Wafer defect recognition method, system, device and medium based on local threshold calculation

Abstract

This invention relates to a wafer defect identification method, system, device, and medium based on local threshold calculation. It includes: employing an image differential defect identification method that considers local region thresholds to identify defects in a wafer image of a die to be identified, obtaining current defect candidate regions, and acquiring the image differential defect identification confidence score corresponding to the current defect candidate regions. The local region threshold is calculated based on the image parameters corresponding to the local region and the true / false defect ratio of the local region historically identified as a defect by the image differential defect identification method; based on the current die image to be identified, the current defect candidate regions, and a defect identification AI model, acquiring the current defect identification result and the model's defect identification confidence score; and based on the image differential defect identification confidence score and the model's defect identification confidence score, acquiring the final confidence score. This improves the reliability of the defect identification result and the defect identification confidence score, thereby ensuring the accuracy and reliability of wafer defect identification.

Description

Technical Field

[0001] Several embodiments of this specification relate to the field of semiconductor manufacturing, specifically to wafer defect identification methods, systems, devices, and media based on local threshold calculation. Background Technology

[0002] As semiconductor process nodes continue to evolve towards 3Xnm, 2Xnm, and even more advanced processes, the geometric feature sizes of wafer surface structures are significantly reduced, with linewidths and spacing approaching the nanometer scale. This makes the impact of minute defects in the wafer manufacturing process (such as particle contamination, pattern collapse, etching residue, and microcracks) on device performance and yield increasingly significant. Therefore, high-resolution, high-sensitivity pattern wafer defect detection technology has become an indispensable key link in semiconductor manufacturing lines.

[0003] Traditional wafer defect detection systems primarily rely on visible light or near-ultraviolet optical microscopy imaging, using contrast enhancement and fixed threshold algorithms for defect identification. However, with increasing process complexity, wafer surface structures become more diverse, and noise patterns more complex. Fixed threshold algorithms and traditional image difference methods have poor adaptability to different layers and process lines, easily leading to over-detection or under-detection.

[0004] To address these challenges, methods for intelligent defect identification using AI models have emerged. However, this approach typically relies on massive amounts of labeled samples, resulting in high training costs and limited generalization capabilities for specific process scenarios. Furthermore, AI models are often considered "black boxes," lacking interpretability in their reasoning mechanisms and being susceptible to image noise, leading to insufficient reliability and making it difficult to fully meet the stringent requirements of production lines for zero false alarms and low false alarms.

[0005] Therefore, how to build a wafer defect identification system that integrates traditional statistical features with the advantages of deep learning has become a pressing technical challenge. Summary of the Invention

[0006] This specification provides a wafer defect identification method, system, device, and medium for local threshold calculation. It utilizes a real-time feedback mechanism to dynamically calibrate the local detection threshold, effectively balancing the contradiction between over-detection and under-detection. It also uses the defect candidate region obtained by the image differential defect identification method as guidance information for AI model defect identification and combines the image differential defect identification confidence and the model defect identification confidence to perform multi-dimensional fusion confidence output, thereby improving the reliability of defect identification results and defect identification confidence, and thus ensuring the accuracy and reliability of wafer defect identification.

[0007] The technical solution is as follows:

[0008] In a first aspect, embodiments of this specification provide a wafer defect identification method based on local threshold calculation, including:

[0009] S1. Obtain wafer image;

[0010] S2. An image differential defect recognition method considering local region thresholds is adopted to perform defect recognition on the die image to be identified in the wafer image to obtain the current defect candidate region, and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region and the true and false ratio of the local region to the image differential defect recognition method in history.

[0011] S3. Based on the current defect image to be identified, the current defect candidate region, and the defect identification AI model, obtain the current defect identification result and the model's defect identification confidence.

[0012] S5. Based on the image difference defect recognition confidence score and the model defect recognition confidence score, obtain the final confidence score;

[0013] S6. Repeat steps S2 to S5 above to perform complete defect identification on the wafer image.

[0014] As a preferred embodiment, the image parameters include grayscale variance and texture complexity.

[0015] As a preferred embodiment, obtaining the local region threshold includes:

[0016] Based on the image parameters corresponding to the local region, obtain the initial threshold corresponding to the local region;

[0017] Based on the true / false ratio of defects identified as defects by the image difference defect recognition method in history corresponding to local regions, and the initial threshold corresponding to local regions, the image difference threshold corresponding to local regions is obtained.

[0018] As a preferred embodiment, obtaining the initial threshold corresponding to the local region based on the image parameters corresponding to the local region includes:

[0019] T = α * σ + β * (1 - C);

[0020] Where T represents the initial threshold corresponding to the local region, α and β are parameters obtained from offline training, σ represents the gray-level variance corresponding to the local region, and C represents the texture complexity corresponding to the local region.

[0021] As a preferred embodiment, the step of obtaining the image difference threshold corresponding to the local region based on the true / false ratio of defects historically identified as defects by the image difference defect recognition method and the initial threshold corresponding to the local region includes:

[0022] T' = (1-λ) * T + λ * Td;

[0023] Where T' represents the image difference threshold corresponding to the local region, T represents the initial threshold corresponding to the local region, and λ is the adaptive step size;

[0024] Wherein, Td is obtained based on the true and false ratios of defects identified as defects by the image differential defect recognition method in the local region in history.

[0025] As a preferred embodiment, the value of Td is inversely proportional to the value of Zy / Jy, where Zy represents the number of true positives of the local region that have been identified as defects by the image difference defect identification method in history, and Jy represents the number of false positives of the local region that have been identified as defects by the image difference defect identification method in history.

[0026] As a preferred embodiment, the step of obtaining the final confidence score based on image differential defect recognition confidence score and model defect recognition confidence score includes:

[0027] pfinal =γ* pai +(1-γ) * pstat;

[0028] Where pfinal represents the final confidence level, pai represents the model defect recognition confidence level, pstat represents the image difference defect recognition confidence level, and γ represents the dynamic adjustment factor, which is obtained based on the model defect recognition confidence level and the real-time performance of the defect recognition AI model.

[0029] Secondly, embodiments of this specification provide a wafer defect identification system based on local threshold calculation, used to implement the wafer defect identification method based on local threshold calculation described in the first aspect, including:

[0030] The image acquisition module acquires wafer images;

[0031] The differential defect recognition module uses an image differential defect recognition method that considers local region thresholds to perform defect recognition on the die image to be identified in the wafer image, so as to obtain the current defect candidate region and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region and the true and false ratio of the local region to the image differential defect recognition method in history.

[0032] The AI ​​defect recognition module obtains the current defect recognition result and the model's defect recognition confidence score based on the current image to be identified, the current defect candidate region, and the defect recognition AI model.

[0033] The confidence score acquisition module obtains the final confidence score based on the confidence score of image differential defect recognition and the confidence score of model defect recognition.

[0034] Thirdly, embodiments of this specification provide an electronic device, including a processor and a memory; the processor is connected to the memory; the memory is used to store executable program code; the processor reads the executable program code stored in the memory to run a program corresponding to the executable program code, so as to perform the steps described in the first aspect of the above embodiments.

[0035] Fourthly, embodiments of this specification provide a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the steps described in the first aspect of the above embodiments.

[0036] The beneficial effects of the technical solutions provided in some embodiments of this specification include at least the following:

[0037] By using a real-time feedback mechanism to dynamically calibrate local detection thresholds, the contradiction between over-detection and under-detection is effectively balanced. Furthermore, the defect candidate regions obtained by the image differential defect recognition method are used as guidance information for AI model defect recognition. By combining the confidence scores of image differential defect recognition and model defect recognition, a multi-dimensional fusion confidence score output is performed, thereby improving the reliability of defect recognition results and defect recognition confidence scores, and thus ensuring the accuracy and reliability of wafer defect recognition. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A schematic flowchart of a wafer defect identification method based on local threshold calculation according to some embodiments of the present disclosure is shown.

[0040] Figure 2 A schematic diagram of the structure of a wafer defect identification system for local threshold calculation according to some embodiments of the present disclosure is shown.

[0041] Figure 3 A schematic block diagram of an electronic device according to some embodiments of the present disclosure is shown. Detailed Implementation

[0042] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings.

[0043] The terms "first," "second," "third," etc., in the description, claims, and accompanying drawings are used to distinguish different objects, not to describe a specific order. 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 includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0044] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of this specification. Various processes or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0045] Figure 1 A flowchart illustrating a wafer defect identification method based on local threshold calculation according to some embodiments of this disclosure is shown. It should be understood that the numbers in the flowchart do not indicate the order in which these steps are performed; some or all of these steps can be performed in parallel, or their order can be interchanged, and this disclosure does not limit this. Furthermore, Figure 1 The methods described may also include additional steps not shown and / or the steps shown may be omitted, and the scope of this disclosure is not limited in this respect.

[0046] like Figure 1 As shown, the wafer defect identification method based on local threshold calculation can include at least:

[0047] S1. Obtain wafer image;

[0048] S2. An image differential defect recognition method considering local region thresholds (Note: die-to-die differential or die-to-database differential can be used) is employed to identify defects in the current die image to be identified in the wafer image, so as to obtain the current defect candidate region and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region (Note: the image parameters may include, but are not limited to, gray mean, gray variance, spectral center energy, and texture complexity, and these image parameters can be used as input to the defect recognition AI model in step S3 as reference information for its defect recognition) and the true / false defect ratio of the local region corresponding to the image differential defect recognition method in history (Note: the true / false defect situation of the local region corresponding to the image differential defect recognition method can be determined based on the AI ​​model described in step S3).

[0049] S3. Based on the current defect image to be identified, the current defect candidate region, and the defect identification AI model, obtain the current defect identification result and the model's defect identification confidence.

[0050] S4. Based on the image difference defect recognition confidence score and the model defect recognition confidence score, obtain the final confidence score;

[0051] S5. Repeat steps S2 to S4 to perform complete defect identification on the wafer image.

[0052] As described in several embodiments of this specification, a real-time feedback mechanism is used to dynamically calibrate the local detection threshold, effectively balancing the contradiction between over-detection and under-detection. Furthermore, the defect candidate region obtained by the image differential defect recognition method is used as guiding information for AI model defect recognition, and a multi-dimensional fusion confidence output is performed by combining the image differential defect recognition confidence and the model defect recognition confidence, thereby improving the reliability of defect recognition results and defect recognition confidence, and thus ensuring the accuracy and reliability of wafer defect recognition.

[0053] In some embodiments of this specification, obtaining the local region threshold includes:

[0054] Based on the image parameters corresponding to the local region, obtain the initial threshold corresponding to the local region;

[0055] Based on the true / false ratio of defects identified as defects by the image difference defect recognition method in history corresponding to local regions, and the initial threshold corresponding to local regions, the image difference threshold corresponding to local regions is obtained.

[0056] Understandable:

[0057] When using the die-to-database differential image defect recognition method, since its core logic is to compare the die image to be detected with a "perfect" standard image, it is only necessary to obtain the initial threshold corresponding to the local region once for the standard image.

[0058] When using the die-to-die differential image defect recognition method, the reference die image changes during the detection process. Therefore, it is necessary to obtain the initial threshold corresponding to the local region for each reference die image.

[0059] The step of obtaining the initial threshold corresponding to the local region based on the image parameters corresponding to the local region includes:

[0060] T = α * σ + β * (1 - C);

[0061] Where T represents the initial threshold corresponding to the local region, α and β are parameters obtained from offline training, σ represents the gray-level variance corresponding to the local region, and C represents the texture complexity corresponding to the local region.

[0062] It's important to note that the local grayscale variance reflects the degree of fluctuation in pixel values ​​within that region. The more complex the wafer structure, the greater the grayscale fluctuation, resulting in a higher background noise floor after differencing. Using a low threshold can easily misjudge these normal structural fluctuations as defects. Therefore, the formula links the local threshold to the corresponding grayscale variance of that region; the larger σ is, the more adaptively the initial threshold T increases. This design ensures high defect tolerance in "dirty" backgrounds, effectively suppressing background noise caused by the wafer structure itself.

[0063] It should also be noted that in wafer differential inspection, the more complex the texture, the richer the high-frequency details and edge information in the image. This directly exacerbates the sensitivity of alignment errors in the differential image, significantly enhancing structural ghosting caused by alignment errors. Simultaneously, complex texture environments often result in relatively weak signals from minute defects, which are easily obscured by these structural ghostings. Therefore, the formula introduces texture complexity C as a correction factor; the larger the C value (the more complex the texture), the more significant the effect of the term β*(1-C) on reducing the threshold. This design complements the adaptive mechanism based on gray-level variance, aiming to balance "background tolerance" and "defect sensitivity" by appropriately reducing the threshold, preventing the omission of weak but crucial minute defects due to excessively complex textures.

[0064] The step of obtaining the image difference threshold corresponding to a local region based on the true / false ratio of defects historically identified as defects by the image difference defect recognition method and the initial threshold corresponding to the local region includes:

[0065] T' = (1-λ) * T + λ * Td;

[0066] Where T' represents the image difference threshold corresponding to the local region, T represents the initial threshold corresponding to the local region, and λ is the adaptive step size (taken as 0.01~0.1).

[0067] Wherein, Td is obtained based on the true and false ratios of defects identified as defects by the image differential defect recognition method in the local region in history.

[0068] Let Zy represent the number of true positives identified as defects by the image differential defect recognition method in history for a local region, and Jy represent the number of false positives identified as defects by the same method in history (note: only the number of true positives and false positives within a preset backoff time can be referenced). When the Zy / Jy value is too small, it indicates an excessive number of false positives, meaning there are over-detections of defects, which increases the computational burden on the system during defect detection. Conversely, when the Zy / Jy value is too large, it indicates a low number of false positives, meaning there may be missed defects, which increases the probability of missed defects. Therefore, as described in several embodiments of this specification, a real-time feedback mechanism is used to dynamically calibrate the local detection threshold, effectively balancing the contradiction between over-detection and missed detection. Therefore, the Zy / Jy value should be inversely proportional to the Td value.

[0069] The process of obtaining the final confidence score based on image differential defect recognition confidence score and model defect recognition confidence score includes:

[0070] pfinal =γ* pai +(1-γ) * pstat;

[0071] Where pfinal represents the final confidence level, pai represents the model defect recognition confidence level, pstat represents the image difference defect recognition confidence level, and γ represents the dynamic adjustment factor. γ is obtained based on the model defect recognition confidence level and the real-time performance of the defect recognition AI model (Note: The model defect recognition confidence level is directly proportional to γ; that is, the higher the model defect recognition confidence level, the larger γ is; the real-time performance of the defect recognition AI model is directly proportional to γ; that is, the better the real-time performance of the defect recognition AI model, the larger γ is).

[0072] As is understood, in the embodiments of this specification, the confidence level of image differential defect recognition and the confidence level of model defect recognition are combined to perform multi-dimensional fusion confidence output, thereby improving the reliability of defect recognition results and defect recognition confidence, and thus ensuring the accuracy and reliability of wafer defect recognition.

[0073] Among them, the confidence level of image differential defect identification can be obtained based on data such as the degree to which the differential value exceeds the local threshold and the image area of ​​the defect.

[0074] The following provides a more detailed description of the wafer defect identification process using the local threshold calculation method described in the embodiments of this specification. Key implementation steps may include:

[0075] Step 1: System initialization (Power-on self-test)

[0076] The motion platform is zeroed, and the encoder is calibrated; the light source is preheated (at a constant temperature); the camera performs a self-test (black level / bright level); after the hardware self-test passes, the current software version and algorithm model (model_hash) are loaded.

[0077] Step 1: Load the task and parameters (Operator)

[0078] The operator enters wafer_id and recipe (including exposure, line speed, focus parameters, and detection strategy) in the GUI.

[0079] Step 2: Motion & Alignment

[0080] The motion controller performs alignment to the initial die, and the Z-axis automatically focuses in a cycle (averaging multiple calibration points). If the focusing error exceeds the preset value, fine-tuning is triggered.

[0081] Once alignment is complete, scanning begins, and the motion controller issues a hardware trigger (TTL) to synchronize camera line acquisition.

[0082] Motion control employs a high-precision motion platform (XY-axis air-bearing platform + Z-axis autofocus), with an FPGA / real-time CPU responsible for trajectory control, trigger signal generation, and high-precision time synchronization.

[0083] Among them, the XY axis position control accuracy is ≤50 nm; dual encoder closed-loop feedback is adopted; and the maximum linear speed is ≥500 mm / s.

[0084] Among them, Z-axis autofocus: based on interferometric focusing or confocal sensor, response time ≤1 ms, compensation accuracy ≤20nm.

[0085] Step 3: Real-time Acquisition

[0086] The TDI camera sends out parallel data line by line. The FPGA performs line merging in real time and writes the data to the host buffer via PCIeDMA (the tag includes a PTP timestamp and encoder position information).

[0087] A preprocessing batch is triggered every N rows (N is set according to the tile size, typically N=256 rows).

[0088] Step 4: Preprocessing and Candidate Generation

[0089] The preprocessing service reads the buffer and performs flat field correction, noise reduction, brightness normalization, and TDI row alignment.

[0090] The candidate generation service performs frequency domain energy analysis and differential analysis, and outputs a candidate list to the AI ​​inference queue.

[0091] Step 5: AI Inference and Result Fusion

[0092] The AI ​​inference service loads the corresponding inference model and performs classification and regression (boundary refinement) on each candidate patch.

[0093] The final defect terms are merged by combining morphological post-processing (removing noise smaller than 3 pixels) and multimodal fusion (frequency domain + spatial features).

[0094] Note: Steps 4 and 5 refer to performing the wafer defect detection using the local threshold calculation method described in several embodiments of this specification.

[0095] Step 6: Anomaly Detection and Protection

[0096] If frame loss occurs, trigger offset occurs, or temperature exceeds limits, the system will automatically enter a safe shutdown, stop scanning, save cached data, switch to safe mode, and notify the operator.

[0097] Automatic diagnostic logs (including PTP time summaries and encoder trajectories) are available for subsequent traceability.

[0098] Step 7: Offline Training and Model Deployment

[0099] At night or under low load, the system pushes the labeled data in main memory to the offline training platform for full training and generates a new model (versioned).

[0100] Once the new model has been validated (regression test) and passed, it will be automatically deployed to the production node after manual approval (supporting A / B upgrades and rollbacks).

[0101] The aforementioned tasks, such as data collection and processing, AI inference, and result persistence, are handled using a distributed computing node + task scheduling model to improve data processing efficiency.

[0102] Figure 2Schematic diagrams of a wafer defect identification system based on local threshold calculation according to some embodiments of this disclosure are shown. The various embodiments in this specification are described in a progressive manner, and similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of the wafer defect identification system based on local threshold calculation are basically similar to the embodiments of the wafer defect identification method based on local threshold calculation, so the description is relatively simple; relevant parts can be referred to in the description of the embodiments of the wafer defect identification method based on local threshold calculation.

[0103] like Figure 2 As shown, a wafer defect identification system based on local threshold calculation can include at least:

[0104] The image acquisition module acquires wafer images;

[0105] The differential defect recognition module uses an image differential defect recognition method that considers local region thresholds to perform defect recognition on the die image to be identified in the wafer image, so as to obtain the current defect candidate region and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region and the true and false ratio of the local region to the image differential defect recognition method in history.

[0106] The AI ​​defect recognition module obtains the current defect recognition result and the model's defect recognition confidence score based on the current image to be identified, the current defect candidate region, and the defect recognition AI model.

[0107] The confidence score acquisition module obtains the final confidence score based on the confidence score of image differential defect recognition and the confidence score of model defect recognition.

[0108] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0109] Figure 3 A block diagram of an electronic device 300 that can implement various embodiments of the present disclosure is shown. For example... Figure 3 As shown, the electronic device 300 includes a processor 310, a disk drive 320, an input / output interface 330, a network interface 340, and a memory 350. The processor 310, disk drive 320, input / output interface 330, network interface 340, and memory 350 can communicate with each other via a communication bus 360.

[0110] The processor 310 can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs in order to implement the technical solution provided in this application.

[0111] The memory 350 can be implemented in the form of ROM (Read Only Memory), RAM (Read Access Memory), static memory, dynamic storage devices, etc. The memory 350 can store the operating system 351 used to control the operation of the electronic device 300, and the basic input / output system (BIOS) 352 used to control the low-level operations of the electronic device 300. Additionally, it can store a web browser 353, a data storage management system 354, etc. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 350 and is called and executed by the processor 310.

[0112] Input / output interface 330 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0113] Network interface 340 is used to connect a communication module (not shown in the figure) to enable communication and interaction between the device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0114] Bus 360 includes a pathway for transmitting information between various components of the device, such as processor 310, disk drive 320, input / input interface 330, network interface 340, and memory 350.

[0115] It should be noted that although the above-described device only shows the processor 310, disk drive 320, input / output interface 330, network interface 340, memory 350, bus 360, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the method of this application, and does not necessarily include all the components shown in the figures.

[0116] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. Furthermore, although operations are depicted in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0118] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A wafer defect identification method based on local threshold calculation, characterized in that, include: S1. Obtain wafer image; S2. An image differential defect recognition method considering local region thresholds is adopted to perform defect recognition on the die image to be identified in the wafer image to obtain the current defect candidate region, and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region and the true and false ratio of the local region to the image differential defect recognition method in history. S3. Based on the current defect image to be identified, the current defect candidate region, and the defect identification AI model, obtain the current defect identification result and the model's defect identification confidence. S4. Based on the image difference defect recognition confidence score and the model defect recognition confidence score, obtain the final confidence score; S5. Repeat steps S2 to S4 to perform complete defect identification on the wafer image.

2. The wafer defect identification method based on local threshold calculation according to claim 1, characterized in that, The image parameters include grayscale variance and texture complexity.

3. The wafer defect identification method based on local threshold calculation according to claim 1, characterized in that, Obtaining the local region threshold includes: Based on the image parameters corresponding to the local region, obtain the initial threshold corresponding to the local region; Based on the true / false ratio of defects identified as defects by the image difference defect recognition method in history corresponding to local regions, and the initial threshold corresponding to the local regions, the image difference threshold corresponding to the local regions is obtained.

4. The wafer defect identification method based on local threshold calculation according to claim 3, characterized in that, The step of obtaining the initial threshold corresponding to the local region based on the image parameters corresponding to the local region includes: T = α * σ + β * (1 - C); Where T represents the initial threshold corresponding to the local region, α and β are parameters obtained from offline training, σ represents the gray-level variance corresponding to the local region, and C represents the texture complexity corresponding to the local region.

5. The wafer defect identification method based on local threshold calculation according to claim 3, characterized in that, The step of obtaining the image difference threshold corresponding to the local region based on the true / false ratio of defects historically identified as defects by the image difference defect recognition method corresponding to the local region and the initial threshold corresponding to the local region includes: T' = (1-λ) * T + λ * Td; Where T' represents the image difference threshold corresponding to the local region, T represents the initial threshold corresponding to the local region, and λ is the adaptive step size; Wherein, Td is obtained based on the true and false ratios of defects identified as defects by the image differential defect recognition method in the local region in history.

6. The wafer defect identification method based on local threshold calculation according to claim 5, characterized in that, The value of Td is inversely proportional to the value of Zy / Jy, where Zy represents the number of true positives in the local region that have been identified as defects by the image difference defect identification method in history, and Jy represents the number of false positives in the local region that have been identified as defects by the image difference defect identification method in history.

7. The wafer defect identification method based on local threshold calculation according to claim 3, characterized in that, The process of obtaining the final confidence score based on image differential defect recognition confidence score and model defect recognition confidence score includes: pfinal =γ* pai +(1-γ) * pstat; Where pfinal represents the final confidence level, pai represents the model defect recognition confidence level, pstat represents the image difference defect recognition confidence level, and γ represents the dynamic adjustment factor, which is obtained based on the model defect recognition confidence level and the real-time performance of the defect recognition AI model.

8. A wafer defect identification system based on local threshold calculation, wherein the wafer defect identification method based on local threshold calculation as described in any one of claims 1 to 7 is characterized in that, include: The image acquisition module acquires wafer images; The differential defect recognition module uses an image differential defect recognition method that considers local region thresholds to perform defect recognition on the die image to be identified in the wafer image, so as to obtain the current defect candidate region and obtain the image differential defect recognition confidence corresponding to the current defect candidate region. The local region threshold is calculated based on the image parameters corresponding to the local region and the true and false ratio of the local region to the image differential defect recognition method in history. The AI ​​defect recognition module obtains the current defect recognition result and the model's defect recognition confidence score based on the current image to be identified, the current defect candidate region, and the defect recognition AI model. The confidence score acquisition module obtains the final confidence score based on the confidence score of image differential defect recognition and the confidence score of model defect recognition.

9. An electronic device, characterized in that, include: One or more processors, and A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method according to any one of claims 1-7.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images