Intelligent adjustment control method and system for semiconductor wafer grabbing manipulator

Abstract

The application relates to the technical field of semiconductor manufacturing and automation, in particular to an intelligent adjustment control method and system for a semiconductor wafer grabbing manipulator. The method comprises the following steps: S1: obtaining contact force data and topography data of a wafer surface, constructing a contact force matrix and a topography state vector according to the contact force data and the topography data, and calculating a contact force gradient matrix and a contact force field curvature matrix according to the contact force matrix; S2: calculating an expected contact force of a current semiconductor wafer according to the topography state vector, and calculating a basic adsorption amount of the current semiconductor wafer according to actual contact force, the expected contact force and the contact force field curvature matrix; S3: calculating a gap confidence factor of the wafer according to the contact force field curvature matrix and the topography state vector, and obtaining a decision matrix of the wafer according to the gap confidence factor. The application corrects the adsorption force of the manipulator on the wafer, so that the manipulator is intelligently adjusted.

Description

Technical Field

[0001] This invention relates to the fields of semiconductor manufacturing and automation technology, and in particular to an intelligent adjustment and control method and system for a semiconductor wafer gripping robot. Background Technology

[0002] In the semiconductor wafer manufacturing process, the wafer gripping robot is the core automated equipment connecting various process links. As the wafer size continues to increase and the thickness continues to decrease, the risk of wafer breakage during the gripping process increases sharply. Yield control in the gripping process has become a major problem in semiconductor wafer manufacturing.

[0003] Currently, wafer gripping robots primarily use vacuum adsorption to pick up and release wafers. However, existing technologies still face several unresolved issues: First, the control of vacuum adsorption force lacks sophisticated sensing capabilities. Existing robots typically only use a single pressure sensor and simple on / off vacuum detection on the end effector, failing to acquire information on the contact force distribution at each adsorption point on the wafer surface, let alone calculate the spatial gradient and curvature distribution of the contact force field. Second, existing control strategies offer fast response but lack global coordination, while also exhibiting global optimization but with significant computational latency, making it difficult to integrate these two approaches. Third, the avoidance of wafer orientation notches heavily relies on external visual recognition systems. The technology requires additional configuration of cameras and image processing units, which not only increases the system cost and complexity, but also causes a serious disconnect between image processing latency and fast timescale control. In addition, visual recognition is prone to failure when the wafer surface is reflective, the pattern is complex, or the edge is damaged, leading to avoidance failure. Fourth, existing optimization control methods only focus on the tracking error of single-point adsorption force and the difference between pairs, ignoring the overall smoothness requirement of the contact force field in space. For example, when the force field gradient and curvature are too large, even if the adsorption force at each point is within the set range, the wafer will still break due to excessive local bending moment. Therefore, there is an urgent need to develop an intelligent adjustment and control method and system for semiconductor wafer gripping robots. Summary of the Invention

[0004] To overcome the shortcomings of lacking detection methods for wafer orientation notches and the inability to globally optimize the correction of adsorption force, this invention provides an intelligent adjustment and control method and system for semiconductor wafer gripping robots.

[0005] The technical solution of this invention is: an intelligent adjustment and control method for a semiconductor wafer gripping robot, comprising the following steps:

[0006] S1: Obtain contact force data and morphology data of the wafer surface, construct contact force matrix and morphology state vector based on contact force data and morphology data, and calculate contact force gradient matrix and contact force field curvature matrix based on contact force matrix;

[0007] S2: Calculate the expected contact force of the current semiconductor wafer based on the morphology state vector; calculate the basic adsorption amount of the current semiconductor wafer based on the actual contact force, expected contact force, and contact force field curvature matrix;

[0008] S3: Calculate the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector; obtain the decision matrix of the wafer based on the notch confidence factor; calculate the effective total contact force of the current wafer based on the decision matrix; and determine the wafer adsorption behavior based on the effective total contact force.

[0009] S4: Construct a loss function based on the expected contact force, decision matrix, gradient square norm, and curvature square norm, and obtain the corrected adsorption force increment based on the loss function;

[0010] S5: Obtain the final corrected adsorption amount based on the basic adsorption amount and the corrected adsorption force increment, and calculate the current wafer health status at the same time. Intelligently adjust the robot arm based on the corrected adsorption amount and health status.

[0011] Preferably, the step of acquiring contact force data and morphology data of the wafer surface, constructing a contact force matrix and morphology state vector based on the contact force data and morphology data, and calculating and acquiring a contact force gradient matrix and a contact force field curvature matrix based on the contact force matrix includes: constructing a wafer coordinate system on the end effector of the robot arm; acquiring contact force data at the coordinates of each adsorption point on the robot arm based on the force sensing unit on the end effector of the robot arm, wherein the contact force data is the mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the suction cup at the corresponding adsorption point; acquiring morphology data at the coordinates of each sampling point on the lower surface of the wafer based on the optical ranging array on the end effector of the robot arm, wherein the morphology data is the height deviation value of the lower surface of the wafer relative to the reference plane at the sampling point; and constructing the current wafer coordinate system based on the contact force data. The contact force matrix of all adsorption points is used to construct the morphological state vector of all adsorption points on the current wafer based on the morphological data. The contact force gradient of the current adsorption point on the current wafer is calculated based on the contact force matrix, and the contact force gradient matrix of all adsorption points on the current wafer is obtained based on the contact force gradient of different adsorption points. The curvature of the contact force field of the current adsorption point on the current wafer is obtained by second-order central difference of the horizontal and vertical axes based on the contact force matrix, and the curvature matrix of the contact force field of all adsorption points on the current wafer is obtained. The curvature of the contact force field represents the degree of bending of the force field at different adsorption points on the current wafer. The gradient square norm and the curvature square norm are calculated based on the contact force gradient matrix and the contact force field curvature matrix, respectively. The gradient square norm and the curvature square norm are the squares of the Euclidean norms of the contact force gradient matrix and the contact force field curvature matrix.

[0012] Preferably, the step of calculating the desired contact force of the current semiconductor wafer based on the morphology state vector includes: calculating the warpage characteristic parameter of the current wafer based on the morphology state vector; presetting a local region radius, obtaining the local region of the current adsorption point based on the local region radius, and calculating the local morphology variation coefficient of the current wafer based on the local region of the current adsorption point, wherein the warpage characteristic parameter is the degree of curvature of all adsorption points on the lower surface of the wafer, and the local morphology variation coefficient is the degree of undulation of the local surface of the wafer near a single chuck; calculating the desired contact force of the current adsorption point based on the warpage characteristic parameter and the local morphology variation coefficient, wherein the desired contact force is the desired mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the chuck at the adsorption point, and the formula for calculating the desired contact force is:

[0013] ;

[0014] ;

[0015] ;

[0016] in, In order to be in The first moment The expected contact force at each adsorption point In order to be in Warp characteristic parameters at time t, In order to be in The first moment The coefficient of variation of the local morphology of each adsorption site This is the reference contact force for the current wafer. This is the warpage weighting coefficient. This is the morphological variation weighting coefficient. This serves as a reference value for wafer warpage. The adsorption point number is... This represents the total number of adsorption sites. In order to be in The first time in the shape state vector The height deviation value at each adsorption point It is a function of standard deviation. It is a mean function. It is a very small positive number. For the first The set of morphological height values ​​at the locations of all adjacent adsorption points within a local region centered on an adsorption point and with a radius equal to the radius of the local region.

[0017] Preferably, the step of calculating the basic adsorption amount of the current semiconductor wafer based on the actual contact force, the desired contact force, and the contact force field curvature matrix includes: calculating the contact force tracking error of the current adsorption point of the current wafer based on the actual contact force and the desired contact force, wherein the contact force tracking error is the difference between the actual contact force and the desired contact force at the adsorption point, and the actual contact force is directly read from the contact force matrix; calculating a first adsorption amount based on the contact force tracking error, wherein the first adsorption amount is the pressure control amount of the current adsorption point under the contact force tracking error; calculating a second adsorption amount of the current adsorption point based on the contact force field curvature matrix, wherein the second adsorption amount is the compensation amount for the adsorption force under the influence of the contact force curvature; and obtaining the basic adsorption amount by adding the first adsorption amount and the second adsorption amount, wherein the formula for calculating the basic adsorption amount is:

[0018] ;

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] in, For the first Adsorption sites at The basic adsorption amount at any given time. For the first Adsorption sites at The initial adsorption amount at a given time. For the first Adsorption sites at The second adsorption amount at time [time]. For proportional control gain, For integral control gain, For differential control gain, In order to be in The first moment The expected contact force at each adsorption point For the first The actual contact force at each adsorption point For the first Contact force tracking error at each adsorption point The curvature-pressure conversion factor. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, This is the gap attenuation compensation coefficient. For the first The adsorption sites at time [time] The gap between the suction cups, The corresponding coordinates in the contact force matrix are Contact force.

[0024] Preferably, the step of calculating the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector, and obtaining the decision matrix of the wafer based on the notch confidence factor, includes: converting the morphology state vector into matrix form, obtaining the meshed morphology deviation matrix of the current wafer, calculating the morphology spatial gradient based on the meshed morphology deviation matrix, wherein the morphology spatial gradient is the gradient value of the morphology deviation at different adsorption points, calculating the notch confidence factor of the current wafer in different coordinates based on the morphology spatial gradient and the contact force field curvature matrix, wherein the notch confidence factor is the confidence level that a notch exists on the current wafer at the adsorption point; and presetting the notch confidence factor. A threshold is set by comparing the notch confidence factor with the notch confidence threshold. If the notch confidence factor at the current coordinate is greater than the notch confidence threshold, the current coordinate is selected. A set of confidence points for the current wafer is constructed based on the selected coordinates. This set of confidence points consists of points with high confidence factors at different coordinates. A decision matrix is ​​constructed based on the coordinates in the confidence point set. This decision matrix is ​​a binary matrix used to represent the notch problem on the surface of the current wafer. A minimum confidence point number threshold is preset. If the number of points in the current confidence point set is less than the minimum confidence point number threshold, the decision matrix for the current wafer is an all-zero matrix. The formula for calculating the notch confidence factor is:

[0025] ;

[0026] in, Coordinates are The gap confidence factor These are the force field characteristic weighting coefficients. The morphological feature weighting coefficient. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, It is a positive part function. The first in the gridded topography deviation matrix Line number The value of the column, Coordinates are The morphological space gradient, For norm functions, The force field curvature threshold, This is the topography gradient threshold.

[0027] Preferably, the step of calculating the effective total contact force of the current wafer based on the decision matrix and determining the wafer adsorption behavior based on the effective total contact force includes: calculating the effective total contact force of the current wafer based on the number of elements, element values, and actual contact forces in the decision matrix, wherein the effective total contact force is the sum of the actual contact forces received by the current wafer at all adsorption points; calculating the safe gripping critical force of the current wafer, wherein the safe gripping critical force is the minimum total contact force that the wafer must provide in all non-avoidance states of adsorption points during dynamic movement; and comparing the effective total contact force with the safe gripping critical force for determination. The determination condition is: if the current effective total contact force is greater than the safe gripping critical force, then the current wafer is allowed to be gripped by the robotic arm; if the current effective total contact force is less than the safe gripping critical force, then the current wafer is not allowed to be gripped by the robotic arm and an alarm is triggered. The calculation formulas for the effective total contact force and the safe gripping critical force are as follows:

[0028] ;

[0029] ;

[0030] in, This represents the current effective total contact force of the wafer. This represents the total number of adsorption sites. The adsorption point number is... The decision matrix is ​​the first... The decision value corresponding to each adsorption point For the first The adsorption sites at time [time] The actual contact force, To secure the critical force for current wafer safety, As for the current safety factor of the wafer, For the current quality of the wafer, This is the standard gravitational acceleration.

[0031] Preferably, the step of constructing a loss function based on the desired contact force, decision matrix, gradient square norm, and curvature square norm, and obtaining the corrected adsorption force increment based on the loss function, includes: presetting an adsorption increment vector; calculating the estimated corrected adsorption force of the current wafer at the current adsorption point based on the adsorption increment vector and the pressure sensitivity coefficient; the estimated corrected adsorption force is the contact force estimate of the current wafer after correction for each adsorption point in the actual environment; presetting a slow-scale time window; constructing loss functions for different slow-scale time windows based on the desired contact force, decision matrix, gradient square norm, and curvature square norm; and performing feedback iterative optimization using an optimization algorithm based on constraints to calculate the optimal adsorption increment vector of the current wafer; using the optimal adsorption increment vector of the current wafer as the corrected adsorption increment of the current wafer; the corrected adsorption increment is the difference in contact force of the current wafer after correction for each adsorption point in the actual environment, wherein the calculation formula of the loss function is:

[0032] ; ;

[0033] ;

[0034] ;

[0035] ;

[0036] in, For loss function, , and These are the first, second, and third constraints for the loss function. To correct the estimated adsorption force, The corresponding coordinates in the contact force matrix are Contact force, This is the pressure contact force sensitivity coefficient. The adsorption increment vector is the first... The increase in adsorption force at each adsorption point , , , and These are the first weighting coefficient, the second weighting coefficient, the third weighting coefficient, the fourth weighting coefficient, and the fifth weighting coefficient. For the first The basic adsorption capacity of each adsorption point For the strength of the pressure valve, To secure the critical force for current wafer safety, This is the minimum gap between the suction cup and the bottom of the wafer. This is the maximum gap between the suction cup and the bottom of the wafer. It is the gradient square norm. Let be the square norm of curvature. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point This serves as a reference threshold for gradient deviation. As a reference threshold for curvature deviation, This serves as a reference threshold for adsorption deviation. This is the reference threshold for adjacent deviations.

[0037] Preferably, the step of obtaining the final corrected adsorption amount based on the basic adsorption amount and the corrected adsorption force increment, simultaneously calculating the current wafer health, and intelligently adjusting the robot arm based on the corrected adsorption amount and health includes: adding the basic adsorption amount and the corrected adsorption force increment to the corresponding adsorption points to obtain the final corrected adsorption amount at the current adsorption point, wherein the corrected adsorption amount is the air pressure value of the robot arm adsorbing the wafer after correction; intelligently adjusting the robot arm based on the corrected adsorption amount, wherein the intelligent adjustment involves driving and adjusting the proportional valve of the robot arm according to the corrected adsorption amount to change the vacuum degree in the suction cup at the adsorption point, thereby changing the contact force acting on the wafer; calculating the current wafer health based on the gradient square norm and the curvature square norm, wherein the health degree is the smoothness of the current wafer surface, and adjusting and optimizing the weight set of the loss function based on the health degree, wherein the formula for calculating the health degree is:

[0038] ;

[0039] in, For the current wafer in Health status at all times For the current wafer in The contact force gradient matrix at time t. For the current wafer in The curvature matrix of the contact force field at time t. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point To correct the estimated adsorption force, For the current wafer in The first moment The expected contact force at each adsorption point This serves as a reference threshold for gradient bias. As a reference threshold for curvature deviation, This is the reference threshold for adsorption deviation.

[0040] Preferably, the step of adjusting and optimizing the weight set of the loss function based on health status includes: presetting a fast-scale time window; calculating the health status of the current wafer based on the time detection window; optimizing the weight set of the loss function for the next fast-scale time window based on the health status of different fast-scale time windows, wherein the weight set consists of the first weight coefficient, second weight coefficient, third weight coefficient, fourth weight coefficient, and fifth weight coefficient in the loss function; and dynamically adjusting the loss function of the robot arm based on health status, wherein the calculation formula for dynamic adjustment is:

[0041] ;

[0042] ;

[0043] ;

[0044] in, For the first The fourth weighting coefficient for each fast-scale time window. For the first The fifth weighting coefficient for a fast-scale time window. This is the base maximum value of the fourth weighting coefficient. This is the base maximum value of the fifth weighting coefficient. For the first Health status within a fast-scale time window For the first The first weighting coefficient of a fast-scale time window.

[0045] Preferably, the intelligent adjustment and control system for the semiconductor wafer gripping robot further includes:

[0046] The data matrix module acquires contact force data and morphology data of the wafer surface, constructs a contact force matrix and morphology state vector based on the contact force data and morphology data, and calculates and obtains the contact force gradient matrix and contact force field curvature matrix based on the contact force matrix.

[0047] The basic adsorption module calculates the expected contact force of the current semiconductor wafer based on the morphology state vector; and calculates the basic adsorption amount of the current semiconductor wafer based on the actual contact force, expected contact force, and contact force field curvature matrix.

[0048] The adsorption determination module calculates the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector, obtains the decision matrix of the wafer based on the notch confidence factor, calculates the effective total contact force of the current wafer based on the decision matrix, and determines the wafer adsorption behavior based on the effective total contact force.

[0049] The adsorption correction module constructs a loss function based on the desired contact force, decision matrix, gradient square norm, and curvature square norm, and obtains the correction adsorption force increment based on the loss function.

[0050] The health adjustment module obtains the final corrected adsorption amount based on the basic adsorption amount and the incremental correction adsorption force, and calculates the current health of the wafer. It then intelligently adjusts the robotic arm based on the corrected adsorption amount and the health.

[0051] The present invention has the following advantages:

[0052] 1. This invention employs a control architecture that works in tandem with two time scales: fast time scale control can respond instantly to sudden vibrations and air pressure fluctuations within microsecond cycles, preventing wafers from detaching instantly and overpressure; slow time scale control coordinates and optimizes the contact force distribution of all suction cups within millisecond cycles, avoiding uneven force distribution across the entire area caused by each suction cup acting independently.

[0053] 2. This invention enables the contact force on the entire wafer surface to smoothly transition from the center to the edge by forcibly constraining the difference slope of the contact force between adjacent chucks and the curvature of the force field distribution in global optimization. By actively suppressing the force field gradient and curvature, the bending stress inside the wafer is minimized.

[0054] 3. The present invention automatically adjusts the control center of gravity according to the smoothness of the current global contact force field: when the force field distribution is healthy, the focus is on maintaining the accurate tracking of each suction cup; when a deterioration trend of the force field is detected, the control center of gravity is automatically transferred to repair the spatial shape of the force field, giving priority to ensuring the overall force balance of the wafer. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the intelligent adjustment and control method for a semiconductor wafer gripping robot according to the present invention.

[0056] Figure 2 This is a schematic diagram of the intelligent adjustment and control system for the semiconductor wafer gripping robot of the present invention. Detailed Implementation

[0057] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0058] Example 1

[0059] Intelligent adjustment and control methods for semiconductor wafer gripping robots, such as Figure 1 As shown, it includes the following steps:

[0060] S1: Obtain contact force data and morphology data of the wafer surface, construct contact force matrix and morphology state vector based on contact force data and morphology data, and calculate contact force gradient matrix and contact force field curvature matrix based on contact force matrix;

[0061] S2: Calculate the expected contact force of the current semiconductor wafer based on the morphology state vector; calculate the basic adsorption amount of the current semiconductor wafer based on the actual contact force, expected contact force, and contact force field curvature matrix;

[0062] S3: Calculate the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector; obtain the decision matrix of the wafer based on the notch confidence factor; calculate the effective total contact force of the current wafer based on the decision matrix; and determine the wafer adsorption behavior based on the effective total contact force.

[0063] S4: Construct a loss function based on the expected contact force, decision matrix, gradient square norm, and curvature square norm, and obtain the corrected adsorption force increment based on the loss function;

[0064] S5: Obtain the final corrected adsorption amount based on the basic adsorption amount and the corrected adsorption force increment, and calculate the current wafer health status at the same time. Intelligently adjust the robot arm based on the corrected adsorption amount and health status.

[0065] A wafer coordinate system is constructed on the end effector of the robotic arm. Contact force data at the coordinates of each adsorption point on the robotic arm is acquired using the force sensing unit on the end effector. This contact force data represents the mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the suction cup at the corresponding adsorption point. Shape data at the coordinates of each sampling point on the lower surface of the wafer is acquired using the optical ranging array on the end effector. This shape data represents the height deviation of the lower surface of the wafer relative to a reference plane at the sampling point. A contact force matrix is ​​constructed for all adsorption points on the current wafer based on the contact force data, and a shape state vector is constructed for all adsorption points on the current wafer based on the shape data. The contact force matrix is ​​calculated to obtain the contact force gradient of the current adsorption point on the current wafer. Based on the contact force gradient of different adsorption points, the contact force gradient matrix of all adsorption points on the current wafer is obtained. Based on the contact force matrix, the curvature of the contact force field of the current adsorption point on the current wafer is obtained through the second-order central difference of the horizontal and vertical axes. The curvature matrix of the contact force field of all adsorption points on the current wafer is obtained. The curvature of the contact force field represents the degree of bending of the force field at different adsorption points on the current wafer. Based on the contact force gradient matrix and the contact force field curvature matrix, the gradient square norm and the curvature square norm are calculated respectively. The gradient square norm and the curvature square norm are the squares of the Euclidean norms of the contact force gradient matrix and the contact force field curvature matrix.

[0066] It needs to be explained that the coordinate system is constructed with the geometric center of the end effector as the origin, the plane where the end effector is located as the reference plane, and the direction of the robot's movement as the positive direction of each coordinate axis; the contact force matrix is ​​constructed based on the coordinates of different adsorption points, with rows corresponding to the horizontal coordinates and columns corresponding to the vertical coordinates; the morphology state vector is a set of vectors with different sampling points sorted by their serial numbers and morphology data as values. Specifically, the morphology data of different sampling points is mapped to a grid coordinate system of the same dimension as the contact force matrix through bilinear interpolation to obtain the morphology data of the adsorption points. The adsorption points are then sorted and arranged according to their serial numbers to construct an initial morphology state vector. The morphology data of the current adsorption point is then used as the value to construct the complete morphology state vector. The specific acquisition of the sampling points... The sampling points are determined by the distribution of the optical ranging array, specifically the edge region of the end effector. The contact force field curvature is obtained as follows: based on the adsorption points within the contact force matrix, the second-order difference in the row direction is calculated as follows: the contact force of the right adjacent point minus twice its own contact force, plus the contact force of the left adjacent point, divided by the square of the row spacing. The second-order difference in the column direction is calculated as follows: the contact force of the lower adjacent point minus twice its own contact force, plus the contact force of the upper adjacent point, divided by the square of the column spacing. The sum of the row and column second-order differences is the contact force field curvature at the current adsorption point. The contact force field curvature matrix is ​​constructed based on the coordinates of different adsorption points, with rows corresponding to the horizontal coordinate and columns corresponding to the vertical coordinate, where the value represents the contact force field curvature at the current adsorption point. The formula for calculating the contact force gradient is as follows:

[0067] ;

[0068] ;

[0069] ;

[0070] in, Here is the contact force gradient matrix. The gradient component matrix is ​​the matrix of the row direction. The column direction gradient component matrix, The first row gradient component matrix is ​​the first row gradient component matrix. Line number Column elements, The column direction gradient component matrix is ​​the first Line number Column elements, The first in the contact force matrix Line number Column elements, The first in the contact force matrix Line number Column elements; The first in the contact force matrix Line number Column elements; The physical distance between adjacent adsorption points in the row direction. This represents the physical spacing between adjacent adsorption points in the column direction.

[0071] The warpage characteristic parameter of the current wafer is calculated based on the morphology state vector; a local neighborhood radius is preset, and the local neighborhood of the current adsorption point is obtained based on the local neighborhood radius. The local morphology variation coefficient of the current wafer is calculated based on the local neighborhood of the current adsorption point. The warpage characteristic parameter is the degree of curvature of all adsorption points on the lower surface of the wafer, and the local morphology variation coefficient is the degree of undulation of the local surface of the wafer near a single chuck; the expected contact force of the current adsorption point is calculated based on the warpage characteristic parameter and the local morphology variation coefficient. The expected contact force is the expected mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the chuck at the adsorption point. The formula for calculating the expected contact force is:

[0072] ;

[0073] ;

[0074] ;

[0075] in, In order to be in The first moment The expected contact force at each adsorption point In order to be in Warp characteristic parameters at time t, In order to be in The first moment The coefficient of variation of the local morphology of each adsorption site This is the reference contact force for the current wafer. This is the warpage weighting coefficient. This is the morphological variation weighting coefficient. This serves as a reference value for wafer warpage. The adsorption point number is... This represents the total number of adsorption sites. In order to be in The first time in the shape state vector The height deviation value at each adsorption point It is a function of standard deviation. It is a mean function. It is a very small positive number. For the first The set of morphological height values ​​at the locations of all adjacent adsorption points within a local region centered on an adsorption point and with a radius equal to the radius of the local region.

[0076] It should be explained that the reference contact force, under ideal working conditions, is the standard bearing capacity of a single adsorption point, measured in Newtons (N). Specifically, it is obtained by dividing the current wafer's weight by the total number of adsorption points to obtain the unit contact force for that adsorption point. Experiments are conducted based on this unit contact force, gradually increasing the contact force from the center to obtain a contact force curve until no wafer slippage occurs. The desired contact force is measured in Newtons. The local morphology variation coefficient, warpage weighting coefficient, and morphology variation weighting coefficient are all dimensionless parameters. The warpage characteristic parameter and the wafer warpage reference value are measured in micrometers. The local radius is obtained by statistically analyzing the distances between all adsorption points and other adsorption points, sorting them in descending order to obtain the distance sequence for the current adsorption point, using the median of the distance sequence as the initial radius of the current adsorption point, and then using the initial local radius of all adsorption points... The average value is calculated as the local radius; the warpage weight coefficient is the correction strength of the warpage feature parameter; the morphology variation weight coefficient is the correction strength of the local morphology variation coefficient. The specific methods for obtaining the warpage weight coefficient and the morphology variation weight coefficient are as follows: collect a large amount of historical data captured from wafers, including warpage, morphology variation coefficient, and breakage rate, establish a function model between the morphology variation weight coefficient and the breakage rate, and a function model between the warpage weight coefficient and the breakage rate, and automatically search for the morphology variation weight coefficient and the warpage weight coefficient with the minimum breakage rate through a genetic algorithm; the wafer warpage reference value is the reference bending degree of all adsorption points on the lower surface of the wafer. The specific method for obtaining the wafer warpage reference value is as follows: select a batch of wafers with normal process for sampling measurement, and use the measured average warpage as the reference value under the current process conditions.

[0077] The contact force tracking error at the current adsorption point on the current wafer is calculated based on the actual contact force and the expected contact force. This contact force tracking error is the difference between the actual and expected contact forces at the adsorption point. The actual contact force is directly read from the contact force matrix. A first adsorption amount is calculated based on the contact force tracking error, which is the pressure control amount at the current adsorption point under the contact force tracking error. A second adsorption amount is calculated based on the contact force field curvature matrix, which is the compensation amount for the adsorption force under the influence of the contact force curvature. The base adsorption amount is obtained by adding the first and second adsorption amounts, where the formula for calculating the base adsorption amount is:

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] in, For the first Adsorption sites at The basic adsorption amount at any given time. For the first Adsorption sites at The initial adsorption amount at a given time. For the first Adsorption sites at The second adsorption amount at time [time]. For proportional control gain, For integral control gain, For differential control gain, In order to be in The first moment The expected contact force at each adsorption point For the first The actual contact force at each adsorption point For the first Contact force tracking error at each adsorption point The curvature-pressure conversion factor. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, This is the gap attenuation compensation coefficient. For the first The adsorption sites at time [time] The gap between the suction cups, The corresponding coordinates in the contact force matrix are Contact force.

[0084] It should be explained that the gridded morphology deviation matrix is ​​composed of morphology data from different sampling points. Specifically, it is constructed as follows: the horizontal and vertical axes represent the number of rows and columns, and the morphology data of the current adsorption point is used as the value. The units for the basic adsorption capacity, the first adsorption capacity, and the second adsorption capacity are Pascals; the unit for the contact force tracking error is Newtons; the unit for the suction cup gap is millimeters; and the unit for the force field curvature is Newtons per square millimeter. The proportional control gain is the ratio of the conversion gain between the contact force tracking error and the adsorption capacity, in Pascals per Newton. The integral control gain is the ratio of the conversion gain between the integral of the contact force tracking error and the adsorption capacity, in Pascals per Newton-second. The differential control gain is the ratio of the conversion gain between the differential of the contact force tracking error and the adsorption capacity, in Pascals per Newton-second. The specific method for obtaining the proportional control gain, integral control gain, and differential control gain is as follows: while calculating the current ratio, the other two ratios are kept constant. The current ratio is gradually increased, and the response curve of the contact force to the desired value is observed. When the contact force response begins to show a continuous oscillation with equal amplitude, the current ratio is the limit of the current ratio. The value is determined by taking half of the current proportional limit based on the Ziegler-Nichols empirical formula. The curvature-pressure conversion coefficient is the proportional relationship between the change in force field curvature and the required pressure compensation, expressed in Pascals per millimeter per Newton. Specifically, it is obtained by selecting a standard test wafer with the same mechanical properties as the target wafer, calculating the linear relationship between the curvature change and the contact force change, obtaining the gas supply pressure change based on the contact force change, dividing the gas supply pressure change by the curvature change to obtain the curvature-pressure conversion coefficient of the current adsorption point, and taking the average value from multiple experiments as the final curvature-pressure conversion coefficient. The gap attenuation compensation coefficient is the rate attenuation of the feedforward compensation effect as the suction cup gap increases, expressed in millimeters. Specifically, it is obtained by fixing the gas supply pressure on an offline experimental platform, gradually changing the gap distance between the suction cup and the wafer, recording the contact force values ​​corresponding to different gap distances, plotting the curve of contact force versus gap, fitting the curve using an exponential function model, and obtaining the coefficient of the fitted exponential term as the gap attenuation compensation coefficient.

[0085] The morphology state vector is converted into a matrix form to obtain the meshed morphology deviation matrix of the current wafer. The morphology spatial gradient is calculated based on the meshed morphology deviation matrix, where the gradient value of the morphology deviation at different adsorption points is the gradient value. The notch confidence factor of the current wafer at different coordinates is calculated based on the morphology spatial gradient and the contact force field curvature matrix. The notch confidence factor represents the confidence level that a notch exists at the adsorption point of the current wafer. A notch confidence threshold is preset, and the notch confidence factor is compared with the notch confidence threshold. If the notch confidence factor at the current coordinate is greater than the notch confidence threshold, the current coordinate is selected. A confidence point set for the current wafer is constructed based on the selected coordinates. The confidence point set is the set of points with high confidence factors at different coordinates. A decision matrix is ​​constructed based on the coordinates in the confidence point set. The decision matrix is ​​a binary matrix used to represent the notch problem on the surface of the current wafer. A minimum confidence point number threshold is preset. If the number of points in the current confidence point set is less than the minimum confidence point number threshold, the decision matrix of the current wafer is a zero matrix. The formula for calculating the notch confidence factor is:

[0086] ;

[0087] in, Coordinates are The gap confidence factor These are the force field characteristic weighting coefficients. The morphological feature weighting coefficient. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, It is a positive part function. The first in the gridded topography deviation matrix Line number The value of the column, Coordinates are The morphological space gradient, For norm functions, The force field curvature threshold, This is the topography gradient threshold.

[0088] It should be explained that the force field curvature threshold is the boundary value of force field curvature normalization, and the topography gradient threshold is the boundary value of topography gradient normalization. Specifically, they are obtained by selecting the maximum force field curvature from historical data as the force field curvature threshold and the maximum topography gradient from historical data as the topography gradient threshold. The specific calculation method for the topography spatial gradient is as follows:

[0089] ;

[0090] in, Coordinates are The morphological space gradient, The first in the gridded topography deviation matrix Line number Column elements, The first in the gridded topography deviation matrix Line number Column elements; The first in the gridded topography deviation matrix Line number Column elements; The physical distance between adjacent adsorption points in the row direction. The physical spacing between adjacent adsorption points in the column direction; the notch confidence threshold is the maximum confidence level for the current coordinate as a notch; the specific implementation of the bilinear interpolation algorithm is as follows: using the weighted average of the four original measurement points closest to the target grid point as the morphology deviation value of the target grid point, thereby ensuring that each adsorption point coordinate position corresponds to a unique morphology height value, and obtaining the coordinates and morphology height values ​​of the adsorption points; the notch confidence factor is a dimensionless parameter, and the notch confidence threshold is the threshold at which the current wafer indicates that a notch has appeared in the current wafer region. Specifically, under the unloaded state of an ideally flat wafer, contact force data and morphology data of each adsorption point are collected to obtain force noise and morphology noise, and the maximum possible confidence factor caused by pure noise is calculated. Different multiples of the maximum possible confidence factor are used as candidate notch confidence thresholds. The candidate notch confidence thresholds are then substituted back into the notch confidence factor calculation formula to obtain candidate force field curvature values, and the candidate force field curvature values ​​are minimized. The candidate notch confidence threshold is used as the notch confidence threshold. The specific method for obtaining the minimum confidence point number threshold is as follows: statistically analyze the confidence point set in historical data, select the minimum confidence point set that determines the current wafer defect status, and use the number of the minimum confidence point set as the minimum confidence point number threshold. The specific method for constructing the decision matrix is ​​as follows: if the current coordinate exists in the confidence point set, the decision value of the current coordinate is one; if it does not exist, it is zero. The number of rows and columns of the matrix elements are the horizontal and vertical coordinates of the current sampling point. The force field feature weight coefficient is the coefficient for constraining the contact force, and the morphology feature weight coefficient is the coefficient for constraining the morphology feature. The specific method for obtaining the force field feature weight coefficient and the morphology feature weight coefficient is as follows: statistically analyze historical data to obtain the number of wafer damages caused by contact force and the number of wafer damages caused by morphology feature differences. Use the quotient of the number of wafer damages caused by contact force and the total number of damages as the force field feature weight coefficient, and use the quotient of the number of wafer damages caused by morphology feature differences and the total number of damages as the morphology feature weight coefficient.

[0091] The effective total contact force of the current wafer is calculated based on the number and value of elements in the decision matrix and the actual contact force. The effective total contact force is the sum of the actual contact forces experienced by the current wafer at all adsorption points. The safe gripping critical force of the current wafer is then calculated. The safe gripping critical force is the minimum total contact force that must be provided by all adsorption points in non-avoidance states during dynamic movement of the wafer. A comparison is made between the effective total contact force and the safe gripping critical force. The determination condition is: if the current effective total contact force is greater than the safe gripping critical force, the current wafer is allowed to be gripped by the robotic arm; if the current effective total contact force is less than the safe gripping critical force, the current wafer is not allowed to be gripped by the robotic arm, and an alarm is triggered. The formulas for calculating the effective total contact force and the safe gripping critical force are as follows:

[0092] ;

[0093] ;

[0094] in, This represents the current effective total contact force of the wafer. This represents the total number of adsorption sites. The adsorption point number is... The decision matrix is ​​the first... The decision value corresponding to each adsorption point For the first The adsorption sites at time [time] The actual contact force, To secure the critical force for current wafer safety, As for the current safety factor of the wafer, For the current quality of the wafer, This is the standard gravitational acceleration.

[0095] When it needs to be explained, the units of effective total contact force and actual contact force are Newtons (N); the safety factor refers to the strictness of the safety boundary, and is specifically obtained as follows: an accelerometer is attached to the wafer surface, and gripping and handling tests are performed with different acceleration curves. The critical acceleration value at which the wafer just slips and shifts is recorded. The minimum total contact force required is obtained based on the critical acceleration value. The minimum total contact force is divided by the wafer weight to obtain the critical safety factor. The final safety factor is calculated based on the 1.5 times standard method and the critical safety factor; the specific method for obtaining wafer quality is to directly read it from the equipment automation system according to the wafer specifications.

[0096] A preset adsorption increment vector is used. Based on the adsorption increment vector and the pressure sensitivity coefficient, a corrected adsorption force estimate of the current wafer at the current adsorption point is calculated. The corrected adsorption force estimate is the contact force estimate of the current wafer after correction for each adsorption point under actual conditions. A preset slow-scale time window is used. Based on the expected contact force, decision matrix, gradient square norm, and curvature square norm, loss functions for different slow-scale time windows are constructed. Based on the constraints, an optimization algorithm is used to perform feedback iterative optimization to calculate the optimal adsorption increment vector of the current wafer. The optimal adsorption increment vector of the current wafer is used as the corrected adsorption increment of the current wafer. The corrected adsorption increment is the contact force difference of the current wafer after correction for each adsorption point under actual conditions. The calculation formula of the loss function is:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] in, For loss function, , and These are the first, second, and third constraints for the loss function. adsorption point Corresponding coordinates The corrected adsorption force estimate, adsorption point Corresponding coordinates The corrected adsorption force estimate, The corresponding coordinates in the contact force matrix are Contact force, This is the pressure contact force sensitivity coefficient. The adsorption increment vector is the first... The increase in adsorption force at each adsorption point , , , and These are the first weighting coefficient, the second weighting coefficient, the third weighting coefficient, the fourth weighting coefficient, and the fifth weighting coefficient. For the first The basic adsorption capacity of each adsorption point For the strength of the pressure valve, To secure the critical force for current wafer safety, This is the minimum gap between the suction cup and the bottom of the wafer. This is the maximum gap between the suction cup and the bottom of the wafer. It is the gradient square norm. Let be the square norm of curvature. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point This serves as a reference threshold for gradient bias. As a reference threshold for curvature deviation, This serves as a reference threshold for adsorption deviation. This is the reference threshold for adjacent deviations.

[0103] It should be explained that the slow-scale time window is the current working cycle of the robotic arm grasping a wafer in a single operation. The initial value of the adsorption increment vector is preset to a zero vector. The final adsorption increment vector of Huqiu is calculated based on the loss function and constraints. The optimization algorithm is the effective set method. The specific implementation is as follows: a feasible initial point is selected and an initial effective set is determined. The search direction and search step size are determined by solving the equality constraints through the initial effective set. The initial point and effective set are updated according to the search direction and search step size. The convergence condition is the maximum number of iterations. The specific method for obtaining the maximum number of iterations is: the adsorption increment vector is calculated using the optimization algorithm based on a known standard wafer. The adsorption increment vector and the actual adsorption... The incremental vector is used for difference calculation, and the number of iterations when the difference is minimized is the maximum number of iterations. The pressure-contact force sensitivity coefficient is the amount by which the contact force at the adsorption point will increase for each unit increase in supply pressure, in Newtons per Pascal. Specifically, it is obtained by applying a known step change to the supply pressure at a certain adsorption point under stable operating conditions, recording the steady-state change in contact force, and dividing the change in contact force by the change in supply pressure to obtain the pressure sensitivity coefficient of the current adsorption point. The first weighting coefficient, second weighting coefficient, third weighting coefficient, fourth weighting coefficient, and fifth weighting coefficient are all dimensionless parameters, where the second and third weighting coefficients are fixed values, and the first weighting coefficient... The coefficients, fourth weighting coefficient, and fifth weighting coefficient are non-fixed values. The specific method for obtaining the second weighting coefficient is as follows: Take a test wafer with the same thickness and mechanical properties as the target wafer, attach strain gauges to key positions on the wafer surface, and artificially create uneven distribution of adsorption forces of varying degrees during the gripping experiment. Record the internal stress of the wafer measured by the strain gauges. By analyzing the correspondence between the degree of force difference and the internal stress of the wafer, obtain the maximum allowable force difference range that ensures the internal stress of the wafer does not exceed the safety threshold. Based on the maximum allowable force difference range, deduce the weight ratio that the global equilibrium term should occupy in the loss function to obtain the calibration value of the second weighting coefficient; the specific method for obtaining the third weighting coefficient is as follows: A wafer chuck contact model was established using finite element analysis software. Simulations were performed where different numbers of adsorption points in the notch region were marked as avoidance points. The minimum weight value that reduces the contact force at the avoided point to a level that does not cause excessive compensation at surrounding points was selected as the calibration value of the third weight coefficient. The fourth and fifth weight coefficients were obtained by calculating their base maximum values ​​and health status. The pressure valve strength was the maximum pressure the wafer could withstand. Specifically, it was obtained by continuously applying pressure at a constant rate to a complete standard wafer until it fractured, recording the critical load value at fracture, and then converting it using the elastic thin-plate theory formula to obtain the pressure valve strength.Because the surface of the wafer is uneven, the maximum gap is the maximum spatial distance between the robot and the wafer when the robot is adsorbing it, and the minimum gap is the minimum spatial distance when the robot is not adsorbing it. The maximum gap is obtained by measuring the contact depth between the robot and the wafer after vacuum adsorption deformation, and the minimum gap is obtained by measuring the maximum height deviation of the wafer relative to the reference plane in its free state. The gradient deviation reference threshold is a reference value for gradient deviation, specifically obtained by: establishing a three-dimensional structural model of the end effector using finite element analysis, applying different force field gradient loads, calculating the deformation of each part of the end effector, and using the force field gradient value corresponding to the maximum allowable elastic deformation of the end effector as the gradient deviation reference threshold. The curvature deviation reference threshold is a reference value for curvature deviation, specifically obtained by: establishing a three-dimensional structural model of the wafer material using fracture mechanics analysis. The stress-strain constitutive model is used to calculate the force field curvature value of the wafer under the ultimate bending moment, and the force field curvature value is used as the curvature deviation reference threshold. The adsorption deviation reference threshold is a reference value for adsorption deviation, specifically obtained by multiplying the number of effective adsorption points by the square of the maximum allowable single-point adsorption deviation, in Newtons. The adjacent deviation reference threshold is a reference value for deviation between adjacent adsorption points, specifically obtained by multiplying the number of pairwise combinations of adsorption points by the maximum two-point force difference. The maximum single-point adsorption deviation is the adsorption deviation value of a single adsorption point, and the maximum two-point force difference is the maximum contact force difference between any two adsorption points. Specifically, the initial value of the maximum single-point adsorption deviation is set to 10% of the reference contact force, and the initial value of the maximum two-point force difference is set to 20% of the reference contact force. The initial values ​​are corrected through standard wafer trial gripping experiments to obtain the final maximum single-point adsorption deviation and maximum two-point force difference.

[0104] The basic adsorption amount and the incremental correction adsorption force are added together at the corresponding adsorption points to obtain the final corrected adsorption amount at the current adsorption point. This corrected adsorption amount is the pressure value at which the robotic arm adsorbs the wafer after correction. The robotic arm is intelligently adjusted based on this corrected adsorption amount. This intelligent adjustment involves driving and adjusting the proportional valve of the robotic arm according to the corrected adsorption amount, changing the vacuum level within the suction cup at the adsorption point, thereby changing the contact force acting on the wafer. The current wafer health level is calculated based on the gradient square norm and the curvature square norm. This health level represents the smoothness of the current wafer surface. The weight set of the loss function is adjusted and optimized based on the health level. The formula for calculating the health level is:

[0105] ;

[0106] in, For the current wafer in Health status at all times For the current wafer in The contact force gradient matrix at time t. For the current wafer in The curvature matrix of the contact force field at time t. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point To correct the estimated adsorption force, For the current wafer in The first moment The expected contact force at each adsorption point This serves as a reference threshold for gradient bias. As a reference threshold for curvature deviation, This is the reference threshold for adsorption deviation.

[0107] A preset fast-scale time window is used to calculate the health of the current wafer based on the time detection window. The weight set of the loss function for the next fast-scale time window is optimized based on the health of different fast-scale time windows. The weight set consists of the first, second, third, fourth, and fifth weight coefficients in the loss function. The loss function of the robot arm is dynamically adjusted based on the health status, and the calculation formula for this dynamic adjustment is as follows:

[0108] ;

[0109] ;

[0110] ;

[0111] in, For the first The fourth weighting coefficient for each fast-scale time window. For the first The fifth weighting coefficient for a fast-scale time window. This is the base maximum value of the fourth weighting coefficient. This is the base maximum value of the fifth weighting coefficient. For the first Health status within a fast-scale time window For the first The first weighting coefficient of a fast-scale time window.

[0112] It should be explained that the fast-scale time window is the time period for refreshing the weight set of the loss function. Specifically, it is obtained by multiplying the current slow-scale time window of the robot. The health of the current fast-scale time window is obtained by averaging the health calculated at all times within the current fast-scale time window. The basic maximum value of the fourth weight coefficient is obtained by determining the maximum allowable force field gradient value of the end effector through finite element analysis, and then inferring the maximum weight ratio of the gradient space regularization term in the loss function based on the maximum force field gradient value, thus obtaining the basic maximum value of the fourth weight coefficient. The basic maximum value of the fifth weight coefficient is obtained by determining the maximum allowable force field curvature value of the wafer material through fracture mechanics analysis, and then inferring the maximum weight ratio of the curvature space regularization term in the loss function based on the maximum force field curvature value, thus obtaining the basic maximum value of the fifth weight coefficient.

[0113] Example 2: Based on Example 1, an intelligent adjustment and control system for a semiconductor wafer gripping robot, such as... Figure 2 As shown, it also includes:

[0114] The data matrix module acquires contact force data and morphology data of the wafer surface, constructs a contact force matrix and morphology state vector based on the contact force data and morphology data, and calculates and obtains the contact force gradient matrix and contact force field curvature matrix based on the contact force matrix.

[0115] The basic adsorption module calculates the expected contact force of the current semiconductor wafer based on the morphology state vector; and calculates the basic adsorption amount of the current semiconductor wafer based on the actual contact force, expected contact force, and contact force field curvature matrix.

[0116] The adsorption determination module calculates the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector, obtains the decision matrix of the wafer based on the notch confidence factor, calculates the effective total contact force of the current wafer based on the decision matrix, and determines the wafer adsorption behavior based on the effective total contact force.

[0117] The adsorption correction module constructs a loss function based on the desired contact force, decision matrix, gradient square norm, and curvature square norm, and obtains the correction adsorption force increment based on the loss function.

[0118] The health adjustment module obtains the final corrected adsorption amount based on the basic adsorption amount and the incremental correction adsorption force, and calculates the current health of the wafer. It then intelligently adjusts the robotic arm based on the corrected adsorption amount and the health.

[0119] The technical principles of the embodiments of the present invention have been described above with reference to specific examples. These descriptions are merely for explaining the principles of the embodiments of the present invention and should not be construed as limiting the scope of protection of the embodiments of the present invention in any way. Based on the explanation herein, those skilled in the art can conceive of other specific embodiments of the present invention without creative effort, and these embodiments will all fall within the scope of protection of the embodiments of the present invention.

Claims

1. An intelligent adjustment and control method for a semiconductor wafer gripping robot, characterized in that, Includes the following steps: S1: Obtain contact force data and morphology data of the wafer surface, construct contact force matrix and morphology state vector based on contact force data and morphology data, and calculate contact force gradient matrix and contact force field curvature matrix based on contact force matrix; S2: Calculate the expected contact force of the current semiconductor wafer based on the morphology state vector; calculate the basic adsorption amount of the current semiconductor wafer based on the actual contact force, expected contact force, and contact force field curvature matrix; S3: Calculate the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector, and obtain the decision matrix of the wafer based on the notch confidence factor; The effective total contact force of the current wafer is calculated based on the decision matrix, and the wafer adsorption behavior is determined based on the effective total contact force. S4: Construct a loss function based on the expected contact force, decision matrix, gradient square norm, and curvature square norm, and obtain the corrected adsorption force increment based on the loss function; S5: Obtain the final corrected adsorption amount based on the basic adsorption amount and the corrected adsorption force increment, and calculate the current wafer health status at the same time. Intelligently adjust the robot arm based on the corrected adsorption amount and health status.

2. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The process of acquiring contact force data and morphology data of the wafer surface, constructing a contact force matrix and morphology state vector based on the contact force data and morphology data, and calculating the contact force gradient matrix and contact force field curvature matrix based on the contact force matrix includes: constructing a wafer coordinate system on the end effector of the robot arm; acquiring contact force data at the coordinates of each adsorption point on the robot arm based on the force sensing unit on the end effector of the robot arm, wherein the contact force data is the mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the suction cup at the corresponding adsorption point; acquiring morphology data at the coordinates of each sampling point on the lower surface of the wafer based on the optical ranging array on the end effector of the robot arm, wherein the morphology data is the height deviation value of the lower surface of the wafer relative to the reference plane at the sampling point; and constructing a contact force gradient matrix and contact force field curvature matrix for the current wafer based on the contact force data. The contact force matrix of the adsorption points is constructed by building a morphology state vector of all adsorption points on the current wafer based on morphology data. The contact force gradient of the current adsorption point on the current wafer is calculated based on the contact force matrix, and the contact force gradient matrix of all adsorption points on the current wafer is obtained based on the contact force gradient of different adsorption points. The curvature of the contact force field of the current adsorption point on the current wafer is obtained by second-order central difference of the horizontal and vertical axes based on the contact force matrix, and the curvature matrix of the contact force field of all adsorption points on the current wafer is obtained. The curvature of the contact force field represents the degree of bending of the force field at different adsorption points on the current wafer. The gradient square norm and the curvature square norm are calculated based on the contact force gradient matrix and the contact force field curvature matrix, respectively. The gradient square norm and the curvature square norm are the squares of the Euclidean norms of the contact force gradient matrix and the contact force field curvature matrix.

3. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The step of calculating the desired contact force of the current semiconductor wafer based on the morphology state vector includes: calculating the warpage characteristic parameter of the current wafer based on the morphology state vector; presetting a local region radius; obtaining the local region of the current adsorption point based on the local region radius; calculating the local morphology variation coefficient of the current wafer based on the local region of the current adsorption point; the warpage characteristic parameter being the degree of curvature of all adsorption points on the lower surface of the wafer; the local morphology variation coefficient being the degree of undulation of the local surface of the wafer near a single chuck; calculating the desired contact force of the current adsorption point based on the warpage characteristic parameter and the local morphology variation coefficient; the desired contact force being the desired mechanical normal contact reaction force between the lower surface of the wafer and the upper surface of the chuck at the adsorption point; wherein the formula for calculating the desired contact force is: ； ； ； in, In order to be in The first moment The expected contact force at each adsorption point In order to be in Warp characteristic parameters at time t, In order to be in The first moment The coefficient of variation of the local morphology of each adsorption site This is the current reference contact force for the wafer. This is the warpage weighting coefficient. This is the morphological variation weighting coefficient. This serves as a reference value for wafer warpage. The adsorption point number is... This represents the total number of adsorption sites. In order to be in The first time in the shape state vector The height deviation value at each adsorption point It is a function of standard deviation. It is a mean function. It is a very small positive number. For the first The set of morphological height values ​​at the locations of all adjacent adsorption points within a local region centered on an adsorption point and with a radius equal to the radius of the local region.

4. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The calculation of the basic adsorption amount of the current semiconductor wafer based on the actual contact force, the expected contact force, and the contact force field curvature matrix includes: calculating the contact force tracking error of the current adsorption point of the current wafer based on the actual contact force and the expected contact force, wherein the contact force tracking error is the difference between the actual contact force and the expected contact force at the adsorption point, and the actual contact force is directly read from the contact force matrix; calculating a first adsorption amount based on the contact force tracking error, wherein the first adsorption amount is the pressure control amount of the current adsorption point under the contact force tracking error; calculating a second adsorption amount of the current adsorption point based on the contact force field curvature matrix, wherein the second adsorption amount is the compensation amount for the adsorption force under the influence of the contact force curvature; and obtaining the basic adsorption amount by adding the first adsorption amount and the second adsorption amount, wherein the calculation formula for the basic adsorption amount is: ； ； ； ； ； in, For the first Adsorption sites at The basic adsorption amount at any given time. For the first Adsorption sites at The initial adsorption amount at a given time. For the first Adsorption sites at The second adsorption amount at time [time]. For proportional control gain, For integral control gain, For differential control gain, In order to be in The first moment The expected contact force at each adsorption point For the first The actual contact force at each adsorption point For the first Contact force tracking error at each adsorption point The curvature-pressure conversion factor. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, This is the gap attenuation compensation coefficient. For the first At time 1 adsorption point The gap between the suction cups, The corresponding coordinates in the contact force matrix are Contact force.

5. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The step of calculating the notch confidence factor of the wafer based on the contact force field curvature matrix and the morphology state vector, and obtaining the wafer's decision matrix based on the notch confidence factor, includes: converting the morphology state vector into matrix form, obtaining the current wafer's gridded morphology deviation matrix, calculating the morphology spatial gradient based on the gridded morphology deviation matrix, wherein the morphology spatial gradient is the gradient value of the morphology deviation at different adsorption points, calculating the notch confidence factor of the current wafer under different coordinates based on the morphology spatial gradient and the contact force field curvature matrix, wherein the notch confidence factor is the confidence level that a notch exists at the adsorption point of the current wafer; and setting a notch confidence threshold. The notch confidence factor and the notch confidence threshold are compared. If the notch confidence factor at the current coordinate is greater than the notch confidence threshold, the current coordinate is selected. A set of confidence points for the current wafer is constructed based on the selected coordinates. This set of confidence points consists of points with high confidence factors at different coordinates. A decision matrix is ​​constructed based on the coordinates in the confidence point set. This decision matrix is ​​a binary matrix used to represent the notch problem on the surface of the current wafer. A minimum confidence point number threshold is preset. If the number of points in the current confidence point set is less than the minimum confidence point number threshold, the decision matrix for the current wafer is an all-zero matrix. The formula for calculating the notch confidence factor is: ； in, Coordinates are The gap confidence factor These are the force field characteristic weighting coefficients. The morphological feature weighting coefficient is used. The contact force curvature matrix at time t Middle and the first The coordinates of each adsorption point are: The force field curvature value, It is a positive part function. The first in the gridded topography deviation matrix Line number The value of the column, Coordinates are The morphological space gradient, For norm functions, The force field curvature threshold, This is the topography gradient threshold.

6. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The step of calculating the effective total contact force of the current wafer based on the decision matrix and determining the wafer adsorption behavior based on the effective total contact force includes: calculating the effective total contact force of the current wafer based on the number of elements, element values, and actual contact forces in the decision matrix, wherein the effective total contact force is the sum of the actual contact forces received by the current wafer at all adsorption points; calculating the safe gripping critical force of the current wafer, wherein the safe gripping critical force is the minimum total contact force that the wafer must provide in all non-avoidance states of adsorption points during dynamic movement; and comparing the effective total contact force and the safe gripping critical force for determination. The determination condition is: if the current effective total contact force is greater than the safe gripping critical force, it is determined that the current wafer is allowed to be gripped by the robotic arm; if the current effective total contact force is less than the safe gripping critical force, it is determined that the current wafer is not allowed to be gripped by the robotic arm and an alarm is triggered. The calculation formulas for the effective total contact force and the safe gripping critical force are as follows: ； ； in, This represents the current effective total contact force of the wafer. This represents the total number of adsorption sites. The adsorption point number is... The decision matrix is ​​the first... The decision value corresponding to each adsorption point For the first At time 1 adsorption point The actual contact force, To secure the critical force for current wafer safety, As for the current safety factor of the wafer, For the current quality of the wafer, This is the standard gravitational acceleration.

7. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The process of constructing a loss function based on the desired contact force, decision matrix, gradient square norm, and curvature square norm, and obtaining the corrected adsorption force increment based on the loss function, includes: pre-setting an adsorption increment vector; calculating the estimated corrected adsorption force of the current wafer at the current adsorption point based on the adsorption increment vector and the pressure sensitivity coefficient; the estimated corrected adsorption force is the contact force estimate of the current wafer after correction for each adsorption point in the actual environment; pre-setting a slow-scale time window; constructing loss functions for different slow-scale time windows based on the desired contact force, decision matrix, gradient square norm, and curvature square norm; and performing feedback iterative optimization using an optimization algorithm based on constraints to calculate the optimal adsorption increment vector of the current wafer; using the optimal adsorption increment vector of the current wafer as the corrected adsorption increment of the current wafer; the corrected adsorption increment is the difference in contact force of the current wafer after correction for each adsorption point in the actual environment. The calculation formula for the loss function is: ； ； ； ； ； in, For loss function, , and These are the first, second, and third constraints for the loss function. adsorption point Corresponding coordinates The corrected adsorption force estimate, adsorption point Corresponding coordinates The corrected adsorption force estimate, The corresponding coordinates in the contact force matrix are Contact force, This is the pressure contact force sensitivity coefficient. The adsorption increment vector is the first... The increase in adsorption force at each adsorption point , , , and These are the first weighting coefficient, the second weighting coefficient, the third weighting coefficient, the fourth weighting coefficient, and the fifth weighting coefficient. For the first The basic adsorption capacity of each adsorption point For the strength of the pressure valve, To secure the critical force for current wafer safety, This is the minimum gap between the suction cup and the bottom of the wafer. This is the maximum gap between the suction cup and the bottom of the wafer. It is the gradient square norm. Let be the square norm of curvature. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point This serves as a reference threshold for gradient bias. As a reference threshold for curvature deviation, This serves as a reference threshold for adsorption deviation. This is the reference threshold for adjacent deviations.

8. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 1, characterized in that, The process involves obtaining the final corrected adsorption amount based on the base adsorption amount and the incremental corrected adsorption force, simultaneously calculating the current wafer health, and intelligently adjusting the robotic arm based on the corrected adsorption amount and health. This includes: adding the base adsorption amount and the incremental corrected adsorption force at corresponding adsorption points to obtain the final corrected adsorption amount at the current adsorption point, where the corrected adsorption amount is the air pressure value of the robotic arm adsorbing the wafer after correction; intelligently adjusting the robotic arm based on the corrected adsorption amount, whereby the intelligent adjustment involves driving and adjusting the proportional valve of the robotic arm according to the corrected adsorption amount to change the vacuum degree within the suction cup at the adsorption point, thereby changing the contact force acting on the wafer; calculating the current wafer health based on the gradient square norm and the curvature square norm, where the health degree is the smoothness of the current wafer surface; and adjusting and optimizing the weight set of the loss function based on the health degree, wherein the formula for calculating the health degree is: ； in, For the current wafer in Health status at all times For the current wafer in The contact force gradient matrix at time t. For the current wafer in The curvature matrix of the contact force field at time t. The square of the norm, The decision matrix is ​​the first... The decision value corresponding to each adsorption point To correct the estimated adsorption force, For the current wafer in The first moment The expected contact force at each adsorption point This serves as a reference threshold for gradient bias. As a reference threshold for curvature deviation, This is the reference threshold for adsorption deviation.

9. The intelligent adjustment and control method for a semiconductor wafer gripping robot according to claim 8, characterized in that, The step of adjusting and optimizing the weight set of the loss function based on health status includes: presetting a fast-scale time window; calculating the health status of the current wafer based on the time detection window; optimizing the weight set of the loss function for the next fast-scale time window based on the health status of different fast-scale time windows, wherein the weight set consists of the first weight coefficient, second weight coefficient, third weight coefficient, fourth weight coefficient, and fifth weight coefficient in the loss function; and dynamically adjusting the loss function of the robot arm based on health status, wherein the calculation formula for dynamic adjustment is: ； ； ； in, For the first The fourth weighting coefficient for each fast-scale time window. For the first The fifth weighting coefficient for a fast-scale time window. This is the base maximum value of the fourth weighting coefficient. This is the base maximum value of the fifth weighting coefficient. For the first Health status within a fast-scale time window For the first The first weighting coefficient of a fast-scale time window.

10. An intelligent adjustment and control system for a semiconductor wafer gripping robot, used to implement the intelligent adjustment and control method for a semiconductor wafer gripping robot as described in any one of claims 1-9, characterized in that it further includes... include: The data matrix module acquires contact force data and morphology data of the wafer surface, constructs a contact force matrix and morphology state vector based on the contact force data and morphology data, and calculates and obtains the contact force gradient matrix and contact force field curvature matrix based on the contact force matrix. The basic adsorption module calculates the expected contact force of the current semiconductor wafer based on the morphology state vector; The basic adsorption amount of the current semiconductor wafer is calculated based on the actual contact force, the expected contact force, and the curvature matrix of the contact force field. The adsorption determination module calculates the notch confidence factor of the wafer based on the contact force field curvature matrix and morphology state vector, and obtains the decision matrix of the wafer based on the notch confidence factor. The effective total contact force of the current wafer is calculated based on the decision matrix, and the wafer adsorption behavior is determined based on the effective total contact force. The adsorption correction module constructs a loss function based on the desired contact force, decision matrix, gradient square norm, and curvature square norm, and obtains the correction adsorption force increment based on the loss function. The health adjustment module obtains the final corrected adsorption amount based on the basic adsorption amount and the incremental correction adsorption force, and calculates the current health of the wafer. It then intelligently adjusts the robotic arm based on the corrected adsorption amount and the health.

Images