleaning recipes and parameters. For example, the machine learning model may discover that using a specific cleaning agent formulation can significantly reduce energy consumption while maintaining cleaning effectiveness. Using a multi-objective optimization algorithm, the system comprehensively considers cleaning effectiveness, predicted energy and material consumption, equipment utilization, production efficiency, and delivery time to construct an objective function that minimizes cleaning costs. In practice, this may mean finding a balance between reducing energy consumption and ensuring cleaning quality. For example, by adjusting the flow rate and temperature of the cleaning fluid, the cleaning process can be optimized to minimize energy and material use without sacrificing wafer surface cleanliness. During the cleaning process, the sensor module monitors the wafer surface contamination level and cleaning fluid parameters such as temperature, concentration, and flow rate in real time. Based on the wafer contamination level, the intelligent control system automatically adjusts the cleaning fluid formula and flow rate. For example, for wafers with high contamination levels, the system may increase the cleaning fluid concentration and flow rate to enhance cleaning effectiveness; for wafers with low contamination levels, these parameters may be reduced to conserve resources and reduce potential damage to the wafers. Ultimately, based on real-time monitoring data and the results of the optimization algorithm, the system generates a final cleaning task scheduling plan and automatically executes the cleaning process through the actuator module. After cleaning is complete, the system collects cleaning performance data, actual energy consumption, and material consumption data and stores this data in a historical database. This data will be used for future predictions and optimization, helping to further reduce cleaning costs, improve cleaning efficiency, and improve wafer production quality. This approach achieves efficiency and automation throughout the cleaning process, ensuring high-efficiency and high-quality wafer cleaning while optimizing resource utilization and reducing production costs.

[0048] The above only lists some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and modifications can be made. As long as the improvements and modifications are made on the basis of the basic principles of the present invention, they should be considered to fall within the scope of protection of the present invention.

Claims

1. A cleaning method for a semiconductor wafer cleaning system, characterized in that: The method comprises the following steps: Step S101, obtaining a wafer surface image, processing the wafer surface image using an image segmentation algorithm, identifying contaminated areas and non-contaminated areas in the wafer surface image, and determining the degree of contamination and contamination type of the wafer surface based on the area, morphology, and grayscale characteristics of the contaminated areas; Step S102, selecting an optimal cleaning recipe from a preset cleaning recipe database based on the contamination level and the contamination type, and dynamically adjusting cleaning time, cleaning agent concentration, and cleaning temperature parameters in the optimal cleaning recipe based on the structural complexity of the wafer surface, the structural complexity including line width, aspect ratio, and pattern density, to obtain a dynamically adjusted cleaning recipe; Step S103: During the wafer cleaning process, the ambient temperature, humidity, and pressure parameters within the cleaning equipment are collected in real time by sensors, and the ambient temperature, humidity, and pressure parameters are compared with preset optimal ranges, which are determined based on historical cleaning data and wafer material properties. If the deviation of the comparison result exceeds a preset threshold, an alarm is triggered, and the process parameters of the cleaning equipment are automatically adjusted to maintain the stability of the cleaning environment. Step S104: After cleaning is completed, the wafer surface is inspected again using machine vision, the inspection image after cleaning is compared with the image before cleaning, the contamination removal rate is calculated, and the contamination removal rate is compared with a preset cleaning qualification standard. If the cleaning qualification standard is not met, the process returns to step S102 and the cleaning process is repeated until the cleaning qualification standard is met. Step S105: After the cleaning reaches the qualified standard, the surface of the wafer is tested using an atomic force microscope to obtain the roughness and defect density parameters of the wafer surface, and the roughness and defect density parameters are compared with the preset wafer surface quality standards to determine whether the cleaning process has caused damage to the surface structure of the wafer. If damage to the surface structure is detected, the type and degree of damage are recorded, and the parameters of the cleaning recipe are adjusted to reduce similar damage. Step S106: Recording cleaning process data for each batch of wafers, including cleaning recipes, cleaning parameters, cleaning time, cleaning agent dosage, contamination removal rate, and surface damage. Applying big data analysis technology to discover optimal cleaning processes for different wafer types, dynamically updating and optimizing the cleaning recipe database to improve cleaning efficiency and yield. In step S107, based on the wafer production plan, feedback information from the equipment management system, historical cleaning effect data and energy consumption data, the energy consumption and material consumption of the cleaning equipment are predicted and optimized. An intelligent scheduling algorithm based on multi-objective optimization is applied to comprehensively consider factors such as cleaning effect, energy consumption, material consumption and equipment utilization, reasonably arrange cleaning tasks, minimize cleaning costs, take into account production efficiency and delivery time, and realize intelligent management and control of the cleaning process.

2. The cleaning method of the semiconductor wafer cleaning system according to claim 1, characterized in that: The step S101 further comprises the following steps: a. Obtain high-definition images of the wafer surface and use image preprocessing algorithms to denoise and enhance the images to improve image quality; b. Based on the characteristics of the wafer surface, the image is divided into different areas using an image segmentation algorithm to distinguish between contaminated and non-contaminated areas; c. Extract the characteristic parameters of the area, shape and grayscale of the polluted area and construct the pollution feature vector; d. Input the contamination feature vector into a pre-trained contamination classification model to determine the contamination level on the wafer surface and obtain the contamination level. e. Input the pollution feature vector into the pre-trained pollution type classification model to determine the pollution type on the wafer surface and obtain the specific pollution type; f. Obtain the corresponding cleaning solution formula and cleaning parameters from the cleaning process database according to the pollution level and pollution type; g. Input the cleaning liquid formula and cleaning parameters into the cleaning equipment control module to automatically configure the process parameters of the cleaning equipment; h. Start the cleaning equipment and clean the wafer according to the configured process parameters; After cleaning is completed, obtain the wafer surface image again and repeat step be to determine the degree and type of contamination after cleaning. If the cleaning requirements are not met, return to step f and adjust the cleaning process until the cleaning is qualified.

3. The cleaning method of the semiconductor wafer cleaning system according to claim 1, wherein: The step S102 further includes: Obtaining contamination degree data and contamination type data on the wafer surface; selecting an optimal cleaning formula from a preset cleaning formula database according to the contamination degree data and the contamination type data; Obtaining the structural complexity of the wafer surface, wherein the structural complexity includes line width, aspect ratio, and pattern density; Dynamically adjusting the cleaning time, cleaning agent concentration, and cleaning temperature parameters in the optimal cleaning formula according to the complexity of the structure to obtain a dynamically adjusted cleaning formula; Obtaining wafer surface data after cleaning; Determining whether the surface data of the cleaned wafer reaches a preset threshold; If not achieved, the cleaning recipe optimization process is triggered.

4. The cleaning method of a semiconductor wafer cleaning system according to claim 1, wherein: The step S103 further includes: Obtain the ambient temperature, humidity, and pressure parameters inside the wafer cleaning equipment collected in real time by the sensor module; Obtain the optimal environmental parameter range that matches the current wafer material characteristics from the historical cleaning database; Comparing the real-time collected environmental temperature, humidity and pressure parameters with the optimal environmental parameter range to obtain a deviation value; If the deviation value exceeds the preset threshold, an alarm is triggered, and the cleaning equipment process parameters that need to be adjusted and the adjustment range are determined; According to the process parameter adjustment instruction, the corresponding process parameters of the cleaning equipment are adjusted to obtain an adjustment result; If the deviation value still exceeds the preset threshold value after multiple consecutive adjustments, it is determined that there is a fault in the cleaning equipment and the equipment self-test program is triggered.

5. The cleaning method of a semiconductor wafer cleaning system according to claim 1, wherein: The step S104 further includes: Acquiring a wafer surface image, wherein the image includes information on the degree of contamination on the wafer surface; Analyzing and processing the image to obtain a contamination parameter representing the degree of contamination on the wafer surface; Determine the cleaning solution formula and cleaning process parameters required for this cleaning from a preset cleaning solution formula library according to the pollution parameters; Cleaning the wafer using a multi-stage cleaning strategy, wherein the multi-stage cleaning includes at least a first-stage cleaning and a second-stage cleaning, wherein the first-stage cleaning and the second-stage cleaning use different cleaning solutions and cleaning process parameters; During the wafer cleaning process, obtaining real-time monitoring data characterizing the cleaning process, wherein the monitoring data includes cleaning solution temperature, cleaning solution concentration, and cleaning solution flow rate; Determining whether the cleaning solution formulation and cleaning process parameters need to be adjusted based on the real-time monitoring data, and dynamically adjusting the cleaning solution formulation and cleaning process parameters if necessary; After the wafer is cleaned, an image of the wafer surface after cleaning is obtained, and the image after cleaning is compared and analyzed with the image before cleaning to obtain the contamination removal rate; Determine whether the pollution removal rate reaches a preset threshold. If not, return to re-execute cleaning until the pollution removal rate reaches the preset threshold.

6. The cleaning method of a semiconductor wafer cleaning system according to claim 1, wherein: The step S105 further includes: Obtaining a preset wafer surface quality standard, where the preset wafer surface quality standard includes a target roughness and a target defect density; Acquire real-time monitoring data from the sensor module, including the degree of contamination on the wafer surface, the temperature of the cleaning solution, the concentration of the cleaning solution, and the flow rate of the cleaning solution; Determine the cleaning fluid formula and cleaning process parameters based on monitoring data; A multi-stage cleaning strategy is used to clean the wafers, which includes using different cleaning fluids and cleaning methods at different stages; Based on monitoring data and historical cleaning data, machine learning algorithms are used to optimize the cleaning process and obtain adjusted cleaning process parameters; The wafer is cleaned again according to the adjusted cleaning process parameters; Atomic force microscopy is used to detect the surface morphology of the cleaned wafer to obtain the roughness and defect density parameters of the wafer surface; Compare the obtained roughness and defect density parameters with the preset wafer surface quality standards to determine whether the cleaning process has caused damage to the wafer surface structure; If the wafer surface structure is judged to be damaged, the damage type and degree are recorded; According to the damage type and damage degree, the formula parameters of the cleaning fluid are adjusted to obtain a new cleaning fluid formula; Clean the wafer again according to the new cleaning solution formula; The cleaned wafer surface is again inspected using an atomic force microscope to obtain new roughness and defect density parameters. The new roughness and new defect density parameters are compared with the preset wafer surface quality standards to determine whether the new cleaning fluid formula effectively reduces damage.

7. A control system for a semiconductor wafer cleaning system, characterized in that: The control system adopts the cleaning method of the semiconductor wafer cleaning system according to any one of claims 1 to 6 to control operation, and the control system includes: Wafer surface contamination detection module, used to identify contaminated areas and determine the degree and type of contamination; Cleaning recipe optimization module, used to select the optimal cleaning recipe and dynamically adjust parameters according to the wafer structure; Cleaning process monitoring module, used to monitor cleaning environment parameters in real time and automatically adjust to maintain stability; Cleaning effect evaluation module, used to detect the wafer surface after cleaning and determine whether it meets the cleaning standards; Data analysis and optimization module, used to record cleaning data, analyze optimal processes and update the cleaning formula database; Intelligent scheduling and control module is used to optimize the energy and material consumption of cleaning equipment and reasonably arrange cleaning tasks.

Images