A method and system for semiconductor package equipment worktable positioning error rectification

Abstract

The application discloses a kind of for semiconductor packaging equipment workbench positioning error rectification method and system, it is related to precision motion control technical field;Load and verify the tool with characteristic array;Collect tool key reference features and solve pose parameters to generate global prediction coordinate point set;Drive vision system to traverse in multiple adaptive search and multiple sampling after synthesis original physical coordinate set according to prediction point set;Original coordinate set is carried out abnormal point identification and reconstruction;Error mapping relationship is established based on cleaned data and smooth compensation model is constructed;Dynamic load compensation model and carry out consistency verification, activate real-time compensation and operating state safety monitoring;Solve the problem that different position positioning is not accurate because of XY guide rail microcosm variation or small error caused by processing and assembly in semiconductor equipment ultra-high precision positioning, realize submicron limit high-precision positioning in whole work area, significantly improve the positioning precision, automated calibration efficiency and motion stability of precision packaging equipment.

Description

Technical Field

[0001] This invention relates to the field of precision motion control technology, specifically to a method and system for correcting positioning errors on a semiconductor packaging equipment stage. Background Technology

[0002] With the rapid development of semiconductor microelectronic packaging technology towards high density and ultra-fine pitch, core precision microelectronic assembly equipment such as automatic wire bonding machines and die bonders have placed extremely stringent requirements on the motion control and positioning accuracy of the XY stage, which has now fully entered the sub-micron era. The positioning accuracy of the stage directly determines the accuracy of chip bonding point placement, process consistency, and the yield of the final packaged product.

[0003] In actual industrial production, regardless of the precision of the XY mechanical worktable, it is inevitably limited by machining tolerances, lead screw pitch errors, guide rail straightness errors, orthogonality deviations between the two axes, as well as mechanical wear and thermal deformation caused by long-term high-frequency operation of the equipment. These physical factors intertwine to result in an extremely complex global nonlinear spatial geometric deviation between the actual physical motion trajectory of the worktable and the theoretical commands issued by the control system throughout the entire two-dimensional working stroke.

[0004] Faced with the aforementioned precision bottlenecks, traditional solutions often limit themselves to continuously improving the machining accuracy of mechanical parts or adopting more advanced motion control hardware platforms. However, this approach, which relies solely on hardware upgrades, not only leads to an exponential increase in equipment manufacturing costs, but also yields negligible marginal benefits in terms of precision improvement under the stringent requirements at the sub-micron level. The fundamental reason is that the errors exhibited by the worktable under actual working conditions are global, nonlinear, and accompanied by highly random spatiotemporal coupling errors. These dynamic errors, by their very nature, cannot be completely eliminated by simply improving the static mechanical or hardware precision.

[0005] Therefore, developing an independent and controllable high-precision "workbench mapping" system and calibration method to accurately compensate for and overcome this complex global nonlinear random error is of great significance for breaking through the precision bottleneck and enhancing the core competitiveness of domestic semiconductor packaging equipment. Summary of the Invention

[0006] This invention proposes a method and system for correcting positioning errors on a semiconductor packaging equipment stage. It aims to overcome the bottleneck of global nonlinear random errors that are common in precision motion platforms, and solve problems such as easy deformation of the reference, easy interference of data acquisition, and mechanical vibration caused by non-smooth compensation in the existing technology, thereby achieving sub-micron level extreme high-precision positioning in the entire working domain.

[0007] The technical solution is as follows: On the one hand, a method for correcting positioning errors on a semiconductor packaging equipment stage is provided, comprising: Step S1: Fix the fixture with the preset feature array onto the bearing surface of the worktable and perform a compatibility check; Step S2: Collect the physical position data of at least one set of key reference features on the fixture in the current motion coordinate system, construct a dynamic motion model, calculate the pose parameters of the fixture relative to the motion axis of the worktable, and generate a global prediction coordinate point set based on the pose parameters; Step S3: Drive the visual imaging system to traverse the path planned by the predicted coordinate point set, and at each predicted node, locate the fixture feature corresponding to the node through multi-level adaptive visual search. After multiple sampling and dynamic stability verification, convert the pixel domain deviation into motion control domain error, and synthesize it with the current predicted coordinates to generate the original physical coordinate dataset. Step S4: Perform outlier identification and reconstruction on the original physical coordinate dataset to obtain cleaned physical location data; Step S5: Establish an error mapping relationship and construct a compensation model based on the cleaned physical location data. If there are no invalid nodes, establish the error mapping relationship and construct the compensation model; if there are invalid nodes, execute the reconstruction algorithm. Step S6: Dynamically load the compensation model and perform consistency verification. If the health assessment is passed, the data is deemed valid and real-time compensation is activated. The coordinates are corrected in real time based on the local neighborhood, and running fault safety monitoring is performed. If the compensation step exceeds the preset motion safety change rate threshold, a safe rollback to the original trajectory mode is executed. If it does not exceed the threshold, the motion compensation operation is completed.

[0008] Optionally, in step S2, the key reference feature includes: The origin datum used to determine the starting position of the grid; A first axial direction reference used to define the direction of the grid rows; A second axial direction reference used to define the direction of the grid column; And a first and a second scale reference used to determine the axial scale of the grid.

[0009] Optionally, in step S2, the method for calculating the pose parameters includes: Calculate the rotational deflection angle of the fixture relative to the worktable's motion axis. : ; in, The origin reference coordinates The first axial direction reference The coordinates; Calculate the anisotropic scaling factor : ; in, It is the Euclidean distance function. , Theoretical grid spacing, , This represents the corresponding step size / span.

[0010] Optionally, in step S2, the method for generating the dynamically predicted coordinate point set is as follows: For the nth matrix line, number Predicted nodes of the column Its coordinates are calculated through the following affine transformation: .

[0011] Optionally, in step S3, the hysteresis suppression includes: When executing motion commands, a convergence direction constraint strategy is adopted to ensure that the final displacement before reaching each predicted node always follows the preset unidirectional vector, so that the mechanical transmission system is in the same stress release state and idle direction at the sampling moment.

[0012] Optionally, in step S3, the multi-level adaptive visual search and localization includes: To predict coordinates Construct a primary search space centered on [the search engine]. Instantaneous observation location for identifying target features ; Calculate the geometric consistency evaluation function between the observed location and the predicted coordinates. If the preset convergence threshold is met Then it shrinks to the second-level search space. Perform sub-pixel level feature centroid reconstruction; If the convergence threshold is not met, the node is marked as a singularity, and the dimensionality reduction convergence to the second-level search space is blocked.

[0013] Optionally, in step S3, the multiple sampling and dynamic stability verification includes: Perform on each grid node Successive exposure sampling Obtain the feature deviation vector for each frame. ; Calculate the statistical mean of the characteristic deviation vector. and standard deviation : ; like Exceeding the preset repeatability threshold If this occurs, a sampling retry will be triggered or the node will be marked as abnormal.

[0014] Optionally, in step S3, the conversion of the actual physical coordinates includes: Using a pre-calibrated camera physical scale matrix The statistical mean of the characteristic deviation vector Error vector converted to the motion control domain : ; Compare the error vector with the current predicted coordinates By combining the coordinates, the true absolute position coordinates of the node can be obtained. : .

[0015] Optionally, in step S4, the anomaly identification and reconstruction includes: Through preset physical security thresholds To retrieve the true absolute position coordinates of a node from the entire dataset... If so, the node is determined to be an invalid node; The invalid nodes are repaired using a neighborhood data reconstruction algorithm to ensure the integrity of the error mapping table and the smoothness of the surface.

[0016] On the other hand, a stage positioning error correction system for semiconductor packaging equipment is provided to implement the above-mentioned stage positioning error correction method for semiconductor packaging equipment, including: The fixture has a preset high-contrast feature array to provide a physical reference; A worktable is used to support and move the fixture or the workpiece to be processed. A vision imaging system, installed above the worktable, is used to acquire images of the fixture features or the workpiece. The control system, electrically connected to the worktable and the vision imaging system, includes: An adaptive mesh construction module is used to perform step S2; An automated sampling control module is used to execute step S3; The data processing and transformation module is used to execute steps S4 and S5 and generate the compensation lookup table. The real-time compensation and monitoring module is used to perform step S6.

[0017] The technical solution includes at least the following technical effects: Adaptive spatial alignment reduces assembly requirements: By collecting limited key reference features, the position parameters (rotation, scaling) of the fixture are automatically calculated, achieving high-precision alignment under non-precision installation conditions, which greatly reduces the dependence on operator experience and installation accuracy.

[0018] Robust signal processing ensures data purity: Through multi-level adaptive search, multiple sampling, and dynamic stability verification, optical noise and mechanical vibration interference in industrial environments are effectively suppressed, guaranteeing data purity from the source. Combined with a built-in data reconstruction algorithm, it possesses data self-healing capabilities.

[0019] Kinematic property optimization enhances operational safety: The generated compensation model, through smoothing optimization, solves the problem of derivative discontinuity at grid boundaries in traditional linear compensation, fundamentally suppressing transient shocks and oscillations, resulting in smoother motion trajectories. Simultaneously, the built-in real-time safety monitoring and fault rollback mechanism provide full lifecycle safety assurance for equipment operation.

[0020] Significantly improved positioning accuracy: Through the aforementioned comprehensive technical means, this invention can effectively eliminate the geometric nonlinear residuals of the precision worktable throughout its entire stroke range. Experiments have shown that the overall positioning error of the system in cold welding mode can be significantly reduced to approximately 1.2 μm, achieving a leap in sub-micron level motion accuracy.

[0021] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit the invention. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0023] Figure 1 This is a flowchart of a method for correcting positioning errors on a semiconductor packaging equipment stage, provided in a preferred embodiment of the present invention. Figure 2 This is an experimental result of closed-loop accuracy verification after the mapping lookup table (LUT) is generated, according to a preferred embodiment of the present invention. Figure 3 shows a set of error heatmaps of the X-axis and Y-axis before and after calibration in the cold welding mode in one embodiment of the present invention, wherein a is the X-axis error heatmap before calibration, b is the Y-axis error heatmap before calibration, c is the X-axis error heatmap after calibration, and d is the Y-axis error heatmap after calibration. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] As attached Figure 1 As shown, this invention provides a method for correcting positioning errors on a semiconductor packaging equipment stage. This method is implemented based on a host computer control system, a vision imaging system, and a Table Mapping fixture mounted on an XY precision worktable, and includes the following steps: Step S1: Table Mapping fixture loading and compatibility verification.

[0027] Select a table mapping fixture that matches the effective working stroke of the current machine tool (for example, for a general heating block machine tool with an effective stroke of 83mm, select a dedicated table mapping fixture with an effective calibration area of ​​not less than 83mm × 83mm). Fix the fixture to the bearing surface of the precision worktable, ensuring that the feature array of the fixture completely covers the preset motion working area of ​​the machine tool.

[0028] Step S2: Adaptive Mesh Model Reconstruction Based on Key Reference Features. Physical position data of at least one set of key reference features on the fixture in the current motion coordinate system are collected. A dynamic motion model is constructed, and the pose parameters of the fixture relative to the worktable's motion axis are calculated. A global prediction coordinate point set is generated based on the pose parameters. The key reference features include: an origin reference for determining the starting position of the mesh; a first axial direction reference for defining the mesh row direction; a second axial direction reference for defining the mesh column direction; and a first scale reference and a second scale reference for determining the axial scale of the mesh.

[0029] This step aims to acquire limited key location information of the Table Mapping fixture in the current physical space to construct a dynamic motion model that can automatically adapt to the actual placement of the fixture, thereby eliminating scan path deviations caused by manual installation errors. The specific process is as follows: Step S201: Acquisition of pose information of key reference features.

[0030] The control system receives the physical position data of at least one set of key reference features on the fixture in the current motion coordinate system. The key feature points preferably include: the origin reference located in the array's initial region. Axial direction references used to define the orientation of grid rows and columns respectively. and and boundary reference pairs used to determine the global boundary envelope. and To further precisely define the independent geometric span of the fixture in different axes, the key reference feature also includes axial dimensional reference pairs for confirming the mesh length and width. and Let the set of core location vectors acquired by the system be: ; The method of obtaining the location data is not limited to human-computer interaction teaching or visual automatic search.

[0031] Step S202: Calculate the spatial transformation relationship.

[0032] The control system, based on the known theoretical geometric specifications of the fixture, uses a spatial attitude analysis operator to calculate the rotational deflection angle of the fixture relative to the motion axis of the worktable. and independent scaling factors on different axes .

[0033] First, the system uses the origin reference. Axial Direction Reference The relative positional relationship is used to determine the rotational deviation of the fixture in the spindle direction. : ; in, The origin reference coordinates The first axial direction reference The coordinates; Secondly, the system utilizes axial dimensional references to... , relative to the origin benchmark The Euclidean distance is used to map and calculate the anisotropic scaling factor that reflects the change of the physical dimensions of the fixture with the environment. and To eliminate dimensional deviations caused by thermal expansion and contraction of materials: ; in, It is the Euclidean distance function. , Theoretical grid spacing, , This corresponds to the step size. Furthermore, through comparison... and The system can pre-extract the non-orthogonal components of the fixture by forming the included angle, thereby transforming complex physical assembly and manufacturing errors into mathematical descriptors with decoupling and compensation capabilities.

[0034] Step S203: Establishment and transformation of adaptive global prediction grid.

[0035] Based on the pose descriptor calculated above By utilizing the control system and multidimensional coordinate mapping operators, the entire domain is generated through deduction. Dynamic prediction coordinate point set of each sampling node For any nth element in the matrix line, number The predicted command position of the target node in the column in the motion control coordinate system is preferably calculated using a two-dimensional affine transformation, as follows: ; Based on the above calculations, the system generates a set of scan path points that are strictly aligned with the actual physical placement of the fixture. This ensures that subsequent visual acquisition processes always occur within the effective field of view of the fixture. The system maps the static theoretical mesh into dynamic path instructions that can automatically follow the real-time status of the fixture. This ensures that in the subsequent fully automated scanning process, the visual imaging system can use the calculated pose parameters as a priori guide to drive the optical center to accurately cover every physical feature of the fixture, thereby establishing a highly reliable automated calibration benchmark in non-precision installation environments.

[0036] Step S3: Automated sampling and hysteresis correction of grid data. The driving vision imaging system traverses the path planned according to the predicted coordinate point set. At each predicted node, the corresponding fixture features are located through multi-level adaptive vision search. After multiple sampling and dynamic stability verification, the pixel domain deviation is converted into motion control domain error and synthesized with the current predicted coordinates to generate the original physical coordinate dataset.

[0037] This step aims to utilize the global prediction coordinate point set generated in step S2. Using the motion benchmark, a raw physical dataset reflecting the true positioning characteristics of the workbench is obtained through closed-loop trajectory execution and adaptive visual perception. This specifically includes the following sub-steps: Step S301: Motion execution and hysteresis suppression based on predicted coordinates. The hysteresis suppression includes: when executing motion commands, adopting a convergence direction constraint strategy to ensure that the final displacement before reaching each predicted node always follows a preset unidirectional vector, so that the mechanical transmission system is in the same stress release state and idle direction at the sampling instant.

[0038] The control system retrieves the dynamic prediction point set containing pose and scale compensation information generated in step S203. As input to the instruction sequence, it drives the actuator to carry the visual imaging device to traverse the path planned by the point set.

[0039] To eliminate mechanical hysteresis errors caused by lead screw backlash, bearing clearance, and transmission chain friction, the system employs a trajectory execution strategy with approach direction constraints when executing motion commands, ensuring that the actuator reaches each predicted node. The final displacement always follows a preset unidirectional vector constraint or polar consistency sequence, ensuring that the mechanical transmission system is in the same stress release state and idle direction at the moment of sampling. During this process, the predicted coordinate point set generated in step S2 acts as the dynamic pre-aiming center of the visual imaging system. By compensating for the geometric offset generated during the installation phase, it ensures that the physical features on the calibration fixture remain stably and accurately within the initial acquisition domain of the optical acquisition system even in a non-precision installation state. This provides a stable spatial benchmark for subsequent sub-pixel level feature extraction.

[0040] S302: Multi-level adaptive feature search and precise localization.

[0041] Because the actuator accumulates geometric errors during long-stroke movements, directly using a high-precision, small-scale window for matching can easily lead to search failure due to feature point offset, stray background, and environmental noise. Therefore, this invention employs an iterative convergence search method, aiming to resolve the physical contradiction between "global capture success rate" in a large-scale field of view and "sub-pixel-level positioning accuracy" in a small-scale field of view through dynamic adjustment of spatial search dimensions.

[0042] The control system uses the predicted coordinates provided in step S2 Construct a first-level search space with the logical center of the current motion coordinate system. The visual alignment module first identifies the instantaneous observation position of the target feature within the primary search space. And utilize the spatial coherence evaluation function Determine the geometric consistency between the observed location and the prediction baseline: ; in, The predicted coordinates are derived from step S2. The instantaneous observation position of the feature identified by the visual imaging system. This is a fidelity operator used to evaluate the degree of matching between physical observations and mathematical predictions. This is the preset system convergence threshold.

[0043] If the above geometric consistency criteria are met, the system will automatically shrink the search dimension to the second-order constraint search space. ,in, Within the secondary search space, the system invokes a high-gain edge operator to perform sub-pixel-level feature centroid reconstruction, achieving sub-micron-level high-precision positioning.

[0044] If the convergence threshold is not met, the node is marked as a singularity, and the dimensionality reduction convergence to the second-level search space is blocked.

[0045] If the spatial coherence score falls below a threshold, the system automatically executes an abnormal branch procedure: First, the system blocks dimensionality reduction convergence to the second-level search space to prevent field-of-view loss when pose uncertainty is high; second, the system activates a sampling validity isolation mechanism, marking the current node as an unsteady singularity to ensure that local observation fluctuations do not contaminate the global dataset. The system continuously monitors the global sampling failure density. To ensure the overall effectiveness of the calibration process: ; in, This represents the percentage of abnormal nodes as calculated. This is a critical density threshold used to determine whether the current environment meets the calibration conditions. If the proportion of abnormal nodes exceeds the threshold, the system will terminate the process and report an error.

[0046] Through the above logic, the present invention ensures that the calibration system can still maintain extremely high data continuity and accuracy under non-precision installation and dynamic environmental interference.

[0047] Step S303: Multiple sampling and dynamic stability verification.

[0048] To filter out photon noise and environmental micro-vibration interference during the instantaneous acquisition of the imaging system, the vision acquisition module performs [further processing] on each grid node after locking the secondary search space. Subsequent exposure sampling The system captures multiple frames of images sequentially over time and extracts the feature deviation vector from each frame. And calculate its statistical mean. As the effective visual offset input for this node: ; in, This represents the number of consecutive exposure samples for a single grid node.

[0049] Meanwhile, the control system calculates the standard deviation of the sampling sequence in real time. To assess the steady-state characteristics of the physical environment: ; Compare this indicator with a preset system repeatability threshold. Perform a comparison. If the sampling standard deviation meets the constraints, the current sampling point data is deemed valid and retained; otherwise... Exceeding the preset repeatability threshold If the threshold is exceeded, the system will determine that there is excessive vibration or optical interference in the current physical environment, and automatically trigger the sampling retry logic of the node or mark it as an abnormal state, thereby ensuring the robustness of the mapping table construction from the data source.

[0050] Step S304: Physical true coordinate conversion and calibration dataset recording After completing multiple sampling, the control system obtains the mean pixel domain deviation derived from the feature search results of multiple frames of images. Subsequently, using the pre-calibrated camera physical scale matrix... The pixel domain mean is mapped to the pulse-level error vector of the motion control domain. .

[0051] ; Subsequently, the system performs a cross-spatial dimension vector synthesis operation, combining the predicted command coordinates provided in step S2. Summing the pulse-level error vector with the actual pulse-level error vector generates the true absolute position coordinates of the grid node in physical space. : ; in, The actual predicted coordinates are generated by step S2. For the pre-calibrated camera physical scale matrix, This is the statistical mean of the feature deviation vector from multi-frame sampling.

[0052] This coordinate point set fully records the actual position of the mechanical actuator under specific commands, including the linear pitch error of the worktable, and also decouples and records the non-orthogonality error and the straightness error of the guide rail, providing a high-precision original input reference for subsequent execution space transformation, mapping table generation and smooth interpolation.

[0053] Step S4: Spatial coordinate transformation and anomaly removal. Anomaly identification and reconstruction are performed on the original physical coordinate dataset to obtain cleaned physical location data.

[0054] This step aims to normalize the dimensions of the original physical coordinate set obtained in step S3 and perform reconstruction and repair on singular data caused by environmental interference or missing visual features during the sampling process. The specific process is as follows: First, collect the true absolute position coordinates. With the corresponding nominal command coordinates One-to-one data binding is performed to establish an error vector matrix covering the entire travel area. Subsequently, the system initiates a singularity identification program based on threshold determination, using a preset physical security threshold. The system performs a search on the entire dataset. The criteria for this search are: ; in, The first one generated by step S3 line, number The original absolute position coordinates of the column grid nodes. The preset physical security threshold is used to determine whether the collected data is distorted due to abnormal interference.

[0055] If a sampling node meets the discrimination criteria, it is determined to be an invalid node, and data repair is performed using a system-preset reconstruction algorithm to ensure the integrity and surface smoothness of the final error mapping table (LUT). Furthermore, to ensure the reliability of the reconstruction process, the system first verifies the validity of the region containing the invalid node. If the region is determined to have connectivity failure, a global error flag is automatically triggered, and the mapping generation process is terminated.

[0056] Through the above reconstruction mechanism, the system achieves mathematical repair of locally missing data without the need for secondary physical sampling, ensuring the continuity and smoothness of the mapping table in the global scope, and laying a clean data foundation for the subsequent generation of a high-precision motion compensation lookup table.

[0057] Step S5: Establishing Error Mapping Relationships and Constructing a Compensation Model. Based on the cleaned physical location data, establish error mapping relationships and construct a compensation model.

[0058] This step aims to transform the aforementioned cleaned physical location data into a compensation model capable of guiding the precise movement of the worktable. Through the integration and optimization of the entire dataset, the accuracy and stability of the compensation commands during actual operation are ensured. The specific process is as follows: Step S501: Logical reorganization and index establishment of sampled data.

[0059] The control system first rearranges the original dataset with time-series association output from step S4. This process transforms discrete observation data points into a matrix index architecture based on a physical grid layout, establishing a topologically consistent connection between the storage mapping relationship of data nodes and their distribution in the actual physical space. This provides a standardized and efficient access interface for real-time data retrieval in the subsequent motion compensation stage.

[0060] Step S502: Analysis and calculation of global installation deviation.

[0061] The system obtains the standardized point matrix dataset output in step S501, the core of which contains the nominal command coordinates of each grid node. Compared with the actual physical coordinates obtained by visual measurement .

[0062] The system establishes a transformation function describing the pose relationship between the calibration fixture and the worktable by performing global geometric fitting on the global point matrix. Subsequently, the system performs error decoupling calculation: the measured position... With the nominal command after pose transformation Vector subtraction is performed to eliminate the linear trend component caused by the tilting or translation of the fixture during installation. The mathematical model is as follows: ; in, For the local geometric residual vector field, For actual measured location, To reflect the transformation function between the fixture's mounting position and the overall geometric trend, These are the nominal command coordinates of the corresponding grid node.

[0063] This step involves calculating the local geometric residual vector field. It eliminates macroscopic installation position deviations and retains only nonlinear microscopic deviations introduced by the intrinsic geometric characteristics of the transmission mechanism (such as screw pitch error and guide rail straightness deviation).

[0064] Step S503: Error data smoothing and motion stability optimization.

[0065] To prevent unnecessary mechanical vibrations during motion compensation, the system processes the residual data generated in step S502 using a numerical optimization operator and generates a smoothed and optimized motion compensation lookup table (LUT).

[0066] This process ensures a smooth gradient transition of the compensation vector between adjacent sampling points, enabling the motion system to have continuous derivative characteristics when executing compensation commands, thereby effectively suppressing the actuator from generating instantaneous acceleration changes or step response shocks.

[0067] Step S504: Data reliability check and security verification.

[0068] Before generating the final compensation model, the system executes a complete quality review procedure. This procedure automatically determines the geometric rationality of the current data by calculating the rate of change and distribution characteristics of the data across the entire domain. If the range of variation between data points (such as local rotation trends or scaling ratios) exceeds the preset safety threshold of the mechanical system, the system will automatically identify it as a "calibration anomaly" and terminate the model's activation.

[0069] If the review is approved, the system will solidify the optimized data into a mapping lookup table. This table supports the generation of smooth positioning corrections in real time for any target coordinates during workbench operation, thereby achieving precise motion control across the entire stroke range.

[0070] Step S6: Real-time model mapping and safety closed-loop verification. The compensation model is dynamically loaded and consistency verification is performed. If the health assessment passes, the data is deemed valid and real-time compensation is activated. The coordinates are corrected in real time based on the local neighborhood, and runtime fault safety monitoring is performed.

[0071] This step aims to formally apply the aforementioned compensation model to production operations, ensuring the reliability of the workbench in the compensated state through real-time data exchange and safety monitoring mechanisms. The specific process is as follows: Step S601: Dynamic loading and consistency verification of the compensation model.

[0072] The control system synchronizes the generated mapping dataset to the motion controller's run register. Before the compensation command is activated, the system calls the built-in safety verification operator to quantitatively evaluate the distribution scale of the global compensation components and determine the physical rationality of the mapping model.

[0073] If the global feature values ​​obtained from the assessment are within the preset safety threshold, the data is deemed valid and the real-time compensation function is activated. If an abnormal change in the feature value is detected or it exceeds the preset threshold, the system automatically executes the compensation isolation strategy, forcibly resetting the pose correction value to the initial zero-point state. This measure aims to eliminate the risk of motion loss of control caused by abnormal calibration environment or data transmission errors, ensuring the operational safety of the actuator.

[0074] Step S602: Real-time coordinate correction based on local neighborhood.

[0075] During equipment production and operation, for any target coordinate command issued by the host computer The system does not execute the instruction directly, but instead calculates the corresponding correction amount in real time through the mapping model. .

[0076] The system will pass the original instructions The generated displacement correction vector is used for vector synthesis to produce the final compensation control command. Its basic logical expression is: ; in, This reflects the intensity of the influence of local mechanical intrinsic errors on the current coordinate point.

[0077] Through this step, the system achieves real-time transparent compensation for nonlinear geometric residuals, ensuring that the actual motion trajectory of the actuator remains highly consistent with the nominal logical coordinate system.

[0078] Step S603, runtime fault safety monitoring.

[0079] During real-time compensation, the system establishes a parallel monitoring logic. This logic continuously monitors the correction amount calculated. If at a certain instant of motion, the calculated compensation step size... Exceeded the system's set dynamic change threshold : ; in, This represents the dynamic variation step of the compensation amount at adjacent sampling times. In order to be in The real-time compensation correction amount calculated by the time system. This is a preset threshold for the rate of change of motion safety.

[0080] The system will immediately identify this as "calculation distortion" or "model failure" and automatically trigger a fail-safe rollback mechanism. At this point, the system will instantly switch the motion mode back to the original, uncompensated trajectory state and output a safety warning to the user. Through this dynamic detection and switching mechanism, this invention ensures that even in the event of a mapping model failure, the equipment can still maintain basic operational safety, avoiding mechanical collisions or workpiece damage.

[0081] This invention provides a positioning error correction system for a semiconductor packaging equipment stage, used to implement the aforementioned positioning error correction method for a semiconductor packaging equipment stage. The system includes: The fixture has a preset high-contrast feature array to provide a physical reference; in this embodiment, a table mapping fixture is selected, which is a calibration medium with specific optical reflection characteristics.

[0082] A worktable is used to support and move the fixture or workpiece to be processed; in this embodiment, an XY precision worktable is used.

[0083] A vision imaging system, installed above the worktable, is used to acquire images of the fixture features or the workpiece. The control system, electrically connected to the worktable and the vision imaging system, includes: An adaptive mesh construction module is used to perform step S2; An automated sampling control module is used to execute step S3; The data processing and transformation module is used to execute steps S4 and S5 and generate the compensation lookup table. The real-time compensation and monitoring module is used to perform step S6.

[0084] Experimental data verification: refer to Figure 2 This embodiment presents an experimental result of closed-loop accuracy verification after the system generates the Map Lookup Table (LUT) in cold welding mode. The chart shows that the nominal command coordinates of the grid point in row 1, column 64 are (-310030.843750, 344722.468750) pulses. Experimental data shows that after enabling real-time compensation, the residual visual deviation captured by the visual imaging system at this point is only 0.119107 pulses in the X-axis and only 0.006380 pulses in the Y-axis. The Xdelta term (107.25 pulses) in the data reveals the physical geometric distortion of the machine in the uncompensated state. However, through the real-time coordinate correction (step S106) described in this invention, this mechanical error of up to 107.25 pulses is successfully offset, reducing the final positioning residual to the sub-pixel level. This strongly demonstrates the extremely high accuracy of the "local contribution weighted algorithm" described in this invention when handling global nonlinear errors.

[0085] Referring to Figure 3, Figure 3 shows a set of error heatmaps for the X-axis and Y-axis before and after calibration in the cold welding mode of an embodiment of the present invention. Figure a shows the X-axis error heatmap before calibration, Figure b shows the Y-axis error heatmap before calibration, Figure c shows the X-axis error heatmap after calibration, and Figure d shows the Y-axis error heatmap after calibration. By comparing the X-axis and Y-axis error heatmaps before and after calibration, it can be seen that without the Table Mapping compensation stage, the worktable exhibits complex multi-dimensional spatial error superposition characteristics throughout its entire stroke. From a macroscopic perspective, the error field shows a significant linear gradient, reflecting the inter-axis orthogonality deviation and rotation angle caused by the mounting posture. From a local detail analysis, the error surface exhibits obvious nonlinear distortion and spatial distortion, reflecting the intrinsic pitch error of the mechanical transmission mechanism and the physical deformation of the guide rail. Specifically, the original X-axis deviation ranges from -7 to +8 μm, and the original Y-axis deviation ranges from -5 to +5 μm. After applying the error calibration algorithm described in this invention, the residual error amplitudes of both the X-axis and Y-axis achieved order-of-magnitude reductions, converging to within ±0.3μm and ±0.2μm respectively. Furthermore, the originally regular error gradient completely disappeared, evolving into uniformly distributed and uncorrelated high-frequency random physical noise. This demonstrates that the proposed solution can accurately isolate macroscopic installation pose components and deeply repair intrinsic geometric residuals. Therefore, this invention, through error compensation of all 4608 nodes in the case study, successfully reduced the overall positioning error of the precision stage to the sub-micron level, achieving extremely high consistency and isotropic positioning accuracy, providing robust accuracy assurance for high-precision motion control scenarios such as semiconductor packaging.

[0086] The primary objective of this invention is to provide a method and system for correcting positioning errors on a semiconductor packaging equipment stage. This method combines a high-contrast optical reference with automated visual calibration. By utilizing a calibration medium with specific optical reflection characteristics (hereinafter referred to as a Table Mapping fixture), along with adaptive mesh reconstruction and closed-loop motion control strategies, this method addresses the problems of insufficient mapping accuracy caused by thermal deformation of the reference object, fixture installation posture deviation, mechanical return hysteresis, and data noise in traditional calibration processes. This significantly improves the positioning accuracy, automated calibration efficiency, and motion stability of precision packaging equipment.

[0087] Another objective of this invention is to provide a highly robust visual feature acquisition mechanism. To address the optical noise interference that may arise from the microstructure of the TableMapping fixture surface, a variable-domain adaptive search algorithm and temporal multi-frame fusion processing are used to effectively suppress nonlinear reflections and sensor noise during the imaging process, achieving sub-pixel-level feature localization and ensuring the purity and accuracy of the mapped data source.

[0088] Furthermore, this invention aims to provide a mapping model construction scheme with adaptive data processing, real-time safety monitoring, and smooth compensation capabilities. By introducing a hysteresis-suppressing approaching motion sequence to eliminate nonlinear errors in the mechanical system, and combining dynamic extreme value monitoring and neighborhood data reconstruction algorithms, singular data during the sampling process is automatically repaired. Simultaneously, a high-order smooth mapping lookup table is generated using a continuous interpolation algorithm to eliminate acceleration abrupt changes at grid boundaries. A fault-safe protection mechanism for the operational state is established, ensuring the dynamic response performance and operational safety of the equipment under high-speed operation through real-time monitoring of the compensation amount and model state verification.

[0089] In practical industrial applications, this invention demonstrates significant technical advantages over traditional algorithms, and its core value is reflected in the following three dimensions: 1. Adaptive spatial alignment and low assembly constraint requirements This solution significantly reduces the physical dependence on the initial installation accuracy of the auxiliary fixture through adaptive geometric inference technology. The system only needs to acquire a limited number of key reference features to automatically identify complex spatial pose parameters (including rotational deviations and anisotropic scaling factors). This means that even under conditions where there is significant fixture misalignment or environmental factors induce dimensional drift, the system can still automatically construct an accurate guide path. This feature achieves "high-precision alignment under non-precision installation," greatly reducing reliance on operator experience and shortening equipment maintenance downtime.

[0090] Robust signal processing and data quality assurance To address the complex industrial imaging environment and mechanical vibration interference, this solution establishes a multi-sampling and adaptive search mechanism. This mechanism effectively suppresses random noise and dynamic interference from the data source through statistical feature extraction, solving the feature convergence problem under long-stroke motion. Combined with the built-in reconstruction algorithm, the system possesses significant data self-healing capabilities, automatically identifying and correcting local abnormal samples to ensure that the dataset input into the mapping model has a sub-pixel-level high signal-to-noise ratio and reliability.

[0091] Kinematic property optimization and real-time operational safety The compensation model produced by this solution possesses optimized spatial continuity, resolving the issue of uneven dynamic response at grid boundaries inherent in traditional linear compensation. By improving the higher-order derivative characteristics of the compensation trajectory, the system fundamentally suppresses transient shocks and oscillations in the actuator during compensation actions, resulting in a smoother table trajectory and significantly enhanced landing stability of precision equipment at high speeds. Simultaneously, the system incorporates a built-in safety closed-loop monitoring mechanism, dynamically evaluating the physical rationality of the compensation vector. Upon detecting model anomalies, the system can achieve millisecond-level safety state switching, providing full lifecycle operational assurance for high-value machines and workpieces.

[0092] By implementing the technical solution described in this invention, the precision stage can achieve sub-pixel-level positioning accuracy with isotropic properties throughout its entire stroke range. Experiments have shown that this algorithm can effectively eliminate the geometric nonlinear residuals of the precision stage throughout its entire stroke range, significantly reducing the overall positioning error of the system to approximately 1.2 μm. This achieves a leap in sub-pixel-level motion accuracy and highly reliable process output, while maintaining excellent dynamic response characteristics, providing solid underlying algorithmic support for the domestic substitution of high-end semiconductor packaging equipment.

[0093] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for correcting positioning errors on a semiconductor packaging equipment stage, characterized in that, include: Step S1: Fix the fixture with the preset feature array onto the bearing surface of the worktable and perform a compatibility check; Step S2: Collect the physical position data of at least one set of key reference features on the fixture in the current motion coordinate system, construct a dynamic motion model, calculate the pose parameters of the fixture relative to the motion axis of the worktable, and generate a dynamic prediction coordinate point set based on the pose parameters; Step S3: Drive the visual imaging system to traverse the path planned by the predicted coordinate point set, and at each predicted node, perform multi-level adaptive visual search and locate the fixture features corresponding to the node. After multiple sampling and dynamic stability verification, convert the pixel domain deviation into motion control domain error, and synthesize it with the current predicted coordinates to generate the original physical coordinate dataset. Step S4: Perform outlier identification and reconstruction on the original physical coordinate dataset to obtain cleaned physical location data; Step S5: Establish an error mapping relationship and construct a compensation model based on the cleaned physical location data; Step S6: Dynamically load the compensation model and perform consistency verification. If the health assessment passes, the data is deemed valid and real-time compensation is activated. The coordinates are corrected in real time based on the local neighborhood, and operational fault safety monitoring is performed.

2. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S2, the key reference features include: The origin datum used to determine the starting position of the grid; A first axial direction reference used to define the direction of the grid rows; A second axial direction reference used to define the direction of the grid column; And a first and a second scale reference used to determine the axial scale of the grid.

3. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 2, characterized in that, In step S2, the method for calculating the pose parameters of the fixture relative to the motion axis of the worktable includes: Calculate the rotational deflection angle of the fixture relative to the worktable's motion axis. : ； in, Origin reference coordinates First axial direction reference The coordinates; Calculate the anisotropic scaling factor : ； in, It is the Euclidean distance function. , Theoretical grid spacing, , This represents the corresponding step size / span.

4. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S2, the method for generating the dynamic prediction coordinate point set is as follows: For the nth matrix line, number Predicted nodes of the column Its coordinates are calculated through the following affine transformation: 。 5. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, Hysteresis suppression is performed in step S3, wherein the hysteresis suppression includes: When executing motion commands, a convergence direction constraint strategy is adopted to ensure that the final displacement before reaching each predicted node always follows the preset unidirectional vector, so that the mechanical transmission system is in the same stress release state and idle direction at the sampling moment.

6. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S3, the multi-level adaptive visual search and localization includes: To predict coordinates Construct a primary search space centered on [the search engine]. Instantaneous observation location for identifying target features ; Calculate the geometric consistency evaluation function between the observed location and the predicted coordinates. If the preset convergence threshold is met Then it shrinks to the second-level search space. Perform sub-pixel level feature centroid reconstruction; If the convergence threshold is not met, the node is marked as a singularity, and the dimensionality reduction convergence to the second-level search space is blocked.

7. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S3, the multiple sampling and dynamic stability verification includes: Perform on each grid node Successive exposure sampling Obtain the feature deviation vector for each frame. ; Calculate the statistical mean of the characteristic deviation vector. and standard deviation : ; like Exceeding the preset repeatability threshold If this occurs, a sampling retry will be triggered or the node will be marked as abnormal.

8. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S3, a real physical coordinate conversion is performed, which includes: Using a pre-calibrated camera physical scale matrix The statistical mean of the characteristic deviation vector Error vector converted to the motion control domain : ; Compare the error vector with the current predicted coordinates By combining the coordinates, the true absolute position coordinates of the node can be obtained. : 。 9. The method for compensating for positioning errors on a semiconductor packaging equipment stage according to claim 1, characterized in that, In step S4, the anomaly identification and reconstruction includes: Through preset physical security thresholds The criteria for retrieving data across the entire domain are as follows: ； in, The first one generated by step S3 line, number The original absolute position coordinates of the column grid nodes. A preset physical security threshold is used to determine whether the collected data is distorted due to abnormal interference; If a sampling node meets the discrimination criterion, then the node is determined to be an invalid node; The invalid nodes are repaired using a neighborhood data reconstruction algorithm.

10. A stage positioning error correction system for semiconductor packaging equipment, used to implement the stage positioning error correction method for semiconductor packaging equipment as described in any one of claims 1 to 9, characterized in that, include: The fixture has a preset high-contrast feature array to provide a physical reference; A worktable is used to support and move the fixture or the workpiece to be processed. A vision imaging system, installed above the worktable, is used to acquire images of the fixture features or the workpiece. The control system, electrically connected to the worktable and the vision imaging system, includes: An adaptive mesh construction module is used to perform step S2; An automated sampling control module is used to execute step S3; The data processing and transformation module is used to execute steps S4 and S5 and generate a compensation lookup table; The real-time compensation and monitoring module is used to perform step S6.

Images