A process for controlling edge particles of 12-inch reclaimed silicon wafers in conjunction with SPC monitoring.

By combining graded pretreatment, zoned polishing, and multi-stage cleaning with SPC monitoring, the problem of unstable edge particle control in regenerated silicon wafers was solved, achieving efficient silicon wafer cleaning and polishing synergy, and improving the quality of regenerated silicon wafers and the stability of the production line.

CN123124755APending Publication Date: 2026-11-17ANHUI FULLERDE CHANGJIANG SEMICON MATERIALS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611277772.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies for controlling edge particles in 12-inch reclaimed silicon wafers suffer from problems such as unstable cleaning effects, inability to coordinate with upstream polishing processes, and lack of real-time monitoring mechanisms. These issues lead to excessive edge particles in reclaimed silicon wafers, affecting the yield of subsequent processes.

Method used

The process employs a combined approach of graded pretreatment, zoned differentiated polishing, multi-stage chemical cleaning, and SPC monitoring. This includes dry purging, wet cleaning, V-Notch precision alignment polishing, multi-stage constant temperature turbulence of polishing fluid, and real-time monitoring. Combined with SPC data binding, a closed-loop process quality system is formed.

Benefits of technology

It significantly improves the edge cleanliness and process stability of recycled silicon wafers, reduces manufacturing costs, adapts to existing production lines to achieve large-scale industrial production, and improves the yield of recycled silicon wafers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123124755A_ABST
    Figure CN123124755A_ABST
Patent Text Reader

Abstract

This application relates to the field of semiconductor manufacturing processes, specifically disclosing a process for controlling edge particles of 12-inch reclaimed silicon wafers in conjunction with SPC monitoring, aiming to solve the technical problem of effectively controlling edge particles in reclaimed silicon wafers. This process establishes a linkage mechanism between edge particle detection and the SPC system to accurately process and monitor the edges of the reclaimed silicon wafer in real time, and dynamically adjusts process parameters based on SPC data analysis results. By adopting the above technical solution, this application can achieve effective control of edge particles in reclaimed silicon wafers, improving product yield and quality stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of semiconductor manufacturing processes, specifically relating to a process for controlling and linking edge particles of a 12-inch reclaimed silicon wafer with SPC monitoring. Background Technology

[0002] As the semiconductor integrated circuit industry continues to evolve towards advanced process nodes, the use of 12-inch silicon wafers as the mainstream substrate material in chip manufacturing is increasing daily. Among them, recycled silicon wafers are widely used because they can effectively reduce silicon wafer procurement costs. Recycled silicon wafers, after undergoing multiple chemical mechanical polishing processes, are reused on the production line. Their surfaces, especially the edge areas, inevitably suffer from accumulated damage, scratches, and various impurities. The edge areas of silicon wafers are most prone to microcracks, chipping, and particle contamination due to the significant mechanical stress they experience during cutting, grinding, and handling. If these surface defects are not effectively controlled in upstream processes, they will directly lead to a decrease in the yield of subsequent critical processes such as photolithography, etching, and ion implantation, and in severe cases, even result in the scrapping of the entire silicon wafer. Therefore, the level of edge particle control has become a core indicator for measuring the quality of recycled silicon wafers and a key aspect of process control for semiconductor manufacturing companies.

[0003] Among these, edge particle control and cleaning processes for regenerated silicon wafers are a crucial technological direction in this field. The goal is to effectively remove particulate contaminants from the edge areas of silicon wafers by optimizing cleaning process parameters and equipment configuration, ensuring that the surface cleanliness of the wafers meets the requirements of subsequent chip manufacturing. In recent years, the industry has conducted extensive research on silicon wafer cleaning technology, disclosing various technical solutions for zoned cleaning using spray devices. These solutions aim to improve the cleaning effect in edge areas by setting nozzles in different areas and employing alternating spraying methods for multiple rounds of cleaning.

[0004] Existing technologies still have significant shortcomings when applied to 12-inch reclaimed silicon wafers. First, existing technologies are primarily designed for virgin wafers or wafers processed using conventional methods, failing to adequately consider the unique conditions of reclaimed wafers, such as uneven edge damage due to repeated use and complex incoming material conditions. This makes it impossible to precisely address the differentiated defects of reclaimed wafers, resulting in inconsistent and unreliable cleaning performance. Second, existing technologies generally focus on the cleaning process itself, lacking coordination with upstream polishing processes. Edge particles and metal deposits generated during polishing cannot be effectively controlled at the source; cleaning often only serves as a post-processing remedy, failing to fundamentally solve the problem of excessive edge particles. Third, existing technologies generally lack a linkage monitoring mechanism with statistical process control (SPC) systems. Adjustments to process parameters still rely on manual judgment and post-processing inspection, resulting in delayed responses and a lack of real-time early warning. This weak process control capability fails to meet the stringent requirements of semiconductor manufacturing for process stability. Therefore, developing a 12-inch reclaimed silicon wafer edge particle control process that achieves coordinated optimization of polishing and cleaning and possesses SPC linkage monitoring capabilities has significant industrial application value for improving reclaimed silicon wafer quality and reducing chip manufacturing costs. Summary of the Invention

[0005] The purpose of this invention is to provide a process for controlling and linking edge particles of a 12-inch regenerated silicon wafer with SPC monitoring, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A process for edge particle control and SPC monitoring of 12-inch reclaimed silicon wafers includes the following specific steps: Step 1: Incoming reclaimed silicon wafer grading pretreatment: Full-area scanning and grading of the 3mm edge ring area of ​​the reclaimed silicon wafer, dry high-pressure nitrogen purging to remove surface dust and metal debris, wet neutral cleaning solution spraying and immersion combined with soft fiber roller wiping, completing the preparation of the silicon wafer to be polished; Step 2: Differentiated CMP main polishing: Using the V-Notch edge precision alignment system to distinguish between the central area and the 3mm edge control area, a two-stage gradient polishing process is implemented. The pressure is controlled at 1.8 to 2.2 psi in the rough polishing stage and the pressure in the fine polishing stage. The temperature of the polishing solution is controlled at 1.0 to 1.2 psi, and the temperature of the polishing solution is controlled at 22±1℃ with dynamic turbulent circulation supply. The polishing parameters are monitored in real time throughout the process. Step 3: Graded chemical cleaning and precision drying: HF micro-etching pretreatment, O3~pure water oxidation cleaning, SC1 alkaline deep particle removal, and SC2 acidic metal fine cleaning are performed in sequence. After cleaning, low-speed rotary spin drying combined with gradient heating nitrogen constant temperature purging drying is used. Step 4: Final inspection and sorting and SPC data binding: Metal content testing and edge particle quantity detection are performed on each piece. The test data are synchronously entered into the weekly monitoring report and automatically linked to the trend curve to form a closed loop for process quality monitoring.

[0008] Preferably, in step 1, the incoming recycled silicon wafer grading pretreatment uses a wafer micro-defect detector to perform a full-area scan of the 3mm edge ring area of ​​each recycled silicon wafer. It is classified into three levels, A / B / C, according to the depth of edge chipping, micro-cracks, and scratches. Level A indicates slight scratches that can be directly entered into the process, Level B indicates moderate edge damage that requires special pretreatment for repair, and Level C indicates severe chipping or deep cracks that are isolated and scrapped and prohibited from entering the polishing process. This grading mechanism effectively avoids the problem of inconsistent results caused by uniform treatment of silicon wafers with different degrees of damage.

[0009] Preferably, in step 1, the dry impurity pretreatment adopts a nitrogen high-pressure uniform purging process with a nitrogen purity of not less than 99.999% and a purging pressure controlled between 0.35 and 0.4 MPa, fully covering the silicon wafer surface and the V-Notch area at the edge. This process parameter design can effectively remove hard impurities such as surface dust, free metal debris, and residual polishing powder particles, avoiding the introduction of hard particles into the polishing process to cause secondary scratches and metal residue, thus controlling the source of particulate contamination from the source.

[0010] Preferably, in step 1, the wet fine cleaning pretreatment uses a low-concentration neutral cleaning solution for spraying and immersion, with the immersion time controlled to be 3 to 5 minutes. A soft microfiber roller is used to gently wipe the edge area, specifically dissolving organic stains, oxide layers, and trace metal deposits on the incoming material surface. After pretreatment, pure water overflow is used for 3 rinses to thoroughly remove the removed impurities. This treatment method removes the old impurity layer on the surface without damaging the silicon wafer substrate, providing a clean surface for subsequent polishing processes.

[0011] Preferably, in step 2, the precise positioning and calibration of the machine adopts a V-Notch edge precision alignment system to lock the silicon wafer edge contour, distinguish the central area of ​​the silicon wafer from the critical 3mm edge control area, and enable a dual-mode partitioned process of conventional polishing of the center and fine polishing of the edge. This partitioned strategy abandons the drawbacks of traditional uniform parameter whole-surface polishing and realizes differentiated process control for different areas. The V-Notch edge precision alignment system consists of a transmission laser light source and a linear CCD receiving array. The laser wavelength is selected as 650nm. After the beam passes perpendicularly through the V-Notch area of ​​the silicon wafer edge, it forms a characteristic projection waveform on the CCD array. The built-in DSP processor performs first-order differential calculation on the waveform and extracts the inflection point coordinates as the precise position mark of the V-Notch. Based on this mark, a polar coordinate system is established. The ring piezoelectric ceramic actuator array mounted on the polishing head is driven by a servo motor. The PLC controller switches the downward pressure of the piezoelectric ceramic unit in real time according to the polar coordinate angle information to realize dynamic partitioned switching of pressure between the central area and the critical 3mm edge control area. The real-time angle tracking and coordinate mapping mechanism is as follows: the silicon wafer spindle encoder outputs a real-time angle signal θ(t) with a resolution of 4096 pulses per revolution. The DSP processor reads the current angle value in each control cycle (500μs) and, combined with the initial locking angle θ0 marked by the V-Notch, calculates the real-time angle range of the 3mm edge control area in the polar coordinate system according to the following formula: [θ(t)-Δθ, θ(t)+Δθ], where Δθ=arcsin(3 / R), R is the silicon wafer radius (150mm), and Δθ≈1.146° is calculated. The piezoelectric ceramic actuator array consists of 48 units evenly distributed along the circumference, with each unit covering a 7.5° angle range. The PLC controller queries the mapping table between the angle range and the piezoelectric unit number in real time based on the current θ(t) value, and sends a pressure boosting command to the piezoelectric unit about to enter the edge control area two control cycles (i.e., 1ms) in advance. This causes the pressure of the unit to linearly switch from the central area reference value of 1.5psi to the edge fine polishing set value of 1.0-1.2psi, with a switching ramp time of 200μs. This ensures that the pressure switching is completed before the edge area of ​​the silicon wafer enters the effective working area of ​​the polishing pad. The response timing is synchronized with the silicon wafer rotation, with a phase lag of less than ±0.5°.

[0012] Preferably, in step 2, the first stage of the two-stage gradient polishing process uses a low-pressure, high-flow-rate polishing slurry supply mode, with the pressure controlled at 1.8 to 2.2 psi, to prioritize and quickly repair obvious edge scratches, burrs, and protrusion defects, and eliminate large particle defect sources. In the second stage, the fine polishing stage, the polishing pressure is reduced to 1.0 to 1.2 psi to improve the uniformity of the turntable speed, and to smooth and repair micro-damage and micro-cracks at the edges, preventing silicon chips and metal microparticles generated by over-polishing. This two-stage processing strategy effectively controls particle generation while ensuring defect repair efficiency.

[0013] Preferably, in step 2, the constant temperature turbulence controlled deposition process of the polishing slurry keeps the temperature of the polishing slurry constant within the range of 22±1℃ and adopts a dynamic turbulence circulation supply method to avoid the stagnant accumulation of metal solutes in the polishing slurry at the edge of the silicon wafer. At the same time, the polishing table automatically performs spray cleaning after each wafer process to remove residual particles and metal precipitates from the disk surface and prevent cross-contamination between batches. This design effectively reduces the generation of Fe source caused by the deposition of metal components in the polishing slurry from the process end. The specific configuration of the dynamic turbulent circulation liquid supply system is as follows: the main liquid supply pipeline is arranged in a ring, and six liquid supply nozzles are installed at equal angles along the outer periphery of the polishing pad. A vortex generator driven by a stepper motor is connected in series at the front end of each nozzle. The device has three spiral guide vanes inside. The stepper motor drives the vane group to rotate and generate tangential turbulent flow. A plate heat exchanger is connected in series in the circulation loop. The PID controller adjusts the flow in real time according to the thermocouple signal of the main liquid supply pipeline. A miniature constant temperature chamber with a buffer volume of 100mL is set at the rear end of each nozzle. The outer wall of the chamber is wrapped with electric heating tape as a heat source for fine adjustment. Each chamber is controlled by an independent temperature sensor in a closed loop to ensure that the temperature of each liquid supply point is stable within the range of 22±1℃.

[0014] Preferably, in step 2, the real-time monitoring of the process covers key parameters such as polishing speed, liquid supply flow rate, and pressure fluctuation. When any parameter deviates from the set threshold, a shutdown fine-tuning mechanism is immediately triggered to ensure the polishing consistency of the edges of each silicon wafer and avoid particle abnormalities and metal residue fluctuations caused by process fluctuations. This real-time monitoring mechanism provides a guarantee for process stability.

[0015] Preferably, in step 3, the HF micro-etching pretreatment involves immersing the silicon wafer in diluted HF solution for a short period of 20 to 30 seconds. This process gently removes the oxide layer on the silicon wafer surface, as well as trace metal impurities and embedded microparticles encapsulated by the oxide layer, exposing a clean silicon substrate. This lays the groundwork for subsequent deep cleaning steps. This pretreatment process effectively solves the problem of removing deep particles caused by oxide layer encapsulation.

[0016] Preferably, in step 3, the O3-pure water oxidation cleaning involves introducing O3-ultrapure water with an ozone concentration of 20 to 25 mg / L and using a circulating spray cleaning method for 4 minutes. The strong oxidizing effect of ozone decomposes the residual organic stains and carbonaceous impurities from polishing, eliminates the carrier effect of organic media adsorbing metal particles, and reduces the adhesion of metal contaminants in subsequent processes. This oxidation cleaning step creates favorable conditions for subsequent alkaline cleaning.

[0017] Preferably, in step 3, the SC1 alkaline deep particle removal uses a standard SC1 cleaning solution, namely a mixture of ammonia, hydrogen peroxide and ultrapure water in a specific ratio, and is immersed and cleaned at a constant temperature of 45 to 50°C for 6 minutes. Through alkaline chelation, the micro-silicon particles and suspended metal debris attached to the surface and edges of the silicon wafer are peeled off, with a focus on removing residual particles in the dead corners of the edges. This process has a highly efficient removal capability for micron-sized particles.

[0018] Preferably, in step 3, the SC2 acidic metal cleaning uses a standard SC2 cleaning solution, which is a mixture of hydrochloric acid, hydrogen peroxide and ultrapure water in a specific ratio. The cleaning is carried out at a constant temperature of 35 to 40°C for 5 minutes. This specifically chelates and dissolves inorganic metal residues such as metals and metal oxides, significantly reducing the metal content on the silicon wafer surface. This achieves the metal control target from the cleaning end, and the acidic cleaning step ensures that the metal residues meet the requirements of subsequent processes.

[0019] Preferably, in step 3, the multi-stage low-residue drying process first uses a low-speed rotation method for preliminary spin drying after cleaning to remove free water droplets on the surface. Then, it uses gradient-heated nitrogen gas for constant-temperature purging. The nitrogen gas temperature is controlled at 40°C and a uniform-speed purging method is used to cover the edges and the entire area. This drying strategy effectively avoids water stains and particle adsorption problems caused by rapid drying. At the same time, the entire process is carried out in a closed dust-free environment to prevent secondary pollution caused by external dust.

[0020] Preferably, in step 4, the final inspection and sorting process and SPC data binding process perform Surface Metal content testing and edge particle quantity detection on each wafer after cleaning and drying, and record the detection data of a single batch and a single machine in detail. This detection mechanism ensures that the quality of each silicon wafer is traceable and provides a complete data foundation for subsequent data analysis.

[0021] Preferably, step 4, SPC data binding, also includes synchronously recording the machine output product data into the weekly monitoring report. The system automatically associates the trend chart curve and promptly detects process parameter drift and abnormal fluctuations through trend analysis. This monitoring mechanism realizes the transformation from passive detection to proactive prevention. In the SPC data binding mechanism, after the detection data is uploaded to the SPC server via the RS-232 interface, the server's built-in statistical calculation engine calculates the moving range and process mean for the continuously collected n samples (n≥5) and compares them with the Western Electric judgment rules. When the discrepancy criterion is triggered, the regression analysis module is started. Using polishing pressure, liquid supply flow rate, and turntable speed as independent variables and edge particle number as dependent variable, a multiple linear regression model is used to calculate the recommended correction direction and correction step size for each parameter. The correction step size is output in three levels, ranging from ±2% to ±5% of the current parameter value. The calculated correction amount is encapsulated in Modbus TCP protocol message format and sent to the data register area of ​​the polishing machine PLC controller via industrial Ethernet. The PLC scans the register in each control cycle (100ms), adds the correction amount to the current set value through the built-in PID calculation module, and converts it into a 4-20mA current signal by the analog output module to drive the electric regulating valve and frequency converter, realizing closed-loop automatic control. The specific mathematical form of the multiple linear regression model is: Y = β0 + β1·P + β2·F + β3·S + ε, where Y is the number of edge particles (dependent variable), P is the polishing pressure (psi), F is the liquid supply flow rate (mL / min), S is the turntable speed (rpm), β0 is the intercept term, β1, β2, and β3 are the regression coefficients of their respective variables, and ε is the residual term. The model training set consists of the most recent 200 batches of historical production data, with each batch containing the measured values ​​of the above four variables. The feature normalization method uses Z-score standardization, which involves subtracting the sample mean from each variable and then dividing by the sample standard deviation to eliminate the influence of dimensions. The regression coefficients are updated online using the least squares method, with an update cycle of one recursive least squares calculation triggered every 10 new batches of data. The step size for correction is determined as follows: when the standardized residual |e| / σ < 0.5, the step size is ±2%; when 0.5 ≤ |e| / σ < 1.0, the step size is ±3.5%; and when |e| / σ ≥ 1.0, the step size is ±5%. To prevent adjustment oscillations, a first-order lag filter is connected in series at the correction output. The filter time constant Tf = 3.0s, meaning the actual correction output for the current period = 0.716 × the calculated correction for this period + 0.284 × the actual correction output for the previous period (corresponding to the smoothing coefficient when Tf = 3.0s and the sampling period is 100ms).

[0022] Preferably, step 4 also includes releasing products that meet the standards for particle and metal residue into the warehouse and shipping them out, isolating and rewashing abnormal batches and reviewing process parameters. This closed-loop management mechanism ensures that non-conforming products do not flow into the next process, and at the same time, continuously optimizes process parameters through process review, forming a complete closed loop for process quality monitoring.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. Targeted adaptation to the characteristics of 12-inch reclaimed silicon wafers: In view of the problem of uneven edge damage and many scratches on reclaimed silicon wafers, the particle source is first removed through pretreatment and fine grinding repair. Then, personalized parameter control of edge polishing is achieved through V-Notch precise positioning, avoiding edge damage and increased particles caused by traditional uniform polishing, and significantly improving the material compatibility and process applicability of reclaimed silicon wafers.

[0025] 2. Multi-stage polishing for particle control: The two-stage edge polishing of coarse and fine polishing is combined with the CMP zone dynamic polishing strategy to remove edge damage and burrs while avoiding over-polishing or under-polishing. At the same time, the real-time cleaning of the polishing table and the constant temperature turbulent liquid supply mechanism reduce the deposition of polishing liquid particles at the edges, thereby controlling the source of particle generation from the polishing process.

[0026] 3. Polishing-cleaning-drying synergistic process to prevent secondary particle adhesion: After polishing, a graded cleaning process of HF micro-etching, O3 oxidation, SC1 alkaline and SC2 acidic is used to specifically remove different types of organic residues, metal contaminants and microparticles. Combined with slow lifting drying and gradient heating nitrogen purging process, water stains and particle residues are reduced, further improving edge cleanliness.

[0027] 4. Stable process and industrial mass production capability: The entire process parameters are clear and controllable, and it can be adapted to existing 12-inch silicon wafer regeneration production lines without complicated equipment modifications. It can achieve large-scale industrial production, significantly improve the yield and quality of regenerated silicon wafers, reduce semiconductor manufacturing costs, and has good prospects for industrial applications. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall technical solution architecture for a process of controlling and linking SPC monitoring of edge particles of a 12-inch reclaimed silicon wafer according to an embodiment of this application.

[0029] Figure 2 This is a schematic diagram of the core principle framework of the linkage mechanism between edge particle control and SPC monitoring in a process of controlling edge particles of a 12-inch reclaimed silicon wafer according to an embodiment of this application.

[0030] Figure 3This is a schematic diagram of the logical flow framework of particle detection, control decision and SPC feedback adjustment in a process of controlling and linking edge particles of a 12-inch reclaimed silicon wafer according to an embodiment of this application. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0032] This invention provides a process for controlling edge particles on 12-inch reclaimed silicon wafers and linking it with SPC monitoring. This process encompasses four core steps: graded pretreatment to eliminate the substrate for particle generation, zoned differentiated polishing to control the source of particle generation, multi-stage chemical cleaning to prevent secondary particle adhesion, and final inspection and sorting linked with SPC data to achieve a closed-loop process quality monitoring system. The specific technical solutions for each step are described in detail below.

[0033] In the above method, step 1 is a graded pretreatment of recycled silicon wafers. This step systematically treats the characteristics of recycled silicon wafers, such as uneven edge damage, hidden scratches, and accumulation of old impurities on the surface, eliminating the formation of microparticles in the substrate after polishing from the source and avoiding embedded particle contamination that cannot be removed by subsequent cleaning.

[0034] Specifically, step 1 comprises three sub-steps forming a complete preprocessing system. First, an incoming material appearance and damage grading and sorting process is performed. A wafer micro-defect detector is used to scan the entire 3mm edge ring area of ​​each recycled silicon wafer. This detector has an optical resolution better than 0.5μm, enabling precise identification of edge chipping, microcracks, scratches, and other defect morphologies in the edge area. Based on quantitative indicators such as edge chipping size, microcrack length and distribution density, and scratch depth, the recycled silicon wafers are classified into three grades: A, B, and C. Grade A represents minor scratches, with a scratch depth of less than 0.5 μm and no microcrack extension. Wafers of this grade proceed directly to subsequent processes. Grade B represents moderate edge damage, with scratches between 0.5 μm and 2 μm in depth or the presence of localized microcracks that do not penetrate the edge. Wafers of this grade undergo enhanced treatment in a dedicated pre-processing repair channel. Grade C represents severe edge chipping or deep cracks, with chipped edges exceeding 50 μm or microcracks exceeding 2 μm in depth and exhibiting a network distribution. Wafers of this grade are isolated and scrapped, and are prohibited from proceeding to the polishing process. This grading mechanism ensures that wafers with different levels of damage receive targeted treatment solutions, avoiding inconsistent results caused by uniform treatment.

[0035] Further, in step 1, the wafer micro-defect detector quantitatively measures the scratch depth and micro-crack depth. The data processing flow is as follows: The detector's built-in linear CCD sensor acquires a two-dimensional grayscale image of the edge ring area. The image is transmitted to the built-in FPGA preprocessing unit for median filtering, noise reduction, and edge enhancement. Subsequently, the ARM processor runs a preset depth inversion algorithm. This algorithm performs least-squares fitting between the grayscale ratio of the defect area under bright and dark illumination and a standard depth sample library, inverting the two-dimensional grayscale signal into a three-dimensional physical depth value. The calculated depth data is compared with the built-in A / B / C three-level thresholds (0.5μm and 2μm, respectively). The judgment result is output to the sorting execution mechanism PLC controller via an RS-232 communication interface. The PLC drives a high-speed solenoid valve to control the airflow nozzles according to the grading result, diverting the silicon wafers to the corresponding A-level direct-through channel, B-level repair channel, or C-level scrap channel, achieving fully automatic grading and sorting. The specific definition of the grayscale ratio is: , where I bright I represents the pixel grayscale value (8-bit grayscale, value range 0-255) of the defect area under bright field illumination. dark The grayscale value of the pixel in the same defect area under dark illumination is given. The method for constructing the standard depth sample library is as follows: Standard reclaimed silicon wafers are selected, and standard samples of scratches and microcracks with known depths (9 levels: 0.1μm, 0.3μm, 0.5μm, 0.8μm, 1.0μm, 1.5μm, 2.0μm, 2.5μm, and 3.0μm) are prepared using focused ion beam (FIB). Bright-field and dark-field grayscale values ​​corresponding to each depth are collected under the same illumination conditions using the same detector, and the R value corresponding to each depth is calculated. gray To establish a standard mapping table of depth-grayscale ratios, at least 30 samples are collected for each depth level, and the average value is taken as the standard response value for that depth. The specific equation form of the least squares fitting is: Let the measured grayscale ratio be R_i, and the corresponding depth value be D. i (Unit: μm), a quadratic polynomial model D(R) = a·R² + b·R + c was used to fit 9 sets of data points from the standard sample library. The objective function was: (j=1,2,…,9), the optimal estimates of coefficients a, b, and c are obtained by solving the normal equation system. The measurement accuracy verification method is as follows: three depth samples (0.2μm, 0.7μm, 1.2μm) that were not used in the training are reserved in the standard sample library as a validation set. The calculated depth values ​​are compared with the FIB measured values. The maximum absolute deviation is required to be ≤0.05μm and the average absolute deviation is required to be ≤0.03μm. Only those that meet this accuracy requirement can be put into online use. Otherwise, the fitting coefficients are recalibrated.

[0036] The next step involves a dry impurity pretreatment process, employing a high-pressure, uniform nitrogen purging technique to clean the surface of the graded and sorted silicon wafers. The nitrogen purity is no less than 99.999%, and the purging pressure is precisely controlled within the range of 0.35 to 0.4 MPa. A specially designed fan-shaped nozzle achieves full-area coverage purging at a scanning speed of 50 mm per second, ensuring uniform nitrogen flow across the silicon wafer surface and the V-Notch edge area. This dry purging process effectively removes surface dust, free metal debris, and residual polishing powder particles, achieving a hard particle removal rate of over 95%. The purpose of this step is clearly to prevent hard particles from being introduced into the polishing process, causing secondary scratches and metal residue, thus controlling particulate contamination at its source.

[0037] Finally, a wet fine cleaning pretreatment process is performed, using a low-concentration neutral cleaning solution for spray immersion. The pH value of the neutral cleaning solution is controlled between 6.5 and 7.5, and the immersion time is precisely set to 3 to 5 minutes. During this process, a soft ultra-fine fiber roller is used for gentle wiping, with the roller speed controlled at 30 to 50 revolutions per minute. The fiber diameter is less than 10 μm. The wet cleaning process specifically dissolves and removes organic stains, oxide layers, and trace metal deposits from the incoming material surface, with an organic stain removal efficiency of over 90%. After pretreatment, a pure water overflow rinse is performed three times, with each overflow lasting 30 seconds, ensuring that the stripped impurities are completely removed, completing the preparation of the silicon wafer to be polished. This treatment method removes the old impurity layer on the surface without damaging the silicon wafer substrate, providing a clean surface for subsequent polishing processes.

[0038] In the above method, step 2 is the partitioned differentiated CMP main polishing. This step abandons the drawbacks of traditional uniform parameter whole-surface polishing. Through the technical solution of combining partitioned dynamic polishing with edge-specific parameter control, it achieves precise control of particles at the edge of the regenerated silicon wafer, while reducing the deposition of metal components in the polishing slurry and reducing the generation of Fe source from the process end.

[0039] Specifically, step 2 comprises four key sub-steps. First, precise positioning and calibration of the equipment are performed using a V-Notch edge precision alignment system to accurately position the silicon wafer. This alignment system identifies the V-shaped notch markings on the wafer edge, locking in the geometric features of the wafer edge contour with a positioning accuracy better than ±0.1mm. Based on the edge contour positioning results, the wafer surface is divided into two independent regions: a central region and a critical 3mm edge control region. The central region uses a conventional polishing mode, while the critical 3mm edge control region uses a refined polishing mode, achieving a dual-mode partitioned process. This partitioning strategy breaks the limitations of traditional uniform parameter polishing, employing differentiated process parameter control strategies for different regions.

[0040] Furthermore, the identification principle and partition switching mechanism of the V-Notch edge precision alignment system are as follows: The alignment system consists of a transmission laser light source and a linear CCD receiving array. The laser wavelength is selected as 650nm. After the beam passes perpendicularly through the V-Notch area at the edge of the silicon wafer, it forms a characteristic projection waveform on the CCD array. The built-in DSP processor performs a first-order differential operation on the waveform and extracts the inflection point coordinates in the waveform as the precise position mark of the V-Notch, with a positioning repeatability accuracy of ±0.05mm. Based on this mark position, the system establishes a polar coordinate system with the center of the silicon wafer as the origin. The system drives the annular piezoelectric ceramic actuator array mounted on the polishing head through a servo motor. The actuator consists of 48 independently controlled piezoelectric ceramic units evenly distributed along the circumference. The PLC controller switches the downward pressure of the piezoelectric ceramic units in the corresponding angle range in real time according to the polar coordinate angle information when the silicon wafer rotates to different angles, realizing dynamic partition switching of pressure between the central area and the 3mm edge control area, with a switching response time of less than 100ms. The real-time angle tracking and coordinate mapping mechanism is as follows: the silicon wafer spindle encoder outputs a real-time angle signal θ(t) with a resolution of 4096 pulses per revolution. The DSP processor reads the current angle value in each control cycle (500μs) and, combined with the initial locking angle θ0 marked by the V-Notch, calculates the real-time angle range of the 3mm edge control area in the polar coordinate system according to the following formula: [θ(t)-Δθ, θ(t)+Δθ], where Δθ=arcsin(3 / R), R is the silicon wafer radius (150mm), and Δθ≈1.146° is calculated. The piezoelectric ceramic actuator array consists of 48 units evenly distributed along the circumference, each covering a 7.5° angle range. The PLC controller queries the mapping table between the angle range and the piezoelectric unit number in real time based on the current θ(t) value. Two control cycles (1 ms) in advance, it sends a pressure boosting command to the piezoelectric unit about to enter the edge control zone, linearly switching the pressure of that unit from the central reference value of 1.5 psi to the edge fine polishing setpoint of 1.0-1.2 psi. The switching ramp time is 200 μs, ensuring that the pressure switch is completed before the silicon wafer edge region enters the effective working area of ​​the polishing pad. The response timing is synchronized with the silicon wafer rotation, with a phase lag of less than ±0.5°.

[0041] The second step involves a two-stage gradient polishing process. The first stage is a rough polishing process, employing a low-pressure, high-flow-rate polishing slurry supply. During this stage, the pressure is precisely controlled at 1.8 to 2.2 psi, the slurry flow rate is controlled at 300 to 400 mL per minute, and the turntable speed is set at 90 to 110 revolutions per minute. The goal of this stage is to prioritize and quickly repair obvious macroscopic defects such as scratches, burrs, and protrusions in the edge areas, eliminating large particle defect sources. The efficiency of edge morphology correction in the rough polishing stage directly affects the processing allowance and particle control effect of the subsequent fine polishing process. The second stage is the fine polishing process, performed after the rough polishing. In this stage, the polishing pressure is reduced to 1.0 to 1.2 psi, while the turntable speed uniformity is increased to 100 to 120 revolutions per minute, and the slurry flow rate is finely adjusted to 250 to 300 mL per minute. The fine polishing stage focuses on finely smoothing and repairing microscopic damage and microcracks at the edges. This stage strictly controls the amount of material removed to avoid silicon chips and metal microparticles generated by over-polishing. The two-stage gradient polishing process effectively controls particle generation while ensuring defect repair efficiency, achieving the dual goals of edge morphology quality and particle contamination control.

[0042] The third sub-step is the constant-temperature turbulence-controlled deposition process for the polishing slurry. This process ensures that the polishing slurry maintains stable physicochemical properties throughout the polishing process. The polishing slurry temperature is constantly controlled within the range of 22±1℃. The supply temperature of the polishing slurry is adjusted in real time through a circulating cooling system to avoid viscosity changes and abnormal metal solute deposition behavior caused by temperature fluctuations. The slurry supply method adopts dynamic turbulence circulation, with a turbulence frequency set to 15 to 20 cycles per minute. The turbulence amplitude covers the entire polishing pad. This slurry supply method can effectively prevent the metal solute in the polishing slurry from stagnating and accumulating at the edge area of ​​the silicon wafer, reducing the metal solute deposition rate to less than 30% of the traditional static slurry supply method. After each wafer process is completed, the polishing table automatically performs a spray cleaning program. The cleaning solution is deionized water with a trace amount of surfactant. The spray duration is 15 seconds, and the pressure is 0.2MPa. This program removes residual particles and metal precipitates from the pad surface and prevents cross-contamination between batches. This design effectively reduces the risk of Fe source formation caused by the deposition of metal components in the polishing slurry from the process end.

[0043] Furthermore, the specific hardware structure and thermal management strategy of the dynamic turbulent circulation liquid supply are as follows: The liquid supply system adopts a ring-shaped main pipeline layout, with six liquid supply nozzles installed at equal angles around the periphery of the polishing pad. Each nozzle has a vortex generator connected in series at its front end, driven by a stepper motor. This device contains three spiral guide vanes. The stepper motor drives the vane assembly to rotate at a speed of 30 to 45 revolutions per minute, causing the sprayed polishing liquid to generate a tangential rotating turbulent flow, thereby achieving periodic shear turbulence over the entire polishing pad area. The turbulence period is achieved by the PLC controlling the forward and reverse rotation frequency of the stepper motor; the forward and reverse rotation frequency is the turbulence frequency. The polishing liquid circulates back... A plate heat exchanger is connected in series along the pipeline. The refrigerant side of the heat exchanger is adjusted in real time by a PID controller based on the temperature signal from the thermocouple installed on the main liquid supply pipe. The PID parameters are proportional gain Kp=2.5, integral time Ti=30s, and derivative time Td=5s. To solve the problem of local temperature fluctuation caused by high-frequency turbulence, a miniature constant temperature chamber with a buffer volume of 100mL is set at the rear end of each liquid supply nozzle. The outer wall of the chamber is wrapped with electric heating tape as a heat source for fine adjustment. The temperature of each chamber is controlled by an independent temperature sensor in a closed loop. When the temperature deviation exceeds ±0.5℃, local compensation is activated to ensure that the actual temperature of each liquid supply point is stable within the range of 22±1℃.

[0044] The fourth sub-step is real-time monitoring of the manufacturing process. This monitoring mechanism covers key process parameters during polishing. Monitored parameters include core indicators such as polishing turntable speed, polishing slurry flow rate, polishing pressure, and silicon wafer temperature. Each parameter is sampled 10 times per second. The monitoring system sets parameter threshold ranges; if any parameter deviates from the set threshold, a shutdown and fine-tuning mechanism is immediately triggered, with a response time of less than 500ms. This real-time monitoring mechanism ensures consistent edge polishing for each silicon wafer, avoids particle abnormalities and metal residue fluctuations caused by process variations, and provides end-to-end assurance for process stability.

[0045] In the above method, step 3 is graded chemical cleaning and precision drying. This step is designed to remove organic residues, metal contamination and edge microparticles on the surface of the polished silicon wafer. Through the synergistic combination of multi-solution graded cleaning and low-residue drying process, the problems of secondary particle adhesion and excessive metal residue are completely solved, and the metal control standards of the final cleaning machine are matched.

[0046] Specifically, step 3 comprises five consecutive cleaning and drying sub-steps. First, an HF micro-etching pretreatment is performed, involving a short immersion in a diluted HF solution. The HF solution concentration is diluted to 0.5% to 1% (volume fraction), and the immersion time is precisely controlled within the range of 20 to 30 seconds. The mechanism of the HF micro-etching process utilizes the selective etching properties of hydrofluoric acid on the silicon dioxide layer to gently remove the oxide layer on the silicon wafer surface, as well as trace metal impurities and embedded microparticles encased in the oxide layer. This treatment exposes the deep particles covered by the oxide layer, creating favorable conditions for subsequent deep cleaning steps. The micro-etching depth is controlled within the range of 3 to 5 nm to ensure that only the surface oxide layer is removed without damaging the underlying silicon substrate.

[0047] The second sub-step is O3 pure water oxidation cleaning, in which ultrapure water with an ozone concentration of 20 to 25 mg / L is circulated and sprayed for cleaning. During the cleaning process, the ultrapure water is circulated and sprayed at a flow rate of 5 to 8 L per minute for 4 minutes. The strong oxidizing effect of ozone can effectively decompose organic stains and carbonaceous impurities remaining after polishing. Its oxidation-reduction potential is 2.07 V, which is sufficient to break the carbon-carbon and carbon-hydrogen bonds in organic molecules. This oxidation cleaning step also eliminates the carrier effect of metal particles adsorbed by organic media, reduces the adhesion of metal contaminants in subsequent processes, and creates favorable conditions for the alkaline cleaning step.

[0048] The third sub-step is SC1 alkaline deep particle removal, which involves constant-temperature immersion cleaning using a standard SC1 cleaning solution. The SC1 cleaning solution is composed of ammonia, hydrogen peroxide, and ultrapure water in a specific ratio, with ammonia volume fraction of 0.5% to 2%, hydrogen peroxide volume fraction of 1% to 5%, and the remainder being ultrapure water. The cleaning temperature is constantly controlled within the range of 45 to 50°C, and the immersion time is set to 6 minutes. The principle of SC1 cleaning is to utilize the chelation effect in an alkaline environment. Through the chelation reaction between ammonia and metal ions and the oxidation effect of hydrogen peroxide, micro-silicon particles and suspended metal debris attached to the surface and edges of the silicon wafer are removed. This process has a highly efficient removal capability for micron- and submicron-sized particles, focusing on removing residual particles in edge dead zones, achieving a particle removal rate of over 99%.

[0049] The fourth sub-step is SC2 acidic metal cleaning, which uses a standard SC2 cleaning solution for deep metal removal. The SC2 cleaning solution is composed of hydrochloric acid, hydrogen peroxide, and ultrapure water in a specific ratio, with hydrochloric acid comprising 0.5% to 2% by volume, hydrogen peroxide comprising 1% to 5% by volume, and the remainder being ultrapure water. The cleaning temperature is constantly controlled within the range of 35 to 40°C, and the cleaning time is set to 5 minutes. The principle of SC2 cleaning is to utilize the strong acidity of hydrochloric acid to specifically chelate and dissolve inorganic metal residues such as metals and metal oxides. This step significantly reduces the metal content on the silicon wafer surface, ensuring that the content of key metal elements such as Fe, Na, and K meets the requirements of subsequent processes, thus achieving metal control objectives from the cleaning stage.

[0050] The fifth sub-step is a multi-stage low-residue drying process, which ensures that no secondary contamination occurs on the surface of the cleaned silicon wafer. The drying process consists of two stages: the first stage is a low-speed rotary initial spin-drying, using centrifugal spin-drying with a speed controlled at 200 to 300 revolutions per minute and a spin-drying time of 60 seconds. This stage removes most of the free water droplets on the surface. The second stage is a gradient-heating nitrogen constant-temperature purging, with the nitrogen temperature constantly controlled at 40°C and a purging flow rate of 50 to 80 L per minute. A uniform linear scanning method is used to cover the edges and the entire area, with a scanning speed of 5 mm per second. The drying strategy of low-speed rotary spin-drying combined with gradient-heating nitrogen purging effectively avoids water stains and particle adsorption problems caused by rapid drying. The entire drying process is carried out in a closed, dust-free environment with a cleanliness level better than Class 100, completely eliminating secondary contamination caused by external dust.

[0051] In the above method, step 4 is final inspection and sorting and SPC data binding. This step establishes a complete quality traceability and process monitoring system, realizes the transformation from passive detection to proactive prevention, and forms a complete closed loop for process quality monitoring.

[0052] Specifically, step 4 comprises three related sub-steps. First, a final inspection and sorting process is performed. After cleaning and drying, each wafer undergoes surface metal content testing and edge particle count detection. The metal content test is conducted using a total reflectance X-ray fluorescence spectrometer, achieving atomic-level sensitivity and quantitatively analyzing the surface concentration of metal elements such as Fe, Na, K, Ca, and Cu. Results are recorded in units of particles per square centimeter. Edge particle count detection is performed using a laser scattering particle counter, covering particles larger than 0.5 μm, with the detection area being a 3 mm ring around the edge of the silicon wafer. Detailed records of the test data for each silicon wafer are maintained, including batch information, machine number, testing time, and test values, ensuring quality traceability.

[0053] The second sub-step is the SPC data binding mechanism, which synchronously enters the inspection data of the products produced by the machine into the weekly monitoring report system. The entered data includes key data items such as metal content test results, edge particle count results, and snapshots of machine process parameters. The system automatically links to a Trend Chart curve and performs trend analysis on historical data using statistical methods, including algorithms such as moving averages and exponentially weighted moving averages, to set control limits and warning limits. When the inspection data shows a continuous upward or downward trend, the system automatically issues a warning signal, indicating a potential risk of process parameter drift. This monitoring mechanism realizes a shift from passive detection to proactive prevention, moving the quality control point forward.

[0054] Furthermore, the complete closed-loop logic of the SPC data binding mechanism, from "detection data → algorithm calculation → parameter correction → PLC instruction issuance," is as follows: After the detection data is uploaded to the SPC server via the RS-232 interface, the server's built-in statistical calculation engine first calculates the moving range and process mean for the n continuously collected samples (n≥5), and compares them with the preset Western Electric judgment rules. The judgment rules include eight discrepancy criteria, such as seven consecutive points being on the same side of the mean and seven consecutive points showing a monotonically increasing or decreasing trend. When a discrepancy criterion is triggered, the calculation engine starts the regression analysis module. This module uses three controllable parameters—polishing pressure, liquid supply flow rate, and turntable speed—as independent variables and the number of edge particles as the dependent variable. It uses a multiple linear regression model to calculate the recommended correction direction and correction step size for each parameter. The correction step size is output in three levels, ranging from ±2% to ±5% of the current parameter value. The calculated correction amount is expressed in Modbus... The TCP protocol message format is encapsulated and sent to the data register address area of ​​the polishing machine PLC controller via industrial Ethernet. The PLC's ladder diagram program scans this register every control cycle (100ms). If a new correction value is detected, the correction value is added to the current set value through its built-in PID calculation module. The analog output module converts the correction value into a 4-20mA current signal to drive the corresponding electric regulating valve and frequency converter, ultimately completing the adjustment of the hydraulic valve opening and the speed of the liquid supply pump for polishing pressure, realizing closed-loop automatic control. The correction result is synchronously recorded in the operation log of the SPC server. The specific mathematical form of the multiple linear regression model is: Y=β0+β1·P+β2·F+β3·S+ε, where Y is the number of edge particles (dependent variable), P is the polishing pressure (psi), F is the liquid supply flow rate (mL / min), S is the turntable speed (rpm), β0 is the intercept term, β1, β2, and β3 are the regression coefficients of their respective variables, and ε is the residual term. The model training set consists of the most recent 200 batches of historical production data, each batch containing the measured values ​​of the four variables mentioned above. Feature normalization employs Z-score standardization, which involves subtracting the sample mean from each variable and then dividing by the sample standard deviation to eliminate the influence of dimensions. Regression coefficients are updated online using the least squares method, with an update cycle triggered every 10 new batches of data. The step size for correction is categorized as follows: ±2% when the standardized residual |e| / σ < 0.5; ±3.5% when 0.5 ≤ |e| / σ < 1.0; and ±5% when |e| / σ ≥ 1.0. To prevent adjustment oscillations, a first-order lag filter is connected in series at the correction output, with a filter time constant Tf = 3.0s. This means the actual correction output for the current period = 0.716 × the calculated correction for this period + 0.284 × the actual correction output for the previous period (corresponding to the smoothing coefficient when Tf = 3.0s and the sampling period is 100ms).

[0055] The third sub-step involves anomaly handling and closed-loop management, classifying products based on final inspection results. Products meeting particle and metal residue standards are released into the warehouse and enter the shipping process; batches with outdated results are immediately isolated, undergo rewashing, and their corresponding process parameters are reviewed to analyze the causes of the anomalies and develop improvement measures. The review includes checking process parameter records, equipment operation logs, and operator records to identify potential systemic or random factors and develop targeted preventative measures. This closed-loop management mechanism ensures that non-conforming products do not flow into the next process and continuously improves process capabilities through ongoing improvements.

[0056] To better understand the technical solution of this invention, a specific application example is provided below. Assume a batch of 50 12-inch recycled silicon wafers are processed using the process described in this invention. First, in step 1, incoming material inspection is performed. The wafer micro-defect detector scan results show that 40 wafers are grade A, 8 are grade B, and 2 are grade C. Grade C wafers are isolated and scrapped, while grade B wafers enter a dedicated pre-treatment and repair channel. After dry impurity pre-treatment and wet fine cleaning pre-treatment, all 48 wafers to be polished enter step 2 for differentiated CMP main polishing. During polishing, the real-time monitoring system records a turntable speed of 95 revolutions per minute, a liquid supply flow rate of 350 mL per minute, and a pressure of 2.0 psi, all parameters within the set threshold range. After polishing, step 3 involves graded chemical cleaning and precise drying. The metal content test results after cleaning show a surface Fe concentration of 1.2 × 10^10 particles / cm² and an edge particle count of 8 particles / wafer, both meeting the process standard requirements. The test data was synchronously entered into the SPC weekly monitoring report. The system trend analysis showed that the quality of this batch of products was stable, the process capability index Cpk was greater than 1.33, and the process was under control.

[0057] The technical solution of this invention constructs a complete edge particle control system for regenerated silicon wafers through the coordinated operation of four major steps. Step 1 eliminates the substrate for particle generation at the source through three-level graded pretreatment; Step 2 controls the source of particle generation through zoned differentiated CMP polishing; Step 3 prevents secondary particle adhesion through multi-level chemical cleaning; and Step 4 achieves closed-loop process quality monitoring through final inspection and sorting linked with SPC data. This process solution has clear and controllable parameters, can be adapted to existing 12-inch silicon wafer regeneration production lines without complex equipment modifications, and has good prospects for industrial application.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the scope of protection of the present invention is not limited to the above embodiments. Any technical solutions obtained by modifying the above embodiments or making equivalent substitutions for certain structures based on the technical solutions of the present invention should be included within the scope of protection of the present invention.

Claims

1. A process for controlling edge particles of a 12-inch reclaimed silicon wafer and linking it with SPC monitoring, characterized in that, Includes the following steps: Step 1: Perform incoming material grading and pretreatment on recycled silicon wafers, including appearance and damage grading and sorting, dry blowing and wet cleaning. Step 2: The silicon wafer is divided into a central area and an edge control area using the V-Notch edge precision alignment system. Two-stage gradient CMP polishing is carried out. The polishing slurry is supplied with dynamic turbulent circulation and constant temperature control. Process parameters are monitored in real time throughout the process. Step 3: After polishing, the silicon wafer is subjected to HF micro-etching pretreatment, O3 pure water oxidation cleaning, SC1 alkaline cleaning, and SC2 acid cleaning in sequence, followed by low-speed spin drying combined with nitrogen purging drying. Step 4: After cleaning and drying, each silicon wafer is tested for metal content and the number of edge particles. The test data is entered into the SPC system for trend analysis, and the products are classified and processed according to the test results. The V-Notch edge precision alignment system establishes a polar coordinate system by identifying the V-shaped notch mark on the edge of the silicon wafer, drives the ring piezoelectric ceramic actuator array, and switches the pressure of the piezoelectric unit in real time according to the rotation angle of the silicon wafer, so as to realize the dynamic zoning and switching of pressure between the central area and the edge control area. The silicon wafer micro-defect detector acquires images through a linear CCD array. After noise reduction and enhancement by an FPGA, the ARM processor runs a depth inversion algorithm. This algorithm performs least-squares fitting between the grayscale ratio of the defect area under bright and dark illumination and a standard depth sample library, converting the grayscale signal into a three-dimensional depth value. After comparison with the built-in threshold, the PLC drives the shunting mechanism to complete automatic grading. The dynamic turbulent circulation liquid supply generates tangential turbulent flow by setting a swirl generator at the front end of the liquid supply nozzle, and the heat exchanger on the circulation loop, together with the independent temperature control of each liquid supply point, stabilizes the polishing liquid temperature at 22±1℃. The SPC system compares continuously collected samples using discrepancy criteria. When triggered, it performs regression analysis using controllable process parameters as independent variables to calculate the correction amount, and sends it to the PLC to be superimposed on the current set value before driving the actuator to complete closed-loop adjustment.

2. The process according to claim 1, characterized in that, In step 1, grade A scratches are less than 0.5 μm deep and have no microcrack extension, and proceed directly to the subsequent process; grade B scratches are between 0.5 μm and 2 μm deep or have local microcracks but do not penetrate the edge, and proceed to the special pretreatment repair channel; grade C scratches are more than 50 μm in size or have microcracks deeper than 2 μm and are distributed in a network pattern, and are isolated and scrapped; the dry purging uses nitrogen with a purity of not less than 99.999% and the pressure is controlled at 0.35-0.4 MPa; the pH value of the neutral cleaning solution for wet cleaning is controlled at 6.5-7.5, and the soaking time is 3-5 minutes.

3. The process according to claim 1, characterized in that, In the two-stage gradient polishing process in step 2, the pressure is controlled at 1.8-2.2 psi in the rough polishing stage, the polishing fluid flow rate is controlled at 300-400 mL / min, and the turntable speed is set at 90-110 rpm; in the fine polishing stage, the pressure is controlled at 1.0-1.2 psi, the polishing fluid flow rate is controlled at 250-300 mL / min, and the turntable speed is increased to 100-120 rpm.

4. The process according to claim 1, characterized in that, In the V-Notch edge precision alignment system, the silicon wafer spindle encoder outputs a real-time angle signal θ(t). The DSP processor calculates the real-time angle range of the edge control zone according to the formula [θ(t)-Δθ, θ(t)+Δθ], where Δθ=arcsin(3 / R) and R is the silicon wafer radius. The piezoelectric ceramic actuator array consists of multiple units evenly distributed along the circumference. The PLC controller queries the mapping relationship between the angle range and the piezoelectric unit based on the current angle value, and sends a pressure boosting command to the piezoelectric unit that is about to enter the edge control zone two control cycles in advance to realize dynamic pressure zone switching.

5. The process according to claim 1, characterized in that, The specific parameters of the dynamic turbulent circulation liquid supply are as follows: there are 6 liquid supply nozzles, which are installed at equal angles along the outer periphery of the polishing pad; a stepper motor drives the spiral guide vanes to rotate at a speed of 30-45 rpm, and the turbulence frequency is 15-20 times / min; a miniature constant temperature chamber with a buffer volume of 100mL is set at the rear end of each liquid supply nozzle, and the temperature of each chamber is controlled by an independent temperature sensor in a closed loop. Local compensation is activated when the temperature deviation exceeds ±0.5℃; a plate heat exchanger is connected in series in the circulation loop, and its PID control parameters are Kp=2.5, Ti=30s, and Td=5s.

6. The process according to claim 1, characterized in that, In the depth inversion algorithm, the grayscale ratio is defined as The standard depth sample library is fitted with a quadratic polynomial model D(R)=a·R²+b·R+c using least squares fitting. The standard depth sample library contains 9 depth levels from 0.1μm to 3.0μm, with no less than 30 samples in each level. The measurement accuracy requirements are a maximum absolute deviation ≤0.05μm and an average absolute deviation ≤0.03μm.

7. The process according to claim 1, characterized in that, In step 3, the HF micro-etching pretreatment involves immersing the patient in a diluted HF solution with a volume fraction of 0.5%-1% for 20-30 seconds, with the micro-etching depth controlled at 3-5 nm. The O3 pure water oxidation cleaning involves circulating and spraying ultrapure water with an ozone concentration of 20-25 mg / L for 4 minutes. The SC1 cleaning consists of ammonia, hydrogen peroxide, and ultrapure water in a volume ratio of 0.5-2%:1-5%:balance, at a temperature of 45-50℃ for 6 minutes. The SC2 cleaning consists of hydrochloric acid, hydrogen peroxide, and ultrapure water in a volume ratio of 0.5-2%:1-5%:balance, at a temperature of 35-40℃ for 5 minutes.

8. The process according to claim 1, characterized in that, In step 3, the low-speed rotary drying speed is 200-300 rpm, the drying time is 60 s; the nitrogen constant temperature purging temperature is 40℃, the purging flow rate is 50-80 L / min, the scanning speed is 5 mm / s, and the cleanliness of the drying environment is better than class 100.

9. The process according to claim 1, characterized in that, In step 4, the metal content test uses a total reflection X-ray fluorescence spectrometer to quantitatively analyze the surface concentration of Fe, Na, K, Ca, and Cu; the edge particle number detection uses a laser scattering particle counter, the detection range covers particles larger than 0.5 μm, and the detection area is a 3 mm ring area at the edge of the silicon wafer; the SPC discrepancy criteria include 7 consecutive points located on the same side of the mean and 7 consecutive points showing a monotonically increasing or decreasing trend.

10. The process according to claim 1, characterized in that, In the regression analysis of the SPC system, the multiple linear regression model is Y=β0+β1·P+β2·F+β3·S+ε, where Y is the number of edge particles, P is the polishing pressure, F is the liquid supply flow rate, and S is the turntable speed. The model training set is the most recent 200 batches of historical data. Feature normalization is performed using Z-score standardization, and the regression coefficients are updated online using the recursive least squares method. The correction step size is output in three levels, ranging from ±2% to ±5% of the current parameter value. The level is based on the standardized residual |e| / σ, with thresholds of 0.5 and 1.0, respectively. A first-order hysteresis filter is connected in series at the correction output terminal, with a filter time constant Tf=3.0s. The correction amount is sent to the PLC via Modbus TCP protocol, and the analog output module converts it into a 4-20mA current signal to drive the actuator.