Method for Measuring and Monitoring Actual Pixel Size of Charged Particle Beam Scanning Imaging Device

Abstract

The present invention discloses a method for measuring and monitoring the actual pixel size of a charged particle beam scanning imaging device, which is used for a wafer with a pattern. The measurement method includes: using a charged particle beam scanning imaging device to perform an offline comprehensive measurement FM on a wafer with a pattern to obtain the actual pixel size of each pixel to be measured; performing a partial measurement QM based on the comprehensive measurement FM method, which is used for the device to measure the actual pixel size of some pixels to be measured online, which does not occupy machine time, has a high speed and low time cost; the monitoring method includes a closed-loop process, including the use of an offline comprehensive measurement FM and an online partial measurement QM method, with different measurement frequencies, and an automatic tuning mechanism for the measurement frequency, which can more accurately grasp the timing of implementing the comprehensive measurement FM of the actual pixel size. It can avoid untimely measurement that affects the main work of the equipment, and can also avoid wasting machine time.

Description

Technical Field

[0001] The invention relates to the technical field of semiconductor equipment, and in particular to a method for measuring and monitoring the actual pixel size of a charged particle beam scanning imaging device. Background Art

[0002] In the front-end manufacturing process of semiconductor large-scale integrated circuits, electron beam scanning imaging equipment is often needed. Charged particle beam scanning imaging equipment is, for example, a type of equipment based on a scanning electron microscope (SEM), which is used to detect or re-inspect semiconductor wafers for defects. This type of equipment includes electron beam initial inspection equipment (E-Beam Inspection, EBI) or electron beam re-inspection equipment (E-Beam Review, EBR), as well as critical dimension (Critical Dimension, CD) measurement equipment, namely CD-SEM, which are all electron beam scanning imaging systems in principle. The following is an explanation using the EBR device as an example, but EBI and CD-SEM are not excluded. In the application of the device to semiconductor wafers, an accurate method is required to obtain the actual pixel size of its imaging, which is used to accurately estimate the object to be measured, such as the actual physical size of the defect or the actual physical size of the morphology.

[0003] refer to Figure 1 , the EBR device 100 usually includes a mechanical motion platform 110, which can move and rotate in the X, Y, and Z directions, and has an electrostatic tray (Electro-Static Chuck, E-Chuck) 120 on which the wafer can be placed, on which the wafer (Wafer) 111 can be placed. The EBR device usually also includes an optical microscope imaging system (Optical Microscope, OM) 130, which has a lower magnification but a larger field of view (Field of View, FOV), and is usually used for auxiliary work such as primary wafer alignment. The EBR device also includes a core task component 140, which is the electron optical system 140, that is, the above-mentioned scanning electron microscope, SEM system, including a lens barrel, which has functional components such as electron emission, focusing, beam limiting, scanning, biasing, etc., and also includes the collection of electrons (mainly secondary electrons) emitted from the wafer surface and the signal amplification processing circuit part. The EBR device also includes a computer 150, on which the software running includes a database, which can be used to process and save data and display images.

[0004] refer to Figure 2A and Figure 2B, when the SEM system in the EBR device forms an image, the electron beam is like 210, the focal plane is 211, the depth of focus is 212, the z - direction is the density of the electron beam 210, and the x - direction is the x - coordinate direction on the wafer. During operation, the SEM system scans a predefined area back and forth. When scanning, the sampling intervals in the X and Y directions on the wafer surface (and within its focal depth / range) are Δx and Δy, and the system needs to stay at each sampling point for a certain time to accumulate enough outgoing electrons (to achieve a certain signal - to - noise ratio). The obtained SEM image is, in theory, roughly equivalent to the convolution of the beam spot shape (approximately Gaussian distribution) of the electron beam when it reaches the wafer surface and the wafer surface topography (including different materials and structures). When the theoretical field of view / FOV of the SEM system (in most applications, the FOVs in the X and Y directions are set to be the same, i.e., FOVx = FOVy = FOV, but in reality, there are differences between FOVx and FOVy, that is, there are differences between Px and Py. If the difference is too large, it may cause image deformation and other problems) and the number of pixels of the image in the X and Y directions are determined, the pixel sizes in the X and Y directions, namely the theoretical pixel size (Nominal Pixel Size, NPS), are also determined. The theoretical pixel size is the theoretical field - of - view size divided by the image size (for example, the image size is 1024 / 2048 pixels in width / height). Therefore, the theoretical pixel size is Δx and Δy. In actual use, the operating parameters and states of the SEM system often drift. For example, due to 1) a slight change in the actual beam spot size of the system or the scanning sampling interval determined by the device control circuit; 2) a change in the working distance from the objective lens to the wafer surface in the SEM system (due to, for example, a drift in the Z - direction of the mechanical motion platform or a change in the thickness of different types of wafers, thus changing the actual sampling interval on the wafer), the theoretical pixel size and the actual pixel size are different. With multiple factors superimposed, the difference between the theoretical pixel size and the actual pixel size is relatively large. Therefore, in actual device applications, it is often necessary to obtain a more accurate actual pixel size.

[0005] Please refer to Figure 3A and Figure 3B , Method 1, a commonly used method in the prior art to obtain the actual pixel size, is: taking pictures of a patterned wafer or sample with known dimensions, and then calculating the actual pixel size. In Method 1, usually, a special sample 311 is placed on the electrostatic tray of the mechanical motion platform 310. The image 313 collected on the sample 311 contains an object 312 with a known size, having a known length, such as 5μm. However, its disadvantages are very obvious. The thickness of the sample 311 cannot be guaranteed to be close to that of the actual working wafer, resulting in a large error in the working distance, that is, the distance from its surface to the lens barrel in the SEM system, compared with the distance from the surface of the actual customer wafer to the lens barrel in the SEM system. In addition, after long - term and multiple scans, the local material of the special sample 311 is prone to damage.

[0006] In the prior art, a common method 2 for obtaining the actual pixel size is to use a special wafer with a known size pattern / object. However, its disadvantages are very obvious. First of all, the special wafer will increase the cost, and its thickness cannot be guaranteed to be close to that of different wafers of actual customers in actual work, resulting in the distance from its surface to the lens barrel in the SEM system, that is, the working distance, being different from the distance from the surface of the actual customer wafer to the lens barrel in the SEM system. Moreover, more importantly, its measurement is non-online. Each time it is used, wafers need to be loaded and unloaded, occupying precious machine time, which is very inconvenient and difficult to implement in actual production lines.

[0007] Therefore, there is currently a method 3 in the industry, that is, during the operation of the EBR equipment, there is also a method of directly using customer wafers (limited to patterned wafers) to obtain the actual pixel size. It seems more convenient and there will be no problem of height / working distance difference in the above methods 1 and 2. The process is roughly as follows:

[0008] Figure 4 Disclosed is a current method for measuring the actual pixel size by collecting images (a total of two frames of images) before and after the wafer moves and performing template matching (Pattern Matching, PM). The coordinate axes of the mechanical motion platform coordinate system are the x-axis and the y-axis, and the field of view 402 when the wafer 401 and the SEM system collect images is shown. The position of the field of view 402 is fixed. The principle is as follows: First, collect the first frame of the image, that is, the template image, and select an image therein as the template 403. The template 403 is located at (Xm, Ym) in the image coordinate system. The displacements of the mechanical motion platform in the X and Y directions are dXs and dYs respectively. In the figure, dXs and dYs are labeled as 404 and 405 respectively. Collect the second frame of the image, that is, the target image, and perform template matching to determine the matching position (Xm′, Ym′). The matching position (Xm′, Ym′) is also the coordinate in the image coordinate system. Among them, the origin of the image coordinate system is, for example, the upper left corner, the lower left corner or the center point of the image, but is not limited thereto. Divide dXs by (Xm′ - Xm), and divide dYs by (Ym′ - Ym), then the actual pixel size can be obtained.

[0009] The following gives a more detailed description of this method. Please refer to Figure 5A In method 3, the steps are as follows:

[0010] 1) Determine a position on the wafer, collect the template image 510, and select a template 511 in the template image 510;

[0011] 2) The moving mechanical platform moves synchronously with the wafer to another position to acquire the target image 512, and records the relative displacement amounts dXs and dYs of the mechanical platform. The moving positions here are limited to ensure that the target corresponding to the template is still within the target image (i.e., within the field of view of the SEM system) in the case where the mechanical platform has a known error range.

[0012] 3) Search for the template in the target image, that is, perform template matching to obtain the matching position 513.

[0013] 4) Obtain the distances dXm and dYm between the template 511 and the matching position 513, that is, dXm and dYm are the target displacements in the x - direction and y - direction respectively, specifically:

[0014] dXm = Xm′ - Xm

[0015] dYm = Ym′ - Ym

[0016] where (Xm, Ym) is the position (initial position) of the template in the template image, and (Xm′, Ym′) is the position where the template is matched in the target image (matching position).

[0017] 5) Obtain the actual pixel sizes Px and Py, specifically:

[0018]

[0019]

[0020] where the actual displacements of the mechanical platform in the x - direction and y - direction are dXs and dYs, that is, the coordinate position change of the mechanical platform is (dXs, dYs). For example, a precise code reader / laser interferometer usually equipped on the mechanical platform can give a more accurate actual arrival position of the mechanical platform. There may be a slight difference between the actual arrival position and the position that the system commands the mechanical platform to reach. For example, the normal accuracy can reach within 0.5μm.

[0021] Please refer to Figure 5A , for an LMSEM image with a relatively high magnification (hereinafter referred to as an LM image, for example, an image with a theoretical pixel size of about 100nm or more can be regarded as an LM), the method is as described above, that is, acquire the template image 520, select the template 521 from it, then move the wafer to acquire the target image 522, and perform template matching to obtain the matching position 523. Please refer to Figure 5B, the method for HM SEM images (hereinafter referred to as HM images) is the same. That is, the template image 520 is collected, the template 521 is selected from it, and then the wafer is moved to collect the target image 522, and the matching position 523 is obtained through template matching. It can be seen that the above method 3) seems to have certain progress compared with methods 1) and 2), but there are still the following problems. That is, when the user collects a template image at a certain place on the wafer, selects a template from it, and performs a single template matching, the mechanical motion platform (Stage) has a limited single-trip moving distance, and the relative error of the measurement result is relatively large.

[0022] In addition to the problems of accuracy, speed, and usage cost existing in the above actual pixel size measurement methods 1, 2, and 3 themselves, there is another more serious problem. That is, currently in the industry, the actual pixel size is measured irregularly on the above-mentioned equipment, so there is a problem of improper measurement timing: a. On the one hand, sometimes it may not be necessary to measure the actual pixel size because they have not changed significantly, and unnecessary measurements occupy the equipment time and cause waste of equipment resources, because the equipment machine time on the IC production line is extremely precious; b. On the other hand, sometimes the measurement may not be timely, so that the actual pixel size drifts / deviates, resulting in inaccurate results of the semiconductor equipment's own work. For example, the defect size of EBR / EBI or the IC topography size of CD-SEM is inaccurate, thus reducing / damaging the equipment performance and causing serious consequences such as a decline in product quality. All these are due to the lack of effective monitoring of the actual pixel size to grasp the timing of actual pixel measurement, and ultimately the existing technology is significantly insufficient in terms of accuracy, reliability, and time cost. Therefore, it is necessary to make a substantial improvement. Summary of the Invention

[0023] One aspect of the object of the present invention is to provide a method for measuring the actual pixel size of a charged particle beam scanning imaging device to solve the problem of high time cost of the measurement method in the prior art; another aspect of the object of the present invention is to provide a method for monitoring the actual pixel size of a charged particle beam scanning imaging device to accurately grasp the timing of actual pixel size measurement, neither delaying the timing of actual pixel size measurement nor wasting precious machine time and resources.

[0024] To achieve the above object, on the one hand, the present invention provides a method for measuring the actual pixel size of a charged particle beam scanning imaging device, including: using the charged particle beam scanning imaging device to perform an offline comprehensive measurement FM on a patterned wafer to obtain the actual pixel size of each pixel to be measured; performing a partial measurement QM based on the comprehensive measurement FM method for the device to online measure the actual pixel size of some pixels to be measured.

[0025] To achieve the above object, on the other hand, the present invention also provides a method for monitoring the actual pixel size of a charged particle beam scanning imaging device, including:

[0026] 1411. After the electronic optical system in the device undergoes maintenance and calibration, determine monitoring working parameters, including a second parameter Nf, a second threshold Thfi, a first parameter Nq and a first threshold Thqi, wherein the second parameter Nf is greater than the first parameter Nq, and the second threshold Thfi is greater than the first threshold Thqi;

[0027] 1412. Execute the comprehensive measurement FM and clear the second counter nf and the first counter nq at the same time;

[0028] 1413. Compare the actual pixel size measured in the comprehensive measurement FM with the second threshold Thfi. If all the actual pixel sizes measured this time are outside the second threshold Thfi, return to and re-execute step 1411; otherwise, proceed to the next step.

[0029] 1414. Save all actual pixel sizes measured in the current comprehensive measurement FM;

[0030] 1415. When the device starts or continues to perform its main work, it counts through the second counter nf and the first counter nq;

[0031] 1416. When the first counter nq reaches the second parameter Nf, return to step 1412, otherwise go to step 1417;

[0032] 1417 When the first counter nq does not reach the first parameter Nq, the process returns to step 1415 and the device continues to work as usual; otherwise, the process proceeds to step 1418;

[0033] 1418. Execute the partial measurement QM and clear the first counter nq;

[0034] 1419. When the result of the actual pixel size is greater than or equal to the second threshold value Thfi, return to step 1411 to re-calibrate and maintain the device; otherwise, when the monitoring value is less than the first threshold value Thqi, return to step 1415, and the device continues to perform its job; otherwise, when the monitoring value is less than the second threshold value Thfi but greater than or equal to the first threshold value Thqi, change the first parameter Nq and the second parameter Nf, and then return to step 1415.

[0035] On the one hand, the present invention provides a measurement method as described above. The partial measurement QM method in the present invention utilizes the device to measure the actual pixel size of some pixels to be measured online when performing its job. It is fast and does not occupy the machine, and can be used to obtain the actual pixel size of some pixels to be measured in the current situation.

[0036] On the other hand, the present invention provides a monitoring method as described above, which can accurately grasp the timing of actual pixel size measurement, neither delaying the timing of actual pixel size measurement nor wasting precious machine time and resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a schematic diagram of a related part of an EBR device in the prior art;

[0038] Figure 2A is a schematic diagram of the density distribution of an electron beam in a SEM system in the prior art;

[0039] Figure 2B is a schematic diagram of the focal length and depth of focus of an electron beam in the prior art;

[0040] Figure 3A is a schematic diagram of a special sample placed on a mechanical moving platform in the prior art;

[0041] Figure 3B is a schematic diagram of a SEM image obtained by taking a picture from a special sample in the prior art;

[0042] Figure 4 is a schematic diagram of a method for obtaining the actual pixel size of an image in the prior art;

[0043] Figure 5A is a schematic diagram of template matching in a target image using a template at a low magnification LM in the prior art;

[0044] Figure 5B is a schematic diagram of template matching in a target image using a template at a high magnification HM in the prior art;

[0045] Figure 6A is a schematic diagram of a wafer alignment method in the prior art;

[0046] Figure 6B is a schematic diagram of another wafer alignment method in the prior art;

[0047] Figure 7 is a schematic diagram of generating a template required for actual pixel size measurement in a template image at the theoretical magnification corresponding to different pixels to be measured in an embodiment of the present invention;

[0048] Figure 8 is a schematic diagram of a multi-template chained weighted template matching method in an embodiment of the present invention;

[0049] Figure 9 is a schematic diagram of a method for selecting a template in an image at a high magnification HM in an embodiment of the present invention;

[0050] Figure 10A It is a schematic diagram of a method for selecting a template for partial measurement QM under a low magnification LM in an embodiment of the present invention;

[0051] Figure 10B It is a schematic diagram of a method for selecting a template for partial measurement QM under a high magnification HM in an embodiment of the present invention;

[0052] Figure 11 It is a schematic diagram for selecting a template for partial measurement QM in a single-frame image under a high magnification HM;

[0053] Figure 12 It is a schematic diagram of a method for obtaining adjacent 2 images for partial measurement QM by using the bias of an electron optical system during actual pixel size measurement under a high magnification HM in an embodiment of the present invention;

[0054] Figure 13 It is a schematic diagram of obtaining different adjacent 2 images by using the bias of an electron optical system during actual pixel size measurement under a high magnification HM in an embodiment of the present invention;

[0055] Figure 14 It is a flowchart of a method for monitoring the actual pixel size on a charged particle beam scanning imaging device in an embodiment of the present invention;

[0056] Figure 15 It is a flowchart of a method for measuring the actual pixel size on a charged particle beam scanning imaging device in an embodiment of the present invention. Detailed implementation manners

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0058] First, an explanation related to background knowledge is given.

[0059] During the operation of the charged particle beam scanning imaging device on the IC production line, most of the work involves patterned wafers. The measurement method and monitoring method in the embodiments of the present invention are also directed to patterned wafers.

[0060] Taking the charged particle beam scanning imaging device as an EBR device as an example, like most semiconductor devices, the operation of the EBR device includes two main steps. First, the user formulates a work menu (Recipe) in advance, and the work menu includes the steps of the main tasks of the device; then during normal operation, for the same type of wafers, the device automatically repeats the execution of the work menu.

[0061] In addition, it is assumed that the above device is in a normal state during the work of determining the actual pixel size.

[0062] In actual use, the above-mentioned device usually has a series of pixels to be tested each time, for example, 6 to 7 pixels in total, with theoretical pixel sizes ranging from 500nm to 2nm. The usual practice is to measure from large theoretical pixel size / large field of view (FOV) / small magnification to small theoretical pixel size / small field of view / large magnification.

[0063] In this embodiment, the range of magnification is determined based on the empirical value of the theoretical pixel size. For example, 500nm to 50nm can be regarded as a low magnification LM, and the magnification corresponding to 50nm down to 2nm can be regarded as a high magnification HM. Of course, it can also be appropriately adjusted according to the image conditions in practice.

[0064] In the embodiments of the present invention, the known information contained in wafer alignment (WA) will be used as much as possible (which will be mentioned later), because WA is extremely important to the main work of the equipment and is a prerequisite for its normal operation. Therefore, it is necessary to add some WA background here. Usually, semiconductor equipment, in order to ensure the accuracy of its main work, usually has 3 levels of WA (2 or 4 levels in a few cases). After the wafer is mounted, the primary WA usually uses the optical image of the optical microscope imaging system (Optical Microscope, OM), and its field of view (Field of View, FOV) is large, while the subsequent WA uses SEM images, and the magnification gradually increases (FOV gradually decreases). When creating a wafer alignment recipe, from low to high magnification, usually start with 1 or more levels of OM images for primary WA, and after success, transition to SEM images for 1 or more levels (usually 2 levels, low magnification LM and high magnification HM) WA, so that the determination of wafer orientation and position meets the required accuracy requirements. The template matching (PM) method is the same as that in the above background, and NCC or feature-based template matching is also used, and the result is a similarity between [0, 1]. The template selection method is also the same as 511, 521. At each level of WA, refer to Figure 6A For example, images are collected along the same row (or column) of multiple die positions for template matching, and matching points 611, 612, 613, 614, 615, and 616 are obtained. The points that are successfully matched (similarity reaches a predetermined threshold, usually above 0.65) are fitted to a straight line to obtain the wafer orientation angle θ, which is then corrected (for example, by rotating the mechanical motion platform) and then transferred to a higher level of wafer alignment until all levels of WA are completed. When creating a WA Recipe, the above successful matching position is maintained, and when the Recipe is executed, the WA Recipe is automatically executed after the wafer is loaded. Reference Figure 6B, some WAs use 4 - 5 points in the X and Y directions, namely the matching points 620, and each level of WA (including those using OM images and SEM images) uses this method until the last level of WA is completed.

[0065] In the embodiments of the present invention, two methods for distinguishing the full measurement (FM) and quick measurement (QM) of the actual pixel size are defined. Among them, the quick measurement QM is also known as partial measurement QM, and generally starts from a small magnification and increases gradually. The full measurement FM can provide high - precision measurement results and is an offline measurement, while QM is fast and can be used for inline measurement without interrupting the normal operation of the device.

[0066] As Figure 15 shown, the embodiments of the present invention provide a method for measuring the actual pixel size of a charged - particle - beam scanning imaging device, including: using the charged - particle - beam scanning imaging device to perform an offline full measurement FM on a patterned wafer to obtain the actual pixel size of each pixel to be measured; performing a partial measurement QM based on the full measurement FM method for the device to perform an inline measurement of the actual pixel size of some pixels to be measured.

[0067] Among them, after first performing the full measurement FM to obtain the actual pixel size of each pixel to be measured, since the actual pixel size may change over time during the use of the device, it is necessary to measure the actual pixel size of the pixels to be measured again. In order not to delay the normal operation of the device, the device is used to perform an inline measurement of the actual pixel size of some pixels to be measured during the execution of its normal work to assist in judging the drift condition of the actual pixel size on the device, which can reduce the frequency of the full measurement FM that requires offline operation, so time can be saved; on the other hand, a monitoring method is provided by comprehensively considering the measurement results of the full measurement FM and the partial measurement QM later.

[0068] The following elaborates on the full measurement FM method for the actual pixel size in this embodiment.

[0069] In this embodiment, the full measurement FM method includes:

[0070] Collect a template image on the wafer, extract one or more templates from the template image, and save them;

[0071] Move the wafer in a preset direction and at a preset distance and collect a target image. Perform template matching in the target image according to the template to obtain matching positions greater than or equal to the similarity threshold, obtain the target displacements between each template and the corresponding matching positions, and obtain the weighted target displacements with the similarity ratio corresponding to each matching position as the normalization weight;

[0072] Use the target image as a template image to perform the next extraction of the template and template matching to obtain another weighted target displacement. After completion, continue in the same manner until the preset stop condition is met, then stop image acquisition and template extraction, and proceed to the next step;

[0073] Obtain the actual pixel size based on the accumulated wafer movement distance and the (accumulated) weighted target displacement.

[0074] The basic idea is also to first select a template image for template matching on the wafer, extract the template from it, then move the wafer to acquire the target image, perform template matching with the template in it, and combine the displacement of the mechanical motion platform to finally obtain the measured values Px and Py of the actual pixel size at a given theoretical magnification. This seems somewhat similar to the steps in the background technology. The difference is that both the wafer movement and the template matching method are more superior, so that the measurement error of the result is lower and the accuracy is higher. As mentioned above, to comprehensively measure FM, it is necessary to cover all the actual pixels to be measured. For example, there are 7 different magnifications involved in the measurement of the actual pixel size. Refer to Figure 7 , for the template image 710 at the theoretical magnification corresponding to a certain pixel to be measured, there is a template area position 711 in it, and its brightness, contrast, and uniqueness in the image / FOV all meet the template conditions. Similar areas that can be used as templates also include areas 712, 713, 714 and more areas (not marked in the figure). When the theoretical magnification corresponding to a certain pixel to be measured is slightly higher, there is a template image 720, and areas such as 722, 723, 724 in it are suitable as templates required for comprehensive measurement of FM. Similarly, when the theoretical magnification corresponding to the pixel to be measured is even higher, there is a template image 730, and area 732 in it can be used as a template for comprehensive measurement of FM. When the theoretical magnification corresponding to the pixel to be measured is even higher, there is a template image 740, and area 742 in it can be used as a template for comprehensive measurement of FM.

[0075] The template matching methods involved above include similarity such as NCC (cross-correlation method) or feature-based template matching. The features include corner points such as those extracted by Harris and FAST algorithms and post-processed, such as non-maximum suppression. Its matching uses its corresponding feature vector / descriptor, and the feature-dense area can be converted into an image-like template, and the matching uses similarity. In addition, the template matching in current commercial / industrial image processing / machine vision software, including feature-point-based template matching, is very mature and can all do the above work. In addition, users can also customize features and corresponding feature vectors, and the matching uses their dot product. These all belong to the category of existing technologies and will not be elaborated.

[0076] In this embodiment, the wafer is moved according to a preset distance. However, affected by the error of the mechanical motion platform, there may be a difference between the actual motion distance of the wafer, i.e., the wafer motion distance, and the preset distance. To reduce the impact of this difference on the actual pixel size during the comprehensive measurement of FM, the actual pixel size is obtained based on the accumulated wafer motion distance and the weighted target displacement.

[0077] Among them, the preset stop condition can be reaching the preset maximum number of matching times. For example, the maximum number of matching times is 19 times; it can also be that the cumulative sum of the wafer motion distance reaches a preset value. For example, the total stroke of the mechanical motion platform reaches the preset maximum value; it can also be that a template satisfying the preset extraction condition cannot be obtained in the current image. For example, the preset extraction condition can be that the brightness, contrast, and uniqueness all meet the corresponding threshold conditions; it can also be that the number of sub-templates satisfying the threshold condition cannot be obtained in the current image. For example, the number of sub-templates is insufficient or the number of unextracted sub-templates is zero. Of course, the preset stop condition is not limited to this. For example, the preset stop condition can also be a combination of at least two of the above conditions. As long as one of them is satisfied, it is considered to meet the preset stop condition.

[0078] In this embodiment, the specific steps for comprehensively measuring FM are described below. Refer to Figure 8, the electro - optical system 140 in the device (in this embodiment, including the SEM system) first acquires the first starting template image 810. Usually, there is at least one (usually multiple) image - type template like the previous region 722 (in addition, the feature - type template mentioned above is not excluded. Of course, in the embodiments of the present invention, the image - type template is preferentially selected). A certain number of regions are selected as templates. For example, there are templates a, b, c, d, e. The wafer / mechanical motion platform is moved, and the next frame of image, that is, the target image 811, is acquired. Templates a, b, c, d, e are used to search in it to obtain the matching positions a’, b’, c’, d’, e’ that meet the established threshold conditions. At the same time, new templates can be detected in this target image 811 (for image - type templates, the conditions are still that when the brightness, contrast, features, and uniqueness all meet the requirements, they can be automatically extracted by a certain algorithm). For example, region f’, all are used as templates. After moving the wafer in a predetermined direction to acquire the new target image 812, template matching is performed in it to obtain the matching positions b”, d”, e”, f” that meet the established threshold conditions. At the same time, new templates such as g”, h” can be detected in this image, which are used for template matching in the new target image acquired after moving the wafer to obtain the next - level target displacement. Of course, in actual work, it is often impossible to keep going like this for too long because it is not certain that new templates can be found to maintain the template matching. However, as long as the single - trip / step size is reasonably planned, and because multiple templates are adopted, usually 5 - 6 such moves + matches can be guaranteed. This method can be called the “multi - template weighted chain matching” method, rather than a single - move like in the prior art. The actual pixel size finally obtained is:

[0079]

[0080]

[0081] Among them,

[0082]

[0083]

[0084] and are respectively the weighted averages of the target displacements (in the X and Y directions) corresponding to all template matches (a total of J) that meet the established threshold conditions in the i - th (between the i - th wafer position / image and the i + 1 - th wafer position / image) template (from the i - th frame of image) matching. (dXmj, dYmj) is the target displacement determined by the j - th template match among the J successful matches in the i - th move; thus (DXmi, DYmi) is the displacement of the i - th single - move of the wafer, which is used for the calculation of the actual pixel size. Here, i = 1, … n, n is the number of wafer moves, n + 1 is the number of image acquisitions on the wafer, and the weight Wj comes from the template - matching score, and the weight Wj is:

[0085]

[0086] Where Sj is the similarity of a single template match that meets the established template matching threshold condition (e.g., 0.60), and its value is between [0, 1]. The target displacements obtained in the X and Y directions for a single match are:

[0087] dXmj = Xmj′ - Xmj

[0088] dYmj = Ymj′ - Ymj

[0089] Where (Xmj, Ymj) is the position of the j-th template among the J templates in the template image, and (Xmj’, Ymj’) is its position in the target image acquired when the wafer moves to the next position.

[0090] In addition, since the movement of the mechanical motion platform only depends on the starting and ending positions, then (the numerator part of the above formula) there is:

[0091]

[0092]

[0093] Where n’ = n + 1. Therefore, the finally obtained actual pixel size can be written as:

[0094]

[0095]

[0096] Due to errors in the actual situation, the formula considering errors is as follows:

[0097]

[0098]

[0099] Among them, the newly added parts are that the maximum error of the mechanical motion platform position degree is δS (usually about 0.5μm), and the maximum error of a single image template match is δM (usually between 0.025 pixels and 0.25 pixels, usually determined by various factors). Further, the Px and Py error parts can be separated, so there is:

[0100]

[0101]

[0102] Since the template matching error δM in the denominator is three orders of magnitude smaller than the single target displacement (Dxmi, DYmi), nδM can be ignored. With δS fixed, examining the error of the above formula / the second part, that is:

[0103]

[0104]

[0105] It can be seen that in the embodiments of the present invention, the relative error of the actual pixel size obtained by multi-template weighted chain matching is greatly reduced compared with the single single-template matching result in the prior art. In addition, using multiple templates can reduce the total number of wafer movements while still having higher statistical accuracy. Thus:

[0106]

[0107]

[0108] In addition, in rare cases when the magnification is extremely high, for example, when the theoretical pixel size is around 2 nm, the features in the SEM image become sparse. At this time, image preprocessing can be performed, including edge extraction. For example, the gradient ΔG or Laplacian ▽G of the Gaussian function G is convolved with the image I, that is, ΔG⊙I, where ⊙ represents the convolution operation. The size of the scale 2σ of the Gaussian function changes (in the opposite direction) with the change of the FOV size and is determined by experience. Then, the traditional thinning method can be used to obtain single-connected curve segments, and then the characteristic regions that do not include the endpoints in the curve, such as the corner parts, are used as templates, and weighted chain multi-template (weighted) matching is performed in the above manner to obtain the actual pixel size. For example, referring to Figure 9 , a high-magnification HM image with a high magnification HM is provided. It is an image after edge extraction, in which there are multiple templates 902, 903 with features (2-direction edges / corners), which can be used as templates for matching. Since the irregularity of the image in the high magnification HM will more or less affect the matching to some extent, the matching results of individual templates can also use the distance between their connecting lines (for example Figure 9 908 in) and their relative orientations to assist in confirming the correctness of the matching results. In addition, since the maximum error range of each movement of the mechanical motion platform is limited, the search range of the above template matching is limited, which helps to determine the correctness of the matching. At this time, the requirements (similarity threshold) of the template matching can be relaxed (reduced). The image processing technologies involved are common in the industry and will not be elaborated here.

[0109] Among them, when extracting a template from the template image, the weighted chain matching method can also be used, and the weight at this time is 1. Therefore, the comprehensive measurement FM method provided in this embodiment uses the weighted chain matching method, which includes two cases: the single-template weighted chain matching method and the multi-template weighted chain matching method.

[0110] It can be seen that the comprehensive measurement FM method in the embodiment of the present invention covers all pixels to be measured. Due to the use of the weighted chain matching method (for example, the multi-template weighted chain matching method), the relative error is lower and the accuracy is higher than that in the prior art, and its accuracy and reliability are greatly improved compared with the methods in the prior art.

[0111] The following describes the partial measurement QM method in the embodiment of the present invention.

[0112] According to the concept of the embodiment of the present invention, the partial measurement QM is a fast actual pixel measurement method with no or almost no additional time consumption. The purpose is to enable the device to quickly measure / evaluate the actual pixel size of the system while performing its own work. It can be carried out online and can be used to further determine the reasonable timing of the comprehensive measurement FM.

[0113] In this embodiment, the partial measurement QM method includes:

[0114] For any one of the partially measured pixels, collect a template image on the wafer, and extract at least a pair of templates that meet the preset template requirements from the template image; obtain the template distance between the two templates in each pair of templates and save it;

[0115] Collect a target image on the wafer, and perform template matching with the template to obtain a matching position that meets the preset target requirements. Obtain the target distance between the two matching positions in each pair of matching positions, compare the template distance and the corresponding target distance, and combine the actual pixel size obtained during the comprehensive measurement FM to obtain the actual pixel size.

[0116] In this embodiment, the measurement principle in QM is described below. As reviewed in the background art above, the relationship between the pixel sizes Δx, Δy and the field-of-view sizes FOVx, FOVy in the X and Y directions is FOVx = NxΔx, FOVy = NyΔy, where Nx and Ny are the pixel sizes in the X and Y directions respectively. Usually, Nx = Ny = N, that is, the width and height of the SEM image are the same, FOVx = NΔx, FOVy = NΔy, and N is the number of pixels in the image in the X / Y direction. Therefore, when Nx and Ny remain unchanged, the actual field-of-view size is proportional to the actual pixel size. Therefore, as long as the distance change between some fixed targets / features in the image can be measured, the change in the actual pixel size can also be estimated. Therefore, after knowing the correct (within the design specifications) pixel size, an area with features (corner points or edges in both the X and Y directions) in the image is selected, such as Figure 10A in the template image 1000 in, there are image regions 1001, 1002, 1003, 1004 as templates. Then the template information is saved, and subsequently, the changes in these distances are examined to obtain information on the change in the actual pixel size. Figure 10A There are several independent feature points 1006, 1007, 1008 as above, which can also be used as templates in the absence of an image-type template. There are many choices for the selection of the independent feature points, which are the same as those in the comprehensive measurement of FM above.

[0117] In this embodiment, continue to refer to Figure 10A , where part of the measurement QM template regions 1002, 1003, 1004 can be defined as image-type templates, and the feature points 1006, 1007, 1008 are the above-mentioned feature-type templates. In Figure 10AIn this case, the sub-images around it can be used to form an image type template. In this embodiment, the template requirements include template conditions (including brightness, contrast, and uniqueness requirements) and distance conditions. Templates that meet the template conditions (the template alone meets the requirements of features, contrast, uniqueness, etc.) and the distance conditions (the distance between 1-1, that is, the template distance reaches or exceeds the established template distance threshold) can be numbered and saved for each pair of templates (each template can only be used once), for example, saved to Recipe. The template distance threshold includes the minimum distance between the template pairs (for example, the distance between their centers needs to be greater than 1 / 5 of the minimum of the image width and height) and there is also a certain distance from their respective regions to the image edge (for example, 10 pixels). In specific implementation, usually when the image type templates that meet the established conditions are insufficient, feature point type templates are supplemented. Otherwise, image type templates are usually preferred. In addition, the dense area of the feature type template can be converted into an image type template, using the area in the image that includes these feature type templates (if they meet the template conditions). Supplementary note, assume that this is exactly the SEM image 1000 during wafer alignment WA. As the template image during WA, there is an area 1001 which is the wafer alignment WA template. The magnification of this SEM image 1000 is exactly the same as the theoretical magnification of the pixels to be measured in part of the measurement QM. Part of the measurement QM can directly use the template image of this wafer alignment WA. However, in general applications, the theoretical magnification of the image used in part of the measurement QM is usually the same as the theoretical magnification of a certain pixel to be measured in the full measurement FM and different from that in the wafer alignment WA. Of course, it is not excluded that in some cases, the theoretical magnification of 1 or more actual pixels to be measured in the full measurement FM is the same as that in the wafer alignment WA.

[0118] In this embodiment, the template requirements include a template distance threshold regarding the template distance, and the target requirements include a target distance threshold regarding the target distance. In this embodiment, the target distance threshold is the corresponding template distance threshold plus a tolerance (for example, it can be a fixed amount such as 3 pixels, or a relative amount such as 5% of the template distance).

[0119] In this embodiment, the method for selecting a template image and template matching in partial measurement QM is described below. Information of each template in the partial measurement QM template set includes some in the form of image types, some as feature points (described by their feature vectors), and the 1-1 distance between each template and the orientation angle of the connection line are also important information and all need to be saved. The selection of the partial measurement QM template can be carried out by independently collecting template images. For example, specifically searching for an area on the wafer with rich template pairs (which can be called an image acquisition area), or referring to the template image information in wafer alignment WA (which needs to be carried out after establishing the Recipe of wafer alignment WA). For example, the position of the template in the template image during a certain level of wafer alignment WA or its surrounding position can be directly selected. For example, i) when and only when all the pixels to be measured contain the same magnification as the SEM image in wafer alignment WA, that is, when the magnification of the SEM image used for wafer alignment is the same as the theoretical magnification of the pixels to be measured, use the template image during wafer alignment WA as the template image, which usually contains multiple image-type templates and independent feature point templates; ii) obtain the template image acquisition position on the wafer during wafer alignment WA and collect the template image, that is, utilize the empirical alignment WA template image acquisition position (because usually the features are relatively rich and the templates are easier to find); iii) based on the template image acquisition position on the wafer during wafer alignment WA, use the device to bias the charged particle beam and collect the template image at the biased position, that is, utilize the template image acquisition position during wafer alignment WA, without moving the mechanical motion platform but using the electron optical system 140 to bias and obtain nearby images, which usually contains multiple image-type templates and independent feature point templates. The biasing is an existing technology in the current electron optical system, and the advantage is that there is no mechanical motion time consumption. In practical applications, the selection of the method depends on actual needs. In addition, the same method needs to be used both when creating the Recipe and when executing the Recipe. For example, if biasing is used when creating the Recipe, biasing should also be used when executing the Recipe; iv) collect the template image in the image acquisition area of a die on the wafer with a template pair greater than or equal to the preset number; this is one of the options, but in the embodiments of the present invention, the previous methods are preferred, that is, as much as possible utilize the information of the wafer alignment template image itself or its image acquisition area, because wafer alignment is the most important preparation work of the device, with high requirements for templates and similar demands.

[0120] In this embodiment, the target image with successful matching is selected according to any of the following methods:

[0121] Obtain the target image corresponding to the template image acquisition method at the position with the closest successful matching to the wafer center in wafer alignment WA;

[0122] Obtain a target image corresponding to the template image acquisition method at the position of the successfully matched position with the highest matching similarity in wafer alignment WA;

[0123] Acquire a target image in the image acquisition area of the same die on the wafer or in the image acquisition area at the same position (having the same position as the image acquisition area of the same die) of other dies.

[0124] Supplementarily explain the bias. For the precisely calibrated electron optical system 140, the bias can basically help to expand the travel to n times that without bias, for example, n > 10, so that the selection range at the periphery is very large. The bias of many electron optical systems 140 can reach the micron level, which is sufficient for the SEM images involved in the embodiments of the present invention.

[0125] In this embodiment, the calculation steps in partial measurement QM are described by way of examples below. Continue to refer to Figure 10A , where the distance between templates 1002 and 1003 is called the template distance d23, and its distances in the X and Y directions are dX23 and dY23, which are saved after determination. During the subsequent partial measurement QM measurement, a target image is acquired at a certain acquisition position above (for example Figure 6A 611 or 612 in

[0126]

[0127]

[0128] , and a pair of templates composed of the saved templates 1002 and 1003 are used for matching therein. The distances dX23’ and dY23’ between the corresponding matching targets are obtained for quickly estimating the change in the actual pixel size,

[0129] where Px and Py are the previously saved correct (i.e., obtained from the most recent full measurement FM) actual pixel sizes, and Px’ and Py’ are the actual pixel sizes currently estimated by the partial measurement QM method. It can be seen that although this method is relative, it is very simple and suitable for monitoring the actual pixel size in the following text.

[0129] When there are multiple templates in the template image and there are n pairs of templates that are non-repetitive and the spacing satisfies the established conditions, the pair with the farthest distance in the image can be selected, for example, the i-th and j-th ones. In a more general form, the distances dXij and dYij in the template image, and then the distances dXij’ and dYij’ in the target image, there are:

[0130]

[0131]

[0132] It is also possible to select m pairs of template matches in the image where the distance reaches a predetermined threshold and the similarity score reaches a predetermined threshold when matching in the target image (each template can only be used once), calculate the average of these distances, and replace dXij, dYij, dXij’, dYij’ in the above formula to make the result more reliable. Additionally, it can be further improved. It is possible to select m pairs of template matches in the image where the distances in both the X and Y directions reach a predetermined threshold and the similarity score reaches a predetermined threshold when matching in the target image, calculate the weighted average of these distances, and replace dXij, dYij, dXij’, dYij’ in the above formula. The normalized weights can be determined by the distances between each pair of matches and / or the minimum / average value of the similarity of each pair of template matches. And using the averaging method will make the accuracy of the result relatively higher. Thus, there can be a basic general formula for a partial measurement QM method:

[0133]

[0134] Among them,

[0135]

[0136]

[0137] where dXij and dYij are the template distances in the X and Y directions of a pair of templates formed by the i-th and j-th templates in the template image respectively, 1 ≤ i ≤ M, 1 ≤ j ≤ M, and M is the total number of templates meeting the template requirements. The target distances in the X and Y directions of the corresponding successful matches in the template image are dXij’ and dYij’ respectively. The summation traverses all (at least 1 pair of templates) successful matches that meet the distance threshold / requirements between templates without repetition. ωij is the set weight. For example, the weight of this pair of matches in all successful matches can be determined by the distances between each pair of matches and / or the minimum / average value of the similarity of each pair of template matches as described above, or it can be a constant, that is, there is no weight difference. Px and Py are still correct, for example, the actual pixel sizes after the last full measurement FM is completed. Px’ and Py’ are the actual pixel sizes currently estimated by the partial measurement QM method.

[0138] Additional supplementary explanation. When Px' and Py' change relative to the original Px and Py, the template image itself will change to the same extent as the overall image, but this does not affect the overall result. Because for the normal device system, the change in Px' and Py' relative to the original Px and Py is still very small. Usually, there is no need to perform a scaled search during template matching (a scaled search slightly increases the calculation time, but generally the algorithm does not increase the running time by more than 10 ms on the current ordinary CPU), so it will not affect the template matching itself. In addition, the said template matching includes not only the above-mentioned image-based template matching such as using the NCC algorithm, but also feature-based template matching, such as template matching using feature points. Many such algorithms are basically not affected by image scaling, which holds true for both low magnification LM and very high magnification HM SEM images.

[0139] When the theoretical magnification of the actual pixels to be measured is high, some methods of measuring QM can be similar to those at low magnification LM, and the method of relative change in template distance is also used. As Figure 10B shown in template image 1010, where templates 1012, 1013, and 1014 can also be selected. The distances between them can also be used to estimate the actual pixel size at this magnification, but the measurement accuracy is slightly worse than that at low magnification LM. Because on the one hand, due to the smaller FOV at high magnification HM, the distance between templates becomes smaller and the local irregular changes in the image increase. On the other hand, because of the changes in the image (the template image and the target image may come from the same type but different wafers, or different positions on the wafer), and the SEM imaging mechanism determines that the SEM image at high magnification HM is always worse than that at low magnification LM. However, the corresponding error should be within 1 pixel, and the distance between templates is at the 10 2 pixel level, so the relative error is still very small, still at the 10 -2 magnitude.

[0140] In rare cases, when the magnification is very high, for example, as shown in the previous Figure 9 , after edge extraction of the image, there are usually still very local curve segments that meet the template matching conditions and can give more than 2 required templates. The distance between the 2 templates 1-1 that meet the template requirements and the template distance threshold / requirements can still be measured using the above method.

[0141] Furthermore, referring to Figure 11 , when the theoretical magnification of the pixels to be measured is extremely high (rare), there is sometimes only one template 1102 in the template image 1101 that meets the template matching conditions, and the above method based on the distance between templates for measuring the actual pixel size cannot be used. Another example is the case where only one usable template 1104 can be generated in the wafer alignment WA template image 1103.

[0142] In this embodiment, when there is only a single template in the template image that meets the preset template threshold, the device is used to bias the charged particle beam to obtain a biased image. A template is extracted from the biased image and at least one pair of templates is formed with the single template in the template image, and the template distance is obtained. A target image is collected at a predetermined position on the wafer, and a biased target image is collected using the same bias as the above-mentioned bias. Template matching is performed on the templates in the unbiased and biased template images respectively to the corresponding unbiased and biased target images to obtain at least one pair of matching positions, and the target distance between the two matching positions in each pair of the matching positions is obtained.

[0143] Among them, the method given in the embodiment of the present invention is to use the biasing function of the electron optical system 140 mentioned above, so that the electron optical system 140 can scan the area within a neighborhood of a predetermined distance from its system center, and finally obtain two templates located in the 2-image, and the distance between them can be used for the calculation in the partial measurement QM. At this time, an appropriate bias voltage needs to be set for the electron optical system 140 without moving the wafer. The bias here has the same working principle as the previous bias but different purposes. The previous bias is to obtain an image as a template image near the WA template image acquisition position, while the bias here is for stitching measurement images. At this time, a bias is set for the electron optical system 140 to obtain images with the same magnification in the adjacent area. It should be noted that there is no need to actually stitch the two images (a stitching method based on the features of the overlapping part in the 2-image). The distance between the two images can be completely obtained through the calibration of the electron optical system 140. For example, the calibration can generate a look-up table (LUT) of bias and distance, and the bias distance can be converted to pixels. The corresponding error should not exceed 1 pixel under the precise calibration of the system. Compared with the distance between a pair of templates that make up the partial measurement QM template, which is still the template distance, it is 2-3 orders of magnitude smaller and can be ignored. The calibration itself is outside the embodiment of the present invention and is prior art, so it will not be elaborated here.

[0144] The following description is also another method for obtaining and matching template images used in partial measurement QM at high magnification HM, which is applicable to the case where the features in the small field of view image at high magnification HM are relatively scarce. Refer to Figure 12, there is a template image 1200 at the original high magnification HM, in which there is a template 1202, and in which there is a feature region 1206. At this time, a bias is set for the electro-optical system 140 to obtain an image 1230 of the same magnification in the adjacent region. In addition, the template feature region 1202 can generate an image-like template 1203, but in the case of such a high magnification, due to interference effects such as feature content, local deformation, and noise, the effect may not be very good. However, other types of templates can also be generated. For example, a certain number of points 1204 on two adjacent sides of the corner are used to fit two straight lines, and a reference point is determined at the intersection 1205 or at a certain distance along its angular bisector, and it is defined as the position of a feature template. And this template can still use some of the aforementioned feature-like templates, such as features (a certain descriptor) extracted by the FAST algorithm. Similarly, the template position is also determined in the same way in the biased image 1230. For example, an image-like template 1232 and the intersection 1235 obtained by sampling and fitting straight lines on two sides of the corner are used as feature points. The template matching of the feature class usually uses algorithms related to vector dot product / descriptor matching, and the result is also between [0, 1], similar to the result of NCC. In this way, the distance between the template 1202 (in the original image) and the template 1233 (in the biased image) can be obtained by using the above method for obtaining the pixel size based on the distance between templates. Thus, there is:

[0145]

[0146]

[0147] Wherein,

[0148] DXod = Xd - Xo + XDef

[0149] DYod = Yd - Yo + YDef

[0150] DXod‘ = Xd’ - Xo‘ + XDef

[0151] DYod’ = Yd‘ - Yo’ + YDef

[0152] Where (Xo, Yo) is the position of template 1202 in the (original / unbiased) template image 1200, (Xd, Yd) is the position of template 1232 in its adjacent biased image 1230, XDef, YDef are the distances in the X and Y directions (in pixels) corresponding to the offsets of the two frames of images, (Xo, Yo) and (Xd, Yd) are both previously retained correct / within the system indicator range, and (Xo', Yo') and (Xd', Yd') are on the target image matched to the template in QM, one is the template matching position on the original / unbiased image, and the other is the template matching position on the biased image. It should be noted that although XDef and YDef contain errors caused by the offset, the maximum value is δD. However, as mentioned above, the δD of the accurately calibrated system is less than 1 pixel, which is 2 to 3 orders of magnitude worse than the template distance, so it can be ignored.

[0153] It should be noted that in the embodiment of the present invention, there is no restriction on the direction and position of the offset image relative to the original image, and the number of offset images, as long as the electronic optical system 140 can support it (within a certain error range), for example Figure 13 In the figure, the original image 1301 when the wafer is aligned with WA may have an offset image 1302 and / or 1303, and there are many other possibilities which will not be elaborated here.

[0154] In this embodiment, a method for monitoring the actual pixel size of a charged particle beam scanning imaging device is also provided, comprising:

[0155] 1411. After the electronic optical system in the device undergoes maintenance and calibration, determine monitoring working parameters, including a second parameter Nf, a second threshold Thfi, a first parameter Nq and a first threshold Thqi, wherein the second parameter Nf is greater than the first parameter Nq, and the second threshold Thfi is greater than the first threshold Thqi;

[0156] 1412. Execute the comprehensive measurement FM and clear the second counter nf and the first counter nq at the same time;

[0157] 1413. Compare the actual pixel size measured in the comprehensive measurement FM with the second threshold Thfi. If all the actual pixel sizes measured this time are outside the second threshold Thfi, return to and re-execute step 1411; otherwise, proceed to the next step.

[0158] 1414. Save all actual pixel sizes measured in the current comprehensive measurement FM;

[0159] 1415. When the device starts or continues to perform its main work, it counts through the second counter nf and the first counter nq;

[0160] 1416. When the first counter nq reaches the second parameter Nf, return to step 1412; otherwise, proceed to step 1417.

[0161] 1417. When the first counter nq does not reach the first parameter Nq, return to step 1415 and the device continues its normal operation; otherwise, proceed to step 1418.

[0162] 1418. Perform the partial measurement QM based on the above, and at the same time clear the first counter nq.

[0163] 1419. When the result of the actual pixel size is greater than or equal to the second threshold Thfi, return to step 1411 to perform device calibration and maintenance again; otherwise, when the monitored value is less than the first threshold Thqi, return to step 1415 and the device continues to perform its normal operation; otherwise, when the monitored value is less than the second threshold Thfi but greater than or equal to the first threshold Thqi, change the first parameter Nq and the second parameter Nf, and then return to step 1415.

[0164] The monitoring method of the actual pixel size in the embodiment of the present invention is described below.

[0165] First, based on the characteristics of the above full measurement FM and partial measurement QM, reasonably plan the monitoring. The full measurement FM must be performed offline, cover all pixels to be measured, and has high result accuracy, which is used for the normal operation of the device. The partial measurement QM can be performed online, can select a few theoretical magnification ratios, is fast, and is used to determine the timing of the full measurement FM. The partial measurement QM is implemented during the normal operation of the device, and the partial measurement QM is performed after the device completes wafer alignment. Specifically, the time when it occurs can be after the device completes wafer alignment WA at low magnification LM and / or high magnification HM, or after the device completes the entire wafer alignment WA and before starting the normal operation. For example, usually the wafer alignment WA includes a first-level low magnification LM and a first-level high magnification HM. If the magnification ratios of the pixels to be measured during the full measurement FM are the same as the magnification ratios of the low magnification LM and high magnification HM in the wafer alignment WA, the partial measurement QM also selects these two magnification ratios for measurement.

[0166] In the embodiments of the present invention, the frequencies for full measurement FM and partial measurement QM in monitoring are specifically planned. The number of wafers processed cumulatively by the equipment, N, or the cumulative working time of the equipment, T, can be used. The two have basically the same meaning. For convenience, the number of wafers processed cumulatively by the equipment, N, is used in the following text. There are multiple choices for the timing of counting. For example, the above-mentioned equipment (computer 150) can record once after each wafer is loaded. In the embodiments of the present invention, the number of wafer processing times Nq required for cumulative partial measurement QM is called the first parameter Nq, and the number of wafer processing times Nf required for cumulative full measurement FM is called the second parameter Nf, where Nf > Nq. Nf and Nq are both working parameters in monitoring, and the two can maintain a fixed proportional relationship. For example, there can be Nf = uNq, where the proportionality factor u is a positive integer. For example, the proportionality factor u = 10, Nq = 1000, so Nf = 10000. This is only an example. In the embodiments of the present invention, these values are usually determined first according to the equipment characteristics and user experience, but can be further optimized during use, and relevant methods are also given in the following text.

[0167] In addition, Px and Py have their initial values before monitoring starts. Usually, they are ideal values determined in advance according to equipment performance and experience. Subsequently, they will be replaced by new values after the full measurement FM is successful. Usually, the actual pixel change will be larger than the initial value. Therefore, the thresholds involved later are all one-way, only considering the case where the actual pixel size is larger than the ideal value.

[0168] Optionally, it also includes examining Pxy = max(Px / Py, Py / Px), because the system often has requirements for the ratio of the two. If the difference between the two is too large, it is also a problem if it is not deliberately designed, resulting in image distortion. Now the quantity to be measured can be uniformly written as Pi, where i = x, y, xy. Correspondingly, the second threshold Thfi and the first threshold Thqi are respectively set to include three thresholds to be compared with the actual pixel size Px in the x direction, the actual pixel size Py in the y direction, and the proportional relationship Pxy respectively. They have different thresholds for the full measurement FM and the partial measurement QM, including the second threshold Thfi, where i = x, y, xy and the first threshold Thqi, where i = x, y, xy, and Thqi < Thfi, where i = x, y, xy. Of course, they all exceed the initial value of Pi. The meaning is that when a certain Pi (i = x, y, x) exceeds the corresponding threshold Thfi, it means that the actual pixel size of the device deviates significantly from the index at this time, and the system must be maintained and calibrated as soon as possible; when all Pis are lower than Thqi, it means that the actual pixel size of the device is normal, and the next partial measurement QM and full measurement FM can be performed according to the established schedule; and when all Thqi ≤ Pi < Thfi (i = x, y, xy), that is, between the two thresholds, it means that although Pi is within the normal range, it deviates from its initial value to a certain extent, and the monitoring needs to be more frequent / closer. The above thresholds Thfi, Thqi, where i = x, y, xy mainly come from system indicators / limitations and are usually based on user experience. In addition, they are different for different charged particle beam scanning imaging systems and different applications. For example, the Thqi used in the EBR device is much higher than that used in the CD-SEM because the EBR device mainly cares about the presence or absence of defects and the classification of defects, and has relatively loose requirements for measurement dimensions.

[0169] The following describes the specific steps of the above monitoring. Please refer to Figure 14 . First, for Figure 14 Explain that it includes a judgment step, and the result is either "yes" or "no". But for convenience, Figure 14 only "yes" is marked in

[0170] In addition, it should also be noted that at the start of the monitoring process, for the device, the parameters Nq, Nf, Thfi, Thqi, where i = x, y, xy, and the values included in the partial measurement QM have all been determined. The parameters required for the full measurement FM and the partial measurement QM have also been determined.

[0171] In this embodiment, Figure 14 shows a closed loop process of the monitoring method in the embodiment of the present invention.

[0172] The process starts at step 1411, which is the normal maintenance calibration of the system. Because usually when a new device or its system / component is updated, proper maintenance calibration is required to ensure that the device can work properly. After the system is first run / maintenance calibration, the values ​​of the first parameter Nq, the second parameter Nf, the first threshold Thqi, the second threshold Thfi, i=x, y, xy, and of course the initial value of Pi are required. As mentioned above, they are determined when the recipe is created.

[0173] Then, step 1412 is performed to perform a comprehensive measurement of FM, the method is as described above, it covers all theoretical pixels / magnifications to be measured, and at the same time, the second counter nf for the comprehensive measurement of FM and the first counter nq for the partial measurement of QM are cleared. At this time, the method for comprehensive measurement of FM is as described above. The second counter nf and the first counter nq are both for counting wafers, for example, counting and recording once when each wafer is placed on the wafer.

[0174] Then, step 1413 is performed to check the actual pixel size at each level / theoretical magnification, using the above-mentioned comprehensive measurement FM method. If any of the actual pixel sizes obtained exceeds the second threshold value Thfi, that is, Pi≥Thfi, i=x, y, xy, a warning is issued, allowing the user to return to step 1411 for system calibration in time, and then the device starts working again.

[0175] Otherwise, go to the next step 1414, retain all the current Pi, i = x, y, for further study of the monitoring process when necessary. Then enter step 1415, the equipment performs its job normally (for example, the job of EBR is to recheck defects and classify defects). At this time, when the equipment fails, the latest Pi, i = x, y result after comprehensive measurement of FM may be used. In addition, the second counter nf and the first counter nq start counting from zero again, for example, as mentioned above, recording once when each wafer is loaded.

[0176] Then, step 1416 is performed to check the second counter nf. When it reaches the predetermined number threshold Nf, the process returns to step 1412 to perform a comprehensive measurement of FM. Otherwise, step 1417 is performed to check the first counter nq. When it reaches the predetermined number threshold Nq, the process proceeds to step 1418 to perform the partial measurement QM, using the method described above. At the same time, the first counter nq is also cleared. Note that the device is still performing its main work at this time, rather than stopping its main work as in the comprehensive measurement of FM.

[0177] Then, proceed to the next step 1419. Use the second threshold Thfi to check the result. If any one of the actual pixel size measurement results included in the partial measurement QM exceeds the second threshold Thfi, that is, Pi≥Thfi, where i = x, y, xy, it indicates that the problem is relatively serious at this time. Then, issue a warning to prompt the user to return to step 1411 in a timely manner for system maintenance and calibration, and then restart the monitoring process.

[0178] Otherwise, use the first threshold Thqi to check the result. If any one of the actual pixel size measurement results included in the partial measurement QM exceeds the first threshold Thqi, that is, Pi≥Thqi, where i = x, y, xy, it indicates that there is a slightly larger change or a tendency to deteriorate (usually the actual pixel size becomes larger) in the actual pixel size of the system at this time. Therefore, it is necessary to perform detection slightly more frequently, which can be achieved by optimizing the first parameter Nq in a timely manner. Then, reach step 1422. Optionally, optimize the first parameter Nq (in the embodiments of the present invention, it is inclined to the proportional relationship between the fixed second parameter Nf and the first parameter Nq. Changing the first parameter Nq also changes the second parameter Nf). After completion, return to step 1415. Otherwise, at this time Pi<Thqi, where i = x, y, xy, it indicates that there is no abnormality, and directly return to step 1415 to continue the loop. The system is still continuing to perform its original work, and the monitoring is also in progress, without changing anything. The second counter nf and the first counter nq continue to count. The method for optimizing the first parameter Nq (and thus the second parameter Nf because the two have a fixed relationship) in a timely manner, the adjustment amplitude can be proportional to the deviation from the previous / previous actual pixel size. For example, set the new first parameter Nq to S% times the previous first parameter Nq, that is:

[0179] Nq‘ = Nq×S%,

[0180] Nf‘ = uNq‘

[0181] Therefore, there is:

[0182] Nf’ = Nf×S%

[0183] Among them, Nf’ and Nf are the adjusted and current second parameters respectively, Nq’ and Nq are the adjusted and current first parameters respectively, λi is the set weight, S is a bounded adjustment factor, which can be determined according to the degree of deviation from the previous actual pixel size value, and the adjustment factor S is calculated by the average value of the deviation degree of the actual pixel size Px in the X direction, the deviation degree of the actual pixel size Py in the Y direction, and the deviation degree of the proportional relationship Pxy. For example:

[0184]

[0185] Among them, S is restricted, S ≤ St, where St is a preset threshold, for example, S ≤ 5. Where Pi is the value of the qualified actual pixel sizes Px and Py and the value of the proportional relationship Pxy saved after the current full measurement FM, and Pi' is the value of the actual pixel sizes Px and Py and the value of the proportional relationship Pxy of this partial measurement QM, i = x, y, xy. As mentioned above, usually the new result is always larger than the previous one, so in fact it doesn't matter whether the absolute value is taken for the numerator of the above formula. Then, similar to the above, since the second parameter Nf is related to the first parameter Nq, and the proportionality factor u between the two remains fixed, when the first parameter Nq is updated, the second parameter Nf is also updated. In addition, the above formula can also include a weight factor, so there can be

[0186]

[0187] where λi (i = x, y, xy) represents the weight, and the sum of all of them is 1. Each weight value is determined by the user's experience. For example, the weights of Px and Py, λx = λy = 0.4, and the weight of Pxy, λxy = 0.2, are somewhat related to the previous two and are relatively less important. S is still restricted, for example, S ≤ 5. There are many similar methods for updating the second parameter Nf and the first parameter Nq, which will not be elaborated here.

[0188] It can be seen from the above description of the monitoring process that the monitoring method is an automatically optimized closed-loop process. Its advantage is that the second parameter Nf and the first parameter Nq of the full measurement FM and the partial measurement QM have different frequencies. In this way, the accuracy of the full measurement FM and the speed of the partial measurement QM are combined, enabling the device to more accurately grasp the timing of actual pixel size measurement during operation, neither delaying the timing of actual pixel size measurement nor wasting precious machine time and resources, solving the problems in the prior art described in the background art, and greatly improving the performance, efficiency, and usage cost of the device.

[0189] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. Method for measuring actual pixel size of charged particle beam scanning imaging device ， Characterized in that it includes: Using a charged particle beam scanning imaging device to perform offline comprehensive measurement FM on a patterned wafer to obtain the actual pixel size of each pixel to be measured; Based on the comprehensive measurement FM method, perform partial measurement QM for the device to online measure the actual pixel size of some pixels to be measured; Wherein, the comprehensive measurement FM method includes: Collect a template image on the wafer, extract one or more templates in the template image, and save them; Move the wafer in a preset direction and at a preset distance and collect a target image, perform template matching in the target image according to the template to obtain matching positions greater than or equal to the similarity threshold, obtain the target displacement between each template and the corresponding matching position, and obtain a weighted target displacement with the similarity ratio corresponding to each matching position as the normalization weight; Take the target image as the template image, perform the next extraction of the template and template matching to obtain another weighted target displacement, and continue in the same way after completion until the preset stop condition is met, then stop collecting images and extracting templates, and enter the next step; Obtain the actual pixel size according to the accumulated wafer movement distance and the weighted target displacement; Wherein, the partial measurement QM method includes: For any one of the pixels to be measured in part, collect a template image on the wafer, and extract at least one pair of templates that meet the preset template requirements in the template image; Obtain the template distance between the two templates in each pair of templates and save it; collect a target image on the wafer, and perform template matching with the template to obtain matching positions that meet the preset target requirements, obtain the target distance between the two matching positions in each pair of matching positions, compare the template distance and the corresponding target distance, and combine the actual pixel size obtained during the comprehensive measurement FM corresponding to it to obtain the actual pixel size.

2. The measurement method according to claim 1, Characterized in that The template requirements include a template distance threshold regarding the template distance, and the target requirements include a target distance threshold regarding the target distance.

3. The measurement method according to claim 1, Characterized in that When there is only a single template that meets the preset template threshold in the template image, use the device to bias the charged particle beam to obtain a biased image, extract a template in the biased image and form at least one pair of templates with the single template in the template image and obtain the template distance; collect a target image at a predetermined position on the wafer, and collect a biased target image using the same bias as the bias, perform template matching with the templates in the unbiased and biased template images into the corresponding unbiased and biased target images respectively to obtain at least one pair of matching positions, and obtain the target distance between the two matching positions in each pair of the matching positions.

4. The measurement method according to claim 1, Characterized in that Select the template image according to any one of the following methods: When the magnification of the SEM image used for wafer alignment is the same as the theoretical magnification of the pixel to be measured, use the template image during wafer alignment as the template image; Obtain the template image acquisition position on the wafer during wafer alignment and acquire the template image; Based on the template image acquisition position on the wafer during wafer alignment, use the device to deflect the charged particle beam and acquire the template image at the deflected position; Acquire the template image in the image acquisition area of a die on the wafer where there are greater than or equal to a preset number of template pairs.

5. The measurement method according to claim 4, wherein, Select the successfully matched target image according to any one of the following methods: Obtain the target image corresponding to the template image acquisition method at the position of the successfully matched position closest to the wafer center during wafer alignment; Obtain the target image corresponding to the template image acquisition method at the position of the successfully matched position with the highest matching similarity during wafer alignment; Acquire the target image in the image acquisition area of the same die on the wafer or in the image acquisition area at the same position in other dies.

6. The measurement method according to claim 1, wherein, Select the template pairing combination and obtain the target distance according to any one of the following methods: Select a pair of templates that meet the preset template requirements and are the farthest apart in the template image for the template matching, and obtain the target distance after the matching; Select multiple pairs of templates that meet the preset template distance threshold in the template image for the template matching, and use the average value of multiple pairs of target distances that meet the preset target requirements after the matching as the target distance.

7. The measurement method according to claim 6, wherein, The basic general formula of the partial measurement QM method is as follows: ； where, ； Wherein, dXij and dYij are respectively the template distances in the X and Y directions of a pair of templates formed by the i-th and j-th templates in the template image, 1 ≤ i ≤ M, 1 ≤ j ≤ M, and M is the total number of templates meeting the template requirements. The target distances in the X and Y directions of the successfully matched corresponding parts in the template image are respectively and , ωij is the set weight, Px and Py are the actual pixel sizes after the last complete measurement of FM is completed, , is the actual pixel size currently estimated by the partial measurement QM method.

8. A method for monitoring the actual pixel size of a charged particle beam scanning imaging device, wherein, includes: 1411. After the electron optical system in the device undergoes maintenance and calibration, determine the monitoring working parameters, including the second parameter Nf, the second threshold Thfi, the first parameter Nq, and the first threshold Thqi, where the second parameter Nf is greater than the first parameter Nq, and the second threshold Thfi is greater than the first threshold Thqi; 1412. Perform the comprehensive measurement FM according to any one of claims 1-7, and at the same time clear the second counter nf and the first counter nq; 1413. Compare the measured actual pixel size in the comprehensive measurement FM with the second threshold Thfi. If all the measured actual pixel sizes this time are outside the second threshold Thfi, return to step 1411 to execute again, otherwise enter the next step; 1414. Save all the measured actual pixel sizes in the current comprehensive measurement FM; 1415. When the device starts or continues to perform its own work, count through the second counter nf and the first counter nq; 1416. When the first counter nq reaches the second parameter Nf, return to step 1412, otherwise enter step 1417; When the first counter nq does not reach the first parameter Nq, return to step 1415 and the device continues its normal operation; otherwise, proceed to step 1418; 1418. Perform the partial measurement QM according to any one of claims 1-7, and at the same time clear the first counter nq; 1419. When the result of the actual pixel size is greater than or equal to the second threshold Thfi, return to step 1411 to perform device calibration and maintenance again; otherwise, when the result of the actual pixel size is less than the first threshold Thqi, return to step 1415 and the device continues to perform its normal operation; otherwise, when the result of the actual pixel size is less than the second threshold Thfi but greater than or equal to the first threshold Thqi, change the first parameter Nq and the second parameter Nf, and then return to step 1415.

9. The monitoring method according to claim 8, characterized in that, the second parameter Nf and the first parameter Nq have a fixed proportional relationship Nf = uNq, where the proportionality factor u is a positive integer; the counting of the first counter nq and the second counter nf is performed after the device wafers are loaded; During the normal operation of the device, the partial measurement QM is implemented, and the partial measurement QM is performed after the device performs wafer alignment.

10. The monitoring method according to claim 8, characterized in that, the second threshold Thfi and the first threshold Thqi are respectively set to include three thresholds to be compared with the actual pixel size Px in the x direction, the actual pixel size Py in the y direction, and the proportionality formula Pxy, where Pxy = max(Px / Py, Py / Px).

11. The monitoring method according to claim 10, characterized in that, the first parameter Nq and the second parameter Nf are changed by the following formula: ； wherein, Nf and Nc are the adjusted and current second parameters respectively, Nq and Np are the adjusted and current first parameters respectively, the scaling factor u is a positive integer, λi is the set weight, and S is the bounded adjustment factor, ； Among them, Pi is the value of the qualified actual pixel sizes Px and Py saved after the current comprehensive measurement FM and the value of the proportional relationship formula Pxy. It is the value of the actual pixel sizes Px and Py of the current partial measurement QM and the value of the proportional relationship formula Pxy, where S ≤ St and St is a preset threshold.

Images