Wafer quantity detection automatic focus tracking method and system based on pattern influence quantity
By identifying smooth regions and training a pattern influence model, the problem of unstable autofocus in wafer optical quantity inspection is solved, improving imaging quality and inspection accuracy. It is suitable for optical quantity inspection of patterned semiconductor wafers.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-26
- Publication Date
- 2026-03-27
AI Technical Summary
In wafer optical quantity inspection, especially for wafers with deeper lithography processes, the autofocus control is unstable, leading to inconsistent imaging quality and affecting the accuracy and stability of the inspection.
By identifying smooth regions suitable for data acquisition, data is collected and processed using a Bare Wafer, a Pattern Influence Model is trained, the wafer rig is controlled to maintain the optimal focal plane, and the U-Net3D deep learning model is used for automatic focus tracking to correct the Pattern Influence in real time.
It improves the accuracy and stability of wafer measurement, reduces the probability of false defects, and is suitable for optical measurement of patterned semiconductor wafers.
Smart Images
Figure CN121742142A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the field of optical control of semiconductor quantity detection, and particularly relates to a wafer quantity detection automatic focus pursuit method and system based on pattern influence quantity, and is particularly suitable for quantity detection of deep photolithography process wafers. BACKGROUND
[0002] Focusing is an important optical control technology in wafer optical quantity detection equipment, and mainly controls the movement of the wafer Z direction through a machine table, so that the surface of the wafer is always on the best focal plane.
[0003] However, when the Z direction of the wafer surface changes greatly, the machine table needs to be adjusted at a faster frequency and needs a larger stroke, which has a high requirement on the stability of the machine table as a whole. In order to ensure the stability of the machine table, the frequency and stroke of the Z axis adjustment are as small as possible, and the path of the Z axis needs to be as smooth as possible and the change amount in a short time needs to be as small as possible.
[0004] For example, in the automatic focusing process, when a wafer with a relatively complex process and a deep groove is encountered, the stage cannot accurately ensure that the surface of the wafer is always on the best focal plane, so that the imaging quality cannot be guaranteed to be the same on different dies. In bright field wafer detection, the wafer is placed on a chuck, and the stage controls the movement of the chuck in the XYZ three directions to detect different positions of the wafer, but the chuck itself has designs such as concave, gas sealing ring and inclination, which will affect the base surface type of the wafer.
[0005] Therefore, in the automatic focusing process, when a wafer with a relatively complex process and a deep groove is encountered, the stage cannot accurately ensure that the surface of the wafer is always on the best focal plane, so that the imaging quality cannot be guaranteed to be the same on different dies, which affects the wafer quantity detection accuracy and stability. SUMMARY
[0006] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a kind of, which can solve the problem of unstable control of automatic focus pursuit function when the current bright field detection equipment detects deep photolithography process wafers.
[0007] Design principle: In order to make the imaging quality of each Die (die or wafer) the same, it is beneficial for subsequent defect detection. When collecting the influence data of the pattern, a relatively smooth area is needed to ensure that the substrate surface type is relatively consistent. The scheme mainly includes: first, selecting the area suitable for collecting data in the chuck and collecting data, then training the model to obtain the model capable of inferring the influence of the pattern, and finally removing the influence when controlling the stage to make the control path of the stage consistent with the substrate surface type of the wafer.
[0008] A wafer quantity detection automatic focus pursuit method based on pattern influence, the wafer quantity detection automatic focus pursuit method comprising: S1, using a Bare Wafer to find a smooth area suitable for collecting data; S2, data collection and processing, collecting automatic focusing data of the substrate smooth Die in the smooth area of the Bare Wafer, and filtering the collected data; S3, data combination to obtain the best focus surface Ztarget, , wherein Ztarget is the best focus surface, Zpos is the actual position, and Zref is the defocus amount from the current position to the best focus surface; S4, model training, scanning the wafer based on the best focus surface and the pattern influence of the chuck surface type data to obtain model training data, and recording the training data in the model in units of Die for training; S5, model use for automatic focus quantity detection.
[0009] Further, the finding of the smooth area comprises: S11, selecting a Bare Wafer reflecting the chuck surface type change; S12, measuring the overall surface type data of the Bare Wafer; S13, one-to-one corresponding the surface type data of the Bare Wafer and the stage coordinates, and setting a smooth threshold T; S14, selecting a smooth area, selecting an area within the smooth threshold T, and marking these areas as smooth areas through stage coordinates.
[0010] Further, the data collection and processing comprises: S21, selecting a substrate smooth Die in the smooth area; S22, collecting focusing data of the substrate smooth Die; S23, filtering the data.
[0011] Further, the model training comprises: S41, calculating a relatively coarse pattern influence: the best focus surface minus the chuck surface type data at the corresponding position; S42, calculating the true value of the pattern at the corresponding position of the reference Die: averaging the coarse pattern influences of different Dies at the same position; S43, inputting data to train the model, scanning the best focus surface data corresponding to multiple rows of swath and taking the data in units of Die as model input data to record in the model for training.
[0012] Further, the model uses include: S51, detecting the region pre-scanning to obtain focusing data; S52, calculating the pattern influence quantity of the to-be-detected region by the model through the scanned data; and S53, scanning the pattern quantity, wherein the model subtracts the pattern influence quantity in real time during scanning, so that the machine table always follows the substrate surface type, and ensures that the defocus quantities of the dies in the same place are the same.
[0013] The application further provides a wafer quantity detection automatic focus pursuit system based on a pattern influence quantity, which comprises a wafer chuck for carrying a wafer, a wafer machine table for controlling the movement of the wafer in XYZ directions, and an automatic focus pursuit control module.
[0014] Further, the automatic focus pursuit control module comprises a smooth region identification module for identifying a smooth region suitable for data collection on the wafer chuck by using a Bare Wafer, wherein the surface type change of the region is within a smooth threshold T, and the region is described by wafer machine table coordinates; a pattern influence quantity calculation module for collecting automatic focusing data on the identified smooth region by using a Pattern Wafer, wherein the automatic focusing data comprises a defocus quantity and a stage real-time position, the data noise is processed by low-pass filtering, and the position of the best focus surface is calculated; and a quantity detection focus pursuit control model for constructing a pattern influence quantity model based on a U-Net3D deep learning model training, and controlling the wafer machine table to follow the wafer substrate surface type during actual quantity detection of the patterned wafer.
[0015] Compared with the prior art, the application has the beneficial effects that: the application performs special smoothing processing on the control data of automatic focus pursuit, so that the imaging effects of the same features in the same row of the wafer are close, thereby reducing the probability of false defects caused by focus pursuit problems, and the application can be widely used in the field of optical quantity detection of semiconductor patterned wafers. BRIEF DESCRIPTION OF DRAWINGS
[0016] Figure 1 FIG. 1 is a flowchart of a wafer quantity detection automatic focus pursuit method according to the application; Figure 2 FIG. 2 is a comparison diagram of surface type changes before and after the pattern influence quantity of a die is removed; Figure 3 FIG. 3 is a comparison diagram of swath scanning in a row considering the pattern influence quantity; Figure 4 FIG. 4 is a diagram of defocus quantities of different dies in the same row of swath considering the pattern influence quantity. DETAILED DESCRIPTION
[0017] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.
[0018] A wafer quantity detection automatic focus pursuit method based on a pattern influence quantity, referring to Figure 1 The wafer quantity detection automatic focus pursuit method comprises the following contents.
[0019] S1, find a smooth area suitable for data collection using a Bare Wafer. The finding of the smooth area comprises the following steps.
[0020] S11, select a Bare Wafer reflecting the chuck surface type change.
[0021] S12, measure the overall surface type data of the Bare Wafer.
[0022] S13, one-to-one correspondence of the surface type data of the Bare Wafer and the stage coordinates, and setting a smooth threshold T.
[0023] S14, selecting a smooth area, selecting an area within the smooth threshold T, and marking these areas as smooth areas through stage coordinates.
[0024] The present application considers that the area of the surface type change within a certain threshold T is relatively smooth, and first finds all independent areas meeting the conditions, and marks these areas through stage coordinates, so as to find the Die in the chuck smooth area after loading the Pattern wafer.
[0025] S2, data collection and processing, collecting the automatic focusing data of the substrate smooth Die in the smooth area of the Bare Wafer, and filtering the collected data.
[0026] Specifically, the data collection and processing comprises the following steps.
[0027] S21, screening the substrate smooth Die in the smooth area. The present solution provides that if the area of the Die in the smooth area reaches 80%, the Die is determined to be a substrate smooth Die. Of course, the numerical value can be adjusted, such as any value within the range of 50%-100%, and the preferred numerical value is one of 80%, 85%, 90% and 95%.
[0028] S22, collect the focus data of the substrate smooth Die. The focus data includes the calculation of the defocus amount and the real-time true position of the stage.
[0029] S23, data filtering processing. In the focus data, the calculation of the defocus amount has sensor self-noise and other interference, and the jitter of the stage also causes unnecessary noise. The collected data needs to be filtered, and the low-pass filter is used in the application.
[0030] S3, data combination to obtain the best focal plane Ztarget, , wherein Ztarget is the best focal plane, Zpos is the actual position, and Zref is the defocus amount from the current position to the best focal plane. Data combination is to combine the two sets of data in the previous step to obtain the final position of the best focal plane.
[0031] S4, model training, based on the best focal plane and the pattern influence amount of the chuck surface type data, the wafer is scanned to obtain the model training data, and the training data is recorded in the model in units of Die for training.
[0032] In a specific example, considering that the deep learning model depends on a large amount of training data, the application selects the U-Net3D model which has less training data requirement. The model training includes the following steps.
[0033] S41, calculate the coarse pattern influence amount: the best focal plane minus the chuck surface type data at the corresponding position.
[0034] S42, calculate the reference Die corresponding position pattern true value: the average of the coarse pattern influence amount of different Dies at the same position.
[0035] The specific expression formula is: ; , wherein D is a set of pattern true values in a Die, there are n points, m Dies are used to calculate the true value of each point, Z is the current best focal plane, and C is the surface type data of the chuck at the corresponding point.
[0036] S43, input data training model, the best focal plane data corresponding to the swath is scanned and used as the model input data in units of Die to record in the model for training, to obtain a model meeting the requirements, and the model can obtain the pattern influence amount of the current wafer on the same chuck.
[0037] S5, model use for automatic focus amount detection. The model use specifically includes the following contents.
[0038] S51, detect the region pre-scanning to obtain the focus data.
[0039] S52, calculate the pattern influence amount of the to-be-detected area through the scanned data.
[0040] S53, wafer map detection, when scanning, the model subtracts the pattern influence amount in real time, so that the stage always follows the substrate surface type, and ensures that the defocus amounts of each Die at the same place are the same.
[0041] In a specific example, before running the wafer, the area to be detected is scanned in advance, and only three consecutive Dies are scanned to reduce the time as much as possible. The pattern influence amount of the to-be-detected area is calculated through the scanned data, and when scanning, the influence of the pattern can be subtracted in real time, so that the stage always follows the substrate surface type, and ensures that the defocus amounts of each Die at the same place are the same.
[0042] The verification effect of the wafer map detection automatic focus pursuit method is as follows.
[0043] Referring to Figure 2 is a comparison before and after removing the pattern influence amount of a Die, wherein the blue color with a larger change is the influence amount of the pattern plus the chuck, and the yellow-brown color with a smaller change is the surface type change after removing the pattern influence. It can be seen that the surface type change based on the pattern influence amount is small, and the subsequent quantity detection will be more stable.
[0044] Referring to Figure 3 is a wafer map of a row of swath, and the result of removing the pattern influence amount is relatively flat, and the original path to be followed is relatively sharp. Through the application, the problem of frequent adjustment of the Z axis can be effectively alleviated.
[0045] Referring to Figure 4 In a row of swath, the defocus amounts of different Dies are different. It can be seen that through the scheme of the application, the defocus amounts of each Die at the same place can be ensured to be the same.
[0046] The application also provides a wafer map detection automatic focus pursuit system based on a pattern influence amount, which comprises: a wafer chuck for carrying a wafer; a wafer stage for controlling the movement of the wafer in XYZ directions; and an automatic focus pursuit control module, which is based on the wafer map detection automatic focus pursuit method and controls the Z direction movement of the wafer stage through the wafer stage to keep the wafer surface at the best focus surface, so as to improve the accuracy and stability of the wafer map detection.
[0047] The automatic focusing control module comprises: a smooth area identification module, which is used for identifying a smooth area suitable for collecting data on a wafer chuck using a Bare Wafer, the surface type of the area changes within a smooth threshold T, and the area is described through wafer stage coordinates; a Pattern influence quantity calculation module, which is used for collecting automatic focusing data on the identified smooth area using a Pattern Wafer, including defocus quantity and stage real-time position, processing data noise through low-pass filtering, and calculating the best focal plane position; and a quantity detection focusing control model, which is constructed based on a U-Net3D deep learning model training Pattern influence quantity model, and controls the wafer stage to follow the wafer base surface type during actual quantity detection of the Pattern Wafer.
[0048] The smooth area identification module comprises: a data processing unit, which is used for smoothing the automatic focusing control data; a surface type measurement subunit, which is used for measuring the overall surface type data of the Bare Wafer and corresponding to the stage coordinates; and a region definition subunit, which is used for defining the area with a surface type change within a threshold T as a smooth area, and describing the areas through stage coordinates.
[0049] The Pattern influence quantity calculation module further comprises: a data acquisition subunit and a data processing subunit.
[0050] The data acquisition subunit is used for screening the Die with a relatively smooth base surface type, which is defined as the smooth area area ratio in the Die being greater than or equal to 80%; and the data processing subunit is used for collecting defocus quantity and stage real-time position data, processing noise through low-pass filtering, and calculating the best focal plane position Ztarget, wherein Zpos is the actual position of the stage, and Zref is the defocus quantity.
[0051] The quantity detection focusing control model specifically comprises an influence quantity calculation subunit, a true value calculation subunit, a model training subunit, a scanning subunit, a model influence quantity calculation subunit, and a control adjustment subunit. The specific introduction is as follows.
[0052] The influence quantity calculation subunit is used for subtracting the corresponding chuck surface type data from the best focal plane data to obtain the Pattern influence quantity.
[0053] The true value calculation subunit is used for corresponding the influence quantity to the Die coordinates, and calculating the mean value of the influence quantity of the same position of multiple Dies as the true value.
[0054] The model training subunit is used for training using a U-Net3D model, and the input is the best focal plane data in units of Die.
[0055] The scanning subunit is used for scanning three continuous Die areas before Run Wafer detection.
[0056] A model impact amount calculation subunit is configured to calculate the Pattern impact amount by using a model.
[0057] A control adjustment subunit is configured to subtract the Pattern impact amount in real time in stage control to ensure that the defocus amounts of dies at the same position are consistent.
[0058] In a specific hardware structure, a sensor module is used to detect the defocus amount, and various data interfaces receive the instructions of the automatic focus control module and control the movement of the stage.
[0059] The defocus amount detection and focus control model is trained based on the U-Net3D model in the following steps.
[0060] The Ztarget data collected by multiple swath lines is divided into a dataset in units of dies.
[0061] The dataset is input into the U-Net3D model for training, and the output is the Pattern impact amount model.
[0062] In the scheme of the present application, the wafer includes a bare wafer (Bare Wafer) and a pattern wafer (Pattern Wafer), wherein the Bare Wafer is used to identify the surface type change of the chuck, and the Pattern Wafer is used to calculate the Pattern impact amount.
[0063] The following terms are explained.
[0064] wafer, refers to a wafer.
[0065] Bare Wafer, bare crystal, bare die, or patternless wafer.
[0066] Pattern Wafer, patterned wafer or patterned wafer.
[0067] chuck, wafer chuck, used to hold the wafer.
[0068] stage, in the semiconductor field, stage refers to a wafer machine, used to control the movement of the wafer in XYZ direction, some can control the angle of rotation or flatness adjustment in X-Y plane.
[0069] Die, die or wafer.
[0070] Base surface type, in semiconductor manufacturing, the base surface type of the wafer usually refers to its crystal orientation or surface structure, common types include single crystal silicon substrate and special structure substrate.
[0071] Swath, in the semiconductor field, "swath" generally means scanning width or scanning path, and in the present application, it specifically refers to the scanning path.
[0072] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A wafer quantity detection automatic focus pursuit method based on a pattern influence quantity, characterized in that, The wafer thickness detection automatic focusing method comprises: S1, using a Bare Wafer to find a smooth area suitable for data collection; S2, data collection and processing, collecting automatic focusing data of the substrate smooth Die in the smooth area of the Bare Wafer, and filtering the collected data; S3, data combination obtains the best focal plane Ztarget, where Ztarget is the best focal plane, Zpos is the actual position, and Zref is the defocus amount from the current position to the best focal plane. S4, model training, obtaining model training data by scanning the wafer based on the best focal plane and the pattern influence amount of the chuck surface data, and recording the training data in units of Die to train the model; S5, model use for automatic focusing amount detection.
2. The wafer amount detection auto-focusing method according to claim 1, characterized by: The finding of the smooth area comprises: S11, selecting a Bare Wafer reflecting the chuck surface type change; S12, measuring the overall surface data of the Bare Wafer; S13, one-to-one corresponding the surface data of the Bare Wafer and the stage coordinates, and setting a smooth threshold T; S14, selecting a smooth area, selecting an area within the smooth threshold T, and marking these areas as smooth areas through stage coordinates.
3. The wafer amount detection auto-focusing method according to claim 1, wherein: The data collection and processing comprises: S21, selecting a substrate smooth Die in the smooth area; S22, collecting focusing data of the substrate smooth Die; S23, filtering the data.
4. The wafer amount detection auto-focusing method according to claim 3, wherein: In S21, if the area of the Die in the smooth area reaches 80%, the Die is determined to be a substrate smooth Die.
5. The wafer amount detection auto-focusing method according to claim 3, wherein: In S22, the focusing data comprises the calculation of the defocusing amount and the real-time true position of the stage.
6. The wafer amount detection auto-focusing method according to claim 3, wherein: The focusing data is processed by low-pass filtering.
7. The wafer amount detection auto-focusing method according to claim 1, wherein: The model training comprises: S41, calculating a relatively coarse pattern influence amount: the best focal plane minus the chuck surface data at the corresponding position; S42, calculating the pattern true value at the corresponding position of the reference Die: averaging the coarse pattern influence amounts of different Dies at the same position; S43, inputting the data to train the model, scanning the best focal plane data of multiple rows of swathes and taking the data in units of Die as model input data to record in the model for training.
8. The wafer amount detection auto-focusing method according to claim 1, wherein: The model use comprises: S51, detecting the scanning area to obtain focusing data; S52, calculating the pattern influence amount of the to-be-detected area by the model through the scanned data; S53, swath detection, when scanning, the model subtracts the pattern influence amount in real time, so that the stage always follows the substrate surface, and ensures that the defocusing amounts of different Dies at the same position are the same.
9. A wafer quantity detection auto-focusing system based on a pattern influence quantity, characterized by, It comprises: A wafer chuck for carrying a wafer; A wafer stage for controlling the movement of the wafer in XYZ directions; An automatic focusing control module based on the wafer thickness detection automatic focusing method of any one of claims 1-8, and controlling the Z-direction movement of the wafer stage through the wafer stage to keep the wafer surface at the best focal plane, so as to improve the accuracy and stability of wafer thickness detection.
10. The wafer amount detection auto-focusing system according to claim 9, wherein The automatic focusing control module comprises: A smooth area identification module for identifying a smooth area suitable for data collection on the wafer chuck using a Bare Wafer, the surface type change of the area being within a smooth threshold T, and the area being marked through wafer stage coordinates; A pattern influence quantity calculation module is configured to use the pattern wafer to collect automatic focusing data, including defocus quantity and stage real-time position, on the identified smooth area, process data noise through low-pass filtering, and calculate the best focal plane position. A quantity detection tracking control model is configured to train and construct a pattern influence quantity model based on a U-Net3D deep learning model, and control the wafer stage to follow the wafer base surface profile when there is a pattern wafer actual quantity detection.