A silicon wafer cutting whole-process dynamic quality closed-loop control method

Abstract

The application discloses a kind of silicon wafer cutting whole-process dynamic quality closed-loop control method, this method includes: obtaining the incoming property data of to-be-cut silicon rod, target silicon wafer quality index and cutting equipment state parameter, and according to the cutting depth of silicon rod from the cutting-in position to the cutting-out position, multiple cutting control sections are divided;Reference cutting trajectory and segmented process window are generated for each cutting control section;Cutting operation state data and online quality data are collected during cutting, current process state is constructed based on cutting depth coordinate, and trajectory residual characteristics are calculated;When the current cutting control section deviates from the quality stable interval, multiple candidate regulation schemes are generated, and the quality results, equipment safety results and production efficiency results of each candidate scheme in the future cutting depth interval are predicted through a virtual trial cutting prediction model;According to the optimal regulation scheme is selected according to comprehensive cost function and is issued to execute.The method can effectively improve the consistency of silicon wafer cutting quality and reduce the risk of cutting abnormality.

Description

Technical Field

[0001] This invention relates to the field of silicon wafer cutting quality control technology, specifically to a dynamic quality closed-loop control method for the entire silicon wafer cutting process. Background Technology

[0002] Silicon wafer dicing is a crucial upstream process in the manufacturing of photovoltaic cells, semiconductor devices, power devices, and integrated circuit substrates. Silicon ingots are bonded, diced, cleaned, debonded, and inspected to form silicon wafers. The quality of silicon wafer dicing directly affects the yield and stability of subsequent processes such as texturing, diffusion, coating, photolithography, etching, and packaging.

[0003] As silicon wafers evolve towards larger sizes, thinner wafers, and higher production volumes, the dicing process places higher demands on wafer thickness consistency, total thickness variation, warpage, surface marks, edge chipping, cracks, material loss, and dicing efficiency. Particularly in the dicing of ultra-thin and large-size silicon wafers, cutting wire tension, wire bowing, feed rate, cutting fluid temperature, cutting fluid flow rate, guide roller speed, and equipment vibration all significantly affect the final quality of the silicon wafer.

[0004] Current silicon wafer dicing processes typically rely on fixed process parameter formulations. Operators preset linear speed, feed rate, linear tension, slurry flow rate, slurry temperature, and guide roller speed based on silicon rod specifications, target wafer thickness, and equipment experience. While this method can meet general production needs, it lacks dynamic analysis of the dicing process, making it impossible to promptly determine whether the current dicing state has deviated from the stable processing range, or to make targeted adjustments to the dicing parameters before quality deviations increase, thus failing to guarantee consistent silicon wafer dicing quality.

[0005] Therefore, it is necessary to propose a dynamic quality closed-loop control method for the entire silicon wafer cutting process with cutting depth as a unified benchmark. This method can improve the consistency of silicon wafer cutting quality, equipment operation safety, and production efficiency by using cutting depth partitioning, reference cutting trajectory, trajectory residual identification, virtual trial cutting prediction, and comprehensive cost optimization. Summary of the Invention

[0006] To address the aforementioned issues, this invention aims to propose a dynamic quality closed-loop control method for the entire silicon wafer dicing process, using dicing depth as a unified benchmark. This method improves the quality consistency and process adaptability of the silicon wafer dicing process by dicing depth partitioning, reference dicing trajectory, trajectory residual identification, virtual trial dicing prediction, and comprehensive cost optimization.

[0007] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows:

[0008] This invention proposes a dynamic quality closed-loop control method for the entire silicon wafer dicing process, comprising the following steps:

[0009] S1. Obtain the incoming material attribute data of the silicon rod to be cut, the quality index of the target silicon wafer, and the status parameters of the cutting equipment, and divide the silicon rod to be cut into multiple cutting control sections according to the cutting depth from the infeed position to the outfeed position.

[0010] S2. Based on the incoming material attribute data, target silicon wafer quality indicators, cutting equipment status parameters, and historical qualified cutting batch data, generate a reference cutting trajectory and segmented process window for each cutting control segment.

[0011] S3. During the silicon wafer dicing process, real-time dicing operation status data and online quality data are collected. The dicing operation status data and online quality data are aligned with the dicing depth coordinates to construct the current process status. The current process status is then compared with the reference dicing trajectory of the current dicing control section to obtain trajectory residual characteristics.

[0012] S4. Determine whether the current cutting control segment deviates from the quality stability range based on the trajectory residual characteristics, and determine the type of deviation;

[0013] S5. When the current cutting control segment deviates from the quality stability range, multiple candidate control schemes are generated based on the current process status, deviation type and corresponding segment process window.

[0014] S6. Input each candidate control scheme into the virtual trial cutting prediction model to predict the quality results, equipment safety results and production efficiency results of each candidate control scheme in the future cutting depth range.

[0015] S7. Calculate the comprehensive cost function value of each candidate control scheme based on the quality results, equipment safety results, and production efficiency results. Among the candidate control schemes that meet the quality constraints, equipment safety constraints, and process boundary constraints, select the candidate control scheme with the smallest comprehensive cost function value as the optimal control scheme. Then, send the optimal control scheme to the cutting control system to correct the cutting process parameters of the current cutting control segment and the next cutting control segment.

[0016] Furthermore, in step S1, the incoming material attribute data includes silicon rod crystal orientation deviation, silicon rod diameter, silicon rod length, silicon rod warpage, end face perpendicularity, surface damage distribution, resistivity, minority carrier lifetime, oxygen content, carbon content, silicon rod bonding status, and batch identifier.

[0017] The target silicon wafer quality indicators include target wafer thickness, wafer thickness tolerance, total thickness variation tolerance, warpage tolerance, surface roughness tolerance, line mark depth tolerance, edge chipping size tolerance, crack quantity tolerance, target pass rate, and target material loss rate.

[0018] The status parameters of the cutting equipment include cutting wire type, cutting wire diameter, cumulative length of cutting wire used, guide wheel wear status, spindle load reference value, cutting fluid circulation status, cutting fluid temperature control capability, and equipment vibration reference value, etc.

[0019] The silicon rod to be cut is divided into the following sections from the entry point to the exit point: the entry stability section, the middle stable cutting section, the material stress release section, and the exit risk section.

[0020] The cutting stabilization zone is used to monitor tension and vibration changes when the cutting wire enters the silicon rod; the middle stabilization cutting zone is used to monitor wire bowing, wafer thickness trend, and spindle load stability; the material stress release zone is used to monitor cracks caused by internal stress and surface damage in the incoming material; and the cutting exit risk zone is used to monitor edge chipping risk and load changes when the cutting wire approaches the end of the silicon rod.

[0021] Furthermore, in step S2, a reference cutting trajectory and segmented process window are generated for each cutting control segment, specifically including the following steps:

[0022] S21. Select historical sample batches from historical cutting batch data that meet the similarity conditions of the current silicon rod in terms of material type, target wafer thickness, silicon rod size, cutting line specification, cutting equipment number and target quality grade.

[0023] S22. Select from the historical sample batches a historical qualified cutting batch with a pass rate higher than a preset pass rate threshold, a total thickness change lower than a preset total thickness change threshold, a chipping rate lower than a preset chipping rate threshold, and a breakage rate lower than a preset breakage rate threshold.

[0024] S23. Statistically analyze the wire bow amount, wire tension, spindle load, feed resistance, cutting fluid temperature, cutting fluid flow rate, vibration characteristics, and online sheet thickness estimation values ​​of historical qualified cutting batches according to the cutting control section to obtain the reference mean trajectory and reference fluctuation range of each cutting control section.

[0025] S24. Based on the current silicon rod's crystal orientation deviation, warpage, surface damage distribution, resistivity fluctuation, and minority carrier lifetime fluctuation, the reference mean trajectory and reference fluctuation range are corrected to generate a reference cutting trajectory adapted to the current silicon rod.

[0026] S25. Based on the reference cutting trajectory, cutting equipment status parameters, and target silicon wafer quality indicators, generate segmented process windows corresponding to each cutting control section;

[0027] The segmented process window includes the linear velocity range, feed rate range, linear tension range, cutting fluid flow rate range, cutting fluid temperature range, guide wheel speed range, feed acceleration range, single adjustment step size, and parameter adjustment priority.

[0028] Furthermore, in step S3, the cutting operation status data includes linear speed, feed speed, cutting line tension, wire bowing amount, spindle current, spindle load, feed resistance, guide wheel speed, guide wheel speed fluctuation, cutting fluid temperature, cutting fluid flow rate, cutting fluid pressure, and vibration signal of the cutting area;

[0029] The online quality data includes online sheet thickness estimates, thickness variation trends, online edge defect characteristics, online surface line mark characteristics, suspected crack characteristics, and warping trends;

[0030] The current process state is represented as follows:

[0031] in, Indicates the current process status; This represents a subset of current process parameters, including linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration. It represents a subset of equipment operating status, including spindle current, spindle load, feed resistance, wire bowing amount, guide wheel speed fluctuation, vibration characteristics, and cumulative used length of the cutting wire; It represents a subset of online quality status, including online sheet thickness estimates, thickness variation trends, edge defect characteristics, surface line mark characteristics, suspected crack characteristics, and warping trends; It represents a subset of historical short-term states, including the rate of change of the pantograph, tension fluctuation rate, load change rate, temperature change rate, vibration energy change rate, and online quality index change rate within a preset sliding time window; This represents a subset of the cutting position status, including the current cutting depth, the current cutting control segment number, the current cutting time, and the estimated output sequence number.

[0032] Furthermore, in step S3, the trajectory residual features include instantaneous residual, residual change rate, segment cumulative residual, and trajectory deviation index;

[0033] For the i-th type of trajectory variable in the r-th cut control segment, the instantaneous residual is:

[0034]

[0035] in, This represents the instantaneous residual of the i-th type of trajectory variable at the cutting depth d in the r-th cutting control segment; This represents the actual trajectory value of the i-th type of trajectory variable at the cutting depth d; This represents the reference trajectory value of the i-th type of trajectory variable at the cutting depth d;

[0036] The rate of change of residuals is: ;in, Represents the rate of change of residuals. Indicates the cutting depth step size;

[0037] The cumulative residual of the section is: ;in, Indicates the cumulative residual of the section. This represents the initial cutting depth of the r-th cutting control segment. This represents the final cutting depth of the r-th cutting control segment;

[0038] The trajectory deviation index is: ;in, represents the trajectory deviation index of the r-th control segment; M represents the number of trajectory variables involved in the calculation; , and These represent the instantaneous residual weight, residual rate of change weight, and segment cumulative residual weight for the i-th type of trajectory variable, respectively. , and These represent the corresponding instantaneous residual threshold, residual change rate threshold, and segment cumulative residual threshold, respectively.

[0039] Furthermore, in step S4, when When the deviation exceeds the corresponding cutting control segment's deviation threshold, the current cutting control segment is determined to have deviated from the quality stability range.

[0040] The deviation types include bow-increase deviation, tension instability deviation, load abrupt change deviation, insufficient cooling deviation, vibration-enhanced deviation, sheet thickness drift deviation, and defect aggregation deviation;

[0041] When the actual pantograph trajectory is higher than the reference pantograph trajectory and the pantograph residual change rate is positive, it is determined to be a pantograph enlarged deviation.

[0042] When the instantaneous residual and the rate of change of the residual of the actual tension trajectory relative to the reference tension trajectory both exceed the threshold, it is determined to be a tension instability type deviation.

[0043] When both the actual spindle load trajectory and the feed resistance trajectory are higher than the reference load trajectory and continue to exceed the preset time, it is determined to be a load mutation type deviation.

[0044] When the actual cutting fluid temperature trajectory is higher than the reference temperature trajectory and the actual cutting fluid flow trajectory is lower than the reference flow trajectory, it is determined to be a deviation due to insufficient cooling.

[0045] When the frequency band energy of the actual vibration signal is higher than the frequency band energy threshold of the reference vibration trajectory, it is determined to be a vibration-enhanced deviation.

[0046] When the online sheet thickness estimation trajectory deviates continuously in one direction from the reference sheet thickness trend trajectory, it is identified as a sheet thickness drift type deviation.

[0047] When online edge defect features, online surface line mark features, and suspected crack features continuously exceed the corresponding defect threshold within the same cutting control section, it is determined to be a defect clustering deviation.

[0048] Furthermore, in step S5, multiple candidate regulation schemes are generated, specifically including the following steps:

[0049] S51. Determine the allowable adjustment range and single adjustment step size of each adjustable process parameter based on the segmented process window of the current cutting control section.

[0050] S52. Determine the parameter adjustment direction and parameter adjustment priority of the candidate control scheme based on the deviation type;

[0051] S53. Generate multiple candidate control schemes that meet the single adjustment range limit based on the current process parameters;

[0052] S54. Eliminate candidate control schemes that exceed the current cutting control segment segmentation process window;

[0053] Among them, when the deviation of the line bow increases, the candidate control schemes include a combination of reducing the feed rate, increasing the line tension, and limiting the increase of the line speed;

[0054] Under tension instability-type deviation, candidate control schemes include a combination of reducing guide wheel speed fluctuations, reducing feed acceleration, and adjusting the line tension to the tension stability window;

[0055] Under load abrupt deviation, candidate control schemes include a combination of adjusting the feed rate, increasing the cutting fluid flow rate, and increasing the cutting fluid pressure;

[0056] In cases of insufficient cooling deviation, candidate control schemes include combinations of increasing the fluid flow rate, decreasing the fluid temperature, and limiting the upper limit of the feed rate;

[0057] Under vibration-enhanced deviation, candidate control schemes include a combination of adjusting the matching relationship between linear velocity and guide wheel speed, reducing feed acceleration, and reducing linear velocity fluctuation.

[0058] Under the condition of sheet thickness drift deviation, candidate control schemes include a combination of simultaneous correction of feed rate, line tension and line bow target value.

[0059] Furthermore, in step S6, the virtual trial cutting prediction model includes a quality prediction sub-model, an equipment safety prediction sub-model, and an efficiency prediction sub-model;

[0060] For the u-th candidate control scheme The output of the virtual trial-cut prediction model is:

[0061]

[0062] in, This represents the set of predicted results for the u-th candidate control scheme in the future control time domain; This represents a virtual trial-cut prediction model; Indicates the current process status; This represents the u-th candidate control scheme; This represents the set of constraints corresponding to the r-th cutting control segment;

[0063] The quality prediction sub-model is used to predict sheet thickness deviation, total thickness change, line mark severity, edge chipping probability, crack probability, and warping trend in the future control time domain.

[0064] The equipment safety prediction sub-model is used to predict the risks of wire breakage, spindle overload, tension instability, guide wheel abnormality, and vibration abnormality in the future control time domain.

[0065] The efficiency prediction sub-model is used to predict the cutting time, material loss rate, cutting line consumption, and energy consumption per unit silicon wafer in the future control time domain.

[0066] The future cutting depth range is a continuous cutting depth range from the current cutting depth to the end depth of the next cutting control segment.

[0067] Furthermore, in step S7, the cutting process parameters include linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel rotation speed, and feed acceleration.

[0068] The comprehensive cost function value is calculated as follows:

[0069]

[0070] in, Let represent the comprehensive cost function value of the u-th candidate control scheme; This represents the quality cost of the u-th candidate control scheme; This represents the equipment safety cost of the u-th candidate control scheme; This represents the production efficiency cost of the u-th candidate control scheme; This represents the cost of parameter changes for the u-th candidate control scheme; , , and These represent the weights of quality cost, equipment safety cost, production efficiency cost, and parameter change cost, respectively.

[0071] The quality cost The safety cost of the equipment is calculated based on the predicted thickness deviation, predicted total thickness change, predicted edge chipping probability, predicted crack probability, predicted line mark severity, and predicted warping trend. The production efficiency cost is calculated based on the predicted risks of wire breakage, spindle overload, tension instability, guide wheel abnormalities, and vibration abnormalities. The cost of changing these parameters is calculated based on predicted cutting time, predicted material loss rate, predicted cutting line consumption, and predicted energy consumption per unit silicon wafer. Calculated based on the adjustment range of the candidate control scheme relative to the current process parameters;

[0072] When the selected optimal control schemes have opposite adjustment directions in multiple consecutive control cycles, the weight of parameter change cost is increased. This extends the future cutting depth range to suppress process parameter oscillations;

[0073] When none of the candidate control schemes meet the equipment safety constraints, the commands to reduce cutting speed, pause cutting, and perform manual review are executed in sequence.

[0074] Furthermore, the dynamic quality closed-loop control method for the entire silicon wafer cutting process also includes: S8, after cutting, acquiring the final inspection quality data of the silicon wafer, mapping the final inspection quality data to the corresponding cutting control section according to the cutting depth, and updating the reference cutting trajectory, segmented process window, virtual trial cutting prediction model and candidate control scheme generation rules according to the mapping results.

[0075] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0076] 1) This invention uses the cutting depth of the silicon rod from the entry position to the exit position as a unified benchmark, dividing the silicon rod to be cut into multiple cutting control segments, and generating a reference cutting trajectory and segmented process window for each cutting control segment. This allows the entry stage, the mid-section stabilization cutting stage, the material stress release stage, and the exit stage to correspond to different process monitoring priorities and process adjustment boundaries, avoiding the problem of existing silicon wafer cutting processes using relatively fixed parameters that are difficult to adapt to different cutting depth states, thereby improving the consistency of silicon wafer cutting quality.

[0077] 2) This invention aligns real-time collected cutting operation status data and online quality data according to the cutting depth coordinates, and compares the current process status with the reference cutting trajectory of the corresponding cutting control section. It calculates instantaneous residuals, residual change rate, cumulative residuals of the section, and trajectory deviation index, thereby identifying process deviation types such as increased wire bow, tension instability, sudden load changes, insufficient cooling, increased vibration, and sheet thickness drift. Compared to methods that rely solely on fixed threshold alarms or final product inspection to detect problems, this solution can generate candidate control schemes and intervene in control before quality deviations expand, reducing batch quality fluctuations and the probability of defective products.

[0078] 3) Before actual control, this invention uses a virtual trial-cutting prediction model to predict the quality, equipment safety, and production efficiency results of each candidate control scheme within the future cutting depth range. Then, based on a comprehensive cost function, the optimal control scheme is selected, and the linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration of the current and next cutting control segments are jointly corrected. After cutting, the final inspection quality data is mapped to the corresponding cutting control segment according to the cutting depth, and the reference cutting trajectory, segmented process window, virtual trial-cutting prediction model, and candidate control scheme generation rules are updated accordingly. This ensures that the final inspection data is not only used for qualification judgment but can also be used to reverse-optimize the cutting control strategy for subsequent batches. Attached Figure Description

[0079] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0080] Figure 1 This is a simplified flowchart of the dynamic quality closed-loop control method for the entire silicon wafer cutting process according to the present invention.

[0081] Figure 2 This is a schematic diagram of the trajectory residual analysis process;

[0082] Figure 3 This is a schematic diagram of the virtual trial cutting prediction and comprehensive cost optimization process. Detailed Implementation

[0083] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] This embodiment provides a dynamic quality closed-loop control method for the entire silicon wafer dicing process. This method is applied to a silicon rod multi-wire dicing equipment, which includes a dicing wire drive unit, a silicon rod feeding unit, a tension control unit, a guide wheel drive unit, a cutting fluid circulation unit, a cutting fluid temperature control unit, a spindle load detection unit, a wire bow detection unit, a vibration detection unit, an online wafer thickness estimation unit, an online defect detection unit, and a dicing control system.

[0085] In this embodiment, the system uses the cutting depth of the silicon rod from the entry position to the exit position as a unified spatial reference to uniformly associate the cutting operation status data, online quality data, reference cutting trajectory, control scheme and final inspection quality data.

[0086] See attached document Figure 1 As shown in this embodiment, the dynamic quality closed-loop control method for the entire silicon wafer dicing process includes:

[0087] S1. Obtain incoming material attribute data, target silicon wafer quality indicators, and cutting equipment status parameters, and divide the cutting control zone; the specific steps are as follows:

[0088] Before cutting begins, the incoming material attribute data of the silicon rod to be cut is first obtained. The incoming material attribute data includes silicon rod crystal orientation deviation, silicon rod diameter, silicon rod length, silicon rod warpage, end face perpendicularity, surface damage distribution, resistivity, minority carrier lifetime, oxygen content, carbon content, silicon rod bonding status, and batch identification. The silicon rod bonding status includes the thickness of the bonding layer between the silicon rod and the carrier plate, bonding flatness, bonding position deviation, etc.

[0089] Among them, the crystal orientation deviation of silicon rods is obtained through crystal orientation detection equipment; the diameter, length, warpage and end face perpendicularity of silicon rods are obtained through size detection equipment; the surface damage distribution is obtained through image detection equipment; the resistivity, minority carrier lifetime, oxygen content and carbon content are obtained through material performance detection equipment; and the bonding status of silicon rods is obtained through bonding detection equipment or bonding process records.

[0090] Simultaneously, the system acquires the target silicon wafer quality indicators, including target wafer thickness, allowable thickness deviation, allowable total thickness variation, allowable warpage, allowable surface roughness, allowable line mark depth, allowable edge chipping size, allowable number of cracks, target pass rate, and target material loss rate.

[0091] The system also acquires the status parameters of the cutting equipment; the status parameters of the cutting equipment include the cutting wire model, cutting wire diameter, cumulative length of cutting wire used, guide wheel wear status, spindle load reference value, cutting fluid circulation status, cutting fluid temperature control capability, equipment vibration reference value, and historical wire breakage records.

[0092] Subsequently, the cutting depth coordinate d is established by taking the position where the cutting line just touches the silicon rod as the starting point of the cutting depth and the position where the cutting line completes the current cutting of the silicon rod as the ending point of the cutting depth. The system divides the silicon rod to be cut into the following sections according to the cutting depth from the entry position to the exit position: the entry stable section, the middle stable cutting section, the material stress release section, and the exit risk section.

[0093] In one specific embodiment, let the total cutting depth of the silicon rod be D. all Then 0 to D all The cutting depth is divided into four continuous intervals. The cutting depth interval [d0, d1] corresponds to the stable entry section, the cutting depth interval [d1, d2] corresponds to the stable middle section, the cutting depth interval [d2, d3] corresponds to the material stress release section, and the cutting depth interval [d3, d4] corresponds to the exit section, where d0 is the entry position and d4 is the exit position.

[0094] The infeed stabilization zone is primarily used to monitor sudden changes in tension and vibration as the cutting wire enters the silicon rod. The mid-section stabilization zone is primarily used to monitor wire bowing, wafer thickness trends, and spindle load stability. The material stress release zone is primarily used to monitor cracks caused by internal stress and surface damage in the incoming material. The exit risk zone is primarily used to monitor edge chipping risk and sudden load changes as the cutting wire approaches the end of the silicon rod.

[0095] S2. Generate the reference cutting trajectory and segmented process window for each cutting control section; the specific steps are as follows:

[0096] Based on incoming material attribute data, target silicon wafer quality indicators, cutting equipment status parameters, and historical qualified cutting batch data, a reference cutting trajectory and segmented process window are generated for each cutting control section.

[0097] First, historical sample batches that meet the similarity criteria of the current silicon rod in terms of material type, target wafer thickness, silicon rod size, cutting line specifications, cutting equipment number, and target quality grade are selected from the historical cutting batch database.

[0098] In one specific implementation, the system calculates the similarity between the current silicon ingot and the h-th historical dicing batch:

[0099]

[0100] in, This indicates the similarity between the current silicon rod and the h-th historical dicing batch; Indicates the similarity of incoming material attributes; Indicates the similarity of the target silicon wafer quality indicators; Indicates the similarity of the cutting equipment states; , and These represent the corresponding weights. .

[0101] The similarity of incoming material attributes, quality indicators, and cutting equipment status is obtained as follows: Normalized differences are compared between the corresponding parameters of the current cutting task and historical cutting batches in terms of incoming material attributes, target silicon wafer quality indicators, and cutting equipment status. Then, a weighted fusion is performed based on the influence weight of each parameter on cutting quality to obtain the similarity of incoming material attributes, quality indicators, and cutting equipment status. The comprehensive similarity is then calculated by weighting the three types of similarity. Select similarity Historical cut batches that exceed a preset similarity threshold are considered as historical sample batches.

[0102] Then, select historical qualified cutting batches from the historical sample batches, where the pass rate is higher than the preset pass rate threshold, the total thickness change is lower than the preset total thickness change threshold, the edge breakage rate is lower than the preset edge breakage rate threshold, and the wire breakage rate is lower than the preset wire breakage rate threshold.

[0103] Based on the cutting control sections, the linear bowing amount, linear tension, spindle load, feed resistance, cutting fluid temperature, cutting fluid flow rate, vibration characteristics, and online sheet thickness estimation values ​​of historical qualified cutting batches are statistically analyzed to obtain the reference mean trajectory and reference fluctuation range for each cutting control section.

[0104] Taking the bow length as an example, the reference bow trajectory for the r-th cutting control segment is represented as:

[0105]

[0106] in, This represents the reference line trajectory of the r-th cutting control segment at cutting depth d; This indicates the number of historical qualified cutting batches included in the statistics; This represents the amount of wire bowing at the cutting depth d in the r-th cutting control section of the h-th historical qualified cutting batch.

[0107] The fluctuation range of the reference line trajectory is expressed as follows:

[0108]

[0109] in, Indicates the lower boundary of the reference line arc; Indicates the upper boundary of the reference line bow; Represents boundary coefficients ( The value is generally taken as 1.0 to 3.0. This represents the standard deviation of the line bow at cutting depth d for historically qualified cutting batches.

[0110] Similarly, the system generates the reference tension trajectory, reference spindle load trajectory, reference feed resistance trajectory, reference cutting fluid temperature trajectory, reference cutting fluid flow trajectory, reference vibration trajectory, and reference online sheet thickness estimation trajectory for the r-th cutting control section.

[0111] Next, based on the current silicon rod's crystal orientation deviation, warpage, surface damage distribution, resistivity fluctuation, and minority carrier lifetime fluctuation, the reference mean trajectory and reference fluctuation range are corrected. For example, when the current silicon rod warpage exceeds a preset warpage threshold, the reference fluctuation range for wire bowing is tightened; when surface damage is concentrated in the material stress release zone, the deviation threshold of the crack-related trajectory in that zone is tightened; when the cumulative length of the cutting wire is large, the upper limit of the wire tension is reduced and the weight related to the risk of wire breakage is increased.

[0112] Finally, based on the reference cutting trajectory, cutting equipment status parameters, and target silicon wafer quality indicators, segmented process windows are generated for each cutting control section. The segmented process windows include the linear speed range, feed speed range, linear tension range, cutting fluid flow rate range, cutting fluid temperature range, guide wheel speed range, feed acceleration range, single adjustment step size, and parameter adjustment priority.

[0113] In one implementation, the parameter adjustment priority is determined based on the current cutting control section, deviation type, sensitivity of each process parameter to the target deviation type, and equipment safety constraints. First, the main control target is determined based on the current cutting control section. Then, the main related process parameters are determined based on the deviation type. Finally, the adjustment priorities of linear speed, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration are determined by combining the impact of each process parameter on quality improvement and equipment safety risks in historical control records.

[0114] Specifically, when the deviation type is a large-scale deviation of the wire bow, the feed rate and wire tension are set as high-priority parameters, and the wire speed is set as a limiting adjustment parameter; when the deviation type is a tension instability type deviation, the wire tension, guide wheel speed, and feed acceleration are set as high-priority parameters; when the deviation type is a sudden load change type deviation, the feed rate, cutting fluid flow rate, and cutting fluid pressure are set as high-priority parameters; when the deviation type is a insufficient cooling type deviation, the cutting fluid flow rate, cutting fluid temperature, and feed rate are set as high-priority parameters; when the deviation type is a vibration-enhanced type deviation, the wire speed, guide wheel speed, and feed acceleration are set as high-priority parameters; when the deviation type is a sheet thickness drift type deviation, the feed rate, wire tension, and wire bow target value are set as high-priority parameters. Under the premise of meeting equipment safety constraints and process boundary constraints, candidate control schemes with high-priority parameters are generated first, and low-priority parameters are used as auxiliary adjustment variables.

[0115] S3. Real-time acquisition of cutting operation status data and online quality data, construction of the current process status, and comparison of the current process status with the reference cutting trajectory of the current cutting control section to obtain trajectory residual characteristics; the specific steps are as follows:

[0116] During the silicon wafer dicing process, the system collects dicing operation status data and online quality data in real time. The dicing operation status data includes linear speed, feed speed, dicing line tension, wire bowing amount, spindle current, spindle load, feed resistance, guide wheel speed, guide wheel speed fluctuation, dicing fluid temperature, dicing fluid flow rate, dicing fluid pressure, and vibration signal of the dicing area. The online quality data includes online wafer thickness estimation, thickness variation trend, online edge defect characteristics, online surface line mark characteristics, suspected crack characteristics, and warping trend.

[0117] Because different sensors and detection devices have different sampling frequencies, the cutting depth coordinate d is used as a unified benchmark to align the cutting operation status data and online quality data. For high-frequency vibration signals, the root mean square value, peak factor, and frequency band energy are extracted within each cutting depth step. For online sheet thickness estimates, online edge defect features, online surface line mark features, and suspected crack features, mapping is performed based on the cutting depth coordinates corresponding to the acquisition time.

[0118] After data alignment, the system constructs the current process state: ;

[0119] in, Indicates the current process status; This represents a subset of current process parameters, including linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration. It represents a subset of equipment operating status, including spindle current, spindle load, feed resistance, wire bowing amount, guide wheel speed fluctuation, vibration characteristics, and cumulative used length of the cutting wire; It represents a subset of online quality status, including online sheet thickness estimates, thickness variation trends, edge defect characteristics, surface line mark characteristics, suspected crack characteristics, and warping trends; It represents a subset of historical short-term states, including the rate of change of the pantograph, tension fluctuation rate, load change rate, temperature change rate, vibration energy change rate, and online quality index change rate within a preset sliding time window; This represents a subset of the cutting position status, including the current cutting depth, the current cutting control segment number, the current cutting time, and the estimated output sequence number.

[0120] In one specific implementation, the current subset of process parameters is represented as follows:

[0121]

[0122] in, Indicates linear velocity; Indicates the feed rate; Indicates the tension of the cutting line; Indicates the flow rate of the cutting fluid; Indicates the temperature of the cutting fluid; Indicates the rotational speed of the guide wheel; This indicates the feed acceleration.

[0123] The subset of equipment operating status is represented as follows:

[0124] in, Indicates the spindle current; Indicates spindle load; Indicates feed resistance; Indicates the bow length; This indicates fluctuations in the guide wheel speed; Indicates vibration characteristics; This indicates the cumulative length of the cutting line used.

[0125] Next, the current process state is compared with the reference cutting trajectory of the current cutting control section to obtain trajectory residual characteristics; among which, trajectory residual characteristics include instantaneous residual, residual change rate, section cumulative residual and trajectory deviation index.

[0126] For the i-th type of trajectory variable in the r-th cut control segment, the instantaneous residual is:

[0127]

[0128] in, Let be the instantaneous residual of the i-th type of trajectory variable in the r-th cutting control segment at the cutting depth d; Let be the actual trajectory value of the i-th type of trajectory variable at the cutting depth d; is the reference trajectory value of the i-th type of trajectory variable at the cutting depth d.

[0129] The rate of change of residuals is: ;in, Represents the rate of change of residuals. This indicates the step size of the cutting depth.

[0130] The cumulative residual of the section is: ;in, Indicates the cumulative residual of the section. This represents the initial cutting depth of the r-th cutting control segment. This represents the final cutting depth of the r-th cutting control segment.

[0131] The trajectory deviation index is: ;in, represents the trajectory deviation index of the r-th control segment; M represents the number of trajectory variables involved in the calculation; , and These represent the instantaneous residual weight, residual rate of change weight, and segment cumulative residual weight for the i-th type of trajectory variable, respectively. , and These represent the corresponding instantaneous residual threshold, residual change rate threshold, and segment cumulative residual threshold, respectively.

[0132] S4. Determine whether the current cutting control segment deviates from the quality stability range based on the trajectory residual characteristics, and determine the type of deviation; the specific steps are as follows:

[0133] When the trajectory deviates from the index When the deviation exceeds the corresponding cutting control segment's deviation threshold, the system determines that the current cutting control segment deviates from the quality stability range and enters the deviation type identification process, as shown in the attached diagram. Figure 2 As shown.

[0134] In this embodiment, the deviation types include bow-increase deviation, tension instability deviation, load abrupt change deviation, insufficient cooling deviation, vibration enhancement deviation, sheet thickness drift deviation, and defect aggregation deviation.

[0135] When the actual bow trajectory is higher than the reference bow trajectory and the bow residual change rate is positive, it is identified as a bow increase deviation. This deviation indicates that the bow amount has a continuous increasing trend, which may lead to an increase in sheet thickness deviation and total thickness variation.

[0136] When both the instantaneous residual and the rate of change of the residual relative to the reference tension trajectory exceed the threshold, it is identified as a tension instability-type deviation. This deviation indicates that the tension control state of the cutting line is unstable, which may cause wire breakage, wire marks, or cracks.

[0137] When both the actual spindle load trajectory and the feed resistance trajectory are higher than the reference load trajectory and remain higher than the preset time, it is identified as a sudden load deviation. This deviation indicates an abnormal increase in cutting resistance, which may lead to edge chipping, cracking, or spindle overload.

[0138] When the actual cutting fluid temperature trajectory is higher than the reference temperature trajectory and the actual cutting fluid flow rate trajectory is lower than the reference flow rate trajectory, it is identified as a deviation due to insufficient cooling. This deviation indicates insufficient cooling and chip removal capacity, which may cause surface marks, thermal stress cracks, and warping.

[0139] When the frequency band energy of the actual vibration signal is higher than the frequency band energy threshold of the reference vibration trajectory, it is identified as a vibration-enhanced deviation. This deviation indicates an enhanced vibration state of the cutting system, which may affect the line marks, sheet thickness consistency, and edge integrity.

[0140] When the online sheet thickness estimation trajectory deviates continuously in one direction from the reference sheet thickness trend trajectory, it is identified as a sheet thickness drift type deviation. This deviation indicates that the sheet thickness has a trend shift, which may cause batch thickness inconsistencies.

[0141] When online edge defect features, online surface line mark features, and suspected crack features continuously exceed their corresponding defect thresholds within the same cutting control section, it is identified as a defect clustering deviation. This deviation indicates that quality degradation occurs concentratedly within that cutting depth section.

[0142] S5. Generate candidate control schemes based on deviation type and segmented process window; the specific steps are as follows:

[0143] See attached document Figure 3 As shown, when the current cutting control segment deviates from the quality stability range, multiple candidate control schemes are generated based on the current process status, deviation type, and corresponding segment process window.

[0144] First, read the segmented process window of the current cutting control section and determine the allowable adjustment range and single adjustment step size of the linear speed, feed speed, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed and feed acceleration.

[0145] Then, the parameter adjustment direction and parameter adjustment priority of the candidate control scheme are determined according to the deviation type.

[0146] When the bow deviates significantly, candidate control schemes include a combination of reducing feed rate, increasing wire tension, and limiting the increase in wire speed. Reducing feed rate decreases the stress on the cutting wire and the amount of bowing, increasing wire tension enhances the wire's ability to maintain straightness, and limiting the increase in wire speed prevents increased vibration caused by changes in wire speed.

[0147] Under tension instability-type deviation, candidate control schemes include a combination of reducing guide wheel speed fluctuations, reducing feed acceleration, and adjusting the wire tension to the tension stability window. Reducing guide wheel speed fluctuations is used to decrease tension disturbances, reducing feed acceleration is used to decrease sudden changes in cutting load, and adjusting the wire tension to the tension stability window is used to restore the stable operating state of the cutting wire.

[0148] Under abrupt load deviations, candidate control measures include a combination of reducing feed rate, increasing fluid flow rate, and increasing fluid pressure. Reducing feed rate lowers instantaneous cutting resistance, while increasing fluid flow rate and pressure improves cooling and chip removal.

[0149] In cases of insufficient cooling deviation, candidate control measures include combinations of increasing fluid flow rate, decreasing fluid temperature, and limiting the upper limit of feed rate. Increasing fluid flow rate enhances cooling and chip removal, decreasing fluid temperature reduces heat accumulation, and limiting the upper limit of feed rate reduces the amount of cutting heat generated per unit time.

[0150] Under vibration-enhanced deviation, candidate control schemes include a combination of adjusting the matching relationship between linear velocity and guide wheel speed, reducing feed acceleration, and reducing linear velocity fluctuations. This combination of adjustments is used to reduce vibration excitation and operational instability of the cutting system.

[0151] In cases of sheet thickness drift-type deviation, candidate control schemes include a combination of adjustments to simultaneously correct feed rate, wire tension, and wire bow target values. This combination is used to simultaneously control the material removal rate, cutting wire stress state, and wire bow state, thereby suppressing sheet thickness trend deviations.

[0152] Subsequently, multiple candidate control schemes that meet the single adjustment range limit are generated based on the current process parameters, and candidate control schemes that exceed the segmented process window of the current cutting control zone are eliminated. After screening, the retained candidate control schemes enter the virtual trial cutting prediction process.

[0153] S6. Predict the results of candidate control schemes using a virtual trial-and-cut prediction model; the specific steps are as follows:

[0154] Each candidate control scheme is input into the virtual trial cutting prediction model to predict the quality results, equipment safety results, and production efficiency results of each candidate control scheme within the future cutting depth range.

[0155] The virtual trial cutting prediction model includes a quality prediction sub-model, an equipment safety prediction sub-model, and an efficiency prediction sub-model.

[0156] For the u-th candidate control scheme The output of the virtual trial-cut prediction model is:

[0157]

[0158] in, This represents the set of predicted results for the u-th candidate control scheme within the future cutting depth interval; This represents a virtual trial-cut prediction model; Indicates the current process status; This represents the u-th candidate control scheme; This represents the set of constraints corresponding to the r-th cutting control segment, which includes quality constraints, equipment safety constraints, process boundary constraints, and production efficiency constraints within the cutting control segment. It is used to limit the feasible range of candidate control schemes in the virtual trial cutting prediction and optimal control scheme selection process.

[0159] The quality prediction sub-model is used to predict the thickness deviation, total thickness change, line mark severity, edge chipping probability, crack probability, and warping trend within the future cutting depth range.

[0160] The equipment safety prediction sub-model is used to predict the risks of wire breakage, spindle overload, tension instability, guide wheel abnormality, and vibration abnormality within the future cutting depth range.

[0161] The efficiency prediction sub-model is used to predict the cutting time, material loss rate, cutting line consumption, and energy consumption per unit silicon wafer within the future cutting depth range.

[0162] The future cutting depth range is defined as a continuous cutting depth range from the current cutting depth to the end depth of the next cutting control segment. This setting allows the system to predict not only the impact of candidate control schemes on the current cutting control segment but also their continued impact on the next cutting control segment, thereby preventing localized adjustments from causing deterioration in the quality or equipment condition of subsequent segments.

[0163] In one implementation, the quality prediction sub-model, equipment safety prediction sub-model, and efficiency prediction sub-model are all trained based on historical cutting batch data. Specifically, the incoming material attribute data, cutting control section number, reference cutting trajectory, actual cutting trajectory, trajectory residual characteristics, process parameter control records, and final inspection quality data from historical cutting batches are correlated to form model training samples. The current process state and candidate control schemes are used as model inputs, and the quality results, equipment safety results, and production efficiency results generated within subsequent cutting depth intervals are used as model output labels.

[0164] The quality prediction sub-model takes the current process status, candidate control schemes and corresponding cutting control section constraints as inputs, and outputs the sheet thickness deviation, total thickness change, line mark degree, edge breakage probability, crack probability and warping trend in the future cutting depth range. It is trained by historical final inspection quality data and online quality data.

[0165] The equipment safety prediction sub-model takes the current process status, candidate control schemes, and constraints of the corresponding cutting control section as inputs, and outputs the risks of wire breakage, spindle overload, tension instability, guide wheel abnormality, and vibration abnormality within the future cutting depth range. It is trained by historical equipment operation alarm records, wire breakage records, spindle load records, tension fluctuation records, and vibration abnormality records.

[0166] The efficiency prediction sub-model takes the current process state, candidate control schemes, and constraints of the corresponding cutting control section as inputs, and takes the cutting time, material loss rate, cutting line consumption, and unit silicon wafer energy consumption within the future cutting depth range as outputs. It is trained by historical production cycle data, silicon rod cutting completion time, material loss records, cutting line usage records, and energy consumption records.

[0167] In terms of model form, the quality prediction sub-model, equipment safety prediction sub-model, and efficiency prediction sub-model can adopt regression models, classification models, gradient boosting tree models, random forest models, neural network models, or time series prediction models. During the model update process, the system iteratively corrects the parameters of the corresponding sub-models based on the deviation between the actual cut response data and the final inspection quality data and the prediction results.

[0168] S7. Calculate the comprehensive cost function value and select the optimal control scheme, then execute the optimal control scheme and jointly correct the cutting process parameters; the specific steps are as follows:

[0169] Based on the quality results, equipment safety results, and production efficiency results output by the virtual trial cutting prediction model, calculate the comprehensive cost function value for each candidate control scheme:

[0170]

[0171] in, Let represent the comprehensive cost function value of the u-th candidate control scheme; This represents the quality cost of the u-th candidate control scheme; This represents the equipment safety cost of the u-th candidate control scheme; This represents the production efficiency cost of the u-th candidate control scheme; This represents the cost of parameter changes for the u-th candidate control scheme; , , and These represent the weights of quality cost, equipment safety cost, production efficiency cost, and parameter change cost, respectively.

[0172] In one specific implementation, quality cost Calculations are based on predicted thickness deviation, predicted total thickness variation, predicted edge chipping probability, predicted crack probability, predicted line mark severity, and predicted warping trend: ;

[0173] in, Indicates the predicted slice thickness deviation; Indicates the thickness deviation threshold; This indicates the predicted total thickness change; Indicates the threshold for total thickness variation; This indicates the predicted probability of edge collapse. This represents the threshold for the probability of edge collapse. This indicates the predicted probability of crack formation. This represents the crack probability threshold. Indicates the predicted degree of line marks; Indicates the threshold for the degree of line marks; This indicates a predicted warping trend; Indicates the warping trend threshold; to This indicates the weight of the corresponding quality indicator.

[0174] Equipment security costs Calculations are made based on the predicted risks of wire breakage, spindle overload, tension instability, guide wheel abnormalities, and vibration abnormalities: ;

[0175] in, To predict the risk of line breakage; To predict the risk of spindle overload; This indicates a predicted risk of tension instability. This indicates the predicted risk of abnormality in the guide wheel; This indicates a predicted risk of abnormal vibration; the corresponding subscript is... The parameter represents the corresponding threshold; to This indicates the corresponding weight.

[0176] Productivity Cost Calculated based on predicted cutting time, predicted material loss rate, predicted cutting line consumption, and predicted energy consumption per unit silicon wafer: ;

[0177] in, Indicates the predicted cutting time; Indicates the predicted material loss rate; This indicates the predicted consumption of the cutting line; This indicates the predicted energy consumption per unit of silicon wafer; the parameter with the subscript th represents the corresponding threshold. to This indicates the corresponding weight.

[0178] Cost of parameter change Calculations are based on the adjustment range of the candidate control scheme relative to the current process parameters:

[0179]

[0180] Where K represents the number of process parameters involved in the control; This represents the adjustment amount of the j-th process parameter in the u-th candidate control scheme; This represents the maximum allowable single adjustment amount for the j-th process parameter; This represents the cost weight of the change in the j-th process parameter.

[0181] Candidate control schemes that do not meet the quality constraints, equipment safety constraints, and process boundary constraints are eliminated, and the candidate control scheme with the smallest comprehensive cost function value is selected as the optimal control scheme from the remaining candidate control schemes.

[0182] When the selected optimal control schemes have opposite adjustment directions in multiple consecutive control cycles, the weight of parameter change cost is increased. The system extends the future cutting depth range to suppress process parameter oscillations. When none of the candidate control schemes meet the equipment safety constraints, the system sequentially executes commands to reduce cutting speed, pause cutting, and perform manual verification.

[0183] Next, the optimal control scheme is sent to the cutting control system. The cutting control system performs joint corrections on the linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration of the current cutting control segment and the next cutting control segment.

[0184] Specifically, the cutting control system converts the optimal control scheme into equipment control commands, including linear speed setpoint update commands, feed speed setpoint update commands, linear tension setpoint update commands, cutting fluid flow rate setpoint update commands, cutting fluid temperature setpoint update commands, guide wheel speed setpoint update commands, and feed acceleration setpoint update commands.

[0185] During execution, the system continues to collect cutting operation status data and online quality data in real time, and constructs the current process status for the next moment based on the new cutting depth coordinates. When the next control cycle arrives, the system continues to execute trajectory residual analysis, candidate control scheme generation, virtual trial cutting prediction, and comprehensive cost optimization processes. Thus, the cutting control system does not perform isolated parameter adjustments at a single moment, but rather performs continuous joint control of the current cutting control segment and the next cutting control segment.

[0186] S8. After dicing, the final inspection quality data of the silicon wafer is mapped according to the dicing depth and updated in a closed loop. The specific steps are as follows:

[0187] After cutting, the final inspection quality data of the silicon wafer is obtained; the final inspection quality data includes the actual wafer thickness, total thickness variation, warpage, surface roughness, line mark depth, chipping size, crack detection results, pass rate and material loss rate.

[0188] Based on the silicon wafer serial number, cutting depth coordinates, cutting time, and batch identifier, the final inspection quality data of each silicon wafer is mapped to the corresponding cutting control section.

[0189] For the final inspection quality data mapped to the r-th cutting control segment, calculate the segment quality deviation:

[0190]

[0191] in, This represents the segment quality deviation of the r-th cutting control segment; This represents the number of silicon wafers mapped to the r-th dicing control segment; This indicates the weight of the quality index for the l-th silicon wafer; This represents the final inspection quality index of the l-th silicon wafer corresponding to the r-th cutting control section; This indicates the target quality index for the l-th silicon wafer.

[0192] When the quality deviation of a section exceeds the preset quality deviation threshold of a section, the reference cutting trajectory, trajectory deviation threshold, segmented process window, virtual trial cutting prediction model, and candidate control scheme generation rules of the corresponding cutting control section are corrected.

[0193] For example, when the chipping size in the final inspection data corresponding to the risky cutting section is consistently large, the system increases the weight of the chipping-related quality cost in that section, reduces the upper limit of the feed rate in that section, and tightens the judgment threshold for load abrupt deviations. When the total thickness change in the final inspection data corresponding to the stable cutting section in the middle is large, the system corrects the reference bow trajectory and reference sheet thickness trend trajectory for that section, and increases the sensitivity to bow enlargement deviations and sheet thickness drift deviations.

[0194] When the quality deviation of a section does not exceed the preset quality deviation threshold of the section, and the trajectory deviation index of the section is lower than the corresponding threshold, the actual cutting trajectory corresponding to the cutting control section is written into the historical qualified cutting batch data for use in the generation of reference cutting trajectories for subsequent batches.

[0195] Through the above methods, this embodiment achieves dynamic closed-loop quality control throughout the entire process, from incoming material attributes, target quality, equipment status, cutting depth zoning, reference cutting trajectory, trajectory residual analysis, virtual trial cutting prediction, optimal control execution to final inspection quality feedback.

[0196] Furthermore, the functional units in the embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0197] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for dynamic quality closed-loop control throughout the entire silicon wafer dicing process, characterized in that, Includes the following steps: S1. Obtain the incoming material attribute data of the silicon rod to be cut, the quality index of the target silicon wafer, and the status parameters of the cutting equipment, and divide the silicon rod to be cut into multiple cutting control sections according to the cutting depth from the infeed position to the outfeed position. S2. Based on the incoming material attribute data, target silicon wafer quality indicators, cutting equipment status parameters, and historical qualified cutting batch data, generate a reference cutting trajectory and segmented process window for each cutting control segment. S3. During the silicon wafer dicing process, real-time dicing operation status data and online quality data are collected. The dicing operation status data and online quality data are aligned with the dicing depth coordinates to construct the current process status. The current process status is then compared with the reference dicing trajectory of the current dicing control section to obtain trajectory residual characteristics. S4. Determine whether the current cutting control segment deviates from the quality stability range based on the trajectory residual characteristics, and determine the type of deviation; S5. When the current cutting control segment deviates from the quality stability range, multiple candidate control schemes are generated based on the current process status, deviation type and corresponding segment process window. S6. Input each candidate control scheme into the virtual trial cutting prediction model to predict the quality results, equipment safety results and production efficiency results of each candidate control scheme in the future cutting depth range. S7. Calculate the comprehensive cost function value of each candidate control scheme based on the quality results, equipment safety results, and production efficiency results. Among the candidate control schemes that meet the quality constraints, equipment safety constraints, and process boundary constraints, select the candidate control scheme with the smallest comprehensive cost function value as the optimal control scheme. Then, send the optimal control scheme to the cutting control system to correct the cutting process parameters of the current cutting control segment and the next cutting control segment.

2. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S1, the incoming material attribute data includes silicon rod crystal orientation deviation, silicon rod diameter, silicon rod length, silicon rod warpage, end face perpendicularity, surface damage distribution, resistivity, minority carrier lifetime, oxygen content, carbon content, silicon rod bonding status, and batch identifier. The target silicon wafer quality indicators include target wafer thickness, wafer thickness tolerance, total thickness variation tolerance, warpage tolerance, surface roughness tolerance, line mark depth tolerance, edge chipping size tolerance, crack quantity tolerance, target pass rate, and target material loss rate. The status parameters of the cutting equipment include cutting wire type, cutting wire diameter, cumulative length of cutting wire used, guide wheel wear status, spindle load reference value, cutting fluid circulation status, cutting fluid temperature control capability, and equipment vibration reference value; The silicon rod to be cut is divided into the following sections from the entry point to the exit point: the entry stability section, the middle stable cutting section, the material stress release section, and the exit risk section. The cutting stabilization zone is used to monitor tension and vibration changes when the cutting wire enters the silicon rod; the middle stabilization cutting zone is used to monitor wire bowing, wafer thickness trend, and spindle load stability; the material stress release zone is used to monitor cracks caused by internal stress and surface damage in the incoming material; and the cutting exit risk zone is used to monitor edge chipping risk and load changes when the cutting wire approaches the end of the silicon rod.

3. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S2, a reference cutting trajectory and segmented process window are generated for each cutting control segment, specifically including the following steps: S21. Select historical sample batches from historical cutting batch data that meet the similarity conditions of the current silicon rod in terms of material type, target wafer thickness, silicon rod size, cutting line specification, cutting equipment number and target quality grade. S22. Select from the historical sample batches a historical qualified cutting batch with a pass rate higher than a preset pass rate threshold, a total thickness change lower than a preset total thickness change threshold, a chipping rate lower than a preset chipping rate threshold, and a breakage rate lower than a preset breakage rate threshold. S23. Statistically analyze the wire bow amount, wire tension, spindle load, feed resistance, cutting fluid temperature, cutting fluid flow rate, vibration characteristics, and online sheet thickness estimation values ​​of historical qualified cutting batches according to the cutting control section to obtain the reference mean trajectory and reference fluctuation range of each cutting control section. S24. Based on the current silicon rod's crystal orientation deviation, warpage, surface damage distribution, resistivity fluctuation, and minority carrier lifetime fluctuation, the reference mean trajectory and reference fluctuation range are corrected to generate a reference cutting trajectory adapted to the current silicon rod. S25. Based on the reference cutting trajectory, cutting equipment status parameters, and target silicon wafer quality indicators, generate segmented process windows corresponding to each cutting control section; The segmented process window includes the linear velocity range, feed rate range, linear tension range, cutting fluid flow rate range, cutting fluid temperature range, guide wheel speed range, feed acceleration range, single adjustment step size, and parameter adjustment priority.

4. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S3, the cutting operation status data includes linear speed, feed speed, cutting line tension, wire bowing amount, spindle current, spindle load, feed resistance, guide wheel speed, guide wheel speed fluctuation, cutting fluid temperature, cutting fluid flow rate, cutting fluid pressure, and vibration signal of the cutting area. The online quality data includes online sheet thickness estimates, thickness variation trends, online edge defect characteristics, online surface line mark characteristics, suspected crack characteristics, and warping trends; The current process state is represented as follows: ; in, Indicates the current process status; This represents a subset of current process parameters, including linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel speed, and feed acceleration. It represents a subset of equipment operating status, including spindle current, spindle load, feed resistance, wire bowing amount, guide wheel speed fluctuation, vibration characteristics, and cumulative used length of the cutting wire; It represents a subset of online quality status, including online sheet thickness estimates, thickness variation trends, edge defect characteristics, surface line mark characteristics, suspected crack characteristics, and warping trends; It represents a subset of historical short-term states, including the rate of change of the pantograph, tension fluctuation rate, load change rate, temperature change rate, vibration energy change rate, and online quality index change rate within a preset sliding time window; This represents a subset of the cutting position status, including the current cutting depth, the current cutting control segment number, the current cutting time, and the estimated output sequence number.

5. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S3, the trajectory residual features include instantaneous residual, residual change rate, segment cumulative residual, and trajectory deviation index; For the i-th type of trajectory variable in the r-th cut control segment, the instantaneous residual is: ； in, This represents the instantaneous residual of the i-th type of trajectory variable at the cutting depth d in the r-th cutting control segment; This represents the actual trajectory value of the i-th type of trajectory variable at the cutting depth d; This represents the reference trajectory value of the i-th type of trajectory variable at the cutting depth d; The rate of change of residuals is: ;in, Represents the rate of change of residuals. Indicates the cutting depth step size; The cumulative residual of the section is: ;in, Indicates the cumulative residual of the section. This represents the initial cutting depth of the r-th cutting control segment. This represents the final cutting depth of the r-th cutting control segment; The trajectory deviation index is: ;in, represents the trajectory deviation index of the r-th control segment; M represents the number of trajectory variables involved in the calculation; , and These represent the instantaneous residual weight, residual rate of change weight, and segment cumulative residual weight for the i-th type of trajectory variable, respectively. , and These represent the corresponding instantaneous residual threshold, residual change rate threshold, and segment cumulative residual threshold, respectively.

6. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 5, characterized in that, In step S4, when When the deviation exceeds the corresponding cutting control segment's deviation threshold, the current cutting control segment is determined to have deviated from the quality stability range. The deviation types include bow-increase deviation, tension instability deviation, load abrupt change deviation, insufficient cooling deviation, vibration-enhanced deviation, sheet thickness drift deviation, and defect aggregation deviation; When the actual pantograph trajectory is higher than the reference pantograph trajectory and the pantograph residual change rate is positive, it is determined to be a pantograph enlarged deviation. When the instantaneous residual and the rate of change of the residual of the actual tension trajectory relative to the reference tension trajectory both exceed the threshold, it is determined to be a tension instability type deviation. When both the actual spindle load trajectory and the feed resistance trajectory are higher than the reference load trajectory and continue to exceed the preset time, it is determined to be a load mutation type deviation. When the actual cutting fluid temperature trajectory is higher than the reference temperature trajectory and the actual cutting fluid flow trajectory is lower than the reference flow trajectory, it is determined to be a deviation due to insufficient cooling. When the frequency band energy of the actual vibration signal is higher than the frequency band energy threshold of the reference vibration trajectory, it is determined to be a vibration-enhanced deviation. When the online sheet thickness estimation trajectory deviates continuously in one direction from the reference sheet thickness trend trajectory, it is identified as a sheet thickness drift type deviation. When online edge defect features, online surface line mark features, and suspected crack features continuously exceed the corresponding defect thresholds within the same cutting control section, they are identified as defect clustering deviations.

7. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S5, multiple candidate regulation schemes are generated, specifically including the following steps: S51. Determine the allowable adjustment range and single adjustment step size of each adjustable process parameter based on the segmented process window of the current cutting control section. S52. Determine the parameter adjustment direction and parameter adjustment priority of the candidate control scheme based on the deviation type; S53. Generate multiple candidate control schemes that meet the single adjustment range limit based on the current process parameters; S54. Eliminate candidate control schemes that exceed the current cutting control segment segmentation process window; Among them, when the deviation of the line bow increases, the candidate control schemes include a combination of reducing the feed rate, increasing the line tension, and limiting the increase of the line speed; Under tension instability-type deviation, candidate control schemes include a combination of reducing guide wheel speed fluctuations, reducing feed acceleration, and adjusting the line tension to the tension stability window; Under load abrupt deviation, candidate control schemes include a combination of adjusting the feed rate, increasing the cutting fluid flow rate, and increasing the cutting fluid pressure; In cases of insufficient cooling deviation, candidate control schemes include combinations of increasing the fluid flow rate, decreasing the fluid temperature, and limiting the upper limit of the feed rate; Under vibration-enhanced deviation, candidate control schemes include a combination of adjusting the matching relationship between linear velocity and guide wheel speed, reducing feed acceleration, and reducing linear velocity fluctuation. Under the condition of sheet thickness drift deviation, candidate control schemes include a combination of simultaneous correction of feed rate, line tension and line bow target value.

8. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, In step S6, the virtual trial cutting prediction model includes a quality prediction sub-model, an equipment safety prediction sub-model, and an efficiency prediction sub-model; For the u-th candidate control scheme The output of the virtual trial-cut prediction model is: ； in, This represents the set of predicted results for the u-th candidate control scheme in the future control time domain; This represents a virtual trial-cut prediction model; Indicates the current process status; This represents the u-th candidate control scheme; This represents the set of constraints corresponding to the r-th cutting control segment; The quality prediction sub-model is used to predict sheet thickness deviation, total thickness change, line mark severity, edge chipping probability, crack probability, and warping trend in the future control time domain. The equipment safety prediction sub-model is used to predict the risks of wire breakage, spindle overload, tension instability, guide wheel abnormality, and vibration abnormality in the future control time domain. The efficiency prediction sub-model is used to predict the cutting time, material loss rate, cutting line consumption, and energy consumption per unit silicon wafer in the future control time domain. The future cutting depth range is a continuous cutting depth range from the current cutting depth to the end depth of the next cutting control segment.

9. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1 or 8, characterized in that, In step S7, the cutting process parameters include linear velocity, feed rate, linear tension, cutting fluid flow rate, cutting fluid temperature, guide wheel rotation speed, and feed acceleration. The comprehensive cost function value is calculated as follows: ； in, Let represent the comprehensive cost function value of the u-th candidate control scheme; This represents the quality cost of the u-th candidate control scheme; This represents the equipment safety cost of the u-th candidate control scheme; This represents the production efficiency cost of the u-th candidate control scheme; This represents the cost of parameter changes for the u-th candidate control scheme; , , and These represent the weights of quality cost, equipment safety cost, production efficiency cost, and parameter change cost, respectively. The quality cost The safety cost of the equipment is calculated based on the predicted thickness deviation, predicted total thickness change, predicted edge chipping probability, predicted crack probability, predicted line mark severity, and predicted warping trend. The production efficiency cost is calculated based on the predicted risks of wire breakage, spindle overload, tension instability, guide wheel abnormalities, and vibration abnormalities. The cost of changing these parameters is calculated based on predicted cutting time, predicted material loss rate, predicted cutting line consumption, and predicted energy consumption per unit silicon wafer. Calculated based on the adjustment range of the candidate control scheme relative to the current process parameters; When the selected optimal control schemes have opposite adjustment directions in multiple consecutive control cycles, the weight of parameter change cost is increased. This extends the future cutting depth range to suppress process parameter oscillations; When none of the candidate control schemes meet the equipment safety constraints, the commands to reduce cutting speed, pause cutting, and perform manual review are executed in sequence.

10. The dynamic quality closed-loop control method for the entire silicon wafer dicing process according to claim 1, characterized in that, Also includes: S8. After the cutting is completed, obtain the final inspection quality data of the silicon wafer, and map the final inspection quality data to the corresponding cutting control section according to the cutting depth. Update the reference cutting trajectory, segmented process window, virtual trial cutting prediction model and candidate control scheme generation rules according to the mapping results.

Images