A probe card needle mark detection method and system based on needle mark photos

Abstract

The present invention discloses a probe card needle mark detection method and system based on needle mark photos, and relates to the field of semiconductor detection technology. The probe card needle mark detection method based on needle mark photos includes the following steps: using a camera to collect needle mark images on pads; identifying needle mark features from the needle mark images, and evaluating the contact quality between the probe and the pad based on the identification results of the needle mark features; marking probes with abnormal contact according to the screening results; using an optimization algorithm to iteratively correct the contact position of the marked probe and the solder joint, generate the best contact point between the probe and the pad, and repeat the probe detection to verify the optimization effect. The present invention uses wavelet transform to extract nonlinear features in needle mark images and combines a time series prediction model to accurately identify and predict abnormal features in the image, and combines data slicing technology to pre-separate abnormal features from normal features, which not only improves the processing speed, but also increases the precision of processing data.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor detection technology, and in particular to a probe card needle mark detection method and system based on needle mark photos. Background Art

[0002] The probe card is a test interface mainly used for bare core testing. The probe on the probe card directly contacts the pad or bump on the chip to bring out the chip signal. This process is coordinated with peripheral test instruments and software control to achieve automated measurement. Therefore, in IC manufacturing, the probe card has a great impact on manufacturing costs and is an important process link.

[0003] In the probe card needle mark detection method, it is impossible to use the optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint and generate the best contact point for repeated testing, which may result in the contact position between the probe and the pad being not optimal, thereby increasing the contact resistance and reducing the signal transmission efficiency, affecting the accuracy and reliability of the overall test. In the absence of optimization algorithm support, the position of the probe may need to be manually adjusted to find the best contact point. This manual adjustment is not only time-consuming and labor-intensive, but may also increase equipment wear and tear, leading to higher maintenance and replacement costs.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies in the prior art, the present invention provides a probe card needle mark detection method and system based on needle mark photos, which has the advantage of accurately and efficiently adjusting the contact position between the probe and the solder pad, thereby solving the problem of poor contact position between the probe and the solder pad.

[0007] (II) Technical solution

[0008] In order to achieve the above advantages of accurately and efficiently adjusting the contact position between the probe and the pad, the specific technical solution adopted by the present invention is as follows:

[0009] According to another aspect of the present invention, a probe card needle mark detection method based on a needle mark photo is provided, and the probe card needle mark detection method based on a needle mark photo comprises the following steps:

[0010] S1, contacting the probe on the probe card with the pad on the chip to form a needle mark on the pad, and using a camera to capture an image of the needle mark on the pad;

[0011] S2, identifying the needle mark feature from the needle mark image, and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature;

[0012] S3, screening out abnormal data from the optimized needle mark image based on the contact quality evaluation result, and marking the probes with abnormal contact according to the screening result;

[0013] S4. Use the optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect.

[0014] Preferably, identifying the needle mark feature from the needle mark image, and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature comprises the following steps:

[0015] S21, using a filter to remove the needle mark image noise, and performing contrast adjustment and grayscale image conversion on the denoised needle mark image in sequence to obtain an optimized needle mark image;

[0016] S22, establishing a needle mark recognition model to identify and optimize needle mark features in the needle mark image;

[0017] S23. Combine the needle mark features into a feature vector, classify the feature vector using a pre-trained learning model, and evaluate the contact quality between the probe and the pad based on the classification result.

[0018] Preferably, establishing a needle mark recognition model to identify and optimize needle mark features in a needle mark image comprises the following steps:

[0019] S221, extracting color information of the needle mark and the background in the optimized needle mark image, and establishing databases on two color planes based on the color information;

[0020] S222, establishing Bayesian probability distribution maps in two color planes respectively, and normalizing the probability values ​​in the Bayesian probability distribution maps to obtain a needle mark distribution map;

[0021] S223, selecting a preset threshold in the Bayesian probability distribution map to classify the needle mark distribution map into two planar binary distribution maps;

[0022] S224, using a pre-built query table to perform needle mark texture detection on the plane binary distribution map to obtain a focus threshold;

[0023] S225. Use the focus threshold as the input of the needle mark recognition model, and use the needle mark recognition model to output the needle mark feature.

[0024] Preferably, screening out abnormal data from the optimized needle mark image based on the contact quality evaluation result, and marking the probe with abnormal contact according to the screening result comprises the following steps:

[0025] S31, extracting nonlinear features in the optimized needle mark image using wavelet transform, and predicting abnormal features in the optimized needle mark image using a pre-built time series prediction model based on the nonlinear features;

[0026] S32, using data slicing technology to pre-separate abnormal features from normal features, and segment the abnormal features into several potential abnormal areas;

[0027] S33. Train the confidence encoder using the confidence distribution rule to detect abnormal data points in potential abnormal areas, and mark the probes identified as abnormal contact points.

[0028] Preferably, using the confidence distribution rule to train the confidence encoder, detecting abnormal data points in the potential abnormal area, and marking the probes identified as abnormal contact points include the following steps:

[0029] S331, taking the abnormal features in the optimized needle mark image as a training set, and using the Mahalanobis distance to measure the distance between the training sample and the training set, and simplifying the distance between the training sample and the training set;

[0030] S332, generating a Mahalanobis distance array based on the training sample and the training set, taking the average value of the Mahalanobis distance array as a reduction threshold, and extending the confidence encoder based on the reduction threshold;

[0031] S333, converting the data in the confidence encoder into a confidence distribution form corresponding to the input and output in the confidence rule;

[0032] S334, adaptively adjusting the confidence rule weights in the confidence encoder based on the confidence distribution form, and using the adjusted confidence encoder to detect and mark abnormal data points in the potential abnormal area.

[0033] Preferably, adaptively adjusting the confidence rule weights in the confidence encoder based on the confidence distribution form, and using the adjusted confidence encoder to detect and mark abnormal data points in the potential abnormal area includes the following steps:

[0034] S3441, respectively calculating the confidence distribution of each abnormal data point in the potential abnormal area, and calculating the individual matching degree of the abnormal data point based on the confidence distribution of each abnormal data point;

[0035] S3442, obtaining activation weights of confidence rules of several abnormal data points based on the individual matching degrees of the confidence rules, and merging all confidence rules whose activation weights of the confidence rules are greater than a preset value to obtain the distributed confidence of the abnormal data points;

[0036] S3443. Use the distributed confidence of the abnormal data point as input data of the confidence encoder, and output the detection result of the abnormal data point in the potential abnormal area through the confidence encoder.

[0037] Preferably, using an optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint, generating the best contact point between the probe and the solder pad, and repeatedly performing probe detection to verify the optimization effect includes the following steps:

[0038] S41, setting optimization goals and constraints based on the labeled probe with abnormal contact, and iteratively updating the contact position parameters between the labeled probe and the pad using an optimization algorithm until a preset optimization goal is reached;

[0039] S42, selecting the best contact point from the preset optimization target, and repeatedly executing steps S1 to S3 according to the best contact point to obtain the iteratively updated labeling probe detection result;

[0040] S43. Compare the probe detection results before and after the iterative update to verify the optimization effect, and generate a probe detection report.

[0041] Preferably, setting optimization goals and constraints based on the labeled probe with abnormal contact, and iteratively updating the contact position parameters between the labeled probe and the pad using an optimization algorithm until the preset optimization goal is reached includes the following steps:

[0042] S411, setting optimization goals and constraints for the labeled probe with abnormal contact, and using Newton iteration algorithm to establish a corresponding relationship between the contact position between the labeled probe and the pad, and obtaining a correction factor for the position between the labeled probe and the pad;

[0043] S412, providing a dynamic search area for the improved whale optimization algorithm based on the correction factor of the position between the marked probe and the pad, and calculating the error weight between the marked probe and the pad as a judgment basis;

[0044] S413. Based on the judgment reference, the contact position parameters between the labeled probe and the solder pad are iteratively solved using the improved whale optimization algorithm to obtain the optimal parameters that satisfy the contact position between the labeled probe and the solder pad.

[0045] Preferably, the calculation formula for the individual matching degree of abnormal data points is:

[0046] ;

[0047] In the formula, Indicates the individual matching degree of the abnormal data point; k Indicates the number of confidence rules; Represents an abnormal data point of the input; i Indicates the number of abnormal data points; Indicates the confidence level of the abnormal data point; represents the confidence of the abnormal data point in the kth confidence rule; J represents the confidence distribution form of the abnormal data point.

[0048] Preferably, the calculation formula of the error weight between the probe and the pad is:

[0049] ;

[0050] In the formula, w represents the error weight between the probe and the pad; n Indicates the number of contacts between the probe and the pad; Indicates the contact data volume between the probe and the pad; Indicates the total amount of contact data between the probe and the pad; Represents the weight parameter of the improved whale optimization algorithm.

[0051] According to another aspect of the present invention, there is also provided a probe card needle mark detection system based on needle mark photos, the probe card needle mark detection system based on needle mark photos comprises a needle mark image acquisition module, a contact quality assessment module, an abnormality marking module and a detection optimization module;

[0052] The needle mark image acquisition module is sequentially connected with the contact quality assessment module, the abnormality marking module and the detection optimization module;

[0053] A needle mark image acquisition module is used to contact the probe on the probe card with the pad on the chip to form a needle mark on the pad, and to collect the needle mark image on the pad by using a camera device;

[0054] A contact quality evaluation module, used for identifying needle mark features from the needle mark image, and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark features;

[0055] An abnormal marking module is used to filter out abnormal data from the optimized needle mark image based on the contact quality evaluation result, and mark the probe with abnormal contact according to the screening result;

[0056] The detection optimization module is used to iteratively correct the contact position between the marked probe and the solder joint using the optimization algorithm, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect.

[0057] (III) Beneficial effects

[0058] Compared with the prior art, the present invention provides a probe card needle mark detection method and system based on needle mark photos, which has the following beneficial effects:

[0059] (1) The present invention optimizes the needle mark image by using image processing technology to more clearly show the details of the needle mark, so that the texture features of the needle mark can be more accurately identified, thereby improving the assessment accuracy of the contact quality between the probe and the pad, and establishing a needle mark recognition model to enhance the automation and standardization of the detection process, thereby not only reducing human errors and improving detection efficiency, but also enabling the contact quality between the probe and the pad to be accurately measured.

[0060] (2) The present invention utilizes wavelet transform to extract nonlinear features in needle mark images and combines it with a time series prediction model to accurately identify and predict abnormal features in the image. It also combines data slicing technology to pre-separate abnormal features from normal features, which not only improves the processing speed but also increases the precision of processed data. At the same time, by constructing a confidence encoder for detection, fault identification is performed based on the confidence distribution and matching degree of each abnormal data point, thereby greatly improving the accuracy and reliability of probe detection.

[0061] (3) The present invention uses the Newton iteration algorithm and the whale optimization algorithm to accurately and efficiently adjust the contact position between the probe and the pad, continuously verify and compare the contact effect before and after the iterative update, ensure the effectiveness of the optimization measures, and thus increase the credibility of the optimization results. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 is a flow chart of a probe card needle mark detection method based on a needle mark photo according to an embodiment of the present invention;

[0064] Figure 2 4 is a principle block diagram of a probe card needle mark detection system based on needle mark photos according to an embodiment of the present invention.

[0065] In the figure:

[0066] 1. Needle mark image acquisition module; 2. Contact quality assessment module; 3. Abnormality marking module; 4. Detection optimization module. DETAILED DESCRIPTION

[0067] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0068] According to an embodiment of the present invention, a probe card needle mark detection method and system based on needle mark photos are provided.

[0069] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the probe card needle mark detection method based on the needle mark photo according to an embodiment of the present invention, the probe card needle mark detection method based on the needle mark photo includes the following steps:

[0070] S1. The probes on the probe card are brought into contact with the pads on the chip to form needle marks on the pads, and an image of the needle marks on the pads is collected using a camera.

[0071] It should be noted that the steps of bringing the probe on the probe card into contact with the pad on the chip to form a needle mark on the pad, and collecting the needle mark image on the pad by using a camera device include:

[0072] Ensure that the probe card is aligned with the chip and lightly touch the probe to the pad, applying appropriate pressure to form a clear needle mark; use a camera (such as a high-resolution camera) to capture the needle mark image on the pad, and ensure that the image is clear so that the needle mark characteristics can be analyzed in detail.

[0073] S2. Identify needle mark features from the needle mark image, and evaluate the contact quality between the probe and the pad based on the identification result of the needle mark features.

[0074] The steps of identifying the needle mark feature from the needle mark image and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature include the following steps:

[0075] S21, using a filter to remove the needle mark image noise, and performing contrast adjustment and grayscale image conversion on the denoised needle mark image in sequence to obtain an optimized needle mark image;

[0076] S22. Establish a needle mark recognition model to identify and optimize needle mark features in the needle mark image.

[0077] Among them, establishing a needle mark recognition model to identify and optimize needle mark features in the needle mark image includes the following steps:

[0078] S221, extracting color information of the needle mark and the background in the optimized needle mark image, and establishing databases on two color planes based on the color information;

[0079] S222. Establishing Bayesian probability distribution maps in two color planes respectively, and normalizing the probability values ​​in the Bayesian probability distribution maps to obtain a needle mark distribution map.

[0080] It should be noted that the Bayesian probability distribution diagrams are established in the two color planes respectively, and the probability values ​​in the Bayesian probability distribution diagrams are normalized to obtain the needle mark distribution diagrams including:

[0081] Select a color plane (such as the R and G planes in the RGB color model, or the hue and saturation planes in the HSV color model), and the selected color plane should be able to clearly reflect the color difference between the needle mark and the surrounding area; segment the area containing the needle mark from the entire image based on color, brightness or other image features; use the Bayesian statistical method to establish a probability distribution model on the selected color plane, wherein the use of the Bayesian statistical method to establish the probability distribution model includes: collecting color data from the needle mark and non-needle mark areas as training samples; using the Bayesian method to estimate the conditional probability of the existence of the needle mark on the training data; normalizing the probability values ​​in the Bayesian probability distribution map to represent the probability in the range of 0 to 1; finally converting the normalized probability values ​​into an image form, and the brightness or color of each pixel represents the probability of the existence of the needle mark at that position, thereby obtaining a needle mark distribution map of the entire pad area.

[0082] S223, selecting a preset threshold value (the preset threshold value is in the range of 0-1) in the Bayesian probability distribution map to classify the needle mark distribution map into two planar binary distribution maps;

[0083] S224, using a pre-built query table to perform needle mark texture detection on the plane binary distribution map to obtain a focus threshold.

[0084] It should be noted that the focus threshold obtained by performing needle mark texture detection on the plane binary distribution map using the pre-built query table includes:

[0085] The lookup table contains a table of output results corresponding to specific input values. In the context of needle mark texture detection, the lookup table usually includes the mapping relationship between needle mark features and their corresponding focus thresholds. The steps to construct the lookup table include:

[0086] Extract key edge strength, texture directionality, and local contrast from a series of training images; analyze the relationship between these features and needle mark quality (such as focus threshold); and fill the lookup table based on the analysis results to contain a direct mapping from features to focus thresholds.

[0087] The binary distribution map converts the image into a map containing only two colors (usually black and white). This conversion simplifies the complexity of the image and makes the texture features easier to analyze. Once the binary distribution map is generated, a lookup table is used to perform needle mark texture detection, and the corresponding focus threshold is found in the lookup table based on the extracted needle mark features.

[0088] S225. Use the focus threshold as the input of the needle mark recognition model, and use the needle mark recognition model to output the needle mark feature.

[0089] It should be noted that the focus threshold is used as the input of the needle mark recognition model, and the needle mark recognition model is used to output the needle mark features. The needle mark recognition model is mainly:

[0090] Convolutional Neural Networks (CNN): Due to their strong performance in image recognition, CNN is particularly suitable for automatically identifying and classifying needle mark features from images. Models can be trained to recognize different types of needle mark features, such as shape, size, and texture.

[0091] Support Vector Machine (SVM): For simpler classification tasks, SVM can be effectively used for binary classification problems, such as distinguishing between needle mark and non-needle mark areas.

[0092] Random Forest: This ensemble learning method is suitable for processing data with a large number of features and can provide an assessment of the importance of needle mark features, thereby optimizing model performance.

[0093] S23. Combine the needle mark features into a feature vector, classify the feature vector using a pre-trained learning model, and evaluate the contact quality between the probe and the pad based on the classification result.

[0094] S3. Based on the contact quality evaluation results, abnormal data is screened out from the optimized needle mark image, and the probes with abnormal contact are marked according to the screening results.

[0095] The steps of screening out abnormal data from the optimized needle mark image based on the contact quality evaluation result and marking the probe with abnormal contact according to the screening result include the following steps:

[0096] S31. Using wavelet transform to extract nonlinear features in the optimized needle mark image, and based on the nonlinear features, using a pre-built time series prediction model to predict abnormal features in the optimized needle mark image.

[0097] It should be noted that the nonlinear features in the optimized needle mark image are extracted by wavelet transform, and the abnormal features in the optimized needle mark image are predicted by using a pre-built time series prediction model based on the nonlinear features, including:

[0098] Wavelet transform feature extraction: select appropriate wavelet function (such as Daubechies, Haar, etc.) according to the characteristics of the image; perform multi-level wavelet decomposition on the needle mark image, so that the details and texture information of the image can be captured at different resolution levels; extract key features from each level of wavelet transform, such as energy distribution, edge information, etc., which can characterize the nonlinear behavior in the image.

[0099] Build a time series prediction model: The time series prediction model is a model that predicts future behavior based on historical data and is applied to anomaly detection in images. Use historical images and known anomalies to establish a training data set. Select a time series prediction model from existing technologies, such as ARIMA, Hidden Markov Model (HMM) or Recurrent Neural Network (RNN), and use the extracted wavelet features to train the time series prediction model. Finally, evaluate the performance and accuracy of the model through methods such as cross-validation.

[0100] Abnormal feature prediction: Once the model is trained, it is used to predict abnormal features in new images and analyze the prediction results to determine whether abnormal features exist in the image, as well as the likelihood and severity of these features.

[0101] S32. Use data slicing technology to pre-separate abnormal features from normal features, and divide the abnormal features into several potential abnormal areas.

[0102] It should be noted that the use of data slicing technology to pre-separate abnormal features from normal features includes: data slicing technology is used to divide image data into multiple small blocks or layers, each layer or block contains complete feature information, which is automatically performed by setting specific thresholds or using machine learning algorithms; based on known abnormal feature definitions, data slices that may contain anomalies are detected and marked, and data that is not marked as abnormal is further verified to confirm that it is normal data.

[0103] S33. Train the confidence encoder using the confidence distribution rule to detect abnormal data points in potential abnormal areas, and mark the probes identified as abnormal contact points.

[0104] Among them, using the confidence distribution rule to train the confidence encoder, detecting abnormal data points in the potential abnormal area, and marking the probes identified as abnormal contact points include the following steps:

[0105] S331, taking the abnormal features in the optimized needle mark image as a training set, and using the Mahalanobis distance to measure the distance between the training sample and the training set, and simplifying the distance between the training sample and the training set;

[0106] S332, generating a Mahalanobis distance array based on the training sample and the training set, taking the average value of the Mahalanobis distance array as a reduction threshold, and extending the confidence encoder based on the reduction threshold;

[0107] S333, converting the data in the confidence encoder into a confidence distribution form corresponding to the input and output in the confidence rule;

[0108] S334, adaptively adjusting the confidence rule weights in the confidence encoder based on the confidence distribution form, and using the adjusted confidence encoder to detect and mark abnormal data points in the potential abnormal area.

[0109] It should be noted that using the confidence distribution rule to train the confidence encoder, detecting abnormal data points in potential abnormal areas, and marking the probes identified as abnormal contact points include:

[0110] Establishment of training set and calculation of Mahalanobis distance: extract abnormal feature data from the optimized needle mark image as the training set; use Mahalanobis distance to measure the distance between the training sample and the entire training set. Mahalanobis distance takes into account the covariance of the data, is highly sensitive to data distribution, and is suitable for similarity measurement in feature space; by analyzing the distribution of Mahalanobis distance, eliminate abnormal samples that are significantly different from most training data to purify the training set.

[0111] Extension of the confidence encoder: organize the Mahalanobis distances calculated in step S331 into an array; calculate the average value of the Mahalanobis distance array and set it as a reduction threshold to determine which data points are normal; adjust the parameters of the confidence encoder based on the reduction threshold to better identify and process data points close to the threshold.

[0112] Convert the data in the confidence encoder into a confidence distribution form corresponding to the input and output. This involves establishing a probability model for each input feature and output result, which reflects the confidence relationship between the input and output; forming a set of rules based on the confidence distribution form, and adaptively adjusting the weights of each confidence rule in the confidence encoder based on the confidence distribution form; using the adjusted confidence encoder to detect data points in potential abnormal areas and identify abnormal data points; marking the detected abnormal data points for further analysis or direct feedback to the production control system.

[0113] The following steps are included: adaptively adjusting the confidence rule weights in the confidence encoder based on the confidence distribution form, and using the adjusted confidence encoder to detect and mark abnormal data points in the potential abnormal area:

[0114] S3441. Calculate the confidence distribution of each abnormal data point in the potential abnormal area respectively, and calculate the individual matching degree of the abnormal data point based on the confidence distribution of each abnormal data point.

[0115] Among them, the calculation formula for the individual matching degree of abnormal data points is:

[0116] ;

[0117] In the formula, Indicates the individual matching degree of the abnormal data point;k Indicates the number of confidence rules; Represents an abnormal data point of the input; i Indicates the number of abnormal data points; Indicates the confidence level of the abnormal data point; represents the confidence of the abnormal data point in the kth confidence rule; J represents the confidence distribution form of the abnormal data point.

[0118] S3442. Based on the individual matching degree of the confidence rules, the activation weights of the confidence rules of several abnormal data points are obtained, and all confidence rules whose activation weights of the confidence rules are greater than the preset value are merged to obtain the distributed confidence of the abnormal data points.

[0119] It should be noted that the individual matching degree of the confidence rule is calculated:

[0120] In this step, each data point first needs to be evaluated to see how well it matches the conditions in the defined confidence rules. This usually involves calculating the similarity or distance between the characteristics of the data point and the conditions of the rule; individual matches can be calculated using a variety of metrics, such as Euclidean distance, cosine similarity, or other methods that are more suitable for the specific characteristics of the data.

[0121] Activate the corresponding confidence rules:

[0122] Based on the matching degree calculated in the previous step, each rule is assigned an activation weight, which reflects the degree to which the data point matches the rule; only when the activation weight of the rule exceeds the preset threshold, the rule is considered valid and enters the next processing flow.

[0123] Fusion activated confidence rules:

[0124] All rules whose activation weights exceed the preset threshold are fused. The fusion can be a simple weighted average or a more complex statistical method, such as weighted voting, fuzzy logic synthesis, etc. The purpose of fusion is to comprehensively consider the information of all relevant rules to obtain an overall confidence assessment, which will represent the possibility that the data point is abnormal; each data point will have a distributed confidence that comprehensively considers multiple rules. This value indicates the degree of credibility of the data point being considered abnormal.

[0125] It should be noted that by integrating the judgments of multiple rules, the bias that may be introduced by relying on a single rule is reduced, making the results of anomaly detection more reliable and robust; the confidence rule system can dynamically adjust the rule weights according to real-time data, making the system more adaptable to emerging abnormal patterns; distributed confidence can more accurately point out the degree of abnormality of data points, helping decision makers to carry out more targeted processing; when monitoring and maintenance resources are limited, high-confidence abnormal data points can be processed first, thereby optimizing resource utilization efficiency.

[0126] S3443. Use the distributed confidence of the abnormal data point as input data of the confidence encoder, and output the detection result of the abnormal data point in the potential abnormal area through the confidence encoder.

[0127] S4, using the optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect;

[0128] Among them, using the optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint, generating the best contact point between the probe and the solder pad, and repeatedly performing the probe test to verify the optimization effect includes the following steps:

[0129] S41. Setting optimization goals and constraints based on the labeled probe with abnormal contact, and iteratively updating the contact position parameters between the labeled probe and the pad using an optimization algorithm until a preset optimization goal is reached.

[0130] The following steps are included, wherein the optimization target and constraint conditions are set based on the labeled probe with abnormal contact, and the contact position parameters between the labeled probe and the pad are iteratively updated using the optimization algorithm until the preset optimization target is reached:

[0131] S411. Setting optimization goals and constraints for the labeled probe with abnormal contact, and using Newton's iteration algorithm to establish a corresponding relationship between the contact position between the labeled probe and the pad, and obtaining a correction factor for the position between the labeled probe and the pad.

[0132] It should be noted that the optimization objectives and constraints are set:

[0133] Optimization goal: usually to minimize the contact error between the probe and the pad, to ensure that the alignment of the probe and the pad is optimal, thereby improving the accuracy and reliability of the test.

[0134] Constraints: These may include the physical limitations of the probe (e.g. range of motion, possible tilt angles, etc.), the operating parameters of the equipment (e.g. pressure, speed, etc.), and safety standards.

[0135] Apply Newton's iterative algorithm:

[0136] The Newton iteration algorithm is a numerical method commonly used in engineering and scientific computing to find the zeros of a real-valued function. In this scenario, the algorithm is used to iteratively find the probe position that minimizes the contact error.

[0137] In the specific implementation, an error function about the probe position is first defined, and then the probe position is continuously adjusted using the Newton iteration method until the error reaches an acceptable minimum value; the position correction factor is obtained based on the deviation between the optimal probe position calculated during the Newton iteration process and the original design position. These correction factors can be used to adjust the probe position so that it can be directly used in future operations, thereby reducing setup time and increasing operation accuracy.

[0138] It should be noted that by accurately calculating and adjusting the contact position between the probe and the pad, the accuracy of measurement and testing can be significantly improved; failures and problems caused by position errors can be reduced, and the service life of the equipment can be extended; by reducing repeated adjustments and tests caused by errors, the overall production efficiency can be improved; the generated position correction factor can be used as a basis for future design improvements, providing data support for more optimized equipment design.

[0139] S412. A correction factor based on the position between the labeled probe and the pad is used to provide a dynamic search area for the whale optimization algorithm, and an error weight between the labeled probe and the pad is calculated as a judgment basis.

[0140] The calculation formula of the error weight between the probe and the pad is:

[0141] ;

[0142] In the formula, w represents the error weight between the probe and the pad; n Indicates the number of contacts between the probe and the pad; Indicates the contact data volume between the probe and the pad; Indicates the total amount of contact data between the probe and the pad; Represents the weight parameter of the whale optimization algorithm.

[0143] S413. Based on the judgment reference, the whale optimization algorithm is used to iteratively solve the contact position parameters between the labeled probe and the pad to obtain the optimal parameters that meet the contact position between the labeled probe and the pad.

[0144] It should be noted that, based on the judgment benchmark, the whale optimization algorithm is used to iteratively solve the contact position parameters of the labeled probe and the pad to obtain the optimal parameters that meet the contact position of the labeled probe and the pad, including:

[0145] Determine the parameters that need to be optimized, such as the distance, angle, and pressure between the probe and the pad; set an objective function, usually minimizing the contact error or maximizing the contact efficiency, as well as corresponding constraints, such as the physical limitations of the equipment and operational safety standards; randomly generate a set of initial solutions in the parameter space, each solution represents the position of a whale, that is, a specific set of probe position parameters; use the iterative rule of the whale optimization algorithm to update the position of each whale. This process includes simulating the social behavior of whales, such as surrounding prey, catching fish with bubble nets, and other strategies to approach the current optimal solution; compare and update the position of each whale in each iteration, and replace the old solution if the new position provides a better solution; set the number of iterations or the objective function to reach a certain threshold as the stopping condition of the algorithm. Once these conditions are met, the algorithm terminates and outputs the current optimal parameter combination.

[0146] It should be noted that by optimizing the contact position between the probe and the pad, the operating accuracy of the equipment can be significantly improved, ensuring the accuracy of measurement and testing; optimizing parameters reduces the time required for equipment adjustment and testing, and improves the smoothness and efficiency of the production line; precise position adjustment can reduce unnecessary friction and pressure between the probe and the pad, extending the service life of the equipment.

[0147] S42, selecting the best contact point from the preset optimization target, and repeatedly executing steps S1 to S3 according to the best contact point to obtain the iteratively updated labeling probe detection result;

[0148] S43. Compare the probe detection results before and after the iterative update to verify the optimization effect, and generate a probe detection report.

[0149] like Figure 2 As shown, according to another embodiment of the present invention, a probe card needle mark detection system based on needle mark photos is also provided, and the probe card needle mark detection system based on needle mark photos includes a needle mark image acquisition module 1, a contact quality assessment module 2, an abnormality marking module 3 and a detection optimization module 4;

[0150] The needle mark image acquisition module 1 is connected with the contact quality assessment module 2, the abnormality marking module 3 and the detection optimization module 4 in sequence;

[0151] The needle mark image acquisition module 1 is used to contact the probe on the probe card with the bonding pad on the chip to form a needle mark on the bonding pad, and to collect the needle mark image on the bonding pad by using a camera device;

[0152] A contact quality evaluation module 2, used for identifying the needle mark feature from the needle mark image, and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature;

[0153] The abnormal marking module 3 is used to screen abnormal data from the optimized needle mark image based on the contact quality evaluation result, and mark the probe with abnormal contact according to the screening result;

[0154] The detection optimization module 4 is used to iteratively correct the contact position between the marked probe and the solder joint using an optimization algorithm, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect.

[0155] In summary, with the help of the above-mentioned technical scheme of the present invention, the present invention optimizes the needle mark image by using image processing technology to more clearly show the details of the needle mark, so that the texture features of the needle mark can be more accurately identified, thereby improving the evaluation accuracy of the contact quality between the probe and the pad, and establishing a needle mark recognition model to improve the automation and standardization of the detection process, thereby not only reducing human errors, but also improving the detection efficiency, so that the contact quality between the probe and the pad can be accurately measured; the present invention uses wavelet transform to extract nonlinear features in the needle mark image and combines the time series prediction model to accurately identify and predict abnormal features in the image, and combines data slicing technology to pre-separate abnormal features from normal features, which not only improves the processing speed, but also increases the precision of processing data. At the same time, by constructing a confidence encoder for detection, fault identification is performed based on the confidence distribution and matching degree of each abnormal data point, thereby greatly improving the accuracy and reliability of probe detection; the present invention uses the Newton iterative algorithm and the whale optimization algorithm to accurately and efficiently adjust the contact position between the probe and the pad, continuously verify and compare the contact effects before and after iterative updates, ensure the effectiveness of the optimization measures, and thereby increase the credibility of the optimization results.

[0156] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A probe card needle mark detection method based on needle mark photos, characterized in that: The probe card needle mark detection method based on needle mark photos comprises the following steps: S1, contacting the probe on the probe card with the pad on the chip to form a needle mark on the pad, and using a camera to capture an image of the needle mark on the pad; S2, identifying the needle mark feature from the needle mark image, and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature; S3, screening out abnormal data from the optimized needle mark image based on the contact quality evaluation result, and marking the probes with abnormal contact according to the screening result; S4, using the optimization algorithm to iteratively correct the contact position between the marked probe and the solder joint, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect; The method of identifying the needle mark feature from the needle mark image and evaluating the contact quality between the probe and the pad based on the identification result of the needle mark feature comprises the following steps: S21, using a filter to remove the needle mark image noise, and performing contrast adjustment and grayscale image conversion on the denoised needle mark image in sequence to obtain an optimized needle mark image; S22, establishing a needle mark recognition model to identify and optimize needle mark features in the needle mark image; S23, combining the needle mark features into a feature vector, classifying the feature vector using a pre-trained learning model, and evaluating the contact quality between the probe and the pad based on the classification result; The method of iteratively correcting the contact position between the marked probe and the solder joint by using the optimization algorithm to generate the best contact point between the probe and the solder pad, and repeatedly performing the probe test to verify the optimization effect includes the following steps: S41, setting optimization goals and constraints based on the labeled probe with abnormal contact, and iteratively updating the contact position parameters between the labeled probe and the pad using an optimization algorithm until a preset optimization goal is reached; S42, selecting the best contact point from the preset optimization target, and repeatedly executing steps S1 to S3 according to the best contact point to obtain the iteratively updated labeling probe detection result; S43, comparing the probe detection results before and after the iterative update to verify the optimization effect, and generating a probe detection report; The step of setting the optimization target and constraint conditions based on the labeled probe with abnormal contact, and iteratively updating the contact position parameters between the labeled probe and the pad using the optimization algorithm until the preset optimization target is reached includes the following steps: S411, setting optimization goals and constraints for the labeled probe with abnormal contact, and using Newton iteration algorithm to establish a corresponding relationship between the contact position between the labeled probe and the pad, and obtaining a correction factor for the position between the labeled probe and the pad; S412, providing a dynamic search area for the improved whale optimization algorithm based on the correction factor of the position between the marked probe and the pad, and calculating the error weight between the marked probe and the pad as a judgment basis; S413. Based on the judgment reference, the contact position parameters between the labeled probe and the solder pad are iteratively solved using the improved whale optimization algorithm to obtain the optimal parameters that satisfy the contact position between the labeled probe and the solder pad.

2. The probe card needle mark detection method based on needle mark photos according to claim 1, characterized in that: The method of establishing a needle mark recognition model to identify and optimize needle mark features in a needle mark image comprises the following steps: S221, extracting color information of the needle mark and the background in the optimized needle mark image, and establishing databases on two color planes based on the color information; S222, establishing Bayesian probability distribution maps in two color planes respectively, and normalizing the probability values ​​in the Bayesian probability distribution maps to obtain a needle mark distribution map; S223, selecting a preset threshold in the Bayesian probability distribution map to classify the needle mark distribution map into two planar binary distribution maps; S224, using a pre-built query table to perform needle mark texture detection on the plane binary distribution map to obtain a focus threshold; S225. Use the focus threshold as the input of the needle mark recognition model, and use the needle mark recognition model to output the needle mark feature.

3. The probe card needle mark detection method based on needle mark photos according to claim 2, characterized in that: The method of screening out abnormal data from the optimized needle mark image based on the contact quality evaluation result and marking the probe with abnormal contact according to the screening result includes the following steps: S31, extracting nonlinear features in the optimized needle mark image using wavelet transform, and predicting abnormal features in the optimized needle mark image using a pre-built time series prediction model based on the nonlinear features; S32, using data slicing technology to pre-separate abnormal features from normal features, and segment the abnormal features into several potential abnormal areas; S33. Train the confidence encoder using the confidence distribution rule to detect abnormal data points in potential abnormal areas, and mark the probes identified as abnormal contact points.

4. The probe card needle mark detection method based on needle mark photos according to claim 3, characterized in that: The method of training a confidence encoder using a confidence distribution rule, detecting abnormal data points in a potential abnormal area, and marking probes identified as abnormal contact points includes the following steps: S331, taking the abnormal features in the optimized needle mark image as a training set, and using the Mahalanobis distance to measure the distance between the training sample and the training set, and simplifying the distance between the training sample and the training set; S332, generating a Mahalanobis distance array based on the training sample and the training set, taking the average value of the Mahalanobis distance array as a reduction threshold, and extending the confidence encoder based on the reduction threshold; S333, converting the data in the confidence encoder into a confidence distribution form corresponding to the input and output in the confidence rule; S334, adaptively adjusting the confidence rule weights in the confidence encoder based on the confidence distribution form, and using the adjusted confidence encoder to detect and mark abnormal data points in the potential abnormal area.

5. The method for detecting probe mark of a probe card based on a needle mark photograph according to claim 4, characterized in that: The method of adaptively adjusting the confidence rule weight in the confidence encoder based on the confidence distribution form and detecting and marking abnormal data points in the potential abnormal area using the adjusted confidence encoder includes the following steps: S3441, respectively calculating the confidence distribution of each abnormal data point in the potential abnormal area, and calculating the individual matching degree of the abnormal data point based on the confidence distribution of each abnormal data point; S3442, obtaining activation weights of confidence rules of several abnormal data points based on the individual matching degrees of the confidence rules, and merging all confidence rules whose activation weights of the confidence rules are greater than a preset value to obtain the distributed confidence of the abnormal data points; S3443. Use the distributed confidence of the abnormal data point as input data of the confidence encoder, and output the detection result of the abnormal data point in the potential abnormal area through the confidence encoder.

6. The method for detecting probe mark of a probe card based on a probe mark photograph according to claim 5, characterized in that: The calculation formula of the error weight between the probe and the pad is: ； In the formula, w represents the error weight between the probe and the pad; n Indicates the number of contacts between the probe and the pad; Indicates the contact data volume between the probe and the pad; Indicates the total amount of contact data between the probe and the pad; Represents the weight parameter of the improved whale optimization algorithm.

7. A probe card needle mark detection system based on needle mark photos, used to implement the probe card needle mark detection method based on needle mark photos according to any one of claims 1 to 6, characterized in that: The probe card needle mark detection system based on needle mark photos includes a needle mark image acquisition module, a contact quality assessment module, an abnormality marking module and a detection optimization module; The needle mark image acquisition module is sequentially connected to the contact quality assessment module, the abnormality marking module and the detection optimization module; The needle mark image acquisition module is used to contact the probe on the probe card with the pad on the chip to form a needle mark on the pad, and use a camera to collect the needle mark image on the pad; The contact quality evaluation module is used to identify the needle mark feature from the needle mark image, and evaluate the contact quality between the probe and the pad based on the identification result of the needle mark feature; The abnormal marking module is used to filter out abnormal data from the optimized needle mark image based on the contact quality evaluation result, and mark the probe with abnormal contact according to the screening result; The detection optimization module is used to iteratively correct the contact position between the marked probe and the solder joint using an optimization algorithm, generate the best contact point between the probe and the solder pad, and repeat the probe detection to verify the optimization effect.

Images