An end effector for a semiconductor wafer transfer robot

Abstract

The present invention relates to the field of semiconductor technology, and in particular to an end effector for a semiconductor wafer transfer robot. It mainly includes an end effector body, a plurality of air path adsorption modules with independent control functions of negative pressure values, and a control module. The present application can realize a variety of adsorption modes by setting a plurality of air path adsorption modules with independent control functions of negative pressure values ​​to adapt to the warping morphology of different areas of the wafer; through the synergy between the air path adsorption modules, the sensor array and the control module, according to the gap value between the target area collected by the optical fiber sensor and the end effector body, the warping area of ​​the wafer is automatically predicted and located through the machine learning classification model, combined with the negative pressure dynamic adjustment strategy, so as to realize the differentiated adsorption force adjustment of different warping areas of the wafer, which not only significantly improves the response speed and adaptability of the end effector of the wafer transfer robot, but also effectively improves the stability and reliability of wafer adsorption.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to an end effector of a semiconductor wafer transfer robot. Background Art

[0002] At present, the end effector of a traditional wafer transfer robot usually uses a single air path structure for negative pressure adsorption control, and is unable to perform differentiated negative pressure adjustment according to the warping conditions of different areas of the wafer. This single air path structure is prone to poor adsorption or stress concentration when encountering deformations such as warping of the wafer edge or depression of the center, which in turn causes wafer slippage or breakage. In the prior art, the relevant technical solutions attempt to detect the contact status between the wafer and the suction cup through a pressure sensor, but in a complex electromagnetic interference environment, the signal of the pressure sensor is easily distorted, and the warping area cannot be accurately located. In addition, the control logic of the adsorption force is usually a fixed threshold judgment, which lacks the ability to dynamically predict the warping trend of the wafer, reducing the response speed and adaptability of the end effector of the wafer transfer robot.

[0003] To this end, the present application provides an end effector for a semiconductor wafer transfer robot. Summary of the Invention

[0004] Based on this, it is necessary to provide an end effector for a semiconductor wafer transfer robot to address the above technical problems.

[0005] The present invention provides an end effector for a semiconductor wafer transfer robot, comprising: an end effector body for carrying wafers; a plurality of air path adsorption modules with independent negative pressure value control functions, wherein the air path adsorption modules are each provided with a sensor array, and each sensor array corresponds to a target area on the surface of the wafer, and is used to collect the gap value between the target area and the end effector body in real time, and transmit it to the control module, and dynamically adjust the negative pressure value of the corresponding target area according to the received dynamic negative pressure value output signal; the control module is used to construct and train a machine learning classification model, pre-process the received gap value between the target area and the end effector body, generate a deformation feature array, input the deformation feature array into the trained machine learning classification model, output a wafer warping type judgment result, and generate a dynamic negative pressure value output signal of each air path adsorption module based on the negative pressure dynamic adjustment strategy, and transmit it to the corresponding air path adsorption module.

[0006] Optionally, each of the air path adsorption modules includes at least one adsorption unit, the adsorption unit includes a boss and an air cushion, the air cushion is sealed with an inner countersunk hole of the boss, and a negative pressure hole is opened on the air cushion.

[0007] Optionally, the end actuator body includes a main body, a connecting part and a cover plate, the interior of the main body is provided with an air path channel connected one-to-one with the negative pressure hole, the rear end of the main body is fixedly connected to the connecting part, and the main body and the connecting part are sealed by the cover plate, the interior of the connecting part is provided with an air path pipeline connected one-to-one with the air path channel, and the air path pipeline is provided with a regulating valve.

[0008] Optionally, each of the sensor arrays includes at least one optical fiber sensor, and an optical fiber sensor is provided at the bottom of the side of each boss, for starting to collect the gap value between the target area and the end effector body in real time when the distance between the end effector body and the wafer surface is less than a preset distance threshold.

[0009] Optionally, the gas path adsorption module also includes a vacuum pump controlled by PWM, and the vacuum pump is provided with a D / A conversion module. The vacuum pump is connected to the control module and is used to receive the dynamic negative pressure value output signal transmitted by the control module. The dynamic negative pressure value output signal is mapped to the rotation speed of the vacuum pump through the D / A conversion module to achieve dynamic control of the negative pressure value of the corresponding target area.

[0010] Optionally, the constructing and training of the machine learning classification model includes: constructing a historical gap value set, the historical gap value set containing multiple training samples, using multi-classification coding technology to label the typical warping type to which each training sample belongs, and obtaining a labeled historical gap value set, wherein the typical warping types include edge warping, center depression, local deformation and normal state; using a lightweight CNN convolutional neural network to construct an initial machine learning classification model; training the initial machine learning classification model based on the labeled historical gap value set, and combining the cross-entropy loss function to obtain a trained machine learning classification model.

[0011] Optionally, the gap value between the received target area and the end effector body is preprocessed to generate a deformation feature array, including: performing sliding window filtering on the gap value between the received target area and the end effector body to obtain gap change filtering data; based on the gap change filtering data, respectively calculating the gap change rate and local curvature of each target area, wherein the gap change rate is used to reflect the speed of local deformation of the wafer, and the local curvature is used to reflect the degree of curvature of the wafer surface; combining the gap change rate and the local curvature to construct an initial deformation feature array; and performing Z-score normalization on the initial deformation feature array to obtain a deformation feature array.

[0012] Optionally, the calculation formula for the gap change rate of each target area is: ,in, is the gap change rate for each target area, d( t+1 ) is the gap value collected by the i-th optical fiber sensor in each target area at time t+1, d ( t ) is the gap value collected by the i-th optical fiber sensor in each target area at time t, and Δt is the time interval for collecting gap values.

[0013] Optionally, the local curvature calculation formula of each target area is: ,in, is the local curvature of each target region, dx is the distance between two adjacent fiber optic sensors in each target area, is the second-order derivative of the gap value collected by the i-th optical fiber sensor in each target area at time t.

[0014] Optionally, the dynamic negative pressure value output signal of each gas path adsorption module is generated based on the negative pressure dynamic adjustment strategy, including: constructing a negative pressure dynamic adjustment strategy based on the mapping relationship between typical warping types and negative pressure adjustment logic rules; determining the corresponding negative pressure adjustment logic rules based on the output wafer warping type judgment result, and calculating and generating the dynamic negative pressure value output signal of each gas path adsorption module.

[0015] The advantages and beneficial effects of the present invention are as follows: the present invention provides an end effector for a semiconductor wafer transfer robot, which can realize multiple adsorption modes to adapt to the warping morphology of different areas of the wafer by setting multiple air path adsorption modules with independent negative pressure value control functions; through the synergy between the air path adsorption modules, the sensor array and the control module, based on the gap value between the target area collected by the optical fiber sensor and the end effector body, the warping area of ​​the wafer is automatically predicted and located through the machine learning classification model, combined with the negative pressure dynamic adjustment strategy, so as to realize differentiated adsorption force adjustment of different warping areas of the wafer, which not only significantly improves the response speed and adaptability of the end effector of the wafer transfer robot, but also effectively improves the stability and reliability of wafer adsorption. At the same time, the use of optical fiber sensors instead of traditional pressure sensors also has the advantages of anti-electromagnetic interference and higher precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a structural diagram of the end effector body of the present invention mounted on a wafer transfer robot.

[0017] Figure 2 The figure is a schematic structural diagram of the end effector of the semiconductor wafer transfer robot according to the present invention.

[0018] Figure 3 It is a schematic diagram of the structure inside the main body of the end effector of the present invention.

[0019] Figure 4 This is a structural diagram of the gas path adsorption module of the present invention arranged in the main body.

[0020] Figure 5 This is another structural schematic diagram of the gas path adsorption module of the present invention arranged on the main body.

[0021] Figure 6 This is another structural schematic diagram of the gas path adsorption module of the present invention arranged on the main body.

[0022] Among them, the end effector body 1, the multi-joint mechanical motion structure 2, the wafer transfer robot 3, the main body 4, the connecting part 5, the cover 6, the boss 7, the air cushion 8, the optical fiber sensor 9, the air path channel 10, the air pipe joint 11, the air path pipeline 12, the regulating valve 13, the air path adsorption module a14, the air path adsorption module b15, the air path adsorption module c16, the air path adsorption module d17, the air path adsorption module e18, the air path adsorption module f19, the air path adsorption module g20, and the air path adsorption module h21. DETAILED DESCRIPTION

[0023] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and are not to be construed as limitations on the present application. In addition, the following embodiments and features in the embodiments may be combined with each other unless there is a conflict. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0024] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.

[0025] The present application provides an end effector for a semiconductor wafer transfer robot, which mainly includes an end effector body, multiple air path adsorption modules with independent negative pressure value control functions, and a control module.

[0026] In some optional implementations of the present application, the end effector body is used to carry a wafer.

[0027] In some optional implementations of this application, please refer to the attached Figure 2-3The end effector body includes a main body 4, a connecting part 5 and a cover plate 6. The interior of the main body 4 is provided with an air path channel 10 connected to the negative pressure holes in a one-to-one correspondence. The rear end of the main body 4 is fixedly connected to the connecting part 5, and the main body 4 and the connecting part 5 are sealed together through the cover plate 6. The interior of the connecting part 5 is provided with an air path pipeline 12 connected to the air path channel 10 in a one-to-one correspondence, and the air path pipeline 12 is provided with a regulating valve 13.

[0028] In some optional implementations of this application, please refer to the attached Figure 1 The end effector body 1 involved in this application is arranged on a multi-joint mechanical motion structure 2 through a joint axis, wherein the multi-joint mechanical motion structure 2 includes N robotic arms controlled by N+1 joint axes on a horizontal plane, and N is an integer greater than 0. At the same time, the multi-joint mechanical motion structure 2 is arranged on a wafer transfer robot 3 through a joint axis. The wafer transfer robot 3 is a body with its own vertical transmission, and is used to realize wafer transfer through the coordinated action of the end effector body 1, the multi-joint mechanical motion structure 2 and the wafer transfer robot 3.

[0029] In some optional implementations of the present application, an air pipe connector 11 is further provided at the connection point between the air channel 10 passing through the main body 4 and the air pipeline 12 for connecting the air channel 10 and the air pipeline 12 in a one-to-one correspondence.

[0030] In some optional implementations of the present application, multiple air path adsorption modules are provided with independent negative pressure value control functions, and each air path adsorption module is equipped with a sensor array, and each sensor array corresponds to a target area on the surface of a wafer, which is used to collect the gap value between the target area and the end effector body in real time, and transmit it to the control module, and output a signal based on the received dynamic negative pressure value to dynamically adjust the negative pressure value of the corresponding target area.

[0031] In some optional implementations of this application, please refer to the attached Figure 2 Each of the air path adsorption modules includes at least one adsorption unit, and the adsorption unit includes a boss 7 and an air cushion 8. The air cushion 8 is sealed with the inner countersunk hole of the boss 7, and a negative pressure hole is opened on the air cushion 8.

[0032] It should be noted that, based on the comprehensive considerations of reducing manufacturing costs, simplifying device integration, the structural shape of the main body 4, and avoiding extreme warping (such as above 5mm), the number and layout positions of the gas path adsorption modules can be flexibly configured, wherein multiple gas path adsorption modules can be arranged in concentric rings on the main body 4, or in a circular array on the main body 4, or in other layout forms on the main body 4. They can be flexibly selected according to actual industrial production conditions and are no longer limited here. At the same time, the adsorption units can be arranged in a circular array on the main body 4, or in other layout forms on the main body 4. They can be flexibly selected according to actual industrial production conditions and are no longer limited here. As long as the multi-gas path negative pressure dynamic control and real-time feedback mechanism provided in this application are combined, the purpose of differentiated negative pressure adjustment for the warping conditions of different areas of the wafer can be achieved.

[0033] In some optional implementations of this application, please refer to the attached Figure 4 For example, the front end adsorption surface of the main body 4 is a semicircular structure, and the number of gas path adsorption modules can be set to three, specifically gas path adsorption module a14, gas path adsorption module b15 and gas path adsorption module c16. The three gas path adsorption modules are arranged in concentric rings on the main body 4, and are used to divide the wafer surface into multiple target areas, such as edge areas, central areas and local areas. The gas path adsorption modules a14, b15 and c16 each include at least one adsorption unit, each adsorption unit includes a boss 7 and an air cushion 8 embedded in the boss 7, each air cushion 8 is provided with a negative pressure hole, and each negative pressure hole is respectively connected to a The gas paths 10 are interconnected, and each gas path 10 is connected to a gas path pipeline 12. Through the independent regulation of the three gas path adsorption modules, a variety of adsorption modes can be achieved, such as only the gas path adsorption module a14 works, only the gas path adsorption module b15 works, only the gas path adsorption module c16 works, the gas path adsorption module a14 and the gas path adsorption module b15 work together, the gas path adsorption module a14 and the gas path adsorption module c16 work together, the gas path adsorption module b15 and the gas path adsorption module c16 work together, the gas path adsorption module a14, the gas path adsorption module b15 and the gas path adsorption module c16 work together, so as to adapt to the warping shape of different areas of the wafer.

[0034] In some optional implementations of this application, please refer to the attached Figure 5For example, the front end adsorption surface of the main body 4 is a circular structure, and the number of gas path adsorption modules can be set to three, specifically gas path adsorption module d17, gas path adsorption module e18 and gas path adsorption module f19. The three gas path adsorption modules are arranged in concentric rings on the main body 4, and are used to divide the wafer surface into multiple target areas, such as edge areas, central areas and local areas. The gas path adsorption modules d17, e18 and f19 each include at least one adsorption unit, each adsorption unit includes a boss 7 and an air cushion 8 embedded in the boss 7, each air cushion 8 is provided with a negative pressure hole, and each negative pressure hole is respectively connected to a gas The gas paths 10 are connected, and each gas path 10 is connected to a gas path pipeline 12. Through the independent regulation of the three gas path adsorption modules, a variety of adsorption modes can be realized, such as only the gas path adsorption module d17 works, only the gas path adsorption module e18 works, only the gas path adsorption module f19 works, the gas path adsorption module d17 and the gas path adsorption module e18 work together, the gas path adsorption module d17 and the gas path adsorption module f19 work together, the gas path adsorption module e18 and the gas path adsorption module f19 work together, the gas path adsorption module d17, the gas path adsorption module e18 and the gas path adsorption module f19 work together, so as to adapt to the warping shape of different areas of the wafer.

[0035] In some optional implementations of this application, please refer to the attached Figure 6 For example, the main body 4 has a Y-shaped structure, and the number of gas path adsorption modules can be set to two, specifically the gas path adsorption module g20 and the gas path adsorption module h21. The two gas path adsorption modules are arranged in concentric rings on the main body 4, and are used to divide the wafer surface into multiple target areas, such as edge areas and local areas. The gas path adsorption module g20 and the gas path adsorption module h21 each include at least one adsorption unit (such as three adsorption units), each adsorption unit includes a boss 7 and an air cushion 8 embedded in the boss 7, each air cushion 8 is provided with a negative pressure hole, each negative pressure hole is respectively connected to an air path channel 10, and each air path channel 10 is connected to an air path pipeline 12. Through the independent regulation of the two air path adsorption modules, multiple adsorption modes can be achieved, such as only the air path adsorption module g20 working, only the air path adsorption module h21 working, and the air path adsorption module g20 and the air path adsorption module h21 working together, so as to adapt to the warping shape of different areas of the wafer.

[0036] In some optional implementations of the present application, each of the sensor arrays includes a plurality of optical fiber sensors 9, and an optical fiber sensor 9 is provided at the bottom of the side of each boss 7, for starting to collect the gap value between the target area and the end effector body in real time when the distance between the end effector body and the wafer surface is less than a preset distance threshold.

[0037] In some optional implementations of the present application, in order to avoid the problem that the pressure sensor signal is easily distorted and cannot accurately locate the warped area in a complex electromagnetic interference environment, the present application adopts a high-precision fiber optic sensor 9 (such as an FBG). By arranging the fiber optic sensor 9 at the bottom of the side of the boss 7, it can collect the gap value between the target area and the end effector body in real time. When the distance between the end effector body and the wafer surface is less than a preset distance threshold (such as 1 mm), the fiber optic sensor 9 begins to collect time series data (i.e., the gap value between the target area and the end effector body). The sampling frequency can be set to 1 kHz to ensure high-resolution capture of the wafer surface deformation. The fiber optic sensor 9 detects the phase difference change of the reflected light from the wafer surface to obtain the gap value between the wafer and the end effector body in real time.

[0038] In some optional implementations of the present application, the gas path adsorption module also includes a vacuum pump controlled by PWM, and the vacuum pump is provided with a D / A conversion module. The vacuum pump is connected to the control module and is used to receive the dynamic negative pressure value output signal transmitted by the control module. The dynamic negative pressure value output signal is mapped to the rotation speed of the vacuum pump through the D / A conversion module to achieve dynamic control of the negative pressure value of the corresponding target area.

[0039] In some optional implementations of the present application, the present application also sets an air pressure sensor on each air path pipeline 12 to monitor the negative pressure value generated in each target area in real time, and form a closed-loop adjustment and verification with the dynamic negative pressure value generated by each air path adsorption module to ensure the accuracy and reliability of the dynamic regulation of the negative pressure value in each target area.

[0040] In some optional implementations of the present application, the control module is used to construct and train a machine learning classification model, preprocess the received gap value between the target area and the end effector body, generate a deformation feature array, input the deformation feature array into the trained machine learning classification model, output the wafer warping type judgment result, and based on the negative pressure dynamic adjustment strategy, generate a dynamic negative pressure value output signal of each gas path adsorption module, and transmit it to the corresponding gas path adsorption module.

[0041] In some optional implementations of the present application, the construction and training of the machine learning classification model includes: constructing a historical gap value set, the historical gap value set containing multiple training samples, using multi-classification coding technology to label the typical warping type to which each training sample belongs, and obtaining a historical gap value set with labels, wherein the typical warping types include edge warping, center depression, local deformation and normal state; using a lightweight CNN convolutional neural network to construct an initial machine learning classification model; training the initial machine learning classification model based on the historical gap value set with labels, and combining the cross-entropy loss function to obtain a trained machine learning classification model.

[0042] In some optional implementations of the present application, the present application uses a lightweight CNN convolutional neural network to construct an initial machine learning classification model. The lightweight CNN convolutional neural network consists of an input layer, a one-dimensional convolutional layer, a maximum pooling layer, and a fully connected layer connected in sequence. The input layer is used to input a deformation feature array of length n, where the value of n corresponds to the number of optical fiber sensors; the one-dimensional convolutional layer contains 64 filters and 3 convolution kernels, and the activation function uses ReLU; the pooling window of the maximum pooling layer is 2; the fully connected layer contains 128 neurons and uses the Softmax activation function to output the wafer warpage type judgment result. During training, the cross entropy loss function and the Adam optimizer are used to continuously optimize the machine learning classification model, and the learning rate can be set to 0.001. In addition, the machine learning classification model can be accelerated by the GPU during the offline training phase and finally deployed in the embedded FPGA of the control module to ensure that the inference delay is less than 10 ms.

[0043] In some optional implementations of the present application, the construction history gap value set involved in the present application includes multiple training samples (i.e., the historical gap values ​​between the target area and the end effector body). According to the preset typical warping type, a multi-classification coding technology is used to label the typical warping type to which each training sample belongs. The typical warping types include edge warping, center depression, local deformation and normal state. Among them, edge warping is represented by the gap change rate of the wafer edge area being significantly higher than the gap change rate of the center area, which can be defined by the formula as: , σ represents the standard deviation of deformation characteristics, Indicates the gap change rate of the wafer edge area; the central depression indicates that the gap change rate of the central area is negative and the local curvature increases significantly, which can be defined by the formula and , Indicates the gap variation rate in the center area of ​​the wafer, It represents the local curvature of the wafer center area; the local deformation is represented by a sudden change in the gap change rate of the single optical fiber sensor acquisition area, which can be defined as , Indicates the gap change rate of a single fiber optic sensor acquisition area; the normal state is indicated by the gap change rate of all areas being within the range of ±σ; the label can be coded using multiple categories, for example: edge warping is coded as 0, center depression is coded as 1, local deformation is coded as 2, and normal state is coded as 3.

[0044] In some optional implementations of the present application, the gap value between the received target area and the end effector body is preprocessed to generate a deformation feature array, including: performing sliding window filtering on the gap value between the received target area and the end effector body to obtain gap change filtering data; based on the gap change filtering data, calculating the gap change rate and local curvature of each target area respectively, wherein the gap change rate is used to reflect the speed of local deformation of the wafer, and the local curvature is used to reflect the degree of curvature of the wafer surface; combining the gap change rate and the local curvature to construct an initial deformation feature array; and performing Z-score normalization on the initial deformation feature array to obtain a deformation feature array.

[0045] In some optional implementations of the present application, the present application adopts a sliding window filtering technique to perform sliding window filtering on the gap value between the target area and the end effector body collected by the optical fiber sensor, with a window length of N and a sampling frequency of , used to eliminate high-frequency noise; in addition, the data after sliding window filtering can also be detrended, specifically: polynomial fitting is used to remove the linear trend in the data to ensure the accuracy of subsequent feature extraction.

[0046] In some optional implementations of the present application, the steps for constructing the initial deformation feature array involved in the present application include: (1) calculating the gap change rate of each target area based on the gap change filtering data to reflect the speed of local deformation of the wafer, and the formula is: ,in, is the gap change rate for each target area, d ( t+1 ) is the gap value collected by the i-th optical fiber sensor in each target area at time t+1, d ( t ) is the gap value collected by the i-th optical fiber sensor in each target area at time t, and Δt is the time interval for collecting gap values; (2) Based on the gap change filtering data, the local curvature of each target area is calculated to reflect the curvature of the wafer surface. The formula is: ,in, is the local curvature of each target region, dx is the distance between two adjacent fiber optic sensors in each target area, is the second-order derivative of the gap value collected by the i-th fiber optic sensor in each target area at time t; (3) the gap change rate and local curvature are combined to construct the initial deformation feature array, which can be expressed as: , X represents the initial deformation feature array set, Represent the gap change rates of target area 1, target area 2, and target area n respectively, They represent the local curvatures of target area 1, target area 2, and target area n respectively.

[0047] In some optional implementations of the present application, the initial deformation feature array is subjected to Z-score normalization, and the formula is: ,in, Represents the deformation feature array after Z-score normalization. μ Represents the mean value of deformation characteristics, which is used to eliminate the dimensional differences between different fiber optic sensors and obtain the deformation feature array.

[0048] In some optional implementations of the present application, the dynamic negative pressure value output signal of each gas path adsorption module is generated based on the negative pressure dynamic adjustment strategy, including: constructing a negative pressure dynamic adjustment strategy based on the mapping relationship between typical warping types and negative pressure adjustment logic rules; determining the corresponding negative pressure adjustment logic rules based on the output wafer warping type judgment result, and calculating and generating the dynamic negative pressure value output signal of each gas path adsorption module.

[0049] In some optional implementations of the present application, the negative pressure dynamic adjustment strategy involved in the present application is constructed based on the mapping relationship between typical warping types and negative pressure adjustment logic rules, specifically: (1) For edge warping, the negative pressure adjustment logic rules include: increasing the negative pressure value of the air path adsorption module at the edge area and the negative pressure value of the air path adsorption module at the center area to keep the reference negative pressure value unchanged. "Increasing the negative pressure value of the air path adsorption module at the edge area" can be defined as the formula: , Indicates the dynamic negative pressure value of the gas path adsorption module in the edge area. Indicates the reference negative pressure value, ω indicates the proportional coefficient, and the experimental calibration of ω for edge warping is 0.5. Indicates the gap change rate of the edge area; (2) For the central depression, the negative pressure adjustment logic rules include: increasing the negative pressure value of the gas path adsorption module in the central area and the negative pressure value of the gas path adsorption module in the edge area to keep the reference negative pressure value unchanged. "Increasing the negative pressure value of the gas path adsorption module in the central area" can be defined as the formula , Indicates the dynamic negative pressure value of the gas path adsorption module in the central area. Indicates the gap change rate in the central area. The experimental calibration of ω for the central depression is 0.7; (3) For the local deformation, the negative pressure regulation logic rule includes: applying a pulse negative pressure to the gas path adsorption module in the deformation area, with a duration of 50ms, which can be defined as , It represents the dynamic negative pressure value of the gas path adsorption module in the deformation area, and t represents the duration of the pulse negative pressure applied to the gas path adsorption module in the deformation area, which is calibrated to 50ms in the experiment. represents the gap change rate of the single optical fiber sensor acquisition area, τ represents the pulse decay time constant, and τ is 20ms; (4) For normal conditions, the negative pressure adjustment logic rules include: applying uniform reference negative pressure to all areas .

[0050] In some optional implementations of the present application, according to the wafer warpage type judgment result output by the machine learning classification model, based on the negative pressure dynamic adjustment strategy, the corresponding negative pressure adjustment logic rules are determined, and the dynamic negative pressure value of each air path adsorption module is calculated in combination with the formula corresponding to the negative pressure adjustment logic rule. The output signal of the dynamic negative pressure value is transmitted to the corresponding air path adsorption module, and the negative pressure value of the corresponding target area is dynamically controlled by the corresponding air path adsorption module.

[0051] In some optional implementations of the present application, the transmission of 12-inch wafers is taken as an example to illustrate the performance of the actuator using a single gas path structure and the present application. The performance comparison results of the two are shown in Table 1.

[0052] Table 1 Performance comparison between the actuator with a single air path structure and this application

[0053]

[0054] As can be seen from Table 1, the end effector for the semiconductor wafer transfer robot provided in this application can achieve the purpose of differentiated negative pressure adjustment for the warping conditions of different areas of the wafer through a strategy combining multi-gas path negative pressure dynamic control and real-time feedback mechanism. It not only significantly improves the response speed and adaptability of the end effector of the wafer transfer robot, but also effectively improves the stability and reliability of wafer adsorption.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. An end effector for a semiconductor wafer transfer robot, characterized in that: include: The end effector body is used to carry the wafer; Multiple air path adsorption modules with independent negative pressure control function, the air path adsorption modules are arranged on the end effector body, each of the air path adsorption modules is correspondingly provided with a sensor array, and each of the sensor arrays corresponds to a target area on the wafer surface, and is used to collect the gap value between the target area and the end effector body in real time, and transmit it to the control module, and output a signal according to the received dynamic negative pressure value to dynamically adjust the negative pressure value of the corresponding target area; A control module is used to build and train a machine learning classification model, pre-process the received gap value between the target area and the end effector body, generate a deformation feature array, input the deformation feature array into the trained machine learning classification model, output a wafer warpage type judgment result, and based on the negative pressure dynamic adjustment strategy, generate a dynamic negative pressure value output signal for each gas path adsorption module and transmit it to the corresponding gas path adsorption module; The step of preprocessing the received gap value between the target area and the end effector body to generate a deformation feature array includes: Performing sliding window filtering on the received gap value between the target area and the end effector body to obtain gap change filtering data; Based on the gap change filtering data, the gap change rate and local curvature of each target area are calculated respectively. The gap change rate is used to reflect the speed of local deformation of the wafer, and the local curvature is used to reflect the degree of curvature of the wafer surface. Combine the gap change rate and local curvature to construct the initial deformation feature array; The initial deformation feature array is subjected to Z-score normalization to obtain the deformation feature array.

2. The end effector for a semiconductor wafer transfer robot according to claim 1, characterized in that: Each of the air path adsorption modules includes at least one adsorption unit, and the adsorption unit includes a boss and an air cushion. The air cushion is sealed with the inner countersunk hole of the boss, and a negative pressure hole is opened on the air cushion.

3. The end effector for a semiconductor wafer transfer robot according to claim 2, characterized in that: The end effector body includes a main body, a connecting part and a cover plate. The interior of the main body is provided with an air path channel connected one-to-one with the negative pressure holes. The rear end of the main body is fixedly connected to the connecting part, and the main body and the connecting part are sealed by the cover plate. The interior of the connecting part is provided with an air path pipeline connected one-to-one with the air path channel, and a regulating valve is provided on the air path pipeline.

4. The end effector for a semiconductor wafer transfer robot according to claim 2, wherein: Each of the sensor arrays includes at least one optical fiber sensor, and an optical fiber sensor is provided at the bottom of the side of each boss, which is used to start real-time collection of the gap value between the target area and the end effector body when the distance between the end effector body and the wafer surface is less than a preset distance threshold.

5. The end effector for a semiconductor wafer transfer robot according to claim 3, characterized in that: The gas path adsorption module also includes a vacuum pump using PWM control. The vacuum pump is provided with a D / A conversion module. The vacuum pump is connected to the control module and is used to receive the dynamic negative pressure value output signal transmitted by the control module. The dynamic negative pressure value output signal is mapped to the rotation speed of the vacuum pump through the D / A conversion module to achieve dynamic control of the negative pressure value of the corresponding target area.

6. The end effector for a semiconductor wafer transfer robot according to claim 1, characterized in that: The construction and training of the machine learning classification model includes: Constructing a historical gap value set, wherein the historical gap value set includes multiple training samples, and using multi-classification coding technology to label the typical warping type to which each training sample belongs, thereby obtaining a labeled historical gap value set, wherein the typical warping types include edge warping, center depression, local deformation, and normal state; Use lightweight CNN convolutional neural network to build the initial machine learning classification model; The initial machine learning classification model is trained based on the historical gap value set with labels, and the trained machine learning classification model is obtained by combining the cross entropy loss function.

7. The end effector for a semiconductor wafer transfer robot according to claim 1, characterized in that: The calculation formula for the gap change rate of each target area is: in, is the gap change rate of each target area, d(t+1) is the gap value collected by the i-th optical fiber sensor in each target area at time t+1, d(t) is the gap value collected by the i-th optical fiber sensor in each target area at time t, and Δt is the time interval for collecting gap values.

8. The end effector for a semiconductor wafer transfer robot according to claim 1, wherein: The calculation formula for the local curvature of each target area is: in, is the local curvature of each target area, dx is the distance between two adjacent fiber optic sensors in each target area, is the second-order derivative of the gap value collected by the i-th optical fiber sensor in each target area at time t.

9. The end effector for a semiconductor wafer transfer robot according to claim 1, wherein: The method of generating a dynamic negative pressure value output signal of each gas path adsorption module based on the negative pressure dynamic adjustment strategy includes: Based on the mapping relationship between typical warping types and negative pressure adjustment logic rules, a negative pressure dynamic adjustment strategy is constructed; Based on the output wafer warpage type judgment result, the corresponding negative pressure adjustment logic rules are determined, and the dynamic negative pressure value output signals of each gas path adsorption module are calculated and generated.

Images