Intelligent thinning grinding method and system for laser delaminated silicon carbide wafer surface

Abstract

The application belongs to the technical field of semiconductor wafer processing. An intelligent thinning grinding method and system for laser stripping silicon carbide wafer surface are proposed. The wafer surface and grinding wheel state data are collected and the feature vector is constructed. The state evaluation value is calculated to distinguish the stripping surface quality grade. The three-stage grinding parameters are matched and adapted by calling the database. The phased dynamic spindle current window is generated by comprehensively considering the wafer grade, grinding parameters, grinding wheel state and historical data. The spindle current is collected in real time in the grinding stage. The current mean, rise and fluctuation amplitude are obtained by means of sliding window operation. The grinding interval is divided by comparing the dynamic window. The feed speed and grinding removal amount are adaptively adjusted. The finished product detection data is collected and the grinding quality score is calculated after grinding. The application realizes the adaptive control of the laser stripping silicon carbide wafer thinning process, and optimizes the grinding stability and wafer processing quality.

Description

Technical Field

[0001] This invention relates to the field of semiconductor wafer processing technology, and specifically to an intelligent thinning grinding method and system for laser lift-off of silicon carbide wafer surfaces. Background Technology

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Silicon carbide (SiC) is a third-generation wide-bandgap semiconductor material with significant application value in the manufacturing of power electronic devices. Wafer thinning is a key process in device fabrication. Laser lift-off can achieve non-destructive separation of the crystal and is often used as a pre-processing step for SiC wafer thinning. This processing method does not involve large-area material loss, and the overall process is non-contact and highly efficient. Various terminal thinning equipment are generally equipped with grinding modules to eliminate uneven areas on the surface after lift-off. The industry's conventional grinding operation logic relies on fixed process parameters to complete the entire process. These parameters are mostly preset by operators based on general processing experience. The equipment does not actively collect wafer surface status information during operation, nor does it adjust its operating strategy based on real-time grinding load signals. The entire processing flow relies on standardized parameters to complete batch wafer thinning operations, adapting to processing objects with stable surface uniformity.

[0004] The surface of silicon carbide wafers after laser lift-off exhibits diverse morphological characteristics. Fixed grinding parameters cannot adapt to the real-time grinding load variations of a single wafer, making it difficult to simultaneously balance processing efficiency and wafer surface quality in the corresponding processing flow. Existing thinning equipment only uses spindle current for basic equipment protection and lacks a linkage adjustment logic between surface condition, process parameters, and real-time current signals. It cannot grade and match initial grinding parameters according to the actual morphology of a single wafer after lift-off, nor can it dynamically adjust feed and removal parameters based on real-time spindle current characteristics. It is also difficult to iteratively optimize process parameters based on the quality of a single batch of processing, and a complete closed-loop grinding control process cannot be formed. Summary of the Invention

[0005] To address the challenges of uneven surface undulations, inconsistent residual modified layer distribution, and significant local damage variations on silicon carbide wafers after laser lift-off, which make it difficult to balance grinding efficiency, surface quality, and wheel wear with fixed grinding parameters, this invention provides an intelligent thinning grinding method and system for laser-lifted silicon carbide wafers. By analyzing the wafer surface condition data after laser lift-off, the initial lift-off surface quality level is determined, and a three-stage initial grinding parameter set is matched. During the grinding process, adaptive feedback adjustment is performed based on changes in the wheel spindle current. After grinding, the process database is updated based on the post-grind quality and wheel wear results, thereby achieving low-damage, high-stability thinning grinding of silicon carbide wafers.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces.

[0007] A smart thinning grinding method for laser-lifted silicon carbide wafer surfaces includes the following processes: Surface state data and grinding wheel state parameters of laser-lifted silicon carbide wafers are obtained, and surface state feature vectors before grinding are constructed. The evaluation value of the peeled surface state is calculated based on the feature vector of the surface state before grinding, and the quality level of the initial peeled surface is determined. Based on the initial peel surface quality level, retrieve the process database to match the three-stage initial grinding parameter set; Based on the initial peeling surface quality grade, grinding parameters of the current grinding stage, and grinding wheel state parameters, historical stable grinding samples are selected from historical processing data, and dynamic spindle current windows corresponding to each grinding stage are calculated and generated based on the selected historical stable grinding samples. During the grinding process, the spindle current of the grinding wheel is collected in real time. Based on the sliding time window, the average spindle current, the increase in spindle current, and the fluctuation range of spindle current are calculated. The average spindle current, the increase in spindle current, and the fluctuation range of spindle current are compared with the dynamic spindle current window to divide the current interval to which the current grinding state belongs. The feed rate and grinding removal amount are adaptively adjusted according to the corresponding range output parameter adjustment command; After grinding is completed, post-grinding quality inspection data is collected, and grinding quality score is calculated based on the post-grinding quality inspection data.

[0008] In one implementation of the first aspect of the present invention, a surface state feature vector before grinding is constructed, including: total thickness variation, wafer warpage, wafer curvature, surface roughness, local peak-to-valley difference, crack area ratio, edge chipping width, and grinding wheel state parameters as feature components to form the surface state feature vector before grinding. The evaluation value of the surface condition of the stripped surface is calculated, including: normalizing the thickness variation, wafer warpage, wafer curvature, surface roughness and local peak-valley difference in the surface condition feature vector before grinding according to the corresponding parameter reference values ​​to obtain each normalized sub-item; Assign corresponding weight coefficients to each normalized component, and multiply and sum the weight coefficients with the corresponding normalized components to obtain the peeling surface condition evaluation value. Based on the peeling surface condition evaluation value, the deviation of each surface condition parameter from the corresponding benchmark value, and in combination with the crack area ratio and edge chipping width, determine the initial peeling surface quality level.

[0009] In one implementation of the first aspect of the present invention, the three-stage initial grinding parameter set includes three-stage feed rates, three-stage grinding removal amount in a sequentially decreasing trend, as well as grinding wheel spindle speed, table speed, and polishing time; the grinding wheel spindle speed and table speed are only used for parameter matching before grinding and grinding stage switching configuration, and no real-time adaptive adjustment is performed during grinding.

[0010] In one implementation of the first aspect of the present invention, generating dynamic spindle current windows corresponding to each grinding stage includes: Based on the initial peeling surface quality grade, the feed rate of the current grinding stage, the amount of material removed and the grinding wheel state parameters, historical stable grinding samples that meet the preset matching conditions are selected from the historical processing database, and the average spindle current of the historical stable grinding samples in the corresponding grinding stage is calculated to obtain the stage reference stable spindle current. Extract the standard deviation of current fluctuation within the historical stable grinding data of the same stage, configure the lower limit adjustment coefficient and the upper limit adjustment coefficient respectively, and calculate the lower limit value and the upper limit value of the stage spindle current by combining the stage reference stable spindle current, the standard deviation of current fluctuation, and the two types of adjustment coefficients. The stage spindle current lower limit value and the stage spindle current upper limit value constitute the dynamic spindle current window.

[0011] In one implementation of the first aspect of the present invention, calculating the spindle current characteristics based on a sliding time window includes: The arithmetic mean of the grinding wheel spindle current at a fixed number of consecutive sampling times is used to obtain the average spindle current. The difference between the currents of the grinding wheel spindle at two adjacent sampling times is used to calculate the increase in spindle current. Calculate the square mean of the deviation between the grinding wheel spindle current and the average spindle current at each sampling time within the sliding window, and take the square root of the square mean of the deviation to obtain the spindle current fluctuation amplitude.

[0012] In one implementation of the first aspect of the present invention, if the average value of the spindle current is less than the lower limit of the stage spindle current within the dynamic spindle current window, and the duration reaches a first preset duration, an additional judgment on the grinding endpoint is performed. If the cumulative grinding removal amount in the current stage does not reach the grinding removal amount in a single stage, the feed rate or the grinding removal amount in the current stage will be increased by the preset adjustment step size. When the cumulative grinding removal amount in the current stage reaches the single-stage grinding removal amount, the grinding stage will be switched or the grinding will be terminated.

[0013] In one implementation of the first aspect of the present invention, if the average value of the spindle current is within the dynamic spindle current window range, the spindle current fluctuation amplitude is compared with the stable grinding fluctuation threshold. When the spindle current fluctuation amplitude is less than or equal to the stable grinding fluctuation threshold, maintain the current grinding parameters unchanged and record the current processing sample; When the spindle current fluctuation exceeds the stable grinding fluctuation threshold, reduce the current stage feed rate or the current stage grinding removal amount, and simultaneously increase the coolant flow rate or output a grinding wheel status check prompt.

[0014] In one implementation of the first aspect of the present invention, if the average value of the spindle current is greater than the upper limit of the spindle current in the dynamic spindle current window and the duration reaches a third preset duration, an additional safety protection judgment is performed. When the average spindle current reaches the stage safety spindle current threshold, grinding is paused directly; when the average spindle current does not reach the stage safety spindle current threshold, the feed rate or the amount of material removed in the current stage is reduced by a preset adjustment step, and the coolant flow rate is increased. If the increase in spindle current exceeds the threshold of the stage increase, distinguish between controllable mutation and dangerous mutation. In the case of controllable mutation, reduce the feed rate or the amount of material removed during grinding. In the case of dangerous mutation, stop grinding directly.

[0015] In one implementation of the first aspect of the present invention, the process database is updated by associating and storing surface state data, initial peeling surface quality grade, three-stage initial grinding parameter set, dynamic spindle current window, average spindle current, spindle current increase, spindle current fluctuation amplitude, and grinding quality score, including: The normalized quality items are obtained by dividing the total thickness change after grinding, the wafer warpage after grinding, the wafer curvature after grinding, and the surface roughness after grinding by the corresponding parameter target values. The grinding wheel wear is obtained by dividing the grinding wheel wear reference value. Corresponding weight coefficients are assigned to all items. The grinding quality score is obtained by multiplying and summing all weight coefficients with the corresponding items. When the grinding quality score meets the standard, the grinding wheel wear is lower than the preset threshold, and the spindle current characteristics of the corresponding grinding stage meet the stable grinding judgment conditions, the corresponding processing parameters and spindle current data are marked as preferred samples and historical stable grinding samples, and their parameter recommendation priority under the same initial peeling surface quality level is increased for subsequent matching of initial grinding parameters of the same type of wafer and calculation of stage reference stable spindle current. When the grinding quality score fails to meet the standard or the grinding wheel wear exceeds the preset threshold, the three-stage initial grinding parameter set that matches the initial peeling surface quality level of the same type is corrected.

[0016] Secondly, the present invention provides an intelligent thinning grinding system for laser-lifted silicon carbide wafer surfaces.

[0017] A smart thinning grinding system for laser lift-off of silicon carbide wafer surfaces includes: The surface state data input unit is configured to: acquire the surface state data and grinding wheel state parameters of the laser-lifted silicon carbide wafer, and construct a surface state feature vector before grinding; The peeling surface quality grade judgment unit is configured to: calculate the peeling surface state evaluation value based on the surface state feature vector before grinding, and determine the initial peeling surface quality grade. The initial grinding parameter matching unit is configured to retrieve the three-stage initial grinding parameter set from the process database based on the initial peel surface quality level. The dynamic spindle current window generation unit is configured to: select historical stable grinding samples from historical processing data based on the initial peeling surface quality level, grinding parameters of the current grinding stage, and grinding wheel state parameters, and calculate and generate the dynamic spindle current window corresponding to each grinding stage based on the selected historical stable grinding samples; The spindle current reading and interval judgment unit is configured to: collect the grinding wheel spindle current in real time during the grinding process, calculate the average spindle current, spindle current increase, and spindle current fluctuation based on the sliding time window, compare the average spindle current, spindle current increase, and spindle current fluctuation with the dynamic spindle current window, and divide the current interval to which the current grinding state belongs. The grinding parameter feedback adjustment unit is configured to adaptively adjust the feed rate and grinding removal amount in the current stage according to the corresponding range output parameter adjustment command. The post-grinding quality evaluation unit is configured to: collect post-grinding quality inspection data after grinding is completed, and calculate the grinding quality score based on the post-grinding quality inspection data.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention relies on a coherent processing flow of pre-grinding surface feature construction, graded parameter matching, dynamic current window generation, real-time current feature discrimination, parameter adaptive adjustment, and post-grinding quality evaluation. This addresses the current situation where fixed grinding parameters are difficult to adapt to the diverse surfaces of laser-lifted silicon carbide wafers. First, this invention classifies surface quality levels based on various wafer surface morphology indicators and grinding wheel operating conditions, matching suitable three-stage initial grinding parameters. Compared to manually preset uniform parameters, this can better match the basic surface conditions of a single wafer after lift-off, reducing load imbalance in the initial processing stage. This invention combines multiple types of pre-existing information to generate a dynamic spindle current window adapted to each grinding stage. Using a sliding window, it continuously extracts multi-dimensional spindle current features and dynamically adjusts the feed rate and removal amount according to the window range. This allows for fine-tuning of the processing intensity based on real-time load changes during grinding, mitigating processing fluctuations caused by local protrusions and residual modified layers. After grinding, a unified post-grinding quality calculation is performed, completely preserving the correlation information of a single wafer from its initial morphology to real-time processing load and final processing quality, providing a valid basis for subsequent processing parameter iterations. The coordinated operation of the above-mentioned links can take into account both the integrity of the wafer surface and the grinding wear during the processing, adapt to the thinning requirements of laser lift-off silicon carbide wafers with different morphologies, and optimize the operational stability of the grinding process.

[0019] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 A schematic diagram of an intelligent thinning grinding system for laser lift-off of silicon carbide wafer surfaces, provided as an exemplary embodiment of the present invention; Figure 2 A flowchart illustrating an exemplary embodiment of the present invention for a smart thinning grinding method for laser-lifted silicon carbide wafer surfaces. Figure 3 A flowchart of spindle current feedback adjustment is provided for an exemplary embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] In existing thinning equipment, the spindle current of the grinding wheel is mostly used for equipment operation monitoring, overload protection, or abnormal alarms, and has not been fully utilized for intelligent matching and closed-loop control of grinding parameters. Furthermore, process parameters such as the three-stage feed rate, grinding removal amount, grinding wheel speed, and table speed mainly rely on manual experience for setting. Operators find it difficult to accurately judge the current grinding state based on the laser-removed surface condition and real-time spindle current changes, and also find it difficult to determine the optimal parameter combination for different surface and current conditions. Especially for the laser-removed silicon carbide surface, due to its significant regional differences and random fluctuations in surface condition, simply relying on fixed parameters or manual experience is insufficient to adapt to changes in grinding load in a timely manner, easily affecting the stability, processing efficiency, and surface quality of the thinning process.

[0025] In view of the problems existing in the existing solutions, the present invention proposes an intelligent thinning grinding system for laser-lifted silicon carbide wafer surfaces. The intelligent thinning grinding system is embedded in the host software of the thinning machine, or communicates with the host control software of the thinning machine, or the algorithm logic corresponding to the intelligent thinning grinding method of the present invention is directly written into the host software of the thinning machine. The initial grinding parameter set is matched based on the surface state data after laser lift-off, and the feed rate and grinding removal amount of the current stage are adaptively adjusted based on the characteristics of the grinding wheel spindle current. The grinding wheel spindle speed and the table speed can be set as initial parameters or stage switching parameters before grinding, so as to achieve stable, low-damage and low-grinding wheel wear thinning grinding of the silicon carbide surface after laser lift-off.

[0026] In this implementation, such as Figure 1 As shown, the intelligent thinning grinding system is embedded in the upper control software of the thinning machine or communicates with the upper control software of the thinning machine. It is used to realize surface condition identification, initial parameter matching, spindle current feedback adjustment, post-grinding quality evaluation and process database update during the laser lift-off silicon carbide wafer thinning grinding process.

[0027] The intelligent thinning grinding system includes a surface condition data input unit, a peeling surface quality grade judgment unit, an initial grinding parameter matching unit, a dynamic spindle current window generation unit, a spindle current reading and interval judgment unit, a grinding parameter feedback adjustment unit, a post-grinding quality evaluation unit, and a process database update unit. The specific functions of each unit are as follows: Surface condition data input unit: Used to receive or import surface condition data of silicon carbide wafers after laser lift-off. The surface condition data includes the total thickness change. wafer warpage wafer curvature Surface roughness Local peak-valley difference Crack area ratio, edge chipping width, and grinding wheel condition parameters (one or more can be selected).

[0028] Peeling surface quality grade judgment unit: used to judge the initial peeling surface quality grade based on surface condition data. The initial peeling surface quality grade includes flat and stable type, uneven type, stress warping type, rough and broken type and local abnormal type (one or more can be selected).

[0029] Initial grinding parameter matching unit: Used to retrieve the corresponding initial grinding parameter set from the process database based on the initial peel surface quality level. The initial grinding parameter set includes the first-stage feed rate. Second stage feed rate Third stage feed rate First stage grinding removal amount Second stage grinding removal amount Third stage grinding removal amount Grinding wheel spindle speed Table speed and polishing time (One or more can be selected).

[0030] Dynamic spindle current window generation unit: It is used to select historical stable grinding samples from historical processing data based on the initial peeling surface quality level, grinding parameters of the current grinding stage, and grinding wheel state parameters, and calculate and generate the dynamic spindle current window corresponding to each grinding stage based on the selected historical stable grinding samples.

[0031] Spindle current reading and interval judgment unit: used to read the grinding wheel spindle current from the upper control software of the thinning machine, and determine whether the current grinding state is in the lower limit zone, stable zone, upper limit zone or sudden change zone of the spindle current based on the average value of the spindle current, the increase of the spindle current and the fluctuation range of the spindle current at the current stage.

[0032] Grinding parameter feedback adjustment unit: It is used to output parameter adjustment commands based on the spindle current range judgment results, and can trigger coolant adjustment, tool retraction protection, grinding pause, grinding wheel dressing prompt or abnormal alarm.

[0033] Post-grinding quality evaluation unit: Used to receive or import post-grinding quality inspection data and determine whether the current three-stage grinding parameter combination meets the processing requirements. Post-grinding quality inspection data includes post-grinding... After grinding After grinding Surface roughness Crack defects, edge chipping defects, subsurface damage, and grinding wheel wear.

[0034] Process database update unit: It is used to associate and store pre-grinding surface condition data, initial peeling surface quality grade, three-stage grinding parameters, spindle current characteristics of each stage, post-grinding quality results and grinding wheel wear data, and update the corresponding process database.

[0035] In some implementations, the intelligent thinning grinding system performs correlation analysis on the peeling surface condition data, three-stage grinding parameters, spindle current characteristics, and post-grinding quality results to help determine the initial peeling surface quality level, recommend initial grinding parameter sets, and optimize parameter adjustment strategies; the dynamic spindle current window is calculated and generated based on the stage reference stable spindle current and current fluctuation standard deviation corresponding to historical stable grinding samples.

[0036] Based on the aforementioned intelligent thinning grinding system, this invention also provides an intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces, such as... Figure 2 As shown, it includes the following steps: S1: Obtain surface state data of silicon carbide wafers after laser lift-off.

[0037] Before grinding, the surface of the laser-lifted silicon carbide wafer is tested or data is imported to obtain parameters characterizing the quality of the lifted surface. Surface condition data includes total thickness variation. wafer warpage wafer curvature Surface roughness Local peak-valley difference The percentage of crack area, the width of edge chipping, and the condition parameters of the grinding wheel.

[0038] In some implementations, the intelligent thinning grinding system constructs the above surface condition data into a surface condition feature vector before grinding: (1); in, This is the feature vector of the surface state before grinding. This represents the percentage of the area affected by cracks or defects. The width of the edge chipping. These are the grinding wheel condition parameters, which include the grinding wheel allowance, the cumulative number of grinding discs, and the grinding wheel wear.

[0039] S2: Determine the initial surface quality level based on surface condition data.

[0040] The intelligent thinning grinding system establishes a mapping relationship between the surface state data of the silicon carbide wafer after laser lift-off and the initial lift-off surface quality grade based on the surface state feature vector before grinding, thereby obtaining the initial lift-off surface quality grade. The initial lift-off surface quality grades include flat and stable type, uneven type, stress warping type, rough and broken type, and local anomaly type.

[0041] In some implementations, the intelligent thinning grinding system calculates the peeling surface state evaluation value based on the pre-grinding surface state feature vector using a peeling surface state evaluation function: (2); in, This is the evaluation value for the surface condition of the peeled surface. to These are the weighting coefficients. , , , and This is the reference value for the corresponding surface condition parameters.

[0042] In one specific embodiment, , , , and The normalized baseline values ​​are 20 μm, 40 μm, 40 μm, 10 μm and 20 μm respectively; the corresponding weight coefficients are 0.25, 0.20, 0.20, 0.25 and 0.10 respectively, and the sum of the weight coefficients is 1.

[0043] in, Used to characterize the overall thickness uniformity of a wafer; Used to characterize the overall warpage of a wafer; Used to characterize the overall bending direction and bending amplitude of a wafer; Used to characterize the roughness of the laser-exfoliated surface; Used to characterize the degree of local peak-valley undulation on a laser-exfoliated surface.

[0044] Based on the evaluation value of the peeled surface condition The quality level of the initial peeled surface is determined by analyzing the deviation of each surface state parameter from the corresponding normalized reference value.

[0045] when , , , and At that time, the initial peeling surface quality grade was determined to be flat and stable.

[0046] when or When the value exceeds the corresponding normalized reference value and is mainly manifested as wafer thickness variation or abnormal local peak-valley fluctuations, the initial peeling surface quality level is determined to be uneven.

[0047] when or When the value exceeds the corresponding normalized baseline value and is mainly manifested as overall wafer warping or bending abnormality, the initial peeling surface quality level is determined to be stress warping type.

[0048] when If the quality of the initial stripped surface exceeds the corresponding normalized reference value, or if the wafer surface is accompanied by obvious cracks or breakage defects, the quality level is determined to be rough and broken.

[0049] Furthermore, the proportion of crack or defect area and edge chipping width As an auxiliary parameter for judging abnormal conditions on the peeled surface. When the area of ​​cracks or defects accounts for... Or the width of edge chipping When the wafer is identified as having a significant peeling anomaly, the initial peeling surface quality level is prioritized as a local anomaly.

[0050] When the same wafer simultaneously meets the criteria for two or more surface quality types, the dominant surface quality type is determined based on the degree of deviation of each surface state parameter from the corresponding normalized reference value; when the criteria for local anomaly type are met, it is preferentially determined to be a local anomaly type.

[0051] In other embodiments, the normalized reference value, weighting coefficient, and corresponding judgment threshold can be adjusted or updated based on the correlation between surface condition parameters and grinding wheel spindle current characteristics, post-grinding quality results, and grinding wheel wear in historical processing data.

[0052] S3: Match the initial grinding parameter set according to the initial peel surface quality level.

[0053] The intelligent thinning grinding system retrieves the corresponding initial grinding parameter set from the process database based on the initial peeling surface quality level. The initial grinding parameter set includes the first-stage feed rate. Second stage feed rate Third stage feed rate First stage grinding removal amount Second stage grinding removal amount Third stage grinding removal amount Grinding wheel spindle speed Table speed and polishing time .

[0054] The first, second, and third stages are divided sequentially according to the grinding depth. Typically, the feed rates for the three stages satisfy the following: Furthermore, the removal amount by the three-stage grinding process meets the following requirements: The three-stage grinding process is divided according to the preset grinding removal amount range. When the cumulative grinding removal amount of the current stage reaches the target grinding removal amount of the corresponding stage, the system switches to the next stage grinding parameter group.

[0055] In some implementations, the grinding wheel spindle speed and table speed Used for initial parameter matching before grinding, or for setting parameters for switching between the first, second, and third stages. To avoid the impact of frequent changes in the grinding wheel spindle speed and table speed on equipment stability, spindle life, and grinding quality, the grinding wheel spindle speed and table speed are not used as high-frequency real-time feedback adjustment parameters.

[0056] In some implementations, when the initial peeling surface quality level is flat and stable, the intelligent thinning grinding system calls the high-efficiency grinding parameter set; when the initial peeling surface quality level is uneven, the intelligent thinning grinding system calls the homogenizing grinding parameter set; when the initial peeling surface quality level is stress-warped, the intelligent thinning grinding system calls the low-stress grinding parameter set; when the initial peeling surface quality level is rough and fragmented, the intelligent thinning grinding system calls the low-damage grinding parameter set; and when the initial peeling surface quality level is locally abnormal, the intelligent thinning grinding system calls the conservative grinding parameter set.

[0057] S4: Generate a three-stage dynamic spindle current window.

[0058] The intelligent thinning grinding system generates dynamic spindle current windows for the first, second, and third stages based on the initial peeling surface quality level, three-stage feed rate, three-stage grinding removal amount, grinding wheel state parameters, and historical machining data.

[0059] The dynamic spindle current window includes the first Stage spindle current lower limit , No. Upper limit of spindle current in stage and the The phased stable spindle current range, in which, These correspond to the first, second, and third stages of grinding, respectively.

[0060] In some implementations, the dynamic current window for stage j satisfies: (3); (4); in, For the first Stage reference to stabilize spindle current, For the first Standard deviation of current fluctuation during a stable grinding stage in history. and For the first The dynamic adjustment coefficient of the segment.

[0061] In some implementation schemes, the first The stage reference stable spindle current is determined based on historical stable grinding data. The intelligent thinning grinding system selects stable grinding samples that meet preset matching conditions from the historical processing database based on the initial peel surface quality level of the current wafer, the current grinding stage, the current stage feed rate, the current stage grinding removal amount, and the grinding wheel state parameters. It then calculates the average spindle current of the stable grinding samples in the j-th stage as the reference value for the j-th stage. Stage reference stable spindle current: (5); in, To meet the preset matching conditions, the number of historical stable grinding samples is required. For the first The historical stable grinding sample in the first Average spindle current during the stage.

[0062] The dynamic spindle current window is not a fixed current range, but is dynamically generated based on the initial surface quality level of the laser-lifted silicon carbide wafer, the three-stage grinding parameters, the grinding wheel condition, and historical stable grinding data.

[0063] S5: Real-time acquisition of grinding wheel spindle current and extraction of current characteristics according to grinding stage.

[0064] During the three-stage thinning grinding process, the system collects the spindle current of the grinding wheel spindle drive motor of the thinning machine in real time. And based on the current grinding stage Extract the spindle current characteristics for the corresponding stage, where... .

[0065] spindle current The data is read or recorded in real time by the thinning machine's host control software to characterize the change in grinding load between the grinding wheel and the laser-lifted silicon carbide wafer surface.

[0066] The intelligent thinning grinding system calculates the average value of the spindle current, the increase in spindle current, and the fluctuation range of spindle current in the current j-th stage based on the sliding time window.

[0067] The average spindle current in stage j satisfies: (6); The spindle current increase in stage j: (7); The spindle current fluctuation amplitude in stage j: (8); in, for The grinding wheel spindle current at any given time, This represents the number of sampling points within the sliding window.

[0068] S6: Adaptively adjust grinding parameters according to the spindle current range of the current grinding stage.

[0069] The intelligent thinning grinding system adaptively adjusts the grinding parameters for the current grinding stage (j) based on the relationship between the spindle current characteristics and the corresponding dynamic spindle current window. Figure 3 As shown.

[0070] In this process, real-time feedback adjustment during grinding takes priority over the current feed rate. and the amount of material removed by grinding at the current stage If the equipment allows, the coolant flow rate can also be adjusted to trigger a grinding pause or an abnormal alarm. Grinding wheel spindle speed. and table speed It is mainly used for initial parameter matching before grinding or parameter setting between different grinding stages; during the grinding process, the grinding wheel spindle speed and table speed It is not used as a high-frequency real-time feedback adjustment parameter.

[0071] In some implementations, the three-stage feed rate meets the following range: , , The three-stage grinding removal amount meets the following requirements: , , Furthermore, during parameter adjustment, the intelligent thinning grinding system applies boundary constraints to the adjusted feed rate and grinding removal amount, ensuring that the adjusted parameters are within acceptable limits. , , and , , Not exceeding the above range, and preferably maintaining: as well as: .

[0072] In some implementations, the intelligent thinning grinding system sets additional judgment conditions based on the spindle current range judgment. These additional judgment conditions include grinding endpoint judgment, fluctuation stability judgment, safety protection judgment, and current change trend judgment (one or more can be selected). Specifically, grinding endpoint judgment is used to avoid misjudging a low current state near the target thickness or stage endpoint as insufficient grinding load; fluctuation stability judgment is used to avoid misjudging a state with normal average current but large fluctuations as smooth grinding; safety protection judgment is used to prioritize pausing grinding when the spindle current exceeds a safety threshold; and current change trend judgment is used to distinguish between actual current abrupt changes and accidental sampling fluctuations.

[0073] (1) When the real-time spindle current satisfies: If the duration exceeds the first preset duration, the system enters the spindle current lower limit zone judgment. At this time, the intelligent thinning grinding system first performs the grinding endpoint additional judgment, that is, to determine whether the current wafer is close to the target thickness, or whether the cumulative grinding removal amount in the current stage has reached the target grinding removal amount in the corresponding stage.

[0074] When the following conditions are met: When the system determines that the current low current state indicates that the grinding process is nearing its end or that the current grinding stage is complete, it will either switch stages or directly determine that the grinding process is complete. For the first Cumulative grinding removal amount per stage For the first The target amount of material removed by grinding in each stage.

[0075] When the grinding endpoint conditions are not met, the system determines that the grinding load in the current j-th stage is insufficient, posing a risk of insufficient wheel contact, insufficient material removal, or dry grinding. At this point, the intelligent thinning grinding system calls the current stage's increased-load grinding parameter set and increases the feed rate according to a preset step size. Or increase the amount of material removed in a single grinding pass. .

[0076] In some implementations, the current stage feed rate is adjusted as follows: Or adjust the current grinding removal amount as follows: .in, The adjusted feed rate for the current stage. This is the adjusted amount of material removed during the current grinding stage. Adjust the feed rate step size for the current stage. Adjust the step size for the amount of material removed by grinding at the current stage.

[0077] (2) When the real-time spindle current satisfies: At this point, the system enters the candidate judgment for the stable region of the spindle current. Then, the intelligent thinning grinding system further performs an additional judgment on fluctuation stability.

[0078] when At that time, the system determines that the current grinding state is in the spindle current stable region. This is the spindle current fluctuation threshold under stable grinding conditions. In the stable spindle current region, the control system maintains the current grinding parameters unchanged and records the grinding parameters, spindle current characteristics, and grinding status for this stage. When the stable state lasts for more than a second preset duration, and the post-grinding quality meets the requirements, the corresponding value for the current stage is... , The spindle current characteristics and post-grinding quality results were marked as preferred samples and used as the basis for recommending parameters for subsequent wafers with similar modified zone morphology.

[0079] When the average spindle current is within the dynamic spindle current window, but the standard deviation of the spindle current exceeds the stability threshold, the system determines that the current state is an abnormal spindle current fluctuation state. At this time, the system executes a steady-state adjustment strategy, which involves slightly reducing the current feed rate. Reduce the amount of material removed during the current grinding stage. Increase coolant flow or prompt to check the condition of the grinding wheel (one or more options can be selected).

[0080] (3) When the following conditions are met: If the duration exceeds the third preset duration, the system enters the spindle current upper limit zone judgment, and determines that the current grinding load is too high, and there is a risk of local bulges, thick residual modified zone, grinding wheel blockage, excessive feed, crack propagation or edge chipping.

[0081] At this point, the intelligent thinning grinding system performs an additional safety protection check. When the real-time spindle current meets: When the system determines that the current grinding state poses a safety risk, it prioritizes pausing the grinding process. For safe spindle current threshold.

[0082] When the real-time spindle current does not reach the safe spindle current threshold, the intelligent thinning grinding system calls the load reduction grinding parameter set and reduces the feed rate of the current stage according to the preset step size. Or reduce the amount of material removed during the current grinding stage. .

[0083] In some implementations, the current stage feed rate is adjusted as follows: Or adjust the current grinding removal amount as follows: When the equipment supports coolant flow rate adjustment, the coolant flow rate can be increased to reduce grinding heat accumulation and the risk of localized overload.

[0084] (4) When the real-time spindle current increase satisfies: At this point, the system enters the spindle current sudden change zone judgment. Even if the current average spindle current has not yet exceeded the upper limit of the spindle current, the intelligent thinning grinding system still determines that there is a sudden change risk in the current j-th stage of grinding state and executes the sudden change control strategy in advance. Among them, The threshold value for the spindle current increase in the j-th stage.

[0085] when When the system determines that the current mutation is a controllable mutation state, the intelligent thinning grinding system executes load reduction control, which includes reducing the feed rate at the current stage. Reduce the amount of material removed during the current grinding stage. Increase coolant flow or automatically adjust parameter combinations for subsequent stages after the current stage ends.

[0086] when When the system determines that the current mutation is a dangerous mutation, it indicates that the current grinding load has changed drastically, posing a high risk of crack propagation, edge chipping, abnormal wheel impact, or machining instability. In this case, the system prioritizes pausing grinding. is the safety mutation threshold for stage j.

[0087] By setting a spindle current mutation threshold and a safety mutation threshold, this invention can distinguish between normal mutation states that can be recovered through load reduction adjustment and dangerous mutation states that require immediate pausing of grinding, thereby improving the safety and stability of the laser lift-off silicon carbide wafer thinning grinding process.

[0088] S7: Quality evaluation is performed after grinding is completed.

[0089] After grinding, the silicon carbide wafer undergoes post-grinding quality inspection to obtain post-grinding quality data. The post-grinding quality data includes post-grinding... After grinding After grinding Surface roughness Crack defects, edge chipping defects, subsurface damage, and grinding wheel wear (one or more can be selected).

[0090] In some implementations, the intelligent thinning grinding system calculates a grinding quality score based on a post-grinding quality evaluation function: (9); Where Q represents the grinding quality score. , , and After grinding After grinding After grinding And the surface roughness after grinding, , , and These are the corresponding target values. This refers to the wear and tear of the grinding wheel. This is the reference value for grinding wheel wear. to These are the weighting coefficients.

[0091] S8: Update the process database.

[0092] The intelligent thinning grinding system associates and stores pre-grinding surface condition data, initial peeling surface quality grade, initial grinding parameter set, dynamic spindle current window for each stage, spindle current characteristics for each stage, parameter adjustment process for each stage, post-grinding quality results, and grinding wheel wear data to form process sample data.

[0093] Process sample data includes: (10); in, This is process sample data.

[0094] When the post-grinding quality meets the requirements and the grinding wheel wear is below the preset threshold, the system marks the corresponding initial grinding parameter set, dynamic spindle current window for each stage, and feedback adjustment strategy as preferred samples and historically stable grinding samples, and increases their recommended priority under the same morphology level of the modified zone.

[0095] When the post-grind quality does not meet the requirements, or the grinding wheel wear exceeds the preset threshold, the system corrects the three-stage grinding parameter combination under the same initial peeling surface quality level based on the post-grind quality deviation and the spindle current characteristics at each stage.

[0096] Thus, the present invention forms a closed-loop control process including pre-grinding surface condition identification, initial peeling surface quality grade judgment, initial grinding parameter matching, spindle current feedback adjustment during grinding, post-grinding quality evaluation, and process database update.

[0097] In summary, this invention improves the matching accuracy of laser lift-off silicon carbide wafer thinning parameters. This invention is based on the wafer after laser lift-off... , , , , This invention uses surface condition data such as defect distribution to determine the initial surface quality level of the peeled surface and match the corresponding three-stage grinding parameters, avoiding the poor adaptability of traditional fixed parameters or manual experience settings. It achieves staged self-feedback adjustment based on spindle current, establishing dynamic spindle current windows for different grinding stages. Based on the lower limit, stable zone, upper limit, or abrupt change zone of the spindle current, it adaptively adjusts the feed rate and grinding removal amount for the current stage, improving the stability of the grinding process. This invention reduces local overload, surface damage, and abnormal wheel wear. It can promptly identify high-load grinding states caused by local protrusions, uneven residual modified layers, or densely cracked areas based on changes in spindle current, and implements load reduction control, reducing crack propagation, edge chipping, subsurface damage, and abnormal wheel wear. This invention achieves continuous optimization of process parameters, storing pre-grind surface condition, three-stage grinding parameters, spindle current characteristics, post-grind quality, and wheel wear in a correlated manner, and updating the process database based on post-grind results, enabling subsequent similar wafers to obtain better parameter recommendations and feedback adjustment strategies.

[0098] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0099] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital cable) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart thinning grinding method for laser-lifted silicon carbide wafer surfaces, characterized in that, Includes the following processes: Surface state data and grinding wheel state parameters of laser-lifted silicon carbide wafers are obtained, and surface state feature vectors before grinding are constructed. The evaluation value of the peeled surface state is calculated based on the feature vector of the surface state before grinding, and the quality level of the initial peeled surface is determined. Based on the initial peel surface quality level, retrieve the process database to match the three-stage initial grinding parameter set; Based on the initial peeling surface quality grade, grinding parameters of the current grinding stage, and grinding wheel state parameters, historical stable grinding samples are selected from historical processing data, and dynamic spindle current windows corresponding to each grinding stage are calculated and generated based on the selected historical stable grinding samples. During the grinding process, the spindle current of the grinding wheel is collected in real time. Based on the sliding time window, the average spindle current, the increase in spindle current, and the fluctuation range of spindle current are calculated. The average spindle current, the increase in spindle current, and the fluctuation range of spindle current are compared with the dynamic spindle current window to divide the current interval to which the current grinding state belongs. The feed rate and grinding removal amount are adaptively adjusted according to the corresponding range output parameter adjustment command; After grinding is completed, post-grinding quality inspection data is collected, and grinding quality score is calculated based on the post-grinding quality inspection data.

2. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, Construct a surface state feature vector before grinding, including: total thickness variation, wafer warpage, wafer curvature, surface roughness, local peak-to-valley difference, crack area ratio, edge chipping width, and grinding wheel state parameters as feature components to form a surface state feature vector before grinding. The evaluation value of the surface condition of the stripped surface is calculated, including: normalizing the thickness variation, wafer warpage, wafer curvature, surface roughness and local peak-valley difference in the surface condition feature vector before grinding according to the corresponding parameter reference values ​​to obtain each normalized sub-item; Assign corresponding weight coefficients to each normalized component, and multiply and sum the weight coefficients with the corresponding normalized components to obtain the peeling surface condition evaluation value. Based on the peeling surface condition evaluation value, the deviation of each surface condition parameter from the corresponding benchmark value, and in combination with the crack area ratio and edge chipping width, determine the initial peeling surface quality level.

3. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, The three-stage initial grinding parameter set includes the three-stage feed rate, the three-stage grinding removal amount in a decreasing trend, as well as the grinding wheel spindle speed, the table speed, and the polishing time. The grinding wheel spindle speed and the table speed are only used for parameter matching before grinding and grinding stage switching configuration. Real-time adaptive adjustment is not performed during grinding.

4. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, Generate dynamic spindle current windows corresponding to each grinding stage, including: Based on the initial peeling surface quality grade, the feed rate of the current grinding stage, the amount of material removed and the grinding wheel state parameters, historical stable grinding samples that meet the preset matching conditions are selected from the historical processing database, and the average spindle current of the historical stable grinding samples in the corresponding grinding stage is calculated to obtain the stage reference stable spindle current. Extract the standard deviation of current fluctuation within the historical stable grinding data of the same stage, configure the lower limit adjustment coefficient and the upper limit adjustment coefficient respectively, and calculate the lower limit value and the upper limit value of the stage spindle current by combining the stage reference stable spindle current, the standard deviation of current fluctuation, and the two types of adjustment coefficients. The stage spindle current lower limit value and the stage spindle current upper limit value constitute the dynamic spindle current window.

5. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, The spindle current characteristics are calculated based on a sliding time window, including: The arithmetic mean of the grinding wheel spindle current at a fixed number of consecutive sampling times is used to obtain the average spindle current. The difference between the currents of the grinding wheel spindle at two adjacent sampling times is used to calculate the increase in spindle current. Calculate the square mean of the deviation between the grinding wheel spindle current and the average spindle current at each sampling time within the sliding window, and take the square root of the square mean of the deviation to obtain the spindle current fluctuation amplitude.

6. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, If the average value of the spindle current is less than the lower limit of the stage spindle current within the dynamic spindle current window, and the duration reaches the first preset duration, perform additional judgment on the grinding end point. If the cumulative grinding removal amount in the current stage does not reach the grinding removal amount in a single stage, the feed rate or the grinding removal amount in the current stage will be increased by the preset adjustment step size. When the cumulative grinding removal amount in the current stage reaches the single-stage grinding removal amount, the grinding stage switch or grinding is terminated.

7. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, If the average value of the spindle current is within the dynamic spindle current window range, compare the spindle current fluctuation amplitude with the stable grinding fluctuation threshold. When the spindle current fluctuation amplitude is less than or equal to the stable grinding fluctuation threshold, maintain the current grinding parameters unchanged and record the current processing sample; When the spindle current fluctuation exceeds the stable grinding fluctuation threshold, reduce the current stage feed rate or the current stage grinding removal amount, and simultaneously increase the coolant flow rate or output a grinding wheel status check prompt.

8. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, If the average spindle current is greater than the upper limit of the spindle current in the dynamic spindle current window and the duration reaches the third preset duration, an additional safety protection judgment is executed. When the average spindle current reaches the stage safety spindle current threshold, grinding is paused directly; when the average spindle current does not reach the stage safety spindle current threshold, the feed rate or the amount of material removed in the current stage is reduced by a preset adjustment step, and the coolant flow rate is increased. If the increase in spindle current exceeds the threshold of the stage increase, distinguish between controllable mutation and dangerous mutation. In the case of controllable mutation, reduce the feed rate or the amount of material removed during grinding. In the case of dangerous mutation, stop grinding directly.

9. The intelligent thinning grinding method for laser-lifted silicon carbide wafer surfaces as described in claim 1, characterized in that, Associate and store surface condition data, initial peeling surface quality grade, three-stage initial grinding parameter set, dynamic spindle current window, average spindle current, spindle current increase, spindle current fluctuation amplitude, and grinding quality score; update the process database, including: The normalized quality items are obtained by dividing the total thickness change after grinding, the wafer warpage after grinding, the wafer curvature after grinding, and the surface roughness after grinding by the corresponding parameter target values. The grinding wheel wear is obtained by dividing the grinding wheel wear reference value. Corresponding weight coefficients are assigned to all items. The grinding quality score is obtained by multiplying and summing all weight coefficients with the corresponding items. When the grinding quality score meets the standard, the grinding wheel wear is lower than the preset threshold, and the spindle current characteristics of the corresponding grinding stage meet the stable grinding judgment conditions, the corresponding processing parameters and spindle current data are marked as preferred samples and historical stable grinding samples, and their parameter recommendation priority under the same initial peeling surface quality level is increased for subsequent matching of initial grinding parameters of the same type of wafer and calculation of stage reference stable spindle current. When the grinding quality score fails to meet the standard or the grinding wheel wear exceeds the preset threshold, the three-stage initial grinding parameter set that matches the initial peeling surface quality level of the same type is corrected.

10. A smart thinning grinding system for laser-lifted silicon carbide wafer surfaces, characterized in that, include: The surface state data input unit is configured to: acquire the surface state data and grinding wheel state parameters of the laser-lifted silicon carbide wafer, and construct a surface state feature vector before grinding; The peeling surface quality grade judgment unit is configured to: calculate the peeling surface state evaluation value based on the surface state feature vector before grinding, and determine the initial peeling surface quality grade; The initial grinding parameter matching unit is configured to retrieve the three-stage initial grinding parameter set from the process database based on the initial peel surface quality level. The dynamic spindle current window generation unit is configured to: select historical stable grinding samples from historical processing data based on the initial peeling surface quality level, grinding parameters of the current grinding stage, and grinding wheel state parameters, and calculate and generate the dynamic spindle current window corresponding to each grinding stage based on the selected historical stable grinding samples; The spindle current reading and interval judgment unit is configured to: collect the grinding wheel spindle current in real time during the grinding process, calculate the average spindle current, the increase in spindle current, and the fluctuation range of spindle current corresponding to the grinding wheel spindle current based on the sliding time window, compare the average spindle current, the increase in spindle current, and the fluctuation range of spindle current with the dynamic spindle current window, and divide the current interval to which the current grinding state belongs; The grinding parameter feedback adjustment unit is configured to adaptively adjust the feed rate and grinding removal amount in the current stage according to the corresponding range output parameter adjustment command. The post-grinding quality evaluation unit is configured to: collect post-grinding quality inspection data after grinding is completed, and calculate the grinding quality score based on the post-grinding quality inspection data.

Images