Method for detecting wrong work piece not subjected to polishing and optical film thickness measurement device

The method uses an autoencoder trained on correct workpiece spectrum data to identify and prevent the polishing of incorrect workpieces, ensuring accurate polishing and end-point detection by adjusting the polishing process based on the correct workpiece identification.

JP2024158610AActive Publication Date: 2024-11-08EBARA CORP
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023073956
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-04-28
Publication Date
2024-11-08
Estimated Expiration
2043-04-28

AI Technical Summary

Technical Problem

Existing polishing equipment fails to detect and prevent the polishing of incorrect workpieces, leading to improper polishing and end-point detection when a different type of wafer is mistakenly introduced, as the polishing recipe is not suitable for the incorrect workpiece.

Method used

A method using an autoencoder trained through machine learning on inspection spectrum data from correct workpieces to identify incorrect workpieces by calculating the difference between input and output data, determining if the workpiece is not intended for polishing based on a threshold, and adjusting the polishing process accordingly.

Benefits of technology

Effectively detects and prevents the polishing of incorrect workpieces by ensuring the correct workpiece is identified before polishing, maintaining proper polishing and end-point detection, thereby enhancing the accuracy and efficiency of the polishing process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024158610000001_ABST
    Figure 2024158610000001_ABST
Patent Text Reader

Abstract

To provide a method capable of detecting a wrong workpiece (for example, a wafer) not subjected to polishing.SOLUTION: A method comprises the steps of: preparing spectral data for inspection of reflected light from a workpiece W before polishing or after starting polishing of the workpiece W; inputting the spectral data for inspection to an autoencoder; calculating difference between output data output from the autoencoder and the spectral data for inspection; and, when the difference is larger than a threshold, determining that the workpiece W used to prepare the spectrum data for inspection is a wrong workpiece not subjected to polishing.SELECTED DRAWING: Figure 5
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a technique for measuring film thickness based on the spectrum of reflected light from workpieces used in the manufacture of semiconductor devices, such as wafers, substrates, and panels, and more particularly to a technique for detecting erroneous workpieces that are not to be polished. [Background technology]

[0002] The manufacturing process of semiconductor devices involves the use of SiO 2 The manufacturing process of back-illuminated CMOS sensors and through-silicon vias (TSVs) includes various processes such as polishing insulating films such as SiO2 and polishing metal films such as copper and tungsten. In addition to the polishing process of insulating films and metal films, the manufacturing process of back-illuminated CMOS sensors and through-silicon vias (TSVs) also includes the polishing process of silicon layers (silicon wafers). Wafer polishing is completed when the thickness of the films (insulating films, metal films, silicon layers, etc.) that make up the surface of the wafer reaches a specified target value.

[0003] Wafer polishing is performed using a polishing machine. In order to measure the thickness of non-metallic films such as insulating films and silicon layers, the polishing machine is generally equipped with an optical film thickness measuring device. This optical film thickness measuring device is configured to detect the film thickness of the wafer by directing light emitted from a light source to the surface of the wafer and analyzing the spectrum of the light reflected from the wafer. When the film thickness of the wafer reaches a target value, the polishing of the wafer is terminated.

[0004] Wafer polishing is performed according to a polishing recipe set in the polishing device. Since this polishing recipe differs for each type of wafer, it is necessary to change the polishing recipe based on the type of wafer to be polished. In addition, since the spectrum of the reflected light from the wafer differs depending on the type of wafer, the conditions for detecting the polishing end point also differ. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2018-194427 A Summary of the Invention [Problem to be solved by the invention]

[0006] If a wafer that is not to be polished (i.e., a different type of wafer that does not correspond to the polishing recipe set in the polishing apparatus) is loaded into the polishing apparatus and the wafer is polished by the polishing apparatus, the polishing recipe will not be suitable for that wafer and polishing endpoint detection will not be performed properly, so the wafer cannot be polished normally.

[0007] Thus, the present invention provides a method and optical film thickness measurement apparatus that can detect erroneous workpieces (eg, wafers) that are not to be polished. [Means for solving the problem]

[0008] In one aspect, a method for detecting an incorrect workpiece that is not to be polished is provided, the method including: creating inspection spectral data of reflected light from the workpiece before polishing the workpiece or after polishing begins; inputting the inspection spectral data into an autoencoder, the autoencoder being a learned model constructed by machine learning using training data including a plurality of index spectral data of a plurality of reflected lights from a correct workpiece that is to be polished; calculating a difference between output data output from the autoencoder and the inspection spectral data; and, when the difference is greater than a threshold, determining that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished.

[0009] In one aspect, the step of creating the inspection spectral data is a step of creating inspection spectral data of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an initial stage of chemical mechanical polishing of the workpiece. In one embodiment, the method further includes classifying a plurality of spectral data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm, creating the training data including a plurality of indicator spectral data belonging to one of the plurality of groups, and performing the machine learning using the training data to construct the autoencoder, which is the trained model, wherein the plurality of indicator spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

[0010] In one aspect, a method for detecting an incorrect workpiece not to be polished is provided, the method including determining reference spectral data from a plurality of index spectral data of a plurality of reflected light from a correct workpiece to be polished, creating inspection spectral data of reflected light from the workpiece before or after polishing of the workpiece begins, calculating a difference between the reference spectral data and the inspection spectral data, and determining that the workpiece used to create the inspection spectral data is an incorrect workpiece not to be polished when the difference is greater than a threshold value.

[0011] In one aspect, the step of creating the inspection spectral data is a step of creating inspection spectral data of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an initial stage of chemical mechanical polishing of the workpiece. In one embodiment, the difference is the Euclidean distance. In one embodiment, the step of determining the reference spectral data includes a step of classifying a plurality of spectral data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm, and determining the reference spectral data from a plurality of index spectral data belonging to one of the plurality of groups, the plurality of index spectral data belonging to one of the plurality of groups being a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished. In one aspect, the method further includes normalizing the plurality of index spectral data to create a plurality of normalized index spectral data, and normalizing the test spectral data to create normalized test spectral data.

[0012] In one aspect, there is provided an optical film thickness measurement device that optically measures the film thickness of a workpiece, the optical film thickness measurement device comprising: a light source that emits light; an optical sensor head that irradiates the workpiece with the light emitted from the light source and receives reflected light from the workpiece; and a processing system that determines the film thickness of the workpiece based on spectral measurement data of the reflected light from the workpiece, wherein the processing system creates inspection spectral data of the reflected light from the workpiece before polishing of the workpiece or after polishing is started, and inputs the inspection spectral data to an autoencoder, the autoencoder being a learned model constructed by machine learning using training data including a plurality of index spectral data of a plurality of reflected lights from a correct workpiece that is to be polished, and is configured to calculate a difference between output data output from the autoencoder and the inspection spectral data, and when the difference is greater than a threshold value, determine that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished.

[0013] In one aspect, the processing system is configured to generate spectral data for inspection of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an initial stage of chemical mechanical polishing of the workpiece. In one embodiment, the processing system is configured to classify multiple spectral data acquired before or after the start of polishing of a past workpiece into multiple groups according to a clustering algorithm, create the training data including multiple indicator spectral data belonging to one of the multiple groups, and perform the machine learning using the training data to construct the autoencoder, which is the trained model, and the multiple indicator spectral data belonging to one of the multiple groups is multiple spectral data of multiple reflected lights from a correct workpiece to be polished.

[0014] In one aspect, there is provided an optical film thickness measurement device that optically measures the film thickness of a workpiece, the optical film thickness measurement device comprising: a light source that emits light; an optical sensor head that irradiates the workpiece with the light emitted from the light source and receives the reflected light from the workpiece; and a processing system that determines the film thickness of the workpiece based on spectral measurement data of the reflected light from the workpiece, wherein the processing system is configured to determine reference spectral data from a plurality of index spectral data of a plurality of reflected lights from a correct workpiece that is to be polished, create inspection spectral data of the reflected light from the workpiece before polishing the workpiece or after polishing starts, calculate a difference between the reference spectral data and the inspection spectral data, and determine that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

[0015] In one aspect, the processing system is configured to generate spectral data for inspection of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an initial stage of chemical mechanical polishing of the workpiece. In one embodiment, the difference is the Euclidean distance. In one embodiment, the processing system is configured to classify a plurality of spectral data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm, and to determine the reference spectral data from a plurality of index spectral data belonging to one of the plurality of groups, the plurality of index spectral data belonging to one of the plurality of groups being a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished. In one aspect, the processing system is configured to normalize the plurality of index spectral data to create a plurality of normalized index spectral data, and to normalize the test spectral data to create normalized test spectral data. Effect of the Invention

[0016] The autoencoder is a trained model constructed by machine learning using training data including a plurality of index spectral data of a plurality of reflected light from a correct workpiece to be polished. More specifically, machine learning is performed so that when the index spectral data is input to the autoencoder, the autoencoder outputs data that is substantially the same as the input index spectral data. When the spectral data of the reflected light from an incorrect workpiece that is not to be polished is input to the autoencoder, the autoencoder outputs data that is significantly different from the input spectral data. Therefore, when the difference between the input inspection spectral data and the output data (also called the reconstruction error) is greater than a threshold value, it can be determined that the workpiece used to generate the inspection spectral data is an incorrect workpiece that is not to be polished.

[0017] Similarly, when the difference between the predetermined reference spectral data and the inspection spectral data is greater than a threshold value, it can be determined that the workpiece used to generate the inspection spectral data is an incorrect workpiece that is not to be polished. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 is a schematic diagram illustrating an embodiment of a polishing apparatus. [Diagram 2] FIG. 2 is a cross-sectional view showing a detailed configuration of an embodiment of an optical film thickness measurement device. [Diagram 3] 1 is a graph showing an example of spectrum measurement data. [Figure 4] FIG. 1 is a schematic diagram illustrating an example of an autoencoder. [Diagram 5] FIG. 13 is a diagram showing an example of the spectral data of reflected light from a correct workpiece to be polished being input to an autoencoder. [Figure 6] FIG. 13 is a diagram showing an example in which spectral data of reflected light from an erroneous workpiece that is not a target for polishing is input to an autoencoder. [Figure 7] 1 is a flow chart illustrating one embodiment of determining whether a workpiece is an incorrect workpiece that is not to be polished. [Figure 8] FIG. 13 is a diagram showing how a plurality of spectral data are divided into three groups by clustering. [Figure 9] 11 is a flow chart illustrating another embodiment of determining whether a workpiece is an incorrect workpiece that is not to be polished. [Figure 10] FIG. 10(a) is a graph showing an example of a plurality of index spectral data before normalization, and FIG. 10(b) is a graph showing an example of a plurality of index spectral data after normalization. [Figure 11]FIG. 11(a) is a graph showing the results of periodically calculating the difference between reference spectral data determined from multiple index spectral data before normalization and test spectral data before normalization, and FIG. 11(b) is a graph showing the results of periodically calculating the difference between reference spectral data determined from multiple index spectral data after normalization and test spectral data after normalization. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a schematic diagram showing one embodiment of a polishing apparatus. As shown in FIG. 1, the polishing apparatus includes a polishing table 3 supporting a polishing pad 2, a polishing head 1 pressing a workpiece W having a film against the polishing pad 2, a table motor 6 rotating the polishing table 3, a liquid supply nozzle 5 for supplying a polishing liquid such as a slurry onto the polishing pad 2, and an operation control unit 9 for controlling the operation of the polishing apparatus. The upper surface of the polishing pad 2 constitutes a polishing surface 2a for polishing the workpiece W. Examples of the workpiece W include wafers, substrates (circular substrates, polygonal substrates), panels, etc. used in the manufacture of semiconductor devices.

[0020] The polishing head 1 is connected to a head shaft 10, which is connected to a polishing head motor (not shown). The polishing head motor rotates the polishing head 1 together with the head shaft 10 in the direction indicated by the arrow. The polishing table 3 is connected to a table motor 6, which is configured to rotate the polishing table 3 and the polishing pad 2 in the direction indicated by the arrow. The polishing head 1, the polishing head motor, the liquid supply nozzle 5, and the table motor 6 are connected to an operation control unit 9.

[0021] The workpiece W is polished as follows. While the polishing table 3 and polishing head 1 are rotated in the direction indicated by the arrow in Fig. 1, a polishing liquid is supplied from a liquid supply nozzle 5 to the polishing surface 2a of the polishing pad 2 on the polishing table 3. While the workpiece W is rotated by the polishing head 1, the workpiece W is pressed against the polishing surface 2a of the polishing pad 2 by the polishing head 1 with the polishing liquid present on the polishing pad 2. The surface of the workpiece W is polished by the chemical action of the polishing liquid and the mechanical action of the abrasive grains contained in the polishing liquid and / or the polishing pad 2. Polishing the workpiece W using such a polishing liquid (e.g., a slurry) is chemical mechanical polishing of the workpiece W.

[0022] The operation control unit 9 includes a storage device 9a in which a program is stored, and a calculation device 9b that executes calculations according to instructions included in the program. The operation control unit 9 includes at least one computer. The storage device 9a includes a main storage device such as a random access memory (RAM), and an auxiliary storage device such as a hard disk drive (HDD) or a solid state drive (SSD). Examples of the calculation device 9b include a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). However, the specific configuration of the operation control unit 9 is not limited to these examples.

[0023] The polishing apparatus includes an optical film thickness measuring device 20 for measuring the thickness of a film on the workpiece W. The optical film thickness measuring device 20 includes a light source 22 for emitting light, an optical sensor head 25 for irradiating the workpiece W with the light from the light source 22 and receiving the light reflected from the workpiece W, a spectrometer 27 connected to the optical sensor head 25, and a processing system 30 for determining the thickness of the film on the workpiece W based on the light reflected from the workpiece W. The optical sensor head 25 is disposed within the polishing table 3 and rotates together with the polishing table 3. Although not shown, a plurality of optical sensor heads 25 connected to the spectrometer 27 and the light source 22 may be provided.

[0024] The processing system 30 includes a storage device 30a in which a program is stored, and an arithmetic device 30b that executes calculations according to instructions included in the program. The processing system 30 is composed of at least one computer. The storage device 30a includes a main storage device such as a random access memory (RAM), and an auxiliary storage device such as a hard disk drive (HDD) or a solid state drive (SSD). Examples of the arithmetic device 30b include a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). However, the specific configuration of the processing system 30 is not limited to these examples.

[0025] Each of the operation control unit 9 and the processing system 30 may be composed of a plurality of computers. For example, each of the operation control unit 9 and the processing system 30 may be composed of a combination of an edge server and a cloud server. In one embodiment, the operation control unit 9 and the processing system 30 may be composed of a single computer.

[0026] 2 is a cross-sectional view showing an embodiment of a detailed configuration of the optical film thickness measurement device 20. The optical film thickness measurement device 20 includes a light source 22 that emits light, a light projecting optical fiber cable 31 connected to the light source 22, and a light receiving optical fiber cable 32 connected to a spectroscope 27. A tip 31a of the light projecting optical fiber cable 31 and a tip 32a of the light receiving optical fiber cable 32 constitute the optical sensor head 25. That is, the light projecting optical fiber cable 31 guides the light of the light source 22 to the workpiece W on the polishing pad 2, and the light receiving optical fiber cable 32 receives the reflected light from the workpiece W and transmits it to the spectroscope 27.

[0027] The spectroscope 27 is connected to the processing system 30. The light projecting fiber optic cable 31, the light receiving fiber optic cable 32, the light source 22, and the spectroscope 27 are attached to the polishing table 3 and rotate together with the polishing table 3 and the polishing pad 2. The optical sensor head 25, which is composed of the tip 31a of the light projecting fiber optic cable 31 and the tip 32a of the light receiving fiber optic cable 32, is disposed facing the surface of the workpiece W on the polishing pad 2. The position of the optical sensor head 25 is a position where it crosses the surface of the workpiece W on the polishing pad 2 every time the polishing table 3 and the polishing pad 2 rotate once. The polishing pad 2 has a through hole 2b located above the optical sensor head 25. The optical sensor head 25 irradiates light onto the workpiece W through the through hole 2b every time the polishing table 3 rotates once, and receives reflected light from the workpiece W through the through hole 2b.

[0028] The light source 22 is a flash light source that repeatedly emits light at short time intervals. An example of the light source 22 is a xenon flash lamp. The light source 22 is electrically connected to the operation control unit 9, and emits light upon receiving a trigger signal sent from the operation control unit 9. More specifically, while the optical sensor head 25 traverses the surface of the workpiece W on the polishing pad 2, the light source 22 receives multiple trigger signals and emits light multiple times. Therefore, each time the polishing table 3 rotates once, light is irradiated onto multiple measurement points on the workpiece W.

[0029] The light emitted by the light source 22 is transmitted to the optical sensor head 25 through the light-projecting fiber optic cable 31 and is emitted from the optical sensor head 25. The light passes through the through hole 2b of the polishing pad 2 and is incident on the workpiece W on the polishing pad 2. The light reflected from the workpiece W passes through the through hole 2b of the polishing pad 2 again and is received by the optical sensor head 25. The reflected light from the workpiece W is transmitted to the spectrometer 27 through the light-receiving fiber optic cable 32.

[0030] The spectrometer 27 is configured to separate the light according to wavelength and measure the intensity of the reflected light at each wavelength over a predetermined wavelength range. That is, the spectrometer 27 separates the light reflected from the workpiece W according to wavelength and measures the intensity of the reflected light at each wavelength over a predetermined wavelength range to generate spectral measurement data. The spectral measurement data is sent to the processing system 30.

[0031] Fig. 3 is a graph showing an example of the spectrum measurement data. In Fig. 3, the vertical axis represents the intensity of the reflected light from the workpiece W, and the horizontal axis represents the wavelength of the reflected light. As shown in Fig. 3, the spectrum measurement data can be expressed as a spectrum (spectral waveform) showing the relationship between the wavelength and intensity of light.

[0032] The processing system 30 is configured to determine the film thickness of the workpiece W based on the spectral measurement data of the reflected light from the workpiece W. A known technique is used to determine the film thickness of the workpiece W based on the spectral measurement data. For example, the processing system 30 determines reference spectral data from a reference spectrum library that has a shape closest to the spectral measurement data, and determines the film thickness associated with the determined reference spectral data. In another example, the processing system 30 performs a Fourier transform on the spectral measurement data and determines the film thickness from the obtained frequency spectrum.

[0033] The workpiece W is polished according to a polishing recipe set in the polishing apparatus. This polishing recipe includes the type of polishing liquid used, the rotation speed of the polishing table, the rotation speed of the polishing head, the pressing force of the polishing head against the workpiece W, film thickness measurement conditions, polishing end point detection conditions, etc. The polishing recipe differs depending on the type of workpiece (i.e., the layered structure that constitutes the surface of the workpiece). In particular, different types of workpieces have different spectra of reflected light from the workpieces, and therefore different conditions for polishing end point detection.

[0034] If a workpiece that is not to be polished (i.e., a different type of workpiece that does not correspond to the polishing recipe set in the polishing apparatus) is loaded into the polishing apparatus and the workpiece is polished by the polishing apparatus, the polishing recipe will not be suitable for that workpiece and polishing endpoint detection will not be performed properly, so the workpiece cannot be polished normally.

[0035] Therefore, in this embodiment, the processing system 30 is configured to determine whether the workpiece W is a correct workpiece to be polished or an incorrect workpiece not to be polished before or after polishing of the workpiece W is started. A correct workpiece to be polished means a workpiece that corresponds to (is suitable for) the polishing recipe set in the polishing apparatus, and an incorrect workpiece that is not to be polished means a workpiece that does not correspond to (is not suitable for) the polishing recipe set in the polishing apparatus.

[0036] The processing system 30 creates inspection spectrum data of the reflected light from the workpiece W before or after polishing of the workpiece W is started, inputs the inspection spectrum data to the autoencoder, and calculates the difference between the output data output from the autoencoder and the inspection spectrum data. The inspection spectrum data is generated in the same manner as the spectrum measurement data described with reference to Figures 2 and 3, and therefore a detailed description thereof will be omitted.

[0037] In one embodiment, the timing for creating the inspection spectrum data is immediately before polishing the workpiece W or immediately after the start of polishing. In one embodiment, the processing system 30 creates inspection spectrum data of reflected light from the workpiece W during water polishing of the workpiece W, which is performed before chemical mechanical polishing of the workpiece W using a polishing liquid such as a slurry. The water polishing of the workpiece W is performed in a state where the polishing head 1 holds the workpiece W on the polishing surface 2a of the polishing pad 2 while water is supplied from the liquid supply nozzle 5 onto the polishing surface 2a of the polishing pad 2 instead of a polishing liquid (e.g., a slurry). Unlike a polishing liquid such as a slurry, water has no etching function and does not contain abrasive grains. In addition, the polishing head 1 does not strongly press the workpiece W against the polishing pad 2 during water polishing. Therefore, the polishing of the workpiece W during water polishing does not progress, and substantial polishing of the workpiece W does not start.

[0038] In one embodiment, the processing system 30 may generate inspection spectrum data of the light reflected from the workpiece W in an initial stage of chemical mechanical polishing of the workpiece W using a polishing liquid such as a slurry. Examples of the initial stage of chemical mechanical polishing include a period from when chemical mechanical polishing of the workpiece W is started until a predetermined time has elapsed, or a period from when chemical mechanical polishing of the workpiece W is started until the polishing table 3 has rotated a predetermined number of times. For example, after the chemical mechanical polishing of the workpiece W is started, the processing system 30 generates inspection spectrum data of the light reflected from the workpiece W when the polishing table 3 is rotating for the first time.

[0039] The processing system 30 inputs the inspection spectrum data generated immediately before or immediately after the start of chemical mechanical polishing of the workpiece W as described above to the autoencoder. The autoencoder is a learned model constructed by machine learning using training data including a plurality of indicator spectrum data of a plurality of reflected lights from at least one correct workpiece to be polished. The plurality of indicator spectrum data included in the training data is a plurality of past spectrum data acquired before or after the start of chemical mechanical polishing of the correct workpiece to be polished. For example, the plurality of indicator spectrum data included in the training data is a plurality of spectrum data acquired during past water polishing or past chemical mechanical polishing of the correct workpiece to be polished.

[0040] The correct workpiece used to generate the multiple index spectral data included in the training data may be one or multiple pieces. In one embodiment, in order to improve the accuracy of machine learning, multiple index spectral data are generated using multiple correct workpieces, and training data including these index spectral data is created.

[0041] The arithmetic unit 30b of the processing system 30 executes machine learning using the training data according to instructions included in the program stored in the storage device 30a to construct an autoencoder. The autoencoder, which is a trained model constructed by machine learning, is stored in the storage device 30a.

[0042] When index spectral data is input to the autoencoder, machine learning is performed so that the autoencoder outputs data that is substantially the same as the index spectral data that was input. When spectral data of reflected light from an incorrect workpiece that is not the object of polishing is input to the autoencoder, the autoencoder outputs data that is significantly different from the input spectral data.

[0043] The processing system 30 is configured to determine that when the difference (also called the reconstruction error) between the input inspection spectrum data and the output data of the autoencoder is greater than a threshold value, the workpiece used to create the inspection spectrum data is an incorrect workpiece that is not to be polished.

[0044] 4 is a schematic diagram showing an example of an autoencoder. The autoencoder includes an encoder 41 that extracts features of input spectral data, and a decoder 42 that reconstructs the input spectral data from the extracted features.

[0045] Fig. 5 is a diagram showing an example when the spectral data of the reflected light from a correct workpiece to be polished is input to an autoencoder. The autoencoder is a trained model constructed by machine learning using training data including a plurality of index spectral data using a correct workpiece to be polished. Therefore, as shown in Fig. 5, when the spectral data of a correct workpiece to be polished is input to the autoencoder, the autoencoder outputs output data that is substantially the same as the input spectral data. Therefore, the difference (reconstruction error) between the input spectral data and the output data is small.

[0046] 6 is a diagram showing an example when the spectrum data of the reflected light from an incorrect workpiece that is not the target of polishing is input to the autoencoder. As shown in FIG. 6, when the spectrum data of the reflected light from the incorrect workpiece that is not the target of polishing is input to the autoencoder, the autoencoder outputs output data that is significantly different from the input spectrum data. Therefore, the difference (reconstruction error) between the input spectrum data and the output data is large.

[0047] In this manner, the processing system 30 determines whether the workpiece W is a correct workpiece to be polished or an incorrect workpiece not to be polished based on the difference between the inspection spectrum data of the workpiece W and the output data obtained from the autoencoder. If the processing system 30 determines that the workpiece W is an incorrect workpiece not to be polished, the processing system 30 sends the determination result to the operation control unit 9. In response to the determination result, the operation control unit 9 does not allow the polishing apparatus to perform chemical mechanical polishing of the workpiece W, or stops the chemical mechanical polishing of the workpiece W by the polishing apparatus. Alternatively, in response to the determination result, the operation control unit 9 may change the polishing recipe set in the polishing apparatus to a polishing recipe prepared in advance for the workpiece W.

[0048] FIG. 7 is a flow chart illustrating one embodiment for determining whether a workpiece W is an incorrect workpiece that is not to be polished. In step 101, the processing system 30 generates spectral data for inspection of reflected light from the workpiece W immediately before or immediately after starting polishing of the workpiece W. For example, the processing system 30 generates spectral data for inspection of reflected light from the workpiece W during water polishing of the workpiece W or during an early stage of chemical mechanical polishing of the workpiece W.

[0049] In step 102, the processing system 30 inputs the spectral data generated in step 101 above into an autoencoder and outputs output data from the autoencoder. In step 103, the processing system 30 calculates the difference between the spectral data generated in step 101 above and the output data output from the autoencoder. In step 104, processing system 30 determines whether the difference is greater than a threshold value. In step 105, the processing system 30 determines that the workpiece W is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

[0050] In step 106, the processing system 30 sends the judgment result of the workpiece W to the operation control unit 9. In response to the judgment result, the operation control unit 9 either does not allow the polishing apparatus to polish the workpiece W, or stops polishing the workpiece W by the polishing apparatus, or changes the polishing recipe set in the polishing apparatus to a polishing recipe prepared in advance for the workpiece W. In step 107, the processing system 30 polishes the workpiece W according to the polishing recipe set in the polishing apparatus if the difference is less than the threshold value.

[0051] The training data used for the machine learning of the autoencoder includes only the indicator spectral data of the correct workpiece to be polished among the many spectral data obtained during the past water polishing or the past chemical mechanical polishing. If the many spectral data obtained during the past water polishing or the past chemical mechanical polishing include the spectral data of the incorrect workpiece that is not the polishing target, it is necessary not to use such spectral data as the training data. In other words, it is necessary to extract the indicator spectral data of the correct workpiece to be polished from the many spectral data. However, such an operation is very laborious.

[0052] Therefore, in one embodiment, the processing system 30 classifies a plurality of spectral data acquired before or after the start of chemical mechanical polishing of a past workpiece into a plurality of groups according to a clustering algorithm. For example, the processing system 30 classifies a large number of spectral data acquired during a past water polishing or a past chemical mechanical polishing into a plurality of groups according to a clustering algorithm. Then, the processing system 30 is configured to create training data including a plurality of indicator spectral data belonging to one of the plurality of groups, and to perform machine learning using the training data to construct an autoencoder, which is a trained model. The plurality of indicator spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

[0053] Clustering is also called cluster analysis, and a clustering algorithm is a type of artificial intelligence algorithm that classifies multiple spectral data based on certain characteristics. A program for dividing multiple spectral data into multiple groups according to the clustering algorithm is stored in the storage device 30a of the processing system 30. The calculation device 30b of the processing system 30 divides the multiple spectral data into multiple groups by executing calculations according to instructions included in the program.

[0054] Fig. 8 is a diagram showing how a plurality of spectral data are divided into three groups by clustering. In the example shown in Fig. 8, the processing system 30 executes machine learning using the spectral data included in the second group, which is a group of index spectral data, as training data. Clustering can quickly divide a huge amount of spectral data acquired in the past into index spectral data and other spectral data. In the example shown in Fig. 8, the plurality of spectral data are divided into three groups, but they may be divided into two groups, or four or more groups.

[0055] Next, another embodiment for determining whether or not the workpiece W is an incorrect workpiece that is not to be polished will be described. The configuration and operation of the optical film thickness measurement device 20 used in this embodiment are the same as those in the above embodiment described with reference to Figures 1 to 8, so a duplicated description will be omitted.

[0056] In this embodiment, the processing system 30 determines reference spectral data from multiple index spectral data of multiple reflected light from a correct workpiece to be polished, creates inspection spectral data of reflected light from the workpiece W before polishing the workpiece W or after polishing begins, calculates the difference between the reference spectral data and the inspection spectral data, and when the difference is greater than a threshold value, determines that the workpiece W used to create the inspection spectral data is an incorrect workpiece that is not to be polished.

[0057] In one embodiment, the timing for creating the inspection spectral data is immediately before or immediately after the start of polishing of the workpiece W. In one embodiment, the processing system 30 creates the inspection spectral data of the light reflected from the workpiece W during water polishing of the workpiece W, which is performed before the chemical mechanical polishing of the workpiece W. In another embodiment, the processing system 30 creates the inspection spectral data of the light reflected from the workpiece W at an early stage of the chemical mechanical polishing of the workpiece W.

[0058] The reference spectral data may be one or an average of a plurality of index spectral data of a plurality of reflected lights from a correct workpiece to be polished. The index spectral data is a plurality of past spectral data obtained before or after polishing of the correct workpiece to be polished has begun. For example, the index spectral data is a plurality of spectral data obtained during a past water polishing or chemical mechanical polishing of the correct workpiece to be polished.

[0059] The correct workpiece used to generate the multiple index spectral data may be one or multiple pieces. In one embodiment, in order to improve the accuracy of the reference spectral data, multiple index spectral data are generated using multiple correct workpieces, and the reference spectral data is determined from the index spectral data. The reference spectral data is stored in the storage device 30a.

[0060] The processing system 30 calculates the difference between the reference spectrum data and the test spectrum data. In this embodiment, the processing system 30 calculates the Euclidean distance as the difference between the reference spectrum data and the test spectrum data.

[0061] When the difference between the reference spectrum data and the test spectrum data is smaller than the threshold value, the processing system 30 determines that the workpiece W used to generate the test spectrum data is a correct workpiece to be polished, whereas when the difference between the reference spectrum data and the test spectrum data is larger than the threshold value, the processing system 30 determines that the workpiece W used to generate the test spectrum data is an incorrect workpiece to be polished.

[0062] FIG. 9 is a flow chart illustrating one embodiment for determining whether a workpiece W is an incorrect workpiece that is not to be polished. In step 201, the processing system 30 determines reference spectral data from a plurality of previously acquired index spectral data. The previously acquired index spectral data are a plurality of spectral data acquired before or after polishing of the correct workpiece to be polished has begun. For example, the previously acquired index spectral data are a plurality of spectral data acquired during a previous water polishing or chemical mechanical polishing of the correct workpiece to be polished. In one embodiment, the reference spectral data is one of the previously acquired index spectral data or an average of the previously acquired index spectral data.

[0063] In step 202, the processing system 30 generates spectral data for inspection of light reflected from the workpiece W immediately before or immediately after polishing of the workpiece W begins. For example, the processing system 30 generates spectral data for inspection of light reflected from the workpiece W during water polishing of the workpiece W or during an early stage of chemical mechanical polishing of the workpiece W. In step 203, the processing system 30 determines the difference between the reference spectral data and the test spectral data. In step 204, processing system 30 determines whether the difference is greater than a threshold value. In step 205, the processing system 30 determines that the workpiece W is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

[0064] In step 206, the processing system 30 sends the judgment result of the workpiece W to the operation control unit 9. In response to the judgment result, the operation control unit 9 either does not allow the polishing apparatus to polish the workpiece W, or stops polishing the workpiece W by the polishing apparatus, or changes the polishing recipe set in the polishing apparatus to a polishing recipe prepared in advance for the workpiece W. In step 207, the processing system 30 polishes the workpiece W according to the polishing recipe set in the polishing apparatus if the difference is less than the threshold value.

[0065] The clustering described with reference to Fig. 8 can also be applied to this embodiment. That is, the processing system 30 may be configured to classify a plurality of previously acquired spectral data into a plurality of groups according to a clustering algorithm, and determine reference spectral data from a plurality of index spectral data belonging to one of the plurality of groups.

[0066] In one embodiment, the processing system 30 may be configured to normalize the plurality of previously acquired index spectral data to create a plurality of normalized index spectral data, and determine the reference spectral data from the plurality of normalized index spectral data. Additionally, the processing system 30 may be configured to normalize the test spectral data to create the normalized test spectral data.

[0067] Fig. 10(a) is a graph showing an example of a plurality of index spectrum data before normalization, and Fig. 10(b) is a graph showing an example of a plurality of index spectrum data after normalization. As can be seen from the comparison between Fig. 10(a) and Fig. 10(b), normalization can reduce the variation in the level of the index spectrum data.

[0068] Fig. 11(a) is a graph showing the results of periodically calculating the difference between reference spectral data determined from multiple index spectral data before normalization and test spectral data before normalization, and Fig. 11(b) is a graph showing the results of periodically calculating the difference between reference spectral data determined from multiple index spectral data after normalization and test spectral data after normalization. In Fig. 11(a) and Fig. 11(b), the vertical axis represents the difference (Euclidean distance) between the reference spectral data and the test spectral data, and the horizontal axis represents time.

[0069] 11(a) and 11(b), the normalization process can reduce the variation of the difference between the reference spectrum data and the test spectrum data, and as a result, the processing system 30 can correctly determine whether the workpiece W is a correct workpiece to be polished.

[0070] In the embodiments described so far, one optical sensor head 25 is provided, but the present invention is not limited to the above embodiments, and multiple optical sensor heads 25 may be provided in the polishing table 3. For example, the multiple optical sensor heads 25 may be arranged so that the multiple optical sensor heads 25 cross different parts of the workpiece W (e.g., a central region and an edge region) as the polishing table 3 rotates.

[0071] The above-described embodiments have been described for the purpose of enabling a person having ordinary skill in the art to practice the present invention. Various modifications of the above-described embodiments are naturally possible for a person skilled in the art, and the technical idea of ​​the present invention can be applied to other embodiments. Therefore, the present invention is not limited to the described embodiments, but is to be interpreted in the broadest scope according to the technical idea defined by the claims. [Explanation of symbols]

[0072] 1 Polishing head 2 Polishing Pads 2a Polished surface 3 Polishing table 5 Liquid supply nozzle 6 Table Motor 9. Operation control section 10 Head shaft 20 Optical film thickness measuring device 22 Light source 25 Optical sensor head 27 Spectrometer 30 Processing System 31 Light-emitting fiber optic cable 32 Receiving optical fiber cable 41 Encoder 42 Decoder W Workpiece

Claims

1. 1. A method for detecting erroneous workpieces not to be polished, comprising: creating inspection spectrum data for reflected light from the workpiece before or after polishing of the workpiece is started; The inspection spectrum data is input to an autoencoder, and the autoencoder is a trained model constructed by machine learning using training data including a plurality of index spectrum data of a plurality of reflected lights from a correct workpiece to be polished; Calculating a difference between the output data output from the autoencoder and the test spectrum data; The method further comprising determining that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

2. The method of claim 1, wherein the step of creating the inspection spectral data is a step of creating inspection spectral data of reflected light from the workpiece during water polishing of the workpiece performed prior to chemical mechanical polishing of the workpiece or during an initial stage of chemical mechanical polishing of the workpiece.

3. Classifying a plurality of pieces of spectrum data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm; creating the training data including a plurality of index spectrum data belonging to one of the plurality of groups; and performing the machine learning using the training data to construct the autoencoder, which is the trained model; The method of claim 1 , wherein the plurality of index spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

4. 1. A method for detecting erroneous workpieces not to be polished, comprising: determining a reference spectrum data from a plurality of index spectrum data of a plurality of reflected lights from a correct workpiece to be polished; creating inspection spectrum data for reflected light from the workpiece before or after polishing of the workpiece is started; determining a difference between the reference spectral data and the test spectral data; The method further comprising determining that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

5. The method of claim 4, wherein the step of creating the inspection spectral data is a step of creating inspection spectral data of reflected light from the workpiece during water polishing of the workpiece performed prior to chemical mechanical polishing of the workpiece or during an initial stage of chemical mechanical polishing of the workpiece.

6. The method of claim 4 , wherein the difference is a Euclidean distance.

7. The step of determining the reference spectral data comprises: Classifying a plurality of pieces of spectrum data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm; determining the reference spectral data from a plurality of index spectral data belonging to one of the plurality of groups; The method of claim 4 , wherein the plurality of index spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

8. normalizing the plurality of index spectral data to generate a plurality of normalized index spectral data; The method of claim 4 , further comprising normalizing the test spectral data to create normalized test spectral data.

9. An optical film thickness measurement apparatus for optically measuring a film thickness of a workpiece, comprising: A light source that emits light; an optical sensor head that irradiates the workpiece with light emitted from the light source and receives reflected light from the workpiece; a processing system for determining a film thickness of the workpiece based on spectral measurement data of light reflected from the workpiece; The processing system includes: creating inspection spectrum data for reflected light from the workpiece before or after polishing of the workpiece is started; The inspection spectrum data is input to an autoencoder, and the autoencoder is a trained model constructed by machine learning using training data including a plurality of index spectrum data of a plurality of reflected lights from a correct workpiece to be polished; Calculating a difference between the output data output from the autoencoder and the test spectrum data; An optical film thickness measurement apparatus configured to determine that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

10. 10. The optical film thickness measurement apparatus of claim 9, wherein the processing system is configured to generate spectral data for inspection of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an initial stage of chemical mechanical polishing of the workpiece.

11. The processing system includes: Classifying a plurality of pieces of spectrum data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm; creating the training data including a plurality of index spectrum data belonging to one of the plurality of groups; The machine learning is performed using the training data to construct the autoencoder, which is the trained model; 10. The optical film thickness measurement device according to claim 9, wherein the plurality of index spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

12. An optical film thickness measurement apparatus for optically measuring a film thickness of a workpiece, comprising: A light source that emits light; an optical sensor head that irradiates the workpiece with light emitted from the light source and receives reflected light from the workpiece; a processing system for determining a film thickness of the workpiece based on spectral measurement data of light reflected from the workpiece; The processing system includes: determining a reference spectrum data from a plurality of index spectrum data of a plurality of reflected lights from a correct workpiece to be polished; creating inspection spectrum data for reflected light from the workpiece before or after polishing of the workpiece is started; determining a difference between the reference spectral data and the test spectral data; An optical film thickness measurement apparatus configured to determine that the workpiece used to create the inspection spectral data is an incorrect workpiece that is not to be polished when the difference is greater than a threshold value.

13. 13. The optical film thickness measurement apparatus of claim 12, wherein the processing system is configured to generate spectral data for inspection of reflected light from the workpiece during water polishing of the workpiece, which is performed prior to chemical mechanical polishing of the workpiece, or during an early stage of chemical mechanical polishing of the workpiece.

14. The optical film thickness measurement device according to claim 12 , wherein the difference is a Euclidean distance.

15. The processing system includes: Classifying a plurality of pieces of spectrum data acquired before or after the start of polishing of a past workpiece into a plurality of groups according to a clustering algorithm; The reference spectral data is determined from a plurality of index spectral data belonging to one of the plurality of groups; 13. The optical film thickness measurement device according to claim 12, wherein the plurality of index spectral data belonging to one of the plurality of groups is a plurality of spectral data of a plurality of reflected lights from a correct workpiece to be polished.

16. The processing system includes: normalizing the plurality of index spectral data to generate a plurality of normalized index spectral data; The optical film thickness measurement apparatus of claim 12 , configured to normalize the test spectral data to create normalized test spectral data.

Citation Information

Patent Citations

  • Polishing device and polishing method

    JP2018194427A