Global calibration and compensation method for high-precision laminating equipment

Abstract

The invention discloses a global calibration and real-time compensation method for high-precision laminating equipment, and the method comprises the steps: firstly, selecting a plurality of calibration points in the whole working region of a calibration substrate, and obtaining the position deviation value of each calibration point through restoring the actual laminating process and measuring, thereby forming an original calibration data set; secondly, based on the data set, constructing a two-dimensional space compensation network by taking theoretical coordinates of each calibration point as nodes; and finally, during actual fitting, for any target fitting coordinate, an adjacent reference node is determined by querying the network, a prediction compensation value is calculated through a spatial interpolation algorithm based on the data of the reference node, and then fitting is performed after the motion instruction is compensated. According to the method, systematic error surveying and mapping are performed on the whole working area of the equipment, and the compensation network is constructed, so that accurate and dynamic compensation of any laminating point is realized, and the global precision and adaptability of the laminating equipment are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of machine vision and precision automation equipment technology, specifically to a global calibration and compensation method for high-precision bonding equipment. Background Technology

[0002] In high-end equipment manufacturing fields such as semiconductor packaging and high-precision mounting, vision alignment is a key step in achieving precision machining and assembly. For example, the current mainstream method for semiconductor bonding alignment is to use a fixed camera to photograph the marking points on the workpiece (i.e., the marking points on the wafer or chip), use pattern recognition (PR) algorithms to calculate its center coordinates, and compare them with the target position to guide the mechanical platform to perform compensation alignment.

[0003] Furthermore, for high-precision bonding equipment, errors inherent in the motion unit, vision positioning unit, and component handling unit during actual operation can lead to systematic errors in the final equipment operation. These errors are not uniformly distributed but vary with position. Traditional calibration methods include:

[0004] The technical solution adopted in patent document CN120926913A is to perform calibration only at the origin of the equipment or a few specific locations.

[0005] The technical solution adopted in patent document CN118522678 is to calibrate only the relative relationship between the vision system and the robotic arm;

[0006] The technical solution adopted in patent document CN118444541A is to perform internal parameter calibration of a camera.

[0007] The aforementioned existing technical solutions cannot acquire and compensate for the systematic spatial error distribution of the equipment throughout its entire working area. Furthermore, the existing local calibration or single-point compensation methods have the following limitations:

[0008] 1. The compensation effect is limited to the vicinity of the calibration point; for non-calibration points, the accuracy cannot be guaranteed.

[0009] 2. Unable to adapt to error field drift caused by long-term operation and temperature changes in the equipment;

[0010] 3. For processes that require high-density, high-precision bonding on the entire substrate (such as wafers or panels), there is a lack of global precision assurance measures.

[0011] Therefore, in view of the shortcomings of existing technical solutions, the inventors propose the following technical solutions. Summary of the Invention

[0012] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a global calibration and compensation method for high-precision bonding equipment. This method can systematically map the error of the entire working area of ​​the equipment and provide real-time, accurate compensation for any bonding point.

[0013] To solve the above technical problems, the positioning method of the present invention adopts the following technical solution: a global calibration and real-time compensation method for a high-precision bonding device, comprising the following steps: Step S1: Global error calibration data acquisition. Multiple calibration points are selected within the entire working area of ​​the calibration substrate. For each calibration point, the drive mechanism is controlled to perform an operation to restore the actual bonding process. For each calibration point, the position deviation value of the actual bonding position relative to its theoretical bonding coordinates is obtained by measurement. The theoretical bonding coordinates of each calibration point and the position deviation value measured at that point are recorded to form an original calibration dataset covering the working area. Step S2: Spatial compensation network construction. Based on the original calibration dataset, a two-dimensional spatial compensation network is constructed, wherein the theoretical bonding coordinates of each calibration point serve as the network... The nodes are associated and their corresponding position deviation values ​​are stored; Step S3: Real-time compensation bonding. In the actual bonding process, for the target bonding coordinates of any bonding element, the following steps are performed: S31: Network query step. Based on the target bonding coordinates, one or more neighboring nodes close to the target bonding coordinates are determined as reference nodes in the spatial compensation network; S32: Interpolation compensation step. Based on the theoretical bonding coordinates and associated position deviation values ​​of the one or more reference nodes, a predicted compensation value corresponding to the target bonding coordinates is calculated through a spatial interpolation algorithm; S33: Motion synthesis step. The predicted compensation value and the target bonding coordinates are synthesized to generate a compensated motion command; S34: Execution step. The drive mechanism is controlled to complete the bonding action according to the compensated motion command.

[0014] Furthermore, in the above technical solution, in step S1, the actual bonding step controls the mounting head to pick up the bonding element from the material picking position and move it to the theoretical bonding coordinates of the bonding element. The position of the bonding element after bonding is measured by the vision positioning unit to obtain the positional deviation measurement value between the theoretical bonding coordinates and the actual bonding coordinates of the bonding element.

[0015] Furthermore, in the above technical solution, in step S1, the complete operation is repeated at least N times, and the average value of the N position deviation measurements is calculated, and the average value is used as the final position deviation value of the calibration point.

[0016] Furthermore, in the above technical solution, in step S1, the position of the bonding element after bonding is measured by the visual positioning unit, and firstly, the visual positioning unit performs a first image recognition for coarse positioning, and then, based on the coarse positioning, performs a second image recognition for fine positioning, and finally calculates the position deviation value based on the result of fine positioning.

[0017] Furthermore, in the above technical solution, in step S1, after obtaining the position deviation value of each calibration point, the position deviation value is verified. The verification method is as follows: using the position deviation value as compensation, the drive mechanism is controlled to repeat the bonding operation once and the new position deviation is measured. If the new position deviation exceeds the allowable threshold, the data of the calibration point is marked as invalid or recalibrated.

[0018] Furthermore, in the above technical solution, before step S1 begins, a preheating step is also included: controlling the high-precision bonding equipment drive mechanism to run unloaded for a unit time until its temperature reaches a stable state.

[0019] Furthermore, in the above technical solution, the multiple calibration points are distributed in a two-dimensional grid pattern to cover the working area.

[0020] Furthermore, in the above technical solution, in step S31, the reference node selection method is as follows: determine the grid cell in which the target fitting coordinates are located in the spatial compensation network, and take the multiple vertex nodes constituting the grid cell as the reference nodes.

[0021] Furthermore, in the above technical solution, in step S32, the spatial interpolation algorithm includes any one of the following methods, and is adaptively selected according to the positional relationship between the target fitting coordinates and the reference nodes: A. When the target fitting coordinates are located inside a grid cell composed of four reference nodes, a bilinear interpolation algorithm is used; B. When the distance between the target fitting coordinates and a certain reference node is less than a preset threshold, the positional deviation value of the reference node is directly used as the predicted compensation value; C. When the target fitting coordinates are located at the edge of the working area and only two reference nodes along the same direction can be obtained, a unilinear interpolation algorithm is used.

[0022] In the technical solution of this invention, a spatial compensation network is constructed through systematic mapping as an "error map," and dynamic compensation is achieved by querying this "error map" and interpolating during actual bonding. In a preferred embodiment, the data acquisition step captures comprehensive errors by simulating the real production process (material acquisition-movement-bonding-measurement), and improves the reliability of single-point calibration data by averaging multiple measurements and using two-level visual positioning (coarse-fine). Simultaneously, an online verification step is introduced to ensure the validity of each calibration data point. The compensation network is preferably constructed based on calibration points distributed in a two-dimensional grid. During actual compensation, the most suitable interpolation algorithm is adaptively selected based on the positional relationship between the target point and network nodes, thereby ensuring both accuracy and efficiency, as well as robustness.

[0023] The present invention, employing the above-described technical solution, has the following beneficial effects:

[0024] (1) By intensively calibrating the entire working area, a spatial error model covering the entire area was established, which can provide accurate compensation for any bonding point and significantly improve the overall bonding accuracy of the equipment.

[0025] (2) Real-time compensation based on interpolation algorithm to achieve dynamic compensation and ensure accuracy and stability.

[0026] (3) Multiple measures such as multiple measurements, two-level positioning, and online verification were adopted to ensure the accuracy and reliability of the calibration data itself. Attached Figure Description

[0027] Figure 1 These are schematic diagrams of embodiments one and two of the spatial interpolation algorithm of the present invention;

[0028] Figure 2 This is a schematic diagram of Embodiment 3 of the spatial interpolation algorithm of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] This invention discloses a global calibration and real-time compensation method for high-precision bonding equipment. It simulates the actual bonding process on a calibration substrate, performing full-disk or regional bonding operations and recording the bonding positions and their corresponding PB (position deviation) results to form a compensation data network for the equipment in physical space. During the actual bonding process, based on the bonding position coordinates, the bonding compensation value is obtained from the compensation data through linear interpolation, achieving high-precision bonding. For simplicity, the following description uses chip mounting alignment and positioning in semiconductor chip mounting processes as an example, but the application of this invention is not limited to this; it can be applied to semiconductor and electronic manufacturing scenarios with extremely high positional accuracy requirements, such as chip flip-chip bonding, precision dispensing, and component mounting.

[0031] Taking high-precision chip mounting equipment as an example, before global calibration, a "calibration substrate" and a "calibration sheet" for simulating the chip should be prepared. This calibration substrate serves as a reference board for calibrating the equipment and is used for calibrating industrial cameras, pick-and-place machines, image measuring instruments, and other similar devices. Typically, the calibration substrate is made of a single piece of glass with precisely machined patterns (such as concentric circles or grids) on its surface. By capturing these patterns with a camera in the vision positioning unit, the conversion relationship between lens distortion, pixels, and physical dimensions can be calculated, thereby achieving high-precision measurement. This type of glass calibration board is commonly used for calibrating industrial cameras, AOI pick-and-place machines, image measuring instruments, microscopes, and other similar equipment. The calibration sheet is a glass plate with positioning marks on its surface, matching the actual size of the chip being mounted.

[0032] In this invention, the calibration substrate is used as the absolute reference for physical space coordinates. During global calibration, the device will perform a restoration operation on this known calibration substrate according to the actual bonding operation, thereby measuring the systematic error of the device itself at various positions.

[0033] The high-precision bonding equipment used in this invention typically includes:

[0034] The work platform is used to support calibration substrates or production substrates.

[0035] A visual positioning unit, comprising at least one camera, is used for image capture during visual alignment. Typically, the camera is positioned above the work platform, capturing images of the field of view below for visual positioning.

[0036] The placement head is used to pick up or extract chip components and place them onto a designated position on a calibration substrate or production substrate.

[0037] The drive mechanism is linked to the work platform, placement head, and vision positioning unit, and executes corresponding actions through the control unit to perform positioning operations according to a preset process.

[0038] Based on the aforementioned high-precision bonding equipment, the method of the present invention includes: a global calibration stage in offline mode and a real-time compensation stage in online mode.

[0039] Before performing the global calibration phase in offline mode, a preheating step should be performed on the equipment: control the motion axes of the high-precision bonding equipment to run unloaded for a unit of time until the system temperature reaches a stable state. For example, control the X, Y, and Z motion axes in the drive mechanism of the bonding equipment to run unloaded for about 0.5 hours, and wait for the system temperature, including the vision positioning unit, to stabilize, in order to eliminate thermal deformation errors caused by cold start. Global calibration phase:

[0040] The global calibration stage of this invention employs offline calibration. In this stage, to reduce waste and avoid production errors, the bonding components do not directly use chips, but instead utilize the calibration sheet described earlier. Following the actual chip bonding process, the calibration sheet is used to simulate the process on a calibration substrate. Specifically, this includes the following steps.

[0041] Step S1: Global error calibration data acquisition. Select multiple calibration points within the entire working area of ​​the calibration substrate. For each calibration point, control the drive mechanism to perform the operation of restoring the actual bonding process. For each calibration point, measure the position deviation value of the actual bonding position relative to its theoretical bonding coordinates. Record the theoretical bonding coordinates of each calibration point and the position deviation value measured at that point to form an original calibration dataset covering the working area.

[0042] The theoretical bonding coordinates refer to the target position coordinates that the mounting head is expected to reach, which are pre-set in the equipment control coordinate system for each calibration point.

[0043] In step S1, the actual bonding step controls the mounting head to pick up the "calibration sheet" from the material picking position and move it to the theoretical bonding coordinates. The position of the calibration sheet after bonding is measured by the vision positioning unit, so that the positional deviation between the theoretical bonding coordinates and the actual bonding coordinates can be calculated.

[0044] To ensure data accuracy and avoid errors caused by a single calibration, in step S1, the complete operation is repeated at least N times (N is not less than 3), and the average value of the N position deviation measurements is calculated, which is then used as the final position deviation value of the calibration point.

[0045] To improve efficiency, in step S1, the position of the calibration piece after bonding is measured by a visual positioning unit. First, the visual positioning unit performs a first image recognition for coarse positioning. Then, based on the coarse positioning, a second image recognition is performed for fine positioning. Finally, the position deviation value is calculated based on the result of the fine positioning. The first image recognition uses a relatively high-tolerance recognition method to quickly locate the target. The second image recognition uses the center position found in the first instance as a basis for a finer search, for example, improving the positioning accuracy to one-tenth of a pixel or even better, thereby obtaining a precise position.

[0046] After completing step S1 above, after obtaining the position deviation value of each calibration point, the position deviation value is verified. The verification method is as follows: using the position deviation value as compensation, the drive mechanism is controlled to repeat the bonding operation once and the new position deviation is measured. If the new position deviation exceeds the allowable threshold, the data of the calibration point is marked as invalid or recalibrated.

[0047] The global calibration stage will be further explained below with reference to a specific embodiment.

[0048] Within the working area of ​​the calibration substrate, a two-dimensional grid of M rows × N columns is planned, selecting a total of M × N calibration points. The grid spacing is determined based on the equipment accuracy requirements and calibration time. For each calibration point P(i,j) in the grid, the following operations are performed:

[0049] a. Control the placement head to pick up a glass calibration sheet (corresponding to the component to be bonded) from the transfer table.

[0050] b. Drive the placement head to move the calibration piece to the theoretical bonding coordinates (X_theory, Y_theory) of point P(i,j).

[0051] c. Control the placement head to perform the bonding action.

[0052] d. Measurement of positional deviation: Visual measurement is performed using a camera in a vision positioning system located above the placement head. This process employs two-stage positioning. A coarse positioning center C1 for the calibration piece markings is quickly obtained through a first image recognition. Then, a second image recognition is performed, using C1 as the center, to obtain the fine positioning center C2.

[0053] Calculate the difference between the precise positioning center C2 and the theoretical fitting coordinates to obtain the single position deviation measurement value (ΔX_meas, ΔY_meas).

[0054] e. Return the calibration plate to the transfer station.

[0055] f. Repeat step ae a total of 5 times. Calculate the average of (ΔX_meas, ΔY_meas) obtained from the 5 measurements, and use it as the final position deviation value (ΔX_final, ΔY_final) of the calibration point P(i,j).

[0056] g. Verification (optional step): Using (ΔX_final, ΔY_final) as compensation values, control the placement head to perform the pick-up to bonding operation again (using the compensated coordinates). Measure the new positional deviation. If its absolute value is less than the set allowable threshold (e.g., 1μm), the verification passes; otherwise, mark the data at that point as suspicious, and you can choose to recalibrate or discard it.

[0057] Finally, the theoretical coordinates (X_theory, Y_theory) of all calibration points P(i,j) and their final position deviation values ​​(ΔX_final, ΔY_final) are recorded to form the original calibration dataset.

[0058] Online compensation phase:

[0059] The online compensation stage of this invention involves real-time compensation during chip bonding, based on the spatial compensation network established in the aforementioned global calibration stage, during actual production. Specifically, it includes the following steps.

[0060] Step S3: Real-time compensation bonding. During the actual bonding process, for any actual bonded chip (bonding element) target bonding coordinate P(X_target, Y_target), the following steps are performed:

[0061] S31: Network query step, based on the target fitting coordinates, determine one or more neighboring nodes in the spatial compensation network that are close to the target fitting coordinates as reference nodes.

[0062] Specifically, the control drive mechanism performs rapid positioning in the stored spatial compensation network based on the target's fitting coordinates (X_target, Y_target). It determines which grid cell contains the target's fitting coordinates and then combines this information with... Figure 1 As shown, the target fitting coordinate P is located within a grid cell consisting of four nodes: P1(x0, y0), P2(x1, y0), P3(x1, y1), and P4(x0, y1).

[0063] S32: Interpolation compensation step, based on the theoretical fitting coordinates and associated position deviation values ​​of the one or more reference nodes, calculates the predicted compensation value corresponding to the target fitting coordinates through a spatial interpolation algorithm;

[0064] In step S32, the spatial interpolation algorithm includes any one of the following methods, and is adaptively selected based on the positional relationship between the target fitting coordinates and the reference node:

[0065] Method A. When the target fitting coordinate P is located inside a grid cell consisting of four reference nodes, a bilinear interpolation algorithm is used.

[0066] Combination Figure 1 As shown, in this first embodiment, the coordinates and deviation values ​​stored in the four vertex nodes P1, P2, P3, and P4 of the grid cell are obtained: the target point P is inside the grid cell, and the deviations in the X and Y directions are interpolated and calculated respectively, as follows: f(x,y)=W1f(x0,y0) + W2f(x1,y0) + W3f(x1,y1) + W4f(x0,y1);

[0067] Where W1 = (x1-x) / (y1-y);

[0068] W2 = (x - x0) / (y1 - y);

[0069] W3 = (y-y0) / (x-x0);

[0070] W4 = (x1-x) / (y-y0);

[0071] Method B. When the distance between the target's fitting coordinates and a certain reference node is less than a preset threshold, the position deviation value of the reference node is directly used as the predicted compensation value.

[0072] Combination Figure 1 As shown in this second embodiment, if the distance between the target point P and a certain network node P1 is found to be less than a preset minimum threshold (e.g., 5 μm), then the point is considered to almost coincide with the calibration point. To simplify the calculation, the deviation value of node P1 is directly used as the prediction compensation value.

[0073] Method C. When the target fitting coordinate P is located at the edge of the working area and only two reference nodes along the same direction can be obtained, a single linear interpolation algorithm is used.

[0074] If, during the process of acquiring the four vertex nodes of a grid cell, incomplete or unidentifiable identification occurs, or if the target point P is located at the edge of the entire working area, its "grid cell" may lack nodes on one or both sides (only having 2 or 3 neighboring nodes). Figure 2 As shown in this embodiment, when only two nodes P1 and P3 along the X direction are available, single linear interpolation (one-dimensional interpolation) is used to calculate the compensation value in the X direction, and the compensation in the Y direction can use the value of the nearest node or be set to zero.

[0075] S33: Motion synthesis step, synthesizing the predicted compensation value with the target fitting coordinates to generate the compensated motion command; that is, adding the calculated predicted compensation value (ΔX_target, ΔY_target) with the original target coordinates (X_target, Y_target) to obtain the compensated motion command coordinates.

[0076] S34: Execution step, controlling the drive mechanism to complete the fitting action according to the compensated motion command.

[0077] Using the above method, the bonding device can utilize a pre-built global "error map" to provide "tailor-made" compensation for each unique bonding position, thereby achieving consistent high-precision bonding throughout the entire working area.

[0078] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of the present invention. All equivalent changes or modifications made to the structure, features and principles described in the claims of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A global calibration and real-time compensation method for high-precision bonding equipment, characterized in that, Includes the following steps: Step S1: Global error calibration data acquisition. Select multiple calibration points within the entire working area of ​​the calibration substrate. For each calibration point, control the drive mechanism to perform the operation of restoring the actual bonding process. For each calibration point, measure the position deviation value of the actual bonding position relative to its theoretical bonding coordinates. Record the theoretical bonding coordinates of each calibration point and the position deviation value measured at that point to form an original calibration dataset covering the working area. Step S2: Spatial compensation network construction. Based on the original calibration dataset, a two-dimensional spatial compensation network is constructed, wherein the theoretical fitting coordinates of each calibration point are used as nodes of the network, and their corresponding position deviation values ​​are stored in association. Step S3: Real-time compensation bonding. During the actual bonding process, for the target bonding coordinates of any bonding element, the following steps are performed: S31: Network query step, based on the target fitting coordinates, determine one or more neighboring nodes that are close to the target fitting coordinates as reference nodes in the spatial compensation network; S32: Interpolation compensation step, based on the theoretical fitting coordinates and associated position deviation values ​​of the one or more reference nodes, calculates the predicted compensation value corresponding to the target fitting coordinates through a spatial interpolation algorithm; S33: Motion synthesis step, synthesizing the predicted compensation value with the target fitting coordinates to generate compensated motion commands; S34: Execution step, controlling the drive mechanism to complete the fitting action according to the compensated motion command.

2. The global calibration and compensation method for a high-precision bonding device according to claim 1, characterized in that, In step S1, the bonding head is controlled to pick up the bonding element from the material picking position and move it to the theoretical bonding coordinates of the bonding element. The position of the bonding element after bonding is measured by the vision positioning unit to obtain the positional deviation measurement value between the theoretical bonding coordinates and the actual bonding coordinates of the bonding element.

3. The global calibration and compensation method for a high-precision bonding device according to claim 2, characterized in that, In step S1, the complete operation is repeated at least N times, and the average value of the N position deviation measurements is calculated, which is then used as the final position deviation value of the calibration point.

4. The global calibration and compensation method for a high-precision bonding device according to claim 2, characterized in that, In step S1, the position of the bonding element after bonding is measured by the visual positioning unit. First, the visual positioning unit performs a first image recognition for coarse positioning. Then, based on the coarse positioning, a second image recognition is performed for fine positioning. Finally, the position deviation value is calculated based on the result of the fine positioning.

5. The global calibration and compensation method for a high-precision bonding device according to claim 1, characterized in that, In step S1, after obtaining the position deviation value of each calibration point, the position deviation value is verified. The verification method is as follows: using the position deviation value as compensation, the drive mechanism is controlled to repeat the bonding operation once and the new position deviation is measured. If the new position deviation exceeds the allowable threshold, the data of the calibration point is marked as invalid or recalibrated.

6. The global calibration and compensation method for a high-precision bonding device according to claim 1, characterized in that, Before step S1 begins, a preheating step is also included: controlling the high-precision bonding equipment drive mechanism to run unloaded for a unit time until its temperature reaches a stable state.

7. A global calibration and compensation method for a high-precision bonding device according to any one of claims 1-6, characterized in that, The multiple calibration points are distributed in a two-dimensional grid pattern, covering the working area.

8. A global calibration and compensation method for a high-precision bonding device according to claim 7, characterized in that, In step S31, the reference node selection method is as follows: determine the grid cell in which the target fitting coordinates are located in the spatial compensation network, and use the multiple vertex nodes constituting the grid cell as the reference nodes.

9. A global calibration and compensation method for a high-precision bonding device according to claim 8, characterized in that, In step S32, the spatial interpolation algorithm includes any one of the following methods, and is adaptively selected based on the positional relationship between the target fitting coordinates and the reference node: A. When the target fitting coordinates are located inside a grid cell consisting of four reference nodes, a bilinear interpolation algorithm is used; B. When the distance between the target's fitting coordinates and a certain reference node is less than a preset threshold, the position deviation value of the reference node is directly used as the predicted compensation value; C. When the target fitting coordinates are located at the edge of the working area and only two reference nodes along the same direction can be obtained, a single linear interpolation algorithm is used.

Images