Deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction

Abstract

The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging. The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging. The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging. The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging. The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging. The application discloses a deep cavity wire bonding method and system based on visual pre-positioning and blind area coordinate reconstruction, and relates to the technical field of microelectronic packaging.

Description

Technical Field

[0001] This invention relates to the field of microelectronic packaging technology, specifically to a deep cavity wire bonding method and system based on visual pre-positioning and blind zone coordinate reconstruction. Background Technology

[0002] Deep cavity packaged devices typically require reliable electrical connections between internal chip pads and external circuit pads within a tiny cavity. For devices employing cup-bottom pressure-bearing structures, glass ring vias, TO housings, or similar deep cavity structures, the internal pads are located at the bottom of the cavity or within the via, while the external pads are located outside the cavity or at another height. Bonding tools must complete cross-cavity wire bonding under conditions of limited aperture, limited field of view, and significant height differences.

[0003] To achieve wire bonding, existing automated vision-guided wire bonding equipment uses cameras and visual positioning and recognition algorithms to automatically identify and locate target pads and guide the precise movement of the wedge to achieve wire bonding. However, most existing automated vision-guided wire bonding equipment is designed for planar conditions where the target can be observed by the vision system during both the positioning and execution phases. It does not consider that the wedge will block the camera's line of sight after entering the deep cavity, making it impossible to observe the positional relationship between the wedge tip and the pad inside the hole in real time at the moment of bonding. This makes it difficult to directly apply conventional real-time vision-guided bonding methods.

[0004] Therefore, there is an urgent need to propose a deep cavity wire bonding method and system based on visual pre-positioning and blind zone coordinate reconstruction to solve the problems that existing visually guided bonding methods cannot handle in deep cavity scenarios, such as visual obstruction after the blade is inserted into the cavity, errors in planar coordinate mapping at different heights, spatial offset between the camera and the blade, and random drift of image coordinates caused by high-temperature hot airflow. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a deep cavity wire bonding method and system based on visual pre-positioning and blind zone coordinate reconstruction. The aim is to decouple target recognition during the camera's visible phase from the cleaving execution during the camera's invisible phase, thereby improving the positioning accuracy, landing point consistency, and batch operation stability of deep cavity small pad bonding.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction. For each target pad, when the wedge has not entered the target deep cavity and the camera's field of view is not obstructed by the wedge, visual pre-positioning and blind zone execution coordinate reconstruction are completed according to steps S1 to S5 to obtain the blind zone execution coordinates corresponding to the target pad. When the wedge enters the deep cavity, the camera's field of view is obstructed, and the target pad cannot be observed in real time, the motion platform and the wedge are driven according to the blind zone execution coordinates to achieve deep cavity wire bonding.

[0008] Step S1: Obtain a deep cavity image containing the deep cavity orifice and the bonding surface where the target pad is located, and extract the edge point set of the deep cavity orifice; obtain the deep cavity contour center by fitting the edge point set of the deep cavity orifice, and establish a local coordinate system of the cavity.

[0009] Step S2: Based on the local coordinate system of the cavity, determine the dynamic region of interest according to the initial estimated position of the target pad, and determine the optimal focal plane by performing an autofocus operation within the dynamic region of interest, and further acquire the target image containing the target pad under the optimal focal plane.

[0010] Step S3: Calculate the similarity between the preset pad template and each image position within the dynamic region of interest to obtain the matching response map; perform threshold filtering and connected component analysis on the matching response map to determine the target pad region and integer pixel extreme points, and perform sub-pixel fitting on the integer pixel extreme points to obtain the sub-pixel image coordinates of the target pad.

[0011] Step S4: Substitute the subpixel image coordinates into the calibration matrix corresponding to the preset reference height plane to obtain the reference motion platform coordinates; select the corresponding height calibration matrix according to the Z-direction height of the bonding surface where the target pad is located, convert the subpixel image coordinates of the target pad into visual positioning coordinates, and obtain the height plane compensation amount according to the difference between the mapping results of the height calibration matrix and the calibration matrix of the reference height plane for the same subpixel image coordinates.

[0012] Step S5: Obtain the fixed offset compensation amount between the camera optical axis and the tip of the cutting tool, as well as the thermal disturbance compensation amount under high temperature bonding conditions. Add the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount to obtain the blind zone execution coordinates.

[0013] Another aspect of the present invention provides a deep cavity wire bonding system based on visual pre-positioning and blind zone coordinate reconstruction, used in the deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction, comprising:

[0014] The microscopic vision module is used to acquire deep cavity images, including the deep cavity opening and the bonding surface where the target pad is located, when the cleaving tool has not entered the deep cavity and the camera's field of view is not obstructed.

[0015] The deep cavity contour constraint positioning module is used to determine the local coordinate system and dynamic region of interest of the cavity based on the edge of the deep cavity opening;

[0016] An automatic focusing module is used to determine the optimal focal plane of the bonding surface where the target pad is located within a dynamically dynamic region of interest.

[0017] The pad recognition module is used to output the subpixel image coordinates of the target pad;

[0018] The coordinate mapping module is used to convert subpixel image coordinates into reference motion platform coordinates according to the calibration matrix of the preset reference height plane, and to determine the height plane compensation amount according to the Z-direction height of the bonding surface where the current target pad is located.

[0019] The thermal disturbance compensation module is used to determine the thermal disturbance compensation amount based on the coordinates of multiple sub-pixel images of the same target pad; the blind zone coordinate reconstruction module is used to synthesize the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount into blind zone execution coordinates.

[0020] The motion control module is used to drive the motion platform and the cutting tool according to the coordinates of the blind zone when the camera's field of view is blocked and the target pad cannot be observed in real time after the cutting tool enters the deep cavity, so as to realize deep cavity wire bonding.

[0021] The host computer scheduling module is used to schedule image acquisition, target recognition, coordinate mapping, compensation calculation, and motion control in conjunction with other modules in the scheduling system.

[0022] The beneficial effects of adopting the above technical solution are as follows:

[0023] (1) The present invention decouples the visual pre-positioning stage from the blind zone execution stage in time and space, so that the bonding operation that cannot be observed in real time after the cleaver enters the deep cavity can be completed according to the pre-reconstructed execution coordinates;

[0024] (2) The present invention uses deep cavity contour constraint positioning and cavity local coordinate system to make target pad search no longer rely solely on global image coordinates, thereby improving the positioning stability in complex deep cavity backgrounds;

[0025] (3) This invention reduces coordinate transformation errors caused by different height planes inside and outside the deep cavity by cross-height spatial mapping and multi-height calibration matrix selection;

[0026] (4) The present invention forms a blind zone coordinate reconstruction mechanism through fixed offset compensation, thermal disturbance compensation and height plane compensation, so that the visual recognition result can be converted into the invisible execution coordinates of the cleaver;

[0027] (5) The present invention effectively suppresses random coordinate drift caused by high temperature hot airflow by multi-frame outlier removal, averaging and matching score weighted correction. Attached Figure Description

[0028] Figure 1 This is a flowchart of the deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction of the present invention;

[0029] Figure 2 This is a schematic diagram of the automatic focus sharpness evaluation curve within the region of interest in this invention.

[0030] Figure 3 This is a schematic diagram of the pad visual recognition and sub-pixel positioning process of the present invention;

[0031] Figure 4 This is a schematic diagram illustrating the relationship between cross-height spatial mapping and multi-plane calibration in this invention;

[0032] Figure 5 This is a schematic diagram of the deep cavity contour constraint positioning and camera-cleaver spatial offset relationship of the present invention;

[0033] Figure 6 This is a schematic diagram of outlier removal and statistical compensation for multi-frame coordinates under high-temperature thermal disturbance according to the present invention;

[0034] Figure 7 This is a schematic diagram of the module composition of the deep cavity wire bonding system based on visual pre-positioning and blind zone coordinate reconstruction of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0036] Example 1:

[0037] refer to Figure 1 A deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction is proposed. For each target pad, when the wedge has not entered the target deep cavity and the camera's field of view is not obstructed by the wedge, steps S1 to S5 are performed to complete visual pre-positioning and blind zone execution coordinate reconstruction, obtaining the blind zone execution coordinates corresponding to the target pad. When the wedge enters the deep cavity, the camera's field of view is obstructed, and the target pad cannot be observed in real time, the motion platform and wedge are driven according to the blind zone execution coordinates to achieve deep cavity wire bonding.

[0038] Step S1: Obtain a deep cavity image containing the deep cavity orifice and the bonding surface where the target pad is located, and extract the edge point set of the deep cavity orifice; by fitting the edge point set of the deep cavity orifice, obtain the deep cavity contour center, and establish a local coordinate system of the cavity.

[0039] In this embodiment, the component to be bonded is fixed on the XY motion platform, the microscope camera is positioned above the component, and the Z-axis chopping mechanism is located on one side of the deep cavity. Upon startup, the motion platform, microscope camera, and Z-axis chopping mechanism are first placed in a preset initial position, ensuring the chopping tool has not yet entered the deep cavity and the microscope camera's field of view is not obstructed by the chopping tool. At this point, the microscope camera acquires an image of the deep cavity, including the cavity opening and the bonding surface where the target pad is located.

[0040] Specifically, in this embodiment, step S1 establishes the local coordinate system of the cavity as follows:

[0041] Step S11: A preprocessed image is obtained by performing grayscale conversion, filtering and noise reduction, and edge enhancement on the deep cavity image, and the edge point set of the deep cavity orifice is extracted from the preprocessed image by an edge detection algorithm;

[0042] Step S12: Fit the set of edge points of the deep cavity orifice using at least one of circle fitting, ellipse fitting and least squares contour fitting, and use the center of the circle, ellipse or contour geometry obtained by fitting as the deep cavity contour center.

[0043] Step S13: Establish a local coordinate system for the cavity, with the center of the deep cavity contour as the origin of the local coordinate system and the u-axis and v-axis directions of the image coordinate system or the major axis and minor axis directions of the deep cavity contour as the coordinate axis directions of the local coordinate system.

[0044] Step S2: Based on the local coordinate system of the cavity, determine the dynamic region of interest according to the initial estimated position of the target pad, and determine the optimal focal plane by performing an autofocus operation within the dynamic region of interest. Then, acquire the target image containing the target pad under the optimal focal plane.

[0045] Specifically, step S2 determines the dynamic region of interest as follows:

[0046] Step S21: Based on the local coordinate system of the cavity, determine the initial estimated position of the target pad in the deep cavity image according to the preset design position of the deep cavity contour center and the target pad relative to the deep cavity contour center.

[0047] Step S22: Centered on the initial estimated position, a search window is determined based on the target pad size, camera magnification, and motion platform positioning error, and this search window is used as a dynamic region of interest. After determining the target pad region, the initial estimated position and search window of the next target pad are updated based on the identified target pad position. Thus, the search range of the target pad is limited to a local region constrained by the deep cavity contour, rather than performing an unconstrained search across the entire image.

[0048] Furthermore, the automatic focusing operation specifically includes:

[0049] Using the initial estimated height of the bonding surface where the target pad is located as the center, the system moves upward and downward by a preset distance along the Z-axis to form a first scanning range. Moving gradually from one end of the first scanning range to the other with a first preset step size, multiple Z-axis height positions are obtained. At each Z-axis height position, a deep cavity image is acquired, and a sharpness evaluation value is calculated. The Z-axis height position with the highest sharpness evaluation value is then selected. In this embodiment, the Z-axis height position is the sampling position.

[0050] Centered on the selected Z-axis height position, a preset distance is extended upwards and downwards along the Z-axis to form a local scanning interval. This local scanning interval is a height range along the Z-axis, not a planar area of ​​the image. Based on a second preset step size smaller than the first preset step size, the system gradually moves to multiple Z-axis height positions within the local scanning interval. At each Z-axis height position, a deep cavity image is acquired, and a sharpness evaluation value is calculated. The Z-axis height position with the highest sharpness evaluation value is determined as the optimal focal plane. In this embodiment, the Laplacian variance is used as the sharpness evaluation index.

[0051] refer to Figure 2 In this embodiment, the coarse focusing stage scans in steps of 0.5mm within the range of -7.5mm to 7.5mm, and the fine focusing stage scans in steps of 0.05mm near the peak of the coarse scan.

[0052] Furthermore, in this embodiment, when the maximum sharpness evaluation value obtained by coarse focusing or fine focusing is lower than the preset sharpness threshold, or when the sharpness evaluation curve shows two or more peak values ​​that do not exceed the preset peak difference threshold, the dynamic region of interest is redefined, and at least one of the following operations, such as adjusting the illumination parameters or expanding the first scanning range, is performed before refocusing; if the preset sharpness threshold is still not met after refocusing, a focusing failure message is output, and the blind bonding of the target pad is stopped.

[0053] Step S3: Calculate the similarity between the preset pad template and each image position within the dynamic region of interest to obtain a matching response map; perform threshold filtering and connected component analysis on the matching response map to determine the target pad region and integer pixel extreme points, and perform subpixel fitting on the integer pixel extreme points to obtain the subpixel image coordinates of the target pad.

[0054] refer to Figure 3 In this embodiment, step S3 determines the target pad area and the integer pixel extreme point according to the following steps:

[0055] Step S31: Use at least one of the following preprocessing methods to preprocess the target image acquired in step S2 under the optimal focal plane: grayscale normalization, background subtraction, and local contrast enhancement.

[0056] Step S32: Within the dynamic region of interest of the preprocessed target image, the preset pad template is moved step by step according to a preset sliding step size. At each moving position, the matching score between the preset pad template and the corresponding local image window is calculated to form a matching response map. Adjacent positions in the matching response map with scores not less than a preset matching threshold are merged to obtain an initial candidate region. In this embodiment, the preset matching threshold ranges from 0.6 to 0.9.

[0057] Step S33: Perform connected component analysis on the initial candidate regions, and filter them by constraints such as area, aspect ratio, roundness, rectangularity, and distance from the initial estimated position of the target pad to obtain candidate pad regions;

[0058] When there are multiple candidate pad regions, the maximum matching scores of each candidate pad region are sorted in descending order, and the candidate pad region with the highest maximum matching score is selected as the target pad region; when two or more candidate pad regions have the same maximum matching score, the candidate pad region whose center position is closest to the initial estimated position of the target pad is selected as the target pad region; the pixel position with the highest matching score in the target pad region is determined as the integer pixel extreme point.

[0059] To improve the positioning accuracy of the pad center, sub-pixel fitting is further performed on the integer pixel extreme points using the following steps:

[0060] Step S31: Take the center of the pixel position where the extreme point of the whole pixel is located as the center of the whole pixel;

[0061] Step S32: In the 3×3 neighborhood of the integer pixel extreme point, obtain the matching score of the integer pixel extreme point and the position of its two adjacent pixels along the u and v directions of the image coordinate system, respectively, and fit it into a one-dimensional parabola. Further, the offset of the peak position of the one-dimensional parabola relative to the center of the integer pixel is taken as the sub-pixel offset in the corresponding direction.

[0062] Taking the extreme point of an integer pixel and its adjacent pixel positions on both sides of the u direction as an example, let the matching scores of the extreme point of the integer pixel and its adjacent pixel positions on both sides of the u direction be Ru(0), Ru(-1), and Ru(+1) respectively. Calculate sub-pixel offset in the u direction And in the same manner, the sub-pixel offset in the v direction is calculated based on the matching scores of the integer pixel extreme points and their adjacent pixel positions on both sides of the v direction. ;

[0063] Step S33: Add the subpixel offset in the u direction and the subpixel offset in the v direction to the corresponding image coordinates of the whole pixel center to obtain the subpixel image coordinates of the target pad.

[0064] In this embodiment, let the image coordinates of the center of the integer pixel be (u, v), then the sub-pixel image coordinates of the target pad are... .

[0065] Preferably, when the image noise is strong or the matching response peak is wide, the sub-pixel offset can be obtained by fitting a two-dimensional quadratic surface in the neighborhood of the integer pixel extreme point.

[0066] Step S4: Substitute the subpixel image coordinates into the calibration matrix corresponding to the preset reference height plane to obtain the reference motion platform coordinates; select the corresponding height calibration matrix according to the Z-direction height of the bonding surface where the target pad is located, convert the subpixel image coordinates of the target pad into visual positioning coordinates, and obtain the height plane compensation amount according to the difference between the mapping results of the height calibration matrix and the calibration matrix of the reference height plane for the same subpixel image coordinates.

[0067] Specifically, the height plane compensation amount is as follows: based on the calibration matrix of each preset height plane and its corresponding calibration height, the Z-axis height of the bonding surface where the target pad is located is compared with each calibration height, and a calibration matrix whose height difference does not exceed the preset height tolerance is selected. When there are multiple calibration matrices that meet the conditions, the calibration matrix with the smallest height difference is selected. The subpixel image coordinates of the same target pad are substituted into the selected height calibration matrix and the calibration matrix of the preset reference height plane, respectively, and the difference in the X-direction and Y-direction coordinates of the motion platform obtained from the two mappings are used as the height plane compensation amount.

[0068] refer to Figure 4 In this embodiment, step S4 uses a multi-point calibration method with nine calibration points to obtain the calibration matrix corresponding to each preset height plane, and the calibration points are arranged in a grid.

[0069] Specifically, calibration feature points that can be recognized by the camera are set on the calibration plate; for each preset height plane, the motion platform is moved sequentially in the XY plane according to the preset row spacing and column spacing, so that the same calibration feature point is located in nine different positions, the XY coordinates of the motion platform at each position are recorded, and the image coordinates of the calibration feature points are extracted from the calibration images collected at each position.

[0070] For any calibration position, establish a two-dimensional affine transformation equation using the image coordinates and XY coordinates of that position; combine the two-dimensional affine transformation equations corresponding to all calibration positions into an overdetermined linear equation system, and solve the parameters using the least squares method to form the calibration matrix corresponding to the preset height plane.

[0071] In this embodiment, let the image coordinates of the i-th calibration point be (u i ,v i ), whose corresponding XY coordinates are (X i ,Y i The two-dimensional affine transformation relationship between the image coordinates and the XY coordinates is as follows:

[0072] ;

[0073] .

[0074] Step S5: Obtain the fixed offset compensation amount between the camera optical axis and the tip of the cutting tool, as well as the thermal disturbance compensation amount under high temperature bonding conditions. Add the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount to obtain the blind zone execution coordinates.

[0075] refer to Figure 5 The fixed offset compensation amount is obtained through indentation calibration: an indentation is formed on the surface of the test piece using a chopping tool, and the motion platform coordinates of the chopping tool's working point are recorded; the indentation is moved to the center of the camera's field of view, and the motion platform coordinates of the camera's visual point are recorded; the difference between the motion platform coordinates of the chopping tool's working point and the motion platform coordinates of the camera's visual point is used as the fixed offset compensation amount from the camera's visual point to the chopping tool's working point. Indentation calibration can be re-performed after chopping tool replacement, camera magnification switching, fixture replacement, or significant changes in equipment temperature.

[0076] refer to Figure 6 The thermal disturbance compensation amount is obtained through thermal disturbance statistical compensation: Under high-temperature bonding conditions, N frames of images containing the same target pad are continuously acquired at the optimal focal plane, and the sub-pixel image coordinates of the target pad in each frame are calculated; the two-dimensional distance of each frame's sub-pixel image coordinates relative to the median coordinate is calculated, and sub-pixel image coordinates whose two-dimensional distance exceeds a preset outlier threshold are removed. The median coordinate is obtained by sorting the sub-pixel coordinates of multiple frames in the X and Y directions and taking the median value. The target pad region matching score corresponding to the frame containing each unremoved sub-pixel image coordinate is used as the weight of that coordinate, and a weighted average is performed on each unremoved sub-pixel image coordinate to obtain the thermal disturbance stable coordinates; the thermal disturbance stable coordinates and the sub-pixel image coordinates obtained in step S3 are mapped using a calibration matrix of a preset reference height plane, and the difference between the motion platform coordinates obtained from the two mappings is used as the thermal disturbance compensation amount. Furthermore, when the standard deviation of the remaining pad coordinates exceeds a preset stability threshold, multiple frames of images are reacquired. In this embodiment, the number of consecutively acquired frames N is 8 to 20.

[0077] In this embodiment, the deep cavity diameter is 1.6 mm, the side length of the pad inside the cavity is 160 μm, the number of pads inside the cavity is 5, the diameter of the bonding wire is 25 μm, and the first bonding point is located on the chip pad inside the deep cavity, while the second bonding point is located on the circuit board pad outside the deep cavity. Following the steps described above, visual pre-positioning, coordinate mapping, multi-level compensation, and blind zone coordinate reconstruction are performed on the first and second bonding points respectively to obtain the first blind zone coordinates corresponding to the first bonding point and the second blade blind zone coordinates corresponding to the second bonding point. When the camera's field of view is obstructed by the chopper or the deep cavity structure, preventing real-time observation of the target pad, a bonding trajectory is generated according to the first blind zone coordinates, the second blind zone coordinates, and preset wire drawing arc parameters, following the process sequence of first bonding, wire drawing arc, and second bonding. The motion platform and chopper are controlled to complete the cross-deep cavity lead interconnection, and the average tensile strength is 5.78 gf, indicating that this embodiment can form a bonding lead with a certain mechanical connection strength.

[0078] Example 2:

[0079] refer to Figure 7 A deep cavity wire bonding system based on visual pre-positioning and blind zone coordinate reconstruction, used in the deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction, includes:

[0080] The microscopic vision module is used to acquire deep cavity images, including the deep cavity opening and the bonding surface where the target pad is located, when the cleaving tool has not entered the deep cavity and the camera's field of view is not obstructed.

[0081] The deep cavity contour constraint positioning module is used to determine the local coordinate system and dynamic region of interest of the cavity based on the edge of the deep cavity opening;

[0082] An automatic focusing module is used to determine the optimal focal plane of the bonding surface where the target pad is located within a dynamically dynamic region of interest.

[0083] The pad recognition module is used to output the subpixel image coordinates of the target pad;

[0084] The coordinate mapping module is used to convert subpixel image coordinates into reference motion platform coordinates according to the calibration matrix of the preset reference height plane, and to determine the height plane compensation amount according to the Z-direction height of the bonding surface where the current target pad is located.

[0085] The thermal disturbance compensation module is used to determine the thermal disturbance compensation amount based on the coordinates of multiple sub-pixel images of the same target pad; the blind zone coordinate reconstruction module is used to synthesize the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount into blind zone execution coordinates.

[0086] The motion control module is used to drive the motion platform and the cutting tool according to the coordinates of the blind zone when the camera's field of view is blocked and the target pad cannot be observed in real time after the cutting tool enters the deep cavity, so as to realize deep cavity wire bonding.

[0087] The host computer scheduling module is used to perform image acquisition, target recognition, coordinate mapping, compensation calculation and motion control through other modules in the scheduling system.

[0088] Furthermore, in this embodiment, the deep cavity wire bonding system based on visual pre-positioning and blind zone coordinate reconstruction also includes a temperature acquisition module, which is used to record the temperature of the bonding area and assist in adjusting the statistical compensation amount of thermal disturbance.

[0089] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction, characterized in that, For each target pad, before the wedge enters the target cavity and the camera's field of view is obstructed by the wedge, visual pre-positioning and blind zone execution coordinate reconstruction are completed according to steps S1 to S5 to obtain the blind zone execution coordinates corresponding to the target pad. Based on the blind zone execution coordinates, when the camera's field of view is obstructed after the wedge enters the cavity and the target pad cannot be observed in real time, the motion platform and wedge are driven according to the blind zone execution coordinates to achieve deep cavity wire bonding. Step S1: Obtain a deep cavity image containing the deep cavity orifice and the bonding surface where the target pad is located, and extract the edge point set of the deep cavity orifice; obtain the deep cavity contour center by fitting the edge point set of the deep cavity orifice, and establish a local coordinate system of the cavity. Step S2: Based on the local coordinate system of the cavity, determine the dynamic region of interest according to the initial estimated position of the target pad, and determine the optimal focal plane by performing an autofocus operation within the dynamic region of interest, and further acquire the target image containing the target pad under the optimal focal plane. Step S3: Calculate the similarity between the preset pad template and each image position within the dynamic region of interest to obtain the matching response map; perform threshold filtering and connected component analysis on the matching response map to determine the target pad region and integer pixel extreme points, and perform sub-pixel fitting on the integer pixel extreme points to obtain the sub-pixel image coordinates of the target pad. Step S4: Substitute the subpixel image coordinates into the calibration matrix corresponding to the preset reference height plane to obtain the reference motion platform coordinates; select the corresponding height calibration matrix according to the Z-direction height of the bonding surface where the target pad is located, convert the subpixel image coordinates of the target pad into visual positioning coordinates, and obtain the height plane compensation amount according to the difference between the mapping results of the height calibration matrix and the calibration matrix of the reference height plane for the same subpixel image coordinates. Step S5: Obtain the fixed offset compensation amount between the camera optical axis and the tip of the cutting tool, as well as the thermal disturbance compensation amount under high temperature bonding conditions. Add the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount to obtain the blind zone execution coordinates.

2. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 1, characterized in that, Step S1 specifically involves: Step S11: Obtain a deep cavity image containing the deep cavity orifice and the bonding surface where the target pad is located. By performing grayscale conversion, filtering and noise reduction, and edge enhancement on the deep cavity image, a preprocessed image is obtained. The edge point set of the deep cavity orifice is extracted from the preprocessed image using an edge detection algorithm. Step S12: Fit the set of edge points of the deep cavity orifice using at least one of circle fitting, ellipse fitting and least squares contour fitting, and use the center of the circle, ellipse or contour geometry obtained by fitting as the deep cavity contour center. Step S13: Establish a local coordinate system for the cavity, with the center of the deep cavity contour as the origin of the local coordinate system and the u-axis and v-axis directions of the image coordinate system or the major axis and minor axis directions of the deep cavity contour as the coordinate axis directions of the local coordinate system.

3. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 1, characterized in that, The specific method for determining the dynamically interested region in step S2 is as follows: Step S21: Based on the local coordinate system of the cavity, determine the initial estimated position of the target pad in the deep cavity image according to the preset design position of the deep cavity contour center and the target pad relative to the deep cavity contour center. Step S22: Using the initial estimated position as the center, determine the search window based on the target pad size, camera magnification, and motion platform positioning error, and use the search window as a dynamic region of interest; after determining the target pad region, update the initial estimated position and search window of the next target pad with the identified target pad position.

4. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 3, characterized in that, The automatic focusing operation is specifically as follows: Using the initial estimated height of the bonding surface where the target pad is located as the center, move a preset distance upward and downward along the Z-axis to form a first scanning range; move from one end of the first scanning range to the other end with a first preset step size to obtain multiple Z-axis height positions, and acquire deep cavity images and calculate the sharpness evaluation value at each Z-axis height position, and select the Z-axis height position with the highest sharpness evaluation value. Centered on the selected Z-axis height position, the local scanning interval is formed by extending a preset distance upwards and downwards along the Z-axis. Based on a second preset step size smaller than the first preset step size, the system gradually moves to multiple Z-axis height positions within the local scanning interval. At each Z-axis height position, a deep cavity image is acquired and a sharpness evaluation value is calculated. The Z-axis height position with the highest sharpness evaluation value is determined as the optimal focal plane.

5. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 1, characterized in that, The specific method for determining the target pad region and the integer pixel extreme point in step S3 is as follows: Step S31: Use at least one of the following preprocessing methods to preprocess the target image acquired in step S2 under the optimal focal plane: grayscale normalization, background subtraction, and local contrast enhancement. Step S32: Within the dynamic region of interest of the preprocessed target image, the preset pad template is moved step by step according to a preset sliding step size. At each moving position, the matching score between the preset pad template and the corresponding local image window is calculated to form a matching response map. Adjacent positions in the matching response map with scores not less than a preset matching threshold are merged to obtain the initial candidate region. Step S33: Perform connected component analysis on the initial candidate regions and filter them based on constraints such as area, aspect ratio, roundness, rectangularity, and distance from the initial estimated position of the target pad to obtain candidate pad regions; When there are multiple candidate pad regions, the maximum matching scores of each candidate pad region are sorted in descending order, and the candidate pad region with the highest maximum matching score is selected as the target pad region; when two or more candidate pad regions have the same maximum matching score, the candidate pad region whose center position is closest to the initial estimated position of the target pad is selected as the target pad region; the pixel position with the highest matching score in the target pad region is determined as the integer pixel extreme point.

6. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 1, characterized in that, The specific method for obtaining the sub-pixel image coordinates of the target pad in step S3 is as follows: Step S31: Take the center of the pixel position where the extreme point of the whole pixel is located as the center of the whole pixel; Step S32: In the 3×3 neighborhood of the integer pixel extreme point, obtain the matching score of the integer pixel extreme point and its adjacent pixel positions on both sides along the u and v directions of the image coordinate system, respectively, and fit it into a one-dimensional parabola. Further, the offset of the peak position of the one-dimensional parabola relative to the center of the integer pixel is taken as the sub-pixel offset in the corresponding direction. Step S33: Add the subpixel offset in the u direction and the subpixel offset in the v direction to the corresponding image coordinates of the whole pixel center to obtain the subpixel image coordinates of the target pad.

7. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 1, characterized in that, Step S4 uses a multi-point calibration method with at least nine calibration points to obtain the calibration matrix corresponding to each preset height plane, and the calibration points are arranged in a grid. Specifically, calibration feature points that can be recognized by the camera are set on the calibration plate; for each preset height plane, the motion platform is moved sequentially in the XY plane according to the preset row spacing and column spacing, so that the same calibration feature point is located in at least nine different positions, the XY coordinates of the motion platform at each position are recorded, and the image coordinates of the calibration feature points are extracted from the calibration images collected at each position. For any calibration position, establish a two-dimensional affine transformation equation using the image coordinates and XY coordinates of that position; The two-dimensional affine transformation equations corresponding to all calibration positions are combined into an overdetermined linear equation system, and the parameters are solved using the least squares method to form the calibration matrix corresponding to the preset height plane.

8. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 7, characterized in that, The height plane compensation amount is specifically as follows: based on the calibration matrix of each preset height plane and its corresponding calibration height, the Z-axis height of the bonding surface where the target pad is located is compared with each calibration height, and a calibration matrix whose height difference does not exceed the preset height tolerance is selected. When there are multiple calibration matrices that meet the conditions, the calibration matrix with the smallest height difference is selected. The subpixel image coordinates of the same target pad are substituted into the selected height calibration matrix and the calibration matrix of the preset reference height plane, respectively, and the X-direction difference and Y-direction difference of the motion platform coordinates obtained from the two mappings are used as the height plane compensation amount.

9. The deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction according to claim 5, characterized in that, The fixed offset compensation amount is obtained through indentation calibration: an indentation is formed on the surface of the test piece using a chopping tool and the motion platform coordinates of the chopping tool's working point are recorded; the indentation is moved to the center of the camera's field of view and the motion platform coordinates of the camera's visual point are recorded; the difference between the motion platform coordinates of the chopping tool's working point and the motion platform coordinates of the camera's visual point is used as the fixed offset compensation amount from the camera's visual point to the chopping tool's working point. The thermal disturbance compensation amount is obtained through thermal disturbance statistical compensation: under high-temperature bonding conditions, N frames of images containing the same target pad are continuously acquired under the optimal focal plane, and the sub-pixel image coordinates of the target pad in each frame are calculated; the two-dimensional distance of each frame's sub-pixel image coordinates relative to the median coordinates is calculated, and sub-pixel image coordinates whose two-dimensional distance exceeds a preset outlier threshold are removed; the target pad region matching score corresponding to the frame where each unremoved sub-pixel image coordinate is located is used as the weight of the coordinate, and a weighted average is performed on each unremoved sub-pixel image coordinate to obtain the thermal disturbance stable coordinates; the thermal disturbance stable coordinates and the sub-pixel image coordinates obtained in step S3 are mapped through the calibration matrix of a preset reference height plane, and the difference between the motion platform coordinates obtained from the two mappings is used as the thermal disturbance compensation amount.

10. A deep cavity wire bonding system based on visual pre-positioning and blind zone coordinate reconstruction, characterized in that, The method for implementing the deep cavity wire bonding method based on visual pre-positioning and blind zone coordinate reconstruction as described in any one of claims 1 to 9 includes: The microscopic vision module is used to acquire deep cavity images, including the deep cavity opening and the bonding surface where the target pad is located, when the cleaving tool has not entered the deep cavity and the camera's field of view is not obstructed. The deep cavity contour constraint positioning module is used to determine the local coordinate system and dynamic region of interest of the cavity based on the edge of the deep cavity opening; An automatic focusing module is used to determine the optimal focal plane of the bonding surface where the target pad is located within a dynamically dynamic region of interest. The pad recognition module is used to output the subpixel image coordinates of the target pad; The coordinate mapping module is used to convert subpixel image coordinates into reference motion platform coordinates according to the calibration matrix of the preset reference height plane, and to determine the height plane compensation amount according to the Z-direction height of the bonding surface where the current target pad is located. The thermal disturbance compensation module is used to determine the thermal disturbance compensation amount based on the coordinates of multiple sub-pixel images of the same target pad; the blind zone coordinate reconstruction module is used to synthesize the reference motion platform coordinates, height plane compensation amount, fixed offset compensation amount and thermal disturbance compensation amount into blind zone execution coordinates. The motion control module is used to drive the motion platform and the cutting tool according to the coordinates of the blind zone when the camera's field of view is blocked and the target pad cannot be observed in real time after the cutting tool enters the deep cavity, so as to realize deep cavity wire bonding. The host computer scheduling module is used to perform image acquisition, target recognition, coordinate mapping, compensation calculation and motion control through other modules in the scheduling system.

Images