CN116183623B - Intelligent wafer surface defect detection method and device - Google Patents

Abstract

The invention discloses an intelligent detection method and device for wafer surface defects, which belong to the technical field of wafer defect detection, and the detection method comprises the following steps: the wafer and the wafer vision photographing module are respectively arranged on the corresponding carriers; controlling the wafer or the wafer vision photographing module to move to a first detection initial relative position; the wafer vision photographing module is used for collecting the surface image of the wafer in a detection mode, and the method comprises photographing by path points in a preset detection path in the detection mode; controlling the wafer or the wafer vision photographing module to move to a second detection initial relative position; the wafer vision photographing module is used for collecting the surface image of the wafer in the two detection modes, and photographing is carried out on path points by path points in the preset detection paths in the two detection modes; and processing the shooting images obtained in the first detection mode and the second detection mode to generate detection results. The invention can accurately and efficiently detect various defects on the surface of the wafer by a polishing mode with high degree of freedom.

Description

A method and device for intelligent detection of wafer surface defects

Technical Field

The present invention relates to the technical field of wafer defect detection, and in particular to a method and device for intelligently detecting wafer surface defects.

Background technique

A wafer refers to a chip used to make semiconductor circuits. Common materials include silicon, germanium, gallium arsenide, silicon carbide, aluminum nitride, zinc oxide, diamond, indium phosphide, gallium nitride, etc.; the corresponding properties are obtained by doping different elements into them; taking silicon material as an example, high-purity polycrystalline silicon is dissolved and then doped with silicon crystal seeds, and the preparation of silicon rods is completed by the pulling method; after grinding, polishing and slicing, the silicon crystal rods are formed into silicon wafers, that is, wafers; in the wafer production process, chemical vapor deposition, optical development, chemical mechanical grinding, single crystal pulling, slicing, grinding, polishing, layer addition, photolithography, doping, heat treatment and dicing are a series of processes and steps; due to the various processing techniques of semiconductor wafers, during the processing process and transportation process, wafer surface defects are very easy to cause due to abnormal operation of instruments and improper operation of workers in the processing process.

Defects caused during the handling process include suction cup marks left on the wafer surface due to improper pressure adjustment when transporting the wafer; claw cracks, edge collapse, scratches and abrasions caused by equipment on the wafer during clamping or processing of the wafer; stains, pinholes and chemical burns on the wafer surface due to unclean removal of photoresist after photoresist coating; and uneven edge removal defects caused by uneven cutting during wafer cutting.

The above defects will cause more serious defects in the subsequent lithography process of the defective wafer; and the defective wafer entering the next processing and manufacturing process will cause a waste of production and manufacturing resources, affecting the subsequent wafer processing yield and overall manufacturing efficiency.

With the advancement of industrial image sensors, computer vision, and image processing hardware computing power, the use of optical inspection equipment and machine vision to detect wafer surface defects and analyze the causes of defects in combination with process analysis has gradually developed and applied to improve the manufacturing process. At present, there are cases of wafer defect detection using machine vision and computer vision methods, most of which are fixed lighting methods, which can only detect single, simple-structured wafer surface defects; and the actual situation in the actual wafer inspection industry is that wafer products are updated and iterated quickly, and the inspection process and materials of wafers are becoming more and more diverse, and the types of defects are diverse. Fixed lighting methods cannot meet the increasing number of types of defects. The above-mentioned detection methods and methods have the disadvantages of single shooting methods, low degree of automation, difficulty in collecting defect data sets, and poor versatility.

Summary of the invention

The present invention aims to solve one of the technical problems in the related art to at least a certain extent. To this end, the present invention provides a method and device for intelligent detection of wafer surface defects, which can achieve accurate and efficient detection through a high degree of freedom lighting method and a highly adaptive detection algorithm, and is a wafer surface defect detection method with high versatility, strong reliability and high ductility.

To achieve the above-mentioned purpose, in a first aspect, the present application provides a method for intelligent detection of wafer surface defects, comprising: mounting a wafer and a wafer visual camera module on corresponding carriers respectively;

Controlling the wafer or the wafer visual camera module to move to a first detection starting relative position;

Using the wafer vision camera module to collect wafer surface images in the detection mode, including taking pictures at each path point in the preset detection path of the detection mode, and looping to the end point of the detection path;

Controlling the wafer or the wafer visual camera module to move to a second detection starting relative position;

Using the wafer vision camera module to collect wafer surface images in the second detection mode, including taking pictures at each path point in the preset detection path of the second detection mode, and looping to the end point of the detection path;

Perform image detection processing on all the captured images obtained in the detection mode 1 and the detection mode 2 to generate wafer surface defect detection results.

Preferably, the detection mode illuminates the wafer surface at different tilt angles through an LED diffuse parallel light source.

Preferably, the second detection mode illuminates the wafer surface at different tilt angles through a high-intensity light source with an illumination greater than a first threshold.

Preferably, the lighting of the wafer surface at different tilt angles in the detection mode 1 and the detection mode 2 includes at least a high lighting angle range and a low lighting angle range, wherein the lighting angle α is the angle between the light source emission direction and the normal of the wafer surface, the high lighting angle range is α=0 ^°° ~30, and the low lighting angle range is α=60 ^°° ~90.

Preferably, the wafer and the wafer vision camera module are respectively installed on corresponding carriers, and the wafer or the wafer vision camera module is controlled to move to the first detection starting relative position, including adsorbing the wafer to be tested on the suction cup on the end bracket of the multi-degree-of-freedom robot, and fixing the wafer vision camera module on the module bracket, and controlling the multi-degree-of-freedom robot to drive the wafer to be tested to move to the first detection starting relative position.

Preferably, the wafer and the wafer vision camera module are respectively installed on corresponding carriers, and the wafer or the wafer vision camera module is controlled to move to the first detection starting relative position, including installing the wafer vision camera module on the end bracket of the multi-degree-of-freedom robot, and placing the wafer to be tested on the module bracket, and controlling the multi-degree-of-freedom robot to drive the vision camera module to move to the first detection starting relative position.

Preferably, image detection processing is performed on all captured images obtained in detection mode 1 and detection mode 2 to generate wafer surface defect detection results, including pre-processing all captured images and then performing region segmentation processing to cut out the wafer images, and using the optimized YOLOV5 model to complete surface defect target detection of the wafer images.

In a second aspect, the present application provides an intelligent detection device for wafer surface defects, comprising:

An inspection workbench, a wafer vision camera module and a multi-degree-of-freedom robot, wherein the wafer vision camera module and the multi-degree-of-freedom robot are symmetrically arranged on the inspection workbench, the wafer vision camera module is used to capture images of wafer surface defects, and the multi-degree-of-freedom robot is used to drive the wafer or the wafer vision camera module to move and transform its position.

Preferably, the wafer visual camera module includes industrial camera 1, industrial camera 2, a high-intensity light source and an LED diffuse parallel light source. The industrial camera 1 and the LED diffuse parallel light source are used to collect the wafer surface image in detection mode 1, and the industrial camera 2 and the high-intensity light source are used to collect the wafer surface image in detection mode 2.

Compared with the prior art, the present invention has the following beneficial effects:

The intelligent wafer surface defect detection method provided by the present invention can accurately and efficiently detect various defects on the wafer surface through a high-degree-of-freedom lighting method, and can achieve the goal of multi-angle, full-size, and full-type surface defect detection of the wafer through the mutual cooperation of different detection modes.

Other features and advantages of the present application will be described in the subsequent description, and in part will become apparent from the description, or it may be understood through the implementation of the present application that the objects and other advantages of the present application can be realized and obtained through the structures particularly pointed out in the written description, claims, and drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for intelligently detecting wafer surface defects according to the present invention;

FIG2 is a schematic diagram of high-angle lighting in a wafer surface defect intelligent detection method of the present invention;

FIG3 is a schematic diagram of low-angle lighting in a wafer surface defect intelligent detection method according to the present invention;

FIG4 is a schematic structural diagram of a first embodiment of an intelligent device for detecting wafer surface defects according to the present invention;

FIG5 is a schematic structural diagram of a second embodiment of a wafer surface defect intelligent detection device according to the present invention;

FIG6 is a schematic diagram of the structure of a wafer visual camera module in a wafer surface defect intelligent detection device according to the present invention;

FIG. 7 is a network structure diagram of a wafer surface defect intelligent detection method according to the present invention.

In the figure: 1. Module bracket; 2. Wafer vision camera module; 3. Wafer; 4. Suction cup; 5. End bracket; 6. Multi-degree-of-freedom robot; 7. Inspection workbench; 8. Wafer bracket; 2-1. Industrial camera 1; 2-2. Industrial lens 1; 2-3. Industrial camera 2; 2-4. Industrial lens 2; 2-5. High-intensity light source; 2-6. LED diffuse parallel light source; 2-7. Sensor bracket.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

As shown in FIG1 , in a first aspect, the present invention provides a first embodiment, a wafer surface defect intelligent detection method, comprising:

The wafer 3 and the wafer vision camera module 2 are mounted on corresponding carriers respectively;

Control the wafer 3 or the wafer visual camera module 2 to move to a first detection starting relative position;

Using the wafer vision camera module 2 to collect the wafer surface image in the detection mode 1, including taking pictures at each path point in the preset detection path of the detection mode 1, and looping to the end point of the detection path;

Control the wafer 3 or the wafer visual camera module 2 to move to a second detection starting relative position;

Using the wafer vision camera module 2 to collect the wafer surface image in the second detection mode, including taking pictures at each path point in the preset detection path of the second detection mode, and looping to the end point of the detection path;

Perform image detection processing on all captured images obtained in detection mode 1 and detection mode 2 to generate wafer surface defect detection results.

The present invention is also equipped with a highly adaptive detection algorithm to complete accurate and efficient detection. The method automatically identifies the image type by calculating the average grayscale value and the uniformity of grayscale distribution, judges the bright and dark fields, uses histogram equalization and a dynamic bilateral filtering algorithm for denoising and sharpening for the bright field, uses a method for removing overexposure images for the dark field to process the original image, uses a dynamic threshold method to filter the image, and then improves YOLOV5 by adding a layer of enhanced feature extractor for defect contrast difference to identify and extract wafer surface defect information. Therefore, the adaptive algorithm has high versatility and strong reliability.

Preferably, the detection mode 1 illuminates the wafer surface at different tilt angles through LED diffuse parallel light sources 2-6, and the detection mode 1 is a full-size detection mode of diffuse reflection of parallel light on the wafer surface. The detection mode 2 illuminates the wafer surface at different tilt angles through a high-intensity light source 2-5 whose illumination is greater than a first threshold. The illumination of the wafer surface at different tilt angles in the detection mode 1 and the detection mode 2 at least includes a high lighting angle range and a low lighting angle range, wherein the lighting angle α is the angle between the light source emission direction and the normal of the wafer surface, the high lighting angle range is α=0°°～30, and the low lighting angle range is α=60°°～90.

As shown in Figures 2 and 3, they are schematic diagrams of the structures of high-angle lighting and low-angle lighting, respectively. The calculation formula is h _c = d _o × tan (α), where h _c is the calculated crack feature depth that can be identified, d _o is the current crack width, and α is the angle between the light source and the camera. According to statistics, the hole depth-to-width ratio of the currently detected silicon wafer cracks and scratch defects is distributed around 2.5, so for low-angle lighting, the camera needs to be facing the object to be tested, and the light source and the camera form an angle α. Experiments have confirmed that the angle between the light source emission direction and the normal to the wafer surface is α = 60°°~90° for the best feature performance. Applying the law of reflection, high-angle lighting obtains a more uniform lighting effect on the silicon wafer. The angle between the light source emission direction and the normal to the wafer surface is in the range of α = 0°°~30°. It has been determined through experiments that it can obtain the most uniform lighting effect.

In the detection mode, in the low lighting angle range, the angle between the light source's emitting direction and the wafer surface normal is α=60°~90, which is the low-angle lighting form of the detection mode. Since the incident light of the low-angle light source enters the camera through the diffuse reflection light of the surface of the object to be measured, this is also called the "relief image" lighting method; in the current form, the collected image can highlight the types of wafer surface defects such as claw cracks, scratches, and abrasions, enhance the contrast between defective areas and non-defective areas, and have better image quality; when the angle between the light source's emitting direction and the wafer surface normal is α=0°~30, this is the high-angle lighting form of the detection mode. After passing through the surface of the object to be measured, most of the incident light of the light source enters the camera through mirror reflection, which is also called the "black and white image" lighting method; in the current form, the collected image can highlight the contrast of the wafer edge image, and the grayscale of the wafer image is more uniform.

In the detection mode 2, when the angle between the high-intensity light source and the normal to the wafer surface is α=60°°～90, this is the low-angle lighting form of the detection mode 2. Under high-intensity light irradiation, defects that are not easy to highlight under the diffuse reflection lighting principle under low-intensity light appear sharper; in the current form, the collected image highlights the contamination, pinholes and other wafer surface defect areas showing bright spot convergence, which also means that the adhesion defects such as contamination and pinholes have a better image quality; in the detection mode 2, when the angle between the high-intensity light source and the normal to the wafer surface is α=0°°～30. This is the high-angle lighting form of the detection mode 2. In this form, the area on the wafer surface burned by the residual liquid is imaged as a bright surface or shadow under high-intensity light irradiation, which is easier to extract by the subsequent defect detection algorithm.

As shown in FIG4 , the first embodiment provided by the present invention, wherein the wafer 3 and the wafer visual camera module 2 are respectively mounted on corresponding carriers, and the wafer 3 or the wafer visual camera module 2 is controlled to move to the first detection starting relative position, comprises: adsorbing the wafer 3 to be tested on the suction cup 4 on the end bracket 5 of the multi-degree-of-freedom robot 6, and fixing the wafer visual camera module 2 on the module bracket 1, and controlling the multi-degree-of-freedom robot 6 to drive the wafer 3 to be tested to move to the first detection starting relative position. Among them, the multi-degree-of-freedom robot 6 is a 6-degree-of-freedom intelligent wafer detection collaborative robot;

The end of the multi-degree-of-freedom robot 6 is equipped with a wafer adsorption end execution module composed of a suction cup 4 and an end bracket 5, and can complete the insertion and adsorption of the wafer through the control of the solenoid valve; the wafers are arranged in sequence in the wafer box, there is a certain gap between the wafers, and they are independent of each other and will not touch each other.

When adsorbing the wafer, the wafer adsorption terminal execution module carried by the multi-degree-of-freedom robot 6 moves to the gap between the wafer to be inspected and the next wafer, and is inserted in a direction parallel to the suction cup 4 surface of the wafer adsorption terminal execution module, and the posture is adjusted so that the plane of the suction cup 4 is parallel to the plane of the wafer and is in a position close to contact. The current position is the position where the multi-degree-of-freedom robot 6 adsorbs the wafer for loading, and the multi-degree-of-freedom robot 6 is in place and sends an in-place signal; the upper computer software system receives the signal sent by the robot, sends an adsorption instruction to the motion controller, the motion controller controls the solenoid valve to respond, and the end suction cup 4 completes the wafer adsorption.

After the end suction cup 4 absorbs a wafer to be inspected, it moves to the inspection area 1. After reaching the inspection starting position, the robot sends an in-position signal; after receiving the signal, the host computer software system controls the light source of the wafer inspection parallel light detection module to turn on, and the light source of the wafer inspection high-intensity light detection module to turn off, and the inspection system enters the inspection mode 1; the end suction cup 4 always keeps the wafer in the state of being adsorbed before the end of the inspection, and the end suction cup 4 carries the wafer to be inspected and moves according to the preset position path; during the inspection process of the inspection mode 1, the host computer software system obtains the position parameters sent by the robot in real time, and after judging that the robot is in place, the host computer software system controls the triggering of the camera to take pictures; the image obtained by taking pictures enters the host computer software system, is processed by the wafer surface defect detection algorithm, and the result is output to the wafer surface defect detection result list; then, the photo detection and processing are carried out at each path point, and it is cycled to the end of the inspection path in sequence;

The end suction cup 4 carries the wafer to be tested and moves to the starting position of the first detection mode. After reaching the position, the robot sends an in-position signal; after receiving the robot in-position signal, the host computer software system turns off the light source of the wafer detection parallel light detection module, and turns on the light source of the wafer detection high-intensity light detection module, and enters the second detection mode; the robot carries the wafer and moves according to the detection path in the second detection area. At each shooting position, the host computer software system obtains the position parameters sent by the robot in real time. After determining that the robot is in position, the host computer software system controls the triggering camera to take pictures; the image obtained by taking pictures is transmitted to the wafer surface defect detection algorithm of the host computer software system, processed by the wafer surface defect detection algorithm, and the result is output to the wafer surface defect detection result list; after that, the photo detection and processing are performed at each path point, and the cycle is cycled to the end of the detection path; after the photo process is completed, the central control system traverses the wafer surface defect detection result list, determines whether there are defects on the current wafer surface, and separates the defective wafer and the normal wafer into the corresponding wafer placement bin, and the detection is completed.

As shown in FIG5 , the second embodiment provided by the present invention is different from the first embodiment in that the wafer 3 and the wafer visual camera module 2 are respectively installed on the corresponding carriers, and the wafer 3 or the wafer visual camera module 2 is controlled to move to the first detection starting relative position, including installing the wafer visual camera module 2 on the end bracket 5 of the multi-degree-of-freedom robot 6, and placing the wafer 3 to be tested on the module bracket 1, and controlling the multi-degree-of-freedom robot 6 to drive the visual camera module to move to the first detection starting relative position. Specifically, in this embodiment, after the end suction cup 4 of the automatic loading module absorbs a wafer to be tested, it is placed on the detection station of the wafer bracket 8. After the detection station sensor detects that a new wafer to be tested is placed, it triggers a signal that the loading is completed.

Preferably, the image detection processing of all captured images to generate wafer surface defect detection results includes preprocessing all captured images and then performing region segmentation processing to cut out the wafer image, and using the optimized YOLOV5 model to complete the surface defect target detection of the wafer image. Preprocessing all captured images includes filtering and histogram equalization processing.

One reason why the original YOLOV5 model is not effective for wafer surface defect detection is that the contrast between wafer defects and the background is weak and the size of the defects on the wafer surface is small. The original YOLOV5 has a large downsampling multiple, and it is difficult for deeper feature maps to learn the feature information of small targets. Therefore, it is proposed to add a small target detection layer to splice the shallower feature map with the deep feature map for detection.

After the last upsampling after the original network feature extraction layer, the feature map continues to be upsampled and other processes to expand the feature map. At the same time, the acquired feature map of size 152X152 is concat fused with the second layer feature map in the backbone network to obtain a larger feature map for small target detection. As for the network structure of each model of the original YOLOV5, for example, the YOLOV5-x with the most complex network structure can only repeat the CSP use layer multiple times in the network structure. It seems that the network structure is deepened, but it does not fit the detection characteristics of the wafer, where the difference between the feature and background is not obvious and the feature size is small, and the detection effect is not ideal. In conjunction with the previous image preprocessing algorithm, adding a layer of enhanced feature extractor and optimizing the YOLOV5 network can achieve better wafer defect extraction results.

As shown in the network structure diagram in Figure 7, compared with the original YOLOV5 model, an upsampling is added, which makes the extraction of wafer surface defect features more obvious.

In a second aspect, the present application provides an intelligent detection device for wafer surface defects, comprising:

An inspection workbench 7, a wafer vision camera module 2 and a multi-degree-of-freedom robot 6, wherein the wafer vision camera module 2 and the multi-degree-of-freedom robot 6 are symmetrically arranged on the inspection workbench 7, the wafer vision camera module 2 is used to capture images of wafer surface defects, and the multi-degree-of-freedom robot 6 is used to drive the wafer 3 or the wafer vision camera module 2 to move and transform its position.

As shown in Figure 6, the wafer vision camera module 2 includes an industrial camera 1 2-1, an industrial camera 2 2-3, a high-intensity light source 2-5 and an LED diffuse parallel light source 2-6. The industrial camera 1 2-1 and the LED diffuse parallel light source 2-6 are used to collect wafer surface images in a detection mode 1, and the industrial camera 2 2-3 and the high-intensity light source 2-5 are used to collect wafer surface images in a detection mode 2. Among them, the industrial camera 1 2-1 and the industrial camera 2 2-3 are respectively equipped with an industrial lens 1 2-2 and an industrial lens 2 2-4, and the industrial camera 1 2-1 and the industrial camera 2 2-3 are both fixedly mounted on a sensor bracket 2-7.

The intelligent wafer surface defect detection method provided by the present invention can accurately and efficiently detect various defects on the wafer surface through a high-degree-of-freedom lighting method, and can achieve the goal of multi-angle, full-size, and full-type surface defect detection of the wafer through the mutual cooperation of different detection modes.

In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Wherein, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

Claims (5)

1. A wafer surface defect intelligent detection method, characterized in that a wafer surface defect intelligent detection device is provided, the wafer surface defect intelligent detection device comprises a detection workbench, a wafer visual camera module and a multi-degree-of-freedom robot, the wafer visual camera module and the multi-degree-of-freedom robot are symmetrically arranged on the detection workbench, the wafer visual camera module is used to take images of wafer surface defects, and the multi-degree-of-freedom robot is used to drive the wafer or the wafer visual camera module to move and transform the position; The detection method comprises: Mounting the wafer and the wafer vision camera module on corresponding carriers respectively; Controlling the wafer or the wafer visual camera module to move to a first detection starting relative position; Using the wafer vision camera module to collect wafer surface images in the detection mode, including taking pictures at each path point in the preset detection path of the detection mode, and looping to the end point of the detection path; Controlling the wafer or the wafer visual camera module to move to a second detection starting relative position; Using the wafer vision camera module to collect wafer surface images in the second detection mode, including taking pictures at each path point in the preset detection path of the second detection mode, and looping to the end point of the detection path; Performing image detection processing on all captured images obtained in the detection mode 1 and the detection mode 2 to generate wafer surface defect detection results; The detection mode illuminates the wafer surface at different tilt angles through LED diffuse parallel light sources; The second detection mode illuminates the wafer surface at different tilt angles using a high-intensity light source with an illumination greater than a first threshold; The illumination of the wafer surface at different tilt angles in the detection mode 1 and the detection mode 2 at least includes a high illumination angle range and a low illumination angle range, wherein the illumination angle α is the angle between the light source emission direction and the normal of the wafer surface, the high illumination angle range is α=0°~30°, and the low illumination angle range is α=60°~90°; In the first detection mode, when the angle between the light source emission direction and the normal direction of the wafer surface is α=60°~90°, the collected image highlights the defects of claw cracks, scratches, and abrasions, and enhances the contrast between the defective area and the non-defective area; when the angle between the light source emission direction and the normal direction of the wafer surface is α=0°~30°, the collected image highlights the contrast of the wafer edge image, and the grayscale of the wafer image is more uniform; In the second detection mode, when the angle between the high-intensity light source's emission direction and the wafer surface normal is α=60°~90°, the collected image highlights the contamination and pinhole defect areas and shows a concentration of bright spots. When the angle between the high-intensity light source's emission direction and the wafer surface normal is α=0°~30°, the areas burned by the residual liquid appear as bright surfaces or shadows under high-intensity light, making it easy for the defect detection algorithm to extract them. 2. A method for intelligent detection of wafer surface defects according to claim 1, characterized in that the wafer and the wafer vision camera module are respectively installed on corresponding carriers, and the wafer or the wafer vision camera module is controlled to move to the first detection starting relative position, including adsorbing the wafer to be tested on the suction cup on the end bracket of the multi-degree-of-freedom robot, and fixing the wafer vision camera module on the module bracket, and controlling the multi-degree-of-freedom robot to drive the wafer to be tested to move to the first detection starting relative position. 3. A method for intelligent detection of wafer surface defects according to claim 1, characterized in that the wafer and the wafer vision camera module are respectively installed on corresponding carriers, and the wafer or the wafer vision camera module is controlled to move to the first detection starting relative position, including installing the wafer vision camera module on the end bracket of a multi-degree-of-freedom robot, and placing the wafer to be tested on the module bracket, and controlling the multi-degree-of-freedom robot to drive the vision camera module to move to the first detection starting relative position. 4. A wafer surface defect intelligent detection method according to claim 3, characterized in that image detection processing is performed on all captured images obtained in detection mode 1 and detection mode 2, and generating wafer surface defect detection results includes pre-processing all captured images and then performing regional segmentation processing to cut out the wafer image, and using an optimized YOLOV5 model to complete surface defect target detection of the wafer image. 5. According to claim 4, a wafer surface defect intelligent detection method is characterized in that the wafer visual camera module includes industrial camera 1, industrial camera 2, a high-intensity light source and an LED diffuse parallel light source, the industrial camera 1 and the LED diffuse parallel light source are used to collect wafer surface images in detection mode 1, and the industrial camera 2 and the high-intensity light source are used to collect wafer surface images in detection mode 2.