Chip pin alignment calibration method based on machine vision

Abstract

The invention relates to the technical field of microelectronic assembly equipment control, and discloses a chip pin alignment calibration method based on machine vision, comprising the following steps: establishing a real-time motion error model representing mechanical positioning errors, and obtaining pin polar coordinates through partition extraction; a tangential margin is deducted from the theoretical region to construct a static tolerance window; performing morphological erosion and parallel simulation on the static window at a multi-stage speed by using an error boundary output by the model, and calculating a full-pin dynamic window geometric intersection; and according to the intersection non-empty state, the optimal speed and the target coordinate are decided, equipment mounting is driven, and thermal drift parameters are updated based on an observation residual closed loop. According to the method, the real-time motion error model containing thermal drift compensation is constructed, dynamic morphological erosion based on the mechanical positioning error boundary value is executed on the static window, the effectiveness of mounting coordinates in different thermal states is guaranteed, and dynamic decoupling of precision and speed is achieved.

Description

Technical Field

[0001] This invention relates to the field of microelectronic assembly equipment control technology, specifically a chip pin alignment and calibration method based on machine vision. Background Technology

[0002] Modern electronics manufacturing is pushing the limits of high density and miniaturization, with fine-pitch chips such as QFP and QFN playing a central role in automotive electronics and communication terminals. To ensure reliable circuit conduction, these components must achieve micron-level placement accuracy during surface mounting.

[0003] Existing visual alignment technologies primarily rely on best-fit algorithms and rigid body transformation logic. These technologies capture images of the chip's bottom pins using an industrial camera, extract the center coordinates of all pins, and match this as a whole geometry with standard pad data. By calculating the least-squares solution, the system can quickly derive a global rotation and translation matrix. This processing method is highly computationally efficient and effectively corrects initial angular deviations and positional offsets during component handling. It ensures that, under low-speed, stable standard production conditions, the chip's geometric center remains consistent with its theoretical mounting position, meeting the basic assembly requirements of most common components.

[0004] However, the processing logic based on the static rigid body assumption has gradually shown its limitations in high-speed production lines that strive for extreme cycle times. Traditional algorithms ignore the dynamic characteristics of physical motion systems, failing to detect the elastic jitter during high-speed sudden stops of the lead screw and the thermal drift after long-term operation. This forces equipment to sacrifice valuable production capacity by forcibly reducing speed to maintain accuracy and gain a safety margin. At the geometric level, existing global fitting strategies lack sensitivity to angular errors in large-size chips. Extremely small center angle deviations, amplified by long lever arms, can generate huge tangential displacements at corner pins, which conventional rectangular box detection often cannot identify. Furthermore, traditional algorithms lack flexibility in dealing with individual pin deformations. Local deviations of a single pin can often deviate the overall alignment coordinates or directly lead to the entire high-value chip being deemed scrap, lacking the ability to tolerate and degrade local defects. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a chip pin alignment and calibration method based on machine vision. This method solves the problem that existing technologies neglect mechanical dynamic errors and thermal drift effects under high-speed motion, and lack flexible fault-tolerant mechanisms for local pin deformation. As a result, it is difficult to balance high precision and high throughput in the mounting process of large-size fine-pitch chips, and the error rate is relatively high.

[0006] To achieve the above objectives, the present invention provides a chip pin alignment and calibration method based on machine vision, comprising the following steps:

[0007] Establish a real-time motion error model, which characterizes the boundary value of the mechanical positioning error of the device under a specific command speed and current thermal state;

[0008] Perform partition feature extraction and polar coordinate calculation to obtain the polar radius and polar angle of all pins of the chip relative to the chip's geometric center;

[0009] A static theoretical tolerance window is constructed by subtracting the tangential displacement safety margin determined by the polar radius and residual angular uncertainty from the theoretical rectangular region that the pin is allowed to fall into.

[0010] Parallel virtual simulation and dynamic erosion are performed at multiple speed levels. The mechanical positioning error boundary value output by the real-time motion error model is used to perform morphological erosion on the static theoretical tolerance window to generate a dynamic engineering window. The geometric intersection of the dynamic engineering windows corresponding to all pins at each speed level is calculated.

[0011] The system executes optimal placement strategy decisions and closed-loop control, selects the optimal command speed and optimal placement target coordinates based on the non-empty state of the geometric intersection, drives the equipment to perform placement actions, and updates the thermal drift parameters in the real-time motion error model based on the observed residuals after placement.

[0012] Preferably, establishing the real-time motion error model includes:

[0013] By calibrating the positioning error distribution of the equipment under different velocity gradient sequences, a basic motion error function is constructed to establish the mapping relationship between velocity variables and basic positioning error boundaries.

[0014] Acquire the physical positioning observation residuals of the equipment relative to the reference point during operation;

[0015] The thermal drift compensation term is iteratively updated based on the physical positioning observation residuals, and then superimposed onto the basic positioning error boundary calculated by the basic motion error function to generate a real-time motion error model.

[0016] Preferably, the step of performing partition feature extraction and polar coordinate calculation includes:

[0017] The image field of view is divided into multiple logical sectors, and the coordinate mapping matrix independently calibrated for each logical sector is applied to extract the absolute physical coordinates of each pin in the machine physical coordinate system.

[0018] Calculate the arithmetic mean of the absolute physical coordinates of all pins to determine the chip's geometric center;

[0019] Convert the absolute physical coordinates of each pin to relative coordinates with respect to the geometric center of the chip;

[0020] Calculate the Euclidean distance from the relative coordinates of each pin to the geometric center of the chip to determine the polar radius, and calculate the angle between the position vector of each pin and the coordinate axis to determine the polar angle.

[0021] Preferably, the construction of the static theoretical tolerance window includes:

[0022] Based on the physical dimensions of the substrate pads and pins, the theoretical maximum physical half-width that the pin center can deviate from the pad center under ideal rigid body alignment conditions is calculated, and the theoretical rectangular area is defined accordingly.

[0023] The tangential displacement safety margin is subtracted from the theoretical maximum physical half-width as a rotational error term to obtain the static effective half-width;

[0024] A rectangular region with the center of the pad as the reference is constructed based on the static effective half-width, which serves as the static theoretical tolerance window.

[0025] Preferably, the parallel virtual simulation and dynamic erosion at multiple speed levels includes:

[0026] Pre-set a discrete set of speed levels;

[0027] For each speed value in the speed level set, the real-time motion error model is called to calculate the corresponding mechanical positioning error boundary value;

[0028] Using the mechanical positioning error boundary value as the structural element size, perform morphological erosion operation on the static theoretical tolerance window of each pin to generate a dynamic engineering window at that velocity value.

[0029] Find the common overlapping region of the dynamic engineering windows of all pins at this speed value, as the geometric intersection.

[0030] Preferably, the execution of the optimal placement strategy decision and closed-loop control includes:

[0031] Iterate through the geometric intersections of all speed levels and filter out all speed candidates that satisfy the non-empty condition;

[0032] Select the speed with the largest value from the speed candidates as the optimal command speed;

[0033] Extract the geometric centroid of the geometric intersection corresponding to the optimal command speed, and use the geometric centroid as the optimal mounting target coordinates.

[0034] Preferably, the optimal placement strategy decision and closed-loop control further include:

[0035] When the geometric intersection at all speed levels is an empty set, the maximum subset filtering process is initiated.

[0036] Based on the preset minimum speed corresponding to the dynamic engineering window, iteratively remove outlier pins that cause the intersection to be empty, until the intersection of the remaining pin subset is not empty;

[0037] If the number of rejected pins is less than the preset process tolerance threshold, the chip is deemed feasible for downgrading, and mounting is performed using the lowest speed and the centroid of the intersection of the remaining pin subset.

[0038] If the number of rejected pins is greater than or equal to the process tolerance threshold, the chip is determined to be scrap and a discard command is generated.

[0039] Preferably, the iterative elimination of outlier pins that result in an empty intersection includes:

[0040] Calculate the geometric center point of the dynamic engineering window corresponding to all currently participating pins;

[0041] Calculate the mean coordinates of all geometric center points as the cluster center;

[0042] Iterate through and calculate the Euclidean distance from the geometric center point of the dynamic engineering window of each pin to the cluster center.

[0043] The pin with the largest Euclidean distance is identified as an outlier and removed from the set.

[0044] Preferably, the iterative update of the thermal drift compensation term based on the physical positioning observation residual includes:

[0045] The initial value of the preset thermal drift compensation term is zero;

[0046] Using the exponentially weighted moving average algorithm, the thermal drift compensation term at the current moment is calculated by weighting the physical positioning observation residual obtained at the current moment with the thermal drift compensation term at the previous moment.

[0047] The smoothing factor used in the weighted calculation has a value range of greater than zero and less than one.

[0048] Preferably, the calculation method for the tangential displacement safety margin includes:

[0049] The product of the polar radius, the residual angular uncertainty, and the absolute value of the sine of the polar angle is determined as the tangential displacement component in the direction of the first coordinate axis.

[0050] The product of the polar radius, the residual angular uncertainty, and the absolute value of the cosine of the polar angle is determined as the tangential displacement component in the direction of the second coordinate axis, which is perpendicular to the first coordinate axis.

[0051] This invention provides a machine vision-based chip pin alignment and calibration method. It offers the following advantages:

[0052] 1. This invention constructs a real-time motion error model that includes thermal drift compensation and performs dynamic morphological erosion on the static window based on the boundary value of mechanical positioning error. This ensures the effectiveness of the mounting coordinates under different thermal states and achieves dynamic decoupling of accuracy and speed. Compared with the shortcomings of traditional rigid body transformation methods that ignore thermal expansion and high-frequency vibration interference, this invention effectively avoids positioning deviations caused by equipment temperature rise or motion jitter and improves system robustness.

[0053] 2. This invention utilizes multi-level parallel virtual simulation and dynamic window intersection operation. The system can autonomously search for the highest instruction speed boundary that satisfies geometric constraints, maximizing the equipment production cycle while ensuring placement yield. Unlike the control strategy in the prior art that forces the adoption of conservative low-speed parameters to retain safety margin, this solution eliminates the production capacity loss caused by excessive speed reduction, and also eliminates the risk of scrap caused by blindly increasing speed.

[0054] 3. This invention addresses the abnormal operating condition where all pin intersections are empty by introducing a maximum subset screening and outlier pin iterative elimination mechanism. By quantifying the process tolerance threshold, the system can salvage chips with slight local deformations and achieve degraded usability. Compared with the rigid mode of conventional template matching algorithms that determine the entire chip as scrapped due to a single pin deviation, this method significantly reduces the false scrap rate of high-value chips and greatly saves production material costs. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram illustrating the construction logic principle of the static theoretical tolerance window in an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram illustrating the multi-level speed parallel virtual simulation and dynamic erosion logic principle of an embodiment of the present invention;

[0059] Figure 5 This is a schematic diagram of the optimal mounting strategy decision and closed-loop control process in an embodiment of the present invention.

[0060] in:

[0061] 10. Motion error modeling module; 20. Visual feature extraction module; 30. Tolerance window construction module; 40. Virtual simulation decision-making module; 50. Motion execution control module. Detailed Implementation

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] See attached document Figure 1 , Figure 1 This is a structural block diagram of a machine vision-based chip pin alignment calibration system according to an embodiment of the present invention. The present invention provides a machine vision-based chip pin alignment calibration system, which operates in the main control unit of a packaging or mounting device. The machine vision-based chip pin alignment calibration system includes: a motion error modeling module 10, a visual feature extraction module 20, a tolerance window construction module 30, a virtual simulation decision module 40, and a motion execution control module 50.

[0064] The motion error modeling module 10 is used to establish and update the motion uncertainty field model of the device. The motion error modeling module 10 is connected to the device's sensor interface to acquire positioning feedback data of the mounting head under different motion speeds and accelerations. The motion error modeling module 10 constructs a motion error function that includes velocity variables and time thermal drift variables, and outputs the mechanical positioning error boundary value at a specific speed at the current moment.

[0065] The visual feature extraction module 20 is used to acquire the geometric position information of the chip pins. The visual feature extraction module 20 receives image data from an industrial camera and divides the image field of view into multiple independent logical sectors. The visual feature extraction module 20 performs edge recognition on the pins within each logical sector, extracts the pin center coordinates, and uses the calibration matrix corresponding to each sector to eliminate lens distortion. The visual feature extraction module 20 is also used to calculate the polar radius and polar angle data of each pin relative to the geometric center of the chip.

[0066] The tolerance window construction module 30 is used to generate the static geometric constraint range of each pin. Based on the physical dimension data of the substrate pads and the physical dimension data of the pins, the tolerance window construction module 30 calculates the theoretical rectangular area that the pins are allowed to fall into. Combining the pin's extreme diameter data and the system's angle measurement accuracy, the tolerance window construction module 30 calculates the tangential displacement of the pins due to rotational errors, and subtracts this tangential displacement as a safety margin from the theoretical rectangular area to generate a static theoretical tolerance window.

[0067] The virtual simulation decision module 40 is used to calculate the optimal mounting parameters. The virtual simulation decision module 40 has a preset set of discrete speed levels. For each speed value in the speed level set, the virtual simulation decision module 40 calls the error boundary value output by the motion error modeling module 10, performs morphological erosion operations on the static theoretical tolerance window generated by the tolerance window construction module 30, and generates a dynamic engineering window. The virtual simulation decision module 40 calculates the geometric intersection of the dynamic engineering windows corresponding to all pins, and selects the maximum feasible speed based on the non-empty state of the geometric intersection.

[0068] The motion execution control module 50 is used to drive the device to perform placement actions. The motion execution control module 50 receives the optimal speed command and geometric intersection centroid coordinates output by the virtual simulation decision module 40. The motion execution control module 50 converts the command into servo drive signals to control the placement head to move to the target position. The motion execution control module 50 is also used to monitor the positioning deviation of the reference point during the placement cycle and feeds the deviation data back to the motion error modeling module 10 to update the thermal drift parameters.

[0069] The machine vision-based chip pin alignment calibration system involves three core coordinate systems during operation. The machine physical coordinate system is defined as the device's global absolute reference system, based on the readings of the grating ruler or encoder provided with the device, and is measured in millimeters. The image pixel coordinate system is defined as the imaging plane coordinate system of the vision acquisition hardware, based on the number of rows and columns of the image sensor array, and is measured in pixels. The chip local coordinate system is defined as a relative coordinate system with the geometric center of the acquired chip body as its origin, used to describe the geometric distribution relationship of the pins relative to the chip center. The visual feature extraction module 20 performs a spatial mapping transformation from the image pixel coordinate system to the machine physical coordinate system.

[0070] See attached document Figure 2 , Figure 2 This is a schematic flowchart of a chip pin alignment and calibration method based on machine vision according to an embodiment of the present invention. The present invention provides a chip pin alignment and calibration method based on machine vision, comprising the following steps:

[0071] S100: Establish a motion error model that includes thermal drift compensation. Construct a basic error function by calibrating the positioning error distribution of the equipment at different speeds. Update the thermal drift compensation term based on the observation residual of the base plate reference point to generate a real-time motion error model.

[0072] S200 performs partition feature extraction and polar radius calculation, divides the chip image field of view into multiple logical sectors, extracts the physical coordinates of the pins in each sector, and calculates the polar radius and polar angle of each pin relative to the geometric center of the chip.

[0073] S300 constructs a static theoretical tolerance window with rotational pre-compensation, defines a theoretical rectangular area based on pad size and pin size, calculates tangential displacement based on pin radius and angle measurement uncertainty, and deducts the tangential displacement as a safety margin from the theoretical rectangular area;

[0074] S400 performs parallel virtual simulation and dynamic erosion at multiple speed levels. For multiple preset speed levels, it uses a real-time motion error model to perform morphological erosion on the static theoretical tolerance window to generate a dynamic engineering window, and calculates the geometric intersection of the dynamic engineering windows of all pins at each speed.

[0075] The S500 executes optimal placement strategy decisions and closed-loop control, selects the maximum speed of the non-empty geometric intersection as the optimal speed, extracts the centroid of the geometric intersection as the placement target coordinates, drives the equipment to perform placement, and updates the thermal drift compensation term based on the execution results.

[0076] Step S100, establishing a motion error model including thermal drift compensation, specifically includes the following sub-steps:

[0077] S101, Construct the basic motion error function. Since the servo drive system of the placement equipment is not an ideal rigid body, the positioning repeatability accuracy of its end effector is... With motion command speed There is a nonlinear positive correlation coupling relationship between them. During the system initialization calibration phase, a laser interferometer or high-precision grating ruler is used as an external reference to perform multiple reciprocating positioning tests on the mounting head under a preset speed gradient sequence. Positioning error data at different speeds are collected, assuming the positioning error is within... shaft and The axes are independent and follow a normal distribution. (Select...) The confidence interval serves as the maximum error boundary at this speed. A basic error function is constructed by performing linear regression fitting on the discrete test data:

[0078] ;

[0079] ;

[0080] in, This indicates the speed of the commands issued by the motion control card; and , which is a dynamic coefficient, characterizing the sensitivity of the servo system's stiffness and damping characteristics to speed changes; and The inherent static positioning noise of the system is characterized by the mechanical assembly clearance and sensor noise floor in a quasi-stationary state. For higher-order nonlinear error characteristics, those skilled in the art can also construct the basic error function using polynomial fitting. Any mathematical model that establishes the mapping relationship between velocity variables and positioning error boundaries falls within the scope of this embodiment.

[0081] S102, Obtain the thermal drift observation residual. During continuous operation of the placement equipment, the linear motor coil heats up and the guide rail friction generates heat, causing slight thermal elongation or deformation of the mechanical structure. This deformation manifests as a systematic drift that accumulates over time or batches. Using the equipment's existing vision acquisition hardware, the fiducial marks on the substrate are observed during each placement cycle or at a preset time interval. The difference between the encoder coordinates fed back by the motion control system and the center coordinates of the fiducial mark image recognized by the vision system is recorded; this difference is the physical positioning observation residual at the current moment. .

[0082] S103, Update thermal drift compensation terms. The system presets the initial value of the thermal drift compensation terms. as well as Since single visual observations are susceptible to fluctuations in illumination or image noise, directly using the single-observation residuals to update the model can lead to system oscillations. Therefore, an exponentially weighted moving average (EWMA) algorithm is used to filter the observation residuals and iteratively update the thermal drift compensation term. Let the first... The observations obtained Axial residual is Then the updated Axial thermal drift compensation term The calculation is as follows:

[0083] ;

[0084] Similarly, we can obtain The update formula for the axial direction. Wherein, This is a smoothing factor, with a value range of [value range missing]. . The choice of value depends on the trade-off between system response speed and stability. The larger the value, the faster the model responds to newly generated thermal drift, but the weaker its ability to suppress noise.

[0085] S104, Generate a real-time motion error model. The velocity-related basic error boundary determined in step S101 is superimposed with the time-related thermal drift compensation term determined in step S103 to generate a real-time motion error model for subsequent dynamic erosion. :

[0086] ;

[0087] ;

[0088] The model outputs and Defined in the current thermal state and specific command speed Below, the maximum positional deviation boundary value that the end effector may produce is determined. This boundary value will subsequently be used as the structuring element size for morphological erosion operations to geometrically constrain the tolerance window.

[0089] Step S200, which involves partition feature extraction and polar coordinate calculation, specifically includes the following sub-steps:

[0090] S201, Constructing the Field of View Partition Mapping Relationship. Considering the optical distortion at the edges of large field-of-view industrial lenses and the potential warping deformation of chip packaging substrates, a single global calibration matrix is ​​insufficient to maintain micron-level linearity across the entire field of view. The imaging field of view of the industrial camera is divided into... The field of view is divided into logical sectors, or according to the physical grouping of chip pins (such as the four sides of a Quad Flat Package (QFP), the field of view is divided into corresponding independent regions. For each logical sector, its corresponding coordinate mapping matrix is ​​pre-calibrated independently using a high-precision calibration board. The mapping matrix It characterizes the affine transformation relationship or perspective transformation relationship between pixels in a specific local region within the image pixel coordinate system and their corresponding points in the machine physical coordinate system, and is used to locally eliminate the effects of nonlinear distortion.

[0091] S202, extract pin pixel features. Based on the chip's CAD design data or a pre-taught template, assign pixel features to each pin in the image coordinate system. Define the region of interest (ROI). Within each ROI, extract the pin's edge contour using edge detection operators (such as the Canny or Sobel operators), and calculate the pin's original center coordinates in the image pixel coordinate system by fitting a rectangle or line using the least squares method, or by employing Blob connectivity analysis. The specific edge detection and geometric fitting algorithms are well-known techniques in the field of computer vision and will not be elaborated upon here.

[0092] S203 performs physical coordinate mapping and centralization processing. (Based on pin...) The logical sector index is used to call the corresponding calibration mapping matrix. , pixel coordinates Convert to absolute coordinates in the machine's physical coordinate system. After obtaining the absolute physical coordinates of all valid pins, calculate the current geometric center coordinates of the chip. The geometric center coordinates can be determined by the arithmetic mean of the absolute coordinates of all pins, or by the coordinates of the chip corner pins. Subsequently, the absolute coordinates of all pins are converted to relative coordinates with respect to the chip's geometric center. :

[0093] ;

[0094] ;

[0095] in, For the first The absolute coordinates of each pin in the machine's physical coordinate system. This step establishes a local coordinate system for the chip with its current physical center as the origin, eliminating the interference of rigid body translation deviations generated during the chip acquisition process on subsequent rotational component calculations.

[0096] S204, Calculate the polar coordinate parameters of the pins. To decouple the rotation angle error into linear tangential displacement in subsequent steps, the Cartesian coordinates of each pin need to be converted to polar coordinates. For each pin... Calculate its polar radius relative to the chip's geometric center. With polar angle :

[0097] ;

[0098] ;

[0099] in, This parameter represents the Euclidean distance from the center of the pin to the geometric center of the chip. This parameter directly determines the sensitivity of the pin to rotational errors. That is, the farther the pin is from the center, the greater the arc length displacement produced by the same amount of rotation. Indicates pin position vector and The angle in the positive direction of the axis is used to determine the tangential displacement. shaft and The directions of the projection components on the axis. Through the above transformation, the mapping from image feature space to physical parameter space is completed, providing a data foundation for the subsequent construction of static theoretical tolerance windows based on geometric constraints.

[0100] See attached document Figure 3 , Figure 3 This is a schematic diagram illustrating the logic principle of constructing a static theoretical tolerance window according to an embodiment of the present invention. Step S300, constructing the rotated pre-compensated static theoretical tolerance window, specifically includes the following sub-steps:

[0101] S301, determine the basic physical geometric tolerances. Based on the PCB substrate manufacturing process specifications and chip packaging parameters, obtain the physical dimensions of the target pads and the physical dimensions of the chip pins. Let the first... The rectangular pads corresponding to each pin are edge The length in the axial direction is ,along The width in the axial direction is The pin itself along The length in the axial direction is ,along The width in the axial direction is Under the ideal rigid body alignment condition assuming no rotational angular deviation between the chip and the substrate, the theoretical maximum physical half-width that the pin center can deviate from the pad center is calculated. This theoretical maximum physical half-width represents the geometric limit of the pin not exceeding the pad boundary when only translational error exists:

[0102] ;

[0103] ;

[0104] in, and The first Each pin in shaft and The basic physical tolerance half-width in the axial direction. For circular or racetrack-shaped pads, they are calculated as equivalent to the circumscribed rectangle or effective contact rectangle area.

[0105] S302, calculates the linearized tangential component of the rotation error. This takes into account the random angular deviation of the mounting head when picking up the chip, and the movement during the process... The servo jitter of the axis results in a residual angular uncertainty in the system. To avoid complex full-degree-of-freedom coordinate transformations and trigonometric function calculations in subsequent high-speed simulations, this step employs a tangential error pre-subtraction strategy, transforming angular uncertainty into linear dimensional constraints along the coordinate axes. The pin radius is calculated based on step S204. and polar angle The pin rotates The resulting displacement is approximately equal to the arc length along the tangent direction of the circle, and this arc length is... shaft and The projected component on the axis is the rotational error safety margin:

[0106] ;

[0107] ;

[0108] in, and They represent the first The pins are affected by the maximum residual angle error. shaft and Potential positional offset in the axial direction. As can be seen from the formula, the polar radius... The larger the value, meaning the farther the pin is from the geometric center of the chip, the greater its corresponding... The larger the value, the greater the safety margin that needs to be deducted. This approach decouples rotational and translational errors, allowing the alignment algorithm to apply differentiated constraint standards to pins at different positions.

[0109] S303, Generate the static theoretical tolerance window. Subtract the rotational error safety margin calculated in step S302 from the basic physical tolerance half-width determined in step S301 to obtain the... Static effective half-width of each pin and :

[0110] ;

[0111] ;

[0112] Based on the static effective half-width, construct a pad-centered structure. Static theoretical tolerance window based on :

[0113] ;

[0114] This window A rectangular region is defined to represent the maximum angular error even when the device is stationary. The pins can still fall entirely within the feasible translation space of the pads. If, during the calculation process... or This indicates that the pin cannot be reliably aligned under the current pad design size and angle accuracy conditions. The system marks the pin as a "geometrically infeasible feature" and removes it from subsequent decision logic or uses it as an alarm basis. Through this step, the multi-dimensional geometric constraint problem is reduced to a two-dimensional planar rectangular region definition problem, providing a standardized geometric input for subsequent introduction of motion error fields for morphological erosion.

[0115] See attached document Figure 4 , Figure 4 This is a schematic diagram of the logic principle of multi-level speed parallel virtual simulation and dynamic erosion according to an embodiment of the present invention. Step S400, which executes parallel virtual simulation and dynamic erosion at multiple speeds, specifically includes the following sub-steps:

[0116] S401 defines a set of discrete velocity levels. A set of discrete velocity levels for simulation calculations is preset in the system configuration parameters. This set covers the entire operating range of the equipment, from low-speed, high-precision mode to high-speed, high-capacity mode, and meets the requirements. The density of speed level divisions depends on the sensitivity requirements to production cycle time (UPH), and typically employs equal-interval sampling or logarithmic-interval sampling. This step, through software-level presets, provides input variables for subsequent parallel computations without requiring the hardware to actually execute the motion.

[0117] S402 calculates the error boundaries at each speed level. (For the set...) Each velocity value in Call the real-time motion error model generated in step S104 Speed and current thermal drift status Substitute into the model and calculate the end effector at this speed. shaft and Positioning error boundary value in the axial direction and :

[0118] ;

[0119] ;

[0120] The error boundary value calculated here characterizes the error when the device is at a speed of During movement, the maximum potential deviation range of the actual landing point from the commanded target point.

[0121] S403 performs morphological erosion operations within a tolerance window. This is to ensure that motion errors are present. and In this case, the pin still does not exceed the static theoretical tolerance window determined in step S303. This requires geometric shrinkage of the window. A morphological erosion algorithm is used, employing the motion error boundary as a structuring element, to perform an erosion operation on the static theoretical tolerance window of each pin, generating a dynamic engineering window at that speed. Mathematically, this is represented by subtracting the corresponding error boundary value from the effective half-width of the window:

[0122] ;

[0123] Through erosion operations, the physical uncertainties of motion are transformed into geometric spatial constraints. (Velocity) The higher the value, the higher the corresponding error boundary. The larger the window, the more dynamic the engineering window after erosion. Smaller area means more stringent requirements for chip alignment accuracy. If the half-width value in any dimension of the calculation result... If the speed exceeds the remaining tolerance of the pin, it indicates that the window has failed to close at that speed level.

[0124] S404 solves for the global intersection of a full-pin dynamic window. For a specific speed... To achieve overall chip alignment, a common mounting coordinate system must be found so that all the chips on the chip can be aligned. Each valid pin simultaneously falls within its respective dynamic engineering window. Calculate the geometric intersection of the dynamic engineering windows corresponding to all pins. :

[0125] ;

[0126] Since the dynamic engineering windows for each pin are axis-aligned rectangles, their intersection... It can also be a rectangle (or an empty set). The boundary coordinates of the intersection region are determined by the extreme values ​​of the boundaries of all windows involved in the calculation. The intersection region is calculated in... Starting coordinates in the axial direction and termination coordinates ,as well as Starting coordinates in the axial direction and termination coordinates :

[0127] ;

[0128] ;

[0129] Similarly, calculate Boundary coordinates in the axial direction. In the above formula... The operation finds the rightmost value of the left boundary of all windows. The operation finds the leftmost value of the right boundary of all windows, thereby determining the extent of the common overlapping area.

[0130] S405, Determine the feasibility of the speed scheme. Based on the intersection boundary coordinates calculated in step S404, determine the speed... Does a feasible mounting solution exist? The criterion is the non-empty condition of the geometric intersection:

[0131] ;

[0132] If all of the above conditions are met, then the geometric intersection is determined. That is, considering speed Even after accounting for the motion error, there is still a physical space that prevents all pins from shifting. Speeds are marked as feasible. Conversely, if the starting coordinate in any direction is greater than the ending coordinate, it indicates that at least two pin tolerance requirements conflict and cannot be simultaneously met under the current speed vibration; this speed is marked as infeasible. The system sets... All speeds execute the above calculation process S402 to S405 in parallel to generate a speed-feasibility mapping table.

[0133] See attached document Figure 5 , Figure 5 This is a schematic diagram of the optimal placement strategy decision and closed-loop control process according to an embodiment of the present invention. Step S500, which executes the optimal placement strategy decision and closed-loop control, specifically includes the following sub-steps:

[0134] S501, Select the maximum feasible speed. Iterate through the geometric intersection calculation results generated in step S400 for all preset speed levels. In the speed set... In the process, select all pairs of pairs that satisfy the non-empty condition of geometric intersection (i.e., The speed candidates are selected. The speed with the highest numerical value is chosen as the optimal command speed for this placement operation. :

[0135] ;

[0136] This selection logic establishes the decision-making principle of "efficiency first, quality constraint," which means that under the premise of ensuring that all pins fall within the pad tolerance range, the highest physical speed of the equipment should be used as much as possible, thereby maximizing the production cycle time (UPH) while ensuring yield. If there is no speed in the set that meets the non-empty condition, the current chip is determined to be in a normal alignment failure state, and the system switches to the subsequent exception handling logic.

[0137] S502, calculate the optimal target coordinates. Based on the selected optimal velocity. Extract its corresponding global geometric intersection region. The boundary coordinates are determined. Since this intersection region represents the safe landing point range for all chip pins after considering the maximum potential motion error, any point within this region is theoretically a feasible solution. To maximize the robustness of the mounting process, the geometric centroid of this rectangular intersection region is selected as the optimal mounting target coordinates. :

[0138] ;

[0139] ;

[0140] Choosing the centroid as the target point ensures that the chip center has the maximum safety margin from each boundary of the intersection region, thereby maximizing the resistance to potential overshooting during movement. Occasional random perturbations at the boundary.

[0141] S503, execute maximum subset filtering anomaly handling. When the geometric intersection of all speed levels in step S401 is an empty set, it indicates that the chip has severe pin deformation and full pin alignment cannot be achieved. At this time, the maximum subset filtering algorithm is activated to determine whether the chip can still meet process requirements by sacrificing the alignment of a few non-critical pins or pins with extremely minor deformation. The algorithm operates at the lowest speed allowed by the system. Using the corresponding dynamic engineering window as a baseline, outlier pins that cause the intersection to be empty are iteratively removed. Specifically, the geometric center point of the dynamic engineering window corresponding to all participating pins is calculated, and the mean coordinates of all geometric center points are calculated as the cluster center. The Euclidean distance from the center of each pin window to the cluster center is calculated, and the pin with the largest distance is identified as an outlier and removed from the set. The intersection of the remaining pin subset is recalculated until the intersection is not empty. The final retained pin subset is recorded. And the number of pins it contains. If the number of pins is removed... Less than the preset process tolerance threshold If the chip is deemed "feasible for downgrading", then the lowest speed will be used. And the centroid of the intersection of subsets are used for mounting; if If the chip is deemed waste, the system generates a discard command, and the drive device places the chip in the waste collection area.

[0142] S504, execute physical motion and closed-loop feedback. The motion execution control module 50 receives the optimal target coordinates determined by step S502 or S503. and optimal speed command (or This is converted into pulse control signals for the servo motor, driving the placement head to perform physical pick-and-place actions. After the placement action is completed, the reference points on the substrate are observed again using the equipment's vision system, or the actual chip placement deviation is obtained using an auto-inspection (AOI) device. This actual deviation is used as a new observation residual. Feedback is sent to step S103 to trigger the thermal drift compensation item. The updates complete the closed-loop control from measurement, decision-making, execution to correction.

[0143] To further clarify the collaborative working process of the technical solution described in this invention, a specific working scenario example will be used for illustration below.

[0144] In this embodiment, a Quad Flat Package (QFP-100) chip is selected as the object to be mounted. This chip has 100 pins with a pin pitch of 0.4 mm, corresponding to a PCB pad width of 0.25 mm. The mounting equipment is configured with three preset speed levels: high-speed mode (1000 mm / s), medium-speed mode (500 mm / s), and low-speed mode (100 mm / s).

[0145] Before the start of the production batch, the motion error modeling module 10 has completed the system initialization calibration. According to the calibration data, the positioning error boundary of the equipment in high-speed mode is ±0.035mm, the positioning error boundary in medium-speed mode is ±0.015mm, and the positioning error boundary in low-speed mode is ±0.005mm.

[0146] When the placement head picks up a QFP-100 chip and moves it to the visual observation position, the visual feature extraction module 20 triggers image acquisition. Through partition mapping and feature extraction, the system identifies a slight physical deformation in pin 35 of the chip, which is offset outward by 0.02 mm relative to its ideal position. Simultaneously, the system calculates the polar radius and polar angle of this pin relative to the chip center, and, combined with the system's residual angular uncertainty (set at 0.1 degrees), calculates the potential displacement component of the pin in the tangential direction.

[0147] The tolerance window construction module 30, based on the pad size and pin physical dimensions, first generates a theoretically permissible rectangular area for each pin. For the deformed pin 35, due to the initial offset of its physical position, its corresponding theoretical rectangular area is translated relative to the chip center. Subsequently, the system subtracts the tangential displacement safety margin caused by angular uncertainty from this theoretical rectangular area to generate a static theoretical tolerance window. At this point, due to the superposition of deformation and rotation error of pin 35, the effective range of its static theoretical tolerance window is significantly smaller than that of a normal pin.

[0148] The virtual simulation decision module 40 then initiates parallel simulation calculations. In the simulation for "high-speed mode (1000mm / s)," the system introduces a ±0.035mm motion error boundary to morphologically erode the static windows of all pins. The calculation results show that for the deformed pin 35, after the ±0.035mm erosion, the remaining effective area of ​​its static window completely disappears (window closes), or the remaining area has no common intersection with the windows of other normal pins. Based on this, the system determines that if the mounting speed is 1000mm / s, it is highly likely that pin 35 will be offset from the pad, therefore the high-speed mode is marked as infeasible.

[0149] Meanwhile, in the simulation for the "medium speed mode (500mm / s)," the system introduced a small ±0.015mm motion error boundary. At this point, the static window of pin 35, after erosion, still retained a small effective area, and this area shared a rectangular intersection with the dynamic engineering windows of the other 99 pins. Although this intersection was small, it physically existed. Based on this, the system determined that if the speed was reduced to 500mm / s, the device vibration would decrease, ensuring that all pins, including deformed pins, would fall onto the pads; therefore, the medium speed mode was marked as feasible.

[0150] Based on the simulation results above, the optimal placement strategy decision automatically discards the high-speed mode, which has the highest efficiency but also the greatest risk, and selects the "medium-speed mode" as the execution parameter. The system extracts the centroid coordinates of the geometric intersection region calculated under the medium-speed mode. Use it as the target mounting position.

[0151] The motion execution control module 50 receives instructions and drives the placement head to precisely place the chip onto the PCB board at a speed of 500 mm / s. After the placement is completed, the equipment detects the reference point position of this motion cycle and finds a thermal drift increment of 0.002 mm. This data is fed back to the motion error modeling module 10, which updates the error model parameters for the next cycle using the EWMA algorithm, ensuring that subsequent calculations can dynamically adapt to changes in the equipment's thermal state. Through this process, the present invention achieves adaptive speed degradation and precise alignment for imperfect materials without sacrificing yield.

[0152] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A chip pin alignment and calibration method based on machine vision, characterized in that, Includes the following steps: Establish a real-time motion error model, which characterizes the boundary value of the mechanical positioning error of the device under a specific command speed and current thermal state; Perform partition feature extraction and polar coordinate calculation to obtain the polar radius and polar angle of all pins of the chip relative to the chip's geometric center; A static theoretical tolerance window is constructed by subtracting the tangential displacement safety margin determined by the polar radius and residual angular uncertainty from the theoretical rectangular region that the pin is allowed to fall into. Parallel virtual simulation and dynamic erosion are performed at multiple speed levels. The mechanical positioning error boundary value output by the real-time motion error model is used to perform morphological erosion on the static theoretical tolerance window to generate a dynamic engineering window. The geometric intersection of the dynamic engineering windows corresponding to all pins at each speed level is calculated. The system executes optimal placement strategy decisions and closed-loop control, selects the optimal command speed and optimal placement target coordinates based on the non-empty state of the geometric intersection, drives the equipment to perform placement actions, and updates the thermal drift parameters in the real-time motion error model based on the observed residuals after placement.

2. The chip pin alignment and calibration method based on machine vision according to claim 1, characterized in that, The establishment of the real-time motion error model includes: By calibrating the positioning error distribution of the equipment under different velocity gradient sequences, a basic motion error function is constructed to establish the mapping relationship between velocity variables and basic positioning error boundaries. Acquire the physical positioning observation residuals of the equipment relative to the reference point during operation; The thermal drift compensation term is iteratively updated based on the physical positioning observation residuals, and then superimposed onto the basic positioning error boundary calculated by the basic motion error function to generate a real-time motion error model.

3. The chip pin alignment and calibration method based on machine vision according to claim 1, characterized in that, The execution of partition feature extraction and polar coordinate calculation includes: The image field of view is divided into multiple logical sectors, and the coordinate mapping matrix independently calibrated for each logical sector is applied to extract the absolute physical coordinates of each pin in the machine physical coordinate system. Calculate the arithmetic mean of the absolute physical coordinates of all pins to determine the chip's geometric center; Convert the absolute physical coordinates of each pin to relative coordinates with respect to the geometric center of the chip; Calculate the Euclidean distance from the relative coordinates of each pin to the geometric center of the chip to determine the polar radius, and calculate the angle between the position vector of each pin and the coordinate axis to determine the polar angle.

4. The chip pin alignment and calibration method based on machine vision according to claim 1, characterized in that, The construction of the static theoretical tolerance window includes: Based on the physical dimensions of the substrate pads and pins, the theoretical maximum physical half-width that the pin center can deviate from the pad center under ideal rigid body alignment conditions is calculated, and the theoretical rectangular area is defined accordingly. The tangential displacement safety margin is subtracted from the theoretical maximum physical half-width as a rotational error term to obtain the static effective half-width; A rectangular region with the center of the pad as the reference is constructed based on the static effective half-width, which serves as the static theoretical tolerance window.

5. The chip pin alignment and calibration method based on machine vision according to claim 1, characterized in that, The parallel virtual simulation and dynamic erosion at multiple speed levels includes: Pre-set a discrete set of speed levels; For each speed value in the speed level set, the real-time motion error model is called to calculate the corresponding mechanical positioning error boundary value; Using the mechanical positioning error boundary value as the structural element size, perform morphological erosion operation on the static theoretical tolerance window of each pin to generate a dynamic engineering window at that velocity value. Find the common overlapping region of the dynamic engineering windows of all pins at this speed value, as the geometric intersection.

6. The chip pin alignment and calibration method based on machine vision according to claim 1, characterized in that, The optimal placement strategy decision-making and closed-loop control include: Iterate through the geometric intersections of all speed levels and filter out all speed candidates that satisfy the non-empty condition; Select the speed with the largest value from the speed candidates as the optimal command speed; Extract the geometric centroid of the geometric intersection corresponding to the optimal command speed, and use the geometric centroid as the optimal mounting target coordinates.

7. The chip pin alignment and calibration method based on machine vision according to claim 6, characterized in that, The optimal placement strategy decision-making and closed-loop control also include: When the geometric intersection at all speed levels is an empty set, the maximum subset filtering process is initiated. Based on the preset minimum speed corresponding to the dynamic engineering window, iteratively remove outlier pins that cause the intersection to be empty, until the intersection of the remaining pin subset is not empty; If the number of rejected pins is less than the preset process tolerance threshold, the chip is deemed feasible for downgrading, and mounting is performed using the lowest speed and the centroid of the intersection of the remaining pin subset. If the number of rejected pins is greater than or equal to the process tolerance threshold, the chip is determined to be scrap and a discard command is generated.

8. The chip pin alignment and calibration method based on machine vision according to claim 7, characterized in that, The iterative elimination of outlier pins that result in an empty intersection includes: Calculate the geometric center point of the dynamic engineering window corresponding to all currently participating pins; Calculate the mean coordinates of all geometric center points as the cluster center; Iterate through and calculate the Euclidean distance from the geometric center point of the dynamic engineering window of each pin to the cluster center. The pin with the largest Euclidean distance is identified as an outlier and removed from the set.

9. The chip pin alignment and calibration method based on machine vision according to claim 2, characterized in that, The iterative update of the thermal drift compensation term based on the physical positioning observation residual includes: The initial value of the preset thermal drift compensation term is zero; Using the exponentially weighted moving average algorithm, the thermal drift compensation term at the current moment is calculated by weighting the physical positioning observation residual obtained at the current moment with the thermal drift compensation term at the previous moment. The smoothing factor used in the weighted calculation has a value range of greater than zero and less than one.

10. The chip pin alignment and calibration method based on machine vision according to claim 4, characterized in that, The calculation method for the tangential displacement safety margin includes: The product of the polar radius, the residual angular uncertainty, and the absolute value of the sine of the polar angle is determined as the tangential displacement component in the direction of the first coordinate axis. The product of the polar radius, the residual angular uncertainty, and the absolute value of the cosine of the polar angle is determined as the tangential displacement component in the direction of the second coordinate axis, which is perpendicular to the first coordinate axis.

Images