Scribing saw cutting pressure control method and control device based on dynamic pressure prediction and scribing saw

Abstract

The embodiment of the invention provides a dicing saw cutting pressure control method and device based on dynamic pressure prediction and a dicing saw. The method comprises the steps that S1, real-time material characteristic parameters on a cutting path are obtained through a multi-source sensor; s2, based on the material characteristic parameters, a dynamic mapping model of material characteristics and cutting pressure is established, and output parameters are predicted pressure compensation values; s3, before a cutting tool reaches a pressure sudden change area, cutting pressure is adjusted in advance according to a predicted pressure compensation value output by the dynamic mapping model, meanwhile, an actual pressure value is detected in real time through a force sensor, and pressure deviation is corrected in combination with a control algorithm; and S4, carrying out staged pressure loading on the non-uniform doping region in the heterogeneous material. According to the method, the pressure sudden change point is predicted through the dynamic mapping model, the pressure is adjusted in advance in combination with a control algorithm, and the characteristics of the adaptive heterogeneous material are loaded in stages, so that the shear mark uniformity is remarkably improved, and the material damage risk is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pressure control, in particular to a wafer saw cutting pressure control method based on dynamic pressure prediction, a control device and a wafer saw. BACKGROUND

[0002] Semiconductor wafer cutting is a key process in chip manufacturing. With the wide application of heterogeneous materials and ultra-thin wafers, traditional cutting technology is facing severe challenges. The internal heterogeneity of the material, due to uneven doping concentration or sudden change in interlayer hardness, can easily cause pressure to change suddenly during cutting, resulting in uneven kerf width, micro-crack propagation or even wafer cracking, which directly affects chip yield and reliability.

[0003] Existing technologies mostly rely on fixed pressure or single feedback control, which cannot predict changes in material properties in real time, and the pressure adjustment lags behind the actual cutting requirements. Especially when dealing with internal heterogeneous interfaces of wafers, edge collapse or kerf deviation can easily occur due to stress concentration.

[0004] The embodiments of the present application provide a wafer saw cutting pressure control method based on dynamic pressure prediction, a control device and a wafer saw, which can effectively overcome the above problems. SUMMARY

[0005] The present application provides a wafer saw cutting pressure control method based on dynamic pressure prediction, a control device and a wafer saw. The method predicts the pressure mutation point through a dynamic mapping model, adjusts the pressure in advance combined with a control algorithm, and loads the heterogeneous material characteristics in stages, which significantly improves the kerf uniformity and reduces the risk of material damage.

[0006] In a first aspect, a wafer saw cutting pressure control method based on dynamic pressure prediction is provided. The method comprises:

[0007] S1: Obtain real-time material property parameters on the cutting path through a multi-source sensor, including material hardness gradient and doping concentration distribution;

[0008] S2: Based on the material property parameters, establish a dynamic mapping model of material properties and cutting pressure. The model is generated by machine learning algorithm training, and the input parameters are material hardness, doping concentration and cutting speed, and the output parameter is the predicted pressure compensation value;

[0009] S3: Before the cutting tool reaches the pressure mutation area, adjust the cutting pressure in advance according to the predicted pressure compensation value output by the dynamic mapping model, and simultaneously detect the actual pressure value in real time through a force sensor, and correct the pressure deviation combined with a control algorithm;

[0010] S4: For the unevenly doped region inside the heterogeneous material, a phased pressure loading is performed, the first phase uses an initial pressure lower than the target pressure to form a pre-cut groove, and the second phase gradually increases to the target pressure based on the depth of the pre-cut groove and the calculation results of the dynamic mapping model.

[0011] It should be understood that by detecting material properties in real time and constructing a dynamic mapping model, accurately predicting pressure mutation regions, using a control algorithm to adjust the pressure in advance, and combining a phased loading strategy to relieve stress concentration at the heterogeneous interface, the uneven cutting marks caused by cutting pressure mutation are solved from the source, and high-precision cutting control is achieved.

[0012] In combination with the first aspect, in some implementations of the first aspect, the dynamic mapping model in step S2 is constructed by:

[0013] S201: Collect historical cutting data, including material hardness, doping concentration, cutting speed, actual pressure value, and cutting quality score;

[0014] S202: Use a machine learning algorithm to train the historical data to generate a nonlinear relationship model of material properties and cutting pressure;

[0015] S203: Input the real-time detected material property parameters into the model to output a predicted pressure compensation value.

[0016] It should be understood that by training a machine learning model with historical data, a nonlinear relationship between material properties and cutting pressure is established, which enhances the prediction accuracy of the dynamic mapping model, makes the pressure compensation value more consistent with the actual variation of heterogeneous materials, and improves the adaptive ability of the cutting process.

[0017] It should be understood that the dynamic mapping model analyzes the material hardness gradient, doping concentration distribution, and cutting speed through a machine learning algorithm to predict the pressure change critical point caused by material property mutation on the cutting path, and then determine the pressure mutation region.

[0018] In combination with the first aspect, in some implementations of the first aspect, the multi-source sensor in step S1 includes:

[0019] An acoustic emission sensor for detecting stress wave signals during cutting to obtain the material hardness gradient;

[0020] An infrared spectrometer for analyzing the doping concentration distribution of the cutting area;

[0021] A piezoelectric force sensor for collecting the actual pressure value of the contact surface between the cutting tool and the material.

[0022] It should be understood that the use of multi-source sensors such as acoustic emission sensors and infrared spectrometers for collaborative detection of material hardness gradient and doping concentration ensures that real-time data comprehensively reflects the characteristics of heterogeneous materials, provides high-precision input for dynamic mapping models, and supports the reliability of pressure prediction.

[0023] In combination with the first aspect, in some implementations of the first aspect, the step S4 includes:

[0024] S401: The first stage pressure is 50%-70% of the target pressure, and the cutting depth is 30%-40% of the total depth;

[0025] S402: The second stage pressure is gradually increased to the target pressure according to the depth of the pre-cut groove and the output value of the dynamic mapping model, and the increase rate is 5%-10% per millimeter of cutting length.

[0026] It should be understood that setting the initial pressure to 50%-70% of the target pressure and cutting to 30%-40% of the total depth in the first stage releases local stress through the pre-cut groove, and gradually increases the pressure in the second stage per millimeter of cutting length to avoid the impact of pressure mutation on the heterogeneous interface and balance the cutting efficiency and quality.

[0027] In combination with the first aspect, in some implementations of the first aspect, the step S3 includes:

[0028] S301: A feedforward control is used, and the generation period of the signal of the feedforward control is related to the cutting tool movement speed, ensuring that the pressure adjustment is synchronized with the tool position;

[0029] S302: A PID control algorithm is used, and the integral term weight of the PID control algorithm is dynamically adjusted according to the brittleness coefficient of the material to reduce the integral term weight for brittle materials to prevent overshoot.

[0030] It should be understood that the feedforward control is used to realize the synchronization of pressure adjustment and tool movement, and the integral weight is dynamically adjusted in combination with the PID algorithm to reduce the integral term for brittle materials to prevent overshoot, ensuring the stability and response speed of pressure correction and adapting to complex cutting scenarios of heterogeneous materials.

[0031] In combination with the first aspect, in some implementations of the first aspect, the method includes a cutting tool compensation step: detecting the wear amount of the cutting tool through an optical sensor, and correcting the predicted pressure compensation value output by the dynamic mapping model according to the wear amount, and the correction coefficient is a linear function of the wear amount.

[0032] It should be understood that by detecting the wear amount of the cutting tool and correcting the pressure compensation value, the cutting force decay caused by the wear of the cutting edge is offset, the prediction accuracy of the dynamic mapping model is maintained, and the long-term stability of the cutting of heterogeneous materials is ensured.

[0033] In combination with the first aspect, in some implementations of the first aspect, the method comprises a vibration suppression step: detecting the spindle vibration amplitude by the acceleration sensor during the cutting pressure adjustment process, and reducing the pressure increase rate until the vibration amplitude returns to the safe range if the vibration amplitude exceeds the threshold value.

[0034] It should be understood that by monitoring the vibration amplitude to dynamically adjust the pressure increase rate, mechanical resonance during cutting is suppressed, vibration interference is prevented from aggravating the kerf deviation, and the straightness consistency of heterogeneous material cutting is improved.

[0035] In combination with the first aspect, in some implementations of the first aspect, the method further comprises a thermal management synergistic control: dynamically adjusting the cooling liquid flow rate according to the cutting area temperature detected by the infrared thermal imager, and the flow rate adjustment and the pressure adjustment have a ratio of 1:0.8 to 1:1.2.

[0036] It should be understood that by synergistically adjusting the cooling liquid flow rate and the cutting pressure, local overheating-induced material thermal stress cracks are avoided, the thermal-mechanical coupling stability of the heterogeneous interface under phased pressure loading is ensured, and the cutting process reliability is enhanced.

[0037] The second aspect provides a control device, which comprises a processor and a memory, the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that any implementation manner of the first aspect is executed.

[0038] The third aspect provides a dicing machine, which comprises the control device according to the second aspect. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is a dynamic pressure prediction-based dicing machine cutting pressure control method implementation flowchart provided by the embodiments of the present application.

[0040] Figure 2 is a dynamic mapping model construction method implementation flowchart provided by the embodiments of the present application.

[0041] Figure 3 is a phased pressure loading method implementation flowchart provided by the embodiments of the present application.

[0042] Figure 4 is a control method implementation flowchart provided by the embodiments of the present application. DETAILED DESCRIPTION

[0043] The terminology used in the following description merely for the purpose of describing particular embodiments and is not intended to limit the application. As used in this description and the accompanying claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or," as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items. The term "or" as used herein is used to mean, and is used in each instance in which it appears, either one of the items in a list of two or more items, or any combination of two or more of the items in the list of two or more items. The term "and / or," as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0044] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrases "in one embodiment" or "in some embodiments" or "in other embodiments" or "in still other embodiments" or similar phrases in various places throughout this specification are not necessarily all referring to the same embodiment, unless otherwise indicated. The terms "including," "comprising," "having," and variations thereof are meant to encompass the items listed thereafter and equivalents thereof as well as additional items. The terms "consisting of" and variations thereof are meant to encompass only the items listed thereafter.

[0045] In the process of cutting a semiconductor wafer by a dicing machine, heterogeneous materials are prone to cause sudden changes in cutting pressure due to internal hardness gradient, sudden change in doping concentration and other characteristics, which may cause uneven cutting marks and micro-cracks, and seriously affect the yield of chips.

[0046] The embodiments of the present application provide a dicing machine cutting pressure control method and device based on dynamic pressure prediction, and a dicing machine.

[0047] The technical solutions provided by the embodiments of the present application will be described below with reference to the accompanying drawings.

[0048] Figure 1 The embodiments of the present application provide a dicing machine cutting pressure control method based on dynamic pressure prediction, and in some examples, the method comprises:

[0049] S1: Obtain real-time material characteristic parameters on the cutting path through a multi-source sensor, including material hardness gradient and doping concentration distribution;

[0050] S2: Based on the material characteristic parameters, a dynamic mapping model of material characteristics and cutting pressure is established, the model is generated by machine learning algorithm training, the input parameters are material hardness, doping concentration and cutting speed, and the output parameter is a predicted pressure compensation value;

[0051] S3: Before the cutting tool reaches the pressure mutation area, the cutting pressure is adjusted in advance according to the predicted pressure compensation value output by the dynamic mapping model, and the actual pressure value is detected in real time through the force sensor, and the pressure deviation is corrected in combination with the control algorithm;

[0052] S4: For the uneven doping area inside the heterogeneous material, the pressure is loaded in stages, the initial pressure lower than the target pressure is used in the first stage to form a pre-cut groove, and the target pressure is gradually increased to the target pressure based on the depth of the pre-cut groove and the calculation result of the dynamic mapping model in the second stage.

[0053] Figure 2 is a dynamic mapping model construction method provided by the embodiment of the application. In some examples, the dynamic mapping model in step S2 is constructed by the following method:

[0054] S201: Collect historical cutting data, including material hardness, doping concentration, cutting speed, actual pressure value and cutting quality score;

[0055] S202: Use a machine learning algorithm to train the historical data to generate a nonlinear relationship model of material characteristics and cutting pressure;

[0056] S203: Input the real-time detected material characteristic parameters into the model to output a predicted pressure compensation value.

[0057] In a possible implementation, the improved random forest algorithm is adopted to construct the dynamic mapping model, and the specific steps are as follows: first, historical cutting data including material hardness, doping concentration, cutting speed, actual pressure value and cutting quality score (taking the uniformity of the cutting mark as the quantitative index) are collected to construct a training data set; second, in view of the nonlinear characteristics of the cutting of heterogeneous materials, an adaptive feature weight distribution mechanism is introduced, higher weights are given to the material hardness, doping concentration and other key features in the splitting process of the random forest, and the interference of redundant parameters is reduced; at the same time, in order to improve the prediction sensitivity of the model to the pressure mutation region, a mutation penalty term is added to the loss function, when the gradient difference between the model predicted pressure value and the actual value exceeds the set threshold, the gradient penalty coefficient is increased to strengthen the learning ability of the model to the mutation point; finally, an online incremental learning strategy is adopted, the material characteristic parameters detected in the real-time cutting process are input into the model as incremental data, and the decision tree node parameters are dynamically updated, so that the model continuously adapts to the new material process changes. Through the above improvement, the average error of the model in predicting the pressure mutation point is reduced, and the generalization ability of the model to unknown heterogeneous materials is significantly improved.

[0058] In some examples, the multi-source sensor in the step S1 comprises:

[0059] An acoustic emission sensor for detecting stress wave signals in the cutting process to obtain the material hardness gradient;

[0060] An infrared spectrometer for analyzing the doping concentration distribution of the cutting region;

[0061] A piezoelectric force sensor for collecting the actual pressure value of the contact surface between the cutting tool and the material.

[0062] In a possible implementation, the acoustic emission sensor is installed in a non-contact manner behind the cutting tool, and real-time acquisition of the stress wave signals generated by the material fracture in the cutting process is realized, and the material hardness gradient change is obtained by analyzing the amplitude, frequency and energy distribution of the stress wave; the infrared spectrometer performs line scanning on the cutting region by using the near-infrared wave band, and the doping concentration distribution is calculated by the intensity of the characteristic absorption peak, and the spatial resolution is controlled to be within 5 μm; the piezoelectric force sensor is integrated at the end of the cutting spindle, and the piezoelectric charge signals of the contact surface between the cutting tool and the material are detected, and the axial pressure value is converted in real time, and the sampling frequency is not less than 10 kHz, and the time stamp is synchronized with the acoustic emission and infrared spectrum data, and the noise interference is eliminated through the multi-sensor data fusion algorithm, and finally the material characteristic parameters with high confidence are output.

[0063] Figure 3 is a flowchart of a phased pressure loading method provided by an embodiment of the present application, and in some examples, the phased pressure loading in the step S4 comprises:

[0064] S401: the first stage pressure is 50%-70% of the target pressure, and the cutting depth is 30%-40% of the total depth;

[0065] S402: the second stage pressure is gradually increased to the target pressure according to the depth of the pre-cut groove and the output value of the dynamic mapping model, and the increase rate is 5%-10% per millimeter of cutting length.

[0066] Figure 4 A control method implementation flowchart is provided for the embodiments of the present application. In some examples, the step S3 includes:

[0067] S301: a feedforward control is adopted, the generation period of the signal of the feedforward control is related to the cutting tool movement speed, and the pressure adjustment is synchronized with the tool position;

[0068] S302: a PID control algorithm is adopted, and the integral term weight of the PID control algorithm is dynamically adjusted according to the material brittleness coefficient, and the integral term weight is reduced for brittle materials to prevent overshoot.

[0069] In some examples, the method includes a cutting tool compensation step: detecting the wear amount of the cutting tool through an optical sensor, and correcting the predicted pressure compensation value output by the dynamic mapping model according to the wear amount, and the correction coefficient is a linear function of the wear amount.

[0070] In some examples, the method includes a vibration suppression step: detecting the spindle vibration amplitude through an acceleration sensor during cutting pressure adjustment, and if the vibration amplitude exceeds a threshold value, the pressure increase rate is reduced until the vibration amplitude returns to a safe range.

[0071] In some examples, the method further includes a thermal management cooperative control: dynamically adjusting the cooling liquid flow according to the cutting area temperature detected by an infrared thermal imager, and the flow adjustment to pressure adjustment ratio is 1:0.8 to 1:1.2.

[0072] The embodiments of the present application provide a control device, which includes a processor and a memory, the processor is coupled with the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method as described in any of the preceding examples is executed.

[0073] The embodiments of the present application also provide a wafer scribing machine, which includes the control device as described in the preceding embodiments.

[0074] The above merely describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments, but any equivalent modifications or changes made by those skilled in the art according to the disclosed content of the present application shall be included in the protection scope recorded in the claims.

Claims

1. A method for controlling the cutting pressure of a dicing machine based on dynamic pressure prediction, characterized in that, The method includes: S1: Real-time material property parameters along the cutting path are obtained through multi-source sensors, including material hardness gradient and doping concentration distribution; S2: Based on the material property parameters, a dynamic mapping model between material properties and cutting pressure is established. The model is generated by training a machine learning algorithm. The input parameters are material hardness, doping concentration and cutting speed, and the output parameter is the predicted pressure compensation value. S3: Before the cutting tool reaches the pressure change zone, the cutting pressure is adjusted in advance according to the predicted pressure compensation value output by the dynamic mapping model. At the same time, the actual pressure value is detected in real time by the force sensor, and the pressure deviation is corrected by the control algorithm. S4: For the uneven doping region inside the heterogeneous material, staged pressure loading is performed. In the first stage, an initial pressure lower than the target pressure is used to form a pre-cut groove. In the second stage, the pressure is gradually increased to the target pressure based on the depth of the pre-cut groove and the calculation results of the dynamic mapping model.

2. The method according to claim 1, characterized in that, The dynamic mapping model in step S2 is constructed in the following way: S201: Collect historical cutting data, including material hardness, doping concentration, cutting speed, actual pressure value, and cutting quality score; S202: Use machine learning algorithms to train the historical data to generate a nonlinear relationship model between material properties and cutting pressure; S203: Input the material property parameters detected in real time into the model and output the predicted pressure compensation value.

3. The method according to claim 1, characterized in that, The multi-source sensor in step S1 includes: An acoustic emission sensor is used to detect stress wave signals during the cutting process to obtain the material hardness gradient; An infrared spectrometer, used to analyze the doping concentration distribution in the cut region; A piezoelectric force sensor is used to collect the actual pressure value of the contact surface between the cutting tool and the material.

4. The method according to claim 1, characterized in that, The staged pressure loading in step S4 includes: S401: The first stage pressure is 50%-70% of the target pressure, and the cutting depth is 30%-40% of the total depth; S402: The second stage pressure is gradually increased to the target pressure based on the depth of the pre-cut groove and the output value of the dynamic mapping model, with an increase rate of 5%-10% per millimeter of cutting length.

5. The method according to claim 1, characterized in that, Step S3 includes: S301: Feedforward control is adopted, and the generation period of the feedforward control signal is related to the moving speed of the cutting tool to ensure that the pressure adjustment is synchronized with the tool position; S302: A PID control algorithm is adopted. The weight of the integral term of the PID control algorithm is dynamically adjusted according to the brittleness coefficient of the material. The weight of the integral term is reduced for brittle materials to prevent overshoot.

6. The method according to claim 1, characterized in that, The method includes a cutting tool compensation step: detecting the wear amount of the cutting tool using an optical sensor, and correcting the predicted pressure compensation value output by the dynamic mapping model based on the wear amount, wherein the correction coefficient is a linear function of the wear amount.

7. The method according to claim 1, characterized in that, The method includes a vibration suppression step: during the cutting pressure adjustment process, the vibration amplitude of the spindle is detected by an acceleration sensor. If the vibration amplitude exceeds a threshold, the pressure increase rate is reduced until the vibration amplitude returns to a safe range.

8. The method according to claim 1, characterized in that, The method also includes thermal management coordinated control: the coolant flow rate is dynamically adjusted according to the temperature of the cutting area detected by the infrared thermal imager, and the ratio of flow rate adjustment to pressure adjustment is 1:0.8 to 1:1.

2.

9. A control device, characterized in that, The device includes a processor and a memory, the processor being coupled to the memory, the memory being used to store computer programs or instructions, and the processor being used to execute the computer programs or instructions in the memory such that the method as claimed in any one of claims 1 to 8 is performed.

10. A dicing machine, characterized in that, The dicing machine includes the control device as described in claim 9.

Images