Grinding disc inclination angle self-adjusting method and device and wafer thinning equipment

Abstract

The invention provides a grinding disc inclination angle self-adjusting method and device and wafer thinning equipment, and relates to the field of wafer processing and manufacturing. The method comprises the following steps: acquiring a first variable and a second variable which influence the thinning effect of a wafer to be thinned; the first variable and the second variable are input into a pre-trained surface shape prediction model, a predicted surface shape of the wafer to be thinned is obtained, and the predicted surface shape at least comprises the thickness change of the wafer to be thinned after being thinned by the wafer thinning equipment; if the predicted surface shape of the to-be-thinned wafer is not consistent with the set target surface shape, the predicted surface shape and the target surface shape of the to-be-thinned wafer are input into an inclination angle adjusting model, and the target inclination angle of the grinding disc is obtained; and the inclination angle of the grinding disc is adjusted from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle executes thinning machining of the wafer to be thinned. According to the method, the inclination angle of the grinding disc can be accurately and automatically adjusted, and the precision and quality of wafer thinning machining are improved.

Description

Technical Field

[0001] The present application relates to the field of wafer processing and manufacturing, and in particular to a method and device for self-adjusting the inclination angle of a grinding disc and a wafer thinning device. Background Art

[0002] In the field of semiconductor manufacturing, wafer thinning is one of the key steps to improve device performance and reduce costs. The thinning technology in related technologies mainly relies on the experience of equipment operators to adjust the inclination angle of the grinding disc to control the thickness uniformity of the wafer. However, this method has many shortcomings: on the one hand, manual adjustment lacks systematicity and precision, and it is difficult to achieve optimal control of the wafer surface shape; on the other hand, manual operation is time-consuming and labor-intensive, which seriously affects production efficiency and the overall operating efficiency of the equipment. In addition, the requirements for wafer thickness uniformity and processing accuracy are increasing, and the traditional empirical adjustment method can no longer meet the needs of modern production. Summary of the invention

[0003] Based on this, it is necessary to provide a method, device and wafer thinning equipment for self-adjusting the inclination angle of a grinding disk in order to solve the above-mentioned technical problems. The method can accurately and automatically adjust the inclination angle of the grinding disk, thereby improving the accuracy and quality of wafer thinning processing.

[0004] In a first aspect, the present application provides a method for self-adjusting the inclination angle of a grinding disc, the method comprising: Obtaining a first variable and a second variable that affect the thinning effect of the wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; Inputting the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, wherein the predicted surface shape at least includes a thickness change of the wafer to be thinned after being thinned by a wafer thinning device; If the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape, the predicted surface shape of the wafer to be thinned and the target surface shape are input into the inclination adjustment model to obtain the target inclination angle of the grinding disc; The inclination angle of the grinding disc in the wafer thinning equipment is adjusted from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle performs the thinning process of the wafer to be thinned.

[0005] In one embodiment, the tilt adjustment model uses an optimization algorithm; the predicted surface shape and the target surface shape of the wafer to be thinned are input into the tilt adjustment model to obtain the target tilt angle of the grinding disc, including: Determine a tilt angle combination set, the tilt angle combination set includes multiple tilt angle combinations, and any tilt angle combination includes a first angle and a second angle; For each inclination angle combination, the predicted surface shape corresponding to the inclination angle combination is determined using the surface shape prediction model, and the difference between the predicted surface shape of the inclination angle combination and the target surface shape is calculated, and the difference is defined as the fitness value; If none of the fitness values ​​meet the preset conditions, the inclination angle combination is updated in the search space and the fitness value is recalculated; Repeat the iteration until there is a fitness value that meets the preset conditions or reaches the maximum number of iterations, and output the optimal inclination angle combination, which is the inclination angle combination with the smallest fitness value.

[0006] In one embodiment, each inclination angle combination is assigned a speed, and the assigned speed includes a unit angle adjusted in the first angle direction and a unit angle adjusted in the second angle direction; if the fitness values ​​do not meet the preset conditions, the inclination angle combination is updated and the fitness value is recalculated, including: Update the inclination angle combination based on the set speed update method and the set angle update method; For each updated inclination angle combination, the fitness value of the updated inclination angle combination is calculated.

[0007] In one embodiment, the speed update method is set as follows: ; in, is the number of iterations, , , , , is a hyperparameter, Combination of inclination angles exist The historical optimal inclination angle in the iteration, The inclination angle combination is The historical optimal inclination angle in the iteration; The angle update method is: ; in, is the position of the previous iteration, The speed of the next iteration.

[0008] In one embodiment, the operating parameters include a spindle speed, a carrier speed, and a spindle feed speed in a wafer thinning device; The first variable also includes wafer parameters of the wafer to be thinned and auxiliary material parameters used in the thinning process of the wafer to be thinned. The wafer parameters at least include the type of wafer to be thinned, and the auxiliary material parameters at least include the type of auxiliary material and the use time of the auxiliary material.

[0009] In one embodiment, the face shape prediction model uses an extreme gradient boosting algorithm, and training the face shape prediction model includes: According to the wafer parameters, auxiliary material parameters, operation parameters and the inclination angle of the grinding disc in each thinning process, the surface shape prediction model to be trained is used as the input variable, and the surface shape measurement result of each wafer in the surface shape measurement device is searched according to the number matching rule as the output variable of the surface shape prediction model to be trained; An extreme gradient boosting algorithm is used to establish a mapping relationship between input and output, and the face shape prediction model to be trained is trained to obtain a face shape prediction model; Among them, the wafer type of the wafer to be thinned is input into the surface prediction model after one-hot encoding processing, and the wafer type is converted into a unique binary feature vector through one-hot encoding processing. There is only one value 1 in the binary feature vector.

[0010] In one embodiment, an extreme gradient boosting algorithm is used to establish a mapping relationship between input and output, and a face shape prediction model to be trained is trained to obtain a face shape prediction model, including: The extreme gradient boosting algorithm is used to establish the mapping relationship between input and output to obtain the face shape prediction model to be trained; Predicting a training set by using the face shape prediction model to be trained to obtain a prediction set that matches the training set; Based on the deviation between the prediction set and the training set, the face shape prediction model to be trained continues to be trained until the number of training times is reached or the deviation is less than a threshold, then the training is stopped to obtain the face shape prediction model.

[0011] In one embodiment, the present application further provides a wafer thinning device, the wafer thinning device comprising: A grinding disc, which is used to thin the wafer to be thinned; a memory storing a computer program; The processor implements the grinding disc inclination self-adjustment method when executing the computer program.

[0012] In a second aspect, the present application also provides a grinding disc inclination self-adjusting device, the device comprising: An acquisition module, the acquisition module is used to acquire a first variable and a second variable that affect a thinning effect of a wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; A prediction module, the prediction module is used to input the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, the predicted surface shape at least including a thickness change of the wafer to be thinned after thinning by a wafer thinning device; The adjustment module is used to input the predicted surface shape and the target surface shape of the wafer to be thinned into the inclination adjustment model to obtain the target inclination angle of the grinding disk if the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape; the adjustment module is also used to adjust the inclination angle of the grinding disk in the wafer thinning equipment from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle performs the thinning process of the wafer to be thinned.

[0013] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the grinding disc inclination self-adjustment method of the first aspect.

[0014] The above-mentioned method for self-adjusting the inclination angle of the grinding disk, the method obtains a first variable and a second variable that affect the thinning effect of the wafer to be thinned, wherein the first variable at least includes the operating parameters of the wafer thinning equipment, and the second variable is the initial inclination angle of the grinding disk. These two variables are input into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, and the predicted surface shape at least includes the thickness change after the wafer is thinned. If the predicted surface shape is inconsistent with the set target surface shape, the predicted surface shape and the target surface shape are input into the inclination adjustment model to obtain the target inclination angle of the grinding disk. The inclination angle of the grinding disk is adjusted from the initial inclination angle to the target inclination angle, and the adjusted wafer thinning equipment performs the thinning process. This method can realize accurate and automatic adjustment of the inclination angle of the grinding disk, and improve the accuracy and quality of the wafer thinning process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flow chart of a method for self-adjusting the inclination angle of a grinding disc in one embodiment; Figure 2 A flow chart of obtaining a target inclination angle of a grinding disc using an optimization algorithm in one embodiment; Figure 3 A flowchart of updating the inclination angle combination and recalculating the fitness value in one embodiment; Figure 4 A flowchart of obtaining a face shape prediction model in one embodiment; Figure 5 A structural diagram of a wafer thinning device in one embodiment; Figure 6 A diagram of a grinding disc inclination self-adjusting device in one embodiment. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0017] In one embodiment, Figure 1As shown, a method for self-adjusting the inclination angle of a grinding disc is provided, the method comprising the following steps: Step 101: obtaining a first variable and a second variable that affect the thinning effect of the wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; The first variable may include operating parameters of the wafer thinning equipment, such as spindle speed, carrier speed, spindle feed speed, etc. The operating parameters of the wafer thinning equipment may determine the basic conditions and processing rate of the thinning operation. Furthermore, the first variable may also include wafer type and auxiliary material information. Among them, the wafer type may be a code pre-defined according to the processing process and wafer size; the auxiliary material information includes but is not limited to the type and life of the grinding wheel during processing.

[0018] The second variable may be the initial inclination angle of the grinding disc in the wafer thinning equipment, which is an adjusted target parameter. By adjusting the initial inclination angle of the grinding disc, the thickness uniformity of the wafer can be optimized, thereby improving the overall quality of the thinning effect. For example, appropriately increasing the initial inclination angle of the grinding disc can enhance the cutting effect of the abrasive on the edge area of ​​the wafer, which helps to improve the uniformity of the edge thickness; while reducing the inclination angle may make the thinning of the center area more sufficient.

[0019] Step 102: inputting the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, wherein the predicted surface shape at least includes a thickness change of the wafer to be thinned after being thinned by a wafer thinning device; It should be noted that the surface shape prediction model can be obtained by training with a large amount of historical data. During the training process, various data in the thinning process history are collected, including wafer type, auxiliary material information, process parameter settings, carrier inclination angle, etc. as input variables, and the corresponding surface shape measurement results as output variables. Furthermore, a machine learning algorithm, such as an extreme gradient boosting algorithm, can be used to establish a mapping relationship between input and output. The extreme gradient boosting algorithm iteratively constructs a tree model by optimizing the objective function, and optimizes the structure of the tree through a greedy algorithm to obtain a powerful integrated model. During the training process, the model continuously learns the relationship between the input variables and the surface shape results to improve the accuracy of the prediction. After sufficient training and optimization, when the fitting error of the model is minimized, the surface shape prediction model is obtained.

[0020] When the first variable and the second variable are input into the pre-trained surface shape prediction model, the surface shape prediction model can predict the surface shape of the wafer after thinning based on its internal algorithm and the knowledge learned in the previous training process. The predicted surface shape can include the thickness change of the wafer after thinning, that is, the total thickness change (TTV). The surface shape prediction model can comprehensively consider the impact of operating parameters and initial tilt angle on the thinning process, and output a predicted surface shape result through complex calculation and analysis.

[0021] For example, if the predicted surface shape is consistent with the target surface shape, the current operating parameters and initial tilt angle settings in Table 4 are reasonable, and the thinning process can continue. On the contrary, if there is a deviation between the predicted surface shape and the target surface shape, the tilt angle of the grinding disc needs to be adjusted to ensure that the final wafer surface shape meets the requirements.

[0022] Step 103: if the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape, the predicted surface shape of the wafer to be thinned and the target surface shape are input into the inclination adjustment model to obtain the target inclination angle of the grinding disc; It should be noted that the tilt adjustment model can be constructed based on the particle swarm optimization algorithm. In order to obtain the tilt adjustment model, a large amount of historical data needs to be collected, including the wafer surface shape results under different tilt angle combinations and the corresponding target surface shapes. A large amount of historical data is used to train the tilt adjustment model so that the tilt adjustment model learns how to adjust the tilt according to the surface shape differences. During the training process, the tilt adjustment model can continuously try different tilt angle combinations, calculate the degree of match between the surface shape and the target surface shape under each combination, and optimize the tilt angle combination according to the degree of match. Through repeated iterations, the tilt adjustment model can learn the optimal tilt adjustment strategy, and then accurately output the optimal grinding disk target tilt angle based on the input predicted surface shape and target surface shape.

[0023] When the predicted surface shape and the target surface shape are input into the pre-trained tilt adjustment model, the tilt adjustment model can quickly calculate a target tilt angle of the grinding disc that can make the predicted surface shape closer to the target surface shape based on the previously learned optimization strategy. The target tilt angle can accurately guide the wafer thinning equipment to automatically adjust the tilt angle of the grinding disc, so that in the subsequent thinning process, the actual surface shape of the wafer to be thinned is closer to or even reaches the target surface shape, effectively improving the processing quality and production efficiency of the wafer.

[0024] Step 104: adjusting the inclination angle of the grinding disc from the initial inclination angle to the target inclination angle, and performing a thinning process on the wafer to be thinned by the wafer thinning equipment adjusted to the target inclination angle.

[0025] After the target inclination angle of the grinding disc is determined, the inclination angle of the grinding disc can be adjusted. The process of adjusting the inclination angle of the grinding disc can be completed by the automatic adjustment system of the wafer thinning equipment. Furthermore, based on the target inclination angle output by the inclination adjustment model, the inclination angle of the grinding disc is adjusted from the initial inclination angle to the target inclination angle. After the adjustment is completed, the wafer thinning equipment is in a new processing condition, and the wafer thinning equipment can perform thinning processing on the wafer to be thinned according to the established operating parameters and the new grinding disc inclination angle setting. During the processing, the grinding disc applies pressure to the wafer at the target inclination angle and grinds it, so that the wafer can remove material according to the expected surface shape changes during the thinning process to achieve a processing effect closer to the target surface shape.

[0026] In this embodiment, the method obtains a first variable and a second variable that affect the thinning effect of the wafer to be thinned, wherein the first variable includes at least the operating parameters of the wafer thinning equipment, and the second variable is the initial inclination angle of the grinding disc in the wafer thinning equipment. These two variables are input into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, and the predicted surface shape includes at least the thickness change after the wafer is thinned. If the predicted surface shape is inconsistent with the set target surface shape, the predicted surface shape and the target surface shape are input into the inclination adjustment model to obtain the target inclination angle of the grinding disc. The inclination angle of the grinding disc is adjusted from the initial inclination angle to the target inclination angle, and the adjusted wafer thinning equipment performs the thinning process. This method can realize accurate and automatic adjustment of the inclination angle of the grinding disc, and improve the accuracy and quality of the wafer thinning process.

[0027] In one embodiment, Figure 2 As shown, the tilt adjustment model adopts an optimization algorithm; the predicted surface shape and the target surface shape of the wafer to be thinned are input into the tilt adjustment model to obtain the target tilt angle of the grinding disc, including the following steps: Step 201: determining a tilt angle combination set, wherein the tilt angle combination set includes a plurality of tilt angle combinations, and any tilt angle combination includes a first angle and a second angle; The bottom of the grinding disc usually has a plurality of legs. By adjusting the height of the legs, the grinding disc can be tilted in different directions. The first angle and the second angle can correspond to the tilt angles of the grinding disc in two orthogonal directions, such as the tilt along the length and width of the grinding disc. It should be noted that the tilt angle combination set can be obtained by collecting historical data, which can come from the past Step 202: for each inclination angle combination, using the surface shape prediction model to determine the predicted surface shape corresponding to the inclination angle combination, and calculating the difference between the predicted surface shape of the inclination angle combination and the target surface shape, and defining the difference as a fitness value; The first angle and the second angle in each tilt angle combination are used as input variables and input into the surface shape prediction model in combination with the first variable. The surface shape prediction model can predict the surface shape of the wafer after thinning processing based on its internal algorithm and the knowledge learned in the previous training process, and output the corresponding predicted surface shape.

[0028] After obtaining the predicted surface shape corresponding to each inclination angle combination, the predicted surface shape is compared with the set target surface shape, the difference between the two is calculated, and the difference is defined as the fitness value.

[0029] It should be noted that the size of the fitness value can reflect the degree of matching between the predicted wafer surface shape under the tilt angle combination and the desired target surface shape. The smaller the fitness value, the more the tilt angle combination can make the actual processed wafer surface shape close to the target surface shape.

[0030] Step 203: If none of the fitness values ​​meet the preset conditions, the inclination angle combination is updated in the search space, and the fitness value is recalculated; Compare these fitness values ​​with the preset conditions. If it is found that the fitness values ​​of all the inclination angle combinations do not meet the preset conditions, it means that there is no suitable solution in the current inclination angle combination set that can make the predicted surface shape close to the target surface shape. The preset conditions can be a variety of specific judgment criteria, such as setting an absolute error threshold, which requires that the difference in thickness change between the predicted surface shape and the target surface shape is less than a preset value, such as 0.01 mm.

[0031] If the fitness values ​​do not meet the preset conditions, it is necessary to update the inclination angle combination in the search space. The search space may refer to a set of all possible inclination angle combinations, including the angle range of the first angle and the second angle. It should be noted that the method for updating the inclination angle combination can be based on a variety of optimization algorithms, such as a particle swarm optimization algorithm. Exemplarily, in the particle swarm optimization algorithm, each inclination angle combination can be regarded as a particle, which moves and adjusts its position in the search space to find a better solution. After the updated inclination angle combination set is generated, the surface shape prediction model can be used again to evaluate each new inclination angle combination, determine its corresponding predicted surface shape, and recalculate the fitness value.

[0032] Step 204: Repeat the iteration until there is a fitness value that meets the preset condition or the maximum number of iterations is reached, and output the optimal tilt angle combination, which is the tilt angle combination with the smallest fitness value.

[0033] In each iteration, the surface shape prediction model is used to evaluate each inclination angle combination in the current inclination angle combination set to obtain its corresponding predicted surface shape, and the difference between the predicted surface shape and the target surface shape is calculated as the fitness value. If in the current iteration, the fitness value of a certain inclination angle combination meets the preset conditions, that is, the difference between its predicted surface shape and the target surface shape is within the allowed range, the iteration process stops, and the inclination angle combination is the optimal inclination angle combination.

[0034] If no inclination angle combination meets the conditions in the current iteration, the inclination angle combination set is updated according to the optimization algorithm (such as particle swarm optimization algorithm), a new inclination angle combination is generated, and their fitness values ​​are recalculated. This process is repeated until a inclination angle combination that meets the conditions is found or the maximum number of iterations is reached. When the maximum number of iterations is reached, even if no inclination angle combination meets the preset conditions, the inclination angle combination with the smallest fitness value in all iterations will be output as the optimal inclination angle combination.

[0035] In this embodiment, the method can achieve precise adjustment of the inclination angle of the grinding disc, thereby improving the quality and efficiency of the wafer thinning process.

[0036] In one embodiment, Figure 3 As shown, each inclination angle combination is assigned a speed, and the assigned speed includes a unit angle adjusted in the first angle direction and a unit angle adjusted in the second angle direction; if the fitness values ​​do not meet the preset conditions, the inclination angle combination is updated and the fitness value is recalculated, including the following steps: Step 301: updating the inclination angle combination based on a set speed update method and a set angle update method; The speed update method can adjust the speed of each inclination angle combination (regarded as a particle), and the speed of each inclination angle can determine the moving direction and distance of the particle in the search space. The speed update takes into account the particle's own best position (that is, the particle's best historical inclination angle combination) and the best position of the entire group (that is, the inclination angle combination with the smallest fitness value among all particles). By combining the best position information of itself and the group, the particle can determine a suitable speed so that it can more effectively explore the possible solution space in subsequent searches.

[0037] The angle update method can be to adjust the position (i.e., angle value) of the tilt angle combination according to the updated speed. The position update means that the new tilt angle combination (position) is the current position plus the updated speed. In this way, each tilt angle combination moves in the search space to explore a better solution.

[0038] Step 302: For each updated inclination angle combination, calculate the fitness value of the updated inclination angle combination.

[0039] After each update of the tilt angle combination, each new tilt angle combination is input into the surface shape prediction model, and the model is used to predict the wafer surface shape corresponding to the tilt angle combination. This predicted surface shape is compared with the target surface shape, and the difference between the two is calculated, which is the fitness value. By calculating the fitness value of each updated tilt angle combination, the advantages and disadvantages of different tilt angle combinations can be quantitatively evaluated, and then the optimal tilt angle combination can be selected in the subsequent iteration process to achieve precise adjustment of the grinding disc tilt angle.

[0040] In this embodiment, by updating the speed and angle and calculating the fitness value, the advantages and disadvantages of each inclination angle combination can be quantitatively evaluated, so that the optimal combination can be screened out during the iteration process to achieve precise adjustment of the grinding disc inclination angle.

[0041] In one embodiment, the speed update method is set as follows: ; in, is the number of iterations, , , , , is a hyperparameter, Combination of inclination angles exist The historical optimal inclination angle in the iteration, The inclination angle combination is The historical optimal inclination angle in the iteration; The angle update method is: ; in, is the position of the previous iteration, The speed of the next iteration.

[0042] The update of the inclination combination is based on the set speed update method and angle update method. The speed update method comprehensively considers the historical optimal position of the particle and the historical optimal position of the group, and determines the moving direction and distance of the particle in the search space by adjusting parameters such as inertia weight, learning factor and random number. The speed update formula is: .

[0043] in, is the number of iterations, , , , , is a hyperparameter, Combination of inclination angles exist The historical optimal inclination angle in the iteration, The inclination angle combination is The historical optimal inclination in the iteration. It should be noted that Used to balance the effect of the particle's current speed on the next step's speed; , are learning factors, representing the particle’s ability to learn its own historical optimal position and the group’s historical optimal position. The angle update method adjusts the position of the inclination angle combination according to the updated speed. The formula is: .

[0044] in, is the position of the last iteration, i.e., the current inclination angle combination; is the speed of the next iteration, that is, the new moving direction and distance calculated according to the speed update formula.

[0045] The process of updating the angle can be understood as the particle moving from the current position to the new position according to its new velocity. This step actually generates new inclination angle combinations in the search space to explore possible better solutions.

[0046] In this embodiment, in this way, it is possible to systematically explore better inclination angle combinations in the search space, gradually approaching the optimal solution that makes the predicted surface shape closest to the target surface shape, thereby achieving precise adjustment of the grinding disk inclination angle and improving the accuracy and quality of wafer thinning processing.

[0047] In one embodiment, the operating parameters include a spindle speed, a carrier speed, and a spindle feed speed in a wafer thinning device; The first variable also includes wafer parameters of the wafer to be thinned and auxiliary material parameters used in the thinning process of the wafer to be thinned. The wafer parameters at least include the type of wafer to be thinned, and the auxiliary material parameters at least include the type of auxiliary material and the use time of the auxiliary material.

[0048] The operating parameters include the spindle speed, the carrier speed and the spindle feed speed in the wafer thinning equipment. The operating parameters can determine the basic conditions and processing rate of the thinning operation. The first variable also includes the wafer parameters of the wafer to be thinned and the auxiliary material parameters used in the thinning process of the wafer to be thinned. The wafer parameters at least include the type of wafer to be thinned. Different types of wafers may have different physical and chemical properties during the thinning process, such as hardness, thickness, material composition, etc. These properties can affect the efficiency and effect of thinning. The auxiliary material parameters at least include the type of auxiliary material and the use time of the auxiliary material. The type of auxiliary material can determine its action mode and effect on the wafer during the thinning process. For example, the particle size and hardness of the abrasive will affect the rate of material removal and the surface quality; the use time of the auxiliary material will affect the stability of its performance. As the use time increases, the auxiliary material may wear or fail, thereby affecting the thinning effect.

[0049] In this embodiment, a plurality of parameters together constitute a first variable, which is closely related to the operating state of the wafer thinning equipment and has an important influence on the thinning process and the final wafer surface shape.

[0050] In one embodiment, the surface shape prediction model adopts an extreme gradient boosting algorithm, and uses wafer parameters, auxiliary material parameters, operating parameters, and the inclination angle of the grinding disc in each thinning process as input variables of the surface shape prediction model to be trained, and obtains the surface shape measurement results of each wafer according to the number matching rule as the output variable of the surface shape prediction model to be trained; the extreme gradient boosting algorithm is used to establish a mapping relationship between the input and the output, and the surface shape prediction model to be trained is trained to obtain the surface shape prediction model; Among them, the wafer type of the wafer to be thinned is input into the surface prediction model after one-hot encoding processing, and the wafer type is converted into a unique binary feature vector through one-hot encoding processing. There is only one value 1 in the binary feature vector.

[0051] The surface shape prediction model can be trained using the extreme gradient boosting algorithm to accurately predict the surface shape of the wafer based on multiple parameters in each thinning process. The wafer parameters, auxiliary material parameters, operating parameters, and the inclination angle of the grinding disc are used as input variables for the surface shape prediction model to be trained. These parameters comprehensively cover the key influencing factors in the thinning process, among which the wafer parameters include the type of wafer to be thinned, the auxiliary material parameters involve the type and usage time of the auxiliary materials, and the operating parameters include the spindle speed, carrier speed, and spindle feed speed of the wafer thinning equipment, and the inclination angle of the grinding disc is also an important influencing factor.

[0052] In order to obtain the output variable of the surface shape prediction model, that is, the surface shape measurement result of the wafer, the surface shape data corresponding to each wafer is obtained according to the number matching rule. The surface shape measurement result can reflect the thickness change of the wafer after thinning processing and is a key indicator for evaluating processing quality.

[0053] The model is trained using the extreme gradient boosting algorithm. The extreme gradient boosting algorithm is a decision tree ensemble learning algorithm based on gradient boosting. It optimizes the objective function, iteratively builds multiple decision trees, and combines the prediction results of these trees to improve the prediction accuracy and generalization ability of the model. During the training process, the extreme gradient boosting algorithm learns the influence pattern of the input parameters on the polygon based on the relationship between the input variables and the output variables in the training samples, thereby establishing a mapping relationship between the input and the output.

[0054] Specifically, the extreme gradient boosting algorithm uses the training set and the true value of the sample to train the first tree, and uses this tree to predict the training set to obtain the predicted value of each sample. Since there is a deviation between the predicted value and the true value, the "residual" can be obtained by subtracting the two. Then the second tree is trained. At this time, the true value is no longer used, but the residual is used as the standard answer. After the two trees are trained, the residual of each sample can be obtained again, and the third tree can be further trained, and so on. The total number of trees can be specified manually, or the training can be stopped by monitoring certain indicators (such as the error on the validation set). When predicting a new sample, each tree will have an output value. The output values ​​of each tree are added together to obtain the final predicted value of the sample.

[0055] Furthermore, for the categorical variable of the wafer type of the wafer to be thinned, one-hot encoding can be used before input into the model. One-hot encoding converts each wafer type into a unique binary feature vector, where only one value is 1 and the rest are 0. For example, if there are three wafer types A, B, and C, A can be encoded as [1,0,0], B as [0,1,0], and C as [0,0,1]. This processing method avoids introducing false order relationships between categories, allowing the model to more accurately identify and utilize the differences between different wafer types, thereby improving the accuracy of predictions.

[0056] In this embodiment, the method can obtain a fully trained surface shape prediction model, which can quickly and accurately predict the surface shape of the wafer to be thinned according to the input processing parameters and wafer type in actual production, thereby achieving precise control of the wafer thinning process.

[0057] In one embodiment, Figure 4 As shown, an extreme gradient boosting algorithm is used to establish a mapping relationship between input and output, and the face shape prediction model to be trained is trained to obtain a face shape prediction model, including the following steps: Step 401: using an extreme gradient boosting algorithm to establish a mapping relationship between input and output, and obtaining a face shape prediction model to be trained; The input variables include the operating parameters of the wafer thinning equipment, the wafer parameters of the wafer to be thinned, the parameters of the auxiliary materials used in the thinning process, and the inclination angle of the grinding disc. The output variables are the surface measurement results of each wafer obtained according to the number matching rules, including at least the thickness change after wafer thinning.

[0058] During the training process, the extreme gradient boosting algorithm can learn the influence pattern of input parameters on the face shape based on the relationship between input variables and output variables, establish a mapping relationship between input and output, and then obtain the face shape prediction model to be trained.

[0059] It should be noted that the extreme gradient boosting algorithm can initialize a decision tree and determine the structure and parameters of the tree by minimizing the objective function. The objective function usually includes a loss function (such as mean square error) and a regularization term to balance the fitting ability and complexity of the face shape prediction model. Furthermore, based on the residual between the prediction result of the current tree and the true output, the next tree is trained, and so on, until the preset number of trees or the convergence condition of the algorithm is reached.

[0060] Step 402: predicting a training set using the face shape prediction model to be trained to obtain a prediction set that matches the training set; The training set may include the operating parameters of the wafer thinning equipment, the wafer parameters of the wafer to be thinned, the auxiliary material parameters used in the thinning process, and the inclination angle of the grinding disc as input variables. The surface measurement results of each wafer obtained according to the number matching rule are used as output variables.

[0061] Through the extreme gradient boosting algorithm, the surface shape prediction model learns the mapping relationship between input variables and output variables. After the training is completed, the surface shape prediction model is used to predict the training set, that is, the input variables of the training set are input into the model, and the surface shape prediction model can output the corresponding predicted surface shape results, which constitute the predicted surface shape results. Furthermore, the predicted set is compared with the output variables of the training set to evaluate the performance and accuracy of the surface shape prediction model, ensuring that the surface shape prediction model can accurately predict the surface shape after wafer thinning.

[0062] Step 403: Based on the deviation between the prediction set and the training set, continue to train the face shape prediction model to be trained until the number of training times is reached or the deviation is less than a threshold, then stop the training and obtain the face shape prediction model.

[0063] The training set is used to perform preliminary training on the surface prediction model, and the surface prediction model is used to predict the training set to obtain a prediction set. The output variables of the prediction set and the training set are compared, and the deviation between the output variables of the prediction set and the training set is calculated. If the deviation is greater than the set threshold and the number of training times has not reached the preset maximum value, the surface prediction model is continued to be trained, and the parameters of the surface prediction model are adjusted to reduce the deviation, where the threshold can be the maximum allowable value of the deviation between the prediction set and the training set. This process will be repeated until the deviation is less than the threshold or the maximum number of training times is reached. At this time, the surface prediction model training is completed, and the final surface prediction model is obtained. The surface prediction model can more accurately predict the surface shape after wafer thinning.

[0064] In this embodiment, this method can effectively improve the prediction accuracy and generalization ability of the surface shape prediction model, and ensure that the surface shape prediction model can accurately predict the surface shape of the wafer after thinning in practical applications.

[0065] In one embodiment, Figure 5 As shown, a wafer thinning device 50 is provided, and the wafer thinning device 50 includes: A grinding disc 51, which is used to thin the wafer to be thinned; A memory 52, the memory stores a computer program; The processor 53 implements the grinding disc inclination self-adjustment method when executing the computer program.

[0066] Specifically, a wafer thinning device 50 is designed to achieve automatic adjustment of the inclination angle of the grinding disc to optimize the wafer thinning effect. The device is mainly composed of a grinding disc 51, a memory 52 and a processor 53. The grinding disc 51 is a key component that directly contacts the wafer to be thinned and performs thinning processing. The precise adjustment of its inclination angle plays a decisive role in the uniformity of the wafer surface shape and the processing quality. The memory 52 is used to store a computer program containing a self-adjustment method for the inclination angle of the grinding disc, which is the core software support for realizing the automatic adjustment function. The processor 43 is the key hardware for executing the program in the memory. When the processor 53 runs the program, the self-adjustment method of the inclination angle of the grinding disc will be implemented according to the preset algorithm and logic.

[0067] Based on the same concept, Figure 6 As shown, the present application also provides a grinding disc inclination self-adjusting device, the device comprising: An acquisition module 601 is used to acquire a first variable and a second variable that affect a thinning effect of a wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; Prediction module 602, the prediction module is used to input the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, the predicted surface shape at least including a thickness change of the wafer to be thinned after thinning by a wafer thinning device; Adjustment module 603, if the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape, the adjustment module is used to input the predicted surface shape and the target surface shape of the wafer to be thinned into the inclination adjustment model to obtain the target inclination angle of the grinding disk; the adjustment module is also used to adjust the inclination angle of the grinding disk from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle performs the thinning process of the wafer to be thinned.

[0068] In one embodiment, the adjustment module 603 inputs the predicted surface shape and target surface shape of the wafer to be thinned into the inclination adjustment model to obtain the target inclination angle of the grinding disk, which is specifically used to: determine a set of inclination angle combinations, the inclination angle combination set includes multiple inclination angle combinations, and any inclination angle combination includes a first angle and a second angle; for each inclination angle combination, use the surface shape prediction model to determine the predicted surface shape corresponding to the inclination angle combination, and calculate the difference between the predicted surface shape and the target surface shape of the inclination angle combination, and define the difference as a fitness value; if the fitness values ​​do not meet the preset conditions, update the inclination angle combination in the search space and recalculate the fitness value; repeat the iteration until there is a fitness value that meets the preset conditions or the maximum number of iterations is reached, and output the optimal inclination angle combination, and the optimal inclination angle combination is the inclination angle combination with the smallest fitness value.

[0069] In one embodiment, the adjustment module 603 assigns a speed to each inclination angle combination, and the assigned speed includes a unit angle adjusted in the first angle direction and a unit angle adjusted in the second angle direction; if the fitness values ​​do not meet the preset conditions, the inclination angle combination is updated and the fitness value is recalculated, specifically for: updating the inclination angle combination based on the set speed update method and the set angle update method; for each updated inclination angle combination, calculating the fitness value of the updated inclination angle combination.

[0070] In one embodiment, the speed update mode set by the adjustment module 603 is: ; in, is the number of iterations, , , , , is a hyperparameter, Combination of inclination angles exist The historical optimal inclination angle in the iteration, The inclination angle combination is The historical optimal inclination angle in the iteration; the angle update method is set as: ;in, is the position of the previous iteration, The speed of the next iteration.

[0071] In one embodiment, the adjustment module 603 is specifically used for: the operating parameters include the spindle speed, the carrier speed and the spindle feed speed in the wafer thinning equipment; the first variable also includes the wafer parameters of the wafer to be thinned and the auxiliary material parameters used in the thinning process of the wafer to be thinned, the wafer parameters at least include the type of wafer to be thinned, and the auxiliary material parameters at least include the type of auxiliary material and the usage time of the auxiliary material.

[0072] In one embodiment, the adjustment module 603 is specifically used for: the surface shape prediction model adopts the extreme gradient boosting algorithm, and uses the wafer parameters, auxiliary material parameters, operating parameters and the inclination angle of the grinding disc in each thinning process as the input variables of the surface shape prediction model to be trained, and searches for the surface shape measurement results of each wafer in the surface shape measurement device according to the number matching rule as the output variable of the surface shape prediction model to be trained; the extreme gradient boosting algorithm is used to establish a mapping relationship between the input and the output, and the surface shape prediction model to be trained is trained to obtain the surface shape prediction model; wherein, the wafer type of the wafer to be thinned is input into the surface shape prediction model after one-hot encoding processing, and the wafer type is converted into a unique binary feature vector through one-hot encoding processing, and only one value in the binary feature vector is 1.

[0073] In one embodiment, the adjustment module 603 uses an extreme gradient boosting algorithm to establish a mapping relationship between input and output, trains the face shape prediction model to be trained, and obtains the face shape prediction model, which is specifically used to: use an extreme gradient boosting algorithm to establish a mapping relationship between input and output, and obtain the face shape prediction model to be trained; predict the training set by the face shape prediction model to be trained to obtain a prediction set with the training set; based on the deviation between the prediction set and the training set, continue to train the face shape prediction model to be trained until the number of training times is reached or the deviation is less than a threshold, then stop training, and obtain the face shape prediction model.

[0074] Based on the same concept, the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for self-adjusting the inclination angle of a grinding disc is implemented.

[0075] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for self-adjusting the inclination angle of a grinding disc, characterized in that: The method comprises: Obtaining a first variable and a second variable that affect the thinning effect of the wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; Inputting the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, wherein the predicted surface shape at least includes a thickness change of the wafer to be thinned after thinning by the wafer thinning equipment; If the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape, the predicted surface shape of the wafer to be thinned and the target surface shape are input into the inclination adjustment model to obtain the target inclination angle of the grinding disc; The inclination angle of the grinding disc is adjusted from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle performs the thinning process on the wafer to be thinned.

2. The grinding disc inclination self-adjusting method according to claim 1, characterized in that: The tilt angle adjustment model adopts an optimization algorithm; Inputting the predicted surface shape of the wafer to be thinned and the target surface shape into the inclination adjustment model to obtain the target inclination angle of the grinding disc comprises: Determine a tilt angle combination set, wherein the tilt angle combination set includes a plurality of tilt angle combinations, and any of the tilt angle combinations includes a first angle and a second angle; For each inclination angle combination, using the surface shape prediction model to determine a predicted surface shape corresponding to the inclination angle combination, and calculating a difference between the predicted surface shape of the inclination angle combination and the target surface shape, and defining the difference as a fitness value; If none of the fitness values ​​meet the preset conditions, the inclination angle combination is updated in the search space, and the fitness value is recalculated; Repeat the iteration until there is a fitness value that meets the preset condition or reaches the maximum number of iterations, and output the optimal inclination angle combination, where the optimal inclination angle combination is the inclination angle combination with the smallest fitness value.

3. The grinding disc inclination self-adjusting method according to claim 2, characterized in that: Each tilt angle combination is assigned a speed, and the assigned speed includes a unit angle adjusted in the first angle direction and a unit angle adjusted in the second angle direction; If the fitness values ​​do not meet the preset conditions, the inclination angle combination is updated and the fitness value is recalculated, including: Update the inclination angle combination based on a set speed update method and a set angle update method; For each updated inclination angle combination, the fitness value of the updated inclination angle combination is calculated.

4. The grinding disc inclination self-adjusting method according to claim 3, characterized in that: The speed update method of the setting is: ； in, is the number of iterations, , , , , is a hyperparameter, Combination of inclination angles exist The historical optimal inclination angle in the iteration, The inclination angle combination is The historical optimal inclination angle in the iteration; The setting angle update method is: ; in, is the position of the previous iteration, The speed of the next iteration.

5. The method for self-adjusting the inclination angle of a crystal grinding disc according to claim 1, characterized in that: The operating parameters include spindle speed, carrier speed and spindle feed speed in the wafer thinning equipment; The first variable also includes wafer parameters of the wafer to be thinned and auxiliary material parameters used in the thinning process of the wafer to be thinned. The wafer parameters at least include the type of the wafer to be thinned, and the auxiliary material parameters at least include the type of auxiliary material and the usage time of the auxiliary material.

6. The grinding disc inclination self-adjusting method according to claim 5, characterized in that: The face shape prediction model adopts an extreme gradient boosting algorithm, and training the face shape prediction model includes: According to the wafer parameters, auxiliary material parameters, operating parameters and the inclination angle of the grinding disc in each thinning process, as the input variables of the surface shape prediction model to be trained, according to the number matching rule, the surface shape measurement result of each wafer in the surface shape measurement device is searched as the output variable of the surface shape prediction model to be trained; The extreme gradient boosting algorithm is used to establish a mapping relationship between input and output, and the face shape prediction model to be trained is trained to obtain the face shape prediction model; Among them, the wafer type of the wafer to be thinned is input into the surface prediction model after one-hot encoding processing, and the wafer type is converted into a unique binary feature vector through one-hot encoding processing, and only one value in the binary feature vector is 1.

7. The grinding disc inclination self-adjusting method according to claim 6, characterized in that: The extreme gradient boosting algorithm is used to establish a mapping relationship between input and output, and the face shape prediction model to be trained is trained to obtain the face shape prediction model, including: The extreme gradient boosting algorithm is used to establish a mapping relationship between input and output to obtain a face shape prediction model to be trained; Predicting a training set by the face shape prediction model to be trained to obtain a prediction set corresponding to the training set; Based on the deviation between the prediction set and the training set, the face shape prediction model to be trained is continuously trained until the number of training times is reached or the deviation is less than a threshold, then the training is stopped to obtain the face shape prediction model.

8. A wafer thinning device, characterized in that: The wafer thinning equipment comprises: A grinding disc, wherein the grinding disc is used to thin the wafer to be thinned; a memory storing a computer program; A processor, wherein when executing the computer program, the processor implements the grinding disc inclination self-adjustment method according to any one of claims 1 to 7.

9. A grinding disc inclination self-adjusting device, characterized in that: The device comprises: An acquisition module, the acquisition module is used to acquire a first variable and a second variable that affect the thinning effect of the wafer to be thinned; wherein the first variable at least includes an operating parameter of a wafer thinning device, the wafer thinning device is configured to perform a thinning process on the wafer to be thinned, and the second variable includes an adjusted target parameter, the target parameter being an initial inclination angle of the grinding disc; A prediction module, the prediction module is used to input the first variable and the second variable into a pre-trained surface shape prediction model to obtain a predicted surface shape of the wafer to be thinned, the predicted surface shape at least including a thickness change of the wafer to be thinned after thinning by the wafer thinning equipment; An adjustment module is provided. If the predicted surface shape of the wafer to be thinned is inconsistent with the set target surface shape, the adjustment module is used to input the predicted surface shape of the wafer to be thinned and the target surface shape into the inclination adjustment model to obtain the target inclination angle of the grinding disk; the adjustment module is also used to adjust the inclination angle of the grinding disk in the wafer thinning equipment from the initial inclination angle to the target inclination angle, and the wafer thinning equipment adjusted to the target inclination angle performs the thinning process of the wafer to be thinned.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the grinding disc inclination self-adjustment method according to any one of claims 1 to 7 is implemented.

Images