JPWO2020144823A1 - Etching end point detection device, substrate processing system, etching end point detection method and classifier - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an etching end point detecting device or the like which does not require the use of a spectroscope when detecting an etching end point. An etching end point detection device (20) includes a process log acquisition unit (21) for acquiring process log data of a substrate processing device (10) for performing a process such as plasma processing on a substrate (W) arranged in a chamber (1), and a process log acquisition unit. Based on the process log data acquired by the above, input data other than the measured values related to the light generated in the chamber is created, and the detection unit 22 is provided to detect the end point of etching in the substrate processing apparatus based on the input data. The detection unit is a classifier 25 generated by using machine learning, and includes a classifier to which input data is input and which is before or after the end point of etching in the substrate processing apparatus.

Description

The present invention relates to an etching end point detection device for detecting an etching end point in a substrate processing device such as a plasma processing device, a substrate processing system including the etching end point detection device, an etching end point detection method, and a classifier. In particular, the present invention relates to an etching end point detection device, a substrate processing system, an etching end point detection method, and a classifier that do not require the use of a spectroscope when detecting the etching end point.

Conventionally, as a substrate processing device for processing a substrate arranged in a chamber, an etching process for etching the substrate using plasma generated in the chamber, a film forming process for forming a film on the substrate, etc. A plasma processing apparatus that performs plasma processing is known.

When etching a substrate in a plasma processing apparatus that executes an etching process, it is important to detect the end point of etching so that the substrate is not excessively etched.
On the other hand, in a plasma processing apparatus that executes a film forming process, generally, after the substrate that has undergone the film forming process is conveyed to the outside of the chamber, the film composition adhering to the inside of the chamber is removed by cleaning. ing. Specifically, the film composition adhering to the inside of the chamber is removed by etching using plasma. It is important to detect the end point of etching (the end point of cleaning) in order to prevent an excessive supply of the processing gas that generates plasma also for the etching performed during this cleaning.

Conventionally, as an apparatus for detecting the end point of etching on a substrate, an apparatus provided with a spectroscope is known (see, for example, Patent Document 1). Specifically, the conventional etching end point detection device as described in Patent Document 1 guides the light generated in the chamber to a spectroscope installed outside the chamber, and the intensity of the light having a predetermined wavelength in this spectroscope. It is a device that detects the end point of etching with respect to the substrate by measuring. For example, in the apparatus described in Patent Document 1, when the Si substrate is etched using SF ₆ gas as the processing gas, the intensity of light having the emission wavelength of SiF, which is a reaction product of Si, becomes equal to or less than the reference value. The time point is detected as the end point of etching.

Similar to the above, a device equipped with a spectroscope is also used when detecting the end point of etching of the film composition adhered to the chamber. For example, when the film composition is etched using C ₄ F ₈ gas as the processing gas, the time when the intensity of the light having the emission wavelength of F becomes equal to or higher than the reference value is detected as the end point of the etching.

As described above, the conventional etching end point detection device has a configuration in which a spectroscope is always required to detect the end point of etching, and a spectroscope is provided for each chamber of the plasma processing device and is always used when detecting the end point. I needed it. For this reason, there is a problem that the manufacturing cost and the labor for maintenance increase.

Japanese Patent No. 4101280

The present invention has been made to solve the above-mentioned problems of the prior art, and is an etching end point detection device that does not require the use of a spectroscope when detecting the end point of etching, a substrate processing system provided with this, and etching. An object of the present invention is to provide an end point detection method and a classifier.

In order to solve the above problems, the present inventor has diligently studied and focused on using the process log data of the substrate processing apparatus. The process log data is a history of various measured values and set values when various processes are executed in the board processing device, and is sequentially obtained during the operation of a general board processing device. This is because I thought that there might be something whose value changes before and after the end point of etching. However, in order to detect the end point of etching based on the magnitude of the values of various types of process log data, it is necessary to study complicated detection logic and complicated adjustment of parameters such as threshold values, which requires enormous labor. .. Therefore, as a result of examining the application of machine learning to the process log data, the present inventor has found that the end point of etching can be detected accurately without any trouble.
The present invention has been completed based on the above findings of the present inventor.

That is, in order to solve the above problems, the present invention has a process log acquisition unit that acquires process log data of a substrate processing apparatus that processes a substrate arranged in a chamber, and a process log acquired by the process log acquisition unit. Based on the data, input data other than the measured values related to the light generated in the chamber is created, and the detection unit includes a detection unit that detects the end point of etching in the substrate processing apparatus based on the input data. Is equipped with a classifier generated by using machine learning, in which the input data is input and which is before or after the end point of etching in the substrate processing device. I will provide a.

In the etching end point detection device according to the present invention, the process log acquisition unit acquires the process log data of the substrate processing device, and the detection unit obtains the light generated in the chamber from the process log data (processing gas supplied into the chamber). Input data other than the measured values related to the measured values (for example, the measured values of light intensity using a conventional spectroscope are excluded) are created.
Then, the detection unit is provided with a classifier generated by using machine learning, and by inputting input data to this classifier, the classifier determines whether it is before or after the end point of etching in the substrate processing apparatus. It is configured to output. Therefore, in the process of executing the etching process, the input data created from the sequentially acquired process log data is input to the classifier of the detection unit, so that the input data is before or after the end point of etching. Is output by the classifier, which makes it possible to detect the end point of etching.
As described above, according to the etching end point detection device according to the present invention, when detecting the end point of etching, input data other than the measured value related to the light generated in the chamber is created based on the process log data, and this input is obtained. The end point of etching can be detected only by using the data. That is, it is not necessary to use a spectroscope when detecting the end point of etching.

The substrate processing apparatus to which the etching end point detecting apparatus according to the present invention is applied is not limited to the plasma processing apparatus. For example, it can be applied to a sacrificial layer etching device using anhydrous HF gas and alcohol, which could only be time-etched (etching performed for a predetermined fixed time), a sacrificial layer etching device using XeF ₂ gas, and the like. Is.
Further, as the classifier, various configurations such as a neural network and a support vector machine can be adopted as long as they can be generated by using machine learning.

Preferably, the classifier outputs that it is before the end point of the etching when the input data created from the process log data acquired before the end point of the etching is given as the input of the teacher data. , Machine learning to output that it is after the end point of the etching when the input data created from the process log data including the one acquired after the end point of the etching is given as the input of the teacher data. Is generated using.

According to the preferred configuration described above, the classifier was given input data created from process log data acquired before the end of etching as input of teacher data (a combination of known inputs and outputs to the classifier). In the case, to output that it is before the end point of etching (specifically, a numerical value indicating that it is before the end point of etching, for example, "0") (that is, as an output of teacher data, for example, " Given "0"), it is generated using machine learning. Further, the classifier shall be after the end point of etching when the input data created from the process log data including the data acquired after the end point of etching is given as the input of the teacher data (specifically, the classifier is after the end point of etching. It is generated using machine learning to output a numerical value indicating that it is after the end point of etching, eg, "1") (ie, giving, for example, "1" as the output of the teacher data). The latter input of teacher data, "input data created from process log data including those acquired after the end point of etching" is input data created only from process log data acquired after the end point of etching. It also means that it may be input data created from process log data straddling before and after the end point of etching.
By performing machine learning using the above-mentioned teacher data to generate a classifier, there is no need to study complicated detection logic or complicated adjustment of parameters such as threshold values, and classification after machine learning. By simply inputting the input data created from the process log data into the container, it is possible to easily detect whether it is before or after the end point of etching.

In the above preferred configuration, it is determined, for example, whether the process log data that is the basis of the input data used as the teacher data is acquired before the end point of etching or after the end point of etching. As in the conventional case, the determination may be made by measuring the intensity of light having a predetermined wavelength with a spectroscope. That is, the end point of etching is detected using a spectroscope, and the end point of etching detected by this spectroscope is set to be true, and the process log data is acquired before the true end point, or the true end point. It suffices to determine whether it was acquired later.
Specifically, for example, when the substrate processing device itself for detecting the end point of etching in the etching end point detection device according to the present invention is provided with a spectroscope, the spectroscope is used as a basis for input data used as teacher data. It suffices to determine whether the process log data to be obtained is before or after the end point of etching. Further, for example, when the substrate processing apparatus for detecting the etching end point in the etching end point detecting apparatus according to the present invention does not have a spectroscope, it is classified by machine learning using another substrate processing apparatus equipped with a spectroscope. It is also possible to generate a device and use this classifier for the etching end point detection device according to the present invention.
Further, the determination is not necessarily limited to the case of using a spectroscope. For example, when acquiring teacher data, the surface of the substrate being etched is observed, and the color difference of the substrate surface before and after the end point of etching is used. It is also conceivable to determine whether the process log data is before or after the end point of etching.

In the etching end point detection device according to the present invention, when the substrate processing device is an etching device that etches the substrate, the etching end point detected by the detection unit is the end point of etching applied to the substrate.

Further, in the etching end point detection device according to the present invention, when the substrate processing device is a film forming device that forms a film on the substrate, the etching end point detected by the detection unit forms a film on the substrate. After that, it is set as the end point of the etching performed to remove the film composition adhering to the inside of the chamber.

The etching end point detection device according to the present invention is preferably used when the substrate processing device is a plasma processing device that applies plasma treatment to the substrate.

According to the results of diligent studies by the present inventors, when the substrate processing device is a plasma processing device, among various process log data, the pressure in the exhaust pipe, the valve opening degree of the automatic pressure control device, and the upper part. The matching position of the matching unit and the matching position of the lower matching unit are particularly liable to change before and after the end point of etching. This is because when the etching is completed, the object to be etched (the substrate or the film composition) disappears, so that the state of the plasma changes. Therefore, it is preferable to use at least these process log data to detect the end point of etching. However, since the pressure in the exhaust pipe and the valve opening degree of the automatic pressure control device change in conjunction with each other, it is considered that only one of them may be used.

That is, the substrate processing apparatus is a plasma processing apparatus that applies a plasma processing apparatus to the substrate placed on a mounting table arranged in the chamber, and the substrate processing apparatus surrounds the chamber. An upper high-frequency power source that applies high-frequency power to the coil or the upper electrode via an upper matching unit, and an upper high-frequency power source that is arranged in the chamber in parallel with the above-mentioned pedestal. A lower high-frequency power source that applies high-frequency power via a lower matching unit, an exhaust pipe that communicates with the chamber, and an exhaust pipe that is provided in the exhaust pipe to control the pressure in the chamber by adjusting the valve opening degree. When the automatic pressure control device is provided, at least the pressure in the exhaust pipe or the valve opening degree of the automatic pressure control device is matched with the upper matching unit and / or the lower matching unit in the process log data. The location and is preferably included.

In the etching end point detection apparatus according to the present invention, as the input data created from the process log data, the process log data can be used as it is without being processed, or the processed process log data can be used. It is possible. For example, when the classifier is a neural network, the advantage of the neural network in image recognition is utilized, and as an example of the latter, there is a case where image data obtained by processing process log data is used as input data. ..
Specifically, in the detection unit, based on the process log data acquired by the process log acquisition unit, one axis is the type of the process log data, and the other axis orthogonal to the one axis is the value of the process log data. It is conceivable to create image data in which a graph is imaged and use the image data as input data to the classifier.

In the etching end point detection device according to the present invention, preferably, the detection unit has a maximum value of 1 and a minimum value of 0 for each type of process log data with respect to the process log data acquired by the process log acquisition unit. Is performed, and input data to the classifier is created based on the process log data after the normalization.

The value of the process log data varies greatly depending on the type of process log data such as pressure, temperature, and flow rate. In addition, the value will differ depending on the unit used. Therefore, if the values of the process log data of each type are used as they are when detecting the end point of etching, the detection accuracy may be affected. In order to avoid this, it is preferable to normalize the values of each type of process log data so that they all fluctuate within a certain range.
Specifically, the detection unit normalizes the process log data acquired by the process log acquisition unit so that the maximum value becomes 1 and the minimum value becomes 0 for each type of the process log data. It is preferable to create input data to the classifier based on the process log data after normalization.

According to the above preferable configuration, the value of the process log data after normalization fluctuates within a certain range of 0 to 1 regardless of the type of process log data. Therefore, the input data created based on this is used. By using it as a classifier, it can be expected that a decrease in detection accuracy can be avoided.

Further, in order to solve the above-mentioned problems, the present invention is characterized by comprising a substrate processing apparatus for processing a substrate arranged in a chamber and an etching end point detecting apparatus according to any one of the above. It is also provided as a processing system.

Further, in order to solve the above problems, the present invention has a process log acquisition step of acquiring process log data of a substrate processing apparatus for processing a substrate arranged in a chamber, and a process log acquired by the process log acquisition step. The detection step includes a detection step of creating input data other than the measured values related to the light generated in the chamber based on the data and detecting the end point of etching in the substrate processing apparatus based on the input data. Then, using the classifier generated by using machine learning, the input data is input to the classifier, and the classifier outputs whether it is before or after the end point of etching in the substrate processing apparatus. It is also provided as an etching end point detection method characterized by.

Further, in order to solve the above-mentioned problems, the present invention has a value other than the measured values related to the light generated in the chamber, which is created based on the process log data of the substrate processing apparatus for processing the substrate arranged in the chamber. It is also provided as a classifier generated by using machine learning, in which input data is input and which is before or after the end point of etching in the substrate processing apparatus.

According to the present invention, it is not necessary to use a spectroscope when detecting the end point of etching.

It is a figure which shows typically the schematic structure of the substrate processing system which concerns on 1st Embodiment of this invention. It is explanatory drawing explaining the operation of the normalization part and the imaging part shown in FIG. It is a figure which shows typically the schematic structure and operation of the classifier shown in FIG. The results of the test using the substrate processing system shown in FIG. 1 are shown. It is a figure which shows typically the schematic structure of the substrate processing system which concerns on 2nd Embodiment of this invention. The results of the test using the substrate processing system shown in FIG. 5 are shown.

Hereinafter, the etching end point detection device according to the embodiment of the present invention and the substrate processing system provided with the etching end point detection device will be described with reference to the accompanying drawings.

<First Embodiment>
FIG. 1 is a diagram schematically showing a schematic configuration of a substrate processing system according to the first embodiment. FIG. 1A is an overall configuration diagram of a substrate processing system, and FIG. 1B is a block diagram showing a schematic configuration of an etching end point detection device. In FIG. 1A, the parameters to be measured are shown surrounded by a broken line rectangle.
As shown in FIG. 1A, the substrate processing system 100 according to the first embodiment includes a substrate processing apparatus 10 and an etching end point detecting apparatus 20.

The substrate processing apparatus 10 of the first embodiment is an apparatus including a chamber 1 and a mounting table 2 arranged in the chamber 1 and performing plasma processing on the substrate W mounted on the mounting table 2. More specifically, the substrate processing apparatus 10 of the first embodiment is an inductively coupled plasma (ICP) type plasma etching apparatus that etches the substrate W as plasma processing.

A processing gas for generating plasma is supplied from a gas supply source (not shown) into the chamber 1 of the substrate processing apparatus 10. In FIG. 1A, the gas No. 1-Gas No. The configuration which can supply 6 kinds of processing gases up to 6 is illustrated. However, when the etching process is performed, the etching process is not limited to the case where all six types of processing gases are used, and the etching process can be performed using any one or more types of processing gases. The flow rate of each processed gas to be supplied is measured by a mass flow controller (MFC) 11 provided in the flow path from the gas supply source to the chamber 1. Further, the chamber 1 is provided with heaters (not shown) for heating the wall surface of the chamber 1 at appropriate locations, and the temperature of the heaters at each location (temperature No. 1-1 shown in FIG. 1A). ~ No. 1-4) is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the pressure in the chamber 1 is measured by the vacuum gauge 12.

The substrate processing apparatus 10 includes a coil 3 arranged in the chamber 1 so as to surround the chamber 1 (in FIG. 1A, for convenience, only a cross section of the coil 3 located on the left side is shown). High-frequency power (upper high-frequency power) is applied to the coil 3 from the upper high-frequency power supply 4 via the upper matching unit 5. By applying the upper high frequency power to the coil 3, the processing gas supplied into the chamber 1 is turned into plasma. The upper high-frequency power applied by the upper high-frequency power supply 4 and the matching position of the upper matching unit 5 (constants such as variable capacitors and variable coils included in the upper matching unit 5) are known measuring instruments (not shown). ).

High-frequency power (lower high-frequency power) is applied to the mounting table 2 from the lower high-frequency power supply 6 via the lower matching unit 7. By applying the lower high-frequency power to the mounting table 2, a bias potential is applied between the mounting table 2 and the plasma in the chamber 1, and the ions in the plasma are accelerated to the substrate W mounted on the mounting table 2. Pull in. As a result, the substrate W is etched. The lower high-frequency power applied by the lower high-frequency power supply 6 and the matching position of the lower matching unit 7 (constants such as variable capacitors and variable coils included in the lower matching unit 7) are known measuring instruments (not shown). ).

During the execution of the plasma treatment, the mounting table 2 is cooled by the chiller 8. The temperature of the chiller 8 is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, during the execution of the plasma treatment, He gas is supplied to the back surface of the substrate W, and the substrate W is cooled by this He gas. At this time, the pressure / flow rate of the supplied He gas is measured by a pressure / flow meter 9 provided in the flow path from the He gas supply source (not shown) to the back surface of the substrate W (upper surface of the mounting table 2). To.

The reaction products and the like generated in the chamber 1 by executing the plasma treatment are exhausted to the outside of the chamber 1 through the exhaust pipe 17 communicating with the chamber 1. The exhaust pipe 17 includes an automatic pressure controller (APC) 13 that controls the pressure in the chamber 1 by adjusting the valve opening degree, and a first pump (turbo molecule) for exhausting the reaction product. A pump) 14 and a second pump (dry pump, rotary pump, etc.) 15 that assists the first pump 14 are provided. The temperature of the automatic pressure control device 13 (temperature No. 1-5 shown in FIG. 1A) and the temperature of the first pump 14 (temperature No. 1-6 shown in FIG. 1A) are different. It is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the exhaust pipe 17 is provided with heaters (not shown) for heating the exhaust pipe 17 at appropriate locations (for example, between the first pump 14 and the second pump 15), and heaters at each location. (Temperatures No. 1-7 and No. 1-8 shown in FIG. 1A) are measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the valve opening degree (APC opening degree) of the automatic pressure control device 13 is measured by a known measuring instrument (not shown) such as an encoder. Further, the pressure (foreline pressure) in the exhaust pipe 17 located between the first pump 14 and the second pump 15 is measured by the vacuum gauge 16.

The etching end point detection device 20 is a device that is electrically connected to the substrate processing device 10 having the above configuration and detects the end point of etching applied to the substrate W in the substrate processing device 10.

As shown in FIG. 1B, the etching end point detection device 20 includes a process log acquisition unit 21 and a detection unit 22, and is composed of, for example, a computer.
The process log acquisition unit 21 is electrically connected to a measuring instrument (for example, a mass flow controller 11) that measures each of the above-mentioned measured values with reference to FIG. 1 (a) by wire or wirelessly (FIG. 1 (a). ), For convenience, shows a state in which only the pressure / flow meter 9, the mass flow controller 11 and the vacuum gauge 12 are connected by wire), and the measurement data sequentially input from each measuring instrument is subjected to a predetermined sampling cycle (). For example, it has a function of acquiring (A / D conversion) in 1 second). The process log acquisition unit 21 is stored in, for example, an A / D conversion board mounted on the computer, a memory such as a ROM or RAM included in the computer, or stored in the memory, and causes the CPU to operate as the process log acquisition unit 21. It consists of a program to be executed. Each acquired measured value and each set value corresponding to each measured value are used as process log data for detecting the end point of etching by the detection unit 22.
In the first embodiment, all the measured values shown in FIG. 1A are used as process log data for detecting the end point of etching, but the present invention is not limited to this. However, it is preferable to use at least the foreline pressure or the APC opening degree and the matching position of the upper matching unit 5 and / or the lower matching unit 7.

When the etching end point detection device 20 also has a function as a control device generally used for controlling the operation of the substrate processing device 10 (when the control device is also used as the etching end point detection device 20). Each setting value constituting the process log data is stored in advance in the etching end point detection device 20 (process log acquisition unit 21). When the etching end point detection device 20 is separate from the above control device and both are electrically connected, each set value stored in advance in the control device is the etching end point detection device 20 (process log acquisition unit). It will be transmitted to 21). Further, in the first embodiment, the case where the etching end point detection device 20 is directly connected to each measuring device is illustrated, but the above control device and each measuring device are directly connected, and each measurement acquired by the control device is illustrated. It is also possible to adopt a configuration in which the value is transmitted to the etching end point detection device 20.

The detection unit 22 creates input data from the process log data sequentially (for example, every second) acquired by the process log acquisition unit 21, and detects the end point of etching in the substrate processing apparatus 10 based on the input data. Is. The detection unit 22 is composed of, for example, a memory such as a ROM or RAM provided in the computer, or a program stored in the memory and causing the CPU to execute the operation as the detection unit 22.

The detection unit 22 includes a classifier 25. The detection unit 22 of the first embodiment further includes a normalization unit 23 and an imaging unit 24 as a preferable configuration. The normalization unit 23, the imaging unit 24, and the classifier 25 are also stored in, for example, a memory such as a ROM or RAM provided in the computer, or a program stored in the memory and causing the CPU to execute operations as the respective units 23 to 25. It is composed.

FIG. 2 is an explanatory diagram illustrating the operation of the normalization unit 23 and the imaging unit 24. FIG. 2A is a diagram for explaining the operation of the normalization unit 23, and FIGS. 2B and 2C are diagrams for explaining the operation of the imaging unit 24.
The left figure of FIG. 2A is a diagram schematically showing the process log data acquired by the process log acquisition unit 21. The parameters 1 to N shown in FIG. 2A are, for example, the gas No. 1 in which the parameter 1 is measured by the mass flow controller 11 shown in FIG. The flow rate of No. 1 and the parameter N is the temperature No. 1 shown in FIG. 1 (a). It means the type of process log data such as 1-8. X _ij (i = 1 to N, j = 1 to M) shown in the left figure of FIG. 2 (a) is acquired for parameter i when the process time (elapsed time from the start of etching) is j [sec]. It means the value of the process log data. For example, X ₁₁ is the value of the process log data acquired when the process time is 1 [sec] for parameter 1, and X _NM is acquired when the process time is M [sec] for parameter N. The value of the process log data.

The normalization unit 23 calculates in advance the maximum value MAX _i and the minimum value MIN _i of the process log data in the total process time (1 to M [sec]) for each type of process log data (for each parameter i). For example, the maximum value for parameter 1 is MAX ₁ , the minimum value is MIN ₁ , the maximum value for parameter N is MAX _N , and the minimum value is MIN _N. The maximum value MAX _i and the minimum value MIN _i are not calculated using the process log data acquired when etching one substrate W, but are not calculated using the process log data acquired when etching one substrate W, but are performed during learning of the classifier 25 described later. It is preferable to calculate in advance using the process log data acquired for a plurality of substrates W etched by the same recipe (plasma treatment conditions). The maximum value MAX _i and the minimum value MIN _{i for} each type (parameter i) of the calculated process log data are stored in the normalization unit 23.

Then, the normalization unit 23 normalizes the process log data X _ij sequentially acquired by the process log acquisition unit 21 so that the maximum value is 1 and the minimum value is 0 for each type of process log data (for each parameter i). Perform the conversion.
Specifically, based on the following equation (1), the normalized process log data _Yij is calculated as shown in the right figure of FIG.
Y _ij = (X _ij- MIN _i ) / (MAX _i- MIN _i ) ... (1)
In the above formula (1), i = 1 to M and j = 1 to N.
From the above equation (1), it is clear that when X _ij = MAX _i , Y _ij = 1 and when X _ij = MIN _i , Y _ij = 0 is normalized.

The imaging unit 24 creates input data to the classifier 25 based on the process log data after normalization.
Specifically, the imaging unit 24 is uniaxial (FIG. 2 (b), as shown in the right figure of FIG. 2 (b), based on the process log data after normalization shown in the left figure of FIG. 2 (b). ) Is the type of process log data (parameters 1 to N) in the example shown on the right, and the other axis (vertical axis in the example shown on the right in FIG. 2B) is normal. Image data that is an image of a graph (bar graph) that is the value _Yij of the process log data after conversion is sequentially created (for example, every second).
The type of image data is not limited to the monochrome shading image as shown in the right figure of FIG. 2B, and any image data such as a binarized image or a color image can be created.

Next, the imaging unit 24 of the first embodiment divides the created image data into a predetermined pixel region composed of a plurality of pixels.
Specifically, as shown in FIG. 2C, the imaging unit 24 divides the image data into K in each of the uniaxial (horizontal axis) direction and the other axis (vertical axis) direction, and the pixel area _Aij (i). = 1 to K, j = 1 to K). Then, the imaging unit 24 (the average value of the density values of a plurality of pixels constituting the pixel region _{A ij)} the average density value for each pixel region _{A ij I ave (A ij)} (i = 1~K, j = 1 to K) are calculated. This average concentration value I _ave (A _ij ) is used as input data to the classifier 25.

When the imaging unit 24 creates a color image (color image of three RGB colors), the imaging unit 24 calculates an average density value for each color image, and all of them are input data to the classifier 25. Used.
Further, in the first embodiment, an example in which the imaging unit 24 divides the image data into pixel regions is shown, but the present invention is not limited to this, and the density values of each pixel constituting the image data are classified as they are. It can also be used as input data to the vessel 25.

FIG. 3 is a diagram schematically showing a schematic configuration and operation of the classifier 25.
As shown in FIG. 3, the classifier 25 of the first embodiment is composed of a neural network having an input layer, an intermediate layer, and an output layer. Although FIG. 3 illustrates a configuration having two intermediate layers, the neural network that can be used as the classifier of the present invention is not limited to this, and a configuration having an arbitrary number of intermediate layers is used. It can be adopted. Further, the number of nodes in each layer shown in FIG. 3 (the portion indicated by “◯” in FIG. 3) is merely an example, and the number of nodes in the neural network that can be used as the classifier of the present invention is limited to the one shown in the figure. Absent.

The classifier 25 is a substrate when the image data created by the imaging unit 24 as input data (specifically, the average density value I _ave (A _ij ) of each pixel area A _ij ) is input to the input layer. It is a configuration generated by machine learning so as to output from the output layer (output the output value OUT) which is before or after the end point of etching in the processing apparatus 10.

Specifically, at the time of learning the classifier 25, input data (image data) created from process log data acquired before the end point of etching is given as input of teacher data, and output of teacher data combined with the input is given. As a result, machine learning is given so that OUT = 0 is output from the output layer when the input is input to the input layer by giving that it is before the end point of etching (OUT = 0 in the first embodiment). I do.
Further, as the input of the teacher data, the input data (image data) created from the process log data acquired after the end point of etching is given, and the output of the teacher data combined with the input is after the end point of etching (No. 1). In one embodiment, when OUT = 1) is given and the input is input to the input layer, machine learning is performed so that OUT = 1 is output from the output layer.

When the substrate processing device 10 is provided with a spectroscope, the teacher data may be acquired using the spectroscope, and when the substrate processing device 10 is not provided with the spectroscope, the spectroscope is provided. Teacher data may be acquired using the substrate processing apparatus of. Also, the machine learning of the classifier 25 is not limited to one time. If necessary, the classifier 25 can be re-learned using the new teacher data, or the classifier 25 can be re-learned by adding new teacher data to the conventional teacher data.

When the classifier 25 after learning as described above detects the end point of etching based on the input data sequentially input, the input data (image data) is sequentially input to the input layer of the classifier 25 and classified. The output value OUT is output from the output layer of the device 25. Unlike the learning time, the output value OUT value at the time of detection is 0 ≦ OUT ≦ 1.
When 0 ≦ OUT <0.5 (when rounded to 0 with one decimal point), the detection unit 22 of the first embodiment determines that it is before the end point of etching, and 0.5 ≦ OUT. When ≦ 1 (when rounded to 1 with one decimal point), it is determined that it is after the end point of etching.

The etching end point detection device 20 having the configuration described above sequentially detects the end points of etching applied to the substrate W in the substrate processing device 10.

Hereinafter, an example of the result of performing a test in which the substrate W is etched by the substrate processing apparatus 10 of the substrate processing system 100 according to the first embodiment and the end point of etching is detected by the etching end point detecting device 20 will be described.

In the above test, first, 19 substrates W (Si substrates) are etched with SF ₆ gas, and the etching time of each substrate W (from the start of etching of the substrate W to the end of etching to the end of overetching). Input data (image data) to the classifier 25 was created for each sampling cycle of 1 second (50 seconds). Whether the input data of each sampling cycle is the input data created from the process log data acquired before the end point of etching or the input data created from the process log data acquired after the end point of etching is described above. Since the substrate processing apparatus 10 used in the test is provided with a spectroscope, the determination was made based on whether or not the intensity of light having the emission wavelength of SiF measured by this spectroscope was equal to or less than the reference value. Using the teacher data collected as described above, the classifier 25 was machine-learned. When the same teacher data input data was input to the classifier 25 after learning and it was determined whether it was before or after the end point of etching, the correct answer rate (number of correct answers / number of judgments x 100) was 99.89%. It was. Since the time required for one determination by the classifier 25 was sufficiently shorter than the overetching time, the etching end point detection device 20 detected the etching end point (determined to be after the etching end point). It can be said that there is no problem even if the etching is completed.

Next, in the above test, another 6 substrates (No. 1-1 to No. 1-6) of substrates W (Si substrates) were etched with SF ₆ gas using the same recipe, and the etching time of each substrate W was In, input data (image data) to the classifier 25 was created for each sampling cycle of 1 second, and the data was input to the classifier 25 after the learning to determine whether it was before or after the end point of etching. At this time, as in the case of the above-mentioned learning, using the spectroscope provided in the substrate processing apparatus 10, the input data of each sampling period is before the end point of etching or after the end point of etching. It was judged whether there was.

FIG. 4 shows the results of the above test. “0” shown in FIG. 4 is determined by the etching end point detecting device 20 to be before the etching end point, and “1” is determined by the etching end point detecting device 20 to be after the etching end point. is there. In FIG. 4, the columns that have been hatched and surrounded by thick lines are determined to be after the end point of etching by using a spectroscope.
As shown in FIG. 4, it was No. 4 that made a judgment different from the judgment using the spectroscope. Judgment of the substrate W of 1-3 at 38 seconds and No. Only the judgment of the substrate W of 1-5 at 38 seconds was made, and the correct answer rate was 99.35% (= 306/308 × 100). Therefore, according to the etching end point detecting device 20 of the substrate processing system 100 according to the first embodiment, it can be said that the etching end point of the substrate W can be detected with high accuracy.

<Second Embodiment>
FIG. 5 is a diagram schematically showing a schematic configuration of a substrate processing system according to a second embodiment. In FIG. 5, as in FIG. 1 (a) described above, the parameters to be measured are shown surrounded by a broken line rectangle. In FIG. 5, the illustration of the configuration corresponding to FIG. 1 (b) described above is omitted.
As shown in FIG. 5, the substrate processing system 200 according to the second embodiment includes a substrate processing device 10A and an etching end point detecting device 20.
Hereinafter, the differences from the substrate processing system 100 according to the first embodiment will be mainly described, and the same components as those of the substrate processing system 100 according to the first embodiment are designated by the same reference numerals and the description thereof will be omitted.

The substrate processing apparatus 10A of the second embodiment is an apparatus including a chamber 1 and a mounting table 2 arranged in the chamber 1 and performing plasma processing on the substrate W mounted on the mounting table 2. More specifically, the substrate processing apparatus 10A of the second embodiment is a capacitively coupled plasma (CCP) type plasma film forming apparatus that forms a film on the substrate W as plasma processing.
Therefore, unlike the substrate processing apparatus 10 of the first embodiment, instead of the coil 3 (see FIG. 1A), an upper electrode 18 is provided in the chamber 1 so as to face parallel to the mounting table 2. ..

A processing gas for generating plasma is supplied from a gas supply source (not shown) into the chamber 1 of the substrate processing apparatus 10A. In FIG. 5, the gas No. 1-Gas No. The configuration which can supply 6 kinds of processing gases up to 6 is illustrated. However, this is not limited to the case where all of the six types of processing gases are used when performing the film forming process or when cleaning the film composition adhering to the chamber 1 after the film forming process, and any one of them is used. It is possible to perform film formation processing and cleaning using more than one type of processing gas.

High-frequency power (upper high-frequency power) is applied to the upper electrode 18 from the upper high-frequency power supply 4 via the upper matching unit 5. Further, high frequency power (lower high frequency power) is applied to the mounting table 2 from the lower high frequency power supply 6 via the lower matching unit 7. As a result, the processing gas supplied into the chamber 1 is turned into plasma, and the generated plasma moves toward the mounting table 2, so that a film is formed on the substrate W mounted on the mounting table 2. .. When the film composition is cleaned after the film forming process, the generated plasma moves toward the inner surface of the chamber 1, so that the film composition adhering to the inside of the chamber 1 is removed by etching.

Unlike the substrate processing apparatus 10 of the first embodiment, the substrate processing apparatus 10A of the second embodiment does not include a chiller 8, a pressure / flow meter 9, a first pump (turbo molecular pump) 14, and a vacuum gauge 16.

In the substrate processing system 200 according to the second embodiment, the flow rate of each processing gas to be supplied is measured by a mass flow controller 11 provided in the flow path from the gas supply source to the chamber 1. Further, the temperature of the heater (not shown) provided at an appropriate position on the wall surface of the chamber 1 (temperatures No. 2-1 to No. 2-3 shown in FIG. 5) is a known measuring instrument such as a thermoelectric pair (not shown). (Not shown). Further, the pressure in the chamber 1 is measured by the vacuum gauge 12.
Further, the upper high-frequency power applied by the upper high-frequency power supply 4 and the matching position of the upper matching unit 5 (constants such as variable capacitors and variable coils included in the upper matching unit 5) are known measuring instruments (not shown). ).
Further, the lower high-frequency power applied by the lower high-frequency power supply 6 and the matching position of the lower matching unit 7 (constants such as variable capacitors and variable coils included in the lower matching unit 7) are known measuring instruments (not shown). ).
Further, the temperature of the automatic pressure control device 13 (temperature No. 2-5 shown in FIG. 5) and the temperature of a heater (not shown) provided at an appropriate position in the exhaust pipe 17 (temperature No. 2 shown in FIG. 5). 2-4, No. 2-6, No. 2-7) are measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the APC opening degree of the automatic pressure control device 13 is measured by a known measuring instrument (not shown) such as an encoder.

The etching end point detection device 20 included in the substrate processing system 200 according to the second embodiment has the same configuration as that of the first embodiment, and is a measuring instrument (for example, for example) that measures each of the above-mentioned measured values with reference to FIG. It is electrically connected to the mass flow controller 11) by wire or wirelessly. The etching end point detection device 20 acquires measurement data sequentially input from each measuring instrument at a predetermined sampling cycle (for example, 1 second), and each acquired measurement value and each set value corresponding to each measurement value are set. , Used as process log data to detect the end point of etching.
In the second embodiment, all the measured values shown in FIG. 5 are used for determination as process log data, but the present invention is not limited to this. However, it is preferable to use at least the APC opening degree and the matching position of the upper matching unit 5 and / or the lower matching unit 7.

Unlike the first embodiment, the etching end point detected by the etching end point detecting device 20 of the second embodiment is executed to remove the film composition adhering to the chamber 1 after forming the film on the substrate W. This is the end point of etching.

Hereinafter, during etching (cleaning) performed to remove the film composition adhering to the inside of the chamber 1 after forming a film on the substrate W by the substrate processing apparatus 10A of the substrate processing system 200 according to the second embodiment. , An example of the result of the test of detecting the end point of etching by the etching end point detection device 20 will be described.

In the above test, firstly, C ₄ F ₈ and 13 cleaning times in the chamber 1 using gas, over-etching is completed through the etching end from the etching start of coating composition adhered to the cleaning of the etching time (chamber 1 Input data (image data) to the classifier 25 was created for each sampling cycle of 1 second (about 150 seconds until the etching). Whether the input data of each sampling cycle is the input data created from the process log data acquired before the end point of etching or the input data created from the process log data acquired after the end point of etching is described above. Since the substrate processing apparatus 10A used in the test is provided with a spectroscope, it was determined by whether or not the intensity of light having an emission wavelength of F measured by this spectroscope was equal to or higher than the reference value. Using the teacher data collected as described above, the classifier 25 was machine-learned. When the input data of the same teacher data was input to the classifier 25 after learning and a judgment was made, the correct answer rate was 99.39%. Since the time required for one determination by the classifier 25 was sufficiently shorter than the overetching time, the etching end point detection device 20 detected the etching end point (determined to be after the etching end point). It can be said that there is no problem even if the etching is completed after that.

Next, in the above test, the inside of the chamber 1 was cleaned with the same recipe 5 times (No. 2-1 to No. 2-5) C ₄ F ₈ gas at different timings, and at the etching time of each cleaning. Input data (image data) to the classifier 25 was created for each sampling cycle of 1 second, and the data was input to the classifier 25 after the learning to determine whether it was before or after the end point of etching. At this time, as in the case of the above-mentioned learning, using the spectroscope provided in the substrate processing apparatus 10A, the input data of each sampling cycle is before the end point of etching or after the end point of etching. It was judged whether there was.

FIG. 6 shows the results of the above test. “0” shown in FIG. 6 is determined by the etching end point detecting device 20 to be before the etching end point, and “1” is determined by the etching end point detecting device 20 to be after the etching end point. is there. In FIG. 6, the diagonally hatched columns and the columns surrounded by thick lines are determined by using a spectroscope after the end point of etching.
As shown in FIG. 6, it was No. 6 that made a judgment different from the judgment using the spectroscope. Judgment at the time of 35 to 39 seconds of cleaning of 2-1 and No. Judgment at 3 seconds of cleaning 2-3 and No. Only the judgment at 44 seconds of 2-5 cleaning was made, and the correct answer rate was 99.1% (= 743/750 × 100). Therefore, according to the etching end point detection device 20 of the substrate processing system 200 according to the second embodiment, it can be said that the end point of etching (cleaning) of the film composition adhering to the chamber 1 can be detected with high accuracy.

In the first embodiment and the second embodiment described above, the configuration in which the detection unit 22 includes the normalization unit 23 has been described as an example, but the present invention is not limited to this. It is also possible that the detection unit 22 does not include the normalization unit 23, and the process log data acquired by the process log acquisition unit 21 is used as it is without normalization to create input data.

Further, in the first embodiment and the second embodiment, the configuration in which the detection unit 22 includes the imaging unit 24 has been described as an example, but the present invention is not limited to this. The detection unit 22 does not include the imaging unit 24, and the process log data acquired by the process log acquisition unit 21 can be used as input data to the classifier 25 without being imaged after being used as it is or after being normalized. is there.
Specifically, when not imaging, for example, the process log data X _ij (i = 1 to N, j = 1 to M) shown in FIG. 2 or after normalization is applied to the input layer of the classifier 25 shown in FIG. Process log data Y _ij (i = 1 to N, j = 1 to M) will be input.

Further, in the first embodiment and the second embodiment, the input data (image data) created in a predetermined sampling cycle (for example, 1 second) is used as the input data to the classifier 25 during both learning and determination. However, the present invention is not limited to this. For example, a graph showing changes in process log data acquired in a predetermined sampling cycle within a predetermined time (within a time corresponding to a plurality of sampling cycles) is imaged, and this image data is used as input data to the classifier 25. It can also be used. Any of the above image data will include a graph created from process log data that spans before and after the end point of etching.

Further, in the first embodiment and the second embodiment, the configuration in which the detection unit 22 includes a single classifier 25 has been described as an example, but the present invention is not limited thereto. A plurality of classifiers 25 are generated by machine learning for each equivalent recipe, and when the end point of etching is detected, the end point is detected by using the classifier 25 generated according to the corresponding recipe among the plurality of classifiers 25. It is also possible to adopt a configuration that does.

Further, according to the etching end point detection device 20 of the first embodiment and the second embodiment, it is not always necessary to provide a spectroscope on the substrate processing devices 10 and 10A to which the etching end point detection device 20 is applied. Even if it is provided, it should be used only when learning the classifier 25. The spectroscope may be removed after learning. However, the present invention is not limited to the mode in which the spectroscope is used when learning the classifier 25.
Generally, an optical window made of a transparent material such as quartz glass is provided on the side wall of the chamber 1 in order to guide the light generated in the chamber 1 to a spectroscope installed outside the chamber 1. The optical window may be etched by the plasma in the chamber 1 to roughen the surface, or may become cloudy due to the adhesion of reaction products in the chamber 1. When the optical window becomes cloudy, the amount of light detected by the spectroscope decreases, which may reduce the accuracy of detecting the end point of etching by the spectroscope. Therefore, for example, when the existing substrate processing devices 10 and 10A are already provided with a spectroscope, the etching end point detection device 20 assists the etching end point detection with the etching end point detection (for example, the spectroscope). It is also possible to adopt a mode used as (use for issuing an alarm).

1 ... Chamber 2 ... Mounting table 10, 10A ... Substrate processing device 20 ... Etching end point detection device 21 ... Process log acquisition unit 22 ... Detection unit 23 ... Normalization unit 24 ... Imaging unit 25 ... Classifiers 100, 200 ... Board processing system W ... Board

0004
In the process of executing the etching process, the input data created from the sequentially acquired process log data is input to the classifier of the detection unit, so that the input data is before or after the end point of etching. Is output by the classifier, which makes it possible to detect the end point of etching.
As described above, according to the etching end point detection device according to the present invention, when detecting the end point of etching, input data other than the measured value related to the light generated in the chamber is created based on the process log data, and this input The end point of etching can be detected only by using the data. That is, it is not necessary to use a spectroscope when detecting the end point of etching.
[0012]
The substrate processing apparatus to which the etching end point detecting apparatus according to the present invention is applied is not limited to the plasma processing apparatus. For example, it can be applied to a sacrificial layer etching device using anhydrous HF gas and alcohol, which could only be time-etched (etching performed for a predetermined fixed time), a sacrificial layer etching device using XeF ₂ gas, and the like. Is.
Further, as the classifier, various configurations such as a neural network and a support vector machine can be adopted as long as they can be generated by using machine learning.
[0013]
Preferably, the detection unit compares the output of the classifier with a predetermined threshold value and detects the end point of etching in the substrate processing apparatus according to the magnitude thereof.
[0014]
Preferably, the process log acquisition unit acquires the process log data at a predetermined sampling cycle, the detection unit creates the input data for each sampling cycle, and the classifier creates the input data for each sampling cycle. The input data is input, and which is before or after the end point of etching in the substrate processing apparatus is output for each sampling cycle.
[0015]
Preferably, the classifier outputs that it is before the end point of the etching when the input data created from the process log data acquired before the end point of the etching is given as the input of the teacher data. , Teacher data

0005
Generated using machine learning to output that it is after the end point of the etching when the input data created from the process log data including the one acquired after the end point of the etching is given as the input. Has been done.
[0016]
According to the preferred configuration described above, the classifier was given input data created from process log data acquired before the end of etching as input of teacher data (a combination of known inputs and outputs to the classifier). In the case, to output that it is before the end point of etching (specifically, a numerical value indicating that it is before the end point of etching, for example, "0") (that is, as an output of teacher data, for example, " Given "0"), it is generated using machine learning. Further, the classifier shall be after the end point of etching when the input data created from the process log data including the data acquired after the end point of etching is given as the input of the teacher data (specifically, the classifier is after the end point of etching. It is generated using machine learning to output a numerical value indicating that it is after the end point of etching, eg, "1") (ie, giving, for example, "1" as the output of the teacher data). The latter input of teacher data, "input data created from process log data including those acquired after the end point of etching" is input data created only from process log data acquired after the end point of etching. It also means that it may be input data created from process log data straddling before and after the end point of etching.
By performing machine learning using the above-mentioned teacher data to generate a classifier, there is no need to study complicated detection logic or complicated adjustment of parameters such as threshold values, and classification after machine learning. By simply inputting the input data created from the process log data into the container, it is possible to easily detect whether it is before or after the end point of etching.
[0017]
In the above preferred configuration, it is determined, for example, whether the process log data that is the basis of the input data used as the teacher data is acquired before the end point of etching or after the end point of etching. Judgment is made by measuring the intensity of light having a predetermined wavelength with a spectroscope as in the past.

0006
Just do it. That is, the end point of etching is detected using a spectroscope, and the end point of etching detected by this spectroscope is set to be true, and the process log data is acquired before the true end point, or the true end point. It suffices to determine whether it was acquired later.
Specifically, for example, when the substrate processing device itself for detecting the end point of etching in the etching end point detection device according to the present invention is provided with a spectroscope, the spectroscope is used as a basis for input data used as teacher data. It may be determined whether the process log data to be obtained is before or after the end point of etching. Further, for example, when the substrate processing apparatus for detecting the etching end point in the etching end point detecting apparatus according to the present invention does not have a spectroscope, it is classified by machine learning using another substrate processing apparatus equipped with a spectroscope. It is also possible to generate a device and use this classifier for the etching end point detection device according to the present invention.
Further, the determination is not necessarily limited to the case of using a spectroscope. For example, when acquiring teacher data, the surface of the substrate being etched is observed, and the color difference of the substrate surface before and after the end point of etching is used. It is also conceivable to determine whether the process log data is before or after the end point of etching.
[0018]
In the etching end point detecting device according to the present invention, when the substrate processing device is an etching device that etches the substrate, the etching end point detected by the detection unit is the etching end point applied to the substrate.
[0019]
Further, in the etching end point detection device according to the present invention, when the substrate processing device is a film forming device that forms a film on the substrate, the etching end point detected by the detection unit forms a film on the substrate. After that, it is set as the end point of the etching performed to remove the film composition adhering to the inside of the chamber.
[0020]
The etching end point detection device according to the present invention is preferably used when the substrate processing device is a plasma processing device that applies plasma treatment to the substrate.
[0021]
According to the results of diligent studies by the present inventors, when the substrate processing device is a plasma processing device, among various process log data, the pressure in the exhaust pipe, the valve opening degree of the automatic pressure control device, and the upper part. Matching position of matching unit and

0007
The matching position of the lower matching unit is particularly liable to change before and after the end point of etching. This is because when the etching is completed, the object to be etched (the substrate or the film composition) disappears, so that the state of the plasma changes. Therefore, it is preferable to use at least these process log data to detect the end point of etching. However, since the pressure in the exhaust pipe and the valve opening degree of the automatic pressure control device change in conjunction with each other, it is considered that only one of them may be used.
[0022]
That is, the substrate processing apparatus is a plasma processing apparatus that applies a plasma processing apparatus to the substrate placed on a mounting table arranged in the chamber, and the substrate processing apparatus surrounds the chamber. An upper high-frequency power source that applies high-frequency power to the coil or the upper electrode via an upper matching unit, and an upper high-frequency power source that is arranged in the chamber in parallel with the above-mentioned pedestal. A lower high-frequency power source that applies high-frequency power via a lower matching unit, an exhaust pipe that communicates with the chamber, and an exhaust pipe that is provided in the exhaust pipe to control the pressure in the chamber by adjusting the valve opening degree. When the automatic pressure control device is provided, at least the pressure in the exhaust pipe or the valve opening degree of the automatic pressure control device is matched with the upper matching unit and / or the lower matching unit in the process log data. The location and is preferably included.
[0023]
In the etching end point detection device according to the present invention, as the input data created from the process log data, the process log data can be used as it is without being processed, or the processed process log data can be used. It is possible. For example, when the classifier is a neural network, the advantage of the neural network in image recognition is utilized, and as an example of the latter, there is a case where image data obtained by processing process log data is used as input data. ..
Specifically, the detection unit is based on the process log data acquired by the process log acquisition unit, and one axis is the type of the process log data.

0008
It is conceivable to create image data that is an image of a graph in which the other axis orthogonal to the one axis is the value of the process log data, and use the image data as input data to the classifier.
[0024]
In the etching end point detection device according to the present invention, preferably, the detection unit has a maximum value of 1 and a minimum value of 0 for each type of process log data with respect to the process log data acquired by the process log acquisition unit. Is performed, and input data to the classifier is created based on the process log data after the normalization.
[0025]
The value of the process log data varies greatly depending on the type of process log data such as pressure, temperature, and flow rate. In addition, the value will differ depending on the unit used. Therefore, if the values of the process log data of each type are used as they are when detecting the end point of etching, the detection accuracy may be affected. In order to avoid this, it is preferable to normalize the values of each type of process log data so that they all fluctuate within a certain range.
Specifically, the detection unit normalizes the process log data acquired by the process log acquisition unit so that the maximum value becomes 1 and the minimum value becomes 0 for each type of the process log data. It is preferable to create input data to the classifier based on the process log data after normalization.
[0026]
According to the above preferable configuration, the value of the process log data after normalization fluctuates within a certain range of 0 to 1 regardless of the type of process log data. Therefore, the input data created based on this is used. By using it as a classifier, it can be expected that a decrease in detection accuracy can be avoided.
[0027]
Further, in order to solve the above-mentioned problems, the present invention is characterized by comprising a substrate processing apparatus for processing a substrate arranged in a chamber and an etching end point detecting apparatus according to any one of the above. It is also provided as a processing system.
[0028]
Further, in order to solve the above-mentioned problems, the present invention has a process log acquisition step of acquiring process log data of a substrate processing apparatus for processing a substrate arranged in a chamber, and a process log acquired by the process log acquisition step. data

0009
Based on the above, input data other than the measured values related to the light generated in the chamber is created, and the detection step includes a detection step of detecting the end point of etching in the substrate processing apparatus based on the input data. , Using the classifier generated by using machine learning, input the input data to the classifier, and output from the classifier whether it is before or after the end point of etching in the substrate processing apparatus. It is also provided as a featured etching end point detection method.
[0029]
Further, in order to solve the above-mentioned problems, the present invention has a value other than the measured values related to the light generated in the chamber, which is created based on the process log data of the substrate processing apparatus for processing the substrate arranged in the chamber. It is also provided as a classifier generated by using machine learning, in which input data is input and which is before or after the end point of etching in the substrate processing apparatus.
Effect of the invention [0030]
According to the present invention, it is not necessary to use a spectroscope when detecting the end point of etching.
Brief description of drawings [0031]
FIG. 1 is a diagram schematically showing a schematic configuration of a substrate processing system according to a first embodiment of the present invention.
FIG. 2 is an explanatory diagram illustrating the operation of the normalization unit and the imaging unit shown in FIG.
FIG. 3 is a diagram schematically showing a schematic configuration and operation of the classifier shown in FIG. 1.
FIG. 4 shows the results of a test using the substrate processing system shown in FIG.
FIG. 5 is a diagram schematically showing a schematic configuration of a substrate processing system according to a second embodiment of the present invention.
FIG. 6 shows the results of a test using the substrate processing system shown in FIG.
A mode for carrying out the invention [0032]
Hereinafter, the etching end point detection device according to the embodiment of the present invention and the substrate processing system provided with the etching end point detection device will be described with reference to the accompanying drawings.
[0033]
<First Embodiment>
FIG. 1 schematically shows a schematic configuration of a substrate processing system according to a first embodiment.

0010
It is a figure. FIG. 1A is an overall configuration diagram of a substrate processing system, and FIG. 1B is a block diagram showing a schematic configuration of an etching end point detection device. In FIG. 1A, the parameters to be measured are shown surrounded by a broken line rectangle.
As shown in FIG. 1A, the substrate processing system 100 according to the first embodiment includes a substrate processing apparatus 10 and an etching end point detecting apparatus 20.
[0034]
The substrate processing apparatus 10 of the first embodiment is an apparatus including a chamber 1 and a mounting table 2 arranged in the chamber 1 and performing plasma processing on the substrate W mounted on the mounting table 2. More specifically, the substrate processing apparatus 10 of the first embodiment is an inductively coupled plasma (ICP) type plasma etching apparatus that etches the substrate W as plasma processing.
[0035]
A processing gas for generating plasma is supplied from a gas supply source (not shown) into the chamber 1 of the substrate processing apparatus 10. In FIG. 1A, the gas No. 1-Gas No. The configuration which can supply 6 kinds of processing gases up to 6 is illustrated. However, when the etching process is performed, the etching process is not limited to the case where all six types of processing gases are used, and the etching process can be performed using any one or more types of processing gases. The flow rate of each processed gas to be supplied is measured by a mass flow controller (MFC) 11 provided in the flow path from the gas supply source to the chamber 1. Further, the chamber 1 is provided with heaters (not shown) for heating the wall surface of the chamber 1 at appropriate locations, and the temperature of the heaters at each location (temperature No. 1-1 shown in FIG. 1A). ~ No. 1-4) is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the pressure in the chamber 1 is measured by the vacuum gauge 12.
[0036]
The substrate processing apparatus 10 includes a coil 3 arranged in the chamber 1 so as to surround the chamber 1 (in FIG. 1A, for convenience, only a cross section of the coil 3 located on the left side is shown). High-frequency power (upper high-frequency power) is applied to the coil 3 from the upper high-frequency power supply 4 via the upper matching unit 5. By applying the upper high frequency power to the coil 3, the processing gauge supplied into the chamber 1

0011
Is turned into plasma. The upper high-frequency power applied by the upper high-frequency power supply 4 and the matching position of the upper matching unit 5 (constants such as variable capacitors and variable coils included in the upper matching unit 5) are known measuring instruments (not shown). ).
[0037]
High-frequency power (lower high-frequency power) is applied to the mounting table 2 from the lower high-frequency power supply 6 via the lower matching unit 7. By applying the lower high-frequency power to the mounting table 2, a bias potential is applied between the mounting table 2 and the plasma in the chamber 1, and the ions in the plasma are accelerated to the substrate W mounted on the mounting table 2. Pull in. As a result, the substrate W is etched. The lower high-frequency power applied by the lower high-frequency power supply 6 and the matching position of the lower matching unit 7 (constants such as variable capacitors and variable coils included in the lower matching unit 7) are known measuring instruments (not shown). ).
[0038]
During the execution of the plasma treatment, the mounting table 2 is cooled by the chiller 8. The temperature of the chiller 8 is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, during the execution of the plasma treatment, He gas is supplied to the back surface of the substrate W, and the substrate W is cooled by this He gas. At this time, the pressure / flow rate of the supplied He gas is measured by a pressure / flow meter 9 provided in the flow path from the He gas supply source (not shown) to the back surface of the substrate W (upper surface of the mounting table 2). To.
[0039]
The reaction products and the like generated in the chamber 1 by executing the plasma treatment are exhausted to the outside of the chamber 1 through the exhaust pipe 17 communicating with the chamber 1. The exhaust pipe 17 includes an automatic pressure control device (Auto Pressure Controller, APC) 13 that controls the pressure in the chamber 1 by adjusting the valve opening degree, and a first pump (turbo molecule) for exhausting the reaction product. A pump) 14 and a second pump (dry pump, rotary pump, etc.) 15 that assists the first pump 14 are provided. The temperature of the automatic pressure control device 13 (temperature No. 1-5 shown in FIG. 1A) and the temperature of the first pump 14 (temperature No. 1-6 shown in FIG. 1A) are different from each other. It is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the exhaust pipe 17 has a heater (not shown) for heating the exhaust pipe 17.

0012
Are provided at appropriate locations (for example, between the first pump 14 and the second pump 15), and the temperature of the heater at each location (temperatures No. 1-7 and No. 1 shown in FIG. 1A). -8) is measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the valve opening degree (APC opening degree) of the automatic pressure control device 13 is measured by a known measuring instrument (not shown) such as an encoder. Further, the pressure (foreline pressure) in the exhaust pipe 17 located between the first pump 14 and the second pump 15 is measured by the vacuum gauge 16.
[0040]
The etching end point detection device 20 is a device that is electrically connected to the substrate processing device 10 having the above configuration and detects the end point of etching applied to the substrate W in the substrate processing device 10.
[0041]
As shown in FIG. 1B, the etching end point detection device 20 includes a process log acquisition unit 21 and a detection unit 22, and is composed of, for example, a computer.
The process log acquisition unit 21 is electrically connected to a measuring instrument (for example, a mass flow controller 11) that measures each of the above-mentioned measured values with reference to FIG. 1 (a) by wire or wirelessly (FIG. 1 (a). ), For convenience, shows a state in which only the pressure / flow meter 9, the mass flow controller 11 and the vacuum gauge 12 are connected by wire), and the measurement data sequentially input from each measuring instrument is subjected to a predetermined sampling cycle (). For example, it has a function of acquiring (A / D conversion) in 1 second). The process log acquisition unit 21 is stored in, for example, an A / D conversion board mounted on the computer, a memory such as a ROM or RAM included in the computer, or stored in the memory, and causes the CPU to operate as the process log acquisition unit 21. It consists of a program to be executed. Each acquired measured value and each set value corresponding to each measured value are used as process log data for detecting the end point of etching by the detection unit 22.
In the first embodiment, all the measured values shown in FIG. 1A are used as process log data for detecting the end point of etching, but the present invention is not limited to this. However, at least the foreline pressure or APC opening and the upper mat

0013
It is preferable to use the matching position of the ching unit 5 and / or the lower matching unit 7.
[0042]
When the etching end point detection device 20 also has a function as a control device generally used for controlling the operation of the substrate processing device 10 (when the control device is also used as the etching end point detection device 20). Each setting value constituting the process log data is stored in advance in the etching end point detection device 20 (process log acquisition unit 21). When the etching end point detection device 20 is separate from the above control device and both are electrically connected, each set value stored in advance in the control device is the etching end point detection device 20 (process log acquisition unit). It will be transmitted to 21). Further, in the first embodiment, the case where the etching end point detection device 20 is directly connected to each measuring device is illustrated, but the above control device and each measuring device are directly connected, and each measurement acquired by the control device is illustrated. It is also possible to adopt a configuration in which the value is transmitted to the etching end point detection device 20.
[0043]
The detection unit 22 creates input data from the process log data sequentially (for example, every second) acquired by the process log acquisition unit 21, and detects the end point of etching in the substrate processing apparatus 10 based on the input data. Is. The detection unit 22 is composed of, for example, a memory such as a ROM or RAM provided in the computer, or a program stored in the memory and causing the CPU to execute the operation as the detection unit 22.
[0044]
The detection unit 22 includes a classifier 25. The detection unit 22 of the first embodiment further includes a normalization unit 23 and an imaging unit 24 as a preferable configuration. The normalization unit 23, the imaging unit 24, and the classifier 25 are also stored in, for example, a memory such as a ROM or RAM provided in the computer, or a program stored in the memory and causing the CPU to execute operations as the respective units 23 to 25. It is composed.
[0045]
FIG. 2 is an explanatory diagram illustrating the operation of the normalization unit 23 and the imaging unit 24. FIG. 2A is a diagram for explaining the operation of the normalization unit 23, and FIGS. 2B and 2C are diagrams for explaining the operation of the imaging unit 24.
The left figure of FIG. 2A shows the process acquired by the process log acquisition unit 21.

0014.
It is a figure which shows the log data schematically. The parameters 1 to N shown in FIG. 2 (a) are, for example, the gas No. 1 in which parameter 1 is measured by the mass flow controller 11 shown in FIG. 1 (a). The flow rate of No. 1 and the parameter N is the temperature No. 1 shown in FIG. 1 (a). It means the type of process log data such as 1-8. X _ij (i = 1 to N, j = 1 to M) shown in the left figure of FIG. 2 (a) is acquired for parameter i when the process time (elapsed time from the start of etching) is j [sec]. It means the value of the process log data. For example, X ₁₁ is the value of the process log data acquired when the process time is 1 [sec] for parameter 1, and X _NM is acquired when the process time is M [sec] for parameter N. The value of the process log data.
[0046]
The normalization unit 23 calculates in advance the maximum value MAX _i and the minimum value MIN _i of the process log data in the total process time (1 to M [sec]) for each type of process log data (for each parameter i). For example, the maximum value for parameter 1 is MAX ₁ , the minimum value is MIN ₁ , the maximum value for parameter N is MAX _N , and the minimum value is MIN _N. The maximum value MAX _i and the minimum value MIN _i are not calculated using the process log data acquired when etching one substrate W, but are not calculated using the process log data acquired when etching one substrate W, but are performed during learning of the classifier 25 described later. It is preferable to calculate in advance using the process log data acquired for a plurality of substrates W etched by the same recipe (plasma treatment conditions). The maximum value MAX _i and the minimum value MIN _{i for} each type (parameter i) of the calculated process log data are stored in the normalization unit 23.
[0047]
Then, the normalization unit 23 normalizes the process log data X _ij sequentially acquired by the process log acquisition unit 21 so that the maximum value is 1 and the minimum value is 0 for each type of process log data (for each parameter i). Perform the conversion.
Specifically, based on the following equation (1), the normalized process log data _Yij is calculated as shown in the right figure of FIG.
Y _ij = (X _ij- MIN _i ) / (MAX _i- MIN _i ) ... (1)
In the above formula (1), i = 1 to M and j = 1 to N.

0015.
From the above equation (1), it is clear that when X _ij = MAX _i , Y _ij = 1 and when X _ij = MIN _i , Y _ij = 0 is normalized.
[0048]
The imaging unit 24 creates input data to the classifier 25 based on the process log data after normalization.
Specifically, the imaging unit 24 is uniaxial (FIG. 2 (b), as shown in the right figure of FIG. 2 (b), based on the process log data after normalization shown in the left figure of FIG. 2 (b). ) Is the type of process log data (parameters 1 to N) in the example shown on the right, and the other axis (vertical axis in the example shown on the right in FIG. 2B) is normal. Image data that is an image of a graph (bar graph) that is the value _Yij of the process log data after conversion is sequentially created (for example, every second).
The type of image data is not limited to the monochrome shading image as shown in the right figure of FIG. 2B, and any image data such as a binarized image or a color image can be created.
[0049]
Next, the imaging unit 24 of the first embodiment divides the created image data into a predetermined pixel region composed of a plurality of pixels.
Specifically, as shown in FIG. 2C, the imaging unit 24 divides the image data into K in each of the uniaxial (horizontal axis) direction and the other axis (vertical axis) direction, and the pixel area _Aij (i). = 1 to K, j = 1 to K). Then, the imaging unit 24 (the average value of the density values of a plurality of pixels constituting the pixel region _{A ij)} the average density value for each pixel region _{A ij I ave (A ij)} (i = 1~K, j = 1 to K) are calculated. This average concentration value I _ave (A _ij ) is used as input data to the classifier 25.
[0050]
When the imaging unit 24 creates a color image (color image of three RGB colors), the imaging unit 24 calculates an average density value for each color image, and all of them are input data to the classifier 25. Used.
Further, in the first embodiment, an example in which the imaging unit 24 divides the image data into pixel regions is shown, but the present invention is not limited to this, and the density values of each pixel constituting the image data are classified as they are. It can also be used as input data to the vessel 25.

0016.
[0051]
FIG. 3 is a diagram schematically showing a schematic configuration and operation of the classifier 25.
As shown in FIG. 3, the classifier 25 of the first embodiment is composed of a neural network having an input layer, an intermediate layer, and an output layer. Although FIG. 3 illustrates a configuration having two intermediate layers, the neural network that can be used as the classifier of the present invention is not limited to this, and a configuration having an arbitrary number of intermediate layers is used. It can be adopted. Further, the number of nodes in each layer shown in FIG. 3 (parts indicated by “◯” in FIG. 3) is merely an example, and the number of nodes in the neural network that can be used as the classifier of the present invention is limited to those shown in the figure. Absent.
[0052]
The classifier 25 is a substrate when the image data created by the imaging unit 24 (specifically, the average density value I _ave (A _ij ) of each pixel area A _ij ) is input to the input layer as input data. It is a configuration generated by machine learning so as to output from the output layer (output the output value OUT) which is before or after the end point of etching in the processing apparatus 10.
[0053]
Specifically, at the time of learning the classifier 25, input data (image data) created from process log data acquired before the end point of etching is given as input of teacher data, and output of teacher data combined with the input is given. As a result, machine learning is given so that OUT = 0 is output from the output layer when the input is input to the input layer by giving that it is before the end point of etching (OUT = 0 in the first embodiment). I do.
Further, as the input of the teacher data, the input data (image data) created from the process log data acquired after the end point of etching is given, and the output of the teacher data combined with the input is after the end point of etching (No. 1). In one embodiment, when OUT = 1) is given and the input is input to the input layer, machine learning is performed so that OUT = 1 is output from the output layer.
[0054]
When the substrate processing device 10 is provided with a spectroscope, the teacher data may be acquired using the spectroscope, and when the substrate processing device 10 is not provided with the spectroscope, the spectroscope is provided. You can get the teacher data using the board processing device of

[0017]
I. Also, the machine learning of the classifier 25 is not limited to one time. If necessary, the classifier 25 can be re-learned using the new teacher data, or the classifier 25 can be re-learned by adding new teacher data to the conventional teacher data.
[0055]
When the classifier 25 after learning as described above detects the end point of etching based on the input data sequentially input, the input data (image data) is sequentially input to the input layer of the classifier 25 and classified. The output value OUT is output from the output layer of the device 25. Unlike the learning time, the output value OUT value at the time of detection is 0 ≦ OUT ≦ 1.
When 0 ≦ OUT <0.5 (when rounded to 0 with one decimal point), the detection unit 22 of the first embodiment determines that it is before the end point of etching, and 0.5 ≦ OUT. When ≦ 1 (when rounded to 1 with one decimal point), it is determined that it is after the end point of etching.
[0056]
The etching end point detection device 20 having the configuration described above sequentially detects the end points of etching applied to the substrate W in the substrate processing device 10.
[0057]
Hereinafter, an example of the result of performing a test in which the substrate W is etched by the substrate processing apparatus 10 of the substrate processing system 100 according to the first embodiment and the end point of etching is detected by the etching end point detecting device 20 will be described.
[0058]
In the above test, first, 19 substrates W (Si substrates) are etched with SF ₆ gas, and the etching time of each substrate W (from the start of etching of the substrate W to the end of etching to the end of overetching). Input data (image data) to the classifier 25 was created for each 1-second sampling cycle (50 seconds). Whether the input data of each sampling cycle is the input data created from the process log data acquired before the end point of etching or the input data created from the process log data acquired after the end point of etching is described above. Since the substrate processing device 10 used in the test is provided with a spectroscope, the determination was made based on whether or not the intensity of light having the emission wavelength of SiF measured by this spectroscope was equal to or less than the reference value. The classifier 25 was machine-learned using the teacher data collected as described above. Same teacher's day on classifier 25 after learning

0018
When the input data of the data was input and it was determined whether it was before or after the end point of etching, the correct answer rate (number of correct answers / number of judgments × 100) was 99.89%. Since the time required for one determination by the classifier 25 was sufficiently shorter than the overetching time, the etching end point detection device 20 detected the etching end point (determined to be after the etching end point). It can be said that there is no problem even if the etching is completed.
[0059]
Next, in the above test, another 6 substrates (No. 1-1 to No. 1-6) of substrates W (Si substrates) were etched with SF ₆ gas using the same recipe, and the etching time of each substrate W was In, input data (image data) to the classifier 25 was created for each sampling cycle of 1 second, and the data was input to the classifier 25 after the learning to determine whether it was before or after the end point of etching. At this time, as in the case of the above-mentioned learning, using the spectroscope provided in the substrate processing apparatus 10, the input data of each sampling cycle is before the end point of etching or after the end point of etching. It was judged whether there was.
[0060]
FIG. 4 shows the results of the above test. “0” shown in FIG. 4 is determined by the etching end point detecting device 20 to be before the etching end point, and “1” is determined by the etching end point detecting device 20 to be after the etching end point. is there. In FIG. 4, the columns that have been hatched and surrounded by thick lines are determined to be after the end point of etching by using a spectroscope.
As shown in FIG. 4, it was No. 4 that made a judgment different from the judgment using the spectroscope. Judgment of the substrate W of 1-3 at 38 seconds and No. Only the judgment of the substrate W of 1-5 at 38 seconds was made, and the correct answer rate was 99.35% (= 306/308 × 100). Therefore, according to the etching end point detecting device 20 of the substrate processing system 100 according to the first embodiment, it can be said that the etching end point of the substrate W can be detected with high accuracy.
[0061]
<Second Embodiment>
FIG. 5 is a diagram schematically showing a schematic configuration of a substrate processing system according to a second embodiment. In addition, also in FIG. 5, the measurement is performed in the same manner as in FIG. 1 (a) described above.

0019
The parameter is shown by enclosing it in a dashed rectangle. In FIG. 5, the illustration of the configuration corresponding to FIG. 1 (b) described above is omitted.
As shown in FIG. 5, the substrate processing system 200 according to the second embodiment includes a substrate processing device 10A and an etching end point detecting device 20.
Hereinafter, the differences from the substrate processing system 100 according to the first embodiment will be mainly described, and the same components as those of the substrate processing system 100 according to the first embodiment are designated by the same reference numerals and the description thereof will be omitted.
[0062]
The substrate processing apparatus 10A of the second embodiment is an apparatus including a chamber 1 and a mounting table 2 arranged in the chamber 1 and performing plasma processing on the substrate W mounted on the mounting table 2. More specifically, the substrate processing apparatus 10A of the second embodiment is a capacitively coupled plasma (CCP) type plasma film forming apparatus that forms a film on the substrate W as plasma processing.
Therefore, unlike the substrate processing apparatus 10 of the first embodiment, instead of the coil 3 (see FIG. 1A), an upper electrode 18 is provided in the chamber 1 so as to face parallel to the mounting table 2. ..
[0063]
A processing gas for generating plasma is supplied from a gas supply source (not shown) into the chamber 1 of the substrate processing apparatus 10A. In FIG. 5, the gas No. 1-Gas No. The configuration which can supply 6 kinds of processing gases up to 6 is illustrated. However, it is not limited to the case where all of the six types of processing gases are used when performing the film forming process or when cleaning the film composition adhering to the chamber 1 after the film forming process, and any one of them is used. It is possible to perform film formation processing and cleaning using more than one type of processing gas.
[0064]
High-frequency power (upper high-frequency power) is applied to the upper electrode 18 from the upper high-frequency power supply 4 via the upper matching unit 5. Further, high frequency power (lower high frequency power) is applied to the mounting table 2 from the lower high frequency power supply 6 via the lower matching unit 7. As a result, the processing gas supplied into the chamber 1 is turned into plasma, and the generated plasma moves toward the mounting table 2, so that a film is formed on the substrate W mounted on the mounting table 2. .. Film composition after film formation

0020
When cleaning the chamber 1, the generated plasma moves toward the inner surface of the chamber 1, so that the film composition adhering to the inside of the chamber 1 is removed by etching.
[0065]
Unlike the substrate processing apparatus 10 of the first embodiment, the substrate processing apparatus 10A of the second embodiment does not include a chiller 8, a pressure / flow meter 9, a first pump (turbo molecular pump) 14, and a vacuum gauge 16.
[0066]
In the substrate processing system 200 according to the second embodiment, the flow rate of each processing gas to be supplied is measured by a mass flow controller 11 provided in the flow path from the gas supply source to the chamber 1. Further, the temperature of the heater (not shown) provided at an appropriate position on the wall surface of the chamber 1 (temperatures No. 2-1 to No. 2-3 shown in FIG. 5) is a known measuring instrument such as a thermoelectric pair (not shown). (Not shown). Further, the pressure in the chamber 1 is measured by the vacuum gauge 12.
Further, the upper high-frequency power applied by the upper high-frequency power supply 4 and the matching position of the upper matching unit 5 (constants such as variable capacitors and variable coils included in the upper matching unit 5) are known measuring instruments (not shown). ).
Further, the lower high-frequency power applied by the lower high-frequency power supply 6 and the matching position of the lower matching unit 7 (constants such as variable capacitors and variable coils included in the lower matching unit 7) are known measuring instruments (not shown). ).
Further, the temperature of the automatic pressure control device 13 (temperature No. 2-5 shown in FIG. 5) and the temperature of a heater (not shown) provided at an appropriate position in the exhaust pipe 17 (temperature No. 2 shown in FIG. 5). 2-4, No. 2-6, No. 2-7) are measured by a known measuring instrument (not shown) such as a thermoelectric pair. Further, the APC opening degree of the automatic pressure control device 13 is measured by a known measuring instrument (not shown) such as an encoder.
[0067]
The etching end point detection device 20 included in the substrate processing system 200 according to the second embodiment has the same configuration as that of the first embodiment, and is described above with reference to FIG.

0021.
It is electrically connected to a measuring instrument (for example, a mass flow controller 11) that measures a measured value by wire or wirelessly. The etching end point detection device 20 acquires measurement data sequentially input from each measuring instrument at a predetermined sampling cycle (for example, 1 second), and each acquired measurement value and each set value corresponding to each measurement value are set. , Used as process log data to detect the end point of etching.
In the second embodiment, all the measured values shown in FIG. 5 are used for determination as process log data, but the present invention is not limited to this. However, it is preferable to use at least the APC opening degree and the matching position of the upper matching unit 5 and / or the lower matching unit 7.
[0068]
Unlike the first embodiment, the etching end point detected by the etching end point detection device 20 of the second embodiment is executed to remove the film composition adhering to the chamber 1 after forming the film on the substrate W. This is the end point of etching.
[0069]
Hereinafter, during etching (cleaning) performed to remove the film composition adhering to the inside of the chamber 1 after forming a film on the substrate W by the substrate processing apparatus 10A of the substrate processing system 200 according to the second embodiment. , An example of the result of the test for detecting the end point of etching by the etching end point detection device 20 will be described.
[0070]
In the above test, firstly, C ₄ F ₈ and 13 cleaning times in the chamber 1 using gas, over-etching is completed through the etching end from the etching start of coating composition adhered to the cleaning of the etching time (chamber 1 Input data (image data) to the classifier 25 was created for each sampling cycle of 1 second (about 150 seconds until the etching). Whether the input data of each sampling cycle is the input data created from the process log data acquired before the end point of etching or the input data created from the process log data acquired after the end point of etching is described above. Since the substrate processing device 10A used in the test is provided with a spectroscope, it was determined by whether or not the intensity of light having an emission wavelength of F measured by this spectroscope was equal to or higher than the reference value. The classifier 25 was machine-learned using the teacher data collected as described above. After learning

0022.
When the input data of the same teacher data was input to the classifier 25 and the judgment was made, the correct answer rate was 99.39%. Since the time required for one determination by the classifier 25 was sufficiently shorter than the overetching time, the etching end point detection device 20 detected the etching end point (determined to be after the etching end point). It can be said that there is no problem even if the etching is completed after that.
[0071]
Next, in the above test, the inside of the chamber 1 was cleaned with the same recipe 5 times (No. 2-1 to No. 2-5) C ₄ F ₈ gas at different timings, and at the etching time of each cleaning. Input data (image data) to the classifier 25 was created for each sampling cycle of 1 second, and the data was input to the classifier 25 after the learning to determine whether it was before or after the end point of etching. At this time, as in the case of the above-mentioned learning, using the spectroscope provided in the substrate processing apparatus 10A, the input data of each sampling cycle is before the end point of etching or after the end point of etching. It was judged whether there was.
[0072]
FIG. 6 shows the results of the above test. “0” shown in FIG. 6 is determined by the etching end point detecting device 20 to be before the etching end point, and “1” is determined by the etching end point detecting device 20 to be after the etching end point. is there. In FIG. 6, the diagonally hatched columns and the columns surrounded by thick lines are determined by using a spectroscope after the end point of etching.
As shown in FIG. 6, it was No. 6 that made a judgment different from the judgment using the spectroscope. Judgment at the time of 35 to 39 seconds of cleaning of 2-1 and No. Judgment at 3 seconds of cleaning 2-3 and No. Only the judgment at 44 seconds of 2-5 cleaning was made, and the correct answer rate was 99.1% (= 743/750 × 100). Therefore, according to the etching end point detection device 20 of the substrate processing system 200 according to the second embodiment, it can be said that the end point of etching (cleaning) of the film composition adhering to the chamber 1 can be detected with high accuracy.
[0073]
In the first embodiment and the second embodiment described above, the detection unit 22

[0023]
Although the configuration including the normalization unit 23 has been described as an example, the present invention is not limited to this. It is also possible that the detection unit 22 does not include the normalization unit 23, and the process log data acquired by the process log acquisition unit 21 is used as it is without normalization to create input data.
[0074]
Further, in the first embodiment and the second embodiment, the configuration in which the detection unit 22 includes the imaging unit 24 has been described as an example, but the present invention is not limited to this. The detection unit 22 does not include the imaging unit 24, and the process log data acquired by the process log acquisition unit 21 can be used as input data to the classifier 25 without being imaged after being used as it is or after being normalized. is there.
Specifically, when not imaging, for example, the process log data X _ij (i = 1 to N, j = 1 to M) shown in FIG. 2 or after normalization is applied to the input layer of the classifier 25 shown in FIG. Process log data Y _ij (i = 1 to N, j = 1 to M) will be input.
[0075]
Further, in the first embodiment and the second embodiment, the input data (image data) created in a predetermined sampling cycle (for example, 1 second) is used as the input data to the classifier 25 during both learning and determination. However, the present invention is not limited to this. For example, a graph showing changes in process log data acquired in a predetermined sampling cycle within a predetermined time (within a time corresponding to a plurality of sampling cycles) is imaged, and this image data is used as input data to the classifier 25. It can also be used. Any of the above image data will include a graph created from process log data that spans before and after the end point of etching.
[0076]
Further, in the first embodiment and the second embodiment, the configuration in which the detection unit 22 includes a single classifier 25 has been described as an example, but the present invention is not limited to this. A plurality of classifiers 25 are generated for each equivalent recipe by machine learning, and when the end point of etching is detected, the end point is detected by using the classifier 25 generated according to the corresponding recipe among the plurality of classifiers 25. It is also possible to adopt a configuration that does.

0024
[0077]
Further, according to the etching end point detection device 20 of the first embodiment and the second embodiment, it is not always necessary to provide a spectroscope on the substrate processing devices 10 and 10A to which the etching end point detection device 20 is applied. Even if it is provided, it should be used only when learning the classifier 25. The spectroscope may be removed after learning. However, the present invention is not limited to the mode in which the spectroscope is used when learning the classifier 25.
Generally, an optical window made of a transparent material such as quartz glass is provided on the side wall of the chamber 1 in order to guide the light generated in the chamber 1 to a spectroscope installed outside the chamber 1. The optical window may be etched by the plasma in the chamber 1 to roughen the surface, or may become cloudy due to the adhesion of reaction products in the chamber 1. When the optical window becomes cloudy, the amount of light detected by the spectroscope decreases, which may reduce the accuracy of detecting the end point of etching by the spectroscope. Therefore, for example, when the existing substrate processing devices 10 and 10A are already provided with a spectroscope, the etching end point detection device 20 assists the etching end point detection with the etching end point detection (for example, the spectroscope). It is also possible to adopt a mode used as (use for issuing an alarm).
Description of Code [0078]
1 ... Chamber 2 ... Mounting table 10, 10A ... Substrate processing device 20 ... Etching end point detection device 21 ... Process log acquisition unit 22 ... Detection unit 23 ... Normalization unit 24 ... Imaging unit 25 ... Classifiers 100, 200 ... Board processing system W ... Board

Claims (11)

A process log acquisition unit that acquires process log data of a substrate processing device that processes a substrate arranged in a chamber, and a process log acquisition unit.
Based on the process log data acquired by the process log acquisition unit, input data other than the measured values related to the light generated in the chamber is created, and based on the input data, the end point of etching in the substrate processing apparatus is detected. With a detector,
The detection unit includes a classifier generated by machine learning, in which the input data is input and which is before or after the end point of etching in the substrate processing apparatus.
An etching end point detection device characterized by this. When the input data created from the process log data acquired before the end point of the etching is given as the input of the teacher data, the classifier outputs that it is before the end point of the etching and outputs the teacher data. When input data created from the process log data including the one acquired after the end point of the etching is given as the input of, machine learning is used to output that it is after the end point of the etching. Have been generated
The etching end point detecting apparatus according to claim 1. The substrate processing apparatus is an etching apparatus that etches the substrate.
The end point of etching detected by the detection unit is the end point of etching applied to the substrate.
The etching end point detecting apparatus according to claim 1 or 2. The substrate processing apparatus is a film forming apparatus that forms a film on the substrate.
The end point of etching detected by the detection unit is the end point of etching executed to remove the film composition adhering to the inside of the chamber after forming a film on the substrate.
The etching end point detecting apparatus according to claim 1 or 2. The substrate processing apparatus is a plasma processing apparatus that applies plasma processing to the substrate.
The etching end point detecting apparatus according to any one of claims 1 to 4. The substrate processing apparatus is a plasma processing apparatus that applies a plasma processing apparatus to the substrate placed on a mounting table arranged in the chamber.
The substrate processing device is
A coil arranged in the chamber so as to surround the chamber or an upper electrode arranged in the chamber in parallel with the above-described pedestal
An upper high frequency power supply that applies high frequency power to the coil or the upper electrode via an upper matching unit, and
A lower high-frequency power supply that applies high-frequency power to the above-mentioned stand via a lower matching unit,
An exhaust pipe that communicates with the chamber
An automatic pressure control device provided in the exhaust pipe and controlling the pressure in the chamber by adjusting the valve opening degree is provided.
The process log data includes at least the pressure in the exhaust pipe or the valve opening degree of the automatic pressure control device and the matching position of the upper matching unit and / or the lower matching unit.
The etching end point detecting apparatus according to any one of claims 1 to 4. Based on the process log data acquired by the process log acquisition unit, the detection unit images a graph in which one axis is the type of the process log data and the other axis orthogonal to the one axis is the value of the process log data. Create the image data, and use the image data as input data to the classifier.
The etching end point detecting apparatus according to any one of claims 1 to 6. The detection unit normalizes the process log data acquired by the process log acquisition unit so that the maximum value becomes 1 and the minimum value becomes 0 for each type of process log data, and the process after the normalization. Create input data to the classifier based on the log data,
The etching end point detecting apparatus according to any one of claims 1 to 7. A substrate processing apparatus for processing a substrate arranged in a chamber, an etching end point detecting apparatus according to any one of claims 1 to 8, and an etching end point detecting apparatus.
A substrate processing system characterized by comprising. A process log acquisition process for acquiring process log data of a substrate processing device that processes a substrate arranged in a chamber, and a process log acquisition process.
Based on the process log data acquired in the process log acquisition step, input data other than the measured values related to the light generated in the chamber is created, and based on the input data, the end point of etching in the substrate processing apparatus is detected. Including the detection step,
In the detection step, using a classifier generated by using machine learning, the input data is input to the classifier, and the classifier outputs whether it is before or after the end point of etching in the substrate processing apparatus. To do,
An etching end point detection method characterized by this. Input data other than the measured values related to the light generated in the chamber, which is created based on the process log data of the substrate processing apparatus that processes the substrate arranged in the chamber, is input, and the etching in the substrate processing apparatus is performed. A classifier generated using machine learning that outputs whether it is before or after the end point.

Images