Search

CN-121600074-B - Chip contact pin accurate positioning control optimization method and system based on machine vision

CN121600074BCN 121600074 BCN121600074 BCN 121600074BCN-121600074-B

Abstract

The invention discloses a chip contact pin accurate positioning control optimization method and system based on machine vision, which relate to the technical field of contact pin accurate positioning and comprise the steps of obtaining initial optical coaxial uniformity based on multi-angle brightness distribution consistency analysis, performing neighborhood expansion sampling on the initial optical coaxial uniformity to generate a plurality of optical coaxial uniformity to be screened, mapping the optical coaxial uniformity to a control target group based on an illumination-imaging coupling response curve, respectively applying the mapped optical coaxial uniformity to perform contact pin positioning, obtaining contact pin positioning data, generating uniformity scores based on machine learning training, constructing a performance curve according to the optical coaxial uniformity to be screened and the corresponding uniformity scores, analyzing and screening the optimal optical coaxial uniformity to perform contact pin positioning, and solving the problems of uneven brightness, fuzzy contour and insufficient contact pin positioning accuracy of a BGA solder ball array in imaging.

Inventors

  • JIN QILONG

Assignees

  • 苏州中芯长宏半导体科技有限公司

Dates

Publication Date
20260505
Application Date
20260128

Claims (8)

  1. 1. The chip contact pin accurate positioning control optimization method based on machine vision is characterized by comprising the following steps of: Acquiring initial optical coaxial uniformity based on multi-angle brightness distribution uniformity analysis; Performing neighborhood expansion sampling on the initial optical coaxial uniformity to generate a plurality of optical coaxial uniformity to be screened; Mapping the optical on-axis uniformity to a control target group based on an illumination-imaging coupling response curve, wherein the control target group comprises an on-axis illumination numerical aperture adjustment amount and an optical axis decentration compensation displacement amount; Respectively applying the mapped optical coaxial uniformity to be screened to execute pin positioning; Obtaining pin positioning data and generating uniformity scores based on machine learning training; Constructing a performance curve according to the optical coaxial uniformity to be screened and the corresponding uniformity score, analyzing and screening the optimal optical coaxial uniformity, and positioning the contact pin; the initial optical coaxial uniformity is obtained based on multi-angle brightness distribution consistency analysis, and specifically comprises the following steps: Acquiring brightness distribution images of a light source under a preset angle set, wherein the angle set is a plurality of observation angles acquired by taking an optical axis as a center according to a fixed angle step length; Calculating a radial gradient sequence under a corresponding angle for the brightness distribution image corresponding to each observation angle, and generating a radial gradient sequence set indexed according to the angle; Calculating relative amplitude of the radial gradient sequence corresponding to each angle in the radial gradient sequence set, and generating a radial gradient relative amplitude sequence indexed according to the angle; selecting a relative amplitude maximum value from the radial gradient relative amplitude sequence to generate radial gradient dispersion; Selecting a circumferential brightness value at a preset radius from brightness distribution images corresponding to each observation angle, and generating a circumferential brightness sequence; Calculating the circumferential relative fluctuation amplitude of the circumferential brightness sequence to generate the circumferential brightness fluctuation degree; and analyzing according to the radial gradient dispersion and the circumferential brightness fluctuation degree to obtain the initial optical coaxial uniformity.
  2. 2. The machine vision-based chip pin accurate positioning control optimization method according to claim 1, wherein the initial optical coaxial uniformity is obtained by analyzing radial gradient dispersion and circumferential brightness fluctuation, specifically: converting the radial gradient dispersion and the circumferential brightness fluctuation degree into a brightness deviation amount pair, wherein the brightness deviation amount pair consists of a radial deviation amount corresponding to the radial gradient dispersion and a circumferential deviation amount corresponding to the circumferential brightness fluctuation degree; Generating a brightness deviation characteristic sequence based on the brightness deviation value pairs, wherein the brightness deviation characteristic sequence is formed by arranging radial deviation values and annular deviation values according to a preset sequence; Selecting a component with larger deviation amount from the brightness deviation characteristic sequence as a main deviation amount to obtain the main deviation amount; Substituting the main deviation value into a preset deviation classification section to obtain the classification position of the main deviation value in the deviation classification section; And determining the corresponding uniformity percentage according to the grading position of the main deviation amount as the initial optical coaxial uniformity.
  3. 3. The machine vision-based chip pin accurate positioning control optimization method according to claim 2, wherein the neighborhood expansion sampling of the initial optical coaxiality is performed to generate a plurality of optical coaxiality to be screened, specifically: Determining a neighborhood expansion step length set based on the initial optical coaxiality, wherein the neighborhood expansion step length set comprises a plurality of expansion step lengths which are generated in an arithmetic mode by taking a preset minimum step length as a reference; Performing addition and subtraction operation on the initial optical coaxial uniformity and a positive expansion step length and a negative expansion step length in the neighborhood expansion step length set respectively to generate a candidate uniformity value sequence arranged according to step length indexes; applying a preset uniformity effective interval constraint to the candidate uniformity value sequence, removing candidate uniformity values exceeding the uniformity effective interval, and generating an effective candidate uniformity sequence; checking repeated values of the effective candidate uniformity sequences and sequencing the effective candidate uniformity sequences in order from small to large to generate ordered candidate uniformity sequences; and forming a neighborhood expansion uniformity set by each uniformity value in the ordered candidate uniformity sequence and the initial optical coaxial uniformity, wherein each uniformity value in the neighborhood expansion uniformity set is the optical coaxial uniformity to be screened.
  4. 4. The machine vision based chip pin precise positioning control optimization method according to claim 3, wherein the mapping of the optical coaxial uniformity to the control target group based on the illumination-imaging coupling response curve is specifically as follows: Acquiring a plurality of numerical aperture adjustment amounts from an illumination end as numerical aperture sampling points; Acquiring a plurality of optical axis eccentric compensation displacement amounts from an imaging end as eccentric displacement sampling points; generating an illumination-imaging sampling point set which is arranged according to a two-dimensional index according to the numerical aperture sampling points and the eccentric displacement sampling points; adjusting the illumination numerical aperture and the optical axis decentration compensation displacement under each sampling point in the illumination-imaging sampling point set, and collecting corresponding imaging brightness distribution images to generate a brightness distribution image set indexed according to the sampling points; acquiring radial brightness response values and circumferential brightness response values for each brightness distribution image in a brightness distribution image set, combining the radial brightness response values and the circumferential brightness response values to form coaxial response values, and generating a coaxial response value set indexed according to sampling points; Performing two-dimensional interpolation based on the illumination-imaging sampling point set and the corresponding coaxial response value set to generate an illumination-imaging coupling response curve taking the illumination numerical aperture adjustment quantity and the optical axis decentration compensation displacement quantity as input variables and taking the coaxial response value as output variables; Associating each optical coaxial uniformity to be screened in the neighborhood expansion uniformity set with the coupling response curve, determining a coaxial response value corresponding to the corresponding optical coaxial uniformity, searching an illumination numerical aperture adjustment quantity and an optical axis eccentric compensation displacement quantity which enable the coaxial response value to be consistent with the corresponding optical coaxial uniformity in the coupling response curve, and generating a control pair set indexed according to the uniformity; and combining the illumination numerical aperture adjustment quantity and the optical axis eccentricity compensation displacement quantity in the control pair set to form a control target group.
  5. 5. The optimization method for precise positioning control of chip pins based on machine vision according to claim 4, wherein the performing pin positioning by applying the mapped optical coaxiality to be screened respectively specifically comprises: respectively acquiring an illumination numerical aperture adjustment amount and an optical axis eccentricity compensation displacement amount corresponding to the optical coaxial uniformity to be screened, and adjusting the illumination numerical aperture adjustment amount and the optical axis eccentricity compensation displacement amount to form a positioning light field for corresponding uniformity; performing point-by-point exposure on the full view field area of the array type contact pins under the positioning light field, collecting a corresponding contact pin imaging frame sequence, and generating a contact pin imaging sequence for corresponding optical coaxial uniformity; performing background suppression and brightness normalization processing on the pin imaging sequence to obtain a pin outline image sequence with uniform brightness; Extracting pin boundaries from each pin contour image, performing sub-pixel center fitting according to the boundary shapes to obtain center coordinates of each pin, and forming pin center coordinate sets for corresponding uniformity; The central offset of each contact pin is calculated based on point-to-point correspondence between the contact pin central coordinate set and standard arrangement coordinates of the array contact pins, and a contact pin offset set is generated; And identifying the maximum offset, the average offset and the offset distribution in the pin offset set, and completing positioning.
  6. 6. The optimization method for accurate positioning control of a chip pin based on machine vision according to claim 5, wherein the obtaining pin positioning data and generating a uniformity score based on machine learning training is specifically as follows: extracting a pin center offset characteristic based on a pin offset set, and acquiring a maximum offset, an average offset and an offset variance for representing offset diffusion degree under a low uniformity condition to form a center offset characteristic group; Calculating the contour gradient amplitude value of each pin based on the pin contour image sequence, and counting the minimum value, the maximum value and the gradient amplitude change rate of the gradient amplitude value in the whole view field to obtain a contour gradient feature group for representing contour gray gradient discontinuity caused by low uniformity; measuring the brightness value of a reflection peak area of a solder ball in each pin profile image, and calculating the peak area, the peak overexposure ratio and the adjacent peak brightness difference to form a reflection peak characteristic group for representing the overexposure and the light spot diffusion degree of the reflection area caused by high uniformity; Calculating the contrast between the solder balls based on the brightness difference between adjacent solder balls in the full-view-field pin profile image, and counting the average value of the contrast and the contrast attenuation rate in each image to generate a contrast characteristic group for representing the phenomenon of weakening the solder ball boundary caused by high uniformity; combining the center shift feature set, the contour gradient feature set, the reflection peak feature set and the contrast feature set according to a fixed sequence to form a positioning feature vector for the uniformity; collecting all positioning feature vectors corresponding to the optical coaxial uniformity to be screened, constructing a positioning feature training set, labeling a positioning quality label based on the pin offset stability and the contour fidelity for each feature vector, and generating a positioning quality label set; Based on the positioning feature training set and the positioning quality label set training regression model, the weighting coefficient of each positioning feature in the uniformity score is determined through training, a machine learning model for calculating the optical coaxial uniformity score is obtained, the machine learning model is in a weighted form of all features, the training object is the weight value of each feature, and the trained machine learning model outputs the uniformity score.
  7. 7. The machine vision-based chip pin accurate positioning control optimization method according to claim 6, wherein the method is characterized in that according to the optical coaxial uniformity to be screened and the corresponding uniformity score, a performance curve is constructed, and the optimal optical coaxial uniformity is analyzed and screened for pin positioning, specifically: Sequencing all the optical coaxial uniformity to be screened according to the numerical value to form a uniformity sequence which is arranged according to the uniformity increment; according to the sequence of the uniformity sequences, corresponding uniformity scores are arranged according to the same sequence to form a scoring sequence; constructing a performance curve for representing the relationship between uniformity variation and grading variation by taking the uniformity sequence as a horizontal axis and the grading sequence as a vertical axis; Calculating adjacent scoring differences in the performance curve, identifying the maximum gain position of the scoring differences, and determining the uniformity corresponding to the position as candidate optimal uniformity; Respectively reading corresponding uniformity scores from the former uniformity and the latter uniformity near the candidate optimal uniformity to form candidate interval evaluation groups; comparing the scoring sizes in the candidate interval scoring groups, and selecting the uniformity with the highest scoring as the final optimal optical coaxial uniformity; and inputting the final optimal optical coaxial uniformity into an illumination-imaging coupling response curve, obtaining a corresponding illumination numerical aperture adjustment amount and an optical axis decentration compensation displacement amount, generating an optimal positioning control target, and positioning the chip contact pin.
  8. 8. A system using the machine vision-based chip pin accurate positioning control optimization method of any one of claims 1-7, comprising an optical coaxial uniformity initialization module, a neighborhood extension sampling module, an optical coaxial uniformity control mapping module, a pin positioning module, a uniformity scoring module, and an optical coaxial uniformity screening module; The optical coaxial uniformity initializing module is used for acquiring initial optical coaxial uniformity based on multi-angle brightness distribution uniformity analysis; the neighborhood expansion sampling module is used for carrying out neighborhood expansion sampling on the initial optical coaxial uniformity to generate a plurality of optical coaxial uniformity to be screened; The optical coaxial uniformity control mapping module is used for mapping the optical coaxial uniformity to a control target group based on an illumination-imaging coupling response curve, wherein the control target group comprises coaxial illumination numerical aperture adjustment quantity and optical axis eccentricity compensation displacement quantity; The pin positioning module is used for respectively applying the mapped optical coaxial uniformity to be screened to perform pin positioning; The uniformity scoring module is used for acquiring the pin positioning data and generating uniformity scores based on machine learning training; And the optical coaxial uniformity screening module is used for constructing a performance curve according to the optical coaxial uniformity to be screened and the corresponding uniformity score, analyzing and screening the optimal optical coaxial uniformity and positioning the contact pin.

Description

Chip contact pin accurate positioning control optimization method and system based on machine vision Technical Field The invention relates to the technical field of accurate positioning of pins, in particular to a chip pin accurate positioning control optimization method and system based on machine vision. Background As the chip packaging form rapidly evolves from traditional DIP, QFP and other metal pin type packages to high density BGA array packages, the bottom of the chip is no longer provided with pluggable metal pins, but rather, micro solder balls which are regularly arranged are adopted as contacts. Therefore, in the surface mounting and bonding process, precise array recognition, geometric positioning and alignment control with the substrate pads are required to be performed on the BGA ball array through machine vision, and this process is regarded as "array pin precise positioning" in industry. The existing accurate positioning method for the chip contact pins based on machine vision generally relies on the steps of bottom surface solder ball development imaging, solder ball grid structure extraction, solder ball center positioning, array registration and the like, so that accurate alignment between the chip contacts and the bonding pads is completed. In the machine vision positioning method for accurately positioning the array type contact pins, optical imaging is a bottom layer core link, wherein the optical coaxial uniformity can influence the brightness consistency of coaxial illumination in the whole imaging view field, and the development definition of the bottom surface solder balls, the reflection characteristic stability and whether the edge outline of the solder balls are completely presented are directly determined. Because the surface of the BGA solder ball has high reflectivity and is in a spherical mirror structure, if the coaxiality is insufficient, bright spots, dark areas or local overexposure can occur in a solder ball reflection area, so that the accuracy of the follow-up steps of solder ball contour extraction, sub-pixel center positioning, array direction identification and the like is obviously reduced. The higher the optical coaxiality is, the better the lower the optical coaxiality is, but there is an optimal interval for BGA solder ball development. When the coaxial uniformity is lower, the brightness difference between the edge of the solder ball and the reflective central area is enlarged, so that the judgment of the visual model on the geometric characteristics of the solder ball is unstable, and when the coaxial uniformity is higher, the reflective channel on the surface of the solder ball is enhanced, the problem of over-exposure or over-smoothing of the reflective peak distribution can occur, and the resolution of the central point and the outline characteristics of the solder ball is reduced. This means that the control of different uniformity can cause different types of positioning errors, which together restrict the positioning accuracy of the machine vision pin. Specifically, when the optical coaxiality is low, the imaging view field can generate uneven brightness or local illumination attenuation, so that the solder balls at the edge of the view field or the array offset area can not obtain stable reflection bright spots and clear contours. In this case, the gray gradient of the solder ball edge is discontinuous, which causes deviation between edge detection and spherical contour fitting, and the sub-pixel center extraction result of the low-brightness solder ball is obviously drifted, so that systematic errors exist in the direction angle of the whole array, the center gravity of the array and the measurement of the distance between the solder balls. Finally, the chip is easy to generate alignment deviation, solder ball dislocation or local warping when aligning with the bonding pad, and the pin accuracy is reduced. When the optical coaxiality is too high, a large-area highlight area is formed by strong reflection on the surface of the solder ball in imaging, so that the 'light spot diffusion' effect is generated between the real outline and the reflection peak distribution of the solder ball, and the problems of edge blurring, overexposure of the reflection center, weakening of the boundary between the solder balls and the like are caused. At this time, the profile fitting of the solder balls is distorted, the diameter recognition is made smaller or larger, and even adjacent solder balls are erroneously determined as connected areas. In addition, the overall brightness of the array tends to be consistent due to the excessively high uniformity, so that the sensitivity of a vision system to local morphological differences (such as decentration, deformation, ball missing and the like) is reduced, and the defect identification and the pin stability are affected. In the prior art, however, a fixed illumination parameter or a simple light intensity increasing/decre