Method and device for detecting end point of wafer film grinding

Abstract

The present invention relates to the field of semiconductors, and specifically relates to a method and device for detecting an end point of wafer film grinding. The method comprises: during the process of grinding a wafer film, acquiring an actually measured spectral curve indicating a relationship between the reflectance and wavelength of the wafer film; determining a plurality of feature points in the actually measured spectral curve; on the basis of wavelength values of the plurality of feature points, respectively determining corresponding reference points in reference spectral curves; using the plurality of feature points and the plurality of reference points to calculate a cosine similarity between the actually measured spectral curve and the reference spectral curves; on the basis of the cosine similarity, determining a reference spectral curve that matches actually measured spectral data; and determining the current thickness of the wafer film on the basis of a thickness corresponding to the matching reference spectral curve. In the present invention, a cosine similarity is selected as a score function to evaluate similarity under a multi-layer medium condition, thereby improving the accuracy of online measurement of the thickness of a wafer film.

Description

End point detection method and equipment for wafer thin film grinding Technical Field

[0001] The present invention relates to the field of semiconductors, and in particular to an endpoint detection method and device for wafer thin film grinding. Background Art

[0002] With the rapid development of the semiconductor industry, the feature size of integrated circuits continues to become increasingly miniaturized. Semiconductor wafers are constantly developing towards smaller size, higher circuit density, higher speed, and lower power consumption. Integrated circuits have now entered the ULSI submicron technology stage. With the gradual increase in silicon wafer diameter, the width of component lines gradually decreases, and the number of metal layers increases, the high polishing of semiconductor film surfaces has a significant impact on the high performance, low cost, and high yield of devices. Therefore, the requirements for silicon wafer surface flatness will become increasingly stringent.

[0003] Chemical Mechanical Polishing (CMP) is a global surface planarization technique used in semiconductor manufacturing to minimize the effects of substrate thickness variation and surface topography. Because CMP can accurately and uniformly planarize substrates to the desired thickness and flatness, it has become the most widely used surface planarization technology in semiconductor manufacturing. The current mainstream technology is in-line endpoint detection, which can better control wafer film thickness variation, reduce repetitive operations, and enable automated CMP operations. This improves polishing equipment utilization and throughput, reduces density defects and non-uniformity in IC devices, and ultimately enhances the stability and reliability of semiconductor devices. The principles of in-line endpoint detection technology are primarily based on optical, electrical, acoustic, vibrational, thermal, frictional, chemical, or electrochemical detection. In the spectral endpoint detection monitoring process, reference spectra are first established at different film thicknesses within a defined wavelength range. Then, using a predefined matching method, the spectra measured in situ during the polishing process are matched against a library of reference spectra to find the best-matching reference spectrum. The film thickness corresponding to the best-matching reference spectrum is used as the film thickness at that point in the polishing process.

[0004] Due to the harsh measurement environment during the actual polishing process, in-situ online measurements are affected by various factors, including unavoidable systematic errors caused by the spectrum itself and random errors caused by measurement conditions such as ambient light and probe distance jitter. Especially during the polishing process of multilayer films, external factors during the actual polishing process can cause the measured reflectivity curve to be compressed, elevated, and exhibit parallel displacement differences. They can also lead to information loss in certain wavelength ranges. These conditions prevent the measured reflectivity curve from matching the theoretical curve well, resulting in a decrease in the accuracy of the final measured film thickness value.

[0005] Summary of the Invention

[0006] According to one aspect of an embodiment of the present application, a method for online thickness detection for wafer thin film grinding is provided, the method comprising:

[0007] During the wafer film grinding process, a measured spectral curve of the relationship between the reflectivity and wavelength of the wafer film is obtained; a plurality of characteristic points are determined in the measured spectral curve; based on the wavelength values ​​of the plurality of characteristic points, corresponding reference points are respectively determined in each reference spectral curve; the cosine similarity between the measured spectral curve and the respective reference spectral curves is calculated using the plurality of characteristic points and the plurality of reference points; a reference spectral curve matching the measured spectral data is determined based on the cosine similarity; and the current thickness of the wafer film is determined based on the thickness corresponding to the matched reference spectral curve.

[0008] Optionally, determining multiple feature points in the measured spectral curve includes: determining fluctuation abnormal points based on the reflectivity of each point in the measured spectral curve and its adjacent points; and selecting a preset number of points from each point in the measured spectral curve except the fluctuation abnormal points as the feature points.

[0009] Optionally, determining multiple characteristic points in the measured spectral curve includes: determining at least one critical point based on the change rate of each point in the measured spectral curve, so that the measured spectral curve is divided into at least two bands; selecting a corresponding preset number of points in each of the bands as the characteristic points, wherein the preset number corresponding to the band with a higher change rate is greater than the preset number corresponding to the band with a lower change rate.

[0010] Optionally, the reference spectrum curve is a theoretical spectrum curve of the relationship between theoretical reflectivity and wavelength calculated based on the refractive index of the wafer film and the pre-detected film thickness range.

[0011] Optionally, in the step of calculating the cosine similarity between the measured spectral curve and each reference spectral curve, the cosine similarity between the measured spectral curve and the reference spectral curve is calculated in the following manner:

[0012] Among them, cosθ represents cosine similarity, A i Indicates the reflectivity of the i-th characteristic point in the measured spectrum curve, B i represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

[0013] Optionally, calculating the cosine similarity between the measured spectral curve and each reference spectral curve includes: processing the multiple feature points and the multiple reference points; and calculating the cosine similarity using the processed multiple feature points and the multiple reference points so that the calculation result is in the interval [-1, 1].

[0014] Optionally, the multiple feature points and the multiple reference points are processed, including: calculating the average of the measured reflectivities of the multiple feature points and the average of the reference reflectivities of the multiple reference points; subtracting the reflectivity of each feature point from the average of the measured reflectivity, and subtracting the reflectivity of each reference point from the average of the reference reflectivity.

[0015] Optionally, the cosine similarity is calculated in the following manner:

[0016] Among them, cos′θ represents cosine similarity, represents the result of subtracting the reflectivity of the i-th feature point in the measured spectral curve from the mean of the measured reflectivity, A represents the result of subtracting the reflectivity of the i-th reference point in the reference spectrum curve from the mean value of the reference reflectivity. i Indicates the reflectivity of the i-th characteristic point in the measured spectrum curve, B i represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

[0017] Optionally, the wafer thin film is a multilayer film.

[0018] Optionally, the characteristic points are all data points in the measured spectrum curve.

[0019] Optionally, the number of the feature points is a preset number.

[0020] Optionally, the number of the characteristic points is determined in real time according to the characteristics of the measured spectrum curve.

[0021] Optionally, the multiple feature points are evenly distributed on the measured spectral curve; and determining the multiple feature points in the measured spectral curve includes: evenly selecting the multiple feature points according to the wavelength range of the measured spectral curve and a preset number of feature points.

[0022] Optionally, the plurality of characteristic points are located in a wavelength band of the measured spectrum curve that is relatively rich in characteristic information.

[0023] Optionally, the reference spectral curve that matches the measured spectral data refers to a reference spectral curve whose cosine similarity is closest to 1.

[0024] Optionally, the reference spectrum curve is collected in advance, and the parameters used when collecting the reference spectrum curve and when collecting the measured spectrum curve are the same, and the parameters include wavelength range, wavelength resolution, and reflectivity resolution.

[0025] Optionally, each of the reference spectral curves is a reflectivity-wavelength reference spectral curve established in advance based on the reflectivity of the wafer film and a pre-detected film thickness range, and includes a plurality of limited wavelength ranges under different wafer film thicknesses.

[0026] Optionally, before determining multiple feature points in the measured spectral curve, the method further includes: filtering the measured spectral curve to filter out noise caused by the grinding environment during the grinding process, and the filtering method is any one of wavelet filtering, Fourier filtering and sliding window average filtering.

[0027] According to another aspect of an embodiment of the present application, an endpoint detection method for wafer film grinding is also provided, including: using the above-mentioned online thickness detection method to monitor in real time whether the thickness of the wafer film reaches the target thickness; when the thickness of the wafer film reaches the target thickness, stopping grinding.

[0028] According to the third aspect of the embodiment of the present application, an online thickness detection device for wafer thin film grinding is also provided, which includes: a processor and a memory connected to the processor; wherein the memory stores instructions that can be executed by the processor, and the instructions are executed by the processor so that the processor executes the above-mentioned online thickness detection method for wafer thin film grinding.

[0029] According to the fourth aspect of the embodiment of the present application, an endpoint detection device for wafer thin film grinding is also provided, which includes: a processor and a memory connected to the processor; wherein the memory stores instructions that can be executed by the processor, and the instructions are executed by the processor so that the processor executes the above-mentioned endpoint detection method for wafer thin film grinding.

[0030] According to the online thickness detection method and apparatus for wafer thin film grinding provided by the present invention, the current thickness of the wafer thin film is determined by calculating the cosine similarity between the measured spectral curve and the reference spectral curve. This can effectively alleviate problems such as compression and elevation of the measured curve caused by multilayer films, as well as decreased matching due to parallel shift differences. In addition, the time required for in-situ measurement and matching with the reference curve is relatively short, enabling real-time in-situ measurement. The present invention uses cosine similarity as a scoring function. Because the cosine similarity evaluation method is sensitive to waveforms, the cosine similarity of two parallel curves in two-dimensional space is 1, which is suitable for the actual measurement situation at this time. Therefore, cosine similarity is selected as a scoring function to evaluate similarity in the case of multilayer media, thereby improving the accuracy of online wafer thin film thickness detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] FIG1 is a schematic flow chart of an online thickness detection method for wafer thin film grinding according to an embodiment of the present invention;

[0033] FIG2 is a schematic diagram of a measured spectrum curve and a reference spectrum curve provided by an embodiment of the present invention;

[0034] FIG3 is a schematic diagram of another measured spectrum curve and a reference spectrum curve provided by an embodiment of the present invention;

[0035] 4 is a schematic diagram of a matching diagram of a measured reflectivity-wavelength curve and a reflectivity-wavelength curve in a reference image library after cosine similarity matching analysis provided by an embodiment of the present invention;

[0036] 5 is a schematic diagram of another matching diagram of a measured reflectivity-wavelength curve and a reflectivity-wavelength curve in a reference image library after cosine similarity matching analysis provided by an embodiment of the present invention;

[0037] 6 is a schematic diagram of a matching diagram of a measured reflectivity-wavelength curve and a reflectivity-wavelength curve in a reference image library after mean square error matching analysis according to an embodiment of the present invention;

[0038] FIG7 is a flow chart of an endpoint detection method for wafer thin film grinding provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0040] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0041] As shown in FIG1 , an online detection method for measuring the thickness of a silicon dioxide film during the grinding process of a silicon-based silicon dioxide wafer film can be executed by an electronic device such as a computer or a server, including:

[0042] S101, during the grinding process of a silicon dioxide wafer thin film on a silicon substrate, obtain a measured spectral curve of the relationship between the reflectivity and wavelength of the silicon dioxide wafer thin film on the silicon substrate. The reflectivity of the silicon dioxide wafer during the grinding process within a limited wavelength range is measured in real time using a spectrometer. During the real-time measurement, the incident light of the spectrometer sequentially encounters the following media: air, glass, polyurethane (PU), polishing fluid, silicon dioxide, and silicon.

[0043] The spectrometer outputs a real-time reflectivity-wavelength curve for the current film thickness within a defined wavelength range. In a specific embodiment, the defined wavelength range is 400-1000 nm, with a wavelength resolution of 0.8 nm, a reflectivity resolution of 1*10*^-4, and a spectral curve acquisition interval of 100 ms.

[0044] As shown in FIG2 , the relatively rough curve above is the measured spectrum curve, where the horizontal axis is wavelength and the vertical axis is reflectivity.

[0045] S102, determining a plurality of characteristic points in the measured spectrum curve. These characteristic points may be automatically selected according to certain rules (e.g., local extreme points, transition points, etc.), or may be manually specified. The number of characteristic points may be fixed or determined in real time based on the characteristics of the current curve. In some embodiments, these characteristic points are evenly distributed on the curve. When performing the operation of selecting characteristic points, characteristic points may be evenly selected on the curve based on the wavelength range of the curve and the preset number of characteristic points.

[0046] Alternatively, multiple feature points are determined in the measured spectral curve. These feature points can be all the data points in the measured spectral curve, or they can be some data points automatically selected according to certain rules (such as local extreme points, jump points, etc.), or they can be some data points specified manually. The number of feature points can be fixed (preset amount) or it can be an amount determined in real time based on the characteristics of the current curve. In some embodiments, these feature points are evenly distributed on the curve. When performing the operation of selecting feature points, feature points can be evenly selected in the curve based on the wavelength range of the curve and the preset number of feature points.

[0047] When selecting feature points, it's important to have a sufficient number of them, for example, hundreds, and to position them within the curve, for example, within a wavelength band rich in characteristic information. Selecting a sufficient number of feature points within an appropriate wavelength range ensures that the subsequently calculated data accurately reflects the characteristics of the spectral curve.

[0048] From the example shown in Figure 2, it can be seen that there may be noise interference data points in the measured spectral curve, such as reflectivity fluctuations at a wavelength of around 670nm. These data points can be called fluctuation anomaly points. When the selected feature points include these fluctuation anomaly points, it may interfere with the accuracy of the subsequent calculated data and cannot accurately reflect the characteristics of the measured spectral curve.

[0049] In order to solve the above-mentioned possible problems, in one embodiment, after obtaining the measured spectral curve, the fluctuation abnormal points are first determined based on the reflectivity of each point and its adjacent points. These fluctuation abnormal points are removed when selecting feature points, and a preset number of points are taken from other normal data points as feature points.

[0050] Through the measured spectral curves at various thicknesses, it can also be found that the spectral characteristic information of the short-wavelength band is richer than that of the long-wavelength band. Taking Figure 2 as an example, for example, the characteristic information of the band before the wavelength of 550nm is richer than the characteristic information of the band after the wavelength. Based on this characteristic, in order to make the data calculated subsequently more accurately reflect the characteristics of the spectral curve, in one embodiment, after obtaining the measured spectral curve, at least one critical point is first determined based on the rate of change of each point in the measured spectral curve. For example, the wavelength of 550nm in Figure 2 is the critical point, and a certain clustering algorithm can be used to determine the critical point. In actual application, since the spectral curve characteristics of different grinding materials are different, the appropriate number of critical points can be set in advance based on the characteristics shown by the reference spectral curve.

[0051] These critical points divide the measured spectrum curve into at least two bands. If there is one critical point, the curve is divided into two bands. If there are two critical points, the curve is divided into three bands. And so on. The number of feature points selected is preset for each band. The preset number of feature points corresponding to the band with a higher rate of change is greater than the preset number corresponding to the band with a lower rate of change. For example, in Figure 2, the preset number of feature points corresponding to the band with a wavelength less than 550nm is greater than the preset number corresponding to the band with a wavelength greater than 550nm.

[0052] After obtaining the critical points and bands, a corresponding preset number of points are selected in each band as feature points, so that the selection density of feature points in bands with richer feature information is higher, thereby improving the accuracy of subsequent calculation data and thus improving the accuracy of thickness detection.

[0053] It should be noted that the above two schemes for selecting feature points can be adopted simultaneously, for example, first excluding abnormal fluctuation points, then determining the above critical points, and then selecting different numbers of feature points in different bands.

[0054] S103: Based on the wavelength values ​​of the multiple characteristic points, corresponding reference points are determined in each reference spectrum curve. The reference spectrum curve is pre-collected. The spectrometer uses the same parameters when collecting the reference spectrum curve and when measuring the spectrum curve, namely, the wavelength range, wavelength resolution, and reflectance resolution are all the same.

[0055] Specifically, a reference library of reflectivity-wavelength reference spectral curves for a number of defined wavelength ranges under different wafer film thicknesses can be pre-established based on the reflectivity n of the wafer film and the pre-detected film thickness range d. In a specific embodiment, the defined wavelength range is 400-1000nm, the wavelength resolution is 0.8nm, the reflectivity resolution is 1*10*^-4 (the same parameters used to collect the measured spectral curve), the pre-detected film thickness range d is 800-200nm, and the film thickness resolution is 0.1nm, that is, a corresponding reflectivity-wavelength spectral curve is established every 0.1nm.

[0056] In this way, the spectral curve of a wafer film of known thickness can be obtained. For example, for a wafer film of known thickness d1, the spectral curve of the relationship between its reflectivity and wavelength can be obtained... For a wafer film of known thickness dn, the spectral curve of the relationship between its reflectivity and wavelength can be obtained. The spectral curve obtained in this way is called a reference spectral curve, and the collection of these reference curves is called a reference library.

[0057] In Figure 2, the relatively smooth dashed curve below is the reference spectrum curve, with wavelength on the horizontal axis and reflectivity on the vertical axis. The reference points corresponding to the characteristic points are the two points on the two curves with the same wavelength value, that is, the two points with the same horizontal coordinate.

[0058] S104, using multiple feature points and multiple reference points, calculate the cosine similarity between the measured spectrum curve and each reference spectrum curve. The feature point (or reference point) has two pieces of information: wavelength and reflectivity. In the embodiment shown in FIG2 , the ordinate of the feature point on the measured spectrum curve is reflectivity, and the abscissa is wavelength. Similarly, the ordinate of the reference point on each reference spectrum curve is reflectivity, and the abscissa is wavelength. A vector can be determined based on the coordinates of the feature point. Similarly, a vector can also be determined based on the coordinates of the reference point. The cosine value of the angle between the two vectors is calculated to obtain the cosine similarity of the two curves.

[0059] S105 , determining a reference spectrum curve that matches the measured spectrum data according to the cosine similarity. Specifically, the reference spectrum curve with the cosine similarity closest to 1 is used as the curve that matches the measured spectrum curve.

[0060] S106: Determine the current thickness of the wafer film according to the thickness corresponding to the matched reference spectrum curve. For example, if the thickness corresponding to the matched reference spectrum curve is di, then di can be determined to be the current thickness of the wafer film.

[0061] This embodiment determines the current thickness of the wafer film by calculating the cosine similarity between the measured spectral curve and the reference spectral curve. This can effectively alleviate problems such as compression and elevation of the measured curve and decreased matching due to parallel shift errors caused by multilayer films and harsh grinding environments. In addition, the time required for in-situ measurement and matching with the reference curve is short, enabling real-time in-situ measurement. The process from real-time acquisition to obtaining the final film thickness can be achieved in just 400 milliseconds. This embodiment uses cosine similarity as the scoring function. Because the cosine similarity evaluation method is sensitive to waveforms, it is more consistent with actual measurement conditions by determining whether the cosine similarity of two parallel curves in two-dimensional space is closest to 1. Therefore, cosine similarity is used as the scoring function to evaluate similarity in the case of multilayer media, thereby improving the accuracy of wafer film thickness.

[0062] Before S102 , the measured spectrum curve may be filtered to remove noise caused by the grinding environment during the grinding process. Filtering methods include but are not limited to wavelet filtering, Fourier filtering, and sliding window average filtering.

[0063] This embodiment improves the accuracy of the measured spectral curve by filtering the noise caused by the grinding environment during the grinding process, thereby improving the accuracy of the wafer film thickness.

[0064] In a preferred embodiment, S102 determines a plurality of characteristic points in the measured spectrum curve, including:

[0065] Get the preset number of feature points;

[0066] According to the preset number of characteristic points and the wavelength range in the measured spectrum curve, multiple characteristic points are evenly selected in the measured spectrum curve.

[0067] This embodiment selects a preset number of feature points within the same wavelength range as the reference curve to calculate the cosine similarity of the feature points within the same wavelength range, thereby improving the accuracy of the calculation and the accuracy of the wafer film thickness.

[0068] In one embodiment, the cosine similarity is calculated in step S104 in the following manner:

[0069] Among them, cosθ represents cosine similarity, A i Indicates the reflectivity of the i-th characteristic point in the measured spectrum curve, B i represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

[0070] Among the 800 selected feature points and reference points, the reflectivity of some points is shown in the following table:

[0071] Among them, the top row is A i Indicates the reflectivity of the i-th characteristic point in the measured spectral curve. The following line is B i Represents the reflectance of the i-th reference point in the reference spectrum curve. Due to the large amount of data, only the values ​​of the first 14 points are shown here.

[0072] Compute the cosine similarity between two curves:

[0073] According to the above calculation method, the cosine similarity between the measured spectrum curve and each reference spectrum curve is obtained, and the reference spectrum curve with the cosine similarity closest to 1 is used as the curve matching the measured spectrum curve.

[0074] In this embodiment, when calculating the cosine similarity, only the reflectivity is used for calculation, thereby reducing the amount of calculation and further reducing the calculation time of matching, further enhancing the real-time performance of film thickness measurement.

[0075] To further improve the accuracy, in another embodiment, the cosine similarity is calculated in step S104 in the following manner:

[0076] Calculate the average of the measured reflectance of multiple feature points and the average of the reference reflectance of multiple reference points;

[0077] Subtract the reflectivity of each feature point from the mean of the measured reflectivity, and subtract the reflectivity of each reference point from the mean of the reference reflectivity;

[0078] The cosine similarity is calculated using the result of the subtraction so that the result is in the interval [-1,1]. The specific calculation formula is:

[0079] Among them, cos′θ represents cosine similarity, It represents the result of subtracting the reflectivity of the i-th feature point in the measured spectral curve from the mean of the measured reflectivity. A represents the result of subtracting the reflectivity of the i-th reference point in the reference spectrum curve from the mean of the reference reflectivity. i Indicates the reflectivity of the i-th characteristic point in the measured spectrum curve, B i represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

[0080] Since the vector representation of the reflectivity curve is always in the first quadrant of the plane coordinate system, the cosine similarity range is always [0,1] regardless of how much the reflectivity curve differs from the theoretical value, and the cosine similarity range [-1,0] is not fully utilized. By using the overall mean subtraction operation in both the measured and theoretical value databases, the measured and theoretical value curves may exist in the first or second quadrant. This fully utilizes the evaluation range of cosine similarity and expands the difference in cosine similarity scores between two curves with large actual differences.

[0081] Here is a more extreme example: if there are two curves with each feature point segment close to 90 degrees, their waveform characteristics are shown in Figure 3, where the reflectivity of the feature points and reference points selected on the two curves is:

[0082] If the data is not subjected to mean subtraction, the cosine similarity is

[0083] Rather than the expected value close to -1, the reason is that the line segment at each feature point is not mapped to the origin, so the cosine similarity cannot reflect the angle between them.

[0084] Using the above processing method, subtract the corresponding mean from the feature points and reference points respectively:

[0085] Then the cosine similarity is

[0086] The cosine similarity calculated in this way is negative, which can better reflect that the similarity between the two curves is extremely low. It utilizes a larger value range [-1,1] and can better reflect the similarity between the two curves, thereby improving the accuracy of matching the measured spectral curve and the reference spectral curve.

[0087] As shown in Figure 4, the curve with greater jitter represents the measured reflectivity-wavelength curve at the current moment, and the smooth straight line represents the reference reflectivity-wavelength curve in the reference library that best matches the measured curve. The reflectivity-wavelength curve in the reference library obtained according to the method of the above embodiment represents a film thickness of 706.9nm. In practice, the accuracy of this film thickness measurement was verified by stopping the grinding and using an Otsuka optical thickness gauge OPTM to select 49 points on the wafer to be ground: the maximum value was measured to be 711.32nm, and the minimum value was measured to be 695.14nm. The measured data is within the actual film thickness range, verifying the accuracy of this algorithm.

[0088] As shown in Figure 5, the curve with greater jitter represents the measured reflectivity-wavelength curve at the current moment, and the smooth straight line represents the reference reflectivity-wavelength curve in the reference library that best matches the measured curve. At this time, according to the method of the above embodiment, the film thickness represented by the reflectivity-wavelength curve in the reference library is 197.4nm. In fact, by stopping the grinding and using the Otsuka optical thickness gauge OPTM to select 49 points on the wafer to be ground for film thickness measurement, the accuracy of this film thickness measurement is verified: the maximum value is measured to be 200.30nm, and the minimum value is measured to be 195.13nm. The measured data is within the actual film thickness range, verifying the accuracy of this algorithm.

[0089] In order to further highlight the beneficial effects of the present invention, this embodiment provides a comparative example using Mean Squared Error (MSE) as the evaluation function of the matching algorithm (this method mainly uses the distance between two curves as the evaluation criterion, and the smaller the distance, the more similar the two curves are considered to be). The specific calculation formula is:

[0090] Except for S103-S105, the rest are the same. At a certain moment in the grinding, the real-time film thickness measurement is performed according to the above process, and the final matching diagram is shown in Figure 6. The curve with larger jitter represents the measured reflectivity-wavelength curve at the current moment, and the smooth straight line represents the reference reflectivity-wavelength curve in the reference library that matches the measured curve best. At this time, the film thickness represented by the reflectivity-wavelength curve in the reference library is 64.2nm. However, by using the Otsuka optical thickness gauge OPTM to select 49 points on the wafer to be ground for film thickness measurement, the maximum value was measured to be 688nm and the minimum value was measured to be 695.13nm. It can be seen that this matching method cannot accurately measure the film thickness during the grinding process.

[0091] As shown in FIG7 , an embodiment of the present invention provides an endpoint detection method for wafer thin film grinding, which can be executed by an electronic device such as a computer or a server. The method includes the following operations:

[0092] Using the online thickness detection method of the above embodiment, the thickness of the wafer film is measured in real time through steps S101-S106;

[0093] S107, judging whether the current thickness of the wafer film reaches the target thickness. When the current thickness of the wafer film reaches the target thickness, executing S108; otherwise, continuing grinding and returning to step S101.

[0094] S108, stop grinding.

[0095] In this embodiment, since the reference spectrum curve matched by cosine similarity is highly accurate, the real-time wafer film thickness is accurate, and grinding is stopped in time when the current thickness reaches the target thickness, thereby improving the accuracy of wafer grinding.

[0096] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0097] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0098] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0100] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for online thickness detection of wafer thin film grinding, characterized in that: include: During the grinding process of the wafer film, a measured spectrum curve of the relationship between the reflectivity and wavelength of the wafer film is obtained; Determining a plurality of characteristic points in the measured spectral curve; Based on the wavelength values ​​of the plurality of characteristic points, respectively determining corresponding reference points in each reference spectrum curve; Calculating the cosine similarity between the measured spectrum curve and each reference spectrum curve using the plurality of feature points and the plurality of reference points; Determine a reference spectrum curve matching the measured spectrum data according to the cosine similarity; The current thickness of the wafer thin film is determined according to the thickness corresponding to the matched reference spectrum curve.

2. The thickness online detection method according to claim 1, characterized in that: Determine a plurality of characteristic points in the measured spectrum curve, including: Determine the fluctuation abnormal point according to the reflectivity of each point and its adjacent points in the measured spectrum curve; A preset number of points are selected from each point in the measured spectrum curve except the fluctuation abnormal point as the characteristic points.

3. The thickness online detection method according to claim 1, characterized in that: Determine a plurality of characteristic points in the measured spectrum curve, including: Determining at least one critical point according to the change rate of each point in the measured spectrum curve, so that the measured spectrum curve is divided into at least two bands; A corresponding preset number of points are selected in each of the bands as the feature points, wherein the preset number corresponding to the band with a higher rate of change is greater than the preset number corresponding to the band with a lower rate of change.

4. The thickness online detection method according to claim 1, characterized in that: In the step of calculating the cosine similarity between the measured spectral curve and each reference spectral curve, the cosine similarity between the measured spectral curve and the reference spectral curve is calculated in the following manner: Among them, cosθ represents cosine similarity, A i represents the reflectivity of the i-th characteristic point in the measured spectral curve, B i surface represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

5. The thickness online detection method according to claim 1, characterized in that: Calculating the cosine similarity between the measured spectral curve and each reference spectral curve includes: Processing the multiple feature points and the multiple reference points; The cosine similarity is calculated using the processed feature points and the reference points, so that the calculation result is in the interval [-1, 1].

6. The online thickness detection method according to claim 5, characterized in that: Processing the plurality of feature points and the plurality of reference points includes: Calculating the average values ​​of the measured reflectances of the plurality of feature points and the average values ​​of the reference reflectances of the plurality of reference points; The reflectivity of each feature point is subtracted from the measured reflectivity average value, and the reflectivity of each reference point is subtracted from the reference reflectivity average value.

7. The thickness online detection method according to claim 6, characterized in that: The cosine similarity is calculated in the following way: Among them, cos′θ represents cosine similarity, represents the result of subtracting the reflectivity of the ith feature point in the measured spectral curve from the mean value of the measured reflectivity, A represents the result of subtracting the reflectivity of the ith reference point in the reference spectrum curve from the mean value of the reference reflectivity. i represents the reflectivity of the i-th characteristic point in the measured spectral curve, B i represents the reflectance of the i-th reference point in the reference spectrum curve, and n represents the number of feature points and reference points.

8. The online thickness detection method according to claim 1, characterized in that: The wafer thin film is a multi-layer film.

9. The thickness online detection method according to claim 1, characterized in that: The characteristic points are all data points in the measured spectrum curve.

10. The online thickness detection method according to claim 1, characterized in that: The number of the feature points is a preset number.

11. The online thickness detection method according to claim 1, characterized in that: The number of the characteristic points is a quantity determined in real time according to the characteristics of the measured spectrum curve.

12. The online thickness detection method according to claim 1, characterized in that: The plurality of characteristic points are evenly distributed on the measured spectrum curve; Determining a plurality of characteristic points in the measured spectral curve includes: uniformly selecting a plurality of characteristic points according to a wavelength range of the measured spectral curve and a preset number of characteristic points.

13. The online thickness detection method according to claim 1, characterized in that: The plurality of characteristic points are located in a band in the measured spectrum curve that is relatively rich in characteristic information.

14. The online thickness detection method according to claim 1, characterized in that: The reference spectral curve that matches the measured spectral data refers to the reference spectral curve whose cosine similarity is closest to 1.

15. The online thickness detection method according to claim 1, characterized in that: The reference spectrum curve is collected in advance, and the parameters used when collecting the reference spectrum curve and collecting the measured spectrum curve are the same, and the parameters include wavelength range, wavelength resolution, and reflectivity resolution.

16. The online thickness detection method according to claim 1, characterized in that: Each of the reference spectral curves is a reflectivity-wavelength reference spectral curve established in advance based on the reflectivity of the wafer film and the pre-detected film thickness range, and includes a number of reflectivity-wavelength reference spectral curves with limited wavelength ranges under different wafer film thicknesses.

17. The online thickness detection method according to claim 1, characterized in that: Before determining a plurality of characteristic points in the measured spectrum curve, the method further includes: The measured spectrum curve is filtered to filter out the noise caused by the grinding environment during the grinding process, and the filtering method is any one of wavelet filtering, Fourier filtering and sliding window average filtering.

18. An endpoint detection method for wafer thin film grinding, characterized in that: include: Using the thickness online detection method described in any one of claims 1 to 17, real-time monitoring is performed to determine whether the thickness of the wafer film reaches the target thickness; When the thickness of the wafer film reaches the target thickness, grinding is stopped.

19. An online thickness detection device for wafer thin film grinding, characterized in that: include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor executes the online thickness detection method for wafer thin film grinding as described in any one of claims 1-17.

20. An endpoint detection device for wafer thin film grinding, characterized in that: include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor executes the endpoint detection method for wafer thin film grinding as described in claim 18.

Images