Wafer defect detection system and method based on multi-modal data fusion

Abstract

The invention discloses a wafer defect detection system and method based on multi-modal data fusion, and the system comprises a multi-source data collection unit, a multi-physics field modeling unit, a wafer integrated detection unit, and a defect interaction management unit. Through five steps of acquiring multi-source detection data, constructing a multi-physics field model, detecting and attributing wafer defects, interactively displaying and performing early warning management, the comprehensiveness and adaptability of data acquisition are improved through a multi-source data acquisition unit, and the relevance between physics fields and defects is accurately described through a multi-physics field modeling unit; the wafer integrated detection unit is used for improving the precision of defect detection and attribution, and then the defect interaction management unit is used for realizing full-process visualization and closed-loop process optimization, so that the wafer defect detection system based on multi-modal data fusion is developed towards the high-precision, low-power-consumption and explainable direction; and core support is provided for intelligent upgrading of the semiconductor industry.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and in particular to a wafer defect detection system and method based on multimodal data fusion. Background Technology

[0002] Wafers are the basic material for manufacturing semiconductor chips. They are usually made of high-purity single-crystal silicon or other semiconductor materials, such as gallium arsenide and gallium nitride, into circular thin films for processing electronic components such as integrated circuits.

[0003] However, traditional wafer defect detection technology suffers from limitations in data acquisition, fragmented physical field analysis, insufficient correlation between macro and micro features, and consequently, delays in interaction and optimization. Traditional inspection methods rely on a single data source and have a fixed data acquisition frequency, making it difficult to adapt to dynamic changes in process conditions. This results in insufficient data timeliness and correlation, making it easy to miss key defect information. Furthermore, since wafer defects are closely related to multiple physical fields such as laser action, temperature changes, and stress distribution, traditional methods lack coupled modeling of laser, temperature, and stress fields, making it impossible to establish a quantitative correlation model of macro and micro parameters. This leads to one-sided defect feature analysis, making it difficult to accurately correlate physical processes with defect causes, affecting defect detection accuracy. As a result, there is a lack of intuitive visualization and digital twin interaction, and defect warnings and process optimization responses are lagging behind. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of data acquisition limitations, fragmented physical field analysis, and insufficient correlation between macro and micro features in traditional wafer defect detection technology, as well as the resulting lag in interaction and optimization.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A wafer defect detection system and method based on multimodal data fusion includes the following steps: Step 1: Collect multi-source detection data: Multi-source detection data includes laser detection images, optical detection images, and wafer process parameters; by analyzing the process status through wafer process parameters, the data acquisition frequency is dynamically adjusted, and a spatiotemporal calibration algorithm for feature point matching is set, thereby establishing a dynamic adaptive alignment multimodal data acquisition strategy; Step 2, construct a multiphysics model: The multiphysics model includes a multimodal fusion physical field and a wafer thermal stress coupling model. By building a multimodal fusion physical field and wafer thermal stress coupling model, the local thermal effects during the laser scanning process are analyzed, and the stress tensor and total strain of the wafer are evaluated. Step 3, wafer defect detection and attribution: By building a wafer digital twin model and generating a heat map of the defect area, the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer are analyzed, and the characteristic parameters of wafer defects are automatically identified and classified. Then, combined with the key points of local stress on the wafer, the causes of wafer defects are comprehensively deduced. Step 4, Interactive Display and Early Warning Management: Display the wafer digital twin model through the digital twin interactive interface, display defect detection results through the defect visualization module, provide defect cause prompts through the defect early warning management module, and output corresponding process optimization design schemes through the process optimization design module.

[0006] Furthermore, the specific process of analyzing the process status through wafer process parameters is as follows: Wafer process parameters include temperature, pressure, rotation speed, and industrial stage designation; Obtain the process state vector St through wafer process parameters: ; Among them, temperature is marked as Tt, pressure as Pt, rotation speed as Rt, and industrial stage identifier as Mt; temperature Tt includes wafer surface temperature and equipment chamber temperature; pressure Pt refers to the pressure in the wafer manufacturing chamber; rotation speed Rt refers to the rotation speed of the wafer stage; industrial stage identifier Mt refers to the discretized identifier of the current step in wafer manufacturing. Set the basic sampling frequency The maximum sampling rate supported by the sensor acquisition hardware is marked as The standard vector of the process state is labeled as Sf: ; Set maximum sampling rate Dynamically adjusted data acquisition frequency Constraints are applied to obtain a dynamically adjusted data acquisition frequency. .

[0007] Furthermore, the specific process of the spatiotemporal calibration algorithm for feature point matching is as follows: By extracting feature points Li from the laser detection image and feature points Ci from the optical detection image, and then performing linear least squares calculations, the optimal transformation matrix of the feature point set is obtained. ; The optimal transformation matrix of the feature point set Data points in the laser detection image that match the feature points in the optical detection image are selected for spatiotemporal alignment.

[0008] Furthermore, the wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model; A laser-temperature field coupled sub-model is established by combining the laser heat source equation and the heat conduction equation; A temperature-stress field coupled sub-model is established by simultaneously solving the fundamental equation of the stress tensor and the total strain decomposition equation.

[0009] Furthermore, the specific process of establishing the laser-temperature field coupled sub-model is as follows: A laser heat source equation is constructed by combining laser power P, wafer surface absorptivity ε, laser spot radius R0, and laser scanning speed v. Temperature field in three-dimensional space and time Two-dimensional space and time heat source field A heat conduction equation is constructed by combining the wafer material density ρ, thermal conductivity k(T), and specific heat capacity c.

[0010] Furthermore, the specific process of establishing the temperature-stress field coupled sub-model is as follows: Through the components of the strain tensor Elasticity coefficient matrix Thermal stress coefficient tensor and the initial temperature of the equipment chamber Combining these, the fundamental equation of the stress tensor is constructed as follows: ; Through elastic strain Plastic strain Thermal strain Thermal expansion coefficient tensor By combining these methods, the total strain decomposition equation can be obtained.

[0011] Furthermore, macro- and micro-parameter correlation equations are constructed to analyze the macro-anomaly characteristic parameters and micro-defect characteristic parameters of the wafer; The macro-micro parameter correlation equations include the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer; Macroscopic anomaly parameters of a wafer include warpage W, induced by bending stiffness D, and normal stress in the z-direction. By combining the Laplace operator, the warpage W can be obtained comprehensively. Microscopic defect characteristic parameters of wafers include dislocation density. Through stress tensor gradient The dislocation density is obtained by combining the dislocation generation coefficient k1 and the dislocation annihilation coefficient k2. .

[0012] Furthermore, a defect detection and attribution model is constructed to automatically identify and classify wafer defects; Through temperature gradient Maximum stress dislocation density Combined with the warp degree W, a feature vector F is created: ; The probability of wafer defects is evaluated by combining the mean function and the covariance function. Then, the defect areas are accumulated to obtain the total number and coordinates of wafer defects, and a heat map of the defect areas is generated.

[0013] Furthermore, the process of attributing defects into categories is as follows: Set yield strength and temperature gradient threshold When the maximum stress > and temperature gradient > If so, it is determined that there are microcrack defects at the local stress critical points of the wafer; Set dislocation density threshold and plastic strain threshold When dislocation density Greater than And plastic strain > If so, it is determined that there is a dislocation accumulation defect at the local stress critical point of the wafer; This generates and labels the digital twin state vector as Vt: ; Where Dq represents the defect distribution; Pq represents the physical field distribution; and Cq represents the defect cause determination.

[0014] The wafer defect detection system based on multimodal data fusion includes a multi-source data acquisition unit, a multi-physics modeling unit, a wafer integrated detection unit, and a defect interaction management unit. The units are interconnected and the system executes the wafer defect detection method based on multimodal data fusion as described above when it is applied. The multi-source data acquisition unit is used to acquire multi-source detection data: the multi-source data acquisition unit includes a laser scattering module, an optical imaging module, and a timing synchronization module; The multiphysics modeling unit is used to construct multiphysics models: the multiphysics modeling unit includes a multimodal fusion physics field and wafer thermal stress coupling model, and the wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model. The wafer integrated inspection unit is used for wafer defect detection and attribution: The wafer integrated inspection unit includes a defect feature extraction module, a defect detection and localization module, a heat map generation module, and a cause deduction module; The Defect Interaction Management Unit is used for interactive display and early warning management: The Defect Interaction Management Unit includes a digital twin interactive interface, a defect visualization module, a defect early warning management module, and a process optimization design module.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention enhances the comprehensiveness and adaptability of data acquisition through a multi-source data acquisition unit, accurately depicts the correlation between physical fields and defects through a multi-physics modeling unit, improves the accuracy of defect detection and attribution through a wafer integrated inspection unit, and finally achieves full-process visualization and closed-loop process optimization through a defect interaction management unit. This invention integrates laser, optical detection parameters, and wafer process parameters through a multi-source data acquisition unit, dynamically adjusts the acquisition frequency through a timing synchronization module, and achieves adaptive alignment of multimodal data through a spatiotemporal calibration algorithm. This solves the limitations of traditional single data source and fixed frequency, ensures the timeliness and relevance of data, and provides more comprehensive basic data for defect detection. This invention constructs a wafer thermal stress coupling model through a multi-physics modeling unit, and builds coupled sub-models of laser and temperature field, and temperature and stress field. By combining the laser heat source equation, heat conduction equation, stress tensor equation, etc., the dynamic relationship between laser action, temperature change and stress distribution can be quantified, and the physical cause of defects can be accurately traced, breaking through the problem of fragmentation in traditional physical field analysis. This invention establishes macro- and micro-parameter correlation equations through wafer integrated detection units, analyzes the quantitative relationship between warpage and dislocation density, and constructs a defect detection model by combining feature vectors such as temperature gradient, stress, and dislocation density. It generates defect heat maps and automatically classifies defect types, realizing the collaborative analysis of macro- and micro-features and improving the accuracy of defect identification and attribution. This invention implements a defect interactive management unit with a digital twin interactive interface to display the wafer model. Combined with a defect visualization module and an early warning management module, it intuitively presents the detection results and cause prompts. At the same time, it outputs targeted solutions through a process optimization design module, realizing closed-loop management from defect detection to process improvement, reducing manual intervention, and improving the stability and efficiency of wafer manufacturing. Attached Figure Description

[0016] Figure 1 A schematic diagram of the steps in the method flow of the present invention is shown; Figure 2 A schematic diagram of the system modules of the present invention is shown. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: like Figures 1-2As shown, the wafer defect detection system based on multimodal data fusion includes a multi-source data acquisition unit, a multi-physics modeling unit, a wafer integrated detection unit, and a defect interaction management unit, with communication connections between the various units; The multi-source data acquisition unit is used to acquire multi-source detection data: the multi-source data acquisition unit includes a laser scattering module, an optical imaging module, and a timing synchronization module; The laser scattering module generates a wafer laser inspection image and acquires laser inspection parameters; the optical imaging module generates a wafer optical inspection image and acquires optical inspection parameters; the timing synchronization module analyzes the process status through wafer process parameters, thereby dynamically adjusting the data acquisition frequency and setting a spatiotemporal calibration algorithm for feature point matching, thereby establishing a dynamic adaptive alignment multimodal data acquisition strategy. The multiphysics modeling unit is used to construct multiphysics models: the multiphysics modeling unit includes a multimodal fusion physics field and wafer thermal stress coupling model, and the wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model. Among them, the laser-temperature field coupling sub-model is used to comprehensively analyze the laser heat source equation and the heat conduction equation; the temperature-stress field coupling sub-model is used to comprehensively analyze the fundamental stress tensor equation and the total strain decomposition equation. The wafer integrated inspection unit is used for wafer defect detection and attribution: The wafer integrated inspection unit includes a defect feature extraction module, a defect detection and localization module, a heat map generation module, and a cause deduction module; The defect feature extraction module is used to analyze the macroscopic anomaly feature parameters and microscopic defect feature parameters of the wafer; the defect detection and localization module is used to assess the wafer defect probability and perform coordinate localization; the heat map generation module is used to generate heat maps of the defect area; and the cause deduction module is used for defect attribution deduction and classification. The Defect Interaction Management Unit is used for interactive display and early warning management: The Defect Interaction Management Unit includes a digital twin interactive interface, a defect visualization module, a defect early warning management module, and a process optimization design module; The system utilizes a digital twin interface to display the wafer digital twin model, a defect visualization module to display defect detection results, a defect early warning management module to provide defect cause prompts, and a process optimization design module to output corresponding process optimization design solutions.

[0019] The work steps are as follows: S1, Collect multi-source detection data: Multi-source detection data includes laser detection parameters, optical detection parameters, and wafer process parameters; Analyze the process status through wafer process parameters, thereby dynamically adjusting the data acquisition frequency and setting a spatiotemporal calibration algorithm for feature point matching, thereby establishing a dynamic adaptive alignment multimodal data acquisition strategy; S1-1, the specific process of analyzing the process status through wafer process parameters is as follows: Wafer process parameters include temperature, pressure, rotation speed, and industrial stage designation; Obtain the process state vector St through wafer process parameters: ; Temperature is marked as Tt, pressure as Pt, rotational speed as Rt, and industrial stage as Mt. Temperature Tt includes wafer surface temperature and equipment chamber temperature; wafer surface temperature is detected by infrared thermal imager or thermocouple array; equipment chamber temperature is detected by platinum resistance sensor or infrared temperature measurement module inside the chamber. Pressure Pt refers to the pressure in the wafer fabrication chamber, which is collected by a capacitive pressure transmitter or a piezoelectric sensor. Rotational speed Rt refers to the rotational speed of the wafer stage, which is monitored and obtained by the speed feedback signal of the photoelectric encoder or motor driver; The industrial stage identifier Mt refers to the discrete marker of the current step in wafer manufacturing, which is read in real time by the equipment control system. The wafer manufacturing equipment predefines the wafer process recipe and its step identifier Mt. The step identifiers of the wafer process recipe include etching pretreatment, main etching, and post-treatment.

[0020] S1-2, Set the basic acquisition frequency The maximum sampling rate supported by the sensor acquisition hardware is marked as The standard vector of the process state is labeled as Sf: ; The standard vector Sf of the process status is preset using the average of historical normal process data. The closer the process status vector St is to the standard vector Sf, the more stable the process parameters are; conversely, the closer it is to the standard vector Sf, the greater the deviation of the process parameters and the higher the degree of anomaly of multiple parameters, requiring an increase in the data acquisition density of key stages. S1-3, thereby obtaining the dynamically adjusted data acquisition frequency. : ; Where e is the natural constant, To adjust the coefficient and , The formula for the comprehensive deviation of multiple parameters is as follows: ; in, The weighting coefficients for each process stage are preset and obtained after calculation using a large amount of experimental data, and when... A higher value indicates a greater process deviation, thus requiring a higher data acquisition frequency. The higher the sampling rate, the denser the data collection; this is achieved through the maximum sampling rate. Dynamically adjusted data acquisition frequency Apply constraints.

[0021] S1-3, the specific process of the spatiotemporal calibration algorithm for feature point matching is as follows: By extracting feature points Li from the laser detection image and feature points Ci from the optical detection image, and then performing linear least squares calculations, the optimal transformation matrix of the feature point set is obtained. :

[0022] Among them, feature point Li includes spatial coordinate features and scattering intensity features; feature point Ci includes spatial coordinate features and optical grayscale features; H is an arbitrary transformation matrix; f is the mapping function between scattering intensity and feature point, which is obtained through testing with a large amount of experimental data. When the scattering intensity is higher, it is more likely to correspond to defects, such as particles or scratches, and thus correspond to feature points of optical images. The optimal transformation matrix of the feature point set Data points in the laser detection image that match the feature points in the optical detection image are selected for spatiotemporal alignment.

[0023] S2, Constructing a multiphysics model: The multiphysics model includes a multimodal fusion physical field and a wafer thermal stress coupling model. By constructing a multimodal fusion physical field and wafer thermal stress coupling model, the local thermal effects during the laser scanning process are analyzed, and the stress tensor and total strain of the wafer are evaluated. S2-1, the multimodal fusion physical field is constructed by spatiotemporally aligning the feature points of laser detection images and optical detection images; S2-2, the wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model; S2-201 establishes a laser-temperature field coupled sub-model by combining the laser heat source equation and the heat conduction equation. The laser heat source equation is: ; Where P is the laser power, adjusted based on the wafer material thickness and experience; ε is the wafer surface absorptivity, related to the material surface condition, and lower when the silicon wafer is uncoated than after coating; R0 is the laser spot radius; v is the laser scanning speed; (x-vt, y) refers to the coordinate system that moves with the spot, with the spot center at the origin when t=0; the laser energy exhibits a two-dimensional Gaussian distribution within the spot, moving along the scanning direction x-axis with time, and the coefficient... Ensure total energy conservation: ; The heat conduction equation is: ; in, It refers to the temperature field in three-dimensional space and time, reflecting the spatial distribution and temporal evolution of heat conduction; It refers to the two-dimensional space and time heat source field, reflecting the planar distribution and time variation of the heat source; ρ is the density of the wafer material; c is the specific heat capacity; k(T) is the thermal conductivity; This is a thermal conduction term; the boundary condition is that the upper and lower surfaces of the wafer satisfy convection heat dissipation. h is the convective heat transfer coefficient; It refers to the change in internal energy per unit volume. This refers to the net inflow of heat conduction. It is the laser heat source input amount; S2-202 establishes a temperature-stress field coupled sub-model by simultaneously solving the fundamental equation of the stress tensor and the total strain decomposition equation. The fundamental equation of the stress tensor is: ; in, i represents the normal direction of the stress surface, and j represents the direction of stress application. k and l are components of the strain tensor. When k = l, it is normal strain; when k ≠ l, it is shear strain. ijkl takes the values ​​1, 2, and 3 respectively, corresponding to the x, y, and z directions. This is the elasticity coefficient matrix; This is the thermal stress coefficient tensor; This refers to the initial temperature of the equipment chamber. The stress generated by elastic strain, Thermal stress caused by temperature changes; The total strain decomposition equation is: ; in, For elastic strain, For plastic strain, For thermal strain, The coefficient of thermal expansion is the tensor; the total strain is influenced by a combination of elastic recovery, plastic deformation, and thermal expansion and contraction.

[0024] S3, Wafer Defect Detection and Attribution: By building a wafer digital twin model and generating a heat map of the defect area, the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer are analyzed, and the characteristic parameters of wafer defects are automatically identified and classified. Combined with the key points of local stress on the wafer, the causes of wafer defects are comprehensively deduced. S3-1, Construct macro-micro parameter correlation equations to analyze the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of wafers; The macro-micro parameter correlation equations include the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer; Macroscopic anomaly parameters of wafers include warpage W: ; Where D is the bending stiffness. The normal stress is in the z-direction. is the Laplace operator and reflects the second-order rate of change of stress; W is the warpage, which refers to the maximum height difference between the center and the edge of the wafer; Microscopic defect characteristic parameters of wafers include dislocation density. : ; in, The stress tensor gradient reflects the spatial rate of change of stress; k1 is the dislocation generation coefficient, k2 is the dislocation annihilation coefficient, and k1 and k2 are material constants; dislocation density. The warpage W corresponds to microscopic defects and macroscopic anomalies, respectively.

[0025] S3-2, Construct a defect detection and attribution model to automatically identify and classify wafer defects; Through temperature gradient Maximum stress dislocation density Combined with the warp degree W, a feature vector F is created: ; This represents the mean function, taking the sum of the normal stresses in the x and y directions; denoted by the covariance function, d is the spatial correlation length, and the characteristic length that controls the rate of correlation decay. The coordinates of the reference point, The coordinates of the target point; The probability of wafer defects is evaluated by combining the mean function and the covariance function. : ; Where N is the symbol for a multivariate normal distribution, used to model the probability distribution of a continuous field; when the wafer defect probability... The larger the value, the higher the probability that there is a defect at that location (x, y); Then, the defect areas are accumulated to obtain the total number and coordinates of wafer defects, and a heat map of the defect areas is generated.

[0026] S3-3, The process of defect attribution derivation and classification is as follows: Set yield strength and temperature gradient threshold When the maximum stress > and temperature gradient > If so, it is determined that there are microcrack defects at the local stress critical points of the wafer; Set dislocation density threshold and plastic strain threshold When dislocation density Greater than And plastic strain > If so, it is determined that there is a dislocation accumulation defect at the local stress critical point of the wafer; This generates and labels the digital twin state vector as Vt: Where Dq is the defect distribution; Pq is the physical field distribution; and Cq is the defect cause determination.

[0027] S4, Interactive Display and Early Warning Management: Displays the wafer digital twin model through the digital twin interactive interface, displays defect detection results through the defect visualization module, provides defect cause prompts through the defect early warning management module, and outputs corresponding process optimization design schemes through the process optimization design module; Among them, the process optimization design scheme pre-sets the input device control system and adjusts and manages wafer fabrication based on the determined defect type results.

[0028] In summary, this invention enhances the comprehensiveness and adaptability of data acquisition through a multi-source data acquisition unit, accurately depicts the correlation between physical fields and defects through a multi-physics modeling unit, improves the accuracy of defect detection and attribution through a wafer integrated inspection unit, and finally achieves full-process visualization and closed-loop process optimization through a defect interaction management unit. This invention integrates laser, optical detection parameters, and wafer process parameters through a multi-source data acquisition unit, dynamically adjusts the acquisition frequency through a timing synchronization module, and achieves adaptive alignment of multimodal data through a spatiotemporal calibration algorithm. This solves the limitations of traditional single data source and fixed frequency, ensures the timeliness and relevance of data, and provides more comprehensive basic data for defect detection. This invention constructs a wafer thermal stress coupling model through a multi-physics modeling unit, and builds coupled sub-models of laser and temperature field, and temperature and stress field. By combining the laser heat source equation, heat conduction equation, stress tensor equation, etc., the dynamic relationship between laser action, temperature change and stress distribution can be quantified, and the physical cause of defects can be accurately traced, breaking through the problem of fragmentation in traditional physical field analysis. This invention establishes macro- and micro-parameter correlation equations through wafer integrated detection units, analyzes the quantitative relationship between warpage and dislocation density, and constructs a defect detection model by combining feature vectors such as temperature gradient, stress, and dislocation density. It generates defect heat maps and automatically classifies defect types, realizing the collaborative analysis of macro- and micro-features and improving the accuracy of defect identification and attribution. This invention implements a defect interactive management unit with a digital twin interactive interface to display the wafer model. Combined with a defect visualization module and an early warning management module, it intuitively presents the detection results and cause prompts. At the same time, it outputs targeted solutions through a process optimization design module, realizing closed-loop management from defect detection to process improvement, reducing manual intervention, and improving the stability and efficiency of wafer manufacturing.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0031] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0032] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0033] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A wafer defect detection method based on multimodal data fusion, characterized in that: Includes the following steps: Step 1: Collect multi-source detection data: Multi-source detection data includes laser detection images, optical detection images, and wafer process parameters; by analyzing the process status through wafer process parameters, the data acquisition frequency is dynamically adjusted, and a spatiotemporal calibration algorithm for feature point matching is set, thereby establishing a dynamic adaptive alignment multimodal data acquisition strategy; Step 2, construct a multiphysics model: The multiphysics model includes a multimodal fusion physical field and a wafer thermal stress coupling model. By building a multimodal fusion physical field and wafer thermal stress coupling model, the local thermal effects during the laser scanning process are analyzed, and the stress tensor and total strain of the wafer are evaluated. Step 3, wafer defect detection and attribution: By building a wafer digital twin model and generating a heat map of the defect area, the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer are analyzed, and the characteristic parameters of wafer defects are automatically identified and classified. Then, combined with the key points of local stress on the wafer, the causes of wafer defects are comprehensively deduced. Step 4, Interactive Display and Early Warning Management: Display the wafer digital twin model through the digital twin interactive interface, display defect detection results through the defect visualization module, provide defect cause prompts through the defect early warning management module, and output corresponding process optimization design schemes through the process optimization design module.

2. The wafer defect detection method based on multimodal data fusion according to claim 1, characterized in that: The specific process of analyzing the process status through wafer process parameters is as follows: Wafer process parameters include temperature, pressure, rotation speed, and industrial stage designation; Obtain the process state vector St through wafer process parameters: ; Among them, temperature is marked as Tt, pressure as Pt, rotation speed as Rt, and industrial stage identifier as Mt; temperature Tt includes wafer surface temperature and equipment chamber temperature; pressure Pt refers to the pressure in the wafer manufacturing chamber; rotation speed Rt refers to the rotation speed of the wafer stage; industrial stage identifier Mt refers to the discretized identifier of the current step in wafer manufacturing. Set the basic sampling frequency The maximum sampling rate supported by the sensor acquisition hardware is marked as The standard vector of the process state is labeled as Sf: ; Set maximum sampling rate Dynamically adjusted data acquisition frequency Constraints are applied to obtain a dynamically adjusted data acquisition frequency. .

3. The wafer defect detection method based on multimodal data fusion according to claim 2, characterized in that: The specific process of the spatiotemporal calibration algorithm for feature point matching is as follows: By extracting feature points Li from the laser detection image and feature points Ci from the optical detection image, and then performing linear least squares calculations, the optimal transformation matrix of the feature point set is obtained. ; The optimal transformation matrix of the feature point set Data points in the laser detection image that match the feature points in the optical detection image are selected for spatiotemporal alignment.

4. The wafer defect detection method based on multimodal data fusion according to claim 3, characterized in that: The wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model. A laser-temperature field coupled sub-model is established by combining the laser heat source equation and the heat conduction equation; A temperature-stress field coupled sub-model is established by simultaneously solving the fundamental equation of the stress tensor and the total strain decomposition equation.

5. The wafer defect detection method based on multimodal data fusion according to claim 4, characterized in that: The specific process of establishing the laser-temperature field coupled sub-model is as follows: A laser heat source equation is constructed by combining laser power P, wafer surface absorptivity ε, laser spot radius R0, and laser scanning speed v. Temperature field in three-dimensional space and time Two-dimensional space and time heat source field A heat conduction equation is constructed by combining the wafer material density ρ, thermal conductivity k(T), and specific heat capacity c.

6. The wafer defect detection method based on multimodal data fusion according to claim 5, characterized in that: The specific process of establishing the temperature-stress field coupled sub-model is as follows: Through the components of the strain tensor Elasticity coefficient matrix Thermal stress coefficient tensor and the initial temperature of the equipment chamber Combining these, the fundamental equation of the stress tensor is constructed as follows: ; Through elastic strain Plastic strain Thermal strain Thermal expansion coefficient tensor By combining these methods, the total strain decomposition equation can be obtained.

7. The wafer defect detection method based on multimodal data fusion according to claim 6, characterized in that: Construct macro-micro parameter correlation equations to analyze the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of wafers; The macro-micro parameter correlation equations include the macroscopic anomaly characteristic parameters and microscopic defect characteristic parameters of the wafer; Macroscopic anomaly parameters of a wafer include warpage W, induced by bending stiffness D, and normal stress in the z-direction. By combining the Laplace operator, the warpage W can be obtained comprehensively. Microscopic defect characteristic parameters of wafers include dislocation density. Through stress tensor gradient The dislocation density is obtained by combining the dislocation generation coefficient k1 and the dislocation annihilation coefficient k2. .

8. The wafer defect detection method based on multimodal data fusion according to claim 7, characterized in that: Construct a defect detection and attribution model to automatically identify and classify wafer defects; Through temperature gradient Maximum stress dislocation density Combined with the warp degree W, a feature vector F is created: ; The probability of wafer defects is evaluated by combining the mean function and the covariance function. Then, the defect areas are accumulated to obtain the total number and coordinates of wafer defects, and a heat map of the defect areas is generated.

9. The wafer defect detection method based on multimodal data fusion according to claim 8, characterized in that: The process of attribution and classification of defects is as follows: Set yield strength and temperature gradient threshold When the maximum stress > and temperature gradient > If so, it is determined that there are microcrack defects at the local stress critical points of the wafer; Set dislocation density threshold and plastic strain threshold When dislocation density Greater than And plastic strain > If so, it is determined that there is a dislocation accumulation defect at the local stress critical point of the wafer; This generates and labels the digital twin state vector as Vt: ; Where Dq represents the defect distribution; Pq represents the physical field distribution; and Cq represents the defect cause determination.

10. A wafer defect detection system based on multimodal data fusion, characterized in that: The system includes a multi-source data acquisition unit, a multi-physics modeling unit, a wafer integration inspection unit, and a defect interaction management unit. The units are interconnected and communicate with each other. When the system is applied, it executes the wafer defect detection method based on multimodal data fusion as described in any one of claims 1-9. The multi-source data acquisition unit is used to acquire multi-source detection data: the multi-source data acquisition unit includes a laser scattering module, an optical imaging module, and a timing synchronization module; The multiphysics modeling unit is used to construct multiphysics models: the multiphysics modeling unit includes a multimodal fusion physics field and wafer thermal stress coupling model, and the wafer thermal stress coupling model includes a laser-temperature field coupling sub-model and a temperature-stress field coupling sub-model. The wafer integrated inspection unit is used for wafer defect detection and attribution: The wafer integrated inspection unit includes a defect feature extraction module, a defect detection and localization module, a heat map generation module, and a cause deduction module; The Defect Interaction Management Unit is used for interactive display and early warning management: The Defect Interaction Management Unit includes a digital twin interactive interface, a defect visualization module, a defect early warning management module, and a process optimization design module.

Images