Macroscopic and microscopic combined wafer defect intelligent positioning system and working method thereof

Abstract

The invention relates to a macroscopic and microcosmic combined wafer defect intelligent positioning system which comprises a macroscopic camera group, a movement mechanism, a computer and a microcosmic camera, the movement mechanism is in transmission connection with the macroscopic camera group, and the macroscopic camera group and the microcosmic camera are both connected with the computer. And the computer processes image data collected by the macroscopic camera group and the microscopic camera group by establishing a defect detection model, an optimal angle prediction model and a macro-microscopic correlation model. The working process of the system comprises macroscopic camera scanning; shooting at an optimal angle; carrying out macroscopic angle shooting; detecting macroscopic defects; microcosmic probability prediction; carrying out microscopic shooting; detecting microdefects; and summarizing results. According to the wafer defect intelligent positioning system based on macroscopic and microcosmic combination, the three models are used in a combined mode, an existing business process is embedded, high-speed, accurate and automatic positioning of defect positions can be achieved, quality inspection efficiency is integrally improved, and enterprises are helped to carry out yield improvement, process control and knowledge accumulation.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor production, and in particular to a macro-micro combined wafer defect intelligent positioning system and a working method thereof. Background Art

[0002] In the semiconductor industry, how to locate tiny defects on large-sized wafers during the online inspection of wafer quality has long been a problem that has plagued the industry. With the continuous improvement of semiconductor technology, the size of defects that need to be identified is getting smaller and smaller. Usually, the minimum defect size can reach about 1um, and traditional wafer quality inspection can no longer use manual visual inspection. Therefore, most manufacturers generally use optical microscopes to detect surface defects on wafers. Optical microscopes can be divided into two categories: macro and micro according to their magnification. Macro usually detects defects larger than 0.1mm, and micro detects defects smaller than 0.1mm.

[0003] In the detection process, the camera is scanned and imaged above the wafer manually or according to the program setting, and the display interface of the microscope is manually observed to mark the defect location. Whether using manual detection or traditional image processing methods, when detecting defects in such images, it is often difficult to identify defects, the judgment is inaccurate, and there are too many false alarms.

[0004] For various defects on the wafer surface, the operator is usually required to switch between different light sources and shooting angles to identify the defect level at the best position. This not only reduces the efficiency of quality inspection, but also causes manufacturers to rely on the operator's experience, and the operator's experience is difficult to form quantifiable expert knowledge.

[0005] There is no data association between macroscopic scanning and microscopic scanning, and they are completed by two independent machines, which results in a certain reduction in efficiency and waste of manpower. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides a macro-micro combined wafer defect intelligent positioning system and a working method thereof, using one system to simultaneously solve the problems of macro and micro detection of wafer defects.

[0007] To achieve the above object, the technical solution of the present invention is as follows:

[0008] A macro-micro combined wafer defect intelligent positioning system comprises a macro camera group, a motion mechanism, a computer and a micro camera. The motion mechanism is connected to the macro camera group in a transmission manner. Both the macro camera group and the micro camera are connected to the computer. The computer processes image data collected by the macro camera group and the micro camera by establishing a defect detection model, an optimal angle prediction model and a macro-micro correlation model.

[0009] As a preferred technical solution, the macro camera group includes three macro cameras, and the three macro cameras scan the wafer at three angles: top view, 0° side view, and 90° side view.

[0010] As a preferred technical solution, a working method of a macro-micro combined wafer defect intelligent positioning system includes the following steps:

[0011] Step 1, macro camera scanning;

[0012] Step 2: Shoot at the best angle;

[0013] Step 3: Shoot from a macro angle;

[0014] Step 4: Macro defect detection;

[0015] Step 5: Micro-probability prediction;

[0016] Step 6: Microscopic photography;

[0017] Step 7: Micro defect detection;

[0018] Step 8: Summarize the results.

[0019] As a preferred technical solution, in step two, the computer determines the best shooting angle of the current position based on the best angle prediction model through three images of overhead view, 0° side view, and 90° side view taken by the macro camera group.

[0020] As a preferred technical solution, in step three, the motion mechanism drives the macro camera group to move and rotate to the best shooting angle. At this time, only the bird's-eye view camera is positioned and only the bird's-eye view camera is used for imaging.

[0021] As a preferred technical solution, in step four, the computer performs defect recognition on the image taken at the optimal angle based on the defect detection model, and records the recognition results in the computer's database.

[0022] As a preferred technical solution, in step five, at the best shooting angle, the best angle image and the three images of the initial scan are used simultaneously to perform micro probability prediction through the macro-micro correlation model to determine whether there is a high probability of micro defects at the current position. The computer uses an algorithm to mark the positions with a high probability of micro defects in the macro scene one by one, and the results of a single field of view are saved in the micro scanning position set of the database.

[0023] As a preferred technical solution, in step six, according to the set of microscopic scanning positions, the microscopic cameras are arranged one by one and the wafer is imaged and photographed, and only one picture is taken at each microscopic position.

[0024] As a preferred technical solution, in step seven, defects are identified on the image at the microscopic position through a defect detection model.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The macro-micro combined wafer defect intelligent positioning system of the present invention aims to use one system to solve the problems of macro and micro detection at the same time. In order to reduce the number of manual re-inspections, the present invention uses a defect detection model to identify defective images, and automatically identifies defects by extracting features from a training set. In order to reduce the number of repeated shootings from multiple angles, the present invention constructs an optimal angle prediction model, which only requires shooting three viewing angles to predict the optimal shooting position, thereby reducing the time for repeated shooting. In order to further reduce the redundancy of repeated shooting at the macro and micro levels, the present invention constructs a macro-micro correlation model, which predicts the high-probability position of micro defects through the results of macro shooting, and realizes precise shooting combining macro and micro. By using the above three models in combination and embedding them into existing business processes, high-speed, accurate and automatic positioning of defect positions can be achieved, which improves the overall quality inspection efficiency and helps companies improve yield, process control and knowledge accumulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the arrangement of the macro camera group in the macro-micro combined wafer defect intelligent positioning system of the present invention;

[0028] Figure 2 It is a distribution diagram of the field of view of macroscopic scanning of a wafer in the working method of the wafer defect intelligent positioning system combining macroscopic and microscopic aspects of the present invention;

[0029] Figure 3 It is a schematic diagram of the best angle shooting position in the working method of the macro-micro combined wafer defect intelligent positioning system of the present invention;

[0030] Figure 4 It is a schematic diagram of the microscopic shooting position in the working method of the wafer defect intelligent positioning system combining macroscopic and microscopic aspects of the present invention;

[0031] Figure 5 It is a flow chart of the working method of the macro-micro combined wafer defect intelligent positioning system of the present invention. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below in conjunction with specific implementation methods:

[0033] A macro-micro combined wafer defect intelligent positioning system comprises a macro camera group, a motion mechanism, a computer and a micro camera. The motion mechanism is connected to the macro camera group in a transmission manner. Both the macro camera group and the micro camera are connected to the computer. The computer processes image data collected by the macro camera group and the micro camera by establishing a defect detection model, an optimal angle prediction model and a macro-micro correlation model.

[0034] Defect Detection Model

[0035] For macro / micro defect detection, the present invention uses a neural network model based on deep learning for processing. The training process is as follows: first, prepare a data set, which includes captured images and annotation information, and the annotation box is a rectangular box. Several sets of data are retained in the training set, and 70% are randomly selected as the training set, 20% as the test set, and 10% as the verification set.

[0036] In a single training process, the loss function in the training set is used as the training direction, the loss function in the test set is used as the basis for parameter adjustment, and the final effect is characterized by the loss function in the validation set. The loss function is the joint loss of the classification and the IOU of the rectangular box.

[0037] Optimal Angle Prediction Model

[0038] The optimal angle prediction model uses a parameter regression model based on a neural network. Its input is three captured images, and its output is a three-dimensional vector representing its optimal shooting posture. The backbone of the model uses the resnet-54 or efficientnet architecture. The present invention sets the convolution output of the final fully connected layer to a 3x1 vector. After normalization, the softmax link is removed. This vector can represent the shooting posture.

[0039] The training process is as follows: the data set includes the captured images and the pose vectors of the best position. In the process of collecting the training set, the machine needs to drive the macro camera group to first take three images at the defect position. Then the personnel continue to operate the machine to rotate, and the personnel observe the field of view until the best shooting angle is reached. At this time, the current pose vector is recorded as the labeling result of this set of data.

[0040] The loss function during training is: the Euclidean distance between the currently calculated pose vector and the annotated pose vector.

[0041] Micro-probabilistic forecasting model

[0042] The micro-probability prediction model uses a parameter regression model based on a neural network, and uses three macro images and four optimal position macro images as a set. The output of the model is a floating point number between 0 and 1, which represents the probability of the occurrence of micro defects, with the highest probability being 1. The backbone of the model uses the resnet-54 or efficientNet architecture. The present invention sets the convolution output of the final fully connected layer to a 1x1 vector, performs normalization, and removes the softmax link. This vector can represent the probability.

[0043] The training process is as follows: first prepare the data set, which includes the captured images and the annotation probabilities. The captured images are three macro images of the current state and a macro image of the optimal position. In the process of collecting the training set, the personnel first confirm the defect position under a microscopic microscope, set the annotation probability to 1, and after locating the defect, manually switch to the macro camera and take three macro images. Then continue to operate the machine to rotate and take a macro image of the optimal position at the optimal position angle. In this way, four images and a probability value constitute a set of training samples. In order to balance the samples, the personnel need to repeat the above data collection operation at the microscopic defect-free position and set the annotation probability to 0, so that a large number of negative samples can be collected.

[0044] The loss function during training is: the square of the difference between the currently calculated probability and the labeled probability.

[0045] like Figure 5 As shown, a working method of a macro-micro combined wafer defect intelligent positioning system includes the following steps:

[0046] Step 1, macro camera scanning;

[0047] In this step, the macro camera group is driven by the motion mechanism to scan and image all wafers. Figure 1 The diagram is a schematic diagram of the arrangement of the macro camera group, which is composed of three macro cameras. The three macro cameras are responsible for the top view, 0° side view, and 90° side view, respectively, and can image the same target position three times at one time. Figure 2 As shown, the field of view distribution of the initial scan on a wafer surface. Since it is a full-field scan, the field of view covers the entire wafer.

[0048] Step 2: Shoot at the best angle;

[0049] After the initial scan, the system obtains all images of the wafer surface, including three images of front view, 0° side view, and 90° side view at each position. The best angle prediction model is used to determine the best shooting angle at the current position through these three images. The function of this model is to predict the best shooting angle through a neural network and predict whether there are defects at this position that are worth shooting at the best angle. This process is the initial screening of wafer images.

[0050] Step 3: Shoot from a macro angle;

[0051] After the system obtains the result of step 2, the motion mechanism drives the macro camera group to move to a certain field of view position, the computer calls the best angle result of step 2, and the motion mechanism drives the macro camera to rotate to the best shooting angle, and then further imaging. At this time, only the top-view camera is positioned, so only the top-view camera is used for imaging. Figure 3 The fields of view that require optimal angle shooting are listed, and after screening, only three positions need to be shot.

[0052] Step 4: Macro defect detection;

[0053] At the best shooting angle, there is only one image from the top-down camera. The system uses a deep learning defect detection model based on a neural network to identify defects in the image at the best position, and the identification results are recorded in the database.

[0054] Step 5: Micro-probability prediction;

[0055] At the best shooting angle, microscopic probability prediction is performed simultaneously. This step uses the best angle image and the three images of the initial scan for joint prediction. It determines whether there is a high probability of microscopic defects at this location. The computer uses an algorithm to mark the high-probability locations in the macroscopic scene one by one and send them to the next step for microscopic shooting. The results of a single field of view are saved in the microscopic scanning position set of the database.

[0056] Step 6: Microscopic photography;

[0057] In the microscopic shooting phase, the imaging device of the machine is replaced with a microscopic camera. According to the microscopic scanning position set obtained in step 5, the system drives the microscopic camera to the position one by one by the motion mechanism for imaging. In this step, only one picture is taken for each microscopic position, which is reserved for the next step of algorithm recognition. Figure 4 The prediction of microscopic shooting positions is explained. This step is a more refined screening, and microscopic shooting is only required at a few positions.

[0058] Step 7: Micro defect detection;

[0059] At each microscopic scanning position, there is only one image from the microscopic camera. The system uses a deep learning defect detection model based on a neural network to identify defects in images at microscopic positions.

[0060] Step 8: Result summary: The system summarizes all previously calculated images and results and provides feedback to subsequent processes.

[0061] This embodiment is only a further explanation of the invention rather than a limitation of the invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as they are within the scope of the claims of the invention, they are protected by the patent law.

Claims

1. A wafer defect intelligent positioning system combining macro and micro, It is characterized in that The system comprises a macro camera group, a motion mechanism, a computer and a micro camera. The motion mechanism is connected to the macro camera group in a transmission manner. Both the macro camera group and the micro camera are connected to the computer. The computer processes the image data collected by the macro camera group and the micro camera by establishing a defect detection model, an optimal angle prediction model and a macro-micro correlation model.

2. According to the macro-micro combined wafer defect intelligent positioning system of claim 1, It is characterized in that The macro camera group includes three macro cameras, and the three macro cameras scan the wafer at three angles: top view, 0° side view, and 90° side view.

3. The working method of the macro-micro combined wafer defect intelligent positioning system according to claim 1, It is characterized in that The method comprises the following steps: Step 1, macro camera scanning; Step 2: Shoot at the best angle; Step 3: Shoot from a macro angle; Step 4: Macro defect detection; Step 5: Micro-probability prediction; Step 6: Microscopic photography; Step 7: Micro defect detection; Step 8: Summarize the results.

4. The working method of the wafer defect intelligent positioning system combining macro and micro according to claim 3, It is characterized in that In the step 2, the computer determines the best shooting angle of the current position based on the best angle prediction model through the three images of the top view, 0° side view, and 90° side view taken by the macro camera group.

5. The working method of the macro-micro combined wafer defect intelligent positioning system according to claim 4, It is characterized in that In step three, the motion mechanism drives the macro camera group to move and rotate to the best shooting angle. At this time, only the top-view camera is positioned and only the top-view camera is used for imaging.

6. The working method of the macro-micro combined wafer defect intelligent positioning system according to claim 5, It is characterized in that In step 4, the computer performs defect recognition on the image taken at the optimal angle based on the defect detection model, and records the recognition result in a database of the computer.

7. The working method of the macro-micro combined wafer defect intelligent positioning system according to claim 6, It is characterized in that In step five, at the best shooting angle, the best angle image and the three images of the initial scan are used simultaneously to perform micro probability prediction through the macro-micro correlation model to determine whether there is a high probability of micro defects at the current position. The computer uses an algorithm to mark the positions with a high probability of micro defects in the macro scene one by one, and the results of a single field of view are saved in the micro scanning position set of the database.

8. The working method of the macro-micro combined wafer defect intelligent positioning system according to claim 7, It is characterized in that In step six, according to the set of microscopic scanning positions, the microscopic cameras are arranged one by one and the wafer is imaged and photographed, and only one picture is taken at each microscopic position.

9. The working method of the wafer defect intelligent positioning system combining macro and micro operations according to claim 8, It is characterized in that In step seven, defects are identified on the image at the microscopic position using a defect detection model.

Images