A method and apparatus for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process.

By acquiring defect data and hole design data of TGV wafers, process risk areas are generated, interference characteristics are calculated, and a failure probability model is constructed. This solves the problem of quantitatively assessing the impact of wafer defects on subsequent processes and final product failures, thereby improving TGV manufacturing yield and reducing production costs.

CN123118451APending Publication Date: 2026-11-17ZHONGJIA MICROVISION (SHENZHEN) SEMICONDUCTOR TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611062047.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot quantitatively assess the impact of defects in through-glass via (TGV) wafers on subsequent processes and final product failures, leading to problems such as delayed response, reliance on experience, and waste of resources.

Method used

By acquiring defect data and hole location design data of TGV blank wafers, process risk areas are generated, interference characteristics are calculated, and a failure probability model is constructed to output hole location risk scores and engineering treatment suggestions.

Benefits of technology

This study enabled the quantitative modeling of the relationship between defects in white film and subsequent failures, thereby improving the manufacturing yield of TGV and reducing production costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118451A_ABST
    Figure CN123118451A_ABST
Patent Text Reader

Abstract

This application discloses a method and apparatus for modeling the quantitative relationship between front-end defects and back-end failures in TGV process technology, relating to the field of TGV defect detection technology. This method acquires defect regions and severity characteristics of various defect types during the AOI stage of TGV wafer fabrication, overlays these characteristics with laser-induced risk areas, hole risk areas, or process-sensitive areas corresponding to CAD hole positions, and calculates interference area, interference ratio, center distance, direction weight, defect severity, and process parameter combinations. These characteristics are then linked to subsequent laser-induced results, etching hole results, filling results, electrical testing results, and reliability results to establish a quantitative relationship between defect interference characteristics and the final failure probability. This solution addresses the problem that wafer defects do not necessarily directly lead to subsequent failures and that quantification based on human experience is difficult, providing a calculable basis for hole position avoidance, re-judgment triggering, continued processing, scrap judgment, and recipe adjustment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of automated optical inspection technology for glass through-hole processes, and more specifically, to a method and apparatus for modeling the quantitative relationship between defects in the front end and failures in the back end of the TGV process. Background Technology

[0002] In the through-glass via (TGV) process, defects such as bubbles, foreign objects, scratches, and cracks may exist in the TGV blank stage. These defects do not necessarily lead to subsequent failures, but when the defects are located in the laser-induced region, the via location risk region, or the region sensitive to subsequent circuitry, they may increase the probability of blind vias, rough via walls, cracks, filled voids, high resistance, or open circuits.

[0003] Current AOI (Automated Optical Inspection) systems typically output wafer defects as independent inspection results. However, the results of subsequent laser induction, etching, filling, and electrical testing are scattered across different workstations. There is a lack of unified coordinate overlay and interference quantization between the defect area and the CAD hole location, forcing engineers to rely on experience to determine whether a wafer defect is fatal. This approach suffers from delayed response, reliance on experience, and inconsistent handling. Furthermore, not every wafer defect will lead to subsequent defects. Simply classifying all wafer defects as high-risk results in unnecessary scrapping; conversely, ignoring wafer defects may lead to wasted resources in subsequent laser, etching, and filling processes. Therefore, a method is needed to quantitatively model the relationship between the degree of defect interference and the probability of eventual failure. Summary of the Invention

[0004] The main objective of this application is to provide a method and apparatus for modeling the quantitative relationship between defects in the front end and failures in the TGV process, so as to solve the problem in the prior art that it is impossible to quantitatively assess the impact of wafer defects on subsequent processes and final product failures.

[0005] To achieve the above objectives, the first aspect of this application proposes a method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process, including: Obtain defect data of TGV white film and hole position design data corresponding to the TGV white film. The defect data includes at least one defect area of ​​white film and its defect severity characteristics. Generate process risk area corresponding to each hole position based on the hole position design data. After mapping the defective area of ​​the blank film and the process risk area to the same coordinate system, the interference features between the defective area of ​​the blank film and the process risk area are calculated, and the interference features include at least the interference area; The interference features are then linked to the laser-induced results, etching results, filling results, electrical test results, and reliability results of each hole in subsequent work stations to generate a final failure label. A failure probability model is constructed and trained based on the interference features, defect severity features, process parameter features, and the final failure label. Based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model, the system outputs a hole location risk score and engineering treatment suggestions.

[0006] In some implementations, defect data of TGV wafers and corresponding via design data are acquired. The defect data includes at least one defective region of the wafer and its defect severity characteristics. A process risk region corresponding to each via is generated based on the via design data, including: Defect detection is performed on TGV blank glass using an automated optical inspection device to obtain AOI images and defect detection results. The defect detection results include at least one defect area of ​​the blank glass, its defect category, and defect severity features. The defect severity features include defect area, maximum length, minimum width, aspect ratio, grayscale contrast, edge sharpness, texture roughness, transparency variation, depth estimate, coordinates, and defect category confidence. The defect category includes at least one of the following: bubbles, foreign matter, scratches, edge chipping, contamination, microcracks, inclusions, thickness abnormalities, or visible abnormalities inside the transparent glass. When the defect is a bubble, calculate the bubble's grayscale gradient, edge closure, and transparency changes; when the defect is a crack, calculate the crack's length, direction, number of branches, and endpoint coordinates; when the defect is a foreign object, calculate the foreign object's area, shape factor, and occlusion strength. The system reads the hole design data corresponding to the TGV wafer and generates process risk areas based on the hole center, designed hole diameter, laser-induced spot diameter, etching extension allowance, hole position tolerance, equipment positioning error, and safety margin. These process risk areas are divided into core lethal zones, transitional risk zones, and safety observation zones. Simultaneously, it outputs hole-level risk, region-level risk, and whole-wafer-level risk. The hole-level risk is used for hole position avoidance or reassessment, the region-level risk is used for process parameter adjustment, and the whole-wafer-level risk is used for determining whether to continue processing or scrap the wafer.

[0007] In some embodiments, after mapping the defective region of the unfinished wafer and the process risk region to the same coordinate system, the interference features between the defective region of the unfinished wafer and the process risk region are calculated. The interference features include at least the interference area, including: After transforming the white film defect area and the process risk area to the same global coordinate system, the interference area Area(D∩R) is calculated, where D is the white film defect area and R is the process risk area. The interference ratio ρ is calculated using ρ=Area(D∩R) / Area(R) or ρ=Area(D∩R) / Area(D), where Area(D∩R) is the interference area, Area(R) is the area of ​​the process risk region, and Area(D) is the area of ​​the blank film defect region. Calculate the crossover ratio, the distance from the defect center to the hole center, the angle between the defect long axis direction and the hole arrangement direction, the minimum distance from the defect to the risk zone boundary, and the number of holes covered by the defect. The crossover ratio is IoU=Area(D∩R) / Area(D∪R), where Area(D∪R) is the total area after the regions are superimposed.

[0008] In some implementations, the final failure label includes at least one of blind hole, abnormal hole diameter, hole wall crack, excessive hole wall roughness, abnormal hole taper, filled void, high resistance, open circuit, short circuit, reliability failure, or final scrap.

[0009] In some embodiments, the step of binding the interference features with the laser-induced results, etching results, filling results, electrical test results, and reliability results of each hole position in subsequent workstations to generate a final failure label further includes: The samples are labeled and corrected. When the defects in the blank film are inconsistent in space, discontinuous in time, or the confidence of the coordinates of the subsequent equipment is lower than the threshold, the training weight of the sample is reduced or the sample is marked as an uncertain sample.

[0010] In some implementations, constructing and training a failure probability model based on the interference features, defect severity features, process parameter features, and the final failure label includes: Construct a failure probability model P(Y=1|X,Z), where Y is the final failure label, Y=1 indicates the final failure of the hole position, X is the interference feature and defect severity feature, and Z is the process parameter feature. The process parameter feature includes at least two of the following: laser energy, pulse width, scanning speed, focal position, etching time, etchant concentration, etching temperature, conductive paste viscosity, filling pressure, vacuum time, sintering temperature, equipment number, recipe version, and batch information. The failure probability model is any one of Logistic regression, Bayesian model, random forest, gradient boosting tree, support vector machine, neural network, graph model, or rule-model fusion model. Experiments were grouped according to defect type, defect area, distance between defect and hole, interference ratio, laser recipe, etching recipe, and filling recipe. The subsequent process results of each group were collected as the training set for the model, and the model parameters were updated.

[0011] In some implementations, the step of outputting a hole location risk score and engineering treatment suggestions based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model includes: Based on the failure probability learning risk threshold output by the failure probability model, if the risk score is lower than the first threshold, it is recommended to continue processing; if the risk score is between the first and second thresholds, it is recommended to sample and re-judge or re-inspect the camera for re-judgment; if the risk score is between the second and third thresholds, it is recommended to avoid the hole position or adjust the recipe; if the risk score is higher than the third threshold, it is recommended to suspend processing, rework, scrap, or manually approve. Based on coordinate confidence, defect detection confidence, defect severity, and process rule threshold adjustment engineering handling suggestions, the engineering handling suggestions include at least one of the following: continue processing, sampling re-judgment, re-inspection camera re-judgment, multi-focal plane re-judgment, hole position avoidance, recipe adjustment, suspension of processing, rework, scrapping, or manual approval.

[0012] In some embodiments, the method further includes: The actual back-end results are continuously written back to the model training set, and the failure probability model is incrementally updated. The model performance is monitored by device, batch, product model and Recipe version. If the model's predicted failure probability deviates from the actual failure rate under a certain device or Recipe version, it is determined that there is model drift or process change. When the model drift exceeds the threshold, prompts will be output to re-experiment, recalibrate or supplement the sample collection task, restrict automatic scrapping and automatic Recipe adjustment.

[0013] Secondly, a quantitative relationship modeling device for the relationship between front-end defects and back-end failures in the TGV process is proposed, including: The generation module is used to acquire defect data of TGV wafers and hole position design data corresponding to the TGV wafers. The defect data includes at least one defect area of ​​the wafer and its defect severity characteristics, and generates a process risk area corresponding to each hole position based on the hole position design data. The calculation module is used to map the defective region of the unfinished film and the process risk region to the same coordinate system, and then calculate the interference features between the defective region of the unfinished film and the process risk region. The interference features include at least the interference area. The binding module is used to bind the interference features with the laser-induced results, etching results, filling results, electrical test results and reliability results of each hole position in subsequent work stations to generate the final failure label; The modeling module is used to construct and train a failure probability model based on the interference features, defect severity features, process parameter features, and the final failure label. The disposal decision module is used to output hole location risk scores and engineering disposal suggestions based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model.

[0014] Thirdly, an electronic device is proposed, comprising: At least one processor; and The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0015] Fourthly, a computer-readable storage medium is proposed, on which a computer program is stored, which, when executed by a processor, implements the method described in the first aspect.

[0016] The technical solutions provided by the embodiments of this application may include the following beneficial effects: By establishing a complete data link between blank wafer defect spatial information, process risk areas, and subsequent process failure results, a quantifiable, interpretable, and implementable failure probability prediction model is established. This provides the production end with quantifiable hole location risk scores and engineering handling basis, ultimately improving TGV manufacturing yield and reducing production costs. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart illustrating a quantitative relationship modeling method between front-end defects and back-end failures in the TGV process provided in this application; Figure 2 A schematic diagram of a device for modeling the quantitative relationship between front-end defects and back-end failures in a TGV process, provided in this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0021] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0022] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linked," and "socketing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0023] AOI inspection of unfinished glass substrates (glass substrates that have not undergone subsequent processes such as laser drilling, etching, and filling) can only identify surface defects and cannot quantitatively correlate defects with subsequent process failures. This leads to either missing risky holes, causing subsequent process scrap, or misjudging qualified products, resulting in wasted production capacity. The embodiments disclosed in this paper aim to establish a complete data chain from unfinished glass substrate defect detection, process risk mapping, to subsequent process results, to build a quantitative failure probability prediction model, and to achieve accurate risk scoring for single holes and actionable engineering solutions.

[0024] Figure 1 A flowchart illustrating a method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process, provided in this application, is shown below. The method includes: S110. Obtain defect data of TGV wafers and hole design data corresponding to the TGV wafers. The defect data includes at least one defect area of ​​the wafers and its defect severity characteristics. Generate a process risk area corresponding to each hole based on the hole design data.

[0025] Specifically, the entire TGV film is scanned using AOI (Automated Optical Inspection) equipment. After outputting the AOI image, industrial defect detection algorithms, such as YOLO instance segmentation and U-Net semantic segmentation, are used to process the image, outputting a complete set of attributes for all defects, namely the defect areas of the film, their defect categories, and defect severity features. Defect areas can be approximate regions such as binary masks, polygonal outlines, minimum bounding rectangles, or ellipses. Defect severity features include defect area, maximum length, minimum width, aspect ratio, grayscale contrast, edge sharpness, texture roughness, transparency variation, depth estimate, coordinates, and defect category confidence. Defect categories include bubbles, foreign objects, scratches, edge chipping, contamination, microcracks, inclusions, thickness anomalies, or visible anomalies within the transparent glass.

[0026] When the defect is a bubble, the grayscale gradient, edge closure and transparency changes of the bubble can be further calculated; when the defect is a crack, the length, direction, number of branches and endpoint coordinates of the crack can be further calculated; when the defect is a foreign object, the area, shape factor and occlusion intensity of the foreign object can be further calculated.

[0027] Simultaneously, the hole position design data corresponding to the TGV wafer is read, and a process risk area is generated based on the hole center, designed hole diameter, laser-induced spot diameter, etching extension allowance, hole position tolerance, equipment positioning error, and safety margin. The risk area is not necessarily equivalent to the designed hole diameter, but rather represents the spatial range of influence that defects may have on the laser-induced, etching-based hole formation, or filling effect.

[0028] In one implementation, the process risk area is divided into a core lethal zone, a transitional risk zone, and a safety observation zone. The core lethal zone characterizes the area where defects, once they enter, have a high probability of affecting hole formation. The transitional risk zone characterizes the area where defects may affect hole formation but require judgment in conjunction with process parameters. The safety observation zone characterizes the area where defects have a lower impact on hole formation but still need to be recorded. For example, three concentric regions are generated outward from the hole center. The core lethal zone R1 is the innermost region, corresponding to the range where defects directly cover the hole center or the laser-induced main energy region. The transitional risk zone R2 is the middle layer, corresponding to the range where defects may affect the hole wall through stress, thermal effects, or etching expansion. The safety observation zone R3 is the outermost region, corresponding to the range where the current impact of defects is low but still needs to be recorded. Hole-position level risk, region-level risk, and whole-wafer level risk can be output simultaneously. Hole-position level risk is used for hole avoidance or re-judgment, region-level risk is used for process parameter adjustment, and whole-wafer level risk is used for continuing processing or scrapping judgment.

[0029] S120. After mapping the defective area of ​​the blank film and the process risk area to the same coordinate system, calculate the interference features between the defective area of ​​the blank film and the process risk area, wherein the interference features include at least the interference area.

[0030] In one embodiment of this disclosure, the AOI coordinates of the defective area of ​​the white film and the CAD coordinates of the process risk area are first unified to the same global coordinate system by calibrating with a reference mark. The coordinate alignment accuracy is controlled within 1μm to avoid deviations in subsequent interference area calculations.

[0031] In this embodiment, the interference area is the overlapping area of ​​the white film defect region and the process risk region, used to characterize the degree of direct interference from the defect. The interference area is A_int = Area(D∩R), where D is the white film defect region and R is the process risk region. When D and R have no intersection, A_int is zero; when D covers the hole center or the core dead zone, A_int and the corresponding partition weights are significantly increased.

[0032] Further calculate the interference ratio ρ using ρ=Area(D∩R) / Area(R) or ρ=Area(D∩R) / Area(D), where Area(D∩R) is the interference area, Area(R) is the area of ​​the process risk region, and Area(D) is the area of ​​the blank defect region. Different normalization methods are selected based on the defect type or hole position type. Calculate the intersection-union ratio IoU=Area(D∩R) / Area(D∪R), where Area(D∪R) is the total area after region superposition. Calculate the Euclidean distance d between the defect geometric center and the corresponding hole center, the angle θ between the defect's major axis and the hole position array arrangement direction, the distance b from the defect geometric center to the nearest risk zone boundary, and the number of holes n simultaneously covered by a single defect. For cases where the same defect covers multiple holes, the interference characteristics of each hole position can be calculated separately, and a defect-hole position bipartite diagram can be generated.

[0033] S130. The interference features are bound to the laser-induced results, etching results, filling results, electrical test results and reliability results of each hole in subsequent work stations to generate a final failure label.

[0034] In this embodiment, final failure refers to the failure result confirmed by subsequent workstations or electrical testing, such as blind vias, voids, high resistance, or reliability failure. The results of subsequent processes are mapped to binary labels: Y=1 indicates final failure of the via location, and Y=0 indicates normal operation. Failure determination covers all process anomalies, including blind vias that have not been penetrated after laser induction, abnormal hole diameter, hole wall cracks, excessively high hole wall roughness, and abnormal hole taper; voids and filling offset after filling; high resistance, open circuits, and short circuits in electrical testing; functional failures after reliability testing; and final batch / wafer scrap. For example, the laser induction stage can output induction continuity, induction offset, and induction anomalies; the etching stage can output blind vias, abnormal hole diameter, hole wall roughness, and hole taper; the conductive paste filling stage can output filling voids, residual bubbles, and insufficient filling; and the electrical testing stage can output high resistance, open circuits, and short circuits.

[0035] Due to the significant noise in historical production line data, sample label correction can be performed. When the defects in the blank wafer and subsequent failures are spatially inconsistent, temporally discontinuous, or the confidence level of the subsequent equipment coordinates is below a threshold, the training weight of the sample is reduced or it is marked as an uncertain sample to prevent the model from learning incorrect associations. For example, if the difference between the global coordinates of the blank wafer defect and the coordinates of the failed hole in the subsequent process exceeds the hole position tolerance + equipment positioning error + 50% safety margin, it indicates that there is no spatial correlation between the two, and the training weight of the sample is reduced or it is marked as an uncertain sample. If there is a batch interval between the blank wafer AOI inspection and the subsequent process, such as equipment switching or no coordinate calibration record, the label confidence level of the sample is reduced. If the failure in the subsequent process cannot be located to a specific hole position (such as the entire batch being scrapped or the electrical measurement positioning error being too large), the sample is only used for batch-level risk assessment and does not participate in single-hole model training.

[0036] S140. Construct and train a failure probability model based on the interference features, defect severity features, process parameter features, and final failure label.

[0037] According to embodiments of this disclosure, a failure probability model P(Y=1|X,Z) is constructed, where Y is the final failure label, Y=1 indicates the final failure of the hole position, X is the interference feature and defect severity feature, and Z is the process parameter feature. The process parameter feature includes at least two of the following: laser energy, pulse width, scanning speed, focal position, etching time, etchant concentration, etching temperature, conductive paste viscosity, filling pressure, vacuum time, sintering temperature, equipment number, Recipe version, and batch information. The failure probability model is any one of Logistic regression, Bayesian model, random forest, gradient boosting tree, support vector machine, neural network, graph model, or rule-model fusion model.

[0038] The appropriate model can be selected based on the sample size and interpretability requirements. For small sample scenarios (<10,000 samples), Logistic Regression and Bayesian models are preferred due to their strong interpretability and ability to directly output the contribution of each feature to the failure probability. For medium to large sample scenarios (10,000 to 100,000 samples), Random Forest and Gradient Boosting Tree (XGBoost / LightGBM) are recommended for their high accuracy and ability to handle nonlinear relationships. For large sample scenarios (>100,000 samples), Neural Networks and Graph Models can be selected (if there is an array correlation between the hole positions) to further improve accuracy. For engineering application scenarios, rule-model fusion models are preferred. Hard process rules, such as core lethal zone defects that directly trigger high risks, are used as a pre-filter layer, and then the data is input into a machine learning model for refined scoring, balancing rule interpretability and model accuracy.

[0039] In one implementation, the system uses a Logistic model as the base model to obtain interpretable weights: P(Y=1|X,Z)=σ(w0+w1·ρ+w2·IoU-w3·d+w4·S+w5·Z+w6·ρ·Z) Where σ is the Sigmoid function, S is the defect severity, and ρ·Z represents the interaction term between defect interference and process parameters.

[0040] To align with the physical logic of the TGV process, two types of prior rule terms can be embedded in the model. The first is a distance attenuation term, which reduces the contribution of the defect to the failure probability as the distance between the defect center and the hole center increases. An attenuation coefficient 'd' is introduced for the defect-hole center distance; the smaller 'd', the higher the weight. For example, the weight is 1 when d < 10 μm, drops to 0.2 when d = 50 μm, and approaches 0 when d > 100 μm, consistent with the physical principle that the farther the defect is from the hole, the smaller its impact. The second is a partitioning weight term, assigning different risk weights when a defect falls into the core lethal zone, the transitional risk zone, and the safe observation zone. The core lethal zone has a weight of 1, the transitional risk zone has a weight of 0.5, and the safe observation zone has a weight of 0.1, allowing the model to automatically match the risk level of different regions.

[0041] To ensure the quality of samples for model training and to distinguish between the effects of defects and random process fluctuations, a quantitative experimental plan was established. Experiments were grouped according to defect type, defect area, distance between defect and hole location, interference ratio, laser recipe, etching recipe, and filling recipe. Subsequent process results for each group were collected as the training set for the model, and the model parameters were updated. For example, for bubble defects, three groups—low interference, medium interference, and high interference—could be set up, and subsequent hole formation results were collected under different laser energies and etching times. According to an embodiment of this disclosure, the experimental groupings are shown in the table below:

[0042] As shown in the table above, the experimental plan includes a blank control group, a low-interference group, a medium-interference group, a high-interference group, and a repeated validation group, used to distinguish the correlation between blank defects and subsequent failures and the randomness caused by process fluctuations. The blank control group is used to estimate the baseline failure rate in the absence of blank defects, and the repeated validation group is used to estimate random process fluctuations. During model training, random process fluctuations can be distinguished from the failure risk caused by blank defects.

[0043] S150. Based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model, output the hole location risk score and engineering treatment suggestions.

[0044] In the embodiments of this disclosure, the risk score Risk is a quantitative value that integrates the model output, confidence level, severity, and process rules. The calculation logic is as follows: Risk = f(P(Y=1|X,Z), Conf_coord, Conf_defect, Severity, Rule), where Conf_coord is the coordinate confidence, Conf_defect is the defect detection confidence, Severity is the defect severity, and Rule is the process rule threshold.

[0045] First, a risk threshold is learned based on the failure probability output by the model. If the risk score is below the first threshold, it is recommended to continue processing; if the risk score is between the first and second thresholds, it is recommended to conduct sampling re-judgment or re-inspection of the camera; if the risk score is between the second and third thresholds, it is recommended to avoid hole positions or adjust the recipe; if the risk score is above the third threshold, it is recommended to suspend processing, rework, scrap, or require manual approval. Here, the recipe is a set of parameters for the inspection or process equipment, including but not limited to ROI, exposure, light source, threshold, model version, and output rules.

[0046] The handling strategy relies not only on probability values ​​but also on coordinate confidence, defect detection confidence, defect category, and process rules. Engineering handling recommendations can be adjusted based on coordinate confidence, defect detection confidence, defect severity, and process rule thresholds. For example, when a defect is located in a critical lethal zone and has high coordinate confidence, the system can increase the risk level based on rule gating, even with a small number of historical samples. When coordinate confidence is low, the system prioritizes triggering a re-evaluation rather than direct scrapping.

[0047] In one implementation, the latest production data and back-end process results are collected periodically, and the actual back-end results are continuously written back to the model training set. The model is incrementally updated, and model performance is monitored by equipment, batch, product model, and recipe version. If the model's predicted probability deviates from the actual failure rate for a long period of time under a certain equipment or recipe version, it is determined that there is model drift or process change. When the model drift exceeds a threshold, automatic scrapping and automatic recipe adjustment are restricted. Only re-judgment suggestions are allowed, and tasks for re-experimentation, recalibration, or supplementary sample collection are generated to avoid batch misjudgments caused by model failure.

[0048] In one specific implementation, an AOI of a TGV white film detected a scratch that is 100μm long and 20μm wide. The center of the defect is 30μm away from the center of the adjacent hole. The designed hole diameter is 50μm. The diameter of the risk area R is 70μm (50μm for the core area, 50~60μm for the transition area, and 60~70μm for the safe area). First, interference calculations were performed, yielding an intersection area of ​​2100 μm² between the scratch and the safe zone, and a total area of ​​3848 μm². The interference ratio ρ = 2100 / 3848 ≈ 54.6%, indicating high interference. The model was then input with features such as a scratch severity (weight 1.5), a detection confidence level of 92%, a coordinate alignment confidence level of 96%, and current process parameters (laser energy at 90% of the baseline value, which is relatively low). The model output a failure probability P(Y=1|X,Z) = 72%, and a comprehensive risk score Risk = 72% × 0.96 × 0.92 × 1.5 × 0.3 ≈ 22%. The recommended action was sampling reassessment, selecting 5% of the holes under the same conditions for electrical testing confirmation.

[0049] As can be seen from the above scheme, the quantitative relationship modeling method between front-end defects and back-end failures in the TGV process provided in this disclosure has the following beneficial effects: First, it transforms the relationship between blank wafer defects and CAD hole positions from empirical descriptions into calculable features such as interference area, interference ratio, and distance; Second, by binding back-end results, it distinguishes between truly fatal and non-fatal defects, reducing excessive scrap; Third, by outputting failure risks through probabilistic models, it provides a quantitative basis for hole position avoidance, re-judgment, and recipe adjustment; Fourth, through incremental updates and drift monitoring, it enables the model to continuously iterate with changes in different equipment, batches, and recipes.

[0050] Based on the same inventive concept, embodiments of the present invention also provide a quantitative relationship modeling device between front-end defects and back-end failures in the TGV process, for performing the aforementioned quantitative relationship modeling method between front-end defects and back-end failures in the TGV process. Figure 2 This application provides a schematic diagram of a device for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process, which is applied to inspection equipment. The device may include: The generation module 210 is used to acquire defect data of TGV blank wafers and hole position design data corresponding to the TGV blank wafers. The defect data includes at least one blank wafer defect area and its defect severity characteristics, and generates a process risk area corresponding to each hole position based on the hole position design data.

[0051] The calculation module 220 is used to map the defective region of the white film and the process risk region to the same coordinate system, and then calculate the interference features between the defective region of the white film and the process risk region. The interference features include at least the interference area.

[0052] The binding module 230 is used to bind the interference features with the laser-induced results, etching results, filling results, electrical test results and reliability results of each hole in subsequent work stations to generate a final failure label.

[0053] Modeling module 240 is used to construct and train a failure probability model based on the interference features, defect severity features, process parameter features and the final failure label.

[0054] The disposal decision module 250 is used to output hole location risk scores and engineering disposal suggestions based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model.

[0055] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3As shown in the embodiment of this application, an electronic device 300 includes a memory 330, a processor 310, and a computer program 320 stored in the memory. The processor 310 executes the computer program 320 to implement the quantitative relationship modeling method between front-end defects and back-end failures in the TGV process described in any of the above embodiments.

[0056] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the quantitative relationship modeling method between front-end defects and back-end failures in the TGV process described in any of the above embodiments.

[0057] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0058] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0059] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for modeling the quantitative relationship between front-end defects and back-end failures in a TGV process, characterized in that, The method includes: Obtain defect data of TGV white film and hole position design data corresponding to the TGV white film. The defect data includes at least one defect area of ​​white film and its defect severity characteristics. Generate process risk area corresponding to each hole position based on the hole position design data. After mapping the defective area of ​​the blank film and the process risk area to the same coordinate system, the interference features between the defective area of ​​the blank film and the process risk area are calculated, and the interference features include at least the interference area; The interference features are then linked to the laser-induced results, etching results, filling results, electrical test results, and reliability results of each hole in subsequent work stations to generate a final failure label. A failure probability model is constructed and trained based on the interference features, defect severity features, process parameter features, and the final failure label. Based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model, the system outputs a hole location risk score and engineering treatment suggestions.

2. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The process involves acquiring defect data for TGV wafers and corresponding via design data. The defect data includes at least one defective region of the wafer and its defect severity characteristics. Based on the via design data, a process risk region is generated for each via, including: Defect detection is performed on TGV blank glass using an automated optical inspection device to obtain AOI images and defect detection results. The defect detection results include at least one defect area of ​​the blank glass, its defect category, and defect severity features. The defect severity features include defect area, maximum length, minimum width, aspect ratio, grayscale contrast, edge sharpness, texture roughness, transparency variation, depth estimate, coordinates, and defect category confidence. The defect category includes at least one of the following: bubbles, foreign matter, scratches, edge chipping, contamination, microcracks, inclusions, thickness abnormalities, or visible abnormalities inside the transparent glass. When the defect is a bubble, calculate the bubble's grayscale gradient, edge closure, and transparency changes; when the defect is a crack, calculate the crack's length, direction, number of branches, and endpoint coordinates; when the defect is a foreign object, calculate the foreign object's area, shape factor, and occlusion strength. The system reads the hole design data corresponding to the TGV wafer and generates process risk areas based on the hole center, designed hole diameter, laser-induced spot diameter, etching extension allowance, hole position tolerance, equipment positioning error, and safety margin. These process risk areas are divided into core lethal zones, transitional risk zones, and safety observation zones. Simultaneously, it outputs hole-level risk, region-level risk, and whole-wafer-level risk. The hole-level risk is used for hole position avoidance or reassessment, the region-level risk is used for process parameter adjustment, and the whole-wafer-level risk is used for determining whether to continue processing or scrap the wafer.

3. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, After mapping the defective region of the unfinished wafer and the process risk region to the same coordinate system, the interference features between the defective region of the unfinished wafer and the process risk region are calculated. The interference features include at least the interference area, including: After transforming the white film defect area and the process risk area to the same global coordinate system, the interference area Area(D∩R) is calculated, where D is the white film defect area and R is the process risk area. The interference ratio is calculated using ρ=Area(D∩R) / Area(R) or ρ=Area(D∩R) / Area(D), where Area(D∩R) is the interference area, Area(R) is the area of ​​the process risk region, and Area(D) is the area of ​​the blank film defect region. Calculate the crossover ratio, the distance from the defect center to the hole center, the angle between the defect long axis direction and the hole arrangement direction, the minimum distance from the defect to the risk zone boundary, and the number of holes covered by the defect. The crossover ratio is IoU=Area(D∩R) / Area(D∪R), where Area(D∪R) is the total area after the regions are superimposed.

4. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The final failure label includes at least one of the following: blind hole, abnormal hole diameter, hole wall crack, excessive hole wall roughness, abnormal hole taper, filling void, high resistance, open circuit, short circuit, reliability failure, or final scrap.

5. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The step of binding the interference features with the laser-induced results, etching results, filling results, electrical test results, and reliability results of each hole position in subsequent work stations to generate a final failure label also includes: The samples are labeled and corrected. When the defects in the blank film are inconsistent in space, discontinuous in time, or the confidence of the coordinates of the subsequent equipment is lower than the threshold, the training weight of the sample is reduced or the sample is marked as an uncertain sample.

6. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The construction and training of the failure probability model based on the interference features, defect severity features, process parameter features, and the final failure label includes: Construct a failure probability model P(Y=1|X,Z), where Y is the final failure label, Y=1 indicates the final failure of the hole position, X is the interference feature and defect severity feature, and Z is the process parameter feature. The process parameter feature includes at least two of the following: laser energy, pulse width, scanning speed, focal position, etching time, etchant concentration, etching temperature, conductive paste viscosity, filling pressure, vacuum time, sintering temperature, equipment number, recipe version, and batch information. The failure probability model is any one of Logistic regression, Bayesian model, random forest, gradient boosting tree, support vector machine, neural network, graph model, or rule-model fusion model. Experiments were grouped according to defect type, defect area, distance between defect and hole, interference ratio, laser recipe, etching recipe, and filling recipe. The subsequent process results of each group were collected as the training set for the model, and the model parameters were updated.

7. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The output of hole location risk score and engineering handling suggestions based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model includes: Based on the failure probability learning risk threshold output by the failure probability model, if the risk score is lower than the first threshold, it is recommended to continue processing; if the risk score is between the first and second thresholds, it is recommended to sample and re-judge or re-inspect the camera for re-judgment; if the risk score is between the second and third thresholds, it is recommended to avoid the hole position or adjust the recipe; if the risk score is higher than the third threshold, it is recommended to suspend processing, rework, scrap, or manually approve. Based on coordinate confidence, defect detection confidence, defect severity, and process rule threshold adjustment engineering handling suggestions, the engineering handling suggestions include at least one of the following: continue processing, sampling re-judgment, re-inspection camera re-judgment, multi-focal plane re-judgment, hole position avoidance, recipe adjustment, suspension of processing, rework, scrapping, or manual approval.

8. The method for modeling the quantitative relationship between front-end defects and back-end failures in the TGV process according to claim 1, characterized in that, The method further includes: The actual back-end results are continuously written back to the model training set, and the failure probability model is incrementally updated. The model performance is monitored by device, batch, product model and Recipe version. If the model's predicted failure probability deviates from the actual failure rate under a certain device or Recipe version, it is determined that there is model drift or process change. When the model drift exceeds the threshold, prompts will be output to re-experiment, recalibrate or supplement the sample collection task, restrict automatic scrapping and automatic Recipe adjustment.

9. A device for modeling the quantitative relationship between front-end defects and back-end failures in a TGV process, characterized in that, The device includes: The generation module is used to acquire defect data of TGV wafers and hole position design data corresponding to the TGV wafers. The defect data includes at least one defect area of ​​the wafer and its defect severity characteristics, and generates a process risk area corresponding to each hole position based on the hole position design data. The calculation module is used to map the defective region of the unfinished film and the process risk region to the same coordinate system, and then calculate the interference features between the defective region of the unfinished film and the process risk region. The interference features include at least the interference area. The binding module is used to bind the interference features with the laser-induced results, etching results, filling results, electrical test results and reliability results of each hole position in subsequent work stations to generate the final failure label; The modeling module is used to construct and train a failure probability model based on the interference features, defect severity features, process parameter features, and the final failure label. The disposal decision module is used to output hole location risk scores and engineering disposal suggestions based on the failure probability, coordinate confidence, defect detection confidence, defect severity, and process rule threshold output by the failure probability model.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1-8.