Defect recognition method and device based on multi-modal data of chip manufacturing process

Abstract

This invention discloses a defect identification method and apparatus based on multimodal data from the chip fabrication process. The method includes: after epitaxial growth, acquiring a graphite carrier image and generating a graphite carrier-wafer mapping map; performing multimodal imaging on the wafer under test, acquiring grayscale images, reflectivity images, and photoluminescence intensity images and synchronizing them at the pixel level, performing normalization processing on each, and calculating a physical normalization factor based on the normalization results to obtain the multimodal image; constructing a wafer grid space database, storing the multimodal image, electrical test data, and abnormal regions in the graphite carrier image into the wafer grid space database; and identifying defect types according to preset rules based on the multi-source data of the mapped multimodal image, electrical test data, and abnormal regions in the graphite carrier image. This invention solves the problem of low efficiency in defect root cause analysis in the prior art due to the independent nature of the data generated during the chip fabrication process.

Description

Technical Field

[0001] This invention relates to the field of chip manufacturing technology, and in particular to a defect identification method and apparatus based on multimodal data from the chip fabrication process. Background Technology

[0002] Micro-LED (micro-light-emitting diode) technology is widely recognized as a key enabling technology for next-generation displays and optoelectronic integration due to its ultra-high brightness, ultra-low power consumption, nanosecond-level response speed, and excellent environmental stability. It is widely used in augmented reality (AR), virtual reality (VR), automotive head-up displays, wearable devices, and micro-projection. However, despite its significant performance advantages, the large-scale industrialization of Micro-LEDs still faces severe challenges. One of the most critical bottlenecks is the difficulty in improving manufacturing yield, and the root cause of the yield problem often lies in the data fragmentation across processes and equipment, as well as the lack of unified spatial standards.

[0003] In the current Micro-LED manufacturing process, a GaN-based epitaxial wafer needs to go through several key process steps in sequence, including metal-organic chemical vapor deposition (MOCVD) epitaxial growth, photolithography, dry etching, transparent electrode preparation, laser lift-off, mass transfer, reflow soldering, probe electrical testing, and automated optical inspection (AOI). Each step is performed by specialized equipment and generates data for specific modes: MOCVD equipment records growth parameters (such as temperature, V / III ratio, and rotation rate), and some advanced production lines can perform in-situ imaging of the graphite susceptor surface at the wafer unloading station; photoluminescence (PL) outputs wafer-level luminescence intensity and wavelength distribution maps; AOI system acquires high-resolution grayscale binary images to identify morphological defects such as surface cracks, scratches, particle contamination, fogging defects, and metal bridging; and the probe station provides electrical parameters for each Micro-LED chip (typically 3–200 μm in size), including forward voltage (VF2), reverse leakage current (leakage current Ir), and turn-on voltage (VF1).

[0004] However, these devices generally use their own independent coordinate systems and spatial reference bases. This "spatial island" phenomenon means that even defects at the same physical location cannot be accurately correlated in the data from different devices. When a chip is determined to have a leakage failure during electrical testing, engineers cannot automatically trace back whether that location shows abnormal luminescence in the PL diagram, surface contamination in the AOI diagram, or a corresponding damaged area on the MOCVD graphite substrate. As a result, defect root cause analysis heavily relies on manual experience, with debugging cycles lasting several days and low accuracy. This severely restricts the efficiency of defect root cause analysis and the ability to optimize the process closed loop. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a defect identification method and apparatus based on multimodal data of the chip manufacturing process, which aims to solve the problem of low efficiency of defect root cause analysis in the prior art due to the independent nature of the data generated during the chip manufacturing process.

[0006] The embodiments of the present invention are implemented as follows:

[0007] A defect identification method based on multimodal data from the chip fabrication process, the method comprising:

[0008] After epitaxial growth is completed, the surface of the graphite carrier is photographed in situ using an industrial camera integrated into the wafer unloading machine to obtain images of the graphite carrier, and the placement azimuth angle of the wafer on the graphite carrier is read from the epitaxial growth equipment controller.

[0009] Based on the graphite carrier image and placement azimuth, a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established, and a graphite carrier-wafer mapping diagram is generated.

[0010] Multimodal imaging is performed on the wafer under test. Grayscale images, reflectance images, and photoluminescence intensity images are acquired and synchronized at the pixel level. The three-channel images of grayscale images, reflectance images, and photoluminescence intensity images are normalized respectively. The physical normalization factor is calculated based on the normalization results to obtain the multimodal image.

[0011] The wafer surface is divided into grid cells of a preset size, and a unique grid identifier is assigned to each grid cell to build a wafer grid space database. Abnormal areas in multimodal images, electrical test data and graphite carrier images are mapped to the corresponding grid identifiers and stored in the wafer grid space database.

[0012] Based on multi-source data from the mapped multimodal images, electrical test data, and abnormal regions in the graphite carrier images, the defect type is identified according to preset rules.

[0013] Furthermore, in the aforementioned defect identification method based on multimodal data from the chip fabrication process, the step of establishing the spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system includes:

[0014] The SIFT feature extraction algorithm was used to identify the contact points between the positioning holes on the graphite carrier and the wafer edge as control point pairs.

[0015] The RANSAC algorithm is used to filter control point pairs and eliminate mismatched points.

[0016] Based on the selected effective control point pairs, the initial affine transformation matrix is ​​obtained by solving, and the spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established according to the initial affine transformation matrix.

[0017] Furthermore, in the aforementioned defect identification method based on multimodal data from the chip fabrication process, the step of normalizing the three-channel images (grayscale image, reflectance image, and photoluminescence intensity image) and calculating the physical normalization factor based on the normalization results to obtain the multimodal image includes:

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024] in, To obtain the normalized grayscale value, For normalized reflectivity, For normalized photoelectric, It is a grayscale image. For reflectance images, Image of photoluminescence intensity. , , , These represent the dynamic extreme values ​​of the current batch of wafers in the corresponding channels. As the physical normalization factor, =0.001 is used to prevent division by zero. This is a numerical truncation operation.

[0025] Furthermore, the aforementioned defect identification method based on multimodal data from the chip fabrication process further includes, before the step of normalizing the three-channel images (grayscale image, reflectance image, and photoluminescence intensity image), the following:

[0026] Obtained by scanning the edge region of the current batch of wafers or a reference standard. , , , Real-time statistics.

[0027] Furthermore, in the aforementioned defect identification method based on multimodal data from the chip fabrication process, the step of assigning a unique grid identifier to each grid cell includes:

[0028] =

[0029] in, For grid cells in wafer coordinate system X Directional coordinates For grid cells in wafer coordinate system Y Directional coordinates This indicates the floor function; For wafers Y Total number of grid cells in each direction.

[0030] Furthermore, the defect identification method based on multimodal data from the chip fabrication process described above further includes:

[0031] Collect at least N historical high-quality wafers GRP-NF Distribution data is used to form a sample set of good products;

[0032] Calculate the good product sample set GRP-NF The mean and standard deviation, according to GRP-NF Set dynamic thresholds for the mean and standard deviation;

[0033] Apply dynamic thresholds to the current batch of wafers. GRP-NF In the process of judgment, the normal and abnormal areas are distinguished.

[0034] Furthermore, in the aforementioned defect identification method based on multimodal data from the chip fabrication process, the step of identifying the defect type according to preset rules includes:

[0035] When there is an anomaly in the graphite carrier image display area A and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be a graphite carrier defect, and the carrier is taken offline for maintenance and repair.

[0036] When the gradient difference of the grayscale image is greater than the first threshold and is located in the wafer edge region, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too low, and the MOCVD temperature field increase process is triggered.

[0037] When the gradient difference of the grayscale image is greater than the first threshold and is located in the center region of the wafer, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too high, and the MOCVD temperature field reduction process is triggered.

[0038] When the physical normalization factor is less than the low threshold, the defect morphology is amorphous, and the leakage current Ir of the corresponding grid cell is normal, it is determined to be wafer surface contamination, and the wafer cleaning process is triggered.

[0039] When the physical normalization factor is less than the low threshold, the defect morphology is circular / ring-shaped, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the MOCVD cavity is shedding dust, and the MOCVD maintenance process is triggered.

[0040] When the physical normalization factor is greater than the high threshold and the intensity of the photoluminescence intensity image is lower than the first preset threshold and the Vf of the corresponding grid cell deviates from the threshold, it is determined that there is a leak in the MOCVD cavity, and the MOCVD hydrogen detection process is triggered.

[0041] When the gradient difference of the grayscale image is greater than the second threshold and there are abnormalities at both the center and edge of the wafer, and the leakage current Ir of the corresponding grid cell is generally abnormal, it is determined that the graphite carrier is aging or the temperature control is malfunctioning, and the carrier replacement and MOCVD calibration process are triggered.

[0042] When the physical normalization factor is less than the low threshold, the intensity of the photoluminescence intensity image is lower than the second preset threshold, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the graphite carrier has started to break down, and the carrier is taken offline for maintenance and repair.

[0043] When the physical normalization factor is less than the low threshold and the defect is linear / striped and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be wafer scratch contamination and the wafer pre-placement inspection process is triggered.

[0044] When the physical normalization factor is greater than the high threshold and the brightness of the grayscale image is higher than the preset threshold, but the intensity of the photoluminescence intensity image decreases and the leakage current Ir is normal, it is determined that the wafer surface reflection is enhanced, and the epitaxial layer thickness detection process is triggered.

[0045] Another object of the present invention is to provide a defect identification device based on multimodal data of the chip fabrication process, the device comprising:

[0046] The acquisition module is used to capture images of the graphite carrier surface in situ using an industrial camera integrated into the wafer unloading machine after epitaxial growth is completed, and to read the placement azimuth angle of the wafer on the graphite carrier from the epitaxial growth equipment controller.

[0047] A module is established to create a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system based on the graphite carrier image and placement azimuth angle, and to generate a graphite carrier-wafer mapping diagram.

[0048] The imaging module is used to perform multimodal imaging on the wafer under test, acquire grayscale images, reflectance images and photoluminescence intensity images and synchronize them at the pixel level, normalize the three-channel images of grayscale images, reflectance images and photoluminescence intensity images respectively, and calculate the physical normalization factor based on the normalization results to obtain the multimodal image.

[0049] The storage module is used to divide the wafer surface into grid cells of a preset size and assign a unique grid identifier to each grid cell to build a wafer grid space database. Abnormal areas in multimodal images, electrical test data and graphite carrier images are mapped to the corresponding grid identifiers and stored in the wafer grid space database.

[0050] The identification module is used to identify the defect type according to preset rules based on multi-source data of abnormal areas in the mapped multimodal images, electrical test data, and graphite carrier images.

[0051] Another object of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0052] Another object of the present invention is to provide an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0053] This invention involves acquiring images of the graphite carrier surface through in-situ imaging using an industrial camera integrated into the wafer unloading machine after epitaxial growth is completed, and reading the wafer's placement azimuth angle on the graphite carrier from the epitaxial growth equipment controller. Based on the graphite carrier image and placement azimuth angle, a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established, generating a graphite carrier-wafer mapping map. Multimodal imaging is performed on the wafer under test, acquiring grayscale images, reflectance images, and photoluminescence intensity images and synchronizing them at the pixel level. The three-channel images are normalized separately, and the physical normalization factor is calculated based on the normalization results to obtain multimodal images. The wafer surface is divided into grid cells of a preset size, and a unique grid identifier is assigned to each grid cell to construct a wafer grid spatial database. Abnormal regions in the multimodal images, electrical test data, and graphite carrier images are mapped to their corresponding grid identifiers and stored in the wafer grid spatial database. Based on the multi-source data of abnormal regions in the mapped multimodal images, electrical test data, and graphite carrier images, defect types are identified according to preset rules. By mapping data from heterogeneous equipment such as MOCVD, AOI, PL, and probe stations to the same physical coordinate system and then storing it in a chip-level high-density spatial database, the analysis efficiency of Micro-LED manufacturing is significantly improved. This solves the problem of low efficiency in defect root cause analysis in existing technologies due to the independent nature of data generated during chip fabrication. Attached Figure Description

[0054] Figure 1 This is a flowchart of the defect identification method based on multimodal data of the chip fabrication process in the first embodiment of the present invention;

[0055] Figure 2 This is a structural block diagram of a defect identification device based on multimodal data from the chip fabrication process, as shown in the third embodiment of the present invention.

[0056] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0057] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0058] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0060] Example 1

[0061] Please see Figure 1 The figure shows a defect identification method based on multimodal data of chip fabrication process in the first embodiment of the present invention, the method including steps S10 to S14.

[0062] Step S10: After the epitaxial growth is completed, the graphite carrier surface is photographed in situ using an industrial camera integrated into the wafer unloading machine to obtain an image of the graphite carrier, and the placement azimuth angle of the wafer on the graphite carrier is read from the epitaxial growth equipment controller.

[0063] After epitaxial growth is completed, an industrial camera integrated into the wafer unmounter is used to take in-situ images of the graphite carrier surface. This imaging method does not require modification of the reaction chamber of the epitaxial growth equipment and can directly acquire images of the carrier surface in real condition before the wafer and carrier are separated. At the same time, the placement azimuth angle of the wafer on the graphite carrier is read from the epitaxial growth equipment controller through the SECS / GEM or OPC UA protocol. This azimuth angle can reflect the placement posture of the wafer on the carrier and provide key posture parameters for subsequent spatial coordinate alignment.

[0064] Step S11: Based on the graphite carrier image and placement azimuth angle, establish the spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system, and generate a graphite carrier-wafer mapping diagram.

[0065] Based on the graphite carrier image and azimuth angle, a spatial mapping relationship between the two coordinate systems is established, and a graphite carrier-to-wafer mapping map is generated. This mapping map can be used to accurately associate abnormal areas of the carrier with the corresponding locations on the wafer.

[0066] Specifically, the SIFT feature extraction algorithm is used to identify control point pairs: "Location pin holes" (an inherent reference structure of the carrier, with fixed positions and significant features) and "wafer edge contact points" (the physical contact boundary between the wafer and the carrier, which can be determined through image edge detection) on the graphite carrier are selected as control point pairs. The SIFT algorithm possesses scale invariance and rotation invariance, enabling stable feature point extraction even under conditions of slight contamination on the carrier surface and varying illumination, avoiding feature matching failures. The RANSAC algorithm is then applied to screen the control point pairs: Image noise, scratches on the carrier surface, or rough wafer edges may lead to "false matching points" (such as misidentifying carrier surface impurities as location hole features). The RANSAC algorithm establishes an initial mapping model by randomly sampling some control point pairs, then calculates the deviation between all control point pairs and the model. Points with deviations less than a threshold are identified as "interior points" (valid matching points). After repeated iterations, the model with the most interior points is selected, thereby eliminating false matching points and ensuring the reliability of the control point pairs. Finally, the initial affine transformation matrix is ​​solved based on the effective control point pairs: the effective control point pairs are fitted by the least squares method to solve the affine transformation matrix containing translation, rotation, and scaling parameters. If non-rigid deformations such as minor warping of the carrier are detected (e.g., slight deformation of the carrier caused by the high temperature of MOCVD), the thin plate spline (TPS) model will be further used to optimize the affine transformation matrix, ultimately making the spatial registration error less than ±3μm, ensuring high-precision mapping between the graphite carrier coordinate system and the wafer coordinate system.

[0067] Step S12: Perform multimodal imaging on the wafer under test, acquire grayscale image, reflectance image and photoluminescence intensity image and synchronize them at the pixel level, normalize the three channels of grayscale image, reflectance image and photoluminescence intensity image respectively, and calculate the physical normalization factor based on the normalization result to obtain the multimodal image.

[0068] When performing multimodal imaging on the wafer under test, a "multimodal synchronous imaging module" (integrated into the AOI inspection station or a dedicated calibration platform) is used, consisting of an industrial camera, a ring-shaped white light source, a 365nm ultraviolet laser excitation source, a filter group, and a trigger controller. The ring-shaped white light source is used for imaging grayscale images (Gray) and reflectance images (Ref), the 365nm ultraviolet laser excitation source is used for imaging photoluminescence intensity images (PD), the filter group can filter out stray light to ensure image purity, and the trigger controller is synchronized with the wafer stage to ensure that the three-channel images are synchronized at the pixel level (i.e., the same pixel corresponds to the same physical position on the wafer). Subsequently, the three-channel images are normalized separately, and then the physical normalization factor is calculated and fused to finally obtain a multimodal image that integrates multimodal information.

[0069] Step S13: Divide the wafer surface into grid cells of a preset size, assign a unique grid identifier to each grid cell, construct a wafer grid space database, and map abnormal regions in multimodal images, electrical test data, and graphite carrier images to the corresponding grid identifiers for storage in the wafer grid space database.

[0070] Specifically, the wafer surface is divided into preset 0.1mm × 0.1mm grid cells (adapting to the 3-200μm size range of Micro-LED chips, balancing accuracy and computational efficiency). Each cell is assigned a unique grid identifier (grid cell ID), and a "Wafer Grid Spatial Database (WGSDB)" based on key-value pairs or spatial indexes (such as R-tree) is constructed. This database is deployed on a distributed storage system, using the grid cell ID as the core index to decouple the native coordinate systems of different devices. Then, the aforementioned multimodal images (mapped to grid cell IDs by pixels), electrical test data (such as leakage current Ir and forward voltage Vf, which are obtained from the probe station and associated with grid cell IDs by chip), and abnormal regions in the graphite carrier images (mapped to grid cell IDs through a Susceptor-to-Wafer Mapping Map) are all stored in the WWGSDB to achieve spatially aligned storage.

[0071] Furthermore, the step of assigning a unique grid identifier to each grid cell includes:

[0072] =

[0073] in, For grid cells in wafer coordinate system X Directional coordinates For grid cells in wafer coordinate system Y Directional coordinates This indicates the floor function; For wafers Y Total number of grid cells in each direction.

[0074] The grid cell ID is calculated using a formula. This encoding rule ensures that each 0.1mm × 0.1mm grid cell has a unique grid cell ID by using the formula "X-direction index × total number of grid cells in the Y-direction + Y-direction index". Furthermore, the cell coordinates can be derived from the grid cell ID, providing efficient support for WGSDB indexing and data retrieval.

[0075] Step S14: Based on the multi-source data of abnormal areas in the mapped multimodal image, electrical test data, and graphite carrier image, the defect type is identified according to preset rules.

[0076] Among them, based on the multi-source data mapped in WGSDB, the root cause reasoning module of the built-in structured reasoning rule library is called to comprehensively judge and identify the defect type according to the preset rules, thus completing the closed loop of the whole link from data collection to defect judgment.

[0077] Specifically, when there is an anomaly in the graphite carrier image display area A and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be a graphite carrier defect, and the carrier is taken offline for maintenance and repair.

[0078] When the gradient difference of the grayscale image is greater than the first threshold and is located in the wafer edge region, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too low, and the MOCVD temperature field increase process is triggered.

[0079] When the gradient difference of the grayscale image is greater than the first threshold and is located in the center region of the wafer, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too high, and the MOCVD temperature field reduction process is triggered.

[0080] When the physical normalization factor is less than the low threshold, the defect morphology is amorphous, and the leakage current Ir of the corresponding grid cell is normal, it is determined to be wafer surface contamination, and the wafer cleaning process is triggered.

[0081] When the physical normalization factor is less than the low threshold, the defect morphology is circular / ring-shaped, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the MOCVD cavity is shedding dust, and the MOCVD maintenance process is triggered.

[0082] When the physical normalization factor is greater than the high threshold and the intensity of the photoluminescence intensity image is lower than the first preset threshold and the Vf of the corresponding grid cell deviates from the threshold, it is determined that there is a leak in the MOCVD cavity, and the MOCVD hydrogen detection process is triggered.

[0083] When the gradient difference of the grayscale image is greater than the second threshold and there are abnormalities at both the center and edge of the wafer, and the leakage current Ir of the corresponding grid cell is generally abnormal, it is determined that the graphite carrier is aging or the temperature control is malfunctioning, and the carrier replacement and MOCVD calibration process are triggered.

[0084] When the physical normalization factor is less than the low threshold, the intensity of the photoluminescence intensity image is lower than the second preset threshold, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the graphite carrier has started to break down, and the carrier is taken offline for maintenance and repair.

[0085] When the physical normalization factor is less than the low threshold and the defect is linear / striped and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be wafer scratch contamination and the wafer pre-placement inspection process is triggered.

[0086] When the physical normalization factor is greater than the high threshold and the brightness of the grayscale image is higher than the preset threshold, but the intensity of the photoluminescence intensity image decreases and the leakage current Ir is normal, it is determined that the wafer surface reflection is enhanced, and the epitaxial layer thickness detection process is triggered.

[0087] Examples include the following defects:

[0088] Graphite carrier defect determination: When there is an anomaly in the graphite carrier image display area A (such as damage or contamination, identified by image edge detection or grayscale contrast), and the leakage current Ir of the corresponding grid cell after mapping this area by the Susceptor-to-Wafer Mapping Map is greater than the standard value (such as leakage current Ir > 10μA), it is determined to be a "graphite carrier defect" - the mechanism is that the carrier abnormality will cause the crystal quality of the epitaxial layer in the corresponding wafer area to decrease, resulting in leakage current; the triggering process is "carrier offline maintenance and repair", and the instruction is pushed to the wafer unloading machine through the MES system to suspend the use of the carrier.

[0089] Wafer temperature field anomaly determination: The first threshold is set to 30 (calibrated based on historical temperature field anomaly data). When the gradient difference of the Gray image > 30 (the gradient difference reflects the gray change rate, uneven temperature field will cause differences in the thickness of the epitaxial layer, and then manifest as abnormal gray gradient), and the abnormal area is located at the edge of the wafer and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined as "low wafer temperature field" - the edge temperature field is easily affected by the cavity heat dissipation, and being low will cause slow epitaxial growth rate and poor crystal quality; trigger the "MOCVD temperature field increase process", and increase the power of the edge heating tube through the equipment controller; if the abnormal area is located in the center of the wafer, it is determined as "high wafer temperature field", trigger the "MOCVD temperature field decrease process", and reduce the power of the center heating tube.

[0090] Wafer surface contamination and MOCVD cavity dust drop determination: When GRP-NF < Threshold_low (the physical normalization factor is less than the low threshold), if the defect morphology has no fixed form (such as irregular spots, identified through AOI image morphology analysis) and the leakage current Ir of the corresponding grid cell is normal (contamination only affects the surface and does not damage the internal structure of the chip), it is determined as "wafer surface contamination", and trigger the "wafer cleaning process" (such as plasma cleaning); if the defect morphology is circular / ring-shaped (the dust drop particles in the cavity block the laser and form a circular shadow) and the leakage current Ir deviates from the threshold (dust drop will cause local absence of the epitaxial layer and lead to leakage), it is determined as "MOCVD cavity dust drop", and trigger the "MOCVD maintenance process" (cavity disassembly and cleaning, filter replacement).

[0091] MOCVD cavity leakage determination: When GRP-NF > Threshold_high (the physical normalization factor is greater than the high threshold), and the PD intensity < 80% of the average value of good products (cavity leakage will cause abnormal hydrogen / ammonia concentration and affect the growth of the light-emitting layer), and the forward voltage Vf of the corresponding grid cell > the standard value (such as Vf > 3.5V, leakage causes uneven doping of the epitaxial layer and deteriorates the electrical parameters), it is determined as "MOCVD cavity has leakage", and trigger the "MOCVD helium detection process" (detect the leakage point through a helium mass spectrometer leak detector).

[0092] Carrier overall aging or temperature control failure determination: The second threshold is set to 50 (more strict than the first threshold, corresponding to a serious temperature field anomaly). When the Gray gradient difference > 50, and the abnormal area covers the center and edge of the wafer (the overall temperature field is out of control), and the leakage current Ir of the corresponding grid cells is generally abnormal (> 80% of the leakage current Ir of the abnormal grid cells deviates from the threshold), it is determined as "graphite carrier overall aging or temperature control failure" - carrier aging will cause uneven heat conduction, and temperature control failure will cause global temperature field disorder; trigger the "carrier replacement + MOCVD calibration process", and recalibrate the temperature sensor and heating system of MOCVD after replacing the carrier.

[0093] Initial determination of vehicle damage and wafer scratch: When GRP-NF < Threshold_low (physical normalization factor is less than the low threshold), and the local PD intensity < 80% of the average value of good products (vehicle damage causes local temperature abnormality, affecting PL luminescence), and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined as "the graphite vehicle starts to be damaged" (only the local area is affected in the initial stage of damage), and the "vehicle offline maintenance and repair process" is triggered; if the defect is linear / striped (linear gray abnormality caused by scratch) and the leakage current Ir deviates from the threshold (the scratch damages the chip electrode or epitaxial layer), it is determined as "wafer scratch and contamination", and the "wafer inspection process before loading" is triggered (manual re-inspection or enhanced AOI scanning to identify the source of the scratch).

[0094] Determination of enhanced reflection on the wafer surface: When GRP-NF > Threshold_high (physical normalization factor is greater than the high threshold), and the brightness of the Gray image is higher than the preset threshold (the reflectivity increase causes the gray value to increase), the PD intensity decreases (enhanced reflection means that the P-type layer thickens, squeezing the space of the light-emitting layer), and the leakage current Ir is normal (the leakage path is not damaged), it is determined as "enhanced reflection on the wafer surface (such as thickening of the P-type layer)", and the "epitaxial layer thickness detection process" is triggered (measure the thickness of the P-type layer by spectroscopic ellipsometry and adjust the P-type doping process parameters of MOCVD).

[0095] In addition, in some optional embodiments of the present invention, the method further includes:

[0096] Collect the GRP-NF distribution data of at least N historical good wafers to form a good product sample set;

[0097] Calculate the mean and standard deviation of GRP-NF in the good product sample set, and set dynamic thresholds according to the mean and standard deviation of GRP-NF;

[0098] Apply the dynamic threshold to the GRP-NF judgment of the current batch of wafers to distinguish normal and abnormal regions.

[0099] Among them, collect the GRP-NF distribution data of at least N (N≥100, ensuring statistical significance) historical good wafers. The good wafers need to pass through the full process inspection (no defects in AOI, electrical parameters meet the standards, and PL luminescence is uniform), and cover different production periods, the same product model, and the same MOCVD equipment (to avoid interference from equipment differences and product model differences on the sample validity). The GRP-NF data of each wafer is split according to the grid cell ID, forming a sample set containing millions of grid cell-level GRP-NF values. Calculate statistical parameters and set thresholds: Based on the good sample set, calculate the mean value of all grid cell-level GRP-NF values (reflecting the GRP-NF central level of good products) and the sample standard deviation, and then set dynamic thresholds according to the "3σ principle", which can effectively distinguish normal and abnormal areas. Apply the dynamic threshold to the GRP-NF judgment of the current batch of wafers. If the GRP-NF of a certain grid cell < Threshold_low (the physical normalization factor is less than the low threshold) or > Threshold_high, it is determined that the grid cell is an abnormal area. At the same time, this threshold model needs to be updated regularly - the update cycle is every M (M≥50) good wafers or once a week to adapt to the GRP-NF baseline shift caused by the long-term process drift of the MOCVD equipment (such as changes in source gas purity and heating tube aging), and avoid threshold failure. In addition, this model needs to be established independently for each product model and each MOCVD equipment. For example, when producing red Micro-LED and blue Micro-LED on the same equipment, threshold models need to be constructed separately to further improve the judgment accuracy.

[0100] In summary, the defect identification method based on multimodal data of the chip fabrication process in the above embodiments of the present invention acquires images of the graphite carrier by taking in-situ photographs of the graphite carrier surface using an industrial camera integrated into the wafer unloading machine after epitaxial growth, and reads the placement azimuth angle of the wafer on the graphite carrier from the epitaxial growth equipment controller; based on the graphite carrier image and placement azimuth angle, a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established to generate a graphite carrier-wafer mapping map; multimodal imaging is performed on the wafer under test, acquiring grayscale images, reflectivity images, and photoluminescence intensity images and synchronizing them at the pixel level, and processing the grayscale images... The three-channel images—rank image, reflectance image, and photoluminescence intensity image—are normalized separately. Based on the normalization results, a physical normalization factor is calculated to obtain a multimodal image. The wafer surface is divided into grid cells of a preset size, and a unique grid identifier is assigned to each grid cell to construct a wafer grid spatial database. Abnormal regions in the multimodal images, electrical test data, and graphite carrier images are mapped to their corresponding grid identifiers and stored in the wafer grid spatial database. Based on the multi-source data of abnormal regions in the mapped multimodal images, electrical test data, and graphite carrier images, defect types are identified according to preset rules. By mapping data from heterogeneous equipment such as MOCVD, AOI, PL, and probe stations to the same physical coordinate system and then storing it in a chip-level high-density spatial database, the analysis efficiency of Micro-LED manufacturing is significantly improved. This solves the problem of low efficiency in defect root cause analysis in existing technologies due to the independent nature of data generated during chip fabrication.

[0101] Example 2

[0102] This embodiment also proposes a defect identification method based on multimodal data of the chip fabrication process. The difference between the defect identification method based on multimodal data of the chip fabrication process in this embodiment and the defect identification method based on multimodal data of the chip fabrication process in Embodiment 1 is as follows:

[0103] The steps of normalizing the three-channel images (grayscale image, reflectance image, and photoluminescence intensity image) and calculating the physical normalization factor based on the normalization results to obtain the multimodal image include:

[0104]

[0105]

[0106]

[0107]

[0108]

[0109]

[0110] in, To obtain the normalized grayscale value, For normalized reflectivity, For normalized photoelectric, It is a grayscale image. For reflectance images, Image of photoluminescence intensity. , , , These represent the dynamic extreme values ​​of the current batch of wafers in the corresponding channels. As the physical normalization factor, =0.01 is used to prevent division by zero. This is a numerical truncation operation.

[0111] Understandably, the first step is to calculate the normalized grayscale, where... The original pixel values ​​(range 0-255) of the grayscale image are normalized to the 0-1 range by dividing by 255, eliminating the influence of differences in grayscale acquisition ranges from different cameras. Next, the reflectance image is processed. ,in, This is the dynamic minimum value of the reflectivity channel for the current batch of wafers. These are the dynamic maxima of the channel (both are real-time statistical values ​​for the current batch, not fixed values); since extreme outliers may exist in actual imaging, the clip function is then used to... Limiting the value to the range of 0-1, the normalized reflectance is finally obtained. The processing logic for the photoluminescence intensity image (PD) is the same as that for the reflectance image; after constraint by the clip function, the normalized photocurrent is obtained. Finally, the physical normalization factor is calculated; furthermore, in practical engineering applications, weighting coefficients can be introduced. α , β To adapt to the differences in optical characteristics of different product models, the final result is... GRP-NF It can eliminate non-essential factors such as uneven light source intensity and sensor response differences, and become a unified physical indicator for characterizing the wafer state, i.e., the core data of multimodal images.

[0112] Furthermore, before the step of normalizing the three-channel images—grayscale image, reflectance image, and photoluminescence intensity image—the following steps are included:

[0113] Obtained by scanning the edge region of the current batch of wafers or a reference standard. , , , Real-time statistics.

[0114] Before normalizing the three-channel images, the aforementioned dynamic extrema need to be obtained through two methods: "edge scanning" or "standard wafer calibration". The first method is to scan the edge region of the current batch of wafers—selecting a 5mm wide annular region at the edge of the wafer (this region is grown using the same MOCVD process as the main wafer, has the same physical characteristics, and avoids the central effective chip area, so as not to affect product inspection), and scanning this region line by line. and Data, statistically obtained for this batch , , , The second method is to use a reference standard wafer—a GaN-based wafer that has undergone traceability calibration and has stable performance. and The parameters are known and uniform. By imaging a standard film, the standard film's... / The calibration range serves as the dynamic extreme value for the current batch. Both methods can adapt in real time to the lighting conditions and equipment parameter fluctuations (such as changes in laser power) of the current batch, avoiding the incomparability of data between different batches caused by traditional fixed thresholds, and ensuring the physical consistency of the normalization results.

[0115] In summary, the defect identification method based on multimodal data of the chip fabrication process in the above embodiments of the present invention acquires images of the graphite carrier by taking in-situ photographs of the graphite carrier surface using an industrial camera integrated into the wafer unloading machine after epitaxial growth, and reads the placement azimuth angle of the wafer on the graphite carrier from the epitaxial growth equipment controller; based on the graphite carrier image and placement azimuth angle, a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established to generate a graphite carrier-wafer mapping map; multimodal imaging is performed on the wafer under test, acquiring grayscale images, reflectivity images, and photoluminescence intensity images and synchronizing them at the pixel level, and processing the grayscale images... The three-channel images—rank image, reflectance image, and photoluminescence intensity image—are normalized separately. Based on the normalization results, a physical normalization factor is calculated to obtain a multimodal image. The wafer surface is divided into grid cells of a preset size, and a unique grid identifier is assigned to each grid cell to construct a wafer grid spatial database. Abnormal regions in the multimodal images, electrical test data, and graphite carrier images are mapped to their corresponding grid identifiers and stored in the wafer grid spatial database. Based on the multi-source data of abnormal regions in the mapped multimodal images, electrical test data, and graphite carrier images, defect types are identified according to preset rules. By mapping data from heterogeneous equipment such as MOCVD, AOI, PL, and probe stations to the same physical coordinate system and then storing it in a chip-level high-density spatial database, the analysis efficiency of Micro-LED manufacturing is significantly improved. This solves the problem of low efficiency in defect root cause analysis in existing technologies due to the independent nature of data generated during chip fabrication.

[0116] Example 3

[0117] Please see Figure 2 The image shows a defect identification device based on multimodal data from the chip fabrication process, as proposed in the third embodiment of the present invention. The device includes:

[0118] The acquisition module 100 is used to acquire images of the graphite carrier surface by taking in-situ photos of the graphite carrier surface using an industrial camera integrated in the wafer unloading machine after the epitaxial growth is completed, and to read the placement azimuth angle of the wafer on the graphite carrier from the epitaxial growth equipment controller.

[0119] Module 200 is established to establish a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system based on the graphite carrier image and placement azimuth angle, and to generate a graphite carrier-wafer mapping diagram.

[0120] The imaging module 300 is used to perform multimodal imaging on the wafer under test, acquire grayscale images, reflectance images and photoluminescence intensity images and synchronize them at the pixel level, normalize the three-channel images of grayscale images, reflectance images and photoluminescence intensity images respectively, and calculate the physical normalization factor based on the normalization results to obtain the multimodal image.

[0121] The storage module 400 is used to divide the wafer surface into grid cells of a preset size and assign a unique grid identifier to each grid cell to build a wafer grid space database. Abnormal areas in multimodal images, electrical test data and graphite carrier images are mapped to the corresponding grid identifiers and stored in the wafer grid space database.

[0122] The identification module 500 is used to identify the defect type according to preset rules based on multi-source data of abnormal areas in the mapped multimodal images, electrical test data and graphite carrier images.

[0123] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0124] Example 4

[0125] In another aspect, the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in any one of the above embodiments one to two.

[0126] Example 5

[0127] In another aspect, the present invention provides an electronic device, the electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any one of the methods described in embodiments one to two above.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0130] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0131] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0132] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0133] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for defect identification based on multi-modal data of a chip manufacturing process, characterized in that, The method includes: After epitaxial growth is completed, the surface of the graphite carrier is photographed in situ using an industrial camera integrated into the wafer unloading machine to obtain images of the graphite carrier, and the placement azimuth angle of the wafer on the graphite carrier is read from the epitaxial growth equipment controller. Based on the graphite carrier image and placement azimuth, a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established, and a graphite carrier-wafer mapping diagram is generated. Multimodal imaging is performed on the wafer under test. Grayscale images, reflectance images, and photoluminescence intensity images are acquired and synchronized at the pixel level. The three-channel images of grayscale images, reflectance images, and photoluminescence intensity images are normalized respectively. The physical normalization factor is calculated based on the normalization results to obtain the multimodal image. The wafer surface is divided into grid cells of a preset size, and a unique grid identifier is assigned to each grid cell to build a wafer grid space database. Abnormal areas in multimodal images, electrical test data and graphite carrier images are mapped to the corresponding grid identifiers and stored in the wafer grid space database. Based on multi-source data from the mapped multimodal images, electrical test data, and abnormal regions in the graphite carrier images, the defect type is identified according to preset rules. The steps of normalizing the three-channel images (grayscale image, reflectance image, and photoluminescence intensity image) and calculating the physical normalization factor based on the normalization results to obtain the multimodal image include: in, To obtain the normalized grayscale value, For normalized reflectivity, For normalized photoelectric, It is a grayscale image. For reflectance images, Image of photoluminescence intensity. , , , These represent the dynamic extreme values ​​of the current batch of wafers in the corresponding channels. As the physical normalization factor, =0.01 is used to prevent division by zero. This is a numerical truncation operation.

2. The defect identification method based on multimodal data of chip fabrication process according to claim 1, characterized in that, The steps for establishing the spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system include: The SIFT feature extraction algorithm was used to identify the contact points between the positioning holes on the graphite carrier and the wafer edge as control point pairs. The RANSAC algorithm is used to filter control point pairs and eliminate mismatched points. Based on the selected effective control point pairs, the initial affine transformation matrix is ​​obtained by solving, and the spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system is established according to the initial affine transformation matrix.

3. The defect identification method based on multimodal data of chip fabrication process according to claim 2, characterized in that, Before the step of normalizing the three-channel images (grayscale image, reflectance image, and photoluminescence intensity image) respectively, the following steps are also included: Obtained by scanning the edge region of the current batch of wafers or a reference standard. , , , Real-time statistics.

4. The defect identification method based on multimodal data of the chip fabrication process according to claim 3, characterized in that, The step of assigning a unique grid identifier to each grid cell includes: = in, For grid cells in wafer coordinate system X Directional coordinates For grid cells in wafer coordinate system Y Directional coordinates This indicates the floor function; For wafers Y Total number of grid cells in each direction.

5. The defect identification method based on multimodal data of chip fabrication process according to claim 4, characterized in that, The method further includes: Collect at least N historical high-quality wafers GRP-NF Distribution data is used to form a sample set of good products; Calculate the good product sample set GRP-NF The mean and standard deviation, according to GRP-NF Set dynamic thresholds for the mean and standard deviation; Apply dynamic thresholds to the current batch of wafers. GRP-NF In the process of judgment, the normal and abnormal areas are distinguished.

6. The defect identification method based on multimodal data of chip fabrication process according to claim 5, characterized in that, The step of identifying the defect type according to preset rules includes: When there is an anomaly in the graphite carrier image display area A and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be a graphite carrier defect, and the carrier is taken offline for maintenance and repair. When the gradient difference of the grayscale image is greater than the first threshold and is located in the wafer edge region, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too low, and the MOCVD temperature field increase process is triggered. When the gradient difference of the grayscale image is greater than the first threshold and is located in the center region of the wafer, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the wafer temperature field is too high, and the MOCVD temperature field reduction process is triggered. When the physical normalization factor is less than the low threshold, the defect morphology is amorphous, and the leakage current Ir of the corresponding grid cell is normal, it is determined to be wafer surface contamination, and the wafer cleaning process is triggered. When the physical normalization factor is less than the low threshold, the defect morphology is circular / ring-shaped, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the MOCVD cavity is shedding dust, and the MOCVD maintenance process is triggered. When the physical normalization factor is greater than the high threshold and the intensity of the photoluminescence intensity image is lower than the first preset threshold and the Vf of the corresponding grid cell deviates from the threshold, it is determined that there is a leak in the MOCVD cavity and the MOCVD hydrogen detection process is triggered. When the gradient difference of the grayscale image is greater than the second threshold and there are abnormalities at both the center and edge of the wafer, and the leakage current Ir of the corresponding grid cell is generally abnormal, it is determined that the graphite carrier is aging or the temperature control is malfunctioning, and the carrier replacement and MOCVD calibration process are triggered. When the physical normalization factor is less than the low threshold, the intensity of the photoluminescence intensity image is lower than the second preset threshold, and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined that the graphite carrier has started to break down, and the carrier is taken offline for maintenance and repair. When the physical normalization factor is less than the low threshold and the defect is linear / striped and the leakage current Ir of the corresponding grid cell deviates from the threshold, it is determined to be wafer scratch contamination and the wafer pre-placement inspection process is triggered. When the physical normalization factor is greater than the high threshold and the brightness of the grayscale image is higher than the preset threshold, but the intensity of the photoluminescence intensity image decreases and the leakage current Ir is normal, it is determined that the wafer surface reflection is enhanced, and the epitaxial layer thickness detection process is triggered.

7. A defect identification device based on multimodal data from the chip fabrication process, characterized in that, The apparatus for implementing the defect identification method based on multimodal data of the chip fabrication process according to any one of claims 1 to 6, the apparatus comprising: The acquisition module is used to capture images of the graphite carrier surface in situ using an industrial camera integrated into the wafer unloading machine after epitaxial growth is completed, and to read the placement azimuth angle of the wafer on the graphite carrier from the epitaxial growth equipment controller. A module is established to create a spatial mapping relationship between the graphite carrier coordinate system and the wafer coordinate system based on the graphite carrier image and placement azimuth angle, and to generate a graphite carrier-wafer mapping diagram. The imaging module is used to perform multimodal imaging on the wafer under test, acquire grayscale images, reflectance images and photoluminescence intensity images and synchronize them at the pixel level, normalize the three-channel images of grayscale images, reflectance images and photoluminescence intensity images respectively, and calculate the physical normalization factor based on the normalization results to obtain the multimodal image. The storage module is used to divide the wafer surface into grid cells of a preset size and assign a unique grid identifier to each grid cell to build a wafer grid space database. Abnormal areas in multimodal images, electrical test data and graphite carrier images are mapped to the corresponding grid identifiers and stored in the wafer grid space database. The identification module is used to identify the defect type according to preset rules based on multi-source data of abnormal areas in the mapped multimodal images, electrical test data, and graphite carrier images.

8. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method as described in any one of claims 1 to 6.

Images